Remotely Sensed Data Characterization Classification and Accuracies
Uploaded by
Julio CovarrubiaRemotely Sensed Data Characterization Classification and Accuracies
Uploaded by
Julio CovarrubiaREMOTE SENSING HANDBOOK
V O LU M E I
REMOTELY SENSED
DATA CHARACTERIZATION,
CLASSIFICATION, AND ACCURACIES
2016 by Taylor & Francis Group, LLC
Remote Sensing Handbook
Remotely Sensed Data Characterization, Classification, and Accuracies
Land Resources Monitoring, Modeling, and Mapping with Remote Sensing
Remote Sensing of Water Resources, Disasters, and Urban Studies
2016 by Taylor & Francis Group, LLC
REMOTE SENSING HANDBOOK
V O LU M E I
REMOTELY SENSED
DATA CHARACTERIZATION,
CLASSIFICATION, AND ACCURACIES
Edited by
Prasad S. Thenkabail, PhD
United States Geological Survey (USGS)
2016 by Taylor & Francis Group, LLC
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
2016 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Version Date: 20150513
International Standard Book Number-13: 978-1-4822-1787-2 (eBook - PDF)
This book contains information obtained from authentic and highly regarded sources. Reasonable efforts have been made to publish reliable data and
information, but the author and publisher cannot assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication and apologize to copyright holders if permission
to publish in this form has not been obtained. If any copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic,
mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or
retrieval system, without written permission from the publishers.
For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact
the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides
licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment
has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation
without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
2016 by Taylor & Francis Group, LLC
I dedicate this work to my revered parents whose sacrifices gave me an education,
aswellas to all those teachers from whom I learned remote sensing over the years.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
Contents
Foreword ........................................................................................................................................................ xi
Preface......................................................................................................................................................... xvii
Acknowledgments ..................................................................................................................................... xxiii
Editor ........................................................................................................................................................ xxvii
Contributors .............................................................................................................................................. xxxi
Section i Satellites and Sensors from Different eras and their characteristics
1 Remote Sensing SystemsPlatforms and Sensors: Aerial, Satellite, UAV, Optical,
Radar,andLiDAR .......................................................................................................................................................3
Sudhanshu S. Panda, Mahesh N. Rao, Prasad S. Thenkabail, and James E. Fitzerald
Section ii Fundamentals of Remote Sensing: evolution, State of the Art,
andFuture Possibilities
2 Fundamentals of Remote Sensing for Terrestrial Applications: Evolution, Current State of
theArt, and Future Possibilities ........................................................................................................................... 61
Natascha Oppelt, Rolf Scheiber, Peter Gege, Martin Wegmann, Hannes Taubenboeck, and Michael Berger
Section iii Remote Sensing Data normalization, Harmonization,
andintersensor calibration
3 Overview of Satellite Image Radiometry in the Solar-Ref lective Optical Domain ................................ 87
Philippe M. Teillet
4 Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors with Emphasis
onTerrestrial Reference Standard Sites ............................................................................................................ 109
Philippe M. Teillet and Gyanesh Chander
5 Remote Sensing Data Normalization ................................................................................................................. 133
Rudiger Gens and Jordi Cristbal Rossell
6 Satellite Data Degradations and Their Impacts on High-Level Products ............................................... 143
Dongdong Wang
vii
2016 by Taylor & Francis Group, LLC
viii Contents
Section iV Vegetation index Standardization and cross-calibration of
Datafrom Multiple Sensors
7 Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced
VegetationIndices ................................................................................................................................................... 155
Tomoaki Miura, Kenta Obata, Javzandulam T. Azuma, Alfredo Huete, and Hiroki Yoshioka
8 Toward Standardization of Vegetation Indices ............................................................................................... 175
Michael D. Steven, Timothy J. Malthus, and Frdric Baret
Section V image Processing Methods and Approaches
9 Digital Image Processing: A Review of the Fundamental Methods and Techniques ........................... 197
Sunil Narumalani and Paul Merani
10 Urban Image Classification: Per-Pixel Classifiers, Subpixel Analysis, Object-Based Image
Analysis, and Geospatial Methods ..................................................................................................................... 219
Soe W. Myint, Victor Mesev, Dale A. Quattrochi, and Elizabeth A. Wentz
11 Image Classification Methods in Land Cover and Land Use ...................................................................... 231
Mutlu zdoan
12 Hyperspectral Image Processing: Methods and Approaches ...................................................................... 247
Jun Li and Antonio Plaza
13 Thermal Infrared Remote Sensing: Principles and Theoretical Background ........................................ 259
Claudia Kuenzer, Jianzhong Zhang, and Stefan Dech
14 Object-Based Image Analysis: Evolution, History, State of the Art, and Future Vision ..................... 277
Thomas Blaschke, Maggi Kelly, and Helena Merschdorf
15 Geospatial Data Integration in OBIA: Implications of Accuracy and Validity ..................................... 295
Stefan Lang and Dirk Tiede
16 Image Segmentation Algorithms for Land Categorization .......................................................................... 317
James C. Tilton, Selim Aksoy, and Yuliya Tarabalka
17 LiDAR Data Processing and Applications ........................................................................................................343
Shih-Hong Chio, Tzu-Yi Chuang, Pai-Hui Hsu, Jen-Jer Jaw, Shih-Yuan Lin, Yu-Ching Lin, Tee-Ann Teo,
Fuan Tsai, Yi-Hsing Tseng, Cheng-Kai Wang, Chi-Kuei Wang, Miao Wang, and Ming-Der Yang
Section Vi change Detection
18 Change Detection Techniques Using Multisensor Data ............................................................................... 377
Daniela Anjos, Dengsheng Lu, Luciano Dutra, and Sidnei SantAnna
Section Vii integrating Geographic information Systems (GiS) and Remote
Sensing in Spatial Modeling Framework for Decision Support
19 Geoprocessing, Workf lows, and Provenance .................................................................................................. 401
Jason A. Tullis, Jackson D. Cothren, David P. Lanter, Xuan Shi, W. Fredrick Limp, Rachel F. Linck, Sean G. Young,
and Tareefa S. Alsumaiti
20 Toward Democratization of Geographic Information: GIS, Remote Sensing, and
GNSSApplications in Everyday Life ..................................................................................................................423
Gaurav Sinha, Barry J. Kronenfeld, and Jeffrey C. Brunskill
2016 by Taylor & Francis Group, LLC
Contents ix
21 Frontiers of GIScience: Evolution, State of the Art, and Future Pathways ..............................................445
May Yuan
22 Object-Based Regionalization for Policy-Oriented Partitioning of Space .............................................. 457
Stefan Lang, Stefan Kienberger, Michael Hagenlocher, and Lena Pernkopf
Section Viii Global navigation Satellite Systems (GnSS) Remote Sensing
23 Global Navigation Satellite Systems Theory and Practice: Evolution, State of
the Art, andFuturePathways .............................................................................................................................. 481
Mohinder S. Grewal
24 Global Navigation Satellite System Ref lectometry for Ocean and Land Applications ........................ 493
Kegen Yu, Chris Rizos, and Andrew Dempster
25 Global Navigation Satellite Systems for Wide Array of Terrestrial Applications .................................509
D. Myszor, O. Antemijczuk, M. Grygierek, M. Wierzchanowski, and K.A. Cyran
Section iX crowdsourcing and Remote Sensing Data
26 Crowdsourcing and Remote Sensing: Combining Two Views of Planet Earth ...................................... 535
Fabio DellAcqua
Section X cloud computing and Remote Sensing
27 Processing Remote-Sensing Data in Cloud Computing Environments ................................................... 553
Ramanathan Sugumaran, James W. Hegeman, Vivek B. Sardeshmukh, and Marc P. Armstrong
Section Xi Google earth for Remote Sensing
28 Google Earth for Remote Sensing ....................................................................................................................... 565
John E. Bailey
Section Xii Accuracies, errors, and Uncertainties of Remote
SensingDerived Products
29 Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data ......... 583
Russell G. Congalton
Section Xiii Space Law and Remote Sensing
30 Remote Sensing Law: An Overview of Its Development and Its Trajectory in the Global Context ....... 605
P.J. Blount
Section XiV Summary
31 Remote Sensing Data Characterization, Classification, and Accuracies: Advances of the Last
50Years andaVision for the Future ................................................................................................................. 625
Prasad S. Thenkabail
Index ........................................................................................................................................................... 663
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
Foreword: Satellite Remote
Sensing Beyond 2015
Satellite remote sensing has progressed tremendously since from the advanced very-high-resolution radiometers (AVHRRs)
Landsat 1 was launched on June 23, 1972. Since the 1970s, sat- that now span more than 33years, although no one anticipated
ellite remote sensing and associated airborne and insitu mea- these valuable climate records from this instrument before the
surements have resulted in vital and indispensible observations launch of NOAA-7 in 1981 (Cracknell 1997).
for understanding our planet through time. These observations The success of data from the AVHRR led to the design of
have also led to dramatic improvements in numerical simula- the moderate-resolution imaging spectroradiometer (MODIS)
tion models of the coupled atmospherelandocean systems instruments on NASAs Earth-Observing System (EOS) of sat-
at increasing accuracies and predictive capabilities. The same ellite platforms that improved substantially upon the AVHRR.
observations document the Earths climate and are driving the The first of the EOS platforms, Terra, was launched in 2000; and
consensus that Homo sapiens is changing our climate through the second of these platforms, Aqua, was launched in 2002. Both
greenhouse gas (GHG) emissions. of these platforms are nearing their operational life, and many of
These accomplishments are the combined work of many the climate data records from MODIS will be continued with the
scientists from many countries and a dedicated cadre of engi- visible infrared imaging radiometer suite (VIIRS) instrument on
neers who build the instruments and satellites that collect Earth the polar orbiting meteorological satellites of NOAA. The first
observation (EO) data from satellites, all working toward the of these missions, the NPOES Preparation Project (NPP), was
goal of improving our understanding of the Earth. This edi- launched in 2012 with the first VIIRS instrument that is operat-
tion of the Remote Sensing Handbook (Remotely Sensed Data ing currently among several other instruments on this satellite.
Characterization, Classification, and Accuracies; Land Resources Continuity of observations is crucial for advancing our under-
Monitoring, Modeling, and Mapping with Remote Sensing; and standing of the Earths climate system. Many scientists feel that
Remote Sensing of Water Resources, Disasters, and Urban Studies) the crucial climate observations provided by remote sensing
is a compendium of information for many research areas of our satellites are among the most important satellite measurements
planet that have contributed to our substantial progress since because they contribute to documenting the current state of our
the 1970s. The remote sensing community is now using multiple climate and how it is evolving. These key satellite observations
sources of satellite and insitu data to advance our studies, what- of our climate are second in importance only to the polar orbit-
ever they may be. In the following paragraphs, I will illustrate ing and geostationary satellites needed for numerical weather
how valuable and pivotal satellite remote sensing has been in cli- prediction.
mate system study over the last five decades. The chapters in the The current state of the art for remote sensing is to combine
handbook provide many other specific studies on land, water, different satellite observations in a complementary fashion for
and other applications using EO data of the last five decades. what is being studied. Let us review climate change as an excel-
The Landsat system of Earth-observing satellites has led the lent example of using disparate observations from multiple sat-
way in pioneering sustained observations of our planet. From ellite and insitu sources to observe climate change, verify that it
1972 to the present, at least one and sometimes two Landsat sat- is occurring, and understand the various component processes:
ellites have been in operation (Irons etal. 2012). Starting with the
launch of the first NOAANASA Polar Orbiting Environmental 1. Warming of the planet, quantified by radar altimetry from
Satellites NOAA-6 in 1978, improved imaging of land, clouds, space: Remotely sensed climate observations provide the
and oceans and atmospheric soundings of temperature was data to understand our planet and what forces our climate.
accomplished. The NOAA system of polar-orbiting meteoro- The primary climate observation comes from radar altim-
logical satellites has continued uninterrupted since that time, etry that started in late 1992 with Topex/Poseidon and has
providing vital observations for numerical weather prediction. been continued by Jason-1 and Jason-2 to provide an unin-
These same satellites are also responsible for the remarkable terrupted record of global sea level. Changes in global sea
records of sea surface temperature and land vegetation index level provide unequivocal evidence if our planet is warming,
xi
2016 by Taylor & Francis Group, LLC
xii Foreword
80
Topex
70 Jason-1
Jason-2
60
60-day smoothing
50
Inverse barometer applied, GIA corrected
40
MSL (mm)
30
20
10
Rate = 3.2 0.4 mm/year
0
Seasonal signals removed
10
University of Colorado 2014
20
1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
Year
FIg u r e F.1 Warming of the planet quantified by radar altimetry from space. Sea level determined from three radar altimeters from late 1992 to
the present shows global sea level increases of ~3 mm/year. Sea level is the unequivocal indicator of the Earths climatewhen sea level rises, the
planet is warming; when sea level falls, the planet is cooling. (From Gregory, J.M. etal., J. Climate, 26(13), 4476, 2013.)
cooling, or staying at the same temperature. Radar altim- (Figure F.2). Thus, we must look to other remotely sensed
etry from 1992 to date has shown global sea level increases of climate observations to explain and confirm sea level rise.
~3mm/year, and hence, our planet is warming (Figure F.1). 3. Sea level rise of 60% is explained by a mass balance of
Sea level rise has two components, thermal expansion and melting of ice measured by GRACE satellites. Since
ice melting in the ice sheets of Greenland and Antarctica, 2002, we have measured gravity anomalies from the
and to a much lesser extent, in glaciers. Gravity Recovery and Climate Experiment Satellite
2. The Sun is not to blame for global warming, based on solar (GRACE) dual satellite system. GRACE data quantify
irradiance data from satellites. Next, we consider two very ice mass changes from the Antarctic and Greenland
different satellite observations and one insitu observing ice sheets (AIS and GIS) and concentrations of gla-
system that enable us to understand the causes of sea level ciers, such as in the Gulf of Alaska (GOA) (Luthcke
variations: total solar irradiance, variations in the Earths et al. 2013). GRACE data are truly remarkabletheir
gravity field, and the Argo floats that record ocean tem- retrieval of variations in the Earths gravity field is
perature and salinity with depth, respectively. quantitatively and directly linked to mass variations.
Observations of total solar irradiance have been made With GRACE data, we are able to determine for the first
from satellites since 1979 and show total solar irradiance time the mass balance with time of the AIS and GIS
has varied only 1 part in 500 over the past 35years, estab- and concentrations of glaciers on land. GRACE data
lishing that our Sun is not to blame for global warming show sea level rise of 60% explained by ice loss from
Total solar irradiance composite
Total solar irradiance (W/m2)
1364 ERB 1364
ACRIM II V3-0111
ACRIM I VI-8907 ACRIM III V-1004
VIRGO V6-1009
ERBS V-0508 SORCE TIM
1362 1362
1360 1360
1980 1990 2000 2010 2014
Year
FIg u r e F.2 The Sun is not to blame for global warming, based on solar irradiance data from satellites. Total solar irradiance reconstructed from
multiple instruments dates back to 1979. The luminosity of our Sun varies only 0.1% over the course of the 11-year solar cycle. (From Froehlich, C.,
Space Sci. Rev., 176(14), 237, 2013.)
2016 by Taylor & Francis Group, LLC
Foreword xiii
500 GT
GIS
AIS
GOA
GT
2003 2004 2005 2006 2007 2008 2009 2010 2011
Year
FIg ur e F.3 Sea level rise of 60% explained by mass balance of melting of ice measured by GRACE satellites. Ice mass variations from 2003
to 2010 for the Antarctic ice sheets (AIS), Greenland ice sheets (GIS) and the Gulf of Alaska (GOA) glaciers using GRACE gravity data. (From
Luthcke, S.B. etal., J. Glaciol., 59(216), 613, 2013.)
land (Figure F.3). GRACE data have many other uses, descend. At 1000 m depth, they drift for 10 days con-
such as indicating changes in groundwater storage, and tinuing their measurements of temperature and salinity.
readers are directed to the GRACE projects website if After 10days, they slowly descend to 3000 m and then
interested (http://www.csr.utexas.edu/grace/). ascend to the surface, all the time recording their mea-
4. Sea level rise of 40% is explained by thermal expansion surements. At the surface, each float transmits all the
in the planets oceans measured by insitu ~3700 drifting data collected on the most recent excursion to a geosta-
floats. The other contributor to sea level rise is thermal tionary satellite and then descends again to repeat this
expansion in the planets oceans. This necessitates using process.
diving and drifting floats in the Argo network to record
temperature with depth (Roemmich et al. 2009 and Argo temperature data show that 40% of sea level rise results
Figure F.4). Argo floats are deployed from ships; they from the warming and thermal expansion of our oceans.
then submerge and descend slowly to 1000 m depth, Combining radar altimeter data, GRACE data, and Argo data
recording temperature, pressure, and salinity as they provides a confirmation of sea level rise and shows what is
3627 floats
60N
September 30, 2014
30N
30S
60S
60E 120E 180 120W 60W 0
FIg u r e F.4 Sea level rise of 40% explained by thermal expansion in the planets oceans measured by insitu ~3700 drifting floats. This is the latest
picture of the 3627 Argo floats that were in operation on September 30, 2014. These floats provide the data needed to document thermal expansion
of the oceans. (From http://www.argo.ucsd.edu/.)
2016 by Taylor & Francis Group, LLC
xiv Foreword
responsible for it and in what proportions. With total solar Satellite-borne sensors provide simultaneous global car-
irradiance being near constant, what is driving global warm- bon cycle observations needed for quantifying carbon cycle
ing can be determined. The analysis of surface in situ air processes, that is, to measure atmospheric CO2 concentra-
temperature coupled with lower tropospheric air tempera- tions and emission sources, to measure land and ocean pho-
ture and stratospheric temperature data from remote sensing tosynthesis, to measure the reservoir of carbon in plants on
infrared and microwave sounders shows that the surface and land and its change, to measure the extent of biomass burn-
near surface are warming while the stratosphere is cooling. ing of vegetation on land, and to measure soil respiration and
This is an unequivocal confirmation that greenhouse gases decomposition, including decomposing carbon in permafrost.
are warming the planet. In addition to the required satellite observations, insitu obser-
Many scientists are actively working to study the Earths vations are needed to confirm satellite-measured CO2 concen-
carbon cycle, and there are several chapters in the handbook trations and determine soil and vegetation carbon quantities.
that deal with the components of this undertaking. Much like Understanding the carbon cycle requires a full court press of
simultaneous observations of sea level, total solar irradiance, satellite and in situ observations because all of these obser-
the gravity field, ocean temperature, surface temperature, and vations must be made at the same time. Many of these mea-
atmospheric temperatures were required to determine if the surements have been made over the past 3040years, but new
Earth is warming and what is responsible; the carbon cycle measurements are needed to quantify carbon storage in vege-
(Figure F.5) will require several complementary satellite and tation, atmospheric measurements are needed to quantify CH4
insitu observations (Cias etal. 2014). and CO2 sources and sinks, better measurements are needed
Carbon cycles through reservoirs on the Earths surface in to quantify land respiration, and more explicit numerical car-
plants and soils exist in the atmosphere as gases, such as car- bon models need to be developed.
bon dioxide (CO2) and methane (CH4), and in ocean water in Similar work needs to be performed for the role of clouds and
phytoplankton and marine sediments. CO2 and CH4 are released aerosols in climate because these are fundamental to under-
into the atmosphere by the combustion of fossil fuels, land cover standing our radiation budget. We also need to improve our
changes on the Earths surface, respiration of green plants, and understanding of the global hydrological cycle.
decomposition of carbon in dead vegetation and in soils, includ- The remote sensing community has made tremendous prog-
ing carbon in permafrost. The atmospheric concentrations of ress over the last five decades as discussed in this edition of
CO2 and CH4 control atmospheric and oceanic temperatures the handbook. Chapters on aerosols in climate, because these
through their absorption of outgoing long-wave radiation and are fundamental, provide comprehensive understanding of
thus also indirectly control sea level via the regulation of plan- land and water studies through detailed methods, approaches,
etary ice volumes. algorithms, synthesis, and key references. Every type of remote
8.9 flux/year Global carbon cycle Atmospheric CO2
*830 (@398 ppm)
st) +3 net change/year
re
fo
096
n
tio
e (9
sis
pira
on
he
els
chang
rati
synt
55 plant res
sil fu
64 soil respi
78
123 photo
80
1.1 land cover
7.8 fos
Incident laser pulse
4 net flux/year
2 net flux/year
Forest carbon 350550
Ocean *39,000
*Stock = billion metric tons
Flux = billion metric tons per year Soils = *15002400 (<2 m)
FIg ur e F.5 Global carbon cycle measurements from a multitude of satellite sensors. A representation of the global carbon cycle showing our
best estimates of carbon fluxes and carbon reservoirs. A series of satellite observations are needed simultaneously to understand the carbon cycle
and its role in the Earths climate system. (From Cias, P. etal., Biogeosciences, 11(13), 3547, 2014.)
2016 by Taylor & Francis Group, LLC
Foreword xv
sensing data obtained from systems such as optical, radar, References
light detection and ranging (LiDAR), hyperspectral, and
hyperspatial is presented and discussed in different chapters. Ciais, P. etal. 2014. Current systematic carbon-cycle observations
Remotely Sensed Data Characterization, Classification, and and the need for implementing a policy-relevant carbon
Accuracies sets the stage with chapters in this book address- observing system. Biogeosciences 11(13): 35473602.
ing remote sensing data characteristics, within and between Cracknell, A. P. 1997. The Advanced Very High Resolution
sensor calibrations, classification methods, and accuracies Radiometer (AVHRR). Taylor & Francis, U.K., 534pp.
taking a wide array of remote sensing data from a wide array Froehlich, C. 2013. Total solar irradiance: What have we learned
of platforms over the last five decades. Remotely Sensed Data from the last three cycles and the recent minimum? Space
Characterization, Classification, and Accuracies also brings in Science Reviews 176(14): 237252.
technologies closely linked with remote sensing such as global Gregory, J. M. et al. 2013. Twentieth-century global-mean sea
positioning system (GPS), global navigation satellite system level rise: Is the whole greater than the sum of the parts?
(GNSS), crowdsourcing, cloud computing, and remote sensing Journal of Climate 26(13): 44764499.
law. In all, the 82 chapters in the 3 volumes of the handbook Irons, J. R., Dwyer, J. L., and Barsi, J. A. 2012. The next Landsat sat-
are written by leading and well-accomplished remote sensing ellite: The Landsat data continuity mission. Remote Sensing
scientists of the world and competently edited by Dr. Prasad of Environment 122: 1121. doi:10.1016/j.rse.2011.08.026.
S. Thenkabail, Research Geographer-15, at the United States Luthcke, S. B., Sabaka, T. J., Loomis, B. D. Arendt, A. A., McCarthy,
Geological Survey (USGS). J. J., and Camp, J. 2013. Antarctica, Greenland, and Gulf of
We can look forward in the next 1020years to improving our Alaska land-ice evolution from an iterated GRACE global
quantitative understanding of the global carbon cycle, under- mascon solution. Journal of Glaciology 59(216): 613631.
standing the interaction of clouds and aerosols in our radiation Roemmich, D. and the Argo Steering Team, 2009. ArgoThe
budget, and understanding the global hydrological cycle. There challenge of continuing 10years of progress. Oceanography
is much work to do. Existing key climate observations must be 22(3): 4655.
continued and new satellite observations will be needed (e.g.,
the recently launched NASAs Orbiting Carbon Observatory-2 Compton J. Tucker
for atmospheric CO2 measurements), and we have many well- Earth Science Division
trained scientists to undertake this work and continue the legacy Goddard Space Flight Center
of the past five decades. National Aeronautics and Space Administration
Greenbelt, Maryland
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
Preface: Remote Sensing
Advances of the Last 50Years
and a Vision for the Future
The overarching goal of the Remote Sensing Handbook 5. Global team of writers, global geographic coverage of
(Remotely Sensed Data Characterization, Classification, study areas, and a wide array of satellites and sensors
and Accuracies; Land Resources Monitoring, Modeling, and 6. Plenty of color illustrations
Mapping with Remote Sensing; and Remote Sensing of Water
Chapters in the books cover the following aspects of remote
Resources, Disasters, and Urban Studies), with 82 chapters
sensing:
and about 2500 pages, was to capture and provide the most
comprehensive state of the art of remote sensing science and State of the art
technology development and advancement in the last 50years, Methods and techniques
by clearly demonstrating the (1) scientific advances, (2) meth- Wide array of land and water applications
odological advances, and (3) societal benefits achieved during Scientific achievements and advancements over the last
this period, as well as to provide a vision of what is to come in 50years
the years ahead. The three books are, to date and to the best of Societal benefits
my knowledge, the most comprehensive documentation of the Knowledge gaps
scientific and methodological advances that have taken place in Future possibilities in the twenty-first century
understanding remote sensing data, methods, and a wide array Great advances have taken place over the last 50 years using
of land and water applications. Written by 300+ leading global remote sensing in the study of the planet Earth, especially using
experts in the area, each chapter (1) focuses on a specific topic data gathered from a multitude of Earth observation (EO) satel-
(e.g., data, methods, and applications), (2) reviews the existing lites launched by various governments as well as private enti-
state-of-the-art knowledge, (3) highlights the advances made, ties. A large part of the initial remote sensing technology was
and (4) provides guidance for areas requiring future devel- developed and tested during the two world wars. In the 1950s,
opment. Chapters in the books cover a wide array of subject remote sensing slowly began its foray into civilian applications.
matter of remote sensing applications. The Remote Sensing During the years of the Cold War, remote sensing applications,
Handbook is planned as a reference material for remote sens- both civilian and military, increased swiftly. But it was also
ing scientists, land and water resource practitioners, natural an age when remote sensing was the domain of a very few top
and environmental practitioners, professors, students, and experts and major national institutes, having multiple skills in
decision makers. The special features of the Remote Sensing engineering, science, and computer technology. From the 1960s
Handbook include the following: onward, there have been many governmental agencies that have
initiated civilian remote sensing. The National Aeronautics and
1. Participation of an outstanding group of remote sensing Space Administration (NASA) and the United States Geological
experts, an unparalleled team of writers for such a book Survey (USGS) have been in the forefront of many of these efforts.
project Others who have provided leadership in civilian remote sensing
2. Exhaustive coverage of a wide array of remote sensing include, but are not limited to, the European Space Agency (ESA)
science: data, methods, and applications of the European Union, the Indian Space Research Organization
3. Each chapter being led by a luminary and most chapters (ISRO), the Centre National dtudes Spatiales (CNES) of
written by teams who further enriched the chapters France, the Canadian Space Agency (CSA), the Japan Aerospace
4. Broadening the scope of the book to make it ideal for Exploration Agency (JAXA), the German Aerospace Center
expert practitioners as well as students (DLR), the China National Space Administration (CNSA),
xvii
2016 by Taylor & Francis Group, LLC
xviii Preface
the United Kingdom Space Agency (UKSA), and the Instituto matching techniques, and automated cropland classification
Nacional de Pesquisas Espaciais (INPE) of Brazil. Many private algorithms); and (8) development of new spectral indices to
entities have launched and operated satellites. These government quantify and study specific land and water parameters (e.g.,
and private agencies and enterprises launched and operated a hyperspectral vegetation indices or HVIs). As a result of these
wide array of satellites and sensors that captured the data of the all-around developments, remote sensing science is today very
planet Earth in various regions of the electromagnetic spectrum mature and is widely used in virtually every discipline of the
and in various spatial, radiometric, and temporal resolutions, earth sciences for quantifying, mapping, modeling, and moni-
routinely and repeatedly. However, the real thrust for remote toring our planet Earth. Such rapid advances are captured in a
sensing advancement came during the last decade of the twenti- number of remote sensing and earth science journals. However,
eth century and the beginning of the twenty-first century. These students, scientists, and practitioners of remote sensing science
initiatives included a launch of a series of new-generation EO and applications have significant difficulty gathering a complete
satellites to gather data more frequently and routinely, release understanding of the various developments and advances that
of pathfinder datasets, web enabling the data for free by many have taken place as a result of their vastness spread across the
agencies (e.g., USGS release of the entire Landsat archives as last 50 years. Therefore, the chapters in the Remote Sensing
well as real-time acquisitions of the world for free dissemina- Handbook are designed to give a whole picture of scientific and
tion by web-enabling), and providing processed data ready to technological advances of the last 50years.
users (e.g., surface reflectance products of moderate-resolution Today, the science, art, and technology of remote sensing are
imaging spectroradiometer [MODIS]). Other efforts like Google truly ubiquitous and increasingly part of everyones everyday
Earth made remote sensing more popular and brought in a new life, often without the user knowing it. Whether looking at your
platform for easy visualization and navigation of remote sens- own home or farm (e.g., see the following figure), helping you
ing data. Advances in computer hardware and software made it navigate when you drive, visualizing a phenomenon occurring
possible to handle Big Data. Crowdsourcing, web access, cloud in a distant part of the world (e.g., see the following figure), mon-
computing, and mobile platforms added a new dimension to itoring events such as droughts and floods, reporting weather,
how remote sensing data are used. Integration with global posi- detecting and monitoring troop movements or nuclear sites,
tioning systems (GPS) and global navigation satellite systems studying deforestation, assessing biomass carbon, addressing
(GNSS) and inclusion of digital secondary data (e.g., digital disasters such as earthquakes or tsunamis, and a host of other
elevation, precipitation, temperature) in analysis have made applications (e.g., precision farming, crop productivity, water
remote sensing much more powerful. Collectively, these initia- productivity, deforestation, desertification, water resources
tives provided a new vision in making remote sensing data more management), remote sensing plays a pivotal role. Already,
popular, widely understood, and increasingly used for diverse many new innovations are taking place. Companies such as
applications, hitherto considered difficult. The free availability the Planet Labs and Skybox are planning to capture very-high-
of archival data when combined with more recent acquisitions spatial-resolution imagery (typically, sub-meter to 5 meters),
has also enabled quantitative studies of change over space and even videos from space using a large number of microsatellite
time. The Remote Sensing Handbook is targeted to capture these constellations. There are others planning to launch a constella-
vast advances in data, methods, and applications, so a remote tion of hyperspectral or other sensors. Just as the smartphone
sensing student, scientist, or a professional practitioner will have and social media connected the world, remote sensing is making
the most comprehensive, all-encompassing reference material in the world our backyard. No place goes unobserved and no event
one place. gets reported without a satellite or other kinds of remote sensing
Modern-day remote sensing technology, science, and appli- images or their derivatives. This is how true liberation for any
cations are growing exponentially. This growth is a result of a technology and science occurs.
combination of factors that include (1) advances and innova- Google Earth can be used to seamlessly navigate and
tions in data capture, access, and delivery (e.g., web enabling, precisely locate any place on Earth, often with very-high-
cloud computing, crowdsourcing); (2) an increasing number of spatial-resolution data (VHRI; submeters to 5 m) from satel-
satellites and sensors gathering data of the planet, repeatedly lites such as IKONOS, QuickBird, and GeoEye (Note: the image
and routinely, in various portions of the electromagnetic spec- below is from one of the VHRI). Here, the editor-in-chief (EiC)
trum as well as in an array of spatial, radiometric, and temporal of this handbook located his village home (Thenkabail) and
resolutions; (3) efforts at integrating data from multiple satel- surroundings that have land covers such as secondary rain-
lites and sensors (e.g., sentinels with Landsat); (4) advances in forests, lowland paddy farms, areca nut plantations, coconut
data normalization, standardization, and harmonization (e.g., plantations, minor roads, walking routes, open grazing lands,
delivery of data in surface reflectance, intersensor calibration); and minor streams (typically, first and second order) (note: land
(5) methods and techniques for handling very large data vol- cover detailed is based on the ground knowledge of the EiC).
umes (e.g., global mosaics); (6) quantum leap in computer hard- The first primary school attended by him is located precisely.
ware and software capabilities (e.g., ability to process several Precise coordinates (13 degree 45 minutes 39.22 seconds north-
terabytes of data); (7) innovation in methods, approaches, and ern latitude, 75 degrees 06 minutes 56.03 seconds eastern lon-
techniques leading to sophisticated algorithms (e.g., spectral gitude) of Thenkabails village house on the planet and the date
2016 by Taylor & Francis Group, LLC
Preface xix
of image acquisition (March 1, 2014) are noted. Google Earth Motivation for the Remote Sensing Handbook started with a
images are used for visualization as well as for numerous sci- simple conversation with Irma Shagla-Britton, acquisitior editor
ence applications such as accuracy assessment, reconnaissance, for remote sensing and GIS books of Taylor & Francis Group/
determining land cover, and establishing land use for various CRC Press, way back in early 2013. Irma was informally get-
ground surveys. It is widely used by lay people who often have ting my advice about doing a new and unique book on remote
no idea on how it all comes together but understand the infor- sensing. Neither the specific subject nor the editor was identi-
mation provided intuitively. This is already happening. These fied. What was clear to me though was that I certainly did not
developments make it clear that we not only need to understand want to lead the effort. I was nearing the end of my third year of
the state of the art but also have a vision of where the future of recovery from colon cancer, and the last thing I wanted to do was
remote sensing is headed. Therefore, in a nutshell, the goal of to take any book project, forget a multivolume remote sensing
this handbook is to cover the developments and advancement of magnum opus, as it ultimately turned out. However, mostly out
six distinct eras in terms of data characterization and process- of courtesy for Irma, I did some preliminary research. I tried to
ing as well as myriad land and water applications: identify a specific topic within remote sensing where there was a
1. Precivilian remote sensing era of the pre-1950s: World sufficient need for a full-fledged book. My research showed that
War I and II when remote sensing was a military tool there was not a single book that would provide a complete and
2. Technology demonstration era of the 1950s and 1960s: comprehensive coverage of the entire subject of remote sensing
Sputnik-I and NOAA AVHRR era of the 1950s and 1960s starting from data capture, to data preprocessing, to data analy-
3. Landsat era of the 1970s: when the first truly operational sis, to myriad land and water applications. There are, of course,
land remote sensing satellite (Earth Resources Technology numerous excellent books on remote sensing, each covering a
Satellite or ERTS, later renamed Landsat) was launched and specific subject matter. However, if a student, scientist, or practi-
operated in the 1970s and early 1980s by United States tioner of remote sensing wanted a standard reference on the sub-
4. Earth observation era of the 1980s and 1990s: when a num- ject, he or she would have to look for numerous books or journal
ber of space agencies began launching and operating satellites articles and often a coherence of these topics would still be left
(e.g., Landsat 4,5 by the United States; SPOT-1,2 by France; uncovered or difficult to comprehend for students and even for
IRS-1a, 1b by India) from the middle to late 1980s onward till many experts with less experience. Guidance on how to approach
the middle of 1990s the study of remote sensing and capture its state of the art and
5. Earth observation and the first decade of the New Millennium advances remained hazy and often required referring to a mul-
era of the 2000s: when data dissemination to users became as titude of references that may or may not be immediately avail-
important as launching, operating, and capturing data (e.g., able, and if available, how to go about it was still hazy to most.
MODIS Terra\Aqua, Landsat-8, Resourcesat) in the late 1990 During this process, I asked myself, several times, what remote
and the first decade of the 2000s sensing book will be most interesting, productive, and useful to
6. Second decade of the New Millennium era starting in a broad audience? The answer, each time, was very clear: A com-
the 2010s: when new-generation micro-\nanosatellites (e.g., plete and comprehensive coverage of the state-of-the-art remote
PlanetLabs, Skybox) are added to the increasing constellation sensing, capturing the advances that have taken place over the
of multiagency sensors (e.g., Sentinels, and the next generation last 50years, which will set the stage for a vision for the future.
of satellites such as SMAP, hyperspectral satellites like NASAs When this became clear, I started putting together the needed
HyspIRI and others from private industry) topics to achieve such a goal. Soon I realized that the only way
2016 by Taylor & Francis Group, LLC
xx Preface
to achieve this goal was through a multivolume book on remote Overall, the preparation of the Remote Sensing Handbook
sensing. Because the number of chapters was more than 80, this took two and a half years, from the time book chapters and
appeared to be too daunting, too overwhelming, and too big a authors were being identified to its final publication. The three
project to accomplish. Yet I sent the initial idea to Irma, who I books are designed in such a way that a reader can have all
thought would say forget it and ask me to focus on a single- three books as a standard reference or have individual books to
volume book. But to my surprise, Irma not only encouraged the study specific subject areas. The three books of Remote Sensing
idea but also had a number of useful suggestions. So what started Handbook are
as intellectual curiosity turned into this full-fledged multivol-
Remotely Sensed Data Characterization, Classification,
ume Remote Sensing Handbook.
and Accuracies: 31 Chapters
However, what worried me greatly was the virtual impossibil-
Land Resources Monitoring, Modeling, and Mapping with
ity (my thought at that time) of gathering the best authors. What
Remote Sensing: 28 Chapters
was also crystal clear to me was that unless the very best were
Remote Sensing of Water Resources, Disasters, and Urban
attracted to the book project, it was simply not worth the effort.
Studies: 27 Chapters
I had made up my mind to give up the book project, unless Igot
the full support of a large number of the finest practitioners of There are about 2500 pages in the 3 volumes.
remote sensing from around the world. So, I spent a few weeks The wide array of topics covered is very comprehensive.
researching the best authors to lead each chapter and wrote to The topics covered in Remotely Sensed Data Characterization,
them to participate in the Remote Sensing Handbook project. Classification, and Accuracies include (1) satellites and sensors;
What really surprised me was that almost all the authors I con- (2) remote sensing fundamentals; (3) data normalization, har-
tacted agreed to lead and write a chapter. This was truly surreal. monization, and standardization; (4) vegetation indices and
These are extremely busy people of great scientific reputation their within- and across-sensor calibration; (5) image classifi-
and achievements. For them to spend the time, intellect, and cation methods and approaches; (6) change detection; (7)inte-
energy to write an in-depth and insightful book chapter spread grating remote sensing with other spatial data; (8) GNSS; (9)
across a year or more is truly amazing. Most also agreed to put crowdsourcing; (10) cloud computing; (11) Google Earth remote
together a writing team, as I had requested, to ensure greater sensing; (12)accuracy assessments; and (13) remote sensing law.
perspective for each chapter. In the end, we had 300+ authors The topics covered in Land Resources Monitoring, Modeling,
writing 82 chapters. and Mapping with Remote Sensing include (1) vegetation and
At this stage, I was somewhat drawn into the project as if by biomass, (2) agricultural croplands, (3) rangelands, (4) phenol-
destiny and felt compelled to go ahead. One of the authors who ogy and food security, (5) forests, (6) biodiversity, (7) ecology,
agreed to lead the chapter mentioned ..whether it was even (8)land use/land cover, (9) carbon, and (10) soils.
possible. This is exactly what I felt, too. But I had reached the The topics covered in Remote Sensing of Water Resources,
stage of no return, and I took on the book project with all the Disasters, and Urban Studies include (1) hydrology and water
seriousness it deserved. It required some real changes to my resources; (2) water use and water productivity; (3) floods;
lifestyle: professional and personal. Travel was reduced to bare (4)wetlands; (5) snow and ice; (6) glaciers, permafrost, and ice;
minimum during most of the book project. Most weekends were (7) geomorphology; (8) droughts and drylands; (9) disasters;
spent editing, writing, and organizing, and other social activi- (10)volcanoes; (11) fire; (12) urban areas; and (13) nightlights.
ties were reduced. Accomplishing such complex work requires There are many ways to use the Remote Sensing Handbook.
the highest levels of discipline, planning, and strategy. But, A lot of thought went into organizing the books and chap-
above all, I felt blessed with good health. By the time the book ters. So you will see a flow from chapter to chapter and book
is published, I will have completed about 5years from my colon to book. As you read through the chapters, you will see how
cancer surgery and chemotherapy. So I am as happy to see this they are interconnected and how reading all of them provides
book released as I am with the miracle of cancer cure (I feel con- you with greater in-depth understanding. Some of you may be
fident to say so). more interested in a particular volume. Often, having all three
But it is the chapter authors who made it all feasible. They books as reference material is ideal for most remote sensing
amazed me throughout the book project. First, the quality and experts, practitioners, or students; however, you can also refer
content of each of the chapters were of the highest standards. to individual books based on your interest. We have also made
Second, with very few exceptions, chapters were delivered attempts to ensure the chapters are self-contained. That way
on time. Third, edited chapters were revised thoroughly and you can focus on a chapter and read it through, without having
returned on time. Fourth, all my requests on various formatting to be overly dependent on other chapters. Taking this perspec-
and quality enhancements were addressed. This is what made tive, there is a slight (~5%10%) material that may be repeated
the three-volume Remote Sensing Handbook possible and if I in some of the chapters. This is done deliberately. For example,
may say so, a true magnum opus on the subject. My heartfelt when you are reading a chapter on LiDAR or radar, you dont
gratitude to these great authors for their dedication. It has been want to go all the way back to another chapter (e.g., Chapter1,
my great honor to work with these dedicated legends. Indeed, Remotely Sensed Data Characterization, Classification, and
Icall them my heroes in a true sense. Accuracies) to understand the characteristics of these sensors.
2016 by Taylor & Francis Group, LLC
Preface xxi
Similarly, certain indices (e.g., vegetation condition index Characterization, Classification, and Accuracies; Land Resources
[VCI], temperature condition index [TCI]) that are defined Monitoring, Modeling, and Mapping with Remote Sensing;
in one chapter (e.g., on drought) may be repeated in another and Remote Sensing of Water Resources, Disasters, and Urban
chapter (also on drought). Such minor overlaps are helpful to Studies). I truly enjoyed the effort. What an honor to work with
the reader to avoid going back to another chapter to under- luminaries in this field of expertise. I learned a lot from them
stand a phenomenon or an index or a characteristic of a sensor. and am very grateful for their support, encouragement, and deep
However, if you want a lot of details on these sensors or indices insights. Also, it has been a pleasure working with outstanding
or phenomena or if you are someone who has yet to gain suf- professionals of Taylor & Francis Group/CRC Press. There is no
ficient expertise in the field of remote sensing, then you will joy greater than being immersed in pursuit of excellence, knowl-
have to read the appropriate chapter where there is in-depth edge gain, and knowledge capture. At the same time, I am happy
coverage of the topic. it is over. The biggest lesson I learned during this project was
Each book has a summary chapter (the last chapter of each that if you set yourself to a task with dedication, sincerity, persis-
book). The summary chapter can be read two ways: (1) either as tence, and belief, you will have the job accomplished, no matter
a last chapter to recapture the main points of each of the previ- how daunting.
ous chapters or (2) as an initial overview to get a feeling for what I expect the books to be standard references of immense value
is in the book. I suggest the readers do it both ways: Read it first to any student, scientist, professional, and practical practitioner
before going into the details and then read it at the end to recol- of remote sensing.
lect what was said in the chapters.
It has been a great honor as well as a humbling experience Prasad S. Thenkabail, PhD
to edit the Remote Sensing Handbook (Remotely Sensed Data Editor-in-Chief
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
Acknowledgments
The Remote Sensing Handbook (Remotely Sensed Data Charac- Chapter 3, Philippe M. Teillet
terization, Classification, and Accuracies; Land Resources Chapter 4, Philippe M. Teillet and Gyanesh Chander
Monitoring, Modeling, and Mapping with Remote Sensing; and Chapter 5, Rudiger Gens and Jordi Cristbal Rossell
Remote Sensing of Water Resources, Disasters, and Urban Studies) Chapter 6, Dongdong Wang
brought together a galaxy of remote sensing legends. The lead Chapter 7, Tomoaki Miura, Kenta Obata, Javzandulam T.
authors and coauthors of each chapter are internationally rec- Azuma, Alfredo Huete, and Hiroki Yoshioka
ognized experts of the highest merit on the subject about which Chapter 8, Michael D. Steven, Timothy Malthus, and
they have written. The lead authors were chosen carefully by me Frdric Baret
after much thought and discussions, who then chose their coau- Chapter 9, Sunil Narumalani and Paul Merani
thors. The overwhelming numbers of chapters were written over Chapter 10, Soe W. Myint, Victor Mesev, Dale Quattrochi,
a period of one year. All chapters were edited and revised over the and Elizabeth A. Wentz
subsequent year and a half. Chapter 11, Mutlu Ozdogan
Gathering such a galaxy of authors was the biggest challenge. Chapter 12, Jun Li and Antonio Plaza
These are all extremely busy people, and committing to a book Chapter 13, Claudia Kuenzer, Jianzhong Zhang, and
project that requires a substantial work load is never easy. However, Stefan Dech
almost all those whom I asked agreed to write the chapter, and only Chapter 14, Thomas Blaschke, Maggi Kelly, and Helena
had to convince a few. The quality of the chapters should convince Merschdorf
readers why these authors are such highly rated professionals and Chapter 15, Stefan Lang and Dirk Tiede
why they are so successful and accomplished in their field of exper- Chapter 16, James C. Tilton, Selim Aksoy, and Yuliya
tise. They not only wrote very high quality chapters but delivered Tarabalka
on time, addressed any editorial comments timely without com- Chapter 17, Shih-Hong Chio, Tzu-Yi Chuang, Pai-Hui
plaints, and were extremely humble and helpful. What was also Hsu, Jen-Jer Jaw, Shih-Yuan Lin, Yu-Ching Lin, Tee-Ann
most impressive was the commitment of these authors for quality Teo, Fuan Tsai, Yi-Hsing Tseng, Cheng-Kai Wang, Chi-
science. Three lead authors had serious health issues and yet they Kuei Wang, Miao Wang, and Ming-Der Yang
delivered very high quality chapters in the end, and there were few Chapter 18, Daniela Anjos, Dengsheng Lu, Luciano Dutra,
others who had unexpected situations (e.g., family health issues) and Sidnei SantAnna
and yet delivered the chapters on time. Even when I offered them Chapter 19, Jason A. Tullis, Jackson D. Cothren, David P.
the option to drop out, almost all of them wanted to stay. They only Lanter, Xuan Shi, W. Fredrick Limp, Rachel F. Linck, Sean
asked for a few extra weeks or months but in the end honored their G. Young, and Tareefa S. Alsumaiti
commitment. I am truly honored to have worked with such great Chapter 20, Gaurav Sinha, Barry J. Kronenfeld, and Jeffrey
professionals. C. Brunskill
In the following list are the names of everyone who contributed Chapter 21, May Yuan
and made possible the Remote Sensing Handbook. In the end, Chapter 22, Stefan Lang, Stefan Kienberger, Michael
we had 82 chapters, a little over 2500 pages, and a little over 300 Hagenlocher, and Lena Pernkopf
authors. Chapter 23, Mohinder S. Grewal
My gratitude to the following authors of chapters in Remotely Chapter 24, Kegen Yu, Chris Rizos, and Andrew Dempster
Sensed Data Characterization, Classification, and Accuracies. The Chapter 25, D. Myszor, O. Antemijczuk, M. Grygierek,
authors are listed in chapter order starting with the lead author. M. Wierzchanowski, and K.A. Cyran
Chapter 26, Fabio DellAcqua
Chapter 1, Drs. Sudhanshu S. Panda, Mahesh Rao, Prasad Chapter 27, Ramanathan Sugumaran, James W. Hegeman,
S. Thenkabail, and James P. Fitzerald Vivek B. Sardeshmukh, and Marc P. Armstrong
Chapter 2, Natascha Oppelt, Rolf Scheiber, Peter Gege, Martin Chapter 28, John Bailey
Wegmann, Hannes Taubenboeck, and Michael Berger Chapter 29, Russell G. Congalton
xxiii
2016 by Taylor & Francis Group, LLC
xxiv Acknowledgments
Chapter 30, P.J. Blount Chapter 18, Thomas W. Gillespie, Andrew Fricker, Chelsea
Chapter 31, Prasad S. Thenkabail Robinson, and Duccio Rocchini
Chapter 19, Stefan Lang, Christina Corbane, Palma
My gratitude to the following authors of chapters in Land Blonda, Kyle Pipkins, and Michael Frster
Resources Monitoring, Modeling, and Mapping with Remote Chapter 20, Conghe Song, Jing Ming Chen, Taehee Hwang,
Sensing. The authors are listed in chapter order starting with the Alemu Gonsamo, Holly Croft, Quanfa Zhang, Matthew
lead author. Dannenberg, Yulong Zhang, Christopher Hakkenberg,
Juxiang Li
Chapter 1, Alfredo Huete, Guillermo Ponce-Campos, Chapter 21, John Rogan and Nathan Mietkiewicz
Yongguang Zhang, Natalia Restrepo-Coupe, Xuanlong Chapter 22, Zhixin Qi, Anthony Gar-On Yeh, and Xia Li
Ma, and Mary-Susan Moran Chapter 23, Richard A. Houghton
Chapter 2, Frdric Baret Chapter 24, Jos A.M. Dematt, Cristine L.S. Morgan,
Chapter 3, Wenge Ni-Meister Sabine Chabrillat, Rodnei Rizzo, Marston H.D.
Chapter 4, Clement Atzberger, Francesco Vuolo, Anja Franceschini, Fabrcio da S. Terra, Gustavo M. Vasques,
Klisch, Felix Rembold, Michele Meroni, Marcio Pupin and Johanna Wetterlind
Mello, and Antonio Formaggio Chapter 25, E. Ben-Dor and Jos A.M. Dematt
Chapter 5, Agns Bgu, Damien Arvor, Camille Lelong, Chapter 26, Prasad S. Thenkabail
Elodie Vintrou, and Margareth Simoes
My gratitude to the following authors of chapters in Remote
Chapter 6, Pardhasaradhi Teluguntla, Prasad S.
Sensing of Water Resources, Disasters, and Urban Studies. The
Thenkabail, Jun Xiong, Murali Krishna Gumma,
authors are listed in chapter order starting with the lead author.
Chandra Giri, Cristina Milesi, Mutlu Ozdogan, Russell
G. Congalton, James Tilton, Temuulen Tsagaan Sankey, Chapter 1, Sadiq I. Khan, Ni-Bin Chang, Yang Hong,
Richard Massey, Aparna Phalke, and Kamini Yadav Xianwu Xue, and Yu Zhang
Chapter 7, David J. Mulla, and Yuxin Miao Chapter 2, Santhosh Kumar Seelan
Chapter 8, Baojuan Zheng, James B. Campbell, Guy Serbin, Chapter 3, Trent W. Biggs, George P. Petropoulos, Naga
Craig S.T. Daughtry, Heather McNairn, and Anna Pacheco Manohar Velpuri, Michael Marshall, Edward P. Glenn,
Chapter 9, Prasad S. Thenkabail, Pardhasaradhi Teluguntla, Pamela Nagler, and Alex Messina
Murali Krishna Gumma, and Venkateswarlu Dheeravath Chapter 4, Antnio de C. Teixeira, Fernando B. T.
Chapter 10, Matthew Clark Reeves, Robert A. Hernandez, Morris Scherer-Warren, Ricardo G. Andrade,
Washington-Allen, Jay Angerer, E. Raymond Hunt, Jr., Janice F. Leivas, Daniel C. Victoria, Edson L. Bolfe, Prasad
Ranjani Wasantha Kulawardhana, Lalit Kumar, Tatiana S. Thenkabail, and Renato A. M. Franco
Loboda, Thomas Loveland, Graciela Metternicht, and R. Chapter 5, Allan S. Arnesen, Frederico T. Genofre,
Douglas Ramsey Marcelo P. Curtarelli, and Matheus Z. Francisco
Chapter 11, E. Raymond Hunt, Jr., Cuizhen Wang, Chapter 6, Sandro Martinis, Claudia Kuenzer, and Andr
D. Terrance Booth, Samuel E. Cox, Lalit Kumar, and Twele
Matthew C. Reeves Chapter 7, Chandra Giri
Chapter 12, Lalit Kumar, Priyakant Sinha, Jesslyn F. Chapter 8, D. R. Mishra, Shuvankar Ghosh, C. Hladik,
Brown, R. Douglas Ramsey, Matthew Rigge, Carson A. Jessica L. OConnell, and H. J. Cho
Stam, Alexander J. Hernandez, E. Raymond Hunt, Jr., and Chapter 9, Murali Krishna Gumma, Prasad S. Thenkabail,
Matt Reeves Irshad A. Mohammed, Pardhasaradhi Teluguntla, and
Chapter 13, Molly E. Brown, Kirsten M. de Beurs, and Venkateswarlu Dheeravath
Kathryn Grace Chapter 10, Hongjie Xie, Tiangang Liang, Xianwei Wang,
Chapter 14, E.H. Helmer, Nicholas R. Goodwin, Valry and Guoqing Zhang
Gond, Carlos M. Souza, Jr., and Gregory P. Asner Chapter 11, Qingling Zhang, Noam Levin, Christos
Chapter 15, Juha Hyypp, Mika Karjalainen, Xinlian Chalkias, and Husi Letu
Liang, Anttoni Jaakkola, Xiaowei Yu, Mike Wulder, Chapter 12, James B. Campbell and Lynn M. Resler
Markus Hollaus, Joanne C. White, Mikko Vastaranta, Chapter 13, Felix Kogan and Wei Guo
Kirsi Karila, Harri Kaartinen, Matti Vaaja, Ville Kankare, Chapter 14, Felix Rembold, Michele Meroni, Oscar Rojas,
Antero Kukko, Markus Holopainen, Hannu Hyypp, and Clement Atzberger, Frederic Ham, and Erwann Fillol
Masato Katoh Chapter 15, Brian Wardlow, Martha Anderson, Tsegaye
Chapter 16, Gregory P. Asner, Susan L. Ustin, Philip A. Tadesse, Chris Hain, Wade T. Crow, and Matt Rodell
Townsend, Roberta E. Martin, and K. Dana Chadwick Chapter 16, Jinyoung Rhee, Jungho Im, and Seonyoung Park
Chapter 17, Sylvie Durrieu, Cdric Vga, Marc Bouvier, Chapter 17, Marion Stellmes, Ruth Sonnenschein, Achim
Frdric Gosselin, and Jean-Pierre Renaud Laurent Rder, Thomas Udelhoven, Stefan Sommer, and Joachim
Saint-Andr Hill
2016 by Taylor & Francis Group, LLC
Acknowledgments xxv
Chapter 18, Norman Kerle Royal Danish Geographical Society, and the Vega Medal from
Chapter 19, Stefan Lang, Petra Freder, Olaf Kranz, the Swedish Society of Anthropology and Geography. Hewas the
Brittany Card, Shadrock Roberts, and Andreas Papp NASA representative to the U.S. Global Change Research Program
Chapter 20, Robert Wright from 2006 to 2009. He was instrumental in releasing the AVHRR
Chapter 21, Krishna Prasad Vadrevu and Kristofer Lasko 32-year (19822013) Global Inventory Monitoring and Modeling
Chapter 22, Anupma Prakash and Claudia Kuenzer Studies (GIMMS) data. I strongly recommend that everyone read
Chapter 23, Hasi Bagan and Yoshiki Yamagata his excellent foreword before reading the book. In the foreword,
Chapter 24, Yoshiki Yamagata, Daisuke Murakami, and Dr. Tucker demonstrates the importance of data from EO sensors
Hajime Seya from orbiting satellites to maintaining a reliable and consistent
Chapter 25, Prasad S. Thenkabail climate record. Dr. Tucker further highlights the importance of
continued measurements of these variables of our planet in the
These authors are who is who in remote sensing and come new millennium through new, improved, and innovative EO sen-
from premier institutions of the world. For author affiliations, sors from Sun-synchronous and/or geostationary satellites.
please see Contributors list provided a few pages after this. I am very thankful to my USGS colleagues for their encourage-
Mydeepest apologies if I have missed any name. But, I am sure ment and support. In particular, I mention Edwin Pfeifer, Dr.Susan
those names are properly credited and acknowledged in indi- Benjamin, Dr. Dennis Dye, Larry Gaffney, Miguel Velasco,
vidual chapters. Dr.Chandra Giri, Dr. Terrance Slonecker, Dr.Jonathan Smith, and
The authors not only delivered excellent chapters, they pro- Dr. Thomas Loveland. There are many other colleagues who made
vided valuable insights and inputs for me in many ways through- my job at USGS that much easier. My thanks to them all.
out the book project. I am very thankful to Irma Shagla-Britton, acquisition editor
I was delighted when Dr. Compton J. Tucker, senior Earth for remote sensing and GIS books at Taylor & Francis Group/CRC
scientist, Earth Sciences Division, Science and Exploration Press. Without her initial nudge, this book would never have even
Directorate, NASA Goddard Space Flight Center (GSFC), agreed been completed. Thank you, Irma. You are doing a great job.
to write the foreword for the book. For anyone practicing remote I am very grateful to my wife (Sharmila Prasad) and daugh-
sensing, Dr. Tucker needs no introduction. He has been a god- ter (Spandana Thenkabail) for their usual unconditional love,
father of remote sensing and has inspired a generation of scien- understanding, and support. They are always the pillars of my
tists. I have been a student of his without ever reallybeingone. life. I learned the values of hard work and dedication from my
I mean,I have not been his student in a classroom but have fol- revered parents. This work wouldnt have come about without
lowed his legendary work throughout my career. I remember their sacrifices to educate their children and their silent bless-
reading his highly cited paper (now with citations nearing 4000!): ings. I am ever grateful to my former professors at The Ohio State
University, Columbus, Ohio, United States: Prof. John G. Lyon,
Tucker, C.J. (1979) Red and photographic infrared linear
Dr. Andrew D. Ward, Prof. (Late) Carolyn Merry, Dr. Duane
combinations for monitoring vegetation, Remote Sensing
Marble, and Dr. Michael Demers. They have taught, encour-
of Environment, 8(2),127150.
aged, inspired, and given me opportunities at the right time.
That was in 1986 when I had just joined the National Remote The opportunity to work for six years at the Center for Earth
Sensing Agency (NRSA; now NRSC), Indian Space Research Observation of Yale University (YCEO) was incredibly impor-
Organization (ISRO). After earning his PhD from the Colorado tant. I am thankful to Prof. Ronald G. Smith, director of YCEO,
State University in 1975, Dr. Tucker joined NASA GSFC as a post- for his kindness. At YCEO, I learned and advanced myself as a
doctoral fellow and became a full-time NASA employee in 1977. remote sensing scientist. The opportunities I got from working
Since then, he has conducted path-finding research. He has used for the International Institute of Tropical Agriculture (IITA),
NOAA AVHRR, MODIS, SPOT Vegetation, and Landsat satel- Africa and International Water Management Institute (IWMI)
lite data for studying deforestation, habitat fragmentation, desert that had a global mandate for water were very important, espe-
boundary determination, ecologically coupled diseases, terrestrial cially from the point of view of understanding the real issues
primary production, glacier extent, and how climate affects global on the ground. I learned my basics of remote sensing mainly
vegetation. He has authored or coauthored more than 170 journal working with Dr. Thiruvengadachari of the National Remote
articles that have been cited more than 20,000 times, is an adjunct Sensing Agency/Center (NRSA/NRSC), Indian Space Research
professor at the University of Maryland, is a consulting scholar Organization (ISRO), India, where I started my remote sens-
at the University of Pennsylvanias Museum of Archaeology and ing career as a young scientist. I was just 25 years old then
Anthropology, and has appeared on more than twenty radio and had joined NRSA after earning my masters of engineer-
and TV programs. He is a fellow of the American Geophysical ing (hydraulics and water resources) and bachelors of engi-
Union and has been awarded several medals and honors, includ- neering (civil engineering). During my first day in the office,
ing NASAs Exceptional Scientific Achievement Medal, the Pecora Dr. Thiruvengadachari asked me how much remote sensing
Award from the U.S. Geological Survey (USGS), the National did I know. I said, zero and instantly thought that I would be
Air and Space Museum Trophy, the Henry Shaw Medal from thrown out of the room. But he said very good and gave me a
the Missouri Botanical Garden, the Galathea Medal from the manual on remote sensing from the Laboratory for Applications
2016 by Taylor & Francis Group, LLC
xxvi Acknowledgments
of Remote Sensing(LARS), Purdue. Those were the days where and personally, and he has always been an inspiration. Prof.
there was no formal training in remote sensing in any Indian E.J. James, former director of the Center for Water Resources
universities. So my remote sensing lessons began working prac- Development and Management (CWRDM), was another origi-
tically on projects and one of our first projects was drought nal guru from whom I have learned the values of a true pro-
monitoring for India using NOAA AVHRR data. This was an fessional. I am also thankful to my good old friend Shri C. J.
intense period of learning remote sensing by actually practicing Jagadeesha, who is still working for ISRO as a senior scientist.
it on a daily basis. Data came on 9mm tapes; data were read on He was my colleague at NRSA/NRSC, ISRO, and encouraged
massive computing systems; image processing was done, mostly me to grow as a scientist. This Remote Sensing Handbook is a
working on night shifts by booking time on centralized com- blessing from the most special ones dear to me. Of course, there
puting; field work was conducted using false color composite are many, many others to thank especially many of my dedi-
outputs and topographic maps (not the days of global position- cated students over the years, but they are too many to mention
ing systems); geographic information system was in its infancy; here. I thank the truly outstanding editing work performed by
and a lot of calculations were done using calculators. So when Arunkumar Aranganathan and his team at SPi Global.
I decided to resign my NRSA job and go to the United States It has been my deep honor and great privilege to have edited
to do my PhD, Dr. Thiruvengadachari told me, Prasad, I am the Remote Sensing Handbook. I am sure that I wont be taking
losing my right hand, but you cant miss opportunity. Those on any such huge endeavors in the future. I will need time for
initial wonderful days of learning from Dr. Thiruvengadachari myself, to look inside, understand, and grow. So thank you all,
will remain etched in my memory. Prof. G. Ranganna of the for making this possible.
Karnataka Regional Engineering College (KREC; now National
Institute of Technology), Karnataka, India, was/is one of my most Prasad S. Thenkabail, PhD
revered gurus. I have learned a lot observing him, professionally Editor-in-Chief
2016 by Taylor & Francis Group, LLC
Editor
Prasad S. Thenkabail, PhD, is currently working as a research In2008, for one of these papers, Prasad (lead author) and coau-
geographer-15 with the U.S. Geological Survey (USGS), United thors (Pardhasaradhi Teluguntala, Trent Biggs, Murali Krishna
States. Currently, at USGS, Prasad leads a multi-institutional Gumma, and Hugh Turral) were the second-place recipients of
NASA MEaSUREs (Making Earth System Data Records for the 2008 John I. Davidson American Society of Photogrammetry
Use in Research Environments) project, funded through and Remote Sensing (ASPRS) Presidents Award for practical
NASA ROSES solicitation. The project is entitled Global Food papers. The paper proposed a novel spectral matching technique
Security-Support Analysis Data at 30 m (GFSAD30) (http:// (SMT) for cropland classification. Earlier, Prasad (lead author)
geography.wr.usgs.gov/science/croplands/index.html also see and coauthors (Andy Ward, John Lyon, and Carolyn Merry), won
https://www.croplands.org/). He is also an adjunct professor the 1994 Autometric Award for outstanding paper on remote
at three U.S. universities: (1) Department of Soil, Water, and sensing of agriculture from ASPRS. Recently, Prasad (seccond
Environmental Science (SWES), University of Arizona (UoA); author) with Michael Marshall (lead author), won the ASPRS
(2) Department of Space Studies, University of North Dakota ERDAS award for best scientific paper on remote sensing for their
(UND); and (3) School of Earth Sciences and Environmental hyperspectral remote sensing work.
Sustainability (SESES), Northern Arizona University (NAU), Earlier to this path-breaking r emote Sensing Handbook,
Flagstaff, Arizona. Prasad has published two seminal books (both published by
Dr. Thenkabail has conducted pioneering scientific research Taylor & Francis Group/CRC Press) related to hyperspectral
work in two major areas: remote sensing and global croplands:
1. Hyperspectral remote sensing of vegetation Thenkabail, P.S., Lyon, G.J., and Huete, A. 2011.
2. Global irrigated and rainfed cropland mapping using Hyperspectral Remote Sensing of Vegetation. CRC Press/
spaceborne remote sensing Taylor & Francis Group, Boca Raton, FL, 781pp.
His research papers on these topics are widely quoted. His hyper- Reviews of this book:
spectral work also led to his working on the scientific advisory
board of Rapideye (2001), a German private industry satellite. http://www.crcpress.com/product/isbn/9781439845370.
Prasad was consulted on the design of spectral wavebands. Thenkabail, P., Lyon, G.J., Turral, H., and Biradar, C.M.
In hyperspectral research, Prasad pioneered in the following: 2009. Remote Sensing of Global Croplands for Food
Security. CRC Press/Taylor & Francis Group, Boca Raton,
1. The design of optimal hyperspectral narrowbands (HNBs) FL, 556pp (48 pages in color).
and hyperspectral vegetation indices (HVIs) for agricul-
ture and vegetation studies. Reviews of this book:
2. Certain hyperspectral data mining and data reduction
http://www.crcpress.com/product/isbn/9781420090093.
techniques such as now widely used concepts of lambda
http://gfmt.blogspot.com/2011/05/review-remote-sensing-
by lambda plots.
of-global.html.
3. Certain hyperspectral data classification methods. This
included the use of a series of methods (e.g., discrimi- He has guest edited two special issues for the American Society
nant model, Wilks lambda, Pillai trace) that demonstrate of Photogrammetry and Remote Sensing (PE&RS):
significant increases in classification accuracies of land
Thenkabail, P.S. 2014. Guest editor of special issue on
cover and vegetation classes as determined using HNBs as
Hyperspectral remote sensing of vegetation and agricul-
opposed to multispectral broadbands.
tural crops. Photogrammetric Engineering and Remote
In global croplands, Prasad conducted seminal research that Sensing 80(4).
led to the first global map of irrigated and rainfed cropland Thenkabail, P.S. 2012. Guest editor for Global croplands
areas using multitemporal, multisensor remote sensing, one special issue. Photogrammetric Engineering and Remote
book, and a series of more than ten novel peer-reviewed papers. Sensing 78(8).
xxvii
2016 by Taylor & Francis Group, LLC
xxviii Editor
He has also guest edited a special issue on global croplands for Over the years, he has been a principal investigator (PI) of
the Remote Sensing Open Access Journal (ISSN 2072-4292): NASA, USGS, IEEE, and other funded projects such as inland
valley wetland mapping of African nations, characterization
Thenkabail, P.S. 2010. Guest editor: Special issue on
of eco-regions of Africa (CERA), which involved both African
Global croplands for the MDPI remote sensing open
savannas and rainforests, global cropland water use for food
access journal. Total: 22 papers. http://www.mdpi.com/
security in the twenty-first century, automated cropland classi-
journal/remotesensing/special_issues/croplands/.
fication algorithm (ACCA) within WaterSMART (Sustain and
Prasad is, currently editor-in-chief, Remote Sensing Open Access Manage Americas Resources for Tomorrow) project, water pro-
Journal, an on-line journal, published by MDPI; editorial board ductivity mapping in the irrigated croplands of California and
member, Remote Sensing of Environment; editorial advisory Uzbekistan using multisensor remote sensing, IEEE Water for
board member, ISPRS Journal of Photogrammetry and Remote the World Project, and drought monitoring in India, Pakistan,
Sensing. and Afghanistan.
Prior to joining USGS in October 2008, Dr. Thenkabail was The USGS and NASA selected Dr. Thenkabail to be on the
a leader of the remote sensing programs of leading institutes Landsat Science Team (20072011) for a period of five years
International Water Management Institute (IWMI), 20032008; (http:// landsat.gsfc.nasa.gov/news/news-archive/pol_0005.
International Center for Integrated Mountain Development html; http://ldcm.usgs.gov/intro.php). In June 2007, his team
(ICIMOD), 19951997; International Institute of Tropical was recognized by the Environmental System Research Institute
Agriculture (IITA), 19921995. (ESRI) for special achievement in GIS (SAG award) for their
He also worked as a key remote sensing scientist for Yale tsunami-related work (tsdc.iwmi.org) and for their innova-
Center for Earth Observation (YCEO), 19972003; Ohio State tive spatial data portals (http://waterdata.iwmi.org/dtView-
University (OSU), 19881992; National Remote Sensing Agency Common.php; earlier http://www.iwmidsp.org). Currently,
(NRSA) (now NRSC), Indian Space Research organization he is also a global coordinator for the Agriculture Societal
(ISRO), 19861988. Beneficial Area (SBA) of the Committee for Earth Observation
2016 by Taylor & Francis Group, LLC
Editor xxix
Sy stems(CEOS). He is active in the Group on Earth Observation open channel in India. He has 100+ publications, mostly peer-
(GEO) agriculture and water efforts through Earth observa- reviewed research papers in major international remote sens-
tion. He was a co-lead of the Water for the World Project (IEEE ing journals: http://scholar.google.com/citations?user=9IO5Y7
effort). He is the current chair of the International Society YAAAAJ&hl=en. Prasad has about 30years experience work-
of Photogrammetry and Remote Sensing (ISPRS) Working ing as a well-recognized international expert in remote sens-
Group WG VIII/7: Land Cover and Its Dynamics, including ing and geographic information systems (RS/GIS) and their
Agricultural & Urban Land Use for the period 20132016. application to agriculture, wetlands, natural resource manage-
Thenkabail earned his PhD from The Ohio State University ment, water resources, forests, sustainable development, and
(1992). His masters degree in hydraulics and water resources environmental studies. His work experience spans over 25+
engineering (1984) and bachelors degree in civil engineer- countries spread across West and Central Africa, Southern
ing (1981) were from India. He began his professional career Africa, South Asia, Southeast Asia, the Middle East, East Asia,
as a lecturer in hydrology, water resources, hydraulics, and Central Asia, North America, South America, and the Pacific.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
Contributors
Selim Aksoy Michael Berger r ussell g . Congalton
Department of Computer Engineering European Space Agency Department of Natural Resources and
Bilkent University ESA secondment at European Commission the Environment
Ankara, Turkey Directorate General Research and University of New Hampshire
Innovation Durham, New Hampshire
Tareefa S. Alsumaiti European Commission
University of Arkansas Brussels, Belgium Jackson D. Cothren
Fayetteville, Arkansas Department of Geosciences
University of Arkansas
Thomas Blaschke
Fayetteville, Arkansas
Daniela Anjos Department of GeoinformaticsZ_GIS
Image Processing Division University of Salzburg K.A. Cyran
National Institute for Space Research Salzburg, Austria Faculty of Automatic Control,
So Jos dos Campos, Brazil Electronics and Computer Science
P.J. Blount Institute of Computer Science
O. Antemijczuk School of Law Silesian University of Technology
Faculty of Automatic Control, University of Mississippi Gliwice, Poland
Electronics and Computer Science Oxford, Mississippi
Institute of Computer Science Stefan Dech
Silesian University of Technology German Aerospace Center (DLR)
Jeffrey C. Brunskill
Gliwice, Poland and
Department of Environmental,
Geographical, and Geological Earth Observation Center (EOC)
Marc P. Armstrong Sciences and
Department of Geographical and University of Pennsylvania German Remote Sensing DataCenter,
Sustainability Sciences Bloomsburg, Pennsylvania DFD of DLR
The University of Iowa Oberpfaffenhofen, Germany
Iowa City, Iowa g yanesh Chander and
Science Data Systems Branch
Javzandulam T. Azuma Goddard Space Flight Center Department for Geography and Geology
Department of Natural Resources and National Aeronautics and Space University of Wuerzburg
Environmental Management Administration Wuerzburg, Germany
University of Hawaii at Manoa Greenbelt, Maryland
Honolulu, Hawaii Fabio DellAcqua
Department of Electrical, Computer and
Shih-Hong Chio
Biomedical Engineering
John e . Bailey Department of Land Economics
University of Pavia
Google Inc. National Chengchi University
Pavia, Italy
Mountain View, California Taipei, Taiwan
Andrew Dempster
Frdric Baret Tzu-Yi Chuang School of Electrical Engineering &
National Institute of Agronomic Department of Civil Engineering Telecommunications
Research National Taiwan University University of New South Wales
Avignon, France Taipei, Taiwan Sydney, New South Wales, Australia
xxxi
2016 by Taylor & Francis Group, LLC
xxxii Contributors
Luciano Dutra Stefan Kienberger Timothy J. Malthus
Image Processing Division Department of GeoinformaticsZ_GIS Commonwealth Scientific and Industrial
National Institute for Space Research Paris-Lodron University of Salzburg Research Organisation
So Jos dos Campos, Brazil Salzburg, Austria Ecosciences Precinct
Dutton Park, Queensland, Australia
James e . Fitzerald Barry J. Kronenfeld
Cobb County, Georgia Department of Geology and Geography Paul Merani
Eastern Illinois University Department of Earth and Atmospheric
Peter g ege Charleston, Illinois Sciences
Earth Observation Center
Claudia Kuenzer University of NebraskaLincoln
and
German Aerospace Center (DLR) Lincoln, Nebraska
Remote Sensing Technology Institute
and and
Helena Merschdorf
German Aerospace Centre (DLR) Earth Observation Center (EOC)
Interfaculty Department of
Oberpfaffenhofen, Germany and
GeoinformaticsZ_GIS
German Remote Sensing DataCenter,
r udiger g ens Paris-Lodron University of Salzburg
DFD of DLR
Geophysical Institute Salzburg, Austria
Oberpfaffenhofen, Germany
University of Alaska Fairbanks Victor Mesev
Fairbanks, Alaska Stefan Lang
Department of GeoinformaticsZ_GIS Department of Geography
Mohinder S. g rewal Paris-Lodron University of Salzburg Florida State University
Electrical Engineering Department Salzburg, Austria Tallahassee, Florida
California State University
David P. Lanter Tomoaki Miura
Fullerton, California
CDM Smith Department of Natural Resources and
M. g rygierek Philadelphia, Pennsylvania Environmental Management
LG Nexera Business Solutions AG University of Hawaii at Manoa
Jun Li Honolulu, Hawaii
Vienna, Austria
School of Geography and Planning
Michael Hagenlocher Sun Yat-Sen University Soe W. Myint
Department of GeoinformaticsZ_GIS Guangzhou, Peoples Republic of China School of Geographical Sciences and
Paris-Lodron University of Salzburg Urban Planning
W. Fredrick Limp
Salzburg, Austria Arizona State University
Department of Geosciences
Tempe, Arizona
James W. Hegeman University of Arkansas
Department of Computer Science Fayetteville, Arkansas D. Myszor
The University of Iowa Faculty of Automatic Control,
Shih-Yuan Lin
Iowa City, Iowa Electronics and Computer Science
Department of Land Economics
Pai-Hui Hsu National Chengchi University Institute of Computer Science
National Taiwan University Taipei, Taiwan Silesian University of Technology
Taipei, Taiwan Gliwice, Poland
Yu-Ching Lin
Alfredo Huete Department of Environmental and
Plant Functional Biology and Climate Information and Engineering LG Nexera Business Solutions AG
Change Cluster Chung Cheng Institute of Technology Vienna, Austria
University of Technology, Sydney National Defense University
Ultimo, New South Wales, Australia Taoyuan, Taiwan Sunil Narumalani
Jen-Jer Jaw r achel F. Linck Department of Earth and Atmospheric
Department of Civil Engineering Department of Geosciences Sciences
National Taiwan University University of Arkansas University of NebraskaLincoln
Taipei, Taiwan Fayetteville, Arkansas Lincoln, Nebraska
Maggi Kelly Dengsheng Lu Kenta Obata
Department of Environmental Science, Center for Global Change and Earth Institute of Geology and Geoinformation
Policy, and Management Observations National Institute of Advanced Industrial
University of California Michigan State University Science and Technology
Berkeley, California East Lansing, Michigan Tsukuba, Japan
2016 by Taylor & Francis Group, LLC
Contributors xxxiii
Natascha Oppelt Vivek B. Sardeshmukh Dirk Tiede
Department of Geography Department of Computer Science Department of GeoinformaticsZ_GIS
Christian-Albrechts-Universitt zu Kiel The University of Iowa Paris-Lodron University of Salzburg
Kiel, Germany Iowa City, Iowa Salzburg, Austria
r olf Scheiber James C. Tilton
Mutlu zdoan
Microwaves and Radar Institute Goddard Space Flight Center
Nelson Institute for Environmental
German Aerospace Center (DLR) National Aeronautics and Space
Studies
Oberpfaffenhofen, Germany Administration
and
Department of Forest and Wildlife Ecology Xuan Shi Greenbelt, Maryland
University of WisconsinMadison Department of Geosciences
Fuan Tsai
Madison, Wisconsin University of Arkansas
Center for Space and Remote Sensing
Fayetteville, Arkansas
Research
Sudhanshu S. Panda
g aurav Sinha National Central University
Institute of Environmental Spatial Analysis
Department of Geography Zhongli, Taiwan
University of North Georgia
Ohio University
Oakwood, Georgia Yi-Hsing Tseng
Athens, Ohio
National Cheng Kung University
Lena Pernkopf Michael D. Steven Tainan City, Taiwan
Department of GeoinformaticsZ_GIS School of Geography
Paris-Lodron University of Salzburg University of Nottingham Jason A. Tullis
Salzburg, Austria Nottingham, United Kingdom Department of Geosciences
University of Arkansas
r amanathan Sugumaran
Antonio Plaza Fayetteville, Arkansas
Department of Geographical and
Hyperspectral Computing Laboratory
Sustainability Sciences Cheng-Kai Wang
University of Extremadura
The University of Iowa Department of Geomatics
Cceres, Spain
Iowa City, Iowa National Cheng Kung University
Dale A. Quattrochi and Tainan City, Taiwan
NASA Marshall Space Flight Center
John Deere Chi-Kuei Wang
National Aeronautics and Space
Moline, Illinois Department of Geomatics
Administration
Huntsville, Alabama Yuliya Tarabalka National Cheng Kung University
Sophia-Antipolis Mditerrane Tainan City, Taiwan
Mahesh N. r ao Sophia-Antipolis, France
Dongdong Wang
Department of Forestry and Wildlife
Hannes Taubenboeck Department of Geographical Sciences
Resources
German Remote Sensing Data Center University of Maryland
Humboldt State University
German Aerospace Centre (DLR) College Park, Maryland
Arcata, California
Oberpfaffenhofen, Germany
Miao Wang
Chris r izos Philippe M. Teillet Department of Geomatics
School of Civil and Environmental Department of Physics and Astronomy National Cheng Kung University
Engineering University of Lethbridge Tainan City, Taiwan
University of New South Wales Lethbridge, Alberta, Canada
Sydney, New South Wales, Australia Martin Wegmann
Tee-Ann Teo
National Chiao Tung University Department for Geography and Geology
Jordi Cristbal r ossell Ludwig-Maximilians-Universitt
Hsinchu City, Taiwan
Geophysical Institute Wrzburg
University of AlaskaFairbanks Prasad S. Thenkabail Wrzburg, Germany
Fairbanks, Alaska Geography, Flagstaff Science Center
and e lizabeth A. Wentz
Sidnei SantAnna Western Geographic Science Center School of Geographical Sciences and
Image Processing Division and Urban Planning
National Institute for Space Research United States Geological Survey Arizona State University
So Jos dos Campos, Brazil Flagstaff, Arizona Tempe, Arizona
2016 by Taylor & Francis Group, LLC
xxxiv Contributors
M. Wierzchanowski Ming-Der Yang Kegen Yu
Faculty of Automatic Control, Department of Geomatics School of Geodesy and Geomatics
Electronics and Computer National Chung Hsing University Wuhan University
Science Taichung, Taiwan Wuhan, Peoples Republic of China
Institute of Computer Science
Silesian University of Hiroki Yoshioka
May Yuan
Technology Department of Information Science and
School of Economic, Political, and Policy
Gliwice, Poland Technology
Sciences
Aichi Prefectural University
University of Texas at Dallas
and Nagakute, Japan
Richardson, Texas
LG Nexera Business Sean g . Young
SolutionsAG College of Liberal Arts and Sciences Jianzhong Zhang
Vienna, Austria University of Iowa Freelance Remote Sensing Consultant
Iowa City, Iowa Kaufering, Germany
2016 by Taylor & Francis Group, LLC
I
Satellites and Sensors
from Different Eras and
Their Characteristics
1 r emote Sensing SystemsPlatforms and Sensors: Aerial, Satellite, u AV, Optical, r adar,
andLiDAr Sudhanshu S. Panda, Mahesh N. Rao, Prasad S. Thenkabail, and James E. Fitzerald ................................... 3
Acronyms and Definitions Introduction Remote Sensing Platforms and Sensor Characteristics Remote
Sensing Platforms Upcoming New Satellite Sensor Platforms Future of Remote Sensing and Evolving
Microsatellites Discussions Summary and Conclusion References
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
1
Remote Sensing Systems
Platforms and Sensors:
Aerial, Satellite, UAV, Optical,
Radar, and LiDAR
Acronyms and Definitions .....................................................................................................................3
1.1 Introduction .................................................................................................................................5
Remote Sensing Definition and History Data Collection by Remote Sensing and
Usage Principles of Electromagnetic Spectrum in Remote Sensing
1.2 Remote Sensing Platforms and Sensor Characteristics .........................................................9
Image Characteristics Categorization of Satellites by Country of Ownership Types of
Sensing Technologies Based on Applications
Sudhanshu S. Panda 1.3 Remote Sensing Platforms........................................................................................................25
University of North Georgia
Aerial Imaging Optical Remote Sensing Hyperspectral Remote Sensing RADAR
and SODAR Remote Sensing LASER and RADAR Altimetry Imaging LiDAR Remote
Mahesh N. Rao Sensing Microwave Remote Sensing SONAR and SODAR Remote Sensing Global
Humboldt State University Positioning System
1.4 Upcoming New Satellite Sensor Platforms ........................................................................... 48
Prasad S. Thenkabail 1.5 Future of Remote Sensing and Evolving Microsatellites .....................................................49
United States Geological Survey
1.6 Discussions .................................................................................................................................51
James E. Fitzerald 1.7 Summary and Conclusion ........................................................................................................51
Department of Land Management References...............................................................................................................................................51
AcronymsandDefinitions AVIRIS Airborne visible/infrared imaging spectrometer
AVNIR Advanced visible and near-infrared radiometer
ACRIM Active cavity radiometer irradiance monitor AWiFS Advanced wide field sensor (IRS)
ADEOS Advanced Earth observing satellite BDRF Bidirectional reflectance function
AEP Architecture evolution plan BV Brightness value
AIRS Atmospheric infrared sounder CASI Compact airborne spectrographic imager
AIRSAR Airborne synthetic aperture radar CBERS-2 ChinaBrazil Earth Resources Satellite
ALADIN Atmospheric laser doppler instrument C-CAP Coastal Change Analysis Program
ALI Advanced land imager CCD Charge-coupled device
ALOS Advanced land observing satellite (from JAXA) CCRS Canadian Center for Remote Sensing
AMS Airborne multispectral scanner CCT Computer-compatible tape
AOI Area of interest CEO Centre for Earth Observation
ASAR Advanced synthetic aperture radar (from ESA) CEOS Committee on Earth Observation Satellites
ASFA Aquatic sciences and fisheries abstracts (FAO) CERES Clouds and the Earths Radiant Energy System
ASTER Advanced spaceborne thermal emission and CIR Color infrared
reflection radiometer CMOS Complementary metal-oxide semiconductor
ATM Airborne thematic mapper CNES Centre National dEtudes Spatiales (French
ATSR Along-track scanning radiometer space agency)
AUV Autonomous underwater vehicle DEM Digital elevation model
AVHRR Advanced very-high-resolution radiometer DGPS Differential global positioning
2016 by Taylor & Francis Group, LLC
4 Remotely Sensed Data Characterization, Classification, and Accuracies
DIDSON Dual-frequency identification sonar GSFC Goddard Space Flight Center
DLR German Aerospace Research Establishment GSTC Geospatial Service and Technology Center
DN Digital number (pixel value) HIRS High-resolution infrared sounder
DOQQ Digital orthophoto (quarter) quad Hyperion First spaceborne hyperspectral sensor
DTM Digital terrain model onboard Earth observing-1 (EO-1)
EDC EROS Data Center IFOV Instantaneous field of view
EME Electromagnetic energy IKONOS High-resolution satellite operated by GeoEye
EMR Electromagnetic radiation IPCC Intergovernmental Panel on Climate Change
EMS Electromagnetic spectrum IRS Indian Remote Sensing Satellite
ENVISAT Polar-orbiting environmental satellite by ESA IVOS Infrared and visible optical sensor
EOS Earth Observing System Jason The satellite series planned as follow on to
EOSAT Earth observation satellite TOPEX/Poseidon (Jason-1 is the first of these)
EOSDIS Earth Observing System Data and Information JAXA Japan Aerospace Exploration Agency
System JERS Japanese Earth Resources Satellite
EOSP Earth Observing Scanning Polarimeter JPSS Joint Polar Satellite System
EROS Earth Resources Observation Systems KOMFOSAT Korean Multipurpose Satellite (data marketed
ERS European Remote Sensing Satellite by SPOT Image)
ERS-1 Earth Remote Sensing Satellite LANDSAT Land Remote Sensing Satellite
ERTS Earth Resources Technology Satellite LASER Light Amplification by Stimulated Emission
ESA European Space Agency of Radiation
ESRI Environmental Systems Research Institute LCCP Land cover characterization program
ETM+ Enhanced Thematic Mapper Plus LCT Laser communication terminal
EUMETSAT European Organisation for the Exploitation LIDAR Laser Image Detection and Ranging
of Meteorological Satellites LISS Linear Imaging Self-Scanning Sensor (IRS)
FAS Foreign Agricultural Service LRA Laser reflectometry array
FGDC Federal Geographic Data Committee LWIR Longwave infrared
FIR Far infrared MERIS Medium-resolution imaging spectrometer
FLIR Forward-looking infrared METEOSAT Meteorological satellite by ESA
FORMOSAT Taiwanese satellite operated by the National MIR Mid-infrared
Space Organization of Taiwan (NSPO) MIRAS Microwave Imaging Radiometer withAperture
FOV Field of view Synthesis
FSA Farm Service Agency (USDA) MISR Multi-angle Imaging SpectroRadiometer
G3OS Global Observing Systems of GCOS, GOOS, MLS Microwave Limb Scanner
and GTOS MODIS Moderate-Resolution Imaging Spectroradiometer
GAP National Gap Analysis Program MOS Marine Observation Satellite
GCP Ground control point MRLC Multi-ResolutionLandCharacteristics
GEMS Global Environment Monitoring System Consortium
GEO Group on Earth Observations MSS Multispectral scanner
GEOSS Global Earth Observation System of Systems MTF Modular transfer function
GIS Geographic information system MTSAT Multipurpose transport satellite by JAXA
GLAS Geoscience Laser Altimeter System NAIP National Agriculture Imagery Program
GNSS Global Navigation Satellite System NALC North American Landscape Characterization
GOES Geosynchronous Orbiting Environmental NAOS North American Atmospheric Observing
Satellite System
GOES Geostationary Operational Environmental NCAR National Center for Atmospheric Research
Satellite (United States)
GOMOS Global Ozone Monitoring by Occultation of NASA National Aeronautics and Space Administration
Stars NAPP National Aerial Photography Program
GOOS Global Ocean Observing System NAWQA National Water Quality Assessment Program
GOS Global Observing System (of the WMO) (a USGS program)
GOMS Geostationary Operational Meteorological NCDC National Climatic Data Center
Satellite nDSM Normalized digital surface model
GPM Global Precipitation Measurement NDVI Normalized difference vegetation index
GPS Global Positioning System NED National Elevation Dataset
GSD Ground sample distance NEXRAD Next-Generation Radar
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 5
NHAP National High Altitude Program TOPEX Ocean Topography Experiment
NIMA National Image and Mapping Agency TOPSAR Topographic Synthetic Aperture Radar
NIR Near infrared TRMM Tropical Rainfall Measuring Mission
NLCD National Land Cover Data UAS Unmanned aircraft system
NOAA National Oceanic and Atmospheric UAV Unmanned aerial vehicle
Administration UHF Ultrahigh frequency
NPOESS National Polar-Orbiting Environmental USGS United States Geological Survey
Satellite System UV Ultraviolet
NRCS National Resources Conservation Service VIS Visible spectrum
(a USDI agency) VMS Vessel monitoring system
NRI National Resources Inventory VNIR Visible near-infrared sensor
NRSC National Remote Sensing Centre (United WFOV Wide field of view
Kingdom) WRS Worldwide Reference System
NSSDC National Space Science Data Center X-SAR X-band synthetic aperture radar
NWS National Weather Service
OCTS Ocean Color and Temperature Scanner
OCX Operational Control System 1.1 introduction
OSCAR Ocean Surface Current Analysis Real Time
1.1.1 Remote Sensing Definition and History
OSTM Ocean Surface Topography Mission
PALSAR Phased ArrayType L-Band Synthetic The American Society of Photogrammetry and Remote
Aperture Radar Sensing defined remote sensing as the measurement or
PBS Public Broadcasting Service acquisition of information of some property of an object or
QuickBird Satellite from DigitalGlobe, a private com- phenomenon, by a recording device that is not in physical
pany in United States or intimate contact with the object or phenomenon under
RADAR Radio detection and ranging study (Colwell etal., 1983). Environmental Systems Research
RADARSAT Canadian radar satellite Institute (ESRI) in its geographic information system (GIS)
RapidEye Satellite constellation from RapidEye, a dictionary defines remote sensing as collecting and inter-
German company preting information about the environment and the surface
RESOURCESAT Satellite launched in India of the earth from a distance, primarily by sensing radiation
RGB Red, green, blue that is naturally emitted or reflected by the earths surface
RKA Russian Federal Space Agency or from the atmosphere, or by sending signals transmitted
S/N Signal-to-noise ratio from a device and reflected back to it (ESRI, 2014). The usual
SAR Synthetic aperture radar source of passive remote sensing data is the measurement of
SeaWiFS Sea-Viewing Wide Field-of-View Sensor reflected or transmitted electromagnetic radiation (EMR)
SEVIRI Spinning Enhanced Visible and Infrared from the sun across the electromagnetic spectrum (EMS);
Imager this can also include acoustic or sound energy, gravity, or the
SIRS Satellite infrared spectrometer magnetic field from or of the objects under consideration. In
SLAR Side-looking airborne radar this context, the simple act of reading this text is considered
SODAR Sonic detection and ranging remote sensing. In this case, the eye acts as a sensor and senses
SONAR Sound navigation and ranging the light reflected from the object to obtain information about
SPOT Systeme Probatoire DObservation De La the object. It is the same technology used by a handheld camera
Terre (France) to take a photograph of a person or a distant scenic view. Active
SSM Special sensor microwave remote sensing, however, involves sending a pulse of energy
SSM/I Special sensor microwave/imager and then measuring the returned energy through a sensor (e.g.,
SSM/T Special sensor microwave/temperature Radio Detection and Ranging [RADAR], Light Detection and
SSM/T-2 Special sensor microwave/water vapor Ranging [LiDAR]). Thermal sensors measure emitted energy
SSS Sea surface salinity by different objects. Thus, in general, passive remote sensing
SWIR Shortwave infrared sensor involves the measurement of solar energy reflected from the
SST Sea surface temperature Earths surface, while active remote sensing involves synthetic
SWAT Soil and Water Assessment Tool (man-made) energy pulsed at the environment and the return
TES Tropospheric Emission Spectrometer signals are measured and recorded.
TIMS Thermal infrared multispectral scanner Remote sensing functions in harmony with other spatial data
TIR Thermal infrared collection techniques or tools of the mapping sciences, includ-
TM Thematic Mapper ing cartography and GIS (Jensen, 2009). Remote sensing is a
TOMS Total Ozone Mapping Spectrometer tool or technique similar to mathematics where each sensor is
2016 by Taylor & Francis Group, LLC
6 Remotely Sensed Data Characterization, Classification, and Accuracies
used to measure from a distance the amount of EMR exiting an aerial platforms, which are used by law enforcement and mili-
object, or geographic area, then extracting valuable information tary sectors for surveillance purposes (Briney, 2014). Unmanned
from the data using mathematical and statistical algorithms aerial vehicles (UAVs) are one of the most important advance-
(Jensen, 2009). In fact, the science and art of obtaining reliable ments in aerial remote sensing in recent times and used even
measurements using photographs is called photogrammetry for fighting stealth wars (Russo etal., 2006). UAVs are now con-
(American Society of Photogrammetry, 1952, 1966). According trolled from home base through onboard Global Positioning
to Aronoff (2004), remote sensing is the technology, science, and System (GPS) and forward-looking infrared (FLIR) and/or
art of obtaining information about objects from a distance over videography (FLIR) technology (Jensen, 2009). Other advances
regions too costly, dangerous, or remote for human observers to in remote sensing technology include RADAR, LiDAR, sound
directly access. This includes target areas beyond the limits of navigation and ranging (SONAR), sonic detection and ranging
human ability. Avery and Berlin (1992) simply defined remote (SODAR), microwave synthetic aperture radar (SAR), infrared
sensing as the technique of reconnaissance from a distance in sensors, hyperspectral imaging, spectrometry, Doppler radar,
which information about objects are obtained through the anal- and space probe sensors, in addition to improvements in con-
ysis of data collected by special instruments that are not in phys- ventional aerial photography, hyperspectral imaging, and imag-
ical contact with the objects of investigation. Therefore, remote ing spectroradiometer.
sensing is different from in situ observation that involves use
of sensing instruments in direct contact with the objects under
1.1.2 Data collection by Remote
consideration but not proximal insitu sensing, where data are
Sensing and Usage
collected from the insitu objects with controlled spectra. Insitu
observations of the Earths land cover and other physical phe- Remote sensing in comparison to other methods of data collec-
nomena are often time-consuming, costly, and impossible to tion is much more advantageous as it provides an overview of the
have a full spatial coverage. In contrast, remote sensing acquires Earths phenomena that allow users to discern patterns and rela-
information about an object or phenomenon on the Earths sur- tionships not apparent from the ground (Aronoff, 2004). EMR
face and even beneath the surface without making physical con- (usually solar energy or any form of energy) is reflected or emit-
tact with the object by use of propagated signals, for example, ted from the object and is detected by the devices called remote
EMR (Lillesand etal., 2004). sensors or simply sensors (similar to cameras or scanners) fitted
The very first successful photographic image of nature, on platforms such as aircrafts, satellites, or ships. Traditional
considered as a permanent photograph or remotely sensed remote sensing involves two basic processes: data acquisition
image, was captured by a French native Joseph Nicephore and data analysis. Figure 1.1 illustrates the data collection and
Niepce (Gernsheim and Gernsheim, 1952). Modern remote processing involved in a typical remote sensing operation.
sensing started in 1858 at Paris, France, with Gaspard-Felix According to Lillesand etal. (2004), the elements of remote
Tournachon first taking aerial photographs of the city of Paris sensing data acquisition involve (1) energy sources such as
from a hot air balloon (Briney, 2014). Remote sensing contin- solar energy, self-produced heat/light energy, or sound energy;
ued to grow from there but at a slow rate. During the U.S. Civil (2) propagation of energy through the atmosphere; (3) energy
War (18611865), messenger pigeons, kites, and unmanned bal- interaction with the Earths surface features; (4) retransmis-
loons were flown over enemy territory with cameras attached sion of energy through the atmosphere; (5) sensing platforms
to them, which is considered as one of the very first planned like airborne, spaceborne, and onboard ship sensors; and (6)
usages of remote sensing technology (Trenear-Harvey, 2009; resultant data in the form of pictorial or digital format. There
Briney, 2014). During World War I and II, the first government- are numerous factors that affect the remote sensing data collec-
organized air photography or photogrammetry missions were tion process such as (1) solar position, (2) atmospheric condi-
developed for military surveillance and later adopted by many tion, (3) weather and meteorology, (4) season of data collection,
other countries for other applications, including land survey- (5) ground condition, (6) sensor characteristics, and (7) sensor
ing (Briney, 2014). After World War II, during the Cold War position. As one can realize, most of the aforementioned factors
era, aerial photo reconnaissance became much more prominent directly relate to the intensity of the solar energy in terms of
and prevalent between the United States and the Soviet Union enhancing or attenuating it. On the other hand, remote sens-
to collect information about each other. In December 1954, ing data analysis involves (1) data interpretation using various
U.S. President Dwight Eisenhower approved the U-2 recon- interpreting/viewing devices like computers and software, (2)
naissance program (Brugioni and Doyle, 1997) for World War analyzing the data in collaboration with other geospatial data,
II reconnaissance from space. In recent years, tremendous and (3) applying the interpreted and analyzed data for Earth
advancements in technology have resulted in a rapid growth management decision support.
in the remote sensing industry (Jensen, 2009). During the last Basically, in terms of the energy source and data collec-
few decades, growth in the civilian sector has far surpassed the tion mode, there are two types of remote sensing: (1) passive
defense and military applications. However, the recent years remote sensing, when reflected or emitted energy from an
have seen new applications of miniature remote sensors or cam- object or phenomenon is recorded by sensors mounted on air-
era systems that are mounted on both manned and unmanned borne or spaceborne platforms and (2) active remote sensing,
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 7
Users and application Sensing systems Source of energy
Agriculture
Forestry Sun
Hydrology
Sensor
Geology
Oceanography
Meteorology Photomechanical
Environment scanner
data
n of
issio EMR
nsm
Tra reflection
Flight direction
Receiving station
Electro-magnetic
Picture element radiation (EMR)
in digital form emission
Photogrammetry Reconstructed image
Pixel
Scan line
Image processing
Data interpretation
and analysis Image data
FIg u r e 1.1 Illustration of remote sensing data collection for user application. Sensors onboard aircraft or satellite platforms collect and record
reflected energy from target features. The reflected data are collected across the electromagnetic spectrum and are used for various studies such as
agriculture and forestry.
when reflectance of synthetic light (nonsolar) that is actively light energy into electrons, which is then measured and con-
pulsed or emitted from an aircraft, satellite, or any other verted into radiometric intensity values (McGlone, 2004).
energy-producing/recording platform is recorded (Schott, 2007; However, the active remote sensing systems such as RADAR,
Schowengerdt, 2007). Passive remote sensing systems such as LiDAR, SONAR, and GPS detect not only the backscattered or
film photography have become obsolete due to several disad- energy reflecting off of Earth objects but also record time lag
vantages, including low exposure time, low spectral range (i.e., and intensity (Schott, 2007; Schowengerdt, 2007). Thus, unlike
usually below 700 nm only), low dynamic range (saturation/ passive remote sensing, which only detects the location of an
minimum limit) of only around 100, and other image charac- object, active remote sensing uses the time delay between trans-
teristic issues, including linearity and photometric accuracy. mission and reception of energy pulse to establish the location,
Currently, most systems use charge-coupled device (CCD) height, depth, speed, and direction of an object. Figure 1.2 illus-
and complementary metal-oxide semiconductor (CMOS) trates the difference between active and passive remote sensing
verycommon in digital cameras and camcorders that convert with examples of different platforms.
Sun
Satellite Ship
Solar Transmitted
radiation signal Airplane
Reflectance/
emittance
Passive remote sensing Active remote sensing
FIg u r e 1.2 Illustration of passive vs. active remote sensing with different platforms. In passive remote sensing, the sun acts as the source of
energy. Sensors then measure the reflected solar energy off the targets. Active remote sensing sends a pulse of synthetic energy and measures the
energy reflected off the targets.
2016 by Taylor & Francis Group, LLC
8 Remotely Sensed Data Characterization, Classification, and Accuracies
Zenith
Satellite, at altitude H
Sun
Atmospheric
Meteorological parameters
optical parameters
Cloud cover
Visual range, V
Surface pressure
Aerosol type (phase function)
Relative humidity
Signal-scattering albedo
1 i r r
180
Background
reflectance, b Sensor IFOV
Target reflectance, t
FIg ur e 1.3 Illustration of factors affecting apparent reflectance in passive remote sensing. (Adapted and modified from Bowker, D.E. etal.,
Spectral reflectances of natural targets for use in remote sensing studies, NASA reference publication # 1139, Hampton, VA, 1985.) Note that the
solar energy passes through the atmosphere and hits the target (any object on the planet). Then, depending on the type of the object, certain portion
of the energy reflects off the target, again passes back through the atmosphere, and is captured by the sensor. Such reflected energy can be captured
in various portions of the electromagnetic spectrum (EMS) (e.g., visible, near infrared, shortwave infrared). Also, there can be several wave bands
within each portion of the EMS.
National Aeronautics and Space Administration (NASA) ref- (Figure 1.4) at the velocity of light (c = 2.998 108 m/s); the dis-
erence publication # 1139 by Bowker et al. (1985) list different tance from one wave peak to another is wavelength, , and the
factors affecting apparent reflectance, which subsequently deter- number of peaks passing a fixed point in space per unit time is
mines the object or phenomena on the Earth or in the atmo- wave frequency, (Lillesand etal., 2004).
sphere. These factors are (1) viewing geometry (solar zenith angle, According to the Murai (1993), EMR consists of four elements
viewing angle, azimuthal angle, relative azimuthal angle, and such as (1) frequency/wavelength, (2) transmission direction, (3)
altitude of the sensor), (2) meteorological parameters (relative amplitude, and (4) polarization. These four features are impor-
humidity, cloud cover, and surface pressure), (3) atmospheric tant for remote sensing as each corresponds to different features.
optical parameters (aerosol optical thickness, atmospheric Wavelength/frequency corresponds to the color of an object in
visual range, aerosol types, and single-/multiple-scattering the visible region, which is represented by a unique characteris-
albedo), and (4) target and background parameters (target size, tic curve, aka reflectance curve correlating wavelength and the
target reflectance, background reflectance, instantaneous field radiant energy from the object (Figure 1.5). Transmission direc-
of view [IFOV], and bidirectional reflectance function [BDRF]). tion and amplitude, corresponding to the direction of propagation
Figure 1.3 extracted from Bowker etal. (1985) pictorially depicts and magnitude of the waves, are influenced by the spatial location
these factors affecting target reflectance measured by a passive and shape of the objects. Finally, the plane of polarization, that is,
remote sensing system.
= wavelength
(distance between
1.1.3 Principles of electromagnetic successive wave
peaks)
Spectrum in Remote Sensing
Light or radiant flux in the form of electromagnetic energy
(EME), which includes visible light, infrared, radio waves, heat,
ultraviolet (UV) rays, and x-rays, is the primary form of energy
used in remote sensing (Lillesand et al., 2004). EMR is a car-
Electric field Magnetic c = velocity
rier of EME by transmitting the oscillation of the electromag- field
= frequency of light
netic field through space or matter based on Maxwells equation Basic wave theory (number of cycles per second
(Murai, 1993). equation passing of a fixed point)
EMR has characteristics of both wave and particle motion. c =
EMR essentially follows the basic wave theory (c = ), which
describes that EME travels in a harmonic or sinusoidal fashion FIg u r e 1 .4 Illustration of an electromagnetic wave.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 9
Visible Near-Infrared Mid-Infrared
80
70
60
Reflectance (%)
50
40
30 Clear river water
Turbid river water
20 Vegetation
Silty clay soil
10 Muck soil
Green vegetation
Dry vegetation
0
0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6
Wavelength () (m)
FIg ur e 1.5 Typical reflectance curve of different objects generated with electromagnetic radiation frequency characteristics that are useful to
distinguish/interpret objects. (Adapted and modified from Murai, S., Remote Sensing Note, JARS, Tokyo, Japan, 1993.) Note how energy is reflected
or absorbed in various portions of the spectrum that forms the basis of remote sensing. For example, healthy vegetation absorbs heavily in the red
band (especially around 0.68 m) but reflects heavily in near-infrared 0.760.90 m.
orientation of the electric field of the radiant energy, is influenced landscape images provided the basis for some of the early
by the geometric shape of the objects under investigation (Murai, photogrammetric and stereo analysis of images, which are
1993). Therefore, EMR provides detailed information about the still considered as foundations for aerial surveys. The air-
object(s) under investigation based on the spectral characteristics crafts used for airborne remote sensing are usually single-
of the objects. or twin-engine propeller or turboprop platforms. However,
NASA reference publication # 1139 by Bowker etal. (1985) is small jets (Learjet) are used for high-altitude reconnaissance
one of the most comprehensive articles on spectral reflectance or mapping surveys. These aerial platforms usually fly at
curves and curve development processes of natural targets for about 1,20012,000 ft above the terrain with flying speeds of
use in remote sensing studies. Most investigations based on about 100 knots. Aerial platforms provide the main advan-
remote sensing data are models that are developed between the tage of responding to the need of the user application. For
amount of electromagnetic or light energy reflected, emitted, example, in an emergency forest fire and flooding event,
transmitted, or backscattered from the object at different fre- an aerial survey can be quickly planned and implemented.
quencies of the EMS and the biophysical/chemical properties Thus, an aerial platform is a quick response system that one
of the object or phenomena under investigation (Jensen, 2009). can easily adapt to changes in weather conditions. Moreover,
The EMS is divided into several wavelength (frequency) regions, airborne platforms enable near-real-time review of acquired
such as gamma rays (106 105 m), x-ray (105102 m), UV data and provide great control of data quality. In contrast to
(0.10.4m), visible (0.40.7m), infrared (0.7103m), micro- aerial platforms, satellite platforms are enormously expen-
waves (103106 m), and radio waves (>106 m). Panchromatic sive and complex, and the development and deployment of
(i.e., grayscale) and color (i.e., red, green, blue [RGB]) imaging sensors as payload on satellites can take 510 years. Unlike
systems have dominated electro-optical sensing in the visible aerial platforms, satellite platforms due to orbital charac-
region of the EMS (Figure 1.6), which describe the efficacy of teristics have limitations of revisiting the same land area at
a remotely sensed imagery (Shaw and Burke, 2003). Figure 1.7 user-needed time interval. This revisit cycle that relates to the
depicts the bandwidth (wavelength) of different bands associ- time taken by the satellite to take a subsequent image acquisi-
ated with the spectral regions, along with the different remote tion of the same land area might not be practical. However,
sensing systems that acquire data in these bands. once a satellite-based imaging sensor becomes operational,
the spaceborne image data are usually very consistent and
1.2 Remote Sensing Platforms allow end users to develop robust applications such as land
and Sensor characteristics cover change analysis, particularly for large tracts of the
Earths surface. Hence, depending on the resolutions pro-
Traditional remote sensing has its roots in aircraft-based plat- vided by the onboard sensors, satellite platforms offer a wide
forms where photographic instruments capture Earth features array of scale-based mapping. Over the past few decades,
as either single images or dual overlapping stereo images. The numerous satellites have been launched by various private
2016 by Taylor & Francis Group, LLC
10 Remotely Sensed Data Characterization, Classification, and Accuracies
0.3 104
104 3 104 105 106
W V O Ka K Xu X C S L P
Wavelength (m)
106 105 104 103 102 101 100 101 102 103 104 105 106 107 108 109 1010 1011
EHF SHF UHF VHF HF MF LF VLF
Gamma ray x-ray Ultraviolet Infrared Microwaves Radio waves
Note:
The visible spectrum EHF = Extremely high frequency
SHF = Super high frequency
Ultraviolet Violet Blue Green Yellow Red Infrared UHF = Ultra high frequency
VHF = Very high frequency
400 480 540 580 700 HF = High frequency
MF = Medium frequency
Wavelength (nm)
LF = Long frequency
VLF = Very long frequency
For microwaves band (W, V, .., P)
Information refer Table 1.
FIg u r e 1.6 Electromagnetic spectrum on which remote sensing systems are based. For example, visible portion of electromagnetic radiation is
in 400700 nm. The data can be captured in narrowbands (e.g., 1 nm) or over broad wave bands (e.g., a single band over 630690 nm).
enterprises and government organizations to acquire remote and nonimaging sensors (Murai, 1993). Microwave radiometer,
sensing data. Remote sensors are energy-sensitive devices magnetic sensor, gravimeter, Fourier spectrometer, microwave
mounted on the particular satellite to view and take images altimeter, laser water depth meter, laser distance meter, etc.,
of the Earth in different bands of the EMS (Avery and Berlin, are examples of nonimaging scanners (Murai, 1993). Figure 1.8
1992). In addition to the two types of sensors (discussed in graphically depicts the sensor types.
Section 1.1.2) used in remote sensingpassive and active According to NASAs Earth Observation System Data
sensorsremote sensors are classified into various types and Information System (EOSDIS), accelerometer, radi-
based on their scanning and imaging mechanism. ometer, imaging radiometer, spectrometer, spectroradiom-
One of the main advantages of using active systems relate eter, hyperspectral radiometer, and sounders are examples
to the characteristics of the active energy (RADAR, LiDAR, of passive remote sensors. Linear Imaging Self-Scanning
etc.) used, which is solar energy independent and least affected Sensor (LISS), advanced very-high-resolution radiometer
byatmospheric constituents. This provides huge advantages (AVHRR), Coastal Zone Color Scanner (CZCS), Sea-Viewing
for remote sensing over regions such as tropical environments Wide Field-of-View Sensor (SeaWiFS), Moderate-Resolution
where clouds and rain are frequent weather events that inter- Imaging Spectroradiometer (MODIS), Active Cavity Radiometer
fere with traditional optical/passive systems. Furthermore, Irradiance Monitor (ACRIM II and III), Advanced Spaceborne
active sensors do not require solar energy and hence can Thermal Emission and Reflection Radiometer (ASTER),
operate during night. Additionally, active systems have better Imaging Infrared Radiometer (IIR), Clouds and the Earths
capability of sensing vegetation and soil attributes that are Radiant Energy System (CERES), special sensor microwave
dependent on moisture content, thus having immense value radiometer (SMMR), airborne visible/infrared imaging spec-
in various applications, including hydrology, geology, glaci- trometer (AVIRIS), Polarization and Directionality of the
ology, forestry, and agriculture. The disadvantages include Earths Reflectance (POLDER), and Atmospheric Infrared
lower spectral characteristics, complicated data processing, Sounders (AIRSs) are some of the prominent passive sen-
massive data volume, and higher cost. sors used in remote sensing data collection in the present day
Passive and active sensors are divided into nonscanning (EOSDIS, 2014). As discussed in the previous section, RADAR,
and scanning types, which in turn are divided into imaging ranging instruments, scatterometer, LiDAR, laser altimeter,
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 11
Very short wave (VHF): 30300 MHz (110 m)
Short wave (HF): 330 MHz (10100 m) Astronomical studies, communication
Radio
Medium wave (MF): 0.33 MHz (0.11 km) applications
Long wave (LF): 30300 kHz (110 km)
Very long wave (VLF): 330 kHz (10100 km)
P band: 0.31 GHz (30100 cm)
L band: 12 GHz (1530 cm)
Microwave
S band: 24 GHz (7.515 cm) GPS satellites, microwave remote
C band: 48 GHz (3.875 cm) sensing (SAR imaging) for cloud cover
X band: 812.5 GHz (2.43.8 cm) image analysis and water studies
Ku band: 12.518 GHz (1.72.4 cm)
K band: 1826.5 GHz (1.11.7 cm)
Near infrared (NIR): 0.7 to 1.5 m
Near IR
Short wave infrared (SWIR): 1.5 to 3 m All multispectral remote sensing
Mid wave infrared (MWIR): 3 to 8 m systems, like Landsat, SPOT, and some
Long wave infrared (LWIR): 8 to 15 m CIR aerial images, etc.
Far infrared (FIR): >15 m
Red: 610700 nm
Orange: 590610 nm Most aerial imaging system, all satellite
Visible
Yellow: 570590 nm and hyperspectral remote sensing
Green: 500570 nm systems for visual analysis
Blue: 450500 nm
Ultra-violet
For computer/lab imaging and food
3 to 400 nm (meat) quality determination
X-ray
3 Medical systems for clinical IT
0.1 to 10 nm
applications
Gamma ray
Food sterilization system, medical
<103 nm
sanitation, and nuclear medicine
FIg u r e 1.7 Different bands (bandwidth) associated with the electromagnetic radiation regions and their remote sensing systems. Gathering
data in various spectral ranges is extremely important for characterizing Earth systems. For example, radar penetrates the clouds and is important
to gather data during cloudy days when optical remote sensing is infeasible.
and sounders are examples of active sensors (EOSDIS, 2014). difference between different bands are some of the spec-
Table 1.1 provides a detailed list of remote sensors with the tral characteristics that define the types of optical sensors
instrument name, type of sensor, platform, data center, and (Murai, 1993). Detection accuracy, signal-to-noise ratio
other descriptions about the sensors. (S/N), dynamic range, quantization level, sensitivity differ-
Remote sensing optical sensors are characterized by spec- ence between pixels, linearity of sensitivity, and noise equiv-
tral, radiometric, and geometric performance (Murai, 1993). alent power are a few radiometric characteristics of optical
Observation range of the electromagnetic wave, center wave- sensors (Murai, 1993). Field of view (FOV), IFOV, registra-
length of a band, changes at both ends of a band, sensitiv- tion between different spectral bands, modular transfer
ity of a band, polarization sensitivity, and ratio of sensitivity function (MTF), and optical distortions are examples of
2016 by Taylor & Francis Group, LLC
12 Remotely Sensed Data Characterization, Classification, and Accuracies
Sensor
type
Passive Active
sensor sensor
Non- Non-
scanning Scanning scanning Scanning
Non-imaging Imaging Imaging Non-imaging Imaging
- Microwave Camera
radiometer Image pane Object plane - Microwave Object plane Image plane
- Magnetic radiometer scanning
scanning scanning scanning
sensor - Microwave
- Gravimeter - Monochrome altimeter
- Fourier - Natural color - Laser water
spectrometer - Infrared depth meter
- Others - TV camera - Optical - Real aperture Passive phase
- Color infrared mechanical - Laser distance
- Solid scanner meter radar array radar
- Others scanner - Synthetic
- Microwave aperture radar
radiometer
FIg u r e 1.8 Remote sensing sensortype classification. (Adapted and modified from Murai, S., Remote Sensing Note, JARS, Tokyo, Japan, 1993.)
geometric characteristics of optical sensors that classify the of Landsat satellite series, Landsat 8, provides 12-bit data
sensors (Murai, 1993). These sensor characteristics deter- with DN values ranging from 0 to 4095. Temporal resolution
mine the spatial, spectral, radiometric, and temporal resolu- refers to the revisit frequency/time of the sensor to a specific
tions of remote sensing data. location on the Earths surface. For example, Landsat revisit
time is 16days over a specific geographic location on Earth.
Table 1.2 provides a list of selected remote sensing systems
1.2.1 image characteristics
(sensors) with detailed data characteristics including the spa-
Different sensors continuously scan the Earths surface to tial, spectral (band ranges), and frequency of revisit over a
produce imagery. Each image is a matrix of pixels or pic- location on Earth.
ture elements defined by columns and rows as shown in
Figure 1.9. Each pixel records a numeric value representative 1.2.1.1 Spatial Resolution
of the brightness or intensity level of the reflected energy as Spatial resolution implies the unit ground area for which the
discussed in Section 1.2. The image data produced by different sensor records the reflected energy providing a unique DN
sensors on satellite systems have unique characteristics that or brightness value (BV) (refer to Figure 1.9). For example,
relate to the sensors resolutionsspatial, spectral, radiomet- a Landsat sensor senses a ground space that is 30 m 30 m.
ric, and temporal. Usually, the image spatial resolution is equal to the ground sam-
Radiometric resolution refers to the data depth indicative ple distance (GSD), which is the smallest discernible detail in an
of the sensitivity of the sensor to incoming energy. About image (Gonzalez and Woods, 2002). The GSD could be smaller
8-bit data have higher contrast (0255 digital number [DN] than the spatial resolution of an image acquired by a remote
range) and higher radiometric resolution than a 6-bit sensor sensing system, and they vary from a fraction of a meter to tens
that provides data in a lower range (063 DN range) and thus of meters (Shaw and Burke, 2003). The image spatial resolution
lower contrast. For example, Landsat 7 ETM+ sensor pro- is recognized primarily by the sensor aperture and platform
vides 8-bit data for individual bands, while the latest version altitude. Platform altitude is loosely constrained by the class of
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 13
TABLe 1 .1 List of Selected Active and Passive Sensors and Their Types
Instrument Type Platform Data Center Description
Passive sensors
Power radiometers and imagers
ACRIM II Total power radiometer UARS LaRC ASDC Measures total solar irradiance.
ACRIM III Total power radiometer ACRIMSAT LaRC ASDC Measures total solar irradiance.
TIM Total power radiometer SORCE GES DISC Measures total solar irradiance.
LIS Imager TRMM GHRC DAAC Detects intracloud and cloud-to-ground lightning, day and night.
WFC Wide Field Camera CALIPSO LaRC ASDC Fixed, nadir-viewing imager with a single spectral channel covering the
620670nm region.
Multispectral instruments
AMPR Microwave radiometer ER-2 and DC-8 GHRC DAAC Cross track scanning total power microwave radiometer with four
channels centered at 10.7, 19.35, 37.1, and 85.5 GHz (FIRE ACE,
Teflon-B, TRMM-LBA, CAMEX-4. TCSP, TC4 Projects).
AMSR-E Multichannel Aqua NSIDC DAAC Measures precipitation, oceanic water vapor, cloud water, near-surface
microwave radiometer GHRC DAAC wind speed, sea and land surface temperature, soil moisture, snow
cover, and sea ice. Provides spatial resolutions of 5.4, 12, 21, 25, 38, and
56km and a 0.25 resolution.
ASTER Multispectral Terra LP DAAC Measures surface radiance, reflectance, emissivity, and temperature.
radiometer ORNL DAAC Provides spatial resolutions of 15, 30, and 90 m.
AVHRR Multispectral NOAA/POES GES DISC Four or six bands, depending on platform. Telemetry resolutions are
radiometer NSIDC DAAC 1.1km (HRPT data), 4km (Pathfinder V5 and GAC data), 5, and
ORNL DAAC 25km spatial resolution.
PO.DAAC
CERES Broadband scanning Aqua LaRC ASDC Four to six channels (shortwave, longwave, total). Measures atmospheric
radiometer Terra and surface energy fluxes.
TRMM Provides 20km resolution at nadir.
NPP
IIR IIR CALIPSO LaRC ASDC Nadir-viewing, nonscanning imager having a 64km swath with a pixel
size of 1km. Provides measurements at three channels in the TIR
window region at 8.7, 10.5, and 12.0mm.
MAS Imaging spectrometer NASA ER-2 GES DISC Fifty spectral bands that provide spatial resolution of 50 m at typical
aircraft GHRC DAAC flight altitudes.
LaRC ASDC
ORNL DAAC
MISR Imaging spectrometer Terra LaRC ASDC Obtains precisely calibrated images in four spectral bands, at nine
ORNL DAAC different angles, to provide aerosol, cloud, and land surface data.
Provides spatial resolution of 250 m to 1.1km.
MODIS Imaging Aqua GES DISC Measures many environmental parameters (ocean and land surface
spectroradiometer Terra GHRC DAAC temperatures, fire products, snow and ice cover, vegetation properties
LP DAAC and dynamics, surface reflectance and emissivity, cloud and aerosol
MODAPS properties, atmospheric temperature and water vapor, ocean color and
NSIDC DAAC pigments, and ocean biological properties).
OBPG Provides moderate spatial resolutions of 250 m (bands 1 and 2), 500 m
ORNL DAAC (bands 37), and 1km (bands 836).
PO.DAAC
SSM/I Multispectral DMSP GHRC DAAC Has seven channels and four frequencies. Measures atmospheric, ocean,
microwave radiometer LaRC ASDC and terrain microwave brightness temperatures, which are used to
NSIDC DAAC derive ocean near-surface wind speed, atmospheric integrated water
PO.DAAC vapor, and cloud/rain liquid water content and sea ice extent and
ORNL DAAC concentration.
SMMR Multispectral NIMBUS-7 GES DISC Ten channels. Measures SSTs, ocean near-surface winds, water vapor
microwave radiometer LaRC ASDC and cloud liquid water content, sea ice extent, sea ice concentration,
NSIDC DAAC snow cover, snow moisture, rainfall rates, and differential of ice types.
PO.DAAC
TMI Multispectral TRMM GES DISC TMI measures the intensity of radiation at five separate frequencies:
microwave radiometer GHRC DAAC 10.7, 19.4, 21.3, 37, and 85.5 GHz. TMI measures microwave
brightness temperatures, water vapor, cloud water, and rainfall
intensity.
(Continued )
2016 by Taylor & Francis Group, LLC
14 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 .1 (c ontinued ) List of Selected Active and Passive Sensors and Their Types
Instrument Type Platform Data Center Description
Hyperspectral instruments
AVIRIS Imaging spectrometer Aircraft ORNL DAAC 224 contiguous channels, approximately 10nm wide. Measurements are
used to derive water vapor, ocean color, vegetation classification,
mineral mapping, and snow and ice cover (BOREAS Project).
SOLSTICE Spectrometer SORCE GES DISC Measures the solar spectral irradiance of the total solar disk in the UV
wavelengths from 115 to 430nm.
Polarimetric instruments
POLDER Polarimeter Aircraft ORNL DAAC Measures the polarization and the directional and spectral
characteristics of the solar light reflected by aerosols, clouds, and the
Earths surface (BOREAS Project).
PSR Microwave polarimeter Aircraft GHRC DAAC Measures wind speed and direction (CAMEX-3 Project).
Ranging and sounding instruments
ACC Accelerometer GRACE PO.DAAC The Onera SuperSTAR Accelerometer measures the nongravitational
forces acting on the GRACE satellites.
AIRS Sounder Aqua GES DISC Measures air temperature, humidity, clouds, and surface temperature.
Provides spatial resolution of ~13.5km in the IR channels and ~2.3km
in the visible. Swath retrieval products are at 50km resolution.
AMSU Sounder Aqua GES DISC Has 15 channels. Measures temperature profiles in the upper
GHRC DAAC atmosphere. Has a cloud filtering capability for tropospheric
temperature observations. Provides spatial resolution of 40km at
nadir.
HAMSR Sounder DC-8 GHRC DAAC Measures vertical profiles of temperature and water vapor, from the
surface to 100mb in 24km layers (CAMEX-4, NAMMA Projects).
HIRDLS Sounder Aura GES DISC Measures infrared emissions at the Earths limb in 21 channels to obtain
profiles of temperature, ozone, CFCs, various other gases affecting
ozone chemistry, and aerosols at 1km vertical resolution. In addition,
HIRDLS measures the location of polar stratospheric clouds.
MLS Sounder Aura GES DISC Five broadband radiometers and 28 spectrometers measure microwave
thermal emission from the Earths atmosphere to derive profiles of
ozone, SO2, N2O, OH group, and other atmospheric gases,
temperature, pressure, and cloud ice.
MOPITT Sounder Terra LaRC ASDC Measures carbon monoxide and methane in the troposphere.
ORNL DAAC Is able to collect data under cloud-free conditions. Provides horizontal
resolution of ~22km and vertical resolution of ~4km.
OMI Multispectral Aura GES DISC 740 wavelength bands in the visible and UV. Measures total ozone and
radiometer profiles of ozone, N2O, SO2, and several other chemical species.
TES Imaging Spectrometer Aura LaRC ASDC High-resolution imaging infrared Fourier transform spectrometer that
operates in both nadir and limb-sounding modes.
Provides profile measurements of ozone, water vapor, carbon monoxide,
methane, nitric oxide, nitrogen dioxide, nitric acid carbon dioxide, and
ammonia.
Active sensors
Radar and laser (LiDAR)
ALT-A, -B Radar altimeter TOPEX/ PO.DAAC Dual-frequency altimeter that measures height of the satellite above the
Poseidon sea (satellite range), wind speed, wave height, and ionospheric
correction.
CALIOP Cloud and aerosol CALIPSO LaRC ASDC Two-wavelength polarization-sensitive LiDAR that provides high-
LiDAR resolution vertical profiles of aerosols and clouds.
Altimeters: Laser altimeter ICESat NSIDC DAAC The main objective is to measure ice sheet elevations and changes in
radar and elevation through time. Secondary objectives include measurement of
laser (LiDAR) cloud and aerosol height profiles, land elevation and vegetation cover,
GLAS and sea ice thickness.
Poseidon-1 Radar altimeter TOPEX/ PO.DAAC Single-frequency altimeter that measures height of the satellite above the
Poseidon sea (satellite range), wind speed, and wave height.
(Continued)
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 15
TABLe 1 .1 (c ontinued ) List of Selected Active and Passive Sensors and Their Types
Instrument Type Platform Data Center Description
Poseidon-2 Radar altimeter Jason-1 PO.DAAC Measures sea level, wave height, wind speed, and ionospheric
correction.
SAR SAR ERS-1 ASF SDC Provides high-resolution surface imagery at 7240 m.
ERS-2 NSIDC DAAC Multiple polarizations are utilized by some SAR instruments.
JERS-1 ORNL DAAC
RADARSAT-1
PALSAR
UAVSAR
KBR Ranging instrument GRACE PO.DACC The dual-frequency KBR instrument measures the range between the
GRACE satellites to extremely high precision.
Scatterometers
NSCAT Radar scatterometer ADEOS-I PO.DAAC Dual Fan-Beam Ku Band that measures ocean vector winds at a nominal
grid resolution of 25km.
SASS Radar scatterometer Seasat PO.DAAC Dual Fan-Beam Ku Band that measures ocean vector winds at a nominal
grid resolution of 25km.
Seawinds Radar scatterometer QuikSCAT PO.DAAC Dual Pencil-Beam Ku Band that measures ocean vector winds at a
ADEOS-II nominal grid resolution of 25km.
Sounding instruments
CLS LiDAR ER-2 LaRC ASDC Determines vertical cloud structure (FIRE Project).
LASE LiDAR DC-8 GHRC DAAC Measures water vapor, aerosols, and clouds throughout the troposphere
(CAMEX-4, TCSP, NAMMA Projects).
PR Phased array radar TRMM GES DISC Measures 3D distribution of rain and ice. Provides horizontal resolution
ORNL DAAC of 250 m and vertical resolution of 5km.
VIL LiDAR Ground LaRC ASDC Determines vertical cloud structure (FIFE, FIRE, and BOREAS
ORNL DAAC Projects).
Source: Adapted and modified from Earth Observing System Data and Information System (EOSDIS), Earth Data, 2014, https://earthdata.nasa.gov/,
accessed on July 5, 2014.
sensor platform, that is, either spaceborne or airborne (Shaw and GSD, is commensurate with the footprint of the targets of inter-
Burke, 2003). Sensor aperture size, particularly for spaceborne est. For example, if the application requires observation of global
systems, determines the cost of the remote sensing system; land cover, then low-resolution imagery is suitable. On the other
many times to cut cost, the aperture size is made bigger, thus hand, if the application requires local land cover information
providing low-spatial-resolution image. According to Shaw and such as water stress in a corn field, then high-resolution imag-
Burke (2003), the best detection performance is expected when ery is ideal. According to Belward and Skoien (2014), land cover
the angular resolution of the sensor, specified in terms of the datasets at 1km resolution (National Oceanic and Atmospheric
55 201 67 200 98 33 211 211 67 98
212 33 66 155 200 203 199 201 215 68
201 201 33 65 65 77 208 199 87 99
56 56 188 215 215 77 77 165 166 211
88 88 33 89 200 177 201 201 98 201
201 201 84 88 74 75 98 97 54 98
200 178 200 33 65 178 69 111 65 201
200 201 33 128 201 33 98 98 201 188
201 178 201 78 34 98 95 111 78 202
168 168 32 65 65 33 200 33 201 165
Individual pixel (gray scale color) Individual pixel (digital number)
Gray scale image
FIg u r e 1.9 Basic matrix configuration of an image showing a pixel (picture element). Each pixel denotes an intensity value (e.g., reflected
energy, radar backscatter). Each pixel can have data values between 0 and 256 for 8-bit (quantization) data, 0 and 4096 for 12-bit data, and so on.
2016 by Taylor & Francis Group, LLC
16 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 .2 Data Characteristics of Selective Satellite Sensors
Image Characteristics
Spectral Radiometric Temporal
Sensor Spatial (m) (#) (bit) (Days) Band Range (m)
c oarse-resolution sensors
AVHRR 1000 4 11 Daily 0.580.68
0.7251.1
5.553.95
10.3010.95
10.9511.65
MODIS 250; 500; 1000 36 12 Daily 0.620.67
0.840.876
0.4590.479
0.5450.565
1.231.25
1.631.65
2.112.16
Multispectral sensors
Landsat 1, 2, 3 MSS 56 79 4 6 16 0.50.6
0.60.7
0.70.8
0.81.1
Landsat 4, 5 TM 30 7 8 16 0.450.52
0.520.60
0.630.69
0.760.90
1.551.74
10.412.5
2.082.35
Landsat 7 ETM+ 30 8 8 16 0.450.52
0.520.60
0.630.69
0.500.75
0.750.90
10.412.5
1.551.75
0.520.90 (p)
Landsat 8 15; 30; 100 11 12 16 0.430.45
0.450.51
0.530.59
0.640.67
0.850.88
1.572.29
0.500.68 (p)
1.361.38
10.6011.19
11.5012.51
ASTER 15; 30; 90 15 8 16 0.520.63
0.630.69
0.760.86
1.601.70
2.1452.185
(Continued )
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 17
TABLe 1 .2 (c ontinued ) Data Characteristics of Selective Satellite Sensors
Image Characteristics
Spectral Radiometric Temporal
Sensor Spatial (m) (#) (bit) (Days) Band Range (m)
2.1852.225
2.2352.285
2.2952.365
2.3602.430
8.1258.475
8.4758.825
8.9259.275
10.2510.95
10.9511.65
ALI 30 10 12 16 0.480.69 (p)
0.4330.453
0.4500.515
0.4250.605
0.6330.690
0.7750.805
0.8450.890
1.2001.300
1.5501.750
2.0802.350
SOPT-5 2.520 5 8 23 0.480.71 (p)
0.50.59
0.610.68
0.780.89
1.581.75
IRS-1A, IRS-1B 36; 73 4 8 22 0.450.52
0.520.68
0.620.68
0.770.86
IRS-1C 5.8 (p); 23.5; 70.5 (B5) 5 8 24 0.500.75 (p)
0.520.59
0.620.68
0.770.86
1.551.70
IRS-P6 AWiFS 56 4 10 16 0.520.59
0.620.68
0.770.86
1.551.70
JERS-1 18 8 8 44 0.520.60
0.630.69
0.760.86 (2 bands)
1.601.71
2.012.12
2.132.25
2.272.40
CBERS-2 20 (p) 5 8 26 0.510.73 (p)
CBERS-3B 20 (MS) 0.450.52
CBERS-3 5 (p) 0.520.59
CBERS-4 20 (MS) 0.630.69
0.770.89
(Continued )
2016 by Taylor & Francis Group, LLC
18 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 .2 (c ontinued ) Data Characteristics of Selective Satellite Sensors
Image Characteristics
Spectral Radiometric Temporal
Sensor Spatial (m) (#) (bit) (Days) Band Range (m)
Hyperspectral sensor
Hyperion 30 196 16 16 196 effective calibrated bands
VNIR (bands 857) 0.4270.925
SWIR (band 79224) 0.932.4
Hyperspatial sensor
IKONOS 14 4 11 5 0.4450.516
0.5060.595
0.6320.698
0.7570.853
QuickBird 0.612.44 4 11 5 0.450.52
0.520.60
0.630.69
0.760.89
RESOURCESAT 5.8 3 10 24 0.520.59
0.620.68
0.770.86
RapidEye AE 6.5 5 12 12 0.440.51
0.520.59
0.630.68
0.690.73
0.770.89
WorldView 0.55 1 11 1.75.9 0.450.51
FORMOSAT-2 28 5 11 Daily 0.450.52
0.520.60
0.630.69
0.760.9
0.450.9 (p)
KOMPSAT-2 14 5 10 328 0.450.52
0.520.60
0.630.69
0.760.9
0.50.9 (p)
Source: Adopted from Melesse et al. (2007).
Note: (p) is the panchromatic band.
(MS) is multispectral bands
Administration [NOAA] AVHRR) are low-resolution data when at two or more spatial resolutions with panchromatic spatial
compared with a 30 m Landsat data product but become accept- resolution being higher than the multispectral scanner (MSS)
able when land cover information is gathered for a global assess- sensors. MODIS, Earth Observing System (EOS) Terra, Landsat,
ment perspective or when it is used in a climate model running and Systeme Probatoire DObservation De La Terre (SPOT) are
with a 100km cell size. In addition to sensor optical character- the examples of dual spatial resolutionbased image acquisi-
istics, the flight height of the satellite or the airplane determines tion. Some SAR missions, such as Canadas RADARSAT, also
the spatial resolution. The higher the location of the satellite in provide multiple-resolution images at different band frequencies
space, the lower will be the image resolution. (Belward and Skoien, 2014). In general, spatial resolution can be
Sensor specifications are usually provided by satellite opera- categorized into five broad classes: 0.54.9 m (very high resolu-
tor and/or sponsoring space agency and/or manufacturer, repro- tion), 5.09.9 m (high resolution), 10.039.9 m (medium reso-
duced by three agencies (National Space Science Data Center lution), 40.0249.9 m (moderate resolution), and 250 m1.5km
[NSSDC], Committee on Earth Observation Satellites [CEOS], (low resolution). Any such grouping is somewhat arbitrary;
and Ocean Surface Current Analysis Real Time [OSCAR]) the low-resolution class acknowledges the threshold of 250 m
(Belward and Skoien, 2014). Some remote sensors provide data established for global monitoring of land transformations
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 19
(Townshend and Justice, 1988), and the moderate-/medium- 1.2.1.2 Spectral Resolution
resolution classes include imagery available through free-and- Satellite sensors measure EMR in different portions of the
open data policies. The high- and very-high-resolution classes EMS. Spectral resolution is the ability of the sensor to resolve
reference commercial distinctions. The upper limit of 50cm is spectral features and bands into their separate components
set because the U.S. Government licensing limits unrestricted in the EMS. It also describes the ability of a sensor to distin-
distribution of spatial data at this resolution. Table 1.2 provides guish between wavelength intervals in EMS. Thus, in general,
the spatial resolution information of a selected list of remote spectral resolution determines the number of bands a satellite
sensing systems (sensors). sensor can sense. And so, higher-spectral-resolution imagery
Generally, panchromatic bands of satellites are of very high will provide more spectral information when compared to
spatial resolution. For example, SPOT 1 acquired the first lower-spectral-resolution imagery. The spectral information is
10m resolution images in 1986; then in 1995, Indian Remote particularly useful in applications dealing with mapping and
Sensing Satellite (IRS-1C) image exceeded the 6 m spatial modeling biophysical properties of objects such as water qual-
resolution. In 1999, IKONOS satellite provided imagery with ity, plant vigor, and soil nutrients. As shown in Figure 1.10,
1 m spatial resolution, the 0.5 m spatial resolution imagery U.S. Landsat has higher spectral resolution than the French
was obtained by WorldView-1 after that, and finally, the high- SPOT sensor.
est resolution imagery of 41 cm was obtained with GeoEye MODIS sensor has a high spectral resolution because it senses
in 2007. All these high-spatial-resolution satellite images are in 36 wavelength regions of the EMS, in comparison to Landsat
acquired with panchromatic bands. TM that senses in only 7 EMR regions.
Visible RGB Near-lR Shortwave infraRed
1
2
3
4
5
6
7 (10.412.5)
1
2
3
4
Pan
0.5 0.7 0.9 1.1 1.3 1.5 1.7 1.9 2.1 2.3 2.5
Landsat Wavelength () SPOT
Landsat 7 ETM+ Applications SPOT 4
Band (m) Band (m)
Band 1Blue IR Coastal water mapping, soil/vegetation differentiation,
(0.450.52) deciduous/coniferous differentiation
Band 2Green Green reflectance from healthy vegetation, iron Band 1(0.500.59)
(0.520.60) content in rocks and soil (Green)
Band 3Red Chlorophyll absorption for plant differentiation Band 2 (0.610.68)
(0.63 0.69) (Red)
Band 4Near IR Biomass survey, water body delineation Band 3 (0.790.89)
(0.76 0.90) (Near-infrared)
Band 5Mid IR Plant moisture content, cloud/snow differentiation Band 4 (1.581.75)
(1.551.75) (Mid-infrared)
Band 6Thermal IR Thermal mapping, soil moisture
(10.412.5)
Band 7Mid IR Soil analysis
(2.082.35)
FIg u r e 1.10 General comparison of Landsat and Systeme Probatoire DObservation De La Terre (SPOT) spectral bands for Earth observation
application.
2016 by Taylor & Francis Group, LLC
20 Remotely Sensed Data Characterization, Classification, and Accuracies
Similarly, Hyperion sensor provides hyperspectral capabili- contrast to the display but also facilitate the numerical analy-
ties of resolving 220 spectral bands in the 0.42.6m range. The sis of the subtle differences in the radiometry. The downside
position, number, and width of spectral bands in an image deter- to this is the high data volumes and storage issues associated
mine the degree to which individual targets can be discrimi- with high-radiometric-resolution imagery. Table 1.2 provides
nated. Multispectral imagery has a higher degree of individual the radiometric resolution of many selective remote sensing
target discrimination power than a panchromatic single-band sensors.
image. Table 1.2 provides a list of spectral bands available for
selective remote sensing sensors. 1.2.1.4 temporal Resolution
Spectral reflectance curves (Figure 1.5) and discussion about Satellite revisit period refers to the length of time it takes
EMS in Section 1.2 explain the advantages of discerning Earth for the satellite to complete one entire orbit cycle around
objects or phenomena from the images using the spectral band the Earth, for example, Landsat satellite takes 16days. The
information. One of the important advantages of separately revisit period of a satellite sensor is usually several days
sensing different spectral bands relates to the ability of combin-
ing the bands in various ways to enhance the visual informa-
tion from the image. These combinations are easily implemented
using software routines that stack three image bands and assign
the reflected data values to three fundamental colors (red, green,
and blue) to create image composites. Thus, when we have
higher spectral resolution in our satellite data, more Earths
surface features can be discerned using different band combi-
nations resulting in different color composites. In addition to
visual enhancements, higher spectral resolution provides higher
capabilities to combine band information using band algebra
to extract additional biophysical information about the Earths 8-bit quantization (256 levels) 6-bit quantization (64 levels)
surface such as plant vigor and soil moisture status. These are
implemented as spectral indices that mathematically combine
band data into derivate, for example, normalized difference veg-
etation index (NDVI). Figure 1.10 shows some of the important
applications of individual bands in satellite images, for example,
Landsat 7 ETM+ and SPOT 4. Thus, higher-spectral-resolution
images are better for feature extraction and decision support.
1.2.1.3 Radiometric Resolution
As satellite sensors capture reflected energy (range of spectral
region) from the Earths surface (ground area unit) and the
energy is quantified into a DN, where the value depends 4-bit quantization (16 levels) 3-bit quantization (8 levels)
on the intensity of the energy sensed. ESRI GIS Dictionary
(2014) defines radiometric resolution as the sensitivity of a
sensor to incoming reflectance. Radiometric resolution refers
to the number of divisions of bit depth (for example, 255 for
8-bit, 65,536 for 16-bit, and so on) in data collected by a sen-
sor. While the pixel arrangement describes the spatial struc-
ture of an image, the radiometric characteristics describe the
actual information content in a remotely sensed image. When
an image is acquired on film or by a sensor, its sensitivity to
the magnitude of the EMR determines the radiometric resolu-
tion (CCRS, 2014). The radiometric resolution of an imaging
system describes its ability to differentiate slight differences 2-bit quantization (4 levels) 1-bit quantization (2 levels)
in energy. The finer the radiometric resolution of a sensor, the
more sensitive it is in detecting small differences in reflected
FIg u r e 1.11 Image captured at different levels of energy quantiza-
or emitted energy (CCRS, 2014). As can be seen in Figure 1.11, tion to illustrate concept of radiometric resolution. Center for Remote
a sensor with low radiometric resolution (1 bit) can resolve the Imaging, Sensing, and Processing (CRISP) (2014). (From http://www.
incoming reflectance into two levels of contrast, while sen- crisp.nus.edu.sg/~research/tutorial/image.htm.) The more the quan-
sors with higher radiometric resolution provide higher levels tization, the greater is the detail of the information captured, thus
of energy quantization. These results not only provide better enabling greater separability between features.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 21
643791.12 656333.62
1989
2018129.75
2016607 .25
2008777.25
Legend
Shaded forest (medium density)
Fairly dense mixed forest
Fairly open mixed forest
2002324.75
Open scrub
Agricultural land
Hilltop bare land
0 2 4 6 km
634003.62 658436.12
(a)
643791.12 656333.62
1996
2018129.75
2016607.25
Legend
2008777.25
Shaded forest (medium density)
Fairly dense mixed forest
Fairly open mixed forest
2002324.75
Open scrub
Agricultural land
Hilltop bare land
0 2 4 6 km
634003.62 658436.12
(b)
FIg u r e 1.12 Land use change analysis within a watershed in Orissa (India) using Indian Remote Sensing Satellite 1 Linear Imaging Self-
Scanning Sensor A satellite imagery: (a) 1989 classified land use map with six classes (From IRS-1A-LISS2 satellite image); (b) 1996 classified land
use map, also with 6 classes (From IRS-1B-LISS2 satellite image).
2016 by Taylor & Francis Group, LLC
22 Remotely Sensed Data Characterization, Classification, and Accuracies
643791.12 656333.62
2018129.75
2016607.25
Legend
2008777.25
11 24 41 54
12 25 42 55
13 26 43 56
14 31 44 61
2002324.75
15 32 45 62
16 33 46 63
21 34 51 64
22 35 52 65
0 2.5 5.0 7.5 km 23 36 53 66
634003.62 658436.12
(c)
FIg u r e 1.12 Land use change analysis within a watershed in Orissa (India) using Indian Remote Sensing Satellite 1 Linear Imaging Self-Scanning
Sensor A satellite imagery: (c) land use change matrix analysis showing 36 classes (6 classes of 1989 6 classes of 1996) in the mapClass 11 means
Class 1 (medium density shaded forest) in year 1989 remains same Class 1 in the year 1996 and so on (From GIS analysis of two date satellite images).
and also known as the temporal resolution of the sensor. and implementing with earnest) through the state depart-
Table 1.2 provides the temporal resolution of many selective ment of forestry and village committees of the watershed
remote sensing sensors. With ever-changing global envi- (Panda etal., 2004).
ronment due to events such as urban sprawl, deforestation,
and natural disasters, including drought, f lood, landslides,
1.2.2 categorization of Satellites
wildfire, avalanches, and earthquake, it is extremely essen-
bycountry of ownership
tial for end users to conduct temporal analysis. There is an
increase in studies pertaining to land use forecasting more It is a matter of national pride to have an Earth-observing pro-
so now than earlier due to increased awareness and implica- gram, and the acquired satellite data are very important for
tions of rapid land coverland use changes occurring in the applications pertaining to resource inventory and manage-
landscape. Temporal image analysis is highly essential for ment. Based on the study of Baker et al. (2001), the civilian
these studies. Better temporal resolution of images such as remote sensing systems are government owned, the commer-
RapidEye, WorldView, MODIS, and even SPOT 5 can pro- cial systems are government licensed but privately owned, and
vide rapid land cover change information in scenarios such the military/intelligence gathering platforms are government
as tracking the spread of wildfire or damages caused by any owned but with highly restricted access to the systems and
natural disaster. Aerial imaging can be conducted daily over products. According to Martin (2012), South Africas scientific
a study area and even more than once in a day. Therefore, and engineering capacities were enhanced by the experience of
aerial imaging does not have a specific temporal resolution. designing, building, and launching SumbandilaSat; at the same
The advantages of temporal image resolution are shown in time, the country obtained a great deal of social benefit due
Figure 1.12a and b, in which IRS-1 LISS-A satellite imager- to the Earth-observing program. Belward (2012) opined that
ies of two different years, 1989 and 1996, were used to clas- national space programs and their success not only enhance
sify the land use of a remote watershed in Orissa (India) and the nations self-esteem but also reinforce national identity and
perform land use change analysis to determine a positive sense of purpose for the country. This is why many countries in
environmental effect (increase in forest cover and increase the world are in the forefront of the Earth observation missions.
in changes from low-density forest to high-density forest Table 1.3 presents a comprehensive list of countries that have
cover) over 7 years using social awareness in forest preser- Earth-observing programs and the associated operational satel-
vation (forming village-level forest preservation committee lite series in the program.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 23
TABLe 1 .3 Selected List of Earth-Observing Programs of Different Countries in the World
Country Agency Name Prominent Satellite Programs
France Centre National dtudes Spatiales (CNES) SPOT
Pliades
TOPEX/Poseidon
United States National Aeronautics and Space Administration (NASA) Thermosphere, Ionosphere Mesosphere, Energetics and Dynamics (TIMED)
TOPEX/Poseidon
Upper Atmosphere Research Satellite
Landsat
NOAAAVHRR
SRTM
Vanguard
Russia Russian Federal Space Agency (RKA; Roscosmos) Elektro-L
Monitor-E
Resurs-DK1
Resurs-P No. 1
Japan Japan Aerospace Exploration Agency (JAXA) MOS-1 (Momo-1)
(former NASDA) MOS-1b (Momo-1b)
JERS-1 (Fuyo-1)
ADEOS (Midori)
ADEOS-II (Midori II)
GOSAT (Ibuki)
ALOS (Daichi) PALSAR, AVNIR-2, and PRISM
India Indian Space Research Organisation (ISRO) Oceansat-2, September 23, 2009
IMS-1, April 28, 2008
Cartosat-2A, April 28, 2008
Cartosat-2, January 10, 2007
IRS-P5 (Cartosat-1), May 5, 2005
IRS-P6 (RESOURCESAT-1), October 17, 2003
IRS-P4, May 27, 1999
IRS-P3, March 21, 1996
IRS-P2, October 15, 1994
IRS-1D, September 29, 1997
IRS-1C, December 28, 1995
IRS-1B, August 29, 1991
IRS-1A, March 17, 1988
Canada Canadian Space Agency RADARSAT
China Chinese Space Agency Yogan-21
Europe European Space Agency (ESA) ENVISAT
ERS (1 and 2)
CryoSat-2
Brazil Brazilian Space Agency (AEB) CBERS-1
CBERS-2
CBERS-2B
Argentina Argentina Space Agency (CONAE) SAC-A
SAC-B
SAC-C
SAC-D
SAOCOM
Indonesia Lembaga Penerbangan dan Antariksa Nasional (LAPAN, Lapan (TUBsat)
Indonesia)
Belarus National Academy of Sciences of the Republic of Belarus BelKA
Pakistan SUPARCO (Space and Upper Atmosphere Research Badr-1
Commission) Badr-B
Paksat-1R
Pakistan Remote Sensing Satellite
Sweden Swedish National Space Board Munin
Bolivia Bolivarian Agency for Space Activities VRSS-1
South Korea KOMPSAT-2
Thailand Thaichote
Turkey Gktrk-2 (2012) (IMINT), Turkish Armed Forces, Intelligence [10]
RASAT (2011), mapping
BILSAT-1 (20032006), part of the Disaster Monitoring Constellation Project
Gktrk-1 (2013) (IMINT), Turkish Ministry of National Defense, Intelligence
2016 by Taylor & Francis Group, LLC
24 Remotely Sensed Data Characterization, Classification, and Accuracies
1.2.3 types of Sensing technologies the satellite can virtually see every part of the Earth as the Earth
Based on Applications rotates beneath the satellite. It takes approximately 90min for the
near-polar-orbiting satellites to complete one orbit and is very
The NASA NSSDCs master catalog (http://nssdc.gsfc.nasa.gov/ useful for atmospheric (stratosphere) measurements like green-
nmc/spacecraftSearch.do) lists 7262 spacecraft that have been house gas concentration, ozone concentration, temperature, and
launched between October 4, 1957, and March 31, 2015 (NSSDC, water vapor (Montenbruck and Gill, 2000). Sun synchronous
2014). They were listed in the following categories based on their satellites like Landsat can pass over a section of the Earth at the
application: astronomy (316), earth science (904), planetary same time of a day due to these orbits (Montenbruck and Gill,
sciences (313), solar physics (202), space physics (650), human 2000). These satellites, in general, orbit the Earth at an altitude
crew (324), life science (97), microgravity (71), communication of approximately 700800 km. As some of these satellites take
(2054), engineering (419), navigation and global positioning pictures of the Earth, they work best with bright sunlight, that is,
(448), resupply/refurbishment/repair (204), and surveillance pass over a section of Earth during late morning to until 2-3PM
and other military (2280), and technology applications (201). or early afternoon. When sun synchronous satellites measure
The number in parentheses depicts the number of spacecraft longwave radiation, they would work best in complete darkness
launched with application mission objective. These numbers are and hence pass over the Earth section at night. In geosynchro-
getting updated on the site frequently. Among these, 919 space- nous orbits, also known as geostationary orbits, satellites circle
craft were launched for earth science application. As discussed the Earth at the same rate as the Earth spins (Montenbruck and
earlier, earth science as a broad discipline encompasses land Gill, 2000). These orbits make the satellites stay over a location
useland cover issues including surface (biophysical or chemi- of the Earth constantly throughout and observe almost the full
cal features) or underground (minerals) feature identification, hemisphere of the Earth.
but clearly, not all 890 missions can be used for global land cover According to Jensen (2009), remote sensing systems can
mapping. All of these near-polar-orbiting Earth science appli- directly measure the fundamental biological and/or physical
cationoriented satellites are searchable in the NSSDC website. characteristics of Earth features without using other ancillary
For example, upon clicking on the satellite Aryabhata in the data. The following are a few example of how remotely sensed
search result, the following description shows: data (MSS, hyperspectral, LiDAR, etc.) can help find water and
nutrient stress in agricultural crop fields and support farmers
This spacecraft, named after the famous Indian astronomer, to schedule irrigation and fertilizer applications for increased
was Indias first satellite and was completely designed and crop yields (Panda et al., 2011a,b): finding the growth stages
fabricated in India. It was launched by a Soviet rocket from of blueberry crops and conducting site-specific crop manage-
a Soviet cosmodrome. The spacecraft was quasispherical in ment to enhance the crop yield (Panda et al., 2009, 2010a,b,
shape containing 26 sides and contained three experiments 2011a,b; Panda and Hoogenboom, 2013), finding and assess-
for the measurement of cosmic X-rays, solar neutrons, and ing the drainage characteristics of low-gradient coastal water-
Gamma rays, and an ionospheric electron trap along with a shed for watershed management decision support (Amatya
UV sensor. The spacecraft weighed 360kg, used solar pan- etal., 2013), and estimating the deciduous forest structure for
els on 24 sides to provide 46 watts of power, used a passive forest management decision support (Defibaugh et al., 2013).
thermal control system, contained batteries, and a spin- Engman (1995) shows the advantage of microwave remote sens-
up gas jet system to provide a spin rate of not more than ing to accurately measure soil moisture amounts. Panda etal.
90 rpm. There was a set of altitude sensors comprised of a (2012a,b) and Amatya etal. (2011) used Landsat data for esti-
triaxial magnetometer, a digital elevation solar sensor, and mating soil moisture, stomatal conductance, leaf area index
four azimuth solar sensors. The data system included a tape (LAI), and canopy temperature of pine forest. Hale etal. (2012),
recorder at 256 b/s with playback at 10 times that rate. The Phillips et al. (2012), Rylee et al. (2012), and Ertberger et al.
PCM-FM-PM telemetry system operated at 137.44MHz. The (2012) used remote sensing and other watershed geospatial data
necessary ground telemetry and telecommand stations were for urban and rural watershed development decision support,
established at Shar Centre in Sriharikota, Andhra Pradesh. while Cash and Panda (2012) used high-resolution National
Agriculture Imagery Program (NAIP) imagery, SSURGO soil
The link page to the satellite also includes an alternate name of data, and LiDAR data to develop a hydrologic model using
the spacecraft, other brief facts, funding agencies names, and dis- Soil and Water Assessment Tool (SWAT). Numerous studies
cipline names for which the spacecraft was launched. The reader have already been conducted with the use of remote sensing
is directed to the NASA website (http://nssdc.gsfc.nasa.gov/nmc) to detect, interpret, and analyze the Earths biophysical char-
for additional metadata information about the Earth-observing acteristics for management decision support. More studies for
satellites. environmental and Earth resource management using remote
On another note, among several types, the three most impor- sensing are provided later in individual sections. Jensen (2009)
tant types of satellite orbits are (1) near-polar orbits, (2) sun syn- has developed a database showing the potential remote sensing
chronous orbits, and (3) geosynchronous orbits. In near-polar systems for detection, interpretation, and analysis of various
orbits, the satellites inclination to the Earth is nearly 90. Thus, biophysical variables (Table 1.4).
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 25
TABLe 1 .4 Potential Remote Sensing Systems for Detection, Interpretation, and Analysis of Biophysical Variables
Biophysical Variables Potential Remote Sensing Systems
Geodetic control GPS
Location from orthocorrected imagery Analog and digital stereoscopic aerial photography, space imaging
IKONOS, DigitalGlobe, QuickBird, OrbView-3, SPOT, Landsat (Thematic Mapper, Enhanced TM+),
Indian IRS-ICD, European ERS-1 and ERS-2 microwave and ENVISAT MERIS, MODIS, LiDAR,
Canadian RADARSAT-1 and -2, etc.
Topography/bathymetry
DEM GPS, stereoscopic aerial photography, and other location identification satellite systems
Digital Bathymetric Model SONAR, bathymetric LiDAR, stereoscopic aerial photography
Vegetation
Chlorophyll a and b Color aerial photography, Landsat ETM+, IKONOS, QuickBird, and most other MSS sensorbased
Canopy structure satellites
Biomass Stereoscopic aerial photography, LiDAR, RADARSAT, IFSAR
LAI CIR aerial photography, most other MSS (Landsat, QuickBird, etc.) and hyperspectral systems
Absorbed photosynthetically active radiation (AVIRIS, HyMap, CASI), AVHRR, multiple imaging spectrometer (MISR)
Evapotranspiration
Surface temperature
Land, water, and atmosphere ASTER, AVHRR, GOES, Hyperian, MISR, MODIS, SeaWiFS, Airborne TIR
Soil and rocks
Moisture ASTER, passive microwave (SSM/1), RADARSAT, MISR, ALMAZ, Landsat, ERS-1 and ERS-2,
Mineral composition Intermap STAR-3i
Taxonomy ASTER, MODIS, hyperspectral systems (AVIRIS, HyMap)
Hydrothermal alteration High-resolution color and CIR aerial photography, airborne hyperspectral systems
Landsat, ASTER, MODIS, airborne hyperspectral systems
Surface roughness Aerial photography, ALMAZ, ERS-1 and ERS-2, RADARSAT, Intermap STAR-3i, IKONOS,
QuickBird, ASTER, etc.
Atmosphere
Aerosols (optical depth) MISR, GOES, AVHRR, MODIS, CERES, MOPITT, MERIS
Clouds GOES, AVHRR, MODIS, CERES, MOPITT, MERIS, UARS
Precipitation TRMM, GOES, AVHRR, SSM/1, MERIS
Water vapor (precipitable) GOES, MODIS, MERIS
Ozone MODIS
Water
Color CIR aerial photography, most other MSS (Landsat, QuickBird, etc.) and hyperspectral systems
Surface hydrology, suspended minerals (AVIRIS, HyMap, CASI), bathymetric LiDAR, MISR, Hyperion, TOPEX/Poseidon, MERIS, AVHRR,
Chlorophyll CERES, etc.
Dissolved organic matter
Snow and sea ice
Extent and characteristics CIR aerial photography, most other MSS (Landsat, QuickBird, etc.) and hyperspectral systems
(AVIRIS, HyMap, CASI)
Bidirectional reflectance distribution function MISR, MODIS, CERES
Source: Adapted and modified from Jensen, J.R., Remote Sensing of the Environment: An Earth Resource Perspective 2/e, Pearson Education India, Delhi,
India, 2009, pp. 1112.
1.3 Remote Sensing Platforms and temperature; (14) proximity and presence; (15) sensor tech-
nology; and (16) other sensors. However, this chapter is focused
Depending on the airplane or the Earth-observing satellite on Earth observation remote sensing, covering both active and
platform, there are hundreds of different sensors available for passive remote sensing technologies as discussed in Section
specific usage. These sensors can be sorted by the sensor type 1.1.2. These Earth-observing satellites and aerial imaging sys-
and are categorized into the following: (1) acoustic, sound, and tems are grouped into several types based on their operational
vibration; (2) automotive and transportation; (3) chemical; principles within the EMS and means of obtaining images. They
(4) electric current, electric potential, magnetic, and radio; are optical, radar, microwave, hyperspectral, sonar, etc.
(5) environment, weather, moisture, and humidity; (6) fluid Typical remote sensing systems operate in the visible
flow; (7) ionizing radiation and subatomic particles; (8) naviga- (0.40.7 m) and IR (0.71000 m) portion of the EMS, known
tion instruments; (9) position, angle, displacement, distance, as optical remote sensing (Aronoff, 2004). Other sensors such
speed, and acceleration; (10) optical, light, imaging, and photon; as microwave sensors operate in the microwave region of the
(11) pressure; (12) force, density, and level; (13) thermal, heat, EMS (0.3mm1 m). RADAR uses radio waves (1041011 m) to
2016 by Taylor & Francis Group, LLC
26 Remotely Sensed Data Characterization, Classification, and Accuracies
determine the range, altitude, direction, and speed of an object and many other natural resource management is abundant. As
or phenomena on Earth. On the other hand, hyperspectral discussed previously, photogrammetry or aerial stereo photogra-
imaging sensors, although operate in the visible/NIR ranges of phy was the earliest form of remote sensing. Even today, image
the EMS, are marked by very high spectral resolutions resulting and terrain analysts use stereographic pairs of aerial photographs
in many numbers of spectral bands (refer to Figure 1.6). to make topographic maps by image and terrain analysts for digi-
Another active sensor LiDAR is a remote sensing technology in tal elevationbased land management decision support such as
which the signal (return) distances are measured based on the lag road construction, watershed delineation, stream flow direction
time of the pulsed signal. Based on the different returns recorded, and flow accumulation mapping, and traffic ability applications.
it is possible to accurately measure Earth objects such as canopy According to Dornaika and Hammoudi (2009) and Bulstrode
height and ground surface elevation. SONAR is another active etal. (1986), stereoscopy is a photogrammetric technique for cre-
remote sensing technology platform. It uses sound propagation ating or enhancing the illusion of depth in an image by means of
usually in water for depth mapping, object detection, and naviga- stereopsis for binocularl vision called Stereogram. Adjacent but
tion underwater. These different platform-based remote sensing overlapping aerial photos are called stereopairs and are needed to
technologies have their individual applications and advantages in determine parallax and stereo/3D viewing. Advances in the hard-
real-world problem solving and decision support. ware and software sector simulate such procedures in the digitial
and visualization domains, called soft-copy photogrammetry.
1.3.1 Aerial imaging
1.3.2 optical Remote Sensing
Aerial photography is a remote sensing system in which photo-
graphs of the Earths surface are taken from an elevated posi- As discussed earlier, optical remote sensing uses visible, near infra-
tion, mostly by airplanes flying within a 10 mile height from the red and short-wave infrared sensors to acquire images of the earths
ground. Platforms for aerial photography include fixed-wing surface (Figure 1.13). Optical remote sensing systems are classified
aircraft, helicopters, multirotor unmanned aircraft systems into several types, depending on the number of spectral bands used
(UASs), balloons, blimps and dirigibles, rockets, kites, parachutes, in the imaging process, such as (1) panchromatic (PAN) imaging
stand-alone telescoping, and vehicle-mounted poles and in system, (2) multispectral (MSS) imaging system, (3) superspectral
recent times military operated drones. Recently, radio-controlled imaging system, and (4) hyperspectral imaging system. In pan-
model aircraftsdrones, for example, are very popular in taking chromatic imaging system, the sensor is a single-channel detec-
images of the Earths surface from a much lower height. These tor sensitive to radiation within a broad wavelength. For example,
aerial images are of high resolution and have a wide application IKONOS PAN image covering a broad bandwidth of 0.450.9m,
in environmental and natural resource management fields. which is the range of multispectral B, G, R, and IR bands. Similarly,
The U.S. Geological Survey (USGS) initiated novel programs the SPOT PAN image has a band range of 0.510.73m, which cov-
called the National High Altitude Program (NHAP) and the ers most of the multispectral region of B and G bands. PAN image
National Aerial Photography Program (NAPP) in the 1980s spatial resolution is mostly better than the multispectral images
to acquire aerial photograph for the conterminous 48 states to of the same sensor, for example, IKONOS PAN and SPOT PAN
support operational and mapping needs of the local, state, and spatial resolutions are 1 and 10 m, respectively, whereas the spatial
federal agencies (USGS, 1992a,b). The black-and-white and resolutions of their MSS bands are 4 and 20 m. In multispectral
color-infrared (CIR) products from these programs provide an imaging system, the sensor uses a multichannel detector with a few
invaluable resource for historical assessment of land cover/use at spectral bands like B, G, R, and IR within a narrow wavelength
map scales ranging from 1:58,000 to 1:40,000. band (refer to Table 1.2). Table 1.2 provides a detailed list of MSS
Building on the efforts of the USGS, the USDA Farm Service and PAN bands of some select optical sensors. In contrast to MSS
Agency (FSA) launched and administered the NAIP aimed at imaging system, in a superspectral imaging system, there are many
acquiring high-resolution color and near-infrared (NIR) imag- more spectral channels (typically >10). The bands have narrower
ery during the agricultural growing seasons in the conterminous bandwidths than MSS imaging system, enabling the finer spectral
United States. These annual image data products are orthorecti- characteristics of the targets to be captured by the sensor. MODIS
fied and GIS-ready, and are primarily used by USDA to maintain and medium-resolution imaging spectrometer (MERIS) are a
the Common Land Unit database. NAIP imagery is one of the few examples of the superspectral systems. MODIS superspectral
major resources for the geospatial community working on veg- imaging system has 36bands ranging from 0.405 to 14.385m with
etation mapping and precision agriculture due to its high reso- each band having narrower bandwidths; MODIS band 11 and band
lution (1 and 2 m). More information on the NAIP imagery can 12 have bandwidths of 0.5260.536 and 0.5460.556m, respec-
be obtained from http://www.fsa.usda.gov/FSA website. Use of tively, whereas the bandwidth of Landsat MSS imaging system has
aerial images in urban management (Hodgson etal., 2003; Cleve little wider bandwidths. For instance, bands 4 and 5 of Landsat
et al., 2008), site-specific crop management (Jhang et al., 2002; MSS have bandwidths of 0.760.90 and 1.551.75m, respectively.
Panda etal., 2009, 2010a,b, 2011a,b), forest and ecological resource However, hyperspectral imaging systems acquire images in about
management (Suarez etal., 2005; Morgan etal., 2010; Rao, 2013), 100 or more contiguous spectral bands. Hyperspectral imaging
water resource management (Ritchie etal., 2003; Jha etal., 2007), system is discussed in a later section.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 27
Airplane
Sun
Satellite
Incident
solar radiation
Reflected
solar radiation
Atmosphere
Forest Grass Bare soil Paved Agriculture Urban area
Water road
FIg u r e 1 .13 Basic image acquisition process of an optical imaging system.
According to Avery and Berlin (1992), unlike photographic 1.3.3 Hyperspectral Remote Sensing
cameras that record radiation reflected from ground scene directly
onto films, electro-optical sensors use nonfilm detectors, which Hyperspectral imaging system is an optical imaging system but is
convert the reflected and/or emitted radiation from the object or known as an imaging spectrometer as the system acquires an image
phenomena on ground/earth to proportional electrical signals that from a satellite or airplane in about 100 or more contiguous spec-
are used to construct 2D images for conventional viewing. Avery tral bands. Imaging spectrometry is discussed in a later section.
and Berlin (1992) categorized electro-optical sensors into video The targets on Earth are better identified and characterized by a
camera, vidicon camera, across-track scanner, and along-track hyperspectral imaging system than other optical imaging systems
scanner. According to Murai (1993), optical sensors are character- because of the systems precise spectral information. Hyperspectral
ized by spectral (spectral band, bandwidth, central wavelength, imaging or field spectrometry has potential applications in preci-
response sensitivity at the edge of band, spectral sensitivity at outer sion agriculture or site-specific cropmanagement (e.g., monitoring
wavelength, and sensitivity of polarization), radiometric (radiom- the health and growth of crops and measuring water stress, nutri-
etry of the sensor, sensitivity in noise equivalent power, dynamic ent stress, or stress from crop diseases in crop fields), intense forest
range, S/N, and other quantification noise), and geometric (FOV, management (e.g., forest species differentiation, timber readiness
IFOV, band-to-band registration, MTF, geometric distortions, and for harvesting determination and analysis, wildfire susceptibility
alignment of optical elements) performance. Table 1.5 lists opti- determination and assessment), and coastal management (e.g.,
cal sensorbased remote sensing satellites with their spectral band monitoring of water quality parameters such as phytoplankton,
information. The table also contains satellites based on other dif- organic matter, sediment pollution, and bathymetry changes).
ferent sensing platforms as discussed in this section. The descrip- Hyperspectral imaging produces an image where each
tion column of the table explains about the sensor platform of the pixel has full spectral information with imaging narrow
satellite. In addition to the mechanics of image acquisition, Figures spectral bands over a contiguous spectral range of the EMS.
1.14a and b illustrate a comprehensive working principle of optical Hyperspectral imaging is a very-fast-growing area in remote
remote sensing, which includes across-track/whiskbroom, push- sensing that expands and improves the efficiency/capability of
broom, and multispectral scanning processes. As the names sug- multispectral satellite imaging (Lees and Ritman, 1991; Goetz,
gest, in a wishbroom system, the sensor detectors scan and record 2009). Hyperspectral imagers are used in various applications
the reflected energy using oscillation mirrors that move across the since the late 1980s including mineralogy, biology (food safety),
satellite track, while the pushbroom system employ an array of defense, and environmental measurements (Clark etal., 1990;
detectors that are perpendicular to the satellite track. Landsat sen- Kim etal., 2001; Chang, 2003; Lu, 2003; Van Wagtendonk etal.,
sor systems until Landsat 8 OLI use the wishbroom system. Sensors 2004; Asner etal., 2007; Gowen etal., 2007). Hyperion, AVIRIS,
that use the push broom design include SPOT, IRS, QuickBird, HyMap, and compact airborne spectrographic imager (CASI)
OrbView, and IKONOS. are examples of a high-spatial-resolution hyperspectral system.
2016 by Taylor & Francis Group, LLC
28 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 .5 List of Selected Optical Remote Sensing Satellites along with Their Spectral Band Information
Name of Sensor Description Mission No. of Bands
ACe -FTS Atmospheric Chemistry Experiment Fourier SCISAT
Transform Spectrometer
ACS Atmospheric Correction Sensor RESOURCESAT-3
ADCS/SAr SAT Advanced Data Collection System/Search and JPSS-1 (NPOESS) 4
Rescue Satellite Aided Tracking
Ae ISS Advanced Earth Imaging Sensor System KOMPSAT-3 (Arirang 3) 9
ALI Advanced Land Imager EO-1 3
AlSat Standard DMC sensor AlSat-1
APS Aerosol Polarimetry Sensor GLORY
Arg OS Metop-A, Metop-B, SARAL/AltiKa 4
AWFI Wide Field Imaging Camera CBERS-3, CBERS-4 4
AWiFS Advanced Wide Field Sensor RESOURCESAT-1, RESOURCESAT-2 3
Be IJINg -1-MS Standard DMC sensor BEIJING-1 4
BILSAT-MS BILSAT 1
BILSAT-PAN BILSAT
Camera Multispectral camera with black-and-white and Svea 4
color imaging
CCD Multispectral camera CBERS, CBERS-2, CBERS-3, CBERS-4
CCD/TDI Selectable-multispectral possibility EROS C 1
CCD/TDI Selectable EROS B
CCSP Cloud Camera Sensor Package GLORY 3
Cere S Clouds and Earths Radiant Energy System Aqua, JPSS-1 (NPOESS), Suomi NPP (SNPP), Terra, 1
TRMM
Chinese Mapping Telescope BEIJING-1 1
CIr C Compact infrared camera ALOS-2 9
COBAN BILSAT 6
CZCS Coastal Zone Color Scanner Nimbus-7 3
DMC2 Standard DMC sensor UK-DMC 2 2
DOr IS Doppler Orbitography and Radiopositioning CryoSat-2, ENVISAT, Jason-1, Jason-2 (OSTM), SPOT 2, 1
Integrated by Satellite SPOT 3, SPOT 4, SPOT 5, TOPEX/Poseidon
e OC Electro-optical camera KOMPSAT-1
e -OP1 Panchromatic and 5 band color imager NigeriaSat-2
e -OP2 Wide area coverage imager NigeriaSat-2 13
er B Earth Radiation Budget Nimbus-7 4
er Be Earth Radiation Budget Experiment NOAA-10 1
er OS A EROS A 8
e TM Enhanced Thematic Mapper Landsat 7
Fr S Strip map mode RISAT-1 2
ger B Geostationary Earth Radiation Budget MSG-1, MSG-2, MSG-3 4
g IS-MS Visible and NIR GeoEye-1 (OrbView-5) 1
g IS-PAN Panchromatic GeoEye-1 (OrbView-5) 36
g LI Global Imager ADEOS-II 5
g Oe S Imager Multichannel instrument to sense radiant and GOES-14 6
solar-reflected energy
g OMe Global Ozone Monitoring Experiment ERS-2 3
g OMOS Global Ozone Monitoring by Occultation of Stars ENVISAT
gu VI Global Ultraviolet Imager TIMED 1
HCS High-sensitivity camera SAC-C 7
He PD High-energy particle detector KOMPSAT-1
Hir I High-resolution optical imager Pleiades HR 1A, Pleiades HR 1B
HPO High-performance optical sensor ASNARO-1 1
Hr C High-resolution camera PROBA-1 4
(Continued )
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 29
TABLe 1 .5 (c ontinued ) List of Selected Optical Remote Sensing Satellites along with Their Spectral Band Information
Name of Sensor Description Mission No. of Bands
Hrg High-resolution geometric SPOT 5 1
Hr S High-resolution stereoscopic SPOT 5 1
Hr TC High-resolution technological camera SAC-C 4
Hr V High-resolution visible SPOT 1, SPOT 2, SPOT 3 4
Hr VIr High-resolution visible and infrared SPOT 4 2
HSr S Hot Spot Recognition Sensor System BIRD 64
HySI Hyper Spectral Camera with 64 fixed bands IMS 1
IDee Instrument for Detection of High-Energy Electrons TARANIS 3
IKAr -De LTA PRIRODA-MIR 5
IKAr -N PRIRODA-MIR 2
IKAr -P PRIRODA-MIR
IMA Multispectral Imager operating in 5 bands Argo (RapidEye 6) 5
Imager GOES-10, GOES-11, GOES-13 = GOES-East, 5
GOES-8, GOES-9
Imager-M GOES-12 = GOES-South America 1
IMg Interferometer Monitor for Greenhouse Gases ADEOS
IMM Instrument for magnetic measurements TARANIS 1
IMS Ionosphere Measurement Sensor KOMPSAT-1 4
IPe I Ionosphere Plasma and Electrodynamics FORMOSAT-1 4
Instrument
Ir MSS Medium-resolution scanner CBERS-3, CBERS-4 4
Ir -MSS Infrared multispectral scanner CBERS, CBERS-2
ISuAL Imager of Sprites and Upper Atmospheric FORMOSAT-2 5
Lightning
JAMI Japanese Advanced Meteorological Imager MTSAR-1R 2
KBr K-Band Ranging System GRACE 4
KOMPSAT-MSC B&W panchromatic, MSS and merged 1 m KOMPSAT-2 1
resolution images
KVr -1000 Cosmos 256
LAC LEISA Atmospheric Corrector, corrects high- EO-1
spatial-resolution multispectral data for
atmospheric effects on surface reflectance
Laser reflector CryoSat-2 6
LIMS Limb infrared monitor of the atmosphere Nimbus-7 1
LIS Lightning Imaging Sensor TRMM 4
LISS-1 Linear Imagine Self-Scanning System IRS-1A, IRS-1B 4
LISS-2 Linear Imagine Self-Scanning System IRS-1A, IRS-1B, IRS-P2 4
LISS-3 Linear Imagine Self-Scanning System IRS-1C, IRS-1D 4
LISS-3* Linear Imagine Self-Scanning System RESOURCESAT-1 3
LISS-4 Linear Imagine Self-Scanning System RESOURCESAT-1 4
LISS-3 Linear Imagine Self-Scanning System RESOURCESAT-2 3
LISS-4 Linear Imagine Self-Scanning System RESOURCESAT-2 4
LISS-III-WS Wide-swath sensor RESOURCESAT-3 4
MAC MS RazakSAT (MACSAT), RazakSAT (MACSAT)
MAe STr O Measurements of Aerosol Extinction in the SCISAT
Stratosphere and Troposphere Retrieved by
Occultation
MCP Micro Camera and Photometer TARANIS 15
Mer IS Medium-resolution imaging spectrometer ENVISAT
Me XIC Multiexperimental Interface Controller TARANIS 5
MMr S Multispectral Medium-Resolution Scanner SAC-C
MODI Moderate-Radiation Visible and NIR Imager FengYun-3A 4
MOMS-2P Modular Optoelectronic Multispectral Scanner PRIRODA-MIR 64
(Continued )
2016 by Taylor & Francis Group, LLC
30 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 .5 (c ontinued ) List of Selected Optical Remote Sensing Satellites along with Their Spectral Band Information
Name of Sensor Description Mission No. of Bands
MOPITT Measurement of Pollution in the Troposphere Terra 4
MS camera Derived from SPOT camera, has thermoelastic Deimos 1, Kanopus-Vulkan, THEOS 12
stability
MSC Multispectral VENUS 13
MSI Multispectral instrument EarthCARE, Sentinel-2 4
MSS Multispectral scanner AlSat-2A, Landsat 4, Landsat 5 5
MSS (LS 13) Multispectral scanner Landsat 1, 2, 3 Landsat 1, Landsat 2, Landsat 3 3
MSu -e 1 Resurs-O1 3
MSu -e 2 PRIRODA-MIR 5
MSu -g S Multispectral Scanner Geostationary Electro-L/GOMS 2 5
MSu -SK PRIRODA-MIR 5
MSu -SK (r esurs) Resurs-O1-3 5
MSu -SK1 Resurs-O1 16
MTI Multispectral Thermal Imager MTI 3
Mu X Multispectral CCD Camera CBERS-3, CBERS-4 4
Mx Four band multispectral CCD camera IMS 1 5
NAOMI New AstroSat Optical Modular Instrument SPOT 6, SPOT 7, VNREDSat-1A 3
NigeriaSat-1 Standard DMC sensor NigeriaSat-1 1
NSCAT NASA Scatterometer ADEOS 7
OCI Ocean Color Imager FORMOSAT-1 8
OCM Ocean Color Monitor IRS-P4 (Oceansat), Oceansat-2 12
OCTS Ocean Color and Temperature Scanner ADEOS 3
OIS Optical Imaging System RASAT
OLCI Ocean and Land Color Instrument Sentinel-3 9
OLI Operational Land Imager Landsat 8 (LDCM) 3
OLS Operational Linescan System DMSP-16 3
OMI Ozone Monitoring Instrument AURA
OMPS Ozone Mapping and Profiler Suite JPSS-1 (NPOESS), Suomi NPP (SNPP) 8
OPS Optical Sensor JERS-1 6
Optical Imaging GOKTURK-2 5
(g OKTur K-2)
OrbView-3 OrbView-3 5
OSA Optical Sensor Assembly IKONOS
OSIr IS Optical Spectrograph and Infrared Imaging System ODIN 6
OSMI Ocean Scanning Multispectral Imager KOMPSAT-1 1
OTD Optical Transient Detector OrbView-1 4
OZON-M PRIRODA-MIR 4
PAMe LA Resurs-DK1 1
PAN Cartosat-2B, CBERS-3, CBERS-4, IRS-1C, IRS-1D 1
PAN Camera Panchromatic camera Cartosat-2 (IRS-P7), Kanopus-Vulkan 1
PAN Telescope THEOS 1
PAN-A Panchromatic aft pointing Cartosat-1 1
PANC Panchromatic camera Cartosat-2A 1
PAN-F Panchromatic forward pointing Cartosat-1 9
POAM-II Polar Ozone and Aerosol Measurement SPOT 3 9
POAM-III Polar Ozone and Aerosol Measurement SPOT 4 15
POLDer Polarization and Directionality of the Earths ADEOS, ADEOS-II, Parasol 2
Reflectance
Poseidon 2 altimeter Jason-1
POSe IDON-3 Altimeter Jason-2 (OSTM) 1
Pr ISM Three panchromatic sensors for stereomapping ALOS 4
(Continued )
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 31
TABLe 1 .5 (c ontinued ) List of Selected Optical Remote Sensing Satellites along with Their Spectral Band Information
Name of Sensor Description Mission No. of Bands
QuickBird High-resolution PAN, 61cm (nadir) to 72cm QuickBird 1
(25 off nadir); MS, 2.442.88 m
r -400 PRIRODA-MIR 1
r adar altimeter GFO
r adar sensor SMAP
r adiometer L-band SMAP 4
r BV Return Beam Vidicon Camera Landsat 1 5
re IS Records in 5 spectral bands on VIR and NIR RapidEye 1
r IS Reflector in space ADEOS 5
r OCSAT-2 FORMOSAT-2 4
r SI Remote sensing instrument FORMOSAT-5 1
SAM II Stratospheric Aerosol Measurement Nimbus-7 1
SASS Seasat-A Satellite Scatterometer Seasat 12
SBu V/2 Solar Backscatter Ultraviolet Instrument NOAA-11, NOAA-12, NOAA-14, NOAA-15, 6
NOAA-16, NOAA-17, NOAA-18 (NOAA-N),
NOAA-19 (NOAA-N Prime)
SBu V/TOMS Solar Backscatter Ultraviolet/Total Ozone Nimbus-7
Mappingfailed in 1993
SCAr AB Scanning Radiative Budget Instrument Megha-Tropiques
Scatterometer Oceansat-2 8
SeaWiFS Sea-Viewing Wide Field-of-View Sensor OrbView-2 1
Seawinds ADEOS-II, QuikSCAT
See Solar Extreme Ultraviolet Experiment TIMED
Se M Space Environment Monitor GOES-14 7
Se M To measure solar radiation in the x-ray and GOES-13 = GOES-East 12
extreme ultraviolet (EUV) region
Se VIr I Spinning Enhanced Visible and Infrared Imager MSG-1, MSG-2, MSG-3 35
Sg LI Second-Generation Global Imager GCOM-C1 1
SIM Spectral Irradiance Monitor SORCE 1
SIr -C SRTM 3
SLIM6 Surrey Linear Imager Multispectral 6 channels NigeriaSat-X 1
but 3 spectral bands
SOLSTICe A and B SolarStellar Irradiance Comparison Experiment SORCE 19
Sounder GOES-10, GOES-11, GOES-12 = GOES-South
America, GOES-13 = GOES-East, GOES-8, GOES-9
SSu LI Ultraviolet Limb Imager DMSP-16
SXI Solar X-ray Imager GOES-14
SXI Solar X-r ay Imager To monitor the suns x-rays GOES-13 = GOES-East 4
TANSO-CAI Thermal and NIR Sensor for Carbon Observation, GOSAT (Ibuki) 1
Cloud and Aerosol Imager
TDI Panchromatic AlSat-2A
TIDI TIMED Doppler Interferometer TIMED 1
TIM Total Irradiance Monitor SORCE
TIP Tiny Ionosphere Photometer FORMOSAT-3 (COSMIC) 2
TIr S Thermal Infrared Sensor Landsat 8 (LDCM) 1
TK-350 Cosmos 7
TM Thematic Mapper Landsat 4, Landsat 5 1
TopS at TopSat
TOr Package Tracking, Occupation, and Ranging TanDEM-X 2
Travers SAr PRIRODA-MIR 1
TV Camera PRIRODA-MIR 3
u K-DMC Standard DMC sensor UK-DMC 4
(Continued )
2016 by Taylor & Francis Group, LLC
32 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 .5 (c ontinued ) List of Selected Optical Remote Sensing Satellites along with Their Spectral Band Information
Name of Sensor Description Mission No. of Bands
Vege TATION SPOT 4, SPOT 5 4
Vg T-P Vegetation instrument on PROBA-V PROBA-V 5
VIr S Visible and Infrared scanner TRMM 2
WAOSS-B Wide-Angle Optoelectronic Stereo Scanner BIRD 2
Water Vapor r adiometer GFO 2
WFI Wide Field Imager CBERS, CBERS-2 3
WiFS Wide Field Sensor IRS-1C, IRS-1D, IRS-P3 5
WindSat Coriolis 1
WorldView-1 Provides highly detailed imagery for precise map WorldView-1 8
creation and in-depth image analysis
WV 3 MSS WorldView-3 multispectral sensor WorldView-3 1
WV 3 PAN WorldView-3 panchromatic sensor WorldView-3 8
WV 3 SW WorldView-3 shortwave infrared sensor WorldView-3 8
WV110 Standard 4 colors + new 4 colors WorldView-2 1
WV60 PAN band for WorldView-2 WorldView-2
Xgre X-ray and Gamma Relativistic Electron Detector TARANIS 1
XPS XUV photometer SORCE
Source: Adapted and modified from ITC, ITCs database of satellites and sensors, 2014, http://www.itc.nl/research/products/sensordb/allsensors.aspx,
accessed on May 2, 2014.
Hyperion high-resolution (30 m) hyperspectral imaging sys- and mineral for environmental assessments including vegeta-
tem acquires images in 220 contiguous spectral bands with tion health studies under field or laboratory conditions (Yang
very high radiometric accuracy within a bandwidth range of et al., 2005; Swayze et al., 2014). Hyperspectral imaging and
0.42.5m. With these large numbers (220) of bands, complex imaging spectroscopy are the technologies used to acquire a
land characteristics can be identified. Kruse etal. (2003) used spectrally resolved image of an object or scene (Butler and
Hyperion imagery to map the minerals in and around Cuprite, Laqua, 1996). Also, spectrometry is used as lab/computer
NV, with high accuracy. imaging technology for food qualitymeat deterioration
Figure 1.15 is an example of AVIRIS hyperspectral imaging (Panigrahi etal., 2003; Savenije etal., 2006), fruit sweetness
system data taken over Yellowstone National Park acquired analysis (Guthrie and Walsh, 1997; Nicolai et al., 2007), leaf
in 224 continuous bands having a 10 nm band pass over the chlorophyll measurement (Shibata, 1957), etc.and thus
spectralwavelength range of 3502500nm (from visible light to images acquired through space imaging spectrometry plat-
NIR). AVIRIS collects 20 m wide pixels at approximately 14m forms would analyze the Earth objects and phenomena in the
spacing. The sensor swath width is approximately 10.5 km. most qualitative means.
Kokaly etal. (2003) used AVIRIS data from Yellowstone National
Park to map the vegetation types in the park with very high
1.3.4 RADAR and SoDAR Remote Sensing
accuracy. Kokaly etal. (1998) used AVIRIS imaging spectrom-
etry data to characterize and map the biology and mineralogy The earliest development of radar technology was in 1886 by a
of Yellowstone National Park. Above all, commercial companies German physicist named Heinrich Hertz. A Russian physicist,
also acquire hyperspectral imageries on demand. Table 1.6 con- Alexander Popov in 1985 showed an application of radar tech-
tains the selective list of hyperspectral imaging system satellites nology in detecting far lightning strikes. While the early appli-
with pertinent specifications on number of bands, spatial reso- cations of RADAR remote sensing were focused on defense
lution, and applications. and military applications primarily for reconnaissance surveys
Over the past years, the demands for better mapping and of tropical environments, recent applications in the past few
characterization of Earth objects and related phenomena have decades have seen more diverse applications. Conventional
increased, resulting in a higher need to better understand the radar such as Doppler radar is mostly associated with weather
biophysical interactions of radiation, atmospheric effects, and forecasting, aerial traffic control, and subsequent early warn-
albedo properties of surface materials. Ground-based remote ing. Doppler radar is used by local law enforcements moni-
sensing using spectroradiometers provide an innovative toring of speed limits and in enhanced meteorological data
approach to measuring surface reflectance of materials while collection such as wind speed, direction, precipitation loca-
removing the interfering effects of atmospheric path radi- tion, and intensity. Interferometric SAR is used to produce
ance, absorption, and scattering effects (Clark et al., 2002). precise digital elevation models (DEMs) of large-scale terrain
Thus, spectroscopy serves as a tool to map specific material using RADARSAT, TerraSAR-X, Magellan, etc. Refer to Table
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 33
Flight direction
Flight direction
Linear detector array
Across track scanner
h
at
sw h Instantaneous
un
d at
o sw field of view
Gr d
un n
ip ro ctio
str G
ire
d d
oun an Angular
Gr Sc field of view
us
o
ne w
n ta vie
f
Ground resolution sta o
In eld Ground resolution
(pixel) fi (pixel)
Pushbroom (along track) scanning system Across track/whiskbroom (across track) scanning system
(a)
Scanning Prism
mirror
Tape
Detectors recorder
Motor Dichroic grating
Airplane or
sensing system body
Field of view
Reflected light from
objects phenomena
gh n
fli tio
t
of rec
i
D
th
wa s
ndu o
Gr
s)
te
sa es
lli
ith lin
te
s w an
rie Sc
(b)
a
(v
FIg u r e 1.14 Basic operating system of an optical sensor: (a) across-track or whiskbroom scanner, and (b) multispectral scanner, onboard an
airplane or satellite acquiring images. (Adapted and modified from Avery, T.E. and Berlin, G.L., Fundamentals of Remote Sensing and Airphoto
Interpretation, Prentice Hall, Upper Saddle River, NJ, 1992.)
2016 by Taylor & Francis Group, LLC
34 Remotely Sensed Data Characterization, Classification, and Accuracies
Forest cover types
Yellowstone
Mammoth National Park
Mount Mammoth hot springs area
hot springs Everest
Explanation
Lodgepole pine
(LP13, LP)
Douglas fir
Whitebark pine
Spruce/fir
Swan 2 km
Lake
Pixels for which mature forest cover
was not detected are depicted in
gray-scale shading.
Mammoth
hot springs
2
3
Obsidian Creek
10 km
AVIRIS Scenes
Study boundary
FIg u r e 1.15 AVIRIS image coverage for Yellowstone National Park collected on August 7, 1996, overlaid on Landsat TM imagery. (From http://
speclab.cr.usgs.gov/national.parks/Yellowstone/ynppaper.html.)
1.3 for specifications about RADAR data. In the United States, 1998; Huisman etal., 2003; Durso and Minacapilli, 2006), for-
the National Weather Service (NWS) uses Next-Generation estry application (Leckie and Ranson, 1998), ecosystem studies
Radar (NEXRAD) satellites to detect atmospheric events such (Waring et al., 1995; Kasischke et al., 1997), geomorphic and
as precipitation and wind movement and thus becomes use- hydrologic applications (Lewis and Henderson, 1998; Glenn and
ful in tracking tornados, thunderstorms, and other weather- Carr, 2004), snow and ice mapping and analysis (Hall etal., 2012),
related hazards. Figure1.16 is an example of a freely accessible urban remote sensing (Dong etal., 1997; Chen etal., 2003), and
NEXRAD data analysis by NWS to detect probable tornado land use and land cover mapping (Haack and Bechdol, 2000).
sighting. NASAs next-generation dual-frequency precipitation The multiple look direction (ascending and descending)
radar whose mission was Global Precipitation Measurement characteristic of RADARSAT is especially helpful in distin-
(GPM) can accurately measure rain and snow worldwide every guishing differently oriented linear features such as the traces
3h. The European Space Agency (ESA) SAR imagery is the most of fracture and faults (Riopel, 2000). Some of the fracture
popular microwave radar imagery used for land cover map- traces that might be hidden in one look direction due to a shad-
ping in cloudy, humid, and wet areas similar to coastal regions. owing effect become visible from the opposite look direction.
Microwaves can penetrate the water vapor in the atmosphere Natural corner reflection is another special property of radar.
and collect discernable images for analysis. SAR imageries are If any lineament and fracture along a small fault is present, the
very useful when conventional multispectral satellite imaging backscatter energy from the corner will be highlighted, reveal-
cannot acquire cloud-free images. Figure1.17 is an example of ing faults or other lineaments that would be otherwise hidden
a SAR image acquired to distinguish the land inundation in (Riopel, 2000).
China due to the Three Gorges Dam. Table1.7 shows a list of
radar-based satellite information.
1.3.5 LASeR and RADAR Altimetry imaging
In addition to applications in weather and climate analysis,
RADAR data are used in various applications such as precision Laser and radar altimeters on satellites have provided a wide
agriculture (Brisco et al., 1989), mapping soil water distribu- range of data on detecting the bulges of water caused by grav-
tion with ground-penetrating RADAR (Dobson and Ulaby, ity in the ocean. It helps map features on the seafloor to a
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 35
TABLe 1 .6 List of Selected Hyperspectral SensorBased Satellites along with Detailed Specification
Platform
(Airborne/
Revisit
Spaceborne)
Name of No. of Spatial Period Operation Country/
Sensor Description Mission Air Space Bands Resolution (Days) Since Agency
AVIr IS Airborne Visible/Infrared JPL Earth Remote X 224 20 m and less At wish 1987 NASA JPL,
Imaging Spectrometer Sensing United States
HSI Hyperspectral imager EnMAP X 244 30 m 4 2008 Germany
ACe -FTS Atmospheric Chemistry SCISAT X n/a 15 times/ 2003 Canada
(Ozone) Experiment Fourier day
Transform Spectrometer
CHr IS Compact High-Resolution PROBA-1 X 153 17 m/34 m 2002 ESA, Europe
Imaging Spectrometer (5 modes)
e PTOMS Earth Probe Total Ozone Earth Probe X 6 n/a n/a 1996 NASA,
Mapping Spectrometer United States
Hyperion High-resolution hyperspectral EO-1 X 220 30 2001 USGS,
imager with 220 spectral bands United States
(from 0.4 to 2.5m)
ILAS Improved Limb Atmospheric ADEOS X 2 2,00013,000 m 2002 JAXA, Japan
Spectrometer (IFOV)
ILAS-II Improved Limb Atmospheric ADEOS-II X 4 1,000 m (IFOV) 2002 JAXA, Japan
Spectrometer-II
ISTOK-1 Infrared Spectrometer PRIRODA-MIR X 64 750 m 1996 RKA, Russia
MISr Multi-angle Imaging Terra X 4 275 m 9 1999 JPL,
SpectroRadiometer United States
MODIS Moderate-Resolution Imaging Aqua, Terra X 36 250, 500, 1,000 2 1999 NASA,
Spectroradiometer (PFM on m United States
Terra, FM1 on Aqua)
MOS Modular Optoelectronic IRS-P3 X 18 500 m 3 (approx.) 1996 DLR, Germany
Scanning Spectrometer
MOS-A Modular Optoelectronic PRIRODA-MIR X 4 2.87km 3 (approx.) 1996 DLR, Germany
Scanning Spectrometer
MOS-B Modular Optoelectronic PRIRODA-MIR X 13 0.7 0.65km 3 (approx.) 1996 DLR, Germany
Scanning Spectrometer
SCIAMACHY Scanning Imaging Absorption ENVISAT X 8 30 2730 3 2002 ESA, Europe
Spectrometer for Atmospheric 240km (IFOV)
Chartography
SSJ/5 Precipitating Particle DMSP-16 X 2003 U.S. Air Force,
Spectrometer United States
SSu SI Special Sensor Ultraviolet DMSP-16 X 2003 U.S. Air Force,
Spectrographic Imager United States
TANSO-FTS Thermal and NIR Sensor for Greenhouse Gas X 4 0.5km 3 2009 JAXA, Japan
Carbon Observation, Fourier Observing Satellite
transform spectrometer (GOSAT) (Ibuki)
Te S Tropospheric Emission AURA X 12 0.5km 16 2004 NASA,
Spectrometer United States
TOMS Total Ozone Mapping ADEOS X 6 47km 3.1km 41 19961997 ILRS, NASA
Spectrometer (IFOV) (out of
(see also SBUV/TOMS) service)
Source: Adapted and modified from ITC, ITCs database of satellites and sensors, 2014, http://www.itc.nl/research/products/sensordb/allsensors.aspx,
accessed on May 2, 2014.
resolution of approximately 1 mile. The Light Amplification by radar altimeter acquires accurate measurements of the thick-
Stimulated Emission of Radiation (LASER) and RADAR altim- ness of floating sea ice for annual variations and surveys the
eters measure the height and wavelength of ocean waves, wind surface of ice sheets accurately enough to detect small changes.
speeds, wind direction, surface ocean currents and their direc- This supports the study of global warming and climate change
tions, etc. (Brenner etal., 2007; Giles etal., 2007; Connor etal., effect on present-day global land cover changes and especially
2009). Connor etal. (2009) used satellite microwave altimeters diminishing snow and ice cover. Figure 1.18 shows a graphical
boarded on ENVISAT/RA-2 to measure the Arctic Sea ice depth depiction of how satellite altimetry works with Jason-2 Ocean
and height accurately. ESA-operated CryoSat-2 interferometric Surface Topography Mission (OSTM). Jason-2 OSTM measures
2016 by Taylor & Francis Group, LLC
36 Remotely Sensed Data Characterization, Classification, and Accuracies
Santa Rosa Country, FL
Approximate
Tornado location
Escambia Escambia 1214 PM CST
FIg u r e 1 .16 Sample of a Next-Generation Radar imagery from a tornado event.
FIg u r e 1.17 On October 21, 2009, a scene of the Three Gorges Dam, China, was captured using the TerraSAR-X sensor. It is an X-band radar sen-
sor, operating in different modes and recording images with different swath widths, resolutions, and polarizations. (From www.geo-airbusds.com.)
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 37
TABLe 1 .7 List of Selected RADAR SensorBased Satellites along with Detailed Specification
Platform
(Airborne/
Spatial
Spaceborne)
No. of Resolution/ Revisit Period
Name of Sensor Description Mission Air Space Bands IFOV (Days) Operation Since Agency/Country
ALMAZ-1 SAR ALMAZ-1 X 1 15 m 19911992 NPO
a Out of Mashinostroyenia,
Service Russia
Active Phased SAR-X band TanDEM-X X 1 (SAR-X) 11 2010 DLR, Germany
Array
ALT Dual-Frequency TOPEX/Poseidon X 2 6km 10 19922005 NASA, United
Radar Altimeter b Proposed States
ASAr Advanced SAR ENVISAT X 1 (SAR-C) 30 (150) m 35 20022012 ESA, Europe
(out of service)
COSI Corea SAR KOMPSAT-5 X 1 1, 3, 20 m 28 2013 KARI, South Korea
Instrument (Arirang 5)
CPr Cloud Profiling EarthCARE X 1 25 2015 (proposed) ESA, Europe
RADAR
DPr Dual-Frequency GPM Core X 2 5km 2014 NASA, United
Precipitation Radar States
g MI Microwave Radar GPM Core X 13 626km 2014 NASA, United
Instrument (IFOV) States
HQSAr High-Quality SAR TerraSAR-X X 1 SAR-X) 1 m 11 2007 DLR, Germany
LBI L-band SAR SAOCOM-1A X 1 (SAR-L) 10 m 16 2015 (proposed) CONAE, Argentina
MMSAr Multimode SAR TecSAR X 1 SAR-X) 0.1 m 2008 IAI, Israel
Mr S, Cr S Scan SAR mode RISAT-1 X 1 (SAR-C) 25 m 25 2012 ISRO, India
PALSAr L-band SAR ALOS X 1 (SAR-L) 10 m 46 20062011 JAXA, Japan
(out of service)
PALSAr -2 L-band SAR-2 ALOS-2 X 1 (SAR-L) 3 m 14 2014 JAXA, Japan
Pr Precipitation radar TRMM X 2 25km 0.5 1997 NASA, United
States
r ADAr SAT-2 Radar RADARSAT-2 X 1 (SAR-C) 3100 m 24 2007 MDA Geospatial
Services Inc.
SAr (Jer S-1) SAR Japanese Earth X 1 (SAR-L) 18 m 44 19921998 JAXA, Japan
Resource satellite (out of service)
(JERS-1)
SAr SAR RADARSAT-1 X 1 8.4100 m 24 19952007 MDA Geospatial
(r ADAr SAT-1) (out of service) Services Inc.
SAr (Seasat) SAR Seasat X 1 25 m 19781978 NASA, United
(out of service) States
SAr C-band SAR on Sentinel-1A X 1 (SAR-C) 5 m 12 2014 ESA, Europe
(Sentinel-1A) Sentinel-1A
SAr 2000 RADAR COSMO-SkyMed X 1 1m 16 2007 ASI, Italy
SAr -L L-band SAR MapSAR X 3m 2011 INPE, Brazil
SIr AL Interferometric CryoSat-2 X 250 m 369 with 30days 2010 ESA, Europe
Radar Altimeter subcycle
Sr AL SAR Radar Sentinel-2 X 13 10, 20, 60 m 5 2007 (proposed) ESA, Europe
Altimeter
Sr AL SAR Radar Sentinel-3 X 0.51km 27 2017 (proposed) ESA, Europe
Altimeter
X-SAr SAR SRTM X 1 2590 m 20002000 NASA, United
(out of service) States
Source: Adapted and modified from ITC, ITCs database of satellites and sensors, 2014, http://www.itc.nl/research/products/sensordb/allsensors.aspx,
accessed on May 2, 2014.
2016 by Taylor & Francis Group, LLC
38 Remotely Sensed Data Characterization, Classification, and Accuracies
sea surface height through imaging that ultimately determines 1.3.6 LiDAR Remote Sensing
ocean circulation, climate change, and sea-level rise. Table 1.8
contains a selected list of LASER and RADAR altimetry imag- LiDAR technology, a type of active remote sensing, was devel-
ing sensor platform with their ancillary information. oped in the early 1960s following the invention of LASER and
was initially used to measure distance by illuminating a target
GPS
Satellite
Jason-2
Orbit
Altimeter
Microwave radiometer
measuring water vapor
Range
Laser Doris
station station
Ocean surface Satellite
topography altitude
Geoid
Sea surface
height
Ocean floor
Ellipsoid
FIg u r e 1.18 Graphical depiction of how Jason-2 satellite altimetry works to monitor the ocean. (From http://www.ppi.noaa.gov/bom_chapter3_
satellite_management/National Oceanic and Atmospheric Administration - Office of Program Planning and Integration. 2015. Chapter 3 - NOAA
Operations - Satellite Management. Retrieved from http://www.ppi.noaa.gov/bom_chapter3_satellite_management/ on 3/30/15) Please read the
foreword to gather the importance of radar altimetry in recording climate impacts such as sea-level rise.
TABLe 1 .8 List of Selected LASER and RADAR Altimetry SensorBased Satellites along with Detailed Specification
Platform
(Airborne/
Spaceborne)
Name of No. of Spatial Revisit Period Agency/
Sensor Description Mission Air Space Bands Resolution (Days) Operation Since Country
ALT Dual-Frequency Radar TOPEX/ X 2 6km 10 19922005 NASA,
Altimeter Poseidon (out of service) United States
ALT Altimeter Seasat X 1 (SAR-K) 1.6km 19781978 NASA,
(Seasat) (out of service) United States
AltiKa High-resolution altimeter SARAL/ X 2 2km 35 2013 CNES, France
including bifrequency AltiKa
radiometric function
g LAS GLAS ICESat X 2 100 m1km Monthly/ 20032010 NASA,
seasonal (out of service) United States
rA Radio altimeter ERS-1 X 2 16km 35 19912000 ESA, Europe
(out of service)
rA Radio altimeter ERS-2 X 2 16km 35 19952011 ESA, Europe
(out of service)
SIr AL SAR Interferometric CryoSat-2 X 1 (Radar 369 with 30 2010 ESA, Europe
Radar Altimeter altimeter) day subcycle
Sr AL SAR Radar Altimeter Sentinel-3 X 1 (Radar 0.51km 27 2017 (proposed) ESA, Europe
altimeter)
Source: Adapted and modified from ITC, ITCs database of satellites and sensors, 2014, http://www.itc.nl/research/products/sensordb/allsensors.aspx,
accessed on May 2, 2014.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 39
with LASER. The first LiDAR application was in the field of mete- converted to a position in a 3D frame of reference with spatially
orology; the National Center for Atmospheric Research (NCAR) coherent cloud of points. In this processing stage, some LiDAR
used LiDAR to measure the cloud distance from ground surface data provide texture or color information for each point Renslow
(Goyer and Watson, 1963). The 1971 use of LiDAR technology 2012, Rao, 2013). The processed 3D spatial and spectral informa-
by the Apollo mission to map the surface of the moon made the tion contained in the dataset allows great flexibility to perform
technology well known among the general public. Soon LiDAR manipulations to extract the required information from the point
became a very common tool for geospatial technology research- cloud data Renslow 2012, Rao, 2013). Thereafter, visualization,
ers and users in the present century when LiDAR was first used segmentation, classification, filtering, transformations, gridding,
by airplanes to map the land topography. and mathematical operations are conducted on the data to obtain
Satellite-based LiDAR data acquisition is spearheaded by NASA, required information of Earth objects or phenomena as discussed.
ESA, Russian Federal Space Agency (RKA), and German Aero- The first return of the LiDAR data is generally from the tall-
space Research Establishment (DLR) with sensors such as laser est features, that is, tallest tree canopy or top of high-rise build-
reflectometry array (LRA) on Jason-2 (OSTM) and Geoscience ings, the intermediate returns are from the canopy of the small
Laser Altimeter System (GLAS) for Ice, Cloud, and landElevation trees and shrubs, and the final return is from the ground surface.
Satellite (ICESat) mission, atmospheric laser Doppler instrument These individual return data are processed to get height infor-
(ALADIN) and ATLID EarthCARE, Priroda-Mir mission, and mation of the features discussed earlier. Figure 1.19 displays
laser communication terminal (LCT) for TanDEM-X mission, the point cloud LiDAR data showing different returns color-
respectively (Table 1.9). Table 1.9also provides the satellite speci- ized based on height as visualized using LiDAR Data Viewer
fications on selected LiDAR remote sensing satellites. of the USGS Fusion software (http://forsys.cfr.washington.edu/
Present-day aerial LiDAR data are collected from airplanes fly- fusion/fusionlatest.html). Using interpolation and smothing
ing around 25 miles above the ground surface and provide a mini- algorithms, the point cloud is rendered as a grid surface, which
mum of 30cm resolution data. The data are collected by focusing can be easily manipulated in a GIS using map algebra operations
and scanning a LASER beam from an airplane to the objects on the to produce canopy height, ground elevations, etc. The normal-
ground surface, which then acquires the data as different returns ized digital surface model (nDSM) is developed in the LiDAR
(Panda etal., 2012a,b; Amatya etal., 2013; Panda and Hoogenboom, processing software Quick Terrain Modeler (Applied Imagery,
2013). Table 1.9 includes a list of LiDAR-based satellite information. Chevy Chase, MD) from the ground return data to produce the
LiDAR technology is becoming popular since the start of the millen- accurate DEM (topography) data of the ground surface (Panda
nium due to its advantage in mapping the Earth topography along etal., 2012a,b). Panda etal. (2012a,b) with their wetland change
with object heights on the Earths surface, thus supporting image and cause recognition study of Georgia, United States, coast
classification process tremendously (Dubayah et al., 2000). The found a major discrepancy with the 10-m DEM data available
NOAA mission of collecting LiDAR data of the entire U.S. coast and through USGSs National Elevation Dataset (NED). The problem
some ecologically sensitive areas in the United States made LiDAR was found while comparing with the DEM created with 30-cm
a well-known technology. NOAA Coastal Services Centers Digital resolution LiDAR for Camden County, GA. They found an aver-
Coast Data Access Viewer (http://www.csc.noaa.gov/dataviewer/#) age change of ground elevation of 0.9 m with a range of 0.17
provides free access to LiDAR data, including some recent acquisi- to 1.5 m when comparing the elevation at permanent locations.
tions of 2009 and later, for a major portion of the United States. Figure 1.20 shows the DEM developed for Camden County, GA,
LiDAR remote sensing instrument provides point cloud data. using the 2010 LiDAR data, the 10 m DEM of the county created
The crude point cloud data are processed and each laser shot is in 2000 (based on the raster metadata), the comparison table of
TABLe 1 .9 Selected LiDAR SensorBased Satellites along with Detailed Specification
Name of Platform No. of Spatial Revisit Period
Sensor Description Mission (Spaceborne) Bands Resolution (Days) Operation Since Agency/Country
ALADIN ALADIN ADM-Aeolus X 7 2015 (planned) ESA, Europe
ALISSA Latmosphere par PRIRODA-MIR X 1 150 m 1996 RKA, Russia
Lidar Sur Saliout
ATLID High-spectral- EarthCARE X 2 (S and X) 25 2016 (planned) ESA, Europe, and
resolution LiDAR JAXA, Japan
g LAS GLAS ICESat X 2 70 m and 91 20032010 NASA, United States
3cm vertical (out of service)
LCT LCT TanDEM-X X 11 2010 DLR, GFZ,
Infoterra, Germany
Lr A Laser Retroreflector Jason-2 X 10 2008 NASA, United States
(Jason) Array (OSTM)
Source: Adapted and modified from ITC, ITCs database of satellites and sensors, 2014, http://www.itc.nl/research/products/sensordb/allsensors.aspx,
accessed on May 2, 2014.
2016 by Taylor & Francis Group, LLC
40 Remotely Sensed Data Characterization, Classification, and Accuracies
Height
153.67
147.92
128.68
103.69
78.70
53.70
Histogram
FIg u r e 1.19 A 3D display of point cloud light detection and ranging (LiDAR) data showing different returns (colorized based on height as
visualized using the LiDAR Data Viewer of the USFS Fusion software). Also, note the interactive measuring cylinder with marker can be moved
about the virtual space for measuring height.
elevation differences, and the algorithm developed to modify the woody debris using threshold filtering of the point cloud data.
10 m DEM of the study area. Furthermore, LiDAR data, owing to its properties to penetrate
Panda and Hoogenboom (2013) used 30 cm LiDAR from a canopy, have high application in improving wildland fire map-
blueberry orchard in Pierce County, GA, to obtain the height ping through better estimates of canopy bulk density and base
of pine trees, existing structures, and blueberry plants with height (Riano et al., 2004; Lee and Lucas, 2007; Erdody and
ground elevation to be used in the object-based image analysis Moskal, 2010), canopy cover (Lefsky etal., 2002; Hall etal., 2005),
(OBIA)based image segmentation with eCognition software shrub height (Riano et al., 2007), and aboveground biomass
for blueberry orchard SSCM. Yu et al. (2010) in their study of (Edson and Wing, 2011; Vuong and Rao, 2013) and carbon stocks
urban building density determination using airborne LiDAR (Asner etal., 2010; Goetz and Dubayah, 2011). Other studies have
data explained the development of nDSM and subsequent DEM documented better mapping and classification outputs using an
from the LiDAR data for decision support. Kelly et al. (2014) integrated approach where multispectral data are fused with
completed a study of coastal marsh management decision sup- LiDAR data (Garcia etal., 2011; Rao and Miller, 2012). Innovative
port by using 30cm LiDAR tiles. The DEM of South Carolina approaches to map natural resources at global scales have been
coast was developed to delineate the areas of saltwater intrusion initiated in the recent past such as using the GLAS (see Figure
(with the highest possible tide heights) calculated with the lat- 1.21) aboard the ICESat in conjunction with MODIS percent tree
est climate change analysis. Additionally, the researchers used cover product (MOD44B) to map forest canopy height globally
nDSM from LiDAR imagery to classify the NAIP imagery using with spaceborne LiDAR (Simard etal., 2011). LiDAR is also used
eCongnitions OBIA image segmentation procedure to produce to detect and measure the concentration of various chemicals
an output of the coastal forest plant species. These applications in the atmosphere including the use of camera LiDAR instru-
show the potential of LiDAR in environmental (forestry, ecol- ment to accurately measure the particulates (Barnes etal., 2003).
ogy, agriculture, urban, and others) management decision sup- These and many other studies show the potential of LiDAR data
port systems. in environmental (forestry, ecology, agriculture, urban, and oth-
Similarly, Rao (2012) demonstrated applications of LiDAR- ers) management and decision support systems, particularly in
based mapping and height estimates for snags and coarse context of climate change issues at regional and global levels.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 41
10 m elevation (m) LiDAR elevation (m) Difference (m) Comparison of elevation reading difference
6.235435 6.4008 0.165365 from 10 m DEM and 30 m DEM developed
8.145018 7.62 0.525018 from LiDAR
8.226008 6.7056 1.520408 12 y = 0.8498x + 0.3488
LiDAR elevation (m)
7.989067 7.62 0.369067 10 R2 = 0.9132
12.744309 11.27 1.474309 8
7.929955 7.3152 0.614755 6
Series1
9.491169 8.8392 0.651969 4
2 Linear (series1)
9.68649 7.9248 1.76169
0
8.050032 6.7056 1.344432
0 5 10 15
5.259323 4.2672 0.992123
10 m DEM elevation (m)
0.9088406 = avg diff
FIg u r e 1.20 Comparison of 10 m digital elevation model (DEM) of National Elevation Dataset and DEM developed with airborne LiDAR of
Camden County, GA. (From Panda, S.S. etal., Wetland change and cause recognition in Georgia coastal plain, in: Presented in the International
American Society of Agricultural and Biological Engineers (ASABE) Conference 2012, July 29August 2, 2012, Dallas, TX, Paper # 1338205, 2012a;
Panda, S.S. etal., Stomatal conductance and leaf area index estimation using remotely sensed information and forest speciation, in: Presented in
the Third International Conference on Forests and Water in Changing Environment 2012, September 1820, 2012, Fukuoka, Japan, 2012b.)
Currently, LiDAR is also extensively used by law enforce- 1.3.7 Microwave Remote Sensing
ment for tracking of vehicle speed military for weapon ranging,
mine detection for countermine warfare, and laser-illuminated Unlike LiDAR remote sensing, microwave sensing encompasses
homing of projectiles; in mining industry to calculate the ore both active and passive forms of remote sensing. Atmospheric
volume (3Dlasermapping, 2014); in physics and astronomy by scattering affects shorter optical wavelengths but longer wave-
measuring the distance to the reflectors placed on the moon and lengths of EMS are not affected by atmospheric scattering
detecting snow on the Mars atmosphere (NASA-Phoenix Mars (Aggarwal, 2003). Therefore, longer-wavelength microwave
Lander, 2014); in the field of robotics for the perception of the radiation can penetrate through cloud cover, haze, dust, and
environment as well as object classification and safe landing of all but the heaviest rainfall, and thus microwave remote sens-
robots and manned vehicles (Amzajerdian etal., 2011); in space- ing can be useful for geospatial community as data can be col-
flight, surveying using mobile LiDAR instruments (Figure 1.22); lected with microwave remote sensing systems in almost all
in adaptive cruise control; and in many other high-end energy- weather and environmental conditions. RADAR remote sensing
producing sources like wind farms and solar farms for wind is the example of active microwave remote sensing. The radar
velocity and turbulence measurement and optimizing solar units transmit short pulses or bursts of microwave radiation
photovoltaic systems by determining shading losses. when sunlight is unavailable and then obtain reflectance signal
from the Earth-based objects (Avery and Berlin, 1992). Therefore,
2016 by Taylor & Francis Group, LLC
42 Remotely Sensed Data Characterization, Classification, and Accuracies
40
Canopy height (m)
30
20
10
0 100 200
km
FIg u r e 1.21 Wall-to-wall mapping of canopy height implemented using the Geoscience Laser Altimeter System (GLAS) instrument on the Ice,
Cloud, and land Elevation Satellite. GLAS data model. The inset shows the sensitivity and capability of the model to disturbance gradient in the
Amazon. The GLAS laser transmits short pulses (4 ns) of infrared light at 1064nm and visible green light at 532nm 40 times/s. (From Simard, M.
etal., J. Geophys. Res., 116, G04021, 2011, doi: 10.1029/2011JG001708.)
SAR antenna
SAR antenna
Incident
Backscattered
radar
rader pulse
pulse
(a) Ground targets (b) Ground targets
FIg u r e 1.23 (a) Active microwave remote sensing: Microwave
pulses are incident upon objects from the synthetic aperture radar
(SAR) satellite in no-sunlight condition. (b) Passive microwave remote
FIg u r e 1.22 Mobile light detection and ranging for land surveying. sensing: Microwave radiations are reflected from the objects on ground
(From Photo Science Geospatial Solutions, Lexington, KY, http://www. water receiving light energy from sun.
photoscience.com/services/mobile-mapping.)
microwave remote sensing system can be operable in the night. 2003), temperature of the estuary and sea surface temperature
When the microwave remote sensors collect naturally emitted (SST) (Blume et al., 1978; Klemas, 2011), and rain, snow, ice,
microwave energy, though small, from the objects without send- and sea surface condition (Fung and Chen, 2010), water vapor,
ing pulse from the system, it is passive microwave sensing, which cloud, oil slick, and different gaseous presence in the atmosphere
is similar in concept to thermal remote sensing (Avery and Berlin, like COx, ozone, and NOx. Microwave scatterometer, microwave
1992). The naturally emitted microwave energy is related to the altimeter, and imaging radar are the examples of active micro-
temperature and moisture properties of the emitting object or sur- wave sensing, while microwave, radiometers, and scanners are
face. Figure 1.23 shows the image acquisition principle of both all examples of passive microwave sensing technology. Table 1.10
active and passive remote sensing systems. shows the typical microwave sensorbased satellites and their
Microwave remote sensors efficiently monitor targets such specifications. Table 1.11 summarizes passive and active sens-
as soil moisture content (Engman and Chauhan, 1995; Wagner ingbased microwave sensors and their target areas of applica-
etal., 2007), salinity (Wilson etal., 2001; Matternicht and Zinck, tions for Earth resources measurement.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 43
TABLe 1 .10 List of Selected Microwave SensorBased Satellites along with Detailed Specification
Platform
(Airborne/
Spaceborne)
Name of No. of Spatial Revisit Period Agency/
Sensor Description Mission Air Space Bands Resolution (Days) Operation Since Country
AMI Active Microwave ERS-1 X 1 30 m 35 19912000 ESA, Europe
(out of service)
AMI Active Microwave ERS-2 X 1 30 m 35 19952011 ESA, Europe
(out of service)
AMr Advanced Microwave Jason-2 (OSTM) X 3.3cm sea-level 10 2001 JPL, NASA,
Radiometer measurement United States
accuracy
AMSr Advanced Microwave ADEOS-II X 8 5, 10 1.6km 4 20022003 JAXA, Japan
Scanning Radiometer (IFOV) (out of service)
AMSr 2 Advanced Microwave GCOM-W1 X 7 0.15, 0.35, 0.75, 2 2002 NASA,
Scanning Radiometer (SHIZUKU) 0.65, 1.2, 1.8km United States
AMSr -e Advanced Microwave Aqua X 6 5, 10 1.4km 4 2002 NASA,
Scanning Radiometer (IFOV) United States
for EOS
AMSu -A Advanced Microwave Aqua, Metop-A, X 15 5 2.343km 29 (Metop) 2002 NOAA,
Sounding Metop-B, (IFOV) and 11 United States
(temperature) NOAA-15, (NOAA)
NOAA-16,
NOAA-17,
NOAA-18
(NOAA-N),
NOAA-19
(NOAA-N
Prime)
AMSu -B Advanced Microwave Aqua, X 5 15 2.343km 11 1998 NOAA,
Sounding NOAA-15, (IFOV) United States
NOAA-16,
NOAA-17
CrIMSS Cross track Infrared and JPSS-1 X 3 0.025, 0.0125, 2017 (proposed) NOAA,
Advanced Technology (NPOESS) 0.0625 m United States
Microwave Sounder
CrIMSS Cross track Infrared and Suomi NPP X 3 0.025, 0.0125, 2011 NASA,
Advanced Technology (SNPP) 0.0625 m United States
Microwave Sounder
g MI Microwave Radar GPM Core X 13 2014 (proposed) NOAA,
Instrument United States
JMr Jason-1 Microwave Jason-1 X 3 10 20012013 NASA,
Radiometer (out of service) United States
MADr AS Microwave Imager Megha- X 9 6 40km 16 times a 2011 CNES, France
Tropiques day
MHS Microwave Humidity Metop-A, X 5 16.3 1.078km 29 (Metop) 2006, 2012, ESA, Europe
Sounder Metop-B, (IFOV) and 11 2005, and 2009,
NOAA-18 (NOAA) respectively
(NOAA-N),
NOAA-19
(NOAA-N
Prime)
MIr AS Microwave Imaging SMOS X 1 (SAR-L) 35km 3 2009 ESA, Europe
Radiometer Aperture
Synthesis
MIS Microwave Imager/ JPSS-1 X 1 (SAR-L) 2017 (proposed) NOAA,
Sounder (NPOESS) United States
(Continued )
2016 by Taylor & Francis Group, LLC
44 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 .10 (c ontinued ) List of Selected Microwave SensorBased Satellites along with Detailed Specification
Platform
(Airborne/
Spaceborne)
Name of No. of Spatial Revisit Period Agency/
Sensor Description Mission Air Space Bands Resolution (Days) Operation Since Country
MIS Microwave Imager/ Suomi NPP X 1 (SAR-L) 2011 NOAA,
Sounder (SNPP) United States
MLS Microwave Limb AURA X 5 3km 16 2004 NASA,
Sounder United States
MSMr Multifrequency IRS-P4 X 8 360 236 m 2 19992010 (out ISRO, India
Scanning Microwave (Oceansat) of service)
Radiometer
MSr Microwave Scanning MOS-1 X 2 32km 17 19871995 (out JAXA, Japan
Radiometer of service)
MSr Microwave Scanning MOS-1b X 2 32km 17 19901996 (out JAXA, Japan
Radiometer of service)
MSu Microwave Sounding NOAA-10, X 4 11 19862001 (all NOAA,
Unit NOAA-11, out of service) United States
NOAA-12,
NOAA-14
MSu -e Microwave Sounding RESURS-O1-3 X 3 34 m 18 1994 SRC Planeta,
Unit Russia
MWr Microwave Radiometer ENVISAT X 2 1040 1200 m 35 20022012 (out ESA, Europe
for marine and of service)
2630 300 m
for land and
coastal
application
MWr Microwave Radiometer Sentinel-3 X 2 0.3km 27 2015 (proposed) ESA, Europe
MWr I Microwave Radiation FengYun-3A X 1km, 250 m 2008 NSMC, China
Imager
SAPHIr Microwave Sounding Megha- X 10km at NADIR 16 times a 2011 CNES, France
Instrument Tropiques day
SMMr Scanning Multichannel Nimbus-7 X 5 27149km 6 19781994 (out NASA,
Microwave Radiometer of service) United States
SMMr Scanning Multichannel Seasat X 5 27149km 6 19781978 (out NASA,
Microwave Radiometer of service) United States
TMI TRMM Microwave TRMM X 5 4.4km 0.5 1997 NASA,
Imager United States
TMr TOPEX Microwave TOPEX/ X 3 23.544.6km 10 19922005 (out NASA,
Radiometer Poseidon of service) United States
Source: Adapted and modified from ITC, ITCs database of satellites and sensors, 2014, http://www.itc.nl/research/products/sensordb/allsensors.aspx,
accessed on May 2, 2014.
1.3.8 SonAR and SoDAR Remote Sensing fish-finder instrument to map the bottom of two retention ponds
to study the advantages of retention pond in soil conservation.
SONAR is a technique that uses sound propagation to navigate Similarly, Dual-Frequency Identification Sonar (DIDSON) uses
and communicate with or detect objects on or under the sur- sound to produce images using high- and low-frequency modes
face of the water (Figure 1.24). Two types of technology share in the range of about 1540 m and have demonstrated benefit
the name sonar: passive sonar is essentially list ening for the in fish inventorying (Belcher etal., 2001; Moursund etal., 2003;
sound made by vessels; active sonar is emitting pulses of sound Tiffan etal., 2004).
energy and listening for the echoes. Blakinton et al. (1983) SODAR (Sonic detection and ranging) is upward-looking, in-
used SONAR technology to map the seafloor with the develop- air sonar used for atmospheric investigations. SODAR remote
mentofSeaMARCII. Kiddetal. (1985) used long-range side-scan sensing systems are like a LiDAR/RADAR system in which
SONAR to map sediment distributions over wide expansions sound waves are propagated instead of light or radio waves in
of the ocean floor. Fairfield and Wettergreen (2008) recently order to detect the object of interest (Bailey, 2000). Doppler
used multibeam SONAR to map the ocean floor. As sound can SODAR (Figure 1.25) is a widely used remote sensing system for
propagate through water, the sensing system uses it to know the weather forecasting (Goel and Srivastava, 1990; Beyrich, 1997).
objects underneath water. Sonar may also be used in air for robot Wind farms are using fulcrum compact-beam multiple-axis 3D
navigation. Skarda etal. (2011) used SONAR technologybased SODAR to increase efficiency of wind power production. Satellite
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 45
TABLe 1 .11 Microwave Remote Sensors and Their Application Areas
Sensor Target
Passive sensor
Near-sea-surface wind
SST
Sea condition
Salinity and sea ice condition
Microwave radiometer Water vapor, cloud water content
Precipitation intensity
Air temperature
Ozone, aerosol, NOx, other atmospheric constituents
Active sensor
Soil moisture content
Surface roughness
Lake and ocean ice distribution
Snow distribution
Biomass
SST
Microwave scatterometer Sea condition
Salinity and sea ice condition
Water vapor
Precipitation intensity
Wind direction and velocity
Sea surface topography, geoid
Ocean wave height
Microwave altimeter Change of ocean current
Mesoscale eddy, tide, etc.
Wind velocity
Image of ground surface
Ocean wave
Imaging radar Near-sea-surface wind
Topography and geology
Submarine topography
Ice monitoring
platformbased SONAR systems are operated by the NOAA, code) in longwaves, L1 = 1575.42MHz (19cm wavelength) and
NASA, U.S. Department of Defense, ESA, and Space Agency of L2 = 1227.6MHz (24.4cm wavelength), are transmitted to the
France (Centre National dEtudes Spatiales [CNES]). Table 1.12 receivers along with accurate time stamps and other satellite
provides selected list of these SONAR satellite systems with their information. These signals are intercepted by the GPS receiv-
specifications. These SONAR systems are mostly used for ocean ers where the timing code is translated to interpret the precise
phenomena study, atmospheric measurement, and ice and snow time it took the signal to reach the receiver. This precise time is
studies. used with the signal speed to calculate distance or range. The
intersection of the three ranges from three satellites is used to
accurately determine the location of the GPS receiver.
1.3.9 Global Positioning System
There are three segments involved in GPS data collection:
The GPS, now known as Global Navigation Satellite System (1) space segment, (2) control segment, and (3) user segment.
(GNSS), is a space-based satellite navigation system that pro- Figure 1.26 shows the GPS geosynchronous satellite constella-
vides location (coordinates and elevation) and time informa- tions and their orbit. The control segment consists of five ground
tion in all weather conditions, anywhere on or near the Earth, monitoring stations that continuously monitor the satellite tra-
where there is an unobstructed line of sight to four or more GPS jectory and atmospheric conditions and help develop an error
satellites (Van Diggelen, 2009). GPS satellites use the trilatera- model for the atmospheric interference. The error correction
tion technique to know the location of a receiver by coordinat- based on the model is uploaded back to the GPS satellites via the
ing signals received from four satellites. It is an application of master control facility at Colorado Springs, Colorado, United
active remote sensing system as unique signals (pseudorandom States. The receiver segment is the most important part, because
2016 by Taylor & Francis Group, LLC
46 Remotely Sensed Data Characterization, Classification, and Accuracies
Active SONAR
Passive SONAR
Radio waves received
Radio waves by the ship
transmitted from ship
transmitted from the
Radio waves submarine
reflected back to ship
from school of fish
FIg ur e 1.24 Example of active and passive sound navigation and ranging (SONAR) detection system emitting ultrasound and radio waves to
find the location of fish in the ocean and receiving radio waves from undersea submarine and detecting its location. SONAR is used in navigation,
communication, and detection of objects on surface and underwater.
While the history of the GPS is rooted in military applica-
Scattering tions, over the past several decades, the civilian applications of
volume GPS have far exceeded the military applications. The literature
is replete with numerous applications of GPS data that include
c Echo telecommunications, commerce and retailing industry, and
recreation, in addition to the traditional applications in natural
Acoustic Z resource management and planning. The GPS.govthe official
pulse U.S. government information portal about GPS and related
topicsprovides a good treatise about the diverse applications
Sodar
of GPS in everyday life limited only by the human imagination.
unit
More importantly, the website provides an important reposi-
tory of information pertaining to the GPS modernization pro-
FIg u r e 1.25 Wind measurement with a phased array SODAR SFAS
gram that includes new series of satellite acquisitions (GPS IIR,
from Scintec Corporation (Louisville, CO) and the working principle of
GPS IIF, and GPS III). The improvements bring about new fea-
SODAR sensing system.
tures such as the Architecture Evolution Plan (AEP) and the
it involves the end user application needs. With the advancement Next-Generation Operational Control System (OCX) that are
of technology, we have centimeter accuracy GPS receivers that geared to deliver enormous improvements in the robustness
are able to process signal correction information (kinematic) and accuracy of the entire GPS system in the coming years
on the fly, thereby facilitating accurate location data collec- (Figure 1.27). These infrastructure and signal developments
tion. Follow this (http://www.gps.gov/) site for more informa- are designed not only to increase operational capacity but also
tion about GPS. GPS is described in detail in another chapter to provide better geometry in shadow environments of urban
of this book. canyons and mountainous terrain.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 47
TABLe 1 .12 List of Selected SONAR TechnologyBased Satellites along with Detailed Specification
Name of Platform No. of Spatial Revisit Period Agency/
Sensor Description Mission (Spaceborne) Bands Resolution (Days) Operation Since Country
AIr S AIS Aqua X 7 13.5km 0.5 2002 NASA,
United States
AMSu -A AIS Aqua X 15 50km 2002 NASA,
United States
AMSu -A AIS Metop-A X 15 50km 29 2006 ESA, Europe
AMSu -A AIS Metop-B X 15 50km 29 2012 ESA, Europe
AMSu -A AIS NOAA-15 X 15 50km 11 1998 NOAA,
United States
AMSu -A AIS NOAA-16 X 15 50km 11 2000 NOAA,
United States
AMSu -A AIS NOAA-17 X 15 50km 11 2002 NOAA,
United States
AMSu -A AIS NOAA-18 X 15 50km 11 2005 NOAA,
(NOAA-N) United States
AMSu -A Advanced Microwave NOAA-19 X 15 50km 11 2009 NOAA,
Sounding (NOAA-N Prime) United States
HSB Humidity Sounder Aqua X 4 13.5km 0.5 2002 NASA,
for Brazil United States
MHS Microwave Humidity Metop-A X 5 50km 29 2006 ESA, Europe
Sounder
MHS Microwave Humidity Metop-B X 5 50km 29 2012 ESA, Europe
Sounder
MHS Microwave Humidity NOAA-18 X 5 50km 11 2005 NOAA,
Sounder (NOAA-N) United States
MHS Microwave Humidity NOAA-19 X 5 50km 11 2009 NOAA,
Sounder (NOAA-N Prime) United States
MLS Microwave Limb AURA X 5 3km 16 2004 NASA,
Sounder United States
MSu Microwave Sounding NOAA-10 X 4 105km 11 19862001 NOAA,
Unit (out of service) United States
MSu Microwave Sounding NOAA-11 X 4 105km 11 19882004 NOAA,
Unit (out of service) United States
MSu Microwave Sounding NOAA-12 X 4 105km 11 19912007 NOAA,
Unit (out of service) United States
MSu Microwave Sounding NOAA-14 X 4 105km 11 19942007 NOAA,
Unit (out of service) United States
SAPHIr Microwave Sounding Megha-Tropiques X 1 16 times/day 2011 CNES, France
Instrument
SSMIS Special sensor DMSP-16 X 4 1343 m 2003 USDD,
microwave imager/ United States
sounder
SSu Stratospheric NOAA-11 X 3 147.3km 11 19882004 NOAA,
Sounder Unit (out of service) United States
SSu Stratospheric NOAA-14 X 3 147.3km 11 19972007 NOAA,
Sounder Unit (out of service) United States
Source: Adapted and modified from ITC, ITCs database of satellites and sensors, 2014, http://www.itc.nl/research/products/sensordb/allsensors.aspx,
accessed on May 2, 2014.
The obvious benefits of location-based information collected include geometric rectification and registration of image data by
through GPS are immense and beyond the scope of this chap- way of ground control points (GCPs) collected through high-
ter to discuss. While these benefits are invaluable for applica- accuracy GPS units. Similarly, location information collected
tions involving field surveys, navigation, and basic GIS mapping through ground-truthing surveys using high-accuracy GPS
of what features are located where, there are additional benefits units plays an important role in image classification and accu-
that directly relate to remote sensing applications. Some of these racy assessment. Using the location information collected using
2016 by Taylor & Francis Group, LLC
48 Remotely Sensed Data Characterization, Classification, and Accuracies
1.4 Upcoming new Satellite
Sensor Platforms
According to NSSDC Master Catalog (2014), since the begin-
ning of 2014, 126 new satellites have been sent to space for Earth
observation and all areas of application. Some of the prominent
Orbits
among them are as follows: (1) ALOS 2 is a RADAR imaging
satellite system operated by the Japan Aerospace Exploration
Agency (JAXA). The primary functions are for land resource
Earth studies, disaster monitoring, and environmental research. (2)
GPM Core Observatory satellite is a core mission of NASA and
Satellites JAXA to provide real-time accurate information of rain and
snow every 3h, along with soil moisture, carbon cycle, winds,
and aerosol estimation. (3) Hodoyoshi 3 and 4 is an experi-
mental Earth-observing microsatellite built by the University
of Tokyo. (4) IRNSS 1B is a satellite-based navigation system
24 satellites in 6 orbital planes developed in India to be compatible with GPS and Galileo
4 satellites in each plane
Max. 12 satellites available at any location on earth (European navigation system). (5) Sentinel-1A is an ESAs two-
20,200 km altitude and 55 inclination satellite constellation with the prime objective of land and
ocean monitoring with C-band SAR continuity. (6) SkySat 2 is
FIg u r e 1.26 Global Positioning System space segment showing a commercial Earth-observing satellite by Skybox Imaging to
24satellites orbiting the Earth at an altitude of 20, 200km in 6orbits acquire very-high-resolution images. (7) TSAT is a 2U CubeSat
with 4 satellites in each. (From http://www.gps.gov/multimedia/images/ microsatellite communication network from Taylor University,
constellation.jpg.) Upland, IN, used to research about a nanosat network to observe
space weather with a plasma probe, three-axis magnetometer,
and three UV photodiodes and is part of the NASA ELaNa 5
GPS, better training sites could be used for algorithm learning CubeSat initiative. (8) UKube is the U.K. Space Agencys pilot
phases during image classification procedures. Furthermore, program with the miniature cube-shaped satellite that allows
as a postclassification routine, the field-collected GPS data are the United Kingdom to test cutting-edge new technologies in
useful as a reference information about the landscape feature space. (9) Velox 1 is a small satellite from Singapores Nanyang
or biophysical phenomenon that could be used in the accuracy Technological University with a high-resolution camera for
assessment exercise of comparing and validating outputs gener- Earth imaging and later split into two smaller satellitesN-
ated from image classification algorithms (Powell and Matzke, sat and P-satto conduct communications experiments. (10)
2004; Congalton and Green, 2008). WorldView-3 is a 0.5 m panchromatic and 1.8 m multispectral
FIg u r e 1.27 Global Positioning System modernization program schedule. The graphic provides links to the important information about
developments in the space segment (http://www.gps.gov/systems/gps/space/#IIRM) and control segment (http://www.gps.gov/systems/gps/
control/), including Architecture Evolution Plan (http://www.gps.gov/systems/gps/control/#AEP) and Operational Control System (http://www.
gps.gov/systems/gps/control/#OCX). (From GPS.gov.)
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 49
Oblique Vertical
FIg u r e 1 .28 Examples of an oblique aerial image and graphics of the image acquisition processes.
resolution Earth observation imaging satellite by DigitalGlobe. 1960s, but its advantages in environmental management and
11. NASAs SMAP (Soil Moisture Active Passsive) mission is other spatial analyses are observed more recently. More images
slated to provide global soil moisture measurement and soil will be acquired with oblique imaging platform in the next years.
freeze/thaw state using a combination of active radar and passive
radiometer. From this present trend of latest satellite launches,
it is presumed that the upcoming satellite launches will be with 1.5 Future of Remote Sensing and
the research-related nanosatellite platforms, satellites doing evolving Microsatellites
dualduties after expiration of successful first job assigned to sat-
ellites, and satellites launched with very-high-resolution imag- According to a study by U.S.-based Forecast International, in
ing capabilities. order to address the need for better imaging data driven by aero-
Newer satellites from older programs like NASA, CNES, space and defense requirements, more satellites will be launched
JAXA, and ESA are being programmed for future space launches from high-resolution imagery acquiring platforms (Anonymous,
with minimum two new improvements, that is, collecting data 2012). For this purpose, UAS and UAV sensor platforms would
with larger swath so that revisit time of a satellite will be faster serve the purpose. As discussed in Section 1.4, the present trends
and including more bands into the system to help collect data for in satellite launches are from very high-resolution, research-
better and improved Earth observation. For example, according oriented nanosatellite platforms; the UAS and UAV platforms
to Astrium Inc. (http://www.astrium-geo.com/en/147-spot-6-7- would suffice the need. According to Everaerts (2008), the ISPRS
satellite-imagery), Frances CNES is expanding SPOT imaging Congress in Istanbul passed a resolution that UAVs provide a
satellite series to 2024, and the latest series of SPOT 6 and SPOT new controllable platform for remote sensing and permit data
7 emphasize on collecting data by covering larger areas, that acquisition in inaccessible and dangerous environments. UASor
is, up to 6 million km2 area/day, an area larger than the entire UAVs are the prominent part of the entire system of flying an
European Union. SPOT 6 and 7 satellites also include 4-day aircraft and acquiring ultraspatial-resolution imagery. The aircraft
weather forecasting to the satellites tasking. NASAs Landsat 8 is controlled from the ground control station with reliable com-
satellite includes two new bands to help in analyzing more Earth munication network through air traffic control. Over the last
resources like coastal resources with more cloud cover. Joint fouryears, the number of UAV systems increased by leaps and
Polar Satellite System (JPSS) is the next-generation satellite sys- bounds in the area of remote sensing and mapping (Everaerts,
tem of the United States and a collaborative program between 2008). Everaerts (2008) also state that much of the work in the
NOAA and its acquisition agent NASA, and its probable launch use of UAVs in remote sensing is in research stage presently and
date is scheduled to be 2017. It will help weather and natural haz- the future of remote sensing in essence depends on the growth
ard forecasting at a newer level. of UAS- and UAV-based remote sensing. As in most countries
Oblique aerial photograph is the latest development in aerial aviation regulations are adapted to include UAS and UAV sys-
imaging area, which has high potential growth in future years. tems into the general airspace, these microsatellite systems will
Figure 1.28 shows an example of oblique aerial photograph. become the preferred platforms of future remote sensing. The
Oblique aerial photographs or a combination of oblique and ver- following is an example of the utility of UAVs for Earth observa-
tical photographs, which are widely used for high-density urban tion and analysis in most advanced manner.
land use mapping, are reviewed by Petrie (2009), and accord- One of the popular fully autonomous UAV eBee (Figure 1.29a)
ing to him, oblique aerial photography have been used since the manufactured by a Swiss sensor manufacturer senseFly (senseFly
2016 by Taylor & Francis Group, LLC
50 Remotely Sensed Data Characterization, Classification, and Accuracies
SA, Cheseaux-Lausanne, Switzerland) can acquire 1.5cm spatial battery, camera, maximum flight time, nominal cruise speed,
resolution aerial images (photos) that can be transformed into 2D radio link range, maximum coverage, wind resistance, GSD,
orthomosaics and 3D models. The eBee can cover up to 12km2 relative orthomosaic/3D model accuracy, absolute vertical/hori-
(4.6 miles2) in a single flight. When it flies over smaller areas at zontal accuracy [with or without GCPs], multidrone operation,
lower altitudes, it can acquire high resolution imagery at 1.5 cm/ automatic 3D flight planning, linear landing accuracy, etc.) about
pixel with overlap. (senseFly, 2014). It uses a built-in 12-megapixel the UAV that can be useful for readers to make decisions to own
cameras, one for visible spectrum (VIS) (R-, G-, and B-bands) and one and use for Earth observation and analysis in very advanced
another for NIR-band image acquisition to create aerial maps for manner.
supporting like wildlife, crops, and traffic management to name The eBees acquired data are processed by senseFlys postflight
a few. Thermal infrared (TIR) cameras can be mounted in the Terra 3D software to produce 2D orthomosaics, 3D point clouds,
UAV for TIR image acquisition. The intuitive eMotion software triangle models, and DEMs. Figure 1.29c is the example of all
of the senseFly makes it easy to plan and simulate your mapping the aerial photos acquired by Sharma and Husley (2014) over
mission, which also generates a full flight path by calculating the university campus, and they mosaicked the acquired images
the UAVs required altitude and projected trajectory (senseFly, using in-house software of the UAV to produce the 2D (Figure
2014). Sharma and Husley (2014) of University of North Georgia 1.29d), stereoscopic analysisbased 3D point cloud (Figure
used eBee as a research and teaching tool. Figure 1.29b shows the 1.29e) and subsequent highly accurate DEM. Sharma and Husley
simulated flight paths of the UAV created by Sharma and Husley (2015) compared the DEM with the DEM created by LiDAR data
(2014) to acquire aerial images at 1 inch spatial resolution for their acquired over the campus (Figure 1.29f). Triangulation between
study. On the site, it is launched by hand, after which it auton- the points together with the onboard GPS camera helped in the
omously follows a flight path along with the simulated image image development with high accuracy (1.54.5cm).
acquisition schedule mapped out in advance by the user, via the The future trend in remote sensing is to have full free-and-open
included eMotion 2 software. Users can take control of the UAV data access, even for higher-resolution data (Belward and Skoien,
at any point and pilot it remotely in real time (senseFly, 2014). 2014). NASA Landsat mission paved the way in providing free
Upon completion of its flight, it can land itself. The eBee web- imagery to the public via the Internet (http://earthexplorer.usgs.
site (https://www.sensefly.com/drones/ebee.html) contains other gov/), and others with commercially driven programs such as
important specifications (weight, wingspan, material, propulsion, Planet Labs and Skybox are following to make their data available
(a) (b) (c)
N
E
W
S
UAV based DEM LiDAR based DEM
(d) (e) (f)
FIg u r e 1.29 Example a unmanned aerial vehicle (UAV) (senseFlys eBee) used for ultrahigh-resolution orthophoto collection and ancillary
derivative data development: (a) eBee UAV, (b) flight path simulated by the Institute of Environmental Spatial Analysis (IESA), (c) all acquired
aerial orthophotos by the UAV over UNG Gainesville Campus, (d) mosaicked 2D orthophoto of the campus (partial), (e) stereoscopic analysis
based 3D point cloud, and (f) comparison of digital elevation models developed by imaging through UAV (eBee) and LiDAR technology. (From
https://www.sensefly.com/drones/ebee.html and Sharma and Husley, 2014.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 51
free to academics and nongovernment organizations, and more 1.7 Summary and conclusion
would follow in the future (Butler, 2014). This approach of free
data sharing is a core element of the Group on Earth Observations This chapter provides an exhaustive overview of the remote
Global Earth Observing System of Systems goal of data sharing sensing satellites, sensors, and their characteristics. First, the
and data management (Withee etal., 2004; GEO, 2012). chapter provides definition and an understanding of remote
sensing from best known sources. This common understand-
1.6 Discussions ing is required for any student and practitioner of remote
sensing. Second, the principles of EMS are enumerated. Third,
In addition to the rich visual information in remote sensing the spatial, spectral, radiometric, and temporal resolution are
data that can be directly interpreted to obtain the geographic explained. Fourth, over 7000 satellites have been launched and
information about landscape features (e.g., the boundary extent operated by various governments and private enterprises from
of forest cover), remote sensing data also provide additional bio- the mid-1950s to the present. The characteristics of many of the
physical information about the object (e.g., forest health, bio- important satellite and sensor systems have been described.
mass). The EMR reflected, emitted, or backscattered from an Fifth, remote sensing is both active and passive and is gathered
object or geographic area is used as a surrogate for the actual in various EMSs such as the visible, NIR, shortwave infrared,
property under investigation (Jensen, 2009). The EME measure- mid-infrared (MIR), TIR, microwave, and radio waves (high-
ments must be calibrated and turned into information using frequency waves). Radar, SAR, LiDAR, SONAR, and SODAR
visual and/or digital image processing techniques. According to technologies operate with microwave and radio wave region
Belward and Skoien (2014), to meet the food, fuel, freshwater, of the EMS, while optical and hyperspectral imaging operates
and fiber requirements of the Earths seven-plus billion humans, with visible to TIR region of EMS. The importance of gather-
obtaining correct land cover information is vital and its impor- ing these data across such wide range of EMS has been high-
tance is enhanced when the world is facing severe consequences lighted. Sixth, remote sensing data gathering is done in various
from global warming and climate change in the form of land modes such as nadir, off nadir, hyperspatial, hyperspectral, and
degradation such as desertification, wildfire, flooding, land- multispectral. All of these have various implications on what we
slides, coastal erosion, and other hazards. study and the level of these accuracies as a result of such acqui-
A major portion of our Earth-orbiting satellites provide a sitions. This aspect has been implied throughout the chapter.
unique vantage point from which to map, measure, and monitor Seventh, remote sensing data are gathered in various platforms:
how, when, and where land resources are changing across the ground based, platform mounted, airborne, spaceborne, under-
globe (Townshend etal., 2008). The importance of remote sens- sea, and UAV. The chapter focuses on spaceborne but does dis-
ing has increased abundantly in the present century due to grow- cuss several other platforms. Reading through these sections,
ing and shifting human population, and therefore, the resulting it becomes clear that remote sensing has truly evolved in both
land use pattern has changed with urbanization and resource sensor design and acquisition platforms. Eighth, the chapter
use for food, fiber, and fuel production (De Castro etal., 2013). provides a window into what to expect in near future through
Additionally, with growing impacts of climate change, newer upcoming newer satellites and sensors. Ninth, the evolution of
approaches to resource conservation aimed at sustainability such microsatellites has been highlighted. Like miniaturization of
as biodiversity offsetting (McKenney and Kiesecker, 2010) and computing technology, satellites and sensors are undergoing
carbon markets (Mollicone et al., 2007) are increasingly using revolutionary technological innovations, which make them
remote sensing data to study and analyze the spatial distribution provide smaller, cheaper, and yet better-quality data. Tenth,
of land use. Land cover mapping using remote sensing can be the chapter captures in a nutshell key developments in remote
implemented at different scales ranging from global (Loveland sensing from 1858 to the present and provides a glimpse on
etal., 1999; Friedl etal., 2002; Bartholome and Belward, 2005; where the future of remote sensing satellites, sensors, and data
Arino etal., 2007; Gong etal., 2013) to more regional and local- are headed. Finally, the authors recommend readers to watch
ized levels (Cihlar, 2000; Panda etal., 2010a,b). the Public Broadcasting Station (PBS) NOVA documentary,
To better plan and balance resource management issues, Earth from Space (http://www.pbs.org/wgbh/nova/earth/earth-
global land cover mapping with remote sensing supports deci- from-space.html) to understand the amazing ability of remote
sion makers and policy analysts at national and international sensing satellites to monitor, measure, and analyze the Earths
scales (Belward and Skoien, 2014). The success of land use man- (oceans, land, and atmosphere) resources for prudent decision
agement plans depends to a large extent on reliable information support to ensure a sustainable life on it.
concerning how, when, and where about resource use and its
changing pattern. A great deal of the reliable information lead- References
ing to land use characterization for ecological habitat assessment
and planning can be accomplished using geospatial technologies 3Dlasermapping. 2014. Volume measuring. http://3dlaser
such as remote sensing and GIS (Osborne etal., 2001; Rao etal., mapping.com/index.php/mining-monitoring-applications/
2005; Rao etal., 2007; Valavanis etal., 2008; Vierling etal., 2008). volume-measuring. Accessed on July 2, 2014.
2016 by Taylor & Francis Group, LLC
52 Remotely Sensed Data Characterization, Classification, and Accuracies
Aggarwal, S. 2003. Principles of remote sensing. In: Satellite Belcher, E. O., Matsuyama, B., and Trimble, G. M. 2001. Object
Remote Sensing and GIS Applications in Agricultural identification with acoustic lenses. In: Proceedings of Oceans
Meteorology, p. 23. 2001 Conference, November 58, Honolulu, HI: Marine
Amatya, D., Panda, S. S., Cheschair, G., Nettles, J., Appleboom,T., Technical Society/IEEE, pp. 611.
and Skaggs, W. 2011. Evaluating evapotranspiration and Belward, A. S. 2012. Europes relations with the wider world
stomatal conductance of matured pine using geospatial A unique view from space. In: Venet, C. and Baranes, B.,
technology. In: American Geophysical Union Conference eds. European Identity through Space; Space Activities and
2011, San Fransisco, CA, December 59. Programmes as a Tool to Reinvigorate the European Identity.
Amatya, D., Trettin, C., Panda, S. S., and Ssegane, H. 2013. Vienna, Austria: Springer, p. 318.
Application of LiDAR data for hydrologic assessments of Belward, A. S. and Skoien, J. O. 2014. Who launched what, when
low-gradient coastal watershed drainage characteristics. and why; trends in global land-cover observation capacity
Journal of Geographic Information System, 5(2), 175191. from civilian earth observation satellites. ISPRS Journal of
doi: 10.4236/jgis.2013.52017. Photogrammetry and Remote Sensing.
American Society of Photogrammetry and Remote Sensing. Beyrich, F. 1997. Mixing height estimation from SODAR data
1952, 1966. Manuals of Photogrammetry. Bethesda, MD: A critical discussion. Atmospheric Environment, 31(23),
ASPRS. 39413953.
Amzajerdian, F., Pierrottet, D., Petway, L., Hines, G., and Roback, Blackinton, J. G., Hussong, D. M., and Kosalos, J. G. 1983. First
V. 2011. LiDAR systems for precision navigation and safe results from a combination side-scan sonar and seafloor
landing on planetary bodies. In: International Symposium on mapping system (SeaMARC II). In: Offshore Technology
Photoelectronic Detection and Imaging 2011, Beijing, China. Conference, Huston, TX.
International Society for Optics and Photonics, p.819202. Blume, H. J. C., Kendall, B. M., and Fedors, J. C. 1978.
Anonymous. 2012. Imaging to drive remote sensing satellite market. Measurement of ocean temperature and salinity via micro-
http://optics.org/news/3/5/43. Accessed on September 1, 2014. wave radiometry. Boundary-Layer Meteorology, 13(14),
Arino, O., Gross, D., Ranera, F., Bourg, L., Leroy, M., Bicheron, 295308.
P., Latham, J. et al. 2007. GlobCover: ESA service for Bowker, D. E., Davis, R. E., Myrick, D. L., Stacy, K., andJones,W.T.
global land cover from MERIS. In: Proceedings of the IEEE 1985. Spectral reflectances of natural targets for use in
International Geoscience and Remote Sensing Symposium, remote sensing studies. NASA reference publication # 1139,
2007 (IGARSS07), Barcelona, Spain, pp. 24122415. http:// Hampton, VA.
dx.doi.org/10.1109/IGARSS.2007.4423328. Brenner, A. C., DiMarzio, J. P., and Zwally, H. J. 2007. Precision
Aronoff, S. 2004. Remote Sensing for GIS Managers. Redlands, CA: and accuracy of satellite radar and laser altimeter data over
ESRI Press. the continental ice sheets. IEEE Transactions on Geoscience
Asner, G. P., Boardman, J., Field, C. B., Knapp, D. E., Kennedy- and Remote Sensing, 45(2), 321331.
Bowdoin, T., Jones, M. O., and Martin, R. E. 2007. Carnegie Briney, A. 2014. An overview of remote sensing. http://geography.
airborne observatory: In-flight fusion of hyperspectral about.com/od/geographictechnology/a/remotesensing.
imaging and waveform light detection and ranging for htm. Accessed on May 2, 2014.
three-dimensional studies of ecosystems. Journal of Applied Brisco, B., Brown, R. J., and Manore, M. J. 1989. Early season crop
Remote Sensing, 1(1), 013536013536. discrimination with combined SAR and TM data. Canadian
Asner, G. P., Powell, G. V., Mascaro, J., Knapp, D. E., Clark, J. K., Journal of Remote Sensing, 15(1): 4454.
Jacobson, J., Kennedy-Bowdoin, T. etal. 2010. High-resolution Brugioni, D. A. and Doyle, F. J. 1997. Arthur C. Lundahl:
forest carbon stocks and emissions in the Amazon. Proceedings Founder of the image exploitation discipline. In:
of the National Academy of Sciences, 107(38), 1673816742. Corona between the Sun and the Earth: The First NRO
Avery, T. E. and Berlin, G. L. 1992. Fundamentals of Remote Reconnaissance Eye in Space, Bethesda, MD: American
Sensing and Airphoto Interpretation. Upper Saddle River, Society for Photogrammetry and Remote Sensing. pp.
NJ: Prentice Hall. 159166.
Bailey, D. T., 2000. Meteorological Monitoring Guidance for Bulstrode, C. J. K., Goode, A. W., and Scott, P. J. 1986. Stereophoto-
Regulatory Modeling Applications. Environmental Protection grammetry for measuring rates of cutaneous healing: A
Agency, Office of Air Quality Planning. EPA document comparison with conventional techniques. Clinical Science,
EPA-454/R-99-005. 71(4), 437443.
Baker, J. C., OConnell, K. M., and Williamson, R., eds. 2001. Butler, D. 2014. Many eyes on earth. Nature, 505, 143144.
Commercial Observation Satellites. Rand Corporation, Santa Butler, L. R. P. and Laqua, K. 1996. Nomenclature, symbols,
Monica, CA, p. 668. units and their usage in spectrochemical analysisIX.
Bartholome, E. M. and Belward, A. S. 2005. GLC.2000. A new Instrumentation for the spectral dispersion and isolation
approach to global land cover mapping from earth observation of optical radiation. Spectrochimica Acta Part B: Atomic
data. International Journal of Remote Sensing, 26(9), 19591977. Spectroscopy, 51(7), 645664.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 53
Canadian Center for Remote Sensing (CCRS). 2014. Defibaugh, Y., Chvez, J., and Tullis, J. A. 2013. Deciduous forest
Fundamentals of remote sensing. https://www.nrcan.gc.ca/ structure estimated with LiDAR-optimized spectral remote
sites/www.nrcan.gc.ca/files/earthsciences/pdf/resource/ sensing. Remote Sensing, 5(1), 155182.
tutor/fundam/pdf/fundamentals_e.pdf. Accessed on July 5, Dobson, M. C. and Ulaby, F. T. 1998. Mapping soil moisture distri-
2014. bution with imaging radar. In: Principles and Applications of
Cash, M. and Panda, S. S. 2012. Urban stream water quality con- Imaging Radar. Manual of Remote Sensing, Vol. 2, New York,
trol with green island provision. In: Presented in Southeast NY: John Wiley, pp. 407433.
Lake and Watershed Management Conference, May 1315, Dong, Y., Forster, B., and Ticehurst, C. 1997. Radar backscatter
2012, Columbus, GA. analysis for urban environments. International Journal of
Center for Remote Imaging, Sensing, and Processing (CRISP). Remote Sensing, 18(6), 13511364.
2014. Electromagnetic waves. http://www.crisp.nus.edu. Dornaika, F. and Hammoudi, K. 2009. Extracting 3D polyhedral
sg/~research/tutorial/em.htm. Accessed on May 2, 2014. building models from aerial images using a featureless and
Chang, C. I. (Ed.). 2003. Hyperspectral imaging: Techniques for direct approach. In: MVA, IAPR Conference on Machine
spectral detection and classification (Vol. 1). Springer Science Vision Applications, May 2022, 2009, Yokohama, JAPAN,
& Business Media: New York, NY. pp. 378381.
Chen, C. M., Hepner, G. F., and Forster, R. R. 2003. Fusion of Dubayah, R. O., Knox, R. G., Hofton, M. A., Blair, J. B., and Drake,
hyperspectral and radar data using the IHS transforma- J. B. 2000. Land surface characterization using lidar remote
tion to enhance urban surface features. ISPRS Journal of sensing. In: M. Hill, and R. Aspinall (Eds.), Spatial infor-
Photogrammetry and Remote Sensing, 58(1), 1930. mation for land use management, Singapore: International
Cihlar, J. 2000. Land cover mapping of large areas from satel- Publishers Direct, pp. 2538.
lites: Status and research priorities. International Journal of Earth Observing System Data and Information System (EOSDIS).
Remote Sensing, 21 (67), 10931114. 2014. Earth Data. https://earthdata.nasa.gov/. Accessed on
Clark, R. N., King, T. V. V., Klejwa, M., and Swayze, G. A. 1990. July 5, 2014.
High spectral resolution spectroscopy of minerals. Journal Edson, C. and Wing, M. G. 2011. Airborne light detection and
of Geophysical Research, 95(B8), 1265312680. ranging (LiDAR) for individual tree stem location, height,
Clark, R. N., Swayze, G. A., Livo, K. E., Kokaly, R. F., King, T. V. and biomass measurements. Remote Sensing, 3, 24942528.
V., Dalton, J. B., Vance, J. S., Rockwell, B. W., Hoefen, T., doi: 10.3390/rs3112494.
and McDougal, R. R. 2002. Surface reflectance calibration Engman, E. T. and Chauhan, N. 1995. Status of microwave soil
of terrestrial imaging spectroscopy data: A tutorial using moisture measurements with remote sensing. Remote
AVIRIS. In: Proceedings of the 10th Airborne Earth Science Sensing of Environment, 51(1), 189198.
Workshop, Pasadena, CA: JPL Publication 02-1. Erdody, T. L. and Moskal, L. M. 2010. Fusion of LiDAR and imag-
Cleve, C., Kelly, M., Kearns, F. R., and Moritz, M. 2008. ery for estimating forest canopy fuels. Remote Sensing of
Classification of the wildlandurban interface: A compari- Environment, 114, 725737.
son of pixel-and object-based classifications using high- Ertberger, C., Jaume, A., and Panda, S. S. 2012. Estimation
resolution aerial photography. Computers, Environment and and evaluation of bacterial loadings of the Upper
Urban Systems, 32(4), 317326. Chattahoochee watershed. In: Presented in Southeast Lake
Colwell, R. N., Ulaby, F. T., Simonett, D. S., Estes, J. E., and Watershed Management Conference, May 1315, 2012,
and Thorley, G. A. 1983. Manual of Remote Sensing. Columbus, GA.
Interpretation and Applications, Vol. 2. American Society ESRI. 2014. GIS Dictionary. http://support.esri.com/en/
of Photogrammetry, Falls Church, VA. 2143pp. knowledgebase/GISDictionary/term/remote%20sensing.
Congalton, R. G. and Green, K. 2008. Assessing the Accuracy Accessed on July 5, 2014.
of Remotely Sensed Data: Principles and Practices. Boca Everaerts, J. 2008. The use of unmanned aerial vehicles (UAVs)
Raton, FL: CRC Press. for remote sensing and mapping. The International
Connor, L. N., Laxon, S. W., Ridout, A. L., Krabill, W. B., and Archives of the Photogrammetry, Remote Sensing and Spatial
McAdoo, D. C. 2009. Comparison of ENVISAT radar and Information Sciences, 37, 11871192.
airborne laser altimeter measurements over Arctic sea ice. Fairfield, N. and Wettergreen, D. 2008. Active localization on
Remote Sensing of Environment, 113(3), 563570. the ocean floor with multibeam sonar. In: Proceedings
Durso, G. and Minacapilli, M. 2006. A semi-empirical approach of the IEEE/MTS OCEANS Conference and Exhibition,
for surface soil water content estimation from radar data pp.110.
without a-priori information on surface roughness. Journal Friedl, M. A., McIver, D. K., Hodges, J. C. F., Zhang, X. Y.,
of Hydrology, 321(1), 297310. Muchoney, D., Strahler, A. H., Woodcock, C. E etal. 2002.
De Castro, P., Adinolfi, F., Capitanio, F., Di Falco, S., Di Mambro, Global land cover mapping from MODIS: Algorithms
A., eds. 2013. The Politics of Land and Food Scarcity. and early results. Remote Sensing of Environment, 83(1),
Abingdon, U.K.: Routledge, p. 154. 287302.
2016 by Taylor & Francis Group, LLC
54 Remotely Sensed Data Characterization, Classification, and Accuracies
Fung, A. K. and Chen, K. S. 2010. Microwave Scattering and Hall, D. K., Fagre, D. B., Klasner, F., Linebaugh, G., and Liston,
Emission Models for Users. Artech House, Norwood, MA. G. E. 2012. Analysis of ERS 1 synthetic aperture radar data
Garca, M., Riao, D., Chuvieco, E., Salas, J., and Danson, F. M. of frozen lakes in northern Montana and implications for
2011. Multispectral and LiDAR data fusion for fuel type climate studies. Journal of Geophysical Research: Oceans
mapping using Support Vector Machine and decision rules. (19782012), 99(C11), 2247322482.
Remote Sensing of Environment, 115(6), 13691379. Hall, S. A., Burke, I. C., Box, D. O., Kaufmann, M. R., and Stoker,
GEO. 2012. The Group on Earth Observation 20122015 Work J. M. 2005. Estimating stand structure using discrete-
Plan, GEO Secretariat. Geneva, Switzerland: WMO, p. 79. return LiDAR: An example from low density, fire prone
http://www.earthobservations.org/. Accessed on July 2, ponderosa pine forests. Forest Ecology and Management,
2014. 208, 189209.
Gernsheim, H. and Gernsheim, A. 1952. Re-discovery of the Hodgson, M. E., Jensen, J. R., Tullis, J. A., Riordan, K. D., and
worlds first photograph. Photographic Journal, 118. Archer, C. M. 2003. Synergistic use of LiDAR and color aer-
Giles, K. A., Laxon, S. W., Wingham, D. J., Wallis, D. W., Krabill, ial photography for mapping urban parcel imperviousness.
W. B., Leuschen, C. J., McAdoo, D. etal. 2007. Combined Photogrammetric Engineering and Remote Sensing, 69(9),
airborne laser and radar altimeter measurements over the 973980.
Fram Strait in May 2002. Remote Sensing of Environment, Huisman, J. A., Hubbard, S. S., Redman, J. D., and Annan, A. P.
111(2), 182194. 2003. Measuring soil water content with ground penetrat-
Glenn, N. F. and Carr, J. R. 2004. The effects of soil moisture on ing radar. Vadose Zone Journal, 2(4), 476491.
synthetic aperture radar delineation of geomorphic sur- ITC. 2014. ITCs database of satellites and sensors. http://www.itc.
faces in the Great Basin, Nevada, USA. Journal of Arid nl/research/products/sensordb/allsensors.aspx. Accessed on
Environments, 56(4), 643657. May 2, 2014.
Goel, M. and Srivastava, H. N. 1990. Monsoon trough boundary Jensen, J. R. 2009. Remote Sensing of the Environment: An Earth
layer experiment (MONTBLEX). Bulletin of the American Resource Perspective 2/e. Pearson Education India, Delhi,
Meteorological Society, 71(11), 15941600. India, pp. 1112.
Goetz, A. F. 2009. Three decades of hyperspectral remote sens- Jha, M. K., Chowdhury, A., Chowdary, V. M., and Peiffer, S. 2007.
ing of the Earth: A personal view. Remote Sensing of Groundwater management and development by integrated
Environment, 113, S5S16. remote sensing and geographic information systems: Prospects
Goetz, S. and Dubayah, R. 2011. Advances in remote sensing technol- and constraints. Water Resources Management, 21(2), 427467.
ogy and implications for measuring and monitoring forest car- Jhang, J., Panigrahi, S., Panda, S. S., and Borhan, M. S. 2002.
bon stocks and change. Carbon Management, 2(3), 231244. Techniques for yield prediction from corn aerial images
Gong, P., Wang, J., Yu, L., Zhao, Y., Zhao, Y., Liang, L., Niu, Z. A neural network approach. International Journal of
etal. 2013. Finer resolution observation and monitoring of Agricultural and Biosystems Engineering, 3(1), 1828.
global land cover: First mapping results with Landsat TM Kasischke, E. S., Melack, J. M., and Craig Dobson, M. 1997. The
and ETM+ data. International Journal of Remote Sensing, use of imaging radars for ecological applicationsA review.
34(7), 26072654. Remote Sensing of Environment, 59(2), 141156.
Gonzalez, R. C. and Woods, R. E. 2002. Digital Image Processing. Kelly, B., Panda, S., Trettin, C., and Amatya, D. 2014. Assessment
Delhi, India: Pearson Education (Singapore) Ltd. of the reach and ecological condition of freshwater tidal
Gowen, A. A., ODonnell, C., Cullen, P. J., Downey, G., and Frias, creeks in the lower coastal plain, Charleston County, South
J. M. 2007. Hyperspectral imagingan emerging process Carolina with advanced geospatial technology application.
analytical tool for food quality and safety control. Trends in In: Poster Presented in South Carolina Water Resources
Food Science and Technology, 18(12), 590598. Conference, October 1516, 2014, Columbia, SC.
Goyer, G. G. and Watson, R. 1963. The laser and its application Kidd, R. B., Simm, R. W., and Searle, R. C. 1985. Sonar acous-
to meteorology. Bulletin of the American Meteorological tic facies and sediment distribution on an area of the deep
Society, 44(9), 564. ocean floor. Marine and Petroleum Geology, 2(3), 210221.
Guthrie, J. and Walsh, K. 1997. Non-invasive assessment of pine- Kim, M. S., Chen, Y. R., and Mehl, P. M. 2001. Hyperspectral
apple and mango fruit quality using near infra-red spec- reflectance and fluorescence imaging system for food
troscopy. Animal Production Science, 37(2), 253263. quality and safety. Transactions of the American Society of
Haack, B. and Bechdol, M. 2000. Integrating multisensor data Agricultural Engineers, 44(3), 721730.
and RADAR texture measures for land cover mapping. Klemas, V. 2011. Remote sensing of coastal plumes and ocean
Computers and Geosciences, 26(4), 411421. fronts: Overview and case study. Journal of Coastal Research,
Hale, J. D., Tamblyn, C., Cash, M., and Panda, S. S. 2012. Fecal 28(1A), 17.
coliform and stream health analysis of Flat Creek with geo- Kokaly, R. F., Clark, R. N., and Livo, K. E. 1998. Mapping the
spatial model development. In: Presented in Southeast Lake biology and mineralogy of Yellowstone National Park
and Watershed Management Conference, May 1315, 2012, using imaging spectroscopy. In: JPL Airborne Earth Science
Columbus, GA. Workshop, Vol. 7, Pasadena, CA, pp. 9721.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 55
Kokaly, R. F., Despain, D. G., Clark, R. N., and Livo, K. E. 2003. Metternicht, G. I. and Zinck, J. A. 2003. Remote sensing of soil
Mapping vegetation in Yellowstone National Park using salinity: Potentials and constraints. Remote Sensing of
spectral feature analysis of AVIRIS data. Remote Sensing of Environment, 85(1), 120.
Environment, 84(3), 437456. Mollicone, D., Achard, F., Federici, S., Eva, H. D., Grassi, G.,
Kruse, F. A., Boardman, J. W., and Huntington, J. F. 2003. Belward, A., Raes, F. etal. 2007. An incentive mechanism
Comparison of airborne hyperspectral data and EO-1 for reducing emissions from conversion of intact and non-
Hyperion for mineral mapping. IEEE Transactions on intact forests. Climatic Change, 83(4), 477493.
Geoscience and Remote Sensing, 41(6), 13881400. Montenbruck, O. and Gill, E. 2000. Satellite Orbits. Springer
Leckie, D. G. and Ranson, K. J. 1998. Forestry applications NewYork, NY.
using imaging radar. In: Principles and Applications Morgan, J. L., Gergel, S. E., and Coops, N. C. 2010. Aerial photog-
of Imaging Radar, John Wiley, New York, NY, Vol. 2, raphy: A rapidly evolving tool for ecological management.
pp.435509. BioScience, 60(1), 4759.
Lee, A. C. and Lucas, R. M. 2007. A LiDAR-derived canopy den- Moursund, R. A., Carlson, T. J., and Peters, R. D. 2003. A fisheries
sity model for tree stem and crown mapping in Australian application of a dual-frequency, identification sonar, acous-
forests. Remote Sensing of Environment, 111, 493518. doi: tic camera. ICES Journal of Marine Science, 60, 678683.
10.1016/j.rse.2007.04.018. Murai, S. 1993. Remote Sensing Note. Tokyo, Japan: JARS.
Lees, B. G. and Ritman, K. 1991. Decision-tree and rule-induction NASA-Phoenix Mars Lander. 2014. http://www.nasa.gov/
approach to integration of remotely sensed and GIS data in mission_pages/phoenix/news/phoenix-20080929.html.
mapping vegetation in disturbed or hilly environments. Accessed on July 2, 2014.
Environmental Management, 15(6), 823831. Nicola, B. M., Beullens, K., Bobelyn, E., Peirs, A., Saeys, W.,
Lefsky, M. A., Cohen, W. B., Parker, G. G., and Harding, D. J. 2002. Theron, K. I., and Lammertyn, J. 2007. Nondestructive
LiDAR Remote Sensing for Ecosystem Studies LiDAR, an measurement of fruit and vegetable quality by means of NIR
emerging remote sensing technology that directly mea- spectroscopy: A review. Postharvest Biology and Technology,
sures the three-dimensional distribution of plant canopies, 46(2), 99118.
can accurately estimate vegetation structural attributes and NSSDC. 2014. The NASA Master Directory held at the NASA
should be of particular interest to forest, landscape, and Space Science Data Center. http://nssdc.gsfc.nasa.gov/nmc/
global ecologists. BioScience, 52(1), 1930. SpacecraftQuery.jsp. Accessed on May 2, 2014.
Lewis, A. J. and Henderson, F. M. 1998. Geomorphic and Osborne, P. E., Alonso, J. C., and Bryant, R. G. 2001. Modelling
hydrologic applications of active microwave remote landscapescale habitat use using GIS and remote sensing:
sensing. In: Principles and Applications of Imaging Radar. A case study with great bustards. Journal of Applied Ecology,
Manual of Remote Sensing, John Wiley, New York, NY, 38(2), 458471.
pp. 567629. Panda, S. S., Ames, D. P., and Panigrahi, S. 2010a. Application
Lillesand, T. M., Kiefer, R. W., and Chipman, J. W. 2004. Remote of vegetation indices for agricultural crop yield prediction
Sensing and Image Interpretation, 5th edn. John Wiley & using neural network. Remote Sensing, 2(3), 673696.
Sons Ltd., New York, NY. Panda, S. S., Andrianasolo, H., Murty, V. V. N., and Nualchawee,
Loveland, T. R., Zhu, Z., Ohlen, D. O., Brown, J. F., Reed, B. C., and K. 2004. Forest management planning for soil conserva-
Yang, L. 1999. An analysis of the IGBP global land-cover tion using satellite images, GIS mapping, and soil erosion
characterization process. Photogrammetric Engineering & modeling. Journal of Environmental Hydrology, 12(13),
Remote Sensing, 65(9), 10211032. 116.
Lu, R. 2003. Detection of bruises on apples using near-infrared Panda, S. S., Burry, K., and Tamblyn, C. 2012a. Wetland change
hyperspectral imaging. Transactions of the American Society and cause recognition in Georgia coastal plain. In: Presented
of Agricultural Engineers, 46(2), 523530. in the International American Society of Agricultural and
Martin, G. 2012. Sumbandilasat beyond Repair, African Defense Biological Engineers (ASABE) Conference 2012, July 29
and Security News Portal Defence web, January 25, 2012. August 2, 2012, Dallas, TX. Paper # 1338205.
http://www.defenceweb.co.za/. Accessed May 3, 2014. Panda, S. S. and Hoogenboom, G. 2013. Blueberry orchard site
McGlone, J. C. 2004. Manual of Photogrammetry, 5th edn. specific crop management with geospatial based yield
Bethesda, MD: ASPRS. modeling. In: Presented in the International American
McKenney, B. A. and Kiesecker, J. M. 2010. Policy development Society of Agricultural and Biological Engineers (ASABE)
for biodiversity offsets: A review of offset frameworks. Conference 2012, July 2124, 2013, Kansas City, MO. Paper
Environmental Management, 45(1), 165176. # 1620890.
Melesse, A. M., Weng, Q., Thenkabail, P., and Senay, G. 2007. Panda, S. S., Hoogenboom, G., and Paz, J. 2009. Distinguishing
Remote Sensing Sensors and Applications in Environmental blueberry bushes from mixed vegetation land-use
Resources Mapping and Modelling. Special Issue of Remote using high resolution satellite imagery and geospatial
Sensing of Natural Resources and the Environment. Sensors techniques. Computers and Electronics in Agriculture,
Journal, 7, 32093241. 67(12), 5159.
2016 by Taylor & Francis Group, LLC
56 Remotely Sensed Data Characterization, Classification, and Accuracies
Panda, S. S., Hoogenboom, G., and Paz, J. 2010b. Remote sensing Rao, M. N., Fan, G., Thomas, J., Cherian, G., Chudiwale, V., and
and geospatial technological applications for site-specific Awawdeh, M. 2007. A web-based GIS decision support sys-
management of fruit and nut crops: A review. Remote tem for managing and planning USDAs Conservation Reserve
Sensing, 2(8), 19731997. Program (CRP). Environmental Modelling and Software, 22,
Panda, S. S., Martin, J., and Hoogenboom, G. 2011a. Blueberry 12701280.
crop growth analysis using climatologic factors and multi- Renslow, M.S. 2012. Manual of Airborne Topographic LiDAR.
temporal remotely sensed imageries. In: Carroll, D., ed. ASPRS publication. ISBN 1-57083-097-5 Stock # 4587.
Published in the Peer Reviewed Proceedings of 2011 Georgia Riao, D., Chuvieco, E., Condes, S., Gonzalez-Matesanz, J., and Ustin,
Water Resources Conference, Athens, GA, April 1113, 2011. S. L. 2004. Generation of crown bulk density for Pinus sylvestris
http://www.gawrc.org/2011proceedings.html. L. from LiDAR. Remote Sensing of Environment, 92, 345352.
Panda, S. S., Nolan, J., Amatya, D., Dalton, K., Jackson, R. M., Riao, D., Chuvieco, E., Ustin, S. L., Salas, J., Rodrguez-Prez, J. R.,
Ssegane, H., and Chescheir, G. 2012b. Stomatal con- Ribeiro, L. M., Viegas, D. X., Moreno, J. M., and Helena, F. 2007.
ductance and leaf area index estimation using remotely Estimation of shrub height for fuel-type mapping combining
sensed information and forest speciation. In: Presented airborne LiDAR and simultaneous color infrared ortho imag-
in the Third International Conference on Forests and Water ing. International Journal of Wildland Fire, 16(3), 341348.
in Changing Environment 2012, September 1820, 2012, Riopel, S. 2000. The Use of RADARSAT-1 Imagery for Lithological
Fukuoka, Japan. and Structural Mapping in the Canadian High Arctic.
Panda, S. S., Steele, D. D., Panigrahi, S., and Ames, D. P. 2011b. Ottawa, Ontario, Canada: University of Ottawa.
Precision water management in corn using automated crop Ritchie, J. C., Zimba, P. V., and Everitt, J. H. 2003. Remote sens-
yield modeling and remotely sensed data. International ing techniques to assess water quality. Photogrammetric
Journal of Remote Sensing Applications, 1(1), 1121. Engineering and Remote Sensing, 69(6), 695704.
Panigrahi, S., Gautam, R., Gu, H., Panda, S. S., Venugopal, Russo, J. C., Amduka, M., Gelfand, B., Pedersen, K., Lethin, R.,
M., and Kizil, U. 2003. Fluorescence imaging for quality and Springer, J. 2006. Enabling cognitive architectures for
assessment of meat. ASAE Paper No. RRV03-0025, St. UAV mission planning. http://www.atl.external.lmco.com/
Joseph, MI. papers/1396.pdf. Accessed on July 5, 2014.
Petrie, G. 2009. Systematic oblique aerial photography using Rylee, J., Panda, S. S., Fitzgerald, J., and Hohnhorst, D. 2012.
multiple digital cameras. Photogrammetric Engineering & Geospatial technology based suitability analysis for new
Remote Sensing, 75(2), 102107. additional reservoirs in Hall County, GA. In: Presented in
Phillips, J., Tamblyn, C., Smith, A., and Panda, S. S. 2012. Impact Southeast Lake and Watershed Management Conference,
of urbanization and point source on changes in water qual- May 1315, 2012, Columbus, GA.
ity in upstream of Upper Chattahoochee River. In: Presented Savenije, B., Geesink, G. H., Van der Palen, J. G. P., and Hemke, G.
in Southeast Lake and Watershed Management Conference, 2006. Prediction of pork quality using visible/near-infrared
May 1315, 2012, Columbus, GA. reflectance spectroscopy. Meat Science, 73(1), 181184.
Powell, R. L. and Matzke, N. 2004. Sources of error in accuracy Schott, J. R. 2007. Remote Sensing: The Image Chain Approach,
assessment of thematic land-cover maps in the Brazilian 2nd edn. Oxford University Press, New York, NY, p. 1.
Amazon. Remote Sensing of Environment, 90(2), 221234. Schowengerdt, R. A. 2007. Remote Sensing: Models and Methods
Rao, M., Awawdeh, M., and Dicks, M. 2005. Spatial allocation for Image Processing, 3rd edn. Academic Press, Chicago,
and environmental benefits: The impacts of the conser- IL, p. 2.
vation reserve program in Texas County Oklahoma. In: senseFly. 2014. eBee: The professional mapping drone. https://www.
Allen, A. W. and Vandever, M. W., eds. The Conservation sensefly.com/drones/ebee.html. Accessed on October 10, 2014.
Reserve ProgramPlanting for the Future: Proceedings of Sharma, J. B. and Hushley, D. 2014. Integrating the UAS in
a National Conference, June 89, 2004, Fort Collins, CO: Undergraduate Teaching and Research-Opportunities and
U.S. Geological Survey, Biological Resources Discipline, Challenges at the University of North Georgia. Proceedings
Scientific Investigations Report 2005-5145, pp. 174182. of the Joint ASPRS Pecora 2014 Conference and the ISPRS
Rao, M. and Miller, E. 2012. Mapping of serpentine soils in Lassen Technical Commision 1 and IAG Commision 4 Meeting in
and Plumas National Forests. CSU Geospatial Review, 10, 6. Denver, CO, Nov 2014.
Rao, M. N. 2012. Mapping snag locations in the blacks mountain Shaw, G. A. and Burke, H. H. K. 2003. Spectral imaging for remote
experimental forest using LiDAR data. Technical Report sensing. Lincoln Laboratory Journal, 14(1), 328.
Submitted to USFS-PSW Research Station, CA, April 19, Shibata, K. 1957. Spectroscopic studies on chlorophyll formation
2012. in intact leaves. Journal of Biochemistry, 44(3), 147173.
Rao, M. N. 2013. Mapping of serpentine soils in the Lassen and Simard, M., Pinto, N., Fisher, J. B., and Baccini, A. 2011.
Plumas National Forest using integrated multispectral and Mapping forest canopy height globally with spaceborne
LiDAR data. Technical Report Submitted to USFS-Mt. LiDAR. Journal of Geophysical Research, 116, G04021. doi:
Hough Ranger District, CA, March 13, 2013. 10.1029/2011JG001708.
2016 by Taylor & Francis Group, LLC
Remote Sensing SystemsPlatforms and Sensors 57
Skarda, R. J., Panda, S. S., and Sharma, J. B. 2011. An assessment Van Diggelen, F. S. T. 2009. A-GPS: Assisted GPS, GNSS, and
of the impact of retention ponds for sediment trapping in SBAS. Artech House Norwood, MA.
the Ada Creek and Longwood Cove using remotely sensed Van Wagtendonk, J. W., Root, R. R., and Key, C. H. 2004. Comparison
data and GIS analysis. In: Published in the Proceedings of AVIRIS and Landsat ETM+ detection capabilities for burn
of International Symposium on Erosion and Landscape severity. Remote Sensing of Environment, 92(3), 397408.
Evolution Hilton Anchorage Hotel, September 1821, 2011, Vierling, K. T., Vierling, L. A., Gould, W. A., Martinuzzi, S., and
Anchorage, AK. ISELE Paper Number 11025. Clawges, R. M. 2008. LiDAR: Shedding new light on habitat
Surez, J. C., Ontiveros, C., Smith, S., and Snape, S. 2005. Use characterization and modeling. Frontiers in Ecology and the
of airborne LiDAR and aerial photography in the estima- Environment, 6(2), 9098.
tion of individual tree heights in forestry. Computers and Vuong, H. and Rao, M. 2013. Using LiDAR to estimate total
Geosciences, 31(2), 253262. aboveground biomass of redwood stands in the Jackson
Swayze, G. A., Clark, R. N., Goetz, A. F. H., Livo, K. E., Breit, G. Demonstration State Forest, Mendocino, California,
N., Kruse, F. A., Stutley, S. J. etal. 2014. Mapping advanced B23B-0557. In: Proceedings of the 2013 Fall Meeting,
argillic alteration at Cuprite, Nevada using imaging spec- December 913, American Geophysical Union (AGU),
troscopy. Economic Geology, 109(5), 11791221. doi: San Francisco, CA.
10.2113/econgeo.109.5.1179. Wagner, W., Bloschl, G., Pampaloni, P., Calvet, J. C., Bizzarri, B.,
Tiffan, K. F., Rondorf, D. W., and Skalicky, J. J. 2004. Imagining fall Wigneron, J. P., and Kerr, Y. 2007. Operational readiness of
Chinook salmon redds in the Columbia River with a dual microwave remote sensing of soil moisture for hydrologic
frequency identification sonar. North American Journal of applications. Nordic Hydrology, 38(1), 120.
Fisheries Management, 24, 14211426. Waring, R. H., Way, J., Hunt, E. R., Morrissey, L., Ranson, K. J.,
Townshend, J. R. and Justice, C. O. 1988. Selecting the spatial Weishampel, J. F., and Franklin, S. E. 1995. Imaging radar
resolution of satellite sensors required for global monitor- for ecosystem studies. BioScience, 45, 715723.
ing of land transformations. International Journal of Remote Wilson, W. J., Yueh, S. H., Dinardo, S. J., Chazanoff, S. L.,
Sensing, 9(2), 187236. Kitiyakara, A., Li, F. K., and Rahmat-Samii, Y. 2001. Passive
Townshend, J. R., Latham, J., Arino, O., Balstad, R., Belward, A., active L-and S-band (PALS) microwave sensor for ocean
Conant, R., Elvidge, C. etal. 2008. Integrated global obser- salinity and soil moisture measurements. IEEE Transactions
vation of the land: An IGOS-P theme, IGOL Report No. 8, on Geoscience and Remote Sensing, 39(5), 10391048.
GTOS 54, FAO, Rome, Italy, p. 74. Withee, G. W., Smith, D. B., and Hales, M. B. 2004. Progress in
Trenear-Harvey, G. S. 2009. Historical Dictionary of Air multilateral Earth observation cooperation: CEOS, IGOS
Intelligence. Scarecrow Press. and the ad hoc group on earth observations. Space Policy,
U.S. Geological Survey. 1992a. The National Aerial Photography 20(1), 3743.
Program (NAPP), factsheet: Reston, VA, U.S. Geological Yang, Z., Rao, M., Elliott, N., Kindler, S., and Popham, T. W.
Survey, p. 1. 2005. Using ground-based multispectral radiometry to
U.S. Geological Survey. 1992b. NHAP and NAPP photographic detect stress in wheat caused by Greenbug (Homoptera:
enlargements, factsheet: Reston, VA, U.S. Geological Aphididae) infestation. Computers and Electronics in
Survey, p. 1. Agriculture, 47, 121135.
Valavanis, V. D., Pierce, G. J., Zuur, A. F., Palialexis, A., Saveliev, Yu, B., Liu, H., Wu, J., Hu, Y., and Zhang, L. 2010. Automated
A., Katara, I., and Wang, J. 2008. Modelling of essential fish derivation of urban building density information using
habitat based on remote sensing, spatial analysis and GIS. airborne LiDAR data and object-based method. Landscape
Hydrobiologia, 612(1), 520. and Urban Planning, 98(3), 210219.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
II
Fundamentals of
Remote Sensing:
Evolution, State of
the Art, and Future
Possibilities
2 Fundamentals of r emote Sensing for Terrestrial Applications: e volution, Current State of the Art, and
Future Possibilities Natascha Oppelt, Rolf Scheiber, Peter Gege, Martin Wegmann, Hannes Taubenboeck,
and Michael Berger ........................................................................................................................................................................ 61
Acronyms and Definitions Introduction Evolution of Terrestrial Applications State of the Art Future
Possibilities Conclusions References
59
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
2
Fundamentals of Remote Sensing
for Terrestrial Applications:
Evolution, Current State of the
Art, and Future Possibilities
Natascha Oppelt
Christian-Albrechts-
Universitt zu Kiel
Rolf Scheiber
German Aerospace Center (DLR) Acronyms and Definitions ...................................................................................................................61
2.1 Introduction ...............................................................................................................................62
Peter Gege
German Aerospace Center (DLR)
2.2 Evolution of Terrestrial Applications .....................................................................................63
From Qualitative Description and Visual Interpretation of the Earth Surface to Digital
Martin Wegmann Data Processing Development of Indices and Quantitative Assessment of Environmental
Ludwig-Maximilians- Parameters SAR Imaging from 2D to 3D and Beyond
Universitt Wrzburg 2.3 State of the Art .......................................................................................................................... 68
Terrestrial Applications Using Multisensor Data Thematic Applications Digital Elevation
Hannes Taubenboeck Models Data Policy Data Availability and Comparability
German Aerospace Center (DLR)
2.4 Future Possibilities ....................................................................................................................76
Enhanced Technology for New Applications Data Availability and Continuity
Michael Berger
ESA secondment at 2.5 Conclusions.................................................................................................................................78
European Commission References...............................................................................................................................................78
Acronyms and Definitions DEM Digital elevation model
D-InSAR Differential interferometry synthetic aperture
ADEOS Advanced Earth Observing Satellite radar
ALS Airborne laser scanning DLR German Aerospace Center
ALOS Advanced Land Observing Satellite DMSP Defense Meteorological Satellite Program
ASAR Advanced SAR DOI US Departments of Interior
ASTER Advanced spaceborne thermal emission and DSM Digital surface model
reflection radiometer DTED Digital terrain elevation data
AVHRR Advanced very-high-resolution DTM Digital terrain model
radiometer EEA European Environment Agency
CBERS China/Brazil Earth Resources Satellite EnMAP EnvironmentalMappingandAnalysis
CEOS Committee on Earth Observation Satellites Program, Germany
CHRIS Compact high-resolution imaging EO Earth observation
spectrometer ERS European Remote Sensing
CIR Color infrared ERTS Earth Resources Technology Satellite
CNES Centre National dEtudes Spatiales ESA European Space Agency
CORINE CoORdination of INformation on the EU European Union
Environment EUMETSAT European Organisation for the Exploitation
DA U.S. Departments of Agriculture of Meteorological Satellites
61
2016 by Taylor & Francis Group, LLC
62 Remotely Sensed Data Characterization, Classification, and Accuracies
fAPAR Fraction of absorbed photosynthetically SRTM Shuttle Radar Topography Mission
active radiation SWIR Short-wave infrared
G8 Group of Eight TIR Thermal or far infrared
GCOS Global Climate Observation System TIROS Television Infrared Observation Satellite
GDEM Global digital elevation model UN United Nations
GEO Group on Earth Observation US United States
GEO-CAPE Geostationary Coastal and Air Pollution USDA United States Department of Agriculture
Events USDOI United States Department of Interior
GEOSS Global Earth Observation System of Systems USGS US Geological Survey
GFW Global Forest Watch VHRR Very-high-resolution radiometer
GIS Geographic information system VIS Visible light
GMES Global Monitoring for Environment and
Security 2.1 introduction
GOCI Geostationary ocean color imager
HICO Hyperspectral imager for the coastal ocean Remote sensing is the acquisition of information about objects
HISUI Hyperspectral imager suite, Japan from a distance. In general, information is acquired by detect-
HRCCD High Resolution Charge Coupled Device ing and measuring an electromagnetic field emitted or reflected
HSE RESURS-P Hyperspectral equipment on the Russian by the objects, but it may also be an acoustic field or gravity/
Resurs-P platform magnetic potential.
HyspIRI Hyperspectral infrared imager Today remote sensing is applied in various fields such as astron-
InSAR Airborne interferometric synthetic aperture omy, medicine, industry, and environmental science; the latter
radar includes atmospheric, oceanic, and terrestrial applicationsthe
IOCCG International Ocean Colour Coordinating subject of our attention. In general, terrestrial remote sensing is
Group related to gravity, magnetic fields, and electromagnetic radiation;
IRMSS InfraRed MultiSpectral Scanner the techniques of the latter cover the electromagnetic spectrum
IRS Indian Remote Sensing from the visible (VIS) to the microwave region. Table2.1 indicates
JPL Jet Propulsion Laboratory the wavelength ranges of different spectral regions of the electro-
LAI Leaf area index magnetic spectrum. There are, however, no clear-cut divisions of
LULC Land use and land cover the spectrum; the partitions have rather developed from different
MERIS Medium-resolution imaging spectrometer application fields and may vary (Campbell and Wynne, 2011).
METI Ministry of Economy, Trade, and Industry Passive remote sensing in all of these regions includes sen-
of Japan sors measuring radiation naturally reflected or emitted from
MIR Mid-infrared the Earth surface, atmosphere, or clouds. The VIS, near-infra-
MODIS Moderate-resolution imaging spectrometer red (NIR), and mid-infrared (MIR) regions correspond to the
MSI Multispectral imager reflective spectral range, because the radiation is essentially
MSS Multispectral scanner system solar radiation reflected from the Earth. In the literature, the
NASA NationalAeronauticsandSpace MIR region may also be referred to as short-wave infrared
Administration (SWIR). The thermal or far-infrared (TIR) consists of wave-
NDVI Normalized difference vegetation index lengths beyond the MIR and extends into regions that border
NIR Near-infrared the microwave region and cover radiation that is emitted by
NOAA NationalOceanicandAtmospheric the Earth; this kind of radiation is often referred to as ther-
Administration mal energy. This self-emitted radiation can be sensed even in
OCM Ocean color monitor the microwave region using passive systems. At this end of the
OLCI Ocean and land color instrument electromagnetic spectrum, active remote sensing is, employing
OLS Operational linescan system an artificial source of radiation, used to analyze the scattering
PALSAR Phased array L-band synthetic aperture radar
Pol-InSAR Polarimetric SAR interferometry TABLe 2 .1 Primary Spectral Regions Used in Terrestrial
PRISMA Hyperspectral Precursor and Application Remote Sensing
Mission, Italy Name Wavelength Range
SAOCOM Satellites for Observation and
VIS 0.380.70m
Communications, Argentina
NIR 0.701.30m
SAR Synthetic aperture radar
MIR, SWIR 1.303m
SeaWiFSS Sea viewing wide field-of-view sensor
TIR 314m
SMAP Soil Moisture Active Passive
Microwave, radar 1mm1 m
SPOT Satellite Pour lObservation de la Terre
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 63
TABLe 2 .2 Subdivisions of Active Microwave Sensory were visual interpretations of prints or transparencies of aerial
Wavelength Wavelength images (Fischer et al., 1976). Traditionally, black-and-white
Band Name Range (cm) Band Name Range (cm) (panchromatic, 400700nm) aerial photography has been com-
mon. With panchromatic images, different patterns of the land
Ka 0.751.18 C 3.757.50
Ku 1.181.67 S 7.5015.0
surface appeared as a series of grey values ranging from black to
K 1.672.40 L 15.030.0
white. In land applications, the development of color infrared
X 2.403.75 P 77107
(CIR) photographs (covering green, red, and NIR wavelengths),
sometimes referred to as false color, may be attributed among
the most significant developments. Colwell (1956) was one of the
properties of the surface. Imaging radars operate within small
first who successfully used CIR images to address crop diseases.
ranges of wavelengths within the broad range indicated in Table
Figure 2.1 demonstrates that besides analysis of shapes, sizes,
2.1; Table 2.2 lists the subdivisions of active microwave regions
pattern, texture, shadow, or spatial relationships, CIR images
(Campbell and Wynne, 2011).
improved interpretation of vegetation due to the coverage of the
The immediate next few chapters discuss the history of terres-
NIR plateau of healthy vegetation (for vegetation spectra, see
trial applications and image processing since the advent of the term
Figure 2.4 later in the chapter). In CIR imagery healthy, dense
remote sensing in the 1960s. Section 2.2 gives an overview of pres-
vegetation appears in bright red. Lighter tones of red gener-
ent terrestrial applications; it is, however, beyond the scope of this
ally represent vegetation with a less distinct NIR plateau such
chapter to provide details for all fields of terrestrial remote sensing,
as mature stands of evergreens. Agricultural fields approaching
but the reader is encouraged to look through other relevant books
the end of growing season, and dead or unhealthy plants often
and chapters of the Remote Sensing Handbook. Therefore, we
appear in less intense reds or green. Soils appear in white, blue-
picked some relevant and representative applications, which will
green, or tan, while water bodies are dark blue to black (USGS,
be discussed briefly. These applications provide good understand-
2001). Image interpretation, however, was limited to visual anal-
ing of how remote sensing data gathered from different portions of
ysis and therefore remained qualitative.
the electromagnetic spectrum is used. The selected examples rep-
In 1957, the former Soviet Union launched the first artificial
resent a broad range of studies at different scales (Section 2.3.1), and
satellite into space: Sputnik, carrying four radio antennas to
applications of certain specific disciplines (Section 2.3.2). These
broadcast radio pulses, triggered the space race and therefore
applications are characteristic for vegetation and Earth sciences,
was the starting signal for the space age (Siddiqi, 2003). NASAs
the hydrosphere, land use and land cover (LULC); moreover, they
(National Aeronautics and Space Administration) Television
cover a variety state of art sensors, including passive reflective mul-
Infrared Observation Satellite, TIROS-1, was launched in 1960
tispectral and hyperspectral (imaging spectrometers), emissive
and provided the first systematic images of the Earth from space
techniques, and active instruments. Multispectral refers to sen-
(Allison and Neil, 1962; see also Table 2.3). A single television
sors, which collect the radiation reflected by the Earths surface in
camera pointed at the Earth surface for a limited time each
a few, predefined and rather broad spectral bands. Hyperspectral
orbit, and mainly collected images of North America. However,
or imaging spectroscopy sensors collect the reflected radiation
TIROS-1 was the first of a series of experimental weather satel-
contiguously across the electromagnetic spectrum in several nar-
lites, which through TIROS-X contained television cameras; four
row wavebands (Campbell and Wynne, 2011). In practice, how-
of them also included IR sensors (Hastings and Emery, 1992).
ever, imaging spectroscopy is more a conceptual term used for
By the late 1960s, scientists in the U.S. Departments of
sensors with more than a few spectral bandsoften, few 10s or
Agriculture (DA) and Interior (DOI) had proposed a space mission
100s but contiguously over a range (e.g., 0.42.5m) of the electro-
dedicated to acquire synoptic, multispectral images of the Earths
magnetic spectrum. After discussing thematic applications, data
surface. These data could be used in a wide range of applications
policy strategies are introduced (Section 2.3.4). Data availability
such as agriculture, forestry, mineral exploration, land resource
and comparability are discussed in Section 2.3.5. Section 2.4 out-
evaluation, land use planning, water resources, mapping and chart-
lines future possibilities including enhanced technologies (Section
ing, and environmental protection. The mission was launched in
2.4.1), data availability, and continuity issues (Section 2.4.2) before
1972 as the Earth Resources Technology Satellite (ERTS), and later
main conclusions are discussed in the last section of this chapter.
renamed Landsat. The launch of the multispectral scanner system
(MSS) onboard the ERTS/Landsat represents the beginning of
2.2 evolution of terrestrial Applications spaceborne terrestrial remote sensing. For the first time an Earth-
orbiting system provided systematic, repetitive observations of the
2.2.1 From Qualitative Description and
Earths land areas. Landsat MSS depicted large areas of the Earth
Visual interpretation of the earth
surface and, although only with four spectral bands and a pixel size
Surface to Digital Data Processing
of 80 m, provided routinely available data (Lillesand etal., 2008).
Aerial photography is the original form of remote sensing; Also in the 1960s, National Oceanic and Atmospheric
throughout its history, proponents of aerial photography have Administration (NOAA) launched the very-high-resolution
sought improved technology to collect information about radiometer (VHRR), which was based on developments
Earths resources. During the 1950s and 1960s, most analyses around the TIROS sensors on board of the NOAA-4 satellite.
2016 by Taylor & Francis Group, LLC
64 Remotely Sensed Data Characterization, Classification, and Accuracies
111030E 11110E 111130E 11120E 111230E 11130E 111330E
48110N 48110N
481030N 481030N
48100N 48100N
48930N 48930N
4890N 4890N
111030E 11110E 111130E 11120E 111230E 11130E 111330E
N
0 500 1000 2000 m
FIg u r e 2.1 CIR aerial image acquired on April 14, 1983. In the East, the image covers the small town Fuerstenfeldbruck (Bavaria, Germany),
a mixed forest stand crosses the image in SWNE direction, surrounded by agricultural fields (bare soil, winter cereals, and pasture). (Geospatial
data Bavarian Surveying and Mapping Authority, Bavaria, Germany, 2014.)
NOAA also introduced the direct reception of digital VHRR The availability of digital data, however, was a seeding point
data free of charge to ground stations installed by users already for the development of image analysis software and computer-
in 1972 (Hastings and Emery, 1992). Being available also in based processing techniques. In 1978, the advanced very-high-
digital form, VHRR and Landsat MSS marked the beginning resolution radiometer (AVHRR) was on board of TIROS-N
of the era of digital image analysis (Campbell and Wynne, (TIROS-NOAA) and the seventh generation of NOAA, which
2011), which, however, was limited to specialized research up to then also carried the VHRR instrument (Hastings and
institutions. The technical standard of personal computers and Emery, 1992). Initially, the NOAA satellites were designed to
limited availability of image analysis software prohibited the observe the Earths weather. Nevertheless, subsequent sen-
analysis of data that now is regarded as commonplace (Jones sors were specifically designed to measure other phenomena
and Vaughan, 2010). including terrestrial vegetation, where the AVHRR-derived
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 65
TABLe 2 .3 Advent of Satellite-Borne Instruments Used for Terrestrial Applications
Revisit Time at
Sensor/Satellite Launch Band Setting Spatial Resolution (m)a Equator (days)b
TIROS-1 1960 4 bands VIS 8000
VHRR/NOAA 1966 1 band VIS, 1 band TIR 900 1
MSS/ETRS (Landsat-1) 1972 2 bands VIS, 3 bands NIR/MIR 68 83 18
AVHRR/TIROS-NOAA 1978 3 bands VIS, 2 bands TIR 1100 3
HRV/SPOT 1986 1 band PAN, 2 bands VIS, 1 band NIR 10 (PAN), 20 (VIS/NIR) 26 (1 with pointing)
LISS/IRS 1988 3 bands VIS, 1 band NIR 72.5 (LISS-I), 36 (LISS-II) 22
ERS-1/2 1991/1995 C-band (5.3 GHz) 30 30 35/1
Radarsat-1/2 1995/2007 C-band (5.4 GHz) 3100 24
AVNIR/ADEOS 1996 1 band PAN, 3 bands VIS, 1 band NIR 8 (PAN), 16 (VISNIR) 41
HRCCD/CBERS 1999 1 band PAN, 3 bands VIS, 1 band NIR 20 26
IRMSS/CBERS 1999 1 band PAN, 2 bands MIR, 1 band TIR 80 (PAN, MIR), 160 (TIR) 26
ASTER/Terra 1999 14 band (VISTIR) 15 (VIS)90 (TIR) 16
MODIS/Terra, Aqua 1999/2002 36 bands (VISTIR) 2501000 12
Hyperion/EO-1 1999 36 bands (VISTIR) 2501000 16
ALI/EO-1 2000 1 band PAN, 4 bands VIS, 5 bands NIR/MIR 10 (PAN), 30 (VISMIR) 16
CHRIS/Proba 2001 up to 63 bands (VISNIR) 17 16 (2 with pointing)
MERIS/Envisat 2002 15 selectable bands (VISNIR) 260 300 (land and coast) 3
1040 1200 (ocean)
ASAR/Envisat 2002 C-band (5.3 GHz) 30 (image mode)150 (wide swath) 35
ALOS 2007 L-band (1.3 GHz) 10100 46
TerraSAR-X/TanDEM-X 2007/2010 X-band (9.6 GHz) 140 11
CosmosSkyMed Constellation 2007/2010 X-band (9.6 GHz) 1100 Up to 1
Sentinel-1a 2014 C-band (5.4 GHz) 5 55 20 12
a With passive, reflective sensors the spatial resolution is provided at nadir.
b For SAR sensors the orbit repeat cycles are indicated instead of revisit time.
Landsat 1
1972
Landsat 2
1975
Landsat 3
1978 Landsat 4
1982
Landsat 5
1984
Landsat 7
1999
LDCM (Landsat 8)
2013
Landsat 9 T80
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
2020
FIg u r e 2 .2 Landsat mission timeline. (Courtesy of NASA Goddard Space Flight Center and U.S. Geological Survey, Greenbelt, MD.)
regional and global indices have proven to be a very robust and (Advanced Earth Observing Satellite [ADEOS]; Shimoda,
useful quantity (Gutman, 1991). Also in the 1970s, the next 1999). This development pushed forward further develop-
two Landsat satellites have been launched (see Figure 2.2). ments in data processing; dedicated software and the rapid
During this time period, other nations decided to operate own advances in computer technology enabled the users to apply
sensors, which led to a rapid increase of data availability; for image processing such as geometric and radiometric correc-
example, in the 1980s, SPOT-1 (Satellite Pour lObservation tion, image enhancement, data classification, or data merging
de la Terre), the first of a series of multispectral satellites, was (Lillesand etal., 2008).
launched by the French Centre National dEtudes Spatiales The launch of the first civil synthetic aperture radar (SAR)
(CNES) in cooperation with Belgium and Sweden (Chevrel in space started a similar rapid development of applications
etal., 1981), and India launched the first sensor from its Indian as with optical imagery. Since that time, a number of satel-
Remote Sensing (IRS) satellite program (Misty, 1998). In the lites have been deployed that carry microwave radiometers
1990s, other nations launched their own Earth Observation and imaging radars (Table 2.3; Elachi, 1988; Kuschel and
(EO) satellites such as China/Brazil (China/Brazil Earth OHagan, 2010; Sullivan, 2000). In spite of data complexity,
Resources Satellite [CBERS]; De Oliveira etal., 2000) or Japan the parallel advances in computer techniques have escalated
2016 by Taylor & Francis Group, LLC
66 Remotely Sensed Data Characterization, Classification, and Accuracies
the number of radar data for terrestrial applications (Jones 2.2.2 Development of indices and Quantitative
and Vaughan, 2010). Prominent examples were the European Assessment of environmental Parameters
Remote Sensing (ERS) satellites, ERS-1 and ERS-2, which
were launched in the 1990s and acquired tandem C-band SAR Computer-based processing of digital data directly led to the
data for many applications for nearly two decades (see also development of quantitative assessment of environmental
Section 2.2.3). parameters, at which the development of indices was an impor-
These data offered the very first opportunity for global cover- tant step. Spectral indices do aim at deriving information about
age every 322days; with two Landsat or ERS sensors in orbit, the land surface by using specific spectral responses. Rouse
the coverage time could be reduced. Nevertheless, true global observed that vegetation shows a typical reflectance behavior in
coverage was limited by a host of issues such as cloud cover, data the VIS and NIR (see also Figure 2.5 later in the chapter) and
downlink limitations, and absence of sufficient ground stations introduced the normalized difference vegetation index (NDVI)
to receive data (Goward et al., 2001). In the 1990s, however, a for MSS data already in 1973 (Rouse, 1974):
series of EO satellites came through with a wide range of spa-
(NIR RED)
tial, spectral, temporal coverage capabilities, which allowed NDVI = (2.1)
(NIR + RED)
for true global coverage. In 1999, Terra was launched, a joint
spaceborne mission of the Ministry of Economy, Trade, and
where
Industry (METI) of Japan and NASA (Running et al., 1998).
NIR reflectance in the NIR (MSS, 0.81.1m)
Terra was the first satellite of a system specifically designed to
RED reflectance in the VIS (MSS, 0.60.7m)
acquire global coverage for monitoring the Earths ecosystems.
Terra carries several sensors such as the advanced spaceborne
Foremost, the NDVI was used to analyze any terrestrial green
thermal emission and reflection radiometer (ASTER) and the
vegetation status and spatiotemporal dynamics of herbaceous
moderate-resolution imaging spectrometer (MODIS) (Xiong
vegetation or forests (Sellers, 1985). The normalization proved
etal., 2011). For global applications, MODIS was a substantial
to be an important advantage for the success of the NDVI as a
improvement over the AVHRR sensor in spatial resolution,
descriptor of vegetation variations in spite of atmospheric effects
band setting, with on-board calibration and enhanced radio-
(Holben et al., 1990). Accordingly, it was used in numerous
metric accuracy (Running etal., 1994). In 2002, the European
regional and global applications for studying vegetation with
Space Agency (ESA) launched ENVISATwith 10 instruments,
various sensors (e.g., Hame etal., 1997; Tucker and Sellers, 1986).
the largest civil EO mission. With the advanced SAR (ASAR)
Although the NDVI suffers from specific deficiencies (e.g., soil
aboard, ENVISAT ensured the data continuity of ERS, but new
background influence), it is probably still the most well-known
instruments supplemented the mission such as two atmospheric
and common index. Nowadays, there are numerous alternative
sensors to monitor trace gases and the medium-resolution
forms of indices that correct for specific deficiencies of the NDVI
imaging spectrometer (MERIS) that operated with 15 bands in
or are adapted to specific sensors such as MODIS (Huete et al.,
the VIS/NIR spectral domain and a spatial resolution of 260 m
2002; see also Section 2.3.1.2). For hyperspectral instruments,
300 m. These systems were designed to support broad-scale
a broad variety of narrow band indices have been developed to
Earth science research and allowed monitoring spatial patterns
assess parameters such as LAI, fractional cover, fAPAR, or to
of environmental and climate changes (Louet and Bruzzi, 1999).
assess and quantify biophysical and biochemical plant parameters
In contrast to the common approach of distributing radiance
such as chlorophyll, accessory pigments, or plant water content
or reflectance images, the broad-scale sensors also opened a new
(Thenkabail et al., 2012). Hyperspectral indices are also com-
era of image-processing techniques, which targeted toward the
monly used for mineral identification (Sabins, 1999; van der Meer
generation of standard data products. A new goal was the pro-
etal., 2012), monitoring of inland water constituents (Odermatt
cessing of consistent, satellite-based records of physical envi-
etal., 2012), and assessment of soil conditions (Pettorelli, 2013).
ronmental parameters (e.g., chlorophyll, leaf area index [LAI]
or fraction of absorbed photosynthetically active radiation
2.2.3 SAR imaging from 2D to 3D and Beyond
[fAPAR]). These products are focused on monitoring tempo-
ral dynamics, and have been successfully integrated into Earth Fostered by the availability of regular SAR observations from
system models to improve the understanding of connections space, using satellite sensors like ENVISAT ASAR (C-band, ESA),
between driver and responsive variables (Pielke, 2005; Sellers Radarsat1 and 2 (C-band, Canada), and ALOS-PALSAR (L-band,
etal., 1997). Japan), interferometric techniques have been developed for sev-
During the first decade of the twenty-first century, inter- eral terrestrial applications, which until then were not supported
net began to influence public access to remote sensing imag- by remote sensing data (Rosen etal., 2000). Primarily dedicated
ery (Goodchild, 2007; Gould et al., 2008; see also Chapters 9 to the generation of digital elevation models (DEMs), culmi-
through11). Nowadays, many image-based products and radi- nating in the Shuttle Radar Topography Mission (SRTM), the
ance or reflectance data are web-enabled and delivered free application range quickly extended toward the measurement of
of cost such as the MODIS (Woodcock etal., 2008) or MERIS displacements with centimeter accuracy using differential inter-
(Bontemps etal., 2011) products. ferometry SAR (D-InSAR; Figure 2.3) techniques. Thus, mapping
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 67
SAR acquisition 1
na
ten
An Topographic
signal
SAR acquisition 2 na component
ten
An
Interferometric signal:
if = topo + atm + subsidence + noise
Atmospheric
delay
htopo
(a) Subsidence
3.14
20
displacement [radian]
Azimuth [km]
0 0.00
20
3.14
30 20 10 0 10 20 30
(b) Ground range [km]
FIg ur e 2.3 Geometry for differential interferometric SAR (D-InSAR) data acquisition (a). Differential interferometric phase signature of
coseismic displacement of Bam earthquake, December 26, 2003, Iran as derived from ASAR data on ENVISAT superimposed on radar reflectivity
(b). One phase/color cycle corresponds to 2.8 cm displacement in radar line of sight. Maximum uplift amounts to 31 cm at [-2, 20] km (geographic
location ~29.05N, 58.36E), maximum subsidence is 17 cm at [-1,5] km.( Data provided by ESA and processed by DLR.)
of co- and postseismic displacements, volcano inflation, and gla- accurate; for example, the measurement of seasonal thermal
cier surface velocities became possible over large areas support- dilation of buildings with accuracies down to few millimeters
ing traditional point-wise GPS measurements. To counteract the was reported. With the availability of spaceborne SAR data
disturbing influence of atmosphere and temporal changes of the with resolutions in the order of 1 m (presently TerraSAR-X
scattering in between acquisitions, D-InSAR approaches quickly [Germany] and CosmoSkyMed [Italy]), the identification of
developed toward the evaluation of time series of SAR images. scattering centers in layover became feasible, allowing the
The common primary objective of all these D-InSAR stack- monitoring of individual building infrastructures (Zhu and
ing approaches is to remove the disturbing signal component of Bamler, 2014). Further applications of the D-InSAR technique
the atmosphere (correlated in space, but random in time), and are for the detection of landslides, measurements of subsid-
of DEM errors (using a geometric model) from the displace- ence due to ground water extraction, mining activities, gas
ment signal (correlated in time and spatially limited). In this and petrol prospection, as well as control of inflation due to
way, D-InSAR measurements became more trustworthy and carbon capture sequestration (Moreira etal., 2013).
2016 by Taylor & Francis Group, LLC
68 Remotely Sensed Data Characterization, Classification, and Accuracies
2.3 State of the Art focuses on a wide range of applications; these include crop yield,
leaf area, and biomass (Doraiswamy etal., 2003; Shanahan etal.,
At present, a large diversity of data sets is available, ranging from 2001; Yang etal., 2000), crop nutrient and water status (Oppelt
very high spatial resolution optical data such as WorldView, and Mauser, 2004; Tilling etal., 2007), weed infestation (Lamb
Ikonos, GeoEye, and Quickbird (commercial sensors operated and Brown, 2001; Thorp and Tian, 2004), plant diseases (Mahlein
by Digital Globe) to Aster, Landsat, SPOT toward medium- etal., 2012), and soil properties such as organic matter, moisture,
resolution MODIS and coarse-scale NOAA AVHRR data; fur- and pH (Christy, 2008) or salinity (Corwin and Lesch, 2003).
thermore, a broad variety of satellite-borne SAR data from To cover the wide range of applications, a variety of sensors
sensors like ALOS, RADARSAT, TerraSAR-X, or Sentinel-1 are and techniques are used. Multispectral and hyperspectral sen-
available. According to their spectral and spatial characteristics, sors are applied as well as thermal imagery. Using multispectral
they support various terrestrial applications (Figure 2.4). and hyperspectral imageries, vegetation analysis generally bases
on the dependence of vegetation spectra on plant pigments and
2.3.1 terrestrial Applications vegetation water content (Blackburn, 2007), mostly by apply-
Using Multisensor Data ing spectral indices (Mulla, 2013). Figure 2.5 depicts how plant
reflectance changes with varying pigment and plant water con-
2.3.1.1 Precision Farming: Local Level tent as modeled with Prospect 5 (Feret etal., 2008).
Precision farming, or information-based management of agri- The most appropriate spatial and spectral resolution for pre-
cultural production systems, emerged in the mid-1980s as a way cision farming depends on crop management, farm equipment,
to manage within and between field spatial and temporal vari- and dominant field sizes. The requirements regarding spatial
abilities associated with all aspects of agricultural production for resolution also differ for the respective application and vary
the purpose of improving crop performance and environmental between 1 m for weed sensing and 10 m for variable rate applica-
quality (Pierce and Nowak, 1999). Apart from field crop produc- tion of fertilizer (Mulla, 2013).
tion, precision farming technologies have been applied success- Satellite instruments such as Ikonos, GeoEye, Quickbird,
fully in viticulture and horticulture, including orchards, and in and WorldView offer high spatial resolution and high revisit
livestock production, as well as pasture and turf management times; combination with (relatively) high geolocation accura-
(Gebbers and Adamchuk, 2010). Due to the need for high spatial cies (Table 2.4) led to a broad range of applications, especially for
resolution, satellite-based remote sensing applications have seen WorldView and GeoEye data. These instruments offer spectral
limited use for quite a long time. Nowadays, remote sensing is a bands most suitable for calculating multiple spectral indices.
key component of precision farming and is used by an increas- Until present, imaging spectroscopy is often confined to airborne
ing number of scientists, engineers, and large-scale farmers and imaging spectrometers; satellite-based hyperspectral instruments
n
Spatial and temporal resolution for selected applications lutio
so
100 re
Geology ral
po
Topography
Forestry tem
10 and
Land-use tial
Land-cover pa Climate
Utility, gs
sin
1 transportation ea
cr
Temporal resolution (years)
infrastructure In
0.1 Crop type,
Precision
agriculture yield
Inland
0.01
waters
Emergency
0.001 response Weather
prediction
0.0001
0.1 m 1m 10 m 100 m 1 km 10 km 100 km
Nominal spatial resolution
FIg u r e 2.4 Terrestrial remote sensing applications. (Modified from a figure originally published by the American Society for Photogrammetry
and Remote Sensing, Bethesda, MD, www.asprs.org; Davis, F. etal., Photogramm. Eng. Remote Sens., 57(6), 689, 1991.)
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 69
Chlab WC layer or man-made structures. Supervised and unsupervised
0.6 30 0.008 classifications of polarimetric SAR data enable, for example, dis-
60 0.008
60 0.024 tinguishing between different crop types (Moreira etal., 2013).
0.4 60 0.032
Reflectance
2.3.1.2 From Space to Species: Assessment of Plant
Species and Monitoring of Biodiversity
0.2 and Habitat Dynamics: Regional Level
Understanding the global distribution of biodiversity has been
0.0
advanced in the past also by means of remote sensing (Gaston,
400 700 1000 1300 1600 1900 2200 2500 2000). Biodiversity, the variety of species, is increasingly affected
Wavelength [nm] by the ongoing global change, and its decrease since decades has
been identified internationally (Balmford et al., 2005; Collen
FIg ur e 2.5 Reflectance signatures of a clover leaf (Trifolium etal., 2013). Monitoring the status and its changes is a high pri-
pratense) with varying contents of chlorophyll a + b (Chlab, in g/cm2) ority for conservation (Myers et al., 2000; Pereira et al., 2013)
and plant water (WC, in g/m2). and can be achieved using remote sensing data (Horning etal.,
2010; Kerr and Ostrovsky, 2003) combined with insitu measure-
are rare. An example may be the compact high-resolution imaging ments and statistical models (Figure 2.6).
spectrometer (CHRIS), launched in 2001 by ESA. CHRIS covers AVHRR time series data has already been used in the mid-
the VIS and NIR with up to 63 spectral bands and a spatial resolu- 1990s for analyzing biodiversity hotspots (Fjelds etal., 1997) and
tion of 17 m at nadir (Barnsley etal., 2004). Using CHRIS, various has been advanced to broad application in environmental man-
studies have been conducted for precision farming applications agement (Duro et al., 2007; Pettorelli et al., 2005; Strand et al.,
(Delegido etal., 2008; Kneubhler etal., 2006; Oppelt, 2010). 2007). Especially the widely applied NDVI has proven to be valu-
In addition to reflectance, plant leaves can emit energy by flu- able to contribute to an explanation of spatial ecological patterns
orescence or thermal emission. The former is used to investigate (Pettorelli etal., 2005). Recent developments of remote sensing for
the photosynthetic activity by solar-induced chlorophyll fluores- biodiversity managed to show statistical correlation of multisensor
cence (Meroni etal., 2009). The latter has been shown to be useful approaches with plant species richness and provide spatial predic-
for the delineation of management zones within fieldsbased on tion models (Buermann etal., 2008; Saatchi etal., 2008). Further
the transpiration of crop canopies related to plant biomassas examples by Vierling etal. (2008) and Nagendra etal. (2013) show
well as for the assessment of plant vigor (Stenzel etal., 2007) and the potential for a variety of environmental applications.
pathogens (Hillnhtter etal., 2011); however, airborne applica-
tions dominate fluorescence measurements and thermography. 2.3.1.3 changing Land Use and Forest
Besides, a new trend toward the use of robust, low-cost, and pref- Monitoring: Global Level
erably, real-time remote sensing systemsunmanned airborne LULC, and alterations of LULC, play a major role in global-
sensorsalso can be observed (Zhang and Kovacs, 2012). scale patterns of the climate and biogeochemistry of the Earth
In recent years also, polarimetric SAR data started to be used for system (Friedl etal., 2002); therefore, monitoring LULC and its
investigating the temporal evolution/phenology of crops as well as changes is one of the most important agents of environmental
to extract soil moisture information. Being in a preoperational change and has significant implications for ecosystem health,
research stadium, reported results are primarily based on airborne water quality, and sustainable land management (Friedl et al.,
SAR data. However, being independent from daylight conditions, 2010). An important goal in developing these products is to meet
operational use starts to emerge based on the polarimetric capa- the needs of the modeling community and to attempt to better
bilities of Radarsat-2, ALOS (Japan), and Risat-1 (India) sensors. understand the role of human impacts on Earth systems through
Related investigations make use of polarimetric decomposition LULC conversions (Hansen etal., 2000). Therefore, many LULC
techniques to distinguish reflections from bare soil, a vegetation classifications are globally available such as the global land cover
TABLe 2 .4 Characteristics of Satellite Instruments Used for Precision Farming,a Geolocation Accuracy is Specified According to the CE90
Standard at Nadir over Flat Terrain
Name Launch Band Setting Spatial Resolution (m) Revisit Time (days) Geolocation Accuracy (m)
IKONOS 1999 PAN, B, G, R, NIR 0.8 (PAN)4 (MS) 5 15
Quickbird 2001 PAN, B, G, R, NIR 0.6 (PAN)2.4 (MS) 13.5 (dep. on latitude) 23
GeoEye 2008 PAN, B, G, R, NIR 0.4 (PAN)1.6 (MS) <3 5
RapidEye 2008 B, G, R, NIR 5.0 1 23.6
WorldView2 2009 PAN, COASTAL, B, G, Y, R, RED EDGE, NIR 0.4 (PAN)3.0 (MS) 1 5
Source: Digital Globe. 2014. Digital Globe satellite information. https://www.digitalglobe.com/resources/satellite-information (accessed March 24, 2014).
a Band setting codes: PAN, panchromatic; MS, multispectral; B, blue; G, green; Y, yellow; R, red; NIR, near-infrared COASTAL, violet/blue; RED EDGE, near
infrared.
2016 by Taylor & Francis Group, LLC
70 Remotely Sensed Data Characterization, Classification, and Accuracies
B C
Random Maxent
forest
High
GAM ...
Probability
Low
FIgu r e 2.6 Workflow of the spatial prediction modeling of species by combining species occurrence points and environmental data sets with
statistical models. Resulting prediction maps (A, B, C) highly depend on chosen input variables and occurrence points as well as selected model.
All models do not consider the biogeography, and hence, areas such as islands might have a high probability of occurrence just based on their
environmental conditions.
products derived from AVHRR data (DeFries etal., 1995, 1998). Being validated independently, there also exist critical differ-
As newer, medium-resolution data sources have emerged, sub- ences (spatial disagreements) between LULC products. For
stantial effort has been made on developing improved charac- GlobCover and MODIS, for example, Fritz etal. (2011) reported
terizations of global land cover. The current generation of global a discrepancy of approximately 20% of the global cropland
land cover products includes the Global Land Cover 2000 prod- area. Differences in classification methodology, training and
uct, produced from SPOT VEGETATION (4 spectral bands in ground reference data, the type of satellite sensors used, and
the VIS/NIR/MIR and a spatial resolution of 1km; Bartholom georeferencing errors may cause these disagreements. A small
and Belward, 2005), the MODIS Land Cover Product (Friedl portion of the differences may also be attributed to differences
et al., 2010), and the GlobCover product produced using data in the date of the satellite data acquisition used in creating land
from MERIS (Arino et al., 2008). Mora et al. (2014) discusses cover maps (Fritz etal., 2011).
present LULC products and future trends in detail. Accuracy assessment and generation of reference data sets
Since LULC maps became increasingly important sources of for accuracy assessments therefore remain difficult. To address
baseline information for a wide variety of applications, valida- these problems, Oloffson et al. (2013) claimed definition of
tion also became increasingly important. LULC maps are com- accuracy standards apart from established quantities such as
monly validated using higher-quality reference data, such as overall accuracy and kappa. They stated the high value of con-
independent validation data sets or regional maps, or they are fusion matrices, which provide valuable information on the
cross validated against field data (Friedl et al., 2002; Hansen magnitude of the classification errors as well as on area esti-
etal., 2013). Global Land Cover 2000, MODIS Land Cover, or mates and provision of confidence intervals for each class. To
MERIS GlobCover are examples for such independently vali- move toward a more efficient use of reference data, Oloffson
dated LULC, which rely on validation based on visual inter- et al. (2012) proposed generation of a new reference data set
pretation of satellite images, regional maps, and geotagged that is created without the use of remote sensing data. An
photographs (Tsendbazar et al., 2014). As a consequence, the improvement in training data coverage or data sharing may
accuracies of the resulting LULC maps vary between 67% and also help to improvefuture global land cover data sets (Fritz
81% (Tsendbazar et al., 2014); in some regions of the world, et al., 2011). Detailed information of methods and reporting
accuracies range between 10% and 50% (Mora et al., 2014). of results may enhance the use of accuracy assessment data
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 71
beyond the present common practice; moreover, it will lead to affected by temporal decorrelation, which otherwise may lead
better estimates of area, accuracy, and the uncertainty associ- to height bias. Having an estimate of forest height, the biomass
ated with these estimates (Oloffson etal., 2013). may be inferred through allometry relations, which are specific
Within the LULC mapping, particular attention has been paid for different tree species. Finally, polarimetric SAR interferom-
to the global observation of forest cover with a special emphasis etry (Pol-InSAR) and SAR tomography offer an increased obser-
on forest loss. Since the highest losses in forest cover have been vational space, allowing for additional information on the forest
documented in tropical and subtropical regions (Hansen et al., structure and thus improved classification possibilities For these
2013), microwave applications play an important role in forest reasons ESA has selected in 2013 BIOMASS, a P-band radar, to
mapping. Not being affected by cloud cover or changing illumina- be implemented as the 7-th Earth Explorer mission. (ESA, 2012;
tion conditions, spaceborne SAR data are best suited for mapping Moreira etal., 2013).
forested areas on a global scale. ERS-1/2 data were used to moni- There exists a variety of initiatives which point towards the
tor deforestation especially in tropical regions and to classify on a importance of remote sensing based forest monitoring such
large-scale boreal forest (Wagner etal., 2003). Identification of the as the United Nations Collaborative Programme on Reducing
response of forested areas to changing environmental conditions, the Emissions from Deforestation and Forest Degradation in
as, for example, water level changes in mangroves, was demon- Developing Countries (REDD+, Angelsen et al. 2012), the Global
strated with L-band data of the PALSAR sensor of the Japanese Forest Resources Assessment (FRA; FAO, 2001) or the Global
ALOS (Lucas et al., 2009). Present day interest is particularly Forest Observation Initiative (GFOI; Mitchell and Hoekman, 2013).
focused on the estimation of above ground forest biomass, due to In the context of global forest monitoring using passive sen-
its impact on climate change. Different evaluation methods may sors, a new initiative was announced in 2014: Global Forest
be applied for obtaining estimates of biomass and its temporal Watch (GFW). The World Resources Institute initiated the proj-
evolution. First, there is a direct relationship of radar backscatter ect together with 40 other partners along with Google to set up a
to above ground biomass. Second, forest height may also be esti- near-real-time monitoring system to provide information for an
mated from the interferometric coherence values. This technique enhanced forest management (Showstack, 2014). GFW represents
can also be applied to small wavelength SAR data, as, for example, the latest iteration of an initiative, which began in 1997, and at that
the ones of the presently ongoing German TanDEM-X mission time was mainly based on reports, static maps, and other materi-
(Krieger et al., 2007), which offers the advantage of not being als not updated frequently (WRI, 2002). The new online tool is
GLOBAL
FOREST FOREST CHANGE FOREST COVER FOREST USE CONSERVATION PEOPLE STORIES
WATCH
FEEDBACK
Global tree cover loss from 20002012
Tree cover loss
FIg u r e 2 .7 Global tree cover loss (20022012). (Courtesy of Global Forest Watch.)
2016 by Taylor & Francis Group, LLC
72 Remotely Sensed Data Characterization, Classification, and Accuracies
based on satellite data, open access to data, cloud computing, and Belward, 2005). Currently, new EO initiatives aim at improving
crowdsourcing (Showstack, 2014). Insupport of this activity, the the geometric level for global urban mapping based on optical
space agencies of the Committee on Earth Observation Satellites (Miyazaki etal., 2012; Pesaresi etal., 2013) or radar data (Esch
(CEOS) established a Space Data Coordination Group (represent- et al., 2012; Gamba and Lisini, 2013) to spatial resolution bet-
ing 12 space agencies) to coordinate the acquisition and provision ter than 100 m; however, to date, a multitemporal global com-
of satellite data (Baltuck etal., 2013). GFW provides standardized ponent is still absent. Many projects such as the CORINE Land
products not only of forest cover, but also of change (gain or loss, Cover (EEA, 2014) produce consistent and thus comparable data
Figure 2.7), use (e.g., logging, oil or wood fiber plantations), and with an urban class at continental scale. Anotherinitiative is the
protected areas online and free of charge (GFW, 2014). European Urban Atlas (Seifert, 2009) that delivers a consistent
and thematically higher-resolution classification of more than
300 cities across Europe. Multitemporal data sets (e.g., Landsat,
2.3.2 thematic Applications SPOT, and DMSP/OLS) have also been applied for macro- to
medium-scale monitoring of urban growth with geometric reso-
EO allows monitoring of patterns, structures, morphology, and
lution up to 10 m (Frolking etal., 2013). Taubenbck etal. (2012)
the relationships of the built environment. In the past decades, the
classified and monitored spatial variations of urbanization
expansion of satellite-derived data products expanded incorpo-
from 1970s to present, taking 27 megacities across the world
rating Earth science products into the mainstream environmental,
(Figure 2.8a and b).
meteorological, and other user communities. This includes move-
On a local scale, active and passive sensor systems are used for
ment of essentially science-produced research products derived
thematically and geometrically highly resolved land cover clas-
from missions into operational observations (Brown etal., 2013).
sifications in 2D (e.g., Blaschke, 2010; Figure 2.8c) and 3D (e.g.,
This results in a vast range of applications where remote sens-
Rottensteiner etal., 2014). Especially multisensorial approaches
ing plays a significant role. Land applications involve mapping,
combining optical satellite data and digital surface models allow
change detection, monitoring, and modeling, and other obser-
derivation of highly detailed 3D city models for characterizing
vations of phenomena at the Earths surface (see other contribu-
urban morphology (Sirmacek etal., 2012; Figure 2.8d).
tions in Remotely Sensed Data Characterization, Classification,
The multiscale capabilities of capturing urban patterns,
and Accuracies). A majority of applications is integrated in a geo-
dynamics, and morphologies have driven researchers toward
graphic information system (GIS) environment (see Chapter 7),
many different geographic fields: Most obvious, studies analyz-
mostly developed to provide thematic maps, which may then be
ing the physical dimension of cities have developed: evaluating
used by specialized groups of scientists, stakeholder, or policy mak-
and comparing spatial urban pattern configurations and their
ers (Barrett and Curtis, 2013). Following are examples for applica-
temporal evolution (Angel et al., 2005) or spatial character-
tions that already are established or mainstream (Section 2.3.2.1)
istics have been suggested to turn qualitative urban concepts
or represent an emerging field of applications (Section 2.3.2.2).
such as megaregions into spatial ones (Taubenbck etal., 2014).
Furthermore, the Remote Sensing Handbook covers a detailed view
Characterizing urban morphology based on EO data has been
on a broad range of land (Land Resources Monitoring, Modeling,
presented for biotope types or particular urban structure types
and Mapping with Remote Sensing) and water (Remote Sensing of
such as Central Business Districts (Taubenbck etal., 2013; Figure
Water Resources, Disasters, and Urban Studies) applications.
2.8d), slums (Baud etal., 2010) or arrangements of green spaces
(Zhou and Wang, 2011). Multitemporal approaches also allow
2.3.2.1 Urban Applications for postdisaster damage assessments of buildings (Wegscheider
Urban applications are a dynamically growing research field etal., 2013). Beyond the capabilities of capturing urban patterns,
in remote sensing (for more detail see Remote Sensing of dynamics, and morphologies, thermal EO data also have been
Water Resources, Disasters, and Urban Studies, Chapter 12). applied for mapping urban climates and gradients of urban heat
Measurements all along the electromagnetic spectrum (optical, islands (Weng, 2009) and its effects toward the rural surround-
radar, thermal, and lidar) enable detection of objects, patterns, ings (Voogt and Oke, 2003).
and conditions of the Earths surface as well as the condition
of the atmosphere. With spatial resolutions capable to capture 2.3.2.2 inland Water
the complex patterns and small objects of urban areas [], the Inland water ecosystems play an essential role for all human life.
question has become Now that we have the technology, how do They serve for water supply, energy production, transport, rec-
we use it? (Weng and Quattrochi, 2006). reation and tourism; maintenance of the hydrological balance;
Naturally, the classification of the built environment at various retention of sediments and nutrients; and provision of habitats
scalesfrom global to local scalesplays a central role in urban for various fauna and flora. Since they are often extensively mod-
remote sensing. Global urban maps relying on optical satellite ified by humans and are among the most threatened ecosystem
data (e.g., MODIS, DMSP/OLS [Defense Meteorological Satellite types of all, sustainable management of freshwater resources
Program/operational linescan system]) are available at compara- has gained importance at regional and global scales (UN, 2002,
tively coarse resolution (Potere and Schneider, 2009) with a max- 2005), and a number of directives regulate their monitoring,
imum spatial resolution of roughly 300 m (e.g.,Bartholom and for example, EU (2000). Although it may appear obvious to use
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 73
28450E 29100E 28570E 2900E 2930E
(a) (c)
41100N
41100N
4100N
4100N
4130N
4130N
40500N
40500N
N
Legend
1975 0 5 10 20 km
28450E 29100E 28570E 2900E 2930E
W - 315
W - 270
28450E 29100E
(d) N - 360
(b)
41100N
41100N
4100N
4100N
E - 45
40500N
40500N
N
Legend
2010 0 5 10 20 km
28450E 29100E E - 90
FIgu r e 2.8 (a) Urban footprint classification of megacity Istanbul for 1975 based on Landsat MSS data; (b) Urban footprint classification of
megacity Istanbul for 2010 based on TerraSAR-X data visualizing spatial urban growth; (c) 2D classification of buildings (red), vegetation (green),
streets (yellow) of urban patterns based on Ikonos data; (d) 3D city model based on a Cartosat-1 digital surface modelthe Central Business
District of Levent, Istanbul.
satellites or airplanes to get a synoptic picture of entire ecosys- tasks. Better suited are multispectral sensors designed for ocean
tems, routine monitoring of inland waters is still based on tradi- color and imaging spectrometers. A number of lake-related studies
tional water sampling and does not make use of remote sensing. were made with both sensor types, but past and current ocean color
Many studies demonstrated the potential of remote sensing sensors like SeaWiFSS (Sea Viewing Wide Field-of-View Sensor,
(overviews in Dekker etal., 1995; Lindell etal., 1999; Papoutsa NASA), OCM (Ocean Color Monitor, Japan), MODIS, MERIS, or
and Hadjimitsis, 2013), but operational services such as known GOCI (Geostationary Ocean Color Imager, Korea) (IOCCG, 2010)
for marine and coastal sectors are still missing. provide spatial resolutions (0.31km) too low for the majority of
Remote sensing of lake water quality from satellite began with inland waters, while the current spaceborne imaging spectrometers
Landsat MSS (Lindell etal., 1999). The first water quality maps were with pixel sizes below 100 m (Hyperion (USGS), CHRIS (ESA),
based on very general concepts (trophic class), but chlorophyll-a was HJ-1A (China), HICO (hyperspectral imager for the Coastal Ocean,
already mapped in 1974 (Strong, 1974). The high spatial resolution NASA), HSE RESURS-P (hyperspectral equipment on the Russian
of multispectral sensors designed for land applications allows map- Resurs-P platform) (Staenz et al., 2013) are experimental instru-
ping of many inland water systems including shallow areas. For this ments and do not provide data on a regular basis.
reason, Landsat MSS, which had a pixelsize of 60 m, and its suc- The parameters detectable in inland waters include concen-
cessors like Landsat TM and ETM+, but also SPOT, IRS, Ikonos, trations of phytoplankton as an indicator of the trophic state,
Quickbird, and the German commercial RapidEye sensors, with dissolved organic components and suspended matter as tracers
pixel sizes down to 2.5 m, were used for inland water studies (Dekker for inflow of pollutants, optical properties like turbidity, attenu-
etal., 1995; Lindell etal., 1999). However, their coarse spectral resolu- ation, Secchi depth and euphotic depth, and the surface tem-
tion and low radiometric dynamics poses strong limitations to water perature (Lindell etal., 1999). In shallow areas, additionally, the
applications (Lindell etal., 1999). The lack of spectral and radiomet- bottom type can be classified and water depth determined. With
ric details restricts data analysis to empirical approaches (Matthews, the exception of chlorophyll-a, for which empirical algorithms
2011), which require calibration against field data and cannot be using two or three bands in the red and NIR provide robust esti-
transferred to other regions and seasons. This impracticality for auto- mates in turbid waters (Gitelson etal., 2011), reliable data analysis
mated data analysis makes them unattractive for routine monitoring requires algorithms that are based on radiative transfer models.
2016 by Taylor & Francis Group, LLC
74 Remotely Sensed Data Characterization, Classification, and Accuracies
For operational ocean color sensors, such algorithms have been sensors using laser scanning and radar (interferometric SAR)
developed and are used to produce maps of water constituents on principles, providing users with a multitude of products with
a routine basis. However, since they are designed to process global different horizontal and vertical resolutions, accuracies, and
oceanic data sets, these generic procedures often fail in coastal spatial coverage. The following terms are distinguished:
and inland waters (Cui etal., 2010; Huth and Gege, 2004).
Digital surface model (DSM): Height information of the
The only way to determine water quality parameters reliably,
scenery envelope including man-made features and veg-
that is, with known uncertainties, is to model the underlying
etation. Usually, all remote sensing systems (laser and
physical processes using the optical properties of all involved
radar) provide this type of data.
components. A number of models and inversion techniques have
Digital terrain model (DTM): Height information of the
been developed over the years, but no one can be considered
bare Earth surface without man-made features and veg-
optimal for all situations (Odermatt etal., 2012). Major problems
etation. In case of InSAR-derived DSMs, this product is
are the ambiguity of the reflectance spectrum (Defoin-Platel and
usually computed using specific value adding techniques.
Chami, 2007); for example, different combinations of water con-
For airborne laser scanning (ALS) DSMs and DTMs are
stituents can lead to the same spectrum, and the large variabil-
generated simultaneously from the set of raw data.
ity of type and composition of water constituents and bottom
substrates. Because of this, accurate results require site-specific Regional scale DEMs mostly are products of lidar and airborne
information. Although many algorithms were developed in the InSAR measurements. ALS is based on lidar principles (light
last decades (Dekker etal., 2011), the first software tools that can detection and ranging). The travel distance of short laser pulses
be tuned by the user were published only recently (Gege, 2014; to objects on ground is measured and evaluated. Different scan-
Giardino etal., 2012). ning principles are possible, and the aircraft position attitude
So far, not only the complex optics of inland waters prohibits is precisely considered to infer the 3D location of the reflection
their routine quality control from satellite, but also the prob- from ground. Due to their exceptional resolution in the decime-
lems of atmosphere and sun glint correction are not yet solved. ter range, the applications of these DEMs are in areas of flood
Compared to the atmosphere, the water body reflects little radia- and coastline protection, mapping of open cast mining activi-
tion and therefore appears dark; from satellite, only 5%10% of ties, archeology, and line routing planning, usually being lim-
the measured signal describes the water body (IOCCG, 2010). An ited to regional scales.
inland water-specific problem is the adjacency effect, where scat- In airborne interferometric synthetic aperture radar (InSAR)
tering in the atmosphere contaminates water pixels with radia- systems, the reflected radar signal is received in a side-looking
tion from the surrounding land. The range of this effect can be in geometry. The data are afterward combined interferometrically
the order of kilometers, and reliable correction is not yet possi- to derive the topography information providing swath width
ble. For inland waters also, the standard algorithm for correcting of several kilometers, thus being more economic during data
the unavoidable reflections at the water surface (Cox and Munk, acquisition compared to ALS. In consequence, during the last
1954) is not applicable, because the underlying statistical rela- decade, companies like Intermap (United States/Germany) and
tionship requires undisturbed wind fields for many kilometers. OrbiSAT (Brazil) were able to generate countrywide DEM prod-
Due to the described challenges, remote sensing of inland ucts, with vertical accuracies in the 12 m range and horizontal
waters is still in its infancy. However, the upcoming new genera- posting of approximately 5 m. Besides these commercial service
tion of hyperspectral satellite sensors with pixel sizes of 30 m providers, airborne InSAR-derived DEMs are being generated
(e.g., HyspIRI (hyperspectral infrared imager, NASA), PRISMA occasionally on request by several national research organi-
(Hyperspectral Precursor and Application Mission, Italy), HISUI zations using their own sensors, especially if the application
(hyperspectral imager suite, Japan), EnMAP (Environmental requires up-to-date topography information.
Mapping and Analysis Program, Germany) (Staenz etal., 2013), There has also been a tremendous development for generat-
and modern multispectral sensors like Sentinel-2 and -3 (ESA) ing global scale DEMs, however, at reduced height resolution
will foster the development of robust algorithms. Synergetic and posting. On a global scale, the most prominent DEMs
usage of sensors with complementary properties, for example, originate from SRTM, ASTER-GDEM, and the TanDEM-X
with high spectral and geometric resolution, bears the poten- mission. The SRTM mission, a joint initiative of American
tial to solve some of the problems and to extend the range of and German space agencies NASA and the German Aerospace
applications. Center (DLR), was the first system fulfilling the requirements
for a global, homogeneous, and reliable DEM with the DTED-2
specification (digital terrain elevation data, level 2 with 30m post-
2.3.3 Digital elevation Models
ing, 12m height accuracy) at least for latitudes up to approx. 60
DEMs are of fundamental importance for a broad range of north and 58south, with polar regions not being accessible due
commercial and scientific applications in areas like hydrology, to the shuttle orbit. While NASA used a C-band radar to record
glaciology, forestry, geology, and urban planning. Nowadays, the entire accessible land surface of the Earth with an height pre-
high-quality DEMs are being derived mostly by means of remote cision of 10 m, DLR used an X-band radar to cover a smaller
sensing from a variety of data acquired by air- and spaceborne swath, but with a precision of 6 m. For more than one decade,
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 75
researchers worldwide use these data sets, as they were made commercial marketing rights for the WorldDEM, which is the
accessible (at least partially) free of cost. adaptation of the TanDEM-X elevation model to the needs of
A second type of global DEM has been computed from commercial users worldwide.
ASTER stereopair images. The improved global digital elevation
model (GDEM V2) was released in 2011 (Fujisada etal., 2012); its
coverage spans from 83 north latitude to 83 south, encompass- 2.3.4 Data Policy
ing 99% of Earths landmass including polar regions. The spatial Remote sensing data policy underwent an alternating history.
resolution is similar to one of the SRTM data set. Although ver- The availability of remote sensing data strongly depends on
sion 2 shows significant improvements over the previous release, national distribution policies and pricing. In general there coex-
users are advised that the data contains anomalies and artifacts ist three models of data distribution policy:
that will impede effectiveness for use in certain applications. An
attempt for improving the quality of this free-of-charge data set 1. Web-enabled, free: Organizations that provide data and
has been made by Intermap that offers the NextMap World30 associated products freely available for every interested
product based on the height values of ASTER GDEM V2, but user (e.g., NASAs/USGSs Landsat data or MODIS prod-
with reduced artifacts. ucts, Brazils CBERS data);
A new worldwide DEM, from pole to pole, with an accuracy 2. Commercial, at a cost: Commercial remote sensing
of a new dimension, has been acquired from 2010 to 2013 by the industry, which mainly is located in the very high spatial
TanDEM-X mission (Krieger etal., 2007). Flying only few hundred resolution domain, began with the launch of Ikonos and
meters apart, the two twin satellites, TerraSAR-X and TanDEM-X Quickbird (see also Section 2.3.1.1). Most commercial sys-
(TerraSAR-X add-on for Digital Elevation Measurement), form tems, however, operate with meter or submeter resolutions
the first configurable SAR interferometer in space. The derived (e.g., WorldView, GeoEye, Pleiades [France]) and allow
DSMs are characterized by a 12 m posting and vertical accuracy of the detection of small-scale objects, such as elements of
2 (relative) and 10 m (absolute, see also Figure 2.9 and Table 2.5). residential housing, commercial buildings, transportation
For scientific use, DLR is offering preliminary sample data systems, and other utilities, which can assist industrial
products (intermediate DEMs) within announcements of oppor- geospatial applications (Navulur etal., 2013), but may also
tunity, while Airbus Defence and Space holds the exclusive be purchased by noncommercial users or scientist.
SRTM TanDEMX
50.94 50.94
50.92 50.92
Latitude [deg]
Latitude [deg]
50.90 50.90
50.88 50.88
50.86 50.86
6.40 6.45 6.50 6.55 6.60 6.40 6.45 6.50 6.55 6.60
(a) Longitude [] (b) Longitude []
FIg u r e 2.9 Example for DEM quality and timeliness for an open cast coal mine in Germany. Shaded relief representation of (a) SRTM on 30 m
grid, 6 m height accuracy, recorded in 2000; (b) TanDEM-X on 12 m grid, 2 m height accuracy, recorded in 2011. ( DLR.)
TABLe 2 .5 Available DEMs, Corresponding Resolutions and Accuracies; Accuracies Are Indicated According to the EC90 Standard
DEM Spatial Resolution Absolute Vertical Accuracy Relative, Point-To-Point Vertical Accuracy
ALS <1 m 1 m <1 m <10cm
Airborne InSAR DEMs (e.g., NextMap) 5m5m <2 m <1 m
ASTER GDEM V2 and World30 30 m 30 m <20 m <17 m
SRTM DTED-1 90 m 90 m (3 arcsec) <30 m <20 m
SRTM DTED-2 30 m 30 m (1 arcsec) <18 m <12 m
TanDEM-X DEM; WorldDEM 12 m 12 m (0.4 arcsec) <10 m <2 m
2016 by Taylor & Francis Group, LLC
76 Remotely Sensed Data Characterization, Classification, and Accuracies
3. Databuy: Publicprivate partnerships support the scien- (Nativi etal., 2013). In 2014, it includes 97 national governments
tific community with cost-free data, but also seek to estab- including the European Commission as members; the numbers
lish commercial EO markets, that is, assist in developing of 87 partizipating organizations includes intergovernmental,
remote sensing business in a way that follow-ups may be international, and regional organizations such as space agencies
financed by industry using the profit from selling data or and United nation bodies (GEO, 2015; Nativi etal., 2013). GEOSS
products to nonscientific clients. Examples for public is conceived as a global network of product providers aiming
private partnerships are the RapidEye fleet as well as the provision of decision-support tools to a great variety of stake-
TerraSAR and TanDEM-X missions. The SPOT sensors holder. Focused community portals have developed to meet the
are a successful example for a publicprivate partner- particular needs of individual communities and to develop solu-
ship, which led to a commercial funding and operation of tions targeted to these specialized users (GEO, 2010). As GEO
SPOT-6 by Airbus (launch in 2014). approaches the end of its first 10-year (2005-2015) mandate, the
ministers from GEO member governments and leaders from
GEO participating organizations decided to extend their politi-
2.3.5 Data Availability and comparability
cal commitment to the GEO vision through 2025. Discussions
In the past two decades, the availability of remote sensing data on the implementation plan for GEOSS are ongoing (GEO,
has been growing enormously, while the real costs decreased. In 2015). This kind of coordinated development of methodologies,
parallel, the costs and efforts to gather/download, process, and processing, and validation may lead to a new era of generating
analyze remote sensing data have decreased. The development standardized data products.
of personal computers and image-processing software was tre- SAR data of the last two decades are available within the
mendous, and at present, sophisticated processing is possible archives of several space agencies. With respect to radar, the
on single computers or laptops. The use of remote sensing data, ESA-owned ERS-1/2 (19912000/19952011) as well ENVISAT
however, not only depends on its availability, but also on its prize. (20022012) data, which are available on request and usually
Since Landsat data are freely available, scientific investigations delivered free of charge for scientific applications, whereas for
and applications have spurred a rapid increase (Wulder et al., ALOS PALSAR (20062011) and Radarsat-2 data (since 2007)
2012); similar experiences have been made for MODIS (Masuoka commercial distribution is favored. Free-of-charge data for
etal., 2011) and MERIS (Curran and Steele, 2005) products. supporting scientific applications of the worldwide EO com-
At present, a diversity of data sets is available, ranging from munity are further available within announcements of opportu-
very high to coarse-scale resolution data. Converting these data nity of the high-resolution SAR satellites (e.g., TerraSAR-X and
into valuable products for terrestrial application is a challenge TanDEM-X, Radarsat-2, and CosmoSkyMed).
and has already been aimed at by Bartholom and Belward
(2005). However, the definition of products for terrestrial appli-
cations is challenging and needs to be improved. Prominent 2.4 Future Possibilities
examples are global products such as LULC classifications,
2.4.1 enhanced technology
which provide a good global representation of general land cover
for new Applications
classes but are neither easily comparable nor congruent to other
land cover classifications (Herold etal., 2008; McCallum etal., Various new sensors including new technologies have been
2006; Section 2.3.1.3). Questions of integrity and reusability of recently launched or are planned: the U.S. National Space
global LULC maps therefore will have to be further discussed. Policy of 2010 stated that a primary goal is the improvement
In response to this need, international bodies such as Group on of spaceborne EO to conduct science, monitor climate and
Earth Observation (GEO), Global Climate Observation System global change, manage natural resources, and to support disas-
(GCOS), and EO communities in Europe have been involved to ter response and recovery; moreover, it included the need of
establish an operational and continuous LULC observing sys- long-term observations (USA, 2010). The latest generation of
tem that includes integration, harmonization, and validation of NASAs flagship, Landsat 8, has been launched in early 2013 and
data sets (Mora et al., 2014). Existing global LULC and forest provides the very valuable continuity of the Landsat missions.
initiatives serve as pioneers for future initiatives providing cloud Other missions such as the Soil Moisture Active Passive (SMAP,
computing and crowd funding. The commitment of these many United States) or the FLuorescence EXplorer (FLEX, ESA) are
agents including space agencies and industrial partners to sup- also highly relevant for environmental monitoring. The SMAP
porting an ongoing program such as GFW was unprecedented has been launched in early 2015 and carries a SAR (L-band, 1.2-
(Baltuck etal., 2013). 1.4 GHz) and a radiometer, aiming at soil moisture measure-
GEO also promotes free and open data within its scope to ment for water and energy cycling analysis. The antenna covers
build a Global Earth Observation System of Systems (GEOSS). a 1000km swath, providing global coverage within 3days at the
GEO has been launched as a response to the World Summit on equator and 2days at boreal latitude (NASA EOSPO, 2014).
Sustainable Development and the Group of Eight (G8) leading FLEX is a candidate for ESAs eighth Earth Explorer and
industrialized countries, which called for action concerning aims to provide global maps of vegetation fluorescence, specifi-
international collaboration for exploiting the potential of EO cally designed for the estimation of vegetation fluorescence on a
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 77
global scale. The FLORIS (Fluorescence Imaging Spectrometer) next decade for replacing the present generation of TerraSAR-X
instrument onboard FLEX will measure the radiance between and CosmoSkyMed sensors.
500 and 800 nm with a bandwidth between 0.1 nm and 2 nm, The pair of Sentinel-2 satellites (2015/2016) will routinely pro-
providing images with a 150 km swath and 300 m pixel size. vide medium spatial resolution optical images over land surfaces
This information will allow a detailed monitoring of vegetation globally. Sentinel-2 will ensure continuity of SPOT type obser-
dynamics, by improving the methods for the estimation of clas- vation with improvements in terms of spectral resolution and
sical biophysical parameters (Moreno and Moran, 2014). temporal coverage. The multispectral imager (MSI) instrument
Another future U.S. mission suitable for land applications aboard Sentinel-2 covers the VIS, NIR, and SWIR in 13 spectral
is the coarse-scale hyperspectral GEO-CAPE (Geostationary bands with a spatial resolution of 10 m 10 m (continuity of
Coastal and Air Pollution Events), which is planned to be in orbit SPOT observation), 6 bands of 20 m 20 m, and with 3 bands
in 2020. GEO-CAPE may be an improved MODIS successor and of 60 m 60 m dedicated to atmospheric corrections and cloud
will provide data for a variety of land applications such as water screening. The increased swath width of about 290km together
resources, agriculture, disasters monitoring, and ecological fore- with having two satellites in orbit at the same time will enable
casting (NASA, 2014). Another hyperspectral, but medium-scale a revisit time of 5days at equator and 23days at midlatitudes.
(60 m at nadir) instrument may be the hyperspectral infrared This will allow the monitoring of rapid changes such as vegeta-
imager (HyspIRI), which at present is at the study stage. HyspIRI tion characteristics during the growing season and improved
includes two instruments: an imaging spectrometer measuring change detection techniques.
from the VIS to TIR with 10nm contiguous bands and a multi- Sentinel-3 is primarily a mission to support services relating
spectral imager measuring from 3 to 12m in SWIR and TIR; to the marine environment, and the first Sentinel-3 satellite is
its data might be used for a wide variety of studies primarily in expected to launch in 2015/2016. The system, however, will also
the carbon cycle and ecosystem and earth surface (JPL, 2014). carry the ocean and land color instrument (OLCI), which is based
To support the highly needed high-resolution environmen- on heritage from the MERIS instrument but with improved
tal monitoring, the European Union (EU) recently imple- wavelength bands (21 compared to 15 on MERIS) with a spatial
mented the Copernicus program (former, Global Monitoring resolution of 300 m 300 m (full resolution mode).
for Environment and Security [GMES]). Copernicus is the National space missions will trump international space
European flagship for EO, tailored to the monitoring of the envi- programs; for example, the German EnMAP (launch in 2018)
ronment and security relevant information; therefore, it will will set a milestone for spaceborne imaging spectroscopy with
provide global information in an operational manner tailored Landsat-like spatial resolution. With 244 bands in the VIS
to the needs of a broad range of applications. The Copernicus and SWIR and a, by pointing increased, revisit time of 4days,
comprises two types of EO missions: the Sentinel families EnMAP will provide highly resolved time-series to decipher
(Sentinel-1 to 6) and contributing missions. There are around the response of ecosystems on natural and man-made pressure
30 existing or planned contributing missions from ESA or their and to push forward multidisciplinary applications on climate
member states, EUMETSAT, and other third party (European change, LULC changes, hazards, and environmental pollution
and international) mission operators, which make part of their (Kaufmann etal., 2012).
data available for Copernicus (Berger and Aschbacher, 2012).
The EO capabilities provided through the Copernicus system
2.4.2 Data Availability and continuity
will cause a paradigm shift in EO with the robustness and
reliability of a long-term operational infrastructure fostering As already depicted, data continuity and availability are two
research and innovation in the field of EO following a similar important decisions of international data policies, which will
development as introduced for meteorological mission and ser- improve EO data for a wide range of applications (Lynch etal.,
vices some years back. 2013). To contribute to standardized measures especially for
With the launch of Sentinel-1A in 2014, new opportunities global monitoring, continuous information of our planet and its
arise in the microwave domain, offering repeated global acqui- availability are of high importance (Pereira etal., 2013). In this
sitions with a repeat cycle of 12days, which will be reduced to context, the Landsat (see also Figure 2.2), SPOT as well as ESAs
6days with the launch of Sentinel-1B. Due to the free and open C-band SAR missions already have set standards.
data policy, these data are expected to boost the application range Copernicus has a free and open data policy, following the
of microwave techniques such as D-InSAR. Further SAR data GEO data principle, which is expected to further boost research
for these applications are provided in L-band by ALOS-2 (Japan, and innovation. To further increase the use of remote sensing
launched in 2014) and will be supplemented by the two upcom- data for new applications, the capability to develop standard-
ing SAOCOM (Satellites for Observation and Communications, ized products based on automatic thematic processors for
Argentina, with expected launches in 2016 and 2017), followed consistent knowledge generation and provision may be a pre-
by the Radarsat Constellation Mission (C-band, Canada, launch requisite. Furthermore, the foreseeable increase in data volume
in 2018). Since the commercial market is demanding high-reso- will induce new challenges in terms of data storage, handling,
lution data, being usually acquired at X-band, new satellites are mining, processing techniques, and accuracy requirements.
likely to be launched as private or national initiatives within the Investment in processing is still comparatively high due to
2016 by Taylor & Francis Group, LLC
78 Remotely Sensed Data Characterization, Classification, and Accuracies
mostly not full automation of information generation proce- References
dures due to, for example, different atmospheric conditions, land
cover types, different users requirements, or the algorithms still Allison, L.J. and Neil, E.A. 1962. Final report on the TIROS 1
being in experimental status. Therefore, robust methodologies meteorological satellite system. NASA technical report
are needed. R-131, Goddard Space Flight Center, Greenbelt, MD. 20771.
As it becomes more and more understood and accepted Angel, S., Sheppard, S., Civco, D. L., and Buckley, R. 2005. The
that single disciplines are decreasingly able to progress indi- dynamics of global urban expansion. Department of
vidually, this paradigm shift may be observed increasingly. Transport and Urban Development, The World Bank:
Multidisciplinary approaches using EO-based data in combi- Washington, DC.
nation with other data sources are developing with the idea for Angelsen, A., Brockhaus, M., Sunderlin, W.D. and Verchot, L.V.
geographic value-adding to these spatial data beyond the sole (eds.). 2012. Analysing REDD+: Challenges and choices.
physical measurements. Besides multidisciplinary approaches, CIFOR, Bogor (Indonesia).
multisensor applications will form a further spreading technique Arino, O., Bicheron, P., Achard, F., Latham, J., and Witt, R. 2008.
to enhance existing approaches, which requires temporally over- The most detailed portrait of Earth. ESA Bulletin, 136, 2531.
lapping and well-calibrated sensors. One example may be the Balmford, A., Bennun, L., ten Brink, B., Cooper, D., Ct, I. M.,
synergistic use of Landsat-8 and Sentinel-2 data, which paves Crane, P., Dobson, A. et al. 2005. The convention on bio-
the way to capture near-daily data (Wulder etal., 2012) and fur- logical diversitys 2010 target. Science, 307, 212213.
ther ensures continuity of data if one sensor fails. International Baltuck, M., Briggs, S., Loyche-Wilkie, M., McGee, A., Muchoney,
cooperation and data sharing will also reduce instrument and D., and Skvseth, P. E. 2013. The Global Forest Observation
launch costs and mission redundancy (Durrieu and Nelson, Initiative: Fostering the use of satellite data in forest man-
2013). Joint commitments between national and international agement, reporting and verification. Carbon Management,
space agencies may therefore be a powerful means of stabilizing 4(1), 1721.
new space initiatives, and hence data availability and continuity Barnsley, M. J., Settle, J. J., Cutter, M. A., Lobb, D. R., and Teston,
(Morel, 2013). F. 2004. The CHRIS/Proba Misison: A low cost Smallsat for
hyperspectral, multi-angle, observations of the earth sur-
2.5 conclusions face and atmosphere. IEEE Transactions on Geoscience and
Remote Sensing, 42(7), 15121520.
From the early beginning of satellite-based remote sensing in Barrett, E. C. and Curtis, L. F. 2013. Introduction to Environmental
the 1970s, the use of remote sensing for terrestrial applications Remote Sensing. New York: Taylor & Francis.
has increased tremendously. Manifold applications from field Bartholom, E. and Belward, A. S. 2005. GLC2000: A new
to global scales have been established since then. The needs approach to global land cover mapping from Earth obser-
and uses of the terrestrial EO community, however, are divers, vation data. International Journal of Remote Sensing, 26(9),
with various user communities, sensor types, and data needs. 19591977.
Different systems are therefore needed, which depend on the Baud, I., Kuffer, M., Pfeffer, K., Sliuzas, R., and Karuppannan,
required information. Redundant developments and the repli- S. 2010. Understanding heterogeneity in metropolitan
cation of existing EO programs have often been perceived as a India: The added value of remote sensing data for analyz-
waste of resources, but may also be seen as potential reservoir ing sub-standard residential areas. International Journal
for innovations. The synergy between missions and the poten- of Applied Earth Observation and Geoinformation, 12(5),
tial to capitalize upon sensors with different focus and output 359374.
is high, as could have been experienced with Landsat, MODIS, Berger, M. and Aschbacher, J. 2012. The sentinel missionsNew
and MERIS. Moreover, multidisciplinary studies that use the opportunities for science. Remote Sensing of Environment,
increased data density to investigate temporal dynamics on var- 120, 1276.
ious scales point toward future development of well-calibrated, Blackburn, A. 2007. Hyperspectral remote sensing of plant pig-
global products. ments. Journal of Experimental Botany, 58(4), 855867.
Besides affordability and sustainability, long-term scenarios Blaschke, T. 2010. Object based image analysis for remote sens-
may aim to satisfy the priorities of continuity, frequency of ing. ISPRS Journal of Photogrammetry and Remote Sensing,
observations, (inter-)calibration of sensors, and evolution of the 65(1), 216.
system:continuity of operational observations drives the nature Bontemps, S., Defourny, P., Bogaert, E. V., Arino, O., Kalogirou,
of the observations and the type of missions; the frequency of V., and Perez, J. R. 2011. GLOBCOVER 2009. Products
observations drives the number of simultaneous sensors in description and validation report, UCV Louvain and ESA,
orbit for each class of observations; (inter-)calibration of sensors Louvain, Belgium.
drives data continuity, and the evolution of the sensors based Brown, M. E., Escobar, V. M., Aschbacher, J., Milagro-Perez, M. P.,
on user needs, which accounts for the demand for new services Doorn, B., Macauley, M. K., and Friedl, L. 2013. Policy for
and products, demand for new observations and technology robust space-based earth science, technology and applica-
improvements. tions. Space Politics, 29(1), 7682.
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 79
Buermann, W., Saatchi, T., Smith, T. B., Zutta, B. R., Chaves, J. mapping techniques in Australian and Caribbean coastal
A., Mil, B., and Graham, C. H. 2008. Predicting species environments. Limnology and Oceanography: Methods, 9,
distributions across the Amazonian and Andean regions 396425.
using remote sensing data. Journal of Biogeography, 35(7), Delegido, J., Fernandez, G., Gandia, S., and Moreno, J. 2008.
11601176. Retrieval of chlorophyll and LAI of crops using hyper-
Campbell, J. B., and Wynne, R. H. 2011. Introduction to Remote spectral techniques: Application to CHRIS/PROBA data.
Sensing. New York: The Guilford Press. International Journal of Remote Sensing, 29(24), 71077127.
Chevrel, M., Courtois, M., and Weill, G. 1981. The SPOT satel- De Oliveira, L. C., Lima, M. G. R., and Hubscher, G. L. 2000.
lite remote sensing mission. Photogrammetric Engineering CBERSAn international space cooperation program.
& Remote Sensing, 47, 11631171. Acta Astronautica, 47, 559564.
Christy, C. D. 2008. Real-time measurement of soil attributes Digital Globe. 2014. Digital Globe satellite information. https://
using on-the-go near infrared reflectance spectroscopy. www.digitalglobe.com/resources/satellite-information
Computers and Electronics in Agriculture, 61(1), 1019. (accessed March 24, 2014).
Collen, B., Pettorelli, N., Baillie, J. E. M., and Durant, S. 2013. Doraiswamy, P. C., Moulin, S., Cook, P. W., and Stern, A. 2003. Crop
Biodiversity Monitoring & Conservation: Bridging the Gap yield assessment from remote sensing. Photogrammetric
between Global Commitment and Local Action. Cambridge, Engineering & Remote Sensing, 69(6), 665674.
U.K.: Wiley-Blackwell. Duro, D., Coops, N. C., Wulder, M. A., and Han, T. 2007.
Colwell, R. 1956. Determining the prevalence of certain cereal Development of a large area biodiversity monitoring driven by
crop diseases by means of aerial photography. Hilgardia, 26, remote sensing. Progress in Physical Geography, 31, 235260.
223286. Durrieu, S. and Nelson, R. F. 2013. Earth observation from
Corwin, D. L. and Lesch, A. M. 2003. Application of soil con- spaceThe issue of environmental sustainability. Space
ductivity to precision agriculture. Agronomy Journal, 95(3), Politics, 29(4), 238250.
455471. EEA (European Environment Agency). 2014. CORINE Land
Cox, C. and Munk, W. 1954. Statistics of the sea surface derived Cover. http://www.eea.europa.eu/publications/COR0-land
from sun glitter. Journal of Marine Research, 13, 198227. cover (accessed March 27, 2014).
Cui, T., Zhang, J., Groom, S., Sun, L., Smyth, T., and Elachi, C. 1988. Spaceborne Radar Remote Sensing: Applications
Sathyendranath,S. 2010. Validation of MERIS ocean-color and Techniques. New York: IEEE Press.
products in the Bohai Sea: A case study for turbid coastal ESA (European Space Agency). 2012. Report for Mission Selection:
waters. Remote Sensing of Environment, 114, 23262336. Biomass. ESA SP-1324/1 (3). Noordwijk, the Netherlands: ESA.
Curran, P. J. and Steele, C. M. 2005. MERIS: The re-branding of Esch, T., Taubenbck, H., Roth, A., Heldens, M., Felbier, A., and
an ocean sensor. International Journal of Remote Sensing, Thiel, M. 2012. TanDEM-X mission: New perspectives for
26(9), 17811798. the inventory and monitoring of global settlement patterns.
Davis, F., Quattrochi, D., Ridd, M., Lam, N. Walsh, S., and Michaelson, International Journal of Applied Earth Observation and
J. 1991. Environmental analysis using integrated GIS and Geoinformation, 6(1), 061702, doi: 10.1117/1.JRS.6.061702.
remotely sensed data: Some research needs and priorities. EU (European Union). 2000. European Water Framework
Photogrammetric Engineering & Remote Sensing, 57(6), 689697. Directive (Directive 2000.60/EC) of the European parlia-
Defoin-Platel, M. and Chami, M. 2007. How ambigu- ment and of the Council establishing a framework for the
ous is the inverse problem of ocean color in coastal Community action in the field of water policy. Official
waters? Journal of Geophysical Research, 112, C03004, Journal L327 of the EU. http://eur-lex.europa.eu/legal-
doi:10.1029/2006JC003847. content/EN/TXT/?uri=CELEX:02000L0060-20090625
DeFries, R., Hansen, M., and Townshend, J. 1995. Global dis- (accessed March 3, 2014).
crimination of land cover types from metrics derived from FAO (Food and agriculture organization of the United nations).
AVHRR pathfinder data. Remote Sensing of Environment, 2001. Global Forest Resources Assessment 2000: Main
54(3), 209222. report. FAO, Rome.
DeFries, R., Hansen, M., and Townshend, J. 1998. Global land cover Feret, J. B., Francois, C., Asner, G. P., Gitelsen, A. A., Martin, R.E.,
classifications at 8km spatial resolution: The use of training Bidel, L. P., Ustin, S. L., le Maire, G., and Jaquemoud, S.
data derived from Landsat imagery in decision tree classes. 2008. Prospect 4 and 5: Advances in leaf optical properties
International Journal of Remote Sensing, 19(16), 31414168. model separating photosynthetic pigments. Remote Sensing
Dekker, A. G., Malthus, T. J., and Hoogenboom, H. J. 1995. The of Environment, 112, 30303043.
remote sensing of inland water quality. In Danson, F. M. and Fischer, W. A., Hemphill, W. R., and Kover, A. 1976. Progress in
Plummer, S. E. (eds.). Advances in Environmental Remote remote sensing. Photogrammetria, 32, 3372.
Sensing. New York: Wiley & Sons Ltd. Fjelds, J., Ehrlich, D., Lambin, E., and Prins, E. 1997. Are bio-
Dekker, A. G., Phinn, S. R., Anstee, J., Bissett, P., Brando, V. diversity hotspots correlated with current ecoclimatic
E., Casey, B., Fearns, P. et al. 2011. Intercomparison of stability? A pilot study using the NOAA-AVHRR remote
shallow water bathymetry, hydro-optics, and benthos sensing data. Biodiversity and Conservation, 6, 401422.
2016 by Taylor & Francis Group, LLC
80 Remotely Sensed Data Characterization, Classification, and Accuracies
Friedl, M. A., McIver, D. K., Hodges, J. C., Zang, X. Y., the Vespucci initiative for the advancement of geographic
Muchoney,D., Strahler, A. H., Woodcock, C. E. etal. 2002. information science. International Journal of Spatial Data
Global land cover mapping from MODIS: Algorithms and Infrastructures Research, 3, 146167.
early results. Remote Sensing of Environment, 83(12), Goward, S. N., Masek, J. G., Williams, D. L., Irons, J. R., and
287302. Thompson, R. J. 2001. The Landsat 7 mission: Terrestrial
Friedl, M. A., Sulla-Menashe, D., Tan, B., and Schneider, A. 2010. research and applications for the 21st century. Remote
MODIS Collection 5 global land cover: Algorithms refine- Sensing of Environment, 78, 312.
ments and characterization of new data sets. Remote Sensing Gutman, G. G. 1991. Vegetation indices from AVHRR: An upgrade
of Environment, 114(1), 168182. and future prospects. Remote Sensing of Environment, 35,
Fritz, S., See, L., McCallum, I., Schill, C., Obersteiner, M., van 121136.
der Velde, M., and Achard, F. 2011. Highlighting continued Hame, T., Salli, K., Andersson, K., and Lohi, A. 1997. A new meth-
uncertainty in global land cover maps for the user com- odology for the estimation of biomass of conifer-dominated
munity. Environmental Research Letters, 6(4), 044005, doi: boreal forest using NOAA AVHRR data. International
10.1088/1748-9326/6/4/044005. Journal of Remote Sensing, 18(15), 32113243.
Frolking, S., Milliman, T., Seto, K., and Friedl, M. A. 2013. A Hansen, M. C., de Fries, R. S., and Townshed, J. R. 2000. Global
global fingerprint of macro-scale changes in urban struc- land cover classification at 1km spatial resolution using a
ture from 1999 to 2009. Environmental Research Letters, 8, classification tree approach. International Journal of Remote
024004, doi: 10.1088/1748-9326/8/2/024004. Sensing, 21(67), 13311364.
Fujisada, H., Urai, M., and Iwasaki, A. 2012. Technical meth- Hansen, M. C., Potapov, P. V., Moore, R., Hancher, M.,
odology for ASTER global DEM. IEEE Transactions on Turubanova, S. A., Tyukavina, A., Thau, S. etal. 2013. High-
Geosciences and Remote Sensing, 50(1), 37253736. resolution global maps of 21st century forest cover change.
Gamba, P. and Lisini, G. 2013. Fast and efficient urban extent Science, 342, 850853.
extraction using ASAR wide swath mode data. IEEE Journal Hastings, D. A. and Emery, W. 1992. The advanced very high
of Selected Topics in Applied Earth Observations and Remote resolution radiometer (AVHRR)A brief reference guide.
Sensing, 6(5), 21842195. Photogrammetric Engineering & Remote Sensing, 58(8),
Gaston, K. J. 2000. Global patterns in biodiversity. Nature, 405, 11831188.
220227. Herold, M., Mayaux, P., Woodcock, C. E., Baccini, A., and
Gebbers, R. and Adamchuk, V. L. 2010. Precision agriculture for Schmullius, C. 2008. Some challenges in global land cover
food security. Science, 327(5967), 828831. mapping: An assessment of agreement and accuracy in
Gege, P. 2014. WASI-2D: A software tool for regionally optimized existing 1 km datasets. Remote Sensing of Environment,
analysis of imaging spectrometer data from deep and shal- 112(5), 25382556.
low waters. Computers & Geosciences, 62, 08215. Hillnhtter, C., Mahlein, A.-K., Sikora, R. A., and Oerke E. C.
GEO (Group on Earth Observations). 2010. Report on Progress 2011. Remote sensing to detect plant stress induced by
Beijing Ministerial SummitObserve, Share, Inform. Heterodera schachtii and Rhizoctonia solani in sugar beet
Geneva, Switzerland: GEO Secretariat. fields. Field Crop Research, 122, 7077.
GEO (Group on Earth Observations). 2015. The GEO Website. Holben, B. N., Kaufman, Y. J., and Kenall, J. D. 1990. NOAA-
https://www.earthobservations.org (accessed March 30, 11 AVHRR visible and near-IR inflight calibration.
2015). International Journal of Remote Sensing, 11(8), 15111519.
GFW (Global Forest Watch). 2014. The Global Forest Watch Horning, N., Robinson, J. A., Sterling, E. J., Turner, W., and Spector,
Website. http://www.globalforestwatch.org/ (accessed April S. 2010. Remote Sensing for Ecology and Conservation. New
7, 2014). York: Oxford University Press.
Giardino, C., Candiani, G., Bresciani, M., Lee, Z. P., Gagliano, S., Huete, A., Didan, D., Miura, T., Rodriguez, E. P., and Gao, X.
and Pepe, M. 2012. BOMBER: A tool for estimating water 2002. Overview of the radiometric and biophysical per-
quality and bottom properties from remote sensing images. formance of MODIS vegetation indices. Remote Sensing of
Computers & Geosciences, 45, 313318. Environment, 83(12), 195213.
Gitelson, A. A., Gurlin, D., Moses, W. J., and Yacobi, Y. Z. 2011. Huth, J. and Gege, P. 2004. Suspended matter dynamics in Lake
Remote estimation of chlorophyll-a concentration in inland, Constance in 2003 derived from MERIS and MODIS sat-
estuarine, and coastal waters. In Weng, Q. (ed.). Advances ellite data. Proceedings of ENVISAT & ERS Symposium,
in Environmental Remote Sensing: Sensors, Algorithms and Salzburg, Austria, September 610. ESA Special publica-
Applications. Boca Raton, FL: CRC Press. tion, SP-572 (CD-ROM).
Goodchild, M. F. 2007. Citizens as sensors: The world of volun- IOCCG (International Ocean Colour Coordinating Group).
teered geography. GeoJournal, 69(4), 211221. 2010. Atmospheric correction for remotely-sensed
Gould, M., Craglia, M., Goodchild, M. F., Annoni, A., Camara,G., ocean-colour products. In Wang, M. (ed.). Reports of
Kuhn, K., Mark, D., Masser, I., Liang, S., and Parsons, E. the IOCCG, No. 10. Dartmouth, Nova Scotia, Canada:
2008. Next Generation Digital Earth: A position paper from IOCCG.
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 81
Jones, H. G. and Vaughan, R. A. 2010. Remote Sensing of Land Remote Sensing and Global Environmental Change,
Vegetation: Principles, Techniques, and Applications. New Remote Sensing and Digital Image Processing Series,
York: Oxford University Press. Vol. 11. Springer, New York.
JPL (Jet Propulsion Laboratory, California Institute of Matthews, M. W. 2011. A current review of empirical proce-
Technology). 2014. HyspIRI mission study http://hyspiri. dures of remote sensing in inland and near-coastal transi-
jpl.nasa.gov/ (accessed April 4, 2014). tional waters. International Journal of Remote Sensing, 32,
Kaufmann, H., Frster, S., Wulf, H., Segl, K., Guanter, L., Bochow, 68556899.
M., Heiden, U. etal. 2012. Science Plan of the Environmental McCallum, I., Obersteiner, M., Nilsson, S., and Shvidenko,A
Mapping and Analysis Program (EnMAP). Potsdam, 2006. A spatial comparison of four satellite derived
Germany: German Research Center for Geosciences, 1 km global land cover datasets. International Journal
Scientific Technical Report. of Applied Earth Observation and Geoinformation, 8(4),
Kerr, J. T. and Ostrovsky, M. 2003. From space to species: 246255.
Ecological applications for remote sensing. Trends in Meroni, M., Rossini, M., Guanter, L., Alonso, L., Rascher, U.,
Ecology & Evolution, 18(6), 299305. Colombo, R., and Moreno, J. 2009. Remote sensing of
Kneubhler, M., Koetz, B., Huber, S., Schpfer, J., Richter, R., solar-induced chlorophyll fluorescence: Review of meth-
and Itten, K. 2006. Monitoring vegetation growth using ods and applications. Remote Sensing of Environment, 113,
multitemporal CHRIS/PROBA data. IEEE Geoscience and 20372051.
Remote Sensing Symposium (IGARSS 2006), IEEE, Denver, Misty, D. 1998. Indias emerging space program. Pacific Affairs,
CO, pp. 26772680. 71, 151174.
Krieger, G., Moreira, A., Fiedler, H., Hajnsek, I., Werner, M., Mitchell, A. and Hoekman, D.H. Global Forest Observation
Younis, M., and Zink, M. 2007. TanDEM-X: A satellite Initiative (GFOI) - Review of priority research and develop-
formation for high resolution SAR interferometry. IEEE ment topics. GEO, Geneva (Switzerland), 159 p.
Transactions on Geosciences and Remote Sensing, 45(11), Miyazaki, H., Shao, X., Iwao, K., and Shibasaki, R. 2012. An auto-
33173341. mated method for global urban area mapping by integrat-
Kuschel, H. and OHagan, D. 2010. Passive radar from his- ing ASTER satellite images and GIS data. IEEE Journal of
tory to future. Proceedings of the 11th International Radar Selected Topics in Applied Earth Observations and Remote
Symposium, June 1618, Vilnuis, Lithuania, pp. 14. Sensing, 6, 10041019.
Lamb, D. W. and Brown, R. B. 2001. Precision agriculture: Mora, B., Tsendbazar, N. E., Herold, M., and Arino, O. 2014.
Remote sensing and mapping of weeds in crops. Journal of Global land cover mapping: Current status and future
Agricultural Engineering Research, 78, 117125. trends. Land Use Land Cover: Remote Sensing and Digital
Lillesand, T., Kiefer, R. W., and Chipma, J. 2008. Remote Sensing Image Processing, 18, 1130.
and Image Interpretation. New York: Wiley & Sons. Moreira, A., Prats-Iraola, P., Yonis, M., Krieger, G., Hajnsek, I.,
Lindell, T., Pierson, D., Premazzi, G., and Zilioli, E. (eds.). 1999. and Papathanassiou, K. P. 2013. A tutorial on synthetic
Manual for Monitoring European Lakes Using Remote aperture radar. IEEE Geoscience and Remote Sensing
Sensing Techniques. EUR 18665 EN, Joint Research Centre, Magazine, 43pp.
Ispra, Italy. Morel, P. 2013. Advancing Earth observation from space: A global
Louet, J. and Bruzzi, S. 1999. ENVISAT mission and system. challenge. Space Policy, 29, 175180.
Geoscience and Remote Sensing Symposium. IGARSS99 Moreno, J. and Moran, S. 2014. Vegetation stress from soil mois-
Proc 3, pp. 16801682. ture and chlorophyll fluorescence: Synergy between SMAP
Lucas, R. M., Bunting, P., Clewley, D., Proisy, C., Filho, P. W. M., and FLEX approaches. Proceedings of the EGU General
Viergever, K., Woodhouse, I. et al. 2009. Characterisation Assembly, 27 April - 2 May, 2014, Vienna, Austria.
and monitoring of mangroves using ALOS PALSAR data. Mulla, D. J. 2013. Twenty five years of remote sensing in precision
JAXA Kyoto & Carbon Initiative, Phase 1 report. JAXA/ agriculture: Key advances and remaining knowledge gaps.
Meti, Japan. Biosystems Engineering, 114, 385371.
Lynch, M., Maslin, H., Balzter, H., and Sweeting, M. 2013. Myers, N., Mittermeier, R. A., Mittermeier, C. G., da Fonseca,G.A.B.,
Sustainability: Choose satellites to monitor deforestation. and Kent, J. 2000. Biodiversity hotspots for conservation pri-
Nature, 496, 293294. orities. Nature, 403, 853858.
Mahlein, A. K., Oerke, E. C., Steiner, U., and Dehne, W. 2012. Nagendra, H., Lucas, R., Honrado, J. P., Jongman, R. H. G.,
Recent advances in sensing plant diseases for precision Tarantino, C., Adamo, M., and Mairota, P. 2013. Remote
crop protection. European Journal of Plant Pathology, 133, sensing for conservation monitoring: Assessing protected
197209. areas, habitat extent, habitat condition, species diversity,
Masuoka, E., Roy, D., Wolfe, R., Morisette, J., Sinno, S., Teague,M., and threats. Ecological Indicators, 33, 4559.
Saleous, N., Devadiga, S., Justice, C. O., and Nickeson, J. NASA (National Aeronautics and Space Administration). 2014.
2011. MODIS Land data products: Generation, quality GEO-Cape website. http://geo-cape.larc.nasa.gov/ocean-
assurance and validation. In Ramachandran et al. (eds.). instrumentdesign.html (accessed April 4, 2014).
2016 by Taylor & Francis Group, LLC
82 Remotely Sensed Data Characterization, Classification, and Accuracies
NASA EOSPSO (NASAs Earth Observing System Project Science Pierce, F. J. and Nowak, P. 1999. Aspects of precision agriculture.
Office). 2014. SMAP mission. http://eospso.nasa.gov/missions/ Advances in Agronomy, 67, 185.
soil-moisture-active-passive (accessed April 4, 2014). Potere, D. and Schneider, A. 2009. Comparison of global urban
Nativi, S., Mazzetti, P., Craglia, M., and Pirrone, N. 2013. The maps. In: Gamba, P. and M. Herold (eds.). Global Mapping
GEOSS solution for enabling data interoperability and of Human Settlements: Experiences, Data Sets, and Prospects.
integrative research. Environmental Science and Pollution CRC Press, Boca Raton, FL, pp. 269308.
Research, 21, 41774192. Rosen, P., Hensley, S., Joughin, I. R., Li, F. K., Madsen, S. N.,
Navulur, K., Pacifici, F., and Baugh, B. 2013. Trends in optical Rodriguez, E., and Goldstein, R. M. 2000. Synthetic aperture
commercial remote sensing industry. IEEE Geoscience and radar interferometry. Proceedings of the IEEE, 88, 333382.
Remote Sensing Magazine, December, 5764. Rottensteiner, F., Sohn, G., Gerke, M., Wegner, J. D., Breitkopf,U.,
Odermatt, D., Gitelson, A., Brando, V. E., and Schaepman, M. and Jung, J. 2014. Results of the ISPRS benchmark on
2012. Review of constituent retrieval in optically deep and urban object detection and 3D building reconstruction.
complex waters from satellite imagery. Remote Sensing of ISPRS Journal of Photogrammetry and Remote Sensing, 93,
Environment, 118, 116126. 256271.
Oloffson, P., Foody, G. M., Stehman, S. V., and Woodcock, C. E. Rouse, J. W. 1974. Monitoring the vernal advancement and ret-
2013. Making better use of accuracy data in land change rogradation of natural vegetation. NASA/GSFCT Type II
studies: Estimating accuracy and area and quantifying Report, Greenbelt, MD.
uncertainty using stratified estimation. Remote Sensing of Running, S. W., Collatz, G. J., Diner, D. J., Kahle, A. B., and
Environment, 129, 122131. Salomonson, V. V. (eds.). 1998. Special issue on EOS AM-1
Oloffson, P., Stehman, S. V., Woodcock, C. E., Sulla-Menashe, D., platform, instruments, and scientific data. IEEE Transactions
Sibley, A. M., Newell, J. D., Friedl, M., and Herold, M. 2012. on Geoscience and Remote Sensing, 36, 10561349.
A global land-cover validation data set, part I: Fundamental Running, S. W., Justice, C. O., Salomonson, V., Hall, D., Barker, J.,
design principles. International Journal of Remote Sensing, Kaufmann, Y. J., Strah-ler, A. H. etal. 1994. Terrestrial remote
33, 57685788. sensing science and algorithms planned for EOS/MODIS.
Oppelt, N. 2010. Monitoring of the biophysical status of veg- International Journal of Remote Sensing, 15, 35873620.
etation using multi-angular, hyperspectral remote sensing Saatchi, S., Buermann, W., ter Steege, H., Mori, S., and Smith,T.B.
for the optimization of a physically-based SVAT model. 2008. Modeling distribution of Amazonian tree species
In Kieler Geogr Schriften 121:123f. Kiel, Germany: Kiel and diversity using remote sensing measurements. Remote
University. Sensing of Environment, 112, 20002017.
Oppelt, N. and Mauser, W. 2004. Hyperspectral monitoring of Sabins, F. F. 1999. Remote sensing for mineral exploitation. Ore
physiological parameters of wheat during a vegetation Geology Reviews, 14, 157183.
period using AVIS data. International Journal of Remote Seifert, F. M. 2009. Improving urban monitoring toward a
Sensing, 25(1), 145160. European Urban Atlas. In: Gamba, P. and Herold, M. (eds.).
Papoutsa, C. and Hadjimitsis, D. G. 2013. Remote sensing for Global Mapping of Human Settlements: Experiences, Data
water quality surveillance in inland waters: The case study of Sets, and Prospects. CRC Press, Boca Raton, FL, pp. 231249.
Asprokremmos dam in Cyprus. In Hadjimitsis, D. G. (ed.). Sellers, P. J. 1985. Canopy reflectance, photosynthesis and tran-
Remote Sensing of EnvironmentIntegrated Approaches, spiration. International Journal of Remote Sensing, 6,
InTech, Croatia, 211pp. 13351372.
Pereira, M., Ferrier, S., Walters, M., Geller, G. N., Jongman, R. H. Sellers, P. J., Dickinson, R. E., Randall, D. A., Betts, A. K., Hall,
G., Scholes, R. J., Bruford, M. W. etal. 2013. Essential biodi- F. G., Berry, J. A., Collatz, G. J. et al. 1997. Modeling the
versity variables. Science, 339, 277278. exchanges of energy, water, and carbon between continents
Pesaresi, M., Guo, H., Blaes, X., Ehrlich, D., Ferri, S., Gueguen, L., and the atmosphere. Science, 275, 502509.
Halkia, M. etal. 2013. A Global Human Settlement Layer Shanahan, J. F., Schepers, J. S., Francis, D. D., Varvel, G. E.,
from optical HR/VHR RS data: Concept and first results. Wilhelm, W. W., Tringe, J. M., Schlemmer, M. R., and Major,
IEEE Journal of Selected Topics in Applied Earth Observations D. J. 2001. Use of remote sensing imagery to estimate corn
and Remote Sensing, 6, 21022131. grain yield. Agronomy Journal, 93, 583589.
Pettorelli, N. 2013. The Normalized Difference Vegetation Index. Shimoda, H. 1999. ADEOS overview. IEEE Transactions on
New York: Oxford University Press. Geoscience and Remote Sensing, 37, 14651471.
Pettorelli, N., Vik, J. O., Mysterud, A., Gaillard, J. M., Tucker, Showstack, R. 2014. Global Forest Watch Initiative provides
C. J., and Stenseth, N. C. 2005. Using the satellite-derived opportunity for worldwide monitoring. Eos, Transactions,
Normalized Difference Vegetation Index (NDVI) to assess American Geophysical Union, 95, 7779.
ecological effects of environmental change. Trends in Siddiqi, A. A. 2003. Sputnik and the Soviet Space Challenge.
Ecology & Evolution, 20, 503510. Gainesville, FL: University of Florida Press.
Pielke, R. A. 2005. Land use and climate change. Science, 310, Sirmacek, B., Taubenboeck, H., and Reinartz, P. 2012. Performance
16251626. evaluation for 3-D city model generation of six different
2016 by Taylor & Francis Group, LLC
Fundamentals of Remote Sensing for Terrestrial Applications 83
DSMs from air- and spaceborne sensors. IEEE Selected Topics Carranza, E. J. M., de Smeth, J. B., and Woldai, T. 2012.
in Applied Earth Observations and Remote Sensing, 5, 5970. Multi-and hyperspectral geologic remote sensing: A review.
Staenz, K., Mueller, A., and Heiden, U. 2013. Overview of ter- International Journal of Applied Earth Observations and
restrial imaging spectroscopy missions. Proceedings of the Remote Sensing, 14, 112128.
International Geoscience and Remote Sensing Symposium Vierling, K. T., Vierling, L. A., Gould, W. A., Martinuzzi, S., and
July 21-26, Melbourne (AUS) IGARSS 2013, pp. 35023505. Clawges, R. M. 2008. Lidar: Shedding new light on habitat
Stenzel, I., Steiner, U., Dehne, H.-W., and Oerke, E. C. 2007. characterization and modeling. Frontiers in Ecology and the
Occurrence of fungal leaf pathogens in sugar beet fields Environment, 6, 9098.
monitored with digital infrared thermography. In Stafford, Voogt, J. A. and Oke, T. R. 2003. Thermal remote sensing of urban
J. V. (ed.). Precis Agric07. Proceedings of the Sixth European climates. Remote Sensing of Environment, 86, 370384.
Conference on Precision Agriculture, October 10-12, Bonn Wagner, W., Luckman, A., Vietmeier, J., Tansey, K., Balzter, H.,
(Germany) Wageningen Academic Publisher, pp. 529535. Schmullius, C., Davidsone, M. Et al. 2003. Large-scale
Strand, H., Hft, R., Strittholt, J., Miles, L., Horning, N., Fosnight,E., mapping of boreal forest in SIBERIA using ERS tandem
and Turner, W. 2007. Sourcebook on remote sensing coherence and JERS backscatter data. Remote Sensing of
and biodiversity indicators. Montreal: Secretariat of the Environment, 85, 125144.
Convention on Biological Diversity, Technical Series no. 32, Wegscheider, S., Schneiderhan, T., Mager, A., Zwenzner, H., Post,
203pp. http://cce.nasa.gov/pdfs/cbd-ts-32_sourcebook.pdf J., and Strunz, G. 2013. Rapid mapping in support of emer-
Strong, A. E. 1974. Remote sensing of algal blooms by aircraft gency response after earthquake events. Natural Hazards,
and satellite in Lake Erie and Utah Lake. Remote Sensing of 68(1), 181195.
Environment, 3, 99107. Weng, Q. 2009. Thermal infrared remote sensing for urban cli-
Sullivan, R. J. 2000. Microwave Radar: Imaging and Advanced mate and environmental studies: Methods, applications,
Concepts. Norwood, MA: Artech House Inc. and trends. ISPRS Journal of Photogrammetry and Remote
Taubenbck, H., Esch, T., Felbier, A., Wiesner, M., Roth, A., Sensing, 64(4), 335344.
and Dech, S. 2012. Monitoring of mega cities from space. Weng, Q. and Quattrochi, A. 2006. Urban Remote Sensing. Boca
Remote Sensing of Environment, 117, 162176. Raton, FL: CRC Press Taylor & Francis.
Taubenbck, H., Klotz, M., Wurm, M., Schmieder, J., Wagner, B., Woodcock, C. E., Allen, R., Anderson, M., Belward, A.,
Wooster, M., and Esch, T. 2013. Central Business Districts: Bindschadler, R., Cohen, W., Gao, F. etal. 2008. Free access
Delineation in mega city regions using remotely sensed to Landsat imagery. Science, 320(5879), 1011. doi: 10.1126/
data. Remote Sensing of Environment, 136, 386401. science.320.5879.1011a.
Taubenbck, H., Wiesner, M., Felbier, A., Marconcini,M.,Esch,T., WRI (World Resources Institute). 2002. Global Forest Watch.
and Dech, S. 2014. New dimensions of urban landscapes: WRI Annual Review, 2002, 2021.
The spatio-temporal evolution from a polynuclei area Wulder, M. A., Masek, J. G., Cohen, W. B., Loveland, T. R.,
to a mega-region based on remote sensing data. Applied and Woodcock, C. E. 2012. Opening the archive: How
Geography, 47, 137153. free Landsat data has enables the science and monitor-
Thenkabail, P. S., Lyon, J. G., and Huete, A. 2012. Hyperspectral ing promise of Landsat. Remote Sensing of Environment,
Remote Sensing of Vegetation. Boca Raton, FL: CRC Press. 122, 210.
Thorp, K. R. and Tian, L. F. 2004. A review on remote sensing of Xiong, X., Wenny, B. N., and Barnes, W. L. 2011. Overview of
weeds in agriculture. Precision Agriculture, 5, 477508. NASA Earth Observing Systems Terra and Aqua moder-
Tilling, S. K., OLeary, G. J., Ferwerda, J. G., Jones, S. D., Fitzgerald, ate resolution spectroradiometer instrument calibration
G. J., and Rodriguez, D. 2007. Remote sensing of nitrogen algorithms and on-orbit performance. Journal of Applied
and water stress in wheat. Field Crops Research, 104, 7785. Remote Sensing, 3(1), 032501. doi: 10.1117/1.3180864.
Tsendbazar, N. E., de Bruin, S., and Herold, M. 2014. Assessing global Yang, G., Everitt, J. H., Bradford, J. M., and Escobar, D. E. 2000.
land cover reference datasets for different user communities. Mapping grain sorghum growth and yield variations using
ISPRS Journal of Photogrammetry and Remote Sensing, available airborne multispectral digital imagery. Transactions of the
online March 17, 2014. org/10.1016/j.isprsjprs.2014.02.008. ASAE, 43, 19271938.
Tucker, C. J. and Sellers, P. J. 1986. Satellite remote sensing of pri- Zhang, C. and Kovacs, J. M. 2012. The application of small
mary production. International Journal of Remote Sensing, unmanned aerial systems for precision agriculture: A
7, 13951416. review. Precision Agriculture, 13, 693712.
USA. 2010. National Space Policy of the United States of America. Zhou, X. and Wang, Y.-C. 2011. Spatialtemporal dynamics of urban
Washington, DC: United States Federal Government. green space in response to rapid urbanization and greening
USGS (US Geological Survey). 2001. Understanding Color- policies. Landscape and Urban Planning, 100, 268277.
Infrared Photography. USGS Fact Sheet, 129(1). Zhu, X. and Bamler, R. 2014. Super-resolving SAR tomography
Van der Meer, F. D., van der Werff, H., vanRuitenbeek, F. J., Hecker, for multi-dimensional imaging of urban areas. IEEE Signal
C. A., Bakker, W. H., Noomen, M. F., van der Meijde, M., Processing Magazine, 10535888.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
III
Remote Sensing
Data Normalization,
Harmonization, and
Intersensor Calibration
3 Overview of Satellite Image r adiometry in the Solar-r eflective Optical Domain Philippe M. Teillet ..................... 87
Acronyms and Definitions Introduction Need for Data Standardization Overview Sensor
Radiometric Calibration Surface Reflectance Retrieval Geometric Effects on Image Radiometry Spectral
Characterization Normalization Approaches Processing Considerations Discussion of Future
Trends Concluding Remarks Acknowledgments References
4 Postlaunch r adiometric Calibration of Satellite-Based Optical Sensors with e mphasis on Terrestrial
r eference Standard Sites Philippe M. Teillet and Gyanesh Chander ................................................................................ 109
Acronyms and Definitions Introduction Postlaunch Sensor Radiometric Calibration Methodologies Vicarious
Calibration via Terrestrial Reference Standard Sites Concluding Remarks Acknowledgments References
5 r emote Sensing Data Normalization Rudiger Gens and Jordi Cristbal Rossell.......................................................... 133
Acronyms and Definitions Introduction Remote Sensing Data Sources of Radiometric Distortion Radiometric
Normalization Conclusions References
6 Satellite Data Degradations and Their Impacts on High-Level Products Dongdong Wang ...................................... 143
Acronyms and Definitions Introduction Common Issues of Data Degradation Impacts of Data Degradations on
High-Level Products Concluding Remarks References
85
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
3
Overview of Satellite
Image Radiometry in the
Solar-Reflective Optical Domain
Acronyms and Definitions ...................................................................................................................87
3.1 Introduction .............................................................................................................................. 88
3.2 Need for Data Standardization ............................................................................................... 88
3.3 Overview .....................................................................................................................................89
3.4 Sensor Radiometric Calibration ............................................................................................. 90
Dynamic Range Converting Digital Counts to At-Sensor Radiance Converting At-Sensor
Radiance to At-Sensor Reflectance Vicarious Calibration
3.5 Surface Reflectance Retrieval...................................................................................................93
Atmospheric Correction
3.6 Geometric Effects on Image Radiometry...............................................................................95
Pixel Illumination and Viewing Geometries through the Atmosphere Atmospheric
Refraction Reflectance Anisotropy Adjacency Effects Topographic Effects Position
of the Sun
3.7 Spectral Characterization.........................................................................................................97
3.8 Normalization Approaches ......................................................................................................97
3.9 Processing Considerations .......................................................................................................97
3.10 Discussion of Future Trends ....................................................................................................97
3.11 Concluding Remarks.................................................................................................................98
Philippe M. Teillet Acknowledgments ................................................................................................................................ 99
University of Lethbridge References.............................................................................................................................................. 99
AcronymsandDefinitions MODTRAN Moderate resolution atmospheric transmission
NASA National Aeronautics and Space Administration
AVHRR Advanced very-high-resolution radiometer NIST National Institute of Standards and Technology
BRDF Bidirectional reflectance distribution function NOAA NationalOceanicandAtmospheric
CEOS Committee on Earth Observation Satellites Administration
CLARREO Climate Absolute Radiance and Refractivity NOMAD Networked online mapping of atmospheric data
Observatory NRC National Research Council
DEM Digital elevation model PSF Point spread function
DTM Digital terrain model QA4EO Quality Assurance Framework for Earth
ETM+ Enhanced Thematic Mapper Plus Observation
GEO Group on Earth Observations SI International system of units
GEOSS Global Earth Observation System of Systems SRBC Solar-radiation-based calibration
GIFOV Ground instantaneous field of view TM Thematic Mapper
GIS Geographic information system TOA Top of atmosphere
IFOV Instantaneous field of view TRUTHS Traceable Radiometry Underpinning Terrestrial
IVOS Infrared Visible Optical Subgroup and Helio Studies
LAI Leaf area index UTM Universal Transverse Mercator
MODIS Moderate resolution Imaging spectroradiometer WGCV Working Group on Calibration and Validation
87
2016 by Taylor & Francis Group, LLC
88 Remotely Sensed Data Characterization, Classification, and Accuracies
3.1 introduction terrestrial variable that serves as an example and a common
thread in this short treatment is that of reflectance, an Earth
Modern-day remote sensing satellite systems yield high-quality surface parameter of wide interest.
digital images that provide both synoptic and detailed obser-
vations of the Earth from space. The steps that have led to this 3.2 need for Data Standardization
unprecedented geospatial technology are many, and even a
summary of that development is beyond the scope of this chap- Environmental monitoring requirements and responsibilities
ter. Details of the key historical elements of Earth observation spanning from local communities to planetary scales continue
remote sensing, such as aviation, rockets, space travel, orbiting to multiply. Accordingly, building to a significant extent on mili-
satellites, imaging, and lunar and planetary exploration, can be tary technologies, government agencies in many countries have
found in numerous books (e.g., Burrows, 1999; Kramer, 2001; developed geostrategic technologies and applications that make
Jensen, 2006). Nonetheless, it is worth noting that it has long use of space-based observations of the Earth. Much of the focus
been known that seeing our planet from above brings significant of Earth sciences that make use of these satellite sensor systems
advantages.* is on improving predictions of Earth system changes, both short
Countless applications of satellite imagery have been and con- term and long term, primarily with respect to climate, popula-
tinue to be developed, the vast majority for qualitative, every- tion growth, land transformation, pollution, and biodiversity. As
day uses that benefit nevertheless from the laboratory geometric such, remote sensing systems help to provide more solid under-
and radiometric quality of the digital images available. That pinnings for decisions that impact us all in terms of quality of
said, there is also tremendous interest in the quantitative use of life and economic consequences.
satellite images to retrieve and monitor information about the The extent to which new Earth observation technologies can
current and changing states of geophysical and biophysical vari- contribute information of significant economic, social, environ-
ables, particularly those involved in climate, Earth resources, mental, strategic, and political value depends critically on develop-
and environment. Applications include, but are not limited to, ments in data standardization and quality assurance (MacDonald,
vegetation analysis (e.g., agriculture, forestry, precision farm- 1997; Teillet et al., 1997a). Thus, remote sensing calibration and
ing), disaster monitoring, environmental monitoring, watershed validation are essential aspects of Earth observation measurements
management, urban growth analysis, bathymetry, geological and methods used to estimate terrestrial variables to ensure that
mapping, mineral exploration, and intelligence data gathering. they are not compromised by sensor effects and data processing
In that light, this chapter outlines the key considerations that artifacts. Accordingly, research progress in remote sensing calibra-
need to be addressed to ensure that terrestrial variables derived tion has been featured periodically in review articles (e.g., Slater,
from satellite sensor systems operating in the solar-reflective 1984, 1985, 1988a; Duggin, 1986, 1987; Fraser and Kaufman, 1986;
optical domain are calibrated radiometrically to a common Teillet, 1986; Price, 1987a; Ahern etal., 1989; Duggin and Robinove,
physical scale. 1990; Che and Price, 1992; Nithianandan etal., 1993; Slater etal.,
Around 30 space agencies worldwide are operating currently, 1995, 2001; Teillet, 1997a; Teillet etal., 1997a; Secker etal., 2001;
or planning to launch over the coming 15 years, 258 Earth Teillet, 2005; Chander etal., 2013a), special journal issues and con-
observation satellite missions carrying 767 sensors (383 differ- ference proceedings (e.g., Markham and Barker, 1985; Price, 1987b;
ent sensors, some being repeats in a series). The data quality Slater, 1988b; Jackson, 1990; Guenther, 1991; Connolly and Tolar,
varies significantly between sensor systems due to differences in 1994; Bruegge and Butler, 1996; Teillet, 1997b; Chander and Teillet,
mission requirements, sensor characteristics, and the different 2010; Chander etal., 2013b), as well as book chapters and reports
calibration methodologies utilized. (e.g., Muench, 1981; Malila and Anderson, 1986; Teillet etal., 2004a;
In this chapter, emphasis is placed on the postlaunch method- Butler etal., 2005). The challenge is to ensure that measurements
ologies used to convert digital image data in the solar reflective and methods yield self-consistent and accurate geophysical and
domain to radiometrically calibrated products for the user com- biophysical data, even though the measurements are made with a
munity. For most targets of interest, the retrieval of land, atmo- variety of different satellite sensors under different observational
sphere, and ocean data and information from satellite image conditions and the parameter retrieval methodologies vary.
data requires sensor radiometric calibration, retrieval of surface The most stringent satellite data requirements are driven by
reflectance (involving correction for atmospheric effects), allow- the need for long-term monitoring of terrestrial variables to
ance for geometric effects on image radiometry, and an under- determine condition and detect change, as well as to provide
standing of scene spectral reflectance behavior. The primary inputs to regional and global carbon, energy, and water pro-
cess models. These requirements have been well documented
* Man must rise above the atmosphere and beyond to understand fully the and include climate and land surface variables such as surface
world in which he lives. Socrates, 700 BC. temperature, albedo, fraction of absorbed photosynthetically
These numbers are according to The Earth Observation Handbook (2014)
active radiation, and net primary productivity (e.g., Guenther
(http://www.eohandbook.com/) of the international Committee on Earth
Observation Satellites (CEOS), which consists of the majority of the space et al., 1996, 1997; Ohring et al., 2007). Similar user require-
agencies worldwide. The numbers exclude missions by military agencies ments have been documented for oceans (e.g., sea surface tem-
and commercial companies. perature and ocean color) (e.g., Gordon, 1987) and atmospheres
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 89
(e.g., temperature, water vapor, precipitation, ozone, clouds and units. Thus, quantitative uses of Earth observation rely on con-
aerosols, radiation, and trace species) (e.g., Guenther etal., 1996, version of the data to physical units, certifiably traceable to the
1997; Edwards etal., 2004). international system of units (SI) standards, in order to enable
Calibration is defined by the international Committeeon the comparison of Earth data from different sensors or from
Earth Observation Satellites (CEOS) Working Group on any given sensor over time, whether it be from days to decades
Calibration and Validation as the process of quantitatively defin- (e.g., Fox, 1999; Helder et al., 2012). Moreover, even for quali-
ing the system response to known, controlled signal inputs tative applications, remote sensing remains in some measure
(www.ceos.org). Generally, calibration-validation, or cal-val, untapped and it will only become a mainstream information
refers to the entire suite of processing algorithms used to con- technology when it provides reliability of supply, consistent data
vert raw data into accurate and useful Earth science data that quality, and plug-and-play capability (Teillet etal., 1997a).
are verified to be self-consistent (Teillet, 1997a). Calibration can
include radiometric, geometric, spectral, temporal, and polari- 3.3 overview
metric aspects. This chapter focuses on radiometric consider-
ations in the solar-reflective optical domain, but it is clear that Radiometry and radiation propagation in the remote sensing set-
all these different aspects are important considerations for any ting are well described in textbooks (e.g., Slater, 1980; Chen, 1996;
measurement device regardless of what part of the electromag- Wyatt etal., 1998; Morain and Budge, 2004; Schott, 2007). It is not
netic spectrum it covers (optical, microwave, or others) if it is to the role of this chapter to cover fundamental radiometric terms
yield useful data and information (Teillet etal., 2004a). Chapters and concepts but rather to provide a brief overview of the key
3 through 6 provide readers with a comprehensive overview of steps involved in converting digital image data from satellite opti-
remote sensing calibration. cal systems to calibrated surface quantities, spectral reflectance in
The Global Earth Observation System of Systems (GEOSS) particular, in the Earth science context. To begin, it is instructive
of the Group on Earth Observations (GEO) aims to deliver to look at the paths photons take from their source, the Sun, to
timely and comprehensive knowledge information products to the entrance aperture of any given satellite sensor in Earth orbit.
meet the needs of its nine societal benefit areas, of which the Figure 3.1 portrays the processes involved, with five main pathways
most demanding, in terms of accuracy, is climate (Ohring etal., and associated interactions, from a phenomenological perspective.
2005, 2007). This vision builds on the synergistic use of a sys-
1. Photons from the Sun propagate through the Earths
tem of disparate sensing systems that were or are being built
atmosphere, are reflected by the target of interest, and
for a multitude of applications, and requires the establishment
then propagate back through the atmosphere to the satel-
of a worldwide coordinated operational framework to facili-
lite sensor (direct sunlight path).
tate interoperability and harmonization. CEOS, considered to
2. Photons from the Sun are scattered by the Earths atmo-
be the space arm of GEO, has led the development of a Quality
sphere (thus generating diffuse sky illumination), are
Assurance Framework for Earth Observation (QA4EO),* which
reflected by the target of interest, and then propagate back
is based on the adoption of key guidelines. These guidelines
through the atmosphere to the satellite sensor (diffuse
have been derived from best practices for implementation by the
skylight path).
community under the auspices of GEO. The guidelines define
3. Photons from the Sun are scattered by the Earths atmo-
the generic processes and activities needed to put in place an
sphere back to the sensor without reaching the surface
operational QA4EO. Their use will facilitate the assignment of
of the Earth (i.e., path radiance, which can be as much
quality indicators to the output of every step in Earth observa-
as 10% of the total signal at visible wavelengths on hazy
tion data-processing chains to demonstrate the level of trace-
days).
ability to internationally agreed reference standards (SI where
4. Photons from the Sun propagate through the Earths
possible). The QA4EO was endorsed by the CEOS Working
atmosphere, are reflected from background objects and
Group for Calibration and Validation (WGCV) at its 29th ple-
subsequently from the target of interest, and then propa-
nary in September 2008 and endorsed by the CEOS Plenary in
gate back through the atmosphere to the satellite sensor
November 2008.
(multiple reflections).
The inadequacy of proper calibration and validation in diverse
5. Photons from the Sun propagate through the Earths
Earth observation applications has been identified at many work-
atmosphere, are reflected from surrounding surfaces, and
shops and in a variety of user need and market studies (e.g., Sweet
then scattered into the line of sight of the sensor (the so-
et al., 1992; Sellers et al., 1995; Hall et al., 1991; Horler, 1996;
called adjacency effect).
Horler and Teillet, 1996; Teillet, 1997c, 1998; Hegyi, 2004). Raw
or uncorrected imagery cannot be used to provide meaningful Atmospheric radiative transfer codes used for satellite image
information for natural resource management, environmental correction typically take (13) into account. Some codes take
monitoring, and climate studies. As Schowengerdt (2007) puts it: (5)into account, but those that include (4) are less common.
raw sensor digital counts are simply numbers, without physical
A recent estimate indicates that 99.5% of newly created digital data of all
* http://qa4eo.org/index.html. kinds are never analyzed (Regalado, 2013).
2016 by Taylor & Francis Group, LLC
90 Remotely Sensed Data Characterization, Classification, and Accuracies
1 4 3
FIg u r e 3.1 Schematic depicting photon pathways from the Sun to the entrance aperture of a satellite sensor: (1) direct solar illumination;
(2)diffuse sky illumination; (3) atmospheric path radiance; (4) background object reflections; and (5) adjacency effects.
Teillet (2005) lists over 60 considerations and factors that affect extraction. The main elements portrayed are sensor radiometric
the radiometry of Earth observation image data, grouped into calibration, surface reflectance retrieval (i.e., atmospheric cor-
seven categories: sensor characteristics, sensor regime, illumina- rection), radiometric normalization, and information extrac-
tion conditions, observation domain, path medium, target char- tion. Information extraction is a separate subject that is not
acteristics, and product generation domain. While some of the treated in this chapter, and normalization is touched on briefly
factors are straightforward, such as the variation in EarthSun only. However, the additional dimensions of spectral character-
distance, others merit extensive treatments in their own right, such ization and georadiometric effects on image radiometry, not eas-
as atmospheric scattering and absorption. This chapter touches on ily shown in the data flow in Figure 3.2 because they enter into
the key postlaunch factors, but it is beyond the scope of the chapter all of the elements, will be discussed.
to provide details on the remaining factors and considerations. Raw image data and calibrated radiances at the top of the
Figure 3.2 presents a data flow scheme that emphasizes the atmosphere (TOA) (Figure 3.2) are referred to as Level 0 and
main radiometric preprocessing steps prior to information Level 1 data, respectively. Level 2 data consisting of radiometri-
cally calibrated, and geolocated physical variables such as sur-
face reflectance, emittance, and temperature are preferred or
Sensor radiometric calibration Surface reflectance retrieval required in applications dependent on quantitative analyses.
Raw Calibrated Calibrated Still higher data product levels consist of information products
image TOA TOA that are spectral, spatial, or temporal integrations or aggregates
data radiance reflectance
of lower-level data. Few remote sensing data product genera-
tion systems and services offer Level 2 or higher-level products.
Therefore, users must typically undertake this processing work
Data Atmospheric
normalization correction
themselves using image analysis software or in collaboration
procedure procedure with private industry.
Other Other
inputs inputs
Normalized Surface 3.4 Sensor Radiometric calibration
image reflectance
data data The most fundamental part of the calibration-validation pro-
cess is sensor radiometric calibration, a broad and complex field
Other
that imposes the greatest limitations on quantitative applications
inputs Applications information extraction of remote sensing (e.g., Teillet, 1997a; Teillet et al., 1997a). The
methodologies and instrumentation involved can be grouped into
FIgu r e 3.2 Radiometric data flow options in preparation for infor- three domains (e.g., Dinguirard and Slater, 1999): (1) on the ground
mation extraction. The additional dimensions of spectral characteriza- before launch; (2) onboard the spacecraft postlaunch, including
tion and georadiometric effects on image radiometry (not shown, but reference to lamp sources and/or solar illumination (e.g.,Markham
discussed in the text) enter into all of the elements portrayed. et al., 1997); and (3) vicarious approaches using suitable Earth
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 91
scenes or extraterrestrial targets imaged in-flight (cf., later in this underestimation of leaf area index (LAI) over the course of a
chapter and the next chapter (Teillet and Chander, 2014)). decade by 4% and 1% for rangeland and grassland, respectively,
As for any scientific instrument, extensive and expensive whereas there would be negligible impact on the LAI of black
efforts are devoted to building stable optical imaging sensors spruce (Teillet etal., 2004a).
and to characterizing them to the fullest extent possible before
launch (e.g., Guenther etal., 1996). Prelaunch calibrations in a
3.4.1 Dynamic Range
laboratory are easier to control and perform than methods used
after launch. Sensor calibration coefficients are determined Radiance describes the flux of energy impinging on a given area
before launch using radiation sources, whose calibrations are from a specified direction in a given spectral band. Conventional
traceable to national laboratory standards. units for radiance, used in this document, are Watts per square
After launch, sensor radiometric response usually degrades meter per steradian per micrometer (Wm2sr1m1), although
over time such that it must be monitored continually and recali- some data suppliers use slightly different units. The dynamic range
brated as necessary throughout the life of the system to ensure the in radiance units is set by sensor design and calibration special-
maintenance of high data quality. In many instances, consider- ists to cover the full radiance range of scenes to be imaged in that
able time and effort go into the development and use of onboard satellite mission. For sensor systems with linear responsein the
systems such as standard lamps (blackbody sources for thermal radiance range of interest, the dynamic range is specified in terms
bands), and/or solar diffuser panels to monitor the postlaunch of the minimum radiance, L*min, which corresponds to a digital
calibration performance of satellite sensors. Nevertheless, even image level of Qcal, min (counts), and the maximum radiance, L*max ,
the status of the onboard calibration systems must be verified which corresponds to a digital image level of Qcal, max (counts) (e.g.,
over time via independent means. Vicarious methods (described Markham and Barker, 1987). The asterisk indicates a TOA quan-
in Section 3.4.4) provide these independent data and yield cali- tity. The four parameters, Qcal, min, Qcal, max, L*min, and L*max , establish
bration information over the mission lifetime. Further discus- a known linear relationship between calibrated digital data, Qcal,
sion of prelaunch and onboard calibration systems is beyond the and radiance at the sensor, L* (cf., next section).
scope of this chapter.
Whereas prelaunch methods encompass a vast array of pains-
3.4.2 converting Digital counts
taking sensor characterizations in the laboratory (and on very
to At-Sensor Radiance
rare occasions outdoors, so-called solar-radiation-based cali-
bration [SRBC] (e.g., Biggar et al., 1993; Mueller et al., 1996; In each spectral band, the outputs of most optical satellite
Dinguirard et al., 1997)), postlaunch radiometric calibration remote sensing systems are quantized raw image data, Q, which
based on onboard systems or vicarious methods is devoted pri- are transmitted to the ground (or recorded onboard and then
marily to the monitoring of the radiometric responsivities or gain subsequently transmitted to the ground). After housekeeping
coefficients for each sensor in each spectral band over time. Bias corrections and relative radiometric calibration, the raw data are
coefficients are determined generally on the basis of deep space archived on the ground as Level-0 data in units of digital counts.
or dark-shutter readings. In recent years, unified approaches have Relative radiometric calibration of satellite imagery primarily
been pursued to link the three domains (prelaunch, onboard, and involves characterizing and correcting the differences between
vicarious calibrations), plus the field instruments deployed at the detector gain and bias levels in sensor systems that utilize mul-
surface in support of vicarious calibration (e.g., Slater etal., 2001; tiple detectors in a given spectral band.
Butler etal., 2005). Error analyses of such approaches indicate that To derive radiometrically calibrated (Level 1) data requires
uncertainties in the 2%3% (1) range with respect to exo-atmo- knowing the relationship between Q and the TOA at-satellite
spheric solar irradiance are attainable in each domain and can be radiance input to the sensor (absolute radiometric calibration).
accurately related to national laboratory standards (Dinguirard Rather than computing radiance directly, image product genera-
and Slater, 1999). In all cases, the objective is traceability of radio- tion systems often transform raw data, Q, to scaled calibrated data,
metric calibration accuracies to the international system of units Qcal, taking care of changes in sensor calibration performance in
(SI) for science users (e.g., Pollock etal., 2003) and data products the process. For such Qcal products, the user can then use a set of
with consistent quality for the broader user community. time-invariant calibration coefficients to convert calibrated digi-
As an example, the prelaunch coefficients of the Landsat-5 tal counts, Qcal, to TOA radiance, L*. Thus, product generation
Thematic Mapper (TM) were revised after launch, because the systems require time-dependent information about the sensors
percent changes in radiometric gain calibration coefficients in radiometric calibration performance, a requirement that remains
the blue, green, and red spectral bands attained 1.0%, 0.85%, one of the most challenging aspects of the production and quan-
and 0.55% per annum, respectively, and the coefficients in the titative use of satellite image data (e.g., Slater and Biggar, 1996;
near-infrared and two shortwave infrared spectral bands changed Slater etal., 2001; Teillet and Chander, 2014).
by 0%, +0.1%, and 0.2% per annum, respectively (Teillet etal., The linear radiometric calibration equation is
2004a). These changes are significant given the almost 30-year
lifetime of the Landsat-5 mission. The impact of these radio- 1
L* = [Qcal Qo]. (3.1)
metric gain degradations is exemplified in one study by the G
2016 by Taylor & Francis Group, LLC
92 Remotely Sensed Data Characterization, Classification, and Accuracies
The gain and bias calibration parameters G and Qo are specified needed to deal with surfaces that are, to any significant extent,
as follows: non-Lambertian and/or sloped with respect to the horizontal.
The TOA or at-sensor reflectance, *, can be defined as a func-
Qcal ,max Qcal ,min tion of the upwelling radiance, L* (Wm2sr1m1), observed by
G= (counts per unit radiance)
(3.2) the satellite sensor divided by the exo-atmospheric solar irradi-
L*max L*min
ance, E0 (Wm2m1), as follows (Schott, 2007):
and
ds2L* ()
* () = , (3.5)
L*min E 0 ()cos(z )
Qo = Qcal ,min (Qcal ,max Qcal,min ) (counts).
(3.3)
L*max L*min
where
indicates quantities that are wavelength dependent
If a particular application involves the utilization of images from
is used to indicate that these variables are integrated over a
multiple sensors, it is necessary to transform the image data from
given spectral band
the different sensors to common physical units such as radiance
before they can be compared. If a particular application involves
As formulated earlier, the reflectance * is a dimensionless factor
the relative intercomparison of several images from a single sen-
between 0 and 1. Equation 3.5 is exact only for a Lambertian sur-
sor, the data can be used in the form of calibrated counts, Qcal.
face, that is, one for which the radiance is constant as a function
In the latter case, it is not necessary to transform to radiances.
of angle (hence the geometric factor ). E0 is the exo-atmospheric
It is worth noting that, while the sensor radiometric calibra-
solar irradiance (Watts per square meter per micrometer), z is
tion equation is as given in Equation 3.1, thus relating the input
the solar zenith angle, and ds is the EarthSun distance in dimen-
to the sensor (at-sensor radiance) to the output (digital counts)
sionless astronomical units. Exo-atmospheric solar irradiance is
in proper engineering terms, users are more interested in the
a difficult quantity to measure, and it does vary a few percent
inverse computation that converts digital counts in their image
over time. Hence, an average model for E0 is always used and
data to at-sensor radiance. Unfortunately, the coefficients for
several such models have been developed (e.g., Thuillier et al.,
this inverse calculation are at times also called gain and offset,
2003). Therefore, care must be taken to ensure that comparisons
which can lead to confusion. These inverse computation gain
between different calibration methods and results are made on
and offset coefficients have units that differ from those of the sen-
the basis of the same E 0 data.
sor radiometric calibration gain and offset coefficients defined
The use of TOA reflectance as opposed to TOA radiance cor-
inEquations 3.2 and 3.3, that is, the inverse computation gain
rects for at least some of the sources of variation that affect satel-
and offset coefficients are in units of radiance per unit count and
lite data, in particular, variations in solar illumination caused by
radiance, respectively.
the diurnal cycle and by cyclical changes in the EarthSun dis-
tance, as well as differences in solar irradiance due to differences
3.4.3 converting At-Sensor Radiance between similar spectral bands from sensor to sensor (discussed
to At-Sensor Reflectance later in this chapter).
A full description of reflectance, whether it be TOA or at the sur-
face, is given by the bidirectional reflectance distribution func- 3.4.4 Vicarious calibration
tion (BRDF), which is a function of all incident (i) and reflected
(r) zenith angles () and azimuthal angles () (Nicodemus, 1965, Earth surfaces with appropriate characteristics have long been
1970): used to provide postlaunch updates of the radiometric cali-
bration of satellite sensors, a methodology often referred to as
L(r , r ) 1 vicarious or ground-look calibration (e.g., Slater et al., 1987,
BRDF = (sr ). (3.4)
E (i , i ) Teillet et al., 1990; Biggar et al., 1994; Thome, 2001). A com-
prehensive summary of this topic is given in Chapter 4 (Teillet
Units for radiance, L, are as already defined (Wm2sr1m1) and Chander, 2014). Accordingly, only a few aspects are men-
and units for irradiance, E, are Watts per square meter per tioned here.
micrometer (Wm2m1). Note that the units of BRDF are per Reflectance-based or radiance-based methods use surface
steradian (sr1). Since full BRDF characterizations of terrestrial and atmospheric measurements to estimate TOA radiance at
surfaces are nontrivial, remote sensing most often makes use of the entrance aperture of a given satellite sensor in order to
simplified representations such as the Lambertian reflectance. monitor and, if needed, provide updates of sensor radiomet-
The radiance from the idealized surface known as a Lambertian ric calibration. Initially, field measurement campaigns by spe-
surface is the same in all directions, that is, the decreasing inten- cialized teams at calibration reference test sites targeted only
sity with angle follows a cosine law and is exactly compensated one sensor per sortie. In subsequent years, with the increase
by a decrease in projected area. More advanced treatments are in the number of sensors that passed over a given test site on a
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 93
given day, it became possible with careful planning to under- and Saleous, 2006). Quantitative Earth observation applications
take vicarious calibrations for several sensors per sortie (e.g., taking advantage of AVHRR synoptic daily coverage found it
Thome etal., 1998). Efforts such as these are resource intensive difficult to keep up with radiometric calibration changes until
and take time to complete. Hence, it has been of considerable various researchers and, eventually, NOAA developed and
interest to develop less expensive complementary approaches implemented appropriate approaches and communication
that can provide more frequent calibration updates, even if tools (e.g., Teillet and Holben, 1994; Cihlar and Teillet, 1995;
they are less accurate individually. The use of standard refer- Rao and Chen, 1995, 1996, 1999; Brest etal., 1997; Tahnk and
ence test sites to compare or transfer radiometric calibration Coakley,2001).
between satellite sensors, that is, cross-calibration, has also Compared to onboard radiometric calibration techniques, the
been investigated (e.g., Teillet etal., 2001a; Thome etal., 2003; main disadvantages of vicarious calibration are that the meth-
Chander et al., 2013a; Helder et al., 2013), with and without ods suffer from lower precision and lower temporal sampling
near-simultaneous or coincident surface measurements. With frequency. The main advantage of vicarious calibration is that
greater timeliness and reduced costs in mind, new methods the sensor typically acquires calibration data in the same modal-
are being explored that yield useful results without near- ity as it acquires Earth image data, that is, with the same source
simultaneous or coincident measurements by field teams (e.g., spectrum, similar illumination conditions, and the same spec-
Teillet etal., 2001b; Rao etal., 2003; Thome, 2005; Thome etal., tral bands.
2008; McCorkel etal., 2013).
Vicarious calibration can also be undertaken with respect to 3.5 Surface Reflectance Retrieval
lunar views (e.g., Kieffer and Wildey, 1985, 1996; Godden and
McKay, 1997; Kieffer etal., 2002; Barnes etal., 2004; Stone etal., Surface reflectance in the solar reflective part of the electromag-
2005, 2013; Stone, 2008; Xiong etal., 2008) and/or bright stars netic spectrum is a primary geophysical variable of interest in
(e.g., Chang etal., 2012). Although such ultrastable targets are many applications, either for its own sake or to generate other
imaged without atmospheric effects, they are relatively low radi- geophysical or biophysical quantities. The Lambertian surface
ance sources and they require high-risk spacecraft platform spectral reflectance, surf (dimensionless between 0 and 1), is
maneuvers to achieve. Some sensors are designed to acquire defined as times the ratio of the radiance, L (Wm2sr1m1),
space views to provide dark target calibration data. upwelling from the target (as measured just above the target)
Other vicarious calibration or cross-calibration methods divided by the downwelling irradiance, E (Wm2m1), that
take advantage of nonland Earth targets, including atmo- illuminates the target (Schott, 2007):
spheric Rayleigh scattering (e.g., Vermote etal., 1992; Kaufman
and Holben, 1993; Dilligeard etal., 1997; Meygret etal., 2000), L()
surf () = . (3.6)
ocean sun-glint (e.g., Kaufman and Holben, 1993; Vermote and E ()
Kaufman, 1995; Luderer etal., 2005), snow/ice fields (e.g., Loeb,
1997; Tahnk and Coakley, 2001; Nieke et al., 2003; Six et al., As already noted (Figure 3.1), the downwelling irradiance
2004), and clouds (e.g., Vermote and Kaufman, 1995; Iwabuchi, signal at the surface, E, is made up of several components in
2003; Doelling etal., 2004, 2010; Hu etal., 2004; Fougnie and addition to the direct solar and diffuse sky contributions, a
Bach, 2009). consideration that must also be taken into account when mak-
Early and ongoing work worth mentioning in the context of ing ground-based radiance measurements. The sky irradiance
vicarious calibration is the case of the series of advanced very- distribution is anisotropic, but it can be modeled reasonably
high resolution radiometer (AVHRR) sensors operating on well to take into account the increase in diffuse radiation in
National Oceanic and Atmospheric Administration (NOAA) the circumsolar region and toward the horizon (Dave, 1977;
satellites since 1978. The AVHRR program did not include Steven, 1977; Temps and Coulson, 1977; Klucher, 1979; Hooper
onboard postlaunch radiometric calibration for the solar- and Brunger, 1980; Kirchner et al., 1982; Hooper et al., 1987;
reflective spectral bands. As a result, a wide variety of vicarious Brunger and Hooper, 1993). This refinement is especially help-
methodologies were developed to generate updates of prelaunch ful where topography obscures parts of the sky from a given
radiometric calibrations during the course of the mission on- point on the ground (Hay and McKay, 1985; Hay etal., 1986;
orbit. AVHRR radiometric gain degradation proved to be Duguay, 1993).
significant postlaunch, and the degradations differed consider-
ably from sensor to sensor in the AVHRR series (e.g., Frouin
3.5.1 Atmospheric correction
and Gautier, 1987; Smith etal., 1988; Holben etal., 1990; Staylor,
1990; Teillet etal., 1990; Whitlock etal., 1990; Brest and Rossow, Scattering and absorption due to aerosols and gases in the
1992; Che and Price, 1992; Mitchell et al., 1992; Teillet, 1992; atmosphere modify radiation, making its way from outside the
Vermote et al., 1992; Abel et al., 1993; Kaufman and Holben, atmosphere down to the target, as well as the surface reflectance
1993; Vermote and Kaufman, 1995; Loeb, 1997; Cabot et al., propagated through the atmosphere to the satellite sensor (e.g.,
2000; OBrien and Mitchell, 2001; Cao and Heidinger, 2002; Chahine, 1983; Schott, 2007). In the solar-reflective spectral
Iwabuchi, 2003; Wu etal., 2003; Doelling etal., 2004; Vermote domain, molecular (Rayleigh) scattering (4) is strongest in
2016 by Taylor & Francis Group, LLC
94 Remotely Sensed Data Characterization, Classification, and Accuracies
the blue (e.g., contributing ~0.07 in reflectance for 1013 mbar), not be used (Teillet, 1989), and bandpass calculations based on
aerosol scattering (2 to +0.6) is strongest in the visible (e.g., relative spectral response profiles with 0.0025m grid spacing
contributing ~0.04 in reflectance for 10km visibility), ozone gas or finer are recommended. Another caution is that Rayleigh
absorption is strongest in the green (e.g., ~10% effect on atmo- scattering, a long-understood phenomenon, is not always com-
spheric transmission for 0.35cmatm), and water vapor absorp- puted accurately in atmospheric correction codes (Teillet, 1990a;
tion is strongest in the near-infrared and shortwave infrared He etal., 2006).
(e.g., ~12% effect on transmission for 3 gcm2). Interestingly, Operational atmospheric correction for surface reflectance
atmospheric source and loss effects roughly cancel out for Earth retrieval depends on ready access to timely and accurate infor-
surface reflectances around 0.25; hence, this is a good reflec- mation on atmospheric variables such as aerosol optical depth
tance for vicarious calibration reference sites, since atmospheric (e.g., Teillet etal., 1994) and water vapor content. Although the
effects are minimized. computational tools for image correction are available, the user
Given that the optical properties of the Earths atmosphere is still left with the problem of obtaining the necessary atmo-
are not uniform spatially or temporally, image corrections for spheric variables for input to the image correction. The main
these effects in the solar-reflective spectral bands are needed to possibilities in this respect are as follows:
put satellite data on the same radiometric scale for investigations
1. Measure the required parameters in the field at the same
intended to monitor terrestrial surfaces quantitatively over time
time as image acquisition.
and space. A compact form of the relationship between surface
2. Query an online source of radiosonde and/or sunpho-
and TOA reflectance is given as (Tanr etal., 1990)
tometer network data that may have acquired parameters
near the time and location of the image acquisition (e.g.,
Holben etal., 1998).
* = g s v surf + a . (3.7)
1 s 3. Assume fixed standard values for the atmospheric param-
eters for the geographic region of interest (e.g., Fedosejevs
etal., 2000).
All quantities are wavelength dependent. g is atmospheric gas
4. Use climatological values developed over time (e.g.,
transmittance, s is downward scattering transmittance in the
Bokoye etal., 2001).
solar illumination direction, v is upward scattering transmit-
5. Estimate the required parameters from the image data
tance in the sensor view direction, a is atmospheric reflectance,
themselves, using techniques such as the dark target
is the average reflectance of surrounding surfaces, and s is the
approach (e.g., Chavez, 1988, 1996; Teillet and Fedosejevs,
spherical albedo (i.e., the reflectance) of the atmosphere. All
1995; Liang etal., 1997; Song etal., 2001) and atmospheric
reflectances and transmittances in Equation 3.6 are dimension-
absorption feature extraction (e.g., Staenz etal., 2002).
less factors between 0 and 1. The atmospheric quantities are
6. Use data assimilation results based on dynamic models
computed using radiative transfer codes.
driven by analyzed meteorological data (e.g., ONeill etal.,
The detailed textbook treatment by Schott (2007) is also
2002).
instructive and worth citing. It explains all of the contributions
to the upwelling spectral radiance reaching the satellite sensor, The challenge is to make optimum use of available ground-
including the thermal energy paths not presented in this chap- based and satellite-based atmospheric optical measurements
ter. The result is the so-called big equation for the TOA spectral and to ensure they are consistent and easily accessible. A pro-
radiance (cf., Schott (2007) for details). totype atmospheric optical parameter estimation system was
The need for and various approaches to atmospheric correc- developed to provide optical parameters online for any time
tion have long been examined in the literature (e.g., Turner and and place across extended regions such as Canada or North
Spencer, 1972; ONeill etal., 1978; Ahern etal., 1979; Richardson, America based on available climate to meteorological scale
1982; Slater and Jackson, 1982; Tanr etal., 1983; Moran etal., data and models (Freemantle etal., 2002; ONeill etal., 2002).
1992; Kaufman etal., 1997), and atmospheric radiative transfer Aside from making atmospheric corrections more operational,
codes are available in modern remote sensing image analysis sys- this concept of networked online mapping of atmospheric data
tems. Some solutions may involve the use of a radiative transfer (NOMAD) transfers the responsibility of making available
code, but many make use of precomputed look-up table results quality atmospheric parameters from the user to the scientific
to save computation time. A widely accepted atmospheric cor- and technical specialists maintaining the atmospheric param-
rection code for research purposes is MODTRAN5 (moderate eter server. Nevertheless, at the end of the day, the significant
resolution atmospheric transmission code version 5) (Berk etal., uncertainties that can arise from the atmospheric correction
2006). Most of the predominantly used codes tend to disagree process with uncertain inputs (e.g., Kaufman etal., 1997; Teillet
significantly only for very large aerosol optical depths and/or etal., 1994), too varied to characterize in a summary way, are
large off-nadir illumination or viewing geometries (>60). Thus, such that users often prefer to seek out imagery acquired on
the choice of code is not an important factor except for the cor- relatively haze-free days, or to apply normalization techniques
rection of high spectral resolution (hyperspectral) data (e.g., (e.g., Schroeder etal., 2006) rather than undertake atmospheric
Staenz etal., 1994, 2002). Monochromatic computations should correction.
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 95
3.6 Geometric effects on image data acquired for high-latitude regions, where sun angles
are typically far from the zenith, are more difficult to correct
image Radiometry for atmospheric effects. In addition, while some atmospheric
3.6.1 Pixel correction codes incorporate simple BRDF models to allow
for multiple surface-atmosphere scatterings (e.g., the 6S code*:
The nature of the remote sensing image pixel has been given Kotchenova etal., 2006; Vermote etal., 2006; Kotchenova and
some but, arguably, not enough attention (Townshend, 1981; Vermote, 2007), computationally intensive multiple scattering
Duggin, 1986; Fisher, 1997; Cracknell, 1998; Townshend etal., in the radiative transfer is not always a component of atmo-
2000). Clearly, a pixel is not a true geographical object, nor does spheric correction algorithms and, hence, is not always taken
it correspond exactly to the spatial resolution of the imaging into consideration when deriving reflectance information
sensor. The pixel is a sampling of the nonuniform point spread about a surface from satellite image data. The inclusion of mul-
function (PSF) of a given instantaneous field-of-view (IFOV) of tiple scattering is a greater concern for ocean color parameter
the sensor. Space does not allow elaboration on the subject of retrieval (e.g., Gordon and Wang, 1994) than for land surface
PSFs of satellite imaging sensors. A good treatment of this topic reflectance retrieval.
can be found in Schowengerdt (2007).
It is easy to forget that a substantial proportion of the signal
apparently coming from the surface area represented by a given 3.6.3 Atmospheric Refraction
pixel comes from surrounding areas and from the atmospheric For typical air densities, the atmospheric refraction correction
path. A significant proportion of any given satellite image con- for the entire atmosphere down to sea level is less 0.1 for zenith
sists of mixed pixels. Moreover, in most imaging systems, pixels angles less than 80 (Scarpace and Wolf, 1973; McCartney, 1976;
at large off-nadir view angles each encompass significantly larger Chu, 1983; Egan, 1985). Accordingly, atmospheric refraction
surface areas than do pixels closer to nadir because of the pro- only becomes important for satellite sensors designed to study
jection effect. Large field-of-view whiskbroom sensors such as the Earths atmosphere at near-grazing angles.
the moderate-resolution imaging spectroradiometer (MODIS)
are subject to the panoramic bow tie effect, yielding scans th at
partially overlap at off-nadir angles (Souri and Azizi, 2013). 3.6.4 Reflectance Anisotropy
Further still, cross-track ground sampling intervals increase with Reflectance anisotropy as a function of illumination and viewing
increasing view angle because of Earth curvature; the flat-Earth geometries is a fundamental property of any terrestrial surface
approximation is good to within 4% for off-nadir view angles less and is best described in terms of BRDF (Equation 3.4). Some ter-
than 23 (Schowengerdt, 2007). These considerations have impor- restrial surfaces have highly anisotropic reflectance properties,
tant ramifications on the per-pixel characterization of regions of and this has an important influence on image radiometric char-
interest and on the integration of remote sensing data into geo- acteristics. Forest canopies are particularly subject to these effects
graphical information systems (GISs). Thus, contrary to common (cf., Figure 3.3). Methodologies to model and deal with BRDF
practice, land cover properties should be reported at spatial reso- effects are well documented in the literature (e.g., Li and Strahler,
lutions coarser than the individual remote sensing pixels. 1992; Chen and Cihlar, 1997; White et al., 2002a; Pinty et al.,
The case of the AVHRR sensor series illustrates this point. 2004). Nevertheless, although anisotropic reflectance effects have
Nominally, the nadir-view size of AVHRR pixels is 1.1 km by been studied extensively, they remain challenging to deal with in
1.1km. With the large off-nadir angles that constitute a signifi- an operational setting and there are many other geometric effects
cant portion of each AVHRR image, the majority of pixels are to consider apart from the BRDF.
greater than 1.1km in size, reaching as much as 10 times the area
of nadir pixels at the largest off-nadir angles (55 at ground level).
Moreover, it has been estimated that the nadir pixel sampling at 3.6.5 Adjacency effects
1.1km represents approximately only 27% of the radiation cap- As already noted, some of the photons reaching the satellite
tured by the ground IFOV (GIFOV) (personal communication, sensor would have propagated through the Earths atmosphere,
NASA Langley Research Center). It has also been found that, reflected from surfaces surrounding any given IFOV, and then
in the case of the Landsat-5 TM, the cross-track GIFOV is on scattered into the line of sight of the sensor (Otterman and
the order of 4045 m instead of the nominal resolution of 30 m Fraser, 1979; Dave, 1980; Dana, 1982; Meckler and Kaufman,
(Schowengerdt etal., 1985). 1982; Deschamps et al., 1983; Otterman et al., 1983; Kaufman
and Fraser, 1984). This adjacency effect can give rise to blurring
and contrast reduction at the boundaries of IFOVs. While the
3.6.2 illumination and Viewing Geometries
preference here is to include it as a georadiometric effect, the
through the Atmosphere
adjacency effect is commonly discussed as part of atmospheric
Except for large angles, the influence of varying illumination correction. However, more often than not, the adjacency effect
and viewing paths through the atmosphere is generally well
handled by most atmospheric correction algorithms. However, * 6S: Second Simulation of the Satellite Signal in the Solar Spectrum.
2016 by Taylor & Francis Group, LLC
96 Remotely Sensed Data Characterization, Classification, and Accuracies
45 0 +45
FIg ur e 3.3 Digital imagery showing the same black spruce canopy taken at the indicated angles from a truck-based platform. The effect of
reflectance anisotropy is clearly evident (solar illumination is from the right). Because these anisotropy effects are at the scale of trees and their
components, they will give rise to sub-pixel variations in satellite sensor data of any spatial resolution.
is ignored or, at best, taken into account in an approximate way resampling (Forster and Trinder, 1984), even using an optimum
(Tanr etal., 1990; Vermote etal., 2006). parametric cubic convolution algorithm (Schowengerdt et al.,
1983). Additionally, the slope-aspect correction procedure has to
keep track of the scan direction to specify view angles properly.
3.6.6 topographic effects
A better approach is to not apply the map transformation to the
A number of approaches have been developed to address the image before slope-aspect correction and, instead, make use of
influence of surface topography on the radiometric properties the map transformation equations in reverse to apply the slope-
of pixels in remotely sensed image data. Terrain elevation varia- aspect correction in image space.
tions cause variable atmospheric path lengths that image correc- A rule of thumb to keep in mind is that the spatial resolution
tion algorithms have to address in the contexts of remote sensing of the digital elevation model (DEM) used to define the slope
from both airborne platforms in the atmosphere and satellite and aspect data in the DTM should be twice as fine as the spa-
platforms above the atmosphere (e.g., Teillet and Santer, 1991; tial resolution of the image data to be corrected for slope-aspect
Teillet and Staenz, 1992). Terrain elevation variations also cause effects (Hodgson, 1995). For example, terrain elevation data with
shadow effects that introduce errors in image understanding in horizontal spatial resolutions of 15 m by 15 m should be used
general (Woodham and Lee, 1985; Woodham and Gray, 1987) to generate the slope and aspect angles to be used to correct
and in surface reflectance retrieval in particular (Giles, 2001; imagery whose spatial resolution is 30 m by 30 m.
Adeline etal., 2013). In principle, topographic slope-aspect corrections should also
Image corrections for terrain slope and aspect variations are involve atmospheric and surface reflectance models to allow for
more difficult to carry out. Some approaches have used normal- the proper treatment of the effects due to terrain elevation varia-
ization techniques (e.g., Holben and Justice, 1981; Allen, 2000), tions, BRDF, irradiance from adjacent slopes, and diffuse sky
with mixed results (Richter, 1997, 1998). Instead, most radiomet- illumination distribution (Kimes and Kirchner, 1981; Bruhl and
ric corrections for topographic effects use digital terrain eleva- Zdunkowski, 1983; Cavayas etal., 1983; Sjoberg and Horn, 1983;
tion, slope, aspect, and other derivatives (i.e., a digital terrain Woodham and Lee, 1985).
model (DTM)) to describe explicitly the surface topography. Two
main categories of these approaches are: (1) slope-aspect correc-
3.6.7 Position of the Sun
tions that use some function of the cosine of the incident solar
angle (Smith etal., 1980; Teillet etal., 1982; Cavayas, 1987; Meyer Uncertainties in knowledge of the position of the Sun can give
et al., 1993; Ekstrand, 1996; Riao et al., 2003; Soenen et al., rise to significant errors in shadow area estimation in rugged
2005, 2008) and (2) model-based corrections that deal with solar terrain (Teillet et al., 1986). The largest error by far is the use
radiation interactions with surface targets (Hugli and Frei, 1983; of a single set of solar zenith and azimuth angles for the radio-
Proy etal., 1989; Gu and Gillespie, 1998; Dymond and Shepherd, metric processing and correction of an entire satellite image.
1999; Shepherd and Dymond, 2003; Li et al., 2012) and with For example, the lack of specification of within-scene location
atmospheric radiative transfer codes (Richter, 1997, 1998). for a Landsat image leads to uncertainties in solar position of
In practice, slope-aspect corrections may involve the prior 4.5 (zenith angle) by 2.5 (azimuth angle). It is not uncommon
rectification of an image to a map-based coordinate system, to use a single set of solar angles for Landsat images, whereas,
typically the Universal Transverse Mercator (UTM) projec- for radiometric corrections applied to wide-area coverage satel-
tion, which matches the map coordinate system of the DTM. lite imaging sensors such as AVHRR or MODIS, it is routine to
Although this simplifies image processing and facilitates image specify solar angles on a subscene basis. Other contributions to
classification studies involving various images in a common the uncertainty in solar position include, in decreasing order of
map projection, there are disadvantages to the image rotation importance: finite size of the solar disk (30min), lack of speci-
and resampling that occur during the map rectification process. fication of within-scene time (15min [azimuth angle] by 8min
Significant radiometric errors can be introduced as a result of [zenith angle]), use of solar ephemeris from an inappropriate
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 97
epoch, atmospheric refraction effects for large solar zenith used to bypass steps in Figure 3.2. Spectral band transformations
angles, and horizontal parallax (target at Earth surface, not (e.g., band ratios or principal component analyses) have long
Earth center). The potential cumulative uncertainty effect in the been used to mitigate atmospheric and/or topographic effects
Landsat case is 5 (zenith angle) by 3.5 (azimuth angle). (e.g., Crist, 1985). A variety of empirical radiometric normaliza-
tion methods have been developed (e.g., Schott etal., 1988; Hall
3.7 Spectral characterization etal., 1991; Yuan and Elvidge, 1996; Schroeder etal., 2006).
Although it has received less attention than other aspects of satel- 3.9 Processing considerations
lite image radiometry, spectral characterization is an important
consideration for proper surface reflectance retrieval, regard- This chapter has so far provided an overview of the principal
less of how wide or narrow the spectral bands may be. Spectral physical effects affecting satellite image radiometry. Additional
bands designed for specific applications and data products are concerns arise in the domain of image processing that, under
susceptible to postlaunch variations in spectral bandpasses and certain circumstances, can affect image radiometry significantly.
they usually differ between different sensors and missions. If Nominally, image-processing software will first calibrate dig-
spectral bands have changed in wavelength position or band- ital counts to TOA radiance and, in some cases, convert TOA
width postlaunch, or if there are uncertainties as to their charac- radiance to TOA reflectance. In any case, the image data are then
teristics, there is a direct impact on radiometric and atmospheric in physical units such that atmospheric correction to estimate
processing, as well as on derived information products (Suits surface reflectance can be done. Adjustments for bidirectional
etal., 1988; Teillet, 1990b; Teillet and Irons, 1990; Flittner and and/or topographic effects should be included in the latter step.
Slater, 1991; Goetz etal., 1995; Teillet etal., 1997b; Steven etal., It is important to note that these radiometry-related image-
2003; Trishchenko, 2009). In practice, there is little that users processing steps involve equations that are linear or nonlinear.
can do to take into account postlaunch changes in spectral band The associated software must be constrained to prevent user
performance. Even when spectral bands perform as designed, application of the various calibration/correction steps in differ-
users should be aware that similar information products derived ent sequences because linear and nonlinear transformations are
from different sensors with analogous spectral bands that do not not commutative (Teillet 1986). In principle, all radiometric cor-
match exactly are not directly comparable (Teillet et al., 2007; rections should be completed prior to any geometric processing.
Teillet and Ren, 2008). These spectral band difference effects are Moreover, it is important to understand how the sequenc-
scene dependent and, hence, not corrected easily. ing of spectral band integration can affect image radiometry,
A limited number of investigations have been undertaken to especially in processing for surface reflectance retrieval (i.e.,
assess radiometric calibration errors due to differences in spec- atmospheric correction). Detailed atmospheric computations
tral response functions between satellite sensors when attempt- can be done monochromatically and the results integrated over
ing cross-calibration based on near-simultaneous imaging the spectral band involved, or band-integrated quantities can
of common ground look targets in analogous spectral bands be used in the atmospheric computations. Depending on the
(Teillet et al., 2001c, 2004b, 2007; Trishchenko et al., 2002; surface reflectance spectrum and the relative spectral response
Rao et al., 2003; Doelling et al., 2012; Chander et al., 2013c; profile of the spectral band involved, the two approaches can
Henry etal., 2013). A specific example is the radiometric cross- yield results that differ significantly, and the proper choice of
calibration of the Landsat-7 Enhanced Thematic Mapper Plus approach depends mainly on whether the image data are mul-
(ETM+) and Landsat-5 Thematic Mapper (TM) sensors based tispectral or hyperspectral in character (Teillet 1989; Teillet and
on early-mission tandem-orbit data sets (Teillet et al., 2001c, Irons 1990).
2004b), which included adjustments for spectral band differences
between the two Landsat sensors. Spectral band difference effects 3.10 Discussion of Future trends
were shown to be significant (an effect that can be 5% or more),
despite the close similarity in the spectral filters and response Although radiometric calibration is a very specialized aspect of
functions of the Landsat sensors. A variety of terrestrial surfaces remote sensing, it constitutes an essential component of Earth
were assessed regarding their suitability for Landsat radiometric observation systems and their utilization. It is crucial for the
cross-calibration in the absence of surface reflectance spectra. generation of useful long-term data records, as well as for inde-
This line of inquiry was extended to Earth observation sensors pendent data quality control. In that light, this chapter outlines
on several satellite platforms, indicating that large differences the key postlaunch methodologies used to convert digital image
can arise (Teillet etal., 2007; Teillet and Ren, 2008). data to calibrated data products. Nevertheless, postlaunch cali-
bration and the associated quality assurance infrastructures,
3.8 normalization Approaches which should be behind-the-scenes activities as with any mature
technology, remain semioperational at best. The CEOS QA4EO
Many applications are not dependent on securing radiometric effort, noted in Section 3.2 in the context of GEOSS, is an
calibration on an absolute scale. Only relative changes between important step in the direction of operational best practices for
images are of interest such that normalization procedures can be traceability and interoperability. Two other calibration-related
2016 by Taylor & Francis Group, LLC
98 Remotely Sensed Data Characterization, Classification, and Accuracies
initiatives for the future are noted here, both representing chal- 3.11 concluding Remarks
lenging space-based undertakings.
The Traceable Radiometry Underpinning Terrestrial and This chapter has provided a broad overview of the key consider-
Helio Studies (TRUTHS) mission concept (Fox et al., 2002, ations involved in the radiometric calibration and correction of
2011) offers a novel approach to the provision of key scientific image data from satellite sensor systems operating in the solar-
data with unprecedented radiometric accuracy for Earth obser- reflective optical domain. Relevant pioneering research over the
vation and solar studies. TRUTHS will calibrate its instrumen- past four decades has been featured, with emphasis on sensor
tation directly to SI on orbit, overcoming the usual uncertainties radiometric calibration, retrieval of surface reflectance (involv-
associated with drifts of sensor calibration by using an electrical ing correction for atmospheric effects), allowance for geomet-
rather than an optical standard as the basis of its radiometric ric effects on image radiometry, and an understanding of target
calibration. A space-based cryogenic radiometer together with spectral reflectance behavior. Surface reflectance served as an
its associated calibration chain to a terrestrial primary standard important Earth surface variable of wide interest. The need for
will provide a space-based standard reference for measurements radiometric calibration and correction is well documented in
of both the Sun and the Earth. The TRUTHS mission has the the literature and growing in importance as Earth science data
potential to improve the performance and accuracy of Earth products are being derived increasingly from a multiplicity of
observation missions by an order of magnitude (to accuracies different satellite sensor systems, and data and information
on the order of 0.3% in the solar reflective domain). As a result, demands for the purposes of sound decision-making require
TRUTHS will provide the necessary advances to ensure that increasingly greater precision and accuracy.
Earth system data sets have sufficient radiometric long-term pre- Sensor radiometric calibration is the most fundamental con-
cision for the reliable detection and evaluation of global change. sideration given that uncertainties in radiometric calibration
One of the missions recommended for earliest possible translate directly into the same amount of radiometric uncer-
implementation by the U.S. National Research Council (NRC) tainty in all products derived downstream. Uncertainties within
Decadal Survey report (NRC, 2007) is the Climate Absolute 2%3% (1) with respect to exo-atmospheric solar irradiance
Radiance and Refractivity Observatory (CLARREO), a joint are attainable routinely, relative to national laboratory stan-
NASA/NOAA mission. NOAAs contribution is to be sensors dards, and missions such as TRUTHS and CLARREO promise
for total and spectral solar irradiance measurements and Earth an order-of-magnitude improvement. Recommendations for
energy budget climate data records. NASAs contribution is to future research on vicarious calibration using terrestrial refer-
be sensors for the measurement of spectrally resolved thermal ence standard sites can be found in the next chapter (Teillet and
infrared and reflected solar radiation with high absolute accu- Chander, 2014).
racy. The CLARREO mission objective pertinent to the topic However, the retrieval of terrestrial surface variables is sub-
of this chapter is the provision of a space-based high-accuracy ject to many other considerations, as outlined extensively in this
calibration standard to enable calibration and intercalibra- chapter. Significant spatial and temporal variations in the opti-
tion of Earth observation satellite sensors. To accomplish this, cal properties of the Earths atmosphere necessitate image cor-
CLARREO will include a Reflected Solar Suite consisting of two rections for these effects in the solar-reflective spectral bands
push-broom hyperspectral imagers covering 3202300 nm in to put satellite data on the same radiometric scale for studies
a single instrument package. Reflectance will be obtained via intended to monitor terrestrial surfaces over time and space.
ratios of Earth-view data to solar-view data. High accuracy will Although the computational tools for image correction for
be achieved by precisely calibrating the instruments via the solar atmospheric effects are available, they require user inputs reli-
view combined with a sensor model to transfer National Institute ant on ready access to timely and accurate data on atmospheric
of Standards and Technology (NIST) laboratory standards to variables such as aerosol optical depth and water vapor content.
orbit (Anderson etal., 2008; Sandford etal., 2010; Thome etal., The net result is that surface reflectance retrieval from remotely
2010; Lukashin et al., 2013). As for TRUTHS, the CLARREO sensed image data is operational for larger programs (e.g.,
mission has the potential to improve the performance and accu- MODISVermote et al., 1997, 2002; Schaaf et al., 2002), but
racy of Earth observation missions by an order of magnitude (on remains laborious for individual user studies. It would be very
the order of 0.3% in the solar reflective domain). helpful if space agencies and data suppliers more often took the
The success of QA4EO is far from guaranteed, and the extra steps to generate and offer surface reflectance and other
TRUTHS and CLARREO missions have yet to be approved. In higher-level products to the user community at reasonable or
the early twenty-first century, innovation in Earth observation no cost. Progress can also come about if an agency with the per-
as well as the subfield of data standardization requires mak- tinent scientific and technical resources and specialists main-
ing judicious choices. As the geostrategic technologies of Earth tained a dynamic open-access atmospheric parameter server
observation seek adoption by mainstream information society, along the lines of NOMAD.
it is hoped that young people pursuing optical engineering, The chapter also described numerous geometric factors and
radiometry, metrology, and remote sensing physics will strive to phenomena that give rise to radiometric effects in satellite image
ensure the data quality and product validation needed for Earth data. Most of these geometric considerations are tractable, but
observation to underpin sound decision making. a few are more challenging to address, especially anisotropic
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 99
reflectance properties of surfaces and topographic effects. A Ahern, F.J., R.J. Brown, J. Cihlar, R. Gauthier, J. Murphy, R.A.
progressive perspective on BRDF is to view angular signatures Neville, and P.M. Teillet, 1989. Radiometric correction
of spectral reflectance as an information source as opposed of visible and infrared remote sensing data at the Canada
to something that has to be corrected (e.g., White et al., 2001, centre for remote sensing, in: A. Cracknell and L. Hayes,
2002b). Indeed, a greater number of satellite sensor systems eds., Remote Sensing Yearbook 1988/89, Taylor and Francis,
than can perform angular remote sensing would yield substan- Philadelphia, PA, pp. 101127.
tial information returns on investment (M. Verstraete, personal Ahern, F.J., P.M. Teillet, and D.G. Goodenough, 1979.
communication). Slope-aspect corrections will benefit from Transformation of atmospheric and solar illumination con-
the highest-resolution DEMs, which are becoming increasingly ditions on the CCRS image analysis system, in: Proceedings
available for the entire surface of the Earth.* of the Fifth International Symposium on Machine Processing
Although spectral considerations have received relatively of Remotely Sensed Data, West Lafayette, IN, pp. 3451.
less attention than other aspects of satellite image radiometry, Allen, T.R., 2000. Topographic normalization of Landsat Thematic
the literature has grown considerable over the past decade. The Mapper data in three mountain environments, Geocarto
increased attention is partly because of the advent of hyperspec- International, 15: 1319.
tral remote sensing (beyond the scope of this chapter to treat), Anderson, D., K.W. Jucks, and D.F. Young, 2008. The NRC decadal
which demands numerous preprocessing corrections. Also, survey climate absolute radiance and refractivity observa-
efforts devoted to radiometric calibration, a first-order consider- tory: NASA implementation, in: Proceedings of the IEEE
ation, have been well advanced by the research community such International Geoscience and Remote Sensing Symposium
that increased attention is now being devoted, belatedly, to spec- (IGARSS), Boston, MA, pp. 911.
tral considerations. Proper surface reflectance retrieval depends Barnes, R.A., R.E. Eplee, Jr., F.S. Patt, H.H. Kieffer, T.C. Stone, G.
on a good understanding of the state of sensor spectral band Meister, J.J. Butler, and C.R. McClain, 2004. Comparison
characteristics postlaunch, given that they have a direct impact of SeaWiFS measurements of the moon with the U.S.
on radiometric and atmospheric processing, as well as on the geological survey lunar model, Applied Optics, 43(31):
spectral properties of the scene (the atmosphere and the surface 58385854.
targets of interest). Berk, A., G.P. Anderson, P.K. Acharya, L.S. Bernstein, L.
Based on best practices and recommended for implementa- Muratov, J. Lee, M. Fox et al., 2006. MODTRAN5: 2006
tion and use throughout the GEO community, QA4EO is the update, in: Proceedings of SPIE Conference on Algorithms
way forward for sensor radiometric calibration. It is hoped that and Technologies for Multispectral, Hyperspectral, and
QA4EO will grow to encompass the other important elements Ultraspectral Imagery XII, Vol. 6233, Orlando, FL, p. 62331F.
of satellite image radiometry, including atmospheric, geometric, Biggar, S.F., P.N. Slater, and D.I. Gellman, 1994. Uncertainties
and spectral influences on Earth science data. in the in-flight calibration of sensors with reference to
measured ground sites in the 0.4 to 1.1 m range, Remote
Acknowledgments Sensing of Environment, 48: 245252.
Biggar, S.F., P.N. Slater, K.J. Thome, A.W. Holmes, and R.A. Barnes,
The author gratefully acknowledges guidance over the years 1993. Preflight solar-based calibration of SeaWiFS, in:
from Philip N. Slater and many substantive discussions on Proceedings of SPIE Conference 1939, Orlando, FL, pp. 233242.
remote sensing radiometry with Kurtis J. Thome, Nigel P. Fox, Bokoye, A.I., A. Royer, N.T. ONeill, P. Cliche, G. Fedosejevs,
Robert P. Gauthier, and Gunar Fedosejevs. The author also P.M. Teillet, and L.J.B. McArthur, 2001. Characterization of
thanks Gyanesh Chander for taking the time to read and pro- atmospheric aerosols across Canada from a ground-based
vide detailed comments on drafts of this chapter. sunphotometer network: AEROCAN, Atmosphere-Ocean,
39(4): 429456.
References Brest, C.L. and W.B. Rossow, 1992. Radiometric calibration and
monitoring of NOAA AVHRR data for ISCCP, International
Abel, P., B. Guenther, R.N. Galimore, and J.W. Cooper, 1993. Journal of Remote Sensing, 13(2): 235273.
Calibration results for NOAA-11 AVHRR channels 1and2 Brest, C.L., W.B. Rossow, and M.D. Roiter, 1997. Update of radi-
from congruent path aircraft observations, Journal of ance calibrations for ISCCP, Journal of Atmospheric and
Atmospheric and Oceanic Technology, 10(4): 493508. Oceanic Technology, 14(5): 10911109.
Adeline, K., M. Chen, X. Briottet, S. Pang, and N. Paparoditis, Bruegge, C. and J. Butler, eds., 1996. Special issue on earth
2013. Shadow detection in very high spatial resolu- observing system calibration, Journal of Atmospheric and
tion aerial images: A comparative study, ISPRS Journal of Oceanographic Technology, 13(2): 273544.
Photogrammetry and Remote Sensing, 80: 2138. Bruhl, C. and W. Zdunkowski, 1983. An approximate calculation
method for parallel and diffuse irradiances on inclined sur-
faces in the presence of obstructing mountains or buildings,
* For example, WorldDEM: http://www.astrium-geo.com/worlddem/ and Archives for Meteorology, Geophysics, and Bioclimatology
OpenTopography: http://www.opentopography.org/index.php. (Series B), 32(23): 111129.
2016 by Taylor & Francis Group, LLC
100 Remotely Sensed Data Characterization, Classification, and Accuracies
Brunger, A.P. and F.C. Hooper, 1993. Anisotropic sky radiance Chang, I.L., C. Dean, Z. Li, M. Weinreb, X. Wu, and P.A.V.B.
model based on narrow field of view measurements of Swamy, 2012. Refined algorithms for star-based monitor-
shortwave radiance, Solar Energy, 51(1): 5364, Erratum: ing of GOES Imager visible channel responsivities, in:
51(6): 523. Proceedings of SPIE Earth Observing Systems XVII, San
Burrows, W.E., 1999. This New Ocean: The Story of the First Space Diego, CA, p. 85100R.
Age, Modern Library Paperbacks, 752p. Chavez, P.S., 1988. An improved dark-object subtraction tech-
Butler, J.J., B.C. Johnson, and R.A. Barnes, 2005. The calibration nique for atmospheric scattering correction of multispec-
and characterization of Earth remote sensing and environ- tral data, Remote Sensing of Environment, 24: 459479.
mental monitoring instruments, in: A.C. Parr, R.U. Datla, Chavez, P.S., 1996. Image-based atmospheric correction
and J.L. Gardner, eds., Optical Radiometry, Experimental Revisited and improved, Photogrammetric Engineering and
Methods in the Physical Sciences, Vol. 41. R. Celotta and Remote Sensing, 62(9): 10251036.
T. Lucatorto, eds., Treatise, Elsevier/Academic Press, San Che, N. and J.C. Price, 1992. Survey of radiometric calibration
Diego, CA, pp. 453534. results and methods for visible and near-infrared channels
Cabot, F., O. Hagolle, and P. Henry, 2000. Relative and multitem- of NOAA-7, NOAA-9, and NOAA-11 AVHRRs, Remote
poral calibration of AVHRR, SeaWiFS, and VEGETATION Sensing of Environment, 41(1): 1927.
using POLDER characterization of desert sites, in: Chen, H.S., 1996. Remote Sensing Calibration SystemsAn
Proceedings of the International Geoscience and Remote Introduction, A. Deepak Publishing, Hampton, VA, ISBN:
Sensing Symposium 2000, Honolulu, HI, pp. 21882190. 0-937194-38-7.
Cao, C. and A.K. Heidinger, 2002. Inter-comparison of the long- Chen, J.M. and J. Cihlar, 1997. A hotspot function in a simple
wave infrared channels of MODIS and AVHRR/NOAA-16 bidirectional reflectance model for satellite applications,
using simultaneous nadir observations at orbit intersec- Journal of Geophysical ResearchAtmospheres, 102(D22):
tions, in: W.L. Barnes, ed., Proceedings of SPIE Conference 2590725913.
on Earth Observing Systems VII, Vol. 4814, Seattle, WA. Chu, W.P., 1983. Calculations of atmospheric refraction for space-
SPIE, Bellingham, WA, pp. 306316. craft remote sensing applications, Applied Optics, 22(5):
Cavayas, F., 1987. Modelling and correction of topographic effect 721725.
using multi-temporal satellite images, Canadian Journal of Cihlar, J. and P.M. Teillet, 1995. Forward piecewise linear calibra-
Remote Sensing, 13(2): 4967. tion model for quasi-real-time processing of AVHRR data,
Cavayas, F., G. Rochon, and P. Teillet, 1983. Estimation des reflec- Canadian Journal of Remote Sensing, 21(1): 2227.
tances bidirectionnelles par analyse des images Landsat: Connolly, J.I. and B. Tolar, 1994. Standards and calibration,
Problmes et possibilits de solutions, in: Comptes ren- Workshop II, in: Proceedings of the First International
dus du 8me Symposium canadien sur la tldtection et Symposium on Spectral Sensing Research (ISSSR), San Diego,
4me Congres de lAssociation qubcoise de tldtection, CA, pp. 87111.
Montral, Qubec, Canada, p. 645. Cracknell, A.P., 1998. Synergy in remote sensingWhats in
Chahine, M.T., 1983. Interaction mechanisms within the atmo- a pixel? International Journal of Remote Sensing, 19(11):
sphere, in: R.N. Colwell, ed., Chapter 5: Manual of Remote 20252047.
Sensing, 2nd edn., American Society of Photogrammetry, Crist, E.P., 1985. A TM Tasseled cap equivalent transformation
Falls Church, VA. for reflectance factor data, Remote Sensing of Environment,
Chander, G., T.J. Hewison, N. Fox, X. Wu, X. Xiong, and W.J. 17: 301306.
Blackwell, 2013a. Overview of intercalibration of satellite Dana, R.W., 1982. Background reflectance effects in Landsat data,
instruments, IEEE Transactions on Geoscience and Remote Applied Optics, 21(22): 41064111.
Sensing, 51(3 SI): 10561080. Dave, J.V., 1977. Validity of the isotropic-distribution approxi-
Chander, G., T.J. Hewison, N. Fox, X. Wu, X. Xiong, and W.J. mation in solar energy estimations, Solar Energy, 19(4):
Blackwell, Guest Editors, 2013b. Special issue on inter- 331333.
calibration of satellite instruments, IEEE Transactions on Dave, J.V., 1980. Effect of atmospheric conditions on remote
Geoscience and Remote Sensing, 51(3 SI): 491p. sensing of a surface nonhomogeneity, Photogrammetric
Chander, G., N. Mishra, D.L. Helder, D. Aaron, A. Angal, T. Choi, Engineering and Remote Sensing, 46: 11731180.
X. Xiong, and D. Doelling, 2013c. Applications of Spectral Deschamps, P.Y., M. Herman, and D. Tanr, 1983. Definitions of
Band Adjustment Factors (SBAF) for cross-calibration, atmospheric radiance and transmittances in remote sens-
IEEE Transactions on Geoscience and Remote Sensing, ing, Remote Sensing of Environment, 13(1): 8992.
51(3 SI): 12671281. Dilligeard, E., X. Briottet, J.L. Deuze, and R.P. Santer, 1997. SPOT
Chander, G. and P.M. Teillet, Guest Editors and N. Coops, Editor- calibration of blue and green channels using Rayleigh scat-
in-Chief, 2010. Special issue on terrestrial reference stan- tering over clear oceans, in: Proceedings of SPIE Conference
dard test sites for postlaunch calibration, Canadian Journal on Advanced Next-Generation Satellites II, Taormina, Italy,
of Remote Sensing, 36(5): 437630, doi:10.5589/cjrs3605fi. pp. 373379.
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 101
Dinguirard M., J. Mueller, F. Sirou, and T. Tremas, 1997. Flittner, D.E. and P.N. Slater, 1991. Stability of narrow-band filter
Comparison of ScaRaB ground calibration in the short radiometers in the solar-reflective range, Photogrammetric
wave and long wave domains, Metrologia, 35: 597601 Engineering and Remote Sensing, 57(2), 165171.
(Special Issue on NEWRAD97, Tucson, AZ). Forster, B.C. and J.C. Trinder, 1984. An examination of the effects
Dinguirard, M. and P.N. Slater, 1999. Calibration of space- of resampling on classification accuracy, in: Proceedings of
multispectral imaging sensors: A review, Remote Sensing of the Third Australasian Remote Sensing Conference, Brisbane,
Environment, 68(3): 194205. Queensland, Australia, pp. 106115.
Doelling, D.R., G. Hong, D. Morstad, R. Bhatt, A. Gopalan, and Fougnie, B. and R. Bach, 2009. Monitoring of radiometric sensi-
X. Xiong, 2010. The characterization of deep convective tivity changes of space sensors using deep convective clouds:
cloud albedo as a calibration target using MODIS reflec- Operational application to PARASOL, IEEE Transactions
tances, in: X. Xiong, C. Kim, and H. Shimoda, eds., on Geoscience and Remote Sensing, 47(3): 851 861.
Proceedings of SPIE Conference on Earth Observing Missions Fox, N., J. Aiken, J.J. Barnett, X. Briottet, R. Carvell, C. Frohlich,
Sensors: Development, Implementation, and Characteriza- S.B. Groom etal., 2002. Traceable radiometry underpinning
tion, Vol. 7862, Incheon, Republic of Korea, p. 78620I. terrestrial- and helio-studies (TRUTHS), in: Proceedings of
Doelling, D.R., C. Lukashin, P. Minnis, B. Scarino, and D. Morstad, SPIE Conference on Sensors, Systems, and Next-Generation
2012. Spectral reflectance corrections for satellite intercali- Satellites VIII, Vol. 4881, Heraklion, Crete, Greece, 12p.
brations using SCHIAMACHY data, IEEE Geoscience and Fox, N., A. Kaiser-Weiss, W. Schmutz, K. Thome, D. Young, B. Wielicki,
Remote Sensing Letters, 9(1): 119123. R. Winkler, and E. Woolliams, 2011. Accurate radiometry
Doelling, D.R., L. Nguyen, and P. Minnis, 2004. On the use of from space: An essential tool for climate studies, Philosophical
deep convective clouds to calibrate AVHRR data, in: W.L. Transactions of the Royal Society AMathematical, Physical,
Barnes and J.J. Butler, eds., Proceedings of SPIE Conference and Engineering Sciences, 369(1953): 40284063.
on Earth Observing Systems IX, Vol. 5542, Denverr, CO, Fox, N.P., 1999. Improving the accuracy and traceability of radio-
pp. 281289. metric measurements to SI for remote sensing instrumen-
Duggin, M.J., 1986. Variance in radiance recorded from hetero- tation, in: Proceedings of the Fourth International Airborne
geneous targets in the optical-reflective, middle-infrared, Remote Sensing Conference and Exhibition/21st Canadian
and thermal-infrared regions, Applied Optics, 25(23): Symposium on Remote Sensing, Vol. I, Ottawa, Ontario,
42464252. Canada. ERIM International, pp. 304311.
Duggin, M.J., 1987. Impact of radiance variations on satellite sen- Fraser, R.S. and Y.J. Kaufman, 1986. Calibration of satellite sen-
sor calibration, Applied Optics, 26(7): 12641271. sors after launch, Applied Optics, 25(7): 11771185.
Duggin, M.J. and C.J. Robinove, 1990. Assumptions implicit in Freemantle, J., N.T. ONeill, P.M. Teillet, A. Royer, J.-P. Blanchet,
remote sensing data acquisition and analysis, International M. Aub, S. Thulasiraman, F. Vachon, S. Gong, and
Journal of Remote Sensing, 11(10): 16691694. M. Versi, 2002. Using web services for atmospheric cor-
Duguay, C.R., 1993. Radiation modeling in mountainous terrain rection of remote sensing data, in: Proceedings of the 2002
review and status, Mountain Research and Development, International Geoscience and Remote Sensing Symposium
13(4): 339357. (IGARSS02) and the 24th Canadian Symposium on Remote
Dymond, J.R. and J.D. Shepherd, 1999. Correction of the topo- Sensing, Vol. V, Toronto, Ontario, Canada, pp. 29392941,
graphic effect in remote sensing, IEEE Transactions on also on CD-ROM.
Geoscience and Remote Sensing, 37: 26182620. Frouin, R. and C. Gautier, 1987. Calibration of NOAA-7 AVHRR,
Edwards, D.P., L.K. Emmons, D.A. Hauglustaine, D.A. Chu, J.C. GOES-5, and GOES-6 VISSR/VAS solar channels, Remote
Gille, Y.J. Kaufman, G. Petron et al., 2004. Observations Sensing of Environment, 22(1): 73101.
of carbon monoxide and aerosols from the Terra satellite: Giles, P., 2001. Remote sensing and cast shadows in mountainous
Northern Hemisphere variability, Journal of Geophysical terrain, Photogrammetric Engineering and Remote Sensing,
Research, 109: D24202, doi:10.1029/2004JD004727. 67(7): 833840.
Egan, W.G., 1985. Photometry and Polarization in Remote Sensing, Godden, G.D. and C.A. McKay, 1997. A strategy for observing the
Elsevier, New York, 503p. moon to achieve precise radiometric stability monitoring,
Ekstrand, S., 1996. Landsat TM-based forest damage assess- Canadian Journal of Remote Sensing, 23(4): 333341.
ment: Correction for topographic effects, Photogrammetric Goetz, A.F.H., K.B. Heidebrecht, and T.G. Chrien, 1995. High
Engineering and Remote Sensing, 62(2): 151161. accuracy in-flight wavelength calibration of imaging
Fedosejevs, G., N.T. ONeill, A. Royer, P.M. Teillet, A.I. Bokoye, spectrometry data, in: Summaries of the Fifth Annual JPL
and B. McArthur, 2000. Aerosol optical depth for atmo- Airborne Earth Science Workshop, AVIRIS Workshop, Vol. 1,
spheric correction of AVHRR composite data, Canadian Pasadena, CA, pp. 6769.
Journal of Remote Sensing, 26(4): 273284. Gordon, H.R., 1987. Calibration requirements and methodology
Fisher, P., 1997. The pixel: A snare and a delusion, International for remote sensors viewing the ocean in the visible, Remote
Journal of Remote Sensing, 18(3): 679685. Sensing of Environment, 22: 103126.
2016 by Taylor & Francis Group, LLC
102 Remotely Sensed Data Characterization, Classification, and Accuracies
Gordon, H.R. and M. Wang, 1994. Retrieval of water-leaving Hodgson, M.E. 1995. What Cell Size Does the Computed Slope/
radiance and aerosol optical thickness over the oceans with Aspect Angle Represent?, Photogrammetric Engineering and
SeaWiFS: A preliminary algorithm, Applied Optics, 33(3): Remote Sensing, 61(5): 513517.
443452. Holben, B.N., T.F. Eck, I. Slutsker, D. Tanr, J.P. Buis, A. Setzer,
Gu, D. and A. Gillespie, 1998. Topographic normalization of E. Vermote et al., 1998. AERONETA federated instru-
Landsat TM images of forest based on sub-pixel sun-can- ment network and data archive for aerosol characterization,
opy-sensor geometry, Remote Sensing of Environment, 64: Remote Sensing of Environment, 66: 116.
166175. Holben, B.N. and C.O. Justice, 1981. An examination of spectral
Guenther, B., W. Barnes, E. Knight, J. Barker, J. Harnden, R. Weber, band rationing to reduce the topographic effect on remotely
M. Roberto, G. Godden, H. Montgomery, and P. Abel, 1996. sensed data, International Journal of Remote Sensing, 2(2):
MODIS calibration: A brief review of the strategy for the 115133.
at-launch calibration approach, Journal of Atmospheric and Holben, B.N., Y.J. Kaufman, and J.D. Kendall, 1990. NOAA-
Oceanographic Technology, 13(2): 274285. 11 AVHRR visible and near-IR inflight calibra-
Guenther, B.W., 1991. Calibration of passive remote observing tion, International Journal of Remote Sensing, 11(8):
optical and microwave instrumentation, in: Proceedings 15111519.
of SPIE, Vol. 1493, Orlando, FL. SPIE, The International Hooper, F.C. and A.P. Brunger, 1980. A model for the angu-
Society for Optical Engineering, Bellingham, WA, 304p. lar distribution of sky radiance, Journal of Solar Energy
Guenther, B.W., J. Butler, and P. Ardanuy, eds., 1997. Workshop Engineering, 102: 196202.
on Strategies for Calibration and Validation of Global Hooper, F.C., A.P. Brunger, and C.S. Chan, 1987. A clear sky model
Change Measurements, May 1012, 1995, NASA Reference of diffuse sky radiance, Journal of Solar Energy Engineering,
Publication 1397, NASA/GSFC, Greenbelt, MD, 125p. 109: 914.
Hall, F.G., D.E. Strebel, J.E. Nickerson, and S.J. Goetz, 1991. Horler, D.N.H., 1996. Framework study on calibration/validation
Radiometric rectification: Toward a common radiometric user requirements as part of the ground infrastructure
response among multidate, multisensor images, Remote program of the Canadian long-term space plan, Final
Sensing of Environment, 35: 1127. Report to the Canada Centre for Remote Sensing, Ottawa,
Hay, J.E. and D.C. McKay, 1985. Estimating solar irradiance Ontario, Canada, by Horler Information Inc. for Contract
on inclined surfaces: A review and assessment of meth- 23413-5-E178/01-SQ, 53p.
odologies, International Journal of Solar Energy, 3(45): Horler, D.N.H. and P.M. Teillet, 1996. Workshop on Canadian
203240. Earth Observation Calibration and Validation, Workshop
Hay, J.E., R. Perez, and D.C. McKay, 1986. Addendum and Errata Report, Canada Centre for Remote Sensing, Ottawa,
to the Paper Estimating solar irradiance on inclined Ontario, Canada, 59p.
surfaces: A review and assessment of methodologies, Hu, Y., B.A. Wielicki, P. Yang, P.W. Stackhouse, Jr., B. Lin, and
International Journal of Solar Energy, 4(1): 321324. D.F. Young, 2004. Application of deep convective cloud
He, X., D. Pan, Y. Bai, and F. Gong, 2006. A general purpose exact albedo observation to satellite-based study of the terres-
Rayleigh scattering look-up table for ocean color remote trial atmosphere: Monitoring the stability of spaceborne
sensing, Acta Oceanologica Sinica, 25(1): 4856. measurements and assessing absorption anomaly, IEEE
Hegyi, F., 2004. Alignment of earth observation calibration and Transactions on Geoscience and Remote Sensing, 42(11):
validation activities, Report delivered by Hegyi Geomatics 25942599.
International Inc. (HGI) to the Canada Centre for Remote Hugli, H. and W. Frei, 1983. Understanding anisotropic reflec-
Sensing, March 2004, Ottawa, Ontario, Canada, 47p. tance in mountainous terrain, Photogrammetric Engineering
Helder, D., K. Thome, D. Aaron, L. Leigh, J. Czapla-Myers, and Remote Sensing, 49(5): 671683.
N. Leisso, S. Biggar, and N. Anderson, 2012. Recent sur- Iwabuchi, H., 2003. Calibration of the visible and near-infrared
face reflectance measurement campaigns with emphasis on channels of NOAA-11 and -14 AVHRRs by using reflections
best practices, SI traceability and uncertainty estimation, from molecular atmosphere and stratus cloud, International
Metrologia, 49(2): S21S28. Journal of Remote Sensing, 24(24): 53675378.
Helder, D., K. Thome, N. Mishra, G. Chander, X. Xiong, A.Angal, Jackson, R.D., ed., 1990. Special issue on coincident satellite, air-
and T. Choi, 2013. Absolute radiometric calibration of craft, and field measurements at the Maricopa Agricultural
Landsat using a pseudo invariant calibration site, IEEE Center (MAC), Remote Sensing of Environment, 32(2&3):
Transactions on Geoscience and Remote Sensing, 51(3 SI): 77228.
13601369. Jensen, J.R., 2006. Remote Sensing of the Environment: An Earth
Henry, P., G. Chander, B. Fougnie, C. Thomas, and X. Xiong, 2013. Resources Perspective, 2nd edn., Prentice Hall, Upper Saddle
Assessment of spectral band impact on inter-calibration River, NJ, 608p.
over desert sites using simulation based on EO-1 Hyperion Kaufman, Y.J. and R.S. Fraser, 1984. Atmospheric effects on clas-
data, IEEE Transactions on Geoscience and Remote Sensing, sification of finite fields, Remote Sensing of Environment, 15:
51(3 SI): 12971308. 95118.
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 103
Kaufman, Y.J. and B.N. Holben, 1993. Calibration of the AVHRR Loeb, N.G., 1997. In-flight calibration of NOAA AVHRR vis-
visible and near-IR bands by atmospheric scattering, ocean ible and near-IR bands over Greenland and Antarctica,
glint and desert reflection, International Journal of Remote International Journal of Remote Sensing, 18(3): 477490.
Sensing, 14(1): 2152. Luderer, G., J.A. Coakley, Jr., and W.R. Tahnk, 2005. Using sun
Kaufman, Y.J., D. Tanr, H.R. Gordon, T. Nakajima, J. Lenoble, glint to check the relative calibration of reflected spectral
R. Frouin, H. Grassl, B.M., Herman, M.D. King, and P.M. radiances, Journal of Atmospheric and Oceanic Technology,
Teillet, 1997. Passive remote sensing of tropospheric aerosol 22(10): 14801493.
and atmospheric correction for the aerosol effect, Journal of Lukashin, C., B. Wielicki, D. Young, K. Thome, Z. Jin, and W. Sun,
Geophysical Research, 102(D14): 1681516830. 2013. Uncertainty estimates for imager reference inter-cali-
Kieffer, H.H., T.C. Stone, R.A. Barnes, S. Bender, R.E. Eplee, bration with CLARREO reflected solar spectrometer, IEEE
Jr., J. Mendenhall, and L. Ong, 2002. On-orbit radiomet- Transactions on Geoscience and Remote Sensing, 51(3 SI):
ric calibration over time and between spacecraft using 14251436.
the moon, in: Proceedings of SPIE Conference on Sensors, MacDonald, J.S., 1997. From space data to information, in:
Systems, and Next-Generation Satellites VI, Crete, Greece, G. Konecny, ed., Proceedings of the ISPRS Joint Workshop
pp. 287298. on Sensors and Mapping From Space, Institute of
Kieffer, H.H. and R.L. Wildey, 1985. Absolute calibration of Photogrammetry and Engineering Surveys, University of
Landsat instruments using the moon, Photogrammetric Hannover, Hannover, Germany, pp. 233240.
Engineering and Remote Sensing, 51(9): 13911393. Malila, W.A. and D.M. Anderson, 1986. Satellite Data Availability
Kieffer, H.H. and R.L. Wildey, 1996. Establishing the moon as and Calibration Documentation for Land Surface
a spectral radiance standard, Journal of Atmospheric and Climatology Studies, Environmental Research Institute of
Oceanic Technology, 13(2): 360375. Michigan, Contract NAS5-28715, Report No. 180300-1-F,
Kimes, D.S. and J.A. Kirchner, 1981. Modeling the effects of NASA/GSFC, Greenbelt, MD, 214p.
various radiant transfers in mountainous terrain on sen- Markham, B.L. and J.L. Barker, eds., 1985. Special issue on
sor response, IEEE Transactions on Geoscience and Remote Landsat image data quality assessment, Photogrammetric
Sensing, 19: 100108. Engineering and Remote Sensing, 51: 12451493.
Kirchner, J.A., S. Youkhana, and J.A. Smith, 1982. Influence of sky Markham, B.L. and J.L. Barker, 1987. Radiometric properties
radiance distribution on the ratio technique for estimating of U.S. processed Landsat MSS data, Remote Sensing of
bidirectional reflectance, Photogrammetric Engineering and Environment, 22: 3971.
Remote Sensing, 48: 955959. Markham, B.L., W.C. Boncyk, D.L. Helder, and J.L. Barker, 1997.
Klucher, T.M., 1979. Evaluation of models to predict insolation on Landsat-7 enhanced thematic mapper plus radiometric cali-
tilted surfaces, Solar Energy, 23(2): 111114. bration, Canadian Journal of Remote Sensing, 23(4): 318332.
Kotchenova, S.Y. and E.F. Vermote, 2007. Validation of a vector McCartney, E.J., 1976. Optics of the Atmosphere, John Wiley, New
version of the 6S radiative transfer code for atmospheric cor- York, 408p.
rection of satellite data. Part II: Homogeneous Lambertian McCorkel, J., K. Thome, and R. Lockwood, 2013. Absolute
and anisotropic surfaces, Applied Optics, 46(20): 44554464. radiometric calibration of narrow-swath imaging sen-
Kotchenova, S.Y., E.F. Vermote, R. Matarrese, and F. Klemm, 2006. sors with reference to non-coincident wide-swath sensors,
Validation of a vector version of the 6S radiative transfer IEEE Transactions on Geoscience and Remote Sensing, 51(3
code for atmospheric correction of satellite data. Part I: Path SI): 13091318.
radiance, Applied Optics, 45: 67626774. Meckler, Y. and Y.J. Kaufman, 1982. Contrast reduction by the
Kramer, H.J., 2001. Observation of the Earth and Its Environment: atmosphere and retrieval of nonuniform surface reflec-
Survey of Missions and Sensors, 4th edn., Springer-Verlag, tance, Applied Optics, 21: 310316.
Berlin, Heidelberg, New York, 1510p. Meyer, P., K.I. Itten, T. Kellenberger, S. Sandmeier, and
Li, F., D.L.B. Jupp, M. Thankappan, L. Lymburner, N. Mueller, R. Sandmeier, 1993. Radiometric corrections of topo-
A.Lewis, and A. Held, 2012. A physics-based atmospheric graphically induced effects on Landsat TM data in an alpine
and BRDF correction for Landsat data over mountainous environment, ISPRS Journal of Photogrammetry and Remote
terrain, Remote Sensing of Environment, 124: 756770. Sensing, 48(4): 1728.
Li, X. and A.H. Strahler, 1992. Geometric-optical bidirectional reflec- Meygret, A., X. Briottet, P.J. Henry, and O. Hagolle, 2000.
tance modeling of the discrete crown vegetation canopy: Effect Calibration of SPOT4 HRVIR and vegetation cameras over
of crown shape and mutual shadowing, IEEE Transactions on Rayleigh scattering, in: W.L. Barnes, ed., Proceedings of SPIE
Geoscience and Remote Sensing, 30(2): 276292. Conference on Earth Observing Systems V, Vol. 4135, San
Liang, S., H. Fallah-Adl, S. Kalluri, J. JaJa, Y.J. Kaufman, and Diego, CA. SPIE, Bellingham, WA, pp. 302313.
J.R.G. Townshend, 1997. An operational atmospheric Mitchell, R.M., D.M. OBrien, and B.W. Forgan, 1992. Calibration
correction algorithmfor Landsat Thematic Mapper imag- of the NOAA AVHRR shortwave channels using split pass
ery over the land, Journal of Geophysical Research, 102: imagery: I. Pilot study, Remote Sensing of Environment,
1717317186. 40(1): 5765.
2016 by Taylor & Francis Group, LLC
104 Remotely Sensed Data Characterization, Classification, and Accuracies
Morain, S.A. and A.M. Budge, eds., 2004. Postlaunch Calibration Otterman, J., M. Dishon, and S. Rehavi, 1983. Point spread func-
of Satellite Sensors, ISPRS Book Series, Vol. 2, Proceedings of tions in imaging a Lambert surface from zenith through
the International Workshop on Radiometric and Geometric a thin scattering layer, International Journal of Remote
Calibration, Gulfport, MS. A.A. Balkema Publishers, New Sensing, 4: 583.
York, 193p. Otterman, J. and R.S. Fraser, 1979. Adjacency effects on imag-
Moran, M.S., R.D. Jackson, P.N. Slater, and P.M. Teillet, 1992. ing by surface reflection and atmospheric scattering: Cross
Evaluation of atmospheric correction procedures for visible radiance to zenith, Applied Optics, 18: 2852.
and near-infrared satellite sensor output, Remote Sensing of Pinty, B., J.-L. Widlowski, M. Taberner, N. Gobron, M.M.
Environment, 41: 169184. Verstraete, M. Disney, F. Gascon etal., 2004. Radiation trans-
Mueller, J., R. Stulhmann, R. Becker, E. Raschke, J.L. Monge, and fer model intercomparison (RAMI) exercise: Results from
P. Burkert, 1996. Ground based calibration facility for the the second phase, Journal of Geophysical Research, 109: D06.
scanner for radiation budget instrument in the solar spec- Pollock, D.B., T.L. Murdock, R.U. Datla, and A. Thompson, 2003.
tral domain, Metrologia, 32: 657660. Data uncertainty traced to SI units. Results reported in
Muench, H.S., 1981. Calibration of geosynchronous satellite the International System of Units, International Journal of
video sensors, Report No. AFGL-TR-81-0050, Air Force Remote Sensing, 24(2): 225235.
Geophysical Laboratory, Hanscom, MA. Price, J.C., 1987a. Radiometric calibration of satellite sensors in
Nicodemus, F.E., 1965. Directional reflectance and emissivity of the visible and near infrared: History and outlook, Remote
an opaque surface, Applied Optics, 4(7): 767775. Sensing of Environment, 22: 39.
Nicodemus, F.E., 1970. Reflectance nomenclature and directional Price, J.C., ed., 1987b. Special issue on radiometric calibration of
reflectance and emissivity, Applied Optics, 9(6): 14741475. satellite data, Remote Sensing of Environment, 22(1): 1158.
Nieke, J., T. Aoki, T. Tanikawa, H. Motoyoshi, M. Hori, and Y. Proy, C., D. Tanr, and P.Y. Deschamps, 1989. Evaluation of topo-
Nakajima, 2003. Cross-calibration of satellite sensors graphic effects in remotely sensed data, Remote Sensing of
over snow fields, in: W.L. Barnes, ed., Proceedings of SPIE Environment, 30(1): 2132.
Conference on Earth Observing Systems VIII, Vol. 5151, San Rao, C.R.N., C. Cao, and N. Zhang, 2003. Inter-calibration of the
Diego, CA. SPIE, pp. 406414. moderate-resolution imaging spectroradiometer and the
Nithianandam, J., B.W. Guenther, and L.J. Allison, 1993. An anec- along-track scanning radiometer-2, International Journal of
dotal review of NASA earth observation satellite remote Remote Sensing, 24(9): 19131924.
sensors and radiometric calibration methods, Metrologia, Rao, C.R.N. and J. Chen, 1995. Intersatellite calibration linkages
30: 207212. for the visible and near-infared channels of the Advanced
NRC, 2007. Earth Science and Applications from Space: National Very High-Resolution Radiometer on the NOAA-7, NOAA-
Imperatives for the Next Decade and Beyond, National Research 9, and NOAA-11 spacecraft, International Journal of Remote
Council, The National Academies, Washington, DC, 428p. Sensing, 16(11): 19311942.
OBrien, D.M. and R.M. Mitchell, 2001. An error budget for cross- Rao, C.R.N. and J.H. Chen, 1996. Postlaunch calibration of the
calibration of AVHRR shortwave channels against ATSR-2, visible and near-infrared channels of the Advanced Very
Remote Sensing of Environment, 75(2): 216229. High-Resolution Radiometer on the NOAA-14 space-
ONeill, N.T., J.R. Miller, and F.J. Ahern, 1978. Radiative transfer craft, International Journal of Remote Sensing, 17(14):
calculations for remote sensing applications, in: Proceedings 27432747.
of the Fifth Canadian Symposium on Remote Sensing, Rao, C.R.N. and J.H. Chen, 1999. Revised postlaunch cali-
Victoria, British Columbia, Canada, p. 572. bration of the visible and near-infrared channels of the
ONeill, N.T., P.M. Teillet, A. Royer, J.-P. Blanchet, M. Aub, J. Advanced Very High-Resolution Radiometer (AVHRR) on
Freemantle, S. Gong, D. Stanley, S. Thulasiraman, and the NOAA-14 spacecraft, International Journal of Remote
F. Vachon, 2002. Concept of a central optical param- Sensing, 20(18): 34853491.
eter server for atmospheric corrections of remote sensing Regalado, A., 2013. The data made me do it, MIT Technology
data, in: Proceedings of the 2002 International Geoscience Review, 116(4): 6364.
and Remote Sensing Symposium (IGARSS02) and the 24th Riao, D., E. Chuvieco, J. Salas, and I. Aguado, 2003.
Canadian Symposium on Remote Sensing, Vol. V, Toronto, Assessment of different topographic corrections in
Ontario, Canada, pp. 29512953, also on CD-ROM. Landsat-TM data for mapping vegetation types (2003),
Ohring, G., J. Tansock, W. Emery, J. Butler, L. Flynn, F. Weng, IEEE Transactions on Geoscience and Remote Sensing,
K.S. Germain et al., 2007. Achieving satellite instrument 41(5): 10561061.
calibration for climate change, EOS, Transactions American Richardson, A.J., 1982. Relating Landsat digital count values to
Geophysical Union, 88(11): 136. ground reflectances for optical thin atmospheric condi-
Ohring, G., B. Wielicki, R. Spencer, B. Emery, and R. Datla, 2005. tions, Applied Optics, 21: 1457.
Satellite instrument calibration for measuring global cli- Richter, R., 1997. Correction of atmospheric and topo-
mate change: Report of a workshop, Bulletin of the American graphic effects for high spatial resolution satellite images,
Meteorological Society, 86: 13031313. International Journal of Remote Sensing, 18: 10991111.
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 105
Richter, R., 1998. Correction of satellite imagery of mountainous Slater, P.N., 1980. Remote Sensing, Optics and Optical Systems,
terrain, Applied Optics, 37(18): 40044014. Addison-Wesley Publishing Company, Reading, MA.
Sandford, S.P., D.F. Young, J.M. Corliss, B.A. Wielicki, M.J. Slater, P.N., 1984. The importance and attainment of absolute
Gazarik, M.G. Mlynczak, A.D. Little et al., 2010. radiometric calibration, in: Proceedings of SPIE Critical
CLARREO: Cornerstone of the climate observing system Review of Remote Sensing, Vol. 475, Arlingrton, VA,
measuring decadal change through accurate emitted infra- pp.3440.
red and reflected solar spectra and radio occultation, in: Slater, P.N., 1985. Radiometric considerations in remote sensing,
R.Meynart, S.P. Neeck, and H. Shimoda, eds., Proceedings of Proceedings of IEEE, 73(6): 9971011.
SPIE Conference on Sensors, Systems, and Next-Generation Slater, P.N., 1988a. Review of the calibration of radiometric
Satellites XIV, Vol. 7826, Toulouse, France, p. 782611. measurements from satellite to ground level, International
Scarpace, F.L. and P.R. Wolf, 1973. Atmospheric refraction, Archives of Photogrammetry and Remote Sensing, 27(B11):
Photogrammetric Engineering, 39: 521. 726724.
Schaaf, C.B., F. Gao, A.H. Strahler, W. Lucht, X. Li, T. Tsang, Slater, P.N., 1988b. Recent advances in sensors, radiometry, and
N.C. Strugnell etal., 2002. First operational BRDF, albedo data processing for remote sensing, in: Proceedings of SPIE,
nadir reflectance products from MODIS, Remote Sensing of Vol. 924, Orlando, FL. SPIE, The International Society for
Environment, 83(12): 135148. Optical Engineering, Bellingham, WA, 341p.
Schott, J.R., 2007. Remote Sensing, The Image Chain Approach, Slater, P.N. and S.F. Biggar, 1996. Suggestions for radiometric cali-
2nd edn., Oxford University Press, New York. bration coefficient generation, Journal of Atmospheric and
Schott, J.R., C. Salvaggio, and W.J. Volchok, 1988. Radiometric Oceanographic Technology, 13(2): 376382.
scene normalization using pseudo-invariant features, Slater, P.N., S.F. Biggar, R.G. Holm, R.D. Jackson, Y. Mao, M.S.
Remote Sensing of Environment, 26: 116. Moran, J.M. Palmer, and B. Yuan, 1987. Reflectance-
Schowengerdt, R.A., 2007. Remote Sensing: Models and Methods and radiance-based methods for the in-flight absolute
for Image Processing, 3rd edn., Academic Press, San Diego, calibration of multispectral sensors, Remote Sensing of
CA, 515p. Environment, 22: 1137.
Schowengerdt, R.A., C. Archwamety, and R.C. Wrigley, 1985. Landsat Slater, P.N., S.F. Biggar, J.M. Palmer, and K.J. Thome, 1995. Unified
thematic mapper image-derived MTF, Photogrammetric approach to pre- and in-flight satellite-sensor absolute
Engineering and Remote Sensing of Environment, 51(9): radiometric calibration, in: Proceedings of the SPIE Europto
13951406. Symposium, Vol. 2583, SPIE, Paris, France, pp. 130141.
Schowengerdt, R.A., S.K. Park, and R.T. Gray, 1983. An optimized Slater, P.N., S.F. Biggar, J.M. Palmer, and K.J. Thome, 2001.
cubic interpolator for image resampling, in: Proceedings of Unified approach to absolute radiometric calibration in the
the 17th International Symposium on Remote Sensing of solar-reflective range, Remote Sensing of Environment, 77:
Environment, Ann Arbor, MI, p. 1291. 293303.
Schroeder, T.A., W.B. Cohen, C. Song, M.J. Canty, and Z. Yang, Slater, P.N. and R.D. Jackson, 1982. Atmospheric effects on radia-
2006. Radiometric correction of multi-temporal Landsat data tion reflected from soil and vegetation as measured by
for characterization of early successional forest patterns in orbital sensors using various scanning directions, Applied
western Oregon, Remote Sensing of Environment, 103: 1626. Optics, 21: 3923.
Secker, J., K. Staenz, R.P. Gauthier, and B. Budkewitsch, 2001. Smith, G.R., R.H. Levin, P. Abel, and H. Jacobowitz, 1988.
Vicarious calibration of hyperspectral sensors in operational Calibration of the solar channels and NOAA-9 AVHRR
environments, Remote Sensing of Environment, 26: 8192. using high altitude aircraft measurements, Journal of
Sellers, P.J., B.W. Meeson, F.G. Hall, G. Asrar, R.E. Murphy, Atmospheric and Oceanic Technology, 5: 631639.
R.A. Schiffer, F.P. Bretherton et al., 1995. Remote sensing Smith, J.A., T.L. Lin, and K.J. Ranson, 1980. The Lambertian
of the land surface for studies of global change: Models assumption and Landsat data, Photogrammetric Engineering
AlgorithmsExperiments, Remote Sensing of Environment, and Remote Sensing, 46: 11831189.
51: 326. Soenen, S., D. Peddle, C. Coburn, R. Hall, and F. Hall, 2008.
Shepherd, J.D. and J.R. Dymond, 2003. Correcting satellite imag- Improved topographic correction of forest image data using
ery for the variance of reflectance and illumination with a 3-d canopy reflectance model in multiple forward mode.
topography, International Journal of Remote Sensing, 24(17): International Journal of Remote Sensing, 29(4): 11071027.
35033514. Soenen, S.A., D.R. Peddle, and C.A. Coburn, 2005. A modified
Six, D., M. Fily, S. Alvain, P. Henry, and J.P. Benoist, 2004. Surface sun-canopy-sensor topographic correction in forested ter-
characterisation of the Dome Concordia area (Antarctica) rain, IEEE Transactions on Geoscience and Remote Sensing,
as a potential satellite calibration site, using SPOT4/ 43(9): 21482159.
VEGETATION instrument, Remote Sensing of Environment, Song, C., C.E. Woodcock, K.C. Seto, M. Pax-Lenney, and S.A.
89(1): 8394. Macomber, 2001. Classification and change detection using
Sjoberg, R.W. and B.K.P. Horn, 1983. Atmospheric effects in satel- Landsat TM data: When and how to correct atmospheric
lite imaging of mountainous terrain, Applied Optics, 22: 1702. effects, Remote Sensing of Environment, 75: 230244.
2016 by Taylor & Francis Group, LLC
106 Remotely Sensed Data Characterization, Classification, and Accuracies
Souri, A.H. and A. Azizi, 2013. Removing bowtie phenomenon Teillet, P.M., 1986. Image correction for radiometric effects in
by correction of panoramic effect in MODIS imagery, remote sensing, International Journal of Remote Sensing, 7:
International Journal of Computer Applications, 68(3): 1216. 16371651.
Staenz, K., J. Secker, B.C. Gao, C. Davis, and C. Nadeau, 2002. Teillet, P.M., 1989. Surface reflectance retrieval using atmo-
Radiative transfer codes applied to hyperspectral data spheric correction algorithms, in: Proceedings of the 1989
for the retrieval of surface reflectance, ISPRS Journal of International Geoscience and Remote Sensing Symposium
Photogrammetry and Remote Sensing, 57(3): 194203. (IGARSS89) and the 12th Canadian Symposium on
Staenz, K., D.J. Williams, G. Fedosejevs, and P.M. Teillet, 1994. Remote Sensing, Vancouver, British Columbia, Canada,
Surface reflectance retrieval from imaging spectrom- pp.864867.
eter data using three atmospheric codes, in: Proceedings Teillet, P.M., 1990a. Rayleigh optical depth comparisons from
of SPIE EUROPTO94, Vol. 2318, Rome, Italy. SPIE, various sources, Applied Optics, 29: 18971900.
pp. 1728. Teillet, P.M., 1990b. Effects of spectral shifts on sensor response,
Staylor, W.F., 1990. Degradation rates of the AVHRR visible in: Proceedings of the ISPRS Commission VII Symposium,
channel for the NOAA 6, 7, and 9 spacecraft, Journal of Victoria, British Columbia, Canada, pp. 5965.
Atmospheric and Oceanic Technology, 7(3): 411423. Teillet, P.M., 1992. An algorithm for the radiometric and atmo-
Steven, M.D., 1977. Standard distribution of clear sky radiance, spheric correction of AVHRR data in the solar reflective
Quarterly Journal of the Royal Meteorological Society, 106: 57. channels, Remote Sensing of Environment, 41: 185195.
Steven, M.D., T.J. Malthus, F. Baret, H. Xu, and M.J. Chopping, Teillet, P.M., 1997a. A status overview of earth observation cali-
2003. Intercalibration of vegetation indices from differ- bration/validation for terrestrial applications, Canadian
ent sensor systems, Remote Sensing Environment, 88(4): Journal of Remote Sensing, 23(4): 291298.
412422. Teillet, P.M., ed., 1997b. Special issue on calibration/validation,
Stone, T.C., 2008. Radiometric calibration stability and inter- Canadian Journal of Remote Sensing, 23(4): 289423.
calibration of solar-band instruments in orbit using the Teillet, P.M., 1997c. Report on the Second Canadian Workshop
moon, in: Proceedings of SPIE Conference on Earth Observing on Earth Observation Calibration and Validation, Canada
Systems XIII, San Diego, CA, 2008, p. 70810X. Centre for Remote Sensing, Ottawa, Ontario, Canada, 66p.
Stone, T.C., H.H. Kieffer, and I.F. Grant, 2005. Potential for cali- Teillet, P.M., 1998. Report on the Third Canadian Workshop on
bration of geostationary meteorological satellite imagers Earth Observation Calibration and Validation, Canada
using the moon, in: Proceedings of SPIE Conference on Earth Centre for Remote Sensing, Ottawa, Ontario, Canada, 74p.
Observing Systems X, San Diego, CA, pp. 19. Teillet, P.M., September/October 2005. Satellite image radiom-
Stone, T.C., W.B. Rossow, J. Ferrier, and L.M. Hinkelmann, 2013. etry: From photons to calibrated earth science data, Physics
Evaluation of ISCCP multi-satellite radiance calibration in Canada, 61(5), 301310.
for geostationary imager visible channels using the moon, Teillet, P.M. and G. Chander, 2014. Post-launch radiometric cal-
IEEE Transactions on Geoscience and Remote Sensing, 51(3): ibration of satellite-based optical sensors with emphasis
12551266. on terrestrial reference standard sites, in: P.S. Thenkabail,
Suits, G.H., W.A. Malila, and T.M. Weller, 1988. The prospects ed., Chapter 4: The Remote Sensing Handbook, Vol. 1, CRC
for detecting spectral shifts due to satellite sensor ageing, Press, Boca Raton, FL.
Remote Sensing of Environment, 26: 1729. Teillet, P.M. and G. Fedosejevs, 1995. On the dark target approach
Sweet, R.J.M, J.C. Elliott, and J.R. Beasley, 1992. Research needs to atmospheric correction of remotely sensed data,
to encourage the growth of the earth observation applica- Canadian Journal of Remote Sensing, 21(4): 374387.
tions market, in: P.A. Cracknell and R.A. Vaughan, eds., Teillet, P.M., G. Fedosejevs, F.J. Ahern, and R.P. Gauthier, 1994.
Proceedings of the 18th Annual Conference of the Remote Sensitivity of surface reflectance retrieval to uncertain-
Sensing Society: Remote Sensing from Research to Operation, ties in aerosol optical properties, Applied Optics, 33(18):
September 1517, University of Dundee, Dundee, U.K., 39333940.
pp. 399407. Teillet, P.M., G. Fedosejevs, R.P. Gauthier, N.T. ONeill, K.J.
Tahnk, W.R. and J.A. Coakley, 2001. Updated calibration coeffi- Thome, S.F. Biggar, H. Ripley, and A. Meygret, 2001a. A
cients for NOAA-14 AVHRR channels 1 and 2, International generalized approach to the vicarious calibration of mul-
Journal of Remote Sensing, 22(15): 30533057. tiple earth observation sensors using hyperspectral data,
Tanr, D., C. Deroo, P. Duhaut, M. Herman, J.J. Morcrette, and Remote Sensing of Environment, 77(3): 304327.
J. Perbos, 1990. Description of a computer code to simu- Teillet, P.M., G. Fedosejevs, R.K. Hawkins, T.I. Lukowski, R.A.
late the satellite signal in the solar spectrum: The 5S code, Neville, K. Staenz, R. Touzi, J. van der Sanden, and J. Wolfe,
International Journal of Remote Sensing, 11: 659668. 2004a. Importance of Data Standardization for Generating
Tanr, R.C., Herman, M., and Deschamps, P.Y., 1983. Influence High Quality Earth Observation Products for Natural
of the atmosphere on space measurements of directional Resource Management, Earth Sciences Sector, Natural
properties, Applied Optics, 22: 733. Resources Canada, Ottawa, Ontario, Canada, 42p.
2016 by Taylor & Francis Group, LLC
Overview of Satellite Image Radiometry in the Solar-Reflective Optical Domain 107
Teillet, P.M., G. Fedosejevs, and K.J. Thome, 2004b. Spectral band Proceedings of SPIE Conference on Sensors, Systems, and
difference effects on radiometric cross-calibration between Next-Generation Satellites V, Vol. 4550, Toulouse, France.
multiple satellite sensors in the Landsat solar-reflective spec- SPIE, pp. 246254.
tral domain, in: R. Meynart, S.P. Neeck, and H. Shimoda, Temps, R.C. and K.L. Coulson, 1977. Solar radiation inci-
eds., Workshop on Inter-Comparison of Large-Scale Optical dent upon slopes of different orientations, Solar Energy,
and Infrared Sensors, Proceedings of SPIE Conference on 19:179.
Sensors, Systems, and Next-Generation Satellites VIII, Vol. Thome, K., 2005. Sampling and uncertainty issues in trending
5570, Maspalomas, Canary Islands, Spain, pp. 307316. reflectance-based vicarious calibration results, in: J.J. Butler,
Teillet, P.M., G. Fedosejevs, K.J. Thome, and J.L. Barker, 2007. ed., Proceedings of SPIE Conference on Earth Observing
Impacts of spectral band difference effects on radiomet- Systems X, Vol. 5882, San Diego, CA. SPIE, Bellingham,
ric cross-calibration between satellite sensors in the solar- WA, pp. 111.
reflective spectral domain, Remote Sensing of Environment, Thome, K., R. Barnes, R. Baize, J. OConnell, and J. Hair, 2010.
110(3): 393409. Calibration of the reflected solar instrument for the cli-
Teillet, P.M., B. Guindon, and D.G. Goodenough, 1982. On the mate absolute radiance and refractivity observatory, in:
slope aspect correction of multi-spectral scanner data, Proceedings of the 2010 IEEE International Geoscience and
Canadian Journal of Remote Sensing, 8: 84106. Remote Sensing Symposium (IGARSS), Honolulu, HI, pp.
Teillet, P.M. and B.N. Holben, 1994. Towards operational radio- 22752278.
metric calibration of NOAA AVHRR imagery in the visible Thome, K., J. Czapla-Myers, N. Leisso, J. McCorkel, and J. Buchanan,
and near-infrared channels, Canadian Journal of Remote 2008. Intercomparison of imaging sensors using automated
Sensing, 20(1): 110. ground measurements, in: Proceedings of IEEE Conference
Teillet, P.M., D.N.H. Horler, and N.T. ONeill, 1997a. Calibration, val- on Remote Sensing: The Next Generation, Boston, MA. IEEE,
idation, and quality assurance in remote sensing: A new para- Piscataway, NJ, pp. 13321335.
digm, Canadian Journal of Remote Sensing, 23(4): 401414. Thome, K., S. Schiller, J. Conel, K. Arai, and S. Tsuchida, 1998.
Teillet, P.M. and J.R. Irons, 1990. Spectral variability effects on Results of the 1996 earth observing system vicarious cali-
the atmospheric correction of imaging spectrometer data bration joint campaign at Lunar Lake Playa, Metrologia, 35:
for surface reflectance retrieval, in: Proceedings of the ISPRS 631638.
Commission VII Symposium, Victoria, British Columbia, Thome, K.J., 2001. Absolute radiometric calibration of Landsat 7
Canada, pp. 579583. ETM+ using the reflectance-based method, Remote Sensing
Teillet, P.M., M. Lasserre, and C.G. Vigneault, 1986. An evaluation of Environment, 78(12): 2738.
of sun angle computation algorithms, in: Proceedings of the Thome, K.J, S.F. Biggar, and W. Wisniewski, 2003. Cross com-
10th Canadian Symposium on Remote Sensing, Edmonton, parison of EO-1 sensors and other earth resources sen-
Alberta, Canada, pp. 91100. sors to Landsat-7 ETM+ using Railroad Valley Playa,
Teillet, P.M. and X. Ren, 2008. Spectral band difference effects IEEE Transactions on Geoscience and Remote Sensing, 41:
on vegetation indices derived from multiple satellite sensor 11801188.
data, Canadian Journal of Remote Sensing, 34(3): 159173. Thuillier, G., M. Hers, D. Labs, T. Foujols, W. Peetermans,
Teillet, P.M. and R.P. Santer, 1991. Terrain elevation and sensor D. Gillotay, P.C. Simon, and H. Mandel, 2003. The solar
altitude dependence in a semi-analytical atmospheric code, spectral irradiance from 200 to 2400 nm as measured by
Canadian Journal of Remote Sensing, 17: 3644. the SOLSPEC spectrometer from the Atlas and Eureca mis-
Teillet, P.M., P.N. Slater, Y. Ding, R.P. Santer, R.D. Jackson, and sions, Solar Physics, 214(1): 122.
M.S. Moran, 1990. Three methods for the absolute calibra- Townshend, J.R.G., 1981. The spatial resolving power of earth
tion of the NOAA AVHRR sensors in-flight, Remote Sensing resources satellites, Progress in Physical Geography, 5(1):
of Environment, 31: 105120. 3255.
Teillet, P.M. and K. Staenz, 1992. Atmospheric effects due to Townshend, J.R.G., C. Huang, S.N.V. Kalluri, R.S. Defries, and
topography on MODIS vegetation index data simulated S. Liang, 2000. Beware of per-pixel characterization of
from AVIRIS imagery over mountainous terrain, Canadian land cover, International Journal of Remote Sensing, 21(4):
Journal of Remote Sensing, 18(4): 283291. 839843.
Teillet, P.M., K. Staenz, and D.J. Williams, 1997b. Effects of spec- Trishchenko, A.P., 2009. Effects of spectral response function on sur-
tral, spatial, and radiometric characteristics on remote sens- face reflectance and NDVI measured with moderate resolu-
ing vegetation indices for forested regions, Remote Sensing tion satellite sensors: Extension to AVHRR NOAA-17, 18 and
of Environment, 61: 139149. METOP-A, Remote Sensing of Environment, 113(2): 335341.
Teillet, P.M., K.J. Thome, N. Fox, and J.T. Morisette, 2001b. Trishchenko, A.P., J. Cihlar, and Z. Li, 2002. Effects of spectral
Earth observation sensor calibration using a global response function on surface reflectance and NDVI mea-
instrumented and automated network of test sites sured with moderate resolution satellite sensors, Remote
(GIANTS), in: H. Fujisada, J.B. Lurie, and K. Weber, eds., Sensing of Environment, 81(1): 118.
2016 by Taylor & Francis Group, LLC
108 Remotely Sensed Data Characterization, Classification, and Accuracies
Turner, R.E. and M.M. Spencer, 1972. Atmospheric model for measurements, IEEE Transactions on Geoscience and
correction of spacecraft data, in: Proceedings of the Eighth Remote Sensing, 40(5): 10381046.
International Symposium on Remote Sensing of Environment, White, H.P., L. Sun, K. Staenz, R.A. Fernandes, and C. Champagne,
Ann Arbor, MI, p. 895. 2002a. Determining the contribution of shaded elements
Vermote, E. and Y.J. Kaufman, 1995. Absolute calibration of of a canopy to remotely sensed hyperspectral signatures,
AVHRR visible and near-infrared channels using ocean in: Proceedings of the First International Symposium on
and cloud views, International Journal of Remote Sensing, Recent Advances on Quantitative Remote Sensing, Torrent,
16(13): 23172340. Valencia, Spain.
Vermote, E.F., N. El Saleous, C.O. Justice, Y.J. Kaufman, J.L. Whitlock, C.H., W.F. Staylor, J.T. Suttles, G. Smith, R. Levin,
Privette, L. Remer, J.C. Roger, and D. Tanre, 1997. R.Frouin, C. Gautier etal., 1990. AVHRR and VISSR sat-
Atmospheric correction of visible to middle-infrared EOS- ellite instrument calibration results for both cirrus and
MODIS data over land surfaces: Background, operational marine stratocumulus IFO periods, in: Proceedings of FIRE
algorithm and validation, Journal of Geophysical Research, Science Meeting, Vail, CO. NASA Langley Research Center,
102: 1713117141. Hampton, VA, pp. 141146.
Vermote, E.F., N.Z. El Saleous, and C.O. Justice, 2002. Atmospheric Woodham, R.J. and M.H. Gray, 1987. Analytic method for radio-
correction of MODIS data in the visible to middle infra- metric correction of satellite multispectral scanner data,
red: First results, Remote Sensing of Environment, 83(12): IEEE Transactions on Geoscience and Remote Sensing,
97111. GE-25: 258271.
Vermote, E.F. and N.Z. Saleous, 2006. Calibration of NOAA16 Woodham, R.J. and T.K. Lee, 1985. Photometric method for
AVHRR over a desert site using MODIS data, Remote radiometric correction of multispectral scanner data,
Sensing of Environment, 105(3): 214220. Canadian Journal of Remote Sensing, 11: 132.
Vermote, E.F., R. Santer, P.Y. Deschamps, and M. Herman, 1992. Wu, A., C. Cao, and X. Xiong, 2003. Intercomparison of the
In-flight calibration of large field of view sensors at short 11- and 12-m bands of Terra and Aqua MODIS using
wavelengths using Rayleigh scattering, International Journal NOAA-17 AVHRR, in: W.L. Barnes, ed., Proceedings of SPIE
of Remote Sensing, 13(18): 34093429. Conference on Earth Observing Systems VIII, Vol. 5151, San
Vermote, E.F., D. Tanr, J.L. Deuz, M. Herman, J.J. Morcrette, Diego, CA. SPIE, Bellingham, WA, pp. 384394.
S.Y. Kotchenova, and T. Miura, 2006. Second simulation Wyatt, C.L., V. Privalsky, and R. Datla, 1998. Recommended
of the satellite signal in the solar spectrum (6S), 6S user Practice: Symbols, Terms, Units and Uncertainty Analysis
guide version 3, November 2006, http://www.6s.ltdri.org. for Radiometric Sensor Calibration, NIST Handbook,
Accessed May 25, 2015. Vol. 152, National Institute of Standards and Technology,
White, H.P., J.R. Miller, and J.M. Chen, 2001. Four-scale linear Gaithersburg, MD, 120p.
model for anisotropic reflectance (FLAIR) for plant cano- Xiong, X., J. Sun, and W. Barnes, 2008. Intercomparison of on-
pies. I: Model description and partial validation, IEEE orbit calibration consistency between Terra and Aqua
Transactions on Geoscience and Remote Sensing, 39(5): MODIS reflective solar bands using the moon, IEEE
10721083. Geoscience and Remote Sensing Letters, 5(4): 778782.
White, H.P., J.R. Miller, and J.M. Chen, 2002b. Four-scale lin- Yuan, D. and C.D. Elvidge, 1996. Comparison of relative radio-
ear model for anisotropic reflectance (FLAIR) for plant metric normalization techniques, ISPRS Journal of
canopies II: Partial validation and inversion using field Photogrammetry and Remote Sensing, 51: 117126.
2016 by Taylor & Francis Group, LLC
4
Postlaunch Radiometric
Calibration of Satellite-Based
Optical Sensors with Emphasis on
Terrestrial Reference Standard Sites
Acronyms and Definitions .................................................................................................................109
4.1 Introduction .............................................................................................................................110
4.2 Postlaunch Sensor Radiometric Calibration Methodologies ...........................................110
Onboard Radiometric Calibration Vicarious Calibration Methodologies
Philippe M. Teillet 4.3 Vicarious Calibration via Terrestrial Reference Standard Sites .......................................113
University of Lethbridge Historical Perspective Current Developments Next Steps
Gyanesh Chander 4.4 Concluding Remarks...............................................................................................................118
Goddard Space Flight Acknowledgments ...............................................................................................................................119
Center (GSFC) References.............................................................................................................................................119
Acronyms and Definitions InSb Indium antimonide
IRMSS Infrared multispectral scanner
ADEOS Advanced Earth observing satellite ISO International Organization for Standardization
ALI Advanced land imager IVOS Infrared Visible Optical Sensors
AVHRR Advanced very-high-resolution radiometer JAXA Japan Aerospace Exploration Agency
CBERS ChinaBrazil Earth Resources Satellite L4 Landsat-4
CEOS Committee on Earth Observation Satellites L5 Landsat-5
CLARREO Climate Absolute Radiance and Refractivity L7 Landsat-7
Observatory L8 Landsat-8
CNES Centre National dEtudes Spatiales MOBY Marine optical BuoY
DC Digital count MODIS MODerate resolution Imaging Spectroradiometer
EO Earth observation MSS Multispectral scanner
EO-1 Earth observing-1 NASA National Aeronautics and Space Administration
EOS Earth Observing System NIST National Institute of Standards and Technology
EROS Earth Resources Observation and Science NOAA National Oceanic and Atmospheric Administration
ESA European Space Agency NPL National Physical Laboratory
ETM+ Enhanced Thematic Mapper Plus OLI Optical land imager
FASC Full-aperture solar calibrator PASC Partial-aperture solar calibrator
GCOS Global Climate Observing System PICS Pseudoinvariant calibration site
GEO Group on Earth Observations PIF Pseudoinvariant feature
GEOSS Global Earth Observation System of Systems QA4EO Quality Assurance Framework for Earth Observation
GIANTS Global instrumented and automated network of QUASAR Quality assurance and stability reference
test sites SBDE Spectral band difference effects
GLI Global imager SI International system of units
GSICS Global Space-Based Inter-Calibration System SPOT Satellite Pour lObservation de la Terre
HgCdTe Mercury cadmium telluride SRCA Spectro-radiometric calibration assembly
IC Internal calibrator TM Thematic Mapper
109
2016 by Taylor & Francis Group, LLC
110 Remotely Sensed Data Characterization, Classification, and Accuracies
TOA Top of atmosphere reference to an Earth target of known reflectance, to the Moon,
TRUTHS Traceable Radiometry Underpinning Terrestrial or to bright stars. Procedure category (3) is usually called vicari-
and Helio Studies ous calibration. Vicarious calibration can also be used for rela-
USDA United States Department of Agriculture tive calibration to compensate for temporal trends in sensor
USGS United States Geological Survey response.
WGCV Working Group on Calibration and Validation Onboard radiometric calibration systems can provide good
WMO World Meteorological Organization temporal sampling with high precision that allows trending
WSNM White Sands, New Mexico of system responses. A disadvantage of onboard calibration
approaches is that they add significantly to the complexity
4.1 introduction and cost of satellite missions. Vicarious calibration tech-
niques involving Earth targets provide independent, full-
Scientists and decision makers addressing local, regional, and/or aperture calibrations with relatively high accuracy. While
global issues rely increasingly on operational use of data and infor- reference to an Earth target requires an accurate determina-
mation obtained from a multiplicity of satellite-based Earth obser- tion of atmospheric optical properties (mainly aerosol optical
vation (EO) sensor systems. It is imperative that they be able to rely depth) at the time of satellite overpass, it has the important
on the accuracy of EO data and information products (e.g., Ohring advantage of replicating actual conditions of image acquisi-
et al., 2007). Accordingly, the characterization and calibration of tion, with a full irradiation of the entrance aperture. The sen-
these sensors are critical elements of EO programs. sor acquires data in the same modality as it acquires Earth
Chapter 3 provided an overview of satellite image radiometry image data, that is, in the same spectral bands, with the same
in the solar-reflective domain (Teillet, 2014). This chapter delves source spectrum and under typical illumination conditions.
more deeply into postlaunch sensor radiometric calibration The main disadvantages of vicarious calibration are that the
methodologies, encompassing optical sensors operating in the methods yield lower precision and have lower temporal sam-
visible, near-infrared, and shortwave infrared spectral regions, pling frequencies compared to onboard radiometric calibra-
with particular emphasis on the use of terrestrial reference stan- tion approaches.
dard sites. Although a lot of time and effort are devoted to pre- Terrestrial surfaces with suitable characteristics have long
launch radiometric calibration, postlaunch changes to satellite served as benchmarks or reference standard sites, via either
sensors and their radiometric calibration performance neces- vicarious calibration or cross-calibration, to assess the post-
sitate ongoing calibration monitoring and maintenance (e.g., launch radiometric calibration performance of satellite optical
Slater etal., 2001; Butler etal., 2005). sensors (e.g., Teillet and Chander, 2010; Chander etal., 2013a,b).
Calibration of satellite sensor radiometric response is gener- The use of such sites is the only practical way of evaluating
ally classified into relative and absolute calibration (e.g., Kastner radiometric calibration biases between sensors, and it also
and Slater, 1982; Slater, 1984; Teillet, 1986). Relative calibration provides a means of bridging gaps in measurement continu-
mainly deals with compensation for detector-to-detector differ- ity. Accordingly, after a brief review of onboard calibration and
ences. Ideally, detectors constructed from the same materials other vicarious calibration methodologies, this chapter focuses
should respond identically to the same incident energy. However, on the use of terrestrial reference standard sites for postlaunch
typically, detectors do not respond identically, resulting in dif- sensor radiometric calibration of optical sensors operating in
ferences in detector gain and bias levels that lead to striping the visible, near-infrared, and shortwave infrared spectral
in the image data. This striping can be corrected by selecting regions.
a stable reference detector, then scaling and shifting the other
detector responses to identical targets to the reference detectors 4.2 Postlaunch Sensor Radiometric
gain and bias. This process is called detector-to-detector rela-
tive radiometric calibration. Absolute radiometric calibration
calibration Methodologies
enables the conversion of image digital counts (DCs) to physical As noted in Chapter 3 (Teillet, 2014), the establishment of
units in the International System of Units (SI) such as at-sensor worldwide coordinated and operational calibration efforts is
spectral radiance (W m2 sr1 m1). Because DCs from one sen- critical to achieving the goals of quantitative EO programs in
sor bear no relation to DCs from a different sensor, conversion general, and the goals of endeavors such as the Global Earth
to at-sensor spectral radiance is a fundamental step that enables Observation System of Systems (GEOSS)* of the Group on Earth
the comparison of similar products from different sensors. The Observations (GEO), as well as the Global Space-Based Inter-
additional step of converting the at-sensor radiance to top-of- Calibration System (GSICS). Accordingly, the international
atmosphere (TOA) reflectance is also important, as outlined
in Chapter 3 (Teillet, 2014). Absolute radiometric calibration
* http://www.earthobservations.org/geoss.shtml.
procedures (Slater, 1985) include (1) prelaunch calibration, with http://gsics.wmo.int/ GSICS is part of the WMO. It is an international
only focal plane calibration after launch; (2) postlaunch refer- collaboration to monitor, improve and harmonize data quality from
ence to onboard standard sources or to the Sun via a diffuser, operational environmental satellites for climate monitoring and weather
usually not illuminating the full aperture; and (3) postlaunch forecasting.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 111
Committee on Earth Observation Satellites (CEOS), considered neutral density filter and onto the focal plane. The neutral den-
to be the space arm of GEO, has led the development of a Quality sity filter was wedge shaped, yielding a variable attenuation as it
Assurance Framework for Earth Observation (QA4EO),* which rotated with the shutter wheel. The IC did not test the full optical
is based on the adoption of key guidelines to help address dif- path of the MSS sensors.
ficult issues of traceability and interoperability, especially in the The Thematic Mapper (TM) sensors on Landsat-4 (L4),
postlaunch environment. Landsat-5 (L5), and the Landsat-7 (L7) Enhanced TM Plus
Whereas prelaunch calibration methodologies typically include (ETM+) have been well documented (e.g., Markham etal., 1998,
radiometric, geometric, spectral, and polarization characteriza- 2012). Briefly, the TMs and the ETM+ have an uncooled pri-
tions, postlaunch calibration typically involves primarily radio- mary focal plane containing 16 silicon (Si) detectors per band
metric and geometric calibration. One exception is National for the four visible and near-infrared bands. The L7 ETM+ also
Aeronautics and Space Administration (NASA)s Terra and Aqua has a panchromatic band with 32 detectors. The TMs and the
MODerate-resolution Imaging Spectroradiometers (MODIS), ETM+ have a cold focal plane containing 16 indium antimonide
which include an onboard spectro-radiometric calibration (InSb) detectors for each of the two short-wave infrared bands,
assembly (SRCA) designed to monitor the visible, near-infrared, plus 4 mercury cadmium telluride (HgCdTe) detectors (8 for the
and short-wave infrared spectral bands on MODIS (e.g., Xiong ETM+) for the thermal emissive band.
et al., 2006a, 2010a; Choi et al., 2013). Postlaunch radiometric The Landsat TM and ETM+ sensors also incorporate IC sys-
calibration can be categorized into onboard calibration and tems for onboard radiometric calibration monitoring for the
vicarious calibration, the topics of the following sections. solar-reflective spectral bands (Mika, 1997). These IC devices
have a shutter arm that oscillates back and forth directly in front
of the primary focal plane. On the TMs, the IC features three
4.2.1 onboard Radiometric calibration
lamps that cycle through an eight-lamp state sequence over the
Some satellite sensor systems utilize onboard devices to help course of a 24 s scene of data, whereas the ETM+ has two lamps
monitor and, if necessary, update radiometric calibration coef- (four lamp states). A complex, synchronized system involving
ficients. Partial-aperture calibrators include standard lamps the shutter, a scan mirror, and photodiodes records IC lamp
and/or solar diffuser panels. If done well, such approaches signals, dark shutter signals, and Earth-reflected light for each
can capture sensor drift, instability, and sensor and electronic scan line. Algorithms to use the IC data for radiometric calibra-
ageing problems. Shutters or deep space views provide dark tion were developed and implemented in operational Landsat
readings. product generation systems around the world. Techniques were
Although there are numerous satellite sensor systems that developed to analyze the IC data, mainly for the long lifetime of
make use of onboard radiometric calibration systems, the the L5 TM (Metzler and Malila, 1985; Helder, 1996; Helder etal.,
Landsat series of whisk-broom sensors is a notable example. 1996, 1997, 1998a,b; Markham etal., 1998), but also for the L4
Tremendous efforts have been dedicated to understanding and TM (Chander etal., 2007a; Markham and Helder, 2012; Helder
using the onboard calibration lamps (Barker, 1985; Markham et al., 2013). Over time, these trend analyses led to a detailed
and Barker, 1985) and partial-aperture or full-aperture dif- understanding of several image artifacts introduced by vari-
fuse reflectance panels (Markham et al., 2003). These efforts ous TM sensor characteristics. Within-scene relative calibration
are a measure of the importance of the Landsat series of sen- algorithms were developed and implemented in product gen-
sors that have provided a readily accessible, global, and seasonal eration systems to remove most of these artifacts, which include
archive of relatively high spatial resolution data spanning over striping, scan-correlated shift, memory effect, coherent noise,
four decades (Goward and Masek, 2001; Goward et al., 2006; and oscillations due to cold focal plane icing (e.g., Helder and
Markham and Helder, 2012; Wulder and Masek, 2012). It should Micijevic, 2004; Helder and Ruggles, 2004; Teillet etal., 2004a).
also be noted that the Landsat radiometric calibration experi- Landsat processing systems make corrections for all artifacts
ence was a critical contribution to the challenging radiometric except the coherent noise effect, which is typically on the order
calibration of the Earth Observing System (EOS) MODIS whisk- of 0.15 DCs or less.
broom sensors. Historically, the TM radiometric calibration procedure used
Key characteristics of the multispectral scanners (MSS) sen- the instruments response to the IC lamps on a scene-by-scene
sors on Landsat missions 15 are documented mainly in the basis to determine the gain and offset of each detector. Since
so-called gray literature (e.g., Bartolucci and Davis, 1983). For May 2003, the L5 TM data processed and distributed by the
the purposes of this chapter, the main MSS onboard calibration U.S. Geological Survey (USGS) Earth Resources Observation
systems were internal calibrators (ICs) consisting of a shutter and Science (EROS) Center were updated to use a lifetime
wheel with a mirror and a neutral density filter, and two redun- gain look-up-table (LUT) model to radiometrically calibrate
dant tungsten-filament lamps (e.g., Ahern and Murphy, 1979; the L5 TM data instead of using scene-specific IC lampbased
Murphy, 1984, 1986; Markham and Barker, 1985; Thome etal., gains (Chander and Markham, 2003). The new procedure for
1997b). The mirror reflected light from the IC lamps through the
An assessment of the Landsat-5 thermal band calibration has been carried
* http://qa4eo.org/index.html. out by Barsi etal. (2003, 2007) and Padula etal. (2010).
2016 by Taylor & Francis Group, LLC
112 Remotely Sensed Data Characterization, Classification, and Accuracies
the reflective bands (15, 7) is based on a lifetime radiometric bands of the L8 OLI sensor* include the L7 ETM+ sensor bands
calibration curve for the instrument derived from the instru- augmented by a deep blue spectral band designed for water
ments IC, cross-calibration with the L7 ETM+, and vicarious resources and coastal zone investigation, and an infrared spec-
measurements. This change has improved absolute calibration tral band at 1.37 m for cirrus cloud detection. A quality assur-
accuracy, consistency over time, and consistency with L7 ETM+ ance band is also included with each data product to indicate the
data (Chander etal., 2004a). With this change, the radiometric presence of clouds, water, and/or snow.
scaling coefficients were updated as well (Chander etal., 2007b, The L8 OLI radiometric calibration system consists of two
2009; Helder etal., 2008). Users need to apply the new L5 TM solar diffusers and a shutter, and two stimulation lamp assem-
rescaling factors to convert calibrated DCs to TOA at-sensor blies (Markham et al., 2010; Reuter et al., 2011). The diffusers
spectral radiance. provide a full-system, full-aperture calibration when the shutter
The L7 ETM+ has a full aperture solar calibrator (FASC). is open and a dark reference when the shutter is closed. Each
This capability features a near-Lambertian YB71 panel of lamp assembly contains three tungsten lamps, operated at con-
known reflectance that is deployed periodically (approximately stant current and monitored by a silicon photodiode, which illu-
monthly) in front of the ETM+ aperture and diffusely reflects minate the OLI detectors through the full optical system. One
solar radiation into the full aperture of the sensor. With knowl- working lamp is used daily for intraorbit calibration, a reference
edge of the solar irradiance and geometric conditions, the FASC lamp is used approximately monthly, and a pristine lamp is used
serves as an independent, full-aperture calibrator. The ETM+ approximately twice a year. L8 sensor characterization and cali-
also has a partial-aperture solar calibrator (PASC) that con- bration performance is reviewed in a special journal issue edited
sists of a small passive device with a set of optics that allow the by Markham etal. (2015).
ETM+ to image the Sun (daily) through small holes. On-orbit
performance of the ETM+ calibrator systems can be found in
4.2.2 Vicarious calibration Methodologies
Markham etal. (2003, 2004a,b, 2012).
While the IC data from the onboard systems played a cen- The predominant vicarious calibration approach is the mea-
tral role in the radiometric calibration of each Landsat mission, surement, on satellite overpass days, of pertinent surface and
the vicarious calibration techniques described in the following atmospheric optical properties at terrestrial sites with suitable
sections made it possible to establish and validate radiometric characteristics to estimate at-sensor radiance or TOA reflec-
calibration across the full family of Landsat sensors going back tance. Comparisons of these estimates with image-based values
to 1972, including the MSS sensors (Helder et al., 2012b). The computed using the canonical calibration coefficients provide
overall effort was undertaken over the past 15 years, approxi- postlaunch monitoring of the radiometric calibration of satel-
mately, and is summarized by Markham and Helder (2012). lite sensors, either individually or by cross-calibration between
The key result from the tremendous amount of work involved sensors (e.g., Slater etal., 1987, Teillet etal., 1990; Biggar etal.,
is a consistently calibrated Landsat data archive that spans more 1994; Thome, 2001; Teillet, 2014). Well-understood or instru-
than four decades with total uncertainties on the order of 10% or mented sites serve as reference standard sites, whereas other
less for most sensors and bands. terrestrial targets that change very little in surface reflectance
In particular (Markham and Helder, 2012), the L1L5 MSS serve as pseudoinvariant features (PIFs), also known as pseudo-
sensors are within 10% absolute radiometric accuracy in all but invariant calibration sites (PICS) in the present context. A com-
near-infrared spectral band 4, and with the other exceptions prehensive review of terrestrial reference standard sites used for
of L2 red spectral band 3 (11%) and L1 visible spectral bands postlaunch radiometric calibration is given in the next section
13 (11%, 11%, and 12%, respectively). MSS band 4 is very broad of this chapter.
spectrally and includes large atmospheric absorption features. Error analyses of vicarious calibration approaches indicate
Thus, the uncertainty for this band in the MSS series grows that uncertainties in the 2%3% (1 ) range with respect to
much more rapidly to the point where the absolute radiomet- exo-atmospheric solar irradiance (W m2 m1) are attainable
ric calibration uncertainty for L1 MSS band 4 approaches 25%. and can be accurately related to national laboratory standards
For the solar reflective spectral bands of the L4 TM, L5 TM, and (Dinguirard and Slater, 1999). The objective is traceability of
L7 ETM+ sensors, Markham and Helder (2012) report abso- radiometric calibration accuracies to SI units for science users
lute radiometric calibration uncertainties of 9%, 7%, and 5%, (e.g., Pollock etal., 2003) and data products with consistent qual-
respectively. ity for the broader user community.
The optical land imager (OLI) on Landsat-8 (L8) uses push- Although different satellite sensors image the Earth in
broom scanning technology. As a result, OLI radiometric analogous spectral bands, the spectral bands almost never
calibration benefits from the long history of calibration experi- match exactly such that data acquired over identical targets
ence with the push-broom sensors on the French Satellite Pour and derived information products are not directly comparable
lObservation de la Terre (SPOT) satellites (e.g., Valorge et al.,
2004), as well as from the advanced land imager (ALI) technol- * http://landsat.usgs.gov/band_designations_landsat_satellites.php.
ogy demonstration sensor flown on NASAs Earth Observing-1 https://director y.eoporta l.org/web/eoporta l/satellite-missions/l/
(EO-1) satellite (e.g., Mendenhall etal., 2003, 2005). The spectral landsat-8-ldcm.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 113
(Teillet etal., 2007a; Teillet and Ren, 2008). Relatively few inves- 4.3.1 Historical Perspective
tigations have been undertaken to assess radiometric calibration
errors due to differences in spectral band response functions Reflectance and other properties of land surfaces and atmo-
between satellite sensors when attempting cross-calibration spheric phenomena now used by many for vicarious calibration
based on near-simultaneous imaging of common ground look were first examined decades ago using surface and airborne
targets in analogous spectral bands (Teillet etal., 2001a, 2004b, measurements in the contexts of the micrometeorology of arid
2007a; Trishchenko etal., 2002; Rao etal., 2003; Teillet and Ren, zones (Ashburn and Weldon, 1956), investigations of natural
2008; Doelling etal., 2012; Chander etal., 2013c; Henry etal., landing areas for aircraft (Molineux etal., 1971), and other anal-
2013). Such spectral band difference effects (SBDEs) can be sig- yses (Davis and Cox, 1982; Smith, 1984; Bowker and Davis, 1992;
nificant. For instance, Teillet et al. (2007a) simulated SBDEs Warren etal., 1998). In scene understanding studies in support
affecting vicarious-calibration-based TOA reflectance compar- of the first meteorological satellite observations, Salomonson
isons between many satellite sensors. The following summary and Marlatt (1968) examined the directional reflectance prop-
examples are for Railroad Valley Playa in Nevada, a frequently erties of cloud tops, snow, and white gypsum sand (hydrous
used vicarious calibration test site, and for an atmospheric calcium sulfate) at White Sands, New Mexico (WSNM) in the
aerosol optical depth of 0.05 and a solar zenith angle of 60. United States, which are three surface target types that have
Comparisons in visible spectral bands were generally within since been used extensively for vicarious calibration of satellite
3%, some were within 5%, and worst cases were around 7% imaging sensors.
between Terra-MISR* and various other sensors. Comparisons The potential use of suitable terrestrial targets on or near the
in near-infrared spectral bands were generally within 3%, some Earths surface for the postlaunch radiometric calibration of
were within 10%, and worst cases were around 21% between satellite sensors was first examined by a number of investigators
advanced very-high-resolution radiometer (AVHRR) and vari- for various satellite sensor systems and for several surface types,
ous other sensors. Comparisons in shortwave infrared spectral including WSNM, deserts, playas, salt flats, ocean, cloud tops,
bands around 1.65 m were generally within 3%, and worst snow or ice fields, PIFs, and uniformly vegetated areas (Coulson
cases were around 8% between TM or ChinaBrazil Earth and Jacobowitz, 1972; Muench, 1981; Kastner and Slater, 1982;
Resources Satellite (CBERS) infrared multispectral scanner Koepke, 1982; Hovis et al., 1985; Mueller, 1985; Fraser and
(IRMSS) and various other sensors. Comparisons in sh ort- Kaufman, 1986; Staylor, 1986; Duggin, 1987; Schott etal., 1988;
wave infrared spectral bands around 2.2 m were generally Biggar etal., 1994; Teillet etal., 1998a. 2010; Le Marshall etal.,
within 6%, and worst cases were around 21% between MODIS 1999; Smith et al., 2002; Kamstrup and Hansen, 2003; Nieke
or advanced Earth observing satellite (ADEOS)-global imager et al., 2004; Martiny et al., 2005; Anderson and Milton, 2006;
(GLI) and various other sensors. Wu etal., 2008a,c). Criteria for terrestrial reference site selec-
tion have been well documented (e.g., Scott etal., 1996):
4.3 Vicarious calibration via terrestrial High spatial uniformity over a large area (within 3%)
Reference Standard Sites Surface reflectance greater than 0.3 across solar-reflective
wavelengths
Publications in the form of research articles, special journal Flat spectral reflectance across solar-reflective wavelengths
issues, reports, and books have periodically provided over- Temporally invariant surface properties (within 2%)
views or reviews of satellite sensor radiometric calibration Horizontal surface with nearly Lambertian reflectance
with some mention of vicarious calibration (Slater, 1980, 1984, High altitude, far from ocean, urban, and industrial areas
1985, 2001; Markham and Barker, 1985; Malila and Anderson, Arid region with low probability of cloud cover or air-
1986; Price, 1987a,b; Ahern et al., 1988; Jackson, 1990; borne particles
Nithianandam et al., 1993; Bruegge and Butler, 1996; Chen,
1996; Slater and Biggar, 1996; Teillet, 1997a,b; Dinguirard and It is also worth noting that the characterization of any given cali-
Slater, 1999; Morain and Budge, 2004; Valorge et al., 2004; bration reference site usually involves nested sites: the specific
Butler etal., 2005; Chander etal., 2013a,b). This section pro- surface measurement site and the overall calibration reference
vides an overview of the use of terrestrial reference standard site to be imaged by satellite sensors, which must be well repre-
sites for postlaunch sensor radiometric calibration from his- sented by the measurement sites spectral reflectance properties.
torical, current, and future perspectives. Emphasis is placed Early research on surface measurement methodologies for
on optical sensors operating in the visible, near-infrared, and land surfaces focused to a considerable extent on the alkali flats
shortwave infrared spectral regions. region of WSNM (cf. Table 4.1 for selected literature references).
These vicarious calibration methodologies were used to provide
postlaunch radiometric calibration updates for satellite sensor
* Terra-MISR: Multi-angle Imaging Spectro-Radiometer on NASAs Terra systems such as the L4 and L5 TM sensors (Slater, 1986; Slater
satellite.
CBERS-IRMSS: ChinaBrazil Earth Resources Satellite Infrared Multi- et al., 1987). A notable focus of attention was the provision of
Spectral Scanner. postlaunch radiometric calibration for the AVHRR sensors on
ADEOS-GLI: Japans Advanced Earth Observing Satellite Global Imager. the series of National Oceanic and Atmospheric Administration
2016 by Taylor & Francis Group, LLC
114 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 4 .1 Selected Literature for Vicarious Calibration Targets Mentioned in the Text
Test Site Literature References
White Sands, New Mexico Castle etal. (1984), Smith and Levin (1985), Smith etal. (1985), Whitlock etal. (1990a,b), Wheeler etal.
(1994)
The Moon Kieffer and Wildey (1985, 1996), Godden and McKay(1997), Kieffer etal. (2002), Barnes etal. (2004), Stone
etal. (2005, 2013), Stone (2008), Xiong etal. (2008)
La Crau, France Xing-Fa etal. (1990), Santer etal. (1992, 1997), Gu etal. (1992), Moran etal. (1995), Richter (1997), Rondeaux
etal. (1998), Six (2002)
Saharan and Arabian Deserts, North Africa Henry etal. (1993), Cosnefroy etal. (1994, 1996, 1997), Cabot (1997), Cabot etal. (1998, 1999, 2000), Miesch
etal. (2003), Rao etal. (2003), Markham etal. (2006), Vermote and Saleous (2006), Wu etal. (2008b), Gamet
etal. (2011), Lachrade etal. (2013)
Atmospheric molecular (Rayleigh) scattering Vermote etal. (1992), Kaufman and Holben (1993), Dilligeard etal. (1997), Meygret etal. (2000)
Saharan and Arabian Deserts, North Africa Henry etal. (1993), Cosnefroy etal. (1994, 1996, 1997), Cabot (1997), Cabot etal. (1998, 1999, 2000), Miesch
etal. (2003), Rao etal. (2003), Markham etal. (2006), Vermote and Saleous (2006), Wu etal. (2008b), Gamet
etal. (2011), Lachrade etal. (2013)
Ocean sun-glint Kaufman and Holben (1993), Vermote and Kaufman (1995), Hagolle etal. (2004), Luderer etal. (2005)
Clouds Desormeaux etal. (1993), Vermote and Kaufman (1995), Iwabuchi (2003), Doelling etal. (2004, 2010, 2013),
Hu etal. (2004), Fougnie and Bach (2009)
Sites in Australia Graetz etal. (1994), Prata etal. (1996), Mitchell etal. (1997), OBrien and Mitchell (2001), De Vries etal. (2007)
Dunhuang, China Min and Zhu (1995), Wu etal. (1994, 1997), Xiao etal. (2001), Hu etal. (2001, 2009, 2010), Liu etal. (2004),
Zhang etal. (2001, 2004, 2008, 2009), Li etal. (2009)
Railroad Valley Playa, USA Scott etal. (1996), Snyder etal. (1997), Thome (2001, 2005), Matsunaga etal. (2001), Thome etal. (2000, 2001,
2003a,b, 2004b, 2006, 2007, 2008), Biggar etal. (2003), Chander etal. (2004a, 2007a), McCorkel etal. (2006),
DAmico etal. (2006), Czapla-Myers etal. (2007, 2008, 2010), Angal etal. (2008), Kerola and Bruegge (2009)
Ivanpah Playa Thome etal. (1996, 2004b), Villa-Aleman etal. (2003a,b), Thome (2005), Rodger etal. (2005)
Snow/ice fields Loeb (1997), Tahnk and Coakley (2001), Nieke etal. (2003), Six etal. (2004)
Negev Desert, Israel Bushlin etal. (1997), Gilead and Karnieli (2004), Faiman etal. (2004), Gilead (2005)
Lunar Lake Playa, USA Thome etal. (1998), Bannari etal. (2005)
Dome-C, Antarctica Six etal. (2004, 2005), Tomasi etal. (2008), Wu etal. (2008b), Wenny and Xiong (2008), Wenny etal. (2009),
Xiong etal. (2009a,b), Bouvet and Ramoino (2010), Cao etal. (2010), Potts etal. (2013)
Frenchman Flat, USA Gross etal. (2007), Helmlinger etal. (2007), Kerola etal. (2009)
Tuz Golu, Turkey Gurol etal. (2008, 2010)
Sonoran Desert, Mexico, aka Yuma, Arizona Angal etal. (2010a,b, 2011), Kim and Lee (2013)
(NOAA) satellites, because of the widespread utilization of systems, given that it is difficult to avoid detector-to-detector dif-
AVHRR image data and because AVHRRs do not have onboard ferences if the whole array cannot be exposed easily to a uniform
calibration systems (Duggin, 1987; Frouin and Gautier, 1987; radiance field. Because significant episodes of airborne particles
Smith etal., 1988; Holben etal., 1990; Staylor, 1990; Teillet etal., can occur in these regions, only the clearest image acquisitions
1990; Brest and Rossow, 1992; Che and Price, 1992; Mitchell are used for radiometric calibration purposes.
etal., 1992; Abel etal., 1993; Kaufman and Holben, 1993; Teillet Additionally, other vicarious calibration or cross-calibration
and Holben, 1994; Cihlar and Teillet, 1995; Rao and Chen, 1995, methods take advantage of nonland Earth targets, including (cf.
1996, 1999; Brest etal., 1997; Tahnk and Coakley, 2001; Cao and Table 4.1 for selected literature references) atmospheric molecu-
Heidinger, 2002; Wu etal., 2003, 2008a,c, 2010; Cao etal., 2005). lar (Rayleigh) scattering, ocean sun-glint, snow/ice fields, and
Other deserts and playas used as reference standard sites clouds. A snowfield of particular interest is the Dome-C site in
include (cf. Table 4.1 for selected literature references) La Crau in Antarctica, where image data from many overpasses are avail-
France, several sites in Australia, Dunhuang in China, Railroad able during the summer months. A key facility for the vicarious
Valley Playa in the United States, Ivanpah Playa in the United calibration of ocean color satellite sensors is the Marine Optical
States, the Negev Desert in Israel, Lunar Lake Playa in the United BuoY (MOBY), an autonomous optically instrumented buoy
States, Frenchman Flat in the United States, the Sonoran Desert moored off the island of Lanai, Hawaii, United States (Clark
in Mexico (also known as Yuma, Arizona, United States), and etal., 1997; Brown etal., 2007).
Tuz Golu in Turkey. About 20 large areas in the Saharan and The Remote Sensing Group at the University of Arizona pioneered
Arabian Deserts of North Africa have been used without surface vicarious calibration methodologies that involved not only WSNM
observations to enable cross-calibration and/or uniformity cali- but also the playas at Edwards Air Force Base, Railroad Valley, Lunar
bration for a variety of sensor systems for several decades. These Lake, and Ivanpah. Many of the key surface measurement method-
large uniform sites are important for the calibration of push- ologies central to vicarious calibration were developed as a result of
broom sensor detector arrays, such as those on the SPOT satellite close collaborations between the University of Arizona and theU.S.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 115
Department of Agriculture (USDA) (Phoenix), focused to a con- The GCOS climate-monitoring principles refer specifically to
siderable extent on field campaigns at the Maricopa Agricultural the critical role of calibrated satellite sensor observations.
Center (e.g., Jackson, 1990), but also involving many field campaigns A generalized approach to the vicarious calibration of mul-
at WSNM and elsewhere. These pioneering efforts, together with tiple EO sensors using hyperspectral data was demonstrated by
leading-edge work by NASA scientists and collaborations with Teillet et al. (1998a,b, 2001b) in a project on quality assurance
scientists at the National Institute for Standards Technology (NIST), and stability reference (QUASAR) monitoring. The approach
played an important role in the calibration programs of NASAs EOS uses spatially extensive airborne or satellite hyperspectral imag-
(e.g., Bruegge and Butler, 1996; Slater etal., 1996; Xiong etal., 2010b). ery of a terrestrial reference standard site as spectroradiometric
Similar methodologies and efforts have also been key elements of reference data to carry out vicarious radiometric calibrations for
other EO systems (e.g., Hill and Aifadopoulou, 1990; Guenther all optical satellite sensors that image the site on the same day.
etal., 1996; Markham etal., 1997; Thome etal., 1997a,b, 2003a,b; The establishment of a global instrumented and automated
Schroeder etal., 2001; Secker etal., 2001; Teillet etal., 2001a, 2006, network of test sites (GIANTS) for postlaunch radiometric
2007a; Biggar etal., 2003; Black etal., 2003; Schiller, 2003; Chander calibration of EO sensors was outlined by Teillet etal. (2001c).
etal., 2004b, 2007a,c, 2013b; Xiong etal., 2006b; Henry etal., 2013). The GIANTS concept proposes that a small number of well-
There have also been systematic and sustained vicarious calibration characterized benchmark test sites and datasets be supported
efforts over many years by the Centre National dEtudes Spatiales for the calibration of all space-based optical sensors imaging
(CNES), the European Space Agency (ESA), the UKs National the Earth. A core set of surface sensors, measurements, and pro-
Physical Laboratory (NPL), the Japan Aerospace Exploration tocols should be standardized across all participating test sites
Agency (JAXA), and the USGS EROS Center. and measurement datasets should undergo identical process-
Vicarious calibration can also be undertaken by imaging ing at a central secretariat. GIANTS is intended to supplement
the Moon, a stable target whose dynamic illumination varia- calibration information already available from other calibration
tions can be computed (cf. Table 4.1 for selected literature refer- systems and efforts, reduce the resources required by individual
ences), and/or by imaging bright stars (e.g., Chang etal., 2012). agencies, and provide greater consistency for terrestrial moni-
Although ultrastable stellar targets are imaged without atmo- toring studies based on multiple sensor systems. The network
spheric effects, they are relatively low radiance sources and, approach is also intended to explore the use of automation, com-
as with imaging the Moon, they require high-risk spacecraft munication, co-ordination, visibility, and education, all of which
platform maneuvers to achieve. Satellite sensors often use data can be facilitated by greater use of advanced ground-based sen-
acquisitions of deep space views without spacecraft platform sor and telecommunication technologies.
maneuvers to provide dark target calibration data.
As the importance of remote sensing calibration was increas-
4.3.2 current Developments
ingly acknowledged and as more terrestrial sites were investi-
gated for use as reference standard test sites for postlaunch More recent initiatives in the domain of reference standard sites
sensor calibration, specialists in the international community include ongoing methodology research (e.g., McCorkel etal., 2013;
endeavored to put together databases of worldwide calibration Wang etal., 2011; Helder etal., 2012a, 2013; Thome, 2012; Thome
facilities, including test sites and instruments. Early pilot efforts and Fox, 2011; Anderson etal., 2013; Govaerts etal., 2013; Thome
were those of the international CEOS in 1993 (spearheaded by etal., 2013), as well as the GEOSS, CEOS, and QA4EO efforts men-
Barton, Guenther, and others) and NASAs EOS in 1995 (spear- tioned in the Introduction. Notably and significantly, the CEOS
headed by Butler, Starr, Reber, and others). Calls for contribu- Working Group on Calibration and Validation (WGCV) subgroup
tions led to editions of such databases (e.g., Butler etal., 2001), on Infrared Visible Optical Sensors (IVOSs) has worked in recent
but these early databases were never fully populated and they years with collaborators around the world to establish a core set
no longer appear to be active. Current efforts in this regard by of CEOS-endorsed, globally distributed, reference standard test
CEOS, ESA, and the USGS are outlined in the next section. sites for the postlaunch calibration of space-based optical imag-
During the 1990s, the role of Earth-monitoring systems of ing sensors. There are now eight CEOS reference instrumented
all kinds was increasingly cast in the context of climate obser- sites (Table 4.2; Figure4.1) and six CEOS reference PICS (Table4.3;
vations. For example, the Global Climate Observing System Figure 4.2). The instrumented sites are mainly used for field cam-
(GCOS) was established in 1992 to focus on satellite and insitu paigns to obtain or update radiometric gain coefficients, and they
observations for climate in the atmospheric, oceanic, and ter- serve as a focus for international efforts, facilitating traceability and
restrial domains.* As part of the process of identifying climate cross-comparison to evaluate biases between current and future
observation requirements in 1999, GCOS established a set of sensors and methods in a harmonized manner. The pseudoinvari-
climate-monitoring principles that were adopted in 2003 by ant reference standard test sites are desert areas characterized by
the World Meteorological Organization (WMO) and CEOS. high surface reflectance, some sand dunes, little or no vegetation,
and low atmospheric aerosol loadings. Such sites can be used to
* http://www.wmo.int/pages/prog/gcos/index.php. evaluate the long-term stability of a given sensor and to facilitate
http://w w w.w mo.int/pages/prog /gcos/index.php?na me= Climate cross-comparisons between multiple sensors (e.g., Helder et al.,
MonitoringPrinciples. 2010; Chander etal., 2013b). They can also be used for the validation
2016 by Taylor & Francis Group, LLC
116 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 4 .2 Core Set of CEOS-Endorsed Instrumented Reference Standard Test Sites
Centre Latitude (Degrees), Longitude
# Site WRS-2 Path/Row (Degrees), and Altitude ASL (m) Point of Contact
1 Dome C, Antarctica 88-89-90/113 74.50, +123.00, Stephen Warren,
3215 University of Washington, USA
2 Dunhuang, China 137/32 +40.13, +94.34, Xiuqing Hu,
1220 National Satellite Meteorological Center, China
3 Frenchman Flat, USA 40/34 +36.81, 115.93, Carol J. Bruegge,
940 NASA Jet Propulsion Laboratory, USA
4 Ivanpah Playa, USA 39/35 +35.5692, 115.3976, Kurtis J. Thome,
813 NASA Goddard Space Flight Center, USA
5 La Crau, France 196/30 +43.47, +4.97, Patrice Henry,
28 CNES, France
6 Negev Desert, Israel 174/39 +30.11, +35.01, Arnon Karnieli,
334 Ben Gurion University, Israel
7 Railroad Valley Playa, USA 40/33 +38.50, 115.69, Kurtis J. Thome,
1435 NASA Goddard Space Flight Center, USA
8 Tuz Golu, Turkey 177/33 +38.83, +33.33, Selime Gurol,
905 Tubitak Uzay (Space Technologies Research Institute), Turkey
Notes: WRS, Landsat Worldwide Reference System; ASL, above sea level. Other characteristics of some of these test sites are given in Teillet etal. (2007b).
Dome C Dunhuang Frenchman Flat
Ivanpah Piaya La Crau Negev
Railroad Valley Tuz Golu
FIg u r e 4.1 Examples of Landsat-7 Enhanced Thematic Mapper Plus (ETM+) imagery of the core set of CEOS-endorsed instrumented reference
standard test sites (N.B., scales differ, north is up in all cases, and site coordinates can be found in Table 4.2).
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 117
TABLe 4 .3 Core Set of CEOS-Endorsed Pseudoinvariant Reference Standard Test Sites
Centre Latitude (Degrees), Longitude (Degrees),
# Site WRS-2 Path/Row and Altitude ASL (m)
1 Libya 4 181/40 +28.55, +23.39, 118
2 Mauritania 1 201/47 +19.40, 9.30, 392
3 Mauritania 2 201/46 +20.85, 8.78, 384
4 Algeria 3 192/39 +30.32, +7.66, 245
5 Libya 1 187/43 +24.42, +13.35, 648
6 Algeria 5 195/39 +31.02, +2.23, 530
Notes: WRS, Landsat Worldwide Reference System; ASL, above sea level. Note that Mauritania 1 and 2 are con-
sidered to be one site. The point of contact for these sites is Patrice Henry, CNES, France. Other characteristics of
these test sites are given in Teillet etal. (2007b).
Libya 4 Mauritania 1 Mauritania 2
Algeria 3 Libya 1 Algeria 5
FIg u r e 4.2 Examples of Landsat-7 Enhanced Thematic Mapper Plus (ETM+) imagery of the core set of CEOS-endorsed pseudo-invariant refer-
ence standard test sites (N.B., scales differ, north is up in all cases, and site coordinates can be found in Table 4.3).
of selected higher-order climate variable products. As a precursor to by ground survey teams. Cost is a significant issue for the instru-
the selection of the CEOS-endorsed core sites, a list of potential can- mentation (and associated performance) needed for test sites for
didates was developed (Teillet etal., 2007b) and an online catalog establishing strategies for the long-term maintenance of CEOS
of worldwide test sites used for sensor characterization was imple- test sites. In addition, having a limited number of sites provides
mented by the USGS (Chander etal., 2007d).* The catalog provides an opportunity for the owners/operators of Earth-viewing opti-
ready access to this vital information by the global community. cal imagers (including commercial operators) to routinely collect
CEOS/WGCV/IVOS is also working toward the establishment data over the CEOS-endorsed reference sites.
of optimum methodologies for the characterization and use of In principle, radiometric calibration coefficients should
the endorsed reference instrumented sites. The principal crite- not depend on the geographic locations of reference standard
rion in selecting these sites, other than spatial uniformity and sites. However, the methodologies involved in vicarious cali-
brightness, was that they were all fully and regularly calibrated by bration via instrumented sites involve characterization of sur-
ground-based instrumentation. In some cases, instrumentation face properties and atmospheric conditions by different teams
is permanently deployed; in others, it is transported to the site for and, pending the adoption of standardized approaches, differ-
specific characterization campaigns. However, in all cases, the ent instrumentation. Similarly, vicarious calibration via PICS
basis for assigning a value to the surface reflectance and its sub- involves methods for atmospheric compensation. These diversi-
sequent propagation to the TOA as radiance (for comparison to ties are deemed beneficial in that they can be used to yield an
satellite imager radiances) is derived from measurements made improved understanding of the role of atmospheric effects and
different modeling approaches on calibration results.
* http://calval.cr.usgs.gov/sites_catalog_map.php.
2016 by Taylor & Francis Group, LLC
118 Remotely Sensed Data Characterization, Classification, and Accuracies
In any given year, there are typically several field campaigns near term: (1) Increase the number of sites from eight to ten. (2)
undertaken by various agencies to carry out vicarious calibra- Gather more complete site characterization data and informa-
tions for a number of satellite optical sensor systems. Thus, a tion, especially with respect to their reflectance anisotropy prop-
logical next step is to formalize and test the GIANTS concept via erties and their temporal stabilities. (3) Define a recommended
the concatenation of CEOS-endorsed core sites (Teillet and Fox, standard set of core measurements. (4) Formalize the set of
2009). The concatenation concept proposes the deployment of as sites into an operational network. (5) Organize local, regional,
many field campaigns at as many of the core sites as possible dur- national, and international field campaigns. (6) Acquire and
ing a given time period (e.g., 1 month) to generate updates for as archive imagery of all of the sites on an ongoing basis. (7) Develop
many satellite optical sensors as possible. The pilot project would online calibration data access infrastructures. (8) Continue to
make a concrete initial test of the GIANTS approach (at this improve vicarious calibration methodologies. The current set of
stage excluding an automated aspect in most if not all instances). endorsed sites is distributed unevenly across the globe, which
The trade-off between automation and the accuracy demands should be kept in mind to the extent possible when selecting
of radiometric calibration is an important consideration that has additional instrumented sites. However, the adoption and use of
yet to be investigated to a significant extent. Several organiza- new sites will depend not only on the suitability of site charac-
tions have begun to explore automated radiometric systems for teristics, but also on the interests, capabilities, and funding of
vicarious calibration (Thome etal., 2004a; Czapla-Myers etal., research groups, as well as logistical and political accessibilities.
2007, 2008; Gross et al., 2007; Helmlinger et al., 2007; Kerola Still with respect to the CEOS-endorsed reference standard
etal., 2009; Meygret etal., 2011). Another topic that has received instrumented sites, the following steps are recommended for the
insufficient attention by the calibration research community is longer term: (1) Continue to acquire and archive imagery on an
the use of data at wavelengths outside the solar-reflective spec- ongoing basis for all of the sites. (2) Continue to improve vicari-
trum to help characterize reference standard sites used for opti- ous calibration methodologies. (3) Establish traceability chains
cal sensor calibration (e.g., Teillet et al., 1995; Blumberg and for site measurement data. (4) Develop recommended guidelines
Freilikher, 2001; Floury etal., 2002; Brogioni etal., 2006). and best practices for using the network of sites. (5) Endorse and
The U.K.s NPL has contributed significantly in a number of advocate compliance with recommended guidelines and best
ways to new methods and systems, as well as to international practices. ISO standards for radiometric calibration of remote
collaboration, for radiometric calibration of satellite sensors sensing data have been proposed (Di, 2004), making a high-level
(Fox, 1999), including the development of bold new space mis- start toward certification.
sion concepts for advanced cryogenic-based radiometric cali-
bration capabilities (Fox etal., 2002). NPL has also coordinated 4.4 concluding Remarks
several CEOS campaigns in the context of GEO tasks to compare
satellite calibration approaches, recently using the Dome-C site This chapter has provided a broad overview of the postlaunch
and also leading major comparison exercises focused on the Tuz radiometric calibration of image data from satellite sensor
Golu site in Turkey.* systems operating in the solar-reflective optical domain. Even
CEOS is providing data and information for calibration and val- though satellite sensors are very well characterized radiometri-
idation (cal/val) of EO data through its Cal/Val Portal. The origi- cally prior to launch, postlaunch changes to satellite sensors and
nal Cal/Val Portal started in 2007 and a new version is now moving their radiometric calibration performance necessitate ongoing
to a completely operational system providing new tools, datasets, calibration monitoring and adjustment.
results, and upcoming campaign information on a regular basis. Onboard radiometric calibration systems have been imple-
Among other initiatives, the enhanced Cal/Val Portal supports mented since the early days of satellite imaging sensors. This
measurement and comparison campaigns by means of a prototype chapter highlighted the rich history of Landsat onboard sensor
database with data and information on reference standard sites. calibration and the significant efforts that were required by the
absence of such capability for the AVHRR sensors, which pro-
vided data for countless broad-scale studies of the Earth (e.g.,
4.3.3 next Steps
Townshend, 1994; Townshend et al., 1994; Teillet et al., 2000).
Advocacy of the use of CEOS-endorsed reference standard The chapter focused extensively on vicarious calibration meth-
sites and the establishment and use of some sort of certifica- odologies, important complements to the use of complex and
tion process as part of best practice methodologies should lead expensive onboard calibration systems that seldom test the
to improved measurement consistency between satellite sen- full optics of sensors. Terrestrial reference standard sites for
sors, reduce costs, and help underpin the accurate monitoring vicarious calibration were featured, and the relevant pioneering
of planetary change. research over many decades was documented.
With respect to the CEOS-endorsed reference standard Todays mainstream information society routinely makes
instrumented sites, the following steps are recommended for the extensive qualitative use of satellite imagery. With respect to quan-
titative use, the significant improvement in the knowledge of the
* http://calvalportal.ceos.org/cvp/web/guest/tuz-golu-campaign. radiometric performance of satellite sensor systems, as described
http://calvalportal.ceos.org. in this chapter, is indicative of the advancement in the state of
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 119
the art of postlaunch radiometric calibration methods. It is also Ahern, F.J. and Murphy, J. l979. Radiometric calibration and cor-
indicative of the need for persistent calibration of satellite sensors rection of LANDSAT l, 2, and 3 MSS Data. Canada Centre
over mission lifetimes, as well as the need for a variety of calibra- for Remote Sensing Research Report 78-4, Canada Centre
tion methodologies in order to assess the true radiometric perfor- for Remote Sensing, Ottawa, Ontario, Canada.
mance of any given sensor as accurately as possible. Accordingly, Anderson, D., Jucks, K.W., and Young, D.F. 2008. The NRC decadal
the use of terrestrial reference standard sites for postlaunch mon- survey climate absolute radiance and refractivity obser-
itoring of satellite sensor radiometric performance increasingly vatory: NASA implementation. Proceedings of the IEEE
constitutes a key component of current and future satellite sen- International Geoscience and Remote Sensing Symposium
sor calibration strategies. With continued international coopera- (IGARSS), Boston, MA, pp. 911.
tion on test site calibration methodologies among remote sensing Anderson, K. and Milton, E.J. 2006. On the temporal stability
agencies and organizations, remotely sensed EO data can be made of ground calibration targets: Implications for the repro-
more accurate and useful to help develop an improved and sound ducibility of remote sensing methodologies. International
understanding of our planet and its climate. Journal of Remote Sensing, 27(16): 33653374.
The success of climate monitoring missions will depend Anderson, N., Czapla-Myers, J., Leisso, N., Biggar, S., Burkhart,C.,
critically on significant advances in postlaunch calibration. Kingston, R., and Thome, K. 2013, Design and calibration
The planned Climate Absolute Radiance and Refractivity of field deployable ground-viewing radiometers. Applied
Observatory (CLARREO) (Anderson etal., 2008; Sandford etal., Optics, 52(2): 231240.
2010; Thome etal., 2010; Lukashin etal., 2013) and the proposed Angal, A., Chander, G., Choi, T., Wu, A., and Xiong, X. 2010a.
Traceable Radiometry Underpinning Terrestrial and Helio The use of the Sonoran Desert as a pseudo-invariant site
Studies (TRUTHS) mission (Fox et al., 2002, 2011) offer novel for optical sensor cross-calibration and long-term stability
approaches to the provision of key scientific data with unprec- monitoring. Proceedings of the 2010 International Geoscience
edented radiometric accuracy for terrestrial and solar studies. and Remote Sensing Symposium (IGARSS), Honolulu, HI,
Missions such as CLARREO and TRUTHS will establish well- pp. 16561659.
calibrated reference targets and standards to support other EO Angal, A., Chander, G., Xiong, X., Choi, T., and Wu, A. 2011.
missions as well as solar and lunar observations. Once launched, Characterization of the Sonoran desert as a radiometric
these sensor systems would in effect provide the de facto pri- calibration target for Earth observing sensors. Journal of
mary reference standards for the global ensemble of EO systems. Applied Remote Sensing, 5(1): 059502.
CLARREO and TRUTHS promise to improve the performance Angal, A., Choi, T.J., Chander, G., and Xiong, X. 2008. Monitoring
and accuracy of EO missions by an order of magnitude (to on-orbit stability of Terra MODIS and Landsat 7 ETM+
accuracies on the order of 0.3% in the solar reflective domain). reflective solar bands using Railroad Valley Playa, Nevada
Postlaunch radiometric calibration based on metrology-quality (RVPN) test site. Proceedings of IEEE Conference on Remote
observatories in space lies at the heart of these proposed mis- Sensing: The Next Generation, Vol. IV, Boston, MA, pp.
sions, which, as a result, will be integral and indispensable tools 13641367. IEEE, Piscataway, NJ.
to help maximize the benefits of all EO systems. Angal, A., Xiong, X., Choi, T.-Y., Chander, G., and Wu, A. 2010b.
Using the Sonoran and Libyan desert test sites to monitor
Acknowledgments the temporal stability of reflective solar bands for Landsat
7 Enhanced Thematic Mapper Plus and Terra MODerate-
The postlaunch radiometric calibration methodologies docu- resolution imaging spectroradiometer sensors. Journal of
mented in this chapter are the culmination of decades of effort Applied Remote Sensing, 4(1): 043525-12.
on the part of many individuals and agencies. The authors thank Ashburn, E.V. and Weldon, R.G. 1956. Spectral diffuse reflectance
CEOS/WGCV/IVOS members in particular for their stewardship of desert surfaces. Journal of the Optical Society of America,
and contributions toward progress in EO calibration and validation. 46(8): 583586.
Bannari, A., Omari, K., Teillet, P.M., and Fedosejevs, G. 2005.
References Potential of Getis statistics to characterize the radiometric
uniformity and stability of test sites used for the calibra-
Abel, P., Guenther, B., Galimore, R.N., and Cooper, J.W. 1993. tion of Earth observation sensors. IEEE Transactions on
Calibration results for NOAA-11 AVHRR channels 1 and Geoscience and Remote Sensing, 43(12): 29182926.
2 from congruent path aircraft observations. Journal of Barker, J.L., ed. 1985. Proceedings of the Landsat-4 Science
Atmospheric and Oceanic Technology, 10(4): 493508. Characterization Early Results Symposium, February 2224,
Ahern, F.J., Brown, R.J., Cihlar, J., Gauthier, R., Murphy, J., Neville, 1983, Greenbelt, MD, NASA Conference Publication 2355.
R.A., and Teillet, P.M. 1988. Radiometric correction of vis- Barnes, R.A., Eplee, R.E. Jr., Patt, F.S., Kieffer, H.H., Stone,
ible and infrared remote sensing data at the Canada Centre T.C., Meister, G., Butler, J.J., and McClain, C.R. 2004.
for Remote Sensing. In Remote Sensing Yearbook 1988, Comparison of SeaWiFS measurements of the moon with
A. Cracknell and L. Hayes, eds. Taylor & Francis Group, the U.S. geological survey lunar model. Applied Optics,
Philadelphia, PA. 43(31): 58385854.
2016 by Taylor & Francis Group, LLC
120 Remotely Sensed Data Characterization, Classification, and Accuracies
Barsi, J.A., Hook, S.J., Schott, J.R., Raqueno, N.G., and Markham, B.L. Bruegge, C. and Butler, J., eds. 1996. Special issue on Earth
2007. Landsat-5 Thematic Mapper thermal band calibration observing system calibration. Journal of Atmospheric and
update. Geoscience and Remote Sensing Letters, 4(4): 552555. Oceanographic Technology, 13(2): 273544.
Barsi, J.A., Schott, J.R., Palluconi, F.D., Helder, D.L., Hook, S.J., Bushlin, Y., Ben-Shalom, A., Sheffer, D., Steinman, A., Dimmeler, A.,
Markham, B.L., Chander, G., and ODonnell, E.M. 2003. Clement, D., and Strobel, R. 1997. Background properties
Landsat TM and ETM+ thermal band calibration. Canadian in arid climates: Measurements and analysis. Proceedings
Journal of Remote Sensing, 29(2): 141153. of SPIE Conference 3062 on Targets and Backgrounds:
Bartolucci, L.A. and Davis, S.M. 1983. The calibration of Landsat Characterization and Representation III, Orlando, FL,
MSS data as an analysis tool. LARS Technical Reports Paper W.R. Watkins and D. Clement, eds., pp. 311321. SPIE,
71, LARS Technical Report 062283. Proceedings of the 1983 Bellingham, WA.
Machine Processing of Remotely Sensed Data Symposium, Butler, J.J., Johnson., B.C., and Barnes, R.A. 2005. The calibration
West Lafayette, IN, pp. 279287. and characterization of Earth remote sensing and environ-
Biggar, S.F., Slater, P.N., and Gellman, D.I. 1994. Uncertainties mental monitoring instruments. In Optical Radiometry,
in the in-flight calibration of sensors with reference to Vol 41: Experimental Methods in the Physical Sciences,
measured ground sites in the 0.4 to 1.1 m range. Remote A.C.Parr, R.U. Datla, and J.L. Gardner, eds., R. Celotta and
Sensing of Environment, 48: 245252. T.Lucatorto, treatise eds. Elsevier/Academic Press, Waltham,
Biggar, S.F., Thome, K.J., and Wisniewski, W. 2003. Vicarious MA, pp. 453534.
radiometric calibration of EO-1 sensors by reference to Butler, J.J., Wanchoo, L., and Truong, L. 2001. CEOS database
high-reflectance ground targets. IEEE Transactions on of worldwide calibration facilities and validation test sites.
Geoscience and Remote Sensing, 41(6): 11741179. Proceedings of SPIE Conference 4169 on Sensors, Systems,
Black, S.E., Helder, D.L., and Schiller, S.J. 2003. Irradiance-based and Next-Generation Satellites IV, Barcelona, Spain,
cross-calibration of Landsat-5 and Landsat-7 Thematic H. Fujisada, J.B. Lurie, A. Ropertz, and K. Weber, eds.,
Mapper sensors. International Journal of Remote Sensing, pp. 202208. SPIE, Bellingham, WA.
24(2): 287304. Cabot, F. 1997. Proposal for the development of a repository
Blumberg, D.G. and Freilikher, V. 2001. Soil water-content and sur- for in-flight calibration of optical sensors over terrestrial
face roughness retrieval using ERS-2 SAR data in the Negev targets. Proceedings of SPIE Conference 3117 on Earth
Desert, Israel. Journal of Arid Environments, 49(3): 449464. Observing Systems II, San Diego, CA, W.L. Barnes, ed.,
Bouvet, M. and Ramoino, F. 2010. Radiometric intercomparison of pp.148155. SPIE, Bellingham, WA.
AATSR, MERIS, and Aqua MODIS over Dome Concordia Cabot, F., Hagolle, O., Cosnefroy, H., and Briottet, X. 1998.
(Antarctica). Canadian Journal of Remote Sensing, 36(5): Intercalibration using desertic sites as a reference target.
464473. Proceedings of International Geoscience and Remote Sensing
Bowker, D.E. and Davis, R.E. 1992. Influence of atmospheric aero- Symposium, Seattle, WA, Vol. 5, pp. 27132715.
sols and desert reflectance properties on satellite radiance Cabot, F., Hagolle, O., and Henry, P. 2000. Relative and multitem-
measurements. International Journal of Remote Sensing, poral calibration of AVHRR, SeaWiFS, and VEGETATION
13(16): 31053126. using POLDER characterization of desert sites. Proceedings
Brest, C.L. and Rossow, W.B. 1992. Radiometric calibration and of International Geoscience and Remote Sensing Symposium,
monitoring of NOAA AVHRR data for ISCCP. International Honolulu, HI, pp. 21882190.
Journal of Remote Sensing, 13(2): 235273. Cabot, F., Hagolle, O., Ruffel, C., and Henry, P.J. 1999. Remote
Brest, C.L., Rossow, W.B., and Roiter, M.D. 1997. Update of radi- sensing data repository for in-flight calibration of opti-
ance calibrations for ISCCP. Journal of Atmospheric and cal sensors over terrestrial targets. Proceedings of SPIE
Oceanic Technology, 14(5): 10911109. Conference 3750 on Earth Observing Systems IV, Denver,
Brogioni, M., Macelloni, G., and Pampaloni, P. 2006. Temporal CO, pp. 514523, W.L. Barnes, ed. SPIE, Bellingham, WA.
and spatial variability of multi-frequency microwave emis- Cao, C. and Heidinger, A.K. 2002. Inter-comparison of the
sion from the east Antarctic plateau. Proceedings of IEEE longwave infrared channels of MODIS and AVHRR/
Conference on Remote Sensing: A Natural Global Partnership, NOAA-16 using simultaneous nadir observations at orbit
Denver, CO, pp. 38203823. IEEE, Piscataway, NJ. intersections. Proceedings of SPIE Conference 4814 on
Brown, S.W., Flora, S.J., Feinholz, M.E., Yarbrough, M.A., Houlihan, Earth Observing Systems VII, Seattle, WA, W.L. Barnes, ed.,
T., Peters, D., Kim, Y.S., Mueller, J., Johnson, B.C., and Clark, pp. 306316. SPIE, Bellingham, WA.
D.K. 2007. The Marine Optical BuoY (MOBY) radiometric Cao, C., Uprety, S., Xiong, X., Wu, A., Jing, P., Smith, D.,
calibration and uncertainty budget for ocean color satellite Chander, G., Fox, N., and Ungar, S. 2010. Establishing
sensor vicarious calibration. Proceedings of SPIE Conference the Antarctic Dome C community reference standard site
6744 on Sensors, Systems, and Next-Generation Satellites XI, towards consistent measurements from Earth observa-
Florence, Italy, R. Meynart, S.P. Neeck, H. Shimoda, and tion satellites. Canadian Journal of Remote Sensing, 36(5):
S.Habib, eds. SPIE, Bellingham, WA, pp. 67441M. 498513.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 121
Cao, C., Xu, H., Sullivan, J., McMillin, L., Ciren, P., and Hou, Y. Chander, G., Meyer, D.J., and Helder, D.L. 2004b. Cross calibration of
2005. Intersatellite radiance biases for the high resolu- the Landsat-7 ETM+ and EO-1 ALI sensors. IEEE Transactions
tion infrared radiation sounders (HIRS) onboard NOAA- on Geoscience and Remote Sensing, 42(12): 28212831.
15, -16, and -17 from simultaneous nadir observations. Chander, G., Mishra, N., Helder, D.L., Aaron, D., Angal, A., Choi,T.,
Journal of Atmospheric and Oceanic Technology, 22(4): Xiong, X., and Doelling, D. 2013c. Applications of spectral
381395. band adjustment factors (SBAF) for cross-calibration. IEEE
Castle, K.R., Holm, R.G., Kastner, C.J., Palmer, J.M., Slater, P.N., Transactions on Geoscience and Remote Sensing, 51(3 SI):
Dinguirard, M., Ezra, C.E., Jackson, R.D., and Savage, 12671281.
R.K. 1984. In-flight absolute radiometric calibration of the Chang, I.L., Dean, C., Li, Z., Weinreb, M., Wu, X., and Swamy,
Thematic Mapper. IEEE Transactions on Geoscience and P.A.V.B. 2012. Refined algorithms for star-based monitoring
Remote Sensing, 22(3): 251255. of GOES Imager visible channel responsivities. Proceedings
Chander, G., Angal, A., Choi, T.J., Meyer, D.J., Xiong, X.J., and of SPIE Earth Observing Systems XVII, San Diego, CA,
Teillet, P.M. 2007c. Cross-calibration of the Terra MODIS, pp.851 00R.
Landsat 7 ETM+ and EO-1 ALI sensors using near- Che, N. and Price, J.C. 1992. Survey of radiometric calibration
simultaneous surface observation over the Railroad Valley results and methods for visible and near-infrared channels
Playa, Nevada, test site. Proceedings of SPIE Conference 6677 of NOAA-7, NOAA-9, and NOAA-11 AVHRRs. Remote
on Earth Observing Systems XII, San Diego, CA, J.J. Butler Sensing of Environment, 41(1): 1927.
and J. Xiong, eds., pp. 66770Y. SPIE, Bellingham, WA. Chen, H.S. 1996. Remote Sensing Calibration Systems An
Chander, G., Christopherson, J.B., Stensaas, G.L., and Teillet, P.M. Introduction. A. Deepak Publishing, Hampton, VA, ISBN
2007d. Online catalogue of worldwide test sites for the post- 0-937194-38-7.
launch characterization and calibration of optical sensors. Choi, T., Xiong, X., Wang, Z., and Link, D. 2013. Terra and aqua
Proceedings of the 58th International Astronautical Congress, MODIS on-orbit spectral characterization for reflective
Hyderabad, India. International Astronautical Federation, solar bands. Proceedings of SPIE Conference 8724, Ocean
pp. 20432051. Sensing and Monitoring V, 87240Y, doi:10.1117/12.2016640.
Chander, G., Helder, D.L., Malla, R., Micijevic, E., and Mettler, Cihlar, J. and P.M. Teillet. 1995. Forward piecewise linear calibra-
C.J. 2007a. Consistency of L4 TM absolute calibration with tion model for quasi-real-time processing of AVHRR data.
respect to the L5 TM sensor based on near-simultaneous Canadian Journal of Remote Sensing, 21(1): 2227.
image acquisition. Proceedings of SPIE Conference 6677 on Clark, D., Gordon, H., Voss, K., Ge, Y., Broenkow, W., and Trees, C.
Earth Observing Systems XII, San Diego, CA, J.J. Butler and 1997. Validation of atmospheric correction over the oceans.
J. Xiong, eds., pp. 66770F. SPIE, Bellingham, WA. Journal of Geophysical Research, 102(D14): 1720917217.
Chander, G., Helder, D.L., Markham, B.L., Dewald, J.D., Kaita,E., Cosnefroy, H., Briottet, X., Leroy, M., Lecomte, P., and Santer, R.
Thome, K.J., Micijevic, E., and Ruggles, T.A. 2004a. 1994. In field characterization of Saharan sites reflectances
Landsat-5 TM reflective-band absolute radiometric calibra- for the calibration of optical satellite sensors. Proceedings
tion. IEEE Transactions on Geoscience and Remote Sensing, of IEEE Conference on Surface and Atmospheric Remote
42: 27472760. Sensing: Technologies, Data Analysis and Interpretation,
Chander, G., Hewison, T.J., Fox, N., Wu, X., Xiong, X., and Pasadena, CA, pp. 15001502. IEEE, Piscataway, NJ.
Blackwell, W.J., Guest eds. 2013a. Special issue on Cosnefroy, H., Briottet, X., Leroy, M., Lecomte, P. and Santer, R.
Intercalibration of satellite instruments. IEEE Transactions 1997. A field experiment in Saharan Algeria for the cali-
on Geoscience and Remote Sensing, 51(3 SI): 491 pp. bration of optical satellite sensors. International Journal of
Chander, G., Hewison, T.J., Fox, N., Wu, X., Xiong, X., and Remote Sensing, 18(16): 33373359.
Blackwell, W.J. 2013b. Overview of intercalibration of sat- Cosnefroy, H., Leroy, M., and Briottet, X. 1996. Selection and
ellite instruments. IEEE Transactions on Geoscience and characterization of Saharan and Arabian desert sites for
Remote Sensing, 51(3 SI): 10561080. the calibration of optical satellite sensors. Remote Sensing of
Chander, G. and Markham, B.L. 2003. Revised Landsat-5 TM Environment, 58(1): 101114.
radiometric calibration procedures and postcalibration Coulson, K.L. and Jacobowitz, H. 1972. Proposed calibration tar-
dynamic ranges. IEEE Transactions on Geoscience and get for the visible channel of a satellite radiometer. NOAA
Remote Sensing, 41: 26742677. Technical Report NESS 62, UDC 551.507.362.2:551.508.21,
Chander, G., Markham, B.L., and Barsi, J.A. 2007b. Revised National Oceanic and Atmospheric Administration,
Landsat-5 Thematic Mapper radiometric calibration. IEEE Washington, DC, 31 pp.
Geoscience and Remote Sensing Letters, 4: 490494. Czapla-Myers, J.S., Thome, K.J., and Buchanan, J.H. 2007.
Chander, G., Markham, B.L., and Helder, D.L. 2009. Summary Implication of spatial uniformity on vicarious calibration
of current radiometric calibration coefficients for Landsat using automated test sites. Proceedings of SPIE Conference
MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sensing of 6677 on Earth Observing Systems XII, San Diego, CA, J.J.
Environment, 113: 893903. Butler and J. Xiong, eds., pp. 66770U. SPIE, Bellingham, WA.
2016 by Taylor & Francis Group, LLC
122 Remotely Sensed Data Characterization, Classification, and Accuracies
Czapla-Myers, J.S., Thome, K.J., Cocilovo, B.R., McCorkel, J.T., Doelling, D.R., Morstad, D., Scarino, B.R., Bhatt, R., and Gopalan,
and Buchanan, J.H. 2008. Temporal, spectral, and spatial A. 2013. The characterization of deep convective clouds as
study of the automated vicarious calibration test site at an invariant calibration target and as a visible calibration
Railroad Valley, Nevada. Proceedings of SPIE Conference technique. IEEE Transactions on Geoscience and Remote
7081 on Earth Observing Systems XIII, San Diego, CA, J.J. Sensing, 51(3 SI): 11471159.
Butler and J. Xiong, eds., pp. 70810I. SPIE, Bellingham, WA. Doelling, D.R., Nguyen, L., and Minnis, P. 2004. On the use of deep
Czapla-Myers, J.S., Thome, K.J., and Leisso, N.P. 2010. Radiometric convective clouds to calibrate AVHRR data. Proceedings
calibration of earth-observing sensors using an automated of SPIE Conference 5542 on Earth Observing Systems IX,
test site at Railroad Valley, Nevada. Canadian Journal of Denver, CO, Barnes, W.L. and Butler, J.J., eds., pp. 281289.
Remote Sensing, 36(5): 474487. Duggin, M.J. 1987. Impact of radiance variations on satellite sen-
DAmico, J., Thome, K., and Czapla-Myers, J. 2006. Validation sor calibration. Applied Optics, 26(7): 12641271.
of large-footprint reflectance-based calibration using Faiman, D., Feuermann, D., Ibbetson, P., Medwed, B., Zemel, A.,
coincident MODIS and ASTER data. Proceedings of SPIE Ianetz, A., Liubansky, V., Setter, I., and Suraqui, S. 2004. The
Conference 6296 on Earth Observing Systems XI, San Diego, Negev radiation survey. Journal of Solar Energy Engineering-
CA, J.J. Butler and J. Xiong, eds., pp. 629612629618. SPIE, Transactions of the ASME, 126(3): 906914.
Bellingham, WA. Floury, N., Drinkwater, M., and Witasse, O. 2002. L-band bright-
Davis, J.M. and Cox, S.K. 1982. Reflected solar radiances from ness temperature of ice sheets in Antarctica: Emission
regional scale scenes. Journal of Applied Meteorology, modelling, ionospheric contribution and temporal sta-
21(11): 16981712. bility. Proceedings of IEEE Conference on Remote Sensing:
De Vries, C., Danaher, T., Denham, R., Scarth, P., and Phinn, S. Integrating Our View of the Planet, Toronto, Ontario,
2007. An operational radiometric calibration procedure for Canada, pp. 21032105. IEEE, Piscataway, NJ.
the Landsat sensors based on pseudo-invariant target sites. Fougnie, B. and Bach, R. 2009. Monitoring of radiometric sensitiv-
Remote Sensing of Environment, 107(3): 414429. ity changes of space sensors using deep convective clouds
Desormeaux, Y., Rossow, W.B., Brest, C.L., and Campbell, G.G. Operational application to PARASOL. IEEE Transactions
1993. Normalization and calibration of geostationary on Geoscience and Remote Sensing, 47(3): 851861.
satellite radiances for the International Satellite Cloud Fox, N., Kaiser-Weiss, A., Schmutz, W., Thome, K., Young, D.,
Climatology Project. Journal of Atmospheric and Oceanic Wielicki, B., Winkler, R., and Woolliams, E. 2011. Accurate
Technology, 10(3): 304325. radiometry from space: An essential tool for climate stud-
Di, L. 2004. A proposed ISO/TC 211 standards project on radio- ies. Philosophical Transactions of the Royal Society A-
metric calibration of remote sensing data. Proceedings of Mathematical, Physical & Engineering Sciences, 369(1953):
the International Workshop on Radiometric and Geometric 40284063.
Calibration, ISPRS Book Series Volume 2: Postlaunch Fox, N.P. 1999. Improving the accuracy and traceability of radio-
Calibration of Satellite Sensors, Gulfport, MS, S.A. Morain metric measurements to SI for remote sensing instru-
and A.M. Budge, eds., pp. 5356. Taylor & Francis Group, mentation. Proceedings of the 4th International Airborne
London, U.K. Remote Sensing Conference and Exhibition/21st Canadian
Dilligeard, E., Briottet, X., Deuze, J.L., and Santer, R.P. 1997. Symposium on Remote Sensing, Ottawa, Ontario, Canada,
SPOT calibration of blue and green channels using Rayleigh Vol. I, pp. 304311. ERIM International, Ann Arbor, MI.
scattering over clear oceans. Proceedings of SPIE Conference Fox, N.P., Aiken, J., Barnett, J.J., Briottet, X., Carvell, R., Froehlich,
on Advanced Next-Generation Satellites II, Taormina, Italy, C., Groom, S.B., Hagolle, O., Haigh, J.D., Kieffer, H.H., Lean,
pp. 373379. J., Pollock, D.B., Quinn, T.J., Sandford, M.C.W., Schaepman,
Dinguirard, M. and Slater, P.N. 1999. Calibration of space- M.E., Shine, K.P., Schmutz, W.K., Teillet, P.M., Thome,
multispectral imaging sensors: A review. Remote Sensing of K.J.,Verstraete, M.M., and Zalewski, E.F. 2002. Traceable radi-
Environment, 68(3): 194205. ometry underpinning terrestrial- and helio-studies (TRUTHS).
Doelling, D.R., Hong, G., Morstad, D., Bhatt, R., Gopalan, A., Advances in Space Research, 32(11): 22532261.
and Xiong, X. 2010. The characterization of deep convec- Fraser, R.S. and Kaufman, Y.J. 1986. Calibration of satellite sen-
tive cloud albedo as a calibration target using MODIS sors after launch. Applied Optics, 25(7): 11771185.
reflectances. Proceedings of SPIE Conference 7862 on Earth Frouin, R. and Gautier, C. 1987. Calibration of NOAA-7 AVHRR,
Observing Missions Sensors: Development, Implementation, GOES-5, and GOES-6 VISSR/VAS solar channels. Remote
and Characterization, Incheon, Republic of Korea, X. Xiong, Sensing of Environment, 22(1): 73101.
C. Kim, and H. Shimoda, eds., 78620I. Gamet, P., Lachrade, S., Fougnie, B., and Thomas, C. 2011. Calibration
Doelling, D.R., Lukashin, C., Minnis, P., Scarino, B., and Morstad,D. of VIS/NIR sensors over desert sites: New results for cross- and
2012. Spectral reflectance corrections for satellite intercali- multitemporal calibration. Proceedings of the Annual CALCON
brations using SCHIAMACHY data. IEEE Geoscience and Conference on Characterization and Radiometric Calibration
Remote Sensing Letters, 9(1): 119123. for Remote Sensing, Logan, UT, pp. 110.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 123
Gilead, U. 2005. Locating and Examining Potential Sites for Hagolle, O., Nicolas, J.M., Fougnie, B., Cabot, F., and Henry, P.
Vicarious Radiometric Calibration of Space Multi-Spectral 2004. Absolute calibration of VEGETATION derived from
Imaging Sensors in the Negev Desert. Ben Gurion University, an interband method based on sunglint over ocean sites.
Beer Sheva, Israel, 105 pp. IEEE Transactions on Geoscience and Remote Sensing, 42(7):
Gilead, U. and Karnieli, A. 2004. Locating potential vicari- 14721481.
ous calibration sites for high-spectral resolution sensors Helder, D., Boncyk, W., and Morfitt, R. 1997. Landsat TM memory
in the Israeli Negev Desert by GIS analysis. In Postlaunch effect characterization and correction. Canadian Journal of
Calibration of Satellite Sensors, ISPRS Book Series Volume 2, Remote Sensing, 23(4): 299308.
Proceedings of the International Workshop on Radiometric Helder, D., Thome, K., Aaron, D., Leigh, L., Czapla-Myers, J.,
and Geometric Calibration, Gulfport, MS, S.A. Morain and Leisso, N., Biggar, S., and Anderson, N. 2012a. Recent sur-
A.M. Budge, eds., pp. 181187. A.A. Balkema Publishers, face reflectance measurement campaigns with emphasis on
Leiden, the Netherlands. best practices, SI traceability and uncertainty estimation.
Godden, G.D. and McKay, C.A. 1997. A strategy for observing the Metrologia, 49(2): S21S28.
moon to achieve precise radiometric stability monitoring. Helder, D., Thome, K., Mishra, N., Chander, G., Xiong, X., Angal,
Canadian Journal of Remote Sensing, 23(4): 333341. A., and Choi, T. 2013. Absolute radiometric calibration
Govaerts Y., Sterckx, S., and Adriaensen, S. 2013. Use of simulated of Landsat using a pseudo invariant calibration site. IEEE
reflectances over bright desert target as an absolute calibra- Transactions on Geoscience and Remote Sensing, 51(3 SI):
tion reference. Remote Sensing Letters, 4(6): 523531. 13601369.
Goward, S.N., Arvidson, T., Williams, D.L., Faundeen, J.L., Irons, Helder, D.L. 1996. A radiometric calibration archive for Landsat
J.R., and Franks, S. 2006. Historical record of Landsat TM. Proceedings of the SPIE Conference 2758 on Algorithms
global coverage-mission operations, NSLRSDA, and inter- for Multispectral and Hyperspectral Imagery, Orlando, FL,
national cooperator stations. Photogrammetric Engineering pp. 273284.
andRemote Sensing, 72(10): 11551169. Helder, D.L., Barker, J., Boncyk, W., and Markham, B.L. 1996.
Goward, S.N. and Masek, J.G., eds. 2001. Special Issue on Landsat Short term calibration of Landsat TM: Recent findings and
7. Remote Sensing of Environment, 78 (12): 222 pp. suggested techniques. Proceedings of 1996 International
Graetz, R.D., Wilson, M.A., Prata, A.J., Barton, I.J., and Mitchell, Geoscience and Remote Sensing Symposium (IGARSS96),
R.M. 1994. A Continental Instrumented Ground Site Lincoln, NE, pp. 12861289.
Network (CIGSN, Australia): A prerequisite for the detec- Helder, D.L., Basnet, B., and Morstad, D.L. 2010. Optimized
tion, interpretation and quantification of global change. identification of worldwide radiometric pseudo-invariant
Proceedings of 1994 International Geoscience and Remote calibration sites. Canadian Journal of Remote Sensing, 36(5):
Sensing Symposium (IGARSS94), Pasadena, CA, pp. 1254 527539.
1256. IEEE, Piscataway, NJ. Helder, D.L., Boncyk, W., and Morfitt, R. 1998a. Absolute cali-
Gross, H.N., Bruegge, C.J., and Helmlinger, M.C. 2007. bration of the Landsat Thematic Mapper using the internal
Unattended vicarious calibration of a low Earth orbit calibrator. Proceedings of 1998 International Geoscience and
visible-near infrared sensor. Proceedings of AIAA Space Remote Sensing Symposium (IGARSS98), Seattle, WA, pp.
2007 Conference, Long Beach, CA, pp. 786793. AIAA, 27162718.
Reston, VA. Helder, D.L., Boncyk, W., and Morfitt, R. 1998b. Landsat TM
Gu, X.F., Guyot, G., and Verbrugghe, M. 1992. Evaluation of mea- memory effect characterization and correction. Canadian
surement errors in ground surface reflectance for satellite Journal of Remote Sensing, 23(4): 299301.
calibration. International Journal of Remote Sensing, 13(14): Helder, D.L., Karki, S., Bhatt, R., Micijevic, E., Aaron, D., and
25312546. Jasinski, B. 2012b. Radiometric calibration of the Landsat
Guenther, B., Barnes, W., Knight, E., Barker, J., Harnden, J., Weber, MSS sensor series. IEEE Transactions on Geoscience and
R., Roberto, M., Godden, G., Montgomery, H., and Abel, P. Remote Sensing, 50(6): 23802399.
1996. MODIS calibration: A brief review of the strategy for Helder, D.L., Markham, B.L., Thome, K.J., Barsi, J.A., Chander,
the at-launch calibration approach. Journal of Atmospheric G., and Malla, R. 2008. Updated radiometric calibra-
and Oceanic Technology, 13(2): 274285. tion for the Landsat-5 Thematic Mapper reflective bands.
Grol, S., Behnert, I., zen, H., Deadman, A., Fox, N., and IEEE Transactions on Geoscience and Remote Sensing, 46:
Lelolu, U.M. 2010. Tuz Gl: New CEOS reference stan- 33093325.
dard test site for infrared visible optical sensors. Canadian Helder, D.L. and Micijevic, E. 2004. Landsat-5 Thematic Mapper
Journal of Remote Sensing, 36(5): 553565. outgassing effects. IEEE Transactions on Geoscience and
Gurol, S., Ozen, H., Leloglu, U.M., and Tunali, E. 2008. Tuz Golu: Remote Sensing, 42: 27172729.
New absolute radiometric calibration test site. Proceedings Helder, D.L. and Ruggles, T.A. 2004. Landsat Thematic Mapper
of the XXI ISPRS Congress: Silk Road for Information from reflective-band radiometric artifacts. IEEE Transactions on
Imagery, Beijing, China, pp. 3540. Geoscience and Remote Sensing, 42: 27042716.
2016 by Taylor & Francis Group, LLC
124 Remotely Sensed Data Characterization, Classification, and Accuracies
Helmlinger, M.C., Bruegge, C.J., Lubka, E.H., and Gross, H.N. Kamstrup, N. and Hansen, L.B. 2003. Improved calibration of
2007. LED spectrometer (LSpec) autonomous vicarious Landsat-5 TM applicable for high-latitude and dark areas.
calibration facility. Proceedings of SPIE Conference 6677 on International Journal of Remote Sensing, 24(24): 53455365.
Earth Observing Systems XII, San Diego, CA, J.J. Butler and Kastner, C.J. and Slater, P.N. 1982. In-flight radiometric cali-
J. Xiong, eds., pp. 66770V. SPIE, Bellingham, WA. bration of advanced remote sensing systems. Proceedings
Henry, P., Chander, G., Fougnie, B., Thomas, C., and Xiong, X. of Field Measurement and Calibration Using Electro-
2013. Assessment of spectral band impact on inter-cali- Optical Equipment: Issues and Requirements, San Diego,
bration over desert sites using simulation based on EO-1 CA, F.M. Zweibaum and H. Register, eds., pp. 18. SPIE,
Hyperion data. IEEE Transactions on Geoscience and Bellingham,WA.
Remote Sensing, 51(3 SI): 12971308. Kaufman, Y.J. and Holben, B.N. 1993. Calibration of the AVHRR
Henry, P., Dinguirard, M., and Bidilis, M. 1993. SPOT multitem- visible and near-IR bands by atmospheric scattering, ocean
poral calibration over stable desert areas. Proceedings of glint and desert reflection. International Journal of Remote
SPIE International Symposium on Aerospace Remote Sensing, Sensing, 14(1): 2152.
Orlando, FL, pp. 6776. Kerola, D.X. and Bruegge, C.J. 2009. Desert test site uniformity
Hill, J. and Aifadopoulou, D. 1990. Comparative-analysis of analysis. Proceedings of SPIE Conference 7452 on Earth
Landsat-5 TM and SPOT HRV-1 data for use in multiple Observing Systems XIV, San Diego, CA, J.J. Butler, X. Xiong,
sensor approaches. Remote Sensing of Environment, 34(1): and X. Gu, eds., pp. 74520C-74528. SPIE, Bellingham, WA.
5570. Kerola, D.X., Bruegge, C.J., Gross, H.N., and Helmlinger, M.C.
Holben, B.N., Kaufman, Y.J., and Kendall, J.D. 1990. NOAA- 2009. On-orbit calibration of the EO-1 hyperion and
11 AVHRR visible and near-IR inflight calibration. advanced land imager (ALI) sensors using the LED spec-
International Journal of Remote Sensing, 11(8): 15111519. trometer (LSpec) automated facility. IEEE Transactions on
Hovis, W.A., Knoll, J.S., and Smith, G.R. 1985. Aircraft mea- Geoscience and Remote Sensing, 47(4): 12441255.
surements for calibration of an orbiting spacecraft sensor. Kieffer, H.H., Stone, T.C., Barnes, R.A., Bender, S., Eplee, R.E. Jr.,
Applied Optics, 24: 407410. Mendenhall, J., and Ong, L. 2002. On-orbit radiometric cal-
Hu, X., Liu, J., Sun, L., Rong, Z., Li, Y., Zhang, Y., Zheng, Z., Wu,R., ibration over time and between spacecraft using the moon.
Zhang, L., and Gu, X. 2010. Characterization of CRCS Proceedings of SPIE Conference on Sensors, Systems, and
Dunhuang test site and vicarious calibration utilization for Next-Generation Satellites VI, Crete, Greece, pp. 287298.
Fengyun (FY) series sensors. Canadian Journal of Remote Kieffer, H.H. and Wildey, R.L. 1985. Absolute calibration of
Sensing, 36(5): 566582. Landsat instruments using the moon. Photogrammetric
Hu, X., Zhang, Y., Liu, Z., Zhang, G., Huang, Y., Qiu, K., Wang,Y., Engineering and Remote Sensing, 51(9): 13911393.
Zhang, L., Zhu, X., and Rong, Z. 2001. Optical characteris- Kieffer, H.H. and Wildey, R.L. 1996. Establishing the moon as
tics of China radiometric calibration site for remote sensing a spectral radiance standard. Journal of Atmospheric and
satellite sensors (CRCSRSSS). Proceeding of SPIE Conference Oceanic Technology, 13(2): 360375.
4151 on Hyperspectral Remote Sensing of the Land and Kim, W. and Lee, S. 2013. Study on radiometric variability of the
Atmosphere, Sendai, Japan, W.L. Smith and Y. Yasuoka, eds., Sonoran Desert for vicarious calibration of satellite sensors.
pp. 7786. SPIE, Bellingham, WA. Korean Journal of Remote Sensing, 29(2): 209218.
Hu, X.Q., Liu, J.J., Qiu, K.M., Fan, T.X., Zhang, Y.X., Rong, Z.G., Koepke, P. 1982. Vicarious satellite calibration in the solar spec-
and Zhang, L.J. 2009. New method study of sites vicarious tral range by means of calculated radiances and its applica-
calibration for SZ-3/CMODIS. Spectroscopy and Spectral tion to Meteosat. Applied Optics, 21(15): 28452854.
Analysis, 29(5): 11531159. Lachrade, S., Fougnie, B., Henry, P., and Gamet, P. 2013. Cross
Hu, Y., Wielicki, B.A., Yang, P., Stackhouse, P.W. Jr., Lin, B., and calibration over desert sites: Description, methodology,
Young, D.F. 2004. Application of deep convective cloud and operational implementation. IEEE Transactions on
albedo observation to satellite-based study of the terres- Geoscience and Remote Sensing, 51(3 SI): 10981113.
trial atmosphere: Monitoring the stability of spaceborne Le Marshall, J.F., Simpson, J.J., and Jin, Z.H. 1999. Satellite cali-
measurements and assessing absorption anomaly. IEEE bration using a collocated nadir observation technique:
Transactions on Geoscience and Remote Sensing, 42(11): Theoretical basis and application to the GMS-5 pathfinder
25942599. benchmark period. IEEE Transactions on Geoscience and
Iwabuchi, H. 2003. Calibration of the visible and near-infrared Remote Sensing, 37(1): 499507.
channels of NOAA-11 and-14 AVHRRs by using reflections Li, Y., Zhang, Y., Liu, J., Rong, Z.G., and Zhang, L.J. 2009.
from molecular atmosphere and stratus cloud. International Calibration of the visible and near-infrared channels of the
Journal of Remote Sensing, 24(24): 53675378. FY-2C/FY-2D GEO meteorological satellite at radiometric
Jackson, R.D., ed. 1990. Special issue on coincident satellite, air- site. Guangxue Xuebao/Acta Optica Sinica, 29(1): 4146.
craft, and field measurements at the Maricopa Agricultural Liu, J.J., Li, Z., Qiao, Y.L., Liu, Y.J., and Zhang, Y.X. 2004. A
Center (MAC). Remote Sensing of Environment, 32(23): new method for cross-calibration of two satellite sensors.
77228. International Journal of Remote Sensing, 25(23): 52675281.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 125
Loeb, N.G. 1997. In-flight calibration of NOAA AVHRR vis- Markham, B.L., Seiferth, J.C., Smid, J., and Barker, J.L. 1998.
ible and near-IR bands over Greenland and Antarctica. Lifetime responsivity behavior of the Landsat-5 Thematic
International Journal of Remote Sensing, 18(3): 477490. Mapper. Proceedings of SPIE Conference 3427, San Diego,
Luderer, G., Coakley, J.A. Jr., and Tahnk, W.R. 2005. Using sun CA, pp. 420431.
glint to check the relative calibration of reflected spectral Markham, B.L., Storey, J.C., and Morfitt, R. 2015. Landsat-8 sen-
radiances. Journal of Atmospheric and Oceanic Technology, sor characterization and calibration. Remote Sensing, 7(3):
22(10): 14801493. 22792282.
Lukashin, C., Wielicki, B., Young, D., Thome, K., Jin, Z., and Sun, Martiny, N., Santer, R., and Smolskaia, I. 2005. Vicarious calibra-
W. 2013. Uncertainty estimates for imager reference inter- tion of MERIS over dark waters in the near infrared. Remote
calibration with CLARREO reflected solar spectrometer. Sensing of Environment, 94(4): 475490.
IEEE Transactions on Geoscience and Remote Sensing, 51(3 SI): Matsunaga, T., Nonaka, T., Sawabe, Y., Moriyama, M., Tonooka,
14251436. H., and Fukasawa, H. 2001. Vicarious and cross calibration
Malila, W.A. and Anderson, D.M. 1986. Satellite data availability methods for satellite thermal infrared sensors using hot
and calibration documentation for land surface climatology ground targets. Proceedings of IEEE Conference on Scanning
studies. NASA/GSFC, Greenbelt, MD, Report No. 180300- the Present and Resolving the Future, Sydney, New South
1-F, 214 pp. Wales, Australia, pp. 18411843. IEEE, Piscataway, NJ.
Markham, B., Storey, J., Williams, D., and Irons, J. 2004a. Landsat McCorkel, J., Thome, K., Biggar, S., and Kuester, M. 2006.
sensor performance: History and current status. IEEE Radiometric calibration of advanced land imager using
Transactions on Geoscience and Remote Sensing, 42(12): reflectance-based results between 2001 and 2005.
26912694. Proceedings of SPIE Conference 6296 on Earth Observing
Markham, B., Thome, K., Barsi, J., Kaita, E., Helder, D., Barker, Systems XI, San Diego, CA, J.J. Butler and J. Xiong, eds.,
J., and Scaramuzza, P. 2004b. Landsat-7 ETM+ on-orbit pp. 62960G. SPIE, Bellingham, WA.
reflective-band radiometric stability and absolute calibra- McCorkel, J., Thome, K., and Ong, L. 2013. Vicarious calibra-
tion. IEEE Transactions on Geoscience and Remote Sensing, tion of EO-1 hyperion. IEEE Journal of Selected Topics in
42(12): 28102820. Applied Earth Observations and Remote Sensing, 6(2 SI):
Markham, B.L. and Barker, J.L., eds. 1985. Special issue on Landsat 400407.
image data quality assessment (LIDQA). Photogrammetric Mendenhall, J.A., Hearn, D.R., Lencioni, D.E., Digenis, C.J., and
Engineering and Remote Sensing, 51(9): 12451493. Ong, L. 2003. Summary of the EO-1 ALI performance for
Markham, B.L., Barker, J.L., Kaita, E., Seiferth, J., and Morfitt, R. the first 2.5years on-orbit. Proceedings of SPIE Conference
2003. On-orbit performance of the Landsat-7 ETM+ radio- 5151 on Earth Observing Systems VIII, San Diego, CA, W.L.
metric calibrators. International Journal of Remote Sensing, Barnes, ed., pp. 574585.
24(2): 265286. Mendenhall, J.A., Lencioni, D.E., and Evans, J.B. 2005. Spectral
Markham, B.L., Barsi, J.A., Helder, D.L., Thome, K.J., and Barker, and radiometric calibration of the advanced land imager.
J.L. 2006. Evaluation of the Landsat-5 TM radiometric cali- Lincoln Laboratory Journal, 15(2): 207224.
bration history using desert test sites. Proceedings of SPIE Metzler, M.D. and Malila, W.A. 1985. Characterization and com-
Conference 6361 on Sensors, Systems, and Next-Generation parison of Landsat-4 and Landsat-5 Thematic Mapper data.
Satellites X, Stockholm, Sweden, R. Meynart, S.P. Neeck, Photogrammetric Engineering and Remote Sensing, 51(9):
and H. Shimoda, eds., pp. 63610. SPIE, Bellingham, WA. 13151330.
Markham, B.L., Boncyk, W.C., Helder, D.L., and Barker, J.L. 1997. Meygret, A., Briottet, X., Henry, P.J., and Hagolle, O. 2000.
Landsat-7 enhanced Thematic Mapper Plus radiometric cali- Calibration of SPOT4 HRVIR and vegetation cameras over
bration. Canadian Journal of Remote Sensing, 23(4): 318332. Rayleigh scattering. Proceedings of SPIE Conference 4135 on
Markham, B.L., Dabney, P.W., Knight, E.J., Kvaran, G., Barsi, J.A., Earth Observing Systems V, San Diego, CA, W.L. Barnes,
Murphy-Morris, J.E., and Pedelty, J.A. 2010. The Landsat ed., pp. 302313. SPIE, Bellingham, WA.
data continuity mission operational land imager (OLI) Meygret, A., Santer, R., and Berthelot, B. 2011. ROSAS: A robotic
radiometric calibration. Proceedings of IEEE International station for atmosphere and surface characterization dedi-
Geoscience and Remote Sensing Symposium (IGARSS) 2010, cated to on-orbit calibration. Proceedings of SPIE Conference
July 2530, Honolulu, HI, 4 pp. 8153 on Earth Observing Systems XVI, 815311, San Diego,
Markham, B.L., Haque, O., Barsi, J.A., Micijevic, E., Helder, CA, 12 pp.
D.L., Thome, K., Aaron, D., and Czapla-Myers, J.S. 2012. Miesch, C., Cabot, F., Briottet, X., and Henry, P. 2003. Assimilation
Landsat-7 ETM+: 12years on-orbit reflective-band radio- method to derive spectral ground reflectance of desert sites
metric performance. IEEE Transactions on Geoscience and from satellite datasets. Remote Sensing of Environment,
Remote Sensing, 50(5): 20562062. 87(23): 359370.
Markham, B.L. and Helder, D.L. 2012. Forty-year calibrated Mika, A.M. 1997. Three decades of Landsat instruments.
record of earth-reflected radiance from Landsat: A review. Photogrammetric Engineering and Remote Sensing, 63(7):
Remote Sensing of Environment, 122: 3040. 839852.
2016 by Taylor & Francis Group, LLC
126 Remotely Sensed Data Characterization, Classification, and Accuracies
Min, X. and Zhu, Y. 1995. Properties of atmospheric aerosol OBrien, D.M. and Mitchell, R.M. 2001. An error budget for cross-
extinction for the satellite radiometric calibration site calibration of AVHRR shortwave channels against ATSR-2.
of Dunhuang, China. Proceedings of IEEE Conference on Remote Sensing of Environment, 75(2): 216229.
Quantitative Remote Sensing for Science and Applications, Ohring, G., Tansock, J., Emery, W., Butler, J., L, Weng, F., Germain,
Firenze, Italy, pp. 18581860. IEEE, Piscataway, NJ. K.S., Wielicki, B., Cao, C., Goldberg, M., Xiong, J., Fraser,
Mitchell, R.M., OBrien, D.M., and Forgan, B.W. 1992. Calibration G., Kunkee, D., Winker, D., Miller, L., Ungar, S., Tobin,
of the NOAA AVHRR shortwave channels using split pass D., Anderson, J.G., Pollock, D., Shipley, S., Thurgood, A.,
imagery: I. Pilot study. Remote Sensing of Environment, Kopp, G., Ardanuy, P., and Stone, T. 2007. Achieving sat-
40(1): 5765. ellite instrument calibration for climate change. EOS,
Mitchell, R.M., OBrien, D.M., Edwards, M., Elsum, C.C., and Transactions American Geophysical Union, 88(11): 136.
Graetz, R.D. 1997. Selection and initial characterization of a Padula, F.P., Schott, J.R., Barsi, J.A., Raqueno, N.G., and Hook,
bright calibration site in the Strzelecki Desert, South Australia. S.J. 2010. Calibration of Landsat 5 thermal infrared chan-
Canadian Journal of Remote Sensing, 23(4): 342353. nel: Updated calibration history and assessment of the
Molineux, C.E., Bliamptis, E.E., and Neal, J.T. 1971. A remote- errors associated with the methodology. Canadian Journal
sensing investigation of four Mojave playas. Environmental of Remote Sensing, 36(5): 617630.
Research Papers, No. 352, AFCRL-71-0235, April 16, 1971. Pollock, D.B., Murdock, T.L., Datla, R.U., and Thompson, A.
Air Force Cambridge Research Laboratories, Hanscom 2003. Data uncertainty traced to SI units. Results reported
Field, Bedford, MA, 70 pp. in the International System of Units. International Journal of
Morain, S.A. and Budge, A.M., eds. 2004. Postlaunch Calibration Remote Sensing, 24(2): 225235.
of Satellite Sensors, ISPRS Book Series Volume 2: Proceedings Potts, D.R., Mackin, S., Muller, J.-P., and Fox, N. 2013. Sensor
of the International Workshop on Radiometric and Geometric intercalibration over dome C for the ESA GlobAlbedo
Calibration, Gulfport, MS. A.A. Balkema Publishers, Project. IEEE Transactions on Geoscience and Remote
Leiden, the Netherlands, 193 pp. Sensing, 51(3 SI): 11391146.
Moran, M.S., Jackson, R.D., Clarke, T.R., Qi, J., Cabot, F., Thome, Prata, A.J., Cechet, R.P., Grant, I.F., and Rutter, G.F. 1996. The
K.J., and Markham, B.L. 1995. Reflectance factor retrieval Australian Continental Integrated Ground-truth Site
from Landsat TM and SPOT HRV data for bright and dark Network (CIGSN)Satellite data calibration and valida-
targets. Remote Sensing of Environment, 52(3): 218230. tion, and first results. Proceedings of the Second SEIKEN
Mueller, J.L. 1985. NIMBUS-7 CZCSConfirmation of its radio- Symposium on Global Environmental Monitoring from
metric sensitivity decay-rate through 1982. Applied Optics, Space, Tokyo, Japan, pp. 245261.
24(7): 10431047. Price, J.C. 1987a. Radiometric calibration of satellite sensors in
Muench, H.S. 1981. Calibration of geosynchronous satellite the visible and near-infraredHistory and outlook. Remote
video sensors. Report No. AFGL-TR-810050, Air Force Sensing of Environment, 22(1): 39.
Geophysical Laboratory, Hanscom, MA. Price, J.C., ed. 1987b. Special issue on radiometric calibration
Murphy, J.M. 1984. Radiometric correction of Landsat Thematic of satellite data. Remote Sensing of Environment, 22(1):
Mapper data. CCRS Digital Methods Division Technical 1158.
Memorandum DMD-TM#84-368, Canada Centre for Rao, C.R.N., Cao, C., and Zhang, N. 2003. Inter-calibration of the
Remote Sensing, Ottawa, Ontario, Canada. moderate-resolution imaging spectroradiometer and the
Murphy, J.M. 1986. Within-scene radiometric correction of along-track scanning radiometer-2. International Journal of
Landsat Thematic Mapper data in Canadian production Remote Sensing, 24(9): 19131924.
systems. Proceedings of SPIE Conference 660 on Earth Rao, C.R.N. and Chen, J. 1995. Intersatellite calibration linkages
Remote Sensing using Landsat Thematic Mapper and SPOT for the visible and near-infrared channels of the advanced
Sensor Systems, Innsbruck, Austria, pp. 2531. very high-resolution radiometer on the NOAA-7, NOAA-9,
Nieke, J., Aoki, T., Tanikawa, T., Motoyoshi, H., and Hori, M. 2004. and NOAA-11 spacecraft. International Journal of Remote
A satellite cross-calibration experiment. IEEE Geoscience Sensing, 16(11): 19311942.
and Remote Sensing Letters, 1(3): 215219. Rao, C.R.N. and Chen, J.H. 1996. Postlaunch calibration of
Nieke, J., Aoki, T., Tanikawa, T., Motoyoshi, H., Hori, M., and the visible and near-infrared channels of the advanced
Nakajima, Y. 2003. Cross-calibration of satellite sensors very high-resolution radiometer on the NOAA-14 space-
over snow fields. Proceedings of SPIE Conference 5151 on craft. International Journal of Remote Sensing, 17(14):
Earth Observing Systems VIII, San Diego, CA, W.L. Barnes, 27432747.
ed., Vol. 5151, pp. 406414. Rao, C.R.N. and Chen, J.H. 1999. Revised postlaunch calibration
Nithianandam, J., Guenther, B.W., and Allison, L.J. 1993. An of the visible and near-infrared channels of the advanced
anecdotal review of NASA Earth observing satellite remote very high-resolution radiometer (AVHRR) on the NOAA-
sensors and radiometric calibration methods. Metrologia, 14 spacecraft. International Journal of Remote Sensing,
30(4): 207212. 20(18): 34853491.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 127
Reuter, D., Irons, J., Lunsford, A., Montanaro, M., Pellerano, F., Scott, K.P., Thome, K.J., and Brownlee, M.R. 1996. Evaluation
Richardson, C., Smith, R., Tesfaye, Z., and Thome, K. 2011. of Railroad Valley Playa for use in vicarious calibration.
Operational land imager (OLI) and the thermal infra- Proceedings of SPIE Conference 2818 on Multispectral
red sensor (TIRS) on the Landsat data continuity mission Imaging for Terrestrial Applications, Denver, CO, B. Huberty,
(LDCM). Proceedings of SPIE Conference 8048 on Algorithms J.B. Lurie, J.A. Caylor, P. Coppin, and P.C. Robert, eds.,
and Technologies for Multispectral, Hyperspectral, and pp. 158166. SPIE, Bellingham, WA.
Ultraspectral Imagery XVII, S.S. Shen and P.E. Lewis, eds., Secker, J., Staenz, K., Gauthier, R.P., and Budkewitsch, P. 2001.
804819. Vicarious calibration of airborne hyperspectral sensors in
Richter, R. 1997. On the in-flight absolute calibration of high operational environments. Remote Sensing of Environment,
spatial resolution spaceborne sensors using small ground 76(1): 8192.
targets. International Journal of Remote Sensing, 18(13): Six, C. 2002. Automatic Station for the In-Flight Calibration of
28272833. the Satellite Sensors: Application to the SPOT/HRV on the
Rodger, A.P., Balick, L.K., and Clodius, W.B. 2005. The perfor- La Crau Test Site. Universite du Littoral-Cote dOpale,
mance of the multispectral thermal imager (MTI) sur- Dunkerque, France, 147 pp.
face temperature retrieval algorithm at three sites. IEEE Six, D., Fily, M., Alvain, S., Henry, P., and Benoist, J.P. 2004.
Transactions on Geoscience and Remote Sensing, 43(3): Surface characterisation of the Dome Concordia area
658665. (Antarctica) as a potential satellite calibration site, using
Rondeaux, G., Steven, M.D., Clark, J.A., and Mackay, G. 1998. La SPOT4/VEGETATION instrument. Remote Sensing of
Crau: A European test site for remote sensing validation. Environment, 89(1): 8394.
International Journal of Remote Sensing, 19(14): 27752788. Six, D., Fily, M., Blarel, L., and Goloub, P. 2005. First aerosol opti-
Salomonson, V.V. and Marlatt, W.E. 1968. Anisotropic solar reflec- cal thickness measurements at Dome C (East Antarctica),
tance over white sand, snow, and stratus clouds. Journal of summer season 20032004. Atmospheric Environment,
Applied Meteorology, 7: 475483. 39(28): 50415050.
Sandford, S.P., Young, D.F., Corliss, J.M., Wielicki, B.A., Gazarik, Slater, P.N. 1980. Remote Sensing, Optics and Optical Systems.
M.J., Mlynczak, M.G., Little, A.D., Jones, C.D., Speth, P.W., Addison-Wesley Publishing Company, Reading, MA.
Shick, D.E., Brown, K.E., Thome, K.J., and Hair, J.H. 2010. Slater, P.N. 1984. The importance and attainment of accurate abso-
CLARREO: Cornerstone of the climate observing sys- lute radiometric calibration. Proceedings of SPIE Conference
tem measuring decadal change through accurate emitted 475 on Remote Sensing, Arlington, VA, P.N. Slater, ed.,
infrared and reflected solar spectra and radio occultation. pp. 3440. SPIE, Bellingham, WA.
Proceedings of SPIE Conference 7826 on Sensors, Systems, Slater, P.N. 1985. Radiometric considerations in remote-sensing.
and Next-Generation Satellites XIV, R. Meynart, S.P. Neeck, Proceedings of the IEEE, 73(6): 9971011.
and H. Shimoda, eds., Toulouse, France, 782611. Slater, P.N. 1986. Variations in in-flight absolute radiometric per-
Santer, R., Gu, X.F., Guyot, G., Deuze, J.L., Devaux, C., Vermote, formance. International Satellite Land-Surface Climatology
E., and Verbrugghe, M. 1992. SPOT calibration at the Project (ISLSCP) Conference, Rome, Italy, pp. 357363.
La-Crau test site (France). Remote Sensing of Environment, European Space Agency, Paris, France.
41(23): 227237. Slater, P.N. and Biggar, S.F. 1996. Suggestions for radiometric cali-
Santer, R., Schmectig, C., and Thome, K.J. 1997. BRDF and bration coefficient generation. Journal of Atmospheric and
surface-surround effects on SPOT-HRV vicarious calibra- Oceanic Technology, 13(2): 376382.
tion. Proceedings of SPIE Conference 2957 on Advanced and Slater, P.N., Biggar, S.F., Holm, R.G., Jackson, R.D., Mao, Y.,
Next-Generation Satellites II, Taormina, Italy, H. Fujisada, Moran, M.S., Palmer, J.M., and Yuan, B. 1987. Reflectance-
G. Calamai, and M.N. Sweeting, eds., pp. 344354. SPIE, based and radiance-based methods for the in-flight abso-
Bellingham, WA. lute calibration of multispectral sensors. Remote Sensing of
Schiller, S.J. 2003. Technique for estimating uncertainties in top- Environment, 22(1): 1137.
of-atmosphere radiances derived by vicarious calibration. Slater, P.N., Biggar, S.F., Palmer, J.M., and Thome, K.J. 2001.
Proceedings of SPIE Conference 5151 on Earth Observing Unified approach to absolute radiometric calibration in
Systems VIII, San Diego, CA, W.L. Barnes, ed., pp. 502516. the solar-reflective range. Remote Sensing of Environment,
SPIE, Bellingham, WA. 77(3): 293303.
Schott, J.R., Salvaggio, C., and Volchok, W.J. 1988. Radiometric Slater, P.N., Biggar, S.F., Thome, K.J., Gellman, D.I., and Spyak, P.R.
scene normalization using pseudo-invariant features. 1996. Vicarious radiometric calibration of EOS sensors. Journal
Remote Sensing of Environment, 26: 116. of Atmospheric and Oceanographic Technology, 13(2): 349359.
Schroeder, M., Poutier, L., Muller, R., Dinguirard, M., Reinartz, P., Smith, D.L., Mutlow, C.T., and Rao, C.R.N. 2002. Calibration
and Briottet, X. 2001. Intercalibration of optical satellites monitoring of the visible and near-infrared channels of the
A case study with MOMS and SPOT. Aerospace Science and along-track scanning radiometer-2 by use of stable terres-
Technology, 5(4): 305315. trial sites. Applied Optics, 41(3): 515523.
2016 by Taylor & Francis Group, LLC
128 Remotely Sensed Data Characterization, Classification, and Accuracies
Smith, G.R. 1984. Surface soil moisture measurements of the White Teillet, P.M. 2014. Overview of satellite image radiometry in the solar-
Sands, New Mexico. NOAA Technical Report NESDIS reflective optical domain. In The Remote Sensing Handbook, P.S.
7, PB85-135754. National Oceanic and Atmospheric Thenkabail, ed., Vol. 1, Chapter 3, CRC Press, Boca Raton, FL.
Administration, Washington, DC, 12 pp. Teillet, P.M., Barker, J.L., Markham, B.L., Irish, R.R., Fedosejevs, G.,
Smith, G.R. and Levin, R.H. 1985. High altitude measured and Storey, J.C. 2001a. Radiometric cross-calibration of the
radiance of White Sands, New Mexico, in the 400- Landsat-7 ETM+ and Landsat-5 TM sensors based on tandem
2000nm band using a filter wedge spectrometer. NOAA data sets. Remote Sensing of Environment, 78(12): 3954.
Technical Report NESDIS 21, PB85-206084, National Teillet, P.M., Barsi, J.A., Chander, G., and Thome, K.J. 2007b.
Oceanic and Atmospheric Administration, Washington, Prime candidate Earth targets for the postlaunch radiomet-
DC, 17 pp. ric calibration of space-based optical imaging instruments.
Smith, G.R., Levin, R.H., Abel, P., and Jacobowitz., H. 1988. Proceedings of the SPIE Conference 6677 on Earth Observing
Calibration of the solar channels and NOAA-9 AVHRR Systems XII, September 26, San Diego, CA, J.J. Butler and
using high altitude aircraft measurements. Journal of J.Xiong, eds., pp. 66770S1-12. SPIE, Bellingham, WA.
Atmospheric and Oceanic Technology, 5: 631639. Teillet, P.M. and Chander, G. 2010. Terrestrial reference standard
Smith, G.R., Levin, R.H., and Knoll, J.S. 1985. An atlas of high sites for postlaunch sensor calibration. Canadian Journal of
altitude aircraft measured radiance of White Sands, New Remote Sensing, 36(5): 437450.
Mexico, in the 4501050 nm band. NOAA Technical Teillet, P.M., El Saleous, N., Hansen, M.C., Eidenshink, J.C.,
Report NESDIS 20, PB85-204501, National Oceanic and Justice, C.O., and Townshend, J.R.G. 2000. An evaluation
Atmospheric Administration, Washington, DC, 29 pp. of the global 1-km AVHRR land data set. International
Snyder, W.C., Wan, Z.M., Zhang, Y.L., and Feng, Y.Z. 1997. Journal of Remote Sensing, 21(10): 19872021.
Requirements for satellite land surface temperature vali- Teillet, P.M., Fedosejevs, G., Gauthier, D., DIorio, M.A., Rivard,B.,
dation using a silt playa. Remote Sensing of Environment, and Budkewitsch, P. 1995. Initial examination of radar
61(2): 279289. imagery of optical radiometric calibration sites. Proceedings
Staylor, W.F. 1986. Site selection and directional models of deserts of SPIE Conference 2583 on Advanced and Next-Generation
used for ERBE validation targets. NASA Technical Paper Satellites, December 15, Paris, France, H. Fujisada and M.N.
2540, NASA, Washington, DC. Sweeting, eds., pp. 154165. SPIE, Bellingham, WA.
Staylor, W.F. 1990. Degradation rates of the AVHRR visible Teillet, P.M., Fedosejevs, G., and Gauthier, R.P. 1998a. Operational
channel for the NOAA 6, 7, and 9 spacecraft. Journal of radiometric calibration of broadscale satellite sensors using
Atmospheric and Oceanic Technology, 7(3): 411423. hyperspectral airborne remote sensing of prairie rangeland:
Stone, T.C. 2008. Radiometric calibration stability and inter-cali- First trials. Metrologia, 35(4): 639641.
bration of solar-band instruments in orbit using the moon. Teillet, P.M., Fedosejevs, G., Gauthier, R.P., ONeill, N.T., Thome,
Proceedings of SPIE Conference on Earth Observing Systems K.J., Biggar, S.F., Ripley, H., and Meygret, A. 2001b. A gen-
XIII, San Diego, CA, 70810X. eralized approach to the vicarious calibration of multiple
Stone, T.C., Kieffer, H.H., and Grant, I.F. 2005. Potential for cali- Earth observation sensors using hyperspectral data. Remote
bration of geostationary meteorological satellite imagers Sensing of Environment, 77(3): 304327.
using the moon. Proceedings of SPIE Conference on Earth Teillet, P.M., Fedosejevs, G., Gauthier, R.P., and Schowengerdt,
Observing Systems X, San Diego, CA, pp. 19. R.A. 1998b. Uniformity characterization of land test sites
Stone, T.C., Rossow, W.B., Ferrier, J., and Hinkelmann, L.M. 2013. used for radiometric calibration of Earth observation sen-
Evaluation of ISCCP multi-satellite radiance calibration sors. Proceedings of the Twentieth Canadian Symposium on
for geostationary imager visible channels using the moon. Remote Sensing, May, Calgary, Alberta, pp. 14. Canadian
IEEE Transactions on Geoscience and Remote Sensing, 51(3): Remote Sensing Society, Ottawa, Canada.
12551266. Teillet, P.M., Fedosejevs, G., and Thome, K.J. 2004b. Spectral band
Tahnk, W.R. and Coakley, J.A. 2001. Updated calibration coeffi- difference effects on radiometric cross-calibration between
cients for NOAA-14 AVHRR channels 1 and 2. International multiple satellite sensors in the Landsat solar-reflective spec-
Journal of Remote Sensing, 22(15): 30533057. tral domain. Workshop on Inter-Comparison of Large-Scale
Teillet, P.M. 1986. Image correction for radiometric effects in Optical and Infrared Sensors. Proceedings of SPIE Conference
remote sensing. International Journal of Remote Sensing, 7: 5570 on Sensors, Systems, and Next-Generation Satellites
16371651. VIII, Maspalomas, Canary Islands, Spain, R. Meynart, S.P.
Teillet, P.M. 1997a. A status overview of Earth observation cali- Neeck, and H. Shimoda, eds., pp. 307316.
bration/validation for terrestrial applications. Canadian Teillet, P.M., Fedosejevs, G., Thome, K.J., and Barker, J.L. 2007a.
Journal of Remote Sensing: Special issue on Calibration/ Impacts of spectral band difference effects on radiometric
Validation, 23(4): 291298. cross-calibration between satellite sensors in the solar-
Teillet, P.M., ed. 1997b. Special issue on calibration/validation. reflective spectral domain. Remote Sensing of Environment,
Canadian Journal of Remote Sensing, 23(4): 289423. 110(3): 393409.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 129
Teillet, P.M. and Fox, N.P. 2009. Concatenation of terrestrial Thome, K., Crowther, B.G., and Biggar, S.F. 1997a. Reflectance-
reference standard sites for systematic postlaunch calibra- and irradiance-based calibration of Landsat-5 Thematic
tion monitoring of multiple space-based imaging sensors. Mapper. Canadian Journal of Remote Sensing, 23(4):
Proceedings of SPIE Conference 7474 on Sensors, Systems, 309317.
and Next-Generation Satellites XIII, August 31September 3, Thome, K., Czapla-Myers J., and Biggar, J. 2004a. Ground-monitor
Berlin, Germany, R. Meynart, S.P. Neeck, and H. Shimoda, radiometer system for vicarious calibration. Proceedings of
eds., pp. 747410. SPIE, Bellingham, WA. SPIE Conference 5546 on Imaging Spectrometry X, October
Teillet, P.M., Helder, D.L., Ruggles, T.A., Landry, R., Ahern, F.J., 15, Denver, CO, S.S. Shen and P.E. Lewis, eds., pp. 223232.
Higgs, N.J., Barsi, J., Chander, G., Markham, B.L., Barker, SPIE, Bellingham, WA.
J.L., Thome, K.J., Schott, J.R., and Palluconi, F.D. 2004a. A Thome, K., Czapla-Myers, J., and McCorkel, J. 2007. Retrieval of
definitive calibration record for the Landsat-5 Thematic surface BRDF for reflectance-based calibration. Proceedings
Mapper anchored to the Landsat-7 radiometric scale. of SPIE Conference 6677 on Earth Observing Systems XII,
Canadian Journal of Remote Sensing, 30(4): 631643. September 26, San Diego, CA, J.J. Butler and J. Xiong, eds.,
Teillet, P.M. and Holben, B.N. 1994. Towards operational radio- pp. 6677066711. SPIE, Bellingham, WA.
metric calibration of NOAA AVHRR imagery in the visible Thome, K., Czapla-Myers, J., Leisso, N., McCorkel, J., and
and near-infrared channels. Canadian Journal of Remote Buchanan, J. 2008. Intercomparison of imaging sensors
Sensing, 20(1): 110. using automated ground measurements. Proceedings of
Teillet, P.M., Markham, B.L., and Irish, R.R. 2006. Landsat cross- IEEE Conference on Remote Sensing: The Next Generation,
calibration based on near simultaneous imaging of common Boston, MA, pp. 13321335. IEEE, Piscataway, NJ.
ground targets. Remote Sensing of Environment, 102(34): Thome, K., DAmico, J., and Hugon, C. 2006. Intercomparison of
264270. Terra ASTER, MISR, and MODIS, and Landsat-7 ETM+.
Teillet, P.M., and Ren, X. 2008. Spectral band difference effects Proceedings of IEEE Conference on Remote Sensing: A
on vegetation indices derived from multiple satellite Natural Global Partnership, Denver, CO, pp. 17721775.
sensor data. Canadian Journal of Remote Sensing, 34(3): IEEE, Piscataway, NJ.
159173. Thome, K. and Fox, N. 2011. 2010 CEOS field reflectance
Teillet, P.M., Ren, X., and Smith, A.M. 2010. Suitability of inter-comparison lessons learned. Proceedings of IEEE
rangeland terrain for satellite remote sensing calibration. International Geoscience and Remote Sensing Symposium
Canadian Journal of Remote Sensing, 36(5): 451463. (IGARSS) 2011, July 2429, Vancouver, British Columbia,
Teillet, P.M., Slater, P.N., Ding, Y., Santer, R.P., Jackson, R.D., and Canada, pp. 38793882.
Moran, M.S. 1990. Three methods for the absolute calibra- Thome, K., Gustafson-Bold, C., Slater, P.N., and Farrand, W.H.
tion of the NOAA AVHRR sensors in-flight. Remote Sensing 1996. In-flight radiometric calibration of HYDICE using a
of Environment, 31(2): 105120. reflectance-based approach. Proceedings of SPIE Conference
Teillet, P.M., Thome, K.J., Fox, N.P., and Morisette, J.T. 2001c. Earth 2821 on Hyperspectral Remote Sensing and Applications,
observation sensor calibration using a global instrumented August 4, Denver, CO, S.S. Shen, ed., pp. 311319. SPIE,
and automated network of test sites (GIANTS). Proceedings Bellingham, WA.
of SPIE Conference 4550 on Sensors, Systems, and Next- Thome, K., Markham, B., Barker, J., Slater, P., and Biggar, S. 1997b.
Generation Satellites V, September 21, Toulouse, France, H. Radiometric calibration of Landsat. Photogrammetric
Fujisada, J.B. Lurie, and K. Weber, eds., pp. 246254. SPIE, Engineering and Remote Sensing, 63(7): 853858.
Bellingham, WA. Thome, K., McCorkel, J., and Czapla-Myers, J. 2013. In-situ trans-
Thome, K. 2005. Sampling and uncertainty issues in trending fer standard and coincident-view intercomparisons for sen-
reflectance-based vicarious calibration results. Proceedings sor cross-calibration. IEEE Transactions on Geoscience and
of SPIE Conference 5882 on Earth Observing Systems X, Remote Sensing, 51(3 SI): 10881097.
August 22, San Diego, CA, J.J. Butler, ed., pp. 111. SPIE, Thome, K., Schiller, S., Conel, J., Arai, K., and Tsuchida, S. 1998.
Bellingham, WA. Results of the 1996 Earth Observing System vicarious
Thome, K. 2012. Characterization approaches to place invariant calibration joint campaign at Lunar Lake Playa, Nevada.
sites on traceable scales. Proceedings of IEEE International Metrologia, 35(4): 631638.
Geoscience and Remote Sensing Symposium (IGARSS) 2012, Thome, K., Smith, N., and Scott, K. 2001. Vicarious calibra-
July 2227, Munich, Germany, pp. 70197022. tion of MODIS using Railroad Valley Playa. Proceedings
Thome, K., Barnes, R., Baize, R., OConnell, J., and Hair, J. 2010. of IEEE Conference on Scanning the Present and Resolving
Calibration of the reflected solar instrument for the climate the Future, July 913, Sydney, New South Wales, Australia,
absolute radiance and refractivity observatory. Proceedings pp.12091211. IEEE, Piscataway, NJ.
of the 2010 IEEE International Geoscience and Remote Thome, K.J. 2001. Absolute radiometric calibration of Landsat 7
Sensing Symposium (IGARSS), July 2530, Honolulu, ETM+ using the reflectance-based method. Remote Sensing
Hawaii, pp. 22752278. of Environment, 78(12): 2738.
2016 by Taylor & Francis Group, LLC
130 Remotely Sensed Data Characterization, Classification, and Accuracies
Thome, K.J., Biggar, S.F., and Wisniewski, W. 2003a. Cross com- Vermote, E.F., Santer, R., Deschamps, P.Y., and Herman, M.
parison of EO-1 sensors and other earth resources sen- 1992. In-flight calibration of large field of view sensors at
sors to Landsat-7 ETM+ using Railroad Valley Playa. IEEE short wavelengths using Rayleigh scattering. International
Transactions on Geoscience and Remote Sensing, 41(6): Journal of Remote Sensing, 13(18): 34093429.
11801188. Villa-Aleman, E., Kurzeja, R.J., and Pendergast, M.M. 2003a.
Thome, K.J., Czapla-Myers, J., and Biggar, S. 2003b. Vicarious Assessment of Ivanpah Playa as a site for thermal vicari-
calibration of Aqua and Terra MODIS. Proceedings of SPIE ous calibration for the MTI satellite. Proceedings of SPIE
Conference 5151 on Earth Observing Systems VIII, November Conference 5093 on Algorithms and Technologies for
13, San Diego, CA, W.L. Barnes, ed., pp. 395405. SPIE, Multispectral, Hyperspectral, and Ultraspectral Imagery IX,
Bellingham, WA. Orlando, FL, S.S. Shen and P.E. Lewis, eds., pp. 331342.
Thome, K.J., Helder, D.L., Aaron, D., and Dewald, J.D. 2004b. SPIE, Bellingham, WA.
Landsat-5 TM and Landsat-7 ETM+ absolute radiomet- Villa-Aleman, E., Kurzeja, R.J., and Pendergast, M.M. 2003b.
ric calibration using the reflectance-based method. IEEE Temporal, spatial, and spectral variability at the Ivanpah
Transactions on Geoscience and Remote Sensing, 42(12): Playa vicarious calibration site. Proceedings of SPIE Conference
27772785. 5093 on Algorithms and Technologies for Multispectral,
Thome, K.J., Whittington, E.E., Smith, N., Nandy, P., and Hyperspectral, and Ultraspectral Imagery IX, Orlando, FL, S.S.
Zalewski, E.F. 2000. Ground-reference techniques for the Shen and P.E. Lewis, eds., pp. 320330. SPIE, Bellingham, WA.
absolute radiometric calibration of MODIS. Proceedings Wang, Y., Czapla-Myers, J., Lyapustin, A., Thome, K., and Dutton,
of SPIE Conference 4135 on Earth Observing Systems V, E. 2011. AERONET-based surface reflectance validation
November 15, San Diego, CA, W.L. Barnes, ed., pp. 5159. network (ASRVN) data evaluation: Case study for Railroad
SPIE, Bellingham, WA. Valley calibration site. Remote Sensing of Environment, 115
Tomasi, C., Petkov, B., Benedetti, E., Valenziano, L., Lupi, A., (10): 27102717.
Vitale, V., and Bonafe, U. 2008. A refined calibration pro- Warren, S.G., Brandt, R.E., and Hinton, P.O. 1998. Effect of sur-
cedure of two-channel sun photometers to measure atmo- face roughness on bidirectional reflectance of Antarctic
spheric precipitable water at various Antarctic sites. Journal snow. Journal of Geophysical Research-Planets, 103(E11):
of Atmospheric and Oceanic Technology, 25(2): 213229. 2578925807.
Townshend, J.R.G. 1994. Global data sets for land applications Wenny, B.N. and Xiong, X. 2008. Using a cold Earth surface tar-
from the advanced very high resolution radiometer: An get to characterize long-term stability of the MODIS ther-
introduction. International Journal of Remote Sensing, mal emissive bands. IEEE Geoscience and Remote Sensing
15(17): 33193332. Letters, 5(2): 162165.
Townshend, J.R.G., Justice, C.O., Skole, D., Malingreau, J.-P., Wenny, B.N., Xiong, X., and Dodd, J. 2009. MODIS thermal emis-
Cihlar, J., Teillet, P., Sadowski, F., and Ruttenberg, S. 1994. The sive band calibration stability derived from surface targets.
1-km resolution global data set: Needs of the International Proceedings of SPIE Conference 7474 on Sensors, Systems,
Geosphere Biosphere Programme. International Journal of and Next-Generation Satellites XIII, Berlin, Germany, R.
Remote Sensing, 15(17): 34173441. Meynart, S.P. Neeck, and H. Shimoda, eds., pp. 74740W.
Trishchenko, A.P., Cihlar, J., and Li, Z. 2002. Effects of spectral SPIE, Bellingham, WA.
response function on surface reflectance and NDVI mea- Wheeler, R.J., Lecroy, S.R., Whitlock, C.H., Purgold, G.C., and
sured with moderate resolution satellite sensors. Remote Swanson, J.S. 1994. Surface characteristics for the Alkali Flats
Sensing of Environment, 81(1): 118. and Dunes regions at White-Sands-Missile-Range, New-
Valorge, C., Meygret, A., Lebgue, L., Henry, P., Bouillon, A., Mexico. Remote Sensing of Environment, 48(2): 181190.
Gachet, R., Breton, E., Lger, D., and Viallefont, F. 2004. Whitlock, C.H., Staylor, W.F., Darnell, W.L., Chou, M.-D.,
40 Years of experience with SPOT in-flight calibration. Dedieu, G., Deschamps, P.Y., Ellis, J., Gautier, C., Frouin, R.,
Proceedings of the International Workshop on Radiometric Pinker, R.T., Laslo, I., Rossow, W.B., and Tarpley, D. 1990a.
and Geometric Calibration: ISPRS Book Series Volume 2, Comparison of surface radiation budget satellite algorithms
Postlaunch Calibration of Satellite Sensors, Gulfport, MS, for downwelled shortwave irradiance with Wisconsin FIRE/
S.A. Morain and A.M. Budge, eds., pp. 119133. A.A. SRB surface truth data. Proceedings of the 7th AMS Conference
Balkema Publishers, Leiden, the Netherlands. on Atmospheric Radiation, San Francisco, CA, pp. 237242.
Vermote, E. and Kaufman, Y.J. 1995. Absolute calibration of Whitlock, C.H., Staylor, W.F., Suttles, J.T., Smith, G., Levin, R.,
AVHRR visible and near-infrared channels using ocean Frouin, R., Gautier, C., Teillet, P.M., Slater, P.N., Kaufman,
and cloud views. International Journal of Remote Sensing, Y.J., Holben, B.N., Rossow, W.B., and LeCroy, S.R. 1990b.
16(13): 23172340. AVHRR and VISSR satellite instrument calibration results
Vermote, E.F. and Saleous, N.Z. 2006. Calibration of NOAA16 for both cirrus and marine stratocumulus IFO periods.
AVHRR over a desert site using MODIS data. Remote Proceedings of FIRE Science Meeting, Vail, Colorado. NASA
Sensing of Environment, 105(3): 214220. Langley Research Center, Hampton, Virginia, pp. 141146.
2016 by Taylor & Francis Group, LLC
Postlaunch Radiometric Calibration of Satellite-Based Optical Sensors 131
Wu, A., Cao, C., and Xiong, X. 2003. Intercomparison of the Xiong, X., Choi, T., Che, N., Wang, Z., Dodd, J., Xie, Y., and
11-and 12- m bands of Terra and Aqua MODIS using Barnes, W. 2010a. Results and lessons from a decade
NOAA-17 AVHRR. Proceedings of SPIE Conference 5151 on of Terra MODIS on-orbit spectral characterization.
Earth Observing Systems VIII, San Diego, CA, W.L. Barnes, Proceedings of SPIE Conference 7862, Earth Observing
ed., pp. 384394. SPIE, Bellingham, WA. Missions and Sensors: Development, Implementation, and
Wu, A., Xiong, X., and Cao, C. 2008b. Examination of calibration Characterization, Incheon, Republic of Korea, X. Xiong, C.
performance of multiple POS sensors using measurements Kim, and H. Shimoda, eds., 78620M, doi:10.1117/12.868930.
over the Dome C site in Antarctica. Proceedings of SPIE Xiong, X., Che, N., Xie, Y., Moyer, D., Barnes, W., Guenther, B.,
Conference 7106 on Sensors, Systems, and Next-Generation and Salomonson, V. 2006a. Four-years of on-orbit spec-
Satellites XII, October 9, Cardiff, Wales, U.K., R. Meynart, tral characterization results for Aqua MODIS reflective
S.P. Neeck, H. Shimoda, and S. Habib, eds., pp. 71060W. solar bands. Proceedings of SPIE Conference 6361, Sensors,
SPIE, Bellingham, WA. Systems, and Next-Generation Satellites X, Stockholm,
Wu, A., Xiong, X., and Cao, C. 2008c. Terra and Aqua MODIS Sweden, R. Meynart, S.P. Neeck, and H. Shimoda, eds.,
intercomparison of three reflective solar bands using 63610S, doi:10.1117/12.687163.
AVHRR onboard the NOAA-KLM satellites. International Xiong, X., Sun, J., and Barnes, W. 2008. Intercomparison of on-
Journal of Remote Sensing, 29(7): 19972010. orbit calibration consistency between Terra and Aqua
Wu, A., Xiong, X., Cao, C., and Angal, A. 2008a. Monitoring MODIS reflective solar bands using the moon. IEEE
MODIS calibration stability of visible and near-IR bands Geoscience and Remote Sensing Letters, 5(4): 778782.
from observed top-of-atmosphere BRDF-normalized Xiong, X., Wu, A., Angal, A., and Wenny, B. 2009a. Recent prog-
reflectances over Libyan Desert and Antarctic surfaces. ress on cross-comparison of Terra and Aqua MODIS cali-
Proceedings of SPIE Conference 7081 on Earth Observing bration using Dome C. Proceedings of SPIE Conference 7474
Systems XIII, San Diego, CA, J.J. Butler and J. Xiong, eds., on Sensors, Systems, and Next-Generation Satellites XIII,
pp. 708113708119. SPIE, Bellingham, WA. Berlin, Germany, R. Meynart, S.P. Neeck, and H. Shimoda,
Wu, D., Yin, Y., Wang, Z., Gu, X., Verbrugghe, M., and Guyot, eds., pp. 747411. SPIE, Bellingham, WA.
G. 1997. Radiometric characterisation of Dunhuang satel- Xiong, X., Wu, A., Sun, J., and Wenny, B. 2006b. An overview
lite calibration test site (China). Proceedings of the Seventh of intercomparison methodologies for Terra and Aqua
International Symposium on Physical Measurements and MODIS calibration. Proceedings of SPIE Conference 6296 on
Signatures in Remote Sensing, Courchevel, France, G. Guyot Earth Observing Systems XI, San Diego, CA, J.J. Butler and
and T. Phulpin, eds., pp. 151160. Taylor & Francis Group, J. Xiong, eds., pp. 62960C. SPIE, Bellingham, WA.
Boca Raton, FL. Xiong, X., Wu, A., Wenny, B., Choi, J., and Angal, A. 2010b.
Wu, D., Zhu, Y., Wang, Z., Ge, B., and Yin, Y. 1994. The build- Progress and lessons from MODIS calibration intercom-
ing of radiometric calibration test site for satellite sensors parison using ground test sites. Canadian Journal of Remote
in China. Proceedings of the Sixth International Symposium Sensing, 36(5): 540552.
on Physical Measurements and Signatures in Remote Sensing, Xiong, X.X., Wu, A.S., and Wenny, B.N. 2009b. Using Dome C for
Val DIsere, France, pp. 167171. moderate resolution imaging spectroradiometer calibration
Wu, X., Sullivan, J.T., and Heidinger, A.K. 2010. Operational cali- stability and consistency. Journal of Applied Remote Sensing,
bration of the advanced very high resolution radiometer 3(1): 033520.
(AVHRR) visible and near-infrared channels. Canadian Zhang, Y., Li, Y., Rong, Z.G., Hu, X.Q., Zhang, L.J., and Liu, J.J.
Journal of Remote Sensing, 36(5): 602616. 2009. Field measurement of Gobi surface emissivity spec-
Wulder, M.A. and Masek, J.G., eds. 2012. Landsat legacy special trum at Dunhuang calibration site of China. Spectroscopy
issue. Remote Sensing of Environment, 122, 202 pp. and Spectral Analysis, 29(5): 12131217.
Xiao, Q., Liu, J., Yu, H., and Zhang, H. 2001. Analysis and evalu- Zhang, Y., Qiu, K., Hu, X., Rong, Z., and Zhang, L. 2004. Vicarious
ation of optical uniformity for Dunhuang calibration site radiometric calibration of satellite FY-1D sensors at vis-
by airborne spectrum survey data. Proceedings of China ible and near infrared channels. Acta Meteorologica Sinica,
Remote Sensing Sensors Radiometric Calibration, Ocean 18(4): 505516.
Press, Beijing, China, pp. 136142. Zhang, Y., Rong, Z., Hu, X., Liu, J., Zhang, L., Li, Y., and Zhang, X. 2008.
Xing-Fa, G., Guyot, G., and Verbrugghe, M. 1990. Evaluation Field measurement of Gobi surface emissivity using CE312
of measurement errors on the reflectance of La and infragold board at Dunhuang calibration site of China.
Crau, the French SPOT calibration area. Proceedings Proceedings of IEEE Conference on Remote Sensing: The Next
of the 10th EARSeL Symposium on New European Generation, Boston, MA, pp. 358360. IEEE, Piscataway, NJ.
Systems, Sensors and Applications, Toulouse, France, Zhang, Y., Zhang, G., Liu, Z., Zhang, L., Zhu, S., Rong, Z., and Qiu,
G.Konecny, ed., pp. 121133. European Association of K. 2001. Spectral reflectance measurements at the China
Remote Sensing Laboratories, Boulogne-Billancourt, radiometric calibration test site for the remote sensing sat-
France. ellite sensor. Acta Meteorologica Sinica, 15(3): 377382.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
5
Remote Sensing Data
Normalization
Acronyms and Definitions .................................................................................................................133
5.1 Introduction .............................................................................................................................133
5.2 Remote Sensing Data.............................................................................................................. 134
5.3 Sources of Radiometric Distortion .......................................................................................135
Sensor Calibration and Degradation Atmospheric Influence on Radiometry Topographic
Influence on Radiometry
Rudiger Gens 5.4 Radiometric Normalization .................................................................................................. 136
University of Alaska, Fairbanks
Absolute Radiometric Correction Relative Radiometric Correction
Jordi Cristbal Rossell 5.5 Conclusions...............................................................................................................................139
University of Alaska, Fairbanks References.............................................................................................................................................140
AcronymsandDefinitions In this chapter, the terms remote sensing data and remote
sensing images are used interchangeably. In order to be able
BRDF Bidirectional reflectance distribution function to use data in any combined fashion, each individual data-
CEOS Committee on Earth Observation Satellites set needs to have the same reference that allows quantitative
DN Digital number comparisons. The most generic term for this is data normal-
JAXA Japan Aerospace Exploration Agency ization and for remote sensing data, it is known as radiometric
LSE Land surface emissivity normalization. Depending on the data source, the processing
LST Land surface temperature and necessary corrections differ. Although the terms and defi-
NDVI Normalized Difference Vegetation Index nitions used for remote sensing data from different sources
SAR Synthetic aperture radar might vary, there are two main categories of radiometric nor-
TIR Thermal infrared malization (Bao etal., 2012): absolute normalization (methods
TOA Top of atmosphere based on radiative transfer methods that account for atmo-
spheric, illumination, and sensor differences) and relative nor-
5.1 introduction malization (techniques that minimize the effects of changing
atmospheric and solar conditions in one or a series of images,
The increasing access to remote sensing data from different relative to a standard image). For optical and thermal imag-
platforms, acquired at different spatial, spectral, and temporal ery, radiometric corrections need to be applied to account for
resolutions, are continuously widening the scope of applica- atmospheric conditions, solar angle, or sensor view angle
tions of these datasets. Parallel advancements in computational (Chen etal., 2005; Du etal., 2002), in addition to the sensor
science and technology have led to the development of more prelaunch and postlaunch calibration. These corrections help
sophisticated data processing and analysis tools. While early to convert the raw signal recorded at the sensor to physically
remote sensing studies were centered on detecting a feature or meaningful and measurable values, such as ground reflec-
phenomena, the current practice is to carry out multitemporal tance or ground temperatures. The quantitative use of syn-
studies and time series analysis based on multiple data sources, thetic aperture radar (SAR) data requires calibrated imagery
including optical, microwave, and thermal imagery. There are (Freeman, 1992). Specifically, the SAR processor used for the
a number of calibration and normalization issues that need to image generation needs to be calibrated and the calibration
be resolved before these more complex monitoring and change parameters then need to be applied to the data to generate cali-
detection studies can be accomplished. brated images.
133
2016 by Taylor & Francis Group, LLC
134 Remotely Sensed Data Characterization, Classification, and Accuracies
Remote sensing data
The visible and infrared regions generally range from 0.4 to
14 m in wavelength, with visible region occupying the shorter
wavelength end from 400 to 700nm. In remote sensing litera-
ture, there is considerable discrepancy and no general consensus
Image type Optical data Thermal data SAR data on how to classify and name the infrared portion of the spec-
trum and where to set the boundaries (Gupta, 2003; Quattrochi
etal., 2009). The disagreement stems from the fact that the dom-
Sources of Sensor Atmospheric Topographic
radiometric calibration and influence on influence on inant process operating in the shorter wavelength end of the
distortion degradation radiometry radiometry infrared region can either be reflection (at temperatures close to
the ambient temperature of Earth), or it can be emission (when
the target is at much higher temperatures) with the amount of
Types
of
Absolute normalization Relative normalization energy emitted being dependent on the temperature of the tar-
(radiometric correction, (applicable to get as guided by the Plancks function (Prakash and Gens, 2010;
normalization
calibration) all wavelengths)
Prakash and Gupta, 1999). For the purpose of data normaliza-
tion, the logical way to categorize the data is not by the absolute
FIg u r e 5.1 Remote sensing data normalization overview for various wavelengths, but by the dominant physical processes of reflec-
types of imagery. tion and emission. Therefore, we classify the data as optical and
In combining image datasets, it is assumed that the images are thermal, where optical data occupy the shorter wavelength por-
corrected for imaging geometry, that is, optical and thermal imag- tion and the thermal data occupy the longer wavelength portion
ery have been orthorectified (Lillesand etal., 2007) and SAR imag- of this range.
ery has been terrain corrected (Small, 2011). In the case of SAR data, Optical and thermal remote sensing depend on incoming
it should be noted the radiometric terrain correction also corrects radiation from the sun and emitted radiation from the Earths
the pixel brightness due to geometric distortions (Small, 2011). surface, respectively. It does not depend on any external source
These geometric corrections are prerequisite for the coregistration of energy and, therefore, is also classified as passive remote sens-
of any multitemporal imagery. Image coregistration and radiomet- ing. Passive remote sensing data can be acquired in broad spec-
ric corrections are considered the most important steps in moni- tral bands, for example, panchromatic visible imagery, as well as
toring activities such as change detection (Hussain etal., 2013). in narrower bands as multispectral (spectral bandwidth in the
Figure 5.1 summarizes the elements associated with the topic order of 100nm) and hyperspectral data (spectral bandwidth in
of data normalization covered in this chapter. The chapter focuses the order of about 10nm) (Gens, 2009).
on three image types in the optical, thermal, and microwave spec- SAR data are acquired in the microwave region of the elec-
tra and the approaches for their absolute and relative normal- tromagnetic spectrum between 0.75 cm and 1 m in wave-
izations. A number of sources of radiometric distortion such as length. As the energy is transmitted as a single frequency from
sensor calibration and degradation as well as the atmospheric and the sensor itself, this technique is classified as active remote
topographic influence on the radiometry are identified that need sensing. The SAR sensor not only transmits energy, but also
to be considered. Depending on the wavelength region, absolute records the energy backscattered from the target. SAR data
normalization may involve radiometric correction or calibration, can be acquired day and night and in all weather conditions
whereas relative normalization is applicable to all wavelengths. (Gens, 2009). Unless used for SAR interferometric processing,
the imagery is not affected by atmospheric effects and does
not need to be corrected for its influence. SAR signals can be
5.2 Remote Sensing Data transmitted and received with horizontal and vertical polar-
Remote sensing data, as addressed in this chapter, refer to images ization. The various combinations of polarizations have dif-
acquired anywhere from the visible to the microwave region of ferent backscatter behaviors and, therefore, provide additional
the electromagnetic spectrum (Figure 5.2). information, complementary to the spectral information. Data
Optical
Infrared Microwave
Visible Near Middle Thermal K X C S L P
400 nm 700 nm 1 m 3 m 7 m 14 m 0.75 cm 2.5 cm 4 cm 8 cm 15 cm 30 cm 1 m
FIg u r e 5.2 Optical, thermal, and SAR data and their respective wavelengths in the electromagnetic spectrum. (Adapted from Gens, R., Spectral
information content of remote sensing imagery, in Li, D., Shan J., and Gong, J., eds., Geospatial Technology for Earth Observation, Springer, New
York, 2009.)
2016 by Taylor & Francis Group, LLC
Remote Sensing Data Normalization 135
acquired in passive mode within the microwave region are the microwave region, and for most application purposes, with
not addressed in this chapter. However, readers should refer the exception of applications that rely on SAR interferometric
Chapter 1 to get greater understanding of various sensors and processing; the atmosphere has an insignificant influence on
their characteristics. SAR data.
Teasing out the pure signal from the target from a mixed
5.3 Sources of Radiometric Distortion signal response coming from the target and atmosphere then
becomes important. Common methods for atmospheric cor-
The main sources of radiometric distortions stem from issues rection in the optical and thermal regions are discussed in
related to sensor calibration and sensor degradation, atmo- Section 5.4.1.1.
spheric interactions, and influences of topographic variations on
image radiometry.
5.3.3 topographic influence on Radiometry
Surface topographic variations influence the radiometric
5.3.1 Sensor calibration and Degradation
response of optical, thermal, and SAR data by influencing the
The Committee on Earth Observation Satellites (CEOS) defines sourcetargetsensor geometry. The influence is most pro-
sensor calibration as the process of quantitatively defining the nounced in high-altitude and high-latitude areas, showing high
system response to known, controlled signal inputs. The purpose topographic variations.
of calibration activities is to ensure that the user can retrieve as
accurate and meaningful quantitative information from the 5.3.3.1 effect of topography on optical Data
remote sensing images as possible. All sensors and onboard Differences in illumination conditions due to solar position
calibration devices undergo a rigorous prelaunch calibration, at the moment of image acquisition with respect to surface
but need to be routinely recalibrated due to degradation over slope and aspect or elevation can produce similar reflectance
time (Chander etal., 2009; Wang etal., 2012). Postlaunch cali- responses for similar terrain features (Vanonckelen etal., 2013).
bration activities help to correct the onboard calibrators and Accounting for a topographic correction is then important to
to verify that the signal response has not drifted away from the calculate surface reflectances accurately, especially in high relief
original response for the controlled signal inputs. Any error in areas (Hantson and Chuvieco, 2011). According to Pons et al.
sensor calibration will propagate and cause errors in quantita- (2014), there are several methodologies that account for topo-
tive retrievals from remote sensing data. Excellent reviews of graphic effects based on the phenological stage (Hantson and
pre- and postlaunch calibration techniques for optical and ther- Chuvieco, 2011; Meyer et al., 1993; Riao et al., 2003; Vincini
mal systems are available in the literature (e.g., Datla etal., 2011; and Reeder, 2000), the bidirectional reflectance distribution
Schott etal., 2012; Xiong etal., 2009) function (BRDF), or the cosine topographic correction model
Typically, the agency responsible for the satellite launch and (Teillet etal., 1982). The cosine topographic correction is most
operation undertakes the calibration tasks and provides calibra- convenient for automated radiometric correction procedures.
tion parameters, also referred to as calibration constants, as part Methodologies accounting for phenological stage require knowl-
of the metadata associated with the image data. An end user edge on the different land cover classes or ground reference
then applies the correct calibration constants in the respective information, while BRDF models have limitations, as they at
radiometric correction algorithms to convert the image digital times fail to remove angular effects in spectral bands sensitive to
numbers (DNs) to derived physical parameters such as ground water vapor absorption and caused by large seasonal oscillation
reflectances, ground temperatures, etc. For the sake of brevity, a (Kim etal., 2012). They also need information that is rarely avail-
further discussion on optical and sensor calibration is not pre- able and extremely difficult to obtain for regional or long-term
sented in this chapter. SAR calibration details are presented in studies (Goslee, 2012).
Section 5.4.1.2 under absolute radiometric correction.
5.3.3.2 effect of topography on SAR Data
SAR systems have a side-looking geometry. This viewing geom-
5.3.2 Atmospheric influence on Radiometry
etry leads to geometric distortions, especially in areas with high
The atmosphere plays a huge role in attenuating the signal topography. The information from slopes facing the sensor is
recorded by the sensor. Atmospheric particles cause selective compressed, resulting in brighter pixels, while slopes facing
scattering, absorption, and emission influencing the signal away from the sensor in shadow regions, not covered by any sig-
from the target. The atmosphere intervenes twice in optical nal at all, appear dark.
images: once when the electromagnetic radiations travel from This geometric distortion can be corrected for using a digital
the source (Sun) to the Earth, and the second time when the elevation model. While this technique, called terrain correction,
radiations travel after reflection from the Earth to the sensor. shifts the pixels into the correct geolocation, it is not able to com-
In thermal images, the atmospheric influence comes to play pletely recover the radiometric information. By calculating the
only once as the emitted signal from the Earth travels up to area that was covered by the signal, the correction factors, later
the satellite sensor. The atmosphere is largely transmissive in multiplied by the geometrically corrected SAR image, can be
2016 by Taylor & Francis Group, LLC
136 Remotely Sensed Data Characterization, Classification, and Accuracies
1135W 1130W 11255W 1135W 1130W 11255W
N N
4555N
4555N
4550N
4550N
0 2.5 5 km 0 2.5 5 km
(a) (b)
FIg ur e 5.3 Example of terrain correction of ALOS PALSAR data in the Rocky Mountains. (a) Uncorrected version, clearly showing the topo-
graphic structure of the mountains. In the image on (b), geometric distortions have been removed, so the topography appears flattened. (Imagery
copyright Japan Aerospace Exploration Agency (JAXA), Tokyo, Japan, 2006.)
determined (Small, 2011). However, the distribution of the indi- on relative change from one time to another, relative radiometric
vidual scatterers and their contribution to the received backscat- normalization may be sufficient. The characteristics of the abso-
tered signal within this area remains unknown (Figure 5.3). lute and relative normalizations are summarized in Table 5.1.
5.4.1 Absolute Radiometric correction
5.4 Radiometric normalization
Techniques for absolute radiometric correction use insitu mea-
There are two types of radiometric normalization: absolute and surements or model data to convert from DN to an absolute
relative. The absolute radiometric normalization, also known as scale. Residual effects might still require a relative radiometric
radiometric correction, helps to derive the absolute reflectance of normalization.
targets at the Earths surface. The relative radiometric normal-
ization adjusts the radiometric properties of targets within an 5.4.1.1 correction of optical and thermal Data
image to match a reference image (Janzen etal., 2006; Yuan and Absolute radiometric correction is a two-step process. In a first
Elvidge, 1996). Depending on the application needs, an absolute step, the sensor-specific calibration parameters (gain and offset),
radiometric correction may not be required in all cases. A relative determined usually prior to launch, are applied to convert the
radiometric normalization based on radiometric information in DNs (dimensionless) into spectral radiance Lsat (in W m2 sr 1
the imagery is sufficient for change detection studies or super- m1) using the equation (Chen etal., 2005)
vised land cover classifications (Canty et al., 2004). In general,
Lsat = DN Gain + Offset.
absolute radiometric corrections are required to derive quan-
titative parameters such as biophysical variables from remote Lsat can then be converted to top of atmosphere (TOA) reflec-
sensing images. For other applications such as studies focused tances or temperatures. Applying a radiometric correction to
TABLe 5 .1 Summary of Absolute and Relative Normalizations for the Various Data Types
Absolute Normalization Relative Normalization
Technique Effect Technique Effect
Radiative transfer methods Accounts for changes in Statistical methods using Minimizes effects of changing
Optical
satellite sensor calibration histograms, linear regression, etc. atmospheric and solar conditions
over time, difference among
in-band solar spectral
irradiance, solar angle,
Thermal
Data
topography, and atmospheric
interferences
Analysis in homogeneous Adjusts the intensity to Statistical methods using Preserves dominating scattering
SAR
areas with known reference backscatter level information from common area mechanism
backscatter
2016 by Taylor & Francis Group, LLC
Remote Sensing Data Normalization 137
retrieve surface reflectance or land surface temperature (LST) over large areas (Themistocleous et al., 2012). Reanalysis data
(kinetic temperature) is needed to be able to analyze long time can also provide atmospheric inputs for atmospheric correction;
series of remote sensing data. however, its current spatial resolution is still too coarse to be
Surface reflectance surface can be determined by applied to medium or coarse resolution imagery.
An example of atmospheric correction in the thermal infrared
surface =
( Lsat Lpath ) , (TIR) using a single-channel algorithm proposed by Pons and Sol-
E Sugraes (1994) and Pons et al. (2014) using Landsat-5 TM data
is shown in Figure 5.4. In remote areas such as Alaska (U.S.), col-
where lection of atmospheric data is challenging because of remoteness,
Lpath is the path radiance winter conditions, and the high costs of maintaining ground-based
E is the exoatmopsheric irradiance on the ground target measurement sensors (Cristbal etal., 2012). Therefore, there is a
is the transmission of the atmosphere (Lillesand etal., 2007) real need of methods requiring few atmospheric inputs. The radio-
metric correction shown in Figure 5.4, based on the dark object
Other methodologies to retrieve surface reflectance can be found subtraction (extracted from the image histogram) and the cosine
in Vicente-Serrano etal. (2008). It is important to note that an topographic correction model (computed though a digital elevation
absolute method should account for changes in satellite sensor model), allows reducing the number of undesired artifacts due to
calibration over time, differences among in-band solar spectral the atmospheric effects or differential illumination that are results
irradiance, solar angle, and variability in EarthSun distance of time of day, location on Earth, and relief (zones being more illu-
and atmospheric interferences (de Carvalho et al., 2013). To minated than others, shadows, etc.), minimizing the effect of these
remove the atmospheric effects, detailed information on atmo- factors on the image data.
spheric parameters such as aerosols, water vapor, or ozone is In the case of the TIR region, a suitable and popular proce-
often required. Local atmospheric measurements or reanalysis dure to retrieve LST is by the inversion of the radiative transfer
data are a common source of atmospheric information. However, equation. The following expression can be then applied to a cer-
atmospheric radiosondes are usually not available at the time of tain sensor channel (or wavelength interval).
satellite pass and a single atmospheric radiosonde might not
be representative of the atmospheric conditions of wide-swath Lsensor , = BTs + (1 ) Latm , + Latm , ,
satellite images such as the one provided by Landsat, NOAA-
AVHRR, or TERRA/AQUA-MODIS sensors, especially in areas
with highly variable relief (Cristbal et al., 2009). AERONET where
network is another important source of atmospheric data, but Lsensor is at-sensor radiance (in W m2 sr 1 m1)
as in the case of radiosonde data, ground network distribution is the land surface emissivity (dimensionless)
might not be wide enough to provide atmospheric parameters is the wavelength (in m)
15012W 1506W 15012W 1506W
N N
6257N
6257N
6254N
6254N
0 1.25 2.5 km 0 1.25 2.5 km
(a) (b)
FIgu r e 5.4 Example of radiometric correction of optical data in Alaska (US) from a Landsat-5 TM image of 07 July 2009 (standard false color
composite). (a) Uncorrected version, clearly showing the topographic structure of the mountains. (b) Radiometrically corrected image where topo-
graphic and atmospheric effects have been removed. For example, areas with a similar land cover that occupy different sides of a hilly terrain show
self-cast shadows on the left panel but present similar reflectivity in the (b). (Images are courtesy of the U.S. Geological Survey.)
2016 by Taylor & Francis Group, LLC
138 Remotely Sensed Data Characterization, Classification, and Accuracies
Ts is the LST (in K) missions, split-window algorithms can be applied (Jimenez-
Latm, is the downwelling atmospheric radiance (hemi- Muoz et al., 2014; Wan and Dozier, 1996). The temperature
spherical flux divided by pi and in W m2 sr1 m1) and emissivity separation algorithm can be applied when five
Latm, is the upwelling atmospheric radiance (path radiance or more bands are available (Gillespie et al., 1998) as in the
at wavelength and in W m2 sr1 m1) ASTER mission. Single-channel and split-window methods
is the atmospheric transmissivity (dimensionless) also require the knowledge of land surface emissivity as well as
water vapor. Land surface emissivity estimates for bare soil and
B term is Plancks law, expressed as follows: vegetation covers can be retrieved for operational processing
using methods based on the Normalized Difference Vegetation
2hc 2 Index (NDVI) (Sobrino and Raissouni, 2000; Sobrino et al.,
B (Ts ) = ,
hc 2008; Valor and Caselles, 2005). Water vapor can be obtained
1
5
(kTs ) by means of a local radiosounding or at regional scale by means
of remote sensing data (Sobrino et al., 1999), or via image
where based water vapor products such as the TERRA/AQUA MODIS
c is the speed of light (2.998 108 m s1) product (MOD/MYD_05).
h is the Plancks constant (6.626076 10 34 Js) An example of atmospheric correction in the TIR using a
k is the Boltzmann constant (1.3806 10 23 J K1) single-channel algorithm proposed by Cristbal et al. (2009)
using Landsat-5 TM data, water vapor estimates from the
Usually, LST retrieval methods are developed based on the MODIS water vapor product and emissivity retrieved using
available thermal sensor bands and can be classified in single- Sobrino and Raissouni (2000) are shown in Figure 5.5b
channel, split-window or temperature and emissivity sepa- (Figure5.5a shows a radiometric correction of optical bands to
ration algorithms. A single-channel algorithm is applied to assist the interpretation of the LST image). This method works
thermal sensors with only one band in the TIR (Cristbal etal., well in areas where only remote sensing estimates of water vapor
2009; Jimnez-Muoz and Sobrino, 2003), such as Landsat-5 products and LSE are available as input for LST estimation.
or Landsat-7 missions. In this case, when two thermal bands The magnitude to which brightness temperature (TOA tem-
are available, such as in NOAA-AVHRR or Landsat-8 TIRS peratures not corrected by atmospheric or emissivity effects)
152W 15130W 152W 15130W 152W 15130W
N N N
LST (k) LST - BT (k)
275 2.20
277 2.40
280 2.50
281 2.65
283 2.80
285 2.95
288 3.15
293 3.65
6140N
6140N
6140N
6120N
6120N
6120N
0 5 10 km 0 5 10 km 0 5 10 km
(a) (b) (c)
FIg ur e 5.5 Example of atmospheric correction of thermal data in Alaska (U.S.) from a Landsat-5 TM image of 04 May 2009 (standard false color
composite). (a) Optical image after radiometric correction. (b) LST derived after correcting for atmospheric effects and emissivity. (c)Difference
image generated by subtracting the LST corrected for atmospheric effects and emissivity and the brightness temperature, BT (not corrected for
emissivity or atmospheric effects). (Images are courtesy of the U.S. Geological Survey, Reston, VA.)
2016 by Taylor & Francis Group, LLC
Remote Sensing Data Normalization 139
are corrected isshown in Figure 5.5a. This magnitude difference currently no consensus on what method is more suitable, more
is computed by subtracting the LST corrected by atmospheric objective and automatic methods with less manual interven-
and emissivity effects (mid panel) and the brightness tempera- tion and that take advantage of long time series of remote sens-
ture. In this case, removing atmospheric and emissivity effects ing data, such as those based on the pseudo-invariant areas, are
lead to a correction around 2 and 4 K: brightness temperature preferred for relative radiometric normalization of optical and
is clearly underestimating LST, and absolute radiometric cor- thermal imagery.
rection methodologies are needed for accurate LST retrievals.
5.4.2.2 Relative correction of SAR Data
5.4.1.2 SAR calibration Most often absolute radiometric correction of SAR data is
In order to calibrate an SAR image, the SAR processor needs to sufficient for analysis. Sometimes a relative radiometric nor-
be calibrated, that is, calibration coefficients have been deter- malization is merited, especially when adjacent scenes have
mined. Applying these coefficients to the SAR data calibrates significant seasonal differences. The normalization of SAR
the SAR image. The DNs of the original image get converted polarimetric data is slightly more complicated, as the cor-
into power scale, a ratio of the power that is backscattered rection is not supposed to change the scattering mecha-
to the power sent within a resolution cell. Calibrated images nism. For relative radiometric normalization, Shimada and
are often transformed from power scale into the logarithmic Ohtaki (2010) used a polygonal curve approximation to sup-
dB scale. press differences in intensity between neighboring image
In order to derive geophysical parameters, the calibration needs strips. Antropov et al. (2012) extended the relative correla-
to meet certain requirements. The absolute calibration needs to be tion approach of Shimada and Ohtaki by using the span of
1 dB, the long-term relative calibration 0.5 dB, the short-term the covariance matrix to calculate the corrective gain. There
relative calibration better than 0.5 dB (Freeman, 1992). are various ways to apply the radiometric correction in this
Another technique that can provide radiometric calibration case. Lee et al. (2004) used only those pixels from the over-
uses so-called permanent scatterers. These are natural targets in lapping areas, where a dominating scattering mechanism is
a stack of SAR images that are stable over time and act as cor- preserved. However, for several applications where the pur-
ner reflectors with an unknown radar cross section and with pose is to generate a thematically classified product based on
a quality that can be estimated by the repeated observations relative clustering of backscatter values within an image, sim-
(DAria etal., 2010). These are also exploited in permanent scat- pler radiometric normalization techniques, such as color bal-
terer interferometry that allows long-term time series analysis ance or histogram matching, used popularly in optical remote
(Ferretti etal., 2001). sensing can prove to be equally efficient.
Figure 5.6a shows an example mosaic from the northern
foothills of Alaska Range generated using two adjacent SAR
5.4.2 Relative Radiometric correction
polarimetric images from different seasons. The images were
A relative correction transforms the DNs to a common scale, terrain corrected prior to generating the mosaic. A general
adjusting the radiometric properties of an image to match a ref- brightness contrast is visible across the image boundary (the
erence image (deCarvalho etal., 2013). red dashed line in Figure 5.6b). Note that the boundary is
not straight, due to the effect of terrain correction. Relative
5.4.2.1 correction of optical and thermal Data radiometric normalization using a histogram match of the
The goal of image normalization for the relative radiometric overlapping area was applied to the terrain corrected image
correction of optical and thermal data is to reduce the radio- pair. The mosaic generated using these normalized images
metric influence of nonsurface factors, so that the differences yielded a superior product (5.6c) for mapping and classifica-
in DN between satellite images from different dates will reflect tion applications.
actual changes on the surface of the Earth (Heo and Fitz-Hugh,
2000). Several techniques for relative radiometric correction 5.5 conclusions
have been developed in the last decades such as robust and
linear regression (El Hajj etal., 2008; Olsson, 1993; Wessman, In general, absolute radiometric correction methods are
1987), histogram matching (Chavez and MacKinnon, 1994; required to derive quantitative parameters such as biophysical
Liang, 2002), the use of invariant areas (Eckhardt etal., 1990; variables from remote sensing images. However, for other appli-
Jensen etal., 1995; Michener and Houhoulis, 1997), the use of cations such as studies focused on relative change from one time
pseudoinvariant areas (Bao etal., 2012; Pons etal., 2014; Schott to another, relative radiometric normalization may be sufficient.
et al., 1988; Zhou et al., 2011), and Gaussian method (Singh, In the case of absolute radiometric correction methods, remov-
1989), among others. Linear regression, invariant and pseudo- ing atmospheric effects often requires detailed information on
invariant, methods also require an appropriate selection of sta- atmospheric parameters such as aerosols, water vapor, or ozone
ble or quasistable radiometric areas, areas that can be selected that are not often available at regional scales. Therefore, absolute
using methods such as the multivariate alteration detection radiometric correction methods that minimize input variables
(Nielsen et al., 1998; Scheidt et al., 2008). Although there is are preferred for optical and thermal imagery.
2016 by Taylor & Francis Group, LLC
140 Remotely Sensed Data Characterization, Classification, and Accuracies
14840W 14830W 14820W 14810W For their study on the radiometric cross-calibration of
N Landsat sensors, Teillet et al. (2001) concluded that the most
6345N
limiting factor of their approach is the need to adjust for spec-
tral band differences between the two sensors, introducing a
dependency on knowledge mainly about the surface reflectance
spectrum of the scene. Particularly, the cross-calibration of the
shortwave infrared bands remains an issue without these spec-
6340N
tra (Teillet etal., 2001).
References
Antropov, O., Rauste, Y., Lonnqvist, A., and Hame, T. 2012.
6335N
PolSAR mosaic normalization for improved land-cover
0 2.5 5 km
mapping. IEEE Geoscience and Remote Sensing Letters, 9,
(a) 10741078.
Bao, N. S., Lechner, A. M., Fletcher, A., Mellor, A., Mulligan, D., and
14840W 14830W 14820W 14810W Bai, Z. K. 2012. Comparison of relative radiometric normal-
N ization methods using pseudo-invariant features for change
6345N
detection studies in rural and urban landscapes. Journal of
Applied Remote Sensing, 6, 063578-063571-063578-063518.
Canty, M. J., Nielsen, A. A., and Schmidt, M. 2004. Automatic
radiometric normalization of multitemporal satellite imag-
ery. Remote Sensing of Environment, 91, 441451.
6340N
Chander, G., Markham, B. L., and Helder, D. L. 2009. Summary
of current radiometric calibration coefficients for Landsat
MSS, TM ETM+ and EO-1 ALI sensors. Remote Sensing of
Environment, 113, 893903.
Chavez, P. S. and MacKinnon, D. L. 1994. Automatic detection of
6335N
0 2.5 5 km vegetation changes in the Southwestern United States Using
Remotely Sensed Images. Photogrammetric Engineering and
(b) Remote Sensing, 60, 571583.
Chen, X. X., Vierling, L., and Deering, D. 2005. A simple and
14840W 14830W 14820W 14810W
effective radiometric correction method to improve land-
N
scape change detection across sensors and across time.
6345N
Remote Sensing of Environment, 98, 6379.
Cristbal, J., JimnezMuoz, J. C., Sobrino, J. A., Ninyerola, M.,
and Pons, X. 2009. Improvements in land surface temperature
retrieval from the Landsat series thermal band using water
vapor and air temperature. Journal of Geophysical Research
6340N
Atmospheres, 114, D08103, doi: 10.1029/2008JD010616.
Cristbal, J., Prakash, A., Starkenburg, D., Fochesatto, J., Anderson,
M. A., Kustas, W. P., Alfieri, J. G., Gens, R., and Kane, D.
2012. Energy fluxes retrieval on an Alaskan Arctic and Sub-
Arctic vegetation by means MODIS imagery and the DTD
6335N
0 2.5 5 km method. AGU Fall Meeting, San Francisco, CA, December
37, 2012.
(c) DAria, D., Ferretti, A., Guarnieri, A. M., and Tebaldini, S. 2010. SAR
calibration aided by permanent scatterers. IEEE Transactions
FIgu r e 5.6 Relative correction of adjacent polarimetric ALOS
on Geoscience and Remote Sensing, 48, 20762086.
PALSAR images. The Yamaguchi decompositions have been terrain
Datla, R. U., Rice, J. P., Lykke, K. R., Johnson, B. C., Butler, J. J.,
corrected. (a) Uncorrected mosaic. (b) Boundary between the two
images that are indicated (not a straight line because of the terrain and Xiong, X. 2011. Best practice guidelines for pre-launch
correction). (c) Relative corrected mosaic. (Imagery copyright Japan characterization and calibration of instruments for passive
Aerospace Exploration Agency (JAXA), Tokyo, Japan, 2007.) optical remote sensing. Journal of Research of the National
Institute of Standards and Technology, 116, 621646.
2016 by Taylor & Francis Group, LLC
Remote Sensing Data Normalization 141
de Carvalho, O. A., Guimaraes, R. F., Silva, N. C., Gillespie, A. R., Jensen, J. R., Rutchey, K., Koch, M. S., and Narumalani, S. 1995. Inland
Gomes, R. A. T., Silva, C. R., and De Carvalho, A. P. F. 2013. wetland change detection in the Everglades Water Conservation
Radiometric normalization of temporal images combin- Area 2A using a time series of normalized remotely sensed data.
ing automatic detection of pseudo-invariant features from Photogrammetric Engineering and Remote Sensing, 61, 199209.
the distance and similarity spectral measures, density scat- Jimnez-Muoz, J. C. and Sobrino, J. A. 2003. A generalized sin-
terplot analysis, and robust regression. Remote Sensing, 5, gle-channel method for retrieving land surface temperature
27632794. from remote sensing data. Journal of Geophysical Research
Du, Y., Teillet, P. M., and Cihlar, J. 2002. Radiometric normaliza- Atmospheres, 108, 4688, doi: 10.1029/2003JD003480.
tion of multitemporal high-resolution satellite images with Jimenez-Munoz, J. C., Sobrino, J. A., Skokovic, D., Mattar, C., and
quality control for land cover change detection. Remote Cristobal, J. 2014. Land surface temperature retrieval meth-
Sensing of Environment, 82, 123134. ods from Landsat-8 thermal infrared sensor data. IEEE
Eckhardt, D. W., Verdin, J. P., and Lyford, G. R. 1990. Automated Geoscience and Remote Sensing Letters, 11, 18401843.
update of an irrigated lands GIS using SPOT HRV imag- Kim, D. S., Pyeon, M. W., Eo, Y. D., Byun, Y. G., and Kim, Y. I.
ery. Photogrammmetric Engineering and Remote Sensing, 56, 2012. Automatic pseudo-invariant feature extraction for the
15151522. relative radiometric normalization of Hyperion hyperspec-
El Hajj, M., Bgu, A., Lafrance, B., Hagolle, O., Dedieu, G., and tral images. Giscience and Remote Sensing, 49, 755773.
Rumeau, M. 2008. Relative radiometric normalization and Lee, J. S., Grunes, M. R., Pottier, E., and Ferro-Famil, L. 2004.
atmospheric correction of a SPOT 5 time series. Sensors, 8, Unsupervised terrain classification preserving polarimetric
27742791. scattering characteristics. IEEE Transactions on Geoscience
Ferretti, A., Prati, C., and Rocca, F. 2001. Permanent scatterers in and Remote Sensing, 42, 722731.
SAR interferometry. IEEE Transactions on Geoscience and Liang, S. 2002. Estimation of land surface biophysical variables.
Remote Sensing, 39, 820. In: Kong, J. A. (ed.) Quantitative Remote Sensing of Land
Freeman, A. 1992. SAR calibrationAn overview. IEEE Transactions Surfaces. Wiley, Hoboken, NJ, pp. 247264.
on Geoscience and Remote Sensing, 30, 11071121. Lillesand, T., Kiefer, R. W., and Chipman, A. 2007. Remote Sensing
Gens, R. 2009. Spectral information content of remote sensing and Image Interpretation. Wiley, Hoboken, NJ.
imagery. In: Li, D., Shan, J., and Gong, J. (eds.) Geospatial Meyer, P., Itten, K. I., Kellenbenberger, T., Sandmeier, S., and
Technology for Earth Observation. Springer, New York. Sandmeier, R. 1993. Radiometric corrections of topo-
Gillespie, A., Rokugawa, S., Matsunaga, T., Cothern, J. S., Hook, S., graphically induced effects on Landsat TM data in an alpine
and Kahle, A. B. 1998. A temperature and emissivity sepa- environment. ISPRS Journal of Photogrammetry and Remote
ration algorithm for advanced spaceborne thermal emission Sensing, 48, 1728.
andreflection radiometer (ASTER) images. IEEE Transactions Michener, W. K. and Houhoulis, P. F. 1997. Detection of vegeta-
on Geoscience and Remote Sensing, 36, 11131126. tion changes associated with extensive flooding in a for-
Goslee S. C. 2012. Topographic corrections of satellite data for ested ecosystem. Photogrammetric Engineering and Remote
regional monitoring. Photogrammetric Engineering and Sensing, 63, 173181.
Remote Sensing, 78, 973981. Nielsen, A. A., Conradsen, K., and Simpson, J. J. 1998. Multivariate
Gupta, R. P. 2003. Remote Sensing Geology, 2nd edn. Springer, alteration detection (MAD) and MAF postprocessing in mul-
Berlin/Heidelberg, 656 pp., ISBN-13: 9783540431855. tispectral, bitemporal image data: New approaches to change
Hantson, S. and Chuvieco, E. 2011. Evaluation of different detection studies. Remote Sensing of Environment, 64, 119.
topographic correction methods for Landsat imagery. Olsson, H. 1993. Regression functions for multi-temporal rela-
International Journal of Applied Earth Observation and tive calibration of thematic mapper data over Boreal forest.
Geoinformation, 13, 691700. Remote Sensing of Environment, 46, 89102.
Heo, J. and FitzHugh, T. W. 2000. A standardized radiometric Pons, X., Pesquer, L., Cristbal, J., and Gonzlez-Guerrero, O.
normalization method for change detection using remotely 2014. Automatic and improved radiometric correction of
sensed imagery. Photogrammetric Engineering an Remote Landsat imagery using reference values from MODIS sur-
Sensing, 66, 173181. face reflectance images. International Journal of Applied
Hussain, M., Chen, D. M., Cheng, A., Wei, H., and Stanley, D. Earth Observation and Geoinformation, 33, 243254.
2013. Change detection from remotely sensed images: Pons, X. and Sol-Sugraes, L. 1994. A simple radiometric cor-
From pixel-based to object-based approaches. ISPRS rection model to improve automatic mapping of vegeta-
Journal of Photogrammetry and Remote Sensing, 80, tion from multispectral satellite data. Remote Sensing of
91106. Environment, 48, 191204.
Janzen, D. T., Fredeen, A. L., and Wheate, R. D. 2006. Radiometric Prakash, A. and Gens, R. 2010. Remote sensing of coal fires. In:
correction techniques and accuracy assessment for Landsat Stracher, G. B., Prakash, A., Sokol, E. V. (eds.) Coal and Peat
TM data in remote forested regions. Canadian Journal of Fires: A Global Perspective, Vol. 1: CoalCombustion and
Remote Sensing, 32, 330340. Geology. Elsevier, Oxford, U.K.
2016 by Taylor & Francis Group, LLC
142 Remotely Sensed Data Characterization, Classification, and Accuracies
Prakash, A. and Gupta, R. P. 1999. Surface fires in Jharia Coalfield, Teillet, P. M., Barker, J. L., Markham, B. L., Irish, R. R., Fedosjevs,
IndiaTheir distribution and estimation of area and tem- G., and Storey, J. C. 2001. Radiometric cross-calibration of
perature from TM data. International Journal of Remote the Landsat-7 ETM+ and Landsat-5 TM sensors based on
Sensing, 20, 19351946. tandem data sets. Remote Sensing of Environment, 78, 3954.
Quattrochi, D. A., Prakash, A., Evena, M., Wright, R., Hall, D. K., Teillet, P. M., Guindon, B., and Goodeonugh, D. G. 1982. On
Anderson, M., Kustas, W. P., Allen, R. G., Pagano, T., and the slope-aspect correction of multispectral scanner data.
Coolbaugh, M. F. 2009. Thermal remote sensing: Theory, Canadian Journal of Remote Sensing, 8, 84106.
sensors, and applications. In: Jackson, M. (ed.) Manual of Themistocleous, K., Hadjimitsis, D. G., Retalis, A., and
Remote Sensing 1.1: Earth Observing Platforms & Sensors. Chrysoulakis, N. 2012. Development of a new image based
ASPRS, Bethesda, MD, 550 pp. atmospheric correction algorithm for aerosol optical thick-
Riao, D., Chuvieco, E., Salas, J., and Aguado, I. 2003. Assessment ness retrieval using the darkest pixel method. Journal of
of different topographic corrections in Landsat-TM data for Applied Remote Sensing, 6, 063538.
mapping vegetation types. IEEE Transactions on Geoscience Valor, E. and Caselles, V. 2005. Validation of the vegetation cover
and Remote Sensing, 41, 10561061. method for land surface emissivity estimation. In: Caselles,
Scheidt, S., Ramsey, M., and Lancaster, N. 2008. Radiometric V., Valor, E., and Coll, C. (eds.) Recent Research Developments
normalization and image mosaic generation of ASTER in Thermal Remote Sensing. Research Signpost, Kerala, India,
thermal infrared data: An application to extensive sand pp. 120.
sheets and dune fields. Remote Sensing of Environment, Vanonckelen, S., Lhermitte, S., and Van Rompaey, A. 2013. The
112, 920933. effect of atmospheric and topographic correction methods
Schott, J. R., Hook, S. J., Barsi, J. A., Markham, B. L., Miller, J., on land cover classification accuracy. International Journal
Padula, F. P., and Raqueno, N. G. 2012. Thermal infrared of Applied Earth Observation and Geoinformation, 24, 921.
radiometric calibration of the entire Landsat 4, 5, and 7 Vicente-Serrano, S. M., Prez-Cabello, F., and Lasanta, T. 2008.
archive (19822010). Remote Sensing of Environment, 122, Assessment of radiometric correction techniques in analyzing
4149. vegetation variability and change using time series of Landsat
Schott, J. R., Salvaggio, C., and Vochok, W. J. 1988. Radiometric images. Remote Sensing of Environment, 112, 39163934.
scene normalization using pseudo-invariant features. Vincini, M. and Reeder, D. 2000. Minnaert topographic nor-
Remote Sensing of Environment, 26, 116. malization of Landsat TM imagery in rugged forest areas.
Shimada, M. and Ohtaki, T. 2010. Generating large-scale high- International Archives of Photogrammetry and Remote
quality SAR mosaic datasets: Application to PALSAR data Sensing, Vol. XXXIII, Part B7, Amsterdam, the Netherlands.
for global monitoring. IEEE Journal of Selected Topics in Wan, Z. and Dozier, J. 1996. A generalized split-window algo-
Applied Earth Observations and Remote Sensing, 3, 637656. rithm for retrieving land-surface temperature from space.
Singh, A. 1989. Digital change detection techniques using IEEE Transactions on Geoscience and Remote Sensing, 34,
remotely sensed data. International Journal of Remote 892905.
Sensing, 10, 9891103. Wang, D. D., Morton, D., Masek, J., Wu, A. S., Nagol, J., Xiong, X.
Small, D. 2011. Flattening Gamma: Radiometric terrain correc- X., Levy, R., Vermote, E., and Wolfe, R. 2012. Impact of sen-
tion for SAR imagery. IEEE Transactions on Geoscience and sor degradation on the MODIS NDVI time series. Remote
Remote Sensing, 49, 30813093. Sensing of Environment, 119, 5561.
Sobrino, J. A., Jimnez-Muoz, J. C., Sria, G., Romaguera, M., Wessman, C. A. 1987. Analysis of Landsat Thematic Mapper imagery
Guanter, L., Moreno, J., Plaza, A., and Martnez, P. 2008. over UW arboretum and Blackhawk Island. PhD Dissertation,
Land surface emissivity retrieval from different VNIR and University of Wisconsin-Madison, Madison, WI.
TIR sensors. IEEE Transactions on Geoscience and Remote Xiong, X. X., Wenny, B. N., and Barnes, W. L. 2009. Overview
Sensing, 46, 316327. of NASA Earth Observing Systems Terra and Aqua mod-
Sobrino, J. A. and Raissouni, N. 2000. Toward remote sensing erate resolution imaging spectroradiometer instrument
methods for land cover dynamic monitoring: Application calibration algorithms and on-orbit performance. Journal
to Morocco. International Journal of Remote Sensing, 21, of Applied Remote Sensing, 3, 032501.
353366. Yuan, D. and Elvidge, C. D. 1996. Comparison of relative
Sobrino, J. A., Raissouni, N., Simarro, J., Nerry, F., and Franois, P. radiometric normalization techniques. ISPRS Journal of
1999. Atmospheric water vapour content over land surfaces Photogrammetry and Remote Sensing, 51, 117126.
derived from the AVHRR data. Application to the Iberian Zhou, Q., Li, B., and Chen, Y. 2011. Remote sensing change detec-
Peninsula. IEEE Transactions and Geoscience and Remote tion and process analysis of long-term land use change and
Sensing, 37, 14251434. human impacts. AMBIO, 40, 807818.
2016 by Taylor & Francis Group, LLC
6
Satellite Data Degradations
and Their Impacts on
High-Level Products
Acronyms and Definitions .................................................................................................................143
6.1 Introduction .............................................................................................................................143
6.2 Common Issues of Data Degradation ..................................................................................144
Radiometric Degradation without Onboard Calibration Systems Radiometric Degradation
with Onboard Calibration Device Orbit Drift of Sun-Synchronous Satellites
6.3 Impacts of Data Degradations on High-Level Products ...................................................148
MODIS/Terra Degradation AVHRR Orbit Drift
Dongdong Wang 6.4 Concluding Remarks.............................................................................................................. 150
University of Maryland References............................................................................................................................................ 150
Acronyms and Definitions traditional observation approaches, one important advantage of
satellite remote sensing is its global mapping capability. Satellite
AOD Aerosol optical depth data have thus become an unprecedented source to study global
AVHRR Advanced very-high-resolution radiometer environmental changes (e.g., Kaufman et al., 2002; Nemani
BRDF Bidirectional reflectance distribution function et al., 2003). Thanks to the continuity of satellite missions, the
ECT Equator crossing time length of remote sensing data record is continuously increasing.
EM Electromagnetic Currently, we have 30+ years archive of the National Oceanic
ETM+ Enhanced thematic mapper plus and Atmospheric Administration (NOAA) advanced very-high-
MODIS Moderate Resolution Imaging Spectroradiometer resolution radiometer (AVHRR) at spatial resolution suitable for
MS Mirror side continental or global studies (Beck etal., 2011) and Landsat TM/
NDVI Normalized difference vegetation index ETM+ data at ecosystem scales (Markham and Helder, 2012).
NIR Near infrared Entering the new century, data records from new-generation
NOAA National Oceanic and Atmospheric Administration remote sensors such as the Satellite Pour lObservation de la
RSB Reflective solar bands Terre (SPOT) VEGETATION are also lengthening. More than
RVS Response versus scan angle one decade of Moderate Resolution Imaging Spectroradiometer
SD Solar diffuser (MODIS) data become available (2000present) and serve as valu-
SDSM Solar diffuser stability monitor able resource for the atmospheric (King et al., 2003), terrestrial
SPOT Satellite pour lObservation de la Terre (Justice etal., 2002), and oceanic (Esaias etal., 1998) research com-
SZA Solar zenith angle munity with well-designed spectral configuration and improved
TIROS Television Infrared Observation Satellite Program radiometric and geometric accuracy (Justice etal., 1998).
TM Thematic mapper With progress in sensor design and manufacturing tech-
TOA Top of atmosphere niques, reliability of radiometric calibration has improved
substantially. Detection of subtle trends in environmen-
6.1 introduction tal variables is extremely sensitive to calibration drifts of
satellite sensors (Datla et al., 2004). Errors in geometric
It has been half a century since the launch of the first-generation coregistration will also affect the analysis of multitemporal
Earth observation satellites, such as former Soviet Unions data and thus detection of changes and trends (Townshend
Sputnik 1 and the United States Television Infrared Observation etal., 1992). Stable, consistent, and reliable time series data
Satellite Program (TIROS)-1 (Tatem etal., 2008). Compared with with extended temporal coverage and accurate geolocation
143
2016 by Taylor & Francis Group, LLC
144 Remotely Sensed Data Characterization, Classification, and Accuracies
information are greatly needed to study interannual variabil- onboard calibration devices as Landsat/TM (Helder et al.,
ity and long-term trends of global environments. 1998) and Terra/MODIS (Wang etal., 2012).
Very-high-level efforts have been devoted to assure the Degradation in satellite signature (i.e., top-of-atmosphere
absolute and relative accuracy of radiative quantities measured [TOA] observations) will eventually translate into errors and bias
by remote sensors. Particularly, various calibration approaches in high-level products through the satellite data processing chain,
(e.g., prelaunch, onboard, vicarious, cross-platform) are employed if the issue of data degradation is not well addressed in the design
to convert sensors signature (e.g., digital number) to accurate of retrieval algorithms. Due to the time-dependence nature of
values of energy emitted and scattered by the Earth system data degradations, their impacts will be especially evident in anal-
(Dinguirard and Slater, 1999). Before launch of satellites, sensors ysis of time series data, such as trend detection and investigation
characteristics of spectral, radiometric, and spatial responses are of interannual variability. In such cases, subtle changes of envi-
measured in labs at various levels to produce prelaunch calibra- ronmental parameters of interest will be obscured by artificial
tion metrics. The actual calibration coefficients hardly remain trends embedded in the time series of degraded data. For instance,
constant because of the impacts from the launching process and analysis of four AVHRR Normalized Difference Vegetation Index
the sensor degradation during the operation in space (Mekler (NDVI) datasets revealed inconsistent trends over Europe, Africa,
and Kaufman, 1995). To monitor sensitivity changes of sensors and the Sahel (Beck et al., 2011). Difference in handling sensor
in radiometric response, many satellite sensors are equipped degradation can be an important factor of such discrepancies.
with onboard calibration units, such as internal lamp, solar dif- Comparison of NDVI trends derived from Aqua/MODIS and
fuser (SD), and blackbody. For sensors without onboard calibra- Terra/MODIS showed contradictory results over boreal North
tors, observations over pseudo-invariant targets are typically America (Wang etal., 2012). Further data analysis suggested this
used to adjust calibration coefficients so that sensor degradation be caused by the degradation issue in Terra/MODIS Collection 5
can be taken into account. Such technique of vicarious calibra- data. Similarly, inconsistent trends were observed in aerosol (Levy
tion is also applied to sensors with onboard calibration devices etal., 2010) and ocean color (Djavidnia etal., 2010) when compar-
as an additional source to verify performance of satellite detec- ing products derived from Aqua/MODIS and Terra/MODIS.
tors and their calibration systems (Wu etal., 2008). In addition to radiometric responses, other characteristics of
Despite the various efforts, it is still a challenge to maintain satellites and sensors may also change with time. For example,
the accuracy and precision of radiometric calibration due to orbits of sun-synchronous satellites may drift and cause shift in
various reasons discussed here. The harsh space environment solar zenith angle (SZA) (Ignatov etal., 2004). This can lead to
where satellite sensors operate makes it extremely hard to keep artifact drifts in time series of some satellite products, similar to
the onboard calibration devices function ideally as designed. effects of sensor degradation. Thus, we here treat orbit drift as
For example, the performance of internal illuminating sources one type of data degradations as well. The issues of sensor degra-
and diffusers may degrade over time as well (Helder et al., dation exist in not only optical sensors (visible and near infrared
1998). Vicarious calibration is independent on degradation of [NIR] channels) but also other remote sensing data such as ther-
onboard instruments, but it has its own uncertainties and lim- mal data (Tonooka etal., 2005). In this chapter, we will mainly
itations. Stable atmospheric and surface conditions are usually focus on the degradation of optical sensors, introduce some
assumed at the calibration sites, although both tend to change. common degradation issues, discuss their impacts on high-level
In addition, undetected cloud or cloud shadow and effects of products, and summarize approaches, methods, and techniques
bidirectional reflectance distribution function (BRDF) are to address these degradation issues.
among other major sources of uncertainty. As a result, satellite
data records are prone to contain time-dependent radiomet- 6.2 common issues of Data Degradation
ric drifts of some levels even after such sophisticated calibra-
tion efforts have been made. Substantial degradations have This section briefly summarizes three types (Table 6.1) of com-
been found to exist in both sensors without onboard calibra- mon degradation issues of satellite data that consist of progres-
tors such as AVHRR (Wu and Zhong, 1994) and those with sive changes in radiometric response and drifts in satellite orbital
TABLe 6 .1 Major Types of Data Degradation and Their Impacts on Satellite Products
Types Summary Impacts Solution Example Reference
Sensor degradation Response of detectors changes with Inaccurate radiance and Using pseudo-invariant targets on the NOAA Molling etal.
without onboard time due to the harsh environment time-dependent ground to adjust calibration AVHRR (2010)
calibration unit where satellites operate. artifacts in time series coefficients (vicarious calibration)
Degradation of Characteristics of onboard Inaccurate radiance and Combining calibration with additional MODIS/ Wu etal.
onboard calibration device themselves time-dependent sources, such as lunar observation Terra (2013)
calibration unit degrade with time. artifacts in time series and pseudo-invariant targets
Orbital drift Orbits of sun-synchronous Gradual shift in SZA and Postcorrection of time series data NOAA Privette etal.
satellites drift and cause changes variations in surface using statistical or BRDF models AVHRR (1995)
in ECT. reflectance due to BRDF
2016 by Taylor & Francis Group, LLC
Satellite Data Degradations and Their Impacts on High-Level Products 145
parameters and two types of radiometric degradations (one with 60
onboard calibration and one without onboard calibration).
55
6.2.1 Radiometric Degradation without 50
Reflectance (%)
onboard calibration Systems
45
Unlike ground-based measuring instruments, spaceborne sensors
are usually physically inaccessible after satellites are launched. 40
It is thus impossible to monitor the stability of their responses
by periodically calibrating them with traditional approaches 35
of laboratory measurement. To achieve reliable performance,
extremely stable materials are typically used to manufacture 30
1984 1985 1986 1987 1988 1989
satellite detectors. Calibration coefficients which convert satel-
Year
lite signature (e.g., digital number) to radiance or other radiative Prelaunch NESDISv2
variables are measured in laboratory before launching. The sen- NESDISv3 ISCCP
CCRS/EODM PALGSFC93
sors without onboard calibration units do not have mechanisms Rao PALGSFC9596
to update the coefficients on orbit. However, the characteristics UWash LTDRv3
PATMOSx
of satellite detectors tend to change since the moment when the
prelaunch coefficients are measured. The radiometric response FIg u r e 6.1 Mean reflectance of July over the Libyan Desert from
of sensors will degrade due to the impacts from the launching NOAA9/AVHRR Channel 1 using various versions of calibration coef-
process and the harsh space environment (Rao and Chen, 1995). ficients. The prelaunch coefficients show a clear decreasing trend, caused
Visible and NIR bands of AVHRR do not have onboard cali- by issue of sensor degradation. Many efforts have been devoted to account
bration systems. Staylor (1990) evaluated the calibration stability for such degradation through vicarious calibration. However, large dis-
of AVHRR visible bands onboard NOAA 6, 7, and 9 satellites crepancies exist among results from different methods. (Adapted from
by monitoring their performance over the Libyan Desert. By Figure 1 of Molling, C.C. etal., Int. J. Remote Sens., 31, 6519, 2010.)
assuming a fixed degradation rate, an exponential degradation
model was used to fit AVHRR time series (Staylor, 1990). The Reflectivity of pseudo-invariant targets is hardly a constant.
annual degradation rates of Channel 1 of NOAA 6, 7, and 9 were Though the selected target is relatively isotropic in terms of its
found to be 0%, 4%, and 6%, respectively (Staylor, 1990). Masonis reflectivity, BRDF effects are a major source of uncertainties.
and Warren (2001) used data of the same bands onboard NOAA Moreover, changes in atmospheric conditions can cause errors,
9, 10, and 11 over Greenland to estimate their degradation rates. especially aerosol from volcano eruption and water vapor effects
They found that the sensor degradation is the reduced sensitiv- for NIR bands. Issues of subpixel or undetected cloud cannot be
ity to electromagnetic (EM) signals. NOAA 11 had an annual negligible either. As a result, the derived calibration coefficients
increased rate of 2.3% in radiometric gain (Masonis and Warren, may contain large errors. Molling etal. (2010) compared AVHRR
2001). Similarly, MODIS is also found to have increased gains in reflectances of a desert region calculated from 10 groups of cali-
Band 1 and 2 (Wu etal., 2013). bration coefficients and found that the difference among them
Due to the existence of radiometric degradation, measure- can be as large as 20% (Figure 6.1).
ments from such sensors cannot be directly used in quantita- Ideally, vicarious calibration is expected to account for all the
tive studies (Gorman and McGregor, 1994; Gutman, 1999) that effects of sensor degradation, so that the satellite data after vicar-
require accurate values of radiance to quantify variables of inter- ious calibration can be free of problems of sensor degradation.
est. For such sensors, in practice, vicarious calibration is usually However, as we mentioned, the vicarious calibration has its dis-
employed to account for the time-variant drifts of radiometric advantages and limitations. An alternative strategy of onboard
response (Dinguirard and Slater, 1999). After launching, calibra- calibration is in great need.
tion coefficients are updated by benchmarking observations over
various types of stable surfaces, such as desert (Rao and Chen,
6.2.2 Radiometric Degradation with
1995), ocean and high cloud (Vermote and Kaufman, 1995), and
onboard calibration Device
snow and ice (Loeb, 1997). In addition, the observation can also
be adjusted by comparing with other well-calibrated satellite Given the previously mentioned problems of vicarious calibration,
images such as MODIS (Heidinger etal., 2002) or airborne data an alternative way is to continuously monitor and correct sensor
(Smith etal., 1988). Besides the calibration of individual bands, degradation on orbit. Actually, many sensors have units designed
variable of interest (e.g., NDVI) can also be directly adjusted by a for onboard calibration, for example, SD for MODIS (Xiong and
similar procedure of vicarious calibration (Los, 1998). Barnes, 2006), internal calibrator for thematic mapper (TM)
Different vicarious calibration practices choose different cali- (Chander and Markham, 2003), and two additional devices (Full
bration sites and use various forms of degradation equations. Aperture Solar Calibrator, Partial Aperture Solar Calibrator) for
Lots of factors contribute to uncertainties in such methods. enhanced thematic mapper plus (ETM+) (Markham etal., 2004).
2016 by Taylor & Francis Group, LLC
146 Remotely Sensed Data Characterization, Classification, and Accuracies
Theoretically, a well-designed onboard calibration system should original strategy, the SD door opens only when onboard calibra-
be able to account for the sensor degradation and correct it by tion is in progress to reduce the exposure of SD to the Sun. Due
using the updated calibration coefficients. to an SD anomaly of Terra/MODIS, the door remains open after
For instance, MODIS is equipped with a sophisticated onboard 2003 (Xiong and Barnes, 2006). Terra/MODIS has thus expe-
calibration system for its visible, NIR, and shortwave infrared bands rienced much worse degradation problems than Aqua/MODIS
(also known as reflective solar bands [RSB]) (Xiong etal., 2010). SD since then. Due to the difference of viewing geometry between
and solar diffuser stability monitor (SDSM) are two key devices of onboard calibration and observing the Earth, stability of SD
this system (Xiong and Barnes, 2006). The radiometric response of BRDF is essential for SDSM to track its degradation (Wu etal.,
the MODIS detector is calibrated by viewing SD, which has known 2013). As a result, bidirectionality of SD reflectance and changes
reflecting characteristics. The degradation of SD is monitored by in response versus scan angle (RVS) are major sources of uncer-
SDSM. It is actually a type of radiometer that determines SD deg- tainties in calibrating MODIS in flight (Xiong et al., 2007).
radation by observing the Sun and SD alternately (Xiong et al., Because of its excessive exposure, the changes in BRDF of SD are
2007). In this process, spectral, angular, and mirror-side (MS) believed to be the major cause of the calibration drift of MODIS/
dependency of reflectance and radiometric response has been well Terra (Wu etal., 2013). This degradation is found to be depen-
taken into account. Moreover, a periodic view of the moon sur- dent on view zenith angle, spectral band, and MS. The shorter
face is used as an additional calibration source to independently wavelength bands have the most series degradation. The stabil-
check the stability of radiometric calibration (Xiong etal., 2010). ity of Terra/MODIS blue band reflectance changed as much as
Normally, such well-designed onboard calibration system is able 7% for nadir observations (Figure 6.2) (Wang etal., 2012). With
to update the calibration coefficients as needed to compensate the the lesson learned, the C6 reprocessing of MODIS/Terra data
degradation of the detector, so that reliable calibration results can will use stable observations of the moon and desert to correct
be generated during the lifetime of the sensor. As a matter of fact, the errors in radiometric calibration (Wenny et al., 2010). The
MODIS/Aqua is proved to achieve a very high level of calibration calibration drift issue of MODIS/Terra is expected to be reduced
stability. A combination of four lines of evidence indicated that substantially through reprocessing (Wu etal., 2013).
RSB reflectance derived from MODIS/Aqua has changed less than Landsat 5/TM is another example of onboard calibration
1% during its decadal operation (Wu etal., 2013). drift. Landsat 5 initially used onboard units to provide scene-
However, the other sensor of twin MODIS, MODIS/Terra, by-scene calibration coefficients. However, the performance of
was found to have a serious issue of calibration drifts in some the internal tungsten lamps of Landsat 5 failed to remain con-
SRB bands (Franz etal., 2008; Wu etal., 2013). According to the stant (Helder etal., 1998). The vicarious calibration and onboard
View zenith angle = 51 View zenith angle = 42 View zenith angle = 32 View zenith angle = 28
1.04 1.04 1.04 1.04
1.02 1.02 1.02 1.02
1 1 1 1
0.98 0.98 0.98 0.98
B1 MS1 B1 MS2
0.96 0.96 0.96 0.96
B2 MS1 B2 MS2
0.94 0.94 B3 MS2 0.94 0.94
B3 MS1
0.92 0.92 0.92 0.92
2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010
View zenith angle = 18 View zenith angle = 11 View zenith angle = 3 View zenith angle = 9
1.04 1.04 1.04 1.04
1.02 1.02 1.02 1.02
1 1 1 1
0.98 0.98 0.98 0.98
0.96 0.96 0.96 0.96
0.94 0.94 0.94 0.94
0.92 0.92 0.92 0.92
2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010
View zenith angle = 17 View zenith angle = 31 View zenith angle = 43 View zenith angle = 52
1.04 1.04 1.04 1.04
1.02 1.02 1.02 1.02
1 1 1 1
0.98 0.98 0.98 0.98
0.96 0.96 0.96 0.96
0.94 0.94 0.94 0.94
0.92 0.92 0.92 0.92
2000 2005 2010 2000 2005 2010 2000 2005 2010 2000 2005 2010
(a)
FIg u r e 6 .2 Degradation of TOA reflectance at three MODIS bands (B1, red; B2, NIR; and B3, blue) for (a) Terra/MODIS. (Continued )
2016 by Taylor & Francis Group, LLC
Satellite Data Degradations and Their Impacts on High-Level Products 147
View zenith angle = 51 View zenith angle = 43 View zenith angle = 32 View zenith angle = 29
1.04 1.04 1.04 1.04
1.02 1.02 1.02 1.02
1 1 1 1
0.98 0.98 0.98 0.98
B1 MS1 B1 MS2
0.96 B2 MS1 0.96 B2 MS2 0.96 0.96
0.94 B3 MS1 0.94 B3 MS2 0.94 0.94
0.92 0.92 0.92 0.92
2002 2004 2006 2008 2010 2002 2004 2006 2008 2010 2002 2004 2006 2008 2010 2002 2004 2006 2008 2010
View zenith angle = 19 View zenith angle = 11 View zenith angle = 2 View zenith angle = 10
1.04 1.04 1.04 1.04
1.02 1.02 1.02 1.02
1 1 1 1
0.98 0.98 0.98 0.98
0.96 0.96 0.96 0.96
0.94 0.94 0.94 0.94
0.92 0.92 0.92 0.92
2002 2004 2006 2008 2010 2002 2004 2006 2008 2010 2002 2004 2006 2008 2010 2002 2004 2006 2008 2010
View zenith angle = 21 View zenith angle = 29 View zenith angle = 41 View zenith angle = 51
1.04 1.04 1.04 1.04
1.02 1.02 1.02 1.02
1 1 1 1
0.98 0.98 0.98 0.98
0.96 0.96 0.96 0.96
0.94 0.94 0.94 0.94
0.92 0.92 0.92 0.92
2002 2004 2006 2008 2010 2002 2004 2006 2008 2010 2002 2004 2006 2008 2010 2002 2004 2006 2008 2010
(b)
FIg u r e 6.2 (c ontinued ) Degradation of TOA reflectance at three MODIS bands (B1, red; B2, NIR; and B3, blue) for (b) Aqua/MODIS. The degrada-
tion rates are dependent on view zenith angle, spectral band, and MS. (Adapted from Figure S1 of Wang, D.D. etal., Remote Sens. Environ., 119, 55, 2012.)
calibration suggested that there existed trends in sensor response several hours during the lifetime of NOAA satellites (Figure6.3).
(Thome et al., 1997). Scene-by-scene calibration coefficients From 1988 to 1994, NOAA 11 is the primary data source of
derived from the onboard calibrator are not reliable anymore AVHRR NDVI long-term data archive, because of the failure
(Chander and Markham, 2003). As a result, a new calibration of its replacement satellite (Gutman, 1999). ECT of NOAA11
method based on monitoring desert and cross-calibration with became 3h later after its 6years use.
Landsat 7/ETM+ is developed (Chander etal., 2007). Compared The change in crossing time will translate into shift of
to Landsat 5/TM, ETM+ is rather stable. Annual degradation SZA. The angular changes are dependent on latitudes. Higher
rates of visible and NIR bands of ETM+ are smaller than 0.4%
(Markham etal., 2004). Equator crossing times for NOAA 09 10 11 12
22
6.2.3 orbit Drift of Sun-Synchronous Satellites NOAA 10 NOAA 12
20
Time (local solar time)
Sun-synchronous is a feature of many polar-orbiting Earth
observation missions. Such satellites will always cross the equa- 18
tor at the same local time to minimize effects of diurnal changes
in variables of interest. Besides, this also assures that the obser-
vations have relatively similar illumination conditions. Due to 16
NOAA 09
BRDF effects, fixed illumination geometry will make reflectance
comparable with each other. However, the designated orbits NOAA 11
14
will change during the operation due to various perturbations
(Ignatov et al., 2004). One straightforward way to address the
12
issue is to accordingly adjust the satellite orbit from time to time. 1/86 1/87 1/88 1/89 1/90 1/91 1/92 1/93 1/94
However, it will be a problem for satellites without the capability Date
of adjusting flight on orbit, for example, NOAA heritage satel-
lites. For NOAA morning satellites, equator crossing time (ECT) FIg ur e 6.3 Changes in ECT for four NOAA satellites. Orbits are
becomes gradually earlier. For afternoon satellites, the overpass- predicted from the historical ephemeris data. (Adapted from Figure 1
ing time moves toward later hours. ECT can change as long as of Privette, J.L. etal., Remote Sens. Environ., 53, 164, 1995.)
2016 by Taylor & Francis Group, LLC
148 Remotely Sensed Data Characterization, Classification, and Accuracies
latitudes see greater seasonal variability but relative smaller shift Terra/MODIS and Aqua/MODIS. Sensor degradation of Terra/
of SZA, whereas equatorial area experiences substantial change MODIS is likely to be a major of such difference. Besides, some
of SZA (Privette etal., 1995). The progressive orbit drift will add products such as NDVI may also be impacted by the degra-
a layer of noise into the temporal signals for some applications. dation issue of blue bands through the processing chain of
The impacts mainly are in two aspects. ECT changes will affect MODIS products, although they do not directly use blue bands
features with diurnal variations, for example, cloud (Devasthale as inputs.
etal., 2012). The other issue is the changes in viewing geometry. In a simulation study, Wang etal. (2012) analyzed how degra-
Correction of atmospheric effects, for example, scattering of mol- dation of Terra/MODIS blue band affects its NDVI product and
ecule and aerosol and absorption of water vapor, is dependent on how the embedded errors in NDVI impact data analysis such
the length of path and viewing geometry. In addition, surface as trend detection. As shown in Figure 6.2, TOA reflectance of
reflectance is also dependent on viewing geometry because of Terra/MODIS red and NIR bands does not have serious issue of
surface BRDF effects. In Section 6.3.2, we use AVHRR data as degradation. However, the MODIS NDVI algorithm needs sur-
example to illustrate how orbit shifts may affect NDVI products. face reflectance of red and NIR bands as inputs. MODIS uses
the dense dark vegetation algorithm to atmospherically cor-
rect reflectance of red and NIR bands (Kaufman et al., 1997),
6.3 impacts of Data Degradations which heavily depends on aerosol optical depth (AOD) informa-
on High-Level Products tion derived from the blue band. Using the actual degradation
rates of blue, red, and NIR bands at TOA as inputs, the induced
6.3.1 MoDiS/terra Degradation
errors in AOD and NDVI are simulated for various combina-
As we learned in Section 6.2.2, SRB of MODIS/Terra has expe- tions of view zenith angle, aerosol loadings, and land surface
rienced noticeable issues of calibration drifts, mainly caused types (Figure 6.4). The simulated results are consistent with the
by changes of bidirectional reflectance of SD. The actual deg- observed difference between Aqua/MODIS and Terra/MODIS
radation rates are dependent on spectral bands, view zenith corresponding products.
angle, and MS. Generally speaking, bands of shorter wave- Although the annual degradation rate in the high-level
length and near nadir observations have the worst problem of NDVI products is only with the magnitude of 10 3, it has great
degradation. Blue bands are important inputs to algorithms to implication for the study of global vegetation change with the
generate products of ocean color and aerosol loadings. Large decadal time series of MODIS NDVI, because this is close to
discrepancies are observed for the two products derived from natural changing rates of vegetation without disturbance such
Tundra Forest
0.66 0.06 0.84 0.06
Low AOD
0.64 0.04 0.04
NDVI
AOD
0.82
0.62 0.02 0.02
0.6 0 0.8 0
0.66 0.25 0.84 0.25
Moderate AOD
0.2 0.2
0.64
NDVI
AOD
0.15 0.82 0.15
0.62
0.1 0.1
0.6 0.05 0.8 0.05
0.66 0.84
0.5 0.5
High AOD
0.64
0.45
NDVI
0.45
AOD
0.82
0.62 0.4 0.4
0.6 0.35 0.8 0.35
2000 2005 2010 2000 2005 2010
Year Year
FIgu r e 6.4 Temporal changes of NDVI (solid lines) and AOD (dashed lines) from 2000 to 2010, simulated with the degradation of Terra/
MODIS as inputs. Two biomes (tundra [left] and forest [right]) and three levels of aerosol loadings (0.05, 0.20, and 0.50) are used in simulation.
(Adapted from Wang, D.D. etal., Remote Sens. Environ., 119, 55, 2012.)
2016 by Taylor & Francis Group, LLC
Satellite Data Degradations and Their Impacts on High-Level Products 149
106
2.5 Terra Aqua
Terra
2
Aqua Positive trends 8.1% 13.1%
1.5
1
Negative trends 17.4% 6.7%
0.5
Terra 0
0.06 0.03 0 0.03 0.06
Aqua
0.05 0.04 0.03 0.02 0.01 0 0.01 0.02 0.03 0.04 0.05
FIg ur e 6.5 Boreal North America maps of NDVI trends between 2002 and 2010 calculated from the two actual MODIS products (Terra, left,
and Aqua, right). White pixels are those recently disturbed by fire, wood harvest, etc. Black is nontundra or forest area. (Adapted from Wang, D.D.
etal., Remote Sens. Environ., 119, 55, 2012.)
as fire, insect burst, and logging. Under such circumstances, drops in NDVI are most significant in boreal winter and the
the Monte Carlo simulation suggested that it is extremely equator. After 5years on orbit, the decrease of NDVI caused
difficult to reliably detect trends of NDVI from the decadal by orbit drift can be as great as 10% at the equator (Figure 6.6)
MODIS data. Figure 6.5 shows the boreal North America maps (Privette etal., 1995).
of NDVI trends between 2002 and 2010 derived from Aqua/ To remove effects of orbit drift, various kinds of algorithms
MODIS and Terra/MODIS, respectively. Historical distur- have been developed. Some are based on statistical analysis to
bance datasets were used to exclude pixels disturbed by vari- isolate signals caused by changes in SZA (Sobrino etal., 2008).
ous natural and anthropogenic factors. The two maps display Some utilize prior knowledge of vegetation BRDF to estimate
similar patterns over some area where the change rates are NDVI difference induced by angular variations (Bacour etal.,
large. However, there exists a systematic difference between 2006). Los et al. (2005) developed a BRDF-based correction
the trends derived from the two datasets. As a result, Aqua approach without assumption of biome-dependent BRDF
data suggest two thirds of significant trends are positive trends shapes. Taking a first-order approximation, the NDVI equa-
where Terra data show opposite statistics (Wang etal., 2012). tion in terms of RossLi kernels (K L and K R) is reduced to (Los
Due to the potential impact of Terra/MODIS C5 data on trend etal., 2005)
analysis, it is suggested that the user should use Aqua/MODIS
data in analyzing interannual variability before reprocessed NDVI (v , s , ) = ki + k g K L + kv K R
MODIS data are available.
Even in this reduced form, it is difficult to estimate the kernel
parameters (ki, kg, and kv) because of the limited temporal sam-
6.3.2 AVHRR orbit Drift pling of AVHRR data. The authors further divided NDVI vari-
Orbit drift will not always be a problem for data users. If the ability into two parts: those caused by phenology and those by
users models can explicitly handle data variations in view- BRDF effects. Using both simulated and actual data, the authors
ing geometry, their results will not be affected by the effects demonstrated that the effects of BRDF on NDVI can be reduced
of orbit drift. For example, in calculation of NDVI, if BRDF- by 50%85% through their method.
adjusted reflectance can be used, orbit drift will not necessar- Impacts of NOAA satellite orbit drift are usually coupled with
ily lead to artificial changes in NDVI time series. However, other issues such as sensor degradation of AVHRR (Staylor,
due to the sparse temporal sampling of AVHRR, it is hard to 1990) and insufficient atmospheric correction (Nagol et al.,
model BRDF from AVHRR observations. The NDVI prod- 2009). So data quality of AVHRR time series is heavily depen-
ucts typically directly use directional reflectance at the view- dent on approaches of data processing and analysis. Large differ-
ing geometry as inputs. As a ratio of two bands, NDVI is less ences in both absolute value and temporal change exist among
affected by the variations in viewing geometry. Nevertheless, analysis using different versions of AVHRR NDVI datasets (Beck
orbit drift still produces spurious variations on top of vege- etal., 2011). Since the aim of data correction is to isolate vegeta-
tation dynamics (Latifovic et al., 2012). The impacts of orbit tion variability from noise and artifacts in NDVI time series, an
drift on NDVI are dependent on latitude, season, and land alternative way is to treat all the artifacts originating from data
cover type. Privette etal. (1995) simulated the changes of TOA degradation together and correct them all at once (Jiang etal.,
NDVI caused by the orbit drift of the NOAA 11 satellite. The 2008; Latifovic etal., 2012).
2016 by Taylor & Francis Group, LLC
150 Remotely Sensed Data Characterization, Classification, and Accuracies
Africa (ON 15E) variables. For sensors without onboard calibration devices,
0.8 0.8 vicarious calibration is a key step to account for such artifacts
in the remote sensing data. However, lots of environmen-
tal factors make it a challenge to eliminate the effects of data
0.6 0.6
degradation only through vicarious calibration. So onboard
calibration systems are designed to address such problems.
0.4 0.4 However, reliability of onboard calibration is also affected by
its own degradation. The onboard calibration system needs
collaboration with other calibration methods. A combination
0.2 0.2
(a) of onboard and vicarious calibration and utilization of multi-
ple independent data sources, such as lunar view (Barnes etal.,
FIFE (39N 96.5W) 2001; Cao etal., 2009), is a promising pathway to improve reli-
0.8 0.8 ability of long-term data archive.
Any measurements have uncertainties, including remote
Projected NDM
0.6 0.6
sensing data. Users should always keep uncertainties of sat-
ellite data in mind when using them in quantitative studies.
Inaccurate absolute calibration leads to bias or errors in satel-
0.4 0.4 lite retrievals. Degradation of satellite data causes time-variant
drift in long-term time series of satellite products. Caution
0.2 0.2 should be exercised when drawing conclusion regarding long-
(b) term trends from data with possible artificial drifts. Meanwhile,
Boreas (55N 102W)
efforts to evaluate accuracy of absolute calibration as well as
long-term stability of calibration are greatly needed for climate
0.8 0.8 change study.
0.6 0.6 References
Bacour, C., Breon, F.M., and Maignan, F. (2006). Normalization of
0.4 0.4
the directional effects in NOAA-AVHRR reflectance mea-
surements for an improved monitoring of vegetation cycles.
0.2 0.2 Remote Sensing of Environment, 102, 402413.
Barnes, R.A., Eplee, R.E., Schmidt, G.M., Patt, F.S., and McClain,
1/89 1/90 1/91 1/92 1/93 1/94
C.R. (2001). Calibration of SeaWiFS. I. Direct techniques.
(c) Date Applied Optics, 40, 66826700.
Beck, H.E., McVicar, T.R., van Dijk, A.I.J.M., Schellekens, J.,
FIgu r e 6.6 Impacts of orbit drift on TOA NDVI time series of deJeu, R.A.M., and Bruijnzeel, L.A. (2011). Global evalu-
NOAA11 AVHRR. Dotted lines are top of canopy NDVI with fixed illu- ation of four AVHRR-NDVI data sets: Intercomparison
mination. Dashed lines are TOA NDVI without orbit drift. Solid lines and assessment against Landsat imagery. Remote Sensing of
are simulated TOA NDVI with NOAA11 orbit drift as input. (Adapted
Environment, 115, 25472563.
from Figure 5 of Privette, J.L. etal., Remote Sens. Environ., 53, 164, 1995.)
Cao, C.Y., Vermote, E., and Xiong, X.X. (2009). Using AVHRR lunar
observations for NDVI long-term climate change detection.
6.4 concluding Remarks Journal of Geophysical Research: Atmospheres, 114, D20105.
Chander, G., and Markham, B. (2003). Revised Landsat-5 TM
Analysis of long-term satellite data record has become an impor- radiometric calibration procedures and postcalibration
tant tool to investigate how the Earth system changes over time dynamic ranges. IEEE Transactions on Geoscience and
and how its components interact with each other. Detection of Remote Sensing, 41, 26742677.
subtle changes of global environment prefers consistent and stable Chander, G., Markham, B.L., and Barsi, J.A. (2007). Revised
time series data without artifact drifts. Unfortunately, spaceborne Landsat-5 thematic mapper radiometric calibration. IEEE
remote sensing data suffer from various issues of degradation dur- Geoscience and Remote Sensing Letters, 4, 490494.
ing satellite operation in the harsh space environment. This chapter Datla, R., Emery, B., Ohring, G., Spencer, R., and Wielicki, B.
summarizes some common phenomena of data degradations and (2004). Stability and accuracy requirements for satellite
discusses their impacts on high-level satellite products. remote sensing instrumentation for global climate change
Changes in radiometric response of spaceborne sensors monitoring. In S.A. Morain, and A.M. Budge (Eds.), Post-
cause time-variant drifts in TOA radiance, which translate Launch Calibration of Satellite Sensors. A.A. Balkema
into artificial trends in high-level products of environmental Publishers, Leiden, the Netherlands.
2016 by Taylor & Francis Group, LLC
Satellite Data Degradations and Their Impacts on High-Level Products 151
Devasthale, A., Karlsson, K.G., Quaas, J., and Grassl, H. (2012). Kaufman, Y.J., Tanre, D., and Boucher, O. (2002). A satellite view
Correcting orbital drift signal in the time series of AVHRR of aerosols in the climate system. Nature, 419, 215223.
derived convective cloud fraction using rotated empirical Kaufman, Y.J., Tanre, D., Remer, L.A., Vermote, E.F., Chu, A.,
orthogonal function. Atmospheric Measurement Techniques, and Holben, B.N. (1997). Operational remote sensing of
5, 267273. tropospheric aerosol over land from EOS moderate reso-
Dinguirard, M. and Slater, P.N. (1999). Calibration of space- lution imaging spectroradiometer. Journal of Geophysical
multispectral imaging sensors: A review. Remote Sensing of Research: Atmospheres, 102, 1705117067.
Environment, 68, 194205. King, M.D., Menzel, W.P., Kaufman, Y.J., Tanre, D., Gao, B.C.,
Djavidnia, S., Melin, F., and Hoepffner, N. (2010). Comparison of Platnick, S., Ackerman, S.A., Remer, L.A., Pincus, R., and
global ocean colour data records. Ocean Science, 6, 6176. Hubanks, P.A. (2003). Cloud and aerosol properties, pre-
Esaias, W.E., Abbott, M.R., Barton, I., Brown, O.B., Campbell, cipitable water, and profiles of temperature and water vapor
J.W., Carder, K.L., Clark, D.K. et al. (1998). An overview from MODIS. IEEE Transactions on Geoscience and Remote
of MODIS capabilities for ocean science observations. Sensing, 41, 442458.
IEEE Transactions on Geoscience and Remote Sensing, 36, Latifovic, R., Pouliot, D., and Dillabaugh, C. (2012). Identification
12501265. and correction of systematic error in NOAA AVHRR long-
Franz, B.A., Kwiatkowska, E.J., Meister, G., and McClain, C.R. term satellite data record. Remote Sensing of Environment,
(2008). Moderate Resolution imaging spectroradiometer 127, 8497.
on Terra: Limitations for ocean color applications. Journal Levy, R.C., Remer, L.A., Kleidman, R.G., Mattoo, S., Ichoku, C.,
of Applied Remote Sensing, 2, 117. Kahn, R., and Eck, T.F. (2010). Global evaluation of the
Gorman, A.J. and McGregor, J. (1994). Some considerations for Collection 5 MODIS dark-target aerosol products over
using AVHRR data in climatological studies 2. Instrument land. Atmospheric Chemistry and Physics, 10, 1039910420.
performance. International Journal of Remote Sensing, 15, Loeb, N.G. (1997). In-flight calibration of NOAA AVHRR vis-
549565. ible and near-IR bands over Greenland and Antarctica.
Gutman, G.G. (1999). On the use of long-term global data of International Journal of Remote Sensing, 18, 477490.
land reflectances and vegetation indices derived from Los, S.O. (1998). Estimation of the ratio of sensor degradation
the advanced very high resolution radiometer. Journal of between NOAA AVHRR channels 1 and 2 from monthly
Geophysical Research: Atmospheres, 104, 62416255. NDVI composites. IEEE Transactions on Geoscience and
Heidinger, A.K., Cao, C.Y., and Sullivan, J.T. (2002). Using mod- Remote Sensing, 36, 206213.
erate resolution imaging spectrometer (MODIS) to cali- Los, S.O., North, P.R.J., Grey, W.M.F., and Barnsley, M.J. (2005).
brate advanced very high resolution radiometer reflectance A method to convert AVHRR Normalized Difference
channels. Journal of Geophysical Research: Atmospheres, Vegetation Index time series to a standard viewing and illumi-
107(D23), 4702. doi:10.1029/2001JD002035. nation geometry. Remote Sensing of Environment, 99, 400411.
Helder, D., Boucyk, W., and Morfitt, R. (1998). Absolute calibra- Markham, B.L. and Helder, D.L. (2012). Forty-year calibrated
tion of the Landsat Thematic Mapper using the internal cal- record of earth-reflected radiance from Landsat: A review.
ibrator. In 1998 IEEE International Geoscience and Remote Remote Sensing of Environment, 122, 3040.
Sensing Symposium Proceedings, Seattle, WA, July 610, Markham, B.L., Thome, K.J., Barsi, J.A., Kaita, E., Helder, D.L.,
pp.27162718. Barker, J.L., and Scaramuzza, P.L. (2004). Landsat-7 ETM+
Ignatov, A., Laszlo, I., Harrod, E.D., Kidwell, K.B., and Goodrum, on-orbit reflective-band radiometric stability and absolute
G.P. (2004). Equator crossing times for NOAA, ERS and calibration. IEEE Transactions on Geoscience and Remote
EOS sun-synchronous satellites. International Journal of Sensing, 42, 28102820.
Remote Sensing, 25, 52555266. Masonis, S.J. and Warren, S.G. (2001). Gain of the AVHRR vis-
Jiang, L., Tarpley, J.D., Mitchell, K.E., Zhou, S., Kogan, F.N., and ible channel as tracked using bidirectional reflectance of
Guo, W. (2008). Adjusting for long-term anomalous trends in Antarctic and Greenland snow. International Journal of
NOAAs global vegetation index data sets. IEEE Transactions Remote Sensing, 22, 14951520.
on Geoscience and Remote Sensing, 46, 409422. Mekler, Y., and Kaufman, Y.J. (1995). Possible causes of calibra-
Justice, C.O., Townshend, J.R.G., Vermote, E.F., Masuoka, E., tion degradation of the advanced very high-resolution radi-
Wolfe, R.E., Saleous, N., Roy, D.P., and Morisette, J.T. (2002). ometer visible and near-infrared channels. Applied Optics,
An overview of MODIS land data processing and product 34, 10591062.
status. Remote Sensing of Environment, 83, 315. Molling, C.C., Heidinger, A.K., Straka, W.C., and Wu, X.Q. (2010).
Justice, C.O., Vermote, E., Townshend, J.R.G., Defries, R., Roy, Calibrations for AVHRR channels 1 and 2: Review and path
D.P., Hall, D.K., Salomonson, V.V. etal. (1998). The mod- towards consensus. International Journal of Remote Sensing,
erate resolution imaging spectroradiometer (MODIS): 31, 65196540.
Land remote sensing for global change research. IEEE Nagol, J.R., Vermote, E.F., and Prince, S.D. (2009). Effects of atmo-
Transactions on Geoscience and Remote Sensing, 36, spheric variation on AVHRR NDVI data. Remote Sensing of
12281249. Environment, 113, 392397.
2016 by Taylor & Francis Group, LLC
152 Remotely Sensed Data Characterization, Classification, and Accuracies
Nemani, R.R., Keeling, C.D., Hashimoto, H., Jolly, W.M., Townshend, J.R.G., Justice, C.O., Gurney, C., and McManus, J. (1992).
Piper, S.C., Tucker, C.J., Myneni, R.B., and Running, S.W. The impact of misregistration on change detection. IEEE
(2003). Climate-driven increases in global terrestrial net Transactions on Geoscience and Remote Sensing, 30, 10541060.
primary production from 1982 to 1999. Science, 300, Vermote, E., and Kaufman, Y.J. (1995). Absolute calibration of
15601563. AVHRR visible and near-infrared channels using ocean and
Privette, J.L., Fowler, C., Wick, G.A., Baldwin, D., and Emery, cloud views. International Journal of Remote Sensing, 16,
W.J. (1995). Effects of orbital drift on advanced very high- 23172340.
resolution radiometer productsNormalized difference Wang, D.D., Morton, D., Masek, J., Wu, A.S., Nagol, J., Xiong,
vegetation index and sea surface temperature. Remote X.X., Levy, R., Vermote, E., and Wolfe, R. (2012). Impact
Sensing of Environment, 53, 164171. of sensor degradation on the MODIS NDVI time series.
Rao, C.R.N., and Chen, J. (1995). Intersatellite calibration linkages Remote Sensing of Environment, 119, 5561.
for the visible and near-infrared channels of the advanced Wenny, B.N., Sun, J., Xiong, X., Wu, A., Chen, H., Angal, A., Choi,T.
very high-resolution radiometer on the NOAA-7, NOAA-9, et al. (2010). MODIS calibration algorithm improvements
and NOAA-11 spacecraft. International Journal of Remote developed for Collection 6 Level-1B. In Proceedings. SPIE
Sensing, 16, 19311942. 7807, Earth Observing Systems XV, 78071F (August 28, 2010).
Smith, G.R., Levin, R.H., Abel, P., and Jacobowitz, H. (1988). Wu, A., Angal, A., Xiong, X., and Cao, C. (2008). Monitoring
Calibration of the solar channels of the NOAA-9 AVHRR MODIS calibration stability of visible and near-IR bands
using high altitude aircraft measurements. Journal of from observed top-of-atmosphere BRDF-normalized reflec-
Atmospheric and Oceanic Technology, 5, 631639. tances over Libyan Desert and Antarctic surfaces. Proceedings
Sobrino, J.A., Julien, Y., Atitar, M., and Nerry, F. (2008). NOAA- of SPIE 7081, 708113.
AVHRR orbital drift correction from solar zenithal angle Wu, A. and Zhong, Q. (1994). A method for determining the sen-
data. IEEE Transactions on Geoscience and Remote Sensing, sor degradation rates of NOAA AVHRR channels 1 and 2.
46, 40144019. Journal of Applied Meteorology, 33, 118122.
Staylor, W.F. (1990). Degradation rates of the AVHRR visible Wu, A.S., Xiong, X.X., Doelling, D.R., Morstad, D., Angal, A., and
channel for the NOAA-6, NOAA-7, and NOAA-9 space- Bhatt, R. (2013). Characterization of Terra and aqua MODIS
craft. Journal of Atmospheric and Oceanic Technology, 7, VIS, NIR, and SWIR spectral bands calibration stability. IEEE
411423. Transactions on Geoscience and Remote Sensing, 51, 43304338.
Tatem, A.J., Goetz, S.J., and Hay, S.I. (2008). Fifty years of earth- Xiong, X., Sun, J., Barnes, W., Salomonson, V., Esposito, J., Erives,H.,
observation satellitesViews from space have led to count- and Guenther, B. (2007). Multiyear on-orbit calibration and
less advances on the ground in both scientific knowledge performance of Terra MODIS reflective solar bands. IEEE
and daily life. American Scientist, 96, 390398. Transactions on Geoscience and Remote Sensing, 45, 879889.
Thome, K., Markham, B., Barker, J., Slater, P., and Biggar, S. (1997). Xiong, X.X., and Barnes, W. (2006). An overview of MODIS
Radiometric calibration of Landsat. Photogrammetric radiometric calibration and characterization. Advances in
Engineering and Remote Sensing, 63, 853858. Atmospheric Sciences, 23, 6979.
Tonooka, H., Palluconi, F.D., Hook, S.J., and Matsunaga, T. (2005). Xiong, X.X., Sun, J.Q., Xie, X.B., Barnes, W.L., and Salomonson,
Vicarious calibration of ASTER thermal infrared bands. V.V. (2010). On-orbit calibration and performance of
IEEE Transactions on Geoscience and Remote Sensing, 43, aqua MODIS reflective solar bands. IEEE Transactions on
27332746. Geoscience and Remote Sensing, 48, 535546.
2016 by Taylor & Francis Group, LLC
IV
Vegetation Index
Standardization and
Cross-Calibration
of Data from
Multiple Sensors
7 Inter- and Intrasensor Spectral Compatibility and Calibration of the e nhanced Vegetation
Indices Tomoaki Miura, Kenta Obata, Javzandulam T. Azuma, Alfredo Huete, and Hiroki Yoshioka ........................ 155
Acronyms and Definitions Introduction EVIs Multisensor Applications of EVI and EVI2 Multisensor
Compatibility ofEVI and EVI2: A Review Atmospheric Impact on Inter- and Intrasensor Spectral Compatibility of EVI
and EVI2 Intersensor Calibration of EVI and EVI2 Discussions and Future Directions Appendix 7.A Inter- and
Intrasensor Spectral Relationships for EVI and EVI2 References
8 Toward Standardization of Vegetation Indices Michael D. Steven, Timothy J. Malthus, and Frdric Baret........... 175
Acronyms and Definitions Introduction: Vegetation Indices and Their Uses The Need for Standards Sources
of Variation in Vegetation Indices Vegetation Index Intercalibration Approaches Vegetation Index
Standardization Discussion Conclusions Reference
153
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
7
Inter- and Intrasensor Spectral
Compatibility and Calibration of
the Enhanced Vegetation Indices
Tomoaki Miura
University of Hawaii at Manoa Acronyms and Definitions .................................................................................................................155
7.1 Introduction .............................................................................................................................155
Kenta Obata 7.2 EVIs........................................................................................................................................... 156
National Institute of Advanced
7.3 Multisensor Applications of EVI and EVI2 ........................................................................157
Industrial Science and Technology
7.4 Multisensor Compatibility of EVI and EVI2: A Review ...................................................157
Javzandulam T. Azuma 7.5 Atmospheric Impact on Inter- and Intrasensor Spectral Compatibility of EVI
University of Hawaii at Manoa and EVI2 ...................................................................................................................................160
Materials and Methods Results
Alfredo Huete 7.6 Intersensor Calibration of EVI and EVI2 ............................................................................166
University of Technology, Sydney
7.7 Discussions and Future Directions .......................................................................................169
Hiroki Yoshioka Appendix 7.A Inter- and Intrasensor Spectral Relationships for EVI and EVI2........................170
Aichi Prefectural University References.............................................................................................................................................171
AcronymsandDefinitions NIR Near-infrared
NOAA National Oceanic and Atmospheric Administration
AERONET Aerosol Robotic Network NPP National Polar-orbiting Partnership
AOT Aerosol optical thickness OLI Operational Land Imager
ARVI Atmospherically resistant vegetation index PAC Partial atmosphere correction
ASTER Advanced Spaceborne Thermal Emission and RMSD Root mean square difference
Reflection Radiometer SAVI Soil-Adjusted Vegetation Index
AVHRR Advanced Very High Resolution Radiometer SeaWiFS Sea-Viewing Wide Field-of-View Sensor
CAI Cloud and aerosol imager SGLI Second-generation Global Imager
EO-1 Earth Observing-1 SPOT Systme Pour lObservation de la Terre
EOS Earth Observing System sRMPD Systematic square root of mean product difference
ESA European Space Agency TM Thematic Mapper
ETM+ Enhanced Thematic Mapper Plus TOA Top of the atmosphere
EVI Enhanced Vegetation Index TOC Top of canopy
EVI2 Two-band Enhanced Vegetation Index (without a uRMPD Unsystematic square root of mean product difference
blue band) VGT VEGETATION
GLI Global imager VIIRS Visible Infrared Imaging Radiometer Suite
GMFR Geometric mean functional regression
GPP Gross primary productivity 7.1 introduction
HRVIR Haute Rsolution dans le Visible et lInfra-Rouge
LAI Leaf Area Index The Enhanced Vegetation Index (EVI), an index developed for
MERIS Medium-Resolution Imaging Spectrometer the National Aeronautics and Space Administration (NASA)
MISR Multi-angle Imaging SpectroRadiometer Earth Observing System (EOS) Moderate Resolution Imaging
MODIS Moderate Resolution Imaging Spectroradiometer Spectroradiometer (MODIS) mission (Huete et al., 2002), has
NASA National Aeronautics and Space Administration been shown useful across a wide range of terrestrial vegetation
NDVI Normalized Difference Vegetation Index studies. These include climatevegetation interactions (Saleska
155
2016 by Taylor & Francis Group, LLC
156 Remotely Sensed Data Characterization, Classification, and Accuracies
etal., 2007; Ponce Campos etal., 2013), gross primary produc- in high biomass region, which complements the conventional,
tivity (GPP) (Sims etal., 2008; Chen etal., 2011b; Guanter etal., Normalized Difference Vegetation Index (NDVI) (Huete etal.,
2014), drought impact assessment (Song etal., 2013), land cover 2002). The EVI (dimensionless) takes the form
classification (Friedl et al., 2010), and phenology (Zhang et al.,
2006). The EVI has also been derived from sensors other than
NIR red
MODIS and used in characterizing terrestrial vegetation condi- EVI = G (7.1)
tions and dynamics. These include Systme Pour lObservation de NIR + C1 red C 2 blue + L
la Terre (SPOT) VEGETATION (VGT) for GPP estimation (e.g.,
Xiao etal., 2003, 2004), Landsat Thematic Mapper (TM) for flo- where
ristic diversity assessment (Cabacinha and de Castro, 2009), and L (reflectance unit, dimensionless) is the canopy background
Landsat Enhanced Thematic Mapper Plus (ETM+) and IKONOS brightness adjustment factor that came from the Soil-
for Leaf Area Index (LAI) estimation (Soudani etal., 2006). Adjusted Vegetation Index (SAVI) (Huete, 1988)
Recently, several new satellite sensors have successfully been C1 (dimensionless) and C2 (dimensionless) are the coef-
launched and placed in orbit as data continuity missions to ficients of the aerosol resistance term adapted from the
existing ones, all of which have spectral bands suitable for the Atmospherically Resistant Vegetation Index (ARVI)
EVI. Suomi National Polar-orbiting Partnership (NPP) Visible (Kaufman and Tanr, 1992)
Infrared Imaging Radiometer Suite (VIIRS) is a moderate- G (dimensionless) is the gain factor that adjusts the EVI
resolution sensor to continue the data streams from the National dynamic range to a comparable one to that of the NDVI
Oceanic and Atmospheric Administration (NOAA) Advanced
Very High Resolution Radiometer (AVHRR) sensor series and The input reflectances in Equation 7.1 need to be corrected for
from EOS MODIS (Changyong et al., 2014). PROBA-V is a a partial atmosphere (i.e., molecular scattering and gaseous
European Space Agency (ESA)-owned satellite mission launched absorption effects) or the total atmospheric effects including
to continue the moderate-resolution data stream of SPOT VGT aerosols. Miura et al. (2001) demonstrated the effectiveness
(Sterckx etal., 2013). Landsat 8 Operational Land Imager (OLI) of the EVI in reducing residual aerosol contaminations in the
provides continuation of Landsat type, medium spatial resolu- total-atmosphere corrected reflectances that arise due to the
tion data with a greater number of spectral bands and higher highly spatially and temporally variable nature of aerosol load-
radiometric capabilities (Irons et al., 2012). In particular, the ings and properties, and uncertainties in the aerosol loading
EVI is included as one of the VIIRS standard operational prod- estimations. The coefficients adopted in the MODIS EVI algo-
ucts (Vargas etal., 2013). rithm are L = 1, C1 = 6, C2 = 7.5, and G = 2.5 (Huete etal., 1997),
It is thus of great importance to develop a good understanding which can be used for both of the aforementioned input reflec-
of EVI spectral compatibilities across sensors and also of signifi- tance types.
cant interest to investigate whether the EVI data record begun While many studies found that the EVI was advantageous
with EOS MODIS can be temporally and spatially extended with in vegetation productivity assessments and characterization of
other satellite sensors for improved monitoring capabilities and vegetation biophysical properties (e.g., an improved surrogate
climate science studies. This chapter focuses on the EVI and measure of GPP; Rahman etal., 2005), several potential issues
discusses intersensor spectral compatibility of the EVI. We also have been reported for the EVI. First, by its very design, the EVI
discuss intersensor spectral compatibility of a two-band version is limited to a sensor system with a blue band in addition to a red
of the EVI without a blue band (EVI2) (Jiang etal., 2008) and and NIR bands (Equation 7.1). The EVI is not applicable to, for
its intrasensor compatibility with the EVI. Specific objectives of example, SPOT Haute Rsolution dans le Visible et lInfra-Rouge
this chapter are to (HRVIR), EOS Advanced Spaceborne Thermal Emission and
Reflection Radiometer (ASTER), or AVHRR. Second, the EVI
1. Present a comprehensive review of inter- and intrasensor
algorithm can produce faulty values for snow-/ice-contaminated
spectral compatibility of the EVI and EVI2
observations that have higher blue and red reflectances than NIR
2. Evaluate the atmospheric impact on spectral compatibil-
reflectance (Huete etal., 2002; Didan and Huete, 2006; Vargas
ity of the EVI and EVI2 across sensors
et al., 2013). Finally, Fensholt et al. (2006) suggested that the
3. Discuss cross-sensor calibration methodologies for EVI
consistency of EVI values across different sensors may be more
and EVI2 spectral compatibility
problematic than that of the NDVI due to variable and more dif-
ficult atmospheric correction schemes of the blue reflectance as
7.2 eVis described later (Section 7.4).
In order to address the aforementioned issues, a two-band
The EVI is a three-band index, requiring reflectances of the version of the EVI without a blue band, or EVI2, was developed,
near-infrared (NIR) (NIR), red (red), and blue (blue) bands. The which is functionally equivalent to the EVI although slightly
index was designed to optimize the vegetation signal through a more prone to aerosol noise (Jiang etal., 2008). Jiang etal. (2008)
decoupling of the canopy background signal, with a reduction noted that the blue band in the EVI is rather aimed at reduc-
in atmospheric aerosol influences and with improved sensitivity ing noise and uncertainties associated with highly variable
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 157
atmospheric aerosols than at providing additional biophysical were made in the Eastern Arc Mountains and their catchment
information on vegetation. areas in Kenya and Tanzania. ETM+ EVI2 and HRVIR EVI2
The EVI2 was derived by optimization of the Linear Vegetation were treated as a single dataset and used together as a predictor
Index (LVI) to attain the best similarity with the EVI, particu- variable to derive an empirical LAI estimation equation. Pfeifer
larly when atmospheric effects are insignificant and data quality et al. (2012) also examined the NDVI and the modified Soil-
is good (Jiang etal., 2008). The LVI was a newly proposed index Adjusted Vegetation Index (MSAVI2) (Qi etal., 1994) along with
for the EVI2 derivation that incorporated the soil-adjustment the EVI2.
factor of SAVI with a linearity-adjustment factor. The derived Kobayashi et al. (2007) compared multisatellite reflectance,
EVI2 (dimensionless) is of the form NDVI, and EVI to elucidate the capabilities of satellite-based
phenology monitoring of larch forests in eastern Siberia.
NIR red Although nearly all the analyses were focused on individual
EVI2 = G (7.2)
NIR + C red + L band reflectances and NDVI, they compared EVI temporal pro-
files from MODIS and VGT at a larch forest site and found that
where MODIS EVI and VGT EVI had very similar seasonality.
G = 2.5 (dimensionless) Although it is not a multisensor EVI application, Huettich
C = 2.4 (dimensionless) etal. (2009) used a Landsat ETM+ and MODIS multiresolution
L = 1.0 (reflectance unit, dimensionless) for the MODIS spec- dataset along with local scale in situ botanical survey data for
tral bands (Jiang etal., 2008) vegetation mapping in a dry savanna ecosystem in Namibia. The
Landsat multispectral data were used first to locate homoge-
Jiang etal. (2008) used MODIS reflectance data extracted from neous patches and then to scale up the insitu botanical informa-
40 globally distributed sites to obtain these coefficients. The tion, which were in turn used to obtain training data in MODIS
consistency between EVI and EVI2 across various land cover EVI temporal signature-based classification.
types demonstrated that their similarity was independent of Recently, Kim etal. (2014) developed a long-term (>30years)
land cover. Comparisons of EVI and EVI2 time series further record of spring frost day (SFD) and spring frost damage day
revealed that their similarity was seasonally independent. The (SFDD) metrics over the conterminous United States by inte-
EVI2 (Equation 7.2) has been adopted as the EVI backup algo- grating a satellite microwave remote sensing record of daily
rithm, which is used as a substitute for the EVI for snow-/ice- landscape freezethaw (FT) status and an EVI2-based phe-
contaminated pixels, in the Collection 6 MODIS VI Product nology record of start of season (SOS) and day of peak (DOP)
suite. canopy cover metrics. This long-term EVI2 dataset used in their
study was derived from calibrated and temporally overlapping
MODIS, AVHRR, and VGT satellite data.
7.3 Multisensor Applications
of eVi and eVi2
7.4 Multisensor compatibility
A growing number of studies have used the EVI from MODIS ofeVi and eVi2: A Review
and from other sensors (Table 7.1). Although not as extensive
as the EVI, the number of studies that used the EVI2 has been As in the case of the NDVI, the EVI and EVI2 from different sen-
growing (Table 7.2). Tables 7.1 and 7.2 list a sample of these stud- sors are likely subject to systematic differences due to differences
ies for the purpose of showing a range of applications of the EVI in sensor/platform characteristics and product generation algo-
and EVI2. rithms. The spectral bandpass is one key sensor characteristic
While most of these studies used EVI or EVI2 data from that varies widely among sensors (Figure 7.1). Although focused
a single sensor, several studies used or evaluated the utilities on the NDVI, Swinnen and Veroustraete (2008) provide a com-
of multisensor EVI or EVI2. Soudani etal. (2006) examined prehensive list of factors that would potentially impact intersen-
the feasibility of IKONOS, ETM+, and SPOT HRVIR for LAI sor compatibility between SPOT VGT and NOAA AVHRR.
estimation and whether retrieved LAI would be comparable Unlike the NDVI, only a limited number of studies have
across the 3 sensors for 28 coniferous and deciduous temper- addressed the issue of multisensor EVI and/or EVI2 compatibil-
ate forest stands in France. Various vegetation indices includ- ity (Table 7.3). Most of these studies used MODIS EVI or EVI2
ing the EVI were used as estimators of LAI. For a lack of a as a reference in examining intersensor EVI and/or EVI2 com-
blue band with the HRVIR sensor, the EVI was computed only patibility, as these indices were originally designed for MODIS.
from ETM+ and IKONOS data, and EVI differences between Many of these studies examined spectral compatibility across
the two sensors were found to be larger for higher LAI (LAI different sensor bandpasses using hyperspectral data (e.g., Miura
range 07). and Yoshioka, 2011).
In Pfeifer et al. (2012), the EVI2 was obtained from atmo- While all these studies primarily investigated compatibility
spherically corrected Landsat ETM+ and SPOT HRVIR data of the EVI across sensors or of the EVI2 among different sen-
over a large number of sites where in situ LAI measurements sors using atmospherically corrected, surface reflectance data,
2016 by Taylor & Francis Group, LLC
158 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 7.1 List of Sensors and Application Fields of EVI
Sensors Application Fields Geographic Regions References
MODIS GPP estimation North America Sims etal. (2008)
Xiao etal. (2005a)
Yang etal. (2007)
Zhang etal. (2013)
Central Asia Li etal. (2007)
Africa Sjostrom etal. (2011)
Global Guanter etal. (2014)
Net ecosystem exchange (CO2 and water vapor) Europe Houborg and Soegaard (2004)
Crop yield prediction Southeast Asia Son etal. (2013)
Evapotranspiration North America Nagler etal. (2005)
Yang etal. (2006)
Australia Guerschman etal. (2009)
Aboveground biomass estimation South America Anaya etal. (2009)
Soil classification Middle East Hengl etal. (2007)
Rooting depth estimation South America Ichii etal. (2007)
Site index North America Waring etal. (2006)
Vegetation type mapping Southern Africa Huettich etal. (2009)
Central Africa Betbeder etal. (2014)
Species richness Eastern Asia John etal. (2008)
Canopy conductance North America, Europe, and Australia Yebra etal. (2013)
Ecosystem respiration North America and Europe Jaegermeyr etal. (2014)
Phenology/drought response North America Mendez-Barroso etal. (2009)
Moran etal. (2014)
South America Huete etal. (2006)
North America and Australia Ponce Campos etal. (2013)
VGT GPP estimation North and South America Xiao etal. (2004)
Xiao etal. (2005b)
Land cover classification East Asia Boles etal. (2004)
MISRa LAI estimation North America Pocewicz etal. (2007)
Landsat TM/ETM+ Vegetation dynamics Western Asia Wittenberg etal. (2007)
LAI, aboveground biomass, yield estimations North America Zhao etal. (2007)
Flux footprint analysis, vegetation type mapping North America Chen etal. (2011a)
Floristic diversity South America Cabacinha and de Castro (2009)
IKONOS LAI estimation Europe Soudani etal. (2006)
a Multiangle imaging spectroradiometer.
TABLe 7.2 List of Sensors and Application Fields of EVI2
Sensors Application Field Geographic Region References
MODIS Crop yield estimation North America Bolton and Friedl (2013)
GPP estimation Northern high latitude Schubert etal. (2012)
North America Wu (2012)
Net ecosystem exchange (CO2) North America Rocha and Shaver (2011)
Phenology North America Yang etal. (2012)
Kim etal. (2014)
South America Morton etal. (2014)
Landsat TM/ETM+ Vegetation type mapping Europe OConnell etal. (2013)
LAI estimation North America Liu etal. (2012)
East Africa Pfeifer etal. (2012)
HRVIR LAI estimation East Africa Pfeifer etal. (2012)
Quickbird Burn severity mapping North America Boelman etal. (2011)
Radiation sensor Greenness and LAI North America Rocha and Shaver (2009)
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 159
1.2 Terra MODIS
Suomi-NPP VIIRS
1.0
GCOM-C SGLI
0.8 SPOT-4 VEGETATION
Landsat-7 ETM+
0.6
0.4
0.2
0.0
400 500 600 700 800 900 1000
1.2 NOAA-14 AVHRR/2
NOAA-19 AVHRR/3
1.0
0.8
0.6
0.4
0.2
0.0
400 500 600 700 800 900 1000
Wavelength (nm)
FIg u r e 7.1 Normalized spectral response curves of blue, red, and NIR bands of selected satellite sensors. The AVHRR sensors do not have a
blue band.
several studies evaluated two additional details. First, Kim etal. blue spectral bands. On the other hand, the consistency of the
(2010) examined intrasensor EVI-to-EVI2 compatibility for EVI from MERIS and VGT with MODIS EVI was not as good
VGT and Sea-Viewing Wide Field-of-View Sensor (SeaWiFS) as was found among the NDVI from these sensors. Fensholt
bandpasses and intersensor MODIS EVI-to-AVHRR EVI2 etal. (2006) attributed this reduced EVI consistency to different,
compatibility. Miura et al. (2008) and Yamamoto et al. (2012) sensor-specific atmospheric correction schemes used in blue band
also examined EVI-to-EVI2 compatibility, but for ASTER and reflectance retrievals.
MODIS. Second, Miura and Yoshioka (2011) assessed the impact Miura et al. (2008) examined the compatibility of ASTER
of different atmospheric correction schemes on cross-sensor EVI2 and MODIS EVI using actual satellite data extracted from
EVI2 compatibility using a scenario where total-atmosphere- randomly located, globally distributed locations. A robust linear
corrected EVI2 from MODIS and VGT were compared with relationship was found between the two satellite indices with a
partial-atmosphere-corrected EVI2 from AVHRR and VIIRS. mean difference of 0.012 EVI units (ASTER minus MODIS).
In the following, we highlight key findings of these studies. Kim et al. (2010) performed a band decomposition analysis
Ferreira etal. (2003) investigated the utility of spectral veg- in which band-pass contributions to observed cross-sensor VI
etation indices in monitoring seasonal dynamics of Brazilian difference were identified. They found that disparities in blue
savanna vegetation formations. Airborne hyperspectral data bandpasses were the primary cause of EVI differences between
were acquired and convolved to the MODIS and ETM+ band- MODIS and other coarse-resolution sensors. The highest com-
passes and converted to the EVI. Intersensor comparisons of patibility was found between VIIRS and MODIS EVI2, while
seasonal dynamics, based on spectral band-pass properties, AVHRR EVI2 was the least compatible to MODIS.
showed that the simulated ETM+ EVI had better seasonal dis- Miura and Yoshioka (2011) also empirically examined
crimination capability than that of MODIS. They attributed this intersensor EVI and EVI2 relationships using MODIS as a ref-
finding to the closer proximity between the ETM+ red and NIR erence, but for a larger number of sensors. They also found
band centers. that spectral band-pass differences resulted in systematic dif-
Fensholt et al. (2006) evaluated the quality of MODIS EVI, ferences on intersensor EVI2 relationships with MODIS, but
Medium-Resolution Imaging Spectrometer (MERIS) EVI, and the relationships were very linear. Intersensor EVI relation-
VGT EVI using insitu measurements and their intersensor com- ships between MODIS and other sensors were linear for most
patibility using actual satellite products at a grassland site in of sensors, although several sensors had either curve-linear
Senegal. A good agreement between the EVI from satellite and or incompatible relationships with MODIS EVI, all of which
from in situ measured MODIS data was found, indicating an have spectrally very different blue band from the MODIS blue
accurate atmospheric correction of the MODIS red, NIR, and band-pass.
2016 by Taylor & Francis Group, LLC
160 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 7.3 List of Multisensor EVI/EVI2 Compatibility Studies
References Data Analyzed Sensor/VI Pair
Ferreira etal. (2003) Airborne hyperspectral data MODIS EVI vs. ETM + EVI
Fensholt etal. (2006) Actual sensor data MODIS EVI vs. MERIS EVI
Actual sensor data MODIS EVI vs. VGT EVI
Miura etal. (2008) Actual sensor data MODIS EVI vs. ASTER EVI2
Kim etal. (2010) EO-1 Hyperion data
a MODIS EVI vs. VIIRS EVI
MODIS EVI vs. VGT EVI
MODIS EVI vs. SeaWiFSb EVI
EO-1 Hyperion data MODIS EVI2 vs. VIIRS EVI2
MODIS EVI2 vs. VGT EVI2
MODIS EVI2 vs. SeaWiFS EVI2
MODIS EVI2 vs. AVHRR/2 EVI2
Miura and Yoshioka (2011) EO-1 Hyperion data MODIS EVI vs. VGT EVI
MODIS EVI vs. GLIc EVI
MODIS EVI vs. VIIRS EVI
MODIS EVI vs. SeaWiFS EVI
MODIS EVI vs. SGLId EVI
EO-1 Hyperion data MODIS EVI2 vs. AVHRR/2 EVI2
MODIS EVI2 vs. AVHRR/3 EVI2
MODIS EVI2 vs. VGT EVI2
MODIS EVI2 vs. GLI EVI2
MODIS EVI2 vs. VIIRS EVI2
MODIS EVI2 vs. SeaWiFS EVI2
MODIS EVI2 vs. CAIe EVI2
MODIS EVI2 vs. SGLI EVI2
EO-1 Hyperion data MODIS EVI2 vs. AVHRR/2 EVI2
MODIS EVI2 vs. VIIRS EVI2
VGT EVI2 vs. AVHRR/2 EVI2
VGT EVI2 vs. VIIRS EVI2
Yamamoto etal. (2012) Actual sensor data MODIS EVI vs. ASTER EVI2
MODIS EVI vs. MODIS/ASTER EVI
Obata etal. (2013) Actual sensor data MODIS EVI vs. VIIRS EVI
a EO-1: Earth observing-1.
b SeaWiFS: Sea-Viewing Wide Field-of-View Sensor.
c GLI: global imager.
d SGLI: second-generation global imager.
e CAI: cloud and aerosol imager.
7.5 Atmospheric impact on inter- 7.5.1 Materials and Methods
and intrasensor Spectral 7.5.1.1 Hyperion Data and Preprocessing
compatibility of eVi and eVi2 The dataset used in Miura et al. (2013) and adapted here
consisted of 37 Level 1R EO-1 Hyperion hyperspectral
In this section, we present a spectral compatibility analysis con-
scenes obtained over 15 globally distributed Aerosol Robotic
ducted to fill in some of the knowledge gaps identified earlier,
Network (AERONET) sites (Holben et al., 2001; Pearlman
namely, the impact of operational atmospheric corrections on
etal., 2003; Middleton etal., 2010). Hyperion is a push broom
inter- and intrasensor spectral compatibility of the EVI and
sensor with 256 pixels in the cross-track direction (~7.7 km
EVI2. We specifically evaluated (1) the effects of residual aero-
swath at 30 m spatial resolution), acquiring Earth-reflected
sols on intersensor EVI-to-EVI, intersensor EVI2-to-EVI2, and
radiation from 400 to 2500 nm at 10 nm nominal sampling
intrasensor EVI2-to-EVI relationships and (2) the effects of
intervals (Pearlman et al., 2003). Level 1R Hyperion scenes
different atmospheric correction schemes on intersensor EVI2
provide radiometrically corrected and calibrated radiance
compatibility between AVHRR and other sensors. The analy-
values and are corrected for the interspectrometer misregis-
sis was conducted by band-pass simulations with a global set of
tration (USGS, 2011).
Hyperion hyperspectral data (Miura etal., 2013).
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 161
16 or top-of-canopy (TOC) reflectances. Second, all the scenes
Min : 0.02 0.4 were corrected for total atmospheric effects, but with incor-
Max : 0.53 rect AOT values (AERONET-measured AOT values 0.05
Mean : 0.13 AOT at 550 nm) to simulate residual aerosol effects in the
12 Median : 0.10 retrieved surface reflectances. Finally, we applied a partial
0.3
atmosphere correction (PAC) on the AVHRR band-pass
Proportion per bar
reflectances where all atmospheric effects except for aerosols
Count
8 were accounted for. The EVI and EVI2 were computed from
0.2 each of these retrieved surface reflectances (i.e., TOC reflec-
tances, TOC reflectances subject to residual aerosol effects,
and PAC reflectances).
4 0.1 Approximately, sixty 1km pixels were randomly selected and
extracted from each scene, totaling ~2000 pixels over the 37
scenes. This random sample dataset was used for the following
analyses.
0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
7.5.1.3 Statistical Difference Analysis
AOT at 550 nm
Three statistical measures of difference were employed to quan-
titatively evaluate and compare inter- and intrasensor compat-
FIg u r e 7.2 Histogram and univariate statistics of atmospheric
AOT at 550nm of Hyperion scenes used in this study. (Adapted and
ibility of the EVI and EVI2: the root mean square difference
reprinted with permission from Miura, T. et al., IEEE Trans. Geosci. (RMSD), systematic square root of mean product difference
Remote Sens., 51, 1349, 2013.) (sRMPD), and unsystematic square root of mean product differ-
ence (uRMPD) (Ji and Gallo, 2006):
The sites/scenes represented 13 out of the 17 International n
(X
Geosphere-Biosphere Programme land cover types, with the 4 1
RMSD = 1,i X 2,i )2 (7.3)
cover types not represented being evergreen broadleaf forests, n i =1
permanent wetlands, snow and ice, and water bodies. Coincident
Level 2 AERONET data (1h of the Hyperion acquisition time)
n
were available for all of the scenes. Aerosol optical thickness
(X
1
sRMPD = X 2,i )2 X 1,i X 1,i X 2,i X 2,i (7.4)
(AOT) at 550nm ranged from 0.02 (clean) to 0.53 (turbid) for n
1,i
i =1
this dataset (Figure 7.2). See Table 1 of Miura et al. (2013) for
additional details of this Hyperion dataset.
The Hyperion scenes were spectrally and spatially convolved
( X )
n
1
to simulate top-of-the-atmosphere (TOA) reflectances of the uRMPD = 1,i X 1,i X 2,i X 2,i (7.5)
n i =1
VIIRS, MODIS, second-generation global imager (SGLI), VGT,
AVHRR/3, AVHRR/2, and ETM+ bands at 1 km spatial reso-
lution. The spectral response functions of these spectral bands where
(Figure 7.1) were interpolated to Hyperion band center wave- X1 and X2 are the EVI or EVI2 for sensor-1 and sensor-2,
lengths for each Hyperion pixel (Pearlman et al., 2003). The respectively
MODIS point spread function, which is, to the first order, trian- X 1 and X 2 are the corresponding points of X1 and X2 on the
gular in the scan direction and rectangular in the track direc- regression line that is the best estimate of the primary
tion, was assumed for the spatial convolution (Wolfe etal., 2002). trend of the X1 vs. X2 scatter
7.5.1.2 Simulation of Atmospheric correction sRMPD and uRMPD decouple RMSD into its systematic and
The 6S radiative transfer code was used to atmospherically unsystematic components, respectively; sRMPD measures how
correct the convolved Hyperion scenes (Vermote et al., 2006; far the scatter (trend or regression line) is from the line of
Kotchenova and Vermote, 2007). For each scene, the 6S code was X1 = X2, or the bias errors, and uRMPD measures the magnitude
constrained with the corresponding in situ AERONET atmo- of data scattering (secondary variation) about the regression line
spheric measurements, view and solar zenith, relative azimuth (Ji and Gallo, 2006). Following the protocol developed by Ji and
angles at the time of the image acquisition, and the site elevation Gallo (2006), regression lines were estimated using the geometric
obtained from GTOPO30. mean functional regression (GMFR), which considers that both
Three atmospheric corrections were applied. First, all the variables of interest are subject to error. Unlike the ordinary least
scenes were corrected for total atmospheric effects including squares regression, the derived regression line is symmetric about
aerosols, that is, TOA reflectances were reduced to surface X1 and X2, or invertible.
2016 by Taylor & Francis Group, LLC
162 Remotely Sensed Data Characterization, Classification, and Accuracies
7.5.2 Results systematic differences (in terms of sRMPD) were observed for
MODIS vs. ETM+ and SGLI vs. VGT4, respectively (Table 7.4). The
7.5.2.1 intersensor eVi magnitudes of unsystematic differences also changed among sen-
EVI differences for all possible combinations among the five sen- sor pairs, and uRMPD were the smallest for VGT4 vs. ETM+ and
sors of VIIRS, MODIS, SGLI, VGT4, and ETM+ for both atmo- the largest for VIIRS vs. SGLI and ETM+ (Figure 7.3 and Table 7.4).
spheric corrections with and without AOT errors are plotted in Overlapped symbols in Figure 7.3 indicate that intersen-
Figure 7.3 and summarized in Table 7.4 (for GMFR equations sor EVI differences remained nearly the same when the input
see Table 7.A.1 in Appendix 7.A). All sensor pairs examined here surface reflectances were subject to residual aerosol contamina-
were subject to systematic differences of which magnitudes varied tions. For all the sensor pairs examined here, RMSD, sRMPD,
among sensor pairs. The systematic differences basically linearly and uRMPD of intersensor EVI were the same whether the input
increased or decreased with EVI values although some curve lin- surface reflectances were subject to AOT errors or not (Table7.4).
earity was seen for those between VIIRS and MODIS, VGT4, or These results indicate that the atmospheric resistance of the EVI
ETM+ (Figure7.3a, c, andd, respectively). The smallest and largest functions well for intersensor EVI.
0.10
No AOT error
0.08 AOT +
AOT
0.06
EVI (VIIRS MODIS)
0.04
0.02
0.00
0.02
0.04
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(a) EVI (MODIS)
0.06 0.10 0.10
EVI (VIIRS ETM+)
EVI (VIIRS VGT4)
EVI (VIIRS SGLI)
0.04 0.08 0.08
0.02 0.06 0.06
0.00 0.04 0.04
0.02 0.02 0.02
0.04 0.00 0.00
0.06 0.02 0.02
0.08 0.04 0.04
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(b) EVI (SGLI) (c) EVI (VGT4) (d) EVI (ETM+)
0.04 0.10 0.10
EVI (MODIS ETM+)
EVI (MODIS VGT4)
EVI (MODIS SGLI)
0.02 0.08 0.08
0.00 0.06 0.06
0.02 0.04 0.04
0.04 0.02 0.02
0.06 0.00 0.00
0.08 0.02 0.02
0.10 0.04 0.04
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(e) EVI (SGLI) (f) EVI (VGT4) (g) EVI (ETM+)
FIg ur e 7.3 Inter-sensor EVI difference across VIIRS, MODIS, SGLI, VGT4, and ETM+ bandpasses subject to atmospheric aerosol
correction error. (Continued )
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 163
0.14 0.10 0.04
EVI (VGT4 ETM+)
EVI (SGLI ETM+)
EVI (SGLI VGT4)
0.12 0.08 0.02
0.10 0.06 0.00
0.08 0.04 0.02
0.06 0.02 0.04
0.04 0.00 0.06
0.02 0.02 0.08
0.00 0.04 0.10
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(h) EVI (VGT4) (i) EVI (ETM+) (j) EVI (ETM+)
FIg u r e 7.3 (c ontinued ) Inter-sensor EVI difference across VIIRS, MODIS, SGLI, VGT4, and ETM+ bandpasses subject to atmospheric
aerosol correction error.
TABLe 7.4 Statistics for Intersensor Spectral Difference: EVI
TOC with No AOT Error TOC with AOT Error of 0.05 Ratio (TOC Error/TOC No Error)
RMSD sRMPD uRMPD RMSD sRMPD uRMPD RMSD sRMPD uRMPD
VIIRS vs. MODIS 0.019 0.019 0.005 0.020 0.019 0.005 1.0 1.0 1.0
VIIRS vs. SGLI 0.021 0.019 0.009 0.021 0.019 0.009 1.0 1.0 1.0
VIIRS vs. VGT4 0.054 0.053 0.008 0.054 0.053 0.008 1.0 1.0 1.0
VIIRS vs. ETM+ 0.033 0.032 0.009 0.034 0.032 0.009 1.0 1.0 1.0
MODIS vs. SGLI 0.031 0.030 0.008 0.031 0.030 0.008 1.0 1.0 1.0
MODIS vs. VGT4 0.035 0.034 0.005 0.035 0.034 0.005 1.0 1.0 1.0
MODIS vs. ETM+ 0.018 0.017 0.006 0.018 0.017 0.006 1.0 1.0 1.0
SGLI vs. VGT4 0.060 0.060 0.008 0.060 0.060 0.008 1.0 1.0 1.0
SGLI vs. ETM+ 0.035 0.035 0.006 0.035 0.035 0.006 1.0 1.0 1.0
VGT4 vs. ETM+ 0.026 0.025 0.003 0.026 0.026 0.003 1.0 1.0 1.0
7.5.2.2 intersensor eVi2 being 1.2). The resultant uRMPD of intersensor EVI2 relationships
Intersensor EVI2 relationships were similar to those of EVI for were nearly the same or larger (0.91.7 times) than the correspond-
the sensor pairs examined here (Figure 7.4). They were, however, ing uRMPD of intersensor EVI relationships that were found insen-
very different for VIIRS vs. MODIS in that the intersensor EVI2 sitive to residual aerosol contaminations (the most right column
relationship was very linear and their biases were, on average, in Table 7.6). The residual aerosol contaminations did not impact
negative and smaller than those of EVI (Figure 7.4a). sRMPD because we used symmetric errors to perturb EVI2. These
For the sensor pairs studied here, intersensor EVI2 differences results indicate that the atmospheric resistance of the EVI with a
were basically smaller than those of EVI when the input reflec- blue band is advantageous in deriving intersensor EVI relationships.
tances were not subject to errors. RMSD, sRMPD, and uRMPD for In order to evaluate and confirm the atmospheric resistance
intersensor EVI2 relationships ranged from 0.003 to 0.035, 0.002 to of the EVI against the EVI2, differences between EVI with
0.034, and 0.002 to 0.008, respectively (Table 7.5) (for GMFR equa- and without residual contaminations as well as those between
tions see Table 7.A.2 in Appendix 7.A), whereas the corresponding EVI2 with and without residual contaminations are plotted in
values for intersensor EVI relationships were 0.020.06 (RMSD), Figure7.5 and their RMSD summarized in Table 7.7. For the five
0.0190.06 (sRMPD), and 0.0030.009 (uRMPD) (Table 7.4). Table sensor bandpasses, the EVI successfully reduced residual aerosol
7.6 lists the ratios of these three different statistics for every sen- errors to less than 0.005 (Figure 7.5) with RMSD of 0.002 or
sor pairs. One can observe that, except for MODIS vs. VGT4 and less (Table 7.7). EVI2 errors exceeded 0.015 with RMSD of 0.005
MODIS vs. ETM+, all the ratios were less than 1.0, indicating or more for the AOT estimation errors of 0.05.
smaller systematic and unsystematic, and overall differences for
intersensor EVI2 relationships. 7.5.2.3 intrasensor eVi2 vs. eVi
On the other hand, unlike the EVI, intersensor EVI2 differences EVI2-to-EVI differences with and without residual aerosol errors
varied when the input reflectances were subject to residual aerosol are plotted as a function of EVI in Figure 7.6 and their statistics
contaminations due to atmospheric correction errors. In Figure 7.4, summarized in Table 7.8 (for GMFR equations see Table 7.A.3
it can be seen that EVI2 differences formed three separate trends, in Appendix 7.A), in order to analyze intrasensor EVI2-to-EVI
corresponding to the three correction error scenarios for all the sen- compatibilities for the sensors other than MODIS. Intrasensor
sor pairs. It can also be seen as a 20% or more increase in uRMPD EVI2 vs. EVI relationships of all sensor bandpasses were sub-
upon the contaminations (the smallest uRMPD ratio in Table 7.5 ject to systematic differences although their magnitudes changed
2016 by Taylor & Francis Group, LLC
164 Remotely Sensed Data Characterization, Classification, and Accuracies
0.04
0.02
EVI2 (VIIRS MODIS)
0.00
0.02
0.04 No AOT error
AOT +
AOT
0.06
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(a) EVI2 (MODIS)
0.04 0.08 0.08
EVI2 (VIIRS ETM+)
EVI2 (VIIRS VGT4)
EVI2 (VIIRS SGLI)
0.02 0.06 0.06
0.00 0.04 0.04
0.02 0.02 0.02
0.04 0.00 0.00
0.06 0.02 0.02
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(b) EVI2 (SGLI) (c) EVI2 (VGT4) (d) EVI2 (ETM+)
0.04 0.08 0.08
EVI2 (MODIS ETM+)
EVI2 (MODIS VGT4)
EVI2 (MODIS SGLI)
0.02 0.06 0.06
0.00 0.04 0.04
0.02 0.02 0.02
0.04 0.00 0.00
0.06 0.02 0.02
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(e) EVI2 (SGLI) (f) EVI2 (VGT4) (g) EVI2 (ETM+)
0.08 0.08 0.04
EVI2 (SGLI ETM+)
EVI2 (VGT4 ETM+)
EVI2 (SGLI VGT4)
0.06 0.06 0.02
0.04 0.04 0.00
0.02 0.02 0.02
0.00 0.00 0.04
0.02 0.02 0.06
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(h) EVI2 (VGT4) (i) EVI2 (ETM+) (j) EVI2 (ETM+)
FIg u r e 7.4 Inter-sensor EVI2 difference across VIIRS, MODIS, SGLI, VGT4, and ETM+ bandpasses subject to atmospheric aerosol correction error.
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 165
TABLe 7.5 Statistics for Intersensor Spectral Difference: EVI2
TOC with No AOT Error TOC with AOT Error of 0.05 Ratio (TOC Error/TOC No Error)
RMSD sRMPD uRMPD RMSD sRMPD uRMPD RMSD sRMPD uRMPD
VIIRS vs. MODIS 0.003 0.002 0.002 0.006 0.002 0.006 2.1 0.8 3.2
VIIRS vs. SGLI 0.015 0.014 0.007 0.016 0.014 0.009 1.1 1.0 1.3
VIIRS vs. VGT4 0.031 0.031 0.007 0.032 0.031 0.009 1.0 1.0 1.3
VIIRS vs. ETM+ 0.026 0.025 0.008 0.027 0.025 0.010 1.0 1.0 1.2
MODIS vs. SGLI 0.013 0.012 0.005 0.014 0.012 0.008 1.1 1.0 1.5
MODIS vs. VGT4 0.033 0.032 0.006 0.034 0.033 0.008 1.0 1.0 1.4
MODIS vs. ETM+ 0.027 0.026 0.006 0.028 0.027 0.008 1.0 1.0 1.3
SGLI vs. VGT4 0.035 0.034 0.006 0.036 0.035 0.008 1.0 1.0 1.3
SGLI vs. ETM+ 0.026 0.026 0.005 0.027 0.026 0.008 1.0 1.0 1.4
VGT4 vs. ETM+ 0.010 0.009 0.002 0.011 0.009 0.006 1.1 1.0 2.6
TABLe 7.6 Statistics for Intersensor Spectral Difference: EVI2 to EVI Ratio
TOC with No AOT Error TOC with AOT Error of 0.05
RMSD sRMPD uRMPD RMSD sRMPD uRMPD
VIIRS vs. MODIS 0.2 0.1 0.4 0.3 0.1 1.2
VIIRS vs. SGLI 0.7 0.7 0.7 0.8 0.7 0.9
VIIRS vs. VGT4 0.6 0.6 0.8 0.6 0.6 1.0
VIIRS vs. ETM+ 0.8 0.8 0.8 0.8 0.8 1.0
MODIS vs. SGLI 0.4 0.4 0.7 0.5 0.4 1.0
MODIS vs. VGT4 1.0 0.9 1.1 1.0 1.0 1.5
MODIS vs. ETM+ 1.5 1.6 1.0 1.6 1.6 1.4
SGLI vs. VGT4 0.6 0.6 0.8 0.6 0.6 1.0
SGLI vs. ETM+ 0.7 0.7 0.9 0.8 0.8 1.2
VGT4 vs. ETM+ 0.4 0.4 0.7 0.4 0.4 1.7
from sensor to sensor. They were the largest for SGLI with all the other sensors when paired with AVHRR/3 or AVHRR/2
sRMPD of 0.0150.016 and the smallest for ETM+ with sRMPD (Table 7.9). AVHRR EVI2 had the largest differences with SGLI
of 0.0060.007 (Table 7.8). These sRMPD values were basically and the smallest differences with VGT4 (Table 7.9).
all less than those observed for intersensor EVI or EVI2 relation- AVHRR/3 and AVHRR/2 EVI2 showed very high compat-
ships, except for that of VIIRS EVI2 vs. MODIS EVI2 that had ibility regardless of the level of atmospheric correction as far as
the smallest sRMPD and that of VGT4 EVI2 vs. ETM+ EVI2. both sensor data had the same levels of atmospheric correction
Unsystematic variations of these intrasensor relationships were (Figure 7.7a and b). Although they were subject to some system-
about the same across all the sensors with uRMPD ranging from atic differences (sRMPD ~ 0.015), they were subject to very small
0.007 to 0.008 without residual aerosols and from 0.008 to 0.01 unsystematic variations (uRMPD ~ 0.002), and their relation-
with residual aerosols (Table 7.8). The magnitudes of uRMPD were ships were nearly independent of atmospheres, indicating the
comparable to those of intersensor EVI2 relationships (Table 7.5). compatible sensitivity of their spectral bandpasses to atmo-
spheric (aerosol) effects.
7.5.2.4 intersensor compatibility with AVHRR eVi2 In summary, intersensor EVI relationships were insensitive
Lastly, we present intersensor compatibility between TOC EVI2 to residual aerosol contaminations due to EVIs atmospheric
and PAC AVHRR EVI2. No operational aerosol correction of resistance. The EVI2 showed higher multisensor compatibility
AVHRR data has been implemented, and thus, this provides a than the EVI, with smaller bias and unsystematic errors, which
realistic scenario on EVI2 compatibility/continuity with AVHRR. was however only when the input reflectances were not subject
For comparisons, AVHRR TOC EVI2 was also analyzed. to residual aerosol contaminations. Intrasensor EVI vs. EVI2
Figure 7.7a and b show example plots for VIIRS TOC EVI2 vs. relationships were subject to the same magnitudes of unsystem-
AVHRR/3 TOC EVI2 and for VIIRS TOC EVI2 vs. AVHRR/3 atic errors as intersensor EVI2 relationships. Intrasensor EVI2-
PAC EVI2 differences, respectively. Whether corrected for total or to-EVI relationships were subject to significant biases for some
partial atmosphere, AVHRR/3 EVI2 had a linear relationship with sensors. This indicated the need of EVI and/or EVI2 coefficient
VIIRS EVI2. However, AVHRR/3 PAC EVI2 had larger systematic adjustments, which were optimized for the MODIS bandpasses
and unsystematic differences than the TOC counterpart against by design. Backward compatibility with AVHRR EVI2 would
VIIRS EVI2. sRMPD and uRMPD for the former were 0.028 and be a feasible option only when aerosol source of variability are
0.007, respectively, whereas those for the latter were 0.041 and reduced in AVHRR EVI2 either by atmospheric correction or by
0.011, respectively (Table 7.9). The same trends were observed for temporal compositing to select the cleanest observations.
2016 by Taylor & Francis Group, LLC
166 Remotely Sensed Data Characterization, Classification, and Accuracies
0.020
0.015
0.010
EVI or EVI2
0.005
0.000
0.005
0.010 EVI: AOT +
EVI: AOT
0.015 EVI2: AOT +
EVI2: AOT
0.020
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(a) EVI or EVI2
0.020 0.02
0.015
0.010 0.01
EVI or EVI2
EVI or EVI2
0.005
0.000 0.00
0.005
0.010 0.01
0.015
0.020 0.02
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(b) EVI or EVI2 (c) EVI or EVI2
0.020 0.020
0.015 0.015
0.010 0.010
EVI or EVI2
EVI or EVI2
0.005 0.005
0.000 0.000
0.005 0.005
0.010 0.010
0.015 0.015
0.020 0.020
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(d) EVI or EVI2 (e) EVI or EVI2
FIg ur e 7.5 Comparison of EVI and EVI2 resistance to atmospheric aerosol correction error for five sensor bandpasses. (a) MODIS, (b) VIIRS,
(c) SGLI, (d) VGT4, and (e) ETM+.
TABLe 7.7 Comparison between EVI and EVI2
Kim et al. (2010) used a simple linear model and derived
Resistance to Residual Aerosol Effects
EVI-to-EVI and EVI2-to-EVI2 spectral translation equations
RMSD EVI EVI2 EVI2-to-EVI Ratio between MODIS and select medium-resolution sensors:
VIIRS 0.002 0.006 3.5
MODIS 0.001 0.006 6.8 X Sensor = a X MODIS + b + (7.6)
SGLI 0.001 0.006 4.5
VGT4 0.001 0.005 7.6 where
ETM+ 0.001 0.005 4.8 X is the EVI or EVI2
The subscript, Sensor, refers to either VIIRS, VGT4, SeaWiFS,
or AVHRR/2
7.6 intersensor calibration is the unexplained residual error
of eVi and eVi2 Kim etal. (2010) also developed intrasensor EVI-to-EVI2 trans-
A small number of studies have evaluated and/or proposed inter- lation equations with the simple linear model for the VIIRS,
sensor calibration for the EVI and EVI2. These cross-calibration VGT4, and SeaWiFS bandpasses:
methods can be divided into empirical (Kim et al., 2010) and
theoretical (Yoshioka etal., 2012; Obata etal., 2013) approaches. EVI2Sensor = a EVISensor + b + (7.7)
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 167
0.06 0.06
No AOT error
0.04 AOT + 0.04
AOT
0.02 0.02
EVI2 EVI
EVI2 EVI 0.00 0.00
0.02 0.02
0.04 0.04
0.06 0.06
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
(a) EVI (b) EVI
0.06 0.06
0.04 0.04
0.02 0.02
EVI2 EVI
EVI2 EVI
0.00 0.00
0.02 0.02
0.04 0.04
0.06 0.06
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(c) EVI (d) EVI
FIg u r e 7.6 Intrasensor EVI vs. EVI2 differences subject to atmospheric aerosol correction error for four sensor bandpasses. (a) VIIRS, (b) SGLI,
(c) VGT4, and (d) ETM+.
TABLe 7.8 Difference Statistics for Intrasensor EVI2 and EVI Compatibility
TOC with No AOT Error TOC with AOT Error of 0.05 Ratio (TOC Error/TOC No Error)
RMSD sRMPD uRMPD RMSD sRMPD uRMPD RMSD sRMPD uRMPD
VIIRS 0.016 0.014 0.008 0.016 0.013 0.010 1.02 0.96 1.14
SGLI 0.018 0.016 0.008 0.018 0.015 0.010 1.01 0.97 1.16
VGT4 0.012 0.010 0.007 0.013 0.010 0.008 1.11 1.05 1.22
ETM+ 0.010 0.007 0.007 0.011 0.006 0.009 1.08 0.93 1.18
Mean absolute deviations/differences (MAD) were used to eval- the equations did not perform as good as those for the intersen-
uate translation results: sor calibration. Reductions in MAD were 35% or less and the
largest MAD of 0.0154 was observed for VGT4 (Table 7.10).
n
Yoshioka et al. (2012) and Obata et al. (2013) analytically
X
1
MAD = X 2,i (7.8)
n
1,i
derived EVI and/or EVI2 spectral translation equations. Both
i =1
studies used the vegetation isoline equations in the derivations,
where Xs are the EVI, EVI2, translated (predicted) EVI, or which are based upon the physics of vegetationphoton interac-
translated (predicted) EVI2 using the derived simple linear equa- tions (Yoshioka, 2004).
tions and two of the four variables are selected and inserted into Yoshioka et al. (2012) used the vegetation isoline equations
Equation 7.8 depending on the quantity of interest. Hyperion first to eliminate NIR and blue reflectances from the EVI equa-
data collected along a tropical forestsavanna ecogradient in tion or NIR reflectance from the EVI2 equation and then to relate
Brazil were used in all of these derivations and evaluations of two EVI or EVI2 using the red reflectances. Derived is a spectral
cross-calibration equations. transformation equation that relates two EVI or two EVI2:
The derived, simple linear spectral translation equations
performed well on the Hyperion-simulated band-pass data h1v b h2
va = G (7.9)
(Table 7.10). For both the EVI and EVI2, MAD were reduced h3v b h4
to 0.005 or less (in EVI/EVI2 units) upon the translations for
most cases, which corresponded to 31%87% reductions in where va and vb are the EVI or EVI2 for sensors a and b, respec-
MAD (Table 7.10). For the EVI-to-EVI2 spectral calibration, tively. The four coefficients, hi (i = 1, , 4), are actually functions,
2016 by Taylor & Francis Group, LLC
168 Remotely Sensed Data Characterization, Classification, and Accuracies
0.12 0.12
EVI2 (VIIRS AVHRR/3)
EVI2 (VIIRS AVHRR/3)
0.10 0.10
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0.00 0.00
0.02 0.02
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(a) EVI2 (AVHRR/3) (b) EVI2 (AVHRR/3)
0.12 0.12
EVI2 (AVHRR/3 AVHRR/2)
EVI2 (AVHRR/3 AVHRR/2)
0.10 0.10
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
0.00 0.00
0.02 0.02
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(c) EVI2 (AVHRR/2) (d) EVI2 (AVHRR/2)
FIg u r e 7.7 Intersensor EVI2 difference under two atmospheric correction scenarios: (a) VIIRS vs. AVHRR/3 for TOC, (b) VIIRS vs. AVHRR/3
for PAC, (c) AVHRR/3 vs. AVHRR/2 for TOC, and (d) AVHRR/3 vs. AVHRR/2 for PAC.
TABLe 7.9 Difference Statistics of Intersensor AVHRR EVI2 Compatibility
TOC vs. TOC TOC vs. PAC Ratio (PAC/TOC)
EVI2 vs. AVHRR EVI2 RMSD sRMPD uRMPD RMSD sRMPD uRMPD RMSD sRMPD uRMPD
VIIRS vs. AVHRR/3 0.029 0.028 0.007 0.042 0.041 0.011 1.5 1.5 1.7
MODIS vs. AVHRR/3 0.031 0.030 0.006 0.044 0.043 0.011 1.4 1.4 1.8
SGLI vs. AVHRR/3 0.035 0.034 0.008 0.049 0.047 0.013 1.4 1.4 1.6
VGT4 vs. AVHRR/3 0.008 0.007 0.004 0.018 0.015 0.010 2.2 2.2 2.2
ETM+ vs. AVHRR/3 0.014 0.013 0.005 0.027 0.025 0.010 1.9 1.9 1.9
VIIRS vs. AVHRR/2 0.043 0.042 0.006 0.055 0.054 0.010 1.3 1.3 1.8
MODIS vs. AVHRR/2 0.045 0.045 0.006 0.058 0.057 0.010 1.3 1.3 1.8
SGLI vs. AVHRR/2 0.050 0.050 0.008 0.063 0.062 0.012 1.3 1.2 1.5
VGT4 vs. AVHRR/2 0.018 0.017 0.005 0.030 0.029 0.010 1.7 1.7 1.9
ETM+ vs. AVHRR/2 0.027 0.027 0.006 0.040 0.038 0.010 1.4 1.4 1.7
AVHRR/3 vs. AVHRR/2 0.015 0.015 0.002 0.014a 0.014a 0.001a 0.9 0.9 1.0
See Appendix 7.A.4 for corresponding GMFR equations.
a These statistics are for AVHRR/3 PAC EVI2 vs. AVHRR/2 PAC EVI2.
and their values change with vegetation amount, and soil and where Ki (i = 1,, 4) are the spectral transformation coefficients.
atmosphere conditions. As hi in Equation 7.9, Ki are functions and change with vegeta-
In Obata etal. (2013), MODIS red, NIR, and blue reflectances tion, soil, and atmosphere conditions.
were expressed as a function of VIIRS red, NIR, and blue reflec- Obata et al. (2013), however, found via optimization for
tances, respectively, using the vegetation isoline equations. both model-simulated and actual satellite datasets that a single
Substituting MODIS reflectances with the functions resulted in set of coefficient values can be derived with a reasonable level
a MODIS-compatible EVI, v m: of accuracy. Their simulation dataset was generated with the
PROSPECT+SAIL canopy reflectance model (Feret etal., 2008)
NIR K 1red + K 2 and the 6S atmospheric radiative transfer model. The actual
v m = G (7.10)
NIR + K 1C1red K 3C 2blue + K 4 MODIS-VIIRS dataset was populated by extracting pairs of
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 169
TABLe 7.10 Comparison of Mean Absolute Deviation before and after Spectral Translation Using
Simple Linear Cross-Calibration Equation
EVI intersensor spectral compatibility with MODIS
MAD for EVISensor vs. EVIMODIS MAD for EVISensor vs. pred. EVISensor Reduction (%)
VIIRS 0.0118 0.0052 56
VGT4 0.0207 0.0042 80
SeaWiFS 0.0108 0.0070 35
EVI2 intersensor spectral compatibility with MODIS
MAD for EVI2Sensor vs. EVI2MODIS MAD for EVI2Sensor vs. pred. EVI2Sensor Reduction (%)
VIIRS 0.0027 0.0018 33
VGT4 0.0255 0.0033 87
SeaWiFS 0.0048 0.0033 31
AVHRR/2 0.0411 0.0059 86
EVI-to-EVI2 intrasensor compatibility
MAD for EVI2Sensor vs. EVISensor MAD for EVI2Sensor vs. pred. EVI2Sensor Reduction (%)
VIIRS 0.0108 0.0077 29
VGT4 0.0236 0.0154 35
SeaWiFS 0.0103 0.0091 12
Source: Adapted from Kim, Y. etal., J. Appl. Remote Sens., 4, 043520, 2010.
Aqua MODIS and VIIRS daily nadir-view surface reflectance VIIRS EVI) were 0.008 and 0.0001, respectively, and vari-
spectra on the same date (in August 2013) at the same coordi- ability of *2 was much smaller than that of 1 (RMSD changed
nates over North America. Refer to Obata etal. (2013) for further from 0.01 to 0.0017, an 83% reduction) (Obata etal., 2013). For
details on these datasets. the actual dataset, the mean and RMSD of *2 were 0.001 and
For the simulated dataset, the mean of 1 (MODIS EVI minus 0.023, a 95% and 41% reduction, respectively, from those of 1
VIIRS EVI) and *2 (MODIS EVI minus MODIS-compatible (0.022 and 0.039) (Figure 7.8).
0.02
0.02
1
0.04
0.06
0 0.1 0.2 0.3 0.4 0.5 0.6
(a) MODIS EVI
0.04
0.02
0
2*
0.02
0.04
0 0.1 0.2 0.3 0.4 0.5 0.6
(b) MODIS EVI
FIg u r e 7.8 Density plot of EVI differences against MODIS EVI with actual MODIS and VIIRS data. (a) MODIS EVI minus VIIRS EVI
(1) against MODIS EVI (v m) and (b) MODIS EVI minus MODIS compatible VIIRS EVI using optimized parameters (*2 ) against MODIS
EVI (v m). Ranges of y axis for the density plots were 0.1. (Adapted and reprinted with permission from Obata et al., J. Appl. Remote Sens. 7,
073467, 2013.)
2016 by Taylor & Francis Group, LLC
170 Remotely Sensed Data Characterization, Classification, and Accuracies
7.7 Discussions and Future Directions Appendix 7.A inter- and intrasensor
In this chapter, inter- and intrasensor compatibilities of the
Spectral Relationships for eVi and eVi2
EVI and EVI2 were reviewed with a focus on spectral, atmo- The following equations were estimated for the Hyperion data-
spheric, and calibration issues. Overall, the EVI2 showed set (without AOT errors) described in Section 7.5.1 by GMFR
higher intersensor spectral compatibilities than the EVI. (Tables 7.A.1 through 7.A.4).
The highest spectral compatibility was observed between
MODIS EVI2 and VIIRS EVI2, and AVHRR/2 EVI2 and
AVHRR/3 EVI2 although both intersensor EVI2 relationships
TABLe 7.A.1 Intersensor EVI vs. EVI Spectral Relationships
were subject to systematic differences. When the residual
aerosol effects were considered, however, intersensor spectral Y vs. X Intercepta Slopea R2
compatibilities of the EVI2 and EVI were at the comparable VIIRS vs. MODIS 0.010 1.033 0.999
level. The blue band in the EVI is advantageous in reducing VIIRS vs. SGLI 0.029 0.893 0.997
residual aerosol effects, but brings an added complication in VIIRS vs. VGT4 0.033 1.083 0.997
intersensor compatibilities. Likewise, the higher intersensor VIIRS vs. ETM+ 0.038 0.975 0.997
EVI2 compatibilities do not necessarily translate into higher MODIS vs. SGLI 0.019 0.864 0.998
absolute accuracies of the EVI2. MODIS vs. VGT4 0.023 1.048 0.999
Among the limited number of studies and available cross- MODIS vs. ETM+ 0.027 0.944 0.999
calibration methods, the approach of Obata et al. (2013) SGLI vs. VGT4 0.005 1.213 0.998
appears to be the most promising. The most significant advan- SGLI vs. ETM+ 0.010 1.092 0.999
tage of the approach would be that the translation technique VGT4 vs. ETM+ 0.004 0.900 1.000
simultaneously optimizes for band-pass differences and for a Useful for trend analysis; use with caution on actual satellite data.
the EVI coefficients, which cannot be accomplished with the
simple linear regression approach. The potential pitfall in the
approach is a nonexistence of the unique set of coefficient val-
ues, as it requires nonlinear regression to find the optimum TABLe 7.A.2 Intersensor EVI2 vs. EVI2 Spectral Relationships
coefficient values.
Y vs. X Intercepta Slopea R2
Intersensor spectral compatibility of AVHRR EVI2 with
other sensors is challenging, in particular with MODIS, VIIRS, VIIRS vs. MODIS 0.001 0.989 1.000
and SGLI of which red and NIR spectral bands are much VIIRS vs. SGLI 0.021 0.918 0.998
narrower than those of AVHRR (i.e., Figure 7.1). Backward VIIRS vs. VGT4 0.026 1.018 0.998
compatibility of these sensors with AVHRR EVI2 would be VIIRS vs. ETM+ 0.031 0.971 0.998
a feasible option only when aerosol source of variability are MODIS vs. SGLI 0.020 0.929 0.999
reduced in AVHRR data either by atmospheric correction MODIS vs. VGT4 0.025 1.030 0.999
or by temporal compositing to select cleanest observations. MODIS vs. ETM+ 0.030 0.982 0.998
Another approach would be to generate more than one EVI/ SGLI vs. VGT4 0.006 1.108 0.999
EVI2 long-term data record in parallel, for example, one SGLI vs. ETM+ 0.011 1.057 0.999
solely from the AVHRR sensor series and another by merg- VGT4 vs. ETM+ 0.005 0.954 1.000
ing MODIS and VIIRS EVI products. Intercomparisons of a Useful for trend analysis; use with caution on actual satellite data.
the two data records over the current overlapping period can
be used to develop a methodology for analyzing historical
AVHRR EVI2 time series.
A large number of empirical investigations with actual satel- TABLe 7.A.3 Intrasensor EVI2 vs. EVI Relationships
lite data are needed and the logical next step. For cross-sensor
EVI2 vs. EVI Intercepta Slopea R2
compatibility and calibration of actual satellite data, additional
factors need to be taken into consideration, including differences VIIRS 0.006 0.973 0.997
in spatial resolution, overpass time (morning vs. afternoon), and SGLI 0.002 0.946 0.998
sun-target-viewing geometry (sun and view zenith angle differ- VGT4 0.000 1.036 0.998
ences due to different sensor platform orbits) (e.g., Sims et al., ETM+ 0.000 0.977 0.998
2011; Morton etal., 2014). a Useful for trend analysis; use with caution on actual satellite data.
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 171
TABLe 7.A.4 Intersensor Relationships for AVHRR EVI2
TOC vs. TOC TOC vs. PAC
Y vs. X Intercepta Slopea R2 Interceptb Slopeb R2
VIIRS vs. AVHRR/3 0.013 1.058 0.998 0.008 1.148 0.999
MODIS vs. AVHRR/3 0.012 1.070 0.998 0.013 1.236 0.997
SGLI vs. AVHRR/3 0.009 1.152 0.997 0.017 1.115 0.999
VGT4 vs. AVHRR/3 0.013 1.039 0.999 0.023 1.169 0.998
ETM+ vs. AVHRR/3 0.018 1.089 0.999 0.007 1.199 0.995
VIIRS vs. AVHRR/2 0.011 1.119 0.994 0.006 1.212 0.995
MODIS vs. AVHRR/2 0.010 1.132 0.994 0.016 1.305 0.993
SGLI vs. AVHRR/2 0.011 1.219 0.993 0.019 1.178 0.995
VGT4 vs. AVHRR/2 0.015 1.099 0.995 0.025 1.234 0.995
ETM+ vs. AVHRR/2 0.021 1.152 0.995 0.004 1.073 1.000
AVHRR/3 vs. AVHRR/2 0.009 1.135 0.998 0.004c 1.071c 1.000
a Useful for trend analysis; use with caution on actual satellite data.
b Useful for trend analysis; do not use on actual satellite data.
c The intercept and slope values are for AVHRR/3 PAC EVI2 vs. AVHRR/2 PAC EVI2.
References Changyong, C., F.J. De Luccia, X. Xiaoxiong, R. Wolfe, and W. Fuzhong.
2014. Early on-orbit performance of the Visible Infrared
Anaya, J.A., E. Chuvieco, and A. Palacios-Orueta. 2009. Imaging Radiometer Suite onboard the Suomi National Polar-
Aboveground biomass assessment in Colombia: A remote Orbiting Partnership (S-NPP) satellite. IEEE Trans. Geosci.
sensing approach. Forest Ecol. Manage. 257:12371246. Remote Sens. 52:11421156. doi:10.1109/TGRS.2013.2247768.
doi:10.1016/j.foreco.2008.11.016. Chen, B., N.C. Coops, D. Fu, H.A. Margolis, B.D. Amiro, A.G. Barr,
Betbeder, J., V. Gond, F. Frappart, N. Baghdadi, G. Briant, T.A. Black etal. 2011a. Assessing eddy-covariance flux tower
and E. Bartholome. 2014. Mapping of central Africa location bias across the Fluxnet-Canada Research Network
forested wetlands using remote sensing. IEEE J. Sel. based on remote sensing and footprint modelling. Agric. For.
Topics Appl. Earth Observ. 7:531542. doi:10.1109/ Meteorol. 151:87100. doi:10.1016/j.agrformet.2010.09.005.
JSTARS.2013.2269733. Chen, M., Q. Zhuang, D.R. Cook, R. Coulter, M. Pekour, R.L.
Boelman, N.T., A.V. Rocha, and G.R. Shaver. 2011. Understanding Scott, J.W. Munger, and K. Bible. 2011b. Quantification of
burn severity sensing in Arctic tundra: Exploring vegeta- terrestrial ecosystem carbon dynamics in the conterminous
tion indices, suboptimal assessment timing and the impact United States combining a process-based biogeochemical
of increasing pixel size. Int. J. Remote Sens. 32:70337056. model and MODIS and AmeriFlux data. Biogeosciences.
doi:10.1080/01431161.2011.611187. 8:26652688. doi:10.5194/bg-826652011.
Boles, S.H., X.M. Xiao, J.Y. Liu, Q.Y. Zhang, S. Munkhtuya, S.Q. Didan, K. and A. Huete. 2006. MODIS Vegetation Index Product
Chen, and D. Ojima. 2004. Land cover characterization of Series Collection 5 Change Summary. http://landweb.nascom.
Temperate East Asia using multi-temporal VEGETATION nasa.gov/QA_WWW/forPage/MOD13_VI_C5_Changes_
sensor data. Remote Sens. Environ. 90:477489. doi:10.1016/j. Document_06_28_06.pdf, last accessed on July 13, 2014.
rse.2004.01.016. Fensholt, R., I. Sandholt, and S. Stisen. 2006. Evaluating MODIS,
Bolton, D.K. and M.A. Friedl. 2013. Forecasting crop yield using MERIS, and VEGETATIONVegetation indices using
remotely sensed vegetation indices and crop phenol- in situ measurements in a semiarid environment. IEEE
ogy metrics. Agric. For. Meteorol. 173:7484. doi:10.1016/ Trans. Geosci. Remote Sens. 44:17741786. doi:10.1109/
j.agrformet.2013.01.007. TGRS.2006.875940.
Cabacinha, C.D. and S.S. de Castro. 2009. Relationships between Feret, J.-B., C. Franois, G.P. Asner, A.A. Gitelson, R.E. Martin,
floristic diversity and vegetation indices, forest struc- L.P.R. Bidel, S.L. Ustin, M. le, Guerric, and S. Jacquemoud.
ture and landscape metrics of fragments in Brazilian 2008. PROSPECT-4 and 5: Advances in the leaf optical prop-
Cerrado. Forest Ecol. Manage. 257:21572165. doi:10.1016/ erties model separating photosynthetic pigments. Remote
j.foreco.2009.02.030. Sens. Environ. 112:30303043. doi:10.1016/j.rse.2008.02.012.
2016 by Taylor & Francis Group, LLC
172 Remotely Sensed Data Characterization, Classification, and Accuracies
Ferreira, L.G., H. Yoshioka, A. Huete, and E.E. Sano. 2003. Seasonal Ichii, K., H. Hashimoto, M.A. White, C. Potters, L.R. Hutyra, A.R.
landscape and spectral vegetation index dynamics in the Huete, R.B. Myneni, and R.R. Nemanis. 2007. Constraining
Brazilian Cerrado: An analysis within the Large-Scale Biosphere- rooting depths in tropical rainforests using satellite data
Atmosphere Experiment in Amaznia (LBA). Remote Sens. and ecosystem modeling for accurate simulation of gross
Environ. 87:534550. doi:10.1016/j.rse.2002.09.003. primary production seasonality. Global Change Biol.
Friedl, M.A., D. Sulla-Menashe, B. Tan, A. Schneider, N. 13:6777. doi:10.1111/j.13652486.2006.01277.x.
Ramankutty, A. Sibley, and X. Huang. 2010. MODIS Irons, J.R., J.L. Dwyer, and J.A. Barsi. 2012. The next Landsat sat-
Collection 5 global land cover: Algorithm refinements and ellite: The Landsat Data Continuity Mission. Remote Sens.
characterization of new datasets. Remote Sens. Environ. Environ. 122:1121.
114:168182. doi:10.1016/j.rse.2009.08.016. Jaegermeyr, J., D. Gerten, W. Lucht, P. Hostert, M. Migliavacca,
Guanter, L., Y. Zhang, M. Jung, J. Joiner, M. Voigt, J.A. Berry, and R. Nemani. 2014. A high-resolution approach to esti-
C. Frankenberg et al. 2014. Global and time-resolved mating ecosystem respiration at continental scales using
monitoring of crop photosynthesis with chlorophyll fluo- operational satellite data. Global Change Biol. 20:11911210.
rescence. Proc. Natl. Acad. Sci. U. S. A. 111:E1327E1333. doi:10.1111/gcb.12443.
doi:10.1073/pnas.1320008111. Ji, L. and K. Gallo. 2006. An agreement coefficient for image com-
Guerschman, J.P., A.I.J.M. Van Dijk, G. Mattersdorf, J. Beringer, parison. Photogramm. Eng. Remote Sens. 72:823833.
L.B. Hutley, R. Leuning, R.C. Pipunic, and B.S. Sherman. Jiang, Z., A.R. Huete, K. Didan, and T. Miura. 2008. Development
2009. Scaling of potential evapotranspiration with MODIS of a two-band enhanced vegetation index without a blue
data reproduces flux observations and catchment water bal- band. Remote Sens. Environ. 112:38333845. doi:10.1016/
ance observations across Australia. J. Hydrol. 369:107119. j.rse.2008.06.006.
doi:10.1016/j.jhydrol.2009.02.013. John, R., J. Chen, N. Lu, K. Guo, C. Liang, Y. Wei, A. Noormets,
Hengl, T., N. Toomanian, H.I. Reuter, and M.J. Malakouti. 2007. K. Ma, and X. Han. 2008. Predicting plant diversity based
Methods to interpolate soil categorical variables from profile on remote sensing products in the semi-arid region of
observations: Lessons from Iran. Geoderma. 140:417427. Inner Mongolia. Remote Sens. Environ. 112:20182032.
doi:10.1016/j.geoderma.2007.04.022. doi:10.1016/j.rse.2007.09.013.
Holben, B.N., D. Tanr, A. Smirnov, T.F. Eck, I. Slutsker, N. Kaufman, Y.J. and D. Tanr. 1992. Atmospherically Resistant
Abuhassan, W.W. Newcomb et al. 2001. An emerging Vegetation Index (ARVI) for EOS-MODIS. IEEE Trans.
ground-based aerosol climatology: Aerosol Optical Depth Geosci. Remote Sens. 30(2):261270.
from AERONET. J. Geophys. Res. 106:1206712097. Kim, Y., A.R. Huete, T. Miura, and Z. Jiang. 2010. Spectral com-
Houborg, R.M. and H. Soegaard. 2004. Regional simulation of patibility of vegetation indices across sensors: A band
ecosystem CO2 and water vapor exchange for agricultural decomposition analysis with Hyperion data. J. Appl. Remote
land using NOAA AVHRR and Terra MODIS satellite data. Sens. 4:043520. doi:10.1117/1.3400635.
Application to Zealand, Denmark. Remote Sens. Environ. Kim, Y., J.S. Kimball, K. Didan, and G.M. Henebry. 2014. Response
93:150167. doi:10.1016/j.rse.2004.07.001. of vegetation growth and productivity to spring climate
Huete, A., K. Didan, T. Miura, E.P. Rodriguez, X. Gao, and indicators in the conterminous United States derived from
L.G. Ferreira. 2002. Overview of the radiometric and satellite remote sensing data fusion. Agric. For. Meteorol.
biophysical performance of the MODIS vegetation indi- 194:132143.
ces. Remote Sens. Environ. 83:195213. doi:10.1016/ Kobayashi, H., R. Suzuki, and S. Kobayashi. 2007. Reflectance sea-
S0034-4257(02)00096-2. sonality and its relation to the canopy leaf area index in an
Huete, A.R. 1988. A soil-adjusted vegetation index (SAVI). eastern Siberian larch forest: Multi-satellite data and radia-
Remote Sens. Environ. 25:295309. tive transfer analyses. Remote Sens. Environ. 106:238252.
Huete, A.R., K. Didan, Y.E. Shimabukuro, P. Ratana, S.R. Saleska, doi:10.1016/j.rse.2006.08.011.
L.R. Hutyra, W.Z. Yang, R.R. Nemani, and R. Myneni. 2006. Kotchenova, S.Y. and E.F. Vermote. 2007. Validation of a vector
Amazon rainforests green-up with sunlight in dry season. version of the 6S radiative transfer code for atmospheric cor-
Geophys. Res. Lett. 33:L06405. rection of satellite data. Part II. Homogeneous Lambertian
Huete, A.R., H.Q. Liu, K. Batchily, and W.J. van Leeuwen. 1997. and anisotropic surfaces. Appl. Opt. 46:44554464.
A comparison of vegetation indices over a global set of TM Li, Z., G. Yu, X. Xiao, Y. Li, X. Zhao, C. Ren, L. Zhang, and
images for EOS-MODIS. Remote Sens. Environ. 59:440451. Y. Fu. 2007. Modeling gross primary production of alpine
doi:10.1016/S00344257(96)001125. ecosystems in the Tibetan Plateau using MODIS images
Huettich, C., U. Gessner, M. Herold, B.J. Strohbach, M. and climate data. Remote Sens. Environ. 107:510519.
Schmidt, M. Keil, and S. Dech. 2009. On the suitability doi:10.1016/j.rse.2006.10.003.
of MODIS time series metrics to map vegetation types Liu, J., E. Pattey, and G. Jego. 2012. Assessment of vegetation indi-
in dry savanna ecosystems: A case study in the Kalahari ces for regional crop green LAI estimation from Landsat
of NE Namibia. Remote Sens. 1:620643. doi:10.3390/ images over multiple growing seasons. Remote Sens.
rs1040620. Environ. 123:347358. doi:10.1016/j.rse.2012.04.002.
2016 by Taylor & Francis Group, LLC
Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices 173
Mendez-Barroso, L.A., E.R. Vivoni, C.J. Watts, and J.C. Rodriguez. Pearlman, J.S., P.S. Barry, C.C. Segal, J. Shepanski, D. Beiso, and
2009. Seasonal and interannual relations between precipita- S.L. Carman. 2003. Hyperion, a space-based imaging spec-
tion, surface soil moisture and vegetation dynamics in the trometer. IEEE Trans. Geosci. Remote Sens. 41:11601173.
North American monsoon region. J. Hydrol. 377:5970. Pfeifer, M., A. Gonsamo, M. Disney, P. Pellikka, and R. Marchant.
doi:10.1016/j.jhydrol.2009.08.009. 2012. Leaf area index for biomes of the Eastern Arc
Middleton, E.M., P.K.E. Campbell, S.G. Ungar, L. Ong, Q. Mountains: Landsat and SPOT observations along pre-
Zhang, K.F. Huemmrich, D.J. Mandl, and S.W. Frye. 2010. cipitation and altitude gradients. Remote Sens. Environ.
Using EO-1 Hyperion images to prototype environmen- 118:103115. doi:10.1016/j.rse.2011.11.009.
tal products for HyspIRI. In 2010 IEEE International Pocewicz, A., L.A. Vierling, L.B. Lentile, and R. Smith. 2007.
Geoscience and Remote Sensing Symposium (IGARSS), View angle effects on relationships between MISR vegeta-
June 2530, Honolulu, HI. pp. 42564259. doi:10.1109/ tion indices and leaf area index in a recently burned pon-
IGARSS.2010.5648946. derosa pine forest. Remote Sens. Environ. 107:322333.
Miura, T., A.R. Huete, H. Yoshioka, and B.N. Holben. 2001. An doi:10.1016/j.rse.2006.06.019.
error and sensitivity analysis of atmospheric resistant veg- Ponce Campos, G.E., M.S. Moran, A. Huete, Y. Zhang, C. Bresloff,
etation indices derived from dark target-based atmospheric T.E. Huxman, D. Eamus et al. 2013. Ecosystem resilience
correction. Remote Sens. Environ. 78:284298. doi:10.1016/ despite large-scale altered hydroclimatic conditions. Nature.
S00344257(01)002231. 494:349352. doi:10.1038/nature11836.
Miura, T., J.P. Turner, and A.R. Huete. 2013. Spectral compatibil- Qi, J., A. Chehbouni, A.R. Huete, Y.H. Kerr, and S. Sorooshian.
ity of the NDVI across VIIRS, MODIS, and AVHRR: An 1994. A modified soil adjusted vegetation index. Remote
analysis of atmospheric effects using EO-1 Hyperion. IEEE Sens. Environ. 48:119126.
Trans. Geosci. Remote Sens. 51:13491359. doi:10.1109/ Rahman, A.F., D.A. Sims, V.D. Cordova, and B.Z. El-Masri.
TGRS.2012.2224118. 2005. Potential of MODIS EVI and surface temperature for
Miura, T. and H. Yoshioka. 2011. Hyperspectral data in long-term, directly estimating per-pixel ecosystem C fluxes. Geophys.
cross-sensor continuity studies. In Hyperspectral Remote Res. Lett. 32:L19404. doi:10.1029/2005GL024127.
Sensing of Vegetation. P.S. Thenkabail, J.G. Lyon, and A. Rocha, A.V. and G.R. Shaver. 2009. Advantages of a two band
Huete, eds. Taylor & Francis, New York. pp. 611633. EVI calculated from solar and photosynthetically active
Miura, T., H. Yoshioka, K. Fujiwara, and H. Yamamoto. 2008. Inter- radiation fluxes. Agric. For. Meteorol. 149:15601563.
comparison of ASTER and MODIS surface reflectance and doi:10.1016/j.agrformet.2009.03.016.
vegetation index products for synergistic applications to Rocha, A.V. and G.R. Shaver. 2011. Burn severity influences post-
natural resource and environmental monitoring. Sensors. fire CO2 exchange in arctic tundra. Ecol. Appl. 21:477489.
8:24802499. doi:10.1890/100255.1.
Moran, M.S., G.E. Ponce-Campos, A. Huete, M.P. McClaran, Y.Zhang, Saleska, S.R., K. Didan, A.R. Huete, and H.R. da Rocha. 2007.
E.P. Hamerlynck, D.J. Augustine et al.. 2014. Functional Amazon forests green-up during 2005 drought. Science.
response of U.S. grasslands to the early 21st-century drought. 318:612612. doi:10.1126/science.1146663.
Ecology. 95:21212133. doi:10.1890/13-1687.1. Schubert, P., F. Lagergren, M. Aurela, T. Christensen, A. Grelle,
Morton, D.C., J. Nagol, C.C. Carabajal, J. Rosette, M. Palace, B.D. Cook, M. Heliasz, L. Klemedtsson, A. Lindroth, K. Pilegaard,
E.F. Vermote, D.J. Harding, and P.R.J. North. 2014. Amazon for- T. Vesala, and L. Eklundh. 2012. Modeling GPP in the
ests maintain consistent canopy structure and greenness during Nordic forest landscape with MODIS time series data
the dry season. Nature. 506:221224. doi:Letter. Comparison with the MODIS GPP product. Remote Sens.
Nagler, P.L., R.L. Scott, C. Westenburg, J.R. Cleverly, E.P. Glenn, Environ. 126:136147. doi:10.1016/j.rse.2012.08.005.
and A.R. Huete. 2005. Evapotranspiration on western U.S. Sims, D.A., A.F. Rahman, V.D. Cordova, B.Z. El-Masri, D.D.
rivers estimated using the Enhanced Vegetation Index from Baldocchi, P.V. Bolstad, L.B. Flanagan et al. 2008. A new
MODIS and data from eddy covariance and Bowen ratio flux model of gross primary productivity for North American
towers. Remote Sens. Environ. 97:337351. doi:10.1016/j. ecosystems based solely on the enhanced vegetation index
rse.2005.05.011. and land surface temperature from MODIS. Remote Sens.
OConnell, J., J. Connolly, E.F. Vermote, and N.M. Holden. 2013. Environ. 112:16331646. doi:10.1016/j.rse.2007.08.004.
Radiometric normalization for change detection in peat- Sims, D.A., A.F. Rahman, E.F. Vermote, and Z. Jiang. 2011.
lands: A modified temporal invariant cluster approach. Int. Seasonal and inter-annual variation in view angle effects on
J. Remote Sens. 34:29052924. doi:10.1080/01431161.2012. MODIS vegetation indices at three forest sites. Remote Sens.
752886. Environ. 115:31123120. doi:10.1016/j.rse.2011.06.018.
Obata, K., T. Miura, H. Yoshioka, and A.R. Huete. 2013. Sjostrom, M., J. Ardo, A. Arneth, N. Boulain, B. Cappelaere,
Derivation of a MODIS-compatible enhanced vegetation L. Eklundh, A. de Grandcourt et al. 2011. Exploring the
index from visible infrared imaging radiometer suite spec- potential of MODIS EVI for modeling gross primary pro-
tral reflectances using vegetation isoline equations. J. Appl. duction across African ecosystems. Remote Sens. Environ.
Remote Sens. 7:073467. doi:10.1117/1.JRS.7.073467. 115:10811089. doi:10.1016/j.rse.2010.12.013.
2016 by Taylor & Francis Group, LLC
174 Remotely Sensed Data Characterization, Classification, and Accuracies
Son, N.T., C.F. Chen, C.R. Chen, L.Y. Chang, H.N. Duc, and L.D. Xiao, X.M., Q.Y. Zhang, D. Hollinger, J. Aber, and B. Moore.
Nguyen. 2013. Prediction of rice crop yield using MODIS 2005a. Modeling gross primary production of an evergreen
EVI-LAI data in the Mekong Delta, Vietnam. Int. J. Remote needleleaf forest using modis and climate data. Ecol. Appl.
Sens. 34:72757292. doi:10.1080/01431161.2013.818258. 15:954969. doi:10.1890/040470.
Song, Y., J.B. Njoroge, and Y. Morimoto. 2013. Drought impact Xiao, X., Q. Zhang, S. Saleska, L. Hutyra, C. De, Plinio, S. Wofsy,
assessment from monitoring the seasonality of vegetation S. Frolking, S. Boles, M. Keller, and B.I.I.I. Moore. 2005b.
condition using long-term time-series satellite images: Satellite-based modeling of gross primary production in
A case study of Mt. Kenya region. Environ. Monit. Assess. a seasonally moist tropical evergreen forest. Remote Sens.
185:41174124. doi:10.1007/s10661-012-2854-z. Environ. 94:105122.
Soudani, K., C. Francois, G. le Maire, V. Le Dantec, and E. Yamamoto, H., T. Miura, and S. Tsuchida. 2012. Advanced
Dufrene. 2006. Comparative analysis of IKONOS, SPOT, Spaceborne Thermal Emission and Reflection Radiometer
and ETM+ data for leaf area index estimation in temper- (ASTER) Enhanced Vegetation Index (EVI) products from
ate coniferous and deciduous forest stands. Remote Sens. Global Earth Observation (GEO) Grid: An assessment
Environ. 102:161175. doi:10.1016/j.rse.2006.02.004. using Moderate Resolution Imaging Spectroradiometer
Sterckx, S., S. Livens, and S. Adriaensen. 2013. Rayleigh, deep convec- (MODIS) for synergistic applications. Remote Sens. 4:2277
tive clouds, and cross-sensor desert vicarious calibration vali- 2293. doi:10.3390/rs4082277.
dation for the PROBA-V mission. IEEE Trans. Geosci. Remote Yang, F., K. Ichii, M.A. White, H. Hashimoto, A.R. Michaelis, P.Votava,
Sens. 51:14371452. doi:10.1109/TGRS.2012.2236682. A.-X. Zhu, A. Huete, S.W. Running, and R.R.Nemani. 2007.
Swinnen, E. and F. Veroustraete. 2008. Extending the SPOT- Developing a continental-scale measure of gross primary pro-
VEGETATION NDVI time series (19982006) back in time duction by combining MODIS and AmeriFlux data through
with NOAA-AVHRR data (19851998) for Southern Africa. Support Vector Machine approach. Remote Sens. Environ.
IEEE Trans. Geosci. Remote Sens. 46:558572. doi:10.1109/ 110:109122. doi:10.1016/j.rse.2007.02.016.
TGRS.2007.909948. Yang, F., M.A. White, A.R. Michaelis, K. Ichii, H. Hashimoto, P. Votava,
USGS. 2011. EO-1 (Earth Observing-1). http://eros.usgs.gov/#/ A.-X. Zhu, and R.R. Nemani. 2006. Prediction of continental-
Find_Data/Products_and_Data_Available/ALI [Jan 31, 2012]. scale evapotranspiration by combining MODIS and AmeriFlux
Vargas, M., T. Miura, N. Shabanov, and A. Kato. 2013. An initial data through support vector machine. IEEE Trans. Geosci.
assessment of Suomi NPP VIIRS vegetation index EDR. J. Remote Sens. 44:34523461. doi:10.1109/TGRS.2006.876297.
Geophys. Res. Atmos. 118:116. doi:10.1002/2013JD020439. Yang, X., J.F. Mustard, J. Tang, and H. Xu. 2012. Regional-scale
Vermote, E., D. Tanr, J.L. Deuz, M. Herman, J.J. Morcrette, and phenology modeling based on meteorological records
S.Y. Kotchenova. 2006. Second Simulation of a Satellite Signal and remote sensing observations. J. Geophys. Res. Biogeo.
in the Solar SpectrumVector (6SV) User Guide Version 3. 117:G03029. doi:10.1029/2012JG001977.
Waring, R.H., K.S. Milner, W.M. Jolly, L. Phillips, and D. Yebra, M., A. Van Dijk, R. Leuning, A. Huete, and J.P. Guerschman.
McWethy. 2006. Assessment of site index and forest growth 2013. Evaluation of optical remote sensing to estimate
capacity across the Pacific and Inland Northwest USA with actual evapotranspiration and canopy conductance. Remote
a MODIS satellite-derived vegetation index. Forest Ecol. Sens. Environ. 129:250261. doi:10.1016/j.rse.2012.11.004.
Manage. 228:285291. doi:10.1016/j.foreco.2006.03.019. Yoshioka, H. 2004. Vegetation isoline equations for an atmo-
Wittenberg, L., D. Malkinson, O. Beeri, A. Halutzy, and N. Tesler. 2007. sphere-canopy-soil system. IEEE Trans. Geosci. Remote Sens.
Spatial and temporal patterns of vegetation recovery following 42:166175.
sequences of forest fires in a Mediterranean landscape, Mt. Carmel Yoshioka, H., T. Miura, and K. Obata. 2012. Derivation of rela-
Israel. CATENA. 71:7683. doi:10.1016/j.catena.2006.10.007. tionships between spectral vegetation indices from mul-
Wolfe, R.E., M. Nishihama, A.J. Fleig, J.A. Kuyper, D.P. Roy, J.C. tiple sensors based on vegetation isolines. Remote Sens.
Storey, and F.S. Patt. 2002. Achieving sub-pixel geolocation 4:583597. doi:10.3390/rs4030583.
accuracy in support of MODIS land science. Remote Sens. Zhang, X.Y., M.A. Friedl, and C.B. Schaaf. 2006. Global veg-
Environ. 83:3149. doi:10.1016/S0034-4257(02)00085-8. etation phenology from Moderate Resolution Imaging
Wu, C. 2012. Use of a vegetation index model to estimate gross Spectroradiometer (MODIS): Evaluation of global patterns
primary production in open grassland. J. Appl. Remote Sens. and comparison with insitu measurements. J. Geophys. Res.
6:063532. doi:10.1117/1.JRS.6.063532. 111:G04017. doi:10.1029/2006JG000217.
Xiao, X., B. Braswell, Q. Zhang, S. Boles, S. Frolking, and Zhang, Y., M. Susan Moran, M.A. Nearing, G.E. Ponce Campos,
B.I.I.I. Moore. 2003. Sensitivity of vegetation indices to A.R. Huete, A.R. Buda, D.D. Bosch et al. 2013. Extreme
atmospheric aerosols: Continental-scale observations in precipitation patterns and reductions of terrestrial ecosys-
Northern Asia. Remote Sens. Environ. 84:385392. tem production across biomes. J. Geophys. Res. Biogeosci.
Xiao, X.M., D. Hollinger, J. Aber, M. Goltz, E.A. Davidson, Q.Y. 118:148157. doi:10.1029/2012JG002136.
Zhang, and B. Moore. 2004. Satellite-based modeling of gross Zhao, D., K.R. Reddy, V.G. Kakani, J.J. Read, and S. Koti. 2007. Canopy
primary production in an evergreen needleleaf forest. Remote reflectance in cotton for growth assessment and lint yield pre-
Sens. Environ. 89:519534. doi:10.1016/j.rse.2003.11.008. diction. Eur. J. Agron. 26:335344. doi:10.1016/j.eja.2006.12.001.
2016 by Taylor & Francis Group, LLC
8
Toward Standardization
of Vegetation Indices
Acronyms and Definitions .................................................................................................................175
8.1 Introduction: Vegetation Indices and Their Uses ...............................................................176
Vegetation Index Applications
8.2 The Need for Standards...........................................................................................................178
Vegetation Index Formula Alternatives to Standard Vegetation Indices
8.3 Sources of Variation in Vegetation Indices..........................................................................179
Soil Background Atmospheric Effects Directional Effects Pixel-Size Effects Spectral
Michael D. Steven Band Effects
University of Nottingham 8.4 Vegetation Index Intercalibration Approaches ...................................................................181
Intercalibration of Vegetation Indices after Steven etal. (2003) Validation of Cross-Sensor
Timothy J. Malthus Conversion
Commonwealth Scientific and 8.5 Vegetation Index Standardization ........................................................................................187
Industrial Research Organisation Proposals for Vegetation Index Standards Limits to Standardization
Frdric Baret 8.6 Discussion .................................................................................................................................188
National Institute of 8.7 Conclusions.............................................................................................................................. 190
Agronomic Research References............................................................................................................................................ 190
AcronymsandDefinitions ISO International Organization for Standardization
Kompsat Korean Multipurpose Satellite
6S Second Simulation of a Satellite Signal in the LAI Leaf area index
Solar Spectruman atmospheric correction Landsat A system of earth observation satellites in
model operation since 1972
AATSR Advanced Along-Track Scanning Radiometer LISS Linear Imaging Self-Scanning Sensor (IRS)
ASTER Advanced Spaceborne Thermal Emission and MERIS MEdium-Resolution Imaging Spectrometer
Reflection Radiometer MISR Multi-angle Imaging SpectroRadiometer
ATSR2 Along-Track Scanning Radiometer 2 MODIS MOderate-Resolution Imaging Spectrometer
AVHRR Advanced very-high-resolution radiometer MODIS-TIP Two-stream inversion package applied to
BRDF Bidirectional reflectance distribution function MODIS products by the European Joint
CEOS Committee on Earth Observation Satellites Research Centre
CHRIS Compact High-Resolution Imaging MODTRAN MODerate-resolution atmospheric
Spectrometer on the PROBA satellite TRANsmissionan atmospheric correction
DMC Disaster Monitoring Constellation model
DN Digital number MSAVI Modified Soil-Adjusted Vegetation Index
ETM+ Enhanced Thematic Mapper (Landsat) MSG/SEVIRI Meteosat Second Generation/Spinning
EVI Enhanced Vegetation Index Enhanced Visible and Infrared Imager
fAPAR The fraction of photosynthetically active radi- MSS Multispectral Scanner (Landsat)
ation absorbed by a vegetation canopy NASA National Aeronautics and Space
Formosat Name of a commercial high-resolution Administration
satellite NDVI Normalized difference vegetation index
IKONOS Name of a commercial high-resolution satellite NOAA National Oceanic and Atmospheric
IRS Indian Remote Sensing Satellite Administration
175
2016 by Taylor & Francis Group, LLC
176 Remotely Sensed Data Characterization, Classification, and Accuracies
OrbView Name of a commercial high-resolution 50
satellite
40
OSAVI Optimized Soil-Adjusted Vegetation Index
Reflectance (%)
PAR Photosynthetically Active Radiation 30
POLDER POLarization and Directionality of the Earths
Reflectancesan optical imaging radiometer 20
PROSAIL A combination of the PROSPECT leaf optical
10
properties model with the SAIL canopy bidi-
rectional reflectance model 0
PROSPECT A model of leaf optical properties 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400
QuickBird Name of a commercial high-resolution Wavelength (nm)
satellite
SAIL Scattering by Arbitrarily Inclined Leavesa FIg u r e 8.1 Laboratory-measured reflectance spectrum of a bean
canopy bidirectional reflectance model leaf. (From the Author.) The jitters at longer wavelengths are instru-
SAVI Soil-Adjusted Vegetation index mental noise.
SeaWIFS Sea-Viewing Wide Field-of-View Sensor
of the NIR and visible bands, of which the best known by far
SPOT Systme Probatoire dObservation de la Terre
(although by far from the best) is the Normalized Difference
SPOT-VEG The VEGETATION sensor on the SPOT
Vegetation Index (NDVI), defined as
satellite
TM Thematic Mapper (Landsat)
NIR red
TSAVI Transformed Soil-Adjusted Vegetation Index NDVI = (8.1)
VEGETATION 1km resolution monitoring instrument on the NIR + red
SPOT satellite
Vens Vegetation and Environment monitoring on a where is the spectral reflectance (dimensionless) in the NIR or
New MicroSatellite red spectral band. The NDVI is functionally equivalent to the
VI Vegetation index simple ratio of the two bands (Perry and Lautenshlager 1984).
However, its formulation ensures that the value of the NDVI
8.1 introduction: Vegetation ranges strictly from 1 to +1, which is computationally more
convenient than the ratio, which has no upper bound. A wide
indices and their Uses variety of alternative formulations have been used, for exam-
The concept of the vegetation index (VI) is one of the lasting ple, the Soil-Adjusted Vegetation Index (SAVI) (Huete 1988),
success stories of terrestrial remote sensing. The physiological definedas
and anatomical characteristics of vegetation give rise to distinc-
tive spectral features that allow its presence to be detected in any (NIR red )
SAVI = (1 + L) (8.2)
environment and with suitable precautions permit the proper- (NIR + red + L)
ties of the vegetation canopy to be inferred from the reflected
spectrum. Healthy vegetation absorbs visible (especially red) The formulation of SAVI differs from NDVI by including a
light via chlorophyll and other pigments. In the near-infrared soil calibration factor L that adjusts for variability in the index
(NIR) where no absorbers are active, light is strongly reflected introduced by soil reflectance characteristics. Although Huete
by foliage because the juxtaposition of cells, essentially contain- (1988) found that the optimal value of L varied with vegetation
ing water, with air spaces between, creates a strongly scatter- density, a midrange value of 0.5 was found to provide effective
ing medium (Gates etal. 1965; Gausman and Allen 1973). The correction for variations due to soil background across the full
resulting reflectance spectrum (Figure 8.1) is highly character- range of densities. Later variants on this approach varied the
istic and is recognizable as the spectral signature of vegetation value of L, with Rondeaux et al. (1996) proposing Optimized
even when distorted by other environmental variables. When SAVI (OSAVI) with L = 0.16 on the basis of an optimization
used to monitor growing vegetation, the spectral signal of veg- across a range of agricultural soils. Others include the Modified
etation is mixed with that of soil or other backgrounds, which Soil-Adjusted Vegetation Index (MSAVI) (Qi etal. 1994), which
tend to show a much flatter response across this spectral region. employs a self-adjusted value of L based on the spectral reflec-
The result is that the NIR tends to increase with vegetation cover, tance data themselves, and the Transformed Soil-Adjusted
for example as a crop grows, while the red reflectance decreases, Vegetation Index (TSAVI) (Baret and Guyot 1991), where L is
ending at 100% cover with a spectrum close (but not identical) based on information about the soil characteristics. Most VIs
to that of a single leaf. are reformulations based on the same two spectral bands, but
The key factor in the development of VIs is the increasing con- some, such as the Enhanced Vegetation Index (EVI) (Huete
trast between the reflectance in the two bands. To encapsulate etal. 1997), introduce additional spectral information, usually
this contrast, early work identified a variety of combinations in an attempt to reduce atmospheric sensitivity. Each index has
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 177
its advocates and many have particular merits, but key features 0.8
are that they are intrinsically dimensionless and that they all 0.7
ultimately share the characteristic of using the contrast between
0.6
NIR and red reflectance as the primary measure of vegetation.
For most of this discussion, the term vegetation index (VI) will 0.5
OSAVI
be used generically to apply to all of them. 0.4
0.3
8.1.1 Vegetation index Applications 0.2
The applications of VIs are based on their ability to measure foli- 0.1
age density in a consistent manner across a wide range of vegeta-
0
tion types. In early field studies VIs were successfully related to 0 1 2 3 4
leaf area index (LAI) defined as the total (single-sided) area of Leaf area index
leaf per unit area of ground (m2/m2, i.e., dimensionless), canopy
chlorophyll (g m2), wet and dry biomass (g m2), the fraction of FIg ur e 8.3 Sensitivity of OSAVI to leaf angle distribution for a range
ground covered by leaves (dimensionless), primary productivity of leaf area indices, for ellipsoidal distributions with mean leaf angles of
(g m2 day1), the fraction of photosynthetically active radiation 30 (), 45 (), 57.3 (), and (65) +. All the parameters are dimension-
absorbed by a vegetation canopy (fAPAR) (dimensionless), and less. (Adapted from Steven, M.D., Remote Sens. Environ., 63, 49, 1998.)
other variables (Tucker 1977; Tucker etal. 1979; Holben etal. 1980;
Steven etal. 1983). As all of these factors are measures of foliage
density, they tend to be highly intercorrelated in any individual 0.8
study. However, these variables are hierarchically linked by the 0.7
process of canopy photosynthesis that converts absorbed PAR to
0.6
fixed energy in biomass (Figure 8.2). The reflection of light by a
plant canopy is largely determined by the total area of leaf and 0.5
OSAVI
the projections of that area toward both the source of illumina- 0.4
tion and the detection device. This introduces a dependence on
0.3
leaf angle distribution (Verhoef 1985) and to a lesser extent other
factors such as clumping (Gower etal. 1999). As they involve sim- 0.2
ilar projections, this is closely connected to the way in which leaf 0.1
area and angle combine to determine leaf cover fraction or PAR
0
absorption (Steven etal. 1986), so that as indicated in Figure 8.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2
the response of a VI relates most directly to these variables. An Leaf cover fraction
illustration of this link is that in spite of variations with time
of day, Pinter (1993) found that the relationship of fAPAR with FIgu r e 8.4 Sensitivity of OSAVI cover estimates to leaf angle dis-
tribution. Parameters as for Figure 8.3. (Adapted from Steven, M.D.,
Remote Sens. Environ., 63, 49, 1998.)
Leaf angle Leaf area index Chlorophyll
distribution (LAI; m2/m2) (g m2)
various VIs was independent of illumination angle. Figure8.3
shows the modeled relationship between OSAVI (representative
Leaf cover here of VIs generally) and LAI for canopies with different leaf
fAPAR
fraction
angle distributions, while Figure8.4 shows that the relationships
essentially collapse onto a single curve when expressed as func-
tions of leaf cover fraction (Steven 1998). The same behavior can
Primary be found with NDVI, where the merger of curves for different
Vegetation
productivity
index
(g m2 day1)
leaf angle distributions is slightly less tight, although not enough
to introduce serious variations. The implication of this merger is
that there is redundancy in the biophysical parameters and that
Biomass VIs can be used to measure foliage density in a way that is largely
(g m2) independent of canopy structure and, to some extent, species.
Moreover as leaf cover fraction is itself closely related, through
FIgu r e 8.2 Hierarchical diagram showing the relationships similar dependence on leaf area and structure, to the intercep-
between various measures of canopy density and their link with vegeta- tion of solar radiation by plant canopies (Steven etal. 1986), VIs
tion indices. lend themselves to the estimation of light capture by vegetation
2016 by Taylor & Francis Group, LLC
178 Remotely Sensed Data Characterization, Classification, and Accuracies
(or fAPAR). Following Monteith (1977) who established the A complementary approach is to combine data from more
model outlined in Figure 8.2, showing that light capture was the than one system, sometimes referred to as the use of a virtual
key determinant of the conversion of solar energy to biomass, constellation (CEOS 2006; Martnez-Beltrn etal. 2009). Long-
remote sensing of this variable leads to direct estimation of pro- term environmental analysis may also require meta-analysis of
ductivity in crops (Steven etal. 1983; Wiegand etal. 1991). The data from a range of systems (Boyd and Foody 2011). Key to these
same argument allows VIs to be applied in large-scale monitor- approaches is the adoption of a set of operating standards for
ing of vegetation dynamics (Goward et al. 1993), with a range the systems to be combined. The increasing focus on long-term
of applications in agriculture (Maselli et al. 2000) or ecology continuity of vegetation observations, particularly for monitor-
(Pettorelli etal. 2005). Direct estimation of fAPAR on a global ing environmental change at larger scales, has led to consider-
scale is now the focus of several major space programs, using able interest in back-calibrating data from earlier systems such
the MOderate-Resolution Imaging Spectrometer (MODIS), as Landsat (operational since 1972) and AVHRR (since 1978),
MEdium-Resolution Imaging Spectrometer (MERIS), Sea- as near as possible to current standards, to establish the long-
Viewing Wide Field-of-View Sensor (SeaWIFS), the two-stream term baseline (Brown etal. 2006; Samain etal. 2006; Swinnen
inversion package applied to MODIS products by the European and Veroustrate 2008). Precise calibration of the instruments
Joint Research Centre (MODIS-TIP), the VEGETATION sen- is required and attention to variations in BRDF associated with
sor on the Systme Probatoire dObservation de la Terre satellite different orbital characteristics (Rder etal. 2005; Teillet etal.
(SPOT-VEG), and Advanced Very-High-Resolution Radiometer 2006; Martnez-Beltrn et al. 2009). Thus, standardization is
(AVHRR) systems (Picket-Heaps etal. 2014). required first for individual sensors, to account for variability
in calibration and other observational parameters over time.
8.2 the need for Standards Second, standardization is required between sensors to allow the
interoperability of different systems. Coupled with these instru-
All sciences commence with a lengthy period of exploration, mental issues is the need to specify standard observing condi-
with a diversity of idiosyncratic approaches until the difficulties tions, with unambiguous correction procedures to account for
of rationalizing different methodologies lead to a movement for deviations from the standard. Achievement of these goals would
standardization. The benefits of standardization that are partic- in principle allow consistent monitoring of vegetation across
ularly pertinent to remote sensing are interoperability of systems time and across spatial scales. However, although standards for
and data continuity. The use of standards also helps to generate many activities within the field of remote sensing are being pro-
authority for standardized products and with increasing use, a moted by Committee on Earth Observation Satellites (CEOS),
greater familiarity with their capabilities and limits. International Organization for Standardization (ISO), National
A persistent issue in vegetation monitoring is the acquisition Aeronautics and Space Administration (NASA), and other agen-
of sufficient data to capture the dynamics of plant growth. Plant cies, no concerted attempt has yet been made to standardize VIs.
growth requires water, usually supplied by rainfall, so the more
productive vegetated regions are frequently cloudy, obstruct-
8.2.1 Vegetation index Formula
ing the view of satellite sensors (Heller 1961). The data acqui-
sition problem for satellite systems has broadly been resolved At present, a wide range of VIs are generated by different earth
in two ways: at higher resolutions by the development of sys- observation systems and data users. Le Maire etal. (2004) evalu-
tems such as SPOT or IKONOS with pointable cameras that ated 61 indices for their sensitivity to chlorophyll across different
can target particular sites several times within the satellite types of leaves, and Agapiou etal. (2012) evaluated 71 established
repeat cycle and at lower resolutions with more frequent data indices for distinguishing archaeological crop marks. Neither
acquisition by selectively compositing over periods of 10 days list is exhaustive. As well as highlighting the inadequacies of
or more (Gutman 1991; van Leeuwen etal. 1999). Alternatively, NDVI, the enormous proliferation of indices indicates both the
vegetation monitoring can be performed by aircraft that can fly need for a standard and the difficulty of achieving one. Although
beneath the cloud deck (Jaggard and Clark 1990) or by monitor- a number of indices have been developed that are claimed to
ing systems mounted on mobile farm machinery for precision represent specific properties of vegetationwater content, chlo-
agriculture. All these solutions introduce their own problems. rophyll, carotenoids, and radiation use efficiency (Peuelas and
In particular, both pointing and compositing tend to increase Filella 1998)these usually require high spectral resolution and
the range of viewing angles and to a lesser extent solar angles, are not the concern here. In general, the different broadband
used in the VI product. Monitoring under cloud requires addi- VIs are highly interrelated and are usually used to represent the
tional data to normalize for incident solar irradiance and mea- same vegetation parameters. In most cases, their differences lie
sures different reflectance characteristics corresponding to the in the generality of the data that have been used to test them
multidirectional diffuse illumination of the target. Corrections and the degree to which they can suppress factors extraneous
can be made for angular effects using a model of the bidirec- to the estimation of vegetation characteristics, such as soil or
tional reflectance distribution function (BRDF), but require a atmospheric effects. However, different indices do have distinct
priori knowledge of the vegetation type (Steven 1998; Bacour responses and may yield differing estimates of derived parame-
etal. 2006). ters: Boyd etal. (2011) found that estimates of phenological event
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 179
dates such as the start and end of the growing season differed by 8.3 Sources of Variation
several days when different indices were used, with uncertainties
from these differences as large as those arising from climatic
inVegetation indices
perturbations. Although VIs respond primarily to foliage density, however
expressed, they are critically affected by a range of other factors.
8.2.2 Alternatives to Standard Van Leeuwen (2006) distinguishes between uncertainties related
Vegetation indices to input parameters, VI formulation, and product generation
issues such as compositing rules. Most of the discussion here
While a standard VI would therefore seem desirable, it is none- relates to the first of these categories. Table 8.1 classifies the sources
theless worth discussing the alternatives. If standards are devel- of extraneous variation into environmental, observational, and
oped prematurely, there is a danger of lock-in to a standard that instrumental categories, as discussed further as follows.
proves to be inadequate. To some extent, this is the case with the
NDVI. Many papers have pointed out its excessive sensitivity to
soil characteristics (Huete 1988; Rondeaux etal. 1996), but hav- 8.3.1 Soil Background
ing been adopted as an operational product from the early 1980s Dependence on soil reflectance is inherent in the formulation
(Townshend and Justice 1986), it has become the de facto stan- of VIs as measures of vegetationsoil contrast. With the NDVI,
dard for later systems; a recent web search (June 2014) reported darker soils will tend to amplify the vegetation component of the
over 5000 academic papers on this index. One alternative to signal, while brighter soils will tend to suppress it (Huete 1988;
VIs is to estimate vegetation parameters directly by inverting Rondeaux etal. 1996). Field studies that relate NDVI to measures
a vegetation canopy reflectance model with earth observation of foliage density are almost invariably conducted on a single soil
data. As the basis for productivity estimation, fAPAR has been type and almost inevitably achieve relationships with high cor-
recognized as an essential climate variable for global model- relations; but subsequent attempts to transfer the relationships to
ing and is already a routine product of the MODIS, MERIS, other environments are often disappointing due to changes in the
VEGETATION, and Meteosat Second Generation/Spinning background effect of soil type. Individual soils also decrease in
Enhanced Visible and Infrared Imager (MSG/SEVIRI) sys- brightness with wetting (Bowers and Hanks 1965; Rondeaux etal.
tems. Gobron etal. (2008) evaluated the effects of radiometric 1996). Soil effects in NDVI can be as high as 50% of the dynamic
uncertainties on the MERIS product and estimated that errors range, but are considerably reduced in the SAVI range of indi-
in fAPAR estimation should be 0.1, but DOdorico etal. (2014) ces (Rondeaux etal. 1996), corresponding to a maximum error of
found inconsistencies, particular in forest, when comparing about 0.05 in the estimation of leaf cover fraction.
three fAPAR algorithms over Europe. Similar issues were found
by Picket-Heaps etal. (2014) who tested six alternative fAPAR
products over Australia. They suggest that current fAPAR 8.3.2 Atmospheric effects
products are not reliable enough to be fed into biogeochemical
process models or used in data fusion approaches. A further dif- Atmospheric effects change the radiances measured so that the
ficulty with these products is that they rely on different types top-of-atmosphere reflectances generate a different VI from
of data input including, in most cases, multiangular reflectance that observed at the surface. It is essential to point out here that
data. While greater reliability may be achievable with future to standardize a VI requires the index to be based on surface
systems or better models, it would not be realistic to generate reflectances as defined in Equation 8.1 or 8.2. Any other mea-
the equivalent product from historic data where key inputs are surement, such as top-of-atmosphere reflectance, represents a
missing. However, a VI, although less directly applicable as an combination of signals from both vegetation and atmosphere,
input to process models (it remains just an index), can within but using equivalent surface reflectance data eliminates many
limits be standardized. of the errors and allows comparison of indices (Guyot and Gu
TABLe 8 .1 Sources of Extraneous Variation in Vegetation Indices
Source of Variation Approach to Overcome Key References
Environmental factors
Soil background Soil-adjusted indices Huete (1988) and Rondeaux etal. (1996)
Atmosphere Atmospheric modeling Vermote etal. (1997) and Berk etal. (1998)
Observation parameters
Solar angle Canopy BRDF modeling Jacquemoud etal. (2009) and Vermote etal. (2009)
Viewing angle
Instrumental factors
Pixel size Aggregation, scale modeling Martnez-Beltrn etal. (2009) and Obata etal. (2012)
Spectral bands VI intercalibration Steven etal. (2003); this chapter
2016 by Taylor & Francis Group, LLC
180 Remotely Sensed Data Characterization, Classification, and Accuracies
1994). Measurements must also be calibrated according to best view angle effects may vary with orbital drift of long-term satel-
practice (Price 1987). Many studies in the past have applied lites (Tucker etal. 2005; Brown etal. 2006). Angular effects also
the VI concept rather loosely, some even using uncalibrated occur as a result of the compositing approaches used to gen-
digital numbers (DNs) to compute the index. Such formula- erate cloud-free vegetation products from global datasets (van
tions can provide strong correlations with vegetation density Leeuwen etal. 1999). Where the vegetation characteristics are
measures in individual studies, but cannot easily be compared known, angular effects on VIs can be estimated by canopy mod-
with studies using different instruments or formulations. Zhou els (Shultis 1991; Steven 1998; Verrelst etal. 2008; Jacquemoud
etal. (2009) found large sensor-dependent differences between etal. 2009), and as the solar and viewing angles are known pre-
NDVIs for various systems depending on whether they were cisely for any observation, the residual errors are not large. For
DN based, radiance based, or reflectance based. Hadjimitsis global-scale vegetation monitoring, where it is impractical to
et al. (2010) found a mean difference of 18% between uncor- apply individual models to specific vegetation types, Vermote
rected and atmospherically corrected NDVI values and more etal. (2009) demonstrated that BRDF corrections could be made
modest, but still troublesome, differences in a range of other on the basis of an assumption that the shape of the BRDF varies
indices. Miura et al. (2013) evaluated atmospheric effects on more slowly than the magnitude of the reflectance. The shape
intersensor compatibility and showed that atmospheric correc- can then be quantified by two parameters R and V (related in
tion to top of canopy led to the greatest consistency between sys- broad terms to vegetation and roughness) that can be derived by
tems. Corrections for atmospheric effects to retrieve the surface inversion of the time series.
reflectance are therefore a strict necessity. A variety of empirical BRDF effects can also occur with cloudiness, topography,
and model-based approaches are available (Mahiny and Turner and atmospheric state. These factors change the relative con-
2007). The Second Simulation of a Satellite Signal in the Solar tribution of diffuse solar irradiance that comprises a distri-
Spectrum (6S) (Vermote etal. 1997) and MODerate-resolution bution of incident angles (Steven 1977; Steven and Unsworth
atmospheric TRANsmission (MODTRAN) (Berk et al. 1998) 1980) interacting with different parts of the BRDF and gen-
atmospheric models are widely used, but difficulties can arise erating spectral and angular reflectances different to those
in acquisition of the atmospheric parameters required to make from the unidirectional direct solar irradiance. By compar-
the correction, particularly in remote locations, with uncertain- ing canopy measurements made under full sunlight and
ties associated mainly with aerosol content (Nagol etal. 2009). simulated cloud (entirely diffuse illumination), Steven (2004)
Fortunately, atmospheric effects in VIs are somewhat limited by found differences up to 0.15 in NDVI between canopies in
the ratio construction of most indices coupled with the fact that sunlight and shade, but demonstrated that the canopy spectra
the visible and NIR bands are to a large extent affected in similar under standard conditions could be reconstructed to a pre-
ways by atmospheric aerosol; Steven (1998) found that OSAVI cision of 10%15% from the shaded measurements. Overall,
estimates of leaf cover fraction for mid-latitude summer moni- these studies indicate that although angular responses are
toring conditions were relatively insensitive to quite large errors problematic and complex, they are susceptible to correction
in modeled atmospheric parameters, resulting in no more than by modeling the BRDF.
45% error in the cover estimate.
8.3.4 Pixel-Size effects
8.3.3 Directional effects
A number of studies have commented on the effects of pixel
The complex structure of vegetation canopies as assemblages of aggregation when integrating data from multiple sources (van
leaves and other components suspended above the surface gives Leeuwen 2006; Tarnavsky et al. 2008; Martnez-Beltrn et al.
rise to a more or less complex BRDF that leads to a VI depen- 2009; Munyati and Mboweni 2013). A fundamental issue is
dence on solar and viewing angle. The BRDF of vegetation is that VIs are not linear functions of reflectance, so that spatial
largely controlled by the relative fractions of illuminated soil, averages of the VI are not precisely equivalent to VI values cal-
foliage, and shadow visible in a given direction (Guyot 1990; culated on the basis of averaging the radiances or reflectances
Schluessel etal. 1994). At large incidence angles, the scene will (Figure8.5). The difference is not large in the example shown,
appear to be fully vegetated, even when LAI is small, whereas but would in principle increase with greater heterogeneity of
the nadir view will show a greater proportion of soil and maxi- the surface and with the number of pixels to be aggregated.
mizes sensitivity to leaf area and canopy structure. In general, Obata etal. (2012) developed a theoretical basis for dealing with
variations of reflected radiance with either solar or viewing scaling effects based on monotonic behavior of the effect as a
angle are greatest for erectophile leaf angle distributions and function of spatial resolution. Martnez-Beltrn et al. (2009)
least for planophile distributions. Verrelst etal. (2008) showed indicate that in practice this nonlinearity has not been found
that VIs showed large variations with viewing angle (and, by the to be the major issue and that geolocation uncertainty is a more
principle of reciprocity, solar angle) that are dependent on both serious source of error in intersensor comparison. However,
index and vegetation type. Sims etal. (2011) and Moura etal. Munyati and Mboweni (2013) found that aggregation in areas
(2012) also found that directional effects were index sensitive, of sparse, patchy vegetation could lead to underestimation of
being greater for EVI than for NDVI. If uncorrected, solar and productivity.
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 181
Red NIR the important sources of variation. Simulation from data has the
0.1 0.15 0.2 0.6 0.55 0.5 advantage in this context in that the spectral data are realistic,
0.15 0.2 0.25 0.6 0.55 0.5
the simulated results are not limited to particular instruments,
0.2 0.25 0.3 0.6 0.55 0.5
and the intercomparisons are all made on a common dataset.
Averages 0.200 0.550
NDVI from average = 0.467
8.4.1 intercalibration of Vegetation
indices after Steven etal. (2003)
NDVI
0.71 0.57 0.43 The analysis by Steven etal. (2003, 2007) used a database of high-
0.60 0.47 0.33 resolution spectra of vegetation canopies to simulate NIR and
0.50 0.38 0.25 Average NDVI = 0.471 visible band responses of particular instruments and then com-
pared VIs as measured by different simulated observing systems.
FIg ur e 8.5 Hypothetical illustration on the effect of NDVI nonlin- The database, described in more detail by Steven et al. (2003),
earity on aggregation of nine pixels into one. The mean of the NDVIs comprised a set of 166 nadir-viewing bidirectional reflectance
across a 3 3 pixel area (lower box) is not identical to the NDVI calcu- spectra measured over canopies of contrasting architecture
lated from the mean reflectance values (upper two boxes). (sugar beet and maize) in the United Kingdom and France in
experiments conducted in 1989 and in 1990. The plant canopies
8.3.5 Spectral Band effects had a full range of LAI and soil backgrounds, while contrasting
leaf colors were achieved by treatments with disease or diluted
In addition to variability of the VI itself, any standardization herbicide. Spectral band responses were simulated by convolving
protocol must also account for differences between measure- the top-of-canopy spectral radiance data with the full spectral
ment systems. Even when instruments are precisely calibrated response function of each of the sensors tested and normalizing
and all the proper corrections are applied for BRDF and atmo- with the corresponding convolved data for the reference panel
spheric effects, indices from the various measurement systems used in the field, adjusted for its true reflectance. The spectral
differ systematically due to differences in the position, width, response functions were found in the literature, or from the web,
and shape of the wavebands used (Gallo and Daughtry 1987; or obtained by personal communication and were digitized every
Guyot and Gu 1994). Band placement does not critically affect 1nm to match the spectral data (Figure 8.6). The operators of
the behavior or strength of relationships with vegetation density, the OrbView-2 and OrbView-3 systems were unwilling to release
but can give rise to large differences in the fitted coefficients (Lee the spectral response functions of their instruments, so Steven
etal. 2004; Soudani etal. 2006). Differences in index are inevi- etal. (2007) tested two alternative models: a box function across
table because different sensors measure different parts of the the nominal wavelength range and a Gaussian fitted so that the
vegetationsoil reflectance spectrum and consequently respond nominal waveband limits were the half-power points. The wave-
in differing degrees to the biophysical variables concerned. In bands for Vens, which had not been precisely defined, were also
addition, Teillet and Ren (2008) found that spectral differences modeled with a Gaussian on the basis of the developers advice.
are generated by the spectral dependence of atmospheric gas The simulated band reflectances were then applied to compute
transmittance. However, it transpires that in most cases these
differences are more quantitative than qualitative, with almost 1
perfect correlations between indices recorded by different sys-
tems (Steven etal. 2003, 2007). This study proposes the applica-
0.8
tion of the relationships established in these papers as a key step
in the standardization of VIs.
Relative sensitivity
0.6
8.4 Vegetation index
intercalibration Approaches 0.4 MISR
IRS
Intercomparison of VIs can be performed by direct comparison
0.2
of measurements made by different instruments, by simulation
from model spectra, or by partial simulation (of band responses)
from field reflectance spectra. Direct comparisons are limited to 0
particular instruments and are subject to errors due to nonsimul- 500 600 700 800 900 1000
taneity, inexact image coregistration, and atmospheric effects; Wavelength (nm)
they also fail to distinguish spectral band effects from sensor
calibration errors. Although model spectra are highly versatile, FIgu r e 8.6 Example of spectral response functions for two of the
particularly for sensitivity tests, their direct application requires systems simulated. (Adapted from Steven, M.D. et al., Remote Sens.
supreme confidence that the model parameterization captures all Environ. 88, 412, 2003.)
2016 by Taylor & Francis Group, LLC
182 Remotely Sensed Data Characterization, Classification, and Accuracies
the VIs NDVI, SAVI, and OSAVI for the different systems as well 1
as for a hypothetically ideal narrow-band sensor pair based on
narrow-bands at 670 and 815nm, proposed as a standard. With
0.8
narrow-bands, the reflectance values are no longer sensitive to
bandwidth or spectral response function. However, to reduce
the effects of instrumental noise in the original database, band- 0.6
widths of 20nm centered on the nominal wavelengths were used
CHRIS (L14)
to determine the reflectances for the standard bands. The VIs
from the different simulated systems were then compared. 0.4
Figures 8.7 and 8.8 show NDVI values for two systems com-
pared with the corresponding NDVIs for the proposed standard
0.2
bands at 670 and 815nm. VIs from all systems were highly lin-
early correlated. NOAA8 and the Compact High-Resolution
Imaging Spectrometer (CHRIS) as shown here represent the 0
extremes of slope, with NDVI for NOAA8 being as much 19% 0.2 0 0.2 0.4 0.6 0.8 1
lower than the standard on the same target. The strength and Standard bands
linearity of the correlations (minimum r2 = 0.984) means that 0.2
NDVIs recorded by different observation systems can be inter-
calibrated to a degree of precision of about 1%. Steven et al. FIgu r e 8.8 Regression of NDVI based on the CHRIS near-infra-
(2003) tabulated two-way intersystem conversion coefficients red (L14) and red bands against the standard bands. (Adapted from
for 15 operational systems as well as the standard narrow band Steven, M.D. etal., Intercalibration of vegetation indicesAn update,
sensor pair. In a later update, Steven etal. (2007) extended the in M.E. Schaepman, eds., 10th International Symposium on Physical
Measurements and Spectral Signatures in Remote Sensing, International
intercalibration of VIs forward to include conversion coefficients
Archives of the Photogrammetry, Remote Sensing and Spatial
for orbiting sensor systems launched since the 2003 paper and
Information Sciences, Vol. XXXVI, Part 7/C50, ISPRS, Davos (CH),
backward to include historical variations in the NOAA AVHRR pp.16821777.)
system, a total of 41 systems (Table 8.2). Two-way interconver-
sion tables are no longer practical for the large number of sys- the optimum, maximizing the NDVI value, although in fact the
tems involved, so for simplicity the conversions are given relative NDVI for the CHRIS system using the L14 NIR band (Figure8.8)
to an NDVI based on the standard pair of narrowbands at 670 does have a slightly greater dynamic range.
and 815nm. These bands were originally chosen to be close to Linear regressions for SAVI, OSAVI, and NDVI differ in slope
and intercept by no more than about 0.0008 and 0.013 respec-
1
tively, across the whole range of indices. As these differences are
considerably less than typical errors of measurement, a single
0.8 conversion table is adequate for the range of VI formulations
considered. It is also possible to convert from one operational
system to another using Table 8.2 to convert first to the standard
NOAA8
0.6 as an intermediate stage and then from the standard to the sec-
ond system. On examples tested, the error in this two-stage pro-
cess, as compared with direct conversion between the systems,
0.4
was up to 0.01 in slope and 0.007 in intercept.
Table 8.2 allows VI data from any of the listed systems to be
0.2 corrected to the corresponding VI for the standard pair of spec-
tral bands. To convert a VI from an operational system VIop to
the standard, VIstd, Equation 8.3 is applied, using the slope and
0 intercept values from column B. To convert from the standard
0.2 0 0.2 0.4 0.6 0.8 1
to a particular operational system, Equation 8.4 is applied, using
Standard bands
the slope and intercept values from column A:
0.2
VIstd = VIop [slope]B + [intercept]B (8.3)
FIg u r e 8.7 Regression of NDVI based on NOAA8 bands against
the standard bands 670 10 and 815 10nm. (Adapted from Steven,
M.D. et al., Intercalibration of vegetation indicesAn update, in M.E. VIop = VIstd [slope]A + [intercept]A (8.4)
Schaepman, eds., 10th International Symposium on Physical Measurements
and Spectral Signatures in Remote Sensing, International Archives of For the OrbView systems where the detailed spectral response
thePhotogrammetry, Remote Sensing and Spatial Information Sciences, functions were unavailable, there are two results in Table 8.2,
Vol. XXXVI, Part 7/C50, ISPRS, Davos (CH), pp. 16821777.) corresponding to the alternative assumptions applied. For
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 183
TABLe 8 .2 Conversion Coefficients for VIs (NDVI, SAVI, and OSAVI) from Different Systems (Steven,
Malthus, and Baret 2007) Based on Simulations from the Database of the Full NIR and Visible Spectral Responses
of Each System Where Alternative Options for a System Exist, Both Are Shown
A B
Sensor vs. Standard Standard vs. Sensor
Satellite Sensor Intercept Slope Intercept Slope
ALI 0.005 0.965 0.006 1.034
ASTER, using band 3B 0.001 0.933 0.003 1.068
ASTER, using band 3N 0.000 0.933 0.002 1.068
ATSR2/AATSR 0.008 0.968 0.006 1.030
CHRIS, using band L14 0.015 1.009 0.016 0.989
CHRIS, using band L15 0.005 0.991 0.004 1.007
DMC 0.006 0.954 0.005 1.046
Formosat 0.002 0.936 0.000 1.065
IKONOS 0.010 0.870 0.015 1.144
IRS 0.005 0.950 0.004 1.050
Kompsat 0.004 0.942 0.003 1.058
Landsat 5 TM 0.005 0.938 0.003 1.063
Landsat 7 ETM+ 0.003 0.957 0.002 1.041
Landsat MSS 0.029 0.883 0.024 1.115
MERIS 0.008 0.983 0.008 1.016
MISR 0.005 0.985 0.005 1.014
MODIS 0.017 0.935 0.015 1.065
MSG/SEVIRI 0.012 0.926 0.010 1.076
NOAA10 0.003 0.854 0.001 1.160
NOAA11 0.015 0.831 0.011 1.188
NOAA12 0.015 0.844 0.012 1.173
NOAA13 0.017 0.835 0.014 1.184
NOAA14 0.016 0.837 0.013 1.180
NOAA15 0.016 0.902 0.014 1.100
NOAA16 0.017 0.897 0.015 1.107
NOAA17 0.016 0.904 0.014 1.098
NOAA18 0.017 0.905 0.014 1.097
NOAA6 0.021 0.850 0.018 1.163
NOAA7 0.015 0.857 0.012 1.155
NOAA8 0.015 0.807 0.012 1.226
NOAA9 0.015 0.839 0.012 1.179
OrbView-2, using block function 0.005 0.989 0.004 1.009
OrbView-2, using Gaussian 0.005 0.982 0.005 1.016
OrbView-3, using block function 0.002 0.937 0.000 1.063
OrbView-3, using Gaussian 0.002 0.857 0.001 1.159
POLDER 0.005 0.985 0.005 1.014
QuickBird 0.000 0.909 0.002 1.096
SeaWIFS 0.005 0.982 0.004 1.016
SPOT2 Hrv2 0.012 0.921 0.011 1.081
SPOT4 Hrv2 0.010 0.917 0.008 1.085
SPOT5 0.010 0.928 0.008 1.073
Vens, using band B10 with Gaussian 0.012 0.984 0.013 1.015
Vens, using band B11 with Gaussian 0.007 0.967 0.006 1.032
2016 by Taylor & Francis Group, LLC
184 Remotely Sensed Data Characterization, Classification, and Accuracies
OrbView-2, both methods give comparable results so that States for identical 16-day compositing periods and found lin-
adjustment to the standard can be made to better than 1% pre- ear relationships between NDVI values from different sensors.
cision. For OrbView-3, however, the difference is about 8%, Their regression slopes differ from Steven et al. (2003) by no
indicating that in the absence of further information, this sys- more than 0.02 indicating good agreement within the limits
tem is unsuitable for applications requiring intercalibration of the data; although the compositing process can introduce
with others. a systematic upward bias in NDVI (Goward et al. 1993), this
would probably be similar for both systems. Fensholt et al.
(2006) compared MERIS, MODIS, and VEGETATION prod-
8.4.2 Validation of cross-Sensor conversion
ucts on grass savannah in Senegal using wide-angle in situ
Although direct intercomparisons of sensors suffer from measurements with band radiometers designed to approximate
greater errors and limitations than the simulation approach the relevant bands. They report generally good agreement with
applied in Section 8.4.1, they are valuable in validating the Steven etal. (2003) but with higher MERIS sensitivity to vege-
findings. Steven et al. (2003) reported image-based com- tation than predicted. However, the accuracy of their compari-
parisons between SPOT High-Resolution Visible (HRV) and sons depends on the degree to which the insitu sensor bands
Landsat Thematic Mapper (TM) data and between Along- match those of the satellite instruments. In addition, as they
Track Scanning Radiometer 2 (ATSR-2) and AVHRR. In both noted in their study, wide-angle measurements would exag-
studies, pixels were aggregated across quasi-uniform targets to gerate VIs, particularly in the middle of the range, although,
minimize coregistration errors and the empirical results were given the extensive overlap of the spectral bands, the angular
in reasonable agreement (0.03) with the simulated cross-sen- response effect is likely to be very similar for the different sys-
sor calibrations. Studies by Hill and Aifadopoulou (1990) and tems compared.
Guyot and Gu (1994) also generated coefficients that supported Trishchenko et al. (2002) and DOdorico et al. (2013) have
the findings, but previous simulations by Gallo and Daughtry applied quadratic correction factors for intersensor correc-
(1987) were significantly different, although similarly based on tions. There is no doubt that a quadratic function will provide
simulation from field data. More recent studies have included a better statistical fit to any given set of data, even data that
both direct comparisons of image data and various forms of appear strongly linear as in Figures 8.7 and 8.8, but fitting a
simulation. The quality of image comparisons in the literature curve to such data makes the coefficients relatively unstable as
is however quite variable, with many suffering from substantial the additional coefficient tends to counteract the previous ones,
differences in image acquisition time, different solar or view- and it is unclear whether the extra coefficient is justified by a
ing angles, insufficient aggregation to overcome coregistration general gain in accuracy when applied to independent data.
errors, or other problems. Such studies may show differences Swinnen and Veroustraete (2008) compared the effect of cor-
between sensors but have insufficient precision to test conver- rections according to Steven etal. (2003) and the polynomial
sion factors. Selected results are discussed as follows. corrections of Trishchenko etal. (2002). In general, the correc-
Martnez-Beltrn etal. (2009) compared Enhanced Thematic tions by Steven etal. (2003) gave the greater reduction of error,
Mapper (ETM+), TM, Linear Imaging Self-Scanning Sensor except for open or sparse grassland with low NDVI, where val-
(LISS), Advanced Spaceborne Thermal Emission and Reflection ues were overcorrected, while the method of Trishchenko etal.
Radiometer (ASTER), QuickBird, and AVHRR data on selected (2002) provided a more consistent correction across all land
sites in southeastern Spain covering a wide range of surface cover classes.
types (Figure 8.9). All images were near-nadir viewing and the Song et al. (2010) found linear conversions between AVHRR
image pairs compared were no more than a few hours apart. Global Inventory Modeling and Mapping Studies (GIMMS) and
Although comparisons were of top-of-atmosphere NDVI, with- SPOT-VGT (VEGETATION instrument) NDVI values, but the
out atmospheric correction, they found that with sufficient coefficients varied regionally across China. Figure 8.10 shows a
aggregation to reduce noise, their cross-sensor relationships map of average differences between the GIMMS and VGT prod-
were linear, reasonably precise, and in good agreement with ucts. Mountainous regions (C) show both positive and negative dif-
the results of Steven et al. (2003). However, Figure 8.9 sh ows ferences, while the borderline between regions A and B, where the
the difficulty of establishing reliable cross-sensor relationships differences are larger, and regions C and D, where they are lower,
by direct comparison. In spite of the use of aggregation and corresponds to a major climatic and geological boundary. When
homogenous areas for comparison, substantial differences in rates of change of seasonally integrated NDVI were computed, there
coefficients remain. were substantial differences between the two systems (Figure8.11).
Ji et al. (2008) evaluated differences between AVHRR and Miura etal. (2006) compared NDVI values for a number of
MODIS in two years of data over the conterminous United systems on Hyperion hyperspectral image data over tropical for-
States. In addition to their own findings that the differences est and savannah in Brazil. They applied a similar approach to
are substantial and 20% systematic, they compare their results Steven et al. (2003, 2007), combining the atmospherically cor-
with 17 previous cross-sensor studies. Gallo etal. (2005) com- rected data with spectral response functions to simulate surface
pared NDVI values for MODIS and AVHRR over the United measured radiance in various bands. Although relationships
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 185
May 20, 2003 July 31, 2003
1.0 1.0
NDVI_ETM + =1.0537 NDVI_QuickBird + 0.0148 NDVI_ETM + =1.0330 NDVI_QuickBird + 0.0244
0.9 R2 = 0.9957 0.9 R2 = 0.9951
0.8 0.8
0.7 Alfalfa 0.7 Alfalfa
NDVI_ETM+
NDVI_ETM+
0.6 0.6
0.5 0.5 Summer irrigated
Spring irrigated crop
0.4 crop 0.4
0.3 0.3 Forestry
Forestry
0.2 0.2
0.1 0.1 Bare soil
Bare soil
Grid 25 m 25 m Grid 25 m 25 m
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
NDVI_QuickBird NDVI_QuickBird
1.0 1.0
NDVI_ETM + =1.0526 NDVI_QuickBird + 0.0146 NDVI_ETM + =1.0360 NDVI_QuickBird + 0.0236
0.9 R2 = 0.9983 0.9 R2 = 0.9989
0.8 0.8
Alfalfa
0.7 Alfalfa 0.7
0.6 0.6
NDVI_ETM+
NDVI_ETM+
0.5 Spring irrigated 0.5 Summer irrigated
crop crop
0.4 0.4
0.3 0.3 Forestry
Forestry
0.2 0.2
Bare soil
0.1 Bare soil 0.1
Grid 100 m 100 m Grid 100 m 100 m
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
NDVI_QuickBird NDVI_QuickBird
FIgu r e 8.9 Intercomparisons of ETM+ and QuickBird NDVI in homogeneous zones at 25 m and 100 m scales. The data for July 31, 2003,
include a prior conversion from TM to ETM+ NDVI. (Adapted from Martnez-Beltrn, C. etal., Int. J. Remote Sens., 301355, 2009.)
between simulated radiances in paired bands were found to be and 6S model to generate synthetic data corresponding to
land cover dependent, the relationships for NDVI were indepen- 21satellite sensors. Their simulations included a range of leaf
dent of land cover. However, they were sufficiently nonlinear to characteristics, LAI from 0 to 6 with a fixed spherical leaf angle
require fitting a quadratic function for an adequate conversion distribution, a range of solar and viewing angles, and eight
between systems. distinct backgrounds. They found good agreement with both
Intersensor conversions have also been evaluated by Steven etal. (2003) and van Leeuwen etal. (2006), but less so
modeling approaches. Van Leeuwen et al. (2006) simulated with other studies that had used quadratic fitting. Soudani etal.
NDVI from AVHRR, MODIS, and Visible Infrared Imaging (2006) also found good agreement with Steven etal. (2003) for
Radiometer Suite (VIIRS) using the scattering by arbitrarily PROSAIL simulations of IKONOS, ETM+, and SPOT NDVIs
inclined leaves (SAIL) model with a wide range of LAI. The from forest canopies; the maximum differences in slope and
model was parameterized with data inputs from spectral intercept were 0.383 and 0.0183, respectively. DOdorico etal.
libraries of vegetation (two spectra), soil (two spectra), and (2013) assessed the general effectiveness of synthetic calibra-
snow (one spectrum). Their result for NOAA16 versus MODIS tion data generated by radiative transfer modeling for cross-
is within 0.01 of values predicted from Table 8.1, but their pre- sensor calibration. In general, airborne spectral measurements
diction for NOAA14 has a slope 0.03 higher. Gonsamo and gave better results, but model-generated data were found to be
Chen (2013) used a coupled PROSPECT + SAIL (or PROSAIL) a good substitute for regional or global monitoring.
2016 by Taylor & Francis Group, LLC
186 Remotely Sensed Data Characterization, Classification, and Accuracies
Heihe
N
C B
D C
e
Qinling aih
C Hu
Tengchong
N: 0.80.080.04 0 0.04 0.08 1.7
500 0 500 km
FIg u r e 8.10 Map of mean annual differences of VGT and GIMMS NDVI for 19982003 in China. (Adapted from Song, Y. etal., Int. J. Remote
Sens., 31, 2377, 2010).
E: 4 0.8 0.3 0.1 0.1 0.3 0.8 4
500 0 500 km
(a)
FIg u r e 8.11 Rate of change (year1) of seasonally integrated NDVI in China from 1998 to 2006 for (a) VGT and (b) GIMMS systems. (Adapted
from Song, Y. etal., Int. J. Remote Sens., 31, 2377, 2010.) (Continued )
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 187
E: 4 0.8 0.3 0.1 0.1 0.3 0.8 4
500 0 500 km
(b)
FIg u r e 8.11 (c ontinued ) Rate of change (year1) of seasonally integrated NDVI in China from 1998 to 2006 for (a) VGT and (b) GIMMS sys-
tems. (Adapted from Song, Y. etal., Int. I. Remote Sens., 31, 2377, 2010.)
8.5 Vegetation index Standardization correctionshould come first, followed by corrections
for illumination and viewing geometry as these may vary
In proposing a standard for VIs, some general principles should to some extent with the particular bands of the earth
be considered. observation system. Finally, the VI should be calculated
and adjustment made for the spectral bands as described
Principle 1Standardization should be applicable to veg-
in Section 8.4.1 (Equation 8.3).
etation indices from all systems, including the very ear-
liest, to meet the aim of providing a long-term baseline
of global vegetation measurements. For this reason, the
8.5.1 Proposals for Vegetation index Standards
index chosen should be based on just two bands, the
red and the NIR, in spite of improvements in perfor- In practice, atmospheric corrections are usually combined in a
mance reported with incorporation of additional band single model with calibration. The 6S (Vermote etal. 1997) and
information. MODTRAN (Berk etal. 1998) models are both widely used and
Principle 2Standards should be capable of evolving over appropriate here. Callieco and DellAcqua (2011) investigated
time. A VI standard must be traceable and, if possible, differences between these models, but found that significant
reversible, as new knowledge or improved modeling effects were confined to wavelengths shorter than 500 nm, so
may require readjustment of past data. With an evolv- the choice of model should have minimal impact on VIs. Such
ing standard, complete metadata are essential to allow models also account for variations in earthsun distance and
later adjustments, and adopters of the standard should the effects of solar and viewing angle on the atmospheric effects.
specify precisely what corrections and procedures have Variations in solar output are not modeled, but are typically of
been applied in adjusting measured data. the order of 0.1% and can safely be ignored.
Principle 3Standardization procedures should be modular, A BRDF model is needed to account for effects of viewing
so that corrections for spectral bands, soil background, and illumination directions. The PROSAIL model (Jacquemoud
BRDF, and other factors are performed separately. etal. 2009) is widely respected and sufficiently versatile for this
Modularity improves the ability to trace (and if neces- purpose. To standardize for bidirectional effects, it is also neces-
sary reverse) processes and errors in the system. The sary to adopt standard solar and viewing angles. Bacour et al.
sequence of correction procedures should also be con- (2006) proposed standardizing on viewing at nadir with a solar
sidered as not all the operations are strictly commutative. zenith angle of 40. Conveniently, these angles approximate to
Radiometric correctionscalibration and atmospheric the average measurement conditions that apply to the canopy
2016 by Taylor & Francis Group, LLC
188 Remotely Sensed Data Characterization, Classification, and Accuracies
spectral database used by Steven et al. (2003, 2007), so adop- further studies over a wider range of soil types may lead to better
tion of these angles as standard would ensure a self-consistent understanding of the errors and an improved general index, but
system. The solar angle proposed is also a reasonable mid-value a perfect soil adjustment is not attainable, and residual errors of
for summer viewing conditions at mid-latitudes. The bidirec- the order of 5% can be expected in the estimation of fractional
tional effects depend on vegetation type that can be modeled by vegetation cover (Rondeaux etal. 1996).
PROSAIL on the basis of leaf angle distribution. If this infor- A summary of these proposals is shown in Table 8.3.
mation is unavailable, or impractical to implement such as in
global-scale applications, corrections based on a spherical leaf
8.5.2 Limits to Standardization
angle distribution are recommended here.
The instrument-specific VI can next be calculated from the Finally, it must be recognized that there are limits to the ability
corrected surface reflectance values. Adjustment should then to correct and standardize VIs. The potential pitfalls of failing
be made to determine the corresponding index for the stan- to apply limits to the operational parameters are illustrated in
dard bands670 and 815 nm are proposed hereby applying Figure 8.12, which represents the average of MODIS EVI obser-
Equation 8.3 with the appropriate conversion coefficients from vations over two-month periods around the solstices. On close
Table 8.2. examination, there appears to be a slight increase in EVI along
Multiple standards for indices may be necessary, at least in the Arctic Circle in the middle of the northern winter. This is
the short term. The NDVI is so well established that its continu- emphatically not a result of increased vegetation! At this time
ance is unavoidable, in spite of its deficiencies. An index adjusted of year in this location, the sun is grazing the horizon and as
to minimize sensitivity to variable soil should also be included, a result of its low altitude is severely depleted in shorter wave-
but there are many claimants. The simplest is the SAVI defined lengths by Rayleigh scattering in the atmosphere. The effect of
in Equation 8.2 (Huete 1988), with a value of 0.5 for the adjust- this selective depletion is that although the sun would appear
ment factor L. The advantage of the SAVI formulation is that the red to a ground observer, it is actually relatively weak in the red
adjustment factor eliminates the need for specific calibration to compared to the NIR. Sunlight reflected from snow therefore
different soils (Huete and Liu 1994). Reevaluation by Rondeaux has an exaggerated NIR: red ratio that enhances the VI. We
etal. (1996) of a range of L values found that a value of 0.16 gave recommend this image as a puzzle and a warning for students
a slightly better performance, particularly when the soil types in of remote sensing! The problem is easily avoided: the spectral
the analysis were restricted to those likely to be found in agricul- balance of solar irradiation is conservative once solar elevation
ture and the OSAVI was proposed incorporating this value. In exceeds 10 (Monteith and Unsworth 1990). Variations in spec-
the analysis by Rondeaux etal. (1996), the variance in the index tral irradiance can probably be modeled with reasonable accu-
due to soil type was reduced from 7.5% for NDVI to 1.1% and racy within a few degrees of the horizon, but until this can be
1.7% for SAVI and OSAVI, respectively. The case for OSAVI was reliably demonstrated, observations beyond a solar zenith angle
largely based on its performance with a restricted set of agricul- of 80 should be excluded. The same limit applies to observation
tural soil types where it reduced the soil signal to 0.06% with angles, as selective Rayleigh scattering works in exactly the same
the residual error distributed evenly across the range of foliage way on the upwelling reflected radiance.
cover. With the wider dataset that might be more applicable for
global application, SAVI and MSAVI were slightly better per- 8.6 Discussion
formers. SAVI (with L = 0.5) is recommended here over MSAVI
for its greater simplicity and its known compatibility with the There is a general difficulty with both atmospheric correction and
conversion coefficients in Table 8.2. Both OSAVI and the origi- BRDF modeling that the models require input data for parameter-
nal SAVI are based on analyses of relatively narrow datasets, and ization. Atmospheric correction parameters are difficult to obtain
TABLe 8 .3 Summary of Specific Proposals Made in This Chapter for Vegetation Index Standards
Standard Parameters Value Remarks
VI formula SAVI Preferred
NDVI Tolerated
Spectral bands
Near infrared 815nm Nominal bandwidth 20nm, centered on these wavelengths
Red 670nm
View zenith angle 0 After Bacour etal. (2006)
Solar zenith angle 40
Atmospheric correction To surface reflectance 6S or MODTRAN routines recommendeda
BRDF corrections To standard angles PROSAIL recommended with spherical leaf angle distribution by defaulta
a These are interim recommendations and do not preclude alternative or improved procedures.
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 189
for remote areas, but Steven (1998) found that OSAVI estimates were cross-sensor effects (Sun etal. 2007; Fisher etal. 2008; Pouliot
affected to a very minor extent by errors in the model assumptions. etal. 2009; Tang etal. 2009; Zhang etal. 2009; Propastin and
In a global monitoring context, it would also be possible to incor- Erasmi 2010; Ouyang etal. 2010, 2012). Others have tested the
porate satellite aerosol observations into a routine correction pro- conversion equations against data or modeling approaches as
cedure. For BRDF modeling, the leaf angle distribution is critical. described in Section 8.4.2. The validation studies discussed
In a simulation of the effect of variable SPOT viewing angles, Steven provide general support for the idea of intercalibrating VIs
(1998) found variations in OSAVI up to 4% relative to vertical view- and, in many cases, support for the linear coefficients pro-
ing. This difference corresponded to 30 leftward tilt of the SPOT vided by Steven et al. (2003) and given in extended form in
camera and larger differences will occur at greater angles. However, Table 8.2. The accuracy suggested by these studies is of the
as the off-nadir effect can be estimated by the model, it should not order of 0.03 in the slope. This is probably about the limit
of itself be a major source of error, while errors associated with the of accuracy for a validation study in this context. More per-
assumed leaf angle distribution can be expected to be second order tinent here is the relative precision of the conversions, which
relative to the overall magnitude of the off-nadir effect. is about 0.01.
Since the paper by Steven etal. (2003), a number of stud- Despite this support, the generality of the intersensor rela-
ies have applied the coefficients provided to convert for tionships should be considered. Song et al. (2010) found that
May 21July 21, 2000
November 21, 2000January 21, 2001
Enhanced vegetation index
0.0 0.2 0.4 0.6 0.8 1.0
FIg u r e 8.12 Composited seasonal MODIS EVI observations illustrating the effects of extreme solar angles on the index. (Adapted from http://
earthobservatory.nasa.gov/IOTD/view.php?id=2033.)
2016 by Taylor & Francis Group, LLC
190 Remotely Sensed Data Characterization, Classification, and Accuracies
differences between AVHRR GIMMS and SPOT VEGETATION atmospheric aerosol amount (Steven 1998); 4% for off-nadir view
NDVI were both land cover and seasonally dependent. However, angle effects up to 30, if poorly characterized (Steven 1998); and
it is difficult to assess whether such variations are fundamentally about 1% for spectral band effects. The corresponding uncertain-
linked to the basic observations or an artifact of the composit- ties in downstream products such as fAPAR will be slightly larger.
ing and processing procedures applied to generate one or other Greater errors can also be expected at larger angles of view or solar
of the datasets. It is also possible that these dependencies, and zenith angles, and if errors combine malignly, but the overall effect
the earlier finding by Swinnen and Veroustraete (2008) that the of atmospheric and BRDF errors can be reduced substantially
equations of Steven etal. (2003) overcorrected low values, may by application of simple well-established models and appropri-
be related to the high variability of NDVI with respect to soil. ate input data. Further studies are needed to extend the database
DOdorico etal. (2013) referred to earlier studies that mention for comparison of sensors and assessment of BRDF effects, to
land cover dependence, but these studies related to differences assess the errors of modeling and to validate the procedures and
in VIs rather than their ratio. Miura etal. (2006), as well as the their effect on VIs and downstream products. It is anticipated
study by DOdorico etal. (2013) itself, showed that conversion to that the errors will become better characterized as procedures
NDVI takes out most of the land cover dependence that is found evolve. Nevertheless, there is sufficient evidence to proceed now.
in the individual bands. Standardization will have immense benefits for vegetation moni-
A further concern is that the cross-sensor relationships have toring, both on short-timescale regional studies where the constel-
been established on a relatively narrow base of data. The database lation approach is applied to monitor the dynamics of specific land
used by Steven et al. (2003, 2007) comprised 166 independent covers and in large-scale studies of environmental history where
measurements of spectral reflectance of strictly agricultural can- records from early satellites are merged with modern data.
opies. Other studies have even fewer measurements, and while
model simulations may generate large synthetic datasets, their References
empirical support base is often very narrow. In addition, many of
the studies discussed in Section 8.4.2 have applied the same mod- Agapiou, A., D. G. Hadjimitsis, and D. D. Alexakis. 2012.
els, so similar findings are to be expected. Forest cover classes, Evaluation of broadband and narrowband vegetation indi-
where deep shadow may be a significant component of the target, ces for the identification of archaeological crop marks.
are not well represented in the datasets, although the study by Remote Sens 4:38923919.
Soudani et al. (2006) suggests that the coefficients of Table 8.2 Bacour, C., F. M. Bron, and F. Maignan. 2006. Normalization of
still provide a good correction. As noted by Steven etal. (2003), the directional effects in NOAAAVHRR reflectance mea-
previous studies have also found senescent biomass, or litter to surements for an improved monitoring of vegetation cycles.
be a significant factor. The results of Miura etal. (2006), which Remote Sens Environ 102:402413.
required a quadratic correction between sensors, suggest that Baret, F. and G. Guyot. 1991. Potentials and limits of vegetation
the relations established in Table 8.2 are not universal. A broader indices for LAI and APAR assessment. Remote Sens Environ
database might help to resolve these issues, but as Miura et al. 35:161173.
(2006) pointed out, observations with different spectral bands Berk, A., L. S. Bernstein, G. P. Anderson, P. K. Acharya, D. C.
are inherently different and may introduce bias into downstream Robertson, J. H. Chetwynd, and S. M. Adler-Golden. 1998.
products. The same caveat applies to correction for BRDF effects. MODTRAN cloud and multiple scattering upgrades with
Nevertheless, standardization is imperative for a range of issues application to AVIRIS. Remote Sens Environ 65:367375.
and does not preclude use of the data in original format for spe- Bowers, S. A. and R. J. Hanks. 1965. Reflection of radiant energy
cific purposes. Steven etal. (2003) argued that restriction of the from soil. Soil Sci 100:130138.
intercalibration database to cultivated vegetation may actually Boyd, D. S., S. Almond, J. Dash, P. J. Curran, and R. A. Hill. 2011.
have allowed a tighter statistical relationship to be established Phenology of vegetation in Southern England from Envisat
than would have occurred with a broader, more representative MERIS terrestrial chlorophyll index (MTCI) data. Int J
dataset. Differences between indices are in most cases quite Remote Sens 32:84218447.
small, and applying conversions based on agricultural datas- Boyd, D. S. and G. M. Foody. 2011. An overview of recent remote
ets may help to maintain relationships with fAPAR and related sensing and GIS based research in ecological informatics.
parameters that have been well validated in such environments. Ecol Informatics 6:2536.
Brown, M., J. E. Pinzn, K. Didan, J. T. Morisette, and C. J.
8.7 conclusions Tucker. 2006. Evaluation of the consistency of long-term
NDVI time series derived from AVHRR, SPOT-Vegetation,
The outcome of this study is that corrections are possible, to an SeaWiFS, MODIS, and Landsat ETM+ sensors. IEEE Trans
acceptable degree of precision, for the main sources of variation in Geosci Remote Sens 44:17871793.
VIs: soil background, atmospheric, BRDF, and spectral band effects. Callieco, F. and F. DellAcqua. 2011. A comparison between two
The uncertainty of these corrections is about 5% for soil, when radiative transfer models for atmospheric correction over a
using a SAVI (Rondeaux etal. 1996); 3% for poorly characterized wide range of wavelengths. Int J Remote Sens 32:13571370.
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 191
CEOS. 2006. The CEOS virtual constellation concept V0.4, Guyot, G. and X. F. Gu. 1994. Effect of radiometric corrections
Committee on Earth Observation Satellites. http://igos- on NDVIDetermined from SPOT HRV and Landsat TM
cryosphere.org/docs/CEOS_constellations.doc (accessed data. Remote Sens Environ 49:169180.
July 25, 2014). Hadjimitsis, D. G., G. Papdavid, A. Agapiou, K. Themistocleous,
DOdorico, P., A. Gonsamo, A. Damm, and M. E. Schaepman. M. G. Hadjimitsis, A. Retalis, S. Michaelides, N.
2013. Experimental evaluation of Sentinel-2 spectral Chrysoulakis, L. Toulios, and C. R. I. Clayton. 2010.
response functions for NDVI time-series continuity. IEEE Atmospheric correction for satellite remotely sensed data
Trans Geosci Remote Sens 51:13361348. intended for agricultural applications: impact on vegetation
DOdorico, P., A. Gonsamo, B. Pinty, N. Gobron, N. Coops, indices. Nat Hazards Earth Syst Sci 10:8995.
E.Mendez, and M. E. Schaepman. 2014. Intercomparison Heller, J. 1961. Catch 22. Simon and Schuster, New York.
of fraction of absorbed photosynthetically active radiation Hill, J. and D. Aifadopoulou. 1990. Comparative analysis of
products derived from satellite data over Europe. Remote Landsat-5 TM and SPOT HRV-1 data for use in multiple
Sens Environ 142:141154. sensor approaches. Remote Sens Environ 34:5570.
Fensholt, R., I. Sandholt, and S. Stisen. 2006. Evaluating MODIS, Holben, B. N., C. J. Tucker, and C. J. Fan. 1980. Spectral assess-
MERIS, and VEGETATION vegetation indices using insitu ment of soybean leaf area and leaf biomass. Photogramm
measurements in a semiarid environment. IEEE Trans Eng Remote Sens 46:651656.
Geosci Remote Sens 44:17741786. Huete, A. R. 1988. A soil adjusted vegetation index (SAVI). Int J
Fisher, J. B., K. P. Tu, and D. D. Baldocchi. 2008. Global esti- Remote Sens 9:295309.
mates of the land-atmosphere water flux based on monthly Huete, A. R. and H. Q. Liu 1994. An error and sensitivity analy-
AVHRR and ISLSCP-II data, validated at 16 FLUXNET sis of the atmospheric- and soil-correcting variants of the
sites. Remote Sens Environ 112:901919. NDVI for the MODIS-EOS. IEEE Trans Geosci Remote Sens
Gallo, K., L. Ji, B. Reed, J. Eidenshink, and J. Dwyer. 2005. Multi- 32:897905.
platform comparisons of MODIS and AVHRR normal- Huete, A. R., H. Q. Liu, K. Batchily, and W. van Leeuwen.
ized difference vegetation index data. Remote Sens Environ 1997. A comparison of vegetation indices over a global
99:221231. set of TM images for EOSMODIS. Remote Sens Environ
Gallo, K. P. and C. S. T. Daughtry. 1987. Differences in vegetation 59:440451.
indices for simulated Landsat-5 MSS and TM, NOAA-9 Jacquemoud, S., W. Verhoef, F. Baret, C. Bacour, P. J. Zarco-Tejada,
AVHRR and SPOT-1 sensor systems. Remote Sens Environ G. P. Asner, C. Franois, and S. L. Ustin. 2009. PROSPECT
23:439452. + SAIL models: a review of use for vegetation characteriza-
Gates, D. M., H. J. Keegan, J. C. Schleter, and V. R. Weidner. 1965. tion. Remote Sens Environ 113:S56S66.
Spectral properties of plants. Appl Opt 4:1120. Jaggard, K. W. and C. J. Clark. 1990. Remote sensing to predict
Gausman, H. W. and W. A. Allen. 1973. Optical parameters of the yield of sugar beet in England. In M. D. Steven and J. A.
leaves of 30 plant species. Plant Physiol 52:5762. Clark (eds), Applications of Remote Sensing in Agriculture.
Gobron, N., B. Pinty, O. Aussedat, M. Taberner, O. Faber, F. Mlin, Butterworths, London, U.K., pp. 201208.
T. Lavergne, M. Robustelli, and P. Snoeij. 2008. Uncertainty Ji, L., K. Gallo, J. C. Eidenshink, and J. Dwyer. 2008. Agreement
estimates for the FAPAR operational products derived from evaluation of AVHRR and MODIS 16-day composite NDVI
MERISImpact of top-of-atmosphere radiance uncertain- data sets. Int J Remote Sens 29:48394861.
ties and validation with field data. Remote Sens Environ Lee, K-S., W. B. Cohen, R. E. Kennedy, T. K. Maiersperger, and
112:18711883. S.T. Gower. 2004. Hyperspectral versus multispectral data
Gonsamo, A. and J. M. Chen. 2013. Spectral response function for estimating leaf area index in four different biomes.
comparability among 21 satellite sensors for vegetation Remote Sens Environ 91:508520.
monitoring. IEEE Trans Geosci Remote Sens 51:13191335. Le Maire, G., C. Franois, and E. Dufrne. 2004. Towards univer-
Goward, S. N., D. G. Dye, S. Turner, and J. Yang. 1993. Objective sal broad leaf chlorophyll indices using PROSPECT simu-
assessment of the NOAA global vegetation index data prod- lated database and hyperspectral reflectance measurements.
uct. Int J Remote Sens 14:33653394. Remote Sens Environ 89:128.
Gower, S. T., C. J. Kucharik, and J. K. M. Norman. 1999. Direct Mahiny, A. S. and B. Turner. 2007. A comparison of four common
and indirect estimation of leaf area index, fAPAR, and net atmospheric correction methods. Photogramm Eng Remote
primary production of terrestrial ecosystems. Remote Sens Sens 73:361368.
Environ 70:2951. Martnez-Beltrn, C., M. A. O. Jochum, A. Calera, and J. Meli.
Gutman, G. 1991. Vegetation indices from AVHRR: an update 2009. Multisensor comparison of NDVI for a semi-arid
and future prospects. Remote Sens Environ 35:121136. environment in Spain. Int J Remote Sens 30:13551384.
Guyot, G. 1990. Optical properties of vegetation canopies. In M. D. Maselli, F., S. Romanelli, L. Bottai, and G. Maracchi. 2000.
Steven and J. A. Clark (eds), Applications of Remote Sensing Processing of GAC NDVI data for yield forecasting in the
in Agriculture. Butterworths, London, U.K., pp. 1943. Sahelian region. Int J Remote Sens 21:35093523.
2016 by Taylor & Francis Group, LLC
192 Remotely Sensed Data Characterization, Classification, and Accuracies
Miura, T., A. Huete, and H. Yoshioka. 2006. An empirical investi- Pouliot, D., R. Latifovic, and I. Olthof. 2009. Trends in vegetation
gation of cross-sensor relationships of NDVI and red/near- NDVI from 1km AVHRR data over Canada for the period
infrared reflectance using EO-1 Hyperion data. Remote 19852006. Int J Remote Sens 30:149168.
Sens Environ 100:223236. Price J. C. 1987. Calibration of satellite radiometers and the com-
Miura, T., J. P. Turner, and A. R. Huete. 2013. Spectral compat- parison of vegetation indices. Remote Sens Environ 21:1527.
ibility of the NDVI across VIIRS, MODIS and AVHRR: an Propastin, P. and S. Erasmi. 2010. A physically based approach
analysis of atmospheric effects using EO-1 Hyperion. IEEE to model LAI from MODIS 250 m data in a tropical region.
Trans Geosci Remote Sens 51:13491359. Int J App Earth Obs Geoinf 12:4759.
Monteith, J. L. 1977. Climate and the efficiency of crop produc- Qi, J., A. S. Chehbouni, A. R. Huete, Y. H. Kerr, and S. Sorooshian.
tion in Britain. Philos Trans R Soc Lond Ser B, 281:277294. 1994. A modified soil adjusted vegetation index. Remote
Monteith, J. L. and M. H. Unsworth. 1990. Principles of Environmental Sens Environ 48:119126.
Physics, 2nd edn., Edward Arnold, London, U.K. Rder, A., T. Kuemmerle, and J. Hill. 2005. Extension of retro-
Moura, Y. M., L. S. Galvo, J. R. dos Santos, D. A. Roberts, and spective datasets using multiple sensors. An approach to
F.M. Breunig. 2012. Use of MISR/Terra data to study intra- radiometric intercalibration of Landsat TM and MSS data.
and inter-annual EVI variations in the dry season of tropi- Remote Sens Environ 95:195210.
cal forest. Remote Sens Environ 127:260270. Rondeaux, G., M. D. Steven, and F. Baret. 1996. Optimization
Munyati, C. and G. Mboyeni. 2013. Variation in NDVI values of soil adjusted vegetation indices. Remote Sens Environ
with change in spatial resolution for semi-arid vegetation: a 55:95107.
case study in Northwestern South Africa. Int J Remote Sens Samain, O., B. Geiger, and J-L. Roujean. 2006. Spectral normalisa-
34:22532267. tion and fusion of optical sensors for the retrieval of BRDF and
Nagol, J., E. F. Vermote, and S. D. Prince. 2009. Effects of atmo- Albedo: Application to VEGETATION, MODIS, and MERIS
spheric variation on AVHRR NDVI data. Remote Sens data sets. IEEE Trans Geosci Remote Sens 44:31663179.
Environ 113:392397. Schluessel, G., R. E. Dickinson, J. L. Privette, W. J. Emery, and
Obata, K., T. Miura, and H. Yoshioka. 2012. Analysis of the scal- R.Kokaly. 1994. Modeling the bidirectional reflectance dis-
ing effects in the area-averaged fraction of vegetation cover tribution function of mixed finite plant canopies and soil.
retrieved using an NDVI-isoline-based linear mixture JGeophys Res-Atmos 99(D5):1057710600.
model. Remote Sens 4:21562180. Shultis, J. K. 1991. Calculated sensitivities of several optical radio-
Ouyang, W., F. H. Hao, A. K. Skidmore, T. A. Groen, A. G. metric indices for vegetation canopies. Remote Sens Environ
Toxopeus, and T. Wang. 2012. Integration of multi-sensor 38:211228.
data to assess grassland dynamics in a Yellow River sub- Sims, D. A., A. F. Rahman, E. F. Vermote, and Z. Jiang. 2011.
watershed. Ecol Indic 18:163170. Seasonal and inter-annual variation in view angle effects on
Ouyang, W., F. H. Hao, C. Zhao, and C. Lin. 2010. Vegetation MODIS vegetation indices at three forest sites. Remote Sens
response to 30 years hydropower cascade exploitation Environ 115:31123120.
in upper stream of Yellow River. Commun Nonlinear Sci Song, Y., M. Ma, and F. Veroustraete. 2010. Comparison and
Numer Simul 15:1281941. conversion of AVHRR GIMMS and SPOT VEGETATION
Peuelas, J. and I. Filella. 1998. Visible and near-infrared tech- NDVI data in China. Int J Remote Sens 31:23772392.
niques for diagnosing plant physiological status. Trends Soudani, K., C. Franois, G. le Maire, V. Le Dantec, and E.
Plant Sci 3:151156. Dufrne. 2006. Comparative analysis of IKONOS, SPOT,
Perry, C. R. and L. F. Lautenshlager. 1984. Functional equiva- and ETM+ data for leaf area index estimation in temper-
lence of spectral vegetation indices. Remote Sens Environ ate coniferous and deciduous forest stands. Remote Sens
14:169182. Environ 102:161175.
Pettorelli, N., J. O. Vik, A. Mysterud, J-M. Gaillard, C. J. Tucker, Steven, M. D. 1977. Standard distributions of clear sky radiance.
and N. C. Stenseth. 2005. Using the satellite-derived NDVI Q J Roy Meteorol Soc 103:457465.
to assess ecological responses to environmental change. Steven, M. D. 1998. The sensitivity of the OSAVI vegetation index
Trends Ecol Evol 20:503510. to observational parameters. Remote Sens Environ 63:4960.
Picket-Heaps, C. A., J. G. Canadell, P. R. Briggs, N. Gobron, Steven, M. D. 2004. Correcting the effects of field of view and
V.Haverd, M. J. Paget, B. Pinty, and M. R. Raupach. 2014. varying illumination in spectral measurements of crops.
Evaluation of six satellite-derived Fraction of Absorbed Precis Agric 5:5168.
Photosynthetic Active Radiation (FAPAR) products Steven, M. D., P. V. Biscoe, and K. W. Jaggard. 1983. Estimation of
across the Australian continent. Remote Sens Environ sugar beet productivity from reflection in the red and infra-
140:241256. red spectral bands. Int J Remote Sens 4:325334.
Pinter, P. J. 1993. Solar angle independence in the relationship Steven, M. D., P. V. Biscoe, K. W. Jaggard, and J. Paruntu. 1986.
between absorbed PAR and remotely sensed data for alfalfa. Foliage cover and radiation interception. Field Crops Res
Remote Sens Environ 46:1925. 13:7587.
2016 by Taylor & Francis Group, LLC
Toward Standardization of Vegetation Indices 193
Steven, M. D., T. J. Malthus, and F. Baret. 2007. Intercalibration of Tucker, C. J., J. H. Elgin, and J. E. McMurtrey. 1979. Temporal spec-
vegetation indicesAn update. In M. E. Schaepman, S.Liang, tral measurements of corn and soybean crops. Photogramm
N. Groot, and M. Kneubhler (eds), 10th International Eng Remote Sens 45:643653.
Symposium on Physical Measurements and Spectral Signatures in Tucker, C. J., J. E. Pinzon, M. E. Brown, D. A. Slayback, E. W.
Remote Sensing, International Archives of the Photogrammetry, Pak, R. Mahoney, E. F. Vermote, and N. El Saleous. 2005.
Remote Sensing and Spatial Information Sciences, Vol. XXXVI, An extended AVHRR 8-km NDVI dataset compatible with
Part 7/C50, ISPRS, Davos (CH), pp. 16821777. MODIS and SPOT vegetation NDVI data. Int J Remote Sens
Steven, M. D., T. J. Malthus, F. Baret, H. Xu, and M. Chopping. 26:44854498.
2003. Intercalibration of vegetation indices from different Van Leeuwen, W. J. D. 2006. Spectral vegetation indices and
sensor systems. Remote Sens Environ 88:412422. uncertainty: Insights from a users perspective. IEEE Trans
Steven, M. D. and M. H. Unsworth. 1980. The angular distribu- Geosci Remote Sens 44:19311933.
tion and interception of diffuse solar radiation below over- Van Leeuwen, W. J. D., A. R. Huete, and T. W. Laing. 1999. MODIS
cast skies. Q J Roy Met Soc 106:5761. vegetation index compositing approach: A prototype with
Sun, Z., Q. Wang, Z. Ouyang, M. Watanabe, B. Matsushita, and AVHRR data. Remote Sens Environ 69:264280.
T. Fukushima. 2007. Evaluation of MOD16 algorithm using Van Leeuwen, W. J. D., B. J. Orr, S. E. Marsh, and S. M. Herrmann.
MODIS and ground observational data in winter wheat 2006. Multi-sensor data continuity: uncertainties and impli-
field in North China plain. Hydrol Proc 21:11961206. cations for vegetation monitoring applications. Remote Sens
Swinnen, E. and F. Veroustraete. 2008. Extending the SPOT- Environ 100:6781.
VEGETATION NDVI time series (19982006) back in Verhoef, W. 1985. Light scattering by leaf layers with application
time with NOAA-AVHRR data (19851998) for Southern to canopy reflectance modeling: the SAIL model. Remote
Africa. IEEE Trans Geosci Remote Sens 46:558572. Sens Environ 16:125141.
Tang, Q., S. Peterson, R. H. Cuenca, Y. Hagimoto, and D. P. Vermote, E., C. O. Justice, and F-M. Bron. 2009. Towards a
Lettenmaier. 2009. Satellite-based near-real-time estima- generalised approach for correction of the BRDF effect in
tion of irrigated crop water consumption. J Geophys Res MODIS directional reflectances. IEEE Trans Geosci Remote
114:D05114, doi:10.1029/2008JD010854. Sens 47:898908.
Tarnavsky, E., S. Garrigues, and M. Brown. 2008. Multiscale geo- Vermote, E. F., D. Tanr, J. L. Deuze, M. Herman, and J. J.
statistical analysis of AVHRR, SPOT-VGT, and MODIS Morcrette. 1997. Second simulation of the satellite signal in
global NDVI products. Remote Sens Environ 112:535549. the solar spectrum: an overview. IEEE Trans Geosci Remote
Teillet, P. M., B. L. Markham, and R. R. Irish. 2006. Landsat cross- Sens 35:675686.
calibration based on near simultaneous imaging of com- Verreslst, J., M. E. Schaepman, B. Koetz, and M. Kneubhler.
mon ground targets. Remote Sens Environ 102:264270. 2008. Angular sensitivity analysis of vegetation indices
Teillet, P. M. and X. Ren. 2008. Spectral band difference effects derived from CHRIS/PROBA data. Remote Sens Environ
on vegetation indices derived from multiple satellite sensor 112:23412353.
data. Can J Remote Sens 34:159173. Wiegand, C. L., A. J. Richardson, D. E. Escobar, and A.H.
Townshend, J. R. G. and C. O. Justice. 1986. Analysis of the Gerbermann. 1991. Vegetation indices in crop assessments.
dynamics of African vegetation using the normalised dif- Remote Sens Environ 35:105119.
ference vegetation index. Int J Remote Sens 7:14351445. Zhang, Y., M. Xu, J. Adams, and X. Wang. 2009. Can Landsat
Trishchenko, A. P., J. Cihlar, and Z. Li. 2002. Effects of spectral imagery detect tree line dynamics. Int J Remote Sens
response function on surface reflectance and NDVI mea- 30:13271340.
sured with moderate resolution satellite sensors. Remote Zhou, X., H. Guan, H. Xie, and J. L. Wilson. 2009. Analysis
Sens Environ 81:118. and optimization of NDVI definitions and areal fraction
Tucker, C. J. 1977. Spectral estimation of grass canopy variables. models in remote sensing of vegetation. Int J Remote Sens
Remote Sens Environ 6:1126. 30:721751.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
V
Image Processing
Methods and
Approaches
9 Digital Image Processing: A r eview of the Fundamental Methods and Techniques Sunil Narumalani
andPaul Merani ........................................................................................................................................................................... 197
Acronyms and Definitions Introduction Image Quality Assessment: Basic Statistics and Histogram
Analysis Image Enhancement Image Preprocessing Principal Component Analysis Spatial Filtering Band
Ratioing and Vegetation Indices Image Classification Future Trends References
10 u rban Image Classification: Per-Pixel Classifiers, Subpixel Analysis, Object-Based Image Analysis, and
g eospatial Methods Soe W. Myint, Victor Mesev, Dale A. Quattrochi, and Elizabeth A. Wentz ................................. 219
Acronyms and Definitions Introduction Remote-Sensing Methods for Urban Classification and
Interpretation Concluding Remarks References
11 Image Classification Methods in Land Cover and Land u se Mutlu zdoan .............................................................. 231
Acronyms and Definitions Introduction Image Classification in the Context of Land-Use/Land-Cover
Mapping The Future References
12 Hyperspectral Image Processing: Methods and Approaches Jun Li and Antonio Plaza.............................................247
Acronyms and Definitions Introduction Classification Approaches Experimental Comparison Conclusions
andFuture Directions Acknowledgments References
13 Thermal Infrared r emote Sensing: Principles and Theoretical Background Claudia Kuenzer, Jianzhong
Zhang, and Stefan Dech...............................................................................................................................................................259
Acronyms and Definitions Introduction Principles, Theoretical Background, and Important Laws Potential of
Diurnal and Time Series of Thermal Infrared Remote-Sensing Data Application Examples of Thermal Infrared Data
Analyses Ground Data and Validating Thermal Infrared Data Discussion and Conclusion References
14 Object-Based Image Analysis: e volution, History, State of the Art, and Future Vision Thomas Blaschke,
Maggi Kelly, and Helena Merschdorf ........................................................................................................................................277
Acronyms and Definitions Introduction History of OBIA OBIA: A Short Summary of the State of the
Art Ongoing Developments: Influences of OBIA to Other Fields and Vice Versa Concluding Perspectives References
15 g eospatial Data Integration in OBIA: Implications of Accuracy and Validity Stefan Lang and Dirk Tiede .........295
Acronyms and Definitions Conditioned Information Object Validity Object Fate Conclusions References
16 Image Segmentation Algorithms for Land Categorization James C. Tilton, Selim Aksoy, and Yuliya Tarabalka.... 317
Acronymsand Definitions Introduction Spectrally Based Segmentation Approaches Spatially Based Segmentation
Approaches Image Segmentation Quality Evaluation Concluding Remarks References
195
2016 by Taylor & Francis Group, LLC
196 Image Processing Methods and Approaches
17 LiDAr Data Processing and Applications Shih-Hong Chio, Tzu-Yi Chuang, Pai-Hui Hsu, Jen-Jer Jaw,
Shih-Yuan Lin, Yu-Ching Lin, Tee-Ann Teo, Fuan Tsai, Yi-Hsing Tseng, Cheng-Kai Wang, Chi-Kuei Wang,
MiaoWang, and Ming-Der Yang ...............................................................................................................................................343
Acronyms and Definitions Introduction LiDAR Data Quality Assessment and Control LiDAR Data
Management LiDAR Point Cloud Feature Extraction Three-Dimensional City Modeling from LiDAR Data Full-
Waveform Airborne LiDAR DEM and DSM Generation from Airborne LiDAR Data Terrestrial/Vehicle-Borne
LiDAR Data Processing Conclusions References
2016 by Taylor & Francis Group, LLC
9
Digital Image Processing:
A Review of the Fundamental
Methods and Techniques
Acronyms and Definitions .................................................................................................................197
9.1 Introduction .............................................................................................................................198
9.2 Image Quality Assessment: Basic Statistics and Histogram Analysis ............................198
Histogram
9.3 Image Enhancement............................................................................................................... 199
9.4 Image Preprocessing ...............................................................................................................201
Radiometric Correction Geometric Correction
9.5 Principal Component Analysis............................................................................................. 207
9.6 Spatial Filtering ....................................................................................................................... 207
9.7 Band Ratioing and Vegetation Indices ................................................................................ 208
Band Ratio Vegetation Index Simple Ratio Normalized Difference Vegetation
Index Enhanced Vegetation Index 1 and 2 Wide Dynamic Range Vegetation
Sunil Narumalani Index Three-Band Model
University of Nebraska
9.8 Image Classification ................................................................................................................212
Paul Merani 9.9 Future Trends ...........................................................................................................................215
University of Nebraska References.............................................................................................................................................216
AcronymsandDefinitions GPS Global positioning system
ISODATA Iterative self-organizing data analysis
ACORN Atmospheric CORrection Now LAI Leaf Area Index
AGB Aboveground Biomass MSS Multispectral scanner
AOI Area of interest NDVI Normalized Difference Vegetation Index
APAR Absorbed photosynthetic active radiation NIR Near-infrared
ATREM ATmospheric REMoval OIF Optimum Index Factor
BR Band ratio OLI Operational land imager
BV Brightness value PC Principal component
CALMIT Center for Advanced Land Management Information PCA Principal component analysis
Technologies RGB Red, green, blue
DEM Digital elevation model SDSS Spatial decision support systems
DOQQ Digital orthophoto quarter quadrangles SR Simple ratio
EM Electromagnetic TbM Three-band Model
EPA Environmental Protection Agency TM Thematic mapper
EVI Enhanced Vegetation Index USGS United States Geological Survey
FLAASH Fast line-of-sight atmospheric analysis of spectral UTM Universal Transverse Mercator
hypercubes VBV Vegetation biophysical variable
FT Fourier transform VI Vegetation Index
GCP Ground control point VNIR Visible near-infrared
GIS Geographic information systems WDRVI Wide Dynamic Range Vegetation Index
197
2016 by Taylor & Francis Group, LLC
198 Remotely Sensed Data Characterization, Classification, and Accuracies
9.1 introduction improper processing of the raw data prior to actual data anal-
ysis (e.g., inaccurate analog-to-digital conversion). Therefore,
The modern era of commercial digital remote sensing began one of the initial tasks of an image analyst should be to assess
with the launch of Landsat 1 in 1972 (Schowengerdt, 1997). For its quality and statistical characteristics (Jensen, 2005). This
the first time ever, a satellite system was able to provide high- is normally accomplished by
spectral-quality (4-bands), medium (80 m)-spatial-resolution
Examining the frequency of occurrence of individual BVs
data that could be analyzed by public experts for numerous ter-
in the image displayed in a histogram
restrial applications. Initially, much of the digital processing of
Sample visual analysis of individual pixel BVs at specific
these data could be done only by large, expensive computer sys-
locations or within a geographic area
tems that lay in the purview of government agencies, and only
Computing univariate descriptive statistics to determine
large-format hard-copy images were made available for visual or
if there are unusual anomalies in the image data
analog analysis. With the technological evolution and computer
Computing multivariate statistics to determine the
revolution in the early 1970s, computing power began to slowly
amount of between-band correlation (e.g., to identify
trickle down to the mainstream such as educational institutions
redundancy)
and research centers across the United States and Europe. As
computing capabilities and their usage increased, their applica- Some of the basic statistical information that an analyst may find
tion to remotely sensed data also increased, and for almost two useful is found in measures of central tendency such as the mean,
decades, from the mid-1970s to the early 1990s, there continued standard deviation, and variance. These statistics provide infor-
to be sustained development and improvement of algorithms for mation about the range of BVs in each band, the relationship of
extracting image data. the BVs in each band, representation of spectral characteristics
Remotely sensed data acquired in or converted to digi- of features being examined, and an indicator of values that can
tal format are most often subject to analyses and information be used for image enhancement (e.g., a histogram stretch using
extraction using image processing techniques developed over minimummaximum values in a given band). However, for an
the past 50 years. The process encompasses a myriad of algo- in-depth analysis, other statistical measures such as variance
rithms that an analyst can apply to get meaningful information and correlation may be required to provide an insight into the
from an image. This chapter describes some of the basic image data quality and redundancy.
processing methods applied to remotely sensed data. Remote sensingderived spectral measurements for each
Remote sensing instruments (or sensors) detect and record pixel often change together in some predictable fashion because
electromagnetic (EM) energy. There are two types of senor sys- objects or features exhibit spectral behavioral patterns across the
tems, active sensors, which emit energy directed toward a target of bands. For example, clear deep water would have low, steadily
interest and then record the return, and passive sensors that record declining BVs across the blue, green, and red portions of the
energy either emitted by or reflected from an object. Data may be spectrum, until it reaches near zero in the near-infrared (NIR). If
acquired in single (panchromatic) or multiple (multi-/hyperspec- there is no relationship between the BVs in one band and that of
tral) bands of the EM spectrum. In either case, the data appear in another for a given pixel, it may imply that the values are mutu-
a matrix of x-columns by y-rows, with each square of the matrix ally independent (e.g., reflective spectra versus temperature as
referenced as a pixel and have a gray-scale value that indicates the observed in thermal infrared) or there may be an anomalous
EM energy recorded for that pixel (commonly referred to as the observation for a given feature (e.g., sedimentation present in
brightness value (BV)). The range of BVs recorded in each band deep water). In most cases, spectral measurements of individual
across the image is dependent on the radiometric resolution of pixels may not be independent, and a measure of their mutual
the sensor system (e.g., values ranging from 0 to 255 would be interaction is reflected in the covariance or the joint variation of
recorded for a sensor system with an 8-bit radiometric resolution). two variables about their common mean.
Jensen (2005) notes that most remote sensing studies are based To estimate the degree of interrelation between variables in
on developing deterministic relationships between the EM signals a manner not influenced by measurement units, the correla-
recorded in various bands of the spectrum and the chemical or tion coefficient is commonly used. The correlation between two
biophysical properties of the features being investigated. bands of remotely sensed data, rk,l, is the ratio of their covariance
(covk,l) to the product of their standard deviations (sksl); thus,
9.2 image Quality Assessment: Basic
cov k ,l
Statistics and Histogram Analysis rk ,l =
s k sl
Many remote sensing data are of high quality. However, on
occasions, errors (or noise) are introduced into the data by If we square the correlation coefficient (rk,l), we obtain the sam-
numerous factors such as the environment (e.g., atmospheric ple coefficient of determination (r2), which expresses the pro-
scattering), random or systematic malfunction of the sensor portion of the total variation in the values of band l that can
system (e.g., an uncalibrated detector creates striping), or be accounted for or explained by a linear relationship with the
2016 by Taylor & Francis Group, LLC
Digital Image Processing 199
values of the random variable band k. Thus, a correlation coef- that aid in enhancing an image include image reduction and
ficient (rkl) of 0.70 results in an R2 value of 0.49, meaning that magnification, spatial and spectral profiles, image contrasting,
49% of the total variation of the values of band l in the sample density slicing, and composite generation (Table 9.1). Higher-
is accounted for by a linear relationship with values of band k. order image enhancement techniques include band ratio-
In order to optimize usage of multiple bands, scientists prefer ing, spatial filtering, edge enhancement, and spectral image
higher variance and lower correlation. Furthermore, the corre- transformation.
lation and covariance information can be used for analysis by Reduction and magnification operations are used to adjust
advanced image processing functions such as principal compo- the image scale visually in order to provide either a regional
nent analysis (PCA) and image classification. perspective (i.e., display of an entire scene at a small scale) or
a zoom-in of an area of interest (AOI) for closer examination
(Figure 9.3). They allow an analyst to derive image coordinates
9.2.1 Histogram
(x, y locations) and per-pixel spectral data of features across,
The histogram is the most fundamental and useful graphical and gain an understanding of the spatial distribution of objects
representation of information content in an image. It tabulates across the landscape.
frequencies of occurrences of each BV, displays them graphically, An analyst may further their understanding of the image
and provides information on contrast within each band. Peaks landscape by deriving spatial and spectral profiles along user-
and valleys often correspond to the dominant land cover types specified transects (Jensen, 2005). The pixels that lie along that
in an image, and the information can be converted into percent transect can be measured and displayed to compare the spec-
representation to highlight information content by masking out tral (BVs) or spatial differences (coordinate space). Multiple
BVs with high frequency (or vice versa) (Figure 9.1). Basically, transects may be used to determine spatial patterns or trends.
histograms may provide the user with information on the qual- Transects can also be used to assist in density slicing an image
ity of the image (e.g., high-contrast, low-contrast, bimodal, and or a portion of it (Figure 9.4).
multimodal) and are often used to enhance imageryfor exam- Density slicing is a pseudocolor enhancement technique nor-
ple, brightening up darker areas in an image or conversely dark- mally applied to a single-band monochrome. It is considered an
ening up extremely bright areas of an image (Figure 9.2). effective way of highlighting different but apparently homoge-
neous areas within an image by slicing the range of grayscale
9.3 image enhancement values (e.g., 0255) and assigning different colors to each of those
slices (Figure 9.5). This technique is often used in conjunction
To improve the appearance of an image for visual analysis or with a vegetation index (VI) such as the Normalized Difference
at times even for subsequent computer analysis, an analyst may Vegetation Index (NDVI) to highlight variations in the density
prefer to apply select algorithms. A basic suite of algorithms of biomass.
IKONOS band 4 (NIR)
Water
Land
FIg u r e 9.1 Histogram of the near-infrared band (band 4) of the IKONOS sensor system. The near-infrared band is useful for land/water delin-
eation, and this is evident in the histogram where clear deep water pixels have very low reflectivity while the land pixels record higher BVs in the
band. (Image courtesy of the Center for Advanced Land Management Information Technologies (CALMIT), University of Nebraska, Lincoln, NE.)
2016 by Taylor & Francis Group, LLC
200 Remotely Sensed Data Characterization, Classification, and Accuracies
Red band
Histogram
Min. Max.
Linear stretch histogram
Min. Max.
Near infrared band
Histogram
Min. Max.
Linear stretch histogram
Min. Max.
FIg u r e 9.2 Example of histogram stretch performed on the red and near-infrared bands of a WorldView 2 image acquired for the Grand
Bay, MS area. (Image courtesy of the Center for Advanced Land Management Information Technologies (CALMIT), University of Nebraska,
Lincoln, NE.)
Another effective method or visual analysis is composite gen- thus generating a true color image (Figure 9.6a). Similarly, to
eration. This method utilizes the three planes of a computers display a false color composite of the same image, OLI bands 5,
display device (red (R), green (G), and blue (B)) and allows the 4, and 3 would be placed in the R, G, and B planes (Figure 9.6b).
analyst to place different bands of a multispectral image into InFigure9.6b, the NIR band (OLI band 5) is placed into the red
various planes. For example, to generate a true color composite plane of the display and is often used for vegetation analysis
from Landsat Operational Land Imager (OLI) image, one would because of high spectral reflectance of vegetation in the near-
insert bands 4, 3, and 2 into the R, G, and B planes respectively, and mid-infrared portions of the spectrum.
2016 by Taylor & Francis Group, LLC
Digital Image Processing 201
TABLe 9 .1 Image Enhancement Techniques, Their Effects, and Examples of Application
Image Enhancement
Technique Effect on Image Example of Application
Image magnification Zooms into the area of interest or closer
observation of feature
Image reduction Zooms out either partially or completely from Enable a geographic or spatial analysis of an entire landscape
an image to provide a large-area perspective
Spatial profiles Draw a transect across the area of interest Changes in features across the transect (e.g., forest to
grasslands, to water)
Spectral profiles Draw a transect across the area of interest Changes in spectral signature observed along the transect as
variations in features occur
Density slicing Color coding a band based on BV ranges Highlight specific features or a rapid visualization of potential
land cover observed in the selected band
Image composites Representation of land cover features in various Geologic highlighting of a mineral may use a combination of
colors based on the band combinations used certain bands based on the spectral properties of that mineral
Reduction
0 1 5 7 3 8 4 4 0 5 3 4 - - - -
2 4 5 5 7 3 8 9 1 6 9 7 - - - -
1 3 6 8 9 2 7 7 0 4 8 5 - - - -
1 2 4 8 7 5 6 5 1 5 4 1 - - - -
0 1 4 9 8 6 5 3 - - - - - - - -
0 3 5 7 6 5 2 1 - - - - - - - -
1 3 5 5 4 3 1 1 - - - - - - - -
2 4 7 5 5 3 2 0 - - - - - - - -
(a)
Magnification
0 1 5 7 3 8 4 4 0 0 1 1 5 5 7 7
2 4 5 5 7 3 8 9 0 0 1 1 5 5 7 7
1 3 6 8 9 2 7 7 2 2 4 4 5 5 5 5
1 2 4 8 7 5 6 5 2 2 4 4 5 5 5 5
0 1 4 9 8 6 5 3 1 1 3 3 6 6 8 8
0 3 5 7 6 5 2 1 1 1 3 3 6 6 8 8
1 3 5 5 4 3 1 1 1 1 2 2 4 4 8 8
2 4 7 5 5 3 2 0 1 1 2 2 4 4 8 8
(b)
FIg u r e 9 .3 Concept diagram showing the pixels displayed when (a) 2 image reduction and (b) 2 image magnification are applied.
9.4 image Preprocessing extract information from the image. Often known as image
restoration, the process produces a corrected image that is as
Fureder (2010) states that the operational use of remote sens- close as possible, both geometrically and radiometrically, to the
ing data is often limited due to sensor variation, atmospheric radiant energy characteristics of the actual scene. This requires
effects, as well as topographically induced illumination effects. that internal and external errors be determined and corrected
Image preprocessing is a preparatory phase that, in principle, for. Internal errors are created by the sensor itself and are gen-
improves image quality as the basis for later analyses that will erally systematic and stationary (i.e., predictable and constant).
2016 by Taylor & Francis Group, LLC
202 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 9.4 Example of spatial profiles acquired for the Savannah River over three transects along the channel. (Adapted from Jensen, J.R.,
Introductory Digital Image Processing: A Remote Sensing Perspective, Prentice Hall, Upper Saddle River, NJ, 2005.)
Land Cover
Lincoln, NE
Water
Urban
Trees
Sparse vegetation
Healthy vegetation
W E
S
Density slice
Landsat 8
OLI Band 5 (NIR)
FIg ur e 9.5 An example of a density-sliced image acquired by the Landsat OLI system of Lincoln, NE. (Image courtesy of the Center for
Advanced Land Management Information Technologies (CALMIT), University of Nebraska, Lincoln, NE.)
2016 by Taylor & Francis Group, LLC
Digital Image Processing 203
0 5 10 20 30 40 50 miles
(a) (b)
FIg u r e 9.6 (a) True color image of the Gibraltar area in the Mediterranean Sea of the Landsat OLI sensor system. (b) Color infrared image of
the same region. Note that healthy green vegetation appears bright red, urban areas shades of white, and water (depending on the turbidity) from
blue (turbid water) to dark gray (deep, clear water). (Image courtesy of the Center for Advanced Land Management Information Technologies
(CALMIT), University of Nebraska, Lincoln, NE.)
External errors are due to perturbations and modulation of Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes
scene characteristicsthat is, they are variable and are corrected (FLAASH) by Exelis (formerly Research Systems, 2003). It is
by relating points on ground to sensor measurements. important to note that application of these algorithms for a given
scene and date requires knowledge of the spectral profile and
atmospheric properties for that date and time. This information
9.4.1 Radiometric correction
is extremely difficult to acquire; however, these models can pro-
As radiation passes through the atmosphere, it undergoes sev- vide a close approximation of the reflectance for the scene sans
eral different processes, including absorption, scattering, atten- the atmosphere, versus an atmospherically uncorrected scene.
uation, and transmission (Schowengerdt, 1997; Jensen, 2005). Relative atmospheric correction is often used if an analyst
Various methods of atmospheric correction can be applied wants to normalize the BV among the various bands of a sin-
ranging from detailed modeling of the conditions during data gle scene or normalize multi-date imagery to a single/standard
acquisition, to simple calculations based solely on the image scene selected among the dataset (Jensen, 2005). An example of
data. Broadly, atmospheric correction can be divided into two the former method is to examine the observed BVs in an area
types: (1) absolute and (2) relative. Absolute radiometric cor- of shadow or for a very dark object (such as a large clear lake or
rections turn the BVs into scaled surface reflectance values (Du an asphalt surface) and determine the minimum value. The cor-
etal., 2002) and attempt to model the atmosphere, as it would rection is applied by subtracting the minimum observed value,
exist at the time of image acquisition. Several radiative transfer determined for each specific band, from all pixel values in each
models have been developed as part of the absolute radiometric respective band. Because scattering is wavelength dependent,
correction efforts, including the Atmospheric CORrection Now the minimum values will vary from band to band. This method
(ACORN) by ImSpec (2002), ATmospheric REMoval (ATREM) is based on the assumption that the reflectance from these fea-
program by the University of Colorado (Tanre etal., 1986), and tures, if the atmosphere is clear, should be very small (if not
2016 by Taylor & Francis Group, LLC
204 Remotely Sensed Data Characterization, Classification, and Accuracies
zero). If values are much greater than zero, then they are consid- objects as well as scale variations are minimized or removed
ered to have resulted from atmospheric scattering. entirely (Aronoff, 2005). Geometrically corrected imagery
Multi-date image normalization techniques involve the can be used to extract accurate distance, area, and direction
selection of a base image and then transforming the spectral (bearing) information, and any information derived from
characteristics of all other images to have the same radio- such images can be related to other thematic information in
metric scale as the base image (Jensen, 2005). The method a geographic information systems or spatial decision support
involves the selection of pseudo-invariant features (i.e., radio- systems (Jensen, 2005).
metric ground control points (GCPs)) from the base image, Remotely sensed imagery collected from airborne or space-
identifying the BVs of the same features across all the multi- borne sensors often contains internal and external geometric
date imagery and normalizing them to the base image. The errors (Jensen, 2005). These can be systematic (predictable) or
pseudo-invariant features need to meet specific spectral and nonsystematic (random), and generally, systematic geometric
spatial criteria (Eckhardt etal., 1990; Hall etal., 1991; Jensen error is easier to identify and correct than random geometric
etal., 1995). error. Some of these errors can be corrected by using ephemeris
In addition to atmospheric effects, the landscape elements such of the platform and known internal sensor distortion charac-
as slope and aspect can cause radiometric distortion of the signal teristics. Commercial satellite data (e.g., SPOT Image, Landsat,
received by the sensor system (Jensen, 2005). Teillet etal. (1982) QuickBird, GeoEye, and others) already have much of the sys-
describe four topographic correction methods, including (1) cosine tematic error removed. Other errors can be corrected only by
correction, (2) Minnaert correction, (3) statisticalempirical cor- matching image coordinates of physical features recorded by
rection, and (4) C-correction. Each correction method is based the image to the geographic coordinates of the same features
on illumination and requires a digital elevation model (DEM) collected from a map or global positioning system (GPS).
to determine how much illumination each pixel receives relative Internal geometric errors are introduced by the remote
to its topography in the landscape. Much research continues on sensing system itself or in combination with Earth rotation or
the removal of topographic effects from the scene. For example, curvature characteristics. These distortions are often system-
Civco (1989) identifies several considerations including matching atic (predictable) and may be identified and corrected using
the DEM spatial resolution to that of the image, overcorrection of prelaunch or in-flight platform ephemeris (i.e., information
topographic effect because of the Lambertian surface assumption, about the geometric characteristics of sensor and the Earth at
ignoring that the diffuse component also illuminates the topog- data acquisition). Geometric distortions in imagery that can
raphy, strong anisotropy of apparent reflectance (Leprieur etal., sometimes be corrected through analysis of sensor character-
1988), and consideration to wavelength and deeply shadowed istics and ephemeris data include
areas (Kawata etal., 1988). Readers should refer to more detailed
Skew caused by Earth rotation effects
discussions of radiometric correction in Chapters 3 and 4.
Scanning systeminduced variation including ground
Noise in an image may be due to irregularities or errors
resolution cell size, relief displacement, and tangential
that occur in the sensor response and/or data recording and
scale distortion
transmission (van der Meer etal., 2009). Common forms of
noise include systematic striping or banding and dropped External geometric errors are usually introduced by phenom-
lines. Early Landsat MSS data had substantial striping due to ena that vary in nature through space and time. The most
variations and drift in the response over time of the six MSS important external variables that can cause geometric error
detectors. The drift was different for each of the six detectors, in remote sensor data are random movements of the remote
causing the same brightness to be represented differently by sensing platform at the time of data collection (i.e., altitude
each detector (Fureder, 2010). The overall appearance was and attitude changes). Unless otherwise processed, however,
thus a striped effect. The corrective process made a relative unsystematic random error remains in the image, making
correction among the six sensors to bring their apparent val- it non-planimetric. To correct for these errors, two common
ues in line with each other. Dropped lines occur when there geometric correction procedures are used to make the digital
are system errors that result in missing or defective data remote sensor data of value:
along a scan line and is often corrected by replacing the line
Image-to-map rectification
with the pixel values in the line above or below, or with the
Image-to-image registration
average of the two.
Image-to-map rectification is the process by which the geome-
try of an image is made planimetric. Whenever accurate area,
9.4.2 Geometric correction
direction, and distance measurements are required, image-to-
The stability of a remote sensing platform, the curvature of the map geometric rectification should be performed; however,
earth, sensor orientation, topography, and other factors cause it may not remove all the distortion caused by topographic
geometric distortion in an image. Consequently, geometric relief displacement in an image. The image-to-map rectifica-
correction is applied to remove these distortions so that the tion process normally involves selecting well-identified GCPs
image is planimetrically (x, y) correct and the displacement of on an image and associating them with their planimetrically
2016 by Taylor & Francis Group, LLC
Digital Image Processing 205
correct map counterparts (i.e., GCPs from a paper or digi- location. Therefore, a new BV has to be assigned to the recti-
tal map for which geographic coordinates can be derived fied pixel. Three methods for such resampling (Figure 9.7) are
e.g., meters northing and easting in a Universal Transverse as follows:
Mercator (UTM) map projection). Alternatives to obtaining
Nearest neighbor
accurate GCP map coordinate information for image-to-map
Bilinear interpolation
rectification include the following:
Cubic convolution
Hard-copy planimetric maps (e.g., USGS 7.5min 1:24,000-
The nearest neighbor algorithm assigns the BV of the clos-
scale topographic maps) where GCP coordinates are
est input x, y to the output x, y. This method maintains the
extracted using simple ruler measurements or a coordi-
integrity of the data and is not computationally intensive.
nate digitizer.
Unfortunately, the output image may not be aesthetically
Digital planimetric maps (e.g., the USGS digital 7.5 min
pleasing as it has a block appearance. Bilinear interpolation
topographic map series) where GCP coordinates are
derives the new BV of the pixel based on the weighted value
extracted directly from the digital map on the screen.
of the four pixels nearest to those in the original image. This
Digital orthophotoquads that are already geometri-
method produces a smoother image, but has a slight impact
cally rectified (e.g., USGS digital orthophoto quarter
on the integrity of the data. Cubic convolution determines the
quadrangles.
output BV based on the weighted values of 16 input pixels sur-
GPS instruments that may be taken into the field to obtain
rounding the location of the original pixel. This method pro-
the coordinates of objects to within 20 cm if the GPS
duces a smoother image but is computationally intensive and
data are differentially corrected.
may have a considerable impact on the integrity of the data
Once GCP map coordinates are obtained for several points on (Figure 9.8; Table 9.2).
an image, spatial interpolation algorithms are applied to trans- Image-to-image registration is the translation and rotation
form the image coordinates to the map coordinates, thus mak- alignment process by which two images of like geometry and
ing the image planimetrically correct. of the same geographic area are positioned coincident with
respect to one another so that corresponding elements of the
Polynomial equations are used to convert the file coordi-
same ground area appear in the same place on the registered
nates into rectified map coordinates.
images (Chen and Lee, 1992; Jensen, 2005). This type of geo-
Depending on the distortion of the imagery, complex
metric correction is used when it is not necessary to have each
(higher-order) polynomial equations may be required to
pixel assigned a unique x, y coordinate in a map projection.
express the needed transformation.
For example, we might want to make a cursory examination
The degree of complexity of the polynomial is expressed
of two images obtained on different dates to see if any change
as the order of the polynomial (i.e., the highest exponent
has taken place. In such a case, we need to register only the
used in the polynomial).
two (or more images) to a base image (selected among the
Intensity interpolation arises from the fact that there is no available dataset) and perform a rapid visual analyses of
one-to-one relationship between the input and output pixel the data.
Outline of rectified output raster
(black outlines), superimposed over
original input image (blue outlines).
Current target pixel
Input cells used for each resampling
method for the current target pixel.
Nearest neighbor
+ Bilinear interpolation
+ + Cubic convolution
FIg u r e 9.7 Schematic diagram showing the comparison between nearest neighbor, bilinear interpolation, and cubic convolution algorithms
for intensity interpolation. (Adapted from http://marswiki.jrc.ec.europa.eu/wikicap/index.php/Resampling_techniques_in_image_processing.
Accessed June 18, 2014.)
2016 by Taylor & Francis Group, LLC
206 Remotely Sensed Data Characterization, Classification, and Accuracies
Original Nearest neighbor
(a)
Bilinear interpolation Cubic convolution
(b) (c)
FIg u r e 9.8 An example of three-intensity interpolation techniques used for the geometric rectification of remotely sensed data: (a) the nearest-
neighbor algorithm that uses the closest pixel location BV to assign to the corrected data; (b) the bilinear interpolation that uses a weighted aver-
age of the four spatially closest pixels to assign a BV to the corrected data; and (c) the cubic convolution algorithm that uses the weighted average
of the 16 closest pixels to determine the BV of the rectified pixel. (Image courtesy of the Center for Advanced Land Management Information
Technologies (CALMIT), University of Nebraska, Lincoln, NE.)
TABLe 9 .2 Geometric Correction Methods and Their Comparative Advantages and Disadvantages
Intensity Interpolation Technique Pixels Used for Interpolation Advantages Disadvantages
Nearest neighbor 1 Maintains data integrity Blocky appearance
Not computationally intensive
Bilinear interpolation 4 Minimal disruption to data integrity Contrast may be reduced
Smooth appearance BVs are interpolated
Cubic convolution 16 Improved appearance BVs are highly manipulated
2016 by Taylor & Francis Group, LLC
Digital Image Processing 207
9.5 Principal component Analysis By computing the correlation between each band and each
PC, it is possible to determine how each band loads or is associ-
The majority of remote sensing data are acquired in many dif- ated with each PC. A linear combination of original BV and fac-
ferent bands leading to the generation of vast quantities of data. tor scores (eigenvectors) produces the new BV for each pixel of
Because of the spectral proximity of some bands in a multispec- every PC. It is often the case that the majority of the information
tral dataset and certainly in the case of hyperspectral imag- contained in a multispectral dataset can be represented by the
ery, there is often a high degree of correlation between bands, first three or four PCA components. Higher-order components
implying that there may be similar information content between may be associated with noise in the original dataset.
them. For example, Landsat Thematic Mapper (TM) bands 2
and 3 (green and red, respectively) typically have similar visual 9.6 Spatial Filtering
appearances because reflectances for the same cover types are
almost equal. Image transformation techniques based on com- For any given image, or part thereof, there are changes in BVs
plex processing of the statistical characteristics of multiband throughout the scene. The number of changes in BVs per unit
datasets can be used to reduce this redundancy and correlation distance for any particular part of the image is called spatial fre-
between bands. One such transformation is called PCA whose quencythat is, the roughness of the tonal variations occurring
objective is to reduce the dimensionality (i.e., the number of in an image (Jensen, 2005). Figure 9.10(a) and (b) demonstrates
bands) in the data and compress as much of the information in the differences between low-frequency (less roughness) and
the original bands into fewer bands. The new bands that result high-frequency (more roughness) images. In a low-frequency
from this statistical procedure are called components. The pro- area, the changes in BVs are subtle over the given area, while the
cess attempts to statistically maximize the amount of informa- opposite is true in a high-frequency image.
tion (or variance) from the original data into the least number of To extract quantitative information, local operations are per-
useful new components. formed (spatial filtering), and the BV of a given pixel is modified
PCA transforms the axes of the multispectral space such that based on the values of neighboring pixels.
it coincides with the directions of greatest correlation. Each of Depending on the features to be extracted, filters can be
these new axes are orthogonal to one another; that is, they are applied to an image. A filter (or a convolution mask/kernel) is a
at right angles and the component images are arranged such moving window function that defines a small sub-window with
that the greatest amount of variance (or information) within a dimension of 3 3 or larger and usually with odd-numbered
the original dataset is contained within the first component and dimensions (e.g., 3 3, 5 5, and 7 7). An example of a 33
the amount of variance decreases with each component (Jensen, window is shown in Figure 9.11, with pixels numbered from the
2005; Figure 9.9). Transformation of original data on X1 and X2 top left. In this example, pixel C2,2 in the window is the center
axes onto PC1 and PC2 axes requires transformation coefficients pixel, and odd-numbered window sizes ensure that there is
that can be applied in linear fashion to original pixel values always a center pixel in the sub-window.
(Figure 9.9). These new axes are called first PC. The second PC is Filtering involves computing a weighted average of the pixels
perpendicular (orthogonal) to PC1 (Gonzalez, 2014). Subsequent in the moving window. The choice of weights determines how the
components contain decreasing amounts of the variance found filter affects the image. A window of weight values is called a con-
in the dataset. volution kernel. Multiplying each pixel in the moving window by
Principal components analysis
X2 X2 PC2
X2
PC1
Brightness values
2 2 X1
X1 X1
1 1
(a) Brightness values (b) Brightness values (c) Brightness values
FIg u r e 9.9 Concept diagram illustrating the PCA process: (a) the cluster of BVs from two bands of an image, (b) a new coordinate system
defined by the X, and (c) the PCA transformation that occurs by rotating to the new axis, which is orthogonal to the original X axis. The new axes
are no longer the bands of the original image, but derivative components from those data. (Adapted from Jensen, J.R., Introductory Digital Image
Processing: A Remote Sensing Perspective, Prentice Hall, Upper Saddle River, NJ, 2005.)
2016 by Taylor & Francis Group, LLC
208 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) (b)
FIg u r e 9 .10 Examples of (a) high-frequency image and (b) low-frequency image.
information) and are useful where lower-frequency information
tends to hide parts of the scene of interest, for example, roads
C1,1 C1,2 C1,3
in an urban scene. When building a high-pass filter, the center
pixel of the kernel is given more weight. Consequently, if it is an
edge, then the pixel will be greatly enhanced because edges have
C2,1 C2,2 C2,3 higher pixel values.
Spatial filtering methods can also be used to remove noise
in the data (e.g., striping or speckles). Jensen (2005) used the
C3,1 C3,2 C3,3 Fourier transform (FT) on Landsat TM data to remove strip-
ing in a coastal area near Al Jubail, Saudi Arabia. The process
involved the computation of an FT for the study area, modifying
FIg u r e 9.11 Schematic of 3 3 convolution kernel. Values are the resulting FT by selectively removing the points associated
assigned to each cell depending on the type of information that an ana- with the striping and then computing a reverse FT to derive a
lyst may wish to extract. clean, destriped image. Destriping of an image can also be done
by running a low-pass and a high-pass filter on an image and
its weight and summing all the products yield a new value for the then adding the filter outputs. The resulting image will have its
center pixel (Figure 9.12). The values used in a convolution kernel striping removed or minimized (USGS, 2007). Table 9.3 broadly
define whether the filter is low pass or high pass. describes the types of filters used on remotely sensed data and
Low-pass filters are designed to emphasize low-frequency their utility.
features and de-emphasize the high-frequency components of an
image. Thus, information changing very fast across a landscape 9.7 Band Ratioing and
(e.g., in an urban area) will be subdued while low-frequency Vegetation indices
information (e.g., grassland, water) is preserved. Low-pass fil-
ters are excellent for retaining low-frequency information and Another useful image processing technique exploits the rela-
are useful for removing noise (such as speckle) in an image tionships among the BVs of different bands, of image fea-
(Jensen, 2005). Low-pass filters make similar cover areas appear tures. Mathematical expressions are applied to image bands
uniform and can be useful for boundary detection (Figure 9.13). in order to extract thematic information. These expressions
Conversely, low-pass filters do not preserve edges, and larger may be a simple ratio (SR), or complex equations, and are gen-
window sizes lead to greater smoothing. erally developed to target a specific feature of interest. Many
High-pass filters emphasize the detailed high-frequency com- such algorithms have been developed to highlight character-
ponents of an image and deemphasize the more general low- istics of land cover such as vegetation, soil, water, and urban
frequency information. They enhance image details (infrequent areas, and the information extracted can be applied to a wide
2016 by Taylor & Francis Group, LLC
Digital Image Processing 209
13 39 42 1 1 1 13 39 42
28 55 85 1 1 1 = 28 55 85
73 84 87 1 1 1 73 84 87
(a) Original image (b) Convolution kernel (c) Resulting image
9
i = 0 ci BVi
BVout = int n
BVout = 1 3+ 3 9+ 4 2+ 2 8+ 5 5+ 8 5+ 7 3+ 8 4+ 8 7= 506
BVout = 506/9 BVout = 56.22
FIg u r e 9.12 Schematic diagram showing the process and results of applying a low filter. The original center cell value was (a) 55.
With the application of a convolution kernel (b) with equal weights for all 9 cells, an output image (c) is created. Further, the formula
described earlier sums the values of the nine cells and produces the average value of 56.22. Rounded off to the nearest whole number,
the new value of the center cell will be 56.
range of analyses (see Table 9.4 for examples of ratios and decision process may be as simple as the analyst displaying the
vegetation indices). results of multiple ratios and choosing the resulting dataset that
appears most visually appealing or informative. However, there
are widely used techniques for determining optimum bands for
9.7.1 Band Ratio ratioing, such as the Optimum Index Factor and the Sheffield
The BVs of specific targets of interest vary from image to Index (Chavez etal., 1984; Sheffield, 1985).
image depending on environmental factors, including topog-
raphy, slope of target surface, aspect ratio, solar angle, seasonal 9.7.2 Vegetation index
changes, atmospheric conditions, water content, substrate con-
ditions, or shadowing. Such factors may significantly increase Vegetation is a critical component of the health and condition
or decrease BVs relative to what would be expected in labo- of the Earths natural environment. The U.S. Environmental
ratory conditions. This may make complex image analysis Protection Agency (EPA) cites vegetation characteristics and
functions such as classification, feature discrimination, and biophysical variables (such as biomass and percentage of
change detection, difficult to perform. However, certain ratio cover) and key indicators of ecosystem health (EPA/600/s-05,
transformations applied to two or more spectral bands can 2010). Because of this, vegetation studies have been a popular
minimize such effects. In addition, these ratios may generate subject of remote sensing research since the 1960s. Scientists
unique information, not otherwise attainable, through visual have modeled biophysical characteristics of vegetation using
image analysis techniques. The mathematical expression of the digital imagery since the data became available and continue
band ratio is to do so today (Hardisky etal., 1986; Gross and Klemas 1988;
Gross et al., 1993; Zhang et al., 1997, 2009; Spanglet et al.,
1998; Mishra etal., 2012). Many of these studies involve the
BV, Px Bx
BR , Px = use of vegetation indices, complex mathematical equations
BV , Px B y applied to image bands to measure the relative greenness of
image features, from which meaningful information may
where BR, Px is the output value for a pixel (Px) using the BVs of be extracted of the composition and characteristics of veg-
two bands: band x (Bx) and band y (By). One obvious problem etation. Such vegetation biophysical variables (VBVs) may
becomes clear that BR, Px = 0 is a possible outcome. There are include, but are not limited to, above-ground green biomass
several methods to address this, however, including assigning a (AGB), absorbed photosynthetic active radiation, concentra-
value of 1 to any BV with a value of 0 or adding a small value to tion of chlorophyll (or other leaf pigment), leaf area index
the denominator if it equals zero (such as 0.1). (LAI), and percent of substrate covered by vegetation (vegeta-
While band ratios provide a new series of BVs to evaluate, it is tion fraction). Often, these biophysical properties can be key
not always easy to determine which bands should be used to pro- indicators of vegetation health, ecosystem health, and other
vide information on specific targets of interest. Sometimes, the critical ecological factors.
2016 by Taylor & Francis Group, LLC
210 Remotely Sensed Data Characterization, Classification, and Accuracies
Original Low pass
High pass Sharpen
FIg u r e 9.13 Comparison of application of a low-pass filter versus a high-pass filter versus a sharpening filter. (Adapted from USGS, The power
of spatial filters, http://astrogeology.usgs.gov, accessed June 20, 2014, 2007.)
TABLe 9 .3 Filtering Techniques and Their Utility in Analyzing Remotely Sensed Data
Filtering Technique What It Does Filter Examples
High frequency Allows high-frequency information to pass through Enhancing structural details
Suppresses low-frequency information Bring out boundaries and edges
Edges are sharp and small features stand out
Large features look suppressed
Low frequency Allows low-frequency information to pass through Highlight larger features
Suppresses high-frequency information Bring out information in larger features
Edges get subdued
Larger features are enhanced
Smaller features begin to get smoothed
Edge enhancement Detect edges/boundaries between features Aid in automated feature extraction
Useful for geologic information, urban areas, boundaries, etc.
Frequency domain Converts data from spatial to frequency domain Enhancement, compression
Noise removal, image restoration
Textural classification, quality assessment
2016 by Taylor & Francis Group, LLC
Digital Image Processing 211
TABLe 9 .4 Examples of Some of the Common Ratio and Vegetation Indices
Index Formula Source
red
Simple ratio SR = Birth and McVey (1968)
nir
NDVI =
( nir red )
Normalized Difference Vegetation Index Rouse etal. (1974)
(nir + red )
Soil-adjusted Vegetation Index SAVI =
( nir red ) Huete (1988)
(nir + red + a ) (1 + a )
Green Normalized Difference Vegetation Index GNDVI =
( nir green ) Buschmann and Nagel (1993)
(nir + green )
nir green ( blue red )
Green Atmospherically Resistant Vegetation Index GARI = Gitelson etal. (1996)
nir green + ( blue red )
EVI = 2.5
( nir red )
Enhanced Vegetation Index Huete etal. (1996)
(nir + 6(red ) 7.5(blue ) + 1)
Visible Atmospherically Resistant Index VARI =
( green red ) Gitelson etal. (2002)
(green + red blue )
a ( nir red )
Wide Dynamic Range Vegetation Index WDRVI = Gitelson (2004)
a (nir + red )
TbM = ( 1 ) ( 2 ) ( 3 )
1 1
Three-band Model Gitelson etal. (2006)
EVI2 = 2.5
( nir red )
Enhanced Vegetation Index 2 Jiang etal. (2008)
(nir + 2.4 ( red ) + 1)
In order to maximize the ability to extract meaningful developed early target the inverse relationship between red and
information, a VI should have four characteristics (Running NIR reflectance. However, algorithms have been successfully
et al., 1994; Huete and Justice, 1999; Gitelson et al., 2006). developed using other characteristics of vegetation reflectance
First, the VI must be sensitive to VBVs of interest. It is help- in a variety of environments and/or sensor specifications (Via
ful if the sensitivity demonstrates a predictable relationship etal., 2011).
between index and VBV, preferably, a linear relationship that
is applicable across a wide range of vegetation conditions,
substrates, and species-types. Second, the impact of exter- 9.7.3 Simple Ratio
nal variables such as atmospheric interaction, solar angle, Birth and McVey (1968) described the SR, one of the first docu-
and viewing angle must be minimized. This is necessary to mented VIs that provides a simple formula for measuring the
compare multiple datasets with consistent spatial and tem- ratio of red reflectance (red in% or dimensionless) to NIR reflec-
poral conditions. Third, the impact of internal variables such tance (nir):
as canopy architecture, substrate, phonological changes, and
nonphotosynthetic canopy components should be minimized. red
Such internal variables can contribute substantially to the SR =
nir
recorded spectral response of vegetation in digital imagery
and coarse spatial resolution may contribute to poor results
Green vegetation strongly reflects incident irradiation in the
when comparing multiple image datasets. Fourth, accuracy
NIR region (40%60%) while absorbing up to 97% in the red
assessment must be tied to a specific measureable VBV such
region (Gitelson, 2004). As vegetation greenness declines,
as AGB, LAI, or vegetation fraction.
red reflectance increases and NIR reflectance decreases. By
Since the 1960s, several vegetation indices have been devel-
computing the ratio of red to NIR, this relationship can be
oped, though some of these provide redundant information
quantified.
content (Jensen, 2005). Subtle differences in algorithms are
often adopted due to variability in sensor specifications and/
or target characteristics. Vegetation, however, is a spectrally
9.7.4 normalized Difference Vegetation index
unique surface feature due to the chlorophyll absorption and
leaf reflectance characteristics in the visible through near- One of the most noticeable problems with the SR is that it is
infrared (VNIR) regions of the spectrum. Many of the VIs not normalized, making it difficult to compare results among
2016 by Taylor & Francis Group, LLC
212 Remotely Sensed Data Characterization, Classification, and Accuracies
different studies. Rouse etal. (1974) addressed this issue with the 9.7.7 three-Band Model
NDVI. NDVI is functionally equivalent to SR, and comparison
plots reveal no scatter between SR and NDVI. Most VIs focus on chlorophyll absorption characteristics.
Gitelson etal. (2006) proposed an index that may be optimizable
(nir red ) for other pigments and potentially other features of interest. This
NDVI = three-band model (TbM) requires the use of three spectral bands
(nir + red )
that must be identified as follows:
NDVI is widely applied to spectral and image data for monitoring, Band 1 (1): The band that is most sensitive to changes in
analyzing, and mapping VBVs. There are several characteristics VBV.
of NDVI that contribute to its utility and continuing popularity Band 2 (2): The band that is the most insensitive to
among vegetation experts: changes in VBV.
Band 3 (3): The band that accounts for backscattering/
Seasonal and phonological changes in vegetation can be
noise among samples.
monitored.
Normalized data make comparisons more reliable.
Ratioing reduces some cases of multiplicative noise cause TbM = ( 1 ) ( 2 ) ( 3 )
1 1
by differences in solar angle, shadows, and topographic
variations.
TbM generates a linear relationship and, by accounting for
Conversely, a major disadvantage to NDVI is the nonlinear backscatter/noise, normalizes the relationship between VBV
nature of the relationship between NDVI values and many and index value. Additionally, TbM has the potential to be
VBVs. The index becomes saturated at high levels, and as VBVs applied to other features of interest, including non-vegetation
increase, NDVI shows little variation. surface types (Figure 9.14).
9.7.5 enhanced Vegetation index 1 and 2
Several VIs are tailored to specific sensors and may be tuned to
9.8 image classification
maximize the results of analysis at specific resolution charac- With the advent of digital image data and computer technology,
teristics. An example of this is the Enhanced Vegetation Index algorithms were developed to enable the extraction of land use/
(EVI), which was developed by Huete et al. (2002) specifically land cover and biophysical information directly from the images
for application to MODIS data. EVI is similar to NDVI; how- (Jensen et al., 2009.) Over the past five decades, numerous
ever, it includes several coefficients in the equation to account algorithms have been developed to aid analysts in their image
for atmospheric scattering and to reduce the saturation effects of interpretation processes. Broadly, these algorithms are based
NDVI at high values. on parametric statistics (assuming normally distributed data),
nonparametric statistics (does not require normally distributed
(nir red ) data), and nonmetric (capable of operating on real-valued data
EVI = 2.5
(nir + 6(red ) 7.5(blue ) + 1) and nominal scaled data) (Jensen etal., 2009). Jensen (2005) and
Lu and Weng (2007) provide an extensive taxonomic overview
An updated version of the index, Enhanced Vegetation Index 2 of several image classification methods, and this section will
(EVI2), was developed later by Jiang et al. (2008) for use with broadly review some of the more common methods used for
datasets that did not have sensitivity in the blue region of the thematic information extraction.
spectrum. In general, image classification can be accomplished using
either supervised or unsupervised techniques. Within each
(nir red ) of these broad categories exist a host of different methods
EVI2 = 2.5
(nir + 2.4 ( red ) + 1) that can be applied toward specific imagery (e.g., high spa-
tial resolution, hyperspectral) and/or have been developed as
technological capabilities have advanced (e.g., artificial neural
9.7.6 Wide Dynamic Range Vegetation index network (ANN) classification and object-oriented classifica-
Gitelson (2003) proposed a simple adjustment to NDVI to compen- tion). Supervised classification categorizes every image pixel
sate for the high-end saturation. The Wide Dynamic Range Vegetation into one of several predefined land-type classes (Jensen etal.,
Index (WDRVI) applies a weighted coefficient (a) to NDVI with a 2009). The process requires several steps and includes select-
value of 0.10.2 to linearize the index relationship to VBVs. ing the land-type categories, preprocessing, defining training
data, automated pixel assignment, and accuracy assessment
(see Sabins, 1987; Jensen, 2005; Jensen et al., 2009). Various
a ( nir red )
WNDVI = supervised classification algorithms exist including the max-
a (nir + red ) imum-likelihood classifier (Strahler, 1980; Foody et al., 1992;
2016 by Taylor & Francis Group, LLC
Digital Image Processing 213
(a) (b)
(c) (d)
(e) (f )
Index value
Low High
FIg u r e 9.14 A suite of selected vegetation indices applied to an image acquired by the Airborne Imaging Spectrometer for Applications (AISA)
Eagle acquired over Grand Bay, MS. (a) Original aerial image, (b) simple ratio, (c) Normalized Difference Vegetation Index, (d) Wide Dynamic
Range Vegetation Index, (e) Enhanced Vegetation Index, and (f) three-band model. (Image courtesy of the Center for Advanced Land Management
Information Technologies (CALMIT), University of Nebraska, Lincoln, NE.)
Atkinson and Lewis, 2000), nearest-neighbor classification unsupervised classification algorithm will search for natu-
(James, 1985; Hardin, 1994), decision tree classifiers (DeFries ral groupings (or clusters) and produce a map of the number
and Chan, 2000; Russell and Norvig, 2003; Jensen 2005), of predefined clusters. These clusters can subsequently be
object-oriented classification (AtMost classification processes assigned to previously defined information classes (e.g., land
(Haralick and Shapiro, 1985; Yan etal., 2006; Chen etal., 2009) cover categories) through an iterative process (Figure 9.16).
(Figure 9.15), and ANNs (Gopal and Woodcock, 1996; Hardin, Because there is a high likelihood that there will be clusters of
2000; Jensen etal., 2000)). mixed categories, the user may have to employ other methods
Unsupervised classification methods generally partition the to parse such clusters and minimize the undefined areas in an
spectral data of an image into feature space with a minimal image. For example, mixed clusters from the original image
input from the analyst. Operating under various constraints may be masked out of the image prior to running an iteration
specified by the user (e.g., number of clusters, spectral and of the unsupervised algorithm using a different combination
spatial search radius, bands used, and iterations defined), an of inputs (e.g., the number of bands included). The user may
2016 by Taylor & Francis Group, LLC
214 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) (b)
(c) (d)
Impervious surface Golf course
Water Roof in direct sunlight Green Bare soil
Vegetation Roof oriented away from direct sunlight Fairway Cart path
Barrenland Flat rooftop and other built-up Rough Sand
Asphalt
FIg u r e 9.15 Example of object-based classification at two different image segmentation scales. The PAN-sharpened QuickBird high-spatial-
resolution (61 61cm) multispectral imagery of Las Vegas, NV, was acquired on May 18, 2003. (a) Segmentation scale 100, (b) segmentation scale
150, (c) classification at scale 100, and (d) classification at scale 150. (Adapted from Jensen, J.R. etal., Image classification, in The SAGE Handbook
of Remote Sensing, Warner, T.A., Nellis, M.D., and Foody, G.M., eds., 2009, pp. 269281.)
2016 by Taylor & Francis Group, LLC
Digital Image Processing 215
ISODATA first iteration
ISODATA initial arbitrary mean vector assignment
mean vector assignment and partition of feature space
1 Cluster 5
Distribution of Cluster 4
Band 4
Band 4
brightness values
Cluster 3
in bands 3 and 4
1
Cluster 2
Cluster 1 Ellipses
depict + 2
1 1
(a) Band 3 (b) Band 3
ISODATA second iteration ISODATA nth iteration
mean vector assignment mean vector assignment
and partition of feature space and partition of feature space
Band 4
Band 4
(c) Band 3 (d) Band 3
FIg u r e 9.16 Conceptual diagram showing how the ISODATA algorithm functions: (a) initial distribution of five hypothetical mean vectors
using 1 in both bands as the beginning and ending points, (b) in the first iteration, each candidate pixel is compared to each cluster mean and
assigned to the cluster whose mean is closest in Euclidean distance, (c) during the second iteration, a new mean is calculated for each cluster based
on the actual spectral locations of the pixels assigned to each cluster, instead of the initial arbitrary calculation. This involves analysis of several
parameters to merge and/or split clusters. After the mean vectors are selected, every pixel in the scene is assigned to one of the new clusters, (d)
this split/mergeassign process continues until there is little change in class assignment between iterations (the T threshold is reached) or the
maximum number of iterations is reached (M). (Adapted from Jensen, J.R., Introductory Digital Image Processing: A Remote Sensing Perspective,
Prentice Hall, Upper Saddle River, NJ, 2005.)
also apply supervised classification algorithms to glean infor- 9.9 Future trends
mation from the mixed clusters. Jensen (2005) describes the
two common clustering methods including the chain method Sensor systems and image processing software/hardware will
and the iterative self-organizing data analysis (ISODATA). continue to evolve and improve. Their utility has already been
The building of clusters by either of these algorithms does not proven by the vast array of applications that image data and
require any a priori knowledge. However, once the analyst its subsequent information extraction have been used in. With
works interactively with the clusters to assign them into the the problems of the twenty-first century being focused on
various land cover categories, considerable familiarity with issues such as climate change, atmospheric conditions, envi-
the study area is needed. ronmental degradation, natural hazards, population growth,
2016 by Taylor & Francis Group, LLC
216 Remotely Sensed Data Characterization, Classification, and Accuracies
urbanization, resource scarcity, and many others, it is inevi- Eckhardt. D. W., Verdin, J. P., and G. R. Lyford, 1990. Automated
table that remote sensing data will aid toward a better under- update of an irrigated lands GIS using SPOT HRV imagery.
standing of these problems. Future sensors will continue the Photogramm Eng Remote Sens, 56(11):15151522.
progression toward more comprehensive and accurate Earth Exelis, 2003. Atmospheric Correction Module: QUAC and
science measurements (Hartley, 2003). Increased spatial, FLAASH Users Guide, https://www.exelisvis.com/portals/0/
spectral, radiometric, and temporal resolutions will provide pdfs/envi/Flaash_Module.pdf (accessed June 8, 2014).
scientists new levels of detail, and new algorithms will be Foody, G. M., Campbell, N. A., Trodd, N. M., and T. F. Wood,
developed to extract the relevant information with improved 1992. Derivation and applications of probabilistic measures
accuracies. Furthermore, technological revolutions in the of class membership form the maximum-likelihood clas-
miniaturization of electronics, stabilization of optical sys- sification. Photogramm Eng Remote Sens, 58(9):13351341.
tems, efficient power sources, minimizing size and weight of Fureder, R., 2010. Topographic correction of satellite images
senor systems, etc., will lead to changes in the design of the for improved LULC classification in alpine areas. In 10th
sensors, uninterrupted data collection, very high data quality, International Symposium on High Mountain Remote Sensing
and orbital stability of systems. Cartography, Kathmandu, Nepal, pp. 187194.
The fusion of close range and in situ remote sensor data Gitelson, A., 2004. Wide Dynamic Range Vegetation Index for
with satellite/airborne imagery and other geospatially derived remote quantification of biophysical characteristics of veg-
information has already transformed analytical capabilities etation. J Plant Phys, 131:165173.
by facilitating multiscale studies of phenomena. With mas- Gitelson, A. A., Kaufman, Y. J., Stark, R., and D. Rundquist, 2002.
sive biogeophysical data volumes being generated, and the Novel algorithms for remote estimation of vegetative frac-
computational challenges for their analyses and visualization, tion. Remote Sens Environ, 80:7687.
image processing is a major player in the Big Data arena where Gitelson, A. A., Keydan, G. P., and M. N. Merzlyak, 2006. Three-
new technologies are being sought to process large quantities band model for noninvasive estimation of chlorophyll,
of data within tolerable time frames. In addition, there is an carotenoids, and anthocyanin contents in higher plant
urgent need for realistic and computationally simple surface leaves. Geophys Res Lett, 33:L11402.
radiation models for inversion of land surface variables from Gonzalez, L., 2014. Principal component analysis, http://
satellite data (Jensen et al., 2007). Liang (2003, 2007) have wiki.landscapetoolbox.org/doku.php/remote_sensing_
introduced several physically based inversion algorithms methods:principal_components_analysis (accessed June
for estimating biogeophysical variables (e.g., LAI, fractional 23, 2014).
vegetation coverage, broadband albedo, etc.), and further Gopal, S. and C. E. Woodcock, 1996. Remote sensing of forest
research continues for the derivation of a suite of algorithms change using artificial neural networks. IEEE Trans Geosci
that will enable a simple but accurate determination of land Remote Sens, 34:398404.
surface variables. Gross, M. and V. Klemas, 1988. Remote sensing of biomass of salt
marsh vegetation in France. Int J Remote Sens, 9(3):397408.
References Hall, F. G., Strebel, D. E., Nickeson, J. E., and S. J. Goetz, 1991.
Radiometric rectification: Toward a common radiometric
Aronoff, S., 2005. Remote Sensing for GIS Managers. Redlands, response among multidate, multisensor images. Remote
CA: ESRI Press. Sens Environ, 35:1127.
Birth, G. S. and G. R. McVey, 1968. Measuring color of growing turf Haralick, R. M. and L. G. Shapiro, 1985. Image segmentation
with a reflectance spectrophotometer. Agro J, 60:640649. techniques. Comp Vis Graph Image Proc, 29:100132.
Chavez, P. C., Guptill, S. C., and J. A. Bowell, 1984. Image pro- Hardin, P. J., 1994. Parametric and nearest-neighbor methods for hybrid
cessing techniques for Thematic Mapper Data. Proc Am Soc classification. Photogramm Eng Remote Sens, 60:14391448.
Photogramm, 2:728743. Hardin, P. J., 2000. Neural networks versus nonparametric
Chen, L. and L. Lee, 1992. Progressive generation of control frame- neighbor-based classifiers for semisupervised classification of
works for image registration. Photogramm Eng Remote Sens, Landsat Thematic Mapper imagery. Opt Eng, 39:18981908.
58(9):13211328. Hardisky, A., Gross, M., and V. Klemas, 1986. Remote sensing of
Civco, D. L., 1989. Topographic normalization of Landsat coastal wetlands. Bioscience, 37(7):453460.
Thematic Mapper digital imagery. Photogramm Eng Remote Hartley, J., 2003. Earth remote sensing technologies in the twenty-
Sens, 55(9):13031309. first century. In Proceedings, International Geoscience and
DeFries, R. S. and J. C. Chan, 2000. Multiple criteria for evaluat- Remote Sensing Symposium, Toulouse, France.
ing machine learning algorithms for land cover classifica- Huete, A., Didan, K., Miura, T., Rodriquez, P., Gao, X., and
tion. Remote Sens Environ, 74:503515. L.Ferreira, 2002. Overview of the radiometric and biophys-
Du, Y., Tiellet, P. M., and J. Cihlar, 2002. Radiometric normaliza- ical performance of the MODIS vegetation indices. Remote
tion of multitemporal high-resolution satellite images with Sens Environ, 83:195213.
quality control for land cover change detection. Remote ImSpec, LLC, 2002. ACORN 4.0 users guide, http://www.aigllc.
Sens Environ, 82:123. com/pdf/acorn4_ume.pdf (accessed June 8, 2014).
2016 by Taylor & Francis Group, LLC
Digital Image Processing 217
James, M., 1985. Classification Algorithms. New York: Wiley-Interscience. Russell, S. J. and P. Norwig, 2003. Artificial Intelligence: A Modern
Jensen, J. R., 2005. Introductory Digital Image Processing: A Remote Approach. Upper Saddle River, NJ: Prentice-Hall, 1080 pp.
Sensing Perspective. Upper Saddle River, NJ: Prentice Hall. Sabins, F. F., Jr., 1987. Remote Sensing: Principles and Interpretation.
Jensen, J. R., Fang, Q., and J. Minhe, 2000. Predictive modeling of New York: W. H. Freeman.
coniferous forest age using statistical and neural network Schowengerdt, R. A., 1997. Remote Sensing: Models and Methods
approaches applied to remote sensor data. Int J Remote Sens, for Image Processing. San Diego, CA: Academic Press.
20:28052822. Sheffield, C., 1985. Selecting band combinations from multispec-
Jensen, J. R., Im, J., Hardin, P., and R. R. Jensen, 2009. Image clas- tral data. Photogramm Eng Remote Sens, 51(6):681687.
sification. In The SAGE Handbook of Remote Sensing, Eds. Spanglet, H., Ustin, S., and E. Rejmankova, 1998. Spectral reflec-
Warner, T. A., Nellis, M. D., and Foody, G. M., pp. 269281. tance characteristics of California subalpine marsh plant
London, U.K.: SAGE. communities. Wetlands, 18(3):307319.
Jensen, J. R., Rutchey, K., Koch, M., and S. Narumalani, 1995. Inland Strahler, A., 1980. The use of prior probabilities in maximum like-
wetland change detection in the Everglades Water Conservation lihood classification of remotely sensed data. Remote Sens
Area 2A using a time series of normalized remotely sensed Environ, 10(2):135183.
data. Photogramm Eng Remote Sens, 61(2):199209. Tanre, D., Deroo, C., Duhaut, P., Herman, M., Morcrette,
Jiang, Z., Juete, A., Didan, R., and T. Miura, 2008. Development J.,Perbos, J., and P. Y. Deschamps, 1986. Second Simulation
of a two-band Enhanced Vegetation Index without a blue of the Satellite Signal in the Solar Spectrum (6S) Users Guide.
band. Remote Sens Environ, 112(10):38333845. Villeneuve dAscq, France: Lab dOptique Atmospherique,
Kawata, Y., Ueno, S., and T. Kusaka, 1988. Radiometric correction U.S.T. de Lille.
for atmospheric and topographic effects on Landsat MSS USGS, 2007. The power of spatial filters, http://astrogeology.usgs.
images. Int J Remote Sens, 9(4):729748. gov (accessed June 20, 2014).
Liang, S., 2003. Quantitative Remote Sensing of Land Surfaces. van der Meer, F., van der Werff, H. M. A., and S. M. de Jong, 2009.
New York: Wiley. Pre-processing of optical imagery. In The SAGE Handbook
Liang, S., 2007. Recent developments in estimating land surface of Remote Sensing, Eds. Warner, T. A., Nellis, M. D., and
biogeophysical variables from optical remote sensing. Prog Foody, G. M. London, U.K.: SAGE, pp. 229243.
Phys Geogr, 31(5):501516. Via, A., Gitelson, A., Nguy-Robertson, A., and Y. Peng, 2011.
Lu, D. and Q. Weng, 2007. A survey of image classification meth- Comparison of different vegetation indices for the remote
ods and techniques for improving classification perfor- estimation of green leaf area index of crops. Remote Sens
mance. Int J Remote Sens, 28(5):823870. Environ, 115:34683478.
Mishra, D., Cho, J., Ghosh, S., Fox, A., Downs, C., Merani, P., Yan, G., Mas, J.-F., Maathuis, B. H. P., Xiangmin, Z., and P. M. Van
Kirui, P., Jackson, N., and S. Mishra, 2012. Post-spill state Dijk, 2006. Comparison of pixel-based and object-oriented
of the marsh: Impact of the Gulf of Mexico oil spill on the image classification approachesA case study in a coal
health and productivity of Louisiana salt marshes. Remote fire area, Wuda, Inner Mongolia, China. Int J Remote Sens,
Sens Environ, 118:176185. 27(18):40394055.
Rouse, J., Haas, R., Schell, J., and D. Deering, 1974. Monitoring Zhang, H., Hu, H., Yao, X., Zheng, K., and Y. Gan, 2009.
vegetation systems in the Great Plains with ERTS. In Third Estimation of above-ground biomass using HJ-1 hyper-
Earth Research Technical Satellite Symposium, Vol. 1: Technical spectral images in Hangzhou Bay, China. In International
Presentations, NASA SP351, NASA, Washington, DC. Conference on Information, Engineering and Computer
Running, S. W., Justice, C. O., Solomonson, V., Hall, D., Barker, Science, Wuhan, China, 2009.
J., Kaufmann, Y. J., Strahler, A. H. et al., 1994. Terrestrial Zhang, M., Ustin, S., Rejmankova, E., and E. Sanderson, 1997.
remote sensing science and algorithms planned for EOS/ Monitoring Pacific coast salt marshes using remote sensing.
MODIS. Int J Remote Sens, 15(17):35873620. Ecol Appl, 7(3):10391053.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
10
Urban Image Classification:
Per-Pixel Classifiers, Subpixel
Analysis, Object-Based
Image Analysis, and
Geospatial Methods
Soe W. Myint
Arizona State University
Victor Mesev
Florida State University
Acronyms and Definitions .................................................................................................................219
Dale A. Quattrochi 10.1 Introduction .............................................................................................................................219
NASA Marshall Space
Flight Center
10.2 Remote-Sensing Methods for Urban Classification and Interpretation ........................ 220
Per-Pixel Methods Subpixel Methods Object-Based Methods Geospatial Methods
Elizabeth A. Wentz 10.3 Concluding Remarks.............................................................................................................. 228
Arizona State University References............................................................................................................................................ 228
AcronymsandDefinitions QuickBird High-resolution satellite operated by DigitalGlobe
TIR Thermal infrared portion of the electromagnetic
ASTER Advanced Spaceborne Thermal Emission and spectrum
Reflection Radiometer USGS United States Geological Survey
ETM+ Enhanced Thematic Mapper Plus V-I-S Vegetation, impervious, and soil
Gearys C A spatial autocorrelation index Vis Visible portion of the electromagnetic spectrum
GEOBIA Geographic Object-Based Image Analysis
GeoEye High-resolution satellite operated by GeoEye company 10.1 introduction
GIS Geographic information system or science
GLCM Gray-level co-occurrence matrix Remote-sensing methods used to generate base maps to analyze
ICAMS Image characterization and modeling system the urban environment rely predominantly on digital sensor data
IKONOS High-resolution satellite operated by GeoEye from spaceborne platforms. This is due in part from new sources of
LiDAR Light Detection and Ranging high-spatial-resolution data covering the globe, a variety of mul-
MESMA Multiple endmember spectral mixture analysis tispectral and multitemporal sources, sophisticated statistical and
MIR Mid-infrared portion of the electromagnetic spectrum geospatial methods, and compatibility with geographic informa-
Morans I A spatial autocorrelation index tion system or science (GIS) data sources and methods. The goal
MSS Multispectral scanner of this chapter is to review the four groups of classification meth-
NIR Near-infrared portion of the electromagnetic ods for digital sensor data from spaceborne platforms; per-pixel,
spectrum subpixel, object-based (spatial-based), and geospatial methods.
NDVI Normalized difference vegetation index Per-pixel methods are widely used methods that classify pixels into
OBIA Object-based image analysis distinct categories based solely on the spectral and ancillary infor-
OLI Operational Land Imager mation within that pixel. They are used for simple calculations
Pan Panchromatic of environmental indices (e.g., normalized difference vegetation
219
2016 by Taylor & Francis Group, LLC
220 Remotely Sensed Data Characterization, Classification, and Accuracies
(1) Per-pixel (2) Sub-pixel
A group of pixels
classifier classifier
(object vs. spatial)
(classical approaches) (spectral models)
Output Output
one class per pixel Two or more classes
per pixel (fractions)
(3) Object-based classifier
(e.g., An objectirregular shape)
(4) Geospatial classifier
(e.g., A local window -3 3)
Output = one class per pixel Output = One class per pixel
(All pixels in an object are used to identify a class (All pixels in a window is used to identify
All pixels in an object are identified as a class) a class in the center of the window)
FIg u r e 1 0.1 Overview of the four main classification groups.
index [NDVI]) to sophisticated expert systems to assign urban demonstrate the overview of the four main classification groups
land covers (Stefanov etal., 2001). Researchers recognize, however, in Figure 10.1, while Table 10.1 details the approaches with
that even with the smallest pixel size, the spectral information respect to classification requirements and procedures (e.g., reflec-
within a pixel is really a combination of multiple urban surfaces. tance conversion, steps before training sample selection, training
Subpixel classification methods therefore aim to statistically quan- samples, spatial approaches commonly used, classifiers, primary
tify the mixture of surfaces to improve overall classification accu- inputs for classification, output structures, number of output
racy (Myint, 2006a). While within-pixel variations exist, there is layers, and accuracy assessment). The chapter concludes with a
also significant evidence that groups of nearby pixels have similar brief summary of the methods reviewed and the challenges that
spectral information and therefore belong to the same classifica- remain in developing new classification methods for improving
tion category. Object-oriented methods have emerged that group the efficiency and accuracy of mapping urban areas.
pixels prior to classification based on spectral similarity and spa-
tial proximity. Classification accuracy using object-based methods 10.2 Remote-Sensing Methods for Urban
shows significant success and promise for numerous urban appli- classification and interpretation
cations (Myint etal., 2011). Like the object-oriented methods that
recognize the importance of spatial proximity, geospatial methods Urban areas comprised a heterogeneous patchwork of land covers
for urban mapping also utilize neighboring pixels in the classifica- and land uses that are juxtaposed so that classification of specific
tion process. The primary difference though is that geostatistical classes using remote-sensing data can be problematic. Derivation of
methods (e.g., spatial autocorrelation methods) are utilized during classification methods for urban landscape features has evolved in
both the pre- and postclassification steps (Myint and Mesev, 2012). tandem with increasing spatial, spectral, and temporal resolutions
Within this chapter, each of the four approaches is described of remote-sensing instruments (e.g., from 90 m Landsat multi-
in terms of scale and accuracy classifying urban land use and spectral scanner [MSS] to 30 m to the Landsat Enhanced Thematic
urban land cover and for its range of urban applications. We Mapper Plus [ETM+] and Operational Land Imager [OLI] data
2016 by Taylor & Francis Group, LLC
Urban Image Classification 221
TABLe 1 0.1 Classification Procedures and Characteristics of the Four Main Classification Groups
Per-Pixel Subpixel Object-Based Geospatial
Reflectance conversion Not required Necessary Not required Not required
Additional step before No No Segment image into No
training sample selection objects
Training samples Irregular polygons that cover Spectra of selected Segmented objects that Square windows that cover
multiple pixels representing endmembers cover multiple pixels multiple pixels representing
selected land cover classes representing selected selected land cover classes
land cover classes
Commonly used spatial GLCM No GLCM Fractal, Gearys C Morans I,
approaches Getis index Fourier
transforms, Lacunarity index,
wavelet transforms
Widely used classifiers Maximum likelihood, Linear spectral mixture, Nearest neighbor Mahalanobis distance,
Mahalanobis distance, multiple regression, minimum distance
minimum distance, regression regression tree, neural
tree, neural network Bayesian network Bayesian, MESMA
Primary input for Training samples are used to End member spectra are used All pixels in each All pixels in each window are
classification identify land cover classes to quantify fractions of object identified as used to identify one class and
classes one of the training the winner class is assigned to
sample classes the center of the local window
No. of output layer One layer Multiple layers One layer One layer
Output structure One class per pixel One fraction per pixel per One class per pixel One class per pixel
class
Accuracy assessment Randomly selected pixels for Correlation between Randomly selected Randomly selected pixels for
error matrix predicted and reference pixels for error matrix error matrix
fractions (or) object-based
accuracy assessment
Note: GLCM, Gray-level (or) spatial co-occurrence matrix; MESMA, Multiple endmember spectral mixture analysis.
and progressing to submeter spatial resolution products available relationship between the pixel and the field of view, which assumes
from commercial systems such as 0.34 m Geoeye) to achieve more reflectance, is recorded entirely and uniformly from within the
robust digital classification schemes. This evolution of classification spatial limits of individual pixels (Figure 10.1).
techniques, however, does not imply that one method is better than Regardless, the appeal of per-pixel or hard classifications
another. As with the type of satellite remote-sensing data that are remains, predominantly because they produce crisp and con-
employed for analyses, the application of a specific algorithm for venient thematic coverages that can be easily integrated with
classification of urban land cover and land use is depend upon what raster-based GIS models (Table 10.1). Composite models and
the users objectives are and what level of detail, frequency, and sen- methodologies containing information from remotely sensed
sors are required for the anticipated or resulting output products. sources are critical for revising databases and for producing
Table 10.2 shows urban remote-sensing applications with regard to comprehensive query-based urban applications. To preserve
spatial, temporal, and sensor resolutions. this relationship with GIS, the quality of per-pixel classifications
must be monitored not only using conventional determination
of accuracy based on comparisons with more reliable reference
10.2.1 Per-Pixel Methods
data but also in relation to levels of suitability or scale of appro-
Scale is indelible when conducting per-pixel classifications. The priateness. Both were evident in the USGS hierarchical scheme
spatial resolution of the sensor dictates the classification type, (Anderson et al., 1976) using the much-cited 85% as a general
range, and accuracy of urban land use and urban land cover. That guideline for the accuracy of urban features, which subsequently
is because individual urban features are rarely the same size as established a benchmark for researchers to attain and supersede
pixels, nor are they conveniently rectangular in shape. Add tem- using a variety of statistical and stochastic per-pixel techniques.
poral scale representing rapid urban activity and per-pixel clas- Some of these focused exclusively on maximizing computational
sifications become even more removed from reality. Refining the class separability, using the traditional maximum likelihood
spatial resolution and reducing the area of the pixel do not nec- algorithm (Strahler, 1980) and the more recent support vector
essarily lead to improvements in classification accuracy and may machines (Yang, 2011), while others developed methodologies
even introduce additional spectral noise, especially when pixels that imported extraneous information when aggregating spec-
are smaller than urban features. In all, the ideal situation that each trally similar pixels (Mesev, 1998), by incorporating contextual
pixel can be identified to represent conclusively one and only one relationships (Stuckens etal., 2000) or by measuring pixel inter-
land cover type has now long been abandoned. So too the perfect connectivity (Barr and Barnsley, 1997). In both, classification
2016 by Taylor & Francis Group, LLC
222 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 0.2 Urban Remote-Sensing Classifications with Regard to Spatial, Temporal, and Sensor Resolutions
Spatial Temporal Sensor
Urban Features Urban Process Resolution Resolution Resolution
Micro scale: Individual measurements Building unit (roofs: flat, pitch) Type and architecture 15 m 15years PanVisNIR
(material: tile, natural/metal, Density
synthetic)
Vegetation unit (tree, shrub) Type and health 0.255 m 15years PanNIR
Nature
Transport unit (width: road Infrastructure 0.2530 m 15years PanVisNIR
lanes, sidewalk) (material: Mobility and access
asphalt, concrete, composite)
Macro scale: Aggregation of imperviousness, Residential neighborhood Suburbanization 15km 110years VISNIRTIR
greenness, soil and water Gentrification, poverty, crime,
racial segregation, etc.
Industrial/commercial zone Land use zoning 15km 110years VISNIRTIR
Storm water flow
Heat island effect
Nonbuilt urban Environmental concerns 15km 110years VISNIRTIR
Beautification
Public space
Urban area Centrality and sprawl 5100km 110years VISNIRTIR
Flow and congestion MIRRadar
Sustainability
accuracy typically improves only marginally, simply because poverty, land use zoning, storm water flow, and heat islands, can
there is an inherent numerical limitation to the extent individual be studied simultaneously across an entire urban area as part of
pixel values can comprehensively represent the multitude of true the search for theories of livability and sustainability. In sum,
urban features within the rigid confines of their regular-sized per-pixel classifications produce simple and convenient thematic
pixel limits (Fisher, 1997). maps of urban land use and land cover that can be incorporated
However, within these numerical limits, per-pixel classifica- into GIS models. The spatial resolution of the remote sensor,
tion accuracy can be consistently high if the appropriate spatial however, limits their accuracy away from mapping individual
resolution (i.e., pixel size) is identified with respect to the suit- urban features with any level of pragmatic precision and toward
able level of urban detail (Table 10.2). Such ideas of scale appro- more traditional macroscales of generalized land cover combi-
priateness can be traced back to Welch (1982) and have since nations reminiscent of the timeless vegetation, impervious, and
been widely accepted as an important part of the class training soil (V-I-S) model (Ridd, 1995).
process. But the decision is far from trivial and must also con-
sider the appropriate scale of analysis (Mesev, 2012). Consider
10.2.2 Subpixel Methods
a continuous scale that can be conceptualized by levels of mea-
surement from remote sensor data, ranging from the representa- If locational and thematic accuracy of urban representation
tion of atomistic urban features (building, tree, sidewalk, etc.) at from remote sensing is paramount, per-pixel classifications can
the microscale to the representation of aggregate urban features be modified statistically to measure spectral mixtures represent-
(residential neighborhoods, industrial zones, or even complete ing multiple land cover classes within individual pixels. These
urban areas) at the macroscale. Microurban remote sensing are termed subpixel algorithms or soft classifications because
by per-pixel classification remains highly tenuous (even using pixels are no longer constrained to representing single classes,
meter and submeter resolutions from the latest sensors), and but instead represent various proportions of land cover classes,
any reliable interpretation is extracted directly from the spatial which are conceptually more akin to the spatial and composi-
orientation of pixelsin a similar vein to conventional inter- tional heterogeneity of urban configurations (Ji and Jensen,
pretation of aerial photography, but with lower clarity and with 1999; Small, 2004). The debate is on which approach, per-pixel
limited stereoscopic capabilities. However, the spectral hetero- or subpixel, can again be tied to the scale of urban analysis.
geneity problem is less restrictive at the macroscale of analysis For example, the measurement of impervious surfaces is par-
where classified pixels, instead of measuring individual urban ticularly amenable to subpixel classification because pixels can
objects, can be aggregated to represent a generalized view of represent a continuum of imperviousness, from total coverage
urban areas, including total imperviousness, approximate lat- (downtown areas and industrial estates) to scant dispersion
eral growth, and overall greenness. It is at this scale of analysis intermingled with biophysical land covers (city parks). Extensive
that many types of urban processes, such as sprawl, congestion, research has been devoted to more precise quantification of
2016 by Taylor & Francis Group, LLC
Urban Image Classification 223
impervious surfaces, and other urban land covers at subpixel Acquire
Select endmembers
level, such as linear mixture models (Rashed etal., 2003; Wu and multispectral image
Murray,2003), background removal spectral mixture analysis (Ji
and Jensen, 1999; Myint, 2006a), Bayesian probabilities (Foody Select multispectral bands
et al., 1992; Mesev, 2001; Eastman and Laney, 2002; Hung and
Ridd, 2002), artificial neural network (Foody and Aurora, 1996; Apply SMA model
Zhang and Foody, 2001), normalized spectral mixture analy-
sis (Wu and Yuan, 2007; Yuan and Bauer, 2007), fuzzy c-means
Evaluate endmember fractions
methods (Fisher and Pathirana, 1990; Foody, 2000), multivariate and RMS error
statistical analysis (Bauer etal., 2004; Yang and Liu, 2005; Bauer
et al., 2007), and regression trees (Yang et al., 2003a,b; Homer
No Yes Output fraction
etal., 2007). Satisfied
images
Among these, linear spectral mixture analysis, regression analy-
sis, and regression trees have had a wider appeal because they are
Accuracy assessment
theoretically and computationally simpler, as well as more preva- (correlation between predicted
lent in many commercial software packages. However, the suc- and reference fractions)
cess of measuring urban land cover types using linear techniques
is dependent on identifying spectrally pure endmembers, prefer- FIg u r e 1 0.2 Spectral mixture analysis.
ably using reference samples collected in the field (Adams etal.,
1995; Roberts etal., 1998, 2012). Although Weng and Hu (2008) representative of urban land cover, from a practical standpoint,
derived moderate accuracy levels from employing linear spec- it nonetheless perpetuates the mixed pixel problem and presents
tral mixture analysis using ASTER and Landsat ETM+ sensor thematic and semantic limitations to urban land classification
imagery, they discovered that artificial neural networks were also schemes. In other words, output from subpixel analysis pro-
capable of performing nonlinear mixing of land cover types at duces fractional classes that are more difficult to integrate with
the subpixel level (Borel and Gerstl, 1994; Ray and Murray, 1996). GIS data and may even limit their portability for comparisons
Another limitation with linear spectral mixture classifiers is that across space and through time.
they do not permit the number of endmembers to be greater than
the number of spectral bands (Myint, 2006a). In response, a mul-
10.2.3 object-Based Methods
tiple endmember spectral mixture analysis (MESMA) has been
developed to identify many more endmember types to represent With the representational limitations of purely spectrally based
the heterogeneous mixture of urban land cover types (Rashed per-pixel and subpixel classifications, it was only a matter of
etal., 2003; Powell etal., 2007; Myint and Okin, 2009). Diagrams time before the shift to the spatial domain gained momen-
demonstrating linear spectral mixture analysis and MESMA are tum. Even from a purely intuitive standpoint, finer resolution
provided in Figures 10.2 and 10.3, respectively. (i.e., smaller pixels or large cartographic scale) imagery exhib-
Two challenges dominate the research efforts to improve sub- its higher levels of detailed features that mimic the heteroge-
pixel analysis methods for urban settings. The first challenge neous nature of urban areas. This greater level of spatial detail
is pixel size. Identifying endmembers for all classes in images invariably also leads to many more uncertain spectral classes
with large to medium pixels in urban areas is difficult given known as noisewhich can be true but potentially unwanted
the heterogeneous nature of urban areas. In small spatial dis- urban features such as chimneys or manhole covers. Assuming
tances (e.g., <30 m), surfaces rapidly change from impervious, spectral noise is reduced, images with spatial resolutions rang-
to grass, to building. The smaller pixel size (e.g., 1 m or subme- ing from about 0.25 to 5 m have the potential to help identify
ter), however, is not always the optimal solution. While pixels urban structures necessary to perform many urban applica-
may not reflect a mixture of the desired endmembers (e.g., a tions, including estimation of population based on the number
combination of asphalt and grass), reflectance from unwanted of dwellings of different housing types, residential water use,
features begins to appear that needs to be filtered (e.g., oil sur- predicting energy consumption, urban heat island, outdoor
faces and automobiles in asphalt, chimneys and air conditions water use, solar energy use, and storm water pollution model-
on rooftops). The second limitation is that it is almost impos- ing (Jensen and Cowen, 1999).
sible to identify all possible endmembers in a study area and Conceptually, spatial- or object-based approaches are most
classification accuracy can be degraded by the potential pres- applicable to high-spatial-resolution remote-sensing data, where
ence of unknown classes or unidentified classes (e.g., the asphalt objects of interest are larger than the ground resolution element,
and rooftop examples mentioned earlier). This is because the or pixel. Urban objects may be vegetated features of urban land-
classifier is based on the assumption that the sum of the frac- scapes (e.g., trees, shrubs, and golf course) or anthropogenic
tional proportions of all possible endmembers in a pixel is equal features (e.g., buildings, pools, sidewalks, roads, and canals).
to one. Although this type of modeling is conceptually more With regard to mapping categorical data or identifying land
2016 by Taylor & Francis Group, LLC
224 Remotely Sensed Data Characterization, Classification, and Accuracies
(I) Select endmembers (II) SMA models with (III) Best fit model (IV) Output (V) Accuracy
and their spectra different endmembers for each pixel layers assessment
Model 1 (m1)
Multispectral
band 1 i1 s1 v1
Band 2 Model 2 (m2)
Band 3 i1 s1 v2 m317 m91 m38
Band 4
m124 m249 m19
Model
m18 m347 m6 numbers
Band n m213 m46 m396
Model 73 (m73) RMS
m97 m113 m72 error
i12 s9 v5
Collect endmember Soil
spectra
Evaluate
(i1, i2, i3, .....in) Impervious fractions
(s1, s2, s3, .....sn) and RMSE
(v1, v2, v3, .....vn)
Vegetation
i = impervious
s = soil
v = vegetation Shade
No
Satisfied
Yes
Accuracy
assessment
Model n (mn) (correlation)
in sn vn
Final
outputs
FIg u r e 1 0.3 Multiple endmember spectral mixture analysis.
use land cover classes, remotely sensed image analysis started to frequency characteristics of multispectral data and then mea-
shift from pixel-based (per-pixel) to object-based image analy- suring the proportions of V-I-S subobjects to identify residential
sis or geospatial object-based image analysis (GEOBIA) around land use in Accra, Ghana (they documented an overall accuracy
the year 2000 (Blaschke, 2010). The object-centered classifica- of 75%). In another study by Zhou etal. (2008), postclassification
tion prototype starts with the generation of segmented objects change detection based on the object-based analysis of multitem-
at multiple scales (Desclee etal., 2006; Navulur, 2007; Im etal., poral high spatial resolution produced even higher accuracies of
2008; Myint et al., 2008). To demonstrate, Walker and Briggs 92% and 94%, while Myint and Stow (2011) demonstrated the
(2007) employed an object-oriented classification procedure to effectiveness of object-based strategies based on decision rules
effectively delineate woody vegetation in an arid urban ecosys- (i.e., membership functions) and nearest neighbor classifiers on
tem using high-spatial-resolution true-color aerial photogra- high-spatial-resolution QuickBird multispectral satellite data
phy (without the near-infrared band) and achieved an overall over the city of Phoenix. These are further supported by Myint
accuracy of 81%. Hermosilla etal. (2012) developed two object- et al. (2011) who directly compared the accuracy from object-
based approaches for automatic building detection and localiza- based classifications (90%) with more traditional spectral-based
tion using high-spatial-resolution imagery and light detection classifications (68%). The land cover classes that the authors
and ranging (LiDAR) data. Stow etal. (2007) further developed identified for this particular study include buildings, other
object-based classification by taking advantage of the spatial impervious surfaces (e.g., roads and parking lots), unmanaged
2016 by Taylor & Francis Group, LLC
Urban Image Classification 225
soil, trees/shrubs, grass, swimming pools, and lakes/ponds. The at the scale that identifies the class with variable options and
study selected 500 sample points that led to approximately 70 analyze them heuristically on the display screen. Given that
points per class (7 total classes) using a stratified random sam- the nearest neighbor classifier and decision rule available in the
pling approach for the accuracy assessment of two different sub- object-based approach are nonparametric approaches, they are
sets of QuickBird over Phoenix. To be consistent and for precise independent of the assumption that data values need to be nor-
comparison purposes, they applied the same sample points gen- mally distributed. This is advantageous, because most data are
erated for the output generated by the object-based classifier as not normally distributed in many real-world situations. Another
the output produced by the traditional classification technique advantage of the object-based approach is that it allows addi-
(i.e., maximum likelihood). tional selection or modification of new objects (training sam-
In general, spectrally similar signatures such as dark/gray ples) at iterative stages, until the satisfactory result is obtained.
soil, dark/gray rooftops, dark/gray roads, swimming pools/blue However, the object-based approach has a significant problem
color rooftops, and red soil/red rooftop remain problematic even when dealing with a remotely sensed data over a fairly large area
with object-based approaches. Furthermore, the most com- since computer memory needs to be used extensively to segment
monly used object-oriented software (Definiens or eCognition) tremendous numbers of objects using multispectral bands. This
is required to perform a tremendous number of segmentations is true even for fine spatial resolution data with fewer bands (e.g.,
of objects from all spectral bands using various scale parame- QuickBird) over a small study area when requiring smaller-
ters. There is no universally accepted method to determine an scale parameters (smaller objects). Figure 10.4 shows segmented
optimal level of scale (e.g., object size) to segment objects, and a images at scale level 25, 50, and 100 using a subset of a QuickBird
single scale may not be suitable for all classes. The most feasible image over Phoenix. Figure 10.5 demonstrates how hierarchical
approach may be to select the bands for membership functions image segmentation delineates image objects at various scales.
(a) (b)
(c) (d)
FIg u r e 10.4 A subset image and segmented images at different scales. (a) Original subset, (b) level 1 (scale parameter 25), (c) level 2 (scale
parameters 50), and (d) level 3 (scale parameter 100).
2016 by Taylor & Francis Group, LLC
226 Remotely Sensed Data Characterization, Classification, and Accuracies
Object scale level 3
(biggest objects)
Object scale level 2
Object scale level 1
(smallest objects)
Original image
(pixel level)
FIg u r e 1 0.5 Image objects at each image scale level. Level 3 = 100, level 2 = 50, and level 1 = 25.
10.2.4 Geospatial Methods rough surfaces in remotely sensed images. In terms of other geo-
spatial techniques, De Jong and Burrough (1995) and Woodcock
Texture plays an important role in the human visual system for etal. (1988) implemented variograms to measurements derived
pattern recognition and interpretation. For image interpretation, from remotely sensed images to quantitatively describe urban
pattern is defined as the overall spatial form of related features, spatial patterns. Myint and Lam (2005a,b) and Myint etal. (2006)
where the repetition of certain forms is a characteristic pattern developed a number of lacunarity approaches to characterize
found in many cultural objects and some natural features. Local urban spatial features with completely different texture appear-
variability in remotely sensed data, which is part of texture or ances that may share the same fractal dimension values. Both
pattern analysis, can be characterized by computing the statis- studies report that lacunarity can be considered more effective in
tics of a group of pixels, for example, standard deviation, coef- comparison to fractal approaches for urban mapping.
ficient of variance or autocovariance, or the analysis of fractal The geospatial methods described so far may not provide sat-
similarities or autocorrelation of spatial relationships. There isfactory accuracies when they are applied to the classification
have been some attempts to improve the spectral analysis of of urban features from fine spatial resolution remotely sensed
remotely sensed data by using texture transforms in which some images. That is, mainly because most of them focus primarily
measure of variability in digital numbers is estimated within on coupling features and objects at a single scale and cannot
local windows, for example, the contrast between neighboring determine the effective representative value of particular texture
pixels (Edwards et al., 1988), standard deviation (Arai, 1993), features according to their directionality, spatial arrangements,
or local variance (Woodcock and Harward, 1992). One com- variations, edges, contrasts, and the repetitive nature of object
monly used statistical procedure for interpreting texture uses an and features. There have been a number of reports in spatial
image spatial co-occurrence matrix, which is also known as a frequency analysis of mathematical transforms, which provide
gray-level co-occurrence matrix (GLCM) (Franklin etal., 2000). solutions using multiresolution analysis. Recent developments
There are a number of texture measures, which could be applied in spatial/frequency transforms such as the Fourier transform,
to spatial co-occurrence matrices for texture analysis (Peddle Wigner distribution, discrete cosine transform, and wavelet
and Franklin, 1991). For instance, Herold etal. (2003) proposed transform have all provided sound multiresolution analytical
a method based on using landscape metrics to classify IKONOS tools (Bovik etal., 1990; Zhu and Yang, 1998).
sensor images, which in turn is compared to a GLCM. Liu etal. Of all transformation approaches, wavelets play the most
(2006) further contrasted spatial metrics, GLCM, and semivar- critical part in texture analysis. Wavelets are part of spatial- and
iograms in terms of urban land use classification. frequency-based classification approaches, and a local window
Lam etal. (1998) demonstrated how fractal dimensions yield plays an important role in measuring and characterizing spa-
quantitative insight into the spatial complexity and informa- tial arrangements of objects and features. Homogeneity, size of
tion contained in remotely sensed data. Quattrochi etal. (1997) regions, characteristic scale, directionality, and spatial period-
went further and created a software package known as the image icity are important issues that should be considered to identify
characterization and modeling system to explore how the fractal local windows when performing wavelet analysis (Myint, 2010).
dimension is related to surface texture. Fractal dimensions were From a computational perspective, the ideal window size is the
also analyzed by Emerson et al. (1999) who used the isarithm smallest size that also produces the highest accuracy (Hodgson,
method and Morans I and Gearys C spatial autocorrelation mea- 1998). The accuracy should increase with a larger local window
sures to observe the differing spatial structure of the smooth and
2016 by Taylor & Francis Group, LLC
Urban Image Classification 227
size since it contains more information than a smaller window Myint and Mesev (2012) employed a wavelet-based classification
size and therefore provides more complete coverage of spatial method to identify urban land use and land cover classes using
variation, directionality, and spatial periodicity of a particular different decision rule sets and spatial measures and demon-
texture. However, minimization of local window size is also strated the effectiveness of wavelets. However, the current wave-
important in spatial-based urban image classification tech- let-based classification system with the dyadic wavelet approach
niques since a larger window size tends to cover more urban is limited by the fact that higher-level subimages are just a quar-
land cover features and consequently creates mixed boundary ter of the preceding image. In general, smaller window size is
pixels or mixed land cover problems. However, some spatial and generally thought to yield higher accuracy in geospatial-based
frequency approaches such as wavelet dyadic decomposition image classification because if the window is too large, much
approaches require large window sizes to capture spatial infor- spatial information from two or more land cover classes could
mation at multiple scales (Myint, 2006b). The potential solution create a mixed boundary problem. Further research is required
to this problem would be to employ a multiscale overcomplete to consider an overcomplete wavelet approach that can generate
wavelet analysis using an infinite scale decomposition proce- spatial arrangements of objects and features at any scale level for
dure. This is because a large spatial coverage or a large local win- urban mapping. Such an approach could potentially be appli-
dow is not needed to describe a spatial pattern. Furthermore, cable to any land use/land cover system at any resolution or scale
this approach can measure different directional information of because it can effectively use any window size. Figure 10.6 shows
anisotropic features at unlimited scales, and it is designed to how wavelet approaches work in comparison to other geospatial
normalize and select effective features to identify urban classes. approaches in urban mapping.
Local (1 index per band used to identify a class)
window
(residential)
Level 1 Level 2 (8 indices per band used to identify a class
Local A H (e.g., entropy values of sub-images)
A H
window
(residential) V D
V D
Level 1 Level 2 Level 3
Local
window A H A H A H
(residential)
V D V D V D
(Unlimited indices per band
Level 4 Level 5 Level N (unlimited) used to identify a class)
(e.g., entropy values of sub-images)
A H A H A H
(6 normalization procedures
(4 classification decision rules)
V D V D V D
FIg u r e 10.6 An example of feature vectors or indices (32 32 window or a subset) used to identify an urban class using other geospatial
approaches, the dyadic wavelet approach, and the overcomplete wavelet approach. Note: Subimages at level two in the dyadic approach reach the
suggested minimum dimension (8 8 pixels) since any subimages smaller than eight pixels may not contain any useful spatial information. A
subimage at a higher level is exactly the same as its original size at the preceding level in the overcomplete approach. It should also be noted that
the level of scale with the overcomplete approach is unlimited. A, approximation texture; H, horizontal texture; V, vertical texture; D, diagonal tex-
ture. (a) Other geospatial approaches (multiscale or singlescale) (e.g., fractal, spatial autoC, Cooccurrence matrix, G index, lacunarity), (b) dyadic
wavelet approach, and (C) overcomplete wavelet approach.
2016 by Taylor & Francis Group, LLC
228 Remotely Sensed Data Characterization, Classification, and Accuracies
10.3 concluding Remarks Borel, C. C. and Gerstl, S. A. W. 1994. Nonlinear spectral mixing
models for vegetative and soil surfaces. Remote Sensing of
Interpreting urban land cover from the data captured by remote Environment, 47, 403416.
sensors remains a conceptual and technical challenge. Accuracy Bovik, A. C., Clark, M., and Geisler, W. S., 1990. Multichannel tex-
levels are typically lower than the interpretation of more natu- ture analysis using localized spatial filters. IEEE Transactions
rally occurring surfaces. However, huge strides have been made on Pattern Analysis and Machine Intelligence, 12, 5573.
with the formulation of statistical models that help disentan- De Jong, S. M. and Burrough, P. A. 1995. A fractal approach to the
gle the spectral and spatial complexity of urban land covers. classification of Mediterranean vegetation types in remotely
Whereas per-pixel classification has stood the test of time (pri- sensed images. Photogrammetric Engineering and Remote
marily for pragmatic reasons, especially when integrated with Sensing, 61, 10411053.
GIS-handled datasets), developments in subpixel, object-based, Descle, B., Bogaert, P., and Defourny, P. 2006. Forest change
and geospatial techniques have begun, at last, to reproduce the detection by statistical object-based method. Remote
geographical configuration and compositional texture of urban Sensing of Environment, 102, 111.
structures. These developments are further tempered by concep- Eastman, J. R. and Laney, R. M. 2002. Bayesian soft classification
tual developments that now consider the appropriateness of scale for sub-pixel analysis: A critical evaluation. Photogrammetric
(understanding the level of urban structural measurements) Engineering and Remote Sensing, 6811, 11491154.
and the appropriateness of time (understanding the lag between Edwards, G., Landry, R., and Thompson, K. P. B. 1988. Texture
urban process and urban structure). Both are critical for mea- analysis of forest regeneration sites in high-resolution SAR
suring the rate of urban change, not simply the amount of lateral imagery. Proceedings of the International Geosciences and
growth but also the juxtaposition of land use within existing Remote Sensing Symposium (IGARSS 88), ESA SP-284,
urban limits. Further research will only improve our use of pp.13551360. European Space Agency, Paris, France.
remote sensor data for measuring urban patterns and in turn Emerson, C. W., Lam, N. S. N., and Quattrochi, D. A. 1999.
will complement our understanding of key urban processes. Multi-scale fractal analysis of image texture and pattern.
Photogrammetric Engineering and Remote Sensing, 65, 5161.
References Fisher, P. 1997. The pixel: A snare and a delusion. International
Journal of Remote Sensing, 18, 679685.
Adams, J. B., Sabol, D. E., Kapos, V., Almeida-Filho, R., Roberts, Fisher, P. F. and Pathirana, S., 1990. The evaluation of fuzzy mem-
D. A., Smith, M. O., and Gillespie, A. R. 1995. Classification bership of land cover classes in the suburban zone. Remote
of multiple images based on fractions of endmembers: Sensing of Environment, 34, 121132.
Application to landcover change in the Brazilian Amazon, Foody, G. M., 2000. Estimation of sub-pixel land cover compo-
Remote Sensing of Environment, 52, 137154. sition in the presence of untrained classes. Computers and
Anderson, J. R., Hardy, E. E., Roach, J. T., and Witmer, R. E. 1976. Geosciences, 26, 469478.
A land use and land cover classification system for use with Foody, G. M. and Aurora, M. K. 1996. Incorporating mixed pixels
remote sensor data. U.S. Geological Survey Professional in the training, allocation and testing of supervised classifi-
Paper, 964. http://landcover.usgs.gov/pdf/anderson.pdf. cation. Pattern Recognition Letters, 17, 13891398.
Accessed March 27, 2015. Foody, G. M., Campbell, N. A., Trodd, N. M., and Wood, T. F.
Arai, K. 1993. A classification method with a spatial-spectral vari- 1992. Derivation and applications of probabilistic measures
ability. International Journal of Remote Sensing, 14, 699709. of class membership from the maximum-likelihood classi-
Barr, S. and Barnsley, M. A. 1997. A region-based, graph-theo- fication. Photogrammetric Engineering and Remote Sensing,
retic data model for the inference of second-order thematic 58, 13351341.
information from remotely-sensed images. International Franklin, S. E., Hall, R. J., Moskal, L. M., Maudie, A. J., and Lavigne,
Journal of Geographical Information Science, 11, 555576. M. B. 2000. Incorporating texture into classification of for-
Bauer, M. E., Heinert, N. J., Doyle, J. K., and Yuan, F. 2004. est species composition from airborne multispectral images.
Impervious surface mapping and change monitoring using International Journal of Remote Sensing, 21, 6179.
satellite remote sensing. Proceedings of the ASPRS 2004 Hermosilla, T., Ruiz, L. A., Recio, J. A., and Cambra-Lpez, M.
Annual Conference, May 2428, Denver, CO. 2012. Assessing contextual descriptive features for plot-
Bauer, M. E., Loeffelholz, B. C., and Wilson, B. 2007. Estimating based classification of urban areas. Landscape and Urban
and mapping impervious surface area by regression Planning, 106, 124137.
analysis of Landsat imagery. In Q. Wang (Ed.), Remote Herold, M., Liu, X., and Clarke, K. C. 2003. Spatial metrics and
Sensing of Impervious Surfaces, pp. 320. Boca Raton, image texture for mapping urban land use. Photogrammetric
FL: CRC Press. Engineering and Remote Sensing, 69, 9911001.
Blaschke, T. 2010. Object-based image analysis for remote sens- Hodgson, M. E. 1998. What size window for image classification?
ing. ISPRS International Journal of Photogrammetry and A cognitive perspective. Photogrammetric Engineering and
Remote Sensing, 65, 216. Remote Sensing, 64, 797807.
2016 by Taylor & Francis Group, LLC
Urban Image Classification 229
Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C., Myint, S. W. and Lam, N. S. N. 2005a. A study of lacunarity-based tex-
Herold, N., McKerrow, A., VanDriel, J. N., and Wickham,J. ture analysis approaches to improve urban image classification.
2007. Completion of the 2001 national land cover data- Computers, Environment, and Urban Systems, 29, 501523.
base for the conterminous United States. Photogrammetric Myint, S. W. and Lam, N. S. N. 2005b. Examining lacunarity
Engineering and Remote Sensing, 73, 337341. approaches in comparison with fractal and spatial autocor-
Hung, M. and Ridd, M. K. 2002. A subpixel classifier for urban relation techniques for urban mapping. Photogrammetric
land-cover mapping based on a maximum-likelihood Engineering and Remote Sensing, 71, 927937.
approach and expert system rules. Photogrammetric Myint, S. W. and Mesev, V. 2012. A comparative analysis of spa-
Engineering and Remote Sensing, 68, 11731180. tial indices and wavelet-based classification. Remote Sensing
Im, J., Jensen, J. R., and Hodgson, M. E. 2008. Object-based land Letters, 3, 141150.
cover classification using high posting density lidar data. Myint, S. W., Mesev, V., and Lam, N. S. N. 2006. Texture analysis
GIScience and Remote Sensing, 45, 209228. and classification through a modified lacunarity analysis
Jensen, J. R. and Cowen, D. C. 1999. Remote sensing of urban/ based on differential box counting method. Geographical
suburban infrastructure and socio-economic attributes. Analysis, 38, 371390.
Photogrammetric Engineering and Remote Sensing, 65, 611622. Myint, S. W. and Okin, G. S. 2009. Modelling land-cover types using
Ji, M. and Jensen, J. R. 1999. Effectiveness of subpixel analysis in multiple endmember spectral mixture analysis in a desert
detecting and quantifying urban impervious from Landsat city. International Journal of Remote Sensing, 30, 22372257.
Thematic Mapper Imagery. Geocarto International, 14, 3341. Myint, S. W. and Stow, D. 2011. An objectoriented pattern rec-
Lam, N. S. N, Quattrochi, D., Qui, H., and Zhao, W. 1998. ognition approach for urban classification. In X. Yang (Ed.),
Environmental assessment and monitoring with image Urban Remote Sensing, Monitoring, Synthesis and Modeling
characterization and modeling system using multiscale in the Urban Environment, pp. 129140. Chichester, U.K.:
remote sensing data. Applied Geographic Studies, 2, 7793. John Wiley & Sons, Ltd. DOI: 10.1002/9780470979563.
Liu. X., Clarke, K. C., and Herold, M. 2006. Population density Navulur, K. 2007. Multispectral Image Analysis Using the Object-
and image texture: A comparison study. Photogrammetric Oriented Paradigm. Boca Raton, FL: CRC Press/Taylor and
Engineering and Remote Sensing, 72, 187196. Frances Group.
Mesev, V. 1998. The use of census data in urban image classifica- Peddle, D. R. and Franklin, S. E. 1991. Image texture process-
tion. Photogrammetric Engineering and Remote Sensing, 64, ing and data integration for surface pattern discrimina-
431438. tion. Photogrammetric Engineering and Remote Sensing, 57,
Mesev, V. 2001. Modified maximum likelihood classifications of 413420.
urban land use: Spatial segmentation of prior probabilities. Powell, R. L., Roberts, D. A., Dennison, P. E., and Hess, L. L. 2007.
Geocarto International, 16, 4148. Sub-pixel mapping of urban land cover using multiple end-
Mesev, V. 2012. Multiscale and multitemporal urban remote sens- member spectral mixture analysis: Manaus, Brazil. Remote
ing. ISPRS International Archives of the Photogrammetry, Sensing of Environment, 106, 253267.
Remote Sensing & Spatial Information Sciences, XXXIX-B2, Quattrochi, D. A., Lam, N. S. N., Qiu, H., and Zhao, W. 1997.
Melbourne, Australia, pp. 1721. Image characterization and modeling system (ICAMS):
Myint, S. W. 2006a. Urban vegetation mapping using sub-pixel A geographic information system for the characterization
analysis and expert system rules: A critical approach. and modeling of multiscale remote sensing data. In D. A.
International Journal of Remote Sensing, 27, 26452665. Quattrochi and M. F. Goodchild (Eds.), Scale in Remote
Myint, S. W. 2006b. A new framework for effective urban land use Sensing and GIS, pp. 295308. Boca Raton, FL: CRC Press.
land cover classification: A wavelet approach. GIScience and Rashed, T., Weeks, J. R., Roberts, D., Rogan J., and Powell, R.
Remote Sensing, 43, 155178. 2003. Measuring the physical composition of urban mor-
Myint, S. W. 2010. Multi-resolution decomposition in relation to phology using multiple endmember spectral mixture mod-
characteristic scales and local window sizes using an opera- els. Photogrammetric Engineering and Remote Sensing, 69,
tional wavelet algorithm. International Journal of Remote 10111020.
Sensing, 31, 25512572. Ray, T. W. and Murray, B. C. 1996. Nonlinear spectral mixing in
Myint, S. W., Giri, C. P., Wang, L., Zhu, Z., and Gillette, S. 2008. desert vegetation. Remote Sensing of Environment, 55, 5964.
Identifying mangrove species and their surrounding land Ridd, M. K. 1995. Exploring a V-I-S vegetation-impervious surface-
use and land cover classes using an object oriented approach soil model for urban ecosystems analysis through remote
with a lacunarity spatial measure. GIScience and Remote sensing: Comparative anatomy for cities. International
Sensing, 45, 188208. Journal of Remote Sensing, 16, 21652186.
Myint, S. W., Gober, P., Brazel, A., Grossman-Clarke, S., and Roberts, D. A., Gardner, M., Church, R., Ustin, S., Scheer, G., and
Weng, Q. 2011. Per-pixel versus object-based classification Green, R. O. 1998. Mapping chaparral in the Santa Monica
of urban land cover extraction using high spatial resolution Mountains using multiple endmember spectral mixture
imagery. Remote Sensing of Environment, 115, 11451161. models. Remote Sensing of Environment, 65, 267279.
2016 by Taylor & Francis Group, LLC
230 Remotely Sensed Data Characterization, Classification, and Accuracies
Roberts, D. A., Quattrochi, D. A., Hulley, G. C., Hook, S. J., and Wu, C. and Murray, A. 2003. Estimating impervious surface dis-
Green, R. O. 2012. Synergies between VSWIR and TIR data tribution by spectral mixture analysis. Remote Sensing of
for the urban environment: An evaluation of the potential for Environment, 84, 493505.
the hyperspectral Infrared Imager (HyspIRI) decadal survey Wu, C. and Yuan, F. 2007. Seasonal sensitivity analysis of impervious
mission. Remote Sensing of Environment, 117, 83101. surface estimation with satellite imagery. Photogrammetric
Small, C. 2004. The Landsat ETM+ spectral mixing space. Remote Engineering and Remote Sensing, 73, 13931401.
Sensing of Environment, 93, 117. Yang, L., Huang, C., Homer, C. G., Wylie, B. K., and Coan, M.J.
Stow, D., Lopez, A., Lippitt, C., Hinton, S., and Weeks, J. 2003a. An approach for mapping large-area impervious
2007. Object-based classification of residential land use surfaces: Synergistic use of Landsat-7 ETM+ and high
within Accra, Ghana based on QuickBird satellite data. spatial resolution imagery. Canadian Journal of Remote
International Journal of Remote Sensing, 28, 51675173. Sensing, 29, 230240.
Stefanov, W. L., Ramsey M. S., and Christensen, P. R. 2001. Yang, L., Xian, G., Klaver, J. M., and Deal, B. 2003b. Urban land-
Monitoring urban land cover change: An expert system cover change detection through sub-pixel imperviousness
approach to land cover classification of semiarid to arid urban mapping using remotely sensed data. Photogrammetric
centers. Remote Sensing of Environment, 77(2), 173185. Engineering and Remote Sensing, 69, 10031010.
Strahler, A. H. 1980. The use of prior probabilities in maximum Yang, X. 2011. Parameterizing support vector machines for land
likelihood classification of remotely sensed data. Remote cover classification. Photogrammetric Engineering and
Sensing of Environment, 10, 135163. Remote Sensing, 77, 2737.
Stuckens, J., Coppin, P. R., and Bauer, M. 2000. Integrating contex- Yang, X. and Liu, Z. 2005. Use of satellite-derived landscape
tual information with per-pixel classification for improved imperviousness index to characterize urban spatial
land cover classification. Remote Sensing of Environment, growth. Computers, Environment and Urban Systems,
71, 282296. 29, 524540.
Walker, J. S. and Briggs, J. M. 2007. An object-oriented approach Yuan, F. and Bauer, M. E. 2007. Comparison of impervious
to urban forest mapping with high-resolution, true-color surface area and normalized difference vegetation index
aerial photography. Photogrammetric Engineering and as indicators of surface urban heat island effects in
Remote Sensing, 73, 577583. Landsat imagery. Remote Sensing of Environment, 106,
Welch, R. A. 1982. Spatial resolution requirements for urban 375386.
studies. International Journal of Remote Sensing, 3, 139146. Zhang, J. and Foody, G. M. 2001. Fully-fuzzy supervised classifica-
Weng, Q. and Hu, X. 2008. Medium spatial resolution satellite tion of sub-urban land cover from remotely sensed imagery:
imagery for estimating and mapping urban impervious Statistical and neural network approaches. Photogrammetric
surfaces using LSMA and ANN. Transactions on Geoscience Engineering and Remote Sensing, 22, 615628.
and Remote Sensing, 46, 23872406. Zhou, W. Q., Troy, A., and Grove, M. 2008. Object-based land
Woodcock, C. and Harward, V. J. 1992. Nested-hierarchical scene cover classification and change analysis in the Baltimore
models and image segmentation. International Journal of metropolitan area using multitemporal high resolution
Remote Sensing, 13, 31673187. remote sensing data. Sensors, 8, 16131636.
Woodcock, C. E, Strahler, A. H., and Jupp, D. L. B. 1988. The use Zhu, C. and Yang, X. 1998. Study of remote sensing image tex-
of variograms in remote sensing: I. Scene models and simu- ture analysis and classification using wavelet. International
lated images. Remote Sensing of Environment, 25, 323348. Journal of Remote Sensing, 13, 31673187.
2016 by Taylor & Francis Group, LLC
11
Image Classification Methods
in Land Cover and Land Use
Acronyms and Abbreviations ............................................................................................................231
11.1 Introduction .............................................................................................................................231
Definitions Advantages and Limitations of Remote Sensing for Map Making
11.2 Image Classification in the Context of Land-Use/Land-Cover Mapping.......................233
Historical Perspective Methods Uncertainty Assessment
Mutlu zdoan 11.3 The Future ................................................................................................................................ 241
University of WisconsinMadison References............................................................................................................................................ 241
Acronyms and Abbreviations USDA United States Department of Agriculture
NDVI Normalized Difference Vegetation Index
GIS Geographic Information System TOA Top of Atmosphere
USGS United States Geological Survey BT Brightness Temperature
LULC Land Use Land Cover WELD Web-Enabled Landsat Data
EMR Electromagnetic Radiation UMD University of Maryland
LACIE Large Area Crop Inventory Experiment
SPOT Satellite Pour lObservation de la Terre (Earth 11.1 introduction
observation satellite)
NLCD National Land Cover Database Today, information on land cover and land use is almost exclu-
MODIS Moderate Resolution Imaging sively derived from remotely sensed observations at various spa-
Spectroradiometer tial and temporal scales. The advantages of these observations
GLCC Global Land Cover Characteristics include, but not limited to, synoptic view, availability of spectral
FAO UN Food and Agricultural Organization bands that help distinguish land surface properties, archived
LCCS Land Cover Classification System temporal record, and digital nature. The principle form of
FROM-GLC Fine Resolution Observation and Modeling deriving land-cover information from remotely sensed images
Global Land Cover is classification. In the context of remote sensing, classification
NIR Near Infrared refers to the process of translating observations into land-cover
MLC Maximum Likelihood Classification categories with clearly defined biogeophysical function. For
MD Minimum Distance Classification example, a typical land-cover map may contain categories like
KMC K-Means Clustering forest, water, agriculture, and so on. These maps are then used in
AVHRR Advanced Very High Resolution Radiometer a growing number of environmental applications, from resource
ANN Artificial Neural Networks management to global change studies. To this end, the purpose
CT Classification Trees of this chapter is to review existing and emerging image clas-
SVM Support Vector Machines sification methods applied to remote sensing.
ART Adaptive Resonance Theory This chapter has the following outline. First definitions for
MLP Multi-layered Perceptron land cover and land use as used throughout the chapter are
ARTMAP Adaptive Resonance Theory Multi-layered given. Then, advantages and disadvantages of remote sensing
Perceptron with respect to land-cover mapping are discussed. In the main
LIBSVM A Library for SVM classification and Regression body of the document, existing as well as emerging methods
SVC Support Vector Classification for land-cover mapping are described. The main conclusions
SVR Support Vector Regression and the future of land-cover mapping are provided in the last
OBIA Object Based Image Analysis section.
231
2016 by Taylor & Francis Group, LLC
232 Remotely Sensed Data Characterization, Classification, and Accuracies
11.1.1 Definitions the need for detailed land surveys. Finally, the (mostly) digital
nature of data make it possible to use computers for fast pro-
Land cover is defined as describing the physical status of the cessing and combining results with other datasets within a
Earths land surface. Land use, on the other hand, describes the Geographic Information system (GIS).
use of that land cover for a particular purpose. For example, in the While there are clear advantages to use remote sensing for
case of a forested area, the forest land-cover label would be used to extracting land-cover information, remote sensing also has impor-
describe the fact that the land is occupied by a group of trees that tant limitations that need to be specified. Perhaps the biggest limi-
constitute a forest stand without any reference to the use of that for- tation of remote sensing is that it is oversold (Jensen, 2006). More
est, while the land-use label could include its usage status, for exam- specifically, remote sensing is often (incorrectly) thought of hav-
ple, industrial forest for wood production. In this context, image ing the ability to observe/identify/locate objects with excessive or
classification methods applied to remotely sensed data allow clas- unwarranted enthusiasm. Secondly, by definition, remotely sensed
sification of Earths surface into only land-cover categories, all of observations are not direct, and provide only a manifestation of
which then can be interpreted into land-use classes. For the rest of the object of interest, made possible by electromagnetic radiation.
this chapter, these definitions will be used and primarily, the land- This requires that the observations be translated into information
cover classification with remotely sensed data will be emphasized. of interest and calibrated against reality. However, this calibration
is never exact: a classification error of 10% is considered excel-
lent. Third, different land surfaces may yield very similar obser-
11.1.2 Advantages and Limitations of
vations, making them difficult to separate. On the other hand, a
Remote Sensing for Map Making
single land-cover category may be manifested in multiple different
In the last four decades, both academic-and application-oriented ways in observations, leading to incorrect classifications. Fourth,
studies have clearly established both the advantages and limi- depending on the observing platform, observations may contain
tations of remote sensing for deriving land-cover information noise associated with atmosphere and sensor characteristics. In
(Table 11.1). In terms of advantages, remote sensing offers a rela- certain cases, these unwanted information must be accounted
tively cheap and rapid method of acquiring information over a for. Finally, the spatial resolution of satellite imagery may be too
large geographical area. For example, a single Landsat scene cov- coarse for detailed mapping and for distinguishing small contrast-
ers over 30,000 km2 of land surface, producing data at several ing areas. Rule of thumb: a land use must occupy at least 16 pix-
spectral bands at fairly high spatial resolutions. When subjected els (picture elements, cells) to be reliably identified by automatic
to classification, these data could reveal land-cover information methods. However, new satellites are being proposed with 1 m
at the scale of a standard topo sheet (~1:25,000 scale). Even with resolution, these will have high data volume but will be suitable for
the cost of ground verification, this is very economical. Second, land-cover mapping at a detailed scale.
remote sensing allows unobtrusive means to acquire observa- Despite these disadvantages, remote sensing continues to
tions, elucidating the Earth surface discretely by means of elec- be the primary data source for deriving land-cover and land-
tromagnetic radiation. Third, remotely sensed observations use information worldwide, from local to global studies. The
offer a synoptic view, which can be described as observations land-cover and land-use maps generated from remotely sensed
that give a broad view of the Earths surface at a particular time. observations also span a wide varieties of themes from crop type
As a result, regional phenomena, which are invisible from the identification to forest composition mapping to wetland detection,
ground, are clearly visible, such as geological structures or for- just to name a few. As our population grows and more pressure is
ests instead of the trees. Fourth, remote sensing provides one of placed on the natural resources, remote sensing will continue to
few means to obtain data from inaccessible areas. Fifth, presence play a significant role to monitor these resources. Both the exist-
of archival observations allows rapid assessment of land-cover ing tools and those to be developed in the near future will allow
change, or cheap ways to update/construct base maps without us to achieve this goal more accurately and efficiently.
TABLe 1 1.1 Advantages and Limitations of Remote Sensing for Land-Cover Studies
Advantages Disadvantages
Low cost and rapid Overemphasis on what it can
Up-to-date Inflated accuracies
Large area coverage Not a direct sample need to be calibrated
Synoptic view Large measurement uncertainty
Regular revisit period Unintended measurements (e.g., clouds)
Digital nature of data for integration in GIS Lack of observations in certain regions
Data acquisition over inaccessible areas Spatial/temporal resolution mismatch
Unobtrusive data collection Difficult to interpret in some cases
Spectral information in the nonvisible Need specific training to process/interpret
Derive a map as the final result
2016 by Taylor & Francis Group, LLC
Image Classification Methods in Land Cover and Land Use 233
11.2 image classification in the context (USGS) released its entire satellite data archive at no cost, mak-
ing it possible to study states land-use/land-cover (LULC) on a
of Land-Use/Land-cover Mapping repeated and low-cost basis. At the same time, new computer-based
11.2.1 Historical Perspective machine learning algorithms have found their way into regional
remote sensing investigations that help to provide improved accu-
Many environmental and natural resource management ques- racies for map products. On the application side, the environment
tions require accurate and timely information on land cover and and the natural resources of our state are under increased pressure
land use. To this end, many states and agencies today systemati- from population growth, pollution, and bioenergy prospects.
cally collect, update, and disseminate this most fundamental form
of land information. However, it is accurate to suggest that remote
11.2.2 Methods
sensing has had a much longer history of data collection than we
have been making maps using these data. While history of remote Numerous classification algorithms have been developed since the
sensing is reserved for elsewhere (e.g., Chapter 1 of Remotely early 1970s when first dedicated land observations became avail-
Sensed Data Characterization, Classification, and Accuracies), able. These algorithms range from visual interpretation of printed
we provide a short history of land-cover mapping using remote or digital color images to advanced machine learning algorithms
sensing as it sets the stage for the following discussion. that emulate human learning behavior. In this section, we divide
In many ways, the scientific establishment of modern-era land- the individual methods into two main subcategories: those that rely
cover and land-use mapping with remote sensing can be traced on manual interpretation and those rely on automation. Note that
back to early 1970s. In anticipation of civilian space-based remote the traditional division of classification methods into supervised
sensing observations, a number of federal agencies in the United and unsupervised logic is not used here as that topic is extensively
States formed an interagency Steering Committee on Land Use treated in many textbooks and tutorials. The decision to use manual
information and Classification early in 1971. The formation and versus automated categories stems the very purpose of this chap-
subsequent work of this committee resulted in first standards for ter: to provide an in-depth review of image classification methods.
mapping land cover (Anderson, 1971). The land-use classification With this in mind, the supervised versus the unsupervised logic is
system to be used with remote sensor data, initially proposed considered as part of the described classification methods.
by James R. Anderson and published in final form in 1976, was
designed to place major reliance on remote sensing and used a 11.2.2.1 Approaches that Rely on
system of hierarchically defined categories (Anderson et al., Manual interpretation
1976). These developments along with the launch of Landsat I, II, Overwhelming majority of land-cover classification performed
and III are considered to have built the foundation of modern- on satellite data today are based on some form of automation.
day land-cover mapping with remote sensing. However, manual interpretation of satellite data still plays an
Several developments occurred from the 1970s to present important role in image categorization, especially in large state
(Table 11.2). On the technical side, the U.S. Geological Survey and international organizations such as the United Nations Food
TABLe 1 1.2 Historical Developments in Land-Cover Land-Use Mapping from Remote Sensing
19201930 Development and early applications of aerial photography
19351945 WWIIExtensive applications of nonvisible portions of the EMR
19601970 First use of the term remote sensing
19701980 Rapid advances in image processing
1971 Interagency Steering Committee on Land Use Information and Classification in the United States
1974 The Large Area Crop Inventory Experiment (LACIE)
1975 Development of the neural network backpropagation algorithm by Paul Werbos
1976 Publication of James R. Andersons landmark land use classification system
19701980 Launch of early Landsat satellites
1983 Development of the Global vegetation and land-use database by Elaine Matthews
1984 Launch of Landsat 4
1986 Launch of SPOT
1992 Development of the first National Land Cover Dataset (NLCD)
1993 Development of the original SVM algorithm by Vladmir N. Vapnik
1997 Release of the Global Land Cover Characteristics (GLCC) database
1999 Launch of Landsat 7 and MODIS
2000 Development of the FAO Land Cover Classification System (LCCS)
2000 Deep Learning in the neural networks community gets traction
2013 Launch of Landsat 8
2013 Release of the Finer Resolution Observation and Monitoring Global Land Cover (FROM-GLC) database by Chinese researchers
2016 by Taylor & Francis Group, LLC
234 Remotely Sensed Data Characterization, Classification, and Accuracies
and Agricultural Organization (FAO). The principal advantage of Forexample, in an agricultural landscape, identification of irri-
manual (or visual) interpretation of satellite data for the purpose gated versus nonirrigated fields may be possible by recognizing
of land-cover mapping is the accuracy. Humans are extremely shapes associated with center pivot irrigation: identification of crop
good at pattern recognition and categorization, even in very types grown on that field is less straightforward. On the other hand,
complex landscapes with multiple, perhaps overlapping catego- manual interpretation of image data appears to be more cost effec-
ries. However, the visual interpretation of satellite images can be tive if using only sampling units and enumeration as the point of
a complex process, going beyond what is contained in an image. area estimates. However, these applications do not provide wall-to-
In fact, successful image interpretation is only possible through wall maps that are often required for environmental analyses.
the iterative process of recognition and interpretation of objects
(Figure 11.1), both of which rely heavily on the knowledge and the 11.2.2.2 Approaches that Rely on
expertise of the analyst in charge of the analysis (Albertz, 2007). Automated classification
In a remotely sensed image, the objects are defined in terms Image classification for the purpose of extracting land-cover infor-
of the way they reflect or emit radiation that give rise to their mation can also be achieved in an automated way with the help
recorded colors and shapes. Thus, visual elements of tone, shape, of computers. While there are many interpretations of the word
size, pattern, texture, and association are the available tools that automatic in the context of classification, what is being referred
aid recognition and interpretation of the objects in an image. to here is the idea that only a sample of the landscape need to be
Perhaps what separates todays manual interpretation of sat- known and not the entire population. In other words, unlike man-
ellite data for mapping land cover is the medium in which the ual interpretation in which the knowledge of the whole landscape
interpretation is made. Early work with satellite imagery relied on is needed, automated classification only requires information on a
visual interpretation of large hard-copy satellite image prints to subsample of the landscape. In this context, the idea of automated
identify and map land-cover elements. Todays visual image inter- classification applies to both unsupervised and supervised clas-
pretation occurs all in digital format, using the idea of digitizing sification logic. In the unsupervised approach, the relationship
boundaries of land-cover elements directly on a computer using between the observed pattern (the feature space) and a predeter-
within a GIS system. Regardless of the analysis environment, sev- mined number of statistical clusters (the information classes) is
eral known factors significantly influence the outcomes. First, the established with the help of an algorithm. It is often applied in sit-
red and the near infrared (NIR) portions of the electromagnetic uations where prior knowledge of the ground cover is not readily
spectrum provide tremendous information for recognizing land- available. Ultimately, the statistical clusters are assigned to land-
cover elements. Second, the use of multiple images of the same cover labels and it is here where idea of automation comes into
area increases classification accuracy, because the phenology of play. That is, the analyst only uses a sample of known associations
the land cover is captured. Third, studies involving visual inter- between the information clusters and actual land-cover labels.
pretation of satellite data benefit from radiometric enhancements In the case of supervised classification, on the other hand, the
and manipulations in the form of spectral indices. For example, user collects a set of learning samples to train a classifier to iden-
compared to automated techniques of image classification, stud- tify the class label of every pixel in the image. Once again this is
ies involving visual interpretation of satellite data report superior what is being referred to automation: only a fraction of the popu-
performance, particularly for area estimation problems. lation is needed to estimate the characteristics of every individ-
On the down side, manual interpretation of satellite data is ual in the population: the pixels that constitute the landscape.
expensive, involving thousands of analyst hours. This is one rea- Another way of looking at this is that there is a large return on
son, why large organizations such as various state mapping agencies investment (training data or known samples): knowledge on less
choose manual interpretation as a method of national land-cover than 5% of the landscape is used to extract information on the
inventories in many developing countries (e.g., Travaglia et al., entire landscape in question. Note that the accuracy of unsuper-
2001). Moreover, land-cover maps derived from visual interpreta- vised approaches is generally lower than that of supervised meth-
tion of satellite data may not contain finest levels of categorical detail ods, especially in complex landscapes where spectral/temporal
due to inherent limitations of analysts knowledge and experience. manifestation of different land-cover categories may be similar.
Previous
knowledge and
experience
Aerial or satellite Recognition of Result of the
Interpretation
images objects interpretation
FIg u r e 11.1 Schematic presentation of the interpretation process. (Modified from Albertz, D. 2007. Einfhrung in die Fernerkundung.
Grundlagen der Interpretation von Luft- & Satellitenbildern. (Darmstadt) 254pp.)
2016 by Taylor & Francis Group, LLC
Image Classification Methods in Land Cover and Land Use 235
The last two decades have seen a growing number of automated In the MLC procedure, the distribution of each class learn-
and advanced image classification methods, both supervised and ing sample (the training data) is assumed to come from a normal
unsupervised, applied to land-cover mapping across various spatial distribution. Using the mean vector and the covariance matrix to
and temporal scales (Table 11.3). These methods often belong to the characterize a class, the probability that a given pixel belongs to
nonparametric statistics domain, defined as a set of statistical tools a specific class is calculated (Richards, 2012). Each pixel is then
that are not based on parameterized families of probability distri- assigned to the class that has the highest probability (hence the
butions, and hence make no assumptions about the probability dis- maximum likelihood). A prototype land-cover database for the
tributions of the variables being evaluated. The principal advantage conterminous United States was created by Loveland etal. (1991) by
of the newer methods is their ability to deal with between-class vari- first stratifying vegetated and barren land, then using an unsuper-
ability that has limited more traditional pattern recognition meth- vised classification of multitemporal greenness data derived from
ods. To this end, the automated approaches are further categorized advanced very-high resolution radiometer (AVHRR) imager, and
into those that rely on parametric and nonparametric statistics. postclassification stratification of classes into homogeneous land-
cover regions using ancillary data. Jia and Richards (1994) intro-
11.2.2.2.1 ParametricTools duced a simplified MLC technique for handling remotely sensed
The image classification tools that are parametric in nature assume image data that reduces the processing time and copes with the
that the observation matrix comes from a known probability dis- training of geographically small classes. Hall and Knapp (2000)
tribution and make inferences about the parameters of the distri- demonstrated the use of the MLC procedure applied to Landsat
bution. Many of the traditional classification algorithms, including data for characterizing the successional and disturbance dynam-
the maximum likelihood classification (MLC), minimum dis- ics of the boreal forest for use in carbon modeling in Canada.
tance (MD), and, to some degree, K-means clustering (KMC), Vogelmann etal. (2001) demonstrated the use of statistical tools to
procedure fall into this category of statistical classifiers. Some of map land cover of the continental United States with Landsat data.
the advantages of parametric algorithms include their simplicity These earlier studies showed that with smaller number of catego-
and assumed higher statistical power. From the practical perspec- ries, achieving accuracies better than 80% was possible with the
tive, all off-the-shelf image-processing softwares include these use of parametric classifiers. However, as the number of categories
algorithms, thus making them readily available and easy to apply and their complexity increased, these methods did not produce
under a wide range of conditions. They also work well, in some image classification results with better than 75%, particularly for
cases even better than the new generation of classification tech- the more refined categories. This notion still holds true today.
niques as long as the assumptions about the probability distribu- In most applications of the MLC, each class is given the
tions of data are valid. However, in many cases, the input satellite same equal likelihood of belonging. However, MLC algorithm
data do not conform to these assumptions, rendering parametric allows incorporation of prior probabilities to increase/decrease
classifiers less robust in their predictions. An additional limitation the likelihood of occurrence for certain classes. For example,
occurs when nontraditional inputs, such as topography or other Strahler (1980) in an early seminal paper shows how prob-
ancillary information, are part of the feature space as these vari- abilities of occurrence of classes based on separate, indepen-
ables never conform to known probability distributions. dent knowledge such as collateral information datasets (e.g.,
TABLe 1 1.3 Comparison of Automated Image Classification Algorithms Described Here
Advantages Limitations Typical Accuracies References
Parametric Long history Normality assumption 85%95% for small Akaike (1973)
tools Widespread use Lack of ancillary input use class (<5) problems Loveland etal. (1991)
Simple formulation Limited class separation 75585% for large Jia and Richards (1994)
Fast computation class problems Hall and Knapp (2000)
Small set of parameters Vogelmann etal. (2001)
Mondal etal. (2012)
Sun etal. (2013)
Nonparametric Nondistributional requirement Lack of standard software 85%98% for small Benediktsson etal. (1990)
tools Improved accuracies Can be a black box class (<5) problems Yoshida and Omatu (1994)
Probability estimates Can be computationally expensive 75%90% for large Gopal and Woodcock (1996)
Ability to ignore unimportant Some optimized for two class class (>5) problems Friedl and Brodley (1997)
features problems Carpenter etal. (1997)
Ability to estimate feature Large parameter space Pal and Mather (2003)
importance Specialized knowledge may be needed Rodriguez-Galiano etal. (2012)
Many-to-one feature
Object-based Minimum mapping unit Lack of standard software 85%95% for small Woodcock and Harward (1992)
tools Natural class boundaries Large parameter space class (<5) problems Blaschke and Hay (2001)
Improved accuracy Computationally expensive 75%90% for large Hay etal. (2001)
class (>5) problems Benz etal. (2004)
2016 by Taylor & Francis Group, LLC
236 Remotely Sensed Data Characterization, Classification, and Accuracies
rock type, soil type, topography) can be used in a Bayesian- The KMC algorithm, among many, attempts to find a predeter-
type classifier (i.e., MLC) to improve classification accuracies. mined number of natural groupings or clusters in the data. It is
Strahler (1980) also recognized that using prior probabili- an iterative approach in which the mean of each information class
ties with supervised classification algorithms could also bias (the cluster) is initialized in some manner (e.g., randomly) and
the posterior probability of a given class especially if there is each pixel in the image matrix is assigned to the cluster that is clos-
uncertainty surrounding the information used to prescribe the est to in the feature space. Then new cluster means are calculated
prior probabilities. Chen et al. (1999) addressed this issue by based on the current assignment, and an attempt is made to reas-
introducing a subjective confidence parameter (c) that condi- sign each pixel to its new category based on some similarity index.
tions the likelihood estimate for the membership of each pixel This procedure is repeated until convergence, which occurs when
in a given class. Ranging from 0.0 to 1.0, the condition param- there is little or no change in assigned cluster means. The KMC
eter controls the influence of the ancillary information on the method falls into the unsupervised classification domain and is
predictions. Although the choice of a value for c is subjective, considered to be semiparametric that lies between fully paramet-
McIver and Friedl (2002) show that a simple heuristic based on ric and fully nonparametric statistical domains (Alpaydin, 2004).
objective criteria can be used to prescribe a value for c in many
situations (Figure 11.2). 11.2.2.2.2 NonparametricTools
The MD procedure is one of the simplest forms of paramet- The nonparametric methods (also called distribution-free meth-
ric classification algorithms in which the unknown image pixels ods) do not require the variables in the image matrix to belong to
are categorized by minimizing the distance between the obser- any particular distribution. While there is cornucopia of nonpara-
vation matrix of the pixel and the class in multifeature space metric classification techniques, the remote sensing community
(Wacker and Landgrebe, 1972). Variants of the MD procedure has adapted the artificial neural networks (ANNs), classification
use different methods to calculate the distance, which is defined trees (CTs), and support vector machines (SVMs), both from the
as an index of similarity. For example, the Euclidian distance algorithmic and practical (i.e., availability as an image processing
measure is used in cases where the variances of the classes are software) perspectives (Hansen et al., 1996; Friedl and Brodley,
very different, while the Mahalanobis distance works better in 1997; Lawrence and Wright, 2001; Pal and Mather, 2003; Rogan
cases where there is correlation between the variables of the fea- etal., 2003; Krishnaswamy etal., 2004). These particular meth-
ture space (Richards, 2012). ods, sometimes referred to as machine learning classifiers, have
DT output Sample bias adjustment Spatial priors, unadjusted
Spatial priors + sample adjustment Final
Spatial priors, adjusted
FIg u r e 11.2 Classification results for MODIS tile v05h10 (south central United States) outcomes at each stage of processing. (Adapted from
Friedl, M.A., et al., Remote Sensing of Environment, 114, 168, 2010. With permission.)
2016 by Taylor & Francis Group, LLC
Image Classification Methods in Land Cover and Land Use 237
been used effectively in a variety of land-cover mapping studies A CT classifier takes a different approach to land-cover clas-
(e.g., Huang and Jensen, 1997; DeFries and Chan, 2000; Friedl sification. It breaks an often very complex classification prob-
etal., 2010). With some exceptions, these classifiers have proven lem into multiple stages of simpler decision-making processes
superior to parametric classifiers (e.g., maximum likelihood), (Breiman etal., 1984; Safavian and Landgrebe, 1991). Depending
with accuracy improving on the order of 10%20% (Rogan etal., on the number of variables used at each stage, there are univari-
2002). Their success can be attributed to (1) ability to incorporate ate and multivariate decision trees (Friedl and Brodley, 1997).
class-relevant categorical and continuous observations into the Univariate decision trees have been used to develop land-cover
features space; (2)not being constrained by parametric statistical classifications at a global scale (Quinlan, 1993; DeFries et al.
assumptions so that multimodal, noisy, or missing observations 1998; Hansen et al., 2000). Though multivariate decision trees
are effectively handled; (3) ability to tease apart complex feature are often more compact and can be more accurate than univari-
spaces; and (4) ability to perform many-to-one classifications ate decision trees (Brodley and Utgoff, 1995), they involve more
where multiple manifestations of the same category are present in complex algorithms and, as a result, are affected by a suite of
the observation matrix. algorithm-related factors (Friedl and Brodley, 1997). An addi-
The ANN technique in remote sensing was introduced over tional benefit of the CT algorithms is their ease of use and compu-
two decades ago by Benediktsson etal. (1990). Since then, a large tational efficiency (Lawrence and Wright, 2001; Pal and Mather,
number of studies have demonstrated their effectiveness in remote 2003; Lawrence et al., 2004). Several studies have found CTs
sensing image classification. Their potential discriminating power to be an acceptable classification method (Hansen etal., 1996;
has attracted a great deal of research effort, so many types of neural Lawrence and Wright, 2001; Rogan etal., 2003; Krishnaswamy
networks have been developed (Lippman, 1987). Among the more etal., 2004) and have shown improvements in accuracy over tra-
popular implementations of ANN are the backpropagation train- ditional parametric classifiers (Friedl and Brodley, 1997; Pal and
ing algorithm and multilayer perceptron (MLP) approach (e.g., Mather, 2003).
Paola and Schowengerdt, 1995; Atkinson and Tatnall, 1997) and SVMs are a supervised nonparametric statistical learning
the adaptive resonance theory (ART) approach (Carpenter etal., technique that is increasingly being used by the remote sensing
1991). MLP networks are architectures in which each node receives community (Huang etal., 2002; Mantero etal., 2005; Mountrakis
inputs from previous layers and information flows in one direction etal., 2011). At the heart of an SVM training algorithm lies the
to the output layer (Pratola etal., 2011). The number of nodes in the concept of a linear hyperplanean optimal boundary found
intermediate layer(s) defines both the complexity and the power through an iterative learning procedure that separates the
of a neural network model to describe underlying relationships training set into a discrete predefined number of classes while
and structures inherent in a training dataset (Kavzoglu, 2009). minimizing misclassifications errors (Vapnik, 1979; Zhu and
Although ANN classification has been shown to greatly improve Blumberg, 2002). Several approaches have been developed to
accuracy over traditional parametric methods with reduced train- improve SVM predictive accuracies using multispectral remote
ing sets, the process to implement classification is not straightfor- sensing data. These include the soft margin approach (Cortes
ward and can be time consuming (Pal and Mather, 2003). and Vapnik, 1995) and kernel-based learning (Scholkopf and
The ART framework describes a number of neural network Smola, 2001) that lead to SVM optimization, although the ker-
models, which use supervised and unsupervised learning meth- nel functions often result in more expensive parameterization
ods for pattern recognition and prediction (Carpenter et al. (Kavzoglu and Colkesen, 2009).
1992). The premise of the ART model is that object identification Prior research has identified at least three benefits of SVMs
and recognition generally occur as a result of the interaction of that make them particularly suitable for remote sensing appli-
top-down observer expectations with bottom-up sensory infor- cations. First, regardless of the size of the learning sample, not
mation, in this case information content of satellite and ancil- all the available examples are used in the specification of the
lary data. Assuming that the difference between sensation and hyperplane. This allows SVMs to successfully handle small
expectation stays below a set threshold (via the vigilance param- training datasets, because only a subset of pointsthe support
eter), a pixel will be considered a member of the expected class, vectorsthat lie on the margin is used to define the hyperplane
thus offering a solution to the problem of plasticity that is, the (Mantero et al., 2005). Second, unlike many statistical classi-
problem of acquiring new knowledge without disrupting exist- fiers, SVMs do not make prior assumptions on the probability
ing one (Carpenter and Grossberg, 2003). Carpenter etal. (1997) distribution of the data, which leads to reduction in classification
applied the Fuzzy ARTMAP, a version of the ART framework, errors when input data do not conform to a required distribution
to Landsat data and terrain features for vegetation classification (e.g., Gaussian). Third, SVM-based classification algorithms have
in a challenging environment and reported a fast, reliable, and been shown to produce generalizable models from a set of input
scalable algorithm overcoming many limitations of backpropa- training data, eliminating the notion of overfitting (Montgomery
gation ANNS, K nearest neighbor algorithms, and MLC. They and Peck, 1992).
also report an additional benefit of the ARTMAP method, in One of the more popular implementations of SVMs is software
which a voting strategy improves prediction and assigns confi- called the LIBSVM implementation that provides linear, poly-
dence estimates by training the system several times on different nomial (cubic), and radial-basis kernels (Chang and Lin, 2011).
orderings of an input set. This implementation includes C-support vector classification
2016 by Taylor & Francis Group, LLC
238 Remotely Sensed Data Characterization, Classification, and Accuracies
(C-SVC), -support vector classification (-SVC), distribu- the performance of several nonparametric classifiers. Shao and
tion estimation (one-class SVM), -support vector regression Lunetta (2012) compared the performance of SVMs, ANNs, and
(-SVR), and -support vector regression (-SVR) formulations. CT for land-cover characterization using MODIS time-series
All SVM formulations supported in LIBSVM are quadratic min- data and investigated the effects of training sample size, sample
imization problems. Using the radial-basis kernel classification variability, and landscape homogeneity (purity). Their results
option, the LIBSVM required only two parameters to be defined: indicate a strong relationship between training sample size and
the kernel parameter and the cost parameter C (Chang and classification accuracy but above a set threshold, increasing the
Lin, 2011). Both of these parameters are data dependent and are number of training data does not lead to an equal increase in per-
identified separately for each footprint/date-pair combination formance. They also show that SVMs had superior generalization
using the grid search option over log-transformed hyperparam- capability, particularly with respect to small training sample sizes
eters as suggested by Hsu etal. (2001). (Figure11.3). There was also less variability of SVM performance
Note that SVMs have been shown to perform well given a cer- when classification trials were repeated using different training
tain level of noise (i.e., mislabeled training data), but they are sets.
not completely impervious to outliers (Vapnik, 1995). While a A growing number of studies continue to show that the non-
number of methods have been developed to mitigate the effects parametric image classification tools come with undisputed
of outliers on SVMs (Lin and Wang, 2002; Suykens etal., 2002; advantages, not least of the improved classification performance.
Tsujinishi and Abe, 2003; Lin and Wang 2004) they show only While the more sophisticated nonparametric algorithms still
incremental improvements over standard SVM methods. remain in the research domain and require specialized software
While there is evidence in the literature to show that nonpara- environment, the rate at which the remote sensing community
metric methods perform better than traditional classifiers for takes up these methods is high and points to the possibility that
mapping land cover, very little attempt has gone into comparing they may become the de facto choice for image classification.
Urban Urban
Natural land N Natural land N
Ag 0 20 40 80 120 160 Ag 0 20 40 80 120 160
Water km Water km
(a) (b)
Urban N Urban
N
Natural land Natural land
Ag 0 20 40 80 120 160 Ag 0 20 40 80 120 160
Water km Water km
(c) (d)
FIg ur e 11.3 Comparison of classification results for SVM (a), ANN (b), and CT (c) algorithms. The NLCD 2001 (d) is also included as reference.
(Adapted from Shao and Lunetta, 2012. With permission.)
2016 by Taylor & Francis Group, LLC
Image Classification Methods in Land Cover and Land Use 239
11.2.2.3 Pixel vs. object Based classifications the size of the objects under consideration on the landscape.
Individual pixels form the smallest unit of analysis when classi- From the theoretical perspective, the ShannonNyquist sam-
fying remotely sensed images for the purpose of land-cover map- pling theorem suggests that to be able reconstruct the original
ping. However, it is well known that as arbitrary objects, pixels do image (i.e., to reconstruct the landscape objects in the image),
not necessarily represent the landscape that is being characterized. the spatial sampling rate (or the spatial resolution) of the image
Contextual classification methods overcome this issue by incor- has to be higher than twice the highest spatial frequency of the
porating the spatial information into the classification process. In original image (i.e., spatial resolution must be finer than at least
the absence of contextualization, pixel-based classification results half the size of the smallest objects on the landscape) (Blaschke,
tend to contain significant noise as no neighborhood information 2010). From the practical perspective however, a minimum of
is being considered (Fisher, 1997). One reason for this is that infor- six pixels per objects has been found to be necessary in order to
mation content of individual pixels is heavily influenced by the accurately identify and map objects.
radiance contributed from surrounding pixels, or the Modulation Returning to different kinds of OBIA, in the first category,
Transfer Function (MTF) effect (Townshend etal., 2000). Moreover, pixel-based image classification is performed independent of the
both the landscape structure and the resolving power of the sensor image segmentation process. More specifically, first a traditional
contribute to our ability to identify and map objects of interest in a pixel-based classification is performed using the best possible
map. To overcome the errors that result from traditional per-pixel tools and inputs with the highest possible accuracy. Then image
classifiers that ignore contextual properties from surrounding segmentation is performed, either on the original image that
pixels, the remote sensing community has developed a number of went into the classification process or on an image with higher
image analysis algorithms that go beyond pixels and take also into spatial resolution of the same area. The key here is to define the
account spatial information (Blaschke, 2010). These new methods minimum mapping unit, defined as the size of the smallest object
include object-based image analyses (OBIA) that have generally to be identified on the landscape, based on the spatial resolution
yielded higher thematic accuracies than the traditional per-pixel of the image used in classification. Finally, the image segments,
methods (Woodcock and Harward, 1992; Blaschke et al., 2006; either in raster or vector format, are merged with the pixel-based
Platt and Rapoza, 2008; Lizarazo and Elsner, 2011). classification using the majority rule, in which the most fre-
In this chapter, we divide OBIA into three categories, primarily quently occurring class label is used to label the entire segment
distinguished by the order and the purpose by which the objects (Figure 11.4). While the majority rule works well, there are other
are utilized. A common denominator of these methods is image ways to collapse individual pixels into objects (polygons).
segmentation, purpose of which is to produce a set of nonoverlap-
11.2.2.4 emerging Methods
ping segments (objects, polygons). Although not new in indus-
trial and medical image processing (Haralick and Shapiro, 1985), 11.2.2.4.1 CloudandCloudShadowMasking
image segmentation has only two decades of history in geospatial The presence of clouds and cloud shadows affects almost all
applications (Ryherd and Woodcock, 1996; Blaschke and Hay, forms and types of analyses conducted with satellite data, and
2001; Hay etal., 2005; Esch etal., 2008; Lizaraso and Elsner, 2009) their detection and subsequent removal is often the initial step
but has been rapidly increasing in the past 10years. in most image-processing chains (Simpson and Stitt, 1998; Irish,
Image segmentation partitions the image into a set of distinct 2000; Arvidson et al., 2001). For example, they selectively alter
and uniform (homogenous) regions, in which the criteria of homo- the reflectance and transmission of radiation through the Earths
geneity are defined by one or more dimensions of the feature space atmosphere and reduce the accuracy of atmospheric correction of
(Blaschke, 2010). It is generally understood that these segments images. Land-cover classification involving multiple image com-
represent meaningful landscape objects (e.g., a forest stand or an posites requires clouds and their shadows to be removed from the
agricultural field), but this translation is not always clear at the seg- scene, before the compositing process. In change detection studies,
mentation stage. What is clear however is that different regions are the presence of clouds and cloud shadows lead to false detection
found at different scales of analysis, placing image segmentation in of land-cover change. Calculation of vegetation indices like the
the realm of multiscale landscape analysis (Hay etal., 2001). The normalized difference vegetation index (NDVI) may be biased in
image segments also contain additional spectral (e.g., mean values locations where clouds and cloud shadows are present. Archiving
per band, and also median values, minimum and maximum values, processes such as scene selection and scene quality assessment too
mean ratios, variance, etc.) and spatial information compared to requires cloud information, not only on the total amount but also
single pixels, and it is often argued that it is this spatial information the location of clouds and cloud shadows in a Landsat scene. In
that provides greater advantage to OBIA using segments (Blaschke parallel with the growing demand and use of Landsat imagery,
and Strobl, 2001; Flanders etal., 2003; Hay and Castilla, 2008). there have also been a number of new developments in detecting
One question surrounding the OBIA is the scale at which the clouds in Landsat images (e.g., Irish etal., 2006; Oreopoulos etal.,
analysis is made. More specifically, at what scale (or spatial reso- 2011; Zhu and Woodcock, 2012; Goodwin etal., 2013).
lution) does the image under consideration lends itself to image One of the more robust methods in detecting cloud and cloud
segmentation for the purpose of object-based analysis? It turns shadows was developed by Zhu and Woodcock (2012). This
out the answer is not related only to spatial resolution but also method uses a series of rules based on calculated probabilities
2016 by Taylor & Francis Group, LLC
240 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 11.4 Object-based analysis example: (a) the original Landsat TM image in true color; (b) the original USDA Cropland Data Layer
(c)object-filled USDA Cropland Data Layer in which the objects (agricultural field boundaries) were used to convert pixel-based classification
results to a polygon based map.
of temperature, spectral variability, and brightness, using top Bartholom and Belward, 2005), the new developments mostly
of atmosphere (TOA) reflectance and brightness temperature apply to medium-resolution (i.e. less than 100 m pixels) images
(BT) as inputs. The clouds and cloud shadows are treated as and rely on automated algorithms (Kalensky, 1998; Gong etal.,
3D objects determined via segmentation of the potential cloud 2013; Sexton et al., 2013; Yu et al., 2013). Also contributing to
layer and an assumption of a constant temperature lapse rate. these developments is the availability of powerful computers
The solar illumination and sensor view angles are used to predict and image classification software, mainly based on nonparamet-
possible cloud shadow locations and select the one that has the ric methods that are able to generalize across space and time.
maximum similarity to cloud shape and size. For example, development of the Web-Enabled Landsat Data
(WELD) (Roy et al., 2010) system has revolutionized the way
11.2.2.4.2 LargeAreaMapping Landsat data are being processed and temporally aggregated for
Free and open access to most satellite data (e.g., Landsat, Sentinel) rapid and efficient large-area applications. However, despite the
are changing the way remotely sensed images are processed. For existence and the availability of maps over large areas, there is
example, there is increased interest in automatically processing still the issue of different land-cover definitions as well as the
large volumes of imagery covering large areas of the Earth. These use of different algorithms. In Figure 11.5, we show forest clas-
methods rely on availability of a large number of images within sification results from three large-area classification projects
and across the years and use both the spectral and the temporal in Kenya. While there is strong correlation between each map
information available in satellite data. While the concept of time product, notable differences remain. For example, the FROM
series observation for large area land-cover mapping is not new GLC product (Gong etal., 2013) has a large forest area than the
(Hansen et al., 2000; Loveland et al., 2000; Friedl et al., 2002; other products. Alternatively, in locations where the UMD tree
(a) (b) (c)
FIg ur e 11.5 Comparison of three land cover maps for the forest category in Kenya. (a) is the land cover map from the AfriCover project; (b)for-
est category of a Landsat based land cover map as part of the FROM GLC project; (c) tree cover map derived from multi-temporal Landsat data.
The black polygon outlines the boundaries of a single Landsat WRS2 footprint (path:168 row:60).
2016 by Taylor & Francis Group, LLC
Image Classification Methods in Land Cover and Land Use 241
cover map (Sexton etal., 2013) shows small tree cover fractions for evaluating maps of past time periods. Finally, maps derived
(i.e., less than 50% threshold), the AfricCover product (Kalensky, from object-based analysis are evaluated at the object level.
1998) maps them as full forest. Of course, this is partly related to That is, accuracy assessment is performed using the objects
class definitions as each of three products uses different descrip- (e.g., polygons) in the independent sample in which the class
tions of what a forest is. Nevertheless, data in Figure 11.5 simply label of the entire polygon (not the label of an individual pixel
highlights the difficulty with which large area land-cover maps inside the polygon) is used for evaluation. In evaluating pixel-
are made using medium- to high-resolution satellite data. based maps, it is recommended that an area corresponding to at
least one pixel shift in all directions be included in the analysis
to address the geolocation-based errors inherent in all satellite
11.2.3 Uncertainty Assessment
data. For example, a map made from Landsat data with 30 m
Accuracy assessment of classification results developed from satel- pixels would require a square-like polygon, at least 90 m on a
lite data (or other sources) is an important but often neglected step in side, centered on the pixel to be evaluated. Information inside
producing land-cover maps. The application of various image clas- this polygon is then used to label the samples in the indepen-
sification methods to satellite data produces a map that is hypoth- dent dataset.
esized to represent the land cover in question. Like all hypothesis Once the independent sample is acquired, development of
testing, the next step is to obtain data, preferably from an indepen- the confusion matrix (or error matrix) is the standard practice.
dent sources, and test the validity of the hypothesis in map form. The confusion matrix is a specific table that allows assessment
One reason map validation is either neglected or underestimated of both class-specific and overall accuracies rapidly. The table is
stems from the fact that the accuracy assessmentif performed also used to quantify the omission and commission errors that
rightis a difficult and costly task. In general, validation is a com- allow the specific performance of the classifier in use. In gen-
plex task as maps made from satellite data cover large areas rela- eral, when reporting map accuracy, it is recommended that the
tive to the spatial sampling unit (i.e., pixel or a polygon). Moreover, confusion matrix along with all the other accuracy measures be
collecting the reference data to be used in validation is expensive reported. This allows the map user to draw his/her own conclu-
and labor intensive. For this reason, the sampling design to acquire sions when using the land-cover map.
accurate, representative, statistically sound, and independent refer-
ence data must be carefully crafted and evaluated against available
resources allocated to the validation procedure. Therefore, the key 11.3 the Future
to an effective accuracy assessment is a sound sampling plan that
balances accuracy and pragmatism while including both the com- It is the authors opinion that the future of land-cover mapping
mon and rare categories (Zhu etal., 2000). from satellite data is bright. Contributing factors include: (1) free
It is generally accepted that accuracy assessment using an and unlimited access to satellite observations; (2) development
independent, stratified random sample produces the best eval- of cheaper and better sensors; (3) availability of dense temporal
uation results with respect to statistical rigor and representa- observations, even at high spatial resolutions; (4) advances in
tives (Congalton, 1991; Stehman and Czaplewski, 1998). In a computer hardware and software rooted in artificial intelligence
stratified random design, the map categories often constitute community; (5) ease of access using web-enabled services; and
the strata so that each class could be evaluated using an inde- (6) availability of ever-growing number of ancillary datasets on
pendent random sample. The allocation strategy could be pro- environmental and social variables that in turn help determine
portionate in which known class areas determine the number the presence or absence of a specific land cover. As our planet is
of samples per class, or disproportionate in which a set num- continually pushed to meet the demands of a growing and afflu-
ber of samples is assigned in category. One disadvantage of the ent population, identifying and mapping of its resources will be
proportional allocation is that small classes receive a small, even more paramount. It is hoped that the new developments
and in some cases unrealistic, number of samples. While the in image processing and satellite data availability will allow us
question of sample size, both per class and total, has received to make more accurate and refined land-cover maps to face this
due attention in the literature (e.g., see Stehman, 2012 for a challenge.
comprehensive review), the number of samples to be used in
the validation effort ultimately depends on the resources avail- References
able. Note that with the development of free, high-resolution,
and accurately located image data, such as those available from Akaike, H. 1973. Information theory and an extension of the
Google Earth, some of the cost associated with field work has maximum likelihood principle, in Proceedings of the Second
been shifted to the office in the form of manual interpretation. International Symposium on Information Theory. Petrov, B.
However, caution must be exercised when using these sources and Cazakil, F. eds. Budapest, Hungary: Aakademiai Kidao.
as the independent sample, because complex land-cover catego- Albertz, D. 2007. Einfhrung in die Fernerkundung. Grundlagen der
ries do not always lend themselves to visual analysis. Moreover, Interpretation von Luft- & Satellitenbildern. (Darmstadt) 254pp.
these online sources often represent a single time period cor- Alpaydin, E. 2004. Introduction to Machine Learning. Cambridge,
responding to current or recent past, rendering them less useful MA: MIT Press.
2016 by Taylor & Francis Group, LLC
242 Remotely Sensed Data Characterization, Classification, and Accuracies
Anderson, J. R. 1971. Land use classification schemes used in Carpenter, G. A., Grossberg, S., Markuzon, N., Reynolds, J. H.,
selected recent geographic applications if remote sensing. and Rosen, D. B. 1992. Fuzzy ARTMAP: A neural network
Photogrammatic Engineering, 37(4), 379387. architecture for incremental supervised learning of ana-
Anderson, J. R., Hardy, E. E., Roach, J. T., and Witmer, R. E. log multidimensional maps. IEEE Transactions on Neural
1976. A land use and land cover classification system for Networks, 3, 698713.
use with remote sensor data. USGS Professional Paper Carpenter, G. A. and Grossberg, S. 2003. Adaptive resonance theory,
964, 28pp. in M. A. Arbib (ed.), The Handbook of Brain Theory and Neural
Arvidson, T., Gasch, J., and Goward, S. N. 2001. Landsat-7s long- Networks, 2nd edn. Cambridge, MA: MIT Press, pp. 8790.
term acquisition planAn innovative approach to building Chang, C.-C. and Lin, J.-C. 2011. LIBSVM: A library for support
a global imagery archive. Remote Sensing of Environment, vector machines. ACM Transactions on Intelligent Systems
78(12), 1326. and Technology, 2, 139.
Atkinson, P. M. and Tatnall, A. R. 1997. Introduction: Neural net- Chen, M.-H., Ibrahim, J. G., and Yianoutsos, C. 1999. Prior elici-
works in remote sensing. International Journal of Remote tation, variable selection, and Bayesian computation for
Sensing, 18, 699709. logistics regression models. Journal of the Royal Statistical
Bartholom, E. and Belward, A. S. 2005. GLC2000: A new approach Society B, 61, 223242.
to global land cover mapping from Earth observation data. Congalton, R. G. 1991. A review of assessing the accuracy of
International Journal of Remote Sensing, 26(9), 19591977. classification of remote sensed data. Remote Sensing of
Benediktsson, J., Swain, P. H., and Ersoy, O. K. 1990. Neural net- Environment, 37, 3546.
work approaches versus statistical methods in classification Cortes, C. and Vapnik, V. 1995. Support-vector networks. Machine
of multisource remote sensing data. IEEE Transactions on Learning, 20, 273297.
Geoscience and Remote Sensing, 28(4), 540552. DeFries, R. S. and Chan, J.C-W. 2000. Multiple criteria for evaluating
Benz, U. C., Hofmann, P., Willhauck, G., Lingenfelder, I., and Heynen, machine learning algorithms for land cover classification from
M. 2004. Multi-resolution, object-oriented fuzzy analysis of satellite data. Remote Sensing of Environment, 74, 503515.
remote sensing data for GIS-ready information. ISPRS Journal DeFries, R. S., Hansen, M., Townshend, J. R. G., and Sohlberg, R.
of Photogrammetry and Remote Sensing, 58, 239258. 1998. Global land cover classifications at 8km spatial resolu-
Blaschke, T., Burnett, C., and Pekkarinen, A. 2006. Image seg- tion: the use of training data derived from Landsat imagery
mentation methods for object-based analysis and classi- in decision tree classifiers. International Journal of Remote
fication, in S. de Jong and F. van der Meer (eds.), Remote Sensing, 19, 31413168.
Sensing Image Analysis: Including the Spatial Domain. Esch, T., Thiel, M., Bock, M., Roth, A., and Dech, S. 2008.
Berlin, Germany: Springer, pp. 211236. Improvement of image segmentation accuracy based on
Blaschke, T. 2010. Object based image analysis for remote sens- multiscale optimization procedure. IEEE Geosciences and
ing. ISPRS Journal of Photogrammetry and Remote Sensing, Remote Sensing Letters, 5(3), 463467.
65, 216. Fisher, P. 1997. The pixel: A snare and a delusion. International
Blaschke, T. and Hay, G. J. 2001. Object-oriented image analy- Journal of Remote Sensing, 18, 679685.
sis and scale-space: Theory and methods for modeling and Flanders, D., Hall-Beyer, M., and Pereverzoff, J. 2003. Preliminary
evaluating multi-scale landscape structure. International evaluation of eCognition object-based software for cut
Archives of Photogrammetry and Remote Sensing, 34(Part 4/ block delineation and feature extraction. Canadian Journal
W5), 2229. of Remote Sensing, 29(4), 441452.
Blaschke, T. and Strobl, J. 2001. Whats wrong with pixels? Some Friedl, M. A., McIver, D. K., Hodges, J. C. F., Zhang, X. Y.,
recent developments interfacing remote sensing and Muchoney, D., Strahler, A. H., Woodcock, C. E. etal. 2002.
GIS. GISZeitschrift fur Geoinformationssysteme, 14(6), Global land cover mapping from MODIS: Algorithms and
1217. early results. Remote Sensing of Environment, 83, 287302.
Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J. Friedl, M. A., Sulla-Menashe, D., Tan, B., Schneider, A.,
1984. Classification and Regression Trees. Belmont, CA: Ramankutty, N., Sibley, A., and Huang, X. 2010. MODIS
Wadsworth International Group. Collection 5 global land cover: Algorithm refinements
Brodley, C. E. and Utgoff, P. E. 1995. Multivariate decision trees. and characterization of new datasets. Remote Sensing of
Machine Learning, 19, 4577. Environment, 114, 168182.
Carpenter, G. A., Gjaja, M. N., Gopal, S., and Woodcock, C. E. 1997. Friedl, M. A. and Brodley, C. E. 1997. Decision tree classification
ART neural networks for remote sensing: Vegetation classifi- of land cover from remotely sensed data. Remote Sensing of
cation from Landsat TM and terrain data. IEEE Transactions Environment, 61(3), 399409.
on Geoscience and Remote Sensing, 35(2), 308325. Gong, P., Wang, J., Yu, L., Zhao, Y. C., Zhao, Y. Y., Liang, L., Niu,
Carpenter, G. A., Grossberg, S., and Reynolds, J. H. 1991. Z.G. etal. 2013. Finer resolution observation and monitor-
ARTMAP: Supervised real-time learning and classification ing of global land cover: First mapping results with Landsat
of nonstationary data by a self-organizing neural network. TM and ETM+ data. International Journal of Remote
Neural Networks, 4, 565588. Sensing, 34(7), 26072654.
2016 by Taylor & Francis Group, LLC
Image Classification Methods in Land Cover and Land Use 243
Goodwin, N. R., Collett, L. J., Denham, R. J., Flood, N., and Tindall, D. Jia, X. and Richards, J. A. 1994. Efficient maximum likelihood classi-
2013. Cloud and cloud shadow screening across Queensland, fication for imaging spectrometer data sets. IEEE Transactions
Australia: An automated method for Landsat TM/ETM+ time on Geoscience and Remote Sensing, 32(2), 274281.
series. Remote Sensing of Environment, 134, 5065. Kalensky, Z. D. 1998. AFRICOVER land cover database and map of
Gopal, S. and Woodcock, C. E. 1996. Remote sensing of forest Africa. Canadian Journal of Remote Sensing, 24(3), 292297.
change using artificial neural networks. IEEE Transactions Kavzgolu, T. 2009. Increasing the accuracy of neural network
on Geoscience and Remote Sensing, 34(2), 398404. classification using refined training data. Environmental
Hall, F. G. and Knapp, D. 2000. BOREAS TE-18 Landsat TM Modelling and Software, 24(7), 850858.
Maximum Likelihood Classification Image of the SSA, Kavzoglu, T. and Colkesen, I. 2009. A kernel functions analysis
Technical Report Series on the Boreal Ecosystem-Atmosphere for support vector machines for land cover classification.
Study (BOREAS), Vol. 175. Greenbelt, MD: NASA. International Journal of Applied Earth Observation and
Hansen, M., DeFries, R., Townshend, J. R. G., and Sohlberg, R. Geoinformation, 11, 352359.
2000. Global land cover classification at 1 km resolution Krishnaswamy, J., Kiran, M. C., and Ganeshaiah, K. N. 2004.
using a decision tree classifier. International Journal of Tree model based ecoclimatic vegetation classification and
Remote Sensing, 21, 13311365. fuzzy mapping in diverse tropical deciduous ecosystems
Hansen, M., Dubayah, R., and DeFries, R., 1996, Classification using multi-season NDVI. International Journal of Remote
trees: An alternative to traditional land cover classifiers. Sensing, 25(6), 11851205.
International Journal of Remote Sensing, 17, 10751081. Lawrence, R., Bunn, A., Powell, S., and M. Zambon. 2004.
Haralick R. M. and Shapiro L. G. 1985. Survey: Image segmen- Classification of remotely sensed imagery using stochastic
tation techniques. Computer Vision, Graphics and Image gradient boosting as a refinement of classification tree anal-
Processing, 29(1), 100132. ysis. Remote Sensing of Environment 90, 331336.
Hay, G. J., Castilla, G., Wulder, M. A., and Ruiz, J. R. 2005. An auto- Lawrence, R. L. and Wright, A. 2001. Rule-based classification
mated object-based approach for the multiscale image seg- systems using classification and regression tree (CART)
mentation of forest scenes. International Journal of Applied analysis. Photogrammetric Engineering and Remote Sensing
Earth Observation and Geoinformation, 7(4), 339359. 67, 11371142.
Hay, G. J., Marceau, D. J., Dube, P., and Bouchard, A. 2001. A Lin, C.-F. and Wang, S.-D. 2004. Training algorithms for fuzzy
multiscale framework for landscape analysis: Object- support vector machines with no data. Pattern Recognition
specific analysis and upscaling. Landscape Ecology, 16(6), Letters, 25, 16471656.
471490. Lin, C.-F. and Wang, S. D. 2002. Fuzzy support vector machines.
Hay, G. J. and Castilla, G. 2008. Geographic Object-Based Image IEEE Transactions on Neural Networks, 13, 464471.
Analysis (GEOBIA): A new name for a new discipline, in Lippman, R. P. 1987. An introduction to computing with neural
T. Blaschke, S. Lang, and G. Hay (eds.), Object Based Image nets. IEEE ASSP Magazine, 4, 222.
Analysis. New York: Springer, pp. 93112. Lizarazo, I. and Elsner, P. 2009. Fuzzy segmentation for object-
Hsu, C.-W., Chang, C.-C., and Lin, C.-J. 2001. A practical guide to based image classification. International Journal of Remote
support vector classification. Available at http://www.csie. Sensing, 30(6), 16431649.
ntu.edu.tw/~cjlin/. Accessed March 20, 2015. Lizarazo, I. and Elsner, P. 2011. Segmentation of Remotely Sensed
Huang, C., Davis, L. S., and Townshend, J. R. G. 2002. An Imagery: Moving from Sharp Objects to Fuzzy Regions, in
assessment of support vector machines for land cover Dr. Pei-Gee Ho (ed.), Image Segmentation, ISBN: 978-953-
classification. International Journal of Remote Sensing, 23, 307-228-9, InTech, Rijeka, Croatia. Available from: http://
725749. www.intechopen.com/books/image-segmentation/segmen-
Huang, X. and Jensen, J. R. 1997. A machine-learning approach tation-of-remotely-sensed-imagery-moving-from-sharp-
to automated knowledge-base building for remote sensing objects-to-fuzzy-regions. Accessed March 20, 2015.
image analysis with GIS data. Photogrammetric Engineering Loveland, T. R., Merchant, J. W., Ohlen, D. O., and Brown, J. F.
and Remote Sensing, 63(10), 11851194. 1991. Development of a land-cover characteristics database
Irish, R., Barker, J. L., Goward, S. N., and Arvidson, T. 2006. for the conterminous U.S. Photogrammatic Engineering and
Characterization of the Landsat-7 ETM+ Automated Cloud- Remote Sensing, 57, 14531463.
Cover Assessment (ACCA) algorithm. Photogrammetric Loveland, T. R., Reed, B. C., Brown, J. F., Ohlen, D. O., Zhu, Z.,
Engineering and Remote Sensing, 72(10), 11791188. Yang, L., and Merchant, J. W. 2000. Development of a global
Irish, R. 2000. Landsat-7 automatic cloud cover assessment algo- land cover characteristics database and IGBP DISCover
rithms for multispectral, hyperspectral, and multraspectral from 1 km AVHRR data. International Journal of Remote
imagery. The International Society for Optical Engineering, Sensing, 21(67), 13031330.
4049, 348355. Mantero, P., Moser, G., and Serpico, S. B. 2005. Partially super-
Jensen, J. R. 2006. Remote Sensing of the Environment: An Earth vised classification of remote sensing images through SVM-
Resource Perspective, 2nd edn. Englewood Cliffs, NJ: based probability density estimation. IEEE Transactions on
Prentice Hall, 608pp. Geoscience and Remote Sensing, 43, 559570.
2016 by Taylor & Francis Group, LLC
244 Remotely Sensed Data Characterization, Classification, and Accuracies
McIver, D. K. and Friedl, M. A. 2002. Using prior probabilities in Ryherd, S. and Woodcock, C. E. 1996. Combining spectral and
decision-tree classification of remotely sensed data. Remote texture data in the segmentation of remotely sensed images.
Sensing of Environment, 81(23), 253261. Photogrammetric Engineering and Remote Sensing, 62(2),
Mondal, A., Kundu, S., Chandniha, S. K., Shukla, R., and Mishr, 181194.
P. K. 2012. Comparison of support vector machine and Safavian, S. R. and Landgrebe, D. 1991. A survey of decision tree
maximum likelihood classification technique using satellite classifier methodology. IEEE Transactions on Systems, Man,
imagery. International Journal of Remote Sensing and GIS, and Cybernetics, 21, 660674.
1(2), 116123. Scholkopf, B. and Smola, A. J. 2001. Learning with Kernels.
Montgomery, D. C. and Peck, E. A. 1992. Introduction to Linear Cambridge, MA: The MIT Press.
Regression Analysis, 2nd edn. New York: Wiley. Sexton, J. O., Song, X.-P., Feng, M., Noojipady, P., Anand, A.,
Mountrakis, G., Im, J., and Ogole, C. 2011. Support vector Huang, C., Kim, D.-H., Collins, K. M., Channan, S., DiMiceli,
machines in remote sensing: A review. ISPRS Journal of C., and Townshend, J. R. G. 2013. Global, 30-m resolution
Photogrammetry and Remote Sensing, 66, 247259. continuous fields of tree cover: Landsat-based rescaling of
Oreopoulos, L., Wilson, M., and Vrnai, T. 2011. Implementation MODIS Vegetation Continuous Fields with lidar-based esti-
on Landsat data of a simple cloud mask algorithm developed mates of error. International Journal of Digital Earth, 6(5),
for MODIS land bands. IEEE Transactions on Geoscience 130321031236007. doi:10.1080/17538947.2013.786146.
and Remote Sensing, 8(4), 597601. Shao, Y. and Lunetta, R. 2012. Comparison of support vector
Pal, M. and Mather, P. M. 2003. An assessment of the effective- machine, neural network, and CART algorithms for the land-
ness of decision tree methods for land cover classification. cover classification using limited training data points. ISPRS
Remote Sensing of Environment, 86, 554556. Journal of Photogrammetry and Remote Sensing, 70, 7887.
Paola, J. D. and Schowengerdt, R. A. 1995. A review and analy- Simpson, J. J. and Stitt, J. R. 1998. A procedure for the detection
sis of backpropagation neural networks for classification and removal of cloud shadow from AVHRR data over land.
of remotely sensed multi-spectral imagery. International Geoscience and Remote Sensing, 36(3), 880890.
Journal of Remote Sensing, 16, 30333058. Stehman, S. V. 2012. Impact of sample size allocation when
Platt, R. V. and Rapoza, L. 2008. An evaluation of an object- using stratified random sampling to estimate accuracy
oriented paradigm for land use/land cover classification. and area of land-cover change. Remote Sensing Letters,
The Professional Geographer, 60(1), 87100. 3(2), 111120.
Pratola, C., Del Frate, F., Schiavon, G., Solimini, D., and Stehman, S. V. and Czaplewski, R. L. 1998. Design and analysis
Licciardi, G. 2011. Characterizing land cover from X-band of thematic map accuracy assessment: Fundamental prin-
COSMO-SkyMed images by neural networks, in Urban ciples. Remote Sensing of Environment, 64, 331344.
Remote Sensing, April 1113. Munich, Germany: IEEE, Strahler, H. 1980. The use of prior probabilities in Maximum
pp. 4952. Likelihood Classification of remotely sensed data. Remote
Quinlan, J. R. 1993. C4.5 Programs for Machine Learning. San Sensing of Environment, 10,135163.
Mateo, CA: Morgan Kaufmann Publishers. Sun, J., Yang, J., Zhang, C., Yun, W., and Qu, J. 2013. Automatic
Richards, J. A. 2012. Remote Sensing Digital Image Analysis: An remotely sensed image classification in a grid environment
Introduction, 5th edn. New York: Springer Science+Business based on the maximum likelihood method. Mathematical
Media and Computer Modeling, 58(34), 573581.
Rodriguez-Galiano, V., Ghimire, B., Rogan, J., Chica-Olmo, M., Suykens, J. A. K., Brabanter, J. D., Lukas, L., and Vandewalle, J. 2002.
and Rigol-Sanchez, J. 2012. An assessment of the effective- Weighted least squares support vector machines: Robustness
ness of a random forest classifier for land-cover classifica- and sparse approximation. Neurocomputing, 48, 85105.
tion. ISPRS Journal of Photogrammetry and Remote Sensing, Townshend, J. R. G., Huang, C., Kalluri, S. N., DeFries, R. S.,
67, 93104. Liang, S., and Yang, K. 2000. Beware of per-pixel charac-
Rogan, J., Franklin, J., and Roberts, D. 2002. A comparison of terization of land cover. International Journal of Remote
methods for monitoring multitemporal vegetation change Sensing, 21(4), 839843.
using Thematic Mapper imagery. Remote Sensing of Travaglia, C., Milenova, L., Nedkov, R., Vassilev, V., Milenov, P.,
Environment, 80(1), 143156. Radkov, R., and Pironkova, Z. 2001. Preparation of land
Rogan, J., Miller, J., Stow, D., Franklin, J., Levien, L., and Fischer, C. cover database of Bulgaria through remote sensing and GIS.
2003. Land cover change mapping in California using classifica- Environment and Natural Resources Working Paper No. 6,
tion trees with Landsat TM and ancillary data. Photogrammetric FAO, Rome, Italy, 57pp.
Engineering and Remote Sensing, 69(7), 793804. Tsujinishi, D. and Abe, S. 2003. Fuzzy least squares support vec-
Roy, D. P., Ju, J., Kline, K., Scaramuzza, P. L., Kovalskyy, V., tor machines for multiclass problems. Neural Networks, 16,
Hansen, M. C., Loveland, T. R., Vermote, E. F., and Zhang, 785792.
C. 2010. Web-enabled Landsat Data (WELD): Landsat Vapnik, V. 1979. Estimation of Dependences Based on Empirical
ETM+ Composited Mosaics of the Conterminous United Data. Moscow, Russia: Nauka, pp. 51655184, 27 (in Russian)
States. Remote Sensing of Environment, 114, 3549. (English translation: New York: Springer Verlag, 1982).
2016 by Taylor & Francis Group, LLC
Image Classification Methods in Land Cover and Land Use 245
Vapnik, V. 1995. The Nature of Statistical Learning Theory, 2nd Yu, L., Wang, J., and Gong, P. 2013. Improving 30 meter
edn. New York: Springer. global land cover map FROM-GLC with time series
Vogelmann, J. E., Howard, S. M., Yang, L., Larson, C. R., Wylie, B. K., MODIS and auxiliary datasets: A segmentation based
and Van Driel, J. N. 2001. Completion of the 1990s National approach. International Journal of Remote Sensing, 34(16),
Land Cover Data Set for the conterminous United States. 58515867.
Photogrammetric Engineering and Remote Sensing, 67, 650662. Zhu, G. and Blumberg, D. G. 2002. Classification using ASTER
Wacker, A. G. and Landgrebe, D. A. 1972. Minimum distance clas- data and SVM algorithms: The case study of Beer Sheva,
sification in remote sensing. LARS Technical Reports. Paper Israel. Remote Sensing of Environment, 80, 233240.
25. http://docs.lib.purdue.edu/larstech/25. Accessed March 20, Zhu, Z., Yang, L., Stehman, S. V., and Czaplewski, R. L. 2000.
2015. Accuracy Assessment for the U.S. Geological Survey
Woodcock, C. E. and Harward, V. J. 1992. Nested-hierarchical Regional Land-Cover Mapping Program: New York and
scene models and image segmentation. International New Jersey Region. Photogrammetric Engineering and
Journal of Remote Sensing, 13, 31673187. Remote Sensing, 66(12), 14251435.
Yoshida, T. and Omatu, S. 1994. Neural network approach to Zhu, Z. and Woodcock, C. E. 2012. Object-based cloud and cloud
land cover mapping. IEEE Transactions on Geoscience and shadow detection in Landsat imagery. Remote Sensing of
Remote Sensing, 32(5), 11031109. Environment, 118, 8394.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
12
Hyperspectral Image Processing:
Methods and Approaches
Acronyms and Definitions ................................................................................................................ 247
12.1 Introduction ............................................................................................................................ 247
12.2 Classification Approaches ..................................................................................................... 248
Supervised Classification SpectralSpatial Classification Subspace-Based
Approaches Semisupervised Classification
Jun Li 12.3 Experimental Comparison .....................................................................................................252
Sun Yat-Sen University
12.4 Conclusions and Future Directions ..................................................................................... 256
Antonio Plaza Acknowledgments .............................................................................................................................. 257
University of Extremadura References............................................................................................................................................ 257
Acronyms and Definitions image-processing algorithms have been presented in the lit-
erature [3,7,8]. It should be noted, however, that most recently
EMPs Extended morphological profiles developed hyperspectral image-processing techniques focus
EMPs Extended morphological profiles on analyzing the spectral and spatial informations contained
LDA Linear discriminant analysis in the hyperspectral data in simultaneous fashion [9]. In other
LogDA Logarithmic discriminant analysis words, the importance of analyzing spatial and spectral infor-
MLR Multinomial logistic regression mation simultaneously has been identified as a desired goal
MLRsubMRF Subspace-based multinomial logistic regression by many scientists devoted to hyperspectral image analy-
followed by Markov random fields sis. This type of processing has been approached in the past
MPs Morphological profiles from various points of view. For instance, several possibilities
MRFs Markov random fields are discussed by Landgrebe [10] for the refinement of results
PCA Principal component analysis obtained by spectral-based techniques through a second step
QDA Quadratic discriminant analysis based on spatial context. Such contextual classification [11]
RHSEG Recursive hierarchical segmentation accounts for the tendency of certain ground cover classes to
ROSIS Reflective optics spectrographic imaging system occur more frequently in some contexts than in others. In cer-
SVMs Support vector machines tain applications, the integration of high spatial and spectral
TSVMs Transductive support vector machines information is mandatory to achieve sufficiently accurate map-
ping and/or detection results. For instance, urban area map-
12.1 introduction ping requires sufficient spatial resolution to distinguish small
spectral classes, such as trees in a park or cars on a street [12]
Hyperspectral imaging is concerned with the measurement, (Figure12.1).
analysis, and interpretation of spectra acquired from a given However, there are several important challenges when per-
scene (or specific object) at a short, medium, or long distance, forming hyperspectral image classification. In particular, super-
typically, by an airborne or satellite sensor [1]. The special vised classification faces challenges related with the unbalance
characteristics of hyperspectral data sets pose different pro- between high dimensionality and the limited number of train-
cessing problems, which must be necessarily tackled under ing samples or the presence of mixed pixels in the data (which
specific mathematical formalisms [2], such as classification may compromise classification results for coarse spatial resolu-
and segmentation [3] or spectral mixture analysis [4]. Several tions). Specifically, due to the small number of training samples
machine learning and image-processing techniques have been and the high dimensionality of the hyperspectral data, reliable
applied to extract relevant information from hyperspectral estimation of statistical class parameters is a very challenging
data during the last decade [5,6]. Taxonomies of hyperspectral goal [13]. As a result, with a limited training set, classification
247
2016 by Taylor & Francis Group, LLC
248 Remotely Sensed Data Characterization, Classification, and Accuracies
Atmosphere
1
0.8
Reflectance
0.6
0.4
Ultraspectral 0.2
(1000s of bands) 0
400 800 1200 1600 2000 2400
Wavelength (nm)
Soil
1
0.8
Reflectance
0.6
0.4
0.2
Hyperspectral 0
(100s of bands) 400 800 1200 1600 2000 2400
Water Wavelength (nm)
Reflectance
0.8
0.6
0.4
Vegetation 0.2
0
400 800 1200 1600 2000 2400
Multispectral Wavelength (nm)
(10s of bands) 1
0.8
Reflectance
0.6
0.4
0.2
0
Panchromatic 400 800 1200 1600 2000 2400
Wavelength (nm)
FIg u r e 1 2.1 The challenges of increased dimensionality in remote-sensing data interpretation.
accuracy tends to decrease as the number of features increases. 12.2 classification Approaches
This is known as the Hughes effect [14]. Another relevant chal-
lenge is the need to integrate the spatial and spectral information In this section, we outline some of the main techniques and
to take advantage of the complementarities that both sources of challenges in hyperspectral image classification. Hyperspectral
information can provide. These challenges are quite important image classification has been a very active area of research in
for future developments and solutions to some of them have recent years [3]. Given a set of observations (i.e., pixel vectors
been proposed. Specifically, supervised [15] and semisupervised in a hyperspectral image), the goal of classification is to assign
[1618] techniques for hyperspectral image classification, strat- a unique label to each pixel vector so that it is well defined by a
egies for integrating the spatial and the spectral information given class. In Figure 12.2, we provide an overview of a popular
[1922], or subspace classifiers [23] that can better exploit the strategy to conduct hyperspectral image classification, which is
intrinsic nature of hyperspectral data have been quite popular in based on the availability of labeled samples. After an optional
the recent literature. dimensionality reduction step, a supervised classifier is trained
Our main goal in this chapter is to provide a seminal view on using a set of labeled samples (which are often randomly selected
recent advances in techniques for hyperspectral image analysis from a larger pool of samples) and then tested with a disjoint set
that can successfully deal with the dimensionality problem and of labeled samples in order to evaluate the classification accuracy
with the limited availability of training samples a priori, while of the classifier.
taking into account both the spectral and spatial properties of Supervised classification has been widely used in hyperspec-
the data. The remainder of the chapter is structured as follows. tral data interpretation [2], but it faces challenges related with
Section 12.2 discusses available techniques for hyperspectral the high dimensionality of the data and the limited availability
image classification, including both supervised and semisuper- of training samples, which may not be easy to collect in pure
vised approaches, techniques for integrating spatial and spectral form. However, mixed training samples can also offer relevant
information and subspace-based approaches. Section 12.3 pro- information about the participating classes [10]. In order to
vides an experimental comparison of the techniques discussed address these issues, subspace-based approaches [23,24] and
in Section 12.2, using a hyperspectral data set collected by the semisupervised learning techniques [25] have been developed.
ROSIS over the University of Pavia, Italy, which is used here In subspace approaches, the goal is to reduce the dimensionality
as a common benchmark to outline the properties of the dif- of the input space in order to better exploit the (limited) train-
ferent processing techniques discussed in the chapter. Finally, ing samples available. In semisupervised learning, the idea is to
Section 12.4 concludes the paper with some remarks and hints at exploit the information conveyed by additional (unlabeled) sam-
the most pressing ongoing research directions in hyperspectral ples, which can complement the available labeled samples with a
image classification. certain degree of confidence. In all cases, there is a clear need to
2016 by Taylor & Francis Group, LLC
Hyperspectral Image Processing 249
Dimensionality reduction
(optional)
Randomly selected
Test Labeled Supervised classifier
samples samples (may approaches)
Test classification accuracy
FIg u r e 1 2.2 Standard approach for supervised hyperspectral image classification.
integrate the spatial and spectral information to take advantage set mapped into a Hilbert space by some mapping, the SVM
of the complementarities that both sources of information can separates the data by an optimal hyperplane that maximizes
provide [9]. An overview of these different aspects, which are the margin (see Figure 12.4). However, the most widely used
crucial to hyperspectral image classification, is provided in the approach in hyperspectral classification is to combine soft mar-
following subsections. gin classification with a kernel trick that allows separation of
the classes in a higher-dimensional space by means of a nonlin-
ear transformation (see Figure 12.5). In other words, the SVM
12.2.1 Supervised classification
used with a kernel function is a nonlinear classifier, where the
Several techniques have been used to perform supervised nonlinear ability is included in the kernel and different kernels
classification of hyperspectral data. For instance, in discrimi- lead to different types of SVMs. The extension of SVM to the
nant classifiers, several types of discriminant functions can multiclass cases is usually done by combining several binary
be applied: nearest neighbor, decision trees, linear functions, classifiers.
or nonlinear functions (see Figure 12.3). In linear discrimi-
nant analysis (LDA) [26], a linear function is used in order to
maximize the discriminatory power and separate the available x2
classes effectively. However, such a linear function may not be Margin
the best choice, and nonlinear strategies such as quadratic dis-
criminant analysis (QDA) or logarithmic discriminant analy-
sis (LogDA) have also been used. The main problem of these
classic supervised classifiers, however, is their sensitivity to the
Hughes effect.
In this context, kernel methods such as the support vector
machine (SVM) have been widely used in order to deal effec-
tively with the Hughes phenomenon [27,28]. The SVM was
first investigated as a binary classifier [29]. Given a training
Support vectors
Nearest Decision Linear Nonlinear
x1
neighbor tree functions functions
FIg ur e 12.3 Typical discriminant functions used in supervised
classifiers. FIg u r e 1 2.4 Soft margin classification with slack variables.
2016 by Taylor & Francis Group, LLC
250 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 1 2.5 The kernel trick allows separation of the classes in a higher-dimensional space by means of a linear or nonlinear transformation.
12.2.2 SpectralSpatial classification spatial characteristics of the objects in remotely sensed images.
Advanced morphological techniques such as morphological
Several efforts have been performed in the literature in order to profiles [31] have been successfully used for feature extrac-
integrate spatialcontextual information in spectral-based clas- tion prior to classification of hyperspectral data by extracting
sifiers for hyperspectral data [3,9]. It is now commonly accepted the first few principal components of the data using principal
that using the spatial and the spectral information simultane- component analysis [13] and then building so-called extended
ously provides significant advantages in terms of improving the morphological profiles (EMPs) on the first few components to
performance of classification techniques. An illustration of the extract relevant features for classification [32].
importance of integrating spatial and spectral information is Another strategy in the literature has been to exploit simulta-
given in Figure 12.6. As shown in this figure, spectralspatial neously the spatial and the spectral information. For instance,
classification (obtained using morphological transformations) in order to incorporate the spatial context into kernel-based
provides a better interpretation of classes such as urban features, classifiers, a pixel entity can be redefined, simultaneously both
with a better delineation and characterization of complex urban in the spectral domain (using its spectral content) and also in
structures. the spatial domain, by applying some feature extraction to its
Some of the approaches that integrate spatial and spec- surrounding area, which yields spatial (contextual) features,
tral information include spatial information prior to the clas- for example, the mean or standard deviation per spectral band.
sification, during the feature extraction stage. Mathematical These separated entities lead to two different kernel matrices,
morphology [30] has been particularly successful for this pur- which can be easily computed. At this point, one can sum spec-
pose. Morphology is a widely used approach for modeling the tral and textural dedicated kernel matrices and introduce the
True color image Spectral classification Spectralspatial classification
FIg u r e 1 2.6 The importance of using spatial and spectral information in classification.
2016 by Taylor & Francis Group, LLC
Hyperspectral Image Processing 251
Pixelwise classification MRF-based Classification
regularization map
Class probabilities
Hyperspectral image
Probabilistic pixelwise
classification
FIg u r e 1 2.7 Standard processing framework using pixel-wise probabilistic classification followed by MRF-based spatial postprocessing.
cross information between textural and spectral features in the basis vectors, while the classification criterion for a new input
formulation. This simple methodology yields a full family of new sample would be the distance from the class subspace [24].
kernel methods for hyperspectral data classification, defined in Recently, several subspace projection methods have been
[33] and implemented using the SVM classifier, thus providing a specifically designed for improving hyperspectral data charac-
composite kernel-based SVM. terization, with successful results. For instance, the subspace-
Another approach to jointly exploit spatial and spectral infor- based multinomial logistic regression followed by Markov
mation is to use Markov random fields (MRFs) for the charac- random fields (MLRsubMRF) method in Ref. [23] first performs
terization of spatial information. MRFs exploit the continuity, a learning step in which the posterior probability distributions
in probability sense, of neighboring labels [19,34]. In this regard, are modeled by a multinomial logistic regression (MLR) [37]
several techniques have exploited an MRF-based regulariza- combined with a subspace projection method. Then, the method
tion procedure, which encourages neighboring pixels to have infers an image of class labels from a posterior distribution built
the same label when performing probabilistic classification of on the learned subspace classifier and on a multilevel logistic
hyperspectral data sets. An example of this type of processing is spatial prior on the image of labels. This prior is an MRF that
given in Figure 12.7, in which a pixel-wise probabilistic classifi- exploits the continuity, in probability sense, of neighboring
cation is followed by an MRF-based spatial postprocessing that labels. The basic assumption is that, in a hyperspectral image,
refines the initial probabilistic classification output. it is very likely that two neighboring pixels will have the class
Several other approaches include spatial information as a same label. The main contribution of the MLRsubMRF method
postprocessing, that is, after a spectral-based classification has is therefore the integration of a subspace projection method with
been conducted. One of the first classifiers with spatial postpro- the MLR, which is further combined with spatialcontextual
cessing developed in the hyperspectral imaging literature was information in order to provide a good characterization of the
the well-known extraction and classification of homogeneous content of hyperspectral imagery in both spectral and the spatial
objects [10]. Another one is the strategy adopted in [35], which domains. As will be shown by our experiments, the accuracies
combines the output of a pixel-wise SVM classifier with the mor- achieved by this approach are competitive with those provided
phological watershed transformation [30] in order to provide a by many other state-of-the-art supervised classifiers for hyper-
more spatially homogeneous classification. A similar strategy spectral analysis (Figure 12.8).
is adopted in Ref. [36], in which the output of the SVM classi-
fier is combined with the segmentation result provided by the
unsupervised recursive hierarchical segmentation (RHSEG)*
12.2.4 Semisupervised classification
algorithm. A relevant challenge for supervised classification techniques is
the limited availability of labeled training samples, since their
collection generally involves expensive ground campaigns [38].
12.2.3 Subspace-Based Approaches
While the collection of labeled samples is generally difficult,
Subspace projection methods [23] have been shown to be a pow- expensive, and time-consuming, unlabeled samples can be gen-
erful class of statistical pattern classification algorithms. These erated in a much easier way. This observation has fostered the
methods can handle the high dimensionality of hyperspectral idea of adopting semisupervised learning techniques in hyper-
data by bringing it to the right subspace without losing the origi- spectral image classification. The main assumption of such tech-
nal information that allows for the separation of classes. In this niques is that new (unlabeled) training samples can be obtained
context, subspace projection methods can provide competitive from a (limited) set of available labeled samples without signifi-
advantages by separating classes that are very similar in spec- cant effort/cost. This can be simply done by selecting new sam-
tral sense, thus addressing the limitations in the classification ples from the spatial neighborhood of available labeled samples,
process due to the presence of highly mixed pixels. The idea of under the principle that it is likely that the new unlabeled sam-
applying subspace projection methods to improve classification ples will have similar class labels as the already available ones.
relies on the basic assumption that the samples within each class In contrast to supervised classification, semisupervised algo-
can approximately lie in a lower-dimensional subspace. Thus, rithms generally assume that a limited number of labeled sam-
each class may be represented by a subspace spanned by a set of ples are available a priori and then enlarge the training set using
unlabeled samples, thus allowing these approaches to address
* http://opensource.gsfc.nasa.gov/projects/HSEG/. ill-posed problems. However, in order for this strategy to work,
2016 by Taylor & Francis Group, LLC
252 Remotely Sensed Data Characterization, Classification, and Accuracies
Multinominal logistic
m(1) regression subspace classifier
(MLRsubMRF)
Modelling of spectral information
m(2)
(use a subspace probabilistic MLR
classifier tuned to deal with mixed pixels)
Class 1
Class 2
Modelling of spatial information
(use a multi-level logistic spatial prior to
m(3) enforce smoothness in the classification)
FIg u r e 1 2.8 Multinomial logistic regression subspace classifier.
several requirements need to be met. First and foremost, the the contextual information of the pixels belonging to the neigh-
new (unlabeled) samples should be obtained without significant borhood system of each training sample in the learning phase to
cost/effort. Second, the number of unlabeled samples required improve the robustness to possible mislabeled training patterns.
in order for the semisupervised classifier to perform properly In [43], two semisupervised one-class (SVM-based) approaches
should not be too high in order to avoid increasing computa- are presented in which the information provided by unlabeled
tional complexity in the classification stage. In other words, as samples present in the scene is used to improve classification
the number of unlabeled samples increases, it may be unbear- accuracy and alleviate the problem of free-parameter selection.
able for the classifier to properly exploit all the available training The first approach models data marginal distribution with the
samples due to computational issues. Further, if the unlabeled graph Laplacian built with both labeled and unlabeled samples.
samples are not properly selected, these may confuse the clas- The second approach is a modification of the SVM cost function
sifier, thus introducing significant divergence or even reducing that penalizes more the errors made when classifying samples of
the classification accuracy obtained with the initial set of labeled the target class. In [44], a new method to combine labeled and
samples. In order to address these issues, it is very important that unlabeled pixels to increase classification reliability and accuracy,
the most highly informative unlabeled samples are identified in thus addressing the sample selection bias problem, is presented
computationally efficient fashion, so that significant improve- and discussed. In [45], an SVM is trained with the linear com-
ments in classification performance can be observed without the bination of two kernels: a base kernel working only with labeled
need to use a very high number of unlabeled samples. examples is deformed by a likelihood kernel encoding similari-
The area of semisupervised learning for remote-sensing data ties between labeled and unlabeled examples and then applied in
analysis has experienced a significant evolution in recent years. the context of urban hyperspectral image classification. In [46],
For instance, looking at machine learningbased approaches, similar concepts to those addressed before are adopted using a
in Ref. [39] transductive support vector machines (TSVMs) are neural network as the baseline classifier. In [17], a semiautomatic
used to gradually search a reliable separating hyperplane (in the procedure to generate land cover maps from remote-sensing
kernel space) with a transductive process that incorporates both images using active queries is presented and discussed.
labeled and unlabeled samples in the training phase. In [40], a Last but not least, we emphasize that the techniques summa-
semisupervised method is presented that exploits the wealth rized in this section only represent a small sample (and some-
of unlabeled samples in the image and naturally gives relative how subjective selection) of the vast collection of approaches
importance to the labeled ones through a graph-based method- presented in recent years for hyperspectral image classification.
ology. In [41], kernels combining spectralspatial information For a more exhaustive summary of available techniques and
are constructed by applying spatial smoothing over the original future challenges in this area, we point interested readers to [47].
hyperspectral data and then using composite kernels in graph-
based classifiers. In [18], a semisupervised SVM is presented 12.3 experimental comparison
that exploits the wealth of unlabeled samples for regularizing
the training kernel representation locally by means of cluster In this section, we illustrate the performance of the techniques
kernels. In [42], a new semisupervised approach is presented described in the previous section by processing a widely used
that exploits unlabeled training samples (selected by means of hyperspectral data set collected by ROSIS optical sensor over the
an active selection strategy based on the entropy of the sam- urban area of the University of Pavia, Italy. The flight was oper-
ples). Here, unlabeled samples are used to improve the estima- ated by the Deutschen Zentrum for Luftund Raumfahrt (DLR,
tion of the class distributions, and the obtained classification is the German Aerospace Agency) in the framework of the HySens
refined by using a spatial multilevel logistic prior. In [16], a novel project, managed and sponsored by the European Union.
context-sensitive semisupervised SVM is presented that exploits The image size in pixels is 610 340, with very-high-spatial
2016 by Taylor & Francis Group, LLC
Hyperspectral Image Processing 253
Asphalt
Meadows
Gravel
Trees
Metal sheets
Bare soil
Bitumen
Self-blocking brick
Shadow
(a) (b) (c)
FIg u r e 12.9 The ROSIS Pavia University scene used in our experiments. The scene was collected by the ROSIS instrument in the framework
of the HySens campaign. It comprises 103 spectral bands between 0.4 and 0.9m and was collected over an urban area at the University of Pavia,
Italy. (a) ROSIS Pavia hyperspectral image. (b) Ground-truth classes. (c) Fixed training set.
resolution of 1.3 m per pixel. The number of data channels in the All these factors have made the scene a standard and an excel-
acquired image is 103 (with spectral range from 0.43 to 0.86m). lent test bed for evaluation of hyperspectral image classification
Figure12.9a shows a false color composite of the image, while algorithms, particularly those integrating the spatial and the
Figure 12.9b shows nine reference classes of interest, which spectral information.
comprise urban features, as well as soil and vegetation features. Table 12.1 illustrates the classification results obtained by
Finally, Figure 12.9c shows a fixed training set available for the different supervised classifiers for the ROSIS University of
scene, which comprises 3,921 training samples (42,776 samples Pavia scene in Figure 12.9a, using the same training data in
are available for testing). This scene has been widely used in Figure 12.9c to train the classifiers and a mutually exclusive
the hyperspectral imaging community to evaluate the perfor- set of labeled samples in Figure 12.9b to test the classifiers. As
mance of processing algorithms [6]. It represents a case study shown by Table 12.1, the SVM classifier obtained comparatively
that integrates a challenging urban classification problem, with superior performance in terms of the overall classification accu-
a data set comprising high spatial and spectral resolution, and a racy when compared with discriminant classifiers such as LDA,
highly reliable ground truth, with a well-established training set. QDA or LogDA.
TABLe 1 2.1 Classification Results Obtained for the ROSIS Pavia University Scene by the SVM Classifier as
Compared with Several Discriminant Classifiers
Class Training Testing LDA QDA LogDA SVM
Asphalt (6,631) 548 6,083 69.45 67.75 70.89 83.71
Meadows (18,649) 532 18,117 81.92 75.73 76.72 70.25
Gravel (2,099) 265 1,834 39.11 59.79 55.31 70.32
Trees (3,064) 231 2,833 95.07 96.64 96.38 97.81
Metal Sheets (1,345) 375 970 99.41 99.93 100 99.41
Bare Soil (5,029) 540 4,489 46.59 73.49 75.06 92.25
Bitumen (1,330) 392 938 63.31 93.53 83.98 81.58
Self-Blocking Bricks (3,682) 524 3,158 88.29 89.52 87.91 92.59
Shadow (947) 514 433 99.79 99.26 99.79 96.62
Overall accuracy 77.95 77.95 78.41 80.99
LDA stands for linear discriminant analysis. QDA stands for quadratic discriminant analysis. LogDA stands for logarith-
mic discriminant analysis. SVM stands for support vector machine. The numbers in the parentheses are the total number of
available samples.
2016 by Taylor & Francis Group, LLC
254 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 2.2 Classification Results Obtained for the ROSIS Pavia University Scene by the SVM Classifier as
Compared with a Composite SVM Obtained Using the Summation Kernel
Class Training Testing SVM Composite SVM
Asphalt (6,631) 548 6,083 83.71 79.85
Meadows (18,649) 532 18,117 70.25 84.76
Gravel (2,099) 265 1,834 70.32 81.87
Trees (3,064) 231 2,833 97.81 96.36
Metal Sheets (1,345) 375 970 99.41 99.37
Bare Soil (5,029) 540 4,489 92.25 93.55
Bitumen (1,330) 392 938 81.58 90.21
Self-Blocking Bricks (3,682) 524 3,158 92.59 92.81
Shadow (947) 514 433 96.62 95.35
Overall accuracy 80.99 87.18
Source: Camps-Valls, L. etal., IEEE Geosci. Remote Sens. Lett., 3, 93, 2006.
The numbers in the parentheses are the total number of available samples.
In a second experiment, we compared the standard SVM clas- the original hyperspectral image as input to the classifier. These
sifier with the composite kernel strategy as defined in Ref. [33], results confirm the importance of using spatial and spectral
which combines spatial and spectral information at the kernel information for classification purposes, as it was already found
level. After carefully evaluating all possible types of composite in the experimental results reported in Table 12.2.
kernels, the summation kernel provided the best performance In order to illustrate other approaches that use spatial infor-
in our experiments as reported in Table 12.2. This table suggests mation as postprocessing, Tables 12.4 and 12.5, respectively,
the importance of using spatial and spectral information in the compare the classification results obtained by the traditional
analysis of hyperspectral data. SVM with those found using the strategy adopted in Ref. [35],
Although the integration of spatial and spectral informa- which combines the output of a pixel-wise SVM classifier with
tion carried out by the composite kernel in Table 12.2 is per- the morphological watershed, and in Ref. [36], in which the out-
formed at the classification stage, the spatial information can put of the SVM classifier is combined with the segmentation
also be included prior to classification. For illustrative pur- result provided by the RHSEG algorithm. As shown in Tables
poses, Table12.3 compares the classification results obtained by 12.4 and 12.5, these strategies lead to improved classification
the SVM applied on the original hyperspectral image to those with regard to the traditional SVM. In fact, Tables 12.2 through
obtained using a combination of the morphological EMP for fea- 12.5 illustrate different aspects concerning the integration of
ture extraction followed by SVM for classification (EMP/SVM). spatial and spectral information. The results reported in Table
As shown in Table 12.3, the EMP/SVM provides good clas- 12.2 are obtained by integrating spatial and spectral informa-
sification results for the ROSIS University of Pavia scene, which tion at the classification stage. On the other hand, the results
represent a good improvement over the results obtained using reported in Table 12.3 are obtained by using spatial information
TABLE 12.3 Classification Results Obtained for the ROSIS Pavia University Scene by the SVM Classifier as
Compared with Those Obtained Using a Combination of the Morphological EMP for Feature Extraction
Followed by SVM for Classification (EMP/SVM)
Class Training Testing SVM EMP/SVM
Asphalt (6,631) 548 6,083 83.71 95.36
Meadows (18,649) 532 18,117 70.25 80.33
Gravel (2,099) 265 1,834 70.32 87.61
Trees (3,064) 231 2,833 97.81 98.37
Metal Sheets (1,345) 375 970 99.41 99.48
Bare Soil (5,029) 540 4,489 92.25 63.72
Bitumen (1,330) 392 938 81.58 98.87
Self-Blocking Bricks (3,682) 524 3,158 92.59 95.41
Shadow (947) 514 433 96.62 97.68
Overall accuracy 80.99 85.22
The numbers in the parentheses are the total number of available samples.
2016 by Taylor & Francis Group, LLC
Hyperspectral Image Processing 255
TABLe 1 2.4 Classification Results Obtained for the ROSIS Pavia University Scene by the SVM Classifier as
Compared with Those Obtained Using the Strategy Adopted in [35], Which Combines the Output of a Pixel-
Wise SVM Classifier with the Morphological Watershed
Class Training Testing SVM SVM + Watershed
Asphalt (6,631) 548 6,083 83.71 94.28
Meadows (18,649) 532 18,117 70.25 76.41
Gravel (2,099) 265 1,834 70.32 69.89
Trees (3,064) 231 2,833 97.81 98.30
Metal Sheets (1,345) 375 970 99.41 99.78
Bare Soil (5,029) 540 4,489 92.25 97.51
Bitumen (1,330) 392 938 81.58 97.14
Self-Blocking Bricks (3,682) 524 3,158 92.59 98.29
Shadow (947) 514 433 96.62 97.57
Overall accuracy 80.99 86.64
The numbers in the parentheses are the total number of available samples.
TABLe 1 2.5 Classification Results Obtained for the ROSIS Pavia University Scene by the SVM Classifier as
Compared with Those Obtained Using the Strategy Adopted in [36], in Which the Output of the SVM Classifier
Is Combined with the Segmentation Result Provided by the Recursive Hierarchical Segmentation (RHSEG)
Algorithm Developed by James C. Tilton at NASAs Goddard Space Flight Center
Class Training Testing SVM SVM + RHSEG
Asphalt (6,631) 548 6,083 83.71 94.77
Meadows (18,649) 532 18,117 70.25 89.32
Gravel (2,099) 265 1,834 70.32 96.14
Trees (3,064) 231 2,833 97.81 98.08
Metal Sheets (1,345) 375 970 99.41 99.82
Bare Soil (5,029) 540 4,489 92.25 99.76
Bitumen (1,330) 392 938 81.58 100
Self-Blocking Bricks (3,682) 524 3,158 92.59 99.29
Shadow (947) 514 433 96.62 96.48
Overall accuracy 80.99 93.85
The numbers in the parentheses are the total number of available samples.
at a preprocessing step prior to classification. Finally, the results the SVM classifier with a subspace-based classifier such as the
reported in Tables 12.4 and 12.5 correspond to cases in which MLRsub [23], followed by an MRF-based spatial regularizer.
spatial information is included at a postprocessing step after The idea of applying subspace projection methods relies on the
conducting spectral-based classification. As a result, the com- basic assumption that the samples within each class can approx-
parison reported in this section illustrates different scenarios imately lie in a lower-dimensional subspace. In the experiments
in which spatial and spectral information are used in comple- reported in Table 12.6, it can also be seen that spatial informa-
mentary fashion but following different strategies, that is, spatial tion (included as an MRF-based postprocessing) can be greatly
information is included at different stages of the classification beneficial in order to improve classification performance.
process (preprocessing, postprocessing, or kernel level). Finally, it is worth noting that the results discussed in the pre-
After evaluating the importance of including spatial and vious text are all based on supervised classifiers that assume the
spectral information, we now discuss the possibility to perform sufficient availability of labeled training samples. In case that no
a better modeling of the hyperspectral data by working on a sub- sufficient labeled samples are available, semisupervised learning
space. This is due to the fact that the dimensionality of the hyper- techniques can be used to generate additional unlabeled samples
spectral data is very high, and often, the data live in a subspace. (from the initial set of labeled samples) that can complement
Hence, if the proper subspace is identified prior to classifica- the available labeled samples. The unlabeled samples can also
tion, adequate results can be obtained. In order to illustrate this be used to enhance subspace-based classifiers in case dimen-
concept, Table 12.6 shows the results obtained after comparing sionality issues are found to be relevant in the considered case
2016 by Taylor & Francis Group, LLC
256 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLE 12.6 Classification Results Obtained for the ROSIS Pavia University Scene by the SVM Classifier as
Compared with a Subspace-Based Classifier Followed by Spatial Postprocessing (MLRsubMRF)
Class Training Testing SVM MLRsubMRF
Asphalt (6,631) 548 6,083 83.71 93.83
Meadows (18,649) 232 18,117 70.25 94.80
Gravel (2,099) 265 1,834 70.32 71.13
Trees (3,064) 231 2,833 97.81 92.17
Metal Sheets (1,345) 375 970 99.41 100
Bare Soil (5,029) 540 4,489 92.25 98.43
Bitumen (1,330) 392 938 81.58 99.32
Self-Blocking Bricks (3,682) 524 3,158 92.59 95.19
Shadow (947) 514 433 96.62 96.20
Overall accuracy 80.99 94.10
The numbers in the parentheses are the total number of available samples.
No Labeled data Yes
available?
Unsupervised Yes Are these samples No
techniques sufficient?
Spectral-spatial No Yes
Dimensionality
classifier
issues?
Semi-supervised Unlabeled data Subspace-based
learning techniques
Improve unlabeled data Improve unlabeled data
FIg u r e 1 2.10 Summary of contributions in hyperspectral image classification discussed in this chapter.
study. If sufficient labeled samples are available, then the use of techniques able to exploit both spatial and spectral information,
a spectralspatial classifier is generally recommended as spatial and techniques able to take advantage of a proper subspace rep-
information can provide a very important complement to the resentation of the hyperspectral data before conducting the clas-
spectral information. Finally, in case that labeled samples are sification in spatialspectral terms. These approaches represent
not available at all, unsupervised techniques need to be used a subjective selection of the wide range of techniques currently
for classification purposes. For instance, a relevant unsuper- adopted for hyperspectral image classification [3], which include
vised method successfully applied to hyperspectral image data other techniques and strategies that have not been covered in
is Tiltons RHSEG algorithm.* The different analysis scenarios detail in this chapter for space considerations. Of particular
for classification discussed in this chapter are summarized in importance is the recent role of sparse classification methods
Figure 12.10. [2,20], which are gaining significant popularity and will likely
play an important role in this research area in upcoming years.
12.4 conclusions and Future Directions Despite the wide arrange of techniques and strategies for
hyperspectral data interpretation currently available, some
In this chapter, we have provided an overview of recent advances unresolved issues still remain. For instance, the geometry of
in techniques and methods for hyperspectral image classifica- hyperspectral data is quite complex and dominated by nonlin-
tion. The array of techniques particularly illustrated in this ear structures. This issue has undoubtedly an impact in the out-
chapter comprises supervised and semisupervised approaches, come of the classification techniques discussed in this section. In
order to mitigate this, manifold learning has been proposed [48].
* http://opensource.gsfc.nasa.gov/projects/HSEG/. An important property of manifold learning is that it can model
2016 by Taylor & Francis Group, LLC
Hyperspectral Image Processing 257
and characterize the complex nonlinear structure of the data 11. L. O. Jimenez, J. L. Rivera-Medina, E. Rodriguez-Diaz,
prior to classification [49]. Another remaining issue is the very E. Arzuaga-Cruz, and M. Ramirez-Velez, Integration of
high computational complexity of some of the techniques avail- spatial and spectral information by means of unsuper-
able for classification of hyperspectral data [50]. In other words, vised extraction and classification for homogenous objects
there is a clear need to develop efficient classification techniques applied to multispectral and hyperspectral data, IEEE
that can deal with the very large dimensionality and complexity Transactions on Geoscience and Remote Sensing, 43(1),
of hyperspectral data. 844851, 2005.
12. P. Gamba, F. DellAcqua, A. Ferrari, J. A. Palmason, J. A.
Acknowledgments Benediktsson, and J. Arnasson, Exploiting spectral and spa-
tial information in hyperspectral urban data with high res-
The authors thank Prof. Paolo Gamba, the University of Pavia, olution, IEEE Geoscience and Remote Sensing Letters, 1(3),
and the HySens project for providing the ROSIS data used in the 322326, 2004.
experiments. 13. J. A. Richards and X. Jia, Remote Sensing Digital Image
Analysis: An Introduction. New York: Springer-Verlag, 2006.
References 14. G. F. Hughes, On the mean accuracy of statistical pattern
recognizers, IEEE Transactions on Information Theory,
1. A. F. H. Goetz, G. Vane, J. E. Solomon, and B. N. Rock, 14(1), 5563, 1968.
Imaging spectrometry for Earth remote sensing, Science, 15. G. Camps-Valls and L. Bruzzone, Kernel Methods for Remote
228(4704), 11471153, 1985. Sensing Data Analysis. Hoboken, NJ: Wiley, 2009.
2. J. M. Bioucas-Dias, A. Plaza, G. Camps-Valls, P. Scheunders, 16. L. Bruzzone and C. Persello, A novel context-sensitive
N. Nasrabadi, and J. Chanussot, Hyperspectral remote sens- semisupervised SVM classifier robust to mislabeled train-
ing data analysis and future challenges, IEEE Geoscience ing samples, IEEE Transactions on Geoscience and Remote
and Remote Sensing Magazine, 1(2), 636, 2013. Sensing, 47(7), 21422154, 2009.
3. M. Fauvel, Y. Tarabalka, J. A. Benediktsson, J. Chanussot, 17. J. Muoz-Mar, D. Tuia, and G. Camps-Valls, Semisupervised
and J. C. Tilton, Advances in spectral-spatial classification classification of remote sensing images with active que-
of hyperspectral images, Proceedings of the IEEE, 101(3), ries, IEEE Transactions on Geoscience and Remote Sensing,
652675, 2013. 50(10), 37513763, 2012.
4. J. M. Bioucas-Dias, A. Plaza, N. Dobigeon, M. Parente, 18. D. Tuia and G. Camps-Valls, Semisupervised remote
Q.Du, P. Gader, and J. Chanussot, Hyperspectral unmixing sensing image classification with cluster kernels, IEEE
overview: Geometrical, statistical and sparse regression- Geoscience and Remote Sensing Letters, 6(2), 224228, 2009.
based approaches, IEEE Journal of Selected Topics in Applied 19. Y. Tarabalka, M. Fauvel, J. Chanussot, and J. Benediktsson,
Earth Observations and Remote Sensing, 5(2), 354379, SVM- and MRF-based method for accurate classification of
2012. hyperspectral images, IEEE Geoscience and Remote Sensing
5. J. Chanussot, M. M. Crawford, and B.-C. Kuo, Foreword to Letters, 7(4), 736740, 2010.
the special issue on hyperspectral image and signal process- 20. B. Song, J. Li, M. Dalla Mura, P. Li, A. Plaza, J. M. Bioucas-
ing, IEEE Transactions on Geoscience and Remote Sensing, Dias, J. A. Benediktsson, and J. Chanussot, Remotely
48(11), 38713876, 2010. sensed image classification using sparse representations
6. A. Plaza, J. M. Bioucas-Dias, A. Simic, and W. J. Blackwell, of morphological attribute profiles, IEEE Transactions on
Foreword to the special issue on hyperspectral image and Geoscience and Remote Sensing, 52(8), 51225136, 2014.
signal processing, IEEE Journal of Selected Topics in Applied 21. J. Li, P. R. Marpu, A. Plaza, J. M. Bioucas-Dias, and J. A.
Earth Observations and Remote Sensing, 5(2), 347353, Benediktsson, Generalized composite kernel framework
2012. for hyperspectral image classification, IEEE Transactions on
7. N. Keshava and J. Mustard, Spectral unmixing, IEEE Signal Geoscience and Remote Sensing, 51(9), 48164829, 2013.
Processing Magazine, 19(1), 4457, 2002. 22. J. Li, J. M. Bioucas-Dias, and A. Plaza, Spectral-spatial clas-
8. W.-K. Ma, J. M. Bioucas-Dias, T.-H. Chan, N. Gillis, P.Gader, sification of hyperspectral data using loopy belief propaga-
A. Plaza, A. Ambikapathi, and C.-Y. Chi, A signal process- tion and active learning, IEEE Transactions on Geoscience
ing perspective on hyperspectral unmixing: Insights from and Remote Sensing, 51(2), 844856, 2013.
remote sensing, IEEE Signal Processing Magazine, 31(1), 23. J. Li, J. Bioucas-Dias, and A. Plaza, Spectral-spatial hyper-
6781, 2014. spectral image segmentation using subspace multino-
9. A. Plaza, J. A. Benediktsson, J. Boardman, J. Brazile, L. Bruzzone, mial logistic regression and Markov random fields, IEEE
G. Camps-Valls, J. Chanussot etal., Recent advances in tech- Transactions on Geoscience and Remote Sensing, 50(3),
niques for hyperspectral image processing, Remote Sensing of 809823, 2012.
Environment, 113(Suppl 1), 110122, 2009. 24. J. Bioucas-Dias and J. Nascimento, Hyperspectral subspace
10. D. A. Landgrebe, Signal Theory Methods in Multispectral identification, IEEE Transactions on Geoscience and Remote
Remote Sensing. Hoboken, NJ: Wiley, 2003. Sensing, 46(8), 24352445, 2008.
2016 by Taylor & Francis Group, LLC
258 Remotely Sensed Data Characterization, Classification, and Accuracies
25. I. Dopido, J. Li, P. R. Marpu, A. Plaza, J. M. Bioucas-Dias, 38. F. Bovolo, L. Bruzzone, and L. Carlin, A novel technique for
and J. A. Benediktsson, Semi-supervised self-learning for subpixel image classification based on support vector machine,
hyperspectral image classification, IEEE Transactions on IEEE Transactions on Image Processing, 19, 29832999, 2010.
Geoscience and Remote Sensing, 51(7), 40324044, 2013. 39. L. Bruzzone, M. Chi, and M. Marconcini, A novel transduc-
26. T. V. Bandos, L. Bruzzone, and G. Camps-Valls, Classification tive SVM for the semisupervised classification of remote
of hyperspectral images with regularized linear discrimi- sensing images, IEEE Transactions on Geoscience and
nant analysis, IEEE Transactions on Geoscience and Remote Remote Sensing, 11, 33633373, 2006.
Sensing, 47(3), 862873, 2009. 40. G. Camps-Valls, T. Bandos, and D. Zhou, Semi-supervised
27. F. Melgani and L. Bruzzone, Classification of hyperspec- graph-based hyperspectral image classification, IEEE
tral remote sensing images with support vector machines, Transactions on Geoscience and Remote Sensing, 45,
IEEE Transactions on Geoscience and Remote Sensing, 42(8), 30443054, 2007.
17781790, 2004. 41. S. Velasco-Forero and V. Manian, Improving hyperspec-
28. G. Camps-Valls and L. Bruzzone, Kernel-based methods tral image classification using spatial preprocessing, IEEE
for hyperspectral image classification, IEEE Transactions on Geoscience and Remote Sensing Letters, 6, 297301, 2009.
Geoscience and Remote Sensing, 43, 13511362, 2005. 42. J. Li, J. Bioucas-Dias, and A. Plaza, Semi-supervised hyper-
29. B. Schlkopf and A. Smola, Learning with Kernels: Support spectral image segmentation using multinomial logis-
Vector Machines, Regularization, Optimization and Beyond. tic regression with active learning, IEEE Transactions on
Cambridge, MA: MIT Press, 2002. Geoscience and Remote Sensing, 48(11), 40854098, 2010.
30. P. Soille, Morphological Image Analysis, Principles and 43. J. Muoz Mar, F. Bovolo, L. Gmez-Chova, L. Bruzzone,
Applications, 2nd edn. Berlin, Germany: Springer-Verlag, and G. Camp-Valls, Semisupervised one-class support
2003. vector machines for classification of remote sensing data,
31. M. Pesaresi and J. Benediktsson, A new approach for the IEEE Transactions on Geoscience and Remote Sensing, 48(8),
morphological segmentation of high-resolution satel- 31883197, 2010.
lite imagery, IEEE Transactions on Geoscience and Remote 44. L. Gmez-Chova, G. Camps-Valls, L. Bruzzone, and
Sensing, 39(2), 309320, 2001. J.Calpe-Maravilla, Mean MAP kernel methods for semis-
32. J. Benediktsson, J. Palmason, and J. Sveinsson, Classification upervised cloud classification, IEEE Transactions on
of hyperspectral data from urban areas based on extended Geoscience and Remote Sensing, 48(1), 207220, 2010.
morphological profiles, IEEE Transactions on Geoscience 45. D. Tuia and G. Camps-Valls, Urban image classifica-
and Remote Sensing, 43(3), 480491, 2005. tion with semisupervised multiscale cluster kernels, IEEE
33. G. Camps-Valls, L. Gomz-Chova, J. Muoz-Mar, J. Vila- Journal of Selected Topics in Applied Earth Observations and
Francs, and J. Calpe-Maravilla, Composite kernels for Remote Sensing, 4(1), 6574, 2011.
hyperspectral image classification, IEEE Geoscience and 46. F. Ratle, G. Camps-Valls, and J. Weston, Semisupervised
Remote Sensing Letters, 3, 9397, 2006. neural networks for efficient hyperspectral image classifica-
34. M. Khodadadzadeh, J. Li, A. Plaza, H. Ghassemian, J. tion, IEEE Transactions on Geoscience and Remote Sensing,
Bioucas-Dias, and X. Li, Spectralspatial classification 48(5), 22712282, 2010.
of hyperspectral data using local and global probabilities 47. G. Camps-Valls, D. Tuia, L. Gmez-Chova, S. Jimnez, and
for mixed pixel characterization, IEEE Transactions on J. Malo, Remote Sensing Image Processing. San Rafael, CA:
Geoscience and Remote Sensing, 52(10), 62986314, 2014. Morgan and Claypool, 2011.
35. Y. Tarabalka, J. Chanussot, and J. Benediktsson, 48. L. Ma, M. Crawford, and J. Tian, Local manifold learning
Segmentation and classification of hyperspectral images based k-nearest-neighbor for hyperspectral image clas-
using watershed transformation, Pattern Recognition, 43, sification, IEEE Transactions on Geoscience and Remote
23672379, 2010. Sensing, 48(11), 40994109, 2010.
36. Y. Tarabalka, J. A. Benediktsson, J. Chanussot, and J. C. 49. W. Kim and M. Crawford, Adaptive classification for hyper-
Tilton, Multiple spectral-spatial classification approach for spectral image data using manifold regularization kernel
hyperspectral data, IEEE Transactions on Geoscience and machines, IEEE Transactions on Geoscience Remote Sensing,
Remote Sensing, 48(11), 41224132, 2011. 48(11), 41104121, 2010.
37. D. Bhning, Multinomial logistic regression algorithm, 50. A. Plaza, J. Plaza, A. Paz, and S. Sanchez, Parallel hyper-
Annals of the Institute of Statistics and Mathematics, 44, spectral image and signal processing, IEEE Signal Processing
197200, 1992. Magazine, 28(3), 119126, 2011.
2016 by Taylor & Francis Group, LLC
13
Thermal Infrared Remote Sensing:
Principles and Theoretical
Background
Acronyms and Definitions ................................................................................................................ 259
Claudia Kuenzer 13.1 Introduction ............................................................................................................................ 259
German Remote Sensing
13.2 Principles, Theoretical Background, and Important Laws ...............................................261
Data Center, DFD of DLR
Thermal Infrared Domain Important Laws: Planck Important Laws: Stefan
Jianzhong Zhang Boltzmann Important Laws: Wien Important Laws: Kirchhoff and the Role of Emissivity
Kaufering, Germany 13.3 Potential of Diurnal and Time Series of Thermal Infrared Remote-Sensing
Data .............................................................................................................................. 268
Stefan Dech 13.4 Application Examples of Thermal Infrared Data Analyses ..............................................271
German Remote Sensing
13.5 Ground Data and Validating Thermal Infrared Data ....................................................... 272
Data Center, DFD of DLR
and 13.6 Discussion and Conclusion ................................................................................................... 272
University of Wuerzburg References.............................................................................................................................................274
Acronyms and Definitions SWIR Shortwave infrared
T Temperature
Degree TERRA Platform on which the MODIS sensor is flown
m Micrometer TIR Thermal infrared
AQUA Platform on which the MODIS sensor is flown TM Thematic Mapper
ASTER Advanced Spaceborne Thermal Emission and VIS Visible
Reflection Radiometer W Watt
ATI Apparent thermal inertia Wavelength
AVHRR Advanced Very-High-Resolution Radiometer
C Celsius 13.1 introduction
DLR German Aerospace Center
DN Digital number It is temperature that defines the habitat boundaries of plants, ani-
Emissivity mals, and humans. Temperatureamong other variablesdefines
ETM+ Enhanced Thematic Mapper Plus global, regional, and local climate and weather, the rate of sea level
K Kelvin rise, the magnitude of forest fire risk, and the reproduction condi-
kg Kilogram tions for bacteria and viruses, and it even defines our cultural pref-
LST Land surface temperature erences. Housing style, preferred means of transport, customs of
m Meter eating, and free-time behavior focusing on the inside or the outside,
MODIS Moderate-Resolution Imaging Spectroradiometer as well as the attractiveness of tourist destinations, are all defined
MSG Meteosat Second Generation by temperature. Temperature strongly impacts our well- or mal-
NASA National Aeronautics and Space Administration being, and extreme cold spills or heat waves have proven to nega-
NIR Near infrared tively impact local economies and even lead to fatalities.
nm Nanometer Remote sensing in the thermal infrared (TIR) wavelength
NOAA National Oceanic and Atmospheric Administration domain is an often neglected subdiscipline within the field of
s Second remote sensing. Reasons for this range from a limited avail-
SEVIRI Spinning Enhanced Visible and Infrared Imager ability of spaceborne sensors acquiring data in the TIR domain
SST Sea surface temperature to the, generally, lower spatial resolution of this data, when
259
2016 by Taylor & Francis Group, LLC
260 Remotely Sensed Data Characterization, Classification, and Accuracies
compared with other, for example, optical data, to last but not Several sensors, such as Landsat-5 Thematic Mapper (TM),
least different data analyses approaches, which are mandatory contain bands (detectors), which are responsive in the ther-
when dealing with TIR data. Thus, it is the goal of this chapter mal domain and record TIR radiation emitted by objects (see
to provide a comprehensive overview of the principles of remote Figure13.1). The thermal imagery then represents the kinetic
sensing in the TIR. We address the theoretic background of the temperature of objects at a certain spatial resolution. We can
TIR domain, including important physical laws; address param- differentiate between different categories of temperature.
eters such as kinetic and radiance temperature, emissivity, and Land surface temperature (LST) describes the temperature at
thermal inertia; discuss the preprocessing, the analyses, as well the land surface. It represents the temperature of objects sur-
as the validation of thermal data; and present a range of illustra- facesbe it meadows, house roofs, forest canopies, or inland
tive application examples. Much of this chapter has been refor- water bodies. Typical LST-based thermal analyses focus on
mulated based on Kuenzer and Dech (2013). LST patterns as global or continental scale, at local scale on
In general, electromagnetic radiation is emitted by all objects urban heat island (UHI) studies, assessments of forest fires,
that have a temperature above 0 K (equals 273C). Depending on grassland fires, gas flares, underground coal fires, or geo-
the temperature of the object, it has its peak of electromagnetic thermal phenomena, the observation of nuclear accident sites
emittance in a certain wavelength domain. The peak of emission (e.g., Chernobyl, Fukushima), or the monitoring of industry.
of our planet, which has an average temperature of about 300K, Contrary to LST, sea surface temperature (SST) is the tem-
is located in the TIR domain at a wavelength of about 9.7 m perature of the upper water layer of an ocean or large inland
(Sabins, 1996; Tipler, 2000). Objects on Earth absorb the suns sea (Dech etal., 1998). The SST datasets are usually exploited
incoming radiation and emit corresponding amount at longer for an improved understanding of global circulations and
wavelengths in the TIR domain. for analyses in the context of algae blooms or, for example,
a Warm
32C
km N
0 2 4 6 8 10 17C
Cold
FIg u r e 13.1 Thermal daytime land surface temperature image of Berlin, acquired on June 2, 2011, by the Landsat-5 Thematic Mapper sensor.
The subset areas depict the Berlin airport Tegel (a), the river Spree (b), and a large lake named Mggelsee and its surrounding forest (c), Water
surfaces and densely vegetated areas make up the cooler places within this daytime thermal image, while areas with a high percentage of artificial
surfaces appear warmest. Coordinates: UL: 523225N, 130959E; LR: 522220N, 134414E.
2016 by Taylor & Francis Group, LLC
Thermal Infrared Remote Sensing 261
FIg u r e 13.2 Landsat Thematic Mapper 5 daytime image subsets depicting an agricultural area in northern Italy, acquired on June 12, 2009 (a);
July 14, 2009 (b); and August 31, 2009 (c). The upper row presents true color composites in RGB, whereas the lower row presents the color-coded
Thermal infrared band 6.
thermal water pollution (seeFigure 13.2 for a handheld cam- 13.2 Principles, theoretical
era example).
However, thermal sensor data are also employed outside of
Background, and important Laws
the geosciences. In medical imaging, inflamed areas can be 13.2.1 thermal infrared Domain
detected due to their higher temperature compared with their
surroundings. In industry, machine performance is monitored Different authors define the TIR domain differently, so an over-
with thermal cameras. Architects use thermal imagery to all valid strict, physical definition does not exist. Sabins (1996)
detect energy leaks in buildings, and also police and the mili- defines the TIR range from about 3 to 13 m as a wavelength
tary use thermal cameras for object detection. Several exam- range, in which two important atmospheric windows are located
ples for the use of thermal camera imagery are illustrated by (see Figure13.5). In the 814m window, only ozone absorption
Figure 13.3. occurs, which is omitted by most sensors. In the 35 m range,
A large variety of TIR sensors data is available in data reflected sunlight can still slightly contaminate the emitted ther-
archives, as well as for current tasking (Kuenzer et al., 2013a). mal signal, so the data have to be interpreted with care. However,
Figure 13.4 presents all platforms and instruments that allow according to Lillesand and Kiefer (1994) and Lffler (1994), the TIR
for data acquisition in the TIR domain. We can see that espe- domain ranges from 3 to 1000m. Common to all authors is that
cially the fleet of Advanced Very-High-Resolution Radiometer they define TIR remote sensing as the field that deals with emitted
(AVHRR), as well as Landsat sensors, enable a long-term moni- radiation, whereas multispectral remote sensing in the visible (VIS)
toring of our planet in the TIR domain (Frey etal., 2012). Most and near-infrared (NIR) domain records reflected radiation.
TIR data are available from American sensors, and good access
is especially granted to data of AVHRR, Moderate-Resolution
13.2.2 important Laws: Planck
Imaging Spectroradiometer (MODIS), Landsat, and Advanced
Spaceborne Thermal Emission and Reflection Radiometer Plancks law, or Plancks blackbody radiation law as it is offi-
(ASTER). Also numerous Chinese sensors record data in the cially termed, describes the electromagnetic radiation emit-
TIR domain; however, here, data access is not as easy. A detailed ted by a blackbody at a given wavelength, M, as a function
overview of TIR-related instruments and their preferred appli- of the blackbodys temperature (Planck, 1900). A blackbody
cation domain can be found in Table 13.1 as well as in Kuenzer is a theoretical concept and does not exist in reality. It is an
etal. (2013a). Abbreviations can be found in the abbreviations ideal radiator, which reemits all energy it absorbs. However,
section of this book. there are surfaces on Earth that show near blackbody-like
2016 by Taylor & Francis Group, LLC
262 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) (b)
(c) (d)
FIg ur e 13.3 Thermal camera images. (a) Daytime image of a high-rise residential area in Beijing, China, acquired during a cold wave in the
winter of 2012. While the outside air and background indicate temperatures of well below 0C (up to 20C), it can be seen that heat from the
inside of the apartments penetrates through the windows, so that temperatures at the outside of the building reach about 3C, (b)daytime image
of a skyscraper faade acquired in the same residential area: the surface temperature of the faade is well below 0C (down to 11C); however,
from two windows, heat penetrates to the outside. This heat originates from air-condition systems people install privately above their windows
(inside), which are used as cooling devices in summer and heating devices in winter. Temperatures of up to 9.8C occur outside, (c) picture of an
opened freezer. Temperatures in the freezer go down to 20C. The lower part of the freezer is colder than the upper part (cold air sinks down). On
the outside of the freezer, temperatures of up to 21.5C are reached, (d) tree-lined pathway in the village of Gilching, near Munich. In the evening
around sunset, cemented surfaces, as well as vegetation, appear warmest. (Photographs: C.Kuenzer, 2012.)
behavior for certain wavelengths. The Planck formula allows T is the absolute temperature (K)
to calculate the emitted radiation, M, by inserting a certain is the wavelength (m)
wavelength as well as the bodys temperature. It also allows
deriving a blackbodys temperature, if M and the wavelength The StefanBoltzmann law (following) as well as Wiens law (fol-
are known (Equation 13.1): lowing as well) allows to calculate the total energy a theoretical
2hc 2 (13.1) blackbody radiates, as well as its wavelength of maximum emit-
M = 5 hc kT
e ( 1 ) tance (Tipler, 2000; Walker, 2008).
where
13.2.3 important Laws: StefanBoltzmann
M is the spectral radiant exitance (W m2 m1)
h is Plancks constant (6.626 1034 J s) The StefanBoltzmann enables to calculate the total energy a
c is the speed of light (2.9979246 108 m s1) theoretical blackbody radiates, as a function of its temperature.
k is Boltzmann constant (1.3806 1023 J K1) As depicted in Figure 13.6, the emitted radiation is described by
2016 by Taylor & Francis Group, LLC
Thermal Infrared Remote Sensing 263
TET-1
Elektro-L N1 - MSU-GS
Meteor-MN1 -MSU-MR
Megha- Tropiques - ScaRaB
Aquarius/SAC-D- NIRST
Suomi NPP - VIIRS
Suomi NPP - Crls
Sensors with thermal COMS - MI
GOSAT Tanso-FTS
infrared bands HJ-1B - IRMSS
(platform - instrument) FY-3A/B - MERSI
FY-3A/B - IRAS
FY-3A/B - VIRR
FY-2C/D/E/F - IVISSR
MTSAT-2 - imager
CALIPSO - IIR
Metop-A/B - lASI
MTSAT-1R - imager
NOAA-18/19, Metop-A/B - HIRS/4
Aura - TES
INSAT-3A - VHRR
Meteosat 8/9/10/11 - GERB
FY-1D - MVISR
MSG-SEVIRl - Meteosat 8/9/10
VHRR - KAPLANA-1
ASTER - Terra
AATSR - Envisat
Sounder - GOES-12-15
Imager - GOES-12-15
BIRD
IRMSS - CBERS-1/2
MODIS Terra, Aqua
ETM+ - Landsat 7
HIRS/3-NOAA15/16/17
CERESTRMM, Terra, Aqua, Suomi NPP
MVIRI - Meteosat 7
VIRSTRMM
OLS - DMSP-F8-F18
TM - Landsat 5
AVHRR/1, AVHRR/2, AVHRR/3 - TIROS-N, NOAA/B-19, MetOp A/B
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
FIg u r e 13.4 Satellite sensors with bands sensitive in the thermal infrared until (documentation until the end of 2012). The chart depicts the
sensors start and ending dates, name of the platform and the instrument, and the country of origin.
the area under the radiation curve. The larger the overall energy where
an object radiates, the higher its temperature. As Formula 13.2 max is the wavelength of maximum spectral radiant exitance (m)
elucidates this, the relationship between temperature and energy A is Wiens constant (2897.8 m K)
is not linear, but is described by a fourth power relationship (see T is the absolute kinetic temperature (K)
the following equation):
Wiens displacement law can be well demonstrated on multispectral
TRadBB = T 4
kin (13.2) remote-sensing imagery depicting areas of different temperature,
including extreme hot spots. Figure 13.7 presents several bands of
where a Landsat-5 TM image of the Etna volcano in Italy, acquired on
TRadBB is the radiant flux of a blackbody (W m) July 29, 2001. In Figure 13.7a, we can see the volcano and its main
T is the absolute kinetic temperature (K) chimney on the top. Clouds of smoke are emanating from the vol-
is StefanBoltzmann constant (5.6697 108 W m2 K4) cano. We can see areas of bare soil on the left flank of the peak, as
well as some vegetated patches. Lava flows cannot be observed. In
13.2.4 important Laws: Wien the temperature image displayed in Figure 13.7 on the lower right
(band6 of Landsat-5 TM, representing the TIR domain from 10.4
The law of Wien, also sometimes called Wiens displacement law, to 12.5 m), we can see that the smoke clouds are warm, whereas
describes the wavelength of maximum spectral exitance (inm) the flanks of the volcano are much colder. Vegetated areas appear
as a function of an objects temperature (see Equation 13.3). The coldest. At the same time, we can observe some linear structures
hotter an object, the further shifts its maximum exitance toward in the upper center of this thermal image, which do not seem to
shorter wavelengths (Tipler, 2000; Heald, 2003; Walker, 2008). belong to the smoke itself, but cannot further differentiate this.
This has also already been demonstrated in Figure 13.6. The sun Here, now band 7the shortwave infrared (SWIR) band 7 (2.09
has an average temperature of 5505C (or 5778 K), and its peak 2.35 m)comes into play. As depicted in Figure 13.7c. Here, we
of emission is located in the VIS domain of the electromagnetic can see that lava flows of extreme temperature lead to an elevated
spectrum at 0.55m. Colder objects, such as Earth, have their signal in this band 7 (Wiens law). The material is so hot that its
peak of emission in the TIR. peak of emission does not occur in the TIR domain from 10.4 to
12.5 m, but rather in the SWIR. At the same time, the clearly
A warm clouds (Figure 13.7d) appear dark in Figure 13.7c. The reason
max = (13.3)
T is the high water vapor content of the clouds. Furthermore, they are
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
264
TABLe 1 3.1 Typical Sensors and Their Characteristic Used for the Analyses of Land Surface Temperature and Related Applications
Sensor Spatial Resolution Revisit Time (Days) Swath Width Platform Agency Launch Date
ETM+ 1560 m 16 185km Landsat 7 USGS/NASA 1999
TM 30120 m 16 185km Landsat 5 USGS/NASA 1984
ASTER 1590 m 416 60km Terra NASA 1999
IRSCAM 4089 m 26 60km CBERS-3 CRESDA/INPE 2012(launch end of 2012)
TIRS 100 m 16 185km Landsat 8 (LDCM) USGS/NASA 2013(launch February 2013)
IRMSS 156 m 26 120km CBERS-2 CRESDA/INPE 2003
BIRD 185370 m 13.5 1440km BIRD DLR 2001
MODIS 2501000 m 12 2330km Terra, Aqua NASA 1999, 2002
Remotely Sensed Data Characterization, Classification, and Accuracies
MERSI 2501.1km 1 2800km FY-3A, FY-3B CMA/CNSA 2008, 2010
NIRST 350 m 7 1821000km Aquarius NASA/CONAE 2011
VIIRS 375750 m 16 3000km Suomi NPP NASA/NOAA 2011
AATSR 1000 m 35 500km Envisat ESA 2002
AVHRR/1 1.1km <1 2600km TIROS-N, NOAA6,8,10 NOAA 19781986
AVHRR/2 1.1km <1 3000km NOAA 7,9,11,12,14 NOAA 19811994
AVHRR/3 1.1km <1 2500km NOAA15-19, MetOp A,B NOAA, EUMETSAT 19982012
MSG-SEVIRI 13km <1 Full earth disk Meteosat-8/9/19 ESA/EUMETSAT 2002, 2005, 2012
MVISR 1.1km 34 3200km FY-1D CMA/NRSCC 2002
IVISSR 5km 1 Full earth disk FY-2D, FY-2E, FY-2F NRSCC/CAST/ 2006,2008,2012
NSMC-CMA
VHRR 8km <1 Full earth disk Kaplana-1, Insat-3A ISRO 2002, 2003
IASI 30km 1 2052km MetOP-A/B EUMETSAT 2006, 2012
HIRS/3 20.3km 2240km NOAA 15-17 NOAA 1998, 2000, 2002
HIRS/4 20.3km 2240km NOAA 18/19, MetOp A/B NOAA, EUMETSAT 2005, 2006, 2009, 2012
Source: Kuenzer, C., Guo, H., Ottinger; M., Dech, S., Spaceborne thermal infrared observationAn overview of most frequently used sensors for applied research, in
Kuenzer, C., and Dech, S., eds., Thermal Infrared Remote SensingSensors, Methods, Applications, Remote Sensing and Digital Image Processing Series, Vol. 17, 572 pp.,
pp. 131148, 2013a.
Thermal Infrared Remote Sensing 265
VIS reflected IR Thermal IR
H2O
O3
CO2
O3
Transmittance
CO2
CO2
H2O
5 10
Wavelength (m)
35 m 10.512.5 m
band band
FIg ur e 13.5 The diagram depicts the thermal infrared wavelength domain, typical absorption bands induced by gasses and water, and atmo-
spheric transmittance (atmospheric windows). (Modified from Kuenzer, C. and Dech, S., Theoretical background of thermal infrared remote
sensing, in Kuenzer, C. and Dech, S., eds., Thermal Infrared Remote SensingSensors, Methods, Applications, Remote Sensing and Digital Image
Processing Series, Vol. 17, 572 pp., Springer, Dordrecht, pp. 126, 2013.)
109
Visible spectral region
108
Solar radiation
K
00
107
Spectral radiant exitance (Wm2 m1)
60
106
K
2000 K Wiens displacement law
4000
105
104 1000 K
K
3000
Earths radiation
103
500 K
102
300 K
101
1
0.1 0.2 0.5 1 2 5 10 20 50 100
Wavelength (m)
FIg u r e 13.6 Blackbody radiation curves for blackbodies with different temperatures, as derived from Plancks equation. StefanBoltzmanns
equation describes the area under the curves. The rainbow-colored bar marks the VIS region. (Modified from Kuenzer, C. and Dech, S., Theoretical
background of thermal infrared remote sensing, in Kuenzer, C. and Dech, S., eds., Thermal Infrared Remote SensingSensors, Methods, Applications,
Remote Sensing and Digital Image Processing Series, Vol. 17, 572 pp., Springer, Dordrecht, pp. 126, 2013.)
not hot enough to elevate the signal in the SWIR; they do have the 13.2.5 important Laws: Kirchhoff
peak of emittance in the TIR. Creating a color composite of bands and the Role of emissivity
7, 5, and 4 now allows to beautifully depict hottest lava (orange),
warm smoke (white), former cooled lava flows (black), and sparse Plancks law describes blackbody radiation, but it has
and dense vegetation (blue). already been stated that blackbodies are fictive objects, as
most objects on our planet emit less energy than would be
2016 by Taylor & Francis Group, LLC
266 Remotely Sensed Data Characterization, Classification, and Accuracies
N N
5 km 5 km
True color composite (RGB, 3 2 1) False color composite (RGB, 7 5 4)
(a) (b)
N N
5 km 5 km
Band 7, grey scale Band 6, grey scale
(c) (d)
FIg u r e 13.7 Landsat-5 Thematic Mapper data of Etna volcano, Italy, acquired on July 29, 2001. (a) True color composite with the red, green,
and blue bands displayed in RGB; (b) false color composite with the two SWIR bands 7 and 5 and NIR band 4 displayed in RGB; (c) grayscale
image of the shortwave infrared band 7 (2.092.35 m); and (d) grayscale image of the thermal infrared band 6 (10.4012.50 m). Coordinates: UL:
375032N, 145211E; LR: 373930N, 150507E.
predicted based on their kinetic temperature. The so-called However, most solid objects (except, e.g., water and leafs) do not
emissivity coefficient, (()), is taking this fact into account transmit radiation. They are opaque, so Equation 13.5 can be
and is defined at the radiant f lux of an object at a given tem- reformulated to
perature over the radiant f lux of a blackbody at the same
temperature. A perfect blackbody would emit all radiation it () + () = 1 (13.6)
absorbed, and Kirchhoff s law states thatfor a blackbody
emittance and absorbance at a given wavelength are equal This means that an objects (blackbodys) reflectance allows
(Kirchhoff, 1860): to calculate its emittance and vice versa. Materials with low
absorb and radiate lower amounts of energy, whereas materials
() = () (13.4)
with a high absorb large amounts of incident energy and radi-
As, according to energy conservation, the sum of absorption (), ate large quantities of energy (Kirchhoff, 1860; Sabins, 1996).
reflection (), and transmission () equals 1, and considering Emissivity varies depending on surface type and wavelength,
Kirchoffs law, we can postulate that but is not temperature dependent (Becker, 1987). Table 13.2
illustrates the varying emissivities of common surfaces for the
() + () + () = 1 (13.5) wavelength region 814m (averaged).
2016 by Taylor & Francis Group, LLC
Thermal Infrared Remote Sensing 267
TABLe 1 3.2 Emissivity of Different Surfaces in the 814m Wavelength Range as Compiled from
Different Sources (Own Measurements and Lillesand etal., 2008; Sabins, 1996; Hulley etal., 2009)
Surface Emissivity at 814m
Carbon powder 0.980.99
Water 0.98
Ice 0.970.98
Plant leaves, healthy 0.960.99
Plant leaves, dry 0.880.94
Asphalt 0.96
Sand 0.93
Basalt 0.92
Wood 0.87
Granite 0.830.87
Polished metals, averaged 0.020.21
Aluminum foil 0.036
Source: Kuenzer, C. and Dech, S., Theoretical background of thermal infrared remote sensing, in Kuenzer, C.
and Dech, S. (eds.) Thermal Infrared Remote SensingSensors, Methods, Applications, Remote Sensing and Digital
Image Processing Series, Vol. 17, 572 pp., pp. 126, 2013.
As can be seen, for all real materials, emissivity is below 1. This
T(rad) = 0.051/4 * 288 K Radiometer T(rad) = 0.981/4 * 288 K
means that the radiance temperature, T(rad), measured by a thermal = 286.5 K
= 136.2 K
sensor (e.g., handheld thermal camera, airborne thermal scanner,
thermal detector on board a satellite), is always lower than the real
kinetic (surface) temperature, T(kin) (in which one would measure
with a contact thermometer), of the object. This also means that
objects with exactly the same kinetic temperature on the ground
can exhibit a very different radiant temperature. If we consider a
typical land surface with different geologic surfaces, different types
of vegetation, and differing moisture conditions and maybe even Emissivity = 0.05 Emissivity = 0.98
containing a large variety of construction materials (e.g., urban
areas), it is obvious that a remotely acquired thermal image does not
represent the true kinetic temperatures of these objects unless the Polished shiny surface Surface painted black
data are corrected for emissivity effects. This is especially crucial
for all TIR applications in urban areas as the emissivities of differ- Aluminum block
ent construction materials vary considerably. Emissivities of metals Kinetic temperature = 288 K (15C)
(e.g., aluminum, tin, copper, in some areas used as roof materi-
als) are extremely low. This will lead to the fact that temperatures FIg u r e 13.8 Emissivity strongly impacts the temperature recorded
appear much lower than the sensed temperatures of the surround- at a (remote, noncontact) sensor. In this figure modified after Sabins
(1996), a block of aluminum with a homogeneous kinetic temperature
ing objects of similar kinetic temperature (seealso Figures13.8 and
and a very low emissivity is covered with carbon-rich dark paint (on
13.9). On the contrary, objects with a very high emissivity, such as
its right side). This paint has a very different emissivity than the pol-
water and vegetated surfaces, allow for a pretty exact assessment of ished aluminum block. Although the object is 15C warm, it appears
their kinetic temperature. Theconversion of radiance temperature as 136.9C on the uncovered side and as 13.3C on the painted side
to kinetic temperature can be undertaken as follows: (calculation based on Equations 13.2 and 13.7).
T(rad) = (1/4) * T(kin) (13.7)
polished metal railing and a male hand with a wedding ring
where on one finger. The picture was acquired on a warm day in late
T(rad) is the radiance temperature summer, with an ambient outside air temperature of 22C. The
T(kin) is the kinetic temperature human hand depicts temperature values of 37C (as would be
is the emissivity expected). Note that even the veins within the hand can be seen
(white, hot lines). Also the stairs in the background appears
Also Figure 13.9 visualized this phenomenon very well, based warm. However, the handrail appears at a radiance tempera-
on an imminent example. A handheld thermal camera photo- ture of 8C (cross at image center), although it is definitely not
graph is taken. The photograph shows a concrete stairs with a that cold. However, as this picture was acquired with a standard
2016 by Taylor & Francis Group, LLC
268 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 13.9 Impact of emissivity differences on radiance temperature recorded at the sensor of a handheld thermal camera. Picture taken by
C. Kuenzer in September, 2012. (Modified from Kuenzer, C. and Dech, S., Theoretical background of thermal infrared remote sensing, in Kuenzer, C.
and Dech, S., eds., Thermal Infrared Remote SensingSensors, Methods, Applications, Remote Sensing and Digital Image Processing Series, Vol. 17,
572 pp., Springer, Dordrecht, pp. 126, 2013.)
preset emissivity value of 1 (so no emissivity correction), the (Kuenzer etal., 2008). Also, AVHRR delivers a 1km nighttime
handrail, as well as the gold ring on the persons hand, appears LST image for each spot on Earth. Sensors such as Landsat can
much colder than they actually are. be tasked to acquire thermal nighttime data on special request.
Now, imagine a remotely sensed thermal image of an area The same procedure could be undertaken with the ASTER sen-
with many metal roofs. The temperatures in the image will all sor. Also, several other TIR sensors depicted in Figure 13.4 can
appear much cooler than reality, if not corrected for this emis- acquire nighttime data upon request with data providers.
sivity effect. Therefore, thermal imagery in artificial environ- Nighttime data have the advantage that influences of solar
ments, as well as in areas with a large variety of exposed rocks uneven heating due to topography are minimized. These data are
and minerals, has to be handled with care. therefore especially suitable to detect thermal anomalies, such as
hot spots induced by forest fires, coal fires (Kuenzer etal., 2007,
13.3 Potential of Diurnal and time 2008, 2013c; Zhang etal., 2007), peat fires, industry-related hot
Series of thermal infrared spots, or geothermal phenomena. Daytime data usually reflect
Remote-Sensing Data uneven solar heating due to varying sun-sensor object geometry,
topography, and thermal inertia and therefore often hinder the
TIR data have an enormous advantage to other multispectral extraction of anomalies, or the analyses of time series of ther-
data; it can be acquired independent of the sun as an illumina- mal data. Especially in predawn, thermal nighttime data are the
tion source. This means that TIR data can be acquired during solar component least accentuated. Figure 13.10 illustrates the
the daytime, as well as during the nighttime. Many multispectral differences in thermal daytime and nighttime data based on two
sensors either automatically acquire data during the night or can subsets of MODIS TIR data acquired at 11:20 a.m. during the day-
be tasked to acquire nighttime data. MODIS, National Oceanic time as well as at 2:30 a.m. during the night. In the upper image,
and Atmospheric Administration (NOAA)-AVHRR, or Meteosat the snow in the Alps appears coldest. But also lakes, such as the
Second GenerationSpinning Enhanced Visible and Infrared clearly VIS Lake Constance in the upper center of the image or
Imager (MSG SEVIRI), for example, all acquire TIR data in an the Swiss and Italian lakes, appear in dark tones (cold). Relief
automated mode all day long. MSG SEVIRI delivers geostation- impacts can be clearly observed in the mountainous area of the
ary data for every 1530min at 3km spatial resolution in the TIR. northwestsoutheast trending chain of the EmiliaRomagna. In
MODIS (as on board the platforms TERRA and AQUA) for each the nighttime data, it is the lakes, as well as the settled and agri-
spot on Earth delivers up to four acquisitions per day at 1km spa- cultural areas, that appear warmest, while the Alpine chain and
tial resolution, which usually cover the morning, the afternoon, other elevated and forested regions appear colder.
an early nighttime images, and a predawn image. Therefore, All objects on Earths surface have their own characteristic
this sensor holds a large potential for diurnal thermal mapping diurnal temperature curve (Eastwood et al., 2011). This curve
2016 by Taylor & Francis Group, LLC
Thermal Infrared Remote Sensing 269
20C
N
9C
100 km
(a)
Warm
12C
24C
100 km
Cold
(b)
FIg u r e 13.10 MYD11_L2 product acquired over northern Italy on February 27 and 28, 2009. (a) Acquired during daytime at 11:20 a.m.,
(b) acquired at 2:30 a.m. during nighttime. Coordinates: UL: 472712N, 54548E; LR: 440574N, 123151E.
represents the temperature behavior of the object within a 24-h
I = c p k (13.8)
cycle and describes how much and how fast an object heats up
and cools down. The diurnal curve of an object depends on the
material of the object in the first placedefined by its so-called The difference between the highest and lowest temperature in
thermal inertia. But also season (sun-object geometry character- a typical diurnal temperature curve is called T (Kahle et al.,
izing strength of illumination), atmospheric disturbances, and 1976). Variations in the thermal inertia (e.g., due to moisture)
complicating the matter for land surfacesits exposure (aspect, lead to changes in T. Small Ts indicate a high thermal inertia:
slope) influence those diurnal curves. Figure 13.11 depicts diur- so a high resistance of the object against temperature changes.
nal temperature curves of dry soil/rock and water. Differences Water is a good representative for such a surface. High Ts indi-
in this example are purely based on material-related properties, cate a low thermal inertiaso little resistance of the material
described by the thermal inertia. This parameter (measured in against temperature change (Cracknell and Xue, 1996).
J/m/K/s0.5) is defined as the resistance of a material to heating Thermal inertia cannot be derived from remote-sensing data
and is the product of three factors: the energy needed to raise the directly, but the idea behind it still allows to be employed with
temperature of a material by 1C (heat capacity c) per mass unit remote-sensing data, as minimum and maximum temperature
of the substance; the density of a material, p; and the thermal of a surface within the diurnal cycle can be derived from day-
conductivity, k, of the object: time and nighttime thermal images, such as the ones presented
2016 by Taylor & Francis Group, LLC
270 Remotely Sensed Data Characterization, Classification, and Accuracies
Local sunrise Solar noon Sunset
Radiant temperature Bare soil/rock Water
T
Time lag
Midnight 6 a.m. 12 noon 6 p.m. Midnight
Time
FIg u r e 13.11 Diurnal temperature variation of water and dry soil/rock. Each object shows a distinct diurnal temperature cycle determined by
the thermal inertia of the object and the history of the incoming solar radiation. (Modified from Lillesand, T.M. etal., Remote Sensing and Image
Interpretation, 6th edn, John Wiley & Sons, New York, 768 pp., 2008.)
in Figure 13.10. T can then be calculated subtracting the night- Highest temperatures are reached on the southward-facing
time from the daytime temperature. In soils, T, for example, slope, and peak temperature here occurs more than 1 h
changes under different moisture conditions. Dry soils have a later than on the eastern slope. While peak temperature on
larger T than wet soils. Thus, already in the 1970s did Idso etal. the eastern and northern slope reaches around 35C, on the
(1975) and Schmugge (1991) employ synthetic T images for southern slope, 45C is reached. This demonstrated that one
surface soil moisture retrieval. The concept of T over time has and the same surface material canat one time stepexhibit
been extended to the concept of the so-called apparent thermal completely different temperature depending on aspect and
inertia (ATI). The ATI is defined as slope. At local Landsat overpass time of 10:30, for example,
the temperature of the same object can differ up to 10C. The
(1 A) sand dune exhibits temperature ranging from 5C to 45C
ATI = (13.9)
T over the course of the day.
Solar uneven heating effects, as well as effects of differing sun-
Here, A represents the albedo of the pixel in the VIS band. sensor-object geometries in LST imagery, thus have to be corrected
Albedo is included as dark materials absorb more sunlight than forespecially when a study focuses on thermal changedetection
light materialsso by including it here, the impact of this effect or even time series analyses (Warner and Chen, 2001).
is compensated for. As one example, Notarnicola et al. (2013) Prior to these, more complex correction TIR data of course
employed ATI data to differentiate varying stages of soil mois- have to be preprocessed like other multispectral data also. Just
ture conditions (Kuenzer etal., 2013b). However, working with like optical imagery, data have to be geo-corrected, sensor cali-
T and ATI is a difficult task as they can vary depending on an bration needs to be undertaken with constantly updated calibra-
objects bidirectional reflectance distribution function (BRDF) tion coefficients (establishing a constant relationship between the
relief-induced variations, shadowing, and impacts of wind. To radiation received at the sensor and the digital number [DN]),
compensate for relief-induced variations in T, topographic and DNs then have to be transferred in object radiance, consid-
data and information on solar elevation and azimuth need to be ering atmospheric effects as well, which is usually undertaken in
integrated into approximated corrections. atmospheric correction software tools (Vidal, 1991).
Diurnal effects of solar uneven heating were observed insitu But, not only diurnal variation is interesting to exploit in
on the ground by Zhang and Kuenzer (2007) as depicted in multiple TIR acquisitions, but also annual temperature curves
Figure 13.12. Temperatures were measured with a handheld of objects can be analyzed (see Figure 13.13). Annual T can be
radiometer at a very dense temporal interval of 10min on a sand calculated from the temperature difference of an object between
dune with slopes to the north, east, south, and west. the coldest (winter) and the warmest (summer) season.
As expected, we can see that the east-exposed slope heats To define, at which point in time and object is the warmest
up earliest and fastest in the morning (as the sun rises in the and the coldest, usually gap-free or gap-filtered annual time
east), and also the peak temperature here is reached earliest. series of daily available data are needed. In higher-resolution
2016 by Taylor & Francis Group, LLC
Thermal Infrared Remote Sensing 271
Warm
Km
0 2 4 6 8 10
N
Cold
FIg u r e 13.12 Diurnal temperature variation as measured on a sand dune, where one and the same material occurs at different aspects (thermal
anisotropy). (From Zhang, J. and Kuenzer, C. J. Appl. Geophys., 63, 117, 2007.)
data (e.g., Landsat data, which are only acquired every 16days in data during August 2004. Gaps exist in areas, which are often
the best casemeaning cloud-free conditions), the search for the cloud covered that they could not be interpolated and filled.
highest and lowest annual temperature would be a large challenge. Poles are not covered due to the near polar orbit of Envisat.
However, even here can images acquired within only a few weeks Time series of these products are freely available from ESA and
difference illustrate the strong variability of temperaturewith can be analyzed with respect to SST averages, minima, mean,
season, but also with season-related land use changes. In Figure variability, anomalies, and the representation of occurrences
13.13, we can clearly see how agricultural areas turn cooler with of, for example, El Nio in TIR-derived SST data. TIR data
the expansion of vegetation covering the underlying soil (Kant acquired over the ocean have the big advantage that scientists
etal., 2009). Urban areas usually appear warmest. do not have to deal with topographic effects of solar uneven
The strong variability of thermal data with acquisition data heating. Furthermore, the water surface is more or less homoge-
and time is often not considerednot even in studies, where it neousat least when compared with the patchy mosaics of the
would be most crucial. Numerous authors study so-called urban land surface. Furthermore, BRDF effects are minimized, and
heat islands (Streutker, 2003; Tiangco etal., 2008; Schwarz etal., the thermal inertia of water is high, and therefore temperature
2011), to assess if a citydue to increased surface sealingis get- changes take place relatively slowly.
ting hotter over time due, but do not really consider that actually Figure 13.14 presents a typical application example of TIR
a large amount of data would be needed, to eliminate the effects data analyses on land. Here, hot spots have been derived over the
of different acquisition dates and times. A comprehensive time area of the Niger Delta, Nigeria, Africa, during three time steps
series of scenes acquired during the same date and time need to and covering the time span from 1986/1987 to 2013. Oil industry
be analyzed to derive a real trend (e.g., possible with MODIS or in the Niger Delta flares enormous amounts of natural gas. This
AVHRR data). practice brings with it the harmful release of climate-relevant
and toxic gasses and substances, contributing to the severe envi-
13.4 Application examples of thermal ronmental degradation in the area. Different approaches for
infrared Data Analyses thermal anomaly extraction exist, and here an automated, mov-
ing windowbased approach (Kuenzer etal., 2007) was applied
In the field of TIR analyses, SST analyses are the furthest to extract local hot spots of different temperaturean approach
advanced (see Figure 13.13) (Iwasaki et al., 2008) and one of clearly superior to simple empirical thresholding. The illustra-
the few domains in TIR remote sensing, where services are tion of hot spot occurrences over time (in this case, gas flares)
already offeredfor example, by the European Space Agency can depict the dynamics of oil exploiting industry development
(ESA). Figure 13.13 depicts a mean product of SST derived from in the region and can support the designation of especially
Envisat Advanced Along-Track Scanning Radiometer (AATSR) threatened natural resource or communities.
2016 by Taylor & Francis Group, LLC
272 Remotely Sensed Data Characterization, Classification, and Accuracies
AATSR dual SST Aug 2004
310.00
50
300.00
Temperature/k
Latitude
0 290.00
280.00
50
270.00
100 0 100
Longitude
FIg u r e 13.13 Monthly average of sea surface temperature as derived from Envisat Advanced Along-Track Scanning Radiometer data at 1km
resolution for August 2004. (Courtesy of the European Space Agency, Noordwijk, the Netherlands.)
The range of TIR-based application studies is very broad, and Validation on past data is of course not possible, as LST varies
published studies are numerous. A comprehensive overview of from hour to hour and from day to day. Therefore, ground data
TIR remote sensing is provided in the book Thermal Infrared collection activities should be performed during the satellites
Remote Sensing: Sensors, Method, Applications (Kuenzer and overpass. As temperatures change within minutes, there is little
Dech, 2013), to which numerous authors contributed and which time to measure the on-ground surface temperatures of several
contains three parts. Sensor-related chapters focus, among oth- objects within the scenes footprint. Usually, several people with
ers, on the geometric calibration of thermographic cameras, TIR intercalibrated radiometers, contact thermometers, and thermal
spectroscopy, challenges and opportunities for unmanned aerial cameras are on the ground to measure the kinetic and radiance
vehicle-borne thermal imaging, or planned new thermal mis- temperatures of the hottest and coldest objects (see Figure 13.15).
sions, such as National Aeronautics and Space Administrations Water surfaces are a good target to establish a relationship
(NASA) Hyperspectral Thermal Emission Spectrometer, NASAs between ground-measured temperatures and satellite imagery
Hyperspectral Infrared Imager, or thermal remote sensing with derived temperatures, as they have a high thermal inertia and do
small satellites such as Bispectral and Infrared Remote Detection exhibit fast or accentuated temperature changes over time.
(BIRD), Technologie-Erprobungstrger (TET), and the next-gen- Indirect validation via Google Earth can be undertaken for
eration Berlin InfraRed Optical System (BIROS). Method-oriented selected applications. For example, the Niger Delta gas flare
chapters present cross-comparisons of daily LST products from locations, which were presented in Figure 13.13, were partially
NOAA-AVHRR and MODIS, compare the advantages and short- validated via checks in high-resolution data Google Earth. The
comings of the thermal sensors of SEVIRI and MODIS for LST high-resolution data available for most parts of the planet allow
mapping, and discuss methods for improving atmospheric cor- to put thermal anomalies detected into the right context. Gas
rection of TIR data, or for time series corrections and analyses flares, for example, are clearly visible also in high-resolution
in TIR data, or novel concepts for the derivation of SST products. optical data. If a flame is obscured by smoke, other indirect indi-
Application chapters address approaches to derive urban struc- cators are burnt vegetation, soot around the flare site, and hot
ture types and address TIR-based mineral mapping, soil moisture spots can sometimes also be validated by the numerous photo-
derivation, the assessment of vegetation fires, analyses of lava graphs available in Google Earth. The same applies for forest fire
flows, thermal analyses of volcanoes, investigations in under- occurrences or grassland fires.
ground coal fire regions, and the analyses of geothermal systems.
13.6 Discussion and conclusion
13.5 Ground Data and Validating
thermal infrared Data Thermal remote-sensing data are used for a variety of applica-
tions such as the assessment of LST, SST, and inland water body
To validate temperatures derived from TIR data, or to confirm temperature; the analyses of land coverrelated thermal pat-
phenomena or patterns observed in this data, ground data col- terns, such as UHI; the detection of hot spots and thermal ano-
lection is a common procedure undertaken (Coll et al., 2005). mies resulting from forest fires, coal fires, and industry-related
2016 by Taylor & Francis Group, LLC
Thermal Infrared Remote Sensing 273
6300N
600N
Detail map A
Detail map B
5300N
500N
4300N
N
Detail map C 0 20 40 80 km
500E 5300E 600E 6300E 700E 7300E 800E 8300E
(a) (c)
5500N
4400N
5400N
4300N
5300N
4200N
0 5 10 20 km 0 5 10 20 km
5100E 5200E 5300E 5400E 6400E 6500E 700E 7100E
5400N
(b) Legend
Flare detection dates
Not in SLC-off data gap In SLC-off data gap
1986/87 1986/87
5300N
2002/03 2002/03
2013
1986/87 and 2002/03 1986/87 and 2002/03
1986/87 and 2013
5200N
2002/03 and 2013
0 5 10 20 km
1986/87, 2002/03 and 2013
6200E 6300E 6400E 6500E
FIg u r e 13.14 Gas flare hot spots derived for different years from thermal infrared Landsat data over the course of 27years (1986/1987 until 2013).
Oil industry in the Niger Delta flares enormous amounts of natural gas. This practice brings with it the harmful release of climate-relevant and toxic
gasses and substances, contributing to the severe environmental degradation in the area. (Adapted from Kuenzer, C. etal., Appl. Geogr., 53, 354, 2014.)
2016 by Taylor & Francis Group, LLC
274 Remotely Sensed Data Characterization, Classification, and Accuracies
Thermal sensor
Coldest
object
Hottest
Handheld object
radiometer
Contact
thermometer
Water body
[object w. high thermal inertia]
FIg u r e 13.15 Ground data collection/validation of thermal infrared data and observed phenomena undertaken during satellite over-
pass. (Modified from Kuenzer, C. and Dech, S., Theoretical background of thermal infrared remote sensing, in Kuenzer, C. and Dech, S., eds.,
Thermal Infrared Remote SensingSensors, Methods, Applications, Remote Sensing and Digital Image Processing Series, Vol. 17, 572 pp., Springer,
Dordrecht, pp. 126, 2013.)
activities; and the derivation of moisture conditions via the shorter-wavelength multispectral data. Astonishingly, the new
investigation of diurnal temperature dynamics. While the fleet thermal band on board of Landsat-8, launched in early 2013,
of satellites with a thermal sensor on board is larger than often only offers 100 m spatial resolutioncompared to 60 m resolu-
assumed, there are only few sensors available, which offer free tion, which was available with Landsat-7-Enhanced Thematic
and easy access to thermal data. However, TIR data offer sev- Mapper Plus (ETM+), launched over a decade earlier in 2001. To
eral advantages. While TIR datajust like multispectral data receive up to data, suitable for monitoring purposes, the TIR LST
are affected by cloud cover, an active illumination source is not community currently has to rely on MODIS and AVHRR data, as
needed. Therefore, the data can be acquired during the daytime well as on Landsat TM and ETM+ data. The soon-to-be-launched
as well as the nighttime. Thermal data depict a direct physical Sentinel 3 satellite (foreseen for 2015), including an instrument
quantity (in K or C) and are thus easy to interpret without too continuing the AATSR time series, is eagerly expected.
much bias. Furthermore, long-term data archives of common
medium-resolution sensors with daily acquisition coverage allow
References
for the analyses as decadal, annual, and monthly means, devia-
tions, variability, and trends. Disadvantages of TIR remote sens- Becker F (1987) The impact of emissivity on the measurement
ing are the relatively low spatial resolution of the data. Currently, of land surface temperature from a satellite. International
the highest-resolution spaceborne data have a 100 m pixel size. Journal of Remote Sensing 8(10):15091522.
Thermal sensorsbe it airborne or spaceborneare more costly Coll C, Caselles V, Galve JM, Valor E, Nicls R, Snchez JM,
than, for example, VIS sensors. Thus, the TIR bandwidth is usu- Rivas R (2005) Ground measurements for the validation
ally the domain that is usually discarded if budget cuts have to be of land surface temperatures derived from AATSR and
made. At the same time, the TIR community is relatively small, MODIS data. Remote Sensing of Environment 97:288300.
and lobby voices for thermal sensors are thus not so powerful. Cracknell AP, Xue, Y (1996) Thermal inertia determination from
This chapter presented an overview of the principles and spaceA tutorial review. International Journal of Remote
theoretical background of remote sensing in the TIR domain. Sensing 17(3):431461.
We addressed data characteristics and important laws of phys- Dech SW, Tungalagsaikhan P, Preusser C, Meisner RE (1998)
ics, presented common past and up-to-date TIR instruments, Operational value-adding to AVHRR data over Europe:
discussed approaches to analyze thermal data, and presented Methods, results, and prospects. Aerospace Science and
selected application examples. Remote sensing of the TIR is an Technology 2:335346.
often neglected discipline of remote sensing. Thermal sensors Eastwood S, Le Borgne P, Pr S, Poulter D (2011) Diurnal vari-
are very expensive, and due to the longer wavelength compare ability in sea surface temperature in the Arctic. Remote
to optical and NIR data, spatial resolution is always inferior to Sensing of Environment 115(10):25942602.
2016 by Taylor & Francis Group, LLC
Thermal Infrared Remote Sensing 275
Frey C, Kuenzer C, Dech S (2012) Quantitative comparison of the Kuenzer C, Zhang J, Tetzlaff A, Dech S (2013c) Thermal infra-
operational NOAA AVHRR LST product of DLR and the red remote sensing of surface and underground coal fires.
MODIS LST product V005. International Journal of Remote In Kuenzer C, Dech S (eds.) Thermal Infrared Remote
Sensing 33(22):71657183. SensingSensors, Methods, Applications, Remote Sensing
Heald, MA (2003) Where is the Wien peak? American Journal of and Digital Image Processing Series, Vol. 17, 572 pp,
Physics 71(12):13221323. Springer, Dordrecht, pp. 429451.
Hulley GC, Hook SJ, Baldridge AM (2009) Validation of the Lillesand TM, Kiefer RW (1994) Remote Sensing and Image
North American ASTER Land Surface Emissivity Database Interpretation, 3rd edn., Wiley, New York, 748 pp.
(NAALSED) version 2.0 using pseudo-invariant sand dune Lillesand TM, Kiefer RW, Chipman JW (2008) Remote Sensing
sites. Remote Sensing of Environment 113(10):22242233. and Image Interpretation, 6th edn., John Wiley & Sons,
Idso SB, Jackson RD, Reginato RJ (1975) Detection of soil mois- NewYork, 768 pp.
ture by remote surveillance. American Scientist 63:549557. Lffler E (1994) Geographie und Fernerkundung, 3rd edn.,
Iwasaki S, Kubota M, Tomita H (2008) Inter-comparison and Teubner, Stuttgart, Germany, 251 p.
evaluation of global sea surface temperature products. Notarnicola C, Lewinska E, Temimi M, Zebisch, M (2013)
International Journal of Remote Sensing 29(21):62636280. Application of the apparent thermal inertia concept for soil
Kahle AB, Gillespie AR, Goetz AFH (1976) Thermal Inertia moisture estimation in agricultural areas. In Kuenzer C,
Imaging: A new geological mapping tool. Geophysical Dech S (eds.) Thermal Infrared Remote SensingSensors,
Research Letters 3(1):2628. Methods, Applications, Remote Sensing and Digital Image
Kant Y, Bharath BD, Mallick J, Atzberger C, Kerle N (2009) Satellite- Processing Series, Springer, the Netherlands, Vol. 17, 572 pp.
based analysis of the role of land use/land cover and vegeta- Planck M (1900) Entropie und Temperatur strahlender Wrme.
tion density on surface temperature regime of Delhi, India. Annalen der Physik 306(4):719737.
Journal of the Indian Society of Remote Sensing 37(2):201214. Sabins FF (1996) Remote Sensing, 3rd edn., Wiley, New York,
Kirchhoff G (1860) Ueber das Verhltniss zwischen dem 450 pp.
Emissionsvermgen und dem Absorptionsvermgen der Schmugge TJ, Becker F, Li ZL (1991) Spectral emissivity varia-
Krper fr Wrme und Licht. Annalen derPhysik und tions observed in airborne surface temperature measure-
Chemie (Leipzig) 109:275301. ments. Remote Sensing of Environment 35(2):95104.
Kuenzer C, Dech S (2013) Theoretical background of thermal Schwarz N, Lautenbach S, Seppelt R (2011) Exploring indicators
infrared remote sensing. In Kuenzer C and Dech S (eds.) for quantifying surface urban heat islands of European cit-
Thermal Infrared Remote SensingSensors, Methods, ies with MODIS land surface temperatures. Remote Sensing
Applications, Remote Sensing and Digital Image Processing of Environment 115(12):31753186.
Series, Vol. 17, 572 pp, Springer, Dordrecht, pp. 126. Streutker D (2003) Satellite-measured growth of the urban heat
Kuenzer C, Gessner U, Wagner W (2013b) Deriving soil moisture island of Houston, Texas. Remote Sensing of Environment
from thermal infrared satellite dataSynergies with micro- 85:282289.
wave data. In Kuenzer C, Dech S (eds.) Thermal Infrared Tiangco M, Lagmay AMF, Argete J (2008) ASTER-based
Remote SensingSensors, Methods, Applications, Remote study of the night-time urban heat island effect in Metro
Sensing and Digital Image Processing Series, Vol. 17, 572 Manila. International Journal of Remote Sensing 29(10):
pp, Springer, Dordrecht, pp. 315330. 27992818.
Kuenzer C, Guo H, Ottinger M, Dech S (2013a) Spaceborne ther- Tipler PA (2000) Physik, 3rd edn., Spektrum Akademischer,
mal infrared observationAn overview of most frequently Verlag, Germany, 1520 pp.
used sensors for applied research. In Kuenzer C, Dech S Vidal A (1991) Atmospheric and emissivity correction of land
(eds.) Thermal Infrared Remote SensingSensors, Methods, surface temperature measured from satellite using ground
Applications, Remote Sensing and Digital Image Processing measurements or satellite data. International Journal of
Series, Vol. 17, 572 pp, Springer, Dordrecht, pp. 131148. Remote Sensing 12(12):24492460.
Kuenzer C, Hecker C, Zhang J, Wessling S, Wagner W (2008) The Walker J (2008) Fundamentals of Physics, 8th edn., John Wiley &
potential of multi-diurnal MODIS thermal bands data for Sons, New York, 891 pp.
coal fire detection. International Journal of Remote Sensing Warner TA, Chen X (2001) Normalization of Landsat thermal
29:923944. imagery for the effect of solar heating and topography.
Kuenzer C, van Beijma S, Gessner U, Dech S (2014) Land sur- International Journal of Remote Sensing 22(5):773788.
face dynamics and environmental Challenges of the Niger Zhang J, Kuenzer C (2007) Thermal surface characteristics of coal
Delta, Africa: Remote sensing based analyses spanning fires 1: Results of in-situ measurements. Journal of Applied
three decades (19862013). Applied Geography 53:354368. Geophysics 63:117134.
Kuenzer C, Zhang J, Li J, Voigt S, Mehl H, Wagner W (2007) Zhang J, Kuenzer C, Tetzlaff A, Oettl D, Zhukov B, Wagner W
Detection of unknown coal fires: Synergy of coal fire risk (2007) Thermal characteristics of coal fires 2: Results of
area delineation and improved thermal anomaly extraction. measurements on simulated coal fires. Journal of Applied
International Journal of Remote Sensing 28:45614585. Geophysics 63:135147.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
14
Object-Based Image Analysis:
Evolution, History, State of
the Art, and Future Vision
Acronyms and Definitions ................................................................................................................ 277
14.1 Introduction ............................................................................................................................ 277
14.2 History of OBIA .......................................................................................................................278
Intellectual Roots Needs and Driving Forces GEOBIA Developments
14.3 OBIA: A Short Summary of the State of the Art ............................................................... 282
Segmentation Is Part of OBIA but Not Married to It Classification Complex
Thomas Blaschke Geo-Intelligence Tasks
University of Salzburg 14.4 Ongoing Developments: Influences of OBIA to Other Fields and Vice Versa .............. 284
GIScience and Remote Sensing Changing Workplace Who Uses OBIA?
Maggi Kelly
University of California, Berkeley 14.5 Concluding Perspectives........................................................................................................ 288
New Paradigm: The Need for a Common Nomenclature and Sound Methodologies Toward a
Helena Merschdorf Civilian Geo-Intelligence Epistemological and Ontological Challenges
University of Salzburg References............................................................................................................................................ 289
Acronyms and Definitions VGI Volunteered geographic information
WoS Web of science
API Application programming interface
ASTER Advanced Spaceborne Thermal Emission and 14.1 introduction
Reflection Radiometer
ECHO Extraction and classification of homogeneous Remote sensing, what it is and what it can be used for, is laid
objects out in various chapters of this comprehensive book. We may
ENVI Environment for visualizing images only state here that remote sensing has a short historywhen
GEOBIA Geographic object-based image analysis compared to traditional disciplines such as mathematics or
GIS Geographic information system physics. Contrarily, we may state that it has a long history
GIScience Geographic information science when we compare it to recent Internet-based technology like
GPL General public license social media or, closer to our field, the tracking of people and
GPS Global positioning system moving objects by means of cell phone signals. Remote sensing
ISI Institute for Scientific Information has been a domain for specialists for many years and to some
LiDAR Light detection and ranging degree it still is. Similarly, geographic information system (GIS)
NGO Nongovernmental organization has for years been a field where professionals worked on desig-
OBIA Object-based image analysis nated workstations while not being fully integrated in standard
OGC Open Geospatial Consortium corporate information technology infrastructures. The latter
PPGIS Public participation geographic information changed more than a decade ago, while for remote sensing only
system recently, one may still witness remnants of historical develop-
RGB Red, green, blue color system ments of Remote Sensing (RS)-specific hardware and software.
RS Remote sensing The dominant concept in remote sensing has been the pixel,
RSGISLib Remote sensing and GIS software library while GIS functionality has always been somehow splintered
SAGA System for automated geoscientific analyses into the raster and vector domains. Blaschke and Strobl (2001)
UAS Unmanned aerial systems provocatively raised the question Whats wrong with pixels?
277
2016 by Taylor & Francis Group, LLC
278 Remotely Sensed Data Characterization, Classification, and Accuracies
Landsat 8 SPOT 5 IKONOS-2 WorldView-2 GeoEye Pansharpened
30 m 10 m 3.3 m 1.8 m 0.5 m
FIg u r e 14.1 Objects and resolutions: OBIA methods are associated with the notion of high resolutionwhereby high has always to be seen in context.
having identified an increasing dissatisfaction with pixel- where this development will lead to in terms of applications,
by-pixel image analysis. Although this critique was not new research questions and needs in education, and training and
(Cracknell 1998; see also Blaschke and Strobl 2001; Burnett professional workforce development, and we conclude with the
and Blaschke 2003; Blaschke 2010; Blaschke et al. 2014 for a main advances and recommendations for future work.
more thorough discussion), these authors described a need for
applications beyond pixels and for specific methods and meth-
odologies that support this (Figure14.1).
14.2 History of oBiA
Over the last years, the number of applications that conceptually 14.2.1 intellectual Roots
aim for objectsstill built on the information of the underlying
pixelsrose quickly. Blaschke etal. (2014) identified a high number 14.2.1.1 conceptual Foundations
of relevant publications that usewith some degree of fuzziness The conceptual foundations of OBIA are rooted in the 1960s with
in their terminologythe concept of object-based image analysis predigital aerial photography. The spatial information found in
(OBIA). They even claim that this concept and its instantiation to digital imagery that is harnessed in the object-based approach,
a particular order of scalethe geographic as opposed to applica- for example, image texture, contextual information, pixel prox-
tions in medical imaging or cell biologyis a new paradigm in imity, and geometric attributes of features, were discussed in
remote sensing. For this level of scale and the geodomain, this par- the 1960s as possible components to yet possible automation of
adigm is then referred to by some scholars as Geographic Object- photo interpretation. In his seminal work on aerial photography
Based Image Analysis (GEOBIA), while the generic principlesthe and early remote-sensing applications, Colwell (1965) describes
multiscale segmentation and object handlingmay generically be the photo interpretation process as the act of examining photo-
called OBIA. Other sources use the more generic term of OBIA graphic images for the purpose of identifying objects and judging
when referring to the geospatial domain also, and Blaschke etal. their significance. He said that photo interpretation involves the
refer to Kuhn (1962) stating that an inconsistent use of terminology observation of the size, shape, shadow, tone, texture, pattern, and
can be expected for a new paradigm. Nevertheless, it is high time location of the features, as well as the significance of the features,
to consolidate this terminology and to support a coherent usage based largely on their interrelationships or association (Colwell
of terms and naming conventionsafter having agreed upon the 1965). His assessment of the potential for automation of an object
concepts and the conception of the overall approach. recognition process depended on the capacities of a digital scan-
This chapter therefore briefly explains OBIA methods as ner and the ability of an algorithm to assess the differences, in
used in the geospatial domain and elsewhere. We will start photographic tone, between a blob and its surroundings (Colwell
from the quest to partitioning geospatial data into meaningful 1964, 1965). Colwell was an important advisor on the Landsat 1
image objects and the needs and possibilities to assessing their mission, and his ideas on extraction of meaningful features trans-
characteristics through spatial, spectral, and temporal scale. ferred to his ambitions for the satellite missions (Colwell 1973).
At its most fundamental level, OBIA requires image segmen-
tation, attribution, classification, and the ability to query and 14.4.1.2 image Segmentation
link individual objects (aka segments) in space and time. We Image segmentation is the division of an image into differ-
will elucidate the evolution of this approach, its relatively short ent regions, each having certain properties, and it provides the
history, and its older origins. Instead of a comprehensive state- building blocks of OBIA (Blaschke 2010). The desire expressed
of-the-art analysis, we refer to the key literature and try to sum- by Colwell and others in the 1960s to more automatically delin-
marize the core concepts for the reader in an understandable eate meaningful features, objects, or blobs in his early termi-
way, with a particular emphasis on a common nomenclature, nology launched numerous approaches to image segmentation
definitions, and reporting procedures. Ultimately, we will ask that rapidly advanced in the 1980s. It is widely agreed that the
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 279
TABLe 1 4.1 Overview of Major Groups of Image Segmentation Techniques
Main Issues Strengths Weaknesses
Thresholding Threshold values are applied globally (to the Most thresholding algorithms are The results depend on the initial set of clusters
and clustering whole image) or locally (applied to computationally simple. Clustering and user values or thresholds, respectively
subregions) an image or a raster may be intuitive
for a given number of clusters
Edge detection Boundaries of object or regions under Discontinuities are identified across Edges identified by edge detection are often
consideration and edges are assumed to be the array of values studied. disconnected. To segment an object from an
closely related, since there are often sharp Particularly suited for internally image, however, one needs closed region
differences in intensity at the region relatively homogeneous objects such boundaries. Typically problematic in objects
boundaries as buildings, roads, or water bodies with high internal heterogeneity such as forests
Region growing Starting from the assumption that the Selection of the similarity criterion significantly
neighboring pixels within one region have influences the results
similar values, a similarity criterion is
defined and applied in to neighboring pixels
segmentation algorithms implemented in the OBIA software of largest boost from the availability of satellite imagery of increas-
today owe a debt to theoretical and applied work in the 1970s and ing spatial resolution such as IKONOS (14 m), QuickBird
1980s that developed and refined numerous methods for image (resolution), and OrbView (resolution) sensors (launched in
segmentation (Blaschke et al. 2004; Blaschke 2010). Early key 1999, 2001, and 2003, respectively) (Blaschke 2010). This ready
papers for the remote-sensing field include Kettig and Landgrebe availability of high-resolution multiband imagery coincided with
(1976) who presented experimental results in segmentation of increasing awareness in the remote-sensing literature that novel
Landsat 1 (ERTS-1) imagery, and McKeown et al. (1989) who methods to extract meaningful and more accurate results were
developed a knowledge-based system with image segmenta- critically needed. The business-as-usual pixel-based algorithms
tion and classification tools designed for semiautomated photo were not reliable with imagery exhibiting high local variability
interpretation of aerial photographs. Key reviews are provided and obvious spatial context (Cracknell 1998; Townshend et al.
in numerous papers (Fu and Mui 1981; Haralick and Shapiro 2000; Blaschke and Strobl 2001).
1985; Pal and Pal 1993). Building on that work, image segmen- Importantly, the software package called eCognition from
tation techniques implemented today include those focused on the company Definiens (subsequently called Definiens Earth
thresholding or clustering, edge detection, region extraction, Sciences) became commercially available from 2000. This event
and growing, and some combination of these has been explored is marked as a milestone in the emergence of a body of work on
since the 1970s (Fu and Mui 1981; Blaschke 2010) (Table 14.1). OBIA, as it was the first commercially available, object-based,
image analysis software (Flanders etal. 2003; Benz etal. 2004;
14.2.2 needs and Driving Forces Blaschke 2010), and many peer-reviewed papers during this
phase relied on the software. The eCognition software is built on
With a focus on geospatial data, OBIA has particular needs to the approach originally known as Fractal Net Evolution (Baatz
that were not anticipated by its antecedents. The OBIA meth- and Schpe 2000; Blaschke 2010) that is not easily nor often
ods were driven first by the need to more accurately map mul- described in detail in early papers that relied on the software.
tiscaled Earth features with high-spatial-resolution imagery Representative papers demonstrating the utility of the newly
such as the tree, the building, and the field. Following that, the released software from this time frame include the follow-
spatial dimension of objects (distances, pattern, neighborhoods, ing. Flanders etal. (2003) evaluated the object-based approach
and topologies) was mined for classification accuracy (e.g., Guo from eCognition software and classified forest clearings and
et al. 2007). Most recently, the OBIA field has been character- forest structure elements in British Columbia, Canada, using a
ized by discussions of object semantics within fixed or emergent Landsat-enhanced thematic mapper plus image. They found that
ontologies (Arvor et al. 2013; Yue et al. 2013) and by the need forest clearings as well as forest growth stage, water, and urban
for interoperability between OBIA and GIS and spatial model- features were classified with significantly higher accuracy than
ing frameworks (Harvey and Raskin 2011; Yue etal. 2013). The using a traditional pixel-based method. With slightly different
OBIA approach has evolved from a method of convenience to results, Dorren et al. (2003) also compared pixel- and object-
what has been called a new paradigm in remote sensing and spa- based classification of forest stands using Landsat imagery in
tial analysis (Blaschke etal. 2014). Austria. They used eCogntion for the object-based approach
and found that while the pixel-based method provided slightly
14.2.3 GeoBiA Developments better accuracies, the object-based approach was more real-
istic and better served the needs of local foresters. Benz et al.
14.2.3.1 emergence (19992003/2004) (2004) used eCognition to update urban maps (buildings, roofs,
The emergence of OBIA has been written about extensively etc.) from high-resolution (0.5 m) RGB aerial orthoimages in
elsewhere (e.g., Blaschke 2010; Blaschke etal. 2014) and had its Austria. Theirs was an early and comprehensive examination of
2016 by Taylor & Francis Group, LLC
280 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 4.2 Overview of Early Application Fields
Application Images Comparison to Pixel-Based/Findings Software
Forest clearings and forest structure Landsat-enhanced thematic mapper Significantly higher accuracy of OBIA compared to pixel eCognition
(Flanders etal. 2003) plus image based
Forests (Dorren etal. 2003) Landsat Pixel-based method provided slightly better accuracies, eCognition
butthe object-based approach was more realistic and
better served the needs of local foresters
Update urban maps (Benz etal. 2004) High-resolution (0.5 m) RGB aerial Comprehensive examination of the use of the software. eCognition
orthoimages Discussed, for example, the importance of semantic
features and uncertainties in representation
Shrub cover and rangeland Historic (19371996) scanned aerial eCognition
characteristics (Laliberte etal. 2004) photos and a contemporary
QuickBird satellite image
Correlate field-derived forest inventory IKONOS-2 imagery The strongest relationships were found for discrete eCognition
parameters and image objects land-cover types, species composition, and crown closure
(Chubey etal. 2006)
Vegetation inventory (Yu etal. 2006) Digital airborne imaging system The object-based approach outperformed the pixel-based
imagery approach
Map surface coal fires (Yan etal. 2006) ASTER image (15 m resolution) The OBIA approach yielded classifications of marked
improvement over the pixel-based approach
the use of the software, and they discussed numerous aspects of for urban features (Cleve etal. 2008), Landsat imagery, and land
the OBIA approach that are still actively discussed todayfor cover (Maxwell 2010).
example, the importance of semantic features and uncertain-
ties in representation. Laliberte etal. (2004) used a combination
14.2.3.2.2 Applications
of historic (19371996) scanned aerial photos and a contempo-
rary QuickBird satellite image to map shrub cover and range- From 2005 to 2010, there was a wide net cast around OBIA
land characteristics over time. The eCognition was critical in application areas. Table 14.3 provides an overview of the various
their workflow. Chubey et al. (2006) used eCognition to seg- application areas, which emerged over these years.
ment IKONOS-2 imagery and decision tree analysis to correlate Capturing, attributing, and understanding changing land-
field-derived forest inventory parameters and image objects for scapes continues to be a primary research area in remote sensing,
forests in Alberta, Canada. They found that the strongest rela- and the use of OBIA methods for studying and understanding
tionships were found for discrete land cover types, species com-
position, and crown closure. While much work focused on the TABLe 1 4.3 Development of OBIA Application Fields
use of eCognition for high-resolution imagery, not all work in Application Area Examples
this phase did. Many papers explored the method using Landsat
Forests Flanders etal. (2003)
imagery (e.g., Dorren etal. 2003) (Table 14.2).
Dorren etal. (2003)
Individual trees Guo etal. (2007)
14.2.3.2 establishment (20052010) De Chant etal. (2009)
Forest stands Radoux and Defourny (2007)
14.2.3.2.1 Accuracy Gergel etal. (2007)
Many papers during this time frame focused on proving the Parklands Rocchini etal. (2006)
utility of the new approach and provided comparisons between Yu etal. (2006)
OBIA and pixel-based classifiers (Yan et al. 2006; Cleve et al. Rangelands Laliberte etal. (2007)
2008; Maxwell 2010). For example, Yu et al. (2006) used high- Wetlands and other critical habitat Bock etal. (2005)
spatial-resolution digital airborne imaging system imagery Urban areas Weeks etal. (2007)
and associated topographic data of the Point Reyes National Cleve etal. (2008)
Seashore in California, United States, for a comprehensive and Durieux etal. (2008)
detailed vegetation inventory at the alliance level. The object- Land use and land cover Maxwell (2010)
based approach outperformed the pixel-based approach. Yan Public health Kelly etal. (2011)
et al. (2006) compared pixel- and object-based classification Disease vector habitats Koch etal. (2007)
of an Advanced Spaceborne Thermal Emission and Reflection Troyo etal. (2009)
Radiometer (ASTER) image (15 m resolution) to map surface Public health infrastructure Lang and Blaschke (2006)
coal fires and coal piles. The OBIA approach yielded classifica- (e.g., refugee camps)
tions of marked improvement over the pixel-based approach. Hazard vulnerability and disaster aftermath Al-Khudhairy etal. (2005)
Similar results were shown using high-resolution aerial imagery Gusella etal. (2005)
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 281
characteristics and hazard potential. Tullis et al. (2010) found
that certain land covers (e.g., forest and herbaceous cover rather
than impervious surface) benefited more from a synergy between
LiDAR and optical imagery. Image fusion has also involved mul-
tiple spatial and spectral resolutions. For example, Walsh etal.
(2008) used both QuickBird and Hyperion hyperspectral imag-
ery to map an invasive plant species in the Galapagos Islands.
The fusion of multi- and hyperspectral imagery was beneficial.
14.2.3.2.4 Software
Pixel-based classification Object-based classification
During this time frame, papers evolved from naive and some-
times simplistic use of complicated software (e.g., we used eCog-
nition to segment and classify our imagery) to more nuanced
descriptions of methodology. Editorial boards of journals with
higher impact factors (e.g., Remote Sensing of Environment)
began demanding more explanation in a method section than the
use of a software package. The success of the suite of eCognition/
Definiens software packages likely prompted rapid development
of alternative software for the OBIA workflow. Berkeley Image
Seg (Clinton et al. 2010), Visual Learning Systems Feature
Analyst extension for ArcGIS (Visual Learning Systems 2008),
System for Automated Geoscientific Analyses (SAGA) (Bhner
etal. 2006), Environment for Visualizing Images (ENVI) Feature
Extraction (Hlbling and Neubert 2008), ERDAS IMAGINEs
objective module (ERDAS 2009), and IDRISI Taigas segmenta-
tion module (Clark Labs 2009) appeared between 2006 and 2010.
Use of additional, external software, particularly for the classi-
FIg u r e 1 4.2 Object-based versus pixel-based classification. fication step of the OBIA workflow, became increasingly com-
mon. For example, many papers discuss the use of decision trees
change were increasingly popular during this period. In a com- such as classification and regression trees (CART), usually run
prehensive review article, Chen etal. (2012) presented a timely externally to a software package such as eCognition, in R (http://
overview of the main issues in remote-sensing change detec- www.r-project.org/) or See5 (Quinlan 2013), to classify objects.
tion and suggested reasons for favoring object-based change Yu etal. (2006) used this approach to map vegetation alliances
detection over pixel-based approaches. They suggested that an in a California reserve; Laliberte etal. (2007) did so with high-
object-based approach to change detection allows for multiscale resolution data over rangelands, as did Chubey etal. (2006) for
analysis to optimize the delineation of individual landscape fea- forest inventory mapping. Green and Lopez (2007) used CART
tures, it reduces spurious changes due to high spectral variability to label polygons created in eCognition for benthic habitat in
in high-spatial-resolution imagery, and the approach also allows Texas, and Stow etal. (2007) used a similar combined approach
for more meaningful ways to evaluate change (Figure 14.2). to map urban areas in Accra, Ghana. Since then, eCognition has
implemented a decision tree algorithm for classification.
Next to the commercial software mentioned, several open-
14.2.3.2.3 DataFusion source software products have been developed. While earlier
Data fusion became increasingly common during this phase. attempts may be considered to be more of an academic, not
The utility of light detection and ranging (LiDAR) data for cap- very user-friendly and not well-documented prototypical soft-
turing height, that could be used in both segmentation and clas- ware such as GeoAida (Bckner etal. 2001), recent open-source
sification, was recognized soon after LiDAR became somewhat developments aim to compete with commercial software such
operational. Pascual et al. (2008) incorporated LiDAR data to as eCognition, ERDAS, or ENVI in respect to a modern user-
help characterize forest stands and structure using OBIA in a friendly GUI and software documentation. InterIMAGE is an
complex Pinus sylvestrisdominated forest in central Spain. open-source and free-access framework for knowledge-based
Zhou and Troy (2008) used LiDAR with high-resolution digital image classification. It is based on algorithms from GeoAida and
aerial imagery to analyze and characterize the urban landscape provides a capacity for customization and extension tools. Costa
structure of Baltimore at the parcel level. Ebert et al. (2009) etal. (2010) describe the InterIMAGE system as a multiplatform
used optical, LiDAR, and digital elevation models to estimate framework, implemented for Linux and Windows operational
social vulnerability indicators through the use of physical systems (http://www.lvc.ele.puc-rio.br/projects/interimage/).
2016 by Taylor & Francis Group, LLC
282 Remotely Sensed Data Characterization, Classification, and Accuracies
A more recent development is the Geographic Data Mining The current global explosion of imagery resources at high-
Analyst. It bridges GIS and image-processing functionality and temporal and high-spatial resolution is actively changing all
includes algorithms for segmentation, feature extraction, feature aspects of the geospatial enterprise. The ways in which we
selection, classification, landscape metrics, and multitemporal acquire, store, serve, and generate information from an increas-
methods for change detection and analysis (Krting etal. 2013). ing supply of imagery across domains necessitate the contin-
Bunting et al. (2014) developed the open-source platform ued development of streamlined OBIA workflows that render
RSGISLib for data-processing techniques. Users interact with the imagery useful through geospatial semantics and shared knowl-
software through an XML script, where XML tags and attributes edge (Harvey and Raskin 2011; Blaschke etal. 2014). The time-
are used to parameterize 300 available commands. The develop- sensitive decision support tasks found in disaster response, for
ers claim that command options are easily recognizable to the example, which typically make use of rapidly acquired imagery
user because of their logical and descriptive names. Through to find targets, are often facilitated currently by human volun-
the XML interface, processing chains and batch processing are teers or distributed thinking (Zook etal. 2010). These tasks in the
supported. More recently, a Python binding has been added to future might be supported by OBIA workflows. And the acceler-
Remote Sensing and GIS Software Library (RSGISLib) allowing ated pace of geospatial work that accompanies disaster response
individual XML commands to be called as Python functions. is increasingly characteristic of science in general than it has
The software has been released under a GPL3 (General Public ever been in the past. Decisions that routinely waited for annual,
License) license and makes use of a number of other open-source seasonal, or monthly data (e.g., forest loss, peak greenness, soil
software libraries (e.g., Geospatial Data Abstraction Library water deficits) can now be made based on data at finer spatial
(GDAL)/OGR); a user guide and the source code are available at and temporal resolutions (e.g., Hansen et al. 2014). Doubtlessly,
http://www.rsgislib.org. future research within OBIA will focus on transferring imagery
quickly into comprehensive and web-enabled geographic knowl-
14.2.3.3 consolidation (Since Around 2010) edge bases to be used for decision making (Table 14.4).
Since around 2010, the field has emerged from its earlier stages
and is displaying more maturity. Blaschke etal. (2014) raise the 14.3 oBiA: A Short Summary
discussion that in some ways, this maturity suggests a label of of the State of the Art
new paradigm. From a workshop on OBIA convened at the 2012
GIScience Conference in Columbus, OH, to discuss key theoreti- This section is kept very short and aims to succinctly summarize
cal and applied aspects of the approach emerged several impor- the main findings from other state-of-the-art reviews, particu-
tant topics for the next decade: integration with GIS, semantics, larly Blaschke (2010) and Blaschke etal. (2014).
accuracy, change, standards, and learning from the past. These
themes are born out in the literature. There have also been some
14.3.1 Segmentation is Part of oBiA
important developments on the software front. For example, in
but not Married to it
2010, Trimble (a company expert in field and mobile technol-
ogy and one of the leading manufacturers of research and sur- A common denominator of OBIA applications was, and still
vey grade GPS systems) purchased Definiens Earth Sciences is, that they are built on image segmentation (see also Burnett
(Trimble Acquires Definiens Earth Sciences Business to Expand and Blaschke 2003; Benz et al. 2004; Liu et al. 2006; Hay and
its GeoSpatial Portfolio: https://www.trimble.com/news/release. Castilla 2008; Lang 2008). Image segmentation is not at all
aspx?id=061110a), with expectations that the OBIA workflow new (Haralick and Shapiro 1985; Pal and Pal 1993) but has its
would be of particular use to mobile mapping, survey, and roots in industrial image processing and was not used exten-
urban environment reconnaissance. Additionally, there has been sively in geospatial applications throughout the 1980s and 1990s
increasing use in the remote-sensing world of unmanned aerial (Blaschke etal. 2004).
systems (UAS) or drones, which provide small footprint, very Interestingly, not only independent from most of the OBIA-
high-resolution imagery (cm to meter pixel size). Once geometric related developments described in Blaschke (2010) but also
and radiometric corrections and mosaicking have been applied, triggered by the advent of high resolution satellite imagery,
these images are routinely being approached with the OBIA Aplin etal. (1999) and Aplin and Atkinson (2001) developed an
workflow. UAS provide the ability for repeated deployment for approach to segment image pixels using vector field boundaries
acquisition of multispectral imagery at high temporal resolution and to assign subpixel land cover labels to the pixel segments.
data at very high-spatial resolution. For example, Laliberte etal. Subsequently, hard per-field classification, the assignment
(2011) acquired multispectral imagery using UAS and obtained of land cover classes to fields (land cover parcels) rather than
orthorectified, radiometrically calibrated image mosaics for the pixels (Aplin et al. 1999), was achieved by grouping and ana-
purpose of rangeland vegetation classification. They relied heav- lyzing all land cover labels for all pixels and pixel segments
ily on an OBIA approach for classification of rangeland classes within each individual field. Their approach was somewhat dif-
and achieved relatively high accuracies. Castro etal. (2013) were ferent in a sense that they aimed to classify predefined objects,
able to generate weed maps early in the growing season for maize namely, fields. These developments coincided later with the
fields by using an unmanned aerial vehicle and OBIA. OBIA community when Paul Aplin and Geoff Smith organized
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 283
TABLe 1 4.4 Summary of Historic Effects and OBIA Developments
External Effects/Triggers OBIA Developments
1972: Landsat 1 and its multispectral sensor set the standard for
civilian remote-sensing applications for the next decades
Late 1970s: image segmentation techniques are developed and are Kettig and Landgrebe (1976) developed the first hybrid classification approach that
subsequently being used in image processing but not much in included neighborhood aspects
geospatial applications
Late 1999 and 2000: advent of the first two civilian 1 m resolution 1999/2000: commercialization of Definiens company and eCognition software
satellites mark a new area of high-resolution spaceborne imaging
1998/1999: commercial LiDAR systems available June 2003: July 2001 first scientific workshop on OBIA methods: FE/GIS2001: Remote sensing:
Orbview-3 high-resolution digital airborne cameras such as the New sensorsinnovative methods, Salzburg, Austria (German language)
Ultracam (Leberl and Gruber 2003) 2002: first book on OBIA in German language based on the 2001 workshop (Blaschke 2002)
20012003: first dozen papers in peer-reviewed journals
2004 onward: more high-resolution satellites, decreasing prices of 2005: First OBIA-related book for the fast developing Brazilian market
data, higher accessibility (Blaschke and Kux 2005)
2005: Google Earth raised public awareness about remote-sensing OBIA workshop at the XII Brazilian remote sensing symposium, June 2005, Goiania, Brazil
imagery and subsequently increased demand for information 2006: first OBIA conference in Salzburg, Austria
products 2007: OBIA workshop at UC Berkeley
2008: GEOBIA international conference in Calgary, Alberta, Canada
2009: Object-based landscape analysis workshop at the University of Nottingham,
United Kingdom
2010: GEOBIA international conference in Ghent, Belgium
2012: GEOBIA international conference in Rio de Janeiro, Brazil
2014: GEOBIA international conference in Thessaloniki, Greece
a symposium on object-based landscape analysis in 2009 in within the classification process in addition to their associated
Nottingham, United Kingdom, and edited a special issue in spectral information.
International Journal of Geographical Information Science (Aplin There is increasing awareness that object-based methods
and Smith 2011). make better use ofoften neglectedspatial information implicit
Although most scientists would associate OBIA with seg- within remote-sensing images. Such approaches allow for a
mentation, recent work has shown that some segmentation steps tightly coupled or even full integration with both vector- and
typically involved in OBIA research do not necessarily play a raster-based GIS. In fact, when studying the early OBIA litera-
major role, as sometimes postulated in the earlier development ture for the geospatial domain, it may be concluded that many
of OBIA. See particularly the discussion of Tiede (2014) who applications were driven by the demand for classifications,
in essence decouples OBIA from image processing and Lang which incorporate structural and functional aspects.
etal. (2010, 2014) and their work on concept-related fiat objects, One good example of a comprehensive review is the paper
geons, and on latent phenomena. by Salehi et al. (2012). They conducted recent literature and
evaluated performances in urban land cover classifications
using high-resolution imagery. They analyzed the classification
14.3.2 classification results for both pixel-based and object-based classifications. In
Blaschke and Strobl (2001) have posed the question Whats general, object-based classification outperformed pixel-based
wrong with pixels? and elucidated some shortcomings of a approaches. These authors reason that the cause for the supe-
pure per-pixel approach. This was certainly not the first time riority was the use of spatial measures and that utilizing spatial
to highlight the limitations of treating pixels individually measures significantly improved the classification performance
based on multivariate statistics. In fact, Kettig and Landgrebe particularly for impervious land cover types.
(1976) developed the first algorithm called Extraction and
Classification of Homogeneous Objects (ECHO), which at least
14.3.3 complex Geo-Intelligence tasks
partially utilizes contextual information. Based on the short his-
tory of OBIA in the section before, we may argue that around Increasingly, OBIA is used beyond simple image analysis tasks
the turn of the millennium, the quest for objects reached a new such as image classification and feature extraction from one
dimension. Particularly for high-resolution image, it seems to image or a series of images from the same sensor.
make much sense to classify segmentsrather than pixels. The Today, terabytes of data are acquired from space- and air-
segments may or may not correspond exactly to the objects of borne platforms, resulting in massive archives with incredible
desire. Burnett and Blaschke (2003) called such segments from information potential. As Hay and Blaschke (2010) argue, it is
initial delimitation steps object candidates. They already offer only recently that we have begun to mine the spatial wealth of
parameters such as size, shape, relative/absolute location, bound- these archives. These authors claim that, in essence, we are data
ary conditions, and topological relationships, which can be used rich but geospatial information poor. In most cases, data/image
2016 by Taylor & Francis Group, LLC
284 Remotely Sensed Data Characterization, Classification, and Accuracies
access is constrained by technological, national, and security representing relevant indicators in a multidimensional variable
barriers, and tools for analyzing, visualizing, comparing, and space. In fact, the geon approach creates spatially exhaustive sets
sharing these data and their extracted information are still in of units, scalable to the level of policy intervention, homogenous
their infancy. In the few years since this publication, big data in their domain-specific response, and independent from any
have fully arrived in many sciences, and this debate seems not to predefined boundaries. Despite its validity for decision making
be OBIA specific from todays point of view. and its transferability across scales and application fields, the
Furthermore, policy, legal, and remuneration issues related to delineation of geons requires further methodological research
who owns (and are responsible for) value-added products result- to assess their statistical and conceptual robustness.
ing from the original data sources, or from products that repre-
sent the culmination of many different users input (i.e., citizen 14.4 ongoing Developments:
sensors), are not well understood and still developing. Thus,
myriad opportunities exist for improved geospatial information
influences of oBiA to other
generation and exploitation. Fields and Vice Versa
OBIA has been claimed to be a subdiscipline of GIScience
14.4.1 GiScience and Remote Sensing
devoted to developing automated methods to partition remote-
sensing imagery into meaningful image objects and assessing OBIA arguably has its roots firmly in the field of remote sens-
their characteristics through scale (Hay and Castilla 2008). Its ing. Developments in remote sensing through the decades of
primary objective is the generation of geographic information the 20002010sincluding most importantly the widespread
(in GIS-ready format) from which new geo-intelligence can be availability of high-resolution imagery globally, but also from
obtained. Based on this argument, Hay and Blaschke (2010) have LiDAR and novel methods of data fusionhave continued
defined geo-intelligence as geospatial content in context. this alliance. However, this early grounding of OBIA in theo-
The final theme is intelligencereferring to geo-intelligence retical and practical aspects of remote sensing is recently being
which denotes the right (geographically referenced) information enhanced through multiple novel interactions with aspects of
(i.e., the content) in the right situation so as to satisfy a specific the GIScience field, and OBIA is poised to develop further from
query or queries within user-specified constraints (i.e., the context). new trends in GIScience.
Moreno etal. (2010) describe a geographic object-based vector Since Goodchild (1992) first coined the term GIScience, sug-
approach for cellular automata modeling to simulate land-use gesting it as a manner of dealing with the issues raised by GIS
change that incorporates the concept of a dynamic neighbor- technology by focusing on the unaddressed theoretical short-
hood. This represents a very different approach for partitioning comings of conventional GIS, the contents and borders have
a scene, compared to the commonly used OBIA segmentation constantly shifted, especially in light of recent advances in geo-
techniques, while producing a form of temporal geospatial spatial technologies, including remote sensing (Blaschke and
information with a unique heritage and attributes. Merschdorf 2014). In order to deal with the special properties of
Lang (2008) provided a more holistic perspective on an image spatial information in an era of Web 2.0 technologies, the field
analysis and the extraction of geospatial information or what he of GIScience has embraced not only classic geographical knowl-
called at this time an upcoming paradigm. He started from a edge and concepts but also increasingly incorporated approaches
review of requirements from international initiatives like Global from other disciplines such as computer science and cognitive
Monitoring of Environment and Security (now Copernicus), sciences (Blaschke and Merschdorf 2014). In turn, other disci-
and he discussed in details the concept of class modeling. Also, plines have recently discovered the potential of GIScience, uti-
such methods may need further advancement of the required lizing its tools and methodologies to serve their own needs and
adaptation of standard methods of accuracy assessment and to drastically advance the knowledge base in their own respec-
change detection. He introduced the term conditioned informa- tive fields. Such is not least the case for remote sensing, which
tion. With this term, he addresses processes that entail the cre- has experienced a drastic shift from purely pixel-based meth-
ation of new geographies as a flexible, yet statistically robust and ods of image interpretation to the identification of objects in
(user-) validated unitization of space. remotely sensed imagery by means of OBIA. Hay and Castilla
Lang etal. (2014) developed the concept of geons as a strategy to (2006) propose that OBIA is a subdiscipline of GIScience, com-
represent and analyze latent spatial phenomena across different bining a unique focus on remote sensing and GI (Hay and
geographical scales (local, national, and regional) incorporating Castilla 2006:1). In this sense, OBIA may be seen as the first in a
domain-specific expert knowledge. The authors exemplified how string of developments leading to the consolidation of GIS and
geons are generated and explored. So-called composite geons remote sensing, facilitated through the common denominator of
represent functional land-use classes, required for regional- GIScience. This implies that current and ongoing developments
planning purposes. They are created via class modeling to in the discipline of GIScience may bare a significant impact on
translate interpretation schemes from mapping keys. Integrated the field of remote sensing. Such developments include but are
geons, on the other hand, address abstract, yet policy-relevant not limited to volunteered geographic information (VGI), ubiq-
phenomena such as societal vulnerability to hazards. They are uitous sensing, indoor sensing, and the integration of in situ
delineated by regionalizing continuous geospatial data sets measurements with classic remote-sensing datasets.
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 285
Web 2.0 technologies have had a significant impact on location and values measured there. Such data are particularly
GIScience, as they have enabled the bidirectional and participa- valuable in the era of very high-resolution satellite imagery, as
tory use of the Internet (Blaschke and Merschdorf 2014). These well as the subsequent surge of urban remote-sensing applica-
technologies go beyond GIS-centered assemblages of hardware, tions, such as the mapping of megacities, the monitoring of fast-
software, and functionalities. Wiki-like collective mapping envi- expanding settlements in developing countries, or the routine
ronments, geovisualization APIs, and geotagging may either be monitoring of informal settlements, conducted either by pub-
based on GIS or they have common denominators in the digital lic administration or by commercial companies, as outlined by
storage, retrieval, and visualization of information based upon Blaschke etal. (2011). Based on an extensive literature review,
its geographic content (Sheppard 2006). Blaschke et al. (2011) conclude that the increased availability
These developments have led to an influx of spatial content, of high-resolution satellite imagery has resulted in a greater
contributed by individual users or groups of users, which now- demand for timely urban mapping and monitoring. However,
adays composes a valuable data source in GIS. Such content remotely sensed imagery, which provides the basis for urban
has been termed as volunteered geographic information, by mapping applications, can only provide the birds-eye view of
Goodchild (2007), and Atzmanstorfer and Blaschke (2013) claim a given location, neglecting ground information such as the
that its full realm of possibilities, in terms of citizens partaking building facades or interiors. With the advent of widely applied
in planning initiatives, yet remains unknown. VGI is not only Open Geospatial Consortium (OGC) standards, in situ mea-
limited to online applications such as the provision of geotagged surement data recorded at ground locations can be integrated
photographs on the photo management service Flickr or geolo- with the remote-sensing imagery, providing a more holistic
cated messages on the online messaging portal Twitter but also approach to urban-mapping applications (Blaschke etal. 2011).
includes the information collected by wireless sensors on com- Blaschke et al. (2011) note that although remote sensing and
mon mobile devices. Due to the proliferation of wireless sensors insitu measurements are currently two separate technologies,
in all sorts of mobile devices, sensory data collection is no longer the strengths of both can be combined by means of sensor webs
constrained to few experts equipped with expensive sensors but and OGC standards, potentially producing new and meaning-
rather has shifted more into the lay domain. In GIScience, this ful information (Blaschke etal. 2011). They conclude that while
notion is referred to as ubiquitous sensing and can be used for available information will always be incomplete, decision mak-
monitoring activities and locations of users, or groups of users, ers can be better informed through such technology integra-
in near real time. The near real-time capabilities of ubiquitous tion, even if loosely coupled (Blaschke etal. 2011:1768).
sensing can assist decision makers in a variety of applications, Another trend enabled by the recent advances in mobile tech-
such as emergency response, public safety, traffic management, nology is the concept of indoor sensing, sometimes referred to
environmental monitoring, or public health (Resch 2013). For as indoor geography (Blaschke and Merschdorf 2014). Naturally,
example, Sagl etal. (2012) utilize the movements of cell phones remote-sensing imagery can only provide a planar view of the
between pairs of radio cellstermed as handoversin order to Earths surface, including natural features, as well as human
analyze spatiotemporal urban mobility patterns and demon- infrastructure. While LiDAR technology complements the
strate how mobile phone data can be utilized to analyze patterns classic 2D imagery with the added dimension of depth, it still
of real-world events using the example of a soccer match, while doesnt provide any insight as to the contents of buildings. In this
Zook etal. (2010) present how a mash-up of various data sources, sense, indoor sensing may be a future trend in indoor position-
including both government data and VGI, significantly contrib- ing and mapping, whereby sensor fusion will evolve to support
uted to disaster relief in Haiti, following the earthquake in 2010. indoor locations, paving the way for geoenabled manufacturing
While VGI is oftentimes a passive by-product, resulting from (Blaschke and Merschdorf 2014).
the use of Web 2.0 technologies and mobile-computing devices,
millions of internet users can nowadays choose to actively utilize
14.4.2 changing Workplace
GIS methodologies and applications by means of public partici-
pation geographic information system (PPGIS). Manifestations In the past, remote sensing and GIS were distinctly separated
of such participation can, for instance, be found in the wide- disciplines, whereby remotely sensed imagery was primarily
spread community of users contributing to virtual globes and considered as a data source for GIS (Jensen 1996). However, in
maps by superimposing new layers, such as street networks or light of more recent technical and theoretical advancements,
landmarks, or even in disaster relief efforts such as the recent these disciplines have begun to consolidate, not least attributed
search for the debris of the missing Malaysian Airline flight to the quest for tangible objects. The emergence of OBIA as a
MH370, which was assisted by tens of thousands of Internet subdiscipline of GIScience laid a foundation for the use of shared
users, who helped in sifting through the vast magnitude of sat- methodologies, and remote sensing was recognized as one ele-
ellite data recorded during the time frame in question. ment of an integrated GIS environment, rather than simply
The contribution of the general public, be it actively by an important data source (Malczewski 1999:20). The bidirec-
uploading data to virtual globes or maps or passively by utilizing tional nature of the relationship between remote sensing and
social media platforms such as Twitter or Flickr, has also fuelled GIS implies that not only advances in remote sensing technol-
the collection of insitu data, such as photos taken at a certain ogy influence the GIS environment but also vice versa. In this
2016 by Taylor & Francis Group, LLC
286 Remotely Sensed Data Characterization, Classification, and Accuracies
sense, we can witness the impact of recent trends in GIScience, geographic object-based image analysis, (GEOBIA), contained
described in Section 14.4.1, on the remote-sensing discipline. in the title, abstract, or keywords, returns a total of 451 articles
Especially, the technological advances brought about by the (April 17, 2014). When analyzing which countries the publica-
Web 2.0, such as VGI, ubiquitous sensing, or PPGIS, call for new tions primarily come from, we determined that the highest num-
approaches of data integration, with the primary aim of devel- ber of publications is contributed by the United States, accounting
oping more comprehensive and accurate datasets. Such integra- for 24% of all publications; followed by the Peoples Republic of
tion can complement the birds-eye view perspective offered by China with a 14% contribution; Germany contributing 12%;
remotely sensed imagery, with insitu information, which in turn Austria 8%; Canada 7%; Australia, Brazil, and Netherlands 6%,
can more efficiently represent dynamic urban environments respectively; and Italy and Spain with 4% each, just to name the
(Blaschke etal. 2011). To this end, OGC standards can provide top 10 contributing countries. This shows that while the United
the necessary interface for data integration, as is the case for the States is the main contributor, accounting for nearly a quarter
Global Earth Observing System of Systems, which seamlessly of all publications returned in the search, many other smaller
integrates remotely sensed imagery with insitu measurements. countries also make a noteworthy contribution. In particularly
One particular example of an OBIA application as a substitute remarkable is the 8% contribution made by Austria, which has
for GIS overlay is provided by Tiede (2014). GIS-overlay rou- only a fraction of the population (approx. 8.5 million) compared
tines usually build on relatively simple data models. Topology to most other countries represented within the top 10. Compared
isif at allcalculated on the fly for very specific tasks only. If, to the leading countryUnited StatesAustria has merely 2.7%
for example, a change comparison is conducted between two or of the population but has 33% as many publications. Such a com-
more polygon layers, the result leads mostly to a complete and parison becomes even more extreme when made with China,
also very complex fromto class intersection. Additional pro- the second largest contributor, whereby Austria has only 0.6%
cessing steps need to be performed to arrive at aggregated and as many inhabitants but accounts for 57% as many publications.
meaningful results. To overcome this problem, Tiede (2014) pre- This shows that there may be certain research clusters in certain
sented an automated geospatial overlay method in a topologically countries, which largely contribute to OBIA/GEOBIA research,
enabled (multiscale) framework. The implementation works with rather than all countries contributing relatively to their popula-
polygon and raster layers and uses a multiscale vector/raster data tion (Figure 14.3).
model developed in the OBIA software eCognition. Advantages A further analysis consisting of the research areas contributing
are the use of the software inherent topological relationships in to OBIA/GEOBIA reveals that the largest contribution is made
an object-by-object comparison, addressing some of the basic by remote sensing, accounting for 61% of all publications. The
concepts of object-oriented data modeling such as classification, second largest contributor, namely, imaging science, accounts
generalization, and aggregation. Results can easily be aggregated for only 31%, followed by geology with a share of 27%. A full
to a change-detection layer; change dependencies and the defini- chart of the top 10 contributing fields is depicted in Figure14.4.
tion of different change classes are interactively possible through When assessing the publication years, it is notable that the num-
the use of a class hierarchy and its inheritance (parentchild class ber of publications on the topic of OBIA/GEOBIA has drastically
relationships). The author demonstrates the flexibility and trans- increased over the last 5years, whereby 22% were written in 2013
ferability of change comparison for Corine Land Cover data sets. alone, as compared to >16% prior to 2008. The first OBIA publica-
This is only one example where OBIA and GIS are fully integrated, tion indexed in ISIs WoS database dates back to 1985, preceding
and although this case may be being an exception so far, one field the second OBIA publication by 10years, and at least 20years prior
may jeopardize the other field if the fields are seen isolated. to a steady incline in the number of publications (Figure 14.5).
When the same search is conducted in the Scopus database
(same phrases searched for in title, abstract, and keywords), a
14.4.3 Who Uses oBiA?
total of 586 publications are returned (April 17, 2014). The dis-
In a recent publication, Blaschke etal. (2014) found an increasing crepancy in terms of numbers of publications as compared to the
number of publications concerned with OBIA in peer-reviewed WoS database can be attributed to the fact that Scopus contains
journals, special issues, books, and book chapters and concluded a broader range of document types, such as notes, short sur-
that OBIA is a new evolving paradigm in remote sensing and to veys, and in press articles, while the WoS database only contains
some degree in GIScience also. However, they also noted that the peer-reviewed journal articles, conference proceedings, reviews,
exact terminology used within these publications is distinctly and editorials, all of which are additionally included in Scopus.
ambiguous, as is characteristic for an emerging multidisciplinary Although including a slightly greater number in overall publi-
field (Blaschke etal. 2014). Therefore, we herein aim to review the cations, the trends revealed in the Scopus data are largely in line
literature databases of the ISIs (Institute for Scientific Information) with those depicted in the WoS data. Some discrepancies were
Web of Science (WoS), as well as Scopus, in an attempt to quantify found in terms of research areas, which, however, may largely be
who uses OBIA, both in terms of countries of origin and contribut- down to the different naming conventions utilized by both data-
ing field, and to track its presence in literature over the past years. bases (e.g., the top contributing discipline to OBIA/GEOBIA in
A search in the WoS database for the phrases object-based the Scopus database is Earth and Planetary Sciences, with a total
image analysis or object-oriented image analysis, or OBIA, or of 315 publications or 54%, which corresponds to the largest WoS
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 287
Top 10 contributing countries to OBIA/GEOBIA literature in the
ISI Web of Science database
Spain
Italy
Netherlands
Country of origin Brazil
Australia
Canada
Austria
Germany
China
USA
0 20 40 60 80 100 120
Total number of publications
FIg u r e 1 4.3 Top 10 contributing countries to the OBIA/GEOBIA literature in the Web of Science database and their respective contributions.
Top 10 research areas which OBIA/GEOBIA publications are assigned to
Geography
Optics
Geochemistry/Geophysics
Computer science
Research area
Engineering
Environmental sciences
Physical geography
Geology
Imaging science
Remote sensing
0 50 100 150 200 250 300
Total number of publications*
FIg u r e 14.4 Main research areas for OBIA/GEOBIA publications in ISIs Web of Science database. *The total numbers add up to more than the
total of 451 publications due to the fact that some multidisciplinary publications may have been assigned to more than one research area.
Number of OBIA/GEOBIA publications by year from 20002013
120
Total number of publications
100
80
60
40
20
0
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Publication year
FIg u r e 1 4.5 Number of OBIA/GEOBIA publications by publication year from 2000 to 2014, as indexed in ISIs Web of Knowledge database.
2016 by Taylor & Francis Group, LLC
288 Remotely Sensed Data Characterization, Classification, and Accuracies
Major EO-satellite
launch
Overall
Software
development
Major publication
OBIA Associations
OBIA conference
Others
Publications per year
Scientific articles
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014
OBIA workshop GEOBIA 3rd GEOBIA
IKONOS eCognition Blaschke & Strobl Orbview SAGA software
Berkeley Book release conference, Ghent
Definiens 1st OBIA, 2nd GEOBIA 4th GEOBIA 5th GEOBIA
Quikbird Feature analyst WorldView-2
commercialization conference, Salzburg conference, Calgary conference, Rio de Janeiro conference, Thessaloniki
FIg u r e 1 4.6 Milestone timeline of OBIA/GEOBIA development from the late 1990s until today.
contributor of Remote Sensing). Furthermore, both the publica- contributed by ordinary people for any number of purposes,
tion year timeline and the contributing countries roughly corre- including emergency response, mobilizing activist efforts, mon-
spond to the results obtained from the analyses of the WoS data. itoring environmental change, filling gaps in existing spatial
In conclusion, when analyzing the literature, and some key databases, or identifying and addressing needs and problems in
milestone events and publications, the rise of OBIA/GEOBIA can urban neighborhoods.
be clearly traced through the course of the last decade and a half. OBIA has developed a rich array of approaches for grappling
This is depicted in the timeline shown in Figure 14.6, which exem- with the challenges associated with high-resolution data. One
plifies how both technological and methodological advances gave remaining task is to standardize terms across methods and
birth to object-oriented approaches and, according to Blaschke methodologies being used. While Blaschke et al. (2014) argue
etal. (2014), to a new paradigm in remote sensing although it must that this is very common for a new paradigm, it is nevertheless
be clearly stated that in absolute terms classic per-pixel methods troublesome. OBIA needs to urgently harmonize and streamline
are represented way more in publications at the moment. the terms being used. Otherwise, a widespread recognition from
other fields may be hindered.
14.5 concluding Perspectives
14.5.2 toward a civilian Geo-intelligence
14.5.1 new Paradigm: the need for a common
We do not exactly know how the future will look like. One pos-
nomenclature and Sound Methodologies
sible development can best be illustrated by the power and the
OBIA has certainly arrived at the sciences. While the first innovative potential of the object-by-object change comparison
years of the development were characterized by a lack of high- framework (Tiede 2014) described before. This framework yields
quality peer-reviewed scientific publications, the last few years flexible, transferable, and highly complex change comparisons
witnessed a sharp increase in such articles. Some of them are that can be visualized or calculated and aggregated to higher
remarkably highly cited such as the review paper by Blaschke level composite objects. Here, the geon concept of Lang et al.
(2010). Much of the excitement about this new methodology or (2014) comes into play: as briefly described, it also allows for the
paradigm has to do with the increasing availability of high-res- creation of more conceptual objects that may represent latent
olution datasets, which can now be used to produce informa- phenomena, which are not directly mappable.
tion and, in particular, information on demand or conditioned Hay and Blaschke (2010) suggested the term (civilian) geo-
information. Some predict that researchers, policy makers, intelligence. Since then, the number of technical developments
citizen groups, and private institutions might use information and the number of documented applications, which may support
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 289
the hypothesis of locational intelligence, have clearly grown. As more or less the only subject of studies and, more importantly, as
discussed earlier, Lang (2008) laid some theoretical foundations long as objects of interest are smaller than or similar in size com-
for the concept of conditioned information, and Lang etal. (2014) pared to the pixels, such questions may not be urgent. With the
developed the concept of geons, which may also serve as units to advent of high-resolution imagery, the question Whats wrong
characterize and delimitate latent phenomena. with pixels? (Blaschke and Strobl 2001) is valid to be asked. In
An area for future research emerges from a wider set of organi- fact, concerns about the appropriate use of technology in the
zational changes within the software industry such as the software application of remote-sensing data suggest that nonexpert users
as a service paradigm. This is a significant development in the involved in interpretation tasks may gain a relatively sophis-
organization and deployment of remote-sensing image analysis ticated understanding not just of what the technology can do
for the professional and advanced users. It may also open oppor- but of the processes involved in visualizing and disseminating
tunities for nonexpert users in remote sensing in general and for findings via interactive representations and WebGIS.
OBIA in particular. Tiede etal. (2012) presented an OBIA geopro- We refer to Pickles (2004) who contends that the contingent
cessing service that integrates OBIA methods into a geoprocess- nature of technical outcomes from GIS use is often overlooked,
ing service. This development wasto our best knowledgethe and the exploitation of some groups, particularly those with less
first integration of an eCognition-based OBIA application into an access to technology, becomes a real possibility. He also empha-
interactive WebGIS geoprocessing environment. sizes how important it is to study maps in human terms, to
Interestingly, the emergence of OBIA has not been generat- unmask their hidden agendas, to describe and account for their
ing a substantial quantity of critical reflection neither about the social embeddedness and the way they function as microphys-
technology as such nor about the wider scientific and technolog- ics (Pickles 2004, p. 181).
ical implications of this paradigm for various user groups, both Lastly, we may call for a relaxation of a potential friction
geographically and seen along an educational ladder (students between OBIA and per-pixel approaches. There are dozens,
graduatesprofessionals in private industry and academia). most likely more than a hundred, of scientific papers that com-
Another future research area concerns the remote sensing and pare both methods. Nevertheless, the future may not be domi-
GIScience practices of nonprofessional actors, not as outsourced nated by an either-or question. Rather, we should be cautious
operatives for research institutions but as private actors or NGOs about abandoning too hastily the concepts and terminologies of
following their own agendas. In remote sensing in general, much the old paradigm with reference to its dazzling object of cogni-
more than in GIScience, the vast majority of existing literature tion in this debatethe pixel. The pixel is a technical construct
investigates widely agreed scientific or commercially interesting that may be useful in many cases from a technical, that is, data
problems and reflects both the focus of an Anglophone research acquisition, point of view but sometimes also as a cognitional
community looking primarily back in time and a focus primar- prerogative. In this sense, the aforementioned question whats
ily on activities in the global North by state actors. Although we wrong with pixels (Blaschke and Strobl 2001) may appear in a
did not carry out a severe literature study, we may speculate that less unfavorable lightfor the latter, the pixels.
OBIA researchers may be a little bit less Anglophone dominated
than the general remote-sensing community. References
Aplin, P., Atkinson, P. M., and Curran, P. J. (1999). Fine spatial
14.5.3 epistemological and
resolution simulated satellite sensor imagery for land cover
ontological challenges
mapping in the UK. Remote Sensing of Environment, 68,
We may claim here for the field of remote sensing that some long- 206216.
known principles about technological determinism (McLuhan Aplin, P. and Atkinson, P. M. (2001). Sub-pixel land cover map-
1964 who basically claimed that humans shape their tools and ping for per-field classification. International Journal of
they in turn shape humans) may become more obvious today Remote Sensing, 22(14), 28532858.
because its practical and theoretical implications are now much Aplin, P. and Smith, G. M. (2011). Introduction to object-based
faster discovered. Nevertheless, the process of the social shap- landscape analysis. International Journal of Geographical
ing of technology can be long term, interactive, and sometimes Information Science, 25(6), 869875.
conflict ridden (Rohracher 2003). Arvor, D., Durieux, L., Andrs, S., and Laporte, M. A. (2013).
Like GIS, which has for some years been decried as onto- Advances in geographic object-based image analysis with
logically shallow and insufficient to the task of comprehend- ontologies: A review of main contributions and limita-
ing the many epistemological points of difference among users tions from a remote sensing perspective. ISPRS Journal of
(Schuurman 1999), remote-sensing literature offers very little Photogrammetry and Remote Sensing, 82, 125137.
in regard to its ontological and epistemological foundation. Atzmanstorfer, K. and Blaschke, T. (2013). Geospatial web:
Without doubt, remote-sensing principles have solid foundations A tool to support the empowerment of citizens through
in physics. Only through the amalgamation with GIS-principles e-participation? In C. Nunes Silva (ed.), Handbook of
with OBIA the need for a theoretical, that is, epistemological Research on E-Planning: ICTs for Urban Development and
and ontological, establishment increases. As long as the pixel is Monitoring (pp. 144171). Hershey, PA: IGI-Global.
2016 by Taylor & Francis Group, LLC
290 Remotely Sensed Data Characterization, Classification, and Accuracies
Baatz, M., Hoffmann, C., and Willhauck, G. (2008). Progressing Blaschke, T. and Merschdorf, H. (2014). Geographic Information
from object-based to object-oriented image analysis. In Science as a multidisciplinary and multi-paradigmatic field.
T. Blaschke, Lang, S., and Hay, G. J. (eds.), Object-Based Cartography and Geographic Information Science, 41(3),
Image Analysis. Berlin, Germany: Springer. 196213.
Baatz, M. and Schpe, A. (2000). Multiresolution segmenta- Blaschke, T. and Strobl, J. (2001). Whats wrong with pixels? Some
tion: an optimization approach for high quality multi- recent developments interfacing remote sensing and GIS.
scale image segmentation. Angewandte Geographische GISZeitschrift fur Geoinformationssysteme, 14(6), 1217.
Informationsverarbeitung XII (pp. 1223). Bock, M., Rossner, G., Wissen, M., Remm, K., Langanke, T., Lang,
Benz, U. C., Hofmann, P., Willhauck, G., Lingenfelder, I., and S., T., Klug, H., Blaschke, T., and Vrscay, B. (2005). Spatial
Heynen, M. (2004). Multi-resolution, object-oriented fuzzy indicators for nature conservation from european to local
analysis of remote sensing data for GIS-ready information. scale. Environmental Indicators, 5(4), 322328.
ISPRS Journal of Photogrammetry and Remote Sensing, Bunting, P., Clewley, D., Lucas, R. M., and Gillingham, S. (2014).
58(3), 239258. The Remote Sensing and GIS Software Library (RSGISLib).
Benz, U. C., Hofmann, P., Willhauck, G., Lingenfelder, I., and Computers & Geosciences, 62, 216226.
Heynen, M. (2008). Multi-resolution, object-oriented fuzzy Burnett, C. and Blaschke, T. (2003). A multi-scale segmentation/
analysis of remote sensing data for GIS-ready information. object relationship modelling methodology for landscape
ISPRS Journal of Photogrammetry and Remote Sensing, analysis. Ecological Modelling, 168(3), 233249.
58(34), 239258. Bhner, J., Blaschke, T., and Montanarella, L. (2008). SAGA
Blaschke, T. (2010). Object based image analysis for remote sens- Seconds out. Hamburger Beitrge zur Physischen
ing. ISPRS International Journal of Photogrammetry and Geographie und Landschaftskologie (vol. 19). Hamburg,
Remote Sensing, 65(1), 216. Germany.
Blaschke, T. (ed.). (2002). Fernerkundung und GIS: Neue Bckner, J., Pahl, M., Stahlhut, O., and Liedtke, C.-E. (2001).
SensorenInnovative Methoden. Karlsruhe, Germany: GEOAIDA a knowledge based automatic image data
Wichmann Verlag. analyser for remote sensing data. Paper presented at the
Blaschke, T., Burnett, C., and Pekkarinen, A. (2004). New contextual ICSC Congress on Computational Intelligence Methods and
approaches using image segmentation for object-based clas- ApplicationsCIMA, Bangor, U.K.
sification. In F. De Meer and de Jong, S. (eds.), Remote Sensing Camargo, F. F., Almeida, C. M., Costa, G. A. O. P., Feitosa, R. Q.,
Image Analysis: Including the Spatial Domain (pp. 211236). Oliveira, D. A. B., Heipke, C., and Ferreira, R. S. (2012).
Dordrecht, the Netherlands: Kluwer Academic Publishers. An open source object-based framework to extract land-
Blaschke, T., Hay, G. J., Kelly, M., Lang, S., Hofmann, P., Addink,E., form classes. Expert Systems with Applications, 39(1),
Feitosa, R., van der Meer, F., van der Werff, H., Van Coillie, 541554.
F., and Tiede, D. (2014). Geographic object-based image Castro, A. I. D., Lpez Granados, F., Gmez-Candn, D., Pea
analysis: A new paradigm in remote sensing and geo- Barragn, J. M., Novella, C., Jos, J., and Jurado-Expsito,
graphic information science. ISPRS International Journal of M. (2013). In-season site-specific control of cruciferous
Photogrammetry and Remote Sensing, 87(1), 180191. weeds at broad-scale using quickbird imagery. 9th European
Blaschke, T., Hay, G. J., Weng, Q., and Resch, B. (2011). Collective Conference on Precision Agriculture ECPA (July 711, 2013).
sensing: Integrating geospatial technologies to under- Chen, G., Hay, G. J., Carvalho, L. M., and Wulder, M. A. (2012).
stand urban systemsAn overview. Remote Sensing, 3(8), Object-based change detection. International Journal of
17431776. Remote Sensing, 33(14), 44344457.
Blaschke, T., Lang, S., and Hay, G. J. (eds.). (2008). Object-Based Chubey, M. S., Franklin, S. E., and Wulder, M. A. (2006). Object-
Image Analysis, Spatial Concepts for Knowledge-Drives Remote based analysis of Ikonos-2 imagery for extraction of forest
Sensing Applications. Lecture Notes in Geoinformation and inventory parameters. Photogrammetric Engineering and
Cartography. Berlin, Germany: Springer-Verlag. Remote Sensing, 72(4), 383394.
Blaschke, T., Tiede, D., and Lang, S. (2006). An object-based Cleve, C., Kelly, M., Kearns, F. R., and Moritz, M. (2008).
information extraction methodology incorporating a-priori Classification of the wildlandurban interface: A compari-
spatial information. Paper presented at the ESA Conference, son of pixel-and object-based classifications using high-
Madrid, Spain. resolution aerial photography. Computers, Environment and
Blaschke, T. and Kux, H. (2005). Sensoriamento Remoto e SIG Urban Systems, 32(4), 317326.
acanados: Novos sistemas sensores mtodos inovadores. Sao Clinton, N., Holt, A., Scarborough, J., Yan, L., and Gong, P. 2010.
Paulo, Brazil: Oficina de Textos. Accuracy assessment measures for object-based image
Blaschke, T. and Lang, S. (2006). Object based image analysis for segmentation goodness. Photogrammetric Engineering and
automated information extractionA synthesis. Paper pre- Remote Sensing, 76, 289299.
sented at the Measuring the Earth II ASPRS Fall Conference, Colwell, R. N. (1964). Aerial photographyA valuable sensor for
San Antonio, TX, November 610. the scientist. American Scientist, 52(1), 1649.
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 291
Colwell, R. N. (1965). The extraction of data from aerial photo- Guo, Q., Kelly, M., Gong, P., and Liu, D. (2007). An object-
graphs by human and mechanical means. Photogrammetria, based classification approach in mapping tree mortality
20(6), 211228. using high spatial resolution imagery. GIScience & Remote
Colwell, R. N. (1973). Remote sensing as an aid to the manage- Sensing, 44(1), 2447.
ment of earth resources. American Scientist, 61(2), 175183. Hansen, M. C., Egorov, A., Potapov, P. V., Stehman, S. V.,
Costa, G. A. O. P., Feitosa, R. Q., Fonseca, L. M. G., Oliveira, D.A.B., Tyukavina, A., Turubanova, S. A., and Bents, T. (2014).
Ferreira, R. S., and Castejon, E. F. (2010). Knowledge-based Monitoring conterminous United States (CONUS) land
interpretation of remote sensing data with the InterIMAGE cover change with web-enabled landsat data (WELD).
system: Major characteristics and recent developments. Remote sensing of Environment, 140, 466484.
Paper presented at the Third International Conference on Haralick, R. M. and Shapiro, L. (1985). Survey: Image segmen-
Geographic Object-Based Image Analysis (GEOBIA 2010), tation techniques. Computer Vision, Graphics, and Image
June 29July 2, Ghent, Belgium. Processing, 29, 100132.
Cracknell, A. P. (1998). Synergy in remote sensingWhats in a pixel? Harvey, F. and Raskin, R. G. (2011). Spatial Cyberinfrastructure:
International Journal of Remote Sensing, 19(11), 20252047. Building New Pathways for Geospatial Semantics on Existing
De Chant, T. and Kelly, M. (2009). Individual object change Infrastructures Geospatial Semantics and the Semantic Web
detection for monitoring the impact of a forest pathogen on (pp. 8796). New York: Springer.
a hardwood forest. Photogrammetric Engineering & Remote Hay, G. J. and Blaschke, T. (2010). Special issue: Geographic
Sensing, 75(8), 10051013. object-based image analysis (GEOBIA). Photogrammetric
Dorren, L. K., Maier, B., and Seijmonsbergen, A. C. (2003). Engineering and Remote Sensing, 76(2), 121122.
Improved Landsat-based forest mapping in steep moun- Hay, G. J. and Castilla, G. (2006). Object-based image analysis:
tainous terrain using object-based classification. Forest Strengths, weaknesses, opportunities and threats (SWOT).
Ecology and Management, 183(13), 3146. International Archives of Photogrammetry, Remote Sensing
Durieux, L., Lagabrielle, E., and Nelson, A. (2008). A method for and Spatial Information Sciences, 36, 45.
monitoring building construction in urban sprawl areas Hay, G. J. and Castilla, G. (2008). Geographic Object-Based Image
using object-based analysis of Spot 5 images and existing Analysis (GEOBIA): A new name for a new discipline. In
GIS data. ISPRS Journal of Photogrammetry and Remote T. Blaschke, Lang, S., and Hay, G. J. (eds.), Object Based
Sensing, 63(4), 399408. Image Analysis (pp. 93112). New York: Springer.
Eastman, J. R. (2009). IDRISI Taiga Guide to GIS and Image Hlbling, D. and Neubert, M. (2008). ENVI Feature Extraction
Processing. Worcester, MA: Clark Labs Clark University. 4.5. Snapshot. In GIS Business (pp. 4851). Heidelberg,
Ebert, A., Kerle, N., and Stein, A. (2009). Urban social vulner- Germany: abcverlag GmbH.
ability assessment with physical proxies and spatial metrics Jensen, J. R. (1996). Introductory Digital Image Processing: A
derived from air-and spaceborne imagery and GIS data. Remote Sensing Perspective. Englewood Cliffs, NJ: Prentice-
Natural Hazards, 48(2), 275294. Hall Inc.
ERDAS. (2009). Remote sensing digital image processing soft- Kettig, R. L. and Landgrebe, D. A. (1976). Classification of mul-
ware Tutorial. tispectral image data by extraction and classification of
Flanders, D., Hall-Beyer, M., and Pereverzoff, J. (2003). homogeneous objects. IEEE Transactions on Geoscience and
Preliminary evaluation of eCognition object-based software Remote Sensing, 14(1), 1926.
for cut block delineation and feature extraction. Canadian Koch, B., Heyder, U., and Weinacker, H. (2006). Detection
Journal of Remote Sensing, 29(4), 441452. of individual tree crowns in airborne LiDAR data.
Fu, K.-S. and Mui, J. (1981). A survey on image segmentation. Photogrammetric Engineering and Remote Sensing, 72(4),
Pattern Recognition, 13(1), 316. 357363.
Gergel, S. E., Stange, Y., Coops, N. C., Johansen, K., and Kirby, Koch, D. E., Mohler, R. L., and Goodin, D. G. (2007). Stratifying
K. R. (2007). What is the value of a good map? An example land use/land cover for spatial analysis of disease ecology
using high spatial resolution imagery to aid riparian resto- and risk: an example using object-based classification tech-
ration. Ecosystems, 10(5), 688702. niques. Geospatial Health, 2(1), 1528.
Goodchild, M. F. (1992). Geographical information science. Kuhn, T. S. (1962). The Structure of Scientific Revolutions. Chicago,
International Journal of Geographic Information Systems, IL: The Chicago University Press.
6, 3145. Krting, T. S., Garcia Fonseca, L. M., and Cmara, G. (2013).
Goodchild, M. F. (2007). Citizens as sensors: The world of volun- GeoDMAGeographic data mining analyst. Computers &
teered geography. GeoJournal, 69(4), 211221. Geosciences, 57, 133145.
Green, K. and Lopez, C. (2007). Using object-oriented classifica- Laliberte, A. S., Fredrickson, E. L., and Rango, A. (2007).
tion of ADS40 data to map the benthic habitats of the state Combining decision trees with hierarchical object-oriented
of Texas. Photogrammetric Engineering and Remote Sensing, analysis for mapping arid rangelands. Photogrammetric
73(8), 861. Engineering and Remote Sensing, 73(2), 197207.
2016 by Taylor & Francis Group, LLC
292 Remotely Sensed Data Characterization, Classification, and Accuracies
Laliberte, A. S., Goforth, M. A., Steele, C. M., and Rango, A. Pascual, C., Garca-Abril, A., Garca-Montero, L. G., Martn-
(2011). Multispectral remote sensing from unmanned Fernndez, S., and Cohen, W. (2008). Object-based semi-
aircraft: Image processing workflows and applications automatic approach for forest structure characterization
for rangeland environments. Remote Sensing, 3(11), using LiDAR data in heterogeneous Pinus sylvestris stands.
25292551. Forest Ecology and Management, 255(11), 36773685.
Laliberte, A. S., Rango, A., Havstad, K. M., Paris, J. F., Beck, R. F., Pickles, J. (2004). A History of Spaces: Cartographic Reason, Mapping,
McNeely, R., and Gonzalez, A. L. (2004). Object-oriented and the Geo-Coded World. New York: Psychology Press.
image analysis for mapping shrub encroachment from Quinlan, R. (2013). Data Mining Tools See5 and C5.0. http://www.
1937 to 2003 in southern New Mexico. Remote Sensing of rulequest.com/see5-info.html.
Environment, 93(12), 198210. Radoux, J. and Defourny, P. (2007). A quantitative assessment of
Lang, S. (2008). Object-based image analysis for remote sensing boundaries in automated forest stand delineation using very
applications: Modeling realityDealing with complexity. high resolution imagery. Remote Sensing of Environment,
In T. Blaschke, Lang, S., and Hay, G. J. (eds.), Object-Based 110(4), 468475.
Image Analysis (pp. 125). New York: Springer. Resch, B. (2013). People as Sensors and Collective Sensing-
Lang, S., Albrecht, F., Kienberger, S., and Tiede, D. (2010). Object Contextual Observations Complementing Geo-Sensor
validity for operational tasks in a policy context. Journal of Network Measurements Progress in Location-Based Services
Spatial Science, 55(1), 922. (pp. 391406). Berlin, Germany: Springer.
Lang, S., Kienberger, S., Tiede, D., Hagenlocher, M., and Pernkopf, Rocchini, D., Perry, G. L., Salerno, M., Maccherini, S., and
L. (2014). Geonsdomain-specific regionalization of space. Chiarucci, A. (2006). Landscape change and the dynam-
Cartography and Geographic Information Science, 41(3), ics of open formations in a natural reserve. Landscape and
214226. Urban Planning, 77(1), 167177.
Lang, S. and Blaschke, T. (2006). Bridging remote sensing and Rohracher, H. (2003). The role of users in the social shaping of
GISWhat are the main supporting pillars? International environmental technologies. Innovation, 16(2), 177192.
Archives of Photogrammetry, Remote Sensing and Spatial Sagl, G., Loidl, M., and Beinat, E. (2012). A visual analytics
Information Sciences, XXXVI-4/C42. approach for extracting spatio-temporal urban mobil-
Leberl, F. and Gruber, M. (2003). Flying the new large format digital ity information from mobile network traffic. ISPRS
aerial camera Ultracam. Photogrammetric Week, 3, 6776. International Journal of Geo-Information, 1(3), 256271.
Liu, Y., Li, M., Mao, L., Xu, F., and Huang, S. (2006). Review of Salehi, B., Ming Zhong, Y., and Dey, V. (2012). A review of the
remotely sensed imagery classification patterns based effectiveness of spatial information used in urban land
on object-oriented image analysis. Chinese Geographical cover classification of VHR imagery. International Journal
Science, 16(3), 282288. of Geoinformatics, 8(2), 3551.
Malczewski, J. (1999). GIS and Multicriteria Decision Analysis. Schuurman, N. (1999). Critical GIS: Theorizing an emerging sci-
New York: John Wiley & Sons. ence. Cartographica, 36(4), 1101.
Maxwell, S. K. (2010). Generating land cover boundaries from Sheppard, E. (2006). Knowledge production through critical GIS:
remotely sensed data using object-based image analysis: Genealogy and prospects. Cartographica, 40, 521.
Overview and epidemiological application. Spatial and Silva, T. S. F., Costa, M. P. F., and Melack, J. M. (2010). Spatial
Spatio-Temporal Epidemiology, 1(4), 231237. and temporal variability of macrophyte cover and produc-
McKeown Jr., D. M., Harvey, W. A., and Wixson, L. E. (1989). tivity in the eastern Amazon floodplain: A remote sensing
Automating knowledge acquisition for aerial image inter- approach. Remote Sensing of Environment, 114(9), 1998
pretation. Computer Vision, Graphics, and Image Processing, 2010. doi: 10.1016/j.rse.2010.04.007.
46(1), 3781. Stow, D., Lopez, A., Lippitt, C., Hinton, S., and Weeks, J.
McLuhan, M. (1964). Understanding Media. New York: (2007). Object-based classification of residential land use
McGraw-Hill. within Accra, Ghana based on QuickBird satellite data.
Moreno, N., Wang, F., and Marceau, D. J. (2010). A geographic International Journal of Remote Sensing, 28(22), 5167.
object-based approach in cellular automata modeling. Tiede, D. (2014). A new geospatial overlay method for the
Photogrammetric Engineering and Remote Sensing, 76(2), analysis and visualization of spatial change patterns using
183191. object-oriented data modeling concepts. Cartography and
Opitz, D. and Blundell, S. (2008). Object recognition and image Geographic Information Science, 41(3), 227234.
segmentation: the Feature Analyst approach. In Blaschke, Tiede, D., Huber, J., and Kienberger, S. (2012). Implementation
T., Lang, S., Hay, G.J. (eds.), Object-Based Image Analysis of an interactive WebGIS-based OBIA geoprocessing ser-
(pp. 153167). Berlin/Heidelberg, Germany: Springer. vice. In Proceedings International Conference on Geographic
Pal, R. and Pal, K. (1993). A review on image segmentation tech- Object-Based Image Analysis (Vol. 4, pp. 402406), Rio de
niques. Pattern Recognition, 26(9), 12771294. Janeiro, Brazil, May 79.
2016 by Taylor & Francis Group, LLC
Object-Based Image Analysis 293
Townshend, J. R. G., Huang, C., Kalluri, S. N. V., Defries, R. S., Yan, G., Mas, J.-F., Maathuis, B. H. P., Xiangmin, Z., and
Liang, S., and Yang, K. (2000). Beware of per-pixel char- Van Dijk, P. M. (2006). Comparison of pixel-based and
acterization of land cover. International Journal of Remote object-oriented image classification approachesA
Sensing, 21(4), 839843. case study in a coal fire area, Wuda, Inner Mongolia,
Troyo, A., Fuller, D. O., Calderon Arguedas, O., Solano, M. E., and China. International Journal of Remote Sensing, 27(18),
Beier, J. C. (2009). Urban structure and dengue incidence 40394055.
in Puntarenas, Costa Rica. Singapore Journal of Tropical Yu, Q., Gong, P., Clinton, N., Kelly, M., and Schirokauer, D.
Geography, 30(2), 265282. (2006). Object-based detailed vegetation classification with
Tullis, J. A., Jensen, J. R., Raber, G. T., and Filippi, A. M. 2010. airborne high spatial resolution remote sensing imagery.
Spatial scale management experiments using optical aerial Photogrammetric Engineering and Remote Sensing, 72(7),
imagery and LiDAR data synergy. GIScience & Remote 799811.
Sensing, 47, 338359. Yue, P., Di, L., Wei, Y., and Han, W. (2013). Intelligent services
Walsh, S. J., McCleary, A. L., Mena, C. F., Shao, Y., Tuttle, J. P., Gonzlez, for discovery of complex geospatial features from remote
A., and Atkinson, R. (2008). QuickBird and Hyperion data sensing imagery. ISPRS Journal of Photogrammetry and
analysis of an invasive plant species in the Galapagos Islands of Remote Sensing, 83, 151164.
Ecuador: implications for control and land use management. Zhou, W. and Troy, A. (2008). An object-oriented approach for
Remote Sensing of Environment, 112(5), 19271941. analysing and characterizing urban landscape at the par-
Weeks, J. R., Hill, A., Stow, D., Getis, A., and Fugate, D. (2007). Can cel level. International Journal of Remote Sensing, 29(11),
we spot a neighborhood from the air? Defining neighbor- 31193135.
hood structure in Accra, Ghana. GeoJournal, 69(12), 922. Zook, M., Graham, M., Shelton, T., and Gorman, S. (2010).
Wulder, M. A., White, J. C., Masek, J. G., Dwyer, J., and Roy, D.P. Volunteered geographic information and crowdsourcing
(2011). Continuity of Landsat observations: Short term con- disaster relief: A case study of the Haitian earthquake.
siderations. Remote Sensing of Environment, 115(2), 747751. World Medical and Health Policy, 2(2), 733.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
15
Geospatial Data
Integration in OBIA:
Implications of Accuracy
and Validity
Acronyms and Definitions ................................................................................................................ 295
15.1 Conditioned Information ...................................................................................................... 295
OBIA in Support to Geospatial Information Needs Enriched Information through OBIA
15.2 Object Validity ........................................................................................................................ 299
Color and Form: Elements of Image Understanding Human vs. Machine Vision Class
Modeling Validity of Object Delineation and Classification Multiple-Stage Validation
Stefan Lang 15.3 Object Fate ............................................................................................................................... 306
From Static to Dynamic Change Concepts Application Scenario #1: Object-Based
University of Salzburg
Information Update Object Fate Analysis Object Linking Application Scenario #2:
Dirk Tiede Rapid Information Extraction
Salzburg University 15.4 Conclusions...............................................................................................................................313
University of Salzburg References.............................................................................................................................................313
Acronyms and Definitions SIA Spatial image analysis
SIAM Satellite image automatic mapper
3-D Three-dimensional SII Spatial Information Infrastructure
AI Artificial intelligence SPOT Satellite Pour lObservation de la Terre
APPS Adaptive per-parcel segmentation SWEET Semantic Web for Earth and Environmental
CBD Convention on Biological Diversity Terminology
EC European Commission UN United Nations
EO Earth observation VH(S)R Very high (Spatial) Resolution
ESA European Space Agency GSD Ground samle distance
ESRI Environmental Systems Research Institute WFD Water Framework Directive
EU European Union
FCA Formal Concept Analysis
GEO Group of Earth Observation 15.1 conditioned information
GEOBIA Geographical Object-Based Image Analysis
GI(S) Geographic Information (System/s) The advancement of feature recognition, scene reconstruction,
GMES Global Monitoring for Environment and Security and advanced image analysis techniques facilitates the extrac-
LCCS Land Cover Classification System tion of thematic information, for policy making support and
LIST Landscape Interpretation Support Tool informed decisions. As a strong driver, the ubiquitous availabil-
OBAA Object-based accuracy assessment ity of image data strives to match the ever-increasing need for
OBCD Object-based change detection updated geo-information. In response, we need methods that
OBIA Object-based image analysis exploit image information not only more comprehensively, but
OFA Object fate analysis also more effectively. Object-based image analysis (OBIA), with
OWL Ontology Web Language its explicit focus on objects and their relationships, enables to
RMS Root mean square address and model new dimensions of target classes, enriching
SDI Spatial Data Infrastructure information extracted from image data.
295
2016 by Taylor & Francis Group, LLC
296 Remotely Sensed Data Characterization, Classification, and Accuracies
Spatially defined
categories (pattern-
oriented, multiple objects)
Class modelling (spectrally
Coarse-scale segmentation distinct objects)
(including noise objects)
OBIA
Spectrally defined
categories
Spatially defined
categories (shape-oriented)
Fine-scale
segmentation
FIg u r e 1 5.1 Extended set of categories (target classes) using spatial in addition to spectral features (see text for further explanations).
15.1.1 oBiA in Support to Geospatial So, although object-based suggests objects to be the funda-
information needs mental elements to base the analysis on, OBIA goes beyond
objects as isolated items distributed over space. In capturing
Object-based image analysis (OBIA), as the name indicates, and utilizing the manifold relationships among these objects, we
operates on objects representing real-world entities as the con- can use additional features in classification that exceed classical
stituents of our geographic reality. Spatial key features (basically quantitative measures. Topological and hierarchical relation-
size and shape) are intrinsic parameters to be considered in the ships are more qualitative features that apply in a more gradual
information extraction process, making spatial characteristics an way than measurable features such as object size, or spectral or
additional feature domain in classification. Using regionalization spatial distance. The trend set by OBIAsome may call it a para-
techniques for image segmentation, there is a complementarity digm shift, cf. Blaschke etal. (2014)from pixels (as technically
(one could say a trade-off) between spectral and spatial similari- defined units), via objects (as spatially manifested and conceptu-
ties, that is, between color and neighborhood. In other words, the ally ascertained entities), toward relational patterns (as multis-
spatial constraint balances the spectral behavior, which leads to calar qualities of objects), is just taking off, opening a plethora of
(scalable) generalization and a reduction of the so-called salt-and- possible new ways of flexible spatial representations.
pepper effect (Bischof et al. 1992, Blaschke and Strobl 2001). In The explicit focus on spatial properties and relationships of
OBIA, the category that a group of picture elements is assigned OBIA* enables to address high-level semantic classes, including
to depends on both spatial and spectral characteristics. This may spatial composite classes that emerge (in terms of relevance) on a
sound like a restriction narrowing down classification power, but certain phenomenon scale. Here we clearly move beyond scien-
it actually extends the set of target classes to be addressed. More tific curiosity, with an enabling tool to meet the ever-increasing
precisely, it opens another dimension of potential target classes demand for geospatial information from the perspective of vari-
(Lang 2008), on two additional semantic levels next to the level of ous space-related policies (Lang et al. 2009b). The term space-
spectral classes (see Figure15.1). This comprises, first, all subcat- related is used in a broad sense to comprise all policies with a
egories or instances of a spectral class that are defined by shape spatial component that aredirectly or indirectlyrelying on
features on individual object level: a class <water> may split into critical geospatial information updates. Such policies entail
<lake | river> depending on its length/width ratio. A class <built- more and more societal and environmental domains across geo-
up> may split into several subcategories of a village typology graphic scale domains (i.e., from local to continental), for exam-
based on footprint physiognomy. Second, and even more crucial, ple, a local policy for climate change adaption, a regional policy
we find spatial properties in terms of relations among objects. We of sustainable urban development, the European Union (EU)
will come back to this issue in Section 15.2, when discussing the
term class modeling. In a nutshell, the integration of object rela-
* In order to highlight the relationships between objects rather than empha-
tionships (including relative coverage, distance, and in particular sizing objects as such, we may use a more generic term: spatial image anal-
topological features) allows for addressing complex, composite ysis. Still, since OBIA is an established term in the community, we use it
target classes on high semantic level. throughout the chapter.
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 297
policy on integrated water management (Water Framework OBIA not only produces objects, but is also able to inte-
Directive, WFD), or a global policy on reducing biodiversity loss grate existing spatial units (i.e., digital geographic fea-
(United Nations Convention on Biological Diversity, UN-CBD). tures) in the analysis process. In many domains, the
The primary source is increasingly supplied by imagery and update of (existing) geospatial information is more
respective information exploitation strategies, summarized by critical than the provision of new, spatially disintegrated
the term earth observation (EO). EO is, say, a (semi-)political units (Tiede et al. 2007). Geodata from SDIs may be
expression (Lang 2008) that highlights the capacity of satellite used as a reference data set, to a lesser degree as fixed
(and other) sensor technology for the purpose of updated infor- and exclusive set of boundaries, but as a pool of potential
mation provision in civil application domains. EO infrastruc- boundaries to adapt results on, in terms of both spatial
ture (i.e., space and ground infrastructure for satellite imagery, reference and generalization level. Remote sensing prod-
sensor networks) and the analysis capabilities are bundled to ucts are often stand-alone or even isolated products, not
support societal benefits and the political ambition of such poli- readily capable to match with existing geodata reposi-
cies. Recent examples of large-scale international endeavors are tories (see Section 15.3). This applies in terms of both
Group of Earth Observation (GEO) and the European ESA-EC nomenclature (thematic agreement) and generalization
conjoint initiative Copernicus (formerly Global Monitoring (geometrical match). Interestingly, by explicitly address-
for the Environment and Security (GMES)). In order to reach ing the spatial component in the analysis process, we can
a consistent approach to monitoring all kinds of environmen- improve both!
tal and societal conditions, considerable efforts are made to
investigate opportunities EO methodologies may hold to facili- Spatial data policies like the European INSPIRE directive
tate the reporting requirements and the evaluation of interven- (2007/2/EC) seeks to provide a common baseline for interop-
tion options (Lang etal. 2009b). EO technology may contribute erable data usage, to overcome technological constraints via
significantly toward achieving the objectives of multilateral standardization and opening the way to a content-driven
agreements by (1) increasing knowledge about the underlying debate. Information provision, whether spatial or nonspatial,
processes, (2) supporting the efficient management and moni- requires compatibility to existing data pools. SDIsor more
toring of tasks, and thereby (3) contributing to more effective- appropriate: spatial information infrastructures (SIIs)have
ness of conventions or treaties. It goes beyond the scope of this the particularity to ensure a spatial match as well, by adher-
chapter, to list all the information needs in the various societal ing to generalization principles, and telling modifications
domains, as well as the phenomenon scale and respective policy from different representations. This not only is limited to spa-
scale and link these to existing (and near-future) EO capacity. tial or thematic disagreement, but also relates to the fitness of
For an overview, see Zeil etal. (2008). Such scoping is done when information products derived from different sources. Decision
designing new satellite missions and the suite of onboard instru- makers should be able to interpret the information unambigu-
ments. The Copernicus Sentinel family (starting with the recent ously, a requirement usually attached with attributes like inter-
launch of Sentinel-1A) is designed in a way to be as flexible and pretability, communicability, spatial explicitness, and clarity
ubiquitous in application as possible. about the assumptions. Often, it seems, basic requirements like
The diverse demand requires not only adequate observation usability are neglected in remote sensing products so that deci-
systems but also highly capable analysis tools and methods to sion makers face a notorious dilemma (Lang etal. 2009b): an
effectively transform the enormous amount of data into mean- overwhelming blessing of satellite data, poured out generously
ingful and ready-to-use information. OBIA may prove this capa- to meet the prescribed monitoring schemes, but hardly fit for
bility in the following (nonexclusive) areas: integration in stakeholders daily businesses and established
workflows (ibid.).
OBIA, through its multiscale option (Marceau 1999, Hay Let us consider an example (see Figure 15.2) that will be
et al. 2001), enables to adapt the scale inherent in the discussed in greater detail in Section 15.3): the representa-
source data to the phenomenon scale and thereby to the tion of different agricultural parcels derived by image seg-
policy scale. In combination with an appropriate resolu- mentation from SPOT-5 data, even if perfectly delineated, is
tion level of the used imagery, this gives high flexibility a no-go for users because the analysis scale does not match
in terms of identifying and working on a commensurate with the policy scale. The regional development plan in the
analysis scale. Stuttgart Metropolitan area in Southwest Germany requires
The object-oriented data model intrinsically linked to the a baseline geometry representing so-called biotope complexes
OBIA approach allows for complex and hierarchical class (Schumacher et al. 2007). These are functional composed
models, transferable and adjustable between application spatial units whose outlines should match digital cadastral
domains and geographic areas (Tiede 2014). This is not to boundaries. The automated segmentation shown in Figure
say that OBIA enables fully automated procedures, but the 15.2a provides neither of it. By means of a combined strategy
adjustment of once established class model architectures called spatially constraint class modeling, Tiede etal. (2010)
is a minor effort as compared to manual grouping and were able to deliver the appropriate geospatial information
reclassifications (Tiede etal. 2010). (Figure 15.2b).
2016 by Taylor & Francis Group, LLC
298 Remotely Sensed Data Characterization, Classification, and Accuracies
(a)
(b)
FIg u r e 15.2 Understanding others object understanding. (a) Unconditioned information, simple delineation of land cover objects, (b) condi-
tioned information, aggregated biotope complex.
15.1.2 enriched information through oBiA information conditioning for the act of turning unstructured
(and redundant) information into ready-to-use information,
As pointed out by several authors (Blaschke etal. 2007) at the whereby ideally a full match between the products provided
launching phase of the Copernicus/GMES initiative, EO-based and the user demand, plus its fitness in existing workflows, is
monitoring systems need to be capable of transforming complex achieved. OBIA, also called GEOBIA (Hay and Castilla 2008)
scene content into ready-to-use information, in a reliable, trans- for highlighting the geographic application context, provides
ferable, effective, anddesirablycost-efficient manner. Global the toolbox, capable of treating users needs individually by uti-
commitments, directives, and policies have a pronounced lizing intelligent tools and algorithms for automatically delin-
demand for timely, accurate, and conditioned geo-spatial infor- eating, extracting, and categorizing geographic objects from
mation (Lang et al. 2008), valorizing the load of raw data col- remotely sensed images (Lang et al. 2010a). Methodologically,
lected from various monitoring systems. Naturally, as it seems, OBIA seeks to accommodate various geo-spatial concepts,
a technically and spatially more literate user community asks remote sensing principles (spectral behavior), and the wealth of
for advanced geo-spatial products and expresses their needs experience from visual interpretation and manual delineation
accordingly. With an increased overall consciousness to prevail- (Hay and Castilla 2008).
ing societal challenges, the need for targeted information, too, Decision makers and other policy implementation bodies do
is raising, both in amount and in ambition (toward efficiency, have a clear mission to follow; in other words: informed deci-
quality, etc.) (Lang etal. 2009a). sions are mostly embedded in any kind of operational workflow,
The advancement in feature recognition and image analysis that is, the decisions to be taken are hardly independent from any
techniques facilitates the extraction of rich thematic informa- existing institutional setting. In addition, each decision has spa-
tion, to support informed decision making irrespective of a tial constraints, often linked to political boundaries. This kind
particular application field. The term rich relates to semantic of double constriction (both space- and mandate/task-related) is
content that relates to requirements from particular application called the policy scope (Lang etal. 2008). Conditioned informa-
fields or policy contexts, beyond generic land cover classification tion (ibid.) is the result of a process to fulfill the user demand
schemes (Lang et al. 2009a). With increasing capacities, both in technological and conceptual sense and has undergone any
on the imaging and on the processing side of observation sys- kind of fitness check for operational use (user validation) (p.3)
tems, high expectations (Lang 2008) were raised from all sides, (Lang et al. 2010a). Conditioned information, in other words,
including academia, service providers, and users to the recent means the full match between geo-spatial information provided
generation of very-high-resolution (VHR) data (Aschbacher and the accommodation of this information in their established
2002) such as (by today) QuickBird, WorldView, and GeoEye. workflows.
Even though shortcomings may exist in data provision and han- Another key aspect in information conditioning is the issue
dling, data acquired as such through EO missions are abundant, of scale. According to hierarchy theorists (Allen and Starr
so that users as well as service providers are challenged by inte- 1982), scale is defined as the period of time or space over which
grating and valorizing them. Lang et al. (2009b) use the term
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 299
signals are integrated to give message. Converting imagery well as representation (including visualization). We subsume the
to information ideally follows the principles of communica- entirety of aspects to be considered in this process, under the
tion models. The role of image interpretation or classification term object validity.
is a kind of translation support channelizing and condition-
ing the oft overwhelming (unstructured) information resid-
15.2.1 color and Form: elements
ing in images. Scale is an intrinsic aspect in communication,
ofimage Understanding
when signals need to be filtered and integrated to give message.
Here we need to match the phenomenon scale in general and The guiding principle of OBIA is to represent complex scene
the policy scale in particular. The flexibility of current digital content in such a way that the imaged reality is best under-
representations turns a fixed cartographic scale on maps to a stood and a maximum of the captured information is revealed,
visualization scale confining dynamically the level of detail extracted, and conveyed to those who need it. The term image
of displayed information (Montello 2001). An important step, understanding (Pinz 1994, Zlatof et al. 2004) refers to the
whether in traditional paper map production or dynamic on- reconstruction of the imaged scene as a complete process of
screen visualizations, is generalization. Generalization refers to transforming reflectance values into symbolic representations
the reduction of redundant detail with coarser scale, a filtering and, ultimately, to valid information. Ideally, the target sym-
effect that makes representations commensurate to the scale of bology is a fixed set of symbols that are to be detected in an
interest (simplification, enhancement, selection). In digital rep- image. This applies, for example, to optical character recogni-
resentations, generalization can be achieved by (1) algorithmic tion where a scanned document is analyzed for the occurrence
smoothing of lines or polygon outlines by dropping a selected of a given set of characters (letters, numbers, and special char-
number of vertices; (2) selective display of raster cells or pixels acters). Their appearance may vary within certain ranges, but
with decreasing zoom level; and (3) manual, context-specific in each language or script, there is a fixed and limited set of
omissions or exaggerations to suppress or emphasize certain characters to target on. We can unambiguously assess the suc-
cartographic aspects. cess rate of this image reconstruction by counting the correctly
OBIA generally supports the idea of multiscale representa- identified characters. Once characters are rightly extracted, a
tion (Hay et al. 2001) by enabling nested hierarchies of image word processor is able to interpret their positions relative to
objects in several scales. In this process, generalization is a par- each other and eventually identify words, which can be fur-
ticular challenge (see later) implying a trade-off between spatial ther understood by a spell-checker or even be read out by the
embeddedness and appropriate scale: either the object hierar- machine. A visually impaired person, not able to read the con-
chy is strictly fitting boundary-wise, that is, outlines of higher- tent of a letter, may consider such image understanding capa-
scale objects are not generalized (scale-adaptive strategy), or bilities as a full match between the information provided and
boundaries are generalized according to the scale level and the his or her need (while any interpretation error can be clearly
object hierarchy is not spatial explicit any longer (scale-specific spotted out as well). Characters to represent letters, words, etc.,
strategy). are standardized representations (just as notes are for musi-
The challenges touched upon briefly in the previous paragraphs cal events) as a means to encode and decode information in
extend to concept and methods of assessing and validating the a (thereby regulated) communication process. In other words,
results. Among others, this applies for specific object-based the process of information exchange uses unequivocal, even
approaches to accuracy assessment and change detection (Lang standardized symbology (fonts), and is thereby well-posed and
2008). Object based implies that accuracy or changes are assessed automatable.
not only thematically but also spatially: this refers to the way how Returning to remotely sensed images, we face the problem
objects are delineated, that is, their spatial representation (posi- that there is no such fixed target symbology through which
tion and precision of boundary), and to what degree these repre- we could communicate content and exchange information in a
sentations change over time. standardized way. As close as possible, we get to this in applying
discrete color levels, which are intersubjectively comprehensible
15.2 object Validity and used consensually. But even if color categories exist, color
remains a continuous phenomenon whose discretization is per
The transformation and reconstruction of an image scene into convention and not standardized. A pixel (picture element, i.e.,
ready-to-use geospatial information matching the conceptual the smallest, technically defined unit) appears in a certain color
understanding and the practical needs of users are the primary representing an integrated signal of the spectral reflectance of
aim of OBIA. In this advanced process of image understand- a small portion of the Earth surface. The higher the spectral
ing, image objects may be correctly classified but not valid. This resolution (number of bands, n) and the number of quantiza-
may sound contradictory, but the enabling by OBIA to delin- tion levels (radiometric resolution, bit-depth), the more colors a
eate spatially flexible units challenges the classical concept of pixel can adopt, as first the dimensionality of the feature space
binary accuracy assessment (right or wrong). In addition to the increases and second the number of possible steps.
assigned label or category, we need to evaluate the way how units But what do colors mean? How do we reach from sub-
are delineated. This implies issues of scale, generalization, as symbolic level (colors) to a symbolic one (label)? The integrated
2016 by Taylor & Francis Group, LLC
300 Remotely Sensed Data Characterization, Classification, and Accuracies
signal stored in a pixel represents physical conditions of the would indicate only a change in these conditions. In reality,
ground (land cover), so we can link color with physical mod- however, all kinds of disturbances occur such as atmospheric
els using spectral signatures. Classification schemes such as the influences and topographic effects (Richter and Schlpfer
land cover classification system (LCCS) (Di Gregorio and Jansen 2002) viewing angle of a VHR sensor, etc., that make the same
2005) use spectral signatures to convert reflectance values (via land cover captured twice look differently. While color is the
colors) into nominal classes, that is, labels. With additional most direct (and unambiguous) property of remote sensing
bands, more and better physical models of geographic features images to be used at a per-pixel basis, other features of geo-
can be incorporated in a classification scheme (e.g., spectral graphic phenomena like size, form, appearance, and orienta-
behavior of vegetation in the IR spectrum). The assignment of tion are likewise important. Such features cannot be drawn
a pixel to a nominal class allows for classical site(i.e., pixel-) from single pixels (Blaschke and Strobl 2001), but need to
specific accuracy assessments (Congalton and Green 1998). The consider spatial context, partly even spatial explicit measures.
remaining ambiguity in the class assignments can be dealt with Quantitative parameters can be utilized (such as area, perim-
using, for example, fuzzy rules (Zadeh 1965) in both the clas- eter, shape, and texture) of pixel aggregates to characterize
sification and evaluation process. Since a pixel is a fixed unit, these features, but a standardized scheme is even more diffi-
the assessment of a class assignment focuses on the color-based cult to obtain due to the interrelated character of such features
labeling of this particular pixel. If true ground data, that is, inde- (see later). Spatial features allow for additional differentiation
pendent reference information from the field, is by any reason among target classes on (higher) symbolic level: a river is an
(inaccessibility, security, and time/costs) impaired or not pos- elongated water body, a lake more compact. Such features
sible at all, visual inspection of the classification is an alterna- can be obtained once pixel aggregates are organized in seg-
tive (Campbell 2002). Visual accuracy assessment assumes there ments or image objects (Benz etal. 2004). When co-assessing
is an agreement (full match) between the color impression on color and form, some subtle variances in color are sacrificed
screen and the evaluators experience of physical models and to the overall spatial property of the aggregate. A lake may be
the spectral behavior of various land cover types in color space. considered a lake due to its prevailing polygonal form even if
Usually, a visual inspector considers, to various degree, spatial there are some subtle variances in color tone or even color due
context in taking the decision (see Figure15.3). to different water depths or presence of vegetation, etc. The
Ideally, physical conditions on the ground are the only automated extraction of boundaries is a function of the image
influence of a pixels integrated signal, and different signals resolution that controls average size, number, and complexity
1 Color only: river and sea merge
2 Color and form: river and sea can be
distinguished in multiple scales:
FIg ur e 15.3 Shape is a highly distinguishable feature of land cover classes similar in color (here large river/sea). (Image source: Landsat 8 OLI
(Nov 2013), US Geological Survey.
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 301
of the boundary of the segmented objects. An interpreter 15.2.2 Human vs. Machine Vision
inherently applies generalization and adapts the delineation
according to the mapping scale and the level of detail pre- Great advances have been made in computer vision, but the
scribed by the classification scheme. Higher-level semantics entire potential of human vision has not been achieved so far
usually imply a coarser scale (Belgiu et al. 2014). In OBIA, it seems. While biophysical principles like retinal structure and
multiresolution segmentation is a strategy to perform image functioning and singular processes such as the cerebral reaction
segmentation at several scale levels so to represent the respec- are analytically known, we still lack the bigger picture on human
tive multilevel content of a scene scale adaptively (i.e., without perception as a whole (Blaschke etal. 2014). OBIA tries to exter-
generalization) or scale specifically (i.e., with generalization). nalize basic principles of human perception (Lang 2008), so it is
As we will discuss later, point-based methods fail to assess the important to understand how we deal with imaged information
spatial dimension of object delineations. Lucieer and Stein in various scales, how we manage to relate recognized objects
(2002) distinguish between existential and extensional uncer- to each other with ease, and how we understand complex scene
tainty of delineated objects. contents readily. The ultimate benchmark of OBIA is human
To treat color and form separately suggests independence of perception (Lang 2008) (page 7), but in fact, it is difficult to
both aspects. A (remote sensing) image understanding system explicitly describe what happens when we look at an image and
(IUS) needs to integrate both aspects, according to models of suddenly see something (Blaschke etal. 2014).
human vision (see following section). An IUS that follows the Pattern recognitionnot necessarily the interpretation
multistage paradigm of human vision by Marr (1982) starts works without major effort (Eysenck and Keane 1995, Tarr and
with pre-attentive vision as a first stage and then works along Cheng 2003). Human perception is a complex matter of filtering
these building blocks to compose more complex target classes. relevant signals from noise (Lang 2008), a selective processing
There are several strategies to accomplish this first stage; two of detailed information, and, finally, experience. With respect to
of themwith more communalities than differencesare visual information processing, Marr (1982) provided a concep-
shortly discussed here. The first is a symbolic (i.e., meaning- tual framework of a three-leveled structure. According to Marrs
ful) (pre-)classification reflecting pre-attentive vision (Baraldi paradigm, we can distinguish between the following explana-
and Broschetti 2012) according to color similarity based on tory levels: (1) the computational level, which is related to the
physical models of spectral behavior as implemented in the purpose, logic, and strategy of perception; (2) the algorithmic
SIAM (Satellite Image Automatic Mapper) software (Baraldi level dealing with issues of implementation and the details of
2011). Pixels are preclassified by rule-based clustering in color a proper transfer between input and output; and (3) the hard-
space. Segments can be built by grouping neighboring pixels ware level being concerned with the physical realization of the
assigned to the same symbolic (i.e., meaningful) color levels. representation identified in (1) and its processing algorithms
The initial preclassification aims to reach on a semantic level derived in (2). The computational level within visual process-
of (descriptive pre)classes, while leaving the spectral discern- ing is characterized according to Marr (1982) by several stages
ibility as high as possible: depending on the aggregation level, of visual representation of an images content. These stages (or
an 8-bit 6-band Landsat-7 imagery with 256 6 quantization lev- sketches) are considered to proceed in a subsequent manner
els can be, for example, transformed in a preclassification layer providing more and more detailed information about the scene
between 18 and 96 classes. This process lowers the number of being viewed. The primal sketch involves a 2-D description of
quantization levels significantly. While this is desirable from the radiometric behavior of pixels and pixel aggregates and
the perspective of deductive classification, it leads to a pre- their geometrical properties. The pure spectral differentiation of
defined segmentation result (by grouping neighboring pixels gray shades and color tones makes up the basic level (raw pri-
of the same (color) class), that no longer adheres to the prin- mal sketch), whereas the grouping of spectrally like pixels into
ciples of regionalization, as described earlier. Here the second geometrical units represents the full primal sketch. The latter
strategy comes into play. Utilizing the strength of regionaliza- is built upon quasi-homogeneous regions (blobs) or bounding
tion to consider both spatial and spectral similarity simulta- areas of high contrast (contours or edges). The raw primal sketch
neously, multiscale segmentation provides image objects by marks the stage where image primitives such as blobs, edge seg-
spatial classification (Wise et al. 2001) at first hand. These ments, and their low-level descriptions are produced. They are
sub-symbolic segments are then classified according to their then ordered and organized into higher-level place tokens (Marr
spectral and spatial behavior into (higher) semantic classes. 1982) or perceptual chunks (Bruce and Green 1990). This is per-
The building blocks used in the first strategy are (strictly) formed within the process of perceptual organization, which
spectrally homogeneous pixel aggregatesno matter of what aims at logically grouping a perceived pattern and to transfer
spatial characteristicswhile in the second strategy, we have it to meaningful symbolic representation. Touching the issue
image regionsfairly homogeneous, but spatially contiguous of scale, we can consider relatively small place tokens to cause
and scaled to feed the IUS (Lang 2005). The first strategy uses a certain granularity (Julesz 1975), which is a visual effect of a
pre-existing knowledge on (potentially exhaustive) spectral regular distribution of small enough elements.
signatures, and the second requires (empirically determined) When moving from image perception to image interpreta-
parameterization. tion, experience comes into play. Findings in neuropsychology
2016 by Taylor & Francis Group, LLC
302 Remotely Sensed Data Characterization, Classification, and Accuracies
(Spitzer 2000) explain that data processing by our senses is a kind 15.2.3 class Modeling
of vector coding of signals in a high-dimensional feature space.
Experience can be thought of as a certain cluster or nexus in the Multiscale segmentation, for example, realized by the multi-
feature space. When signals are perceived, they are compared to resolution segmentation algorithm (Baatz and Schpe 2000),
this experience nexus, which compare to a model to reach from generates homogeneous image objects in a nested hierarchy.
(1) raw data, through (2) patterns or aggregates of color and form Spectral information is aggregated while loss of detail is mini-
that are structured in various levels, to (3) relationships between mized (Drgut et al. 2014) and higher-level objects internal
object concepts in an image. Still, human perception is likely far heterogeneity is increasing, leading to more functional homo-
from linear (Gorte 1998), and this may suggest that more than geneity in the sense of Spence (1961). Using this strategy, some
one model is used to construct meaning from an image (Lang composite classes (e.g., forest stands composed by tree species
etal. 2004). Image interpretation, when dealing with unfamil- with similar spectral behavior) can be directly delineated by
iar perspective and scale, requires multiobject recognition in a multiresolution segmentation (Strasser and Lang 2015). More
rather abstracted mode, and the interpreter needs to understand often, however, composite classes such as orchard field, mixed
the whole scene. This includes the physical properties of objects arable land, riparian forest, suburban area, and informal settle-
viewed in an image, but also their affordance which refers to an ment, bear a degree of internal heterogeneity that exceeds the
objects values and meanings towards an interpreter. The skilled capability of state-of-the-art segmentation algorithms (Lang
visual interpreter may recognize some features instantly and et al. 2010a). The internal homogeneous building blocks con-
others by matching the visual impression against experience or sist of different land cover, thus being spectrally diverse (e.g.,
examples listed in an interpretation key (Blaschke etal. 2014) grassland and crop patches), and there is no a priori true
The gestalt approach (Wertheimer 1925) has established gen- delineation. Lang etal. (2014) use the term composite geon for
eral principles (factors, laws) that can be observed when certain such target classes, whose composition is scale depended. Body
patterns or figures are examined. Following the Ehrenfels cri- plans are used to represent the targeted composite objects that
terion, a gestalt shows emergent properties. Gestalt theorems are intersubjectively perceived by experts as functional bona
have a strong predictive (though not an explanatory) potential fide units (Smith 1994). Class modeling (Tiede etal. 2010) can
of how we perceive structures and patterns. Based on the fac- be employed to topologically describe spatial constellations of
tor of proximity, single elements within a scene are grouped a set of subunits in a way that image information is structured
together if they are close enough to each other, depending on into hierarchical divisions based on ontology-like rule sets that
the scale of observation. The factor of good gestalt assumes that employ relational features.
simple, well-shaped geometric figures are more readily to be Tiede etal. (2010) use a class modeling approach as a super-
perceived than more complex ones. Both factors suggest that vised regionalization technique based on spectrally homoge-
a manual interpreter tends to group similar elements into a neous elementary units (spectrally bona fide), which implies
larger one and to close lines straight over a gap, something that iterative segmentation and classification steps. The initial seg-
is also reflected by the factor of good continuation (Lang 2005) mentation and preliminary classification of the basic units
(Figure 15.4). is only the first step in the modeling process. Based on these
Factor of good gestalt Factor of good continuation
Group of trees
(e.g., orchard)
1 Pathway
2 3
Adjacent
meadow
Conceptual
Linear feature
boundary Transition
zones
Meadow
FIg u r e 1 5.4 Gestalt theorems applicable to image analysis problems.
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 303
(a)
Fraxinus_excelsior (L2) Populus_balsamifera (L2)
Picea_abies (L2)
meadow
Alnus_Fraxinus (L2) Picea_abies (L2)
Fraxinus_excelsior (L2)
meadow
Picea_abies (L2)
Alnus_Fraxinus (L2) Non_vegetated
Fraxinus_excelsior (L2)
Fraxinus_Picea (L2) Non_vegetated
Picea_abies (L2)
Non_vegetated
Picea_Populus (L2)
Picea_Populus (L2)
Alnus_Fraxinus (L2) Fraxinus_Populus (L2)
Alnus_Populus (L2) Alnus_Fraxinus (L2)
Fraxinus_Salix (L2) Populus_sp (L2) Non_vegetated
(b)
FIg u r e 1 5.5 (a) SPOT-derived land use units to fully validated composite geons; (b) initial forest patch/generalized forest patch.
initial classes, additional expert knowledge (here based on the of limited, well-distinguishable subunits, class modeling aims
mapping key) is formalized in rule sets and/or the specific use at (re-)assembling familiar (i.e., from everyday experience) or
of auxiliary data sets is defined. The result is a cyclic process expert-known (i.e., trained) target units, according to the gestalt
for a generation of composite objects (functionally homoge- principle of good gestalt. Otherwise, our perception works in a
neous, structurally heterogeneous, see Figure 15.5a), overcom- more subtle way, when we intrinsically generalize detail. For
ing the limitations of object delineations using region-based example, when analyzing a piece of (natural) forest, we would
segmentation algorithms. In the described building block case rather perceive quasi-homogeneous forest stands according to
2016 by Taylor & Francis Group, LLC
304 Remotely Sensed Data Characterization, Classification, and Accuracies
their specific composition of prevailing and accompanying tree Keane 1995). Semantic classification and naming are subsequent
species. This can be seen as an intermediary stage between a stages and involve retrieval of functions and object associates
pure texture-based impression (where individual elements are (Lang et al. 2009a). One of the striking capabilities of human
difficult to discern) and a building block composition (where perception is to tell signal from noise, in other words to distin-
elements are few in number and intersubjectively perceivable). guish information against a simplified environment (Bruce and
This is what gestalt theorists would call the factor of granularity. Green 1990). By experience or training, we continuously feed
When fine granularity enables the segmentation algorithm to our implicit knowledge with explicit knowledge derived from
find higher hierarchical boundaries (Strasser and Lang 2015), formal learning situations (e.g., spectral behavior of stressed
the number of elementary units can increase significantly vegetation). Artificial intelligence distinguishes knowledge in
(Figure 15.5b). procedural and structural knowledge. Procedural knowledge
Interpretation keys for mapping and survey are designed by is concerned with the specific computational functions and can
experts and for experts (Lang et al. 2004). Yet an automated be represented by a set of rules. Structural knowledge implies
interpretation of an entire scene in a wall-to-wall manner is the way concepts of a domain are interrelated: in our case, that
still away from being operational. The intrinsic and intuitive means how far links between image objects and real-world
knowledge of a skilled interpreter is hard to force into rule sys- geographic features (Hay and Castilla 2008) are established.
tems. The criterion of being relatively homogeneous governs any It is characterized by high semantic contents and therefore by
mental organization of reality, manual interpretation, or image far more difficult to tackle with. Structural knowledge can be
analysis, whereas it remains a relative measure and a matter of organized in knowledge organizing systems such as semantic
disposition depending on the target scale dimension. networks (Liedtke et al. 1997) as graphical representations, or
more mathematically by, for example, formal concept analysis
(Ganter and Wille 1996). Within image analysis, semantic nets
15.2.4 Validity of object Delineation
and frames (Pinz 1994) offer a formal framework for semantic
and classification
knowledge representation using an inheritance concept (is part
Object identification impliesnext to the cognitive skills of of, is more specific than, is instanceof ). In recent years, ontologies
interpreting color and form and linking them to biophysical have gained a lot of intention to formalize knowledge. Semantic
propertiesa grasp of the relevance of the extracted informa- webs allow for a representation of structural knowledge in dif-
tion for a certain purpose. Object validity (Lang etal. 2010a) has ferent semantic levels, that is, hierarchical complexities as a top-
been defined as the degree of fitness of object delineations for level ontology. Ontologies are increasingly used in the GIS and
operational tasks in a policy context. Comprising object repre- remote sensing community, including geospatial knowledge rep-
sentations in both the bona fide as well as the fiat domain, OBIA resentation (De Martino and Albertoni 2011), geographic infor-
poses new challenges for evaluating object validity in an opera- mation retrieval (Lutz and Klien 2006), image processing and
tional context (geon concept; also see Chapter 22, Volume I). analysis (Tnjes et al. 1999), semantic sensor networks (Kuhn
According to Smith (1994), objects can be differentiated into (1) 2009), and geo-data interoperability (Reitsma et al. 2009). The
concrete, tangible objects with a visible physical boundary (bona Semantic Web for Earth and Environmental Terminology ontol-
fide) and (2) objects that lack a physical border, thus not visible ogies, for example, provides an upper-level ontology for Earth
in the landscape (fiat). Examples are administrative boundaries system science comprising thousands of terms from the Earth
or artificial, concept-related object boundaries constructed by system science domain using OWL (Ontology Web Language)
human fiat. In remote sensing, we treat a spectrally homogeneous (Raskin and Pan 2003).
image segment as a bona fide object. This, of course, implies a While in visual interpretation, cognitive processes run more
scale issue. As we leave crisp bona fide objects behind, the binary or less simultaneously, within OBIA, this process is split and
decision between correct and false labeling gets blurred, while organized along a procedural processing line, and these com-
the use of, for example, fuzzy rule sets (Zadeh 1965) only partly partments can be decomposed and controlled individually (Lang
solves this. The more complex or concept-related a specific cat- etal. 2010) (see Figure 15.6).
egory is, the more likely a binary assessment will fail. Thus, Lang Machinebased knowledge representation and hierarchy
etal. (2010a) propose to use the terms appropriateness or validity theory (i.e., the principal understanding of decomposability of
to accommodate the fact that such quality criteria are more a complex scene content) complement each other by enriching
continuum with certain thresholds to be set, instead of a binary automated image analysis with geospatial scale concepts. The
decision between right or wrong. rulebased intelligence of a production system can be enriched
Object recognition has been described by theoretical con- by learning algorithms, empowering the classification system to
cepts ranging from template matching theories over feature improve itself (with increasing number of classification tasks),
analysis to structural descriptions (Bruce and Green 1990), the but to some degree obscuring the transparency (and potentially
latter being used to describe the components of a configuration the transferability) of the rule base as such. The spatial relation-
and to make explicit the arrangement of these. The matching of ships among (basic) objects are useful in modeling higherlevel
extracted visual information with the stored structural descrip- (i.e., composed) object classes. These have to match with the
tion is referred to as perceptual classification (Eysenck and truth in the sense of matching the epistemological significance
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 305
Quality Validation
(accuracy,
stability,
transferability,
level of
automation)
Spatial analysis
Epistemological
significance
Object fate
Decomposability Pattern recognition
Hierarchy theory
Machine-based image analysis
(automated)
(multi-scale) Production system,
Classification
Segmentation rule bases
Image object
Scaled representation
(generalization) Image
Pixel aggregate understanding
image region
Visual perception/
human expert/
user instance
FIg u r e 15.6 Object validity in object-based image analysis (Lang etal. 2010). Starting from pixel aggregates (image regions) as initial products
of image portioning, a set of routines may turn suitable segments into image objects that (1) are considered appropriate for the respective context
of use and (2) are adhering to general principles of image segmentation such as scale domain, spatial coherence, nonoverlap, and the like. (Adapted
from Blaschke, T. etal., Int. J. Photogramm. Remote Sens., 87(1), 180, 2014.)
that underlies the classification problem. Simple quality criteria external process of confirming the assigned category. The vali-
(like any accuracy measure) should be applied under consider- dation of boundary delineation as such is scale dependent, as
ation of the process lineage of the object provision and a gen- well as policy related. It includes the issue of degree of freedom
eral understanding on the epistemological nature of the target in delineation when it comes to the grouping of elementary
objects (Lang etal. 2010b). units. Finally, the automation of the delineation as such is
Table 15.1 summarizes particularities of validation tasks characterized.
within OBIA for bona fide objects and composite. Specific
details are discussed by Lang etal. (2010a). Main methodologi-
15.2.5 Multiple-Stage Validation
cal challenges are the ones attached to optimizing the process
of OBIA-specific automation and machine assistance within Semiautomated classification of complex land use/cover units
each category. The issue of reference data varies between dif- faces a high number of degrees of freedom (Lang etal. 2010a).
ferent types of object classes. Label verification refers to the Therefore, the validation of modeled composite classes stands
TABLe 1 5.1 Aspects of Object Validity
Bona Fide Objects Modeled Composite Objects
Example(s) Land cover classes, for example, meadow, field, forest Biotope complexes, for example, mixed arable land and grassland area
Main methodological challenge Scale-specific delineation, boundary generalization Meet requirements in terms of appropriateness and matching key
Real-world reference Existing and recordable in field Perceived by expert vision, agreement with mapping key, and existing
planning units (here cadastral data)
Label verification Binary or fuzzified In stages, appropriateness as a term to be operationalized
Boundary validation Scale-dependent, low degree of freedom Expert-based, high degree of freedom
Policy driven On an elementary level: yes Yes, high expectations in EO-based techniques
Delineation Automated Automated, partly manual regrouping
Source: Modified from Lang, S. etal., J. Spat. Sci., 55(1), 9, 2010a.
2016 by Taylor & Francis Group, LLC
306 Remotely Sensed Data Characterization, Classification, and Accuracies
Improved Reliable elementary
Elementary elementary units, quality assured
units units
2
Final
Biotope 4
Reliable results,
quality assured
3 usability assured
Stratified random
point-based Field evaluation OFA
accuracy boundary
assessment assessment
Final
Preliminary biotope
biotope complexes
complexes
FIg u r e 1 5.7 Multistage validation of composite classes. (Adapted from Tiede, D. etal., Photogramm. Eng. Remote Sens., 193, 2010.)
between traditional methods of quantifying accuracy and 15.3 object Fate
more qualitative validation measures (matching scale and rela-
tionship patterns). Validating OBIA class modeling faces the Image objects as real-world objects are subject to change.
problem of which step in the cyclic approach to validate. The Changes may occur more gradually (varying quality) or more
validation sequence, as proposed by Tiede etal. (2010), includes abruptly (altering category or label). At the same time, image
verification, improvement, quality assurance, reliability, and objects may change their extent, location, or composition. Even
even usability to ensure the product fully satisfies the user topological relationships between neighboring objects may be
requirements. That results in a four-stage validation procedure modified over time. Object-based change detection (OBCD)
(Figure 15.7): strives to assess such changes in terms of type, intensity, size,
and shape, as well as horizontal and vertical relations of image
Statistical point-based accuracy assessment based on
objects. Thereby, OBIA faces the challenge to tell real changes
stratified random sampling (Congalton and Green 1998)
from generalization effects or any kind of noise or disturbance
to validate the elementary (bona fide) units, which were
occurring in the imagery. A similar problem arises in object-
based on spectral homogeneous units.
based accuracy assessment (OBAA) when evaluating the quality
A first expert-based validation cycle including both the
(accuracy, robustness, etc.) of OBIA classifications, as the same
composite classes (in this case biotope complexes) with
effects may render an OBIA result and a reference data layer dif-
spectral heterogeneity and their dependent basic units.
ficult to compare. The concept of object fate analysis (OFA) pro-
This was part of a hybrid modeling approach (on-screen
vides a strategy to deal with such challenges.
expert assessment).
A second expert-based validation cycle included the qual-
ity of the modeling approach not only regarding the given
15.3.1 From Static to Dynamic
mapping key as compared to field mappings by domain
change concepts
experts, but also considering geometric accuracy of the
delineated boundaries. Geographic features, when observed over time, are subject to
Evaluation of inaccuracies introduced by the policy- changeand so are image objects that represent them. So while
demanded matching of delineated biotope complex it seems a straightforward task to extend established change
boundaries to cadastral data (scale-gap). Object-based fate detection methods to the object-based domain, there are some
analysis (OFA, see later) has been applied on individual additional challenges and specifics to what we may call object-
object level. based change detection (OBCD) (Lang 2008, Chen etal. 2012).
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 307
Change, in fact, is a more challenging issue in OBIA than in Image data of different spatial resolutions are combined
pixel-based analysis, due to the reasons already discussed with and/or different segmentation algorithms were employed.
object validity: an image object can change in label and extent, Segmentation results are compared to visual interpreta-
usually in both: the case that a particular geographic feature (a tions and manual delineations, applying different scales
lake, a forest, a town) changes its category within exactly the and outline complexities (partly ignoring subtle changes
same boundary is rather rare. So the static raster we apply in a due or outliers due to generalization effects).
per-pixel analysis is no longer valid. There is not the single pixel
This has implications on the assessment of OBIA outcomes,
converted from meadow to sealed, but the expanding settlement
for example, when comparing a segmented and classified
that eats up grassland patches in the surrounding. Blaschke
result with a visual interpretation map. Here the issue of spa-
(2005) has proposed a typology of object geometry changes
tial disagreement or spatial error (Radoux and Defourny 2008)
considering different categories (existence-related, size- and
arises, as object boundaries may mismatch by simply applying
shape-related, and location-related changes). In reality, we are
a diverging (or no) generalization strategy. Obviously, the qual-
facing a combination of all of these basic categories of geomet-
ity of the segmentation, that is, segmentation goodness (Clinton
ric changes (Lang et al. 2009a). In addition, when an object is
etal. 2010, Hernando etal. 2012), is a critical factor in this pro-
shrinking or expanding, the surrounding is inevitably affected.
cess. Van Coillie et al. (2008) proposed a Purity Index mea-
Raza and Kainz (2001) include subdivision and amalgamation
suring the common area between segmentation objects and
of objects in their list of spatiotemporal characteristics of par-
reference objects. Vector-based measures (area-based, location-
cels. Other studies specifically investigate topological relations
based, and a combination of both) were analyzed by Clinton
to mathematically evaluate the identity of two object hypoth-
et al. (2010), calculating the similarities between segments
eses when comparing two different data sets (Straub and Heipke
and training objects. Two goodness measures for optimal
2004). They present eight topological relations being reducible to
segmentation were proposed by Johnson and Zhixiao (2011):
the principal relationships: disjoint, equal, overlap, and contain-
global intrasegment homogeneity (the variance of each object)
ment. The latter (including touch and cover) utilizes a margin
and intersegment heterogeneity (how similar a region is to its
concept.
neighbors).
Practically, when changes are detected in multitemporal
In the following section, we shall have a closer look into how
imagery, such spatial implications are often neglected so that
to analyze differences in objects by investigating spatial rela-
established methods for spatially explicit object-based change
tionships among corresponding objects. We thereby abstract
analysis are still rare (Schpfer etal. 2008). Map-to-map com-
from the issue whether this is a product of object transition
parisons using raster overlay techniques are site specific, but not
(change over time) or an outcome of different object represen-
object specific; they refer to changes in pixels but not of objects.
tations or delineations. Since spatial relations are various and
Vector overlays with physical intersections of the boundaries
appear in reality in different combinations, there is a demand
produce complex geometry with sliver polygons (Chen et al.
for ready-to-use solutions that are able to structure and catego-
2012). Visual comparisons, on the other hand, are powerful but
rize these.
subjective and time consuming (Schpfer etal. 2008).
Next to the simultaneity of change in label and extent, there
are other reasons why there is no straightforward solution for
15.3.2 Application Scenario #1: object-
object-based change. Image objects can vary due to different
Based information Update
representations, while those geographic features they represent
remain constant. A forest patch may be represented in differ- The need for updating existing information through remote
ent complexities in terms of the outer limit boundaries (Weinke sensingbased technology is undoubted and still increases
et al. 2008) and its internal structure, and the same applies to with the availability of especially high spatial resolution data.
other features like settlements, wetlands, and the like. The In operations workflows, the update of information from
question arises whether a lack of spatial congruence of corre- image data is hampered by the lack of full integration into
sponding image objects is caused by data mismatches, different existing geo-spatial infrastructures. This issue is discussed
representations, or real change. More specific, there are several for decades (Ehlers 1990), and the integration of GIS and
reasons for spatially inconsistent features, which are inherent to remote sensing is progressing. But some problems are still not
the representation itself (Lang etal. 2009a) and not related to the solved for many cases. For example, the operational workflow
geographic features under concern (pseudo-changes): for a planning authority to integrate aeven very accurate
and based on VHSR dataland cover classification in exist-
Image misregistrations due to poor spatial referencing ing vector data sets (e.g., cadastral data) is still limited, if the
may cause shifts in corresponding objects and pseudo- boundaries of the land cover classes do not match the spatial
boundaries when overlaying those. characteristics of the geo-data in use. The geometric match-
Differing image characteristics of multitemporal or mul- ing of the different data sets is an issue of scale/resolution,
tiseasonal data sets in terms of viewing angles, sun illu- the related issue of boundary complexity, but also a principal
mination, etc. problem of conversion between different data models (usually
2016 by Taylor & Francis Group, LLC
308 Remotely Sensed Data Characterization, Classification, and Accuracies
raster to vector) (Merchant and Narumalani 2009). To over- are grouped into three categories. Boundaries refer to existing
come the latter, expectations have been placed on the ability polygon or line data that are used for adaptive parcel-based
of segmentation-based approaches to overcome the problem segmentation. A specific case of using cadastre information
of mismatches that are due to the use of different data models for generating polygon data sets has been dubbed adaptive per
(raster vs. vector) by delivering GIS-ready information (Benz parcel segmentation by Tiede etal. (2007).
etal. 2004) from remote sensing data. However, as Tiede etal. In a study for a planning association (Verband Region
(2010) pointed out, such a full integration of image-derived Stuttgart, an association of local authorities in the Stuttgart
spatial information is not a trivial task: the problem is resid- region, Germany), both problems were addressed (Tiede et al.
ing lesser in the rastervector conversion itself, but more 2010). The requirements to support regional planning based
in the integration of information from different (spatial) on EO data encompassed the delineation of biotope complexes
resolutions. Coarser image resolution than the existing geo- as target units. To be fed into the existing data structure of the
data to be updated results in objects with boundaries often planning authority (cadastre-conform), the biotope complexes
not matching existing ones. Also the contrary, very high should preserve boundaries of the digital cadastre map in cases
resolution imagery (VHRI) imagery as information source where a change in biotope complex type is also reflected in the
to update lower-resolution (i.e., coarser scale, generalized) cadastral data, but remove boundaries within the same biotope
vector data results in objects often too complex regarding complex. In addition, biotope boundaries not reflected in the
shape and boundaries. Smoothening and generalization rou- cadastral data need to be integrated.
tines need to be applied, which can leadaccording to their As a trade-off between costs and quality, image data of 5 m
nondeterministic operationto unsatisfactory match with GSD were used for the analysis, which introduced a certain
existing boundaries. Refer to Walter and Fritsch (1999) and scale-gap, especially for newly introduced boundaries, not yet
Buthenuth et al. (2007) for a discussion about matching of reflected in the cadastral map. A twofold strategy was applied
different (vector) data models and the integration of raster to tackle the problem, by using (1) the digital cadastral data
and vector data. as spatial constraints in the initial object building step (using
To summarize, specific challenges are faced when existing an adaptive per-parcel approach) and (2) the adjustment and
boundaries are considered in a process of information update. validation of newly introduced rasterized boundaries by com-
We shortly presented a collection of instances on how to per- parison with the target vector geometry. In the first step, we
form OBIA for monitoring purposes, aiming at the provision flagged either boundaries of the digital cadastral map to be
of up-to-date information under the explicit consideration preserved if they were also representing biotope complex
of existing geo-spatial data. In Figure 15.9 the different cases boundaries or boundaries to be removed if they are within
Biotope
complexes Cadastral conform final result
Cadastral
(redundant cadastral outlines
data
within biotope complexes
removed)
Newly introduced
boundaries
Cadastral conform borders
between biotope complexes
FIg u r e 15.8 Cadastre-conform extraction of biotope complex units from EO data: Cadastral data are retained, where biotope complexes are
changing; redundant cadastral boundaries within the same biotope complex are eliminated.
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 309
the same biotope complex. By this, the scale-gap of the image 15.3.3 object Fate Analysis
data compared to the existing data is avoided. In the second
stepto integrate newly introduced boundaries from the OFA is a method proposed by Lang (2005) and discussed by
image-based biotope complex delineationit was necessary Schpfer et al. (2008) for investigating spatial relationships
to perform an adjustment in regard to the same generaliza- between corresponding objects in two different representations.
tion/smoothness level of the existing data (instead of a raster- Rephrased and put from an individual image objects perspec-
ized boundary). The latter was achieved using a combination tive, this would reflect the fate of this particular object and its
of established GI-routines and additionally programmed solu- representation (see Figure 15.10). Object fate may be caused by
tions. Figure 15.8 schematically shows the three occurring real change captured in data from different points of time. Or
problems that were addressed to reach the requirements for an otherwise, it may root in differences in object generation by
accurate and compatible data set for administrative purposes: using segmentation algorithms or visual analysis, heterogeneous
(1) The replacement of biotope complex boundaries with cor- data reference data sets from other sources (Schpfer etal. 2008),
responding cadastral boundaries by considering a spatial dis- etc. In reality, we often face a combination of change and repre-
placement tolerance (scale-gap) according to the pixel size of sentation-induced divergences.
the image data. (2) Merging of cadastral polygons within a OFA has been implemented in a tool called Landscape
biotope complex (removing of cadastral boundaries). (3) The Interpretation Support Tool (LIST) (Weinke etal. 2007, Lang etal.
introduction of new boundaries (based on an OBIA-based bio- 2009a), as an extension for ESRIs ArcGIS software. Following
tope complex delineation), not reflected in the cadastral data the concept of parent and child relationships, two vector layers are
set but which represents changes of biotope complexes. The used to represent the specific fate of corresponding objects. Thereby
newly introduced boundaries were finally smoothed and gen- only the geometry is considered, assuming that corresponding
eralized using standard GI procedures (Douglas and Peucker objects retain their label (see later for an extension of OFA to the
1973, Bodansky etal. 2002). thematic dimension). To overcome spatial uncertainty in image
High
Original boundary Retained
boundary
Internal homogeneity
Low
Introduced
boundary
High Removed
boundary
FIg u r e 15.9 Illustration of differentiated boundary treatment. Additional cases (not shown): boundaries are generalized, fractalized, shifted,
found in the surroundings. Case 1: existing boundaries are retained (limited internal heterogeneity); case 2: new boundaries are introduced
internal variance larger than (given) threshold; and case 3: boundaries are removedinternal variance larger than (given) threshold.
2016 by Taylor & Francis Group, LLC
310 Remotely Sensed Data Characterization, Classification, and Accuracies
Same representation
T0 Two different time slots T0
and T1 with the same
Target layer T1 representation, for example R0
OBCD: T1 (present time slot) R0 Constant timing
OBAA: R1 (classification results)
Two different representations
R1 R0 and R1 at the same time
Reference layer slot, for example T0
OBCD: T0 (past time slot) T0R1 Reality: combination
OBAA: R1 (e.g., field mapping,
manual interpretation) T1R1 Different representations at
different time slots
FIg u r e 1 5.10 Object fate considering aspects of different status (change) and representation.
B: Example
A(S1)
Inside I Inside II Expanding
pAIIa
C2
similar to
equal
pAIa
C1 pAIb pAIIIc
similar to Reference (categories) Classification (classes)
disjoint
A,B,C,D.(T) a,b,c,d(t)
Outside I Outside II Invading
(a) (b)
FIg u r e 15.11 Spatial and thematic assessment in extended OFA. (a) OFA categories depending on mutual spatial relations of two objects (out-
lined, shaded). (b) Differentiated OFA categories considering object classification. (Modified from Hernando, A. etal., Int. J. Appl. Earth Obs. Geoinf.,
19, 214, 2012.)
objects, an error band (positive and negative buffer around the Hernando etal. (2012) extended OFA to the thematic dimension
object boundary) is used. The size of the buffer is specified either by assessing results from a detailed OBIA classification of forest
manually or as a function relative to object size. A virtual overlay stands as compared to an existing forest stand mapping derived
characterizes spatial relationships via the relative positions of cen- from visual air-photo interpretation. By combining spatial and the-
troids, retaining the original object boundaries. This distinguishes matic aspects (see Figure 15.11b), an OFA matrix was established.
between weak overlap (similar to disjoint) and strong overlap (simi- First, each classified object (classes a, b, c, d, t) was compared
lar to identical). The resulting categories of object relationships are to the corresponding reference object with ST the overlapping area
discussed by Schpfer etal. (2008) and shown in a simplified ver- for each of the reference categories (A, B, C, DT) and STotal the
sion in Figure 15.11. Object relationships are divided into either C1 total reference area for all w reference categories (A, B, Cw.T).
cluster or C2 cluster by evaluating whether the centroid of the clas- The matrix is established by listing all relationships between all cat-
sification object fell inside the reference object (upper row: inside egory/class shares for each of the five relationship types. Each cell
or good I, inside II, expanding) or not (lower row: outside or not of this matrix is filled by a value r TCt as the sum of the relative areas
interfering I, outside II, invading). There are six relationships; five of the t classified objects considering an OFA type.
of them are relevant for further analysis when two correspond-
15.3.4 object Linking
ing objects at least partly overlap. The relationship not interfering
I assumes that objects do not correspond any longer. Two stability Traditional automated change detection approaches are lim-
measures, object (or offspring) loyalty and interference (Lang etal. ited if, for example, co-registration errors are present or if an
2009a), were derived based on the relationships between objects. object changed over time (changes especially in size/form or
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 311
Pre-disaster Post-disaster
Comparison of extracted Area: 40 m2
shadows of a building. Due to Shape Index: 2.53
geometrical shifts between
the pre- and post-disaster
images, the comparison is
based on area and form
descriptors to compare
shadow objects which are
located within 10 10 pixel
buffer in northwest direction Area: 46 m2
(symbolized by the dashed Shape Index: 2.76
line on the right).
FIg u r e 15.12 Conceptual illustration of the object-linking approach. Here, comparing extracted shadow objects from an earthquake pre- and
post-disaster imagesdespite a geometrical shift between the imagesto indicate if the shadow is considered to be still in existence (here, no
indication of destruction of the building since the area and form descriptors are quite similar). (Adapted from Tiede, D. etal., Photogramm. Eng.
Remote Sens., 9, 933, 2011.)
it disappeared). Such change detection approaches, looking at Figure 15.12 shows the implementation of an object-linking
pixel value changes only neglecting potential geometrical shifts, approach for shadows casted by buildings, before and after an
require precise coregistration of multitemporal images. (Lu etal. earthquake (Tiede etal. 2011). The class-constrained object-by-
2004). Following a study about the robustness of different pixel- object comparison is able to compare objects of different clas-
based change detection algorithms by Sundersan et al. (2007), sifications including geometrical buffering in x or y directions,
the registration error between the images should not exceed 0.2 that is, a geometrical shift between the dates can be compensated
pixels (RMS) for satisfactory performance of the algorithms. (or to track moving objects). Since such a linking of objects
When comparing objects created from different sensor data including the neighborhood and not only the direct spatial
or time epochs. we are able to compensate to some degree the overlapcan lead to 1:n or even n:n matches, not only class con-
requirement for image co-registration. This includes a gen- straints (matching of objects of the same classes only) are impor-
eralization issue in the object generation process, comparing tant, but also the comparison of geometrical object properties
statistical values per object instead of single pixels, compar- (e.g., form and size). By this, the degree of change or the degree
ing also geometric properties). Nevertheless, a strict vertical of movement can be estimated. In the following, an application
object hierarchy, as often used in OBIA frameworks, hampers scenario is described using an object-linking approach in the
sophisticated analysis, because of the intersection of boundar- case of automated rapid information extraction after a disaster
ies hampers treating multitemporal data similar as kind of a (here damage indication after an earthquake)
pseudo-hierarchy.
Human interpreters usually compensate positional shifts
15.3.5 Application Scenario #2: Rapid
between data sets when recognizing similar objects. Such a loose
information extraction
coupling can be mimicked using an object-linking approach
to track objects over time or to analyze changes per object Specific automated image analysis techniques based on VHR
instead of, for example, intersections of objects (Tiede et al. images, for example, satellite images (QuickBird, GeoEye, and
2011, Hofmann and Blaschke 2012). Object linking relies on GI Worldview), have reached a status of maturity to be utilized in
methods adopted in a topologically enabled OBIA framework It application domains crucially depending on reliability and time-
establishes spatial relationships between objects incorporating liness (Lang etal. 2010a). Several studies have shown that differ-
shifts introduced through positional errors using (multi)direc- ent approaches can lead to successful applications in the field of
tional buffering and can be enriched by the comparison of geo- damage assessment using change detection methodologies, for
metric object properties (like form and size). example, comparing of pre- and post-event images. Gusella etal.
2016 by Taylor & Francis Group, LLC
312 Remotely Sensed Data Characterization, Classification, and Accuracies
(2005) demonstrated the use of QuickBird data for quantifying often a lack of high-quality pre- and post-disaster VHSR imag-
the number of buildings that collapsed after the Bam earthquake ery (Rathje and Adams 2008). After the Haiti earthquake in
in Iran, 2003. Also using QuickBird data, Pesaresi etal. (2007) January 2010, the availability of VHSR satellite data was out-
showed that very high accuracies allow for rapid damage assess- standing compared to previous disasters. Data have even been
ment of built-up structures in tsunami-affected areas based on a provided to the public for free (also outside the International
multicriteria recognition system within a restricted set of gen- Charter Space and Major Disasters) by companies like Google
eral assumptions (e.g., built-up structures cast a shadow, and or DigitalGlobe. One of the effects was that a tremendous
collapsed built-up structures no longer cast a shadow, and they amount of crowd-sourced information was provided in the
leave debris on the ground). Vu and Ban (2010) developed a con- aftermath of the earthquake. Nevertheless, these data sets are
text-based automated approach for earthquake damage mapping often not well suited for automated analysis methods, due to
relying on debris area identification. In addition, the calculation reasons such as lacking metadata, non-documented or insuf-
has been speed-optimized by parallel processing implementa- ficient preprocessing, and missing spectral NIR bands.
tion showing good results for the test area, after the Sichuan A strategy for automated extraction of damage indication
earthquake in China, 2008. Ehrlich et al. (2008) and Brunner from very highspatial resolution satellite imagery was presented
et al. (2010) demonstrate the usefulness of applications with for the Haitian town of Carrefour after the January 2010 earth-
the additional use of synthetic aperture radar data for damaged quake (Tiede et al. 2011). Damaged buildings are identified by
building assessment also for the Sichuan earthquake area. Three- changes in their shadows compared in pre- and post-event data.
dimensional information has been taken into account to detect The approach builds on OBIA concepts to extract the relevant
damaged built-up structures after earthquakes by Turker and information about damage distribution (collapsed buildings).
Cetinkaya (2005) with the help of stereo-images, but for rapid The lacking quality of the pre- and post-disaster imagery usually
damage assessment shortly after an earthquake, the necessity of available directly after an event (in this case, the shift between
capturing a pair of fitting stereo-images is usually not feasible. A the multitemporal images was up to 5 m and varied through-
broad overview of automated techniques for earthquake damage out the images) was bypassed using the described object-linking
detection is given by Chini (2009) and Rathje and Adams (2008). approach between objects extracted from the different images.
Still, such approaches are rarely applied on an operational This approach allows to overcome sub-pixel co-registration of
level in the relief phase shortly after a disaster. One reason images (especially if the image source and the conducted pre-
might be that the process critically depends on a good qual- processing steps are not known) in rapid information extraction
ity of the preprocessed data, that is, mainly a high reliabil- workflows as in the context of disaster events (see Figure15.13).
ity of the image (co)registration. For most of the automated For the area of Carrefour, the new methodology produced posi-
change detection algorithms, this is a crucial point (Lu et al. tively validated results in acceptable processing time and was,
2004, Sundersan et al. 2007). Right after a disaster, there is to our best knowledge, the only automated damage assessment
Pre-images
Post-images
FIg u r e 1 5.13 Object linking for relating collapsed buildings.
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 313
method that delivered relevant results to requesting relief organi- Bischof, H., W. Schneider, and A. Pinz. 1992. Multispectral clas-
zations in the first days after the Haiti earthquake. sification of Landsat-images using neural networks. IEEE
Transactions on Geoscience and Remote Sensing 30: 482490.
15.4 conclusions Blaschke, T. 2005. Towards a framework for change detection based
on image objects. Gttinger Geographische Abhandlungen.
In this chapter, we have discussed implications of geodata In Remote Sensing and GIS for Environmental Studies,
integration in the context of OBIA and geospatial informa- S. Erasmi, B. Cyffka, and M. Kappas (eds.). University of
tion update from an interdisciplinary, applied perspective. We Gttingen, Gttingen, Germany, p. 113.
have shown how the advancement of feature recognition and Blaschke, T. and J. Strobl. 2001. Whats wrong with pixels? Some
advanced image analysis techniques facilitates the extraction recent developments interfacing remote sensing and GIS.
of thematic information for decision support. OBIA techniques Zeitschrift fur Geoinformationssysteme 14(6): 1217.
were discussed in light of transforming and reconstructing Blaschke, T. et al. 2007. GMES: From research projects to
image scenes into ready-to-use geospatial information matching operational environmental monitoring services. ISPRS
the conceptual understanding and the practical needs of users. Workshop on High-Resolution Earth Imaging for Geospatial
This implies issues of scale, generalization, as well as representa- Information, Hannover, Germany.
tion (including visualization); thus, we subsumed the entirety of Blaschke, T. etal. 2014. Geographic object-based image analysis:
aspects to be considered in this process, under the term object A new paradigm in remote sensing and geographic infor-
validity. Since image objects mimicking real-world objects are mation science. International Journal of Photogrammetry
subject to change, OBCD was reviewed in terms of type, inten- and Remote Sensing 87(1): 180191.
sity, size and shape, as well as horizontal and vertical relations of Bodansky, E., A. Griboy, and M. Pilouck. 2002. Smoothing and com-
image objects. OBAA faces similar challenges when evaluating pression of lines obtained by raster-to-vector: Conversion.
the quality (accuracy, robustness, etc.) of OBIA classifications. Graphics Recognition: Algorithms and Applications 2390:
The concept of OFA and the object linking approach provide 256265.
strategies on how to deal with such challenges. We have rounded Bruce, V. and P. R. Green. 1990. Visual Perception. East Sussex,
up the chapter with two representative application scenarios to U.K.: Lawrence Erlbaum Associates.
illustrate the maturity of the discussed conceptual items in a Brunner, D., G. Lemoine, and L. Bruzzone. 2010. Earthquake
practical context. damage assessment of buildings using VHR optical and
SAR imagery. IEEE Transactions on Geoscience and Remote
References Sensing 48(5): 24032420.
Buthenuth, M., G. Gsseln, M. Tiedge, C. Heipke, U. Lipeck, and
Allen, T. F. H. and T. B. Starr. 1982. Hierarchy. Chicago, IL: M. Sester. 2007. Integration of heterogeneous geospatial data
University of Chicago Press. in a federated database. ISPRS Journal of Photogrammetry
Aschbacher, J. 2002. Monitoring environmental treaties using and Remote Sensing 62(5): 328346.
earth observation. In VERTIC Verification Yearbook. Campbell, J. B. 2002. Introduction to Remote Sensing. New York:
London, U.K.: VERTIC, pp. 171186. The Guilford Press.
Baatz, M. and A. Schpe. 2000. Multiresolution Segmentation: An Chen, G., G. J. Hay, and B. St-Onge. 2012. A GEOBIA frame-
Optimization Approach for High Quality Multi-Scale Image work to estimate forest parameters from lidar transects,
Segmentation. Salzburg, Austria: Wichmann Verlag. Quickbird imagery and machine learning: A case study
Baraldi, A. 2011. Satellite Image Automatic Mapper (SIAM). in Quebec, Canada. International Journal of Applied Earth
A turnkey software button for automatic near-real-time Observations and Geoinformation 15: 2837.
multi-sensor multi-resolution spectral rule-based prelimi- Chini, M. 2009. Earthquake damage mapping techniques using
nary classification of spaceborne multi-spectral images. SAR and optical remote sensing satellite data. In Advances in
Recent Patents on Space Technology 2011(1): 81106. Geoscience and Remote Sensing, G. Jedlovec (ed.). InTech
Baraldi, A. and L. Broschetti. 2012. Operational automatic remote Open Access Publisher, pp. 269278.
sensing image understanding systems: Beyond Geographic Clinton, N., A. Holt, J. Scarborough, L. Yan, and P. Gong. 2010.
Object-Based and Object-Oriented Image Analysis (GEOBIA/ Accuracy assessment measures for object-based image
GEOOIA). Part 1: Introduction. Remote Sensing 4: 26942735. segmentation goodness. Photogrammetric Engineering and
Belgiu, M., B. Hofer, and P. Hofmann. 2014. Coupling formalized Remote Sensing 76(3): 289299.
knowledge bases with object-based image analysis. Remote Congalton, R. G. and K. Green. 1998. Assessing the Accuracy of
Sensing Letters 5(6): 530538. Remotely Sensed Data: Principles and Practices. Boca Raton,
Benz, U., P. Hofmann, G. Willhauck, I. Lingenfelder, and M. FL: Lewis Publishers.
Heynen. 2004. Multi-resolution, object-oriented fuzzy De Martino, M. and R. Albertoni. 2011. A multilingual/multi-
analysis of remote sensing data for GIS-ready information. cultural semantic-based approach to improve data sharing
ISPRS Journal of Photogrammetry and Remote Sensing 58: in an SDI for nature conservation. International Journal of
239258. Spatial Data Infrastructures Research 6: 206233.
2016 by Taylor & Francis Group, LLC
314 Remotely Sensed Data Characterization, Classification, and Accuracies
Di Gregorio, A. and L. J. M. Jansen. 2005. Land Cover Classification Lang, S. 2005. Image objects and landscape objects: Interpretation,
System (LCCS): Classification Concepts and User Manual. Rome, hierarchical representation and significance. Salzburg
Italy: Food and Agriculture Organization of the United Nations. University, Salzburg, Austria.
Douglas, D. H. and T. K. Peucker. 1973. Algorithms for the reduc- Lang, S. 2008. Object-based image analysis for remote sensing
tion of the number of points required to represent a digi- applications: Modeling realityDealing with complex-
tized line or its caricature. Cartographica 10(2): 112122. ity. In Object-Based Image AnalysisSpatial Concepts for
Drgut, L., O. Csillik, C. Eisank, and D. Tiede. 2014. Automated Knowledge-Driven Remote Sensing Applications, T. Blaschke,
parameterisation for multi-scale image segmentation on S. Lang, and G. J. Hay (eds.). Berlin, Germany: Springer,
multiple layers. ISPRS Journal of Photogrammetry and pp.328.
Remote Sensing 88: 119127. Lang, S., F. Albrecht, S. Kienberger, and D. Tiede. 2010a. Object
Ehlers, M. 1990. Remote sensing and geographic information validity for operational tasks in a policy context. Journal for
systems: Towards integrated spatial information process- Spatial Science 55(1): 922.
ing. IEEE Transactions on Geoscience and Remote Sensing Lang, S., C. Burnett, and T. Blaschke. 2004. Multi-scale object-
28(4): 763766. based image analysis: A key to the hierarchical organisation
Ehrlich, D., H. D. Guo, K. Molch, J. W. Ma, and M. Pesaresi. 2008. of landscapes. Ekologia 23(Suppl.): 19.
Identifying damage caused by the 2008 Wenchuan earth- Lang, S., S. Kienberger, D. Tiede, M. Hagenlocher, and L.
quake from VHR remote sensing data. International Journal Pernkopf. 2014. GeonsDomain-specific regionalization
of Digital Earth 4: 309326. of space. Cartography and Geographic Information Science
Eysenck, M. W. and M. T. Keane. 1995. Cognitive Psychology: A 41(3): 214226.
Students Handbook. East Sussex, U.K.: Psychology Press. Lang, S., E. Schpfer, and T. Langanke. 2009a. Combined object-
Ganter, B. and R. Wille. 1996. Die Formale Begriffsanalyse. Berlin, based classification and manual interpretationSynergies
Germany: Springer. for a quantitative assessment of parcels and biotopes.
Gibson, J. J. 1979. The Ecological Approach to Visual Perception. Geocarto International 24(2): 99114.
Boston, MA: Houghton Mifflin. Lang, S., D. Tiede, F. Albrecht, and P. Freder. 2010b.
Gorte, B. 1998. Probabilistic Segmentation of Remotely Sensed Automated techniques in rapid geospatial reporting
Images. ITC Publication Series, Twente, the Netherlands. Issues of object validity. In VALgEO 2010, Ispra, Italy,
Gusella, L., B. J. Adams, G. Bitelli, C. K. Huyck, and A. Mognol. C. Corbane, D. Carrion, M. Broglia, and M. Pesaresi
2005. Object-oriented image understanding and post- (eds.), pp. 6575.
earthquake damage assessment for the 2003 Bam, Iran, Lang, S., D. Tiede, D. Hlbling, P. Freder, and P. Zeil. 2009b.
earthquake. Earthquake Spectra 21(S1): 225238. Conditioning land-use information across scales and bor-
Hay, G. J. and G. Castilla. 2008. Geographic object-based image ders. In Geospatial Crossroads @ GI_Forum 09, A. Car,
analysis (GEOBIA): A new name for a new discipline. G. Griesebner, and J. Strobl (eds.). Heidelberg, Germany:
In Object-Based Image Analysis: Spatial Concepts for Wichmann, pp. 100109.
Knowledge-Driven Remote Sensing Applications, T. Blaschke, Lang, S., P. Zeil, S. Kienberger, and D. Tiede. 2008. Geons
S. Lang, and G. J. Hay (eds.). Berlin, Germany: Springer. Policy-relevant geo-objects for monitoring high-level indi-
Hay, G. J., D. J. Marceau, P. Dub, and A. Buchard. 2001. A mul- cators. Proceedings of the GI Forum, Geospatial Crossroads @
tiscale framework for landscape analysis: Object-specific GI_Forum 08, Salzburg, Austria, A. Car and J. Strobl (eds.).
analysis and upscaling. Landscape Ecology 16(6): 471490. pp. 180186.
Hernando, A., D. Tiede, F. Albrecht, and S. Lang. 2012. Spatial and Liedtke, C. E., J. Bckner, O. Grau, S. Growe, and R. Tnjes. 1997.
thematic assessment of object-based forest stand delinea- AIDA: A system for the knowledge based interpretation of
tion using an OFA-matrix. International Journal of Applied remote sensing data. Third International Airborne Remote
Earth Observations and Geoinformation 19: 214225. Sensing Conference, Copenhagen, Denmark.
Hofmann, P. and T. Blaschke. 2012. Object based change detec- Lu, D., P. Mausel, and E. Moran. 2004. Change detection tech-
tion using temporal linkages. Proceedings of the 4th GEOBIA niques. International Journal of Remote Sensing 25:
Conference 2012, Rio de Janeiro, Brazil. 23652401.
Johnson, B. and X. Zhixiao. 2011. Unsupervised image seg- Lucieer, A. and A. Stein. 2002. Existential uncertainty of spa-
mentation evaluation and refinement using a multi-scale tial objects segmented from satellite sensor imagery. IEEE
approach. ISPRS Journal of Photogrammetry and Remote Transactions on Geoscience and Remote Sensing 40(11):
Sensing 66: 473483. 25182521.
Julesz, B. 1975. Experiments in the visual perception of texture. Lutz, M. and E. Klien. 2006. Ontology based retrieval of geo-
Scientific American 212: 3848. graphic information. International Journal of Geographical
Kuhn, W. 2009. A functional ontology of observation and mea- Information Science 20(3): 233260.
surement. In GeoSpatial Semantics, K. Janowicz, M. Raubal, Marceau, D. J. 1999. The scale issue in the social and natural sci-
and S. Levashkin (eds.). Berlin, Germany: Springer, Vol. ences. Canadian Journal of Remote Sensing 25: 347356.
5892, pp. 2643. Marr, D. 1982. Vision. New York: W.H. Freeman.
2016 by Taylor & Francis Group, LLC
Geospatial Data Integration in OBIA 315
Merchant, J. W. and S. Narumalani. 2009. Integrating remote Spitzer, M. 2000. Geist im Netz. Modelle fur Lernen, Denken und
sensing and geographic information system. In The SAGE Handeln. Heidelberg, Germany: Spektrum Akademischer
Handbook of Remote Sensing, T. A. Warner, M. D. Nellis, Verlag.
and G. M. Foody (eds.). London, U.K.: SAGE Publications Strasser, T. and S. Lang. In press. Object-based class modelling for
Ltd., pp. 257268. multi-scale riparian forest habitat mapping. International
Montello, D. R. 2001. Scale in Geography: International Journal of Applied Earth Observations and Geoinformations
Encyclopedia of the Social & Behavioural Sciences. Oxford, 37: 2937.
U.K.: Pergamon Press, pp. 1350113504. Straub, B. M. and C. Heipke. 2004. Concepts for internal and
Pesaresi, M., A. Gerhardinger, and F. Haag. 2007. Rapid damage external evaluation of automatically delineated tree tops.
assessment of built-up structures using VHR satellite data IntArchPhRS 26(8): 6265.
in tsunami-affected areas. International Journal of Remote Sundersan, A., P. K. Varshney, and M. K. Arora. 2007. Robustness
Sensing 28(13): 30133036. of change detection algorithms in the presence of regis-
Pinz, A. 1994. Bildverstehen. Vienna, Austria: Springer. tration errors. Photogrammetric Engineering and Remote
Radoux, J. and P. Defourny. 2008. A framework for the quality Sensing 73: 165174.
assessment of object-based classification. In Object-Based Tarr, M. J. and Y. D. Cheng. 2003. Learning to see faces and
Image AnalysisSpatial Concepts for Knowledge-Driven objects. Trends in Cognitive Science 7(1): 2330.
Remote Sensing Applications, T. Blaschke, S. Lang, and G. J. Tiede, D. 2014. A new geospatial overlay method for the anal-
Hay (eds.). Berlin, Germany: Springer, pp. 257271. ysis and visualization of spatial change patterns using
Raskin, R. and M. Pan. 2003. Semantic web for earth and environ- object-oriented data modeling concept. Cartography and
mental terminology (sweet). Proceedings of the Workshop Geographic Information Science 41: 227234.
on Semantic Web Technologies for Searching and Retrieving Tiede, D., S. Lang, F. Albrecht, and D. Hlbling. 2010. Object-
Scientific Data, Sanibel Island, FL. based class modeling for cadastre constrained delineation
Rathje, E. M. and B. J. Adams. 2008. The role of remote sensing of geo-objects. Photogrammetric Engineering and Remote
in earthquake science and engineering: Opportunities and Sensing 76(2): 193202.
challenges. Earthquake Spectra 24(2): 471492. Tiede, D., S. Lang, P. Freder, D. Hlbling, C. Hoffmann, and
Raza, A. and W. Kainz. 2001. An object-oriented approach for P.Zeil. 2011. Automated damage indication for rapid geo-
modeling urban land-use changes. Journal of the Urban and spatial reporting. An operational object-based approach to
Regional Information Association 14(1): 3755. damage density mapping following the 2010 Haiti earth-
Reitsma, F. L. J., S. Ballard, W. Kuhn, and A. Abdelmoty. 2009. quake. Photogrammetric Engineering and Remote Sensing
Semantics, ontologies and eScience for the geosciences. 9: 933942.
Computers and Geosciences 35: 706709. Tiede, D., M. Mller, S. Lang, and D. Hlbling. 2007. Adapting,
Richter, R. and D. Schlpfer. 2002. Geo-atmospheric processing of splitting and merging cadastral boundaries according to
airborne imaging spectrometry data, part 2: Atmospheric/ homogenous LULC types derived from Spot 5 dat. ISPRS
topographic correction. International Journal of Remote Journal of Photogrammetry and Remote Sensing 36(3):
Sensing 23: 26312649. 99104.
Schpfer, E., S. Lang, and F. Albrecht. 2008. Object-fate analysis Tnjes, R., S. Growe, J. Bckner, and C. E. Liedtke. 1999.
Spatial relationships for the assessment of object transition Knowledge-based Interpretation of remote sensing images
and correspondence. In Object-Based Image Analysis using semantic nets Photogrammetric Engineering and
Spatial Concepts for Knowledge-Driven Remote Sensing Remote Sensing 65: 811821.
Applications, T. Blaschke, S. Lang, and G. J. Hay (eds.). Turker, M. and B. Cetinkaya. 2005. Automatic detection of
Berlin, Germany: Springer, pp. 785801. earthquake-damaged buildings using DEMs created
Schumacher, J., S. Lang, D. Tiede, D. Hlbling, J. Rietzke, and from pre- and post-earthquake stereo aerial photo-
J. Trautner. 2007. Einsatz von GIS und objekt-basierter graphs. International Journal of Remote Sensing 26(4):
Analyse von Fernerkundungsdaten in der regionalen 823832.
Planung Methoden und erste Erfahrungen aus dem Van Coillie, F. M. B., R. P. C. Verbeke, and R. R. De Wulff. 2008.
Biotopinformations- und Managementsystem (BIMS) Semi-automated forest stand delineation using wavelet
Region Stuttgart. In Angewandte Geoinformatik 2007, based segmentation of very high resolution optical imag-
J.Strobl, T.Blaschke, and G. Griesebner (eds.). Heidelberg, ery. In Object-Based Image Analysis for Remote Sensing
Germany: Wichmann, pp. 703708. Applications: Modeling Reality-Dealing with Complexity, T.
Smith, B. 1994. Fiat objects. Eleventh European Conference on Blaschke, S. Lang, and G. J. Hay (eds.). Berlin, Germany:
Artificial Intelligence, Amsterdam, the Netherlands, N. Springer, pp. 237256.
Guarino, L. Vieu, and S. Pribbenow (eds.). Vu, T. T. and Y. Ban. 2010. Context-based mapping of dam-
Spence, N. A. 1961. A multifactor uniform regionalization of aged buildings from high-resolution optical satellite
British Countries on the basis of employment data for 1961. images. International Journal of Remote Sensing 31(13):
Regional Studies II 2: 87104. 34113425.
2016 by Taylor & Francis Group, LLC
316 Remotely Sensed Data Characterization, Classification, and Accuracies
Walter, V. and D. Fritsch. 1999. Matching spatial data sets: A Wise, S., R. Haining, and J. Ma. 2001. Providing spatial statisti-
statistical approach. International Journal of Geographical cal data analysis functionality for the GIS user: The SAGE
Information Science 13(5): 445473. project. International Journal of Geographical Information
Weinke, E., S. Lang, and M. Preiner. 2008. Strategies for semi- Science 3(1): 239254.
automated habitat delineation and spatial change assess- Zadeh, L. A. 1965. Fuzzy sets. Information and Control 8(3):
ment in an Alpine environment. In Object-Based Image 338353.
AnalysisSpatial Concepts for Knowledge-Driven Remote Zeil, P., H. Klug, and I. Niemeyer. 2008. GIS and remote sensing:
Sensing Applications, T. Blaschke, S. Lang, and G. J. Hay Monitoring Environmental conventions and agreements.
(eds.). Berlin, Germany: Springer, pp. 711732. BICC Brief, Bonn, 37: 5056.
Weinke, E., S. Lang, and D. Tiede. 2007. Landscape Interpretation Zlatof, N., B. Tellez, and A. Baskurt. 2004. Image understanding
Support Tool (LIST). Shaker Verlag, Graz, Austria. and scene models: A generic framework integrating domain
Wertheimer, M. 1925. Drei Abhandlungen zur Gestalttheorie (in knowledge and Gestalt theory. International Conference on
German). Erlangen, Germany: Palm & Enke. Image Processing, ICIP 04, Singapore.
2016 by Taylor & Francis Group, LLC
16
Image Segmentation Algorithms
for Land Categorization
Acronyms and Definitions .................................................................................................................317
16.1 Introduction .............................................................................................................................317
16.2 Spectrally Based Segmentation Approaches .......................................................................318
James C. Tilton Thresholding-Based Algorithms Clustering-Based Algorithms Support Vector Machines
Goddard Space Flight Center 16.3 Spatially Based Segmentation Approaches ......................................................................... 320
Region-Growing Algorithms Texture-Based Algorithms Morphological
Selim Aksoy Algorithms Graph-Based Algorithms MRF-Based Algorithms
Bilkent University
16.4 Image Segmentation Quality Evaluation .............................................................................335
Yuliya Tarabalka 16.5 Concluding Remarks...............................................................................................................338
Sophia-Antipolis Mditerrane References.............................................................................................................................................338
Acronyms and Definitions determining the level of performance of these image analysis
approaches. It is expected that an appropriately designed image
AA Average accuracy segmentation approach will provide a better understanding
BSMSE Band sum mean squared error of a landscape and/or significantly increase the accuracy of a
DMP Derivative of the morphological profile landscape classification. An image can be partitioned in several
HSeg Hierarchical segmentation ways, based on numerous criteria. Whether or not a particular
HSWO Hierarchical step-wise optimization image partitioning is useful depends on the goal of the image
ICM Iterated conditional modes analysis application that is fed by the image segmentation result.
IOER Institute for Ecological and Regional Development The focus of this chapter is on image segmentation algorithms
MP Morphological profile for land categorization. Our image analysis goal will generally
MRF Markov random field be to appropriately partition an image obtained from a remote
MSE Mean squared error sensing instrument on board a high flying aircraft or a satellite
MSF Minimum spanning forest circling the earth or other planet. An example of an earth remote
NDVI Normalized Difference Vegetation Index sensing application might be to produce a labeled map that
OA Overall accuracy divides the image into areas covered by distinct earth surface
PV Plurality vote covers such as water, snow, types of natural vegetation, types of
RBF Radial basis function rock formations, types of agricultural crops, and types of other
RHSeg Recursive hierarchical segmentation man-created development. Alternatively, one can segment the
ROSIS Reflective optics system imaging spectrometer land based on climate (e.g., temperature, precipitation) and ele-
SAM Spectral angle mapper vation zones. However, most image segmentation approaches do
SAR Synthetic aperture Radar not directly provide such meaningful labels to image partitions.
SE Structuring element Instead, most approaches produce image partitions with generic
SVM Support vector machines labels such as region 1, region 2, and so on, which need to be con-
verted into meaningful labels by a post-segmentation analysis.
16.1 introduction An early survey on image segmentation grouped image seg-
mentation approaches into three categories (Fu and Mui, 1981):
Image segmentation is the partitioning of an image into related (1) characteristic feature thresholding or clustering, (2) bound-
meaningful sections or regions. Segmentation is the key first ary detection, and (3) region extraction. Another early survey
step for a number of image analysis approaches. The nature and (Haralick and Shapiro, 1985) divides region extraction into
quality of the image segmentation result is a critical factor in several region growing and region split-and-merge schemes.
317
2016 by Taylor & Francis Group, LLC
318 Remotely Sensed Data Characterization, Classification, and Accuracies
Both of these surveys note that there is no general theory of Selection of the threshold values is often done in an ad hoc
image segmentation, most image segmentation approaches are manner usually when a single property is involved, but optimal
ad hoc in nature, and there is no general algorithm that will values can also be found by employing exhaustive or stochastic
work well for all images. This is still the case even today. search procedures that look for the values that optimize some
We start our image segmentation discussion with spectrally criteria on the shape or the statistics of the histogram such as
based approaches, corresponding to Fu and Muis characteristic minimization of the within-class variance and maximization
feature thresholding or clustering category. We include here a of the between-class variance (Otsu, 1979). A stochastic search
description of support vector machines (SVMs), as a supervised procedure is particularly needed for finding multiple thresholds
spectral classification approach that has been a popular choice where an exhaustive search is not computationally feasible due
for analyzing multispectral and hyperspectral images (images to the combinatorial increase in the number of candidate val-
with several tens or even hundreds of spectral bands). We then ues. For example, a recent use of multilevel thresholding for the
go on to describe a number of spatially based image segmen- segmentation of Earth observation data is described in Ghamisi
tation approaches that could be appropriate for land categori- etal. (2014), where a particle swarm optimization-based stochas-
zation applications, generally going from simpler approaches tic search algorithm was used to obtain a multilevel threshold-
to more complicated and more recently developed approaches. ing of each spectral channel independently by maximizing the
Here our emphasis is guided by the prevalence of reported corresponding between-class variance.
use in land categorization studies. We then take a brief look at Even when the selected thresholds are obtained by optimizing
various approaches to image segmentation quality evaluation some well-defined criteria on the distributions of the properties
and include a closer look at a particular empirical discrepancy of the pixels, they do not necessarily produce operational image
approach with example quality evaluations for a particular segmentation results because they suffer from the lack of the use
remotely sensed hyperspectral data set and selected image seg- of spatial information as the decisions are independently made
mentation approaches. We wrap up with some concluding com- on individual pixels. Thus, thresholding is usually applied as a
ments and discussion. preprocessing algorithm, and various postprocessing methods
such as morphological operations are often applied to the results
16.2 Spectrally Based of pixel-based thresholding algorithms as discussed in the fol-
lowing section.
Segmentation Approaches
The focus of this section is on approaches that are mainly based 16.2.2 clustering-Based Algorithms
on analyses of individual pixels. These approaches use an initial
labeling of pixels using unsupervised or supervised classification The clustering-based approaches to image segmentation aim to
methods, and then try to group neighboring pixels with similar make use of the rich literature on data grouping and/or parti-
labels using some form of postprocessing to produce segmenta- tioning techniques for pattern recognition (Duda etal., 2001). It
tion results. is intuitive to pose the image segmentation problem as the clus-
tering of pixels, and thus, pixel-based image analysis techniques
in the remote sensing literature have found natural extensions
16.2.1 thresholding-Based Algorithms
to image segmentation. In the most widely used methodology,
Thresholding has been one of the oldest and most widely used first, the spectral feature space is partitioned, and the individual
techniques for image segmentation. Thresholding algorithms pixels are grouped into clusters without regard to their neigh-
used for segmentation assume that the pixels that belong to the bors, and then, a postprocessing step is applied to form regions
objects of interest have a property whose values are substan- by merging neighboring pixels having the same cluster label by
tially different from those of the background, and they aim to using a connected components labeling algorithm.
find a good set of thresholds that partition the histogram of this The initial clustering stage commonly employs well-known
property into two or more nonoverlapping regions (Sezgin and techniques such as k-means (Aksoy and Akcay, 2005), fuzzy
Sankur, 2004). While the spectral channels can be directly used c-means (Shankar, 2007), and their probabilistic extension using
for thresholding, other derived properties of the pixels are also the Gaussian mixture model estimated via expectationmaxi-
commonly used in the literature. For example, Akcay and Aksoy mization (Fauvel et al., 2013). Since no spatial information is
(2011) used thresholding of the red band to identify buildings used during the clustering procedure, pixels with the same
with red roofs, Bruzzone and Prieto (2005) performed change cluster label can either form a single connected spatial region
detection by thresholding the difference image, Rosin and Hervas or can belong to multiple disjoint regions that are assigned dif-
(2005) used thresholding of the difference image for determining ferent labels by the connected components labeling algorithm.
landslide activity, Aksoy etal. (2010) applied thresholding to the This reduces the significance of the difficulty of the users a
normalized difference vegetation index (NDVI) for segment- priori selection of the number of clusters in many popular clus-
ing vegetation areas, and Unsalan and Boyer (2005) combined tering algorithms as there is no strict correspondence between
thresholding of NDVI and a shadowwater index to identify the initial number of clusters and the final number of image
potential building and street pixels in residential regions. regions. However, it still has a high potential of producing an
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 319
oversegmentation consisting of noisy results with isolated pixels estimation where the local maxima (i.e., modes) of the density
having labels different from those of their neighbors due to the can be assumed to correspond to clusters. The algorithm does
lack of the use of spatial data. not require a priori knowledge of the number of clusters in the
Therefore, a follow-on postprocessing step is used to produce data and can identify the locations of the local maxima by a set
a smoother and spatially consistent segmentation by convert- of iterations. These iterations can be interpreted as the shift-
ing the pixel-based clustering results into contiguous regions. ing of points toward the modes where convergence is achieved
A popular approach is to use an additional segmentation result when a point reaches a particular mode. The shifting proce-
(often also an oversegmentation) and to use a majority voting dure uses a kernel with a scale parameter that determines the
procedure for spatial regularization by assigning each region in amount of local smoothing performed during density estima-
the oversegmentation a single label that is determined according tion. The application of the mean shift procedure to image seg-
to the most frequent cluster label among the pixels in that region mentation uses a separate kernel for the feature (i.e., spectral)
(Fauvel etal., 2013). An alternative approach is to use an itera- domain and another kernel for the spatial (i.e., pixel) domain.
tive split-and-merge procedure as follows (Aksoy et al., 2005; The scale parameter for the spectral domain can be esti-
Aksoy, 2006): mated by maximizing the average likelihood of held-out data.
The scale parameter for the spatial domain can be selected
1. Merge pixels with identical labels to find the initial set of
according to the amount of compactness or oversegmentation
regions and mark these regions as foreground.
desired in the image, or can be determined by using geospa-
2. Mark regions with areas smaller than a threshold as back-
tial statistics (e.g., by using semivariogram-based estimates)
ground using connected components analysis.
(Dongping et al., 2012). Furthermore, agglomerative cluster-
3. Use region growing to iteratively assign background pix-
ing of the mode estimates can be used to obtain a multiscale
els to the foreground regions by placing a window at each
segmentation.
background pixel and assigning it to the class that occurs
the most in its neighborhood.
This procedure corresponds to a spatial smoothing of the clus- 16.2.3 Support Vector Machines
tering results. The resulting regions can be further processed
Output of supervised classification of pixels can also be used
using mathematical morphology operators to automatically
as input for segmentation techniques. In recent years, SVMs
divide large regions into more compact subregions as follows:
and the use of kernels to transform data into a new feature
1. Find individual regions using connected components space where linear separability can be exploited have been pro-
analysis for each cluster. posed. The SVM method attempts to separate training samples
2. For all regions, compute the erosion transform and repeat belonging to different classes by tracing maximum-margin
the following: hyperplanes in the space where the samples are mapped. SVM
a. Threshold erosion transform at steps of three pixels in have shown to be particularly well suited to classify high-
every iteration. dimensional data (e.g., hyperspectral images) when a lim-
b. Find connected components of the thresholded image. ited number of training samples are available (Vapnik, 1998;
c. Select subregions that have an area smaller than the Camps-Valls, 2005). The success of SVM for pixel-based classi-
threshold. fication has led to its subsequent use as part of image segmenta-
d. Dilate these subregions to restore the effects of tion methods. Thus, we discuss the SVM approach in detail in
erosion. the following text.
e. Mark these subregions in the output image by mask- SVMs are primarily designed to solve binary tasks, where
ing the dilation using the original image. the class labels take only two values: 1 or 1. Let us consider a
Repeat the steps a-e until no more subregions are found. binary classification problem in a B-dimensional space R B , with
3. Merge the residues of previous iterations to their smallest N training samples, x i R B , and their corresponding class labels
neighbors. yi = 1 available. The SVM technique consists in finding the
hyperplane that maximizes the margin, that is, the distance to
Even though we focused on producing segmentations using
the closest training data points in both classes (see Figure 16.1).
clustering algorithms in this section, similar postprocessing
Noting w R Bas the vector normal to the hyperplane and b R
techniques for converting the pixel-based decisions into con-
as the bias, the hyperplane H is defined as
tiguous regions can also be used with the outputs of pixel-based
thresholding (Section 16.2.1) and classification (Section 16.2.3) w x +b = 0, x H .
procedures (Aksoy etal., 2005).
It is also possible to pose clustering, and the corresponding
If x H, then
segmentation, as a density estimation problem. A commonly
used algorithm that combines clustering with density estima-
|w x +b|
tion and segmentation is the mean shift algorithm (Comaniciu f (x )=
and Meer, 2002). Mean shift is based on nonparametric density ||w||
2016 by Taylor & Francis Group, LLC
320 Remotely Sensed Data Characterization, Classification, and Accuracies
functions of the pairs of vectors, essentially projecting the pixel
vectors in a higher-dimensional space H and thus improving
yi = +1 linear separability of data:
i R B H,
x i
x (x ), (16.5)
x i x j (x i ) (x j ) = K (x i , x j ).
j yj = 1
Here, () is a nonlinear function to project feature vectors
xj into a new space, K() is a kernel function, which allows one to
avoid the computation of scalar products in the transformed
space [(xi)(xj)]and thus reduces the computational complex-
b ity of the algorithm. The kernel K must satisfy Mercers condition
w
(Burges, 1998). The Gaussian radial basis function (RBF) kernel
is the most widely used for remote sensing image classification:
Margin = 2
w
K Gaussian (x i , x j ) = exp[ ||x i x j ||2 ], (16.6)
FIg u r e 16.1 Schematic illustration of the SVM binary classification
method. There is one nonlinearly separable sample in each class. where is the spread of the RBF kernel.
To solve the K-class problem, various approaches have been
defines the distance of the sample x to H. In the linearly sepa- proposed. Two main approaches combining a set of binary clas-
rable case, such a hyperplane must satisfy sifiers are defined as (Smola, 2002) follows:
y i (w x i + b) > 1, i [1, N ]. (16.1) One versus all: K binary classifiers are applied on each
class against others. Each pixel xiis assigned to the class
The optimal hyperplane is the one that maximizes the margin with the maximum output f(xi).
2/||w||. This is equivalent to minimizing ||w||/2and leads to the One versus one: K(K 1)/2 binary classifiers are applied
following quadratic optimization problem: on each pair of classes. Each pixel is assigned to the class
winning the maximum number of binary classification
procedures.
||w||2
min , subject to (16.1). (16.2)
As a conclusion, SVMs directly exploit the geometrical proper-
2
ties of data, without involving a density estimation procedure.
This method has proven to be more effective than other non-
To take into account nonlinearly separable data, slack variables
parametric classifiers (such as neural networks or the k-nearest
are introduced to deal with misclassified samples (see Figure 16.1).
neighbor classifier (Duda et al., 2001)) in terms of classifica-
Equation 16.1 becomes
tion accuracies, computational complexity, and robustness to
parameter setting. SVM can efficiently handle high-dimen-
y i (w x i +b) > 1 i , i 0, i [1, N ]. (16.3)
sional data, exhibiting low sensitivity to the Hughes phenom-
enon (Hughes, 1968). Finally, it exhibits good generalization
The final optimization problem is formulated as capability, fully exploiting the discrimination capability of
available training samples.
||w||2 N
Pixel-based supervised classification results, such as those
min
2
+C ,
i =1
i subject to (16.3), (16.4)
obtained using an SVM classifier, are often given as input to seg-
mentation procedures that aim to group the pixels to form con-
tiguous regions as discussed in the following sections.
where the constant C is a regularization parameter that controls
the amount of penalty. This optimization problem is typically
solved by quadratic programming (Vapnik, 1998). The classifi- 16.3 Spatially Based Segmentation
cation is further performed by computing yu = sign(wxu + b), Approaches
where (w, b) are the hyperplane parameters found during the
training process, and xuis an unseen sample. We cannot possibly discuss myriad of spatially based image
One can notice that the pixel vectors in the optimization and segmentation approaches that have been proposed and devel-
decision rule equations always appear in pairs related through oped over the years. Instead, we will focus on approaches that
a scalar product. These products can be replaced by nonlinear have achieved demonstrated success in remote sensing land
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 321
TABLe 1 6.1 Algorithmic Basis of Image Segmentation Approaches in Remote Sensing Oriented Image Analysis Packages
Name Website or Reference Algorithmic Basis
BerkeleyImgseg Clinton et al. (2010) Region growing, region merging
Ceasar Cook etal. (1996) Simulated annealing
eCognition Developera Trimble Geospatial Imaging (2014) and Baatz and Schape (2000) Region growing
ENVI feature extraction Exelis Visual Information Solutions (2015) and Robinson etal. (2002) Edge-based (full lambda-schedule algorithm for
region merging)
Extended watershed EWS Li and Xiao (2007) Multichannel watershed transformation
Image WS for Erdas Imagine Sramek and Wrbka (1997) Hierarchical watershed
InfoPACK Cook etal. (1996) Simulated annealing
PARBAT Lucieer (2015) and Lucieer (2004) Region growing
RHSeg Tilton (1998) and Tilton etal. (2012) Region growing and spectral clustering
SCRM Castilla etal. (2008) Watershed and region merging
SegSAR Sousa (2014) Hybrid (edge/region oriented)
SEGEN Gofman (2006) Region growing
GRASS GIS GRASS Development Team (2015) Region growing and watershed
IDRISI Clark Labs (2015) Watershed
Orfeo Toolbox CNES (2015) Region growing, watershed, level sets, mean shift
Note: Most of these image segmentation approaches were evaluated in a series of papers by the Leibnitz IOER group.
aWas Definiens Developerbut the remote sensing package is now marketed by Trimble, and the Definiens product is now oriented to biomedical image
analysis.
categorization applications. A compilation of such approaches initial small region objects are often single image pixels, but
can be found in a series of papers published by a research group can also be n n blocks of pixels or another partitioning
based at the Leibniz Institute for Ecological and Regional of the image into small spatially connected region objects.
Development (IOER) that present comparative evaluations of Then, pairs of spatially adjacent region objects are compared
image segmentation approaches implemented in various image and merged together if they are found to be similar enough
analysis packages (Meinel and Neubert, 2004; Neubert et al., according to some comparison criterion. The underlying
2006, 2008; Marpu etal., 2010). Table 16.1 provides a summary assumption is that region objects of interest are several image
listing of most of the remote sensingoriented image analysis pixels in size and relatively homogeneous in value. Most
packages whose image segmentation approach was evaluated in region-growing approaches can operate on either grayscale,
these papers, plus image segmentation approaches from three multispectral, or hyperspectral image data, depending on the
additional notable remote sensingoriented image analysis criterion used to determine the similarity between neighbor-
packages (GRASS GIS, IDRISI, and the Orfeo Toolbox). ing region objects.
We note from Table 16.1 that region growing is the most fre- A very early example of region growing was described in
quent image segmentation approach utilized by these remote Muerle and Allen (1968). Muerle and Allen experimented
sensingoriented image analysis software packages. Further, with initializing their region-growing process with region
several packages combine region growing with other tech- objects consisting of 2 2 up to 8 8 blocks of pixels. After
niques (RHSeg with spectral clustering, SCRM and GRASS GIS initialization, they started with the region object at the upper
with watershed, and SegSAR with edge detection). Watershed left corner of the image and compared this region object
segmentation (an approach based on region boundary detec- with the neighboring region objects. If a neighboring region
tion) is the next most popular approach. Simulated annealing object was found to be similar enough, the region objects
is often utilized in analysis packages oriented toward analyz- were merged together. This process was continued until no
ing synthetic aperture radar (SAR) imagery data (Ceasar and neighboring region objects could be found that were similar
InfoPACK). enough to be merged into the region object that was being
The next several sections describe various spatially based image grown. Then, the image was scanned (left-to-right, top-to-
segmentation approaches, starting with region-growing algorithms bottom) to find an unprocessed region object, that is, a region
and continuing with texture-based algorithms, morphological object that not yet been considered as an initial object for
algorithms, graph-based algorithms, and MRF-based algorithms. region growing or merged into a neighboring region. If an
unprocessed region object was found, they conducted their
region-growing process from that region object. This con-
16.3.1 Region-Growing Algorithms
tinued until no further unprocessed region objects could be
In the region-growing approach to image segmentation, an found, upon which point the region-growing segmentation
image is initially partitioned into small region objects. These process was considered completed.
2016 by Taylor & Francis Group, LLC
322 Remotely Sensed Data Characterization, Classification, and Accuracies
Many early schemes for region growing, such as Muerle and
Allens, can be formulated as logical predicate segmentation, Image
defined in Zucker (1976) as follows:
A segmentation of an image X can be defined as a partition
of X into R disjoint subsets X1, X2,, XR, such that the following Level n
conditions hold:
1. iR=1 X i = X .
2. Xi, i = 1, 2,, R are connected.
Level n1
3. P(Xi) = TRUE for i = 1, 2,, R.
4. P(Xi Xj)= FALSE for i j, where Xi and Xj are adjacent.
P(Xi) is a logical predicate that assigns the value TRUE or FALSE
to Xi, depending on the image data values in Xi. Level n2
These conditions are summarized in Zucker (1976) as follows: the
first condition requires that every picture element (pixel) must be in
a region (subset). The second condition requires that each region
must be connected, that is, composed of contiguous image pixels. Level n3
The third condition determines what kind of properties each region
must satisfy, that is, what properties the image pixels must satisfy
to be considered similar enough to be in the same region. Finally, FIg u r e 16.2 The last four levels of an n-level segmentation hier-
the fourth condition specifies that any merging of adjacent regions archy produced by a region-growing segmentation process. Note that
would violate the third condition in the final segmentation result. when depicted in this manner, the region-growing process is a bottom-
Several researchers in this early era of image segmentation up approach.
research, including Muerle and Allen, noted some problems with
logical predicate segmentation. For one, the results were very down to the final trivial one-region segmentation (if allowed to
dependent on the order in which the image data were scanned. proceed that far). For practical applications, however, a subset
Also, the statistics of a region object can change quite dramatically of segmentations needs to be selected out from this exhaustive
as the region is grown, making it possible that many adjacent cells segmentation hierarchy. At a minimum, such a subset can be
that were rejected for merging early in the region-growing process defined by storing the results only after a preselected number of
would have been accepted in later stages based on the changed regions are reached and then storing selected iterations after that
statistics of the region object. The reverse was also possible, where such that no region is involved in more than one merge between
adjacent cells that were accepted for merging early in the region- stored iteration until a two-region segmentation is reached.
growing process would have been rejected in later stages. A portion of such a segmentation hierarchy is illustrated in
Subsequently, an alternate approach to region growing was Figure16.2. (The selection of a single segmentation from a seg-
developed that avoids these problems, and that approach eventu- mentation hierarchy is discussed in Section 16.3.1.5.)
ally came to be referred to as best merge region growing. An early A unique feature of the segmentation hierarchy produced by
version of best merge region growing, hierarchical step-wise HSWO and related region-growing segmentation approaches is
optimization (HSWO), is an iterative form of region growing, in that the segment or region boundaries are maintained at the full
which the iterations consist of finding the most optimal or best image spatial resolution for all levels of the segmentation hierar-
segmentation with one region less than the current segmenta- chy. The region boundaries are coarsened in many other multi-
tion (Beaulieu and Goldberg, 1989). The HSWO approach can be level representations.
summarized as follows: Many variations on best merge region growing have been
described in the literature. As early as 1994, Kurita (1994) described
1. Initialize the segmentation by assigning each image pixel
an implementation of the HSWO form of best merge region grow-
a region label. If a pre-segmentation is provided, label
ing that utilized a heap data structure (Williams, 1964) for effi-
each image pixel according to the pre-segmentation.
cient determination of best merges and a dissimilarity criterion
Otherwise, label each image pixel as a separate region.
based on minimizing the mean squared error between the region
2. Calculate the dissimilarity criterion value, d, between all
mean image and original image. Further, several of the image seg-
pairs of spatially adjacent regions, find the smallest dis-
mentation approaches studied in the previously referenced series
similarity criterion value, Tmerge, and merge all pairs of
of papers published by the Leibniz Institute are based on best
regions with d = Tmerge.
merge region growing (see Table 16.1). As we will discuss in more
3. Stop if no more merges are required. Otherwise, return to
detail later, the main differences between most of these region-
step 2.
growing approaches are the dissimilarity criterion employed and,
HSWO naturally produces a segmentation hierarchy consist- perhaps, some control logic designed to remove small regions or
ing of the entire sequence of segmentations from initialization otherwise tailor the segmentation output.
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 323
16.3.1.1 Seeded Region Growing 16.3.1.4 Dissimilarity criterion and
Seeded region growing is a variant of best merge region grow- Specialized control Logic
ing in which regions are grown from preselected seed pixels or Muerle and Allen (1968) experimented with various criteria for
seed regions. Adams and Bischof (1994) is an early example of determining whether or not pairs of region objects were similar
this approach. In seeded region growing, the best merges are enough to be merged together. They concluded that an optimal
found by examining the pixels adjacent to each seed pixel of criterion would be a threshold of a function of the mean and
region formed around each seed pixel. As described by Adams standard deviation of the gray levels of the pixels contained in
and Bischof (1994), the region-growing process continues until the compared region objects.
each image pixel is associated with one of the preselected seed In our image segmentation research, we have implemented
pixels or regions. and studied several region merging criteria in the form of dis-
similarity criteria (Tilton, 2013). Included among these dissimi-
16.3.1.2 Split-and-Merge Region Growing larity criteria are criteria based on vector norms (the 1-, 2-, and
In the split-and-merge approach, the image is repeatedly subdi- -norms), criteria based on minimizing the increase in mean
vided until each resulting region has a sufficiently high homo- squared error between the region mean image and the original
geneity. Examples of measures of homogeneity are the mean image data, and a criterion based on the Spectral Angle Mapper
squared error between the region mean and the image data val- (SAM) criterion (Kruse et al., 1993). We briefly describe these
ues, or the region standard deviation. After the region-splitting dissimilarity criteria here.
process converges, the regions are grown using one of the pre- The dissimilarity criterion based on the 1-norm of the differ-
viously described region-growing approaches. This approach ence between the region mean vectors, ui and uj, of regions Xi
is more efficient when large homogeneous regions are present. and Xj, each with B spectral bands, is
However, some segmentation detail may be lost. See Horowitz B
and Pavlidis (1974), Cross et al. (1988), and Strasters and
Gerbrands (1991) for examples of this approach.
d1-norm (X i , X j ) = ||ui u j ||1 = |
b =1
ib jb |, (16.7)
16.3.1.3 Hybrid of Spectral clustering where ib and jb are the mean values for regions i and j, respec-
and Region Growing tively, in spectral band b, that is, ui = (i1, i2, , iB)T and uj =
Tilton (1998) and Tilton et al. (2012) describe a hybridization (j1, j2, , jB)T.
of HSWO best merge region growing with spectral clustering, The dissimilarity criterion based on the 2-norm is
called HSeg (for hierarchical segmentation). We remind the
1/ 2
reader that HSWO is performed by finding a threshold value, B
Tmerge, equal to the value of a dissimilarity criterion of the most
d2-norm (X i , X j ) = ||ui u j ||2 = (ib jb )2 ,
b =1
(16.8)
similar pair of spatially adjacent regions, and then merging all
pairs of regions that have dissimilarity equal to Tmerge. HSeg adds
and the dissimilarity criterion based on the -norm is
to HSWO a step following each step of adjacent region merges
in which all pairs of spatially nonadjacent regions are merged d -norm (X i , X j ) = ||ui u j || = max(|ib jb |,b = 1, 2,, B). (16.9)
that have dissimilarity SwTmerge, where 0.0 Sw 1.0 is a factor
that sets the priority between spatially adjacent and nonadjacent As noted here, a criterion based on mean squared error mini-
region merges. Note that when Sw = 0.0, HSeg reduces to HSWO. mizes the increase in mean squared error between the region
Unfortunately, the inclusion of the step in HSeg of merging mean image and the original image data as regions are grown.
spatially nonadjacent regions adds significantly to the compu- The sample estimate of the mean squared error for the segmen-
tational requirements of this image segmentation approach. tation of band b of the image X into R disjoint subsets X1, X2,,
This is because comparisons must now be made between all XR is given by
pairs of regions instead of just between pairs of spatially adja-
cent regions. A recursive divide-and-conquer approximation of R
MSE (X ),
1
HSeg (called RHSeg), with a straightforward parallel implemen- MSEb (X ) = b i (16.10)
tation, was developed to help overcome this problem (Tilton, N 1 i =1
2007). The computational requirements of this approach were
further reduced by a refinement in which the nonadjacent region where N is the total number of pixels in the image data and
merging was limited to between regions of a minimum size,
Pmin (Tilton etal., 2012). In this refinement, the value of Pmin is
dynamically adjusted to keep the number of large regions (those
MSEb (X i ) = (
x p X i
pb ib )2 (16.11)
of at least Pmin in size) to a range that substantially reduces the
computational requirements without significantly affecting the is the mean squared error contribution for band b from segment
image segmentation results. Xi. Here, xp is a pixel vector (in this case, a pixel vector in data
2016 by Taylor & Francis Group, LLC
324 Remotely Sensed Data Characterization, Classification, and Accuracies
subset Xi), and pb is the image data value for the bth spectral A problem that can often occur with basic best merge region-
band of the pixel vector, xp. The dissimilarity function based on a growing approaches is that the segmentation results contain
measure of the increase in mean squared error due to the merge many small regions. We have found this to be the case when
of regions Xi and Xj is given by employing dissimilarity criteria based on vector norms or SAM,
but not a problem for dissimilarity criteria based on minimizing
B
the increase in mean squared error. This is because the mean
dBSMSE (X i , X j ) = MSE (X , X ),
b =1
b i j (16.12) squared error criterion has a factor, ninj/(ninj), where ni and
nj are the number of pixels in the two compared regions, that
biases toward merging small regions into larger ones. We have
where
found it useful to add on a similar small region merge accelera-
tion factor to the vector norm and SAM-based criterion when
MSEb (X i , X j ) = MSEb (X i X j ) MSEb (X i ) MSEb (X j ). (16.13)
one of the compared regions is smaller than a certain size. See
Tilton etal. (2012) for more details.
BSMSE refers to band sum MSE. Using (16.11) and exchanging Implementations of best merge region growing often add
the order of summation, (16.13) can be manipulated to produce special control logic to reduce the number of small regions or
an efficient dissimilarity function based on aggregated region otherwise improve the final classification result. An example of
features (for the details, see Tilton (2013)): this is SEGEN (Gofman, 2006), which uses the vector 2-norm
(otherwise known as Euclidean distance) for the dissimilarity
B
( criterion. As noted in Tilton etal. (2012), SEGEN is a relatively
nin j
dBSMSE (X i , X j ) = ib jb )2 . (16.14)
(ni + n j ) pure implementation of best merge region growing, optimized
b =1
for efficiency in performance, memory utilization, and image
The dimensionality of the dBSMSE dissimilarity criteria is equal to segmentation quality. SEGEN adds a number of (optional) pro-
the square of the dimensionality of the image pixel values, while cedures to best merge region growing, among them a low-pass
the dimensionality of the vector norm-based dissimilarity cri- filter to be applied on the first stage of the segmentation and
teria is equal to the dimensionality of the image pixel values. To outlier dispatching on the last stage. The latter removes outlier
keep the dissimilarity criteria dimensionalities consistent, the pixels and small segments by embedding them in neighborhood
square root of dBSMSE is often used. segments with the smallest dissimilarity. SEGEN also provides
The SAM criterion is widely used in hyperspectral image several parameters to control the segmentation process. A set of
analysis (Kruse etal., 1993). This criterion determines the spec- good in average control values is suggested in Gofman (2006).
tral similarity between two spectral vectors by calculating the The best merge region-growing segmentation approach
angle between the two spectral vectors. An important prop- employed in eCognition Developer utilizes a dissimilarity func-
erty of the SAM criterion is that poorly illuminated and more tion that balances minimizing the increase in heterogeneity, f, in
brightly illuminated pixels of the same color will be mapped to both color and shape (Baatz and Schape, 2000; Benz etal., 2004):
the same spectral angle despite the difference in illumination.
The spectral angle between the region mean vectors, ui and uj, f = w color hcolor + w shape hshape , (16.17)
of regions Xi and Xj is given by
where wcolor and wshape are weights that range in value from 0 to 1
ui u j and must mutually sum up to 1.
(ui ,u j ) = arccos The color component of heterogeneity, hcolor, is defined as
||ui ||2 ||u j ||2 follows:
B
= arccos
b =1
1/ 2
ib jb
1/ 2
,
(16.5) hcolor = w (n c merge c ,merge (n1 c ,1 + n2 c ,2 )), (16.18)
ib2 2jb
B B c
b =1 b =1
where nx is the number of pixels, with the subscript x = merge
where ib and jb are the mean values for regions i and j, respec- referring to the merged object, and subscripts x = 1 or 2 referring
tively, in spectral band b, that is, ui = (i1, i2,, iB)T and uj = to the first and second objects considered for merging. c refers to
(j1, j2, , jB)T. The dissimilarity function for regions Xi and Xj, the standard deviation in channel (spectral band) c, with the same
based on the SAM distance vector measure, is given by additional subscripting denoting the merged or pair of considered
objects. wc is a channel weighting factor.
The shape component of heterogeneity, hshape, is defined as
dSAM (X i , X j ) = (ui ,u j ). (16.6)
follows:
Note that the value of dSAM ranges from 0.0 for similar vectors up
hshape = w compt hcompt + w smooth hsmooth , (16.19)
to /2 for the most dissimilar vectors.
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 325
where regions and by tracking the behavior of these characteristics
throughout different levels of detail. A manual approach for
doing this using a graphical user interface is described in Tilton
lmerge l l
hcompt = nmerge n1 1 + n2 2 , (16.20) (2003, 2013). A preliminary study on automating this approach
nmerge n1 n2 is described in Plaza and Tilton (2005), where it was proposed
to automate this approach using joint spectral/spatial homoge-
lmerge l l neity scores computed from segmented regions. An alternate
hsmooth = nmerge n1 1 + n2 2 , (16.21)
bmerge b1 b2 approach for making spatially localized selections of segmen-
tation detail based on matching region boundaries with edges
produced by an edge detector is described in Le Moigne and
where Tilton (1995).
l is the perimeter of the object Tarabalka et al. (2012) proposed a modification of HSeg
b is the perimeter of the objects bounding box through which a single segmentation output is automatically
selected for output from the usual segmentation hierarchy. The
The weights wc , wcolor , wshape, wsmooth, and wcompt can be selected to idea is similar to the previously described seeded region grow-
best suit a particular application. ing. The main idea behind the marker-based HSeg algorithm
The level of detail, or scale, of the segmentation is set by stop- consists in automatically selecting seeds, or markers, for image
ping the best merge region-growing process when the increase if regions and then performing best merge region growing with an
heterogeneity, f, for the best merge reaches a predefined thresh- additional condition: two regions with different marker labels
old value. A multiresolution segmentation can be created by per- cannot be merged together (see Figure 16.3). The authors pro-
forming this process with a set of increasing thresholds. posed to choose markers of spatial regions by analyzing results
In eCognition Developer, other segmentation procedures can of a probabilistic supervised classification of each pixel and by
be combined with the multiresolution approach. For example, retaining the most reliably classified pixels as region seeds.
spectral difference segmentation merges neighbor objects that An alternative algorithm that produces a final segmentation
fall within a user-defined maximum spectral difference. This by automatically selecting subsets of regions appearing in differ-
procedure can be used to merge spectrally similar objects from ent levels in the segmentation hierarchy is described in Section
the segmentation produced by the multiresolution approach. 16.3.3.2.
16.3.1.5 Selection of a Single Segmentation
from a Segmentation Hierarchy 16.3.2 texture-Based Algorithms
Some best merge region-growing approaches such as HSWO Along with spectral information, textural features have also been
and HSeg do not produce a single segmentation result, but heavily used for various image analysis tasks, including their use
instead produce a segmentation hierarchy. The best merge as features in thresholding-based, clustering-based, and region-
region-growing segmentation approach employed in eCogni- growing-based segmentation in the literature. In this section, we
tion Developer can also produce a segmentation hierarchy. A focus on the use of texture for unsupervised image segmentation.
segmentation hierarchy is a set of several image segmentations In the following, first, we discuss some particular examples that
of the same image at different levels of detail in which the seg- involve texture modeling for segmenting natural landscapes and
mentations at coarser levels of detail can be produced from sim- man-made structures using local image properties and, then,
ple merges of regions at finer levels of detail. In such a structure, present recent work on generalized texture algorithms that aim
an object of interest may be represented by multiple segments to model complex image structures in terms of the statistics and
in finer levels of detail and may be merged into a surrounding spatial arrangements of simpler image primitives.
region at coarser levels of detail. A single segmentation level One of the common uses of texture in remote sensing image
can be selected out of the segmentation hierarchy by analyz- analysis is the identification of natural landscapes. For example,
ing the spatial and spectral characteristics of the individual Epifanio and Soille (2007) used morphological transformations
Marker 1 Region 1
Dontt merge Region 2
Marker 2 regions with
different
Marker 3 markers Region 3
FIg u r e 1 6.3 Scheme illustrating the marker-based HSeg algorithm.
2016 by Taylor & Francis Group, LLC
326 Remotely Sensed Data Characterization, Classification, and Accuracies
such as white and black top-hat by reconstruction and thresh- regions and then performed hierarchical texture segmentation
olded image gradients with an unsupervised clustering proce- by assuming that frequent neighboring regions are strongly
dure where the texture prototypes were automatically selected related. These relations were represented using Markov chain
based on the dissimilarities between the feature vectors of models computed from quantized region labels, and the image
neighboring image windows to segment vegetation zones and regions that exhibit similar transition probabilities were clus-
forest stands, and Wang and Boesch (2007) combined an ini- tered to construct a hierarchical set of segmentations. Zamalieva
tial color-based oversegmentation step with a threshold-based et al. (2009) used a similar frequency-based approach by find-
region merging procedure that used wavelet feature statistics ing the significant relations between neighboring regions as the
inside candidate image regions to delineate forest boundaries. modes of a probability distribution estimated using the continu-
Man-made structures also exhibit particular characteris- ous features of region co-occurrences. Then, the resulting modes
tics that can be modeled using textural features. For example, were used to construct the edges of a graph where a graph min-
Pesaresi etal. (2008) proposed a rotation invariant anisotropic ing algorithm was used to find subgraphs that may correspond
texture model that used contrast features computed from gray to atomic texture primitives that form the heterogeneous struc-
level co-occurrence matrices and used this model to produce a tures. The final segmentation was obtained by using the histo-
built-up presence index. The motivation behind the use of the grams of these subgraphs inside sliding windows centered at
contrast-based features was to exploit the relationships between individual pixels and by clustering the pixels according to these
the buildings and their shadows. Sirmacek and Unsalan (2010) histograms. As an alternative to graph-based grouping, Akcay
and Gueguen etal. (2012) used a similar idea and modeled urban etal. (2010) performed Gaussian mixture-based clustering of the
areas using spatial voting (smoothing) of local feature points region co-occurrence features to identify frequent region pairs
extracted from the local maxima of Gabor filtering results and that are merged in each iteration of a hierarchical texture seg-
corner detection results, respectively. All of these results can be mentation procedure.
converted to a segmentation output by thresholding the corre- In addition to using co-occurrence properties of neighbor-
sponding urban area estimates. ing regions to exploit statistical information, structural features
An open problem in image segmentation is to identify the can also be extracted to represent the spatial layout for texture
boundaries of regions that are not necessarily homogeneous modeling. For example, Akcay and Aksoy (2011) described a
with respect to low-level features such as color and texture that procedure for finding groups of aligned objects by performing
are extracted from individual pixels or from small local pixel a depth-first search on a graph representation of neighboring
neighborhoods. Recent literature includes several examples primitive objects. After the search procedure identified aligned
that can be considered as generalized texture algorithms that groups of three or more objects that have centroids lying on a
aim to model heterogeneous image content in terms of the spa- straight line with uniform spacing, an agglomerative hierarchi-
tial arrangements of relatively homogeneous image primitives. cal clustering algorithm was used to find larger groups of primi-
Some of this work has considered the segmentation of particular tive objects that have similar spatial layouts. The approach was
structures that have specific textural properties. For example, illustrated in the finding of groups of buildings that have dif-
Dogrusoz and Aksoy (2007) aimed the segmentation of regular ferent statistical and spatial characteristics that cannot be mod-
and irregular urban structures by modeling the image content eled using traditional segmentation methods. Another approach
using a graph where building objects and the Voronoi tessella- for modeling urban patterns using hierarchical segmentations
tion of their locations formed the vertices and the edges, respec- extracted from multiple images of the same scene at various
tively, and the graph was clustered by thresholding its minimum resolutions was described by Kurtz et al. (2012) where binary
spanning tree so that organized (formal) and unorganized partition trees were used to model image data and tree cuts were
(informal) settlement patterns were extracted to model urban learned from user-defined segmentation examples for interac-
development. Some agricultural structures such as permanent tive partitioning of images into semantic heterogeneous regions.
crops also exhibit specific textural properties that can be useful
for segmentation. For example, Aksoy et al. (2012) proposed a
16.3.3 Morphological Algorithms
texture model that is based on the idea that textures are made
up of primitives (trees) appearing in a near-regular repetitive Mathematical morphology has been successfully used for vari-
arrangement (planting patterns) and used this model to com- ous tasks such as image filtering for smoothing or enhance-
pute a regularity score for different scales and orientations by ment, texture analysis, feature extraction, and detecting objects
using projection profiles of multiscale isotropic filter responses with certain shapes in the remote sensing literature (Soille and
at multiple orientations. Then, they illustrated the use of this Pesaresi, 2002; Soille, 2003). Morphological algorithms have
model for segmenting orchards by iteratively merging neighbor- also been one of the most widely used techniques for segmenting
ing pixels that have similar regularity scores at similar scales and remotely sensed images. These approaches view the two-dimen-
orientations. sional image data that consist of the spectral channels or some
More generic approaches have also been proposed for seg- other property of the pixels as an imaginary topographic relief
menting heterogeneous structures. For example, Gaetano where higher-pixel values map to higher imaginary elevation lev-
etal. (2009) started with an oversegmentation of atomic image els (see Figure 16.4). Consequently, differences in the elevations
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 327
(a) (b) (c)
FIg u r e 16.4 Illustration of mapping of the two-dimensional image data that consist of the spectral channels or some other property of the
pixels as an imaginary topographic relief so that higher-pixel values map to higher imaginary elevation levels. (a) An example spectral band.
(b)Thespectral values viewed as a three-dimensional topographic relief. (c) Gradient of the spectral data at each pixel viewed as a three-dimen-
sional topographic relief.
of the pixels in a spatial neighborhood can be exploited to par- computed using derivative filters. Multivariate extensions of the
tition those pixels into different regions. Two morphological gradient function can be used to apply watershed segmentation
approaches for segmentation have found common use in the to multispectral and hyperspectral images (Aptoula and Lefevre,
literature: watershed algorithms and morphological profiles 2007; Li and Xiao, 2007; Noyel etal., 2007; Fauvel etal., 2013).
(MPs). These approaches are described in the following sections. A potential problem in the application of watershed segmen-
Other approaches using mathematical morphology for image tation to images with high levels of detail is oversegmentation
segmentation, and particularly for producing segmentation when the watersheds are computed from raw image gradient
hierarchies, can be found in Soille (2008), Soille and Najman where an individual segment is produced for each local mini-
(2010), Ouzounis etal. (2012), and Perret etal. (2012). mum of the topographic relief. Preprocessing or postprocessing
methods can be used to reduce oversegmentation. For exam-
16.3.3.1 Watershed Algorithms ple, smoothing filters such as the mean or median filters can
The watershed algorithm divides the imaginary topographic be applied to the original image data as a preprocessing step.
relief into catchment basins so that each basin is associated with Alternatively, the oversegmentation produced by the watershed
one local minimum in the image (i.e., individual segments), and algorithm can be given as input to a region merging procedure
the watershed lines correspond to the pixel locations that sepa- for postprocessing (Haris etal., 1998).
rate the catchment basins (i.e., segment boundaries). Watershed Another commonly used alternative to reduce the overseg-
segmentation can be simulated by an immersion process mentation is to use the concept of dynamics that are related to
(Vincent and Soille, 1991). If we immerse the topographic sur- the regional minima of the image gradient. A regional minimum
face in water, the water rises through the holes at the regional is composed of a group of neighboring pixels with the same
minima with a uniform rate. When two volumes of water com- value where the pixels on the external boundary of this group
ing from two different minima are about to merge, a dam is have a greater value. When we consider the image gradient as a
built at each point of contact. Following the immersion process, topographic surface, the dynamic of a regional minimum can
the union of all those dams constitutes the watershed lines. A be defined as the minimum height that a point in the minimum
graph-theoretical interpretation of the watershed algorithm can has to climb to reach a lower regional minimum (Najman and
be found in Meyer (2001). Couprie etal. (2011) describe a com- Schmitt, 1996). The h-minima transform can be used to sup-
mon framework that unifies watershed segmentation and some press the regional minima with dynamics less than or equal
other graph-based segmentation algorithms that are described to a particular value h by performing geodesic reconstruction
in Section 16.3.4. by erosion of the input image f from f + h (Soille, 2003). When
The most commonly used method for constructing the top- it is difficult to select a single h value, it is common to create a
ographic relief from the image data to be segmented is to use multiscale segmentation by using an increasing sequence of h
the gradient function at each pixel. This approach incorporates values. The multiscale watershed segmentation generates a set of
edge information in the segmentation process and maps homo- nested partitions where the partition at scale s is obtained as the
geneous image regions with low-gradient values into the catch- watershed segmentation of the image gradient whose regional
ment basins and the pixels in high-contrast neighborhoods with minima with dynamics less than or equal to s are eliminated
high-gradient values into the peaks in the elevation function. by using the h-minima transform. First, the initial partition is
The gradient function for single-channel images can easily be calculated as the classical watershed corresponding to all local
2016 by Taylor & Francis Group, LLC
328 Remotely Sensed Data Characterization, Classification, and Accuracies
minima. Next, the two catchment basins having a dynamic of of the pixels with high confidence values at the end of pixel-
1 are merged with their neighbor catchment basins at scale 1. based supervised classification (Tarabalka etal., 2010a). Given
Then, at each scale s, the minima with dynamics less than or a marker image f mthat consists of pixels whose value is 0 at the
equal to s are filtered, whereas the minima with dynamics marker locations and a very large value in the rest of the image,
greater than s remain the same or are extended. This continues the minima in the input image f can be rearranged by using
until the last scale corresponding to the largest dynamic in the minima imposition. First, minima can be created only at the
gradient image. Figure16.5 illustrates the use of the h-minima locations of the markers by taking the point-wise minimum
transform for suppressing regional minima for obtaining a mul- between f + 1 and f m . Note that the resulting image is lower
tiscale watershed segmentation. than or equal to the marker image. The second step of the min-
Yet another popular approach for computing the water- ima imposition is the morphological reconstruction by ero-
shed segmentation without a significant amount of over- sion of the resulting image from the marker image f m . Finally,
segmentation is to use markers (Meyer and Beucher, 1990). watershed segmentation is applied to the resulting image. It is
Marker-controlled watershed segmentation can be defined as also possible to produce a multiscale segmentation by applying
the watershed of an input image transformed to have regional marker-controlled watershed segmentation to the input image
minima only at the marker locations. Possible methods for by using a decreasing set of markers. Marker selection is also
identifying the markers include manual selection or selection discussed in Section 16.3.4.
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)
(k) (l) (m) (n) (o)
FIg ur e 16.5 Illustration of the h-minima transform for suppressing regional minima for obtaining a multiscale watershed segmentation. The
columns represent increasing values of h, corresponding to decreasing amount of detail in the gradient data. The first row shows the gradient
information at each pixel as a topographic relief. The second row shows the gradient data as an image. Brighter values represent higher gradient.
The third row shows the segmentation boundaries obtained by the watershed algorithm in red.
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 329
16.3.3.2 Morphological Profiles inverse top-hat, or bot-hat) transform, can be used to isolate
The image representation called morphological profiles was the structures that are brighter (respectively, darker) than
popularized in the remote sensing literature by Pesaresi and their surroundings.
Benediktsson (2001). The representation uses the morphological However, to determine the shapes and sizes of all objects
residuals between the original image function and the composition present in the image, it is necessary to use a range of different
of a granulometry constructed at multiple scales. The proposed SE sizes. This concept is called granulometry. The MP of size
approach makes use of both classical morphological operators, (2k + 1) can be defined as the composition of a granulometry of
such as opening and closing, and recent theoretical advances such size k constructed with opening by reconstruction (opening pro-
as leveling and morphological spectrum to build the MP. file), the original image, and an antigranulometry of size k con-
The fundamental operators in mathematical morphology structed with closing by reconstruction (closing profile) using a
are erosion and dilation (Soille, 2003). Both of these operators sequence of k SEs with increasing sizes. Then, the derivative of
use the definition of a pixel neighborhood with a particular the morphological profile (DMP) is defined as a vector where the
shape called a structuring element (SE) (e.g., a disk of radius measure of the slope of the openingclosing profile is stored for
of three pixels). The erosion operator can be used to identify every step of an increasing SE series (see Figures 16.6 and 16.7
the image locations where the SE fits the objects in the image for the illustration of opening and closing profiles, respectively).
and is defined as the infimum of the values of the image func- Pesaresi and Benediktsson (2001) used DMP for image seg-
tion in the neighborhood defined by SE. The dilation operator mentation. They defined the size of each pixel as the SE size at
can be used to identify the pixels where the SE hits the objects which the maximum DMP is achieved. Then, they defined an
in the image and is defined as the supremum of the image image segment as a set of connected pixels showing the greatest
values in the neighborhood defined by SE. These two opera- value of the DMP for the same SE size. That is, the segment label
tors can be combined to define other operators. For example, of each pixel is assigned according to the scale corresponding
the opening operator, which is defined as the result of erosion to the largest derivative of its profile. This scheme works well
followed by dilation using the same SE, can be used to cut the in images where the structures are mostly flat so that all pixels
peaks of the topographic relief that are smaller than the SE. in a structure have only one derivative maximum. A potential
On the other hand, the closing operator, which is defined as drawback of this scheme is that neighborhood information is not
the result of dilation followed by erosion using the same SE, used at the final step of assigning segment labels to pixels. This
can be used to fill the valleys that are smaller than the SE. may result in an oversegmentation consisting of small noisy seg-
The morphological operations are often used with the non- ments in very high spatial resolution images with non-flat struc-
Euclidean geodesic metric instead of the classical Euclidean met- tures where the scale with the largest value of the DMP may not
ric (Pesaresi and Benediktsson, 2001). The elementary geodesic correspond to the true structure.
dilation of f (called the marker) under g (called the mask) based Akcay and Aksoy (2008) proposed to consider the behavior of
on SE is the infimum of the elementary dilation of f (with SE) and the neighbors of a pixel while assigning the segment label for that
g. Similarly, the elementary geodesic erosion of f under g based pixel. The method assumes that pixels with a positive DMP value
on SE is the supremum of the elementary erosion of f (with SE) at a particular SE size face a change with respect to their neighbor-
and g. A geodesic dilation (respectively, erosion) of size k can hoods at that scale. As opposed to Pesaresi and Benediktsson (2001)
also be obtained by performing k successive elementary geode- where only the scale corresponding to the greatest DMP is used, the
sic dilations (respectively, erosions). Next, the reconstruction by main idea is that a neighboring group of pixels that have a similar
dilation (respectively, erosion) of f under g is obtained by the itera- change for any particular SE size is a candidate segment for the final
tive use of an elementary geodesic dilation (respectively, erosion) segmentation. These groups can be found by applying connected
of f under g until idempotence is achieved. Then, the opening by components analysis to the DMP at each scale. For each opening
reconstruction of an image f can be defined as the reconstruction and closing profile, through increasing SE sizes from 1 to m, each
by dilation of the erosion under the original image. Similarly, the morphological operation reveals connected components that are
closing by reconstruction of the image f can be defined as the dual contained within each other in a hierarchical manner where a pixel
reconstruction by erosion of the dilation above the original image. may be assigned to more than one connected component appear-
The advantage of the reconstruction filters is that they do ing at different SE sizes. Each component is treated as a candidate
not introduce discontinuities and, therefore, preserve the meaningful segment (see Figures 16.6 and 16.7). Using these seg-
shapes observed in the input images. Hence, the opening and ments, a tree is constructed where each connected component is
closing by reconstruction operators can be used to identify a node and there is an edge between two nodes corresponding to
the sizes and shapes of different objects present in the image two consecutive scales if one node is contained within the other.
such that opening (respectively, closing) by reconstruction Leaf nodes represent the components that appear for SE size1. Root
preserves the shapes of the structures that are not removed nodes represent the components that exist for SE size m.
by erosion (respectively, dilation), and the residual between After forming a tree for each opening and closing pro-
the original image and the result of opening (respectively, file, the goal is to search for the most meaningful connected
closing) by reconstruction, called the top-hat (respectively, components among those appearing at different scales in
the segmentation hierarchy. Ideally, a meaningful segment
2016 by Taylor & Francis Group, LLC
330 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)
(k) (l) (m) (n) (o)
(p) (q) (r) (s) (t)
FIg u r e 16.6 Illustration of the opening profile obtained using increasing SE sizes. Each row shows the results for an increasing SE series. The
first column shows the input spectral data as a topographic relief. The second column shows the SEs used. The third column shows the result of
opening by reconstruction of the topographic relief with the corresponding SEs. The fourth column shows the derivative of the opening morpho-
logical profile. The fifth column shows the boundaries of the connected components having a nonzero derivative profile for the corresponding SE
for a multiscale segmentation
is expected to be spectrally as homogeneous as possible. of interest corresponds to the scale right before this change.
However, in the extreme case, a single pixel is the most homo- In other words, if the nodes on a path in the tree stay homo-
geneous. Hence, a segment is also desired to be as large as geneous until some node n, and then the homogeneity is lost
possible. In general, a segment stays almost the same (both in the next level, it can be said that n corresponds to a mean-
in spectral homogeneity and in size) for some number of ingful segment in the hierarchy. With this motivation, to
SEs and then faces a large change at a particular scale either check the meaningfulness of a node, Akcay and Aksoy (2008)
because it merges with its surroundings to make a new struc- defined a measure consisting of two factors: spectral homo-
ture or because it is completely lost. Consequently, the size geneity, which is calculated in terms of the difference of the
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 331
(a) (b) (c) (d) (e)
(f) (g) (h) (i) (j)
(k) (l) (m) (n) (o)
(p) (q) (r) (s) (t)
FIg u r e 16.7 Illustration of the closing profile obtained using increasing SE sizes. Each row shows the results for an increasing SE
series.Thefirst olumn shows the input spectral data as a topographic relief. The second column shows the SEs used. The third column shows the
result of closing by reconstruction of the topographic relief with the corresponding SEs. The fourth column shows the derivative of the closing mor-
phological profile. The fifth column shows the boundaries of the connected components having a nonzero derivative profile for the corresponding
SE for a multiscale segmentation.
standard deviation of the spectral features of the node and its After the tree is finalized, each node is regarded as a candidate
parent, and neighborhood connectivity, which is calculated segment for the final segmentation. Given the goodness measure
using sizes of connected components. Then, starting from the of each node in the hierarchy, the segments that optimize this
leaf nodes (level 1) up to the root node (level m), this measure measure are selected by using a two-pass algorithm that satisfies
is computed at each node, and a node is selected as a mean- the following conditions. Given N as the set of all nodes and P as
ingful segment if it is highly homogeneous and large enough the set of all paths in the tree, the algorithm selects N * Nas the
on its path in the hierarchy (apath corresponds to the set of final segmentation such that any node in N *must have a measure
nodes from a leaf to the root). greater than all of its descendants, any two nodes in N *cannot
2016 by Taylor & Francis Group, LLC
332 Remotely Sensed Data Characterization, Classification, and Accuracies
be on the same path (i.e., the corresponding segments cannot G such that the sum of the edge weights of T *is minimal among
overlap in the hierarchical segmentation), and every path must all the possible spanning trees of G. The minimum spanning for-
include a node that is in N * (i.e., the segmentation must cover est (MSF) rooted on a set of m distinct vertices {t1, , tm}consists
the whole image). The first pass finds the nodes having a measure in finding a spanning forest F* = (V,EF * )of G, such that each
greater than all of their descendants in a bottom-up traversal. The distinct tree of F * is grown from one root ti, and the sum of the
second pass selects the most meaningful nodes having the largest edge weights of F * is minimal among all the spanning forests of
measure on their corresponding paths of the tree in a top-down G rooted on {t1, , tm}.
traversal. The details of the algorithm can be found in Akcay The MSF-based segmentation typically consists of two steps:
and Aksoy (2008). Even though the algorithm was illustrated
using a tree constructed from a DMP, it is a generic selection 1. The objective of this step is to select a marker, or region
algorithm in the sense that it can be used with other hierarchi- seed, for each spatial object in the image. Such region
cal image partitions, such as the ones described in Section 16.3.1, seeds {t1, , tm} can be manually selected from image
and can be applied to specific applications by defining different pixels via interactive image analysis software; however,
goodness measures for desired image segments (e.g., see Genctav automatic marker selection is highly desirable. Markers
etal. (2012) for an application of this selection algorithm to an are often defined by automatically searching flat zones
hierarchical segmentation produced by a multiscale watershed (i.e., connected components of pixels of constant intensity
procedure). value), zones of homogeneous texture, or image extrema
(Soille, 2003). Tarabalka et al. (2010a) proposed to per-
form a supervised probabilistic classification of each pixel
16.3.4 Graph-Based Algorithms
(i.e., compute probabilities for each pixel to belong to each
Graph-based segmentation techniques gained popularity in of the land categories of interest) and to choose the most
recent years. In the graph-based framework, the image is mod- reliably classified pixels as markers of spatial regions.
eled by a graph, where nodes typically represent individual pix- 2. Image pixels are grouped into an MSF, where each tree is
els or regions, while edges connect spatially adjacent nodes. The rooted on a marker. To compute an MSF, an additional root
weights of the edges reflect the (dis)similarity between the neigh- vertex r is added and is connected by the null-weight edges
boring pixels/regions linked by the edge. The general idea is then to the marker vertices ti. The minimum spanning tree of
to find subgraphs in this graph, which correspond to regions in the constructed graph induces an MSF in G, where each
the image scene. The early graph-theoretic approaches for image tree is grown on a marker vertex ti; the MSF is obtained
segmentation were described in Zahn (1971), where a minimum after removing the vertex r. The two most commonly used
spanning tree was used to produce connected groups of vertices, algorithms for computing a minimum spanning tree are
and Narendra and Goldberg (1977), where directed graphs have Prims (1957) and Kruskals (1956) algorithms.
been employed to define regions in edge-detected images. In this
section, we will review two algorithms that have been success- The watershed transform described in the previous section
fully applied for remote sensing applications: optimal spanning can be efficiently built by computing an MSF rooted on the
forests and normalized cuts. image minima (Cousty etal., 2009). For this purpose, an ultra-
metric flooding distance has to be used to compute weights of
16.3.4.1 optimal Spanning Forests edges (Meyer, 2005). This distance is defined as the minimal
The optimal spanning forest segmentation is based on the mini- level of flooding for which two pixels belong to the same lake.
mum spanning tree algorithm introduced by Kruskal (1956) and Meyer (2005) showed that an MSF can also be efficiently
Prim (1957). It was employed for segmentation of remote sensing computed from a minimum spanning tree of image pixels,
images in Tarabalka etal. (2009) and Skurikhin (2010). without introducing an additional root vertex r, as depicted in
We denote an image undirected graph as G = (V, E, W), Figure 16.8. This algorithm is useful if the initial markers can
where each pixel is considered as a vertex v V, each edge be modified (e.g., suppression and addition of markers during
ei,j Econnects a couple of vertices i and j corresponding to interactive segmentation). Given the minimum spanning tree
the neighboring pixels. Furthermore, a weight wi,jis assigned Tof a graph G = (V, E, W) and a set of markers {t1,,tm}, the
to each edge ei,j, which indicates the degree of dissimilarity edges of T are first sorted in the order of their decreasing weights
between two pixels connected by this edge. Different dissimi- and are considered one after another. Suppose that e is the edge
larity measures can be used to compute weights of edges, such currently under consideration. The edge e belongs to a sub-tree
as vector norms, SAM, or spectral information divergence of T. Suppressing e will cut this tree into two smaller sub-trees; if
(Tarabalka etal., 2010a). each of them contains at least one marker, then the suppression
Given a connected graph G = (V, E), a spanning tree T= (V,ET)of of e is validated (this is the case in Figure 16.8c through e, where
G is a connected graph without cycles such that ET E. A span- an edge has been suppressed each time); if at least one of the sub-
ning forest F = (V, EF) of G is a nonconnected graph without trees does not contain a marker, then the edge e is reintroduced.
cycles such that EF E. Given a graph G = (V, E, W), the mini- The process stops when each of the created sub-trees contains
mum spanning tree is defined as a spanning tree T * = (V,ET * )of one and only one marker. This algorithm outputs an MSF with
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 333
4 4
0 0 0 0 0 0
3 3 3 3
3 7 1 2 3 7 1 2
0 0 0 0 0 1 0 0 2 0
9 5 9 5
1 1
0 0 0 0 0 0
2 4 2 4
8 3 2 8 3 2
0 0 0 0 0 0 0 0 3 0
2 3 4 2 3 4
2 2
0 0 0 0 0 0 4 0
(a) (b)
4 4
0 0 0 0 0 0
3 3 3 3
3 7 1 2 3 7 1 2
1 0 0 2 0 1 1 0 2 0
9 5 9 5
1 1
0 0 0 1 0 0
2 4 2 4
8 3 2 8 3 2
0 4 0 3 0 1 4 0 3 0
2 3 2 3 4
2 4 2
0 0 4 4 1 1 4 4
(c) (d)
2 4 2 2
3 3
3 7 1 2
1 1 2 2 2
9 5
1
1 3 3
2 4
8 3 2
1 4 3 3 3
2 3 4
2
1 1 4 4
(e)
FIg u r e 16.8 Example of construction of an MSF rooted on markers from a minimum spanning tree. (a) The initial minimum spanning tree.
(b) Four markers defined by the colored nodes. (c)(d) Illustration of the construction of the MSF from the four markers by highest weight edge
suppression. (e) Final MSF, where each tree has the color of its marker.
one tree rooted in each marker. It was applied in Bernard etal. maximizing the similarity within the regions. 16.9 shows an
(2012) for segmentation of hyperspectral remote sensing images. example graph, where the pixels in group X are strongly con-
If the markers of the image regions cannot be reliably found, a nected with high similarities between adjacent pixels, shown
similar algorithm can be iteratively applied, by suppressing the as thick red lines, as are the pixels in group Y. The connections
edge of the minimum spanning tree with the highest weight at between groups X and Y, shown as blue lines, are much weaker.
each iteration until convergence. A threshold for the edge weight A normalized cut between these two groups separates them into
can control in this case the convergence. two clusters.
The cut between two groups X and Y is computed as the sum
of the weights of all edges being cut:
16.3.4.2 normalized cuts
The normalized cuts segmentation method introduced by Shi
and Malik (2000) aims at partitioning the image in the way cut (X ,Y ) = w i,j , (16.22)
to minimize the similarity between adjacent regions while iX , j Y
2016 by Taylor & Francis Group, LLC
334 Remotely Sensed Data Characterization, Classification, and Accuracies
of minimization of suitable energy functions. While earlier algo-
X Y rithms for optimizing MRF energy, such as iterated conditional
modes (ICM) and simulated annealing (Solberg et al., 1996;
Y Tarabalka etal., 2010b), were time consuming, more advanced
X
methods, such as graph cuts (Boykov etal., 2001; Li etal., 2012),
Y provided powerful alternatives from both theoretical and com-
X Y putational viewpoints, resulting in a growing use of the MRF-
based segmentation techniques.
For land categorization applications, MRFs are usually
applied in the framework of image classification, where the out-
FIg u r e 16.9 Example of a weighted graph and its normalized cut put is a landcover map, with every region assigned to one of the
(shown as a dashed line). thematic classes. The commonly used MRF energy function in
this case is computed as a linear combination of two terms:
where the weights of the edges between two vertices i and j
measure the similarity between the corresponding pixels (or E (L) = E data + E smooth . (16.24)
regions). The optimal bipartitioning of the graph is the one
n
that minimizes this cut value, that is, to find the minimum cut The first term E data = Vi (Li ) is related to pixelwise informa-
of a graph. However, because the value of the cut computed by i =1
tion, and it measures for each pixel the disagreement between a
Equation 16.22 increases with the number of edges separating
prior probabilistic model and the observed data. Thus, individ-
two partitions, using a minimum cut as a segmentation criterion
ual potentials Vi(Li)measure a penalty for a pixel i (i = 1, 2, , n)
favors keeping in one group isolated pixels or small sets of iso-
to have a label Li. This term is often formulated in terms of the
lated nodes in the graph.
probability density function of feature vectors conditioned to
To avoid such partitioning of small sets of nodes into sepa-
the related class label:
rate groups, Shi and Malik (2000) proposed a new measure of
disassociation between two groups. The new measure, called a Vi (Li ) = ln(p(x i | Li )). (16.25)
normalized cut, computes the cut cost as a fraction of the total
edge connections to all the nodes in the graph: This probability density can be estimated based on the available
training samples for each landcover class. A discussion of the
cut (X ,Y ) cut (X ,Y ) main approaches for this estimation problem can be found in
Ncut (X ,Y ) = + , (16.23)
assoc (X ,V ) assoc (Y ,V ) Duda etal. (2001).
The second contribution E smooth =
Wi , j (Li , L j ) expresses
i~j
where assoc (X ,V ) = wi ,v is the sum of all weights from interaction between neighboring pixels, thus exploiting image
iX ,v V
nodes in X to all nodes in the graph, and assoc(Y, V) is similarly spatial context. i ~ j denotes a pair of spatially adjacent pixels,
defined. The normalized cut defined by Equation 16.23 better and Wi,j(Li,Lj) is an interaction term for these pixels. Most
reflects the fitness of a particular segmentation, because it seeks works in remote sensing image classification use a Potts model
to cut a collection of edges that are weak relatively to all of the (Tarabalka etal., 2010b; Li etal., 2012) to compute this spatial
edges both inside and emanating from each of the regions. Shi term, which favors spatially adjacent pixels to belong to the same
and Malik (2000) proposed an efficient algorithm based on a class (or spatial region):
generalized eigenvalue problem to optimize the normalized cut
Wi , j (Li , L j ) = (1 (Li , L j )), (16.26)
criterion.
Numerous works employed a normalized cut algorithm for where () is the Kronecker function ((a, b) = 1 for a = b and (a, b)
segmentation of remote sensing images, such as Grote et al. = 0 otherwise), and is a positive constant parameter that controls
(2007) and Sung and He (2009). Normalized cuts were also the importance of spatial smoothing. This model tends to deterio-
applied in combination with other techniques. For instance, Jing rate classification results at the edges between landcover classes
et al. (2010) first used watershed algorithm to find initial seg- and near small-scale details. In order to preserve the border in the
ments, and then the normalized cuts technique grouped these output thematic map, edge functions have been proposed and inte-
segments into the final regions. grated in the spatial energy term, such as Tarabalka etal. (2010b):
t
16.3.5 MRF-Based Algorithms Wi , j (Li , L j ) = (1 (Li , L j )) , (16.27)
t +|i | j |
Markov random fields (MRFs) are probabilistic graphical mod-
els that conceptually generalize the notion of Markov chain where
(Moser etal., 2013; Wang etal., 2013). They provide a flexible tool i|j is the gradient value of the pixel i in the direction of j
to include spatial context into image analysis schemes in terms t is a parameter controlling the fuzzy edge threshold
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 335
To optimize the MRF energy, different methods were proposed
and applied for remote sensing applications, as described in the
Source
following subsections.
t-links
16.3.5.1 Simulated Annealing and
iterated conditional Modes 2D Image
plane
The main idea of simulated annealing described in Kirkpatrick
etal. (1983) is to iteratively propose a change from the current
configuration of pixel labels (region or class labels) L to the new e-links
randomly generated configuration of labels, and to probabilis-
tically decide if this change is accepted or not. This procedure
yields configurations of labels with the lower MRF energy.
Stewart etal. (2000) applied MRF-based method with simu-
lated annealing optimization for segmentation and classifica-
tion of SAR data. They used Gamma distribution to model the Sink
data energy term and a Potts model to compute the spatial term.
Tarabalka etal. (2010b) applied simulated annealing technique FIg u r e 16.10 Mapping of an image to the graph for the graph-cut
for the classification of hyperspectral remote sensing images. optimization.
They performed probabilistic SVM classification to derive the
data term and employed the spatial energy contribution with
the fuzzy edge function described previously. At each itera- MRF energy (see Equation 16.24) is optimized by solving the
tion, a new class label Lnew
i was randomly selected for a randomly minimum-cut/maximum-flow problem, yielding a binary seg-
selected pixel xi, and a new local energy Enew(xi)was computed. mentation map.
If the variation of the energy E = Enew(xi)E(xi)<0, the new Boykov et al. (2001) proposed two efficient approximation
class label is accepted. Otherwise, the new class assignment is algorithms based on graph-cuts (-expansion and -swap)
accepted with the probability p = exp(E/T), where T is a global to solve multi-label classification problems. In remote sensing,
parameter called temperature. The optimization begins at a high Denis etal. (2009) applied a graph-cut-based algorithm for regu-
temperature, which is gradually lowered as the iteration proce- larization of SAR images. Li etal. (2012) and Tarabalka and Rana
dure proceeds. This helps to avoid converging to local minima. (2014) used an -expansion optimization of the MRF energy for
The ICM optimization algorithm proposed in Besag (1986) the segmentation of hyperspectral images.
is conceptually simpler and computationally less expensive. It
iteratively considers every pixel xiand assigns a (region or class) 16.4 image Segmentation
label to this pixel, which minimizes the local energy centered at Quality evaluation
xi. Solberg etal. (1996) applied the ICM optimization for multi-
source classification of optical, SAR, and geographic information We noted in the introduction to this chapter that there is no gen-
systems images. Farag etal. (2005) employed the ICM method eral theory of image segmentation, and most image segmenta-
for hyperspectral image classification. The density functions in tion approaches are ad hoc in nature. This is also the case with
this case were estimated by mean field-based SVM regression. image segmentation quality evaluation. In practice, image seg-
The drawback of the ICM algorithm is that it is suboptimal and mentation quality is best evaluated by how well the goals of the
converges only to a local minimum of the energy function. analysis utilizing the segmentation are served.
Several general approaches to quantitative segmentation
16.3.5.2 Graph cuts quality evaluation are listed and described in the series of papers
published by the research group based at the Leibniz IOER men-
Boykov and Jolly (2001) were the first to apply the graph-cut
tioned in the introduction to this chapter, in particular Neubert
optimization technique proposed by Greig et al. (1989) to
etal., (2006). Among the evaluation approaches discussed and
binary image segmentation. This algorithm is computationally
employed are the following:
efficient, and it gives a globally optimal solution to the binary
segmentation problem. In the approach of Boykov and Jolly 1. Fragmentation and area-fit-index (Strasters and
(2001), all pixels are connected by t-links to the additional two Gerbrands, 1991; Lucieer, 2004): These measures address
nodes, called source and sink, respectively, as depicted in Figure over- and under-segmentation by analyzing the number
16.10. The weights of these edges are computed as the potentials of segmented and reference regions.
Vi(Li)(see Equation 16.25), so that pixels that are more compat- 2. Geometric features circularity and geometric features shape
ible with the foreground or background region get stronger con- index (Yang etal., 1995; Neubert and Meinel, 2003): These
nections to the respective source or sink. The e-links between measures address the shape conformity between segmen-
adjacent pixels are assigned weights Wi,j(Li,Lj). The resulting tation and reference regions.
2016 by Taylor & Francis Group, LLC
336 Remotely Sensed Data Characterization, Classification, and Accuracies
3. Empirical evaluation function (Borsotti etal., 1998): This of 1.3 m. The ROSIS sensor has 115 spectral channels, with a
measure addresses how uniform a feature is within seg- spectral range of 0.430.86m. The 12 noisiest channels were
mented regions. removed, and the remaining 103 spectral bands were used in
4. Entropy-based evaluation function and a weighted disor- this experiment. See Tarabalka et al. (2010) for more details
der function (Zhang and Gerbrands, 1994): This measure on this data set. The reference data contain nine ground cover
addresses the uniformity within segmented regions using classes: asphalt, meadows, gravel, trees, metal sheets, bare soil,
entropy as a criterion of disorder. bitumen, bricks, and shadows. A three-band false color image
5. Fitness function (Everingham et al., 2002): This measure of this data set and the ground reference data are shown in
addresses multiple criteria and parameterizations of algo- Figure 16.11.
rithms by a probabilistic fitness/cost analysis. We note that this data set is not rigorously geo-referenced.
North is roughly toward the bottom of the image, and the area
Neubert etal. (2006) also note that the vast majority of quantita- covered is 793 m 442 m.
tive image segmentation approaches are empirical discrepancy The classification accuracy results are listed in Table 16.2
methods that analyze the number of misclassified pixels in rela- for the pixelwise SVM classification and the PV region-based
tion to a reference classification. In the remainder of this sec- classification approach for several region-growing approaches.
tion, we describe and demonstrate such an approach to image Figure 16.12 shows the corresponding classification maps.
segmentation quality evaluation. The first step in this evaluation The classification maps for the PV region-based classification
approach is to perform a pixelwise classification of an image data approach appear smoother than the pixelwise SVM classifica-
set. In our example, we create our pixelwise classification using tion, with the D8- and SEGEN-based classification appearing
the SVM classifier. Then, a region classification is obtained by the smoothest. All of the PV region-based classification accu-
assigning each spatially connected region from the segmenta- racies are substantially higher that the accuracy of the SVM
tion result to the most frequently occurring class within the pixelwise classification, with the D8, SEGEN, and HSeg PV
region. The SVM classifier and this plurality vote (PV) region- region-based classifications notably more accurate than the PV
based classification approach are described in more detail in region-based classifications based on Muerle Allen or HSWO
Tilton etal. (2012). segmentation. The HSeg-based classification has a marginally
Our test data set is the University of Pavia data set that higher classification than the other region-growing approaches.
was recorded by the Reflective Optics System Imaging Table 16.2 and Figure 16.6 also show classification results
Spectrometer (ROSIS) over the University of Pavia, Pavia, Italy. obtained by using the graph-based segmentation approaches.
The image is 610 340 pixels in size, with a spatial resolution The MSF segmentation with PV was performed as described in
(a) (b)
Asphalt Bare Gravel Bricks Metal Meadows Bitumen Trees Shadows Unlabeled
soil sheets
(c)
FIg u r e 16.11 (a) Three-band false color image of the University of Pavia hyperspectral data set (RGB = bands 56, 33, and 13). (b) Reference data.
(c) Color key. The training data for the SVM classifier and segmentation optimization are two separate randomly selected subsets of the reference data.
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 337
TABLe 1 6.2 Comparison of Percentage Classification Accuracies on the University of Pavia Hyperspectral Data Set for Per-Pixel SVM and
with the Plurality Vote (PV) Region-Based Classification Method for the Region Muerle Allen, HSWO, Definiens 8.0 (D8), SEGEN, and HSeg
SVM Muerle Allen HSWO + PV D8 + PV SEGEN + PV HSeg + PV (Sw = 0.3) MSF + PV Graph-Cut + PV
OA 89.03 95.35 95.38 97.54 98.09 98.35 96.99 98.49
AA 89.56 95.26 95.50 97.26 97.95 98.15 97.01 97.73
85.46 93.78 93.83 96.71 97.45 97.79 95.99 98.47
HSeg was performed with small region merge acceleration. Results are also included for two graph-based approachesthe minimum spanning forest (MSF)
with plurality vote (PV) method and MRF-based method using the -expansion graph-cut algorithm. Percentage classification accuracies in terms of OA, AA,
and Kappa coefficient ().
(a) (b) (c) (d)
(e) (f) (g) (h)
Asphalt Bare soil Gravel Bricks Metal sheets Meadows Bitumen Trees Shadows Unlabeled
(i)
FIg u r e 16.12 (a) SVM classification. (b) PV classification with Muerle Allen segmentation. (c) PV classification with HSWO segmentation.
(d)PV classification with D8. (e) PV classification with SEGEN segmentation. (f) PV classification with HSeg segmentation (Sw = 0.3). (g) Minimum
spanning forest (MSF) with plurality vote (PV) classification map. (h) Classification map of the -expansion graph-cut algorithm. (i) Color key.
2016 by Taylor & Francis Group, LLC
338 Remotely Sensed Data Characterization, Classification, and Accuracies
Tarabalka etal. (2010a), by using the L1 norm between spectral References
vectors to compute weights of edges between adjacent pixels.
For the graph-cut method, probabilistic SVM classification was Adams, R. and L. Bischof. 1994. Seeded region growing. IEEE
applied to derive the data energy term, and a Potts model was Trans. Pattern Anal. Mach. Intell., 16(6):641646.
used to express the spatial term. We fixed the parameter = 1.5 Akcay, H. G., S. Aksoy, and P. Soille. 2010. Hierarchical segmen-
as suggested in Tarabalka etal. (2010b). The MRF-based method tation of complex structures. Proceedings of ICPR, Istanbul,
using the -expansion optimization yields the highest overall Turkey.
classification accuracy. Akcay, H. G. and S. Aksoy. 2008. Automatic detection of geospa-
The classifications are evaluated in terms of overall accuracy tial objects using multiple hierarchical segmentations. IEEE
(OA), average accuracy (AA), and the kappa coefficient (). OA Trans. Geosci. Remote Sens., 46(7):20972111.
is the percentage of correctly classified pixels, AA is the mean Akcay, H. G. and S. Aksoy. 2011. Detection of compound struc-
of the class-specific accuracies, and is the percentage of agree- tures using hierarchical clustering of statistical and struc-
ment (correctly classified pixels) corrected by the number of tural features. Proceedings of IGARSS, Vancouver, British
agreements that would be expected purely by chance (e.g., see Columbia, Canada.
Richards and Jia, 1999). Aksoy, S., H. G. Akcay, and T. Wassenaar. 2010. Automatic map-
ping of linear woody vegetation features in agricultural
16.5 concluding Remarks landscapes using very-high resolution imagery. IEEE Trans.
Geosci. Remote Sens., 48(1):511522.
Land categorization applications are quite varied. They range Aksoy, S., I. Z. Yalniz, and K. Tasdemir. 2012. Automatic detec-
from land cover monitoring (e.g., croplands, forests, wetlands, tion and segmentation of orchards using very high-
and urbanization), snow and ice mapping, geology and mineral resolution imagery. IEEE Trans. Geosci. Remote Sens.,
exploration, and even wildfire and agricultural burn monitor- 50(8):31173131.
ing. Each land categorization will have its particular image Aksoy, S., K. Koperski, C. Tusk, G. Marchisio, and J. C. Tilton.
analysis needs that will be best served by a particular class 2005. Learning Bayesian classifiers for scene classification
of image segmentation algorithm. Generally, region-grow- with a visual grammar. IEEE Trans. Geosci. Remote Sens.,
ing approaches may be best suited for finding homogeneous 43(3):581589.
regions, while if locating region borders are more important, Aksoy, S. 2006. Spatial techniques for image classification. In:
watershed approaches may be more effective. Region-growing C. H. Chen, ed. Signal and Image Processing for Remote
and morphological approaches that produce segmentation Sensing. Boca Raton, FL: CRC Press, pp. 491513.
hierarchies would be most appropriate if analysis of the image Aksoy, S. and H. G. Akcay. 2005. Multi-resolution segmentation
data at different scales is desired. and shape analysis for remote sensing image classification.
We have demonstrated through a simple PV classification Proceedings of the Second International Conference of Recent
approach that utilizing image segmentation can dramatically Advances in Space Technologies, Istanbul, Turkey.
improve image classification results both qualitatively and Aptoula, E. and S. Lefevre. 2007. A comparative study on mul-
quantitatively. Another advantage of image segmentation is tivariate mathematical morphology. Pattern Recognit.,
that image segmentation defines spatial objects with particular 40(11):29142929.
shapes and sizes that can be utilized to perform a deeper, object- Baatz, M. and A. Schape. 2000. Multiresolution segmentation:
based analysis of the image scene than that is possible with pixel- An optimizing approach for high quality multi-scale seg-
based analysis. In the context of land categorization, this form of mentation. In: J. Strobl and T. Blaschke, eds. Angewandte
analysis has led to the establishment of the new subdiscipline of Geographich Informationsverarbeitung, XII. Heidelberg,
Geographic Object-Based Image Analysis (GEOBIA) and a pop- Germany: Wichmann, pp. 1223.
ular series of biennial conferences starting with GEOBIA 2006 Beaulieu, J.-M. and M. Goldberg. 1989. Hierarchy in picture
in Salzburg, Austriaand most recently with GEOBIA 2014 in segmentation: A stepwise optimal approach. IEEE Trans.
Thessaloniki, Greece (Gitas etal., 2014). Pattern Anal. Mach. Intell., 11:150163.
We have described and discussed a wide range of image Benz, U., P. Hofmann, G. Wilhauck, I. Lingenfelder, and
segmentation approaches developed over the past century, M. Heynen. 2004. Multi-resolution, objected-oriented
focusing on those approaches most applicable to the analysis fuzzy analysis of remote sensing data for GIS-ready
of remotely sensed imagery data for land categorization. The information. ISPRS J. Photogramm. Remote Sens.,
wide range of approaches included in our discussion and the 58(3/4):239258.
ongoing active research are a clear indication that there still Bernard, K., Y. Tarabalka, J. Angulo, J. Chanussot, and J. A.
is no general theory of segmentation and that most of the suc- Benediktsson. 2012. Spectral-spatial classification of hyper-
cessful image segmentation approaches are rather ad hoc in spectral data based on a stochastic minimum spanning for-
nature. But the wide range of approaches provides a rich menu est approach. IEEE Trans. Image Proc., 21(4):20082021.
to choose from for image analysis practitioners in tailoring Besag, J. 1986. On the statistical analysis of dirty pictures. J. R.
their image analysis approach to their particular application. Stat. Soc., 48(3):259302.
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 339
Borsotti, M., P. Campadelli, and R. Schettini. 1998. Quantitative Duda, R. O., P. E. Hart, and D. G. Stork. 2001. Pattern Classification,
evaluation of color image segmentation results. Pattern 2nd edn. New York: Wiley.
Recognit. Lett., 19(8):741747. Epifanio, I. and P. Soille. 2007. Morphological texture features for
Boykov, Y., O. Veksler, and R. Zabih. 2001. Fast approximate unsupervised and supervised segmentations of natural land-
energy minimization via graph cuts. IEEE Trans. Pattern scapes. IEEE Trans. Geosci. Remote Sens., 45(4):10741083.
Anal. Mach. Intell., 23(11):12221239. Everingham, M., H. Muller, and B. Thomas. 2001. Evaluating
Boykov, Y. and M.-P. Jolly. 2001. Interactive graph cuts for opti- image segmentation algorithms using monotonic hulls
mal boundary and region segmentation of objects in N-D in fitness/cost space. In: T. Cootes and C. Taylor, eds.
images. Proceedings of ICCV, Vol. 1, Vancouver, British Proceedings of the 12th British Machine Vision Conference,
Columbia, Canada, pp. 105112. Manchester, U.K., pp. 363372.
Bruzzone, L. and D. F. Prieto. 2005. Automatic analysis of the Exelis Visual Information Solutions. 2015. ENVI Environment
difference image for unsupervised change detection. IEEE for Visualizing Images (using ENVI). http://www.exelisvis.
Trans. Geosci. Remote Sens., 38(3):11711182. com/docs/using_envi_Home.html, last accessed March 31,
Castilla, G., G. J. Hay, and J. R. Ruiz. 2008. Size-constrained 2015.
region merging (SCRM): An automated delineation tool for Farag, A. A., R. M. Mohamed, and A. El-Baz. 2005. A uni-
assisted photointerpretation. PE&RS, 74(4):409419. fied framework for MAP estimation in remote sensing
Clark Labs. 2015. Clark Labs Image Processing. http:// image segmentation. IEEE Trans. Geosci. Remote Sens.,
www.clarklabs.org/products/idrisi-image-processing.cfm, 43(7):16171634.
last accessed March 31, 2015. Fauvel, M., Y. Tarabalka, J. A. Benediktsson, J. Chanussot, and
Clinton, N., A. Holt, J. Scarborough, L. Yan, and P. Gong. 2010. J. C. Tilton. 2013. Advances in spectral-spatial classifi-
Accuracy assessment measures for object-based image cation of hyperspectral images. Proceedings of the IEEE,
segmentation goodness. Photogrammetric Engineering and 101(3):652675.
Remote Sensing, 76(3):289-299. Fu, K. S. and J. K. Mui. 1981. A survey on image segmentation.
CNES. 2015. Start using Orfeo Toolbox | Orfeo ToolBox. https:// Pattern Recognit., 13:316.
www.orfeo-toolbox.org/start/, last accessed March 31, 2015. Gaetano, R., G. Scarpa, and G. Poggi. 2009. Hierarchical texture-
Comaniciu, D. and P. Meer. 2002. Mean shift: A robust approach based segmentation of multiresolution remote-sensing
toward feature space analysis. IEEE Trans. Pattern Anal. images. IEEE Trans. Geosci. Remote Sens., 47(7):21292141.
Mach. Intell., 24(5):603619. Genctav, A., S. Aksoy, and S. Onder. 2012. Unsupervised seg-
Cook, R., I. McConnell, D. Stewart, and C. J. Oliver. 1996. Segmentation mentation and classification of cervical cell images. Pattern
and simulated annealing. In: G. Franceschetti, F. S. Rubertone, Recognit., 45(12):41514168.
C. J. Oliver, and S. Tajbakhsh, eds. Microwave Sensing and Ghamisi, P., M. S. Couceiro, F. M. L. Martins, and J. A.
Synthetic Aperture Radar, Taormina, Italy; Proceedings of SPIE, Benediktsson. 2014. Multilevel image segmentation based
Vol. 2958. Bellingham, WA: SPIE, pp. 3035. on fractional-order Darwinian particle swarm optimiza-
Couprie, C., L. Grady, L. Najman, and H. Talbot. 2011. tion. IEEE Trans. Geosci. Remote Sens., 52(5):23822394.
Power watershed: A unifying graph-based optimiza- Gitas, I., G. Mallinis, P. Patias, D. Stathakis, and G. Zalidis (Guest
tion framework. IEEE Trans. Pattern Anal. Mach. Intell., eds). 2014. GEOBIA 2014, Advancements, Trends and
33(7):13841399. Challenges, Fifth Geographic Object-Based Image Analysis
Cousty, J., G. Bertrand, L. Najman, and M. Couprie. 2009. Conference, Thessaloniki, Greece, May 2124, 2014.
Watershed cuts: Minimum spanning forests and the drop Special issue of the S East. Eur. J. Earth Observ. Geomat.,
of water principle. IEEE Trans. Pattern Anal. Mach. Intell., 3(2S):1768.
31(8):13621374. Gofman, E. 2006. Developing an efficient region growing engine
Cross, A. M., D. C. Mason, and S. J. Dury. 1988. Segmentation of for image segmentation. Proceedings of ICPR, Hong Kong,
remotely sensed image by a split-and-merge process. Int. J. China, pp. 24131416.
Remote Sens., 9:13291345. GRASS Development Team. 2015. GRASS GIS Home. http://
Denis, L., F. Tupin, J. Darbon, and M. Sigelle. 2009. SAR image grass.osgeo.org/, last accessed March 31, 2015.
regularization with fast approximate discrete minimization. Greig, D., B. Porteous, and A. Seheult. 1989. Exact maximum a
IEEE Trans. Image Proc., 18(7):15881600. posteriori estimation for binary images. J. R. Stat. Soc. B,
Dogrusoz, E. and S. Aksoy. 2007. Modeling urban structures 51(2):271279.
using graph-based spatial patterns. Proceedings of IGARSS, Grote, A., M. Betenuth, M. Gerke, and C. Heipke. 2007.
Barcelona, Spain, pp. 48264829. Segmentation based on normalized cuts for the detection
Dongping, M., C. Tianyu, C. Hongyue, L. Longxiang, Q. Cheng, of suburban roads in aerial imagery. Urban Remote Sensing
and D. Jinyang. 2012. Semivariogram-based spatial band- Joint Event, Paris, France, pp. 15.
width selection for remote sensing image segmentation Gueguen, L., P. Soille, and M. Pesaresi. 2012. A new built-up
with mean-shift algorithm. IEEE Geosci. Remote Sens. Lett., presence index based on density of corners. Proceedings of
9(5): 813817. IGARSS, Munich, Germany.
2016 by Taylor & Francis Group, LLC
340 Remotely Sensed Data Characterization, Classification, and Accuracies
Haralick, R. M. and L. G. Shapiro. 1985. Survey: Image seg- Meyer, F. 2005. Grey-weighted, ultrametric and lexicographic
mentation techniques. Comput. Vis. Graph. Image Proc., distances. In: C. Ronse, L. Najman, and E. Decenciere, eds.
29(1):100132. Mathematical Morphology: 40 Years On. Dordrecht, the
Haris, K., S. N. Efstratiadis, N. Maglaveras, and A. Katsaggelos. Netherlands: Springer, pp. 289298.
1998. Hybrid image segmentation using watersheds and fast Meyer, F. and S. Beucher. 1990. Morphological segmentation.
region merging. IEEE Trans. Image Proc., 7(12): 16841699. J.Vis. Commun. Image Represent., 1(1):2146.
Horowitz, S. L. and T. Pavlidis. 1974. Picture segmentation Moser, G., S. B. Serpico, and J. A. Benediktsson. 2013. Landcover
by a directed split-and-merge procedure. Proceedings mapping by Markov modeling of spatial-contextual infor-
of the Second International Joint Conference on Pattern mation in very-high-resolution remote sensing images.
Recognition, Copenhagen, Denmark, pp. 424433. Proc. IEEE, 101(3):631651.
Jing, W., J. Hua, and W. Yubin. 2010. Normalized cut as basic Muerle, J. L. and D. C. Allen. 1968. Experimental evaluation of
tool for remote sensing image. Proceedings of International techniques for automatic segmentation of objects in a com-
Conference on ICISS, Gandhinagar, India, pp. 247249. plex scene. In: G. C. Cheng, ed. Pictorial Pattern Recognition.
Kirkpatrick, S., C. D. Gelatt, Jr., and M. P. Vecchi. 1983. Optimization Washington, DC: Thompson, pp. 313.
by simulated annealing. Science, 220(4598):671680. Najman, L. and M. Schmitt. 1996. Geodesic saliency of water-
Kruse, F. A., A. B. Lefkoff, J. W. Boardman, K. B. Heidebrecht, shed contours and hierarchical segmentation. IEEE Trans.
A. T. Shapiro, P. J. Barloon, and A. F. H. Goetz. 1993. The Pattern Anal. Mach. Intell., 18(12):11631173.
spectral image processing system (SIPS)Interactive visu- Narendra, P. M. and M. Goldberg. 1977. A graph-theoretic
alization and analysis of imaging spectrometer data. Remote approach to image segmentation. Proceedings of the IEEE
Sens. Environ., 44(2/3):145163. Computer Society Conference on Pattern Recognition and
Kruskal, J. 1956. On the shortest spanning tree of a graph and the Image Processing, Troy, NY, pp. 248256.
traveling salesman problem. Proc. Am. Math. Soc., 7:4850. Neubert, M., H. Herold, and G. Meinel. 2006. Evaluation of
Kurita, T. 1994. An efficient agglomerative clustering algorithm remote sensing image segmentation qualityFurther
for region growing. Proceedings of MVA, IAPR Workshop results and concepts. Proceedings of the First International
on Machine Vision Applications, Kawasaki, Japan, Conference on OBIA, Salzburg, Austria.
pp.210213. Neubert, M., H. Herold, and G. Meinel. 2008. Assessing
Kurtz, C., N. Passat, P. Gancarski, and A. Puissant. 2012. Extraction image segmentation qualityConcepts, methods and
of complex patterns from multiresolution remote sensing application. In: T. Blaschke, S. Lang, and G. J. Hay,
images: A hierarchical top-down methodology. Pattern eds. Object-Base Image Analysis: Spatial Concepts for
Recognit., 45(2):685706. Knowledge-Driven Remote Sensing Applications. Berlin,
Le Moigne, J. and J. C. Tilton. 1995. Refining image segmentation Germany: Springer-Verlag.
by integration of edge and region data. IEEE Trans. Geosci. Neubert, M. and G. Meinel. 2003. Evaluation of segmentation
Remote Sens., 33(3):605615. programs for high resolution remote sensing applica-
Li, J., J. M. Bioucas-Dias, and A. Plaza. 2012. spectral-spatial tions. Proceedings of Joint ISPRS/EARSel Workshop on
hyperspectral image segmentation using subspace multi- High Resolution Mapping from Space 2003, Hannover,
nomial logistic regression and Markov random fields. IEEE Germany.
Trans. Geosci. Remote Sens., 50(3):809823. Noyel, G., J. Angulo, and D. Jeulin. 2007. Morphological seg-
Li, P. and B. Xiao. 2007. Multispectral image segmentation by a mentation of hyperspectral images. Image Anal. Stereol.,
multichannel watershed-based approach. Int. J. Remote 26(3):101109.
Sens., 28(19):44294452. Otsu, N. 1979. A threshold selection method from gray-level his-
Lucieer, A. 2004. Uncertainties in segmentation and their visu- tograms. IEEE Trans. Syst. Man Cybernet., SMC-9(1):6266.
alisation. PhD thesis, International Institute for Geo- Ouzounis, G. K., M. Pesaresi, and P. Soille 2012. Differential area
Information Science and Earth Observation (ITC) and profiles: Decomposition properties and efficient computa-
University of Utrecht, Utrecht, the Netherlands. tion. IEEE Transactions on Pattern Analysis and Machine
Lucieer, A. 2015. Homepage of Arko Lucieer Parbat. http:// Intelligence, 34(8):15331548.
parbat.lucieer.net/, last accessed March 31, 2015. Perret, B., S. Lefevre, C. Collet, and E. Slezak. 2012.
Marpu, P. R., M. Neubert, H. Herold, and I. Niemeyer. 2010. Hyperconnections and hierarchical representations for
Enhanced evaluation of image segmentation results. J. Spat. grayscale and multiband image processing. IEEE Trans.
Sci., 55(1):5568. Image Proc., 30(7):1427.
Meinel, G. and M. Neubert. 2004. A comparison of segmenta- Pesaresi, M., A. Gerhardinger, and F. Kayitakire. 2008. A robust
tion programs for high resolution remote sensing data. built-up area presence index by anisotropic rotation-invari-
Proceedings of Commission IV, XXth ISPRS Congress, ant textural measure. IEEE JSTARS, 1(3):180192.
Instanbul, Turkey, pp. 10971102. Pesaresi, M. and J. A. Benediktsson. 2001. A new approach for
Meyer, F. 2001. An overview of morphological segmentation. Int. the morphological segmentation of high-resolution satellite
J. Pattern Recognit. Artif. Intell., 15(7):10891118. imagery. IEEE Trans. Geosci. Remote Sens., 39(2):309320.
2016 by Taylor & Francis Group, LLC
Image Segmentation Algorithms for Land Categorization 341
Plaza, A. J. and J. C. Tilton. 2005. Automated selection of results in Stewart, D., B. Blacknell, A. Blake, R. Cook, and C. Oliver. 2000.
hierarchical segmentations of remotely sensed hyperspec- Optimal approach to SAR segmentation and classification.
tral images. Proceedings of IGARSS, Vol. 7, Seoul, Korea, IEE Proc. Radar Sonar Navig., 147(3):134142.
pp.49464949. Strasters, K. and J. Gerbrands. 1991. Three-dimensional segmen-
Prim, R. 1957. Shortest connection networks and some general- tation using a split, merge and group approach. Pattern
izations. Bell Syst. Tech. J., 36:13891401. Recognit. Lett., 12:307325.
Richards, J. A. and X. Jia. 1999. Remote Sensing Digital Image Sung, F. and J. He. 2009. The remote-sensing image segmenta-
Analysis: An Introduction. New York: Springer-Verlag. tion using textons in the Normalized Cuts framework.
Robinson, D. J., N. J. Redding, and D. J. Crisp. 2002. Mechatronics and Automation: International Conference on
Implementation of a fast algorithm for segmenting SAR ICMA, Changchun, China, pp. 18771881.
imagery. In: Scientific and Technical Report. Defence Tarabalka, Y., J. A. Benediktsson, J. Chanussot, and J. C. Tilton. 2010.
Science and Technology Organization, Edinburgh, South Multiple spectral-spatial classification approach for hyperspec-
Australia, Australia. tral data. IEEE Trans. Geosci. Remote Sens., 48(11):41224132.
Rosin, P. L. and J. Hervas. 2005. Remote sensing image thresh- Tarabalka, Y., J. C. Tilton, J. A. Benediktsson, and J. Chanussot.
olding methods for determining landslide activity. Int. J. 2012. A marker-based approach for the automated selection
Remote Sens., 26(6):10751092. of a single segmentation from a hierarchical set of image
Sezgin, M. and B. Sankur. 2004. Survey over image threshold- segmentations. IEEE JSTARS, 5(1):262272.
ing techniques and quantitative performance evaluation. J. Tarabalka, Y., J. Chanussot, and J. A. Benediktsson. 2009.
Electron. Imaging, 13(1):146168. Classification of hyperspectral images using automatic
Shankar, B. U. 2007. Novel classification and segmentation tech- marker selection and minimum spanning forest. Proceedings
niques with application to remotely sensed images. In: V. of IEEE WHISPERS, Grenoble, France, pp. 14.
W. Marek, E. Orlowska, R. Slowinski, and W. Ziarko, eds. Tarabalka, Y., J. Chanussot, and J. A. Benediktsson. 2010a.
Transactions on Rough Sets VII. New York: Springer-Verlag, Segmentation and classification of hyperspectral images
pp. 295380. using minimum spanning forest grown from automatically
Shi, J. and J. Malik. 2000. Normalized cuts and image segmenta- selected markers. IEEE Trans. Syst. Man Cybern. B: Cybern.,
tion. IEEE Trans. Pattern Anal. Mach. Intell., 22(8):888905. 40(5):12671279.
Sirmacek, B. and C. Unsalan. 2010. Urban area detection using Tarabalka, Y., M. Fauvel, J. Chanussot, and J. A. Benediktsson.
local feature points and spatial voting. IEEE Geosci. Remote 2010b. SVM- and MRF-based method for accurate classi-
Sens. Lett., 7(1):146150. fication of hyperspectral images. IEEE Geosci. Remote Sens.
Skurikhin, A. N. 2010. Patch-based image segmentation of sat- Lett., 7(4):736740.
ellite imagery using minimum spanning tree construction. Tarabalka, Y. and A. Rana. 2014. Graph-cut-based model for
Proceedings of GEOBIA, Ghent, Belgium. spectral-spatial classification of hyperspectral images.
Soille, P. 2003. Morphological Image Analysis. New York: Proceedings of IGARSS, Quebec City, Quebec, Canada.
Springer-Verlag. Tilton, J. C., Y. Tarabalka, P. Montesano, and E. Gofman. 2012.
Soille, P. 2008. Constrained connectivity for hierarchical image Best merge region-growing segmentation with integrated
partitioning and simplification. IEEE Trans. Pattern Anal. nonadjacent region-object aggregation. IEEE Trans. Geosci.
Mach. Intell., 30(7):11321145. Remote Sens., 50(11):44544467.
Soille, P. and L. Najman. 2010. On morphological hierarchical Tilton, J. C. 1998. Image segmentation by region growing and
representations for image processing and spatial data clus- spectral clustering with a natural convergence criterion.
tering. International Workshop on Applications of Discrete Proceedings of IGARSS, Seattle, WA.
Geometry and Mathematical Morphology, Istanbul, Turkey, Tilton, J. C. 2003. Analysis of hierarchically related image seg-
pp. 4367. mentations. Proceedings of IEEE Workshop on Advances in
Soille, P. and M. Pesaresi. 2002. Advances in mathematical mor- Techniques for Analysis of Remotely Sensed Data, Greenbelt,
phology applied to geoscience and remote sensing. IEEE MD, pp. 6069.
Trans. Geosci. Remote Sens., 40(9):20422055. Tilton, J. C. 2007. Parallel implementation of the recursive approxima-
Solberg, A. H. S., T. Taxt, and A. K. Jain. 1996. A Markov random tion of an unsupervised hierarchical segmentation algorithm.
field model for classification of multisource satellite imag- In: A. J. Plaza and C. Chang, eds. High Performance Computing
ery. IEEE Trans. Geosci. Remote Sens., 34(1):100113. in Remote Sensing. New York: Chapman & Hall, pp. 97107.
Sousa, Jr., M. A. 2014. SegSAR Image Segmentation. https:// Tilton, J. C. 2013. RHSeg users manual: Including HSWO, HSeg,
sites.google.com/site/segsar2/home, last accessed March HSegExtract, HSegReader, HSegViewer and HSegLearn,
31, 2015. Version 1.59. Available via email request to James.C.Tilton@
Sramek, M. and T. Wrbka. 1997. Watershed based image segmen- nasa.gov
tationAn effective tool for detecting landscape structure. Trimble Geospatial Imaging. 2014. eCognition Developer.
Digital Image Processing and Computer Graphics (DIP97), http://www.ecognition.com/suite/ecognition-developer,
Proceedings of SPIE, Vol. 3346, Vienna, Austria, pp. 227235. last accessed March 31, 2015.
2016 by Taylor & Francis Group, LLC
342 Remotely Sensed Data Characterization, Classification, and Accuracies
Unsalan, C. and K. L. Boyer. 2005. A system to detect houses and Yang, L., F. Albregsten, T. Lonnestad, and P. Grottum. 1995. A
residential street networks in multispectral satellite images. supervised approach to the evaluation of image segmentation
Comput. Vis. Image Underst., 98(3):423461. methods. Proceedings of CAIP, Lecture Notes on Computer
Vincent, L. and P. Soille. 1991. Watersheds in digital spaces: An Science, Vol. 970, Prague, Czech Republic, pp.759765.
efficient algorithm based on immersion simulations. IEEE Zahn, C. T. 1971. Graph-theoretic methods detecting and describ-
Trans. Pattern Anal. Mach. Intell., 13(6):583598. ing gestalt clusters. IEEE Trans. Comput., C-20(1):6886.
Wang, C., N. Komodakis, and N. Paragios. 2013. Markov Random Zamalieva, D., S. Aksoy, and J. C. Tilton. 2009. Finding com-
field modeling, inference and learning in computer vision pound structures in images using image segmentation and
and image understanding: A survey. Comput. Vis. Image graph-based knowledge discovery. Proceedings of IGARSS,
Understand., 117:16101627. Cape Town, South Africa.
Wang, Z. and R. Boesch. 2007. Color- and texture-based image Zhang, Y. J. and J. J. Gerbrands. 1994. Objective and quantitative seg-
segmentation for improved forest delineation. IEEE Trans. mentation evaluation and comparison. Signal Proc., 39:354.
Geosci. Remote Sens., 45(10):30553062. Zucker, S. W. 1976. Region growing: Childhood and adolescence.
Williams, J. W. 1964. Heapsort. Commun. ACM, 7(12):347348. Comput. Graph. Image Proc., 5:382399.
2016 by Taylor & Francis Group, LLC
17
LiDAR Data Processing
and Applications
Shih-Hong Chio
National Chengchi University
Tzu-Yi Chuang
National Taiwan University
Pai-Hui Hsu
National Taiwan University
Jen-Jer Jaw
National Taiwan University
Shih-Yuan Lin
National Chengchi University
Acronyms and Definitions ................................................................................................................ 343
Yu-Ching Lin 17.1 Introduction ............................................................................................................................ 344
National Defense University
17.2 LiDAR Data Quality Assessment and Control .................................................................. 347
System Components LiDAR Error Budget Quality Assessment Quality Control
Tee-Ann Teo
National Chiao Tung University 17.3 LiDAR Data Management ..................................................................................................... 350
Storage of LiDAR Point Cloud Data Organization and Generalization of LiDAR Point
Fuan Tsai Cloud LiDAR Database
National Central University 17.4 LiDAR Point Cloud Feature Extraction ...............................................................................352
Spatial Features in LiDAR Data Methods for the Extraction of Spatial Features
Yi-Hsing Tseng 17.5 Three-Dimensional City Modeling from LiDAR Data ..................................................... 356
National Cheng Kung University
Properties of LiDAR Data in a City Area Object Reconstruction Strategies Building
Cheng-Kai Wang Extraction Road Extraction
National Cheng Kung University 17.6 Full-Waveform Airborne LiDAR ......................................................................................... 360
Introduction Waveform Data Analysis Applications
Chi-Kuei Wang 17.7 DEM and DSM Generation from Airborne LiDAR Data ................................................ 365
National Cheng Kung University Introduction Data Processing for DEM and DSM Generation DEM Quality Assessment
and Control
Miao Wang 17.8 Terrestrial/Vehicle-Borne LiDAR Data Processing .......................................................... 367
National Cheng Kung University
Terrestrial LiDAR Data Processing Object Extraction from Vehicle-Borne LiDAR Data
Ming-Der Yang 17.9 Conclusions.............................................................................................................................. 369
National Chung Hsing University References............................................................................................................................................ 369
Acronyms and Definitions DBMS Database management system
DEM Digital elevation model
3D Three dimensional DG Direct georeferencing
ALS Airborne laser scanning DSM Digital surface model
ASPRS American Society of Photogrammetry and Remote DTM Digital terrain model
Sensing FDNs Fixed distance neighbors
CBH Crown base height GCPs Ground control points
CF Constant fraction GIS Geographical information system
CG Center of gravity GLAS Geoscience Laser Altimeter System
DBH Diameter at breast height GNSS Global Navigation Satellite System
343
2016 by Taylor & Francis Group, LLC
344 Remotely Sensed Data Characterization, Classification, and Accuracies
GPS Global positioning system Rotating
mirror Lens
GRCS Ground reference coordinate system
Laser
ICESat Ice Cloud and Land Elevation Satellite rangefinder
ICP Iterative Closest Point
IMU Inertial measurement unit Vertical
rotation
INS Inertial navigation system
ITS Intelligent transportation system Range
Object
k-NNs k-nearest neighbors
LiDAR Light Detection and Ranging
LS3D Least squares 3D surface matching
MA Maximum
OGC Open Geospatial Consortium
PCA Principal component analysis Horizontal
POS Positioning and orientation systems (azimuth)
QA/QC Quality assessment and quality control rotation
RANSAC Random sample consensus
RMSD Root mean square difference FIg u r e 17.1 The measurement mechanism of a terrestrial laser
RMSE Root mean square error scanner.
SDMBS Spatial DBMS
TH Threshold The second category includes airborne LiDAR systems
TIN Triangulated irregular networks (also called airborne laser scanning [ALS]), mobile terrestrial
TLS Terrestrial laser scanning LiDAR systems, or other LiDAR systems boarded on mov-
VLR Variable length record ing platforms (Baltsavias, 1999b; Petrie and Toth, 2008a,b).
WB Wavelet based These systems offer the capability of acquiring 3D informa-
VMNS Voxel-based marked neighborhood searching tion about the scanned objects at the mapping coordinate sys-
Voxel Volume element tem by means of direct georeferencing (DG), which is enabled
ZC Zero crossing of the first deviation by the combination of a global positioning system (GPS) unit
and an inertial measurement unit (IMU) unit (together, these
are also known as positioning and orientation systems [POS])
17.1 introduction (Baltsavias, 1999a; Krabill etal., 2000), as shown in Figure17.2.
The spaceborne LiDAR systems can be regarded in the second
Light detection and ranging (LiDAR), also called laser scan- category. However, data processing consideration for such
ning, is an effective remote sensing system for acquiring three- systems is somewhat different. One of the reasons is that the
dimensional (3D) information about scanned objects, which footprint size and the sampling interval of the spaceborne sys-
has been widely applied in a broad range of disciplines since it tems are significantly larger, for which they have been called
was first developed less than two decades ago (Baltsavias, 1999a; large footprint systems. For example, the Geoscience Laser
Krabill etal., 2000). A LiDAR system integrates several accurate Altimeter System aboard the Ice Cloud and Land Elevation
optical, electronic, mechanic, timing, and georeferencing units Satellite has the footprint size of 60 m and sampling interval
that make it enable to acquire 3D point measurements rapidly. of 172 m (Zwally etal., 2002). The geometry details obtained
LiDAR systems are mainly classified into two types based on from spaceborne LiDAR systems have smoother appearance,
the motions that occur when they collect data. The first cate- which prevents many of the processing logic designed for the
gory includes static terrestrial (or ground-based) LiDAR sys- small-footprint systems to be directly applied. For brevity and
tems (usually mounted at a tripod) with a variety of scanning conciseness, these large footprint systems are not included in
ranges, which are also called 3D laser scanners. The primary this chapter.
component of a terrestrial LiDAR system is a laser ranging LiDAR measures the surfaces of objects by accurately record-
unit to obtain the distance to the target. Integrating with one ing the time of flight of the laser pulse, which is transmitted from
or two rotating mirror or prism that changes the direction of the laser ranging system and then returned back as it reflects
emitting laser pulses, the LiDAR system can scan and measure off the scanned surface, and the scan angle of the laser pulse,
the distances to the surrounding objects. By calculating the which the laser is directed to, with respect to the laser ranging
obtained range and the scanning angle, the 3D coordinates of system. The distance between the laser ranging system and the
the scanned target are obtained. Because the mirror rotates in scanned object can be obtained by halving the multiplication of
rapid speed with small-angle variation when the LiDAR oper- the speed of light and the round trip time of the laser pulse, and
ates, a huge number of dense and accurate point measurements, when this is combined with the scan angle of the laser pulse,
often called point cloud, of the surfaces of scanned objects are the 3D coordinates of the point measurement at a local coordi-
obtained (Figure 17.1). nate system can be obtained. For a LiDAR system installed on a
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 345
INS
Z Laser
Y 360
GPS Z Y scanner
Z X
Y
X
GPS X
X
INS
Digital Y
camera Laser Z
scanner
FOV
FOV
Z
Y
GPS X
(a) (b)
FIg u r e 1 7.2 (a) Airborne and (b) vehicle-borne LiDAR systems.
moving platform, the POS provides highly accurate data on the In Section 17.2, the concept of error budget is introduced
position and attitude of the platform for DG. The resulting 3D and implemented as the main tool for the QA/QC of the LiDAR
information about the scanned objects is then registered in the point measurements. Internal and external controls are then
same mapping coordinate system as the POS. The choice of the suggested to ensure that the data derived from the LiDAR sys-
mapping coordinate system depends on the scope of utility for tem meet the requirement of a specific application.
the LiDAR data (Fornaciai etal., 2010). Figure 17.3a and b shows The number of point measurements resulting from a normal
the point clouds of a city area and a forest area, and Table17.1 LiDAR scan project, either airborne or terrestrial, can easily
lists the characteristics of ALS, terrestrial laser scanning (TLS), exceed several tens of millions. Due to the nature of these 3D
and mobile TLS. Table17.2 lists the main applications, strengths, point measurements, which lack an efficient spatial index to eas-
and limitations of ALS, TLS, and mobile TLS. ily identify neighboring points and quickly access any points of
Airborne LiDAR systems have been widely employed for interest, the LiDAR data are difficult to visualize, edit, and pro-
large area or nationwide applications, in many cases replacing cess. Section 17.3 thus presents some concepts that can be used
photogrammetric approaches, as the process flow of LiDAR to establish a spatial index for point clouds.
data is relatively straightforward and less time consuming. Assigning meaningful attribute to the point clouds can sig-
There are two major technological benefits of airborne LiDAR nificantly increase the usage of the LiDAR data. In an urban
that have contributed to its popularity. The first is that LiDAR setting, extracting spatial features from the point clouds, such
can acquire 3D information of homogeneous surfaces, which is as points, lines, and surfaces, can be quite useful as the results
a difficult task for photogrammetry due to the lack of textures can be readily adopted by 3D city modeling. Section 17.4 gives
for successful stereo matching. The other is that it is capable detailed information regarding this kind of feature extraction
of acquiring multiple measurements at different distances with from LiDAR data. Furthermore, several examples of using
only one pulse. For forested areas, it is very useful to be able LiDAR point clouds for 3D city modeling are presented in
to acquire 3D information about not only the top of the can- Section 17.5.
opy but also the ground surfaces beneath it and the structures The complete temporal history of the laser return sig-
within it. As a comparison, photogrammetry tends to be used nal is called the full waveform. In practice, the waveform is
only to measure the top canopy, because the ground surfaces processed on the fly, and only a few meaningful signals (i.e.,
are only partially visible or completely invisible from aerial returns) are extracted while the waveform data are then dis-
photographs. carded. Some of the early LiDAR systems were only able to
Careful data processing must be carried out in order to fully extract two returns for each waveform. More r systems nor-
exploit the point measurements obtained from the scanned mally provide four returns, that is, the first, second, third, and
objects. This chapter covers the issues of data quality assessment last, for each laser pulse. An unlimited number of returns for
and quality control (QA/QC), data management, point cloud each laser pulse is possible, provided the full-waveform data
feature extraction, and full-waveform data processing of the and a decent postprocessing algorithm are available. However,
LiDAR data. In addition, applications for digital elevation model handling such a massive amount of data is rather difficult, and
(DEM) and digital surface model (DSM) generation and 3D city thus this approach has not been widely embraced by the com-
modeling are also described. munity (e.g., manufacturers, service providers, and end users).
2016 by Taylor & Francis Group, LLC
346 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) (b)
(c) (d)
FIg u r e 1 7.3 (Up) Point clouds and (down) profile of (a) a city area and (b) a forest areas. Color represents the height of the points.
TABLe 1 7.1 Characteristics and Applications of ALS, TLS, and Mobile TLS
System ALS TLS Mobile TLS
Footprint size Small (~2040cm in diameter). Extreme small (~2cm in diameter). Extreme small (~2cm in diameter).
Waveform data Available from more recent systems. Available from limited systems. Available from limited systems.
Georeferencing Direct. Indirect. Direct.
Measuring distance 2003000 m. <1km. <1km.
Data characteristic The sunlit surfaces are scanned, Point measurements are distributed on the side of Point measurements are distributed on
where most of the point the scanned objects. Usually extremely high detail the side of the scanned objects, which
measurements are distributed on of the surface can be obtained within the scan are close to the path of the moving
the top surfaces of scanned objects. range of the static station. platform.
Applications DEM, DSM, DBM, CHM, DBH, building modeling. DBH, city modeling, road network.
land-cover classification, city
modeling, road network, biomass.
Survey area Nationwide. Single or multiple objects with small vicinity. City-wide.
Nevertheless, processing the full-waveform data is suggested been designed to achieve this. The methodology used can be as
for areas where the penetration of LiDAR is unsatisfactory, simple as treating all of the last returns as ground points. While
and Section 17.6 explains the benefits provided by the full- this simple solution may be sufficient for a bare surface or in a
waveform data. city setting, it is likely to fail in forested areas, since many of the
Generating DEM using LiDAR point clouds is the main last returns may not represent the ground surface (being merely
purpose of obtaining airborne LiDAR data. To ensure a good- close to it). To obtain good-quality DEM in an efficient manner,
quality DEM, the ground points must be identified from the the LiDAR points are filtered by a morphology-based algorithm,
whole LiDAR point clouds and several algorithms that have which considers the connectivity of the nearby points, and this
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 347
TABLe 1 7.2 Recording Fields for Point Format 10 of LAS 1.4
Item Format Size
X Long 4bytes
Y Long 4bytes
Z Long 4bytes
Intensity Unsigned short 2bytes
Return number 4bits (bits 03) 4bits
Number of returns (given pulse) 4bits (bits 47) 4bits
Classification flags 4bits (bits 03) 4bits
Scanner channel 2bits (bits 45) 2bits
Scan direction flag 1 bit (bit 6) 1 bit
Edge of flight line 1 bit (bit 7) 1 bit
Classification Unsigned char 1byte
User data Unsigned char 1byte
Scan angle Short 2bytes
Point source ID Unsigned short 2bytes
GPS time Double 8bytes
Red Unsigned short 2bytes
Green Unsigned short 2bytes
Blue Unsigned short 2bytes
NIR Unsigned short 2bytes
Wave packet descriptor index Unsigned char 1byte
Byte offset to waveform data Unsigned long 8bytes
Waveform packet size in bytes Unsigned long 4bytes
Return point waveform location Float 4bytes
X(t) Float 4bytes
Y(t) Float 4bytes
Z(t) Float 4bytes
is then followed by manual inspection/editing to ensure all of LiDAR systems and relevant error budgets, and this is fol-
ground points are reasonably identified. The production of DEM lowed by a discussion of common approaches to the quality
is finally realized by interpolating the identified ground points. assessment and control of LiDAR data.
Section 17.7 provides a detailed explanation of DEM generation
from LiDAR data.
The characteristics of the point clouds obtained from the 17.2.1 System components
ground-based and mobile LiDAR systems, for example, the pat- The components of LiDAR systems differ with regard to differ-
tern of point distribution, the variation of point density within ent LiDAR platforms, which are classified into two categories,
the LiDAR data, and the lack of echo information, are different namely, mobile platforms, for example, airborne or vehicle-
from those from airborne LiDAR systems. The processing strat- based LiDAR systems, and static platforms, for example, ter-
egy used for such data should take into account these differences restrial LiDAR mounted on a tripod. The system structure of
and make any necessary adjustments. The processing of the static LiDAR platforms, which integrate state-of-the-art laser
ground-based and mobile LiDAR point clouds is thus described ranging and scanning for the rapid, highly dense, and pre-
in Section 17.8. cise acquisition of 3D point clouds (Petrie and Toth, 2008a), is
relatively simple as compared to mobile ones. Mobile LiDAR
17.2 LiDAR Data Quality systems are based on the combination of a laser scanning
Assessment and control system and a DG system. The laser scanning system can be
subdivided into three key units: the optomechanical scanner,
Due to the nature of LiDAR systems, every single point that is the ranging unit, and the control processing unit; the DG sys-
generated usually contains no redundant measurements, and tem composed of a GPS/Global Navigation Satellite System
there is no associated measure that can be used to evaluate the (GNSS) and an inertial navigation system (INS) measures the
quality of point clouds, except for the nominal precision that sensors position and orientation directly with respect to a
manufacturers claim. Therefore, a well-defined set of QA/QC referenced coordinate system. The combination of GPS/INS
procedures is needed before embarking on any aspect of LiDAR technologies may also be referred to as a GPS-aided INS or
data processing. This chapter thus introduces basic components a POS. Each of these technologies alone has limitations, but
2016 by Taylor & Francis Group, LLC
348 Remotely Sensed Data Characterization, Classification, and Accuracies
the integration of GPS and INS is a powerful solution for DG. propagation in the atmosphere, causing both distortion and
The DG system and scanning sensor must be in a rigidly fixed attenuation of the returned signal. The degree of attenuation
position with respect to each other and calibrated within depends on the wavelength, temperature, pressure, micro-
the reference frame of the platform for meaningful results. scopic particles in the air, and weather conditions. Other fac-
During a scanning task, the DG system records position and tors influencing laser beam propagation are reflection and
orientation data and also records a corresponding time tag atmospheric turbulence, caused by the beam wandering from
for each laser scan. The DG postprocessing software interpo- its initial direction and Gaussian wave-front distortion, called
lates the position and orientation of the laser reference point beam intensity fluctuation. Based on these characteristics, it
at each time tag. The 3D ground coordinates of every laser can be understood that the longer the range, the greater the
return can then be computed using these data and the range expected errors.
measured by the laser.
Direct georeferencing errors: With regard to positioning per-
formance, most errors in a GPS/GNSS system are dependent
17.2.2 LiDAR error Budget
on the operating conditions and setup (Morin, 2002). These
The LiDAR systems introduced earlier face problems with errors, such as atmospheric errors, multipath effects, poor sat-
regard to the random and systematic errors that may occur ellite geometry, baseline length, and loss of lock, have a direct
in each component, as well as due to their integration (Filin, impact on the resulting positioning accuracy of a GPS/GNSS
2001). Random errors include those related to position and ori- system, and most of the related factors are difficult to predict
entation measurements from the DG system, mirror angles, and and describe via a mathematic model. Positioning noise leads
ranges, and these are based on the precision of the instrumen- to similar amounts of noise in the derived point cloud. As for
tal measurements within a mobile system. On the other hand, orientation errors, the overall accuracy of navigation attitude
systematic errors are mainly caused by biases in the mounting depends on the quality of IMU. Errors in the LiDAR return
parameters related to the system components, as well as those position due to attitude errors are directly proportional to the
in the system measurements resulting from biases related to the range from the scanner to target. As a result, an IMU with
range and mirror angle in a laser scanner system. Moreover, sys- higher accuracy is normally required for fixed wing operations
tematic errors are also caused by differences in the troposphere compared to helicopter or ground-based data collection, due
and ionosphere, as well as multipath, INS initialization and to the increased target range (Glennie, 2007). When taking the
misalignment errors, and gyro drifts in a DG system. Lichti and integration of the various sensor components into consider-
Licht (2006) presented a systematic error model that consisted ation, the boresighting angles and the physical mounting angles
of 19 coefficients grouped into two categories: physical and between an IMU and a laser scanning system are usually the
empirical. The physical group comprises known error sources main sources of systematic errors, and thus careful calibration
such as rangefinder offset, cyclic errors, collimation axis error, of these is required. Moreover, due to the fact that the center of
trunnion axis error, and vertical circle index error. The empiri- observations from the laser scanner and the origin of the navi-
cal terms lack ready physical explanation but nonetheless model gation system cannot be colocated, the precise offset or lever
significant errors. Typically, errors of a static LiDAR platform arm between the two centers must be known in order to accu-
come principally from the laser scanning system, while those rately determine the laser scanner measurements. Since the
of a mobile LiDAR platform are the combined effects of the physical measurement origin of the navigation system or laser
laser scanning system and DG system. A detailed description scanner assembly cannot be directly observed, the lever-arm
of random and systematic errors of LiDAR systems is given in offset must be obtained indirectly by a calibration procedure.
Huising and Pereira (1998), Baltsavias (1999b), Schenk (2001), Biases in the lever-arm offsets can then lead to constant shifts
Latypov and Zosse (2002), and Habib etal. (2008, 2009). A sum- in the derived point cloud.
mary of laser scanning and DG errors is given in the following
paragraphs. 17.2.3 Quality Assessment
Laser scanner errors: There are a number of factors affecting Quality assessment encompasses management activities that
the accuracy of a laser scanner. Reshetyuk (2006) classified are carried out prior to data collection to ensure that the
scanning errors into three major categories, which highlight derived data are of the quality demanded by the user. These
the influence of the scanned object on the accuracy of the management controls cover the calibration, planning, imple-
related point clouds. The first category is instrumental system mentation, and review of data collection activities. In consid-
errors, based on the range and angular measurements, and eration of the potential errors mentioned earlier, calibration
these vary with different scanning devices. The second cate- of LiDAR systems is prerequisite for reducing the effects of
gory is object-related errors, and these are related to the reflec- these on the accuracy of acquired point clouds. Ground-based
tance properties of the objects surface, due to several factors LiDAR systems are calibrated by scanning a field of distrib-
such as material properties, laser wavelength, polarization, uted control targets, while airborne LiDAR systems are cali-
surface color, moisture, roughness, and temperature. The last brated by flying over a set of ground calibration targets that
category is environmental errors, which affect laser beam have been precisely surveyed and well mapped. However, there
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 349
is currently no standard procedure for a calibration task, and
thus each LiDAR manufacturer may have its own calibration DLi
scheme for its own products, based on specific parameters.
Further discussions on LiDAR system calibration can be found DPLi
in Schenk (2001), Burman (2002), Filin (2003), Kager (2004), DLi1
and Skaloud and Lichti (2006). One problem is that careful cal-
ibrations require some raw measurements, such as navigation
data, mirror angles, and ranges, and in general, LiDAR systems
do not supply such raw measurements, and thus quality control
procedures are needed to assess system performance after the ALi
data have been collected.
ALi1 APli
17.2.4 Quality control
Quality control procedures can be realized by internal and
external checks. The internal measures are used to check the
Distance of differences
relative consistency of the LiDAR data, while the external mea-
Angle of differences
sures verify the absolute quality of the LiDAR data by check-
ing its compatibility with an independently collected and more
accurate data set. FIg u r e 17.4 Illustration of distance and angle differences when
using line and plane features.
17.2.4.1 internal Quality control As shown in Figure 17.4, in addition to the conventional way
LiDAR data are usually acquired from different scanning view- of assessing positional discrepancies for point features, angle
points or different strips; so that the relative consistency of point and distance measures can also be utilized to assess the qual-
clouds within the overlapping areas can be assessed through the ity, including both internal and external indicators for line and
correspondence of conjugate features. This check is commonly plane features, if applicable (Chuang, 2012).
conducted by comparing interpolated range or intensity images Registration of terrestrial LiDAR data sets and strip adjust-
derived from the overlapping areas or by comparing the conju- ment of airborne LiDAR point clouds are two major ways to
gate features extracted from corresponding data sets. The degree provide QA/QC for final geospatial products. In the case of
of coincidence of the extracted features can be used as a mea- TLS, it is normal to acquire multiple-point clouds from differ-
sure of the data quality and to detect the presence of systematic ent standpoints for a complete scene of objects. The manipu-
biases. A well-presented demonstration in internal quality con- lation of these data should thus be preceded by having them
trol can be found in Habib etal. (2010). registered relative to the same reference frame. On the other
hand, strip adjustment of airborne LiDAR data serves as a
17.2.4.2 external Quality control means of producing a best-fit surface through an adjustment
process that compensates for small misalignments between
A common approach to external checks involves checkpoint
adjacent data sets (Filin and Vosselman, 2004; Tao and Li,
analysis using specially designed LiDAR targets. The targets
2007). The discrepancies among strips are typically caused by
are then extracted from the range and intensity LiDAR imag-
the varied performance of the georeferencing components and
ery using a segmentation procedure. The coordinates of the
thus show systematic patterns that are more visible in overlap
extracted targets are then compared with the surveyed coor-
areas rich in objects of simpler geometric shapes, as illustrated
dinates using a root mean square error (RMSE) or root mean
in Figure 17.5.
square difference (RMSD) indicator. The former usually refers
In summary, the quality assessment of LiDAR systems is
to comparing the estimated results with error-free data, while
restricted to the availability of the raw measurements, which
the latter compares the estimated results with erroneous refer-
ence data, and both share the same expression. For more details
regarding this approach, one can refer to Csanyi and Toth
(2007). On the other hand, in addition to employing special con-
trol targets, features within point clouds can also be used as a
measure of external quality control. In such case, the derived
features can be compared with independently collected control
entities, such as point, line, and planes over the same area. To
this end, checkpoints situated on smoothing surfaces are usually
adopted, and the quality of the vertical components is the main FIg ur e 17.5 Illustration of strip discrepancies of airborne LiDAR
factor to be evaluated. point clouds.
2016 by Taylor & Francis Group, LLC
350 Remotely Sensed Data Characterization, Classification, and Accuracies
are not usually revealed to the end user. Consequently, quality export the required attributes of the point data into a text file in
control is an essential procedure to ensure that the data derived order to carry out data processing using custom-developed soft-
from a LiDAR system is able to meet the requirements of a spe- ware. In practice, ground-based LiDAR data may also be stored
cific application. in LAS format, and the instrument-related data may be stored
as VLRs in the related data file. However, users must know the
format of the special VLRs in advance to correctly read and use
17.3 LiDAR Data Management the recorded data.
For the convenience of data management, airborne LiDAR
17.3.1 Storage of LiDAR Point cloud Data
data are usually stored in one file per strip, and ground-based
A LiDAR data set mainly contains the 3D coordinates of observed LiDAR data are stored in one file per scan. However, the way
points, which are commonly called point cloud data represent- in which mobile LiDAR data is stored is usually decided by
ing the spatial properties of scanned objects. The reflected laser the user.
intensity of each point is also recorded as the radiometric prop-
erty. Many LiDAR systems, particularly for airborne LiDAR sys-
17.3.2 organization and Generalization
tems, may record additional information for each point, such as
of LiDAR Point cloud
return number, number of returns, scan direction, scan angle,
and GPS time tag. The data of scanned points in a LiDAR data file are originally
Each manufacturer has developed a particular LiDAR saved in a sequence based on the scanning time. Retrieving a
data format for storage, and a standard format is thus needed point cloud of a local area or a cluster of neighboring points can
for data exchange. The simplest way to exchange data is to thus be very inefficient, especially when there is a large amount
rewrite it in a text (ASCII) file, although this requires much of data. Organization and generalization of point clouds are thus
more storage space than a binary file. In 2003, the American important practical issues for LiDAR data processing.
Society of Photogrammetry and Remote Sensing (ASPRS) A scheme of spatial indexing is usually proposed for the orga-
published a format standard, known LAS 1.0, for LiDAR data nization of LiDAR data. Point clouds in a specified region of
exchange (ASPRS, 2013). The LAS format is now used as the interest (ROI) or a cluster of neighboring points at a location can
standard format of LiDAR data for both users and hardware be retrieved efficiently through the use of spatial indices. In order
and software manufacturers. Waveform data can also be to visualize a huge LiDAR data set efficiently, the spatial index-
included in the two latest versions of this format, announced ing approach should be extended to allow a data set to be orga-
in 2010 and 2011. nized in a hierarchical, multiresolution fashion. Furthermore, a
The LAS file format provides several optional ways of record- database management system (DBMS) may be needed to handle
ing different point data contents. The optional fields include the long-term collection of LiDAR data.
those for the data related to GPS time tag, color (red, green, blue),
waveform package, and near infrared. Users can choose appro- 17.3.2.1 Spatial indexing
priate formats depends on their specific aims. The LAS 1.4 speci- The purpose of using spatial index of LiDAR point clouds is to
fication provides up to 11 point formats (ID: 010). Table 17.2 establish the relationships among neighboring points, so that it
shows the recorded fields of point format 10, which contains all is easier to search for and retrieve of points of interest in a large
optional fields. data set. LiDAR data processing, such as feature extraction and
In addition to point data, the LAS format also contains differ- DEM generation, often requires the derivation of meaningful
ent numbers of variable length records (VLRs) for storing addi- information from the relationships among neighboring points
tional information about the point cloud. VLRs contain various or closely distributed points in a 3D space. However, searching
types of data, including projection information, metadata, wave- for specific points and their neighbors in a sequentially stored
form packet information, and user application data. point data set is a time-consuming and inefficient task. Some
Although LAS files store data in binary format and consume ground-based LiDAR systems store points in 2D array fash-
less storage space than text files, they are not compact. Isenburg ion. While this means that points obtained from adjacent laser
(2013) thus developed a lossless compression scheme for LAS pulses can be easily accessed from the file, these points may not
files to reduce the storage space down to less than 25% of the be located near to each other in 3D space. Beside time concern,
original. memory space of computer is another important issue. In addi-
Although the LAS format was specially designed for recording tion, the computer being used to process these data may not have
airborne LiDAR data, it is also perfectly good for recording the enough memory to load all the point data in a file to search and
data produced by more recent, mobile LiDAR systems. However, process it. However, these problems may be overcome if a spatial
ground-based LiDAR data are often stored in specially designed index for points is used.
proprietary file formats, in order to take advantage of the par- Building a spatial index for a point cloud requires the assis-
ticular features of the instruments and to aid in the subsequent tance of spatial data structures, such as triangulated irregu-
data processing that occurs using on the software that came with lar networks (TINs), trees, and grids. The distribution of
the hardware. Such software usually allows advanced users to airborne LiDAR point clouds is similar to a 2.5D data set,so
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 351
TIN (Chenetal., 2006; Pu and Vosselman, 2006) and 2D grids multiresolution subimages of the original image, a hierarchi-
(Chen etal., 2007) are often used to handle the neighborhood cal representation of LiDAR point clouds can be achieved using
of points. While ground-based and mobile LiDAR point clouds a similar approach. At a grid-organized point cloud, a repre-
have a 3D distribution, they are usually handled using 3D sentative point, whose coordinates are the average of the points
grids (or volume elements, known as voxels) (Bucksch et al., inside the grid, can be obtained for each grid. These representa-
2009; Gorte and Pfeifer, 2004; Wang and Tseng, 2011). Region tive points form the first level of generalization of the original
quadtrees and octrees, which decompose the space of point point cloud, and the complete generalization hierarchy is then
distribution into regular subspaces, are commonly used to pro- established level by level based on this. Figure 17.6 shows the
vide regular spatial indexing for data access and to determine example of hierarchical representation of a point cloud at dif-
the fixed distance neighbors of points. On the other hand, a ferent levels.
point k-d tree is often used to determine the k-nearest neigh- An additional index file is required to store the resulting gen-
bors of points (Rabbani, 2006). In practice, because most grids eralized hierarchy and spatial index. Based on the viewing scale,
are empty in a 3D grid for a ground-based or mobile LiDAR representative points of a certain interest level can be loaded
point cloud, a region octree structure may be used to record the from the index file and displayed individually. To view the origi-
3D grids to reduce the storage space (Wang and Tseng, 2011). nal point cloud at a large scale, the visible points inside the view-
ing window can be loaded efficiently through the spatial index
17.3.2.2 Hierarchical Representation of LiDAR Data stored in the index file. In practice, a 2D hierarchy can be simply
Before starting the data processing task, it is always helpful for saved like an image pyramid. In contrast, the 3D hierarchy orga-
users to realize the contents of the LiDAR point clouds by visu- nized with an octree structure can be saved as a file in the form
ally inspecting both the outlines and details of the data. Software of a linear tree (Samet, 1990).
for viewing LiDAR data thus need the ability to efficiently dis- Organizing and generalizing a LiDAR point cloud requires
play a huge amount points on different scales, with a small scale extra time, and the results require additional storage space.
used for the outlines of the point cloud, and a large one for the However, this work only needs to be done one time, and it can
details. As with an image pyramid that provides hierarchically then benefit all subsequent data processing and viewing.
(a) (b)
(c) (d)
FIg u r e 1 7.6 Results of point cloud generalization at different resolutions. (a) Original point cloud and (b) 4 m, (c) 8 m, and (d) 16 m.
2016 by Taylor & Francis Group, LLC
352 Remotely Sensed Data Characterization, Classification, and Accuracies
17.3.3 LiDAR Database On the other hand, the features may also directly represent
objects of interest (such as a tree canopy, building model, and
As the use of LiDAR systems for collecting spatial data getting traffic sign) and physical properties of the objects.
easier and popular, more and more LiDAR data are collected In urban areas, ALS and TLS are effective data sources for city
and accumulated. Thus, the management of the huge amount of modeling, especially for the reconstruction of 3D building mod-
LiDAR point clouds becomes an important and inevitable issue. els. An important issue in building reconstruction from LiDAR
The most intuitive way to manage such huge data is using the well- data is to filter out ground and occlusion (such as trees) points.
developed DBMS, which is equipped with many handy functions Several algorithms have been developed specifically for ground
such as query, backup, and sorting for manipulating data. filtering of LiDAR data, and there are thoroughly reviewed in
There are two major different ways in using DBMS for point Meng etal. (2010). This filtering is often achieved by employing
cloud management. The first method partitions point clouds partitioning algorithms to segment the point clouds and then
into unified grids and stores the points in each grid separately. collecting the segmentation attributes to classify the segments
The metadata of each grid is also recorded as intermediate for (Chen etal., 2008; Rabbani etal., 2006; Sampath and Shan, 2007;
accessing point data. The second method employs the spatial Vosselman, 2009; Zhang and Whitman, 2005).
DBMS (SDMBS) to store point clouds. Each point is recorded In addition to 3D coordinates, LiDAR data also consist of
individually as a spatial element in the SDBMS that makes intensity information. Some applications may treat the intensity
allowance for optimization of LiDAR processing (Lewis et al., data as images (after rasterization and resampling) and employ
2012). The generalized hierarchy of point clouds that is used for spectral or spatial image analysis algorithms to extract features.
visual inspection purpose can also be stored at DBMS. For example, a spectral analysis of LiDAR points was used to
identify and map volcano lava flows (Mazzarini et al., 2007).
17.4 LiDAR Point cloud Spectral and spatial analyses of LiDAR intensity data were also
Feature extraction successfully applied to land-cover classification (Im etal., 2008),
extraction of building footprints (Chen et al., 2008; Zhao and
Being the points sampling from the object surfaces, LiDAR point Wang, 2014), and other applications.
clouds contain rich spatial information about the scanned targets.
Feature extraction of LiDAR point clouds is the recognition of 17.4.1 Spatial Features in LiDAR Data
specific geometric shapes or more general smooth surfaces in the
point data (Vosselman etal., 2004) and is the primary procedure to Because of the blind operating manner of LiDAR systems, the
the interpretation of the contents of the point clouds. The purpose spatial features of the scanned objects are implicitly contained
of LiDAR point cloud feature extraction is to improve the auto- in the point cloud. Before they can be used in GIS systems, for
mation of identifying objects of interest and their characteristics example, explicit and simple geometrical spatial elements like
from complicated LiDAR data sets. As LiDAR data are composed points, lines, and surfaces must be extracted from the point
of discrete point clouds, and there are no relationships among clouds for object reconstruction and modeling.
neighboring points, conventional spectral, spatial (texture-based), The shapes of 3D objects, especially man-made ones, are
and photogrammetric feature extraction methods designed for composed of simple geometric elements, like points, lines, and
images may not produce satisfactory results if directly applied to surfaces. Being blind remote sensing instruments, the point and
the processing and analysis of LiDAR point cloud data. Several line features of objects are difficult to directly and accurately
algorithms have thus been developed specifically for the feature measure by LiDAR systems. In contrast, points distributed on
extraction of LiDAR point clouds and full-waveform LiDAR the surfaces of such objects can aid in the extraction of surface
data. In addition, the appropriate method for extracting features features. Point and line features can then be obtained indirectly
from LiDAR data may vary significantly according to the targets from the intersection of neighboring surface features.
of interest and the characteristics of the data. For example, the Methods to extract different spatial features from LiDAR data
features required to reconstruct tree canopy models may be very should thus be designed according to the characteristics of the
different from those required for building model reconstruction. features of interest. Based on the extraction method used, the
Similarly, full-waveform LiDAR data consist of features that are spatial features in LiDAR point clouds are classified into three
not available in conventional LiDAR point clouds. categories, including fitting features, intersection features, and
The features embedded in LiDAR data may be derived from boundary features. Fitting features are obtained by fitting a
the following properties of LiDAR point clouds: point set to geometric elements, like surfaces, lines, and points.
Intersection features are obtained from the intersection of
Location (x, y, z) neighboring surfaces or line features. Boundary features appear
Intensity at the boundary of surface or line features. All three of these are
Echo (return) number described in more detail, in the following:
Waveform
Fitting features: Fitting features are formed if points evenly
The features that are derived or extracted may simply be points, distribute close to a specific geometric model, such as sur-
lines, plans, surfaces, shapes, or other geometric characteristics. faces or lines. The least squares estimation that minimizes
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 353
(a) (b)
(a) (b)
FIg u r e 17.9 Basic ideas of boundary features. (a) Boundary lines
and points of a fitting plane and (b) end points of a fitting line.
(c)
FIg u r e 17.7 Basic ideas of (a) fitting plane, (b) fitting line, and 17.4.2 Methods for the extraction
(c)gravity center of points. of Spatial Features
the squared sum of the normal distance from each point LiDAR point clouds represent the outside appearances of the
to the selected geometric model is a reasonable method to scanned objects. The shapes of some objects are simple, like
extract fitting features from a point cloud. Least squares pipelines in an industrial installation (Rabbani, 2006), and
estimation can be used to determine cosurface or colinear some are complex, like the terrain of the earth and buildings in
features. The centroid of a point set, which is the center posi- a city. If the shape of an object, like cylinder or sphere, can be
tion of the points, can also be categorized as a particular formulated with a simple equation, it can be extracted directly
fitting feature. However, this can be obtained directly by by determining its geometric parameters using the least squares
calculating the average coordinates of points, without the fitting algorithm (Shakarji, 1998). Complex shapes that cannot
need for least squares estimation. Figure 17.7 shows the basic be easily formulated are often disassembled into several plane
ideas of fitting surfaces, fitting lines, and gravity centers. features using segmentation algorithms. The objects are then
Intersection features: Intersection features are the inter- reconstructed and modeled using the extracted features.
sections of existing features, which are usually fitting In the context of building reconstruction, lines and sur-
features. Fitting features should thus be extracted before faces are fundamental elements to shape 3D building models.
obtaining intersection features. Intersection features However, extracting line and surface features from LiDAR data
include lines and points. Intersection lines can be gen- is a difficult task, because the point data are usually randomly
erated by intersecting two nonparallel surfaces, while distributed in the 3D point clouds and lack any topological con-
intersection points can be obtained by intersecting three nections. There are two common approaches to extracting 2D or
nonparallel surfaces, a surface and a line, or two lines. 3D line features from LiDAR data. The first is to extract 3D line
Figure 17.8 shows the basic ideas of several cases of inter- segments directly from the point cloud using scale and rotation
section features. invariant point and object features, as demonstrated in Gross
Boundary features: Boundary features appear at the and Thoennessen (2006) and Brenner etal. (2013). The other is to
boundary of a point set. If the point set fits to a surface intersect identified adjacent 3D planar surfaces to generate line
feature, the boundary lines and points form the boundary segments (Habib etal., 2005). Since the inherit properties of the
feature. If the point set fits to a line feature, the two end distribution of LiDAR points benefit the extraction of surface
points of the line are boundary features. Therefore, fitting features, the rest of this section will discuss methods for surface,
features must be extracted first, and then the boundary and especially planar surface, extraction.
features are determined based on the information con-
tained in the point set, as shown in Figure 17.9. 17.4.2.1 Determining Feature Parameters Using
the Least Squares Fitting Algorithm
The spatial features of simple shapes in LiDAR point clouds can
be extracted using the model-based method that simply deter-
mines their geometric parameters. Because LiDAR points con-
tain random noises, points are distributed close to the object
(a) (b) (c) (d) surfaces with small undulations. The least squares fitting algo-
rithm is often used to determine the geometric parameters.
FIg u r e 17.8 Basic ideas of intersection of (a) two surfaces, (b) three The object is to minimize the square sums of the normal dis-
surfaces, (c) a surface and a line, and (d) two lines. tance between points and the shape. Linearization and iterated
2016 by Taylor & Francis Group, LLC
354 Remotely Sensed Data Characterization, Classification, and Accuracies
calculation are required if the equation of the shape is nonlinear. while the extraction of a sphere can also utilize the normal vec-
The linearization of some shapes can be found in Shakarji (1998). tor and other constraints (Vosselman et al., 2004). One thing
Among the various shapes of geometric elements, planar sur- to note is that this type of feature extraction scheme usually
faces are the simplest case of 3D surfaces. The determination of requires a priori knowledge about the objects of interest.
plane parameters for a point set using the least squares fitting The other strategy is to directly fit the points into parametric
algorithm can be transformed to the principal component anal- surfaces. Figure 17.10 demonstrates an example of constructing
ysis of points. The solution is then solved directly using eigen- curved roof surfaces from TLS and ALS point clouds. TLS data
system analysis, without iteration (Shakarji, 1998; Weingarten acquired from multiple stations (Figure 17.10a) and airborne
etal., 2004). LiDAR point clouds (Figure 17.10b) were registered first. The
Two different types of strategies can be applied for curved roof boundaries were then generated as cubic spline curves using
lines and surfaces. The first is utilizing a semiautomatic, model- RANSAC and curve fitting algorithms. Finally, the roof surfaces
driven (or a combination of model- and data-driven) approach were approximated as ruled surfaces from the boundary curves
to compare point segments with known primitives based on the (Figure 17.10c). The advantage of this type of approach is that
shapes or characteristic parameters. For example, a cylinder can it is flexible and can deal with complicated shapes and objects,
be described by five parameters, while a sphere can be defined although usually with simplified results.
by four parameters. However, directly analyzing data in the The features within LiDAR point clouds may be implicit and
parametric space may be time consuming and the results may difficult to extract directly. Therefore, incorporating additional
not be reliable. Therefore, processes for reducing the paramet- information in the feature extraction process by fusing LiDAR
ric dimensionality or incorporating additional information are point clouds with other data sets has also been proposed, with
often introduced to improve the performance of curved feature successful results reported in different applications. Pu and
extraction. For example, extraction of a cylinder can be sepa- Vosselman (2009b) combined TLS point clouds with images
rated into two parts: cylinder axis direction and circle plane, to reconstruct detailed building facade models. Integration
(a)
(b) (c)
FIg u r e 17.10 Parametric surface reconstruction from point clouds. (a) TLS data, (b) ALS point clouds, and (c) curved roofs constructed as
ruled surfaces.
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 355
of ground plans or large-scale vector maps and ALS data was Many successful segmentation algorithms have been
developed to reconstruct polyhedral building models (e.g., Chen developed for extracting spatial features form digital images
etal., 2008; Vosselman and Dijkman, 2001). Similar techniques (Gonzalez and Woods, 1992), and the segmentation algorithms
were also applied to generate detailed 3D road models (Chen and for extracting spatial features from LiDAR point clouds are
Lo, 2009). The idea of the data fusion approach is to take advan- mainly adopted from these methods. Based on the strategy used
tage of the explicit information (building layouts, road bound- to deal with the coherence and proximity of the points, LiDAR
aries, and so on) in images and vector data sets to help extract point cloud segmentation algorithms can be classified into three
corresponding features from LiDAR point clouds. categories: clustering, region growing, and split and merge.
Clustering algorithms perform at the attribute space to group
17.4.2.2 extraction of Plane Features Using points of similar properties into clusters (Filin, 2004). The nor-
Segmentation Algorithm mal vector of the plane is the necessary attribute for plane feature
In most cases, the appearances of objects are too complex to be extraction. Some algorithms also employ additional attributes to
formulated by a simple equation. Reconstruction of these objects raise the success rate. Because the grouping of points does not con-
from LiDAR point clouds requires disassembling the surfaces of sider the distance between points, the points in each of the clusters
an object into several simple geometric elements and then model- have to be separated into neighbor point groups according to the
ing the object using the extracted results. Many studies use plane proximity criterion. The tensor voting algorithm (Schuster, 2004)
features as the primary geometric elements for reconstructing and the 3D Hough transform (Vosselman et al., 2004) are two
3D objects from LiDAR point clouds, for the following reasons: types of clustering algorithm.
Region growing algorithms (Hoover et al., 1996; Rabbani,
1. The distribution of LiDAR point clouds benefits the
2006; Vosselman etal., 2004) start with the selection of a set of
extraction of surface features.
coherent and neighboring points as the seed of a point group.
2. A general surface is too complex to model using a math-
The neighboring points of the seed are added to the point group
ematical function.
one by one if they satisfy the coherence criterion, until no more
3. The shapes of most artificial objects are composed of pla-
neighboring points can be added to the group. Then, another
nar surfaces.
seed is selected and the growing procedure continues until all
4. Detecting the lines and vertices of structures is not trivial,
points are processed.
due to the scanning mechanism of LiDAR systems.
Split-and-merge algorithms (Wang and Tseng, 2010) include
5. The lines and vertices of structures can be obtained from
two parts: the split and merge processes. The split process starts
the intersections of neighboring planes.
with the examination of the coherence of all points of the point
6. General surfaces can be obtained from the union of neigh-
cloud. If the points cannot satisfy the coherence criterion, the
boring planar patches.
space is split into eight equal-size subspaces. The same proce-
Although determining a plane feature for a point set is easy, dure is then performed on the points contained in each subspace,
as described previously, automatic extraction of all possible until all points satisfy the coherence criterion. During the merge
plane features from a large amount of points is difficult, and is step, the points contained in neighboring subspaces are merged
not a trivial task. Segmentation, a data-driven method, is the if they satisfy the coherence criteria. The merge procedure then
most popular approach for the extraction of plane features from continues until no more points can be merged and the segmenta-
LiDAR point clouds (Filin and Pfeifer, 2006; Rabbani etal., 2006; tion process is complete. In this method, the proximity criterion
Sithole, 2005; Wang and Tseng, 2010). The strategy of partitioning of points is used during both the split and merge processes.
or segmenting point clouds and extracting linear and planar fea- The profile segmentation algorithm (Sithole, 2005), which is
tures has been proved to be an effective approach for identifying adopted from the scan line segmentation algorithm designed for
building parts, such as roof facets and facades. For example, after range image segmentation (Jiang and Bunke, 1994), is a special
segmenting mobile TLS point clouds into planar faces, vertical kind of split-and-merge algorithm. In this method, the space of
wall features can be further extracted according to the inclina- a point cloud is sliced into connected or cross profiles in dif-
tion angles of the segmented planar faces (Rutzinger etal., 2009). ferent directions. Each profile is treated as a scan line, and the
Secondary features, such as windows, doors, and curtains, can adopted scan line segmentation algorithm is performed on the
also be extracted successfully and accurately based on the sizes, contained points to obtain collinear point groups. The collinear
relative positions, orientations, and other characteristics of the points at neighboring profiles are then merged to form coplanar
segmented planar surfaces (Pu and Vosselman, 2009a). point groups.
Segmentation algorithms deal with the coherence and prox- After the plane features are extracted, line and point features
imity of points (Melzer, 2007). In other word, points that are can be obtained by the intersection of neighboring plane fea-
distributed closely and have similar geometric properties are tures. With some constraints, for example, the included angle,
grouped together by the segmentation algorithm. Coplanarity neighboring planes can also be merged to form curved surfaces.
is used as the coherence property of points for the extraction However, a general curved surface, like a terrain, is difficult to
of plane features. The segment results are groups of coplanar express using a simple equation. TIN meshes are thus used with
points, each of which represents a plane feature. most GIS software to represent general curved surfaces.
2016 by Taylor & Francis Group, LLC
356 Remotely Sensed Data Characterization, Classification, and Accuracies
17.4.2.3 extraction of Line and Plane Features The disadvantage of the classical RANSAC algorithm is that it
Using the Hough transform may be very time consuming. However, limiting the selection of
The Hough transform is a classical method of surface extraction. samples from presegmented regions of point clouds will signifi-
Points belonging to a 2D straight line can be represented as a cantly improve the efficiency of RANSAC process. For example,
group of (r, ) values in the Hough space, where r is the distance when extracting the roof facets or wall facades of a building, the
between the line and the origin and is the angle of the vector selection of points should be focused on regions with the same
orthogonal to the line and pointing toward the half upper plane. normal orientation.
Similarly, points on a 3D planar surface can be described as a In conclusion, feature extraction is an essential step of LiDAR
collection of (, , ) in a 3D Hough space represented in spheri- processing and analysis. The features of interest vary with differ-
cal coordinates (Equation 17.1): ent targets and applications. From the feature extraction point
of view, LiDAR point clouds provide abundant information, and
(, ) = cos cos x + sin cos y + sin z (17.1) the methods used for extraction are as important as the data
itself. The algorithm chosen depends on the characteristics of
17.4.2.4 extraction of Plane Features Using RAnSAc the data, the objects of interest, and the aim of the application.
There is unlikely to be a single method or piece of software that
Another popular approach for plane extraction from LiDAR can adequately address the varied needs of all users, and thus it
point clouds is to employ the random sample consensus is necessary to explore the possibility of different algorithms to
(RANSAC) algorithm, which was proposed by Fischler and identify the appropriate methodology for the feature extraction
Bolles (1981). RANSAC is a resampling technique to estimate of LiDAR point clouds, as well as data fusion with other geospa-
model parameters and is designed to deal with data sets contain- tial data sets.
ing a large portion of outliers. RANSAC starts with the smallest
set of the data and proceeds to enlarge this data set with consis-
tent data points (Fischler and Bolles, 1981). A typical RANSAC- 17.5 three-Dimensional city
based algorithm for extracting a planar feature from a LiDAR Modeling from LiDAR Data
point cloud is outlined in Algorithm 1.
Algorithm 1 RANSAC for plane extraction 17.5.1 Properties of LiDAR Data in a city Area
1. Randomly select three points. The importance of 3D city modeling is increasing due to rapid
2. Construct the plane model (solve the parameters of the urbanization and the need for accurate 3D spatial information for
plane equation). urban planning, construction, and management. City modeling
3. Calculate the distance of a point to the plane, di, for all is mainly based on images and LiDAR point clouds. The unique-
points. ness of the data characteristics indicates different perspectives
4. Find inliers (points whose di is less than a predefined for city modeling. Image sensors provide spectral information
threshold). that can be used to derive the well-defined 3D corner and 3D lin-
5. If the ratio of the inliers to the total number of points ear features, which are implicit in stereo images. LiDAR systems
is greater than a predefined threshold, reconstruct the provide abundant 3D shape information for reconstruction of
plane based on all the identified inliers and terminate the city model. Figure 17.11 compares the images of a building taken
process. by an image sensor and produced by a LiDAR system. The ridge
6. Otherwise, repeat steps 1 through 5 until reaching the line is oversaturated in the image, while the ridge line can eas-
maximum number of iterations, N. ily be distinguished from shaded LiDAR triangles. The benefits
of using LiDAR point clouds for city modeling include (1) high
The final plane reconstruction (step 5 in Algorithm 1) is usu-
vertical accuracy for even low texture surfaces, (2) the ability to
ally based on a least squares estimation with all identified inliers.
directly and accurately obtain the 3D shape information, (3) the
The MA number of iterations, N, should be high enough that
ability to establish the nonplanar surfaces of objects in a city,
there is a probability p, which is usually set as 0.99, that at least
and (4) the ability to provide different viewpoints (e.g., top view
one set of randomly selected samples does not include any out-
and front view) for object interpretation. LiDAR thus has great
lier. Let u be the probability that a selected point is an inlier and
potential with regard to producing 3D spatial information for a
v = 1 u is the probability of an outlier, then
city model.
Airborne LiDAR acquires data from the air to the ground and
1 p = (1 u 3 )N (17.2) is used to obtain the 3D points on building rooftop and object
surface. On the other hand, terrestrial LiDAR usually acquire
and the 3D points on building faade and object surface. The scan-
ning distance and beam divergence angle of airborne LiDAR is
log(1 p) larger than terrestrial LiDAR and consequently the point density
N= (17.3)
log(1 (1 v )3 ) of airborne LiDAR is lower than terrestrial LiDAR. Therefore,
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 357
(a) (b)
(c)
FIg u r e 17.11 An example of a building ridge line in an aerial image and airborne LiDAR point clouds: (a) aerial image, (b) LiDAR triangle
mesh, and (c) horizontal view of LiDAR triangle mesh.
airborne LiDAR is more suitable for city-scale object recon- (Brenner, 2005). The model-driven approach is a topdown
struction, while terrestrial LiDAR is suitable for building-scale strategy that starts with a hypothetical building model, which is
detailed object reconstruction. For example, the OGC CityGML verified by the consistency of the model with the existing data.
(OGC, 2012) LOD1 and LOD2 city objects can be reconstructed This method needs to define a database of building primitives,
by airborne LiDAR, while the terrestrial LiDAR is usually used and thus a parametric model is generally used, as shown in
for LOD3 and LOD4 city objects. Figure 17.12a. The data-driven approach is a bottomup strategy
in which the building features, such as point, linear, and pla-
nar features, are extracted at the beginning, then grouped into
17.5.2 object Reconstruction Strategies
a building model through a hypothesizing process, as shown in
Object reconstruction strategies can be classified into three cate- Figure 17.12b. Finally, the hybrid approach integrates the ideas of
gories, that is, model-driven, data-driven, and hybrid approaches both model-driven and data-driven methods.
2016 by Taylor & Francis Group, LLC
358 Remotely Sensed Data Characterization, Classification, and Accuracies
Primitives database
Flat Gable Hipped
Model fitting
Image Generated model
(a)
Image 1 Edges of image 1 Lines of image 1 Inference
engine
3D lines Generated model
Image 2 Edges of image 2 Lines of image 2
(b)
FIg u r e 1 7.12 Illustration of (a) model-driven and (b) data-driven strategies.
17.5.3 Building extraction 17.5.3.1 Building Detection
Buildings are the most attractive elements in a 3D city. There The role of building detection is to extract the location of regions
are three major steps to establish a building model, and these where there are buildings. Once a buildings location is estab-
are detection, reconstruction, and attribution (Gruen, 2005). lished, the reconstruction process can be focused in a specific
Building detection is composed of various methods to detect region rather than considering the whole data set. This advan-
ROIs for subsequent building reconstruction. Building recon- tage not only saves computing time but also reduces ambigu-
struction is the determination of the 3D geometrical description ity. The idea of building detection is carried out using different
of buildings located in a given ROI. Finally, building attribu- characteristics, such as spectral, texture, shape, and roughness,
tion assigns the depiction of building properties, such as type of to separate buildings and nonbuildings. A general procedure
building, semantic attribution, and textures. All three steps are to obtain a building region from LiDAR includes the following
described in more detail, in the following. steps: (1) generate aboveground objects by subtracting the digital
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 359
terrain model (DTM) from the DSM; (2) calculate the different 165
features for each object (an objects features can be shape, tex- 2,741,400 160
155
ture, roughness, echo ratio, and so on); and (3) separate building
150
and nonbuilding regions based on the objects features using a 2,741,300 145
classifier. 140
135
If LiDAR point clouds are combined with a multispectral 130
2,741,200
image, then building regions can be detected by simultane- 125
ously considering spectral and shape information. The detection 120
2,741,100 115
rate may reach 80% when the automatic approach is adopted, 110
although this figure could be worse if the point cloud density 105
2,741,000
is too low or the area of the building is too small. For exam-
250,400 250,500 250,600 250,700 250,800 250,900 251,000 251,100
ple, Rottensteiner et al. (2005) used LiDAR data with an aver- (a)
age point distance of 1.2 m and multispectral images with 0.5
m spatial resolution to perform building detection. For a build-
ing larger than 40 m2, the detection rate was between 50% and
90%, but for a building smaller than 40 m2, the detection rate
was lower than 50%. Point density and building size are two
major factors that will influence the detection rate. Increasing
point density may thus improve the accuracy of building detec-
tion. Figure 17.13 shows an example of building detection using
a LiDAR and multispectral image.
17.5.3.2 Building Reconstruction
The objective of building reconstruction is to build up the geom- (b)
etry of a building. The irregular point clouds need to be structur-
alized for further processing. There are different data structures
for LiDAR data in building reconstruction. The first one uses
point data to perform the 3D Hough transform to extract the
building models from LiDAR point clouds (Hoffman, 2004).
The second one applies TIN data to analyze the planar param-
eters (Chen etal., 2008). The third one groups and linearizes the
building models based on grid data.
The general methods used for building reconstruction
based on LiDAR data can be classified into two types. In the
first type, planar features are extracted and then the extracted
planes are used to derive the line features (Sampath and Shan,
(c)
2007). The building model is obtained by integrating the pla-
nar and linear features. In the second type, linear features are FIg u r e 17.13 An example of building detection: (a) LiDAR digi-
first extracted and then the extracted lines are used to trace tal surface model, (b) multispectral image, and (c) detected building
the building polygons. The building model is obtained by shap- regions.
ing the top of the building polygons (Hu, 2003). Each approach
itself, like floor number, area, and volume. The second are the
has less planimetric accuracy when a regular density (e.g., 12
semantic attributes, which can be obtained from exterior data,
pts./m 2) is employed. Figure 17.14 shows examples of building
like a topographic map. Semantic attributes include the names
reconstruction using LiDAR contours and the boundary regu-
of buildings and materials of buildings. The third attributes are
larization method.
based on the building texture, which make the building model
17.5.3.3 Building Attribution more photo-realistic.
A building model is an object in a geographical information
system (GIS). It consists of spatial information and attributes.
17.5.4 Road extraction
Building reconstruction is used to shape the spatial informa-
tion of a building, while building attribution is used to assign A 3D road network is one of the important infrastructures needed
the attributes of a building model. Building attributes enable the for an intelligent transportation system and GIS, and these are
3D analysis to produce more valuable results. There are three applied to transportation management, maintenance, planning,
kinds of attributes. The first can be generated from the data analysis, and navigation. The traditional 2D road centerlines and
2016 by Taylor & Francis Group, LLC
360 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) (b)
(c)
FIg u r e 17.14 Examples of building reconstruction: (a) building contours, (b) building boundaries after regularization, and (c) building models.
2D road boundaries are insufficient to represent the 3D reality of 2013). Once the road regions are detected, the 3D road surface
actual road systems, especially multilayered ones. Since LiDAR modeling process will focus on them. A general procedure of 3D
data may provide a large number of 3D points on the road surface, road surface modeling includes the following steps: (1) remove
these can be used to generate 3D road models (Oude Elberink the nonroad points based on road profile fitting, (2) carry out 3D
and Vosselman, 2009). Road extraction also includes three major road surface fitting by using polynomial functions, and (3) check
steps, that is, detection, reconstruction, and attribution. Road the consistency of 3D road segments. The main challenge in road
detection is used to detect the 2D road region, while road recon- reconstruction is the shaping of multilayer roads. Because of the
struction is to shape the 3D road surface. Road attribution is then occlusion of multilayer roads, an inference engine is needed to
used to assign the attributes to the road models. restore the missing parts under the upper roads. Figure 17.15
Road detection from LiDAR data usually analyzes the LiDAR shows an example of multilayer road extraction using LiDAR
intensity, flatness, and continuity. The near infrared of the LiDAR points (Chen and Lo, 2009).
system may have a low return signal in asphalt-covered region,
and this is an important attribute that can be used to detect roads.
The road design should meet the related regulations, which means 17.6 Full-Waveform Airborne LiDAR
the flatness and slope of roads can be estimated to detect road
17.6.1 introduction
regions. Finally, the continuity of a road network is another con-
straint that should be considered (Hu, 2003). Note that LiDAR Recent advances in laser scanning technology have led to the
points are irregular points that do not always model the step edge development of new airborne LiDAR commercial systems, called
or roadside. The road boundaries extracted from LiDAR points waveform LiDAR systems. A waveform LiDAR system is capable
are thus not as accurate as the road boundaries obtained from of recording the complete waveform of a return signal (Mallet and
vector maps. An alternative approach to road detection is adopt- Bretar, 2009), which is the digitized intensity of the backscattered
ing the reliable 2D road boundaries from existing topographic signals obtained from the surfaces illuminated by an emitted laser
maps, with LiDAR points then used for road surface shaping. beam. In addition to range measurements, other physical features
In road reconstruction, the major tasks are surface modeling of the illuminated surfaces can also be derived from the waveform
and making connections between road segments (Chiu et al., data (Guo etal., 2011). For example, the echo width, amplitude, and
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 361
backscatter cross section obtained from waveforms have proven so that it improves the interpretation of physical measurements. It
to be useful for object classification of urban (Alexander et al., is also possible to retrieve some missing points by detecting sur-
2010) and vegetation (Heinzel and Koch, 2011; Neuenschwander face responses known as echoes in waveforms, and the detection
etal., 2009) areas. Compared with conventional LiDAR systems, process is especially useful with regard to weak and overlapping
a waveform scanner provides additional information about the echoes. This ability is very significant for some applications of air-
physical characteristics and geometric structures of the illumi- borne LiDAR and especially for the study of forest areas.
nated surfaces. A waveform data set can be used to generate finer With improvements in data storage capacity and process-
point clouds than the data set originally provided by the system, ing speed, waveform airborne LiDAR systems have become
104
103
102
101
100
99
98
97
96
95
94
93
92
91
90
89
88
87
86
85
84
83
82
81
80
79
78
77
(a)
104
103
102
101
100
99
98
97
96
95
94
93
92
91
90
89
88
87
86
85
84
83
82
81
80
79
78
77
(b)
FIg u r e 1 7.15 An example of road extraction: (a) LiDAR points, (b) road regions. (Continued )
2016 by Taylor & Francis Group, LLC
362 Remotely Sensed Data Characterization, Classification, and Accuracies
(c)
FIg u r e 1 7.15 (c ontinued ) An example of road extraction: (c) road models.
increasingly available and first appeared on the commercial 17.6.2 Waveform Data Analysis
market in 2004 (Hug etal., 2004). Table 17.3 shows the techni-
cal specifications of waveform airborne LiDAR systems manu- The waveform data record a series of return signals related to
factured by Leica, Riegl, Optech, and Trimble. Such systems each individual laser shot. The data sampling interval is usually
store the entire waveform of each received pulse, and thus the 1 ns, which corresponds to a 15 cm interval along the path of
sum of reflections from all intercepted surfaces within the laser the laser beam. The intensity of the backscattered signal is then
footprint is preserved. Compared to discrete-return airborne quantified within a range of specified digital levels (e.g., 8bits),
systems, which only provide a single range distance to a tar- so that it is finally converted to a digital data stream. In general,
get, waveform systems record the entire time history of laser it is understood that the higher the sampling rate and the signal-
pulses with a high-resolution sampling interval. Moreover, quantified levels are, the more details of the raw analogue signal
such waveforms are able to store more details about illumi- that are kept.
nated targets than discrete systems are. Users can then apply As the waveform data contain valuable information about
their own pulse detection methods to identify return signals the object distribution and reflectance along the laser path, a
in a more effective manner. The waveform attributes, which specially designed process of waveform data analysis is needed
represent characteristics of each scanned object, can then be to extract the information contained in the waveform data.
extracted and used as criteria for distinguishing the points Echo detection and waveform feature extraction are two com-
into different types. mon processes of waveform data analysis, which will both be
TABLe 1 7.3 Technical Specifications of Waveform Laser Scanners Produced by Leica, Riegl, Optech, and Trimble
Leica Leica Riegl LMS Riegl LMS Optech ALTM Optech ALTM Trimble Trimble
Manufacturer ALS70 ALS60 Q780i Q680i Pegasus 3100 Harrier 68i Harrier 56
Beam deflection Oscillating Oscillating Rotating Rotating Oscillating Oscillating Rotating Rotating
mirror mirror polygon polygon mirror mirror polygon polygon
Flying height (m) 16005000 2005000 <4700 <1600 3002500 <2500 301600 <1000
Laser wavelength (nm) 1064 1064 1064 1550 1064 1064 1556 1550
Pulse width (ns) 56 5 3 3 7 8 3 <4
Pulse rate (kHz) 250500 <50 100400 <400 <400 <200 <400 <240
Beam divergence (mrad) 0.22 0.22 0.25 0.5 0.2 0.3 or 0.8 0.5 0.3 or 1.0
Field of view (degrees) 075 75 060 30 75 25 4560 45 or 60
Sampling interval (ns) 1 1 1 1 1 1 1 1
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 363
introduced in this section. However, before this the shape of multiple returns within the travel path of the laser pulse, the
the received waveform is first discussed. deformation of the received waveforms will be more complex
than for those affected by surface geometry alone. As a result,
17.6.2.1 Shape of Received Waveform since reflected laser pulses are distorted by surface variations
As the shape of a received waveform varies subject to differences within the footprint, the shapes of received waveforms impor-
in the target structures (e.g., the distribution of illuminated tant information in relation to surface roughness, slope, and
objects and surface reflectance along the laser path), it is neces- reflectivity (Gardner, 1992).
sary to study the types of waveforms caused by some typical tar-
get structures. Based on the size of the focused surface geometry 17.6.2.2 echo Detection
within the laser beam (footprint, d; wavelength, ), Jutzi et al. As full-waveform systems provide an entire time history of laser
(2005) categorized illuminated targets into three types, which pulses, users are able to apply their own pulse detection meth-
were macro, meso, and micro structures. Macro structures are ods to detect successive returns and determine the elapsed time.
considered to have a more extended surface than the footprint It even becomes possible to design a combined pulse detection
d (e.g., building roofs). Meso structures are considered to have a method becomes possible to extract information of interest for
less extended surface than the footprint d, but much greater than specific applications (Wagner et al., 2004). For this purpose,
wavelength . Such structures result in a mixture of different Harding (2009) pointed out that the laser pulse width, detec-
range values, for example, in an area with small elevated objects, tor sensitivity, and response time, the systems signal-to-noise
a slanted plane, large roughness, or vegetation. Finally, micro performance, detection TH, and implementation of the ranging
structures are considered to have a surface less extended than electronics all influence the ability to detect discrete returns.
the wavelength . It is also noted that the shapes of the received Wagner etal. (2004) reported that the characteristics of the effec-
waveforms vary depending on different surface structures illu- tive scattering cross section, object distance, and noise level can
minated, especially for meso structures. affect the performance of the detectors. For received pulses with
Jutzi and Stilla (2006) further examined the relation- a complex form, the number and arrival timing of the returns
ship between surface structures and the shape of waveforms are critically dependent on the detection methods employed.
(Figure17.16). A sloped surface or an area with randomly dis- Many echo detection algorithms are available for waveform
tributed small objects causes the deformation (widening) of the data. Wagner etal. (2004) compared the results of using methods
backscattered waveform with one distinct peak (Figure 17.16b based on the TH, center of gravity (CG), MA, zero crossing of
and e). With an increase in the height difference between two the first deviation (ZC), and constant fraction (CF). It was found
elevated surfaces, the two peaks of a waveform are increasingly that the ZC algorithm has the best discrimination with regard to
close together (Figure 17.16d). detecting overlapping echoes among the five algorithms, espe-
Moreover, Wagner et al. (2006) pointed out that when cially when the noise is at minimum. The CF algorithm achieved
the two scattering clusters are at a distance comparable to or worse results than the ZC algorithm, while the TH, CG, and MA
smaller than the range resolution, which is the pulse duration algorithms failed at resolving overlapping echoes.
of the LiDAR system, a waveform with a scattering cluster is Current state-of-the-art echo detection techniques use
presented. Jutzi and Stilla (2006) claimed that the waveform of small-footprint full-waveform data, for example, the correla-
overlapping responses at the distance of 0.85 times pulse width tion (Levanon and Mozeson, 2004), deconvolution (Jutzi and
becomes a single widened peak and is unlikely to be separated Stilla, 2006), wavelet-based (WB), and Gaussian decomposition
as two returns by peak detection. It is also apparent that when methods. Since the waveforms can be considered as a convolu-
considering the effects from various reflectance properties of tion between system waveform and apparent cross section of
Amplitude
Amplitude
Amplitude
Amplitude
Amplitude
Time Time Time Time Time
(a) (b) (c) (d) (e)
FIg u r e 17.16 Surface structure and waveform: (a) plane surface, (b) sloped surface, (c) two significantly different elevated areas, (d) two slightly
different elevated areas, and (e) randomly distributed small objects. (Adapted from Jutzi, B. and Stilla, U., ISPRS J. Photogramm. Remote Sens.,
61(2), 95, 2006.)
2016 by Taylor & Francis Group, LLC
364 Remotely Sensed Data Characterization, Classification, and Accuracies
scatters (Wagner et al., 2006), the surface response (cross sec- the results. Typical surface features extracted from Gaussian
tion of scatters) can be recovered by deconvolution algorithms parameters are range, roughness, and reflectance (Jutzi and
(Jutzi and Stilla, 2006; Wu etal., 2011). However, prior knowl- Stilla, 2005), and thus the estimated parameters provide a direct
edge such as the original laser pulse shape needs to be known, physical interpretation of the surface targets. Additionally,
and sampling of the return signal also needs to be sufficient for Wagner etal. (2006) adapted the radar equation order to convert
such approaches to be applied, and for many of todays LiDAR the received power into a backscatter cross section. As this is a
systems, these prerequisites are not met (Morsdorf etal., 2009). measure of how much energy is scattered backward toward the
The WB echo detector is based on a wavelet transformation sensor and is very useful for comparing the physical quantities
algorithm, which measures the similarity between the signal between different surveys, Wagner et al. (2008) suggested that
and the wavelet. A wavelet (i.e., a Gaussian wavelet for detecting backscatter cross sections should be also considered as a stan-
echoes of LiDAR waveforms) is thus chosen if its shape is similar dard product in laser scanning. The related waveform attributes
to that of the echoes being detected. The wavelet can be scaled are as follows:
by a scale factor and shifted along the signal time domain by
1. Pulse width: The pulse width can be described by the
a translation factor, and so echoes can be located by detecting
standard deviation Wj in Equation 17.4. It can be a quan-
significant peaks in the wavelet coefficients (the results of wave-
titative indicator to evaluate the extent of the pulse broad-
let transform). Gaussian decomposition is widely used to model
ening. Several studies into small-footprint waveforms
small-footprint waveforms. With this method, it is assumed that
have observed that the pulse width of vegetation points
pulses are transmitted with a Gaussian-like distribution (i.e., the
is generally larger than that of terrain points (Persson
impulse response is Gaussian) and that the received signal is a
etal., 2005; Wagner etal., 2008). Forest terrain also tends
sum of individual Gaussian distributions (Hofton et al., 2000;
to exhibit larger pulse widths than open terrain (Mcke,
Wagner et al., 2006). Fitting Gaussian functions to waveform
2008), with trees and meadows also generating larger
data provides each return with a parametric description, which
pulse widths than buildings (Stilla and Jutzi, 2008). The
can be used to store pulse shape information and decrease the
distinction of smooth surfaces from plants, bushes, trees,
effect of noise. In this way, the Gaussian parameters for each
and even short hedges thus seems to be relatively clear
return, which are the amplitude, temporal position of Gaussian
(Wagner etal., 2008). This implies that pulse width can be
peak, and Gaussian width, can be extracted.
an additional factor used to discriminate between ground
The Gaussian model is a classical curve fitting algorithm
returns and vegetation returns, although Stilla and Jutzi
(Wagner etal., 2006) as described in the following equation:
(2008) suggested that additional information, such as the
3D geometrical relationships of the returns, is required to
(x )2
f j (x i ) = A j exp (17.4) classify each return pulse as a specific surface type.
2
2W j 2. Amplitude: The amplitude can be represented by Aj in
Equation 17.4. The estimated amplitude for each return
where by postprocessing waveforms provides reflectance infor-
Aj and Wj are the pulse amplitude and width of the laser pulse mation about the illuminated targets. However, without
j is the center of the echo calibration of such data, the information is too noisy to be
utilized for classification purposes. It has been observed
Roncat etal. (2011) and Cheng and Tsai (2014) both proposed that there is a large overlap among the amplitude histo-
fitting waveforms into higher-order spline curves to better grams obtained from different land-cover classes (Ducic
describe the laser waveform patterns. However, high-order et al., 2006; Mcke, 2008). At low amplitudes, the reli-
spline curve fitting requires more intensive computation, and ability for the estimate of pulse width can be problematic
waveform fitting is usually based on the peak locations of the (Wagner etal., 2008). Since the emitted power is constant,
echo signals. However, since most full-waveform laser scanners pulse broadening of received returns also results in a
provide only a limited number of signal peak locations, certain reduction of the amplitude.
subtle but useful waveform features may be overlooked. Lu and 3. Backscatter cross section: The backscatter cross section is
Tsai (2013) thus adopted derivative analysis (Tsai and Philpot, one of the unique features available from full-waveform
1998, 2002) to detect the occurrence of local peak positions LiDAR data. The backscatter cross section provides infor-
more accurately and to generate a more complete waveform of mation in relation to the range and scattering properties of
the returned laser pulse. the targets (Wagner etal., 2006), and it depends strongly
on the number of returns (Wagner etal., 2008). For each
17.6.2.3 Waveform Feature extraction individual backscatter cross section for each return, it
With the development of full-waveform LiDAR, waveform- has been observed that single returns produce stronger
related characteristics have also become valuable features for estimates than multiple returns. Grass and gravel can be
LiDAR analysis and applications. Once the waveform has been distinguished well, with a small overlay of cross-sectional
decomposed, related waveform features can be derived from values estimated. Vegetation returns generally produce
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 365
TABLe 1 7.4 Strengths/Limitations of Waveform and Multireturn LiDAR Point Cloud
Waveform Point Cloud Multireturn Point Cloud
Number of return is unlimited. Number of return is limited.
Detection of error or missing echoes become possible. Echo detection is a black box operation. Detection of error or missing cannot be
evaluated.
Dead zone problem can be eased by a user developed detector. Multireturn system often suffers from LiDAR dead zone problem
(overlapping echoes problem).
More features (waveform features) can be attached to the point cloud. Only one additional information (intensity) can be attached to each point.
lower cross-sectional values than terrain returns, which Section 17.6.2.3 have been demonstrated in 3D point cloud clas-
may be useful to discriminate terrain from vegetation sification. Integrations with other information (e.g., geometric
points (Wagner etal., 2008). In addition, the backscatter and spectral information) provide even more criteria that can
cross section varies within forest canopies, which may be aid in achieving advanced applications, such as forestry investi-
useful for tree species classification (Wagner etal., 2008). gation (Buddenbaum etal., 2013) and building modeling.
The backscatter cross-sectional feature () of a laser echo Comparing between waveform and multireturn LiDAR point
can be computed from the reflectance (), range (R), and clouds, the main applications are not much difference. For
laser beam divergence angle () of the target (Wagner example, DEM generation, building reconstruction, and forest
etal., 2006), as described in the following equation: parameter estimation can be done by using both kinds of point
clouds. However, a major improvement for waveform point
cloud is that the landscape or the target can be described in more
= R 22 (17.5) detail. Thus, the detection or the reconstruction of the interested
targets can be better accomplished. Besides, the waveform fea-
Equation 17.5 represents the general form of the apparent ture is another advantage for waveform point cloud applications,
cross section of each surface within the laser footprint. For especially on the purpose of point cloud classification. Table17.4
a Gaussian waveform, the equation can be rewritten as lists the strengths and limitations of waveform LiDAR point
cloud and multireturn point cloud.
= C cal R 4 A jW j (17.6)
17.7 DeM and DSM Generation
2 from Airborne LiDAR Data
C cal = (17.7)
R 2 A jW j 17.7.1 introduction
A DEM is a digital model (or the height information) repre-
where Ccal is a calibration constant that can be obtained senting the ground surface of a terrain. In general, the ground
through laboratory experiments (Alexander etal., 2010). surface means the earths surface without any objects like
4. Backscatter coefficient : The backscatter coefficient is plants, buildings, and other man-made structures. However,
the backscatter cross section per unit-illuminated area. permanent earthwork structures, such as dams and road
Alexander etal. (2010) found that the backscatter coeffi- embankments, are usually counted as the ground surface. In
cient is more useful than the amplitude and backscatter contrast, a DSM, in general, represents the visual surface of the
cross section for discriminating road and grass. Wagner earth, including all the objects on it. However, in most cases,
(2010) stated that it is helpful to employ the backscattering we do not treat the surface of a water body as part of DSM.
coefficient when comparing data sets acquired by differ- Figure17.17 shows the general definitions of DEM andDSM.
ent sensors and/or different flight campaigns.
DSM
17.6.3 Applications DEM
Waveform LiDAR data has been used to explore weak and over-
lapping returns that were missed by the on-the-fly detection
process of LiDAR systems. The point clouds extracted from the
waveform data can then better describe the sensed landscape.
The canopy height (Hancock etal., 2011; Wang etal., 2013), for-
est biomass (Clark etal., 2011; Yao etal., 2011), and DTM can
thus be improved.
The land-cover classification is another important application
of waveform LiDAR data. The waveform features introduced in FIg u r e 17.17 The general definitions of DEM and DSM.
2016 by Taylor & Francis Group, LLC
366 Remotely Sensed Data Characterization, Classification, and Accuracies
ADEM or DSM can be represented as a raster (a grid of squares of DEMs. Given the decreased penetration capability of LiDAR
data structure) or as a vector-based TIN. systems in dense building or forest areas, the filtering process
Airborne LiDAR data have proven effective for the genera- may mistakenly register nonground points as ground points,
tion of high-resolution and high-accuracy DEMs and DSMs of or vice versa. This error depends on the parameter setup of the
various terrestrial environments. In general, a DEM is formed specific filter employed. To remedy the misclassification of point
with the ground points extracted from a data set of LiDAR point clouds and obtain a reliable DEM, a manual editing procedure is
clouds, while a DSM is formed with the surface (first-return) conducted in the registration of point clouds to the correct class.
points extracted from LiDAR data. The extracted ground and However, this manual editing procedure requires visual inspec-
surface points are treated as measured points to form DEMs and tion of the thin profiles of point clouds for the whole survey area
DSMs (Mongus and alik, 2012). and thus is extremely time consuming and costly.
Accurate DEMs are often used to derive morphometric param- Different levels of DEM quality can therefore be related to
eters (e.g., slope, aspect, and surface ruggedness) for the mor- the LiDAR DEM generation procedures, which are subject to
phological analysis of an area of interest. Morphological analysis the processes of calibrating possible systematic errors, check-
is an important tool for characterizing environmental changes in ing misclassification errors, and maintaining topographical fea-
both temporal and spatial scales. For example, the data derived tures. In accordance with the typical demands of users, DEM
from DEMs allow the identification of morphologically homoge- products are usually categorized into three levels, as follows:
neous zones, for example, the identification of different volcanic
1. Level 1 DEM: DEMs generated by means of a fast and
edifices, structural domains, and evolutionary stages (Norini
automatic procedure, without any quality control process,
etal., 2004). DEMs have been applied to reveal the variation in
belong to this category. The products are not verified via
elevation of hills and river beds for landslide and inundation
a careful strategy for the removal of systematic errors and
investigations focusing on Chenyulan River (Yang etal., 2011).
are thus likely to involve a number of mistakes. As the
High-resolution DEMs generated by LiDAR have now become
land-cover objects of the observed region become denser
more readily available, and advances in computing technology
and more complicated, so the proportion of the inappro-
make it now possible to produce spatially explicit, fully distrib-
priate filtered points of the DEM could become higher.
uted hydraulic and hydrological models and hydrogeomorpho-
2. Level 2 DEM: DEMs generated by means of a careful
logical assessments (Biron etal., 2013; Vaze etal., 2010).
double-check procedure with a quality assessment and
control process belong to this category. The procedures of
17.7.2 Data Processing for DeM LiDAR system calibration and strip adjustment are usu-
and DSM Generation ally required to reduce the systematic errors. In addition, a
procedure of manual editing is also needed to improve the
DEM generation usually employs a filter (or a classifier) to sepa-
quality of ground point extraction (filtering) and ensure
rate ground points from nonground points. Popular filters, such
the reliability of final products.
as those based on TINs (Axelsson, 2000) and morphology (Pingel
3. Level 3 DEM: If the level 2 DEMs are generated with con-
etal., 2013), are mainly developed by exploiting the detailed geo-
sideration of water boundaries and geomorphological
metrical description of the ground topology in LiDAR point clouds.
features, then they belong to this category. To meet this
For DSM generation, the surface points can be obtained by choos-
requirement, the filtered point cloud should be manually
ing all of the first-returning echoes of the LiDAR data. The process
inspected and edited to improve the precision of interpo-
of DSM generation is much easier than that of DEM generation.
lation by adding certain points of water boundaries and
The overall data processing for DEM and DSM genera-
geomorphology features (e.g., ridges, valleys, steep slopes).
tion using airborne LiDAR data includes preprocessing, bore-
sight calibration, and point cloud generation, strip adjustment,
ground and surface point filtering, and interpolation to form
17.7.3 DeM Quality Assessment and control
DEM and DSM. A LiDAR data set may be contaminated by
systematic errors, if the procedures of boresight calibration and The process of DEM quality assessment and control (QA/QC) is
strip adjustment are not well performed. The point density of the to ensure the quality of the generated DEMs through a series of
source data constrains the resolution of the generated DEMs and verification procedures. The quality of a DEM may be affected by
DSMs (Florinsky, 1998). Accordingly, the resolution of DEM and three major factors, including the quality of the original LiDAR
DSM is constrained mostly by the density of the input terrain point cloud, the quality of the filtering, and the quality of the
data. McCullagh (1988) suggested that the number of grid cells DEM interpolation (if applicable, including thinning and break
should be similar to the number of terrain data in a covered area. line modeling) (Pfeifer and Mandlburger, 2008). These three fac-
Since the extracted ground points are treated as measured tors affect the DEM quality at different stages of the generation
points to form DEMs, the accuracy of the generated DEMs is procedure. The QA/QC must be performed step by step on each
subject to the correctness of ground point extraction. Under this stage of the DEM generation workflow. For many practical proj-
circumstance, ground objects, which prevent laser beams from ects, aerial photos of the survey area are also acquired during
reaching the ground surface, may significantly affect the quality the LiDAR point cloud acquisition. An orthophoto of the survey
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 367
area can then be produced based on these photos and the LiDAR The main advantages of TLS are as follows: (1) the direct mea-
point clouds, which can provide helpful visual clues for manual surement of 3D coordinates, (2) the high degree of automation,
point cloud classification. The QA/QC procedure should thus (3) the easy-to-use hardware, and (4) a massive sampling capa-
take into account the use of aerial photos. An effective the DEM bility (Monserrat and Crosetto, 2008).
QA/QC procedure may include the following six items:
17.8.1.1 Properties of terrestrial LiDAR Data
1. Verification of airborne LiDAR mission plan: to be per-
The measurement mechanism of a TSL is similar to that of a
formed before the scanning operation, including inspec-
total station, a surveying instrument used for angle and distance
tion of flight plan and LiDAR system calibration report.
measurement. The difference is that a TSL allows automated and
The nominal point density should be checked according
near-simultaneous measurements of hundreds or thousands of
to the flight plan.
nonspecific points in the area surrounding the position where
2. Verification of ground control survey: the quantity, distri-
the instrument is set up. In Figure 17.1, a slant range by the laser
bution, and quality of ground control points (GCPs) used
rangefinder and two orthogonal angles by angular encoders in
for LiDAR strip adjustment.
the horizontal and vertical directions are measured simultane-
3. Verification of LiDAR points and strip adjustment: to
ously. The point clouds are then captured point by point by rotat-
check the LiDAR strip coverage and overlaps, overall
ing the reflecting mirrors whose horizontal/vertical angles are
point density achieved, and external and internal errors
gradually increased, with a measuring rate of 1000Hz or more.
of the strips.
These simultaneous measurements of distance and angle are car-
4. Verification of LiDAR point cloud filtering: including
ried out in a highly automated manner (Petrie and Toth, 2008b).
visual inspections and field survey checks.
The TSL can be mounted on a tripod for fixed positions or on a
5. Verification of aerial photography acquisition: to check
vehicle (or a moving platform) for mobile mapping. The first case
overall image coverage and quality.
also refers to static scanning, in which the exterior orientation
6. Verification of orthophoto: to check the quality of the
of the platform is constant for one scan position, and 2D cover-
orthoimages, including the overall image continuity and
age in the angular domain is performed by rotating components
possible distortion of ground objects.
of the device. In the second case, the scanning is performed by
The quality index of a DEM production depends on the require- a univariate beam deflection unit, and area-wise data acquisi-
ments of the project. For general purposes, the quality index tion is established by the movement of the scanning platform,
provided in the USGS LiDAR Base Specification Version 1.0 that is, dynamic scanning is achieved (Pfeifer and Briese, 2007).
(Heidemann, 2012) is an appropriate reference. Many different kinds of static and dynamic TSLs that are used
At forest area, airborne LiDAR laser may penetrate the gap for topographic mapping and modeling operations are described
between leaves and detect the ground points under tree canopy, in Petrie and Toth (2008b).
which is difficult to be obtained from aerial photos. Therefore,
DEMs produced from LiDAR data may have higher accuracy 17.8.1.2 Data Processing Strategy
than that from photogrammetry. At an earlier study, the height The 3D point cloud captured with a TLS contains detailed geomet-
accuracy of DEM derived from airborne LiDAR data is about ric information, but further data processing is still needed for cer-
10 cm while that derived from photogrammetry is about tain applications. In general, the processing of TLS LiDAR data can
25cm (Kraus and Pfeifer, 1998). be divided into three procedures: georeferencing, feature extrac-
tion, and visualization. The task of georeferencing is to transform
the original point cloud, registered in a local coordinate system,
17.8 terrestrial/Vehicle-Borne into a ground reference coordinate system (GRCS). The task of
LiDAR Data Processing feature extraction converts the point cloud data to be meaningful
information for further processing, and the types of derivative fea-
17.8.1 terrestrial LiDAR Data Processing
tures are subject to the involved applications, such as object clas-
Terrestrial laser scanners (TLSs) are stationary, ground-based sification and surface and building reconstruction. Visualization
LiDAR systems that scan surrounding objects. The data cap- and interactive operations implemented on a computer are usually
tured are also point clouds, which are distributed as layers needed for manual interpretation and extraction of meaningful
corresponding to the surfaces of scanned objects. A growing object features. The traditional field work of mapping surveys can
number of applications, such as change detection and deforma- now be performed on a computer system with the TLS data.
tion analysis (Lim etal., 2005; Monserrat and Crosetto, 2008;
Santibanez etal., 2012), rapid modeling of industrial infrastruc- 17.8.1.2.1 RegistrationandGeoreferencing
ture, architecture, agriculture, construction or maintenance The point clouds needed for 3D object modeling may be acquired
of tunnels and mines, facility management, and urban and from different scan stations, although the local point clouds
regional planning, have been carried out using TLS and dem- must be transformed into a common coordinate system for fur-
onstrated that it is a key surveying tool for capturing and mod- ther processing. If only the focal object itself is of interest, it is
eling highly detailed geospatial data. sufficient to determine the relative orientation between scans
2016 by Taylor & Francis Group, LLC
368 Remotely Sensed Data Characterization, Classification, and Accuracies
using registration. If the object also has to be placed in a supe- clustering of features (Biosca and Lerma, 2008; Filin, 2002; Filin
rior coordinate system, absolute orientation is also needed. If the and Pfeifer, 2006; Hoffman, 2004; Lerma and Biosca, 2005), or
superior coordinate system is earth fixed, this then becomes a model fitting (Bauer etal., 2003; Boulaassal etal., 2007; Bretar and
georeferencing task (Pfeifer and Briese, 2007). Roux, 2005).
One of the most popular methods of registration is the ICP
(iterative closest point) algorithm developed by Besl and McKay 17.8.1.2.3 Visualization
(1992). Some variations and improvements based on ICP meth- The visualization system is the foundation for several interac-
ods have been proposed by Chen and Medioni (1992), Zhang tive analysis tools for quality control, extraction of survey mea-
(1994), Okatani and Deguchi (2000), and Segal etal. (2009). The surements, and the extraction of isolated point cloud features
ICP is based on the search for pairs of nearest points between (Kreylos etal., 2008). Staiger (2003) treated the visualization of
two scans, and the rigid body transformation is estimated and point cloud data into six different ways:
applied to the points of one scan. The ICP procedure is iter-
1. Point clouds in a 3D projection, with a color- or grayscale-
ated until convergence is achieved. Another method for point
coded representation of intensity, are often used as a first
cloud registration using least squares matching was proposed
visual check of the acquired data.
by Gruen and Akca (2005). The least squares 3D surface match-
2. Point clouds can also be combined with derived geometri-
ing (LS3D) algorithm estimates the transformation parameters
cal elements.
between two or more fully 3D surfaces, using the generalized
3. True orthophotos are realized by the fusion of digi-
GaussMarkov model, minimizing the sum of the squares of
tal images (point information) and the registered point
the Euclidean distances between the surfaces. This formula-
clouds (geometry).
tion makes it possible to match arbitrarily oriented 3D surfaces
4. 3D contour plans.
simultaneously, without using explicit tie points.
5. 3D models.
The georeferencing of TLS data is a procedure of coordinate
6. Virtual flights through the modeled scene.
transformation in which the 3D point cloud is transformed to
a GRCS, and the method of georeferencing used may be direct
17.8.2 object extraction from
or indirect (dos Santos etal., 2013). The DG method is based on
Vehicle-Borne LiDAR Data
integrating additional POS sensors, such as a GNSS and an IMU,
so that the platforms position and orientation at each moment Vehicle-borne LiDAR technology enables the real-time capture
of data acquisition can be determined accurately. The DG of the of high-resolution 3D spatial information, which is not possible
TLS data has been investigated primarily for terrestrial mobile with static terrestrial LiDAR scanning survey technology. This
mapping (Haala et al., 2008; Hunter et al., 2006; Talaya et al., approach is an important supplement to photogrammetry and
2004). In indirect georeferencing, a set of presurveyed GCPs are remote sensing.
required to transform one or multiple scans into a superior coor- Vehicle-borne and static terrestrial LiDAR data are different
dinate system. In this case, the 3D similarity transformation in three respects (Boulaassal etal., 2011). First, they are different
model is regularly utilized, and the transformation parameters in terms of the level of accuracy that can be obtained. Because
are estimated using least squares adjustment. the vehicle-borne LiDAR data require synchronization of the
positioning and orientation components, the resulting accuracy
17.8.1.2.2 PointCloudOrganizationandSegmentation is less than that of static LiDAR data. Secondly, because the sys-
TSL scans contain detailed geometric information but still tem operates on a moving platform, the density and resolution
require interpretation of the data to make it useable for map- of vehicle-borne LiDAR data are significantly lower than those
ping purposes. Point cloud organization and segmentation are seen with the stationary method, and so fewer details of objects
the early steps in LiDAR data processing, which are not directly are obtained. Moreover, the number of points acquired by vehi-
linked to an application (Pfeifer and Briese, 2007). cle-borne LiDAR systems is often higher than that acquired by
Modern static and kinematic TLS have the ability to acquire several successive stationary terrestrial LiDAR stations.
point cloud data with large amount of points. A well-organized Because vehicle-borne LiDAR systems are operated on roads
data structure for TLS point clouds will be helpful for the accel- when collecting data, they record numerous points of various
eration of the data storage, processing, and visualization. The objects on and nearby the road. This point data can then be used
commonly used data structures include TIN and grid and octree to reconstruct and model road surfaces, guardrails, pavements,
data structures (Elseberg etal., 2013; Wang and Tseng, 2011). utility facilities or polelike features (e.g., power poles, traffic sign
Segmentation and clustering can also be used to organize poles, and light poles), building faades, bushes, and trees, and
discrete points into homogeneous groups (Pfeifer and Briese, this is especially useful for obtaining details of objects in an
2007), and many algorithms have been proposed for extracting urban area. The features on the road surface, for example, road
planar surfaces from point clouds using segmentation methods markings, can also be extracted from these data.
for model reconstruction. Usually, one of three distinct methods Since vehicle-borne LiDAR systems move on a road, all the
is employed for segmenting points: region growing (Dold and objects around the road that appear in the field of view of the
Brenner, 2004; Hoffman et al., 2002; Pu and Vosselman, 2006), scanner can reflect the laser beam and generate points, although
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 369
this also means that the system has limited views in some direc- steps: (1) generation of ROIs in the image, (2) feature extraction
tions. Consequently, the collected data may often not fully cover within each ROI, and (3) feature-based classification.
the target objects of interest, making feature extraction and In addition, organizing LiDAR data using voxels before fur-
identification very difficult. Furthermore, when the vehicle is ther data processing is another good strategy. Aijazi etal. (2013)
moving at high speed during data collection, the point and scan presented a method to classify urban scenes based on a super-
line intervals may become rather large and make further data voxel segmentation of sparse 3D data obtained from LiDAR sen-
processing very complex. Certain strategies designed to handle sors. The 3D point cloud is first segmented into voxels, which
vehicle-borne LiDAR data thus need to be adopted to extract, are then characterized by several attributes transforming them
reconstruct, and model the objects along or nearby roads. into supervoxels. These are joined together by using a link-chain
First, LiDAR data should be organized in advance using aux- method rather than the usual region growing algorithm to
iliary data structures, such as scan lines, 2D grids, or 3D grids explore objects. These objects are then classified using geometri-
(voxels). The LiDAR data are then classified into point clusters of cal models and local descriptors. Schmitt and Vogtle (2009) con-
various objects, like road surfaces, building faces, and utility poles. verted a raw irregular point cloud into regular voxels and then
It is often useful to classify LiDAR data into road surface points extracted planar features extracted by merging adjacent voxels
and nonroad surface points based on the knowledge of the actual with collinear normal vectors. Wu etal. (2013) presented a new
scene. For example, the point height on road surfaces should usu- voxel-based marked neighborhood searching (VMNS) method
ally be lower than that on nonroad surfaces. Additionally, most for efficiently identifying street trees and deriving their mor-
of the points collected from the road surface will have the same phological parameters from vehicle-borne LiDAR point cloud
height, and the density of road points is higher than that seen in data. The VMNS method consists of six technical components:
other places due to the small range from the scanned road surface voxelization, calculating values of voxels, searching and mark-
to the scanner. The height of the road surface varies smoothly, with ing neighborhoods, extracting potential trees, deriving mor-
a very small deviation along the width of the road (Manandhar phological parameters, and eliminating polelike objects other
and Shibasaki, 2001). Based on the road surface points, the road than trees. The method was validated and evaluated through
surface can then be modeled by a TIN or road boundaries. two case studies, with the results showing that the completeness
Nonroad surface points may be located on the surfaces of and correctness of the proposed method for street tree detec-
polelike features (e.g., power poles, traffic sign poles, and light tion are both over 98%. The derived morphological parameters,
poles), building faades, or trees. Some important knowledge including tree height, crown diameter, diameter at breast height
about the objects in the scanning scene should thus be gener- (DBH), and crown base height, were also in a good agreement
alized in order to extract and model the objects from discrete with the field measurements. This method provides an effective
and incomplete LiDAR data in a complicated scanning environ- tool for extracting various morphological parameters for indi-
ment. In general, geometrical constraints for objects of interest vidual street trees from vehicle-borne LiDAR point cloud data.
are often employed. For example, the building faades should
be vertical to the ground or at least be planar. To avoid the 17.9 conclusions
interference from irrelevant points with regard to the extrac-
tion of object points for modeling, the algorithms developed for LiDAR is a promising approach for the fast and robust acqui-
extracting object points for object model reconstruction have to sition of 3D information from scanned surfaces and has been
be robust, and knowledge about the scanner can sometimes be widely used in various applications. For example, airborne
used to achieve this. For example, the scanning mechanism of LiDAR systems can obtain nationwide DSM and DEM data,
the scanner can be considered. If the scan line is perpendicular while ground-based and mobile LiDAR systems have been used
to the ground, the points on the scan lines can be segmented into more often in urban areas for 3D city modeling.
several vertical lines belonging to the walls. These vertical lines Similar to all other remote sensing methods, careful handling
can then be grouped into one plane belonging to the wall. and processing of LiDAR data are required to ensure high quality
Manandhar and Shibasaki (2001, 2002), Goulette etal. (2006), results. The standard procedures provided in this chapter should be
and Li et al. (2004) discussed the aforementioned concepts suitable for general applications. However, for more specific appli-
with regard to feature extraction, reconstruction, and model- cations, for example, very-high-density point clouds and DEM
ing. More specifically, Manandhar and Shibasaki (2001, 2002) data for a very dense forest, it is recommended that consultations
and Goulette etal. (2006) organized LiDAR data by scan lines with professional service providers are carried out ahead of time, as
before processing, while Li etal. (2004) organized LiDAR data some technical issues may arise that are not considered here.
into 2D grids before processing it. In addition to organizing
LiDAR data with scan lines or 2D grids, additional informa- References
tion can be attributed into 2D grids for advanced classification.
Douillard et al. (2009) developed a cell-wise semantic classifi- Aijazi, A.K., Checchin, P., and Trassoudaine, L., 2013.
cation approach for ground cells. This involves the use of color Segmentation based classification of 3D urban point clouds:
imagery to classify cells into one of two classes: asphalt or grass. A super-voxel based approach with evaluation. Remote
Classification of ground cells proceeds according to three main Sensing, 5(4): 16241650.
2016 by Taylor & Francis Group, LLC
370 Remotely Sensed Data Characterization, Classification, and Accuracies
Alexander, C., Tansey, K., Kaduk, J., Holland, D., and Tate, N.J., Bucksch, A., Lindenbergh, R., and Menenti, M., 2009. SkelTre-Fast
2010. Backscatter coefficient as an attribute for the classi- Skeletonisation for Imperfect Point Cloud Data of Botanic
fication of full-waveform airborne laser scanning data in Trees, Eurographics Workshop on 3D Object Retrieval.
urban areas. ISPRS Journal of Photogrammetry and Remote Eurographics, Mnchen, Germany, p. 8.
Sensing, 65(5): 423432. Buddenbaum, H., Seeling, S., and Hill, J., 2013. Fusion of full-
ASPRS, 2013. LASer (LAS) file format exchange activities, waveform LiDAR and imaging spectroscopy remote sensing
http://www.asprs.org/Committee-General/LASer-LAS- data for the characterization of forest stands. International
File-Format-Exchange-Activities.html (last date accessed: Journal of Remote Sensing, 34(13): 45114524.
November 11, 2013). Burman, H., 2002. Laser strip adjustment for data calibration and
Axelsson, P., 2000. DEM generation from laser scanner data verification. International Archives of Photogrammetry and
using adaptive TIN models. International Archives of Remote Sensing, 34(Part 3A/B): 6772.
Photogrammetry and Remote Sensing, 33(Part B4): 110117, Chen, L., Teo, T., Kuo, C., and Rau, J., 2008. Shaping polyhedral
Amsterdam, the Netherlands. buildings by the fusion of vector maps and LiDAR point
Baltsavias, E.P., 1999a. Airborne laser scanning: Basic relations clouds. Photogrammetric Engineering and Remote Sensing,
and formulas. ISPRS Journal of Photogrammetry and Remote 74(5): 11471157.
Sensing, 54(23): 199214. Chen, L.-C. and Lo, C.-Y., 2009. 3D road modeling via the inte-
Baltsavias, E.P., 1999b. Airborne laser scanning: Existing sys- gration of large-scale topomaps and airborne LIDAR data.
tems and firms and other resources. ISPRS Journal of Journal of the Chinese Institute of Engineers, 32(6): 811823.
Photogrammetry and Remote Sensing, 54(23): 164198. Chen, L.C., Teo, T.A., Hsieh, C.H., and Rau, J.Y., 2006.
Bauer, J., Karner, K., Klaus, A., Zach, C., and Schindler, K., 2003. Reconstruction of building models with curvilinear bound-
Segmentation of building models from dense 3D point- aries from laser scanner and aerial Imagery. Lecture Notes in
clouds. 27th Workshop of the Ausman Association for Computer Science, 4319: 2433.
Pattern Recognition, 253258. Chen, Q., Gong, P., Baldocchi, D., and Xie, G., 2007. Filtering
Besl, P.J. and McKay, N.D., 1992. A method for registration of Airborne Laser Scanning Data with Morphological
3-D shapes. IEEE Transactions on Pattern Analysis and Methods. Photogrammetric Engineering and Remote
Machine Intelligence, 14(2): 239256. Sensing, 73(2): 175185.
Biosca, J.M. and Lerma, J.L., 2008. Unsupervised robust pla- Chen, Y. and Medioni, G., 1992. Object modelling by registra-
nar segmentation of terrestrial laser scanner point clouds tion of multiple range images. Image and Vision Computing,
based on fuzzy clustering methods. ISPRS Journal of 10(3): 145155.
Photogrammetry and Remote Sensing, 63(1): 8498. Cheng, Y-H and F. Tsai, 2014. Spline curve fitting of full-wave-
Biron, P.M., Chon, G., BuffinBlanger, T., Demers, S., and form Lidar data and feature extraction for land-cover classi-
Olsen, T., 2013. Improvement of streams hydrogeomor- fication, in Proc. 2014 International Symposium on Remote
phological assessment using LiDAR DEMs. Earth Surface Sensing, Apr. 1618, 2014, Busan, Korea.
Processes and Landforms, 38(15): 18081821. Chiu, C.M., Teo, T.A., and Chen, C.T., 2013. Three-dimensional
Boulaassal, H., Landes, T., and Grussenmeyer, P., 2011. 3D modelling of multilayer road networks using road cen-
modelling of facade features on large sites acquired by terlines and airborne LiDAR data. The International
vehicle based laser scanning. Archives of Photogrammetry, Symposium on Mobile Mapping Technology 2013, May 13,
Cartography and Remote Sensing edited by Polish Society for Tainan, Taiwan, pp. CD-ROM.
Photogrammetry and Remote Sensing, 22: 215226. Chuang, T.Y., 2012. Feature-Based Registration of LiDAR Point
Boulaassal, H., Landes, T., Grussenmeyer, P., and Tarsha-Kurdi, Clouds, National Taiwan University, Taipei, Taiwan.
F., 2007. Automatic segmentation of building facades Clark, M.L., Roberts, D.A., Ewel, J.J., and Clark, D.B., 2011.
using terrestrial laser data. International Archives of Estimation of tropical rain forest aboveground biomass
Photogrammetry, Remote Sensing and Spatial Information with small-footprint LiDAR and hyperspectral sensors.
Sciences, 36(Part 3) : 6570. Remote Sensing of Environment, 115(11): 29312942.
Brenner, C., 2005. Building reconstruction from images and laser Csanyi, N. and Toth, C.K., 2007. Improvement of LIDAR data accu-
scanning. International Journal of Applied Earth Observation racy using LIDAR specific ground targets. Photogrammetric
and Geoinformation, 6(34): 187198. Engineering and Remote Sensing, 73(4): 385396.
Brenner, M., Wichmann, V., and Rutzinger, M., 2013. Eigenvalue and Dold, C. and Brenner, C., 2004. Automatic matching of terres-
Graph-based object extraction from mobile laser scanning trial scan data as a basis for the generation of detailed 3D
point clouds, ISPRS Annals of the Photogrammetry, Remote city models. International Archives of Photogrammetry,
Sensing and Spatial Information Sciences, II-5(W2): 5560. Remote Sensing and Spatial Information Sciences, 35(B3):
Bretar, F. and Roux, M., 2005. Hybrid image segmentation using 10911096.
LiDAR 3D planar primitives. International Archives of dos Santos, D.R., Dal Poz, A.P., and Khoshelham, K., 2013. Indirect
Photogrammetry, Remote Sensing and Spatial Information georeferencing of terrestrial laser scanning data using con-
Sciences, 16: 7278. trol lines. The Photogrammetric Record, 28(143): 276292.
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 371
Douillard, B., Brooks, A., Ramos, F., and Durrant-Whyte, H., Goulette, F., Nashashibi, F., Abuhadrous, I., Ammoun, S., and
2009. Combining laser and vision for 3D urban classifica- Laurgeau, C., 2006. An integrated on-board laser range
tion. Proceedings of Neural Information Processing Systems sensing system for on-the-way city and road modelling.
Conference (NIPS), Vancouver, B.C., Canada. Proceedings of the ISPRS Commission I Symposium, From
Ducic, V., Hollaus, M., Ullrich, A., Wagner, W., and Melzer, T., Sensors to Imagery, Paris, France, Vol. 3, p. 43.
2006. 3D vegetation mapping and classification using full- Gross, H. and Thoennessen, U., 2006. Extraction of lines from laser
waveform laser scanning. Workshop on 3D Remote Sensing point clouds. International Archives of Photogrammetry, Remote
in Forestry, Vienna, Austria, pp.211217. Sensing and Spatial Information Sciences, Vol. 36, pp. 8691.
Elseberg, J., Borrmann, D., and Nchter, A., 2013. One billion Gruen, A., 2005. Towards photogrammetry 2025. Photogrammetric
points in the cloudAn octree for efficient processing Week, Stuttgart, Germany.
of 3D laser scans. ISPRS Journal of Photogrammetry and Gruen, A. and Akca, D., 2005. Least squares 3D surface and curve
Remote Sensing, 76: 7688. matching. ISPRS Journal of Photogrammetry and Remote
Filin, S., 2001. Calibration of airborne and spaceborne laser Sensing, 59(3): 151174.
altimeters using natural surfaces. PhD thesis, The Ohio Guo, L., Chehata, N., Mallet, C., and Boukir, S., 2011. Relevance
State University, Columbus, OH, 140pp. of airborne LiDAR and multispectral image data for urban
Filin, S., 2002. Surface clustering from airborne laser scanning scene classification using Random Forests. ISPRS Journal of
data. International Archives of Photogrammetry and Remote Photogrammetry and Remote Sensing, 66(1): 5666.
Sensing, 34(Part 3A/B): 119124. Haala, N., Peter, M., Kremer, J., and Hunter, G., 2008. Mobile LiDAR
Filin, S., 2003. Recovery of systematic biases in laser altim- mapping for 3D point cloud collection in urban areasA per-
eters using natural surfaces. International Archives of formance test. The International Archives of the Photogrammetry,
Photogrammetry and Remote Sensing, 35(WG III/3), Graz, Remote Sensing and Spatial Information Sciences, 37: 11191127.
Austria, 6570. Habib, A., Al-Durgham, M., Kersting, A., and Quackenbush, P.,
Filin, S., 2004. Surface classification from airborne laser scanning 2008. Error budget of LiDAR systems and quality control
data. Computers & Geosciences, 30: 10331041. of the derived point cloud. Proceedings of the XXI ISPRS
Filin, S. and Pfeifer, N., 2006. Segmentation of airborne laser Congress, Commission I, Vol. 37, Beijing, China, pp. 203209.
scanning data using a slope adaptive neighborhood. ISPRS Habib, A., Bang, K., Kersting, A.P., and Lee, D.-C., 2009. Error
Journal of Photogrammetry and Remote Sensing, 60(2): budget of LiDAR systems and quality control of the derived
7180. data. Photogrammetric Engineering and Remote Sensing,
Filin, S. and Vosselman, G., 2004. Adjustment of airborne laser 75(9): 10931108.
altimetry strips. International Archives of Photogrammetry Habib, A., Ghanma, M., Morgan, M., and Al-Ruzouq, R., 2005.
and Remote Sensing, 35(B3): 285289. Photogrammetric and LiDAR data registration using linear
Fischler, M.A. and Bolles, R.C., 1981. Random sample consensus: features. Photogrammetric Engineering and Remote Sensing,
A paradigm for model fitting with applications to image 71(6): 699707.
analysis and automated cartography. Communications of the Habib, A., Kersting, A.P., Bang, K.I., and Lee, D.-C., 2010.
ACM, 24(6): 381395. Alternative methodologies for the internal quality control
Florinsky, I.V., 1998. Combined analysis of digital terrain mod- of parallel LiDAR strips. IEEE Transactions on Geoscience
els and remotely sensed data in landscape investigations. and Remote Sensing, 48(1): 221236.
Progress in Physical Geography, 22(1): 3360. Hancock, S., Disney, M., Muller, J.P., Lewis, P., and Foster, M.,
Fornaciai, A., Bisson, M., Landi, P., Mazzarini, F., and Pareschi, 2011. A threshold insensitive method for locating the for-
M.T., 2010. A LiDAR survey of Stromboli volcano (Italy): est canopy top with waveform LiDAR. Remote Sensing of
Digital elevation model-based geomorphology and inten- Environment, 115(12): 32863297.
sity analysis. International Journal of Remote Sensing, Harding, D., 2009. Pulsed laser altimeter ranging techniques and
31(12): 31773194. implications for terrain mapping. In: J. Shan and C.K. Toth
Gardner, C.S., 1992. Ranging performance of satellite laser altim- (eds.), Topographic Laser Ranging and Scanning Principles
eters. IEEE Transactions on Geoscience and Remote Sensing, and Processing, CRC Press, Boca Raton, FL, pp. 173194.
30(5): 10611072. Heidemann, H.K., 2012. Lidar Base Specification Version 1.0.
Glennie, C., 2007. Rigorous 3D error analysis of kinematic scan- US Geological Survey Techniques and Methods, U.S.
ning LIDAR systems. Journal of Applied Geodesy, 1(3): Geological Survey, Reston, VA, 63pp.
147157. Heinzel, J. and Koch, B., 2011. Exploring full-waveform LiDAR
Gonzalez, R.C. and Woods, R.E., 1992. Digital Image Processing, parameters for tree species classification. International Journal of
Addison-Wisley, Reading, MA, 716pp. Applied Earth Observation and Geoinformation, 13(1): 152160.
Gorte, B. and Pfeifer, N., 2004. Structuring laser-scanned Hoffman, A.D., 2004. Analysis of TIN-structure parameter spaces in
trees using 3D mathematical morphology. International airborne laser scanner data for 3-D building model generation.
Archives of Photogrammetry and Remote Sensing, 35(B5): International Archives of Photogrammetry, Remote Sensing
929933. andSpatial Information Sciences, 35(Part B3): 302307.
2016 by Taylor & Francis Group, LLC
372 Remotely Sensed Data Characterization, Classification, and Accuracies
Hoffman, A.D., Maas, H.G., and Streilein, A., 2002. Knowledge- Kreylos, O., Bawden, G., and Kellogg, L., 2008. Immersive visu-
based building detection based on laser scanner data and alization and analysis of LiDAR data. Advances in Visual
topographic map information. International Archives Computing, 5358: 846855.
of Photogrammetry and Remote Sensing, 34(Part 3A/B) Latypov, D. and Zosse, E., 2002. LIDAR data quality control and
September 913, Graz, Austria, pp. 169174. system calibration using overlapping flight lines in commer-
Hofton, M.A., Minster, J.B., and Blair, J.B., 2000. Decomposition of cial environment. ACSM-ASPRS 2002 Annual Conference,
laser altimeter waveforms. IEEE Transactions on Geoscience April 2226, Washington, DC.
and Remote Sensing, 38(4): 19891996. Lerma, J. and Biosca, J., 2005. Segmentation and filtering of
Hoover, A. et al., 1996. An experimental comparison of range laser scanner data for cultural heritage. CIPA 2005 XX
image segmentation algorithms. IEEE Transactions on International Symposium, Torino, Italy.
Pattern Analysis and Machine Intelligence, 18(7): 673689. Levanon, N. and Mozeson, E., 2004. Radar Signals, John Wiley &
Hu, Y., 2003. Automated extraction of digital terrain models, Sons, Hoboken, NJ.
roads and buildings using airborne LiDAR data. PhD the- Lewis, P., McElhinney, C.P., and McCarthy, T., 2012. LiDAR
sis, University of Calgary, Calgary, Alberta, Canada, 223pp. data management pipeline; from spatial database popula-
Hug, C., Ullrich, A., and Grimm, A., 2004. Litemapper-5600A tion to web-application visualization. Proceedings of the
waveform-digitizing LiDAR terrain and vegetation mapping Third International Conference on Computing for Geospatial
system. International Archives of Photogrammetry, Remote Research and Applications, ACM, New York, NY, p. 16.
Sensing and Spatial Information Sciences, 36(Part 8): W2. Li, B., Li, Q., Shi, W., and Wu, F., 2004. Feature extraction and
Huising, E.J. and Pereira, L.M.G., 1998. Errors and accuracy modeling of urban building from vehicle-borne laser scan-
estimates of laser data acquired by various laser scanning ning data. International Archives of Photogrammetry, Remote
systems for topographic application. ISPRS Journal of Sensing and Spatial Information Sciences, 35: 934939.
Photogrammetry and Remote Sensing, 53(5): 245261. Lichti, D.D. and Licht, M.G., 2006. Experiences with terres-
Hunter, G., Cox, C., and Kremer, J., 2006. Development of a trial laser scanner modelling and accuracy assessment.
commercial laser scanning mobile mapping system International Archives of the Photogrammetry, Remote
StreetMapper. International Archives of the Photogrammetry, Sensing and Spatial Information Sciences, 36(5): 155160.
Remote Sensing and Spatial Information Sciences, 36. Lim, M. et al., 2005. Combined digital photogrammetry and
Im, J., Jensen, J.R., and Hodgson, M.E., 2008. Object-based land timeofflight laser scanning for monitoring cliff evolution.
cover classification using high-posting-density LiDAR data. The Photogrammetric Record, 20(110): 109129.
GIScience & Remote Sensing, 45(2): 209228. Lu, Y.-H. and Tsai, F., 2013. Analysis of full-waveform LiDAR
Isenburg, M., 2013. LASzip: Lossless compression of LiDAR data. data for land-cover classification. 2013 International
Photogrammetric Engineering and Remote Sensing, 79(2): Symposium on Remote Sensing May 1517, Chiba, Japan,
209217. pp. CDROM.
Jiang, X. and Bunke, H., 1994. Fast segmentation of range images Mallet, C. and Bretar, F., 2009. Full-waveform topographic
into planar regions by scan line grouping. Machine Vision LiDAR: State-of-the-art. ISPRS Journal of Photogrammetry
and Applications, 7(2): 115122. and Remote Sensing, 64(1): 116.
Jutzi, B., Neulist, J., and Stilla, U., 2005. High-resolution wave- Manandhar, D. and Shibasaki, R., 2001. Feature extraction from
form acquisition and analysis of laser pulses. Measurement range data. Paper presented at the 22nd Asian Conference on
Techniques, 2(3.1): 2. Remote Sensing, Vol. 5, Singapore, p. 9.
Jutzi, B. and Stilla, U., 2005. Measuring and processing the wave- Manandhar, D. and Shibasaki, R., 2002. Auto-extraction of
form of laser pulses. In: Proc. 7th Optical 3-D Measurement urban features from vehicle-borne laser data. International
Techniques. FIG/IAG/ISPRS, Vienna, Austria, 35 October Archives of Photogrammetry Remote Sensing And Spatial
2005, pp. 194203. Information Sciences, 34(4): 650655.
Jutzi, B. and Stilla, U., 2006. Range determination with waveform Mazzarini, F. et al., 2007. Lava flow identification and aging by
recording laser systems using a Wiener Filter. ISPRS Journal means of LiDAR intensity: Mount Etna case. Journal of
of Photogrammetry and Remote Sensing, 61(2): 95107. Geophysical Research: Solid Earth (19782012), 112(B2).
Kager, H., 2004. Discrepancies between overlapping laser scanner McCullagh, M., 1988. Terrain and surface modelling systems: Theory
stripsSimultaneous fitting of aerial laser scanner strips. and practice. The Photogrammetric Record, 12(72): 747779.
International Archives of Photogrammetry, Remote Sensing Melzer, T., 2007. Non-parametric segmentation of ALS point clouds
and Spatial Information Sciences, 35(B1): 555560. using mean shift. Journal of Applied Geodesy, 1(3): 159170.
Krabill, W. et al., 2000. Airborne laser II.Assateague National Meng, X., Currit, N., and Zhao, K., 2010. Ground filtering algo-
Seashore Beach. Photogrammetric Engineering & Remote rithms for airborne LiDAR data: A review of critical issues.
Sensing, 66(1): 6571. Remote Sensing, 2(3): 833860.
Kraus, K. and Pfeifer, N., 1998. Determination of terrain models in Mongus, D. and alik, B., 2012. Parameter-free ground filtering of
wooded areas with airborne laser scanner data. ISPRS Journal LiDAR data for automatic DTM generation. ISPRS Journal
of Photogrammetry and Remote Sensing, 53(4): 193203. of Photogrammetry and Remote Sensing, 67: 112.
2016 by Taylor & Francis Group, LLC
LiDAR Data Processing and Applications 373
Monserrat, O. and Crosetto, M., 2008. Deformation measurement Pu, S. and Vosselman, G., 2006. Automatic extraction of build-
using terrestrial laser scanning data and least squares 3D ing features from terrestrial laser scanning. International
surface matching. ISPRS Journal of Photogrammetry and Archives of Photogrammetry, Remote Sensing and Spatial
Remote Sensing, 63(1): 142154. Information Sciences, 36(Part 5): 5pp. (on CD-ROM).
Morin, K.W., 2002. Calibration of airborne laser scanners. MSc the- Pu, S. and Vosselman, G., 2009a. Building facade reconstruction
sis, University of Calgary, Calgary, Alberta, Canada, 134pp. by fusing terrestrial laser points and images. Sensors, 9(6):
Morsdorf, F., Nichol, C., Malthus, T., and Woodhouse, I.H., 2009. 45254542.
Assessing forest structural and physiological information Pu, S. and Vosselman, G., 2009b. Knowledge based reconstruc-
content of multi-spectral LiDAR waveforms by radia- tion of building models from terrestrial laser scanning
tive transfer modelling. Remote Sensing of Environment, data. ISPRS Journal of Photogrammetry and Remote Sensing,
113(10): 21522163. 64(6): 575584.
Mcke, W., 2008. Analysis of full-waveform airborne laser scan- Rabbani, T., 2006. Automatic reconstruction of industrial
ning data for the improvement of DTM generation. MSc installations using point clouds and images. PhD thesis,
thesis, Institute of Photogrammetry and Remote Sensing, Delft University of Technology, Delft, the Netherlands,
Vienna University of Technology, Vienna, Austria. 175pp.
Neuenschwander, A.L., Magruder, L.A., and Tyler, M., 2009. Rabbani, T., van den Heuvel, F.A., and Vosselman, G., 2006.
Landcover classification of small-footprint, full-waveform Segmentation of point clouds using smoothness constraint.
LiDAR data. Journal of Applied Remote Sensing, 3(1). International Archives of Photogrammetry, Remote Sensing
Norini, G., Groppelli, G., Capra, L., and De Beni, E., 2004. and Spatial Information Sciences, 36(5): 24853.
Morphological analysis of Nevado de Toluca volcano (Mexico): Reshetyuk, Y., 2006. Calibration of terrestrial laser scanners
New insights into the structure and evolution of an andesitic to Callidus 1.1, Leica HDS 3000 and Leica HDS 2500. Survey
dacitic stratovolcano. Geomorphology, 62(1): 4761. Review, 38(302): 703713.
OGC, 2012. OGC City Geography Markup Language (CityGML) Roncat, A., Bergauer, G. and Pfeifer, N., 2011. B-spline decon-
Encoding Standard, Version 2.0. volution for differential target cross-section determi-
Okatani, I.S. and Deguchi, K., 2002. A method for fine registra- nation in full-waveform laser scanning data. ISPRS
tion of multiple view range images considering the mea- Journal of Photogrammetry and Remote Sensing, 66(4):
surement error properties. Computer Vision and Image 418428.
Understanding, 87(13): 6677. Rottensteiner, F., Trinder, J., Clode, S., and Kubik, K., 2005. Using
Oude Elberink, S.J. and Vosselman, G., 2009. 3D information the DempsterShafer Method for the fusion of LIDAR
extraction from laser point clouds covering complex road data and multi-spectral images for building detection.
junctions. The Photogrammetric Record, 24(125): 2336. Information Fusion, 6(4): 283300.
Persson, ., Sderman, U., Tpel, J., and Ahlberg, S., 2005. Rutzinger, M., Elberink, S.O., Pu, S., and Vosselman, G., 2009.
Visualization and analysis of full-waveform airborne laser scan- Automatic extraction of vertical walls from mobile and
ner data. International Archives of Photogrammetry, Remote airborne laser scanning data. The International Archives of
Sensing and Spatial Information Sciences, 36(3/W19): 103108. Photogrammetry, Remote Sensing and Spatial Information
Petrie, G. and Toth, C.K., 2008a. Airborne and spaceborne laser Sciences, 38(Part 3): W8.
profilers and scanners. In: J. Shan and C.K. Toth (eds.), Samet, H., 1990. Applications of Spatial Data Structures: Computer
Topographic Laser Ranging and Scanning Principles and Graphics, Image Processing, and GIS, Addison-Wesley,
Processing, CRC Press, Boca Raton, FL, pp. 2985. Reading, MA, 507pp.
Petrie, G. and Toth, C.K., 2008b. Terrestrial laser scanners. In: Sampath, A. and Shan, J., 2007. Building boundary tracing
J.Shan and C.K. Toth (eds.), Topographic Laser Ranging and and regularization from airborne LiDAR point clouds.
Scanning Principles and Processing, CRC Press, Boca Raton, Photogrammetry Engineering & Remote Sensing, 73(7):
FL, pp. 87127. 805812.
Pfeifer, N. and Briese, C., 2007. Geometrical aspects of airborne Santibanez, S.F., dos Santos, D.R., and Faggion, P.L., 2012.
laser scanning and terrestrial laser scanning. International Influence of fitting models and point density sample in the
Archives of Photogrammetry, Remote Sensing and Spatial detection of deformations of structures using terrestrial
Information Sciences, 36(3/W52): 311319. laser scanning. Applied Geomatics, 4(1): 1119.
Pfeifer, N. and Mandlburger, G., 2008. LiDAR data filtering Schenk, T., 2001. Modeling and analyzing systematic errors in air-
and DTM generation. In: J. Shan and C.K. Toth (eds.), borne laser scanners. Technical Report. Department of Civil
Topographic Laser Ranging and Scanning Principles and and Environmental Engineering and Geodetic Science, The
Processing, CRC Press, Boca Raton, FL, pp. 307333. Ohio State University, OH.
Pingel, T.J., Clarke, K.C., and McBride, W.A., 2013. An improved Schmitt, A. and Vogtle, T., 2009. An advanced approach for
simple morphological filter for the terrain classification of automatic extraction of planar surfaces and their topol-
airborne LIDAR data. ISPRS Journal of Photogrammetry ogy from point clouds. Photogrammetrie-Fernerkundung-
and Remote Sensing, 77: 2130. Geoinformation, 2009(1): 4352.
2016 by Taylor & Francis Group, LLC
374 Remotely Sensed Data Characterization, Classification, and Accuracies
Schuster, H.-F., 2004. Segmentation of LIDAR data using the tensor Wagner, W., Ullrich, A., Ducic, V., Melzer, T., and Studnicka, N.,
voting framework. International Archives of Photogrammetry, 2006. Gaussian decomposition and calibration of a novel
Remote Sensing and Spatial Information Sciences, 35: 10731078. small-footprint full-waveform digitising airborne laser
Segal, A., Haehnel, D., and Thrun, S., 2009. Generalized-ICP. scanner. ISPRS Journal of Photogrammetry and Remote
Robotics: Science and Systems, 2: 4. Sensing, 60(2): 100112.
Shakarji, C.M., 1998. Least-squares fitting algorithms of the NIST Wagner, W., Ullrich, A., Melzer, T., Briese, C., and Kraus, K., 2004.
algorithm testing system. Journal of Research of National From single-pulse to full-waveform airborne laser scanners:
Institute of Standards and Technology, 103: 633641. Potential and practical challenges. International Archives of
Sithole, G., 2005. Segmentation and classification of airborne laser Photogrammetry and Remote Sensing, 35(B3): 201206.
scanner data. PhD thesis, Delft University of Technology, Wang, C. et al., 2013. Wavelet analysis for ICESat/GLAS wave-
Delft, the Netherlands, 203pp. form decomposition and its application in average tree
Skaloud, J. and Lichti, D., 2006. Rigorous approach to bore-sight height estimation. IEEE Geoscience and Remote Sensing
self-calibration in airborne laser scanning. ISPRS Journal of Letters, 10(1): 115119.
Photogrammetry and Remote Sensing, 61(1): 4759. Wang, M. and Tseng, Y.-H., 2010. Automatic segmentation of
Staiger, R., 2003. Terrestrial laser scanning technology, sys- LiDAR data into coplanar point clusters using an octree-
tems and applications. Second FIG Regional Conference based split-and-merge algorithm. Photogrammetric
Marrakech, Marrakech, Morocco, p. 1. Engineering and Remote Sensing, 76(4): 407420.
Stilla, U. and Jutzi, B., 2008. Waveform analysis for small- Wang, M. and Tseng, Y.-H., 2011. Incremental segmentation of
footprint pulsed laser systems. In: J. Shan and C.K. Toth LiDAR point clouds with an octree-structured voxel space.
(eds.), Topographic Laser Ranging and Scanning Principles The Photogrammetric Record, 26(133): 3257.
and Processing, CRC Press, Boca Raton, FL, pp. 215234. Weingarten, J.W., Gruener, G., and Siegwart, R., 2004. Probabilistic
Talaya, J. et al., 2004. Integration of a terrestrial laser scanner plane fitting in 3D and an application to robotic mapping.
with GPS/IMU orientation sensors. Proceedings of the XXth Proceedings ICRA04. 2004 IEEE International Conference on
ISPRS Congress, Vol. 35, Istanbul, Turkey, pp. 10491055. Robotics and Automation, Vol. 1, pp. 927932.
Tao, C.V. and Li, J., 2007. Advances in Mobile Mapping Technology, Wu, B. etal., 2013. A voxel-based method for automated identi-
ISPRS Series, Vol. 4, Taylor & Francis: London, UK. fication and morphological parameters estimation of indi-
Tsai, F. and Philpot, W., 1998. Derivative analysis of hyperspectral vidual street trees from mobile laser scanning data. Remote
data. Remote Sensing of Environment, 66(1): 4151. Sensing, 5(2): 584611.
Tsai, F. and Philpot, W.D., 2002. A derivative-aided hyperspectral Wu, J.Y., van Aardt, J.A.N., and Asner, G.P., 2011. A compari-
image analysis system for land-cover classification. IEEE son of signal deconvolution algorithms based on small-
Transactions on Geoscience and Remote Sensing, 40(2): 416425. footprint LiDAR waveform simulation. IEEE Transactions
Vaze, J., Teng, J., and Spencer, G., 2010. Impact of DEM accu- on Geoscience and Remote Sensing, 49(6): 24022414.
racy and resolution on topographic indices. Environmental Yang, M.-D. et al., 2011. Landslide-induced levee failure by high
Modelling & Software, 25(10): 10861098. concentrated sediment flowA case of Shan-An levee at
Vosselman, G., 2009. Advanced point cloud processing. Chenyulan River, Taiwan. Engineering Geology, 123(1):
Photogrammetric Week, Enschede, the Netherlands, 9, 9199.
pp.137146. Yao, T. et al., 2011. Measuring forest structure and biomass
Vosselman, G., Gorte, B.G.H., Sithole, G., and Rabbani, T., in New England forest stands using Echidna ground-
2004. Recognising structure in laser scanner point clouds. based LiDAR. Remote Sensing of Environment, 115(11):
International Archives of Photogrammetry and Remote 29652974.
Sensing, 36(Part 8/W2): 3338. Zhang, K. and Whitman, D., 2005. Comparison of three algo-
Vosselman, G. and Dijkman, S., 2001. 3D building model recon- rithms filtering airborne LiDAR data. Photogrammetric
struction from point clouds and ground plans. International Engineering and Remote Sensing, 71(3): 313324.
Archives of Photogrammetry and Remote Sensing, 34(3/W4): Zhang, Z., 1994. Iterative point matching for registration of free-
3743, Annapolis, MD. form curves and surfaces. International Journal of Computer
Wagner, W., 2010. Radiometric calibration of small-footprint Vision, 13(2): 119152.
full-waveform airborne laser scanner measurements: Basic Zhao, T. and Wang, J., 2014. Use of LiDAR-derived NDTI and
physical concepts. ISPRS Journal of Photogrammetry and intensity for rule-based object-oriented extraction of build-
Remote Sensing, 65(6): 505513. ing footprints. International Journal of Remote Sensing, 35(2):
Wagner, W., Hollaus, M., Briese, C., and Ducic, V., 2008. 3D 578597.
vegetation mapping using smallfootprint fullwaveform Zwally, H. etal., 2002. ICESats laser measurements of polar ice,
airborne laser scanners. International Journal of Remote atmosphere, ocean, and land. Journal of Geodynamics,
Sensing, 29(5): 14331452. 34(3): 405445.
2016 by Taylor & Francis Group, LLC
VI
Change Detection
18 Change Detection Techniques u sing Multisensor Data Daniela Anjos, Dengsheng Lu, Luciano Dutra, and
Sidnei SantAnna ..........................................................................................................................................................................377
Acronyms and Definitions Introduction Brief Description of Key Steps Used in a Change
Detection Procedure Case Study of Land-Cover Change Detection Using Optical and Radar
Data Conclusions Acknowledgments References
375
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
18
Change Detection Techniques
Using Multisensor Data
Daniela Anjos Acronyms and Definitions ................................................................................................................ 377
National Institute for 18.1 Introduction ............................................................................................................................ 377
Space Research
18.2 Brief Description of Key Steps Used in a Change Detection Procedure .........................378
Dengsheng Lu Selection of Suitable Remote Sensing Variables Selection of Suitable Algorithms
Michigan State University for Change Detection Impacts of Scale Issues on the Design of a Change Detection
Procedure Challenge of Using Multisensor Data for Change Detection
Luciano Dutra 18.3 Case Study of Land-Cover Change Detection Using Optical and Radar Data............. 384
National Institute for Study Area Data Collection and Preprocessing Methodology Results Summary of
Space Research the Case Study
Sidnei SantAnna 18.4 Conclusions.............................................................................................................................. 394
National Institute for Acknowledgments .............................................................................................................................. 394
Space Research References............................................................................................................................................ 394
Acronyms and Definitions SS1 Initial secondary succession
SS1BS Initial secondary succession to bare soil
AG Agriculture SS1SS2 Initial secondary succession to intermediate sec-
AGSS1 Agriculture to initial secondary succession ondary succession
BS Bare soil SS2 Intermediate secondary succession
BSAG Bare soil to agriculture SVM Support vector machine
CP Clean pasture TM Thematic Mapper
CPDP Clean pasture to dirty pasture UTM Universal Transverse Mercator
CPSS1 Clean pasture to initial secondary succession VH Cross polarization, transmit vertical signal and
DP Dirty pasture receive horizontal signal
DPCP Dirty pasture to clean pasture VV Single polarization, transmit and receive vertical
DPSS1 Dirty pasture to initial secondary succession signal
EO-1 ALI Earth Observing-1 Advanced Land Imager WT Water
instrument
ETM+ Enhanced Thematic Mapper Plus
FQ13 Fine-resolution quad-polarization beam mode 13 18.1 introduction
HH Single polarization, transmit and receive hori-
zontal signal Land-cover change has long been an active research topic because
HV Cross polarization, transmit horizontal signal of its impacts on climate and environmental changes, as well
and receive vertical signal as its close interactions with human activities (Lu et al., 2014;
IS Impervious surface Volpi et al., 2013). Early publications related to change detec-
MODIS Moderate Resolution Imaging Spectroradiometer tion occurred in the 1960s. Lillestrand (1972) and Singh (1989)
MSS Multispectral scanner overviewed the major change detection techniques developed
PF Primary forest between the 1960s and 1980s. Since then, change detection tech-
SPOT HRG Satellite Pour lObservation de la Terrehigh niques have been applied to different fields such as forest change
resolution geometrical (Collins and Woodcock, 1996; Descle etal., 2006; Huang etal.,
SPOT XS Satellite Pour lObservation de la Terre 2010; Reiche etal., 2013), urban expansion (Borghys etal., 2007;
multispectral Du etal., 2012; Lo and Shipman, 1990; Yeh and Li, 2001), wetland
377
2016 by Taylor & Francis Group, LLC
378 Remotely Sensed Data Characterization, Classification, and Accuracies
change (Baker etal., 2007; Houhoulis and Michener, 2000), and land-cover change detection provides a new but also challenging
even biomass change (Coppin et al., 2001). A large number of opportunity (Lu etal., 2014). This chapter aims to provide brief
change detection techniques have been developed and are sum- descriptions of key steps used in a change detection procedure,
marized in the literature (e.g., Bhagat, 2012; Chen et al., 2012; to discuss the challenges of using multisensor data for land-cover
Coppin etal., 2004; Hussain etal., 2013; Lu etal., 2004, 2014). change detection, and to provide a case study to explore the use
Change detection is defined as the process of identifying of multisensor data for land-use/land-cover change detection in
changes in the state of an object or phenomenon at different the Brazilian Amazon.
times (Lu et al., 2004; Singh, 1989). This process is critical to
understand the relationships and interactions between human
and natural phenomena to promote better decision making in 18.2 Brief Description of Key Steps Used
various situations. Identification and analysis of changes occur- in a change Detection Procedure
ring in a particular region during a time interval require use of
multitemporal datasets. Because of the unique characteristics of The process for conducting land-cover change detection con-
remote sensing data in data collection and presentation, it has sists of a chain of the following steps: (1) description of research
become the main data source for detecting land-cover change problems and objectives, (2) determination of the study area and
(Chen et al., 2012). The basic premise of using remote sensing datasets to be used, (3) preprocessing of multitemporal remote
data is that changes in land covers are expressed in changes of sensing data, (4) extraction of suitable variables and selection of
the responses obtained by passive or active sensors (Mas, 1999). proper algorithms for implementing change detection, and (5)
However, these changes must be separated from changes in evaluation of the change detection results (see Figure 18.1). Lu
spectral responses caused by other factors such as weather, view et al. (2014) provide a brief description of each step used in a
angle, and moisture content at the time of image acquisition change detection procedure. The following subsections discuss
(Singh, 1989). four important aspects in change detectionselection of suit-
In general, change detection applications are directly able remote sensing variables and algorithms, impacts of scale
related to the concepts of land-use/land-cover change. In this issues on change detection methods, and challenges of using
context, it is noteworthy that there is an important distinction multisensor data.
between the terms land use and land cover. The term land cover
refers only to the physical footprints in a region, and land use
18.2.1 Selection of Suitable Remote
refers to human activities performed in a place, consisting of a
Sensing Variables
single coverage or a mosaic for a particular purpose (Barnsley
et al., 2001). Treitz and Rogan (2004) claim that land use is Before selecting potential remote sensing variables for change
an abstract concept related to social, cultural, economic, and detection, it is critical to make sure that proper image prepro-
political factors, and it is weakly associated with the reflec- cessing is conducted, including atmospheric and topographic
tance properties in remote sensing. Therefore, change detec- correction, image-to-image registration, and speckle reduction
tion handles the land-cover changes that present results if radar data are used. Detailed descriptions of image prepro-
caused by conversions and modifications. Land-cover conver- cessing issues can be found in earlier literature (Chander etal.,
sion is characterized by a full replacement of the land cover by 2009; Lu etal., 2008b; Shi and Hao, 2013; Stow and Chen, 2002;
another, such as deforestation or urbanization. Land modifi- Vicente-Serrano etal., 2008).
cation is characterized by subtle changes without the change Remote sensing systems and environmental characteristics
of land-cover types, such as those caused by selective logging, have important impacts on the design of a change detection pro-
drought, and forest insects. A large number of previous stud- cedure (Biging etal., 1999; Jensen, 2005; Lu etal., 2004). Ideally,
ies have investigated land-cover conversion (e.g., Guild etal., the same sensor data with the same radiometric and spatial
2004; Lambin and Strahler, 1994; Li and Yeh, 1998; Mubea and resolution at the same time of the year are needed to eliminate
Menz, 2012), but in the past decade, detection of forest distur- the effects of external sources such as sun angle and phenologi-
bances has received increasing attention due to its importance cal differences. However, in reality, selection of the same sensor
in reducing carbon estimation uncertainty and its impacts on data in a specific study is difficult, especially in moist tropical
management of environmental conditions and biodiversity regions due to cloud cover (Asner, 2001) and low revisit peri-
(Huang etal., 2013; Kennedy etal., 2007; Thomas etal., 2011; ods of satellites. If a change detection requires a long period,
Wilson and Sader, 2002). satellite images may not be available; thus, multisource data
Although much research work related to land-cover change consisting of different satellite images, aerial photographs, and
detection has been published, the majority of it has used the existing thematic maps may be used (Groen et al., 2012; Ichii
same sensor data. In reality, collection of the same sensor data at etal., 2003; Li, 2010; Petit and Lambin, 2001; Tian etal., 2013;
a multitemporal scale is a challenge, especially in moist tropical Walter, 2004). Table18.1 provides potential variables that may be
regions due to the problem of cloud cover (Asner, 2001). Since used in change detection analysis. Selection of suitable variables
different sensor data from optical to radar data with different is a prerequisite for successfully conducting change detection for
spatial resolutions are available, use of multisensor data for a specific study (Lu etal., 2014).
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 379
Major steps Main contents
Research problems and objectives
Geographic location and size
Nature of change
Time periods
detection problems
Change detection system
Accuracy requirement
Characteristics of different kinds of sensor data
Selection of remotely Consideration of atmospheric and environmental
sensed data conditions
Complexity of landscapes under investigation
Geometric rectification/registration
Radiometric and atmospheric correction
Image preprocessing
Topographic correction if needed
Speckle reduction if radar data is used
Per-pixel-based image processing such as
image transformation and vegetation indices
Selection of suitable Subpixel-based image processing such as
variables from spectral mixture analysis
remote sensing data
Image segmentation or classification
Some specific attributes such as biomass
Selection of suitable The characteristics of change detection
algorithms algorithms
corresponding to the Selection of suitable algorithms
selected variables Comparison of different algorithms if needed
Determination of sampling strategy and sample size
Evaluation of change
Collection of reference data
detection results
Accuracy assessment
FIg u r e 1 8.1 Major steps used in a general change detection procedure.
Spectral responses (e.g., spectral bands, vegetation indices, if sufficient training samples are available for conducting land-
and transformed images) may be the most common variables cover classification, the change detection approaches based
used for change detection, in particular, when medium- on classified images may provide better results than spectral
spatial-resolution images such as Landsat and Satellite Pour responses.
lObservation de la Terre (SPOT) are used. As more sensor In the past decade, multisource data are easily obtained;
data with different spatial resolutions are available, spectral thus, effective use of different source data may improve change
signatures may not be the optimal variables; other remote detection performance. Much previous research is based on
sensing features such as spatial and subpixel may better rep- detection of land-cover conversion, but vegetation disturbance
resent the change features. For example, textures and seg- is now regarded as an important source of carbon emission to
ments developed from very-high-spatial-resolution images the atmosphere. Accurate detection of vegetation disturbance
could provide better change detection results than spectral has gained increasing attention. Due to special characteristics
responses, whereas subpixel features developed from coarse- of vegetation types, the change of some forest stand attributes
spatial-resolution images may be better suitable for land-cover such as leaf area index and biomass may better represent the
change detection in a large area due to mixed-pixel problems. change of vegetation disturbance due to natural disasters or
On the other hand, because of the impacts of external factors human-induced activities such as selective logging. Therefore,
such as atmosphere, soil moisture, and vegetation phenology many potential variables can be used for change detection
on the spectral signatures, use of spectral responses may pro- studies. The key is to identify the variables that can best rep-
duce large spurious change detection results. In this situation, resent the change detection purpose. Integration of different
2016 by Taylor & Francis Group, LLC
380 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 8.1 Summary of Potential Variables Used in Land-Cover Change Detection Analysis
Variables Major Characteristics Advantages Disadvantages
Spectral Spectral responses are the most common Remote sensors capture spectral Some land-cover types have similar spectral
features variables for change detection. Vegetation features of land-cover surfaces. features and thus cannot be effectively used to
indices and transformed images may further The change in spectral features detect this kind of land-cover change. External
improve change detection performance. The represents the change of land factors such as the atmosphere, vegetation
majority of change detection techniques are covers and thus may provide the phenology, and soil moisture affect the spectral
based on the use of spectral features. best change detection results. responses.
Spatial Texture and segmentation are common Use of spatial features between Pure spatial features have relatively poor
information methods to use the spatial information different land-cover surfaces and capability in separating land-cover types, thus
inherent in remote sensing data. The spatial reduction of spatial heterogeneity lessening effective detection of land-cover
features are especially valuable for high- in the same land covers are changes. The spatial resolution of remote
spatial-resolution images and radar data. valuable in improving land-cover sensing data and complexity of land-cover
classification and change surfaces in a study area affect the spatial
detection performance. features.
Subpixel Mixed pixels are a major problem affecting the Subpixel features can provide more Application of subpixel features for change
information change detection because changed areas are accurate results for detecting detection is very limited due to the difficulty
small and scattered in different locations. Use land-cover modification, indeveloping algorithms to conduct change
of subpixel features such as the fraction images especially vegetation disturbance detection based on the fractional images.
through unmixing the multispectral images caused by natural disasters or Thecomplexity of a study area may affect
has the potential to provide better change human-induced activities. thedevelopment of fraction data.
detection results.
Thematic Separate land-cover classification for each date The change detection based on Accurate classification result for each image is
information of imagery is required before conducting classified images can provide required but may be difficult for historical
change detection. The classified images are detailed change trajectories. The remote sensing data due to lack of reference data.
then used to examine land-cover change external factors do not affect the
trajectories using the postclassification change detection quality.
comparison approach.
Biophysical Some biophysical attributes such as impervious The land-cover change, especially To identify a suitable parameter that better
attributes surface area in urban landscapes and leaf area vegetation change, can be more reflects the change of a specific land-cover type
index in forest ecosystems can be derived from quantitatively evaluated and is critical; meanwhile, this parameter can be
remotely sensed data. These variables may detected using certain biophysical accurately extracted from remotely sensed data.
better suit specific change detection purposes. parameters.
Multisource Different variables from remote sensing or Availability of multisource data Different sources of data have various formats,
data ancillary data (e.g., previously developed provides an alternative method to quality, and spatial resolutions. Improper use of
thematic maps) may be used in a change conduct land-cover change multisource data may produce a large error due
detection procedure. detection, especially when remote to the data quality problems.
sensing data are not available.
variables in a change detection procedure may provide some per-pixel-based change detection techniques and 3 object-based
new insights for more accurately developing the change detec- techniques. Recently, the combination or fusion of different
tion results. change detection methods has received increasing attention for
improving change detection performance (Du etal., 2012, 2013).
Table 18.2 groups the change detection algorithms into six cat-
18.2.2 Selection of Suitable Algorithms
egories: per-pixel thresholding, per-pixel classification, subpixel,
for change Detection
object-oriented, hybrid, and indirect methods, according to the
In the past four decades, many change detection techniques have use of variables and corresponding change detection techniques.
been developed, and the majority of them are summarized in the Image-processing techniques can be based on per-pixel, sub-
literature (e.g., Bhagat, 2012; Hussain etal., 2013; Lu etal., 2004; pixel, and segment (or object-oriented) scales and combina-
Singh, 1989). For example, Lu etal. (2004) summarized over 30 tions of them; thus, change detection also can be based on these
techniques and grouped them into 7 categories: algebra, trans- scales. The majority of change detection techniques are based
formation, classification, advanced models, approaches to geo- on pixel-based methods (Bhagat, 2012; Hussain et al., 2013;
graphic information science, visual analysis, and others. Bhagat Lu etal., 2004) because of the unique characteristics of remote
(2012) summarized 29 techniques grouped into 8 categories: sensing data. Depending on the purpose of the change detection
spectral classification, multidate radiometric change, support analysis, many techniques such as image differencing, regression
vector analysis, the hybrid approach, artificial neural network, analysis, and principal component analysis are used to detect
fusion, object comparison, and the triangle model. Hussain etal. binary change and nonchange categories; thus, the key step is
(2013) summarized over 20 approaches, which were grouped to identify the optimal thresholds (Lu etal., 2005). The per-pixel
into 10 subclasses, and provided a brief description of 15 thresholding-based approaches can provide only change and
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 381
TABLe 1 8.2 Summary of Six Categories of Change Detection Algorithms
Algorithms Major Characteristics Advantages Disadvantages
Per-pixel thresholding- Many change detection techniques, such Simple and easy to implement and can Selection of thresholds is subjective
based methods as image differencing, vegetation index provide valuable information for and scene dependent, depending on
differencing, principal component designing a proper procedure for analysts skills and familiarity with
analysis, and regression analysis, belong further examining the detailed the study area. Different external
to this category. The critical steps are to land-cover change trajectories. factors such as phenology and soil
identify suitable variables that can moistures may affect change
effectively reflect the change of types of detection results. This category of
interest and to determine suitable methods can only provide change
thresholds in both tails of the histogram and nonchange detection.
representing the changed areas.
Per-pixel classification- The postclassification comparison may be The impacts of external factors on These methods are criticized for their
based methods the most common approach to detect change detection results can be incapability of solving mixed-pixel
land-cover trajectories based on the avoided. This category of methods can problem in medium- or coarse-
per-pixel-based classification for each provide detailed land-cover change spatial-resolution images and the
date of remote sensing data. Another trajectories. high spectral variation within the
approach is to conduct direct same land covers in high-spatial-
classification using per-pixel-based resolution images. The quality of
algorithms based on the combined fromto change detection results
multitemporal remote sensing data. depends on the classification
Detailed training sample data for each accuracy for each date being
land cover or each change trajectory are analyzed.
required.
Object-based methods These detection methods can be This category of methods can reduce the High-quality segmentation image is
conducted in three ways: (1) direct spectral variation within the same land required. It is important to identify
comparison of segmentation images at covers; thus, they are preferable for optimal parameters (e.g., minimum
different dates, (2) comparison of high-spatial-resolution images. value distance, variance factor,
classified objects, and (3) image minimum size of pixels in a segment)
segmentation and classification based on used in a segmentation procedure.
stacked multitemporal images.
Subpixel-based methods Different subpixel methods, such as fuzzy This method is valuable for some specific They lack suitable subpixel-based
classification and spectral mixture change detection studies such as forest algorithms to detect detailed fromto
analysis, may be used. disturbance and urban expansion. land-cover change trajectories.
Hybrid methods Hybrid methods may be conducted in two The merits of different methods may be It is critical to select suitable variables
ways: (1) combination of different integrated into a change detection and corresponding algorithms to
methods into one change detection procedure to produce more accurate conduct the change detection
procedure and (2) combination of results than individual methods. procedure, but these methods lack
change detection results from different general rules to guide this process.
methods into a new result using
developed rules or fusion methods at
feature or decision level.
Indirect methods Indirect methods identify some This method could be valuable for global The key is to develop accurate
biophysical attributes that can effectively land-cover change detection using biophysical attributes from remote
reflect the land-cover change. These coarse-spatial-resolution images, sensing data where the attribute is
attributes can be derived from remote especially for deforestation and suitable for use as a variable for
sensing data through modeling. urbanization. Another important change detection analysis. It is also
Common attributes are, for example, application may be the detection of critical to make sure that the
impervious surface in urban landscapes forest disturbance caused by natural or uncertainty from the attribute
and leaf area index in forest ecosystems. anthropogenic factors by analyzing the estimation is much less than the true
change in vegetation structures such as change amount of the corresponding
leaf area index or biomass. attribute.
nonchange categories, but in reality, detailed fromto change urban landscapes. Therefore, subpixel-based and object-based
trajectories are needed for many studies (Lu etal., 2012, 2013). approaches are proposed. The spatial resolution of selected
In this case, the per-pixel-based classification approaches such remote sensing data, complexity of the study area, and the users
as the postclassification comparison can meet this requirement. needs may significantly affect the determination of the change
Although per-pixel-based approaches are most commonly used detection method. For high-spatial-resolution image, object-
in practice, its accuracy for calculation of change detection areas based methods may be preferable because of the high spectral
may be poor, especially when coarse-spatial-resolution images variation within the same land cover. In contrast, the subpixel-
are used due to the complex land-cover composition such as in based methods may be preferable for coarse-spatial-resolution
2016 by Taylor & Francis Group, LLC
382 Remotely Sensed Data Characterization, Classification, and Accuracies
images due to the mixed-pixel problem (Lu et al., 2014). Since At a regional scale, medium-spatial-resolution images such as
different change detection methods have their own merits and Landsat and SPOT are commonly used for land-cover change
characteristics, a hybrid approach consisting of different algo- detection. Most previous research was based on regional scale,
rithms may provide better change detection results, but have not and different change detection techniques as summarized in
been fully examined yet. Table 18.2 can be used. It is difficult to select an optimal change
detection technique for a specific study because of the following
problems: Which variables as summarized in Table 18.1 should
18.2.3 impacts of Scale issues on the Design
be selected? Which change detection technique should be used
of a change Detection Procedure
considering the complexity of the study area such as urban-
Scale is an important concern in designing a proper change dominated or forest-dominated ecosystems? Although much
detection procedure. Different extents of a study area will research work has been done to explore remote sensing variables
directly influence the selection of remote sensing data and asso- and algorithms for different ecosystems such as forest, urban,
ciated change detection methods. Table 18.3 provides a sum- wetland, and agriculture, as reviewed in previous literature
mary of potential variables and algorithms suitable for local, (e.g., Bhagat, 2012; Hussain etal., 2013; Lu etal., 2004), it is still
regional, national, and global land-cover change detection and unclear which one should be used for a specific study. In prac-
discusses major problems that occur at different scales. tice, a comparative analysis of different variables and algorithms
At a local scale, accurate change detection results are is used to identify the best change detection procedure for a spe-
required because these results may be used as reference data cific purpose (Lu etal., 2014). Recently, hybrid approaches that
for evaluation of other data from medium- or coarse-spatial- consist of different algorithms or remote sensing data have been
resolution images. Thus, high-spatial-resolution images are used to combine their merits and generate the best change detec-
needed for this purpose. In practice, use of multitemporal tion result.
high-spatial-resolution images for change detection is a chal- Compared with the change detection research at regional
lenge. As shown in Figure 18.2, the major problems include (1) scale, studies at national and global scales face greater challenges
different shadow sizes between two QuickBird images caused due to the complexity of land-cover types across large areas and
by different sun elevation angles at image acquisition dates; the mixed-pixel problem in the coarse-spatial-resolution images,
(2) different colors for the same impervious surfaces, as the as illustrated in Figure 18.3. The majority of change detection
locations labeled in A, B, C, and D in Figure 18.2a1 and a2; techniques suitable for medium-spatial-resolution images at
and (3) displacement caused by tall buildings, labeled in A and a regional scale are not feasible for the coarse-spatial-resolu-
B in Figure 18.2b1 and b2. Because of these problems, direct tion images at national or global scale; thus, researchers need
use of high-spatial-resolution images for change detection to develop new change detection techniques that are suitable
may produce large errors based on traditional per-pixel-based for coarse-spatial-resolution images. Integration of multiscale
change detection techniques. It is necessary to develop suit- and multisource data may provide new insights for developing
able change detection techniques by taking these problems detection techniques suitable for large-area changes but has not
into account. received sufficient attention (Lu etal., 2014).
TABLe 1 8.3 Impacts of Scale Issues on Change Detection Analysis
Scales Potential Data Potential Algorithms Major Problems
At local Submeter spatial resolution Application of textures and segmentation-based The following factors will significantly affect the
scale satellite images such as methods will be valuable. change detection results: shadows; confusion of
QuickBird and WorldView are shadow, dark impervious surfaces, water, and
preferable to provide accurate wetland; high spectral variation of the same
and detailed land-cover urban land cover; and spectral confusion
change detection results. among impervious surfaces and bare soils.
At regional Medium-spatial-resolution Almost all change detection techniques are Depending on research purpose and datasets,
scale images such as Landsat and developed at regional scale and can be used to selection of suitable variables and algorithms is
SPOT data; integration of detect land-cover change. Data fusion techniques at important to provide accurate change detection
medium-spatial-resolution different levels such as pixel, feature, and decision results.
image and limited high- can be used to integrate different remote sensing
spatial-resolution images; data or combine change detection results from
multisource data. different algorithms to improve change detection
performance.
At national Coarse-spatial-resolution Integration of MODIS and Landsat images has been The complexity of land-cover composition in
and global images such as time-series used to detect land-cover change. coarse-spatial-resolution image makes change
scales MODIS or SPOT datasets. detection a challenge. The critical task is to solve
the mixed-pixel problem and the complexity of
land-cover distribution in a large area.
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 383
A (a1) A (a2)
C C
B B
D D
2008 Quickbird 2004 Quickbird
band 432 as red, band 432 as red,
green, and blue green, and blue
2007 (b1) 2008 (b2)
B
B
A A
FIg u r e 18.2 Problems in using very-high-spatial-resolution images such as QuickBird for land-cover change detection analysis (a1 and a2 show
the impacts of different sun elevation angles on the size of shadows cast by tall buildings; b1 and b2 show the impacts of different sun elevation
angles on the displacement caused by tall buildings too.)
(a) (b)
Hangzhou
Hangzhou
Lake
River
Lake
Forest
River
Forest
FIg ur e 18.3 Problems in using (a) medium- and (b) coarse-spatial-resolution images (Landsat TM vs. MODIS surface reflectance images) for
land-cover change detection analysis.
2016 by Taylor & Francis Group, LLC
384 Remotely Sensed Data Characterization, Classification, and Accuracies
18.2.4 challenge of Using Multisensor of the differences in spatial, spectral, and radiometric resolu-
Data for change Detection tions, as well as view angles and polarization options for radar
data, traditional change detection techniques such as image
Change detection is usually conducted using the same sen- differencing and principal component analysis are not suitable
sor data with anniversary or near-anniversary dates in order for different sensors. Previous research has mainly examined
to reduce the impacts of remote sensing data per se and envi- the use of multiple optical sensor data for land-cover change
ronmental conditions on the change detection results (Lu etal., detection, such as Landsat Multispectral Scanner (MSS) and
2014). However, sometimes, the same sensor data for a specific Thematic Mapper (TM) (Serra etal., 2003), Landsat Enhanced
study area are not available, especially in situations where the Thematic Mapper Plus (ETM+), and Satellite Pour lObservation
revisit period of sensors does not fit the date of interest well or de la Terre multispectral (SPOT XS) (Deng etal., 2008), Landsat
when images suffer from cloud cover (Lu et al., 2008a, 2014). TM, and Satellite Pour lObservation de la Terre high resolu-
Microwave data represent an alternative to overcome this lim- tion geometrical (SPOT HRG) (Lu etal., 2008a). However, rarely
itation in tropical regions because they enable the imaging in has research examined the integration of optical and radar data
adverse weather conditions, in addition to other advantages for land-cover change detection (Alberga, 2009; Lu etal., 2014;
such as the production of images with different types of polar- Reiche etal., 2013) due to their significantly different features. In
ized energy (HH, HV, VV, VH). However, we should take into practice, use of optical and radar data may be needed, especially
account that unlike optical images containing information in tropical moist regions. The challenge is that current change
about the reflectance and the physicochemical properties of detection techniques cannot effectively handle this problem;
the targets, microwave data carry information about the struc- thus, new techniques to implement the change detection are
ture and dielectric properties of the targets (Jensen, 2007) and required.
emphasize complementary information between optical sensors
and microwave data. One possibility for dealing with the data
availability problem, particularly in tropical regions, is the use 18.3 case Study of Land-cover
of multisensor data, which greatly increases the acquisition of change Detection Using
cloud-free data and may also add information due to the differ- optical and Radar Data
ent characteristics of the sensors involved in the process.
Integration of multisensory data is extensively used for In order to explore the use of multitemporal multisensor data
improving land-cover classification (Lu etal., 2011; Pereira etal., for land-cover change detection, Radarsat-2 C-band in 2009 and
2013). The use of different sensor data at multitemporal scales EO-1 ALI (Earth Observing-1 Advanced Land Imager instru-
in a change detection procedure presents an alternative to solve ment) data from 2013 were used in this research. Figure 18.4
the data acquisition problem but presents a great challenge in illustrates the flowchart for conducting the change detection.
the development of techniques (Akiwowo and Eftekhari, 2013; The major steps include image preprocessing of both datasets
Zeng etal., 2010) because different data can have various quali- and change detection using the direct classification approach.
ties and spatial resolutions. The different sensor data could be As a comparison of the change detection results, a postclassifica-
from optical sensors, radar, or a combination of them. Because tion approach is also conducted.
The 2009 Radarsat image The 2013 EO-1 ALI image
Speckle reduction using Image-to-image Put both registered
gamma filtering method registration images into one file
Training samples for non
Training samples for the
change classes and change-
2009 and 2013 images
trajectory classes
Classification using
Evalution of the
SVM for the 2009 and Change detection
classified images
2013 image separately analysis using the direct-
classification approach
Change detection analysis
using the post classification Evaluation and comparison
comparison approach of change detection results
FIg u r e 1 8.4 Framework of land-cover change detection analysis used in this case study.
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 385
18.3.1 Study Area activities associated with this process such as wood extraction
and expansion of agricultural fields and pasture areas started
The study area with an extent of approximately 600,000 ha is along this highway and adjoining areas. Currently, agriculture
located near the Tapajos National Forest along the Cuiab- and cattle ranching are the main economic activities in this
Santarm highway BR-163 in Belterra Municipality, Par State, region, generating a variety of land-cover classes including pas-
Brazil (Figure 18.5). In the 1970s, the Brazilian government tures, soybeans, and other grains such as rice, maize, and sor-
initiated a policy of encouraging the occupation of Brazilian ghum. This study area is relatively flat with elevations ranging
Amazonia and financed the construction of BR-163. Because from 50 to 200 m and has a warm and humid climate. The maxi-
of this policy and highway construction, deforestation and mum annual average temperature varies between 31C and 33C,
Brazil Par Belterra
Meters
Meters Meters 0 300,000 600,000
0 1,500,000 3,000,000 0 450,000 900,000
724000 726000 728000 730000 732000 734000
9657000
9657000
9655000
9655000
9653000
9653000
9651000
9651000
9649000
9649000
724000 726000 728000 730000 732000 734000
N
0 1500 3000
Meters
FIg u r e 1 8.5 Location of the study area in Belterra Municipality, Par State, Brazil.
2016 by Taylor & Francis Group, LLC
386 Remotely Sensed Data Characterization, Classification, and Accuracies
and the minimum average ranges from 24C to 25C. Annual TABLe 1 8.6 Sample Pixels for Use in the Support Vector
rainfall varies between 1800 and 2800mm with a clear division MachineBased Direct Classification
in rainfall distribution. The rains are plentiful in the first half of Class Training Test
the year and low in intensity during the rest of the year.
No change
AGAG 1246 1106
18.3.2 Data collection and Preprocessing BSBS 1850 1278
CPCP 1070 1299
Radarsat-2 C-band, EO-1 ALI, and field survey data were col-
DPDP 966 1103
lected and summarized in Table 18.4. More detailed samples
ISIS 881 1145
collected during the field survey are summarized in Tables18.5
PFPF 1610 4415
and 18.6. The Radarsat-2 and EO-1 ALI data were coregistered SS2SS2 1111 1264
into the Universal Transverse Mercator (UTM) coordinate sys- WTWT 2422 1713
tem and resampled to a cell size of 5 m. The registration error
Change
was 0.49 pixels. The Radarsat-2 data in amplitude with three
AGSS1 522 221
bandsHH, VV, and (HV + VH)/2were filtered using a stan-
BSAG 1568 1665
dard gamma filter with a window size of 3 3 pixels.
CPDP 2078 1933
CPSS1 892 1082
18.3.3 Methodology DPCP 814 555
DPSS1 900 1195
18.3.3.1 Land-Use/Land-cover classification
SS1BS 1039 1157
with a Support Vector Machine
SS1SS2 416 460
Research on classification algorithms has received great atten- Notes: AG, agriculture; BS, bare soil; CP, clean pasture; DP, dirty pas-
tion in the past four decades. A large number of algorithms, ture; IS, impervious surface; PF, primary forest; SS1, initial secondary suc-
from traditional statistical-based methods such as minimum cession; SS2, intermediate secondary succession; WT, water.
distance and maximum likelihood to complex nonstatistical
TABLe 1 8.4 Datasets Used in Research algorithms such as neural network, decision tree, and support
vector machine (SVM), have been developed for land-cover
Datasets Major Characteristics Data Acquisition Dates
classification (see Lu and Weng, 2007; Tso and Mather, 2009).
Radarsat-2 A full, polarized Radarsat-2 scene in September 9, 2009 In the last decade, nonstatistical-based classification algorithms
fine-resolution quad-polarization have been regarded as having more advantages for providing
beam mode 13 (FQ13) mode with
a pixel size of 4.733 4.748 m.
better land-cover classification than traditional statistical-
EO-1 ALI Nine spectral bands with 30 m October 5, 2013
based algorithms, especially when multisensor/source data
spatial resolution. are used (Li etal., 2012b; Lu etal., 2011). Of the many nonsta-
Field A large number of samples for September 2009 tistical algorithms, SVM is a fairly recent pattern recognition
survey training and test purposes were August 2013 method that has been drawing attention due to its data distribu-
collected (see Tables 18.5 and 18.6). tion independence, good generalization ability, and robustness
against the Hughes phenomenon (Camps-Valls and Bruzzone,
TABLe 1 8.5 Sample Pixels Collected in 2009 and 2013 for Use in the 2009; Camps-Valls etal., 2008; Mountrakis etal., 2011; Perumal
Classification of Individual Sensor Data and Bhaskaran, 2009; Vapnik et al., 1996). Therefore, SVM is
Samples from 2009 Samples from 2013 used in this research for land-cover classification in the moist
tropical region.
Class Training Test Training Test
This method, in its original conception, consists of finding
AG 385 269 2095 1978 a separation hyperplane between training samples with larger
BS 742 1786 1747 2211 margins. The separating hyperplane is a geometric place where
CP 2134 2115 2360 1900
the following linear function is zero:
DP 1004 1952 1720 1702
IS 1063 1063 698 1101
PF 1391 3042 1812 2314 f (x) = w , x + b , (18.1)
SS1 1272 1281 1059 858
SS2 828 796 844 929
WT 877 1445 1448 1105 where
Notes: AG, agriculture; BS, bare soil; CP, clean pasture; DP, dirty pasture; IS,
w represents the orthogonal vector to the hyperplane f(x) = 0
impervious surface; PF, primary forest; SS1, initial secondary succession; SS2, b/||w|| is the distance from the hyperplane to the origin
intermediate secondary succession; WT, water. , denotes the inner product
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 387
The parameters of Equation 18.1 are obtained from the following To apply this method to a multiclass problem (problems with
quadratic optimization problem (Theodoridis and Koutrombas, more than two classes), it is necessary to adopt multiclass strate-
2008): gies, such as one against all or one against one (Theodoridis and
Koutrombas, 2008).
m m m
Before implementing land-cover classification using the SVM
y y
1
i i j i j (x i ), (x j ) ,
2 approach, it is necessary to determine a suitable classification
i =1 i =1 j =1
system. Based on our fieldwork in 2009 and 2013, a classification
0 i C ; i = 1,,m, (18.2) system consisting of nine land-use/land-cover classesagricul-
m ture (AG), bare soil (BS), clean pasture (CP), dirty pasture (DP),
subject to
i y i = 0, impervious surface (IS), primary forest (PF), initial secondary
i =1 succession (SS1, up to 7years of regrowth), intermediate second-
ary succession (SS2, from 7 to 15years of regrowth), and water
where (WT)was designed for this study area. Training samples as
i represents Lagrange multipliers summarized in Table 18.5 were used separately for the land-
yi = {1, +1} defines the class of xi, since SVM is a binary cover classification based on the 2009 Radarsat C-band and the
classifier 2013 EO-1 ALI data. Based on our previous exploration using
C (cost) acts as an upper bound of values SVM for land-cover classification in the moist tropical region
(x) is a function adopted to remap the input vectors into a (Negri et al., 2011, 2012, 2013, 2014) and this experiment, the
higher dimensionality space parameters used in this research were finally selected as follows:
penalty parameter of 100 for all classification cases and gamma
The inner product (xi),(xj) is known as the kernel function. value in the RBF kernel function of 0.33 for Radarsat classifica-
The radial basis function (RBF) is the kernel function used in tion, 0.11 for EO-1 ALI classification, and 0.083 for the 12-band
this study (Webb, 2002): stacked data using the SVM-based direct classification approach.
(x i ), (x j ) = e
( xi x 2j / 2 2 )
, . (18.3) 18.3.3.2 change Detection with
Postclassification comparison
The optimization problem in Equation 18.2 is solved consider- Although many change detection techniques have been developed,
ing a training set D = {(x i , y i ) : i = 1,,l }, where x i Rd . Let SV = as summarized in previous literature (e.g., Bhagat, 2012; Chen etal.,
{xi:i 0;i = 1,,l}, known as the support vector set. The param- 2012; Coppin etal., 2004; Hussain etal., 2013; Lu etal., 2004, 2014),
eters w and b are computed by the postclassification comparison approach is commonly used
to examine land-cover change trajectories and thus used in this
w= y (x ),
xi SV
i i i (18.4) research. In order to conduct the land-cover change detection, one
critical step is to design the change detection system for this study.
Based on our classification results with 9 land-cover types for 2009
l l
and 2013, the postclassification comparison approach can produce
y y
1 9 nonchange classes and 72 potential change classes. However, for a
b= yi + i j i j (x i ), (x j ) . (18.5)
# SV xi SV i =1 j =1 specific study area, some of the change trajectories are not possible
within a 4-year interval or do not exist, as illustrated in Table18.7.
TABLe 1 8.7 Potential Land-Cover Change Trajectories at the Study Site in Belterra
Municipality, Par State, Brazil, 20092013
AG BS CP DP FP SS1 SS2 WT IS
AG N P P P I P I P P
BS P N P P I P I P P
CP P P N P I P I P P
DP P P P N I P I P P
FP P P P P N P I U P
SS1 P P P P I N P U P
SS2 P P P P I P N U P
WT P P P P I P P N P
IS U U U U I U I P N
Notes: AG, agriculture; BS, bare soil; CP, clean pasture; DP, dirty pasture; IS, impervious surface;
PF, primary forest; SS1, initial secondary succession; SS2, intermediate secondary succession; WT,
water; P, possible; I, impossible; N, nonchange; U, improbable.
2016 by Taylor & Francis Group, LLC
388 Remotely Sensed Data Characterization, Classification, and Accuracies
In this table, all transitions in the postclassification comparison in collecting reference data at multitemporal periods (Foody,
labeled I or U are left as unclassified. 2010; Morisette and Khorram, 2000; Olofsson etal., 2013).
Based on our field survey, only eight change classes were In this research, test samples of nine land-cover types for
observed: AGSS1, BSAG, CPDP, CPSS1, DPCP, DPSS1, 2009 and 2013 were collected during the field survey and are
SS1BS, and SS1SS2. In 4years, the SS1SS1 nonchange class summarized in Table 18.5. The test samples for eight change
was not found. This is natural because it is very unlikely that SS1 classes and eight nonchange classes are summarized in Table
will remain stable for more than 3years. Also, no deforestation 18.6. During implementation of the accuracy assessment of
was found in this area between 2009 and 2013. the change detection results using postclassification compari-
son, many pixels fell into an unclassified/nonobserved class.
18.3.3.3 change Detection with SVM-Based Therefore, it was necessary to take the unclassified/nonobserved
Direct classification class into account in the accuracy assessment. A modified con-
Although the postclassification comparison approach can pro- fusion matrix in which the unclassified/nonobserved pixels
vide detailed land-cover change trajectories, the change detec- were disregarded was used. This kappa value, denominated here
tion result depends on the accuracy of individual classification as a partial kappa coefficient, could not be directly compared
results. This means that the classification results for both dates with standard kappa values.
should have high accuracy in order to produce reliable change
detection results (Lu etal., 2014). However, in reality, land-cover 18.3.4 Results
classification for historical remote sensing data is often difficult
if training sample data are not available or the required land- 18.3.4.1 Analysis of classification Results
cover types cannot be effectively separated from the remote sens- Using a Support Vector Machine
ing data (Lu etal., 2014). In this study, Radarsat-2 C-band data A comparison of the land-cover classification results from 2009
were used, but previous research has indicated that Radarsat-2 and 2013, as illustrated in Figure 18.6, indicates that obviously,
C-band cannot provide reliable land-cover classification results the Radarsat data cannot effectively separate the land-cover types
and has very poor classification accuracy compared to optical and has much poorer classification accuracy than the EO-1 ALI
sensor data (Li etal., 2012a; Lu etal., 2011). Therefore, an SVM- data. Based on the error matrices, the kappa coefficient for the
based direct classification approach was used in this study to 2009 Radarsat-based result is only 0.19 and is 0.76 for the 2013
detect land-cover change based on detailed field survey data. EO-1 ALIbased classification result. The error matrices drawn
After image preprocessing for both the 2009 Radarsat and in Figure 18.7 indicate that only CP can be separated from other
the 2013 EO-1 ALI data, these images were stacked into one file. land-cover types with an accuracy of 77%, but all other land
Training samples for the eight change classes and eight non- covers have very low accuracy based on the Radarsat data; in
change classes, as summarized in Table 18.6, were used to train contrast, the EO-1 ALI data can provide good classification for
the SVM classifier. The nonchange classes were used as coun- AG, BS, DP, PF, and AT with accuracies of more than 83%, but
terexamples in the classification stage to help avoid the errors of SS1 and SS2 are highly confused with each other. This conclu-
inclusion. This approach has the advantage of avoiding the sum sion is similar to our previous research conclusions that optical
of errors produced when comparing two land-use/land-cover sensor data can provide much better classification results than
maps. Also, transitions that are not feasible, or impossible, are radar data in the moist tropical regions of the Brazilian Amazon
avoided once only valid transitions are used as reference. The (Lu etal., 2011).
drawback of such a procedure is finding a sufficient and large
enough set of all representative transitions and considering all 18.3.4.2 Analysis of change Detection Results
types of nonchange classes at the same time. Using Postclassification comparison
The change detection result using postclassification compari-
18.3.3.4 evaluation and comparison of son shows very poor accuracy, as illustrated in Figure 18.8, and
change Detection Results many pixels are not detected or wrongly detected. The major
The error matrix approach provides detailed assessment of the problem is the poor classification result from the 2009 Radarsat
agreement between the classified result and reference data and image, thus resulting in a large number of unclassified pixels in
provides information about how the misclassification happened which several pairs of fromto classes are considered invalid
(Congalton and Green, 2008). Therefore, this approach is usually (i.e., impossible or very unlikely, as indicated in Table 18.7), as
used for evaluating land-cover classification based on reference observed in black in Figure 18.8. Another problem is the pix-
data (Congalton and Green, 2008; Foody, 2002; Van Oort, 2007). els in yellow that correspond to a supposed valid transition but
From the error matrix, producers accuracy, users accuracy, over- was not observed during fieldwork. These problems imply the
all classification accuracy, and kappa coefficient are computed difficulty of using postclassification comparisons for land-cover
(Congalton, 1991; Congalton and Green, 2008; Foody, 2002). The change detection based on optical and radar data.
accuracy assessment for historical land-cover classification is dif- Figure 18.9 represents a special confusion matrix where two lines
ficult due to lack of test samples, and evaluation of the change are added at the bottom to show the percentage of known test pix-
detection results is especially challenging because of the difficulty els that are supposed to be a certain type of change. Based on this
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 389
724000 726000 728000 730000 732000 734000
9656000
9656000
9654000
9654000
9652000
9652000
9650000
9650000
724000 726000 728000 730000 732000 734000
(a)
724000 726000 728000 730000 732000 734000
9656000
9656000
9654000
9654000
9652000
9652000
9650000
9650000
724000 726000 728000 730000 732000 734000
(b)
N
AG CP PF SS1 WT
Meters
BS DP IS SS2 0 1500 3000
FIg u r e 18.6 Land-cover classification using the Radarsat data for 2009 (a) and the EO-1 ALI data for 2013 (b). Classification types are agricul-
ture (AG), bare soil (BS), clean pasture (CP), dirty pasture (DP), impervious surfaces (IS), primary forest (PF), initial secondary succession (SS1),
intermediate secondary succession (SS2), and water (WT).
2016 by Taylor & Francis Group, LLC
390 Remotely Sensed Data Characterization, Classification, and Accuracies
Reference 2013
Reference 2009
AG BS CP DP IS PF SS1 SS2 WT
AG BS CP DP IS PF SS1 SS2 WT
AG 24.2 22.2 AG 85.9 16.7
BS 28.6 36.9 BS 83.3
CP 32.0 23.6 77.0 56.7 57.8 47.6 36.9 28.4 27.8 CP 77.2
Classified 2013
DP
Classified 2009
DP 10.2 90.2 12.0
IS IS 60.0
PF 13.5 10.4 16.1 18.4 20.5 PF 93.6
SS1 15.2 17.2 26.3 19.2 31.6 41.1 47.4 SS1 51.4 28.6
SS2 SS2 12.0 40.0 37.5 59.0
WT 49.6 WT 98.5
(a) (b)
FIg u r e 18.7 Confusion matrices for land-cover maps using support vector machine classification based on the Radarsat data in 2009 (a) and
the EO-1 ALI data in 2013 (b).
724000 726000 728000 730000 732000 734000
9656000
9656000
9654000
9654000
9652000
9652000
9650000
9650000
724000 726000 728000 730000 732000 734000
N
No change CPDP DPSS1 Nonobserved
AGSS1 CPSS1 SS1BS Nonclassified Meters
0 1500 3000
BSAG DPCP SS1SS2
FIg u r e 18.8 Land-cover change detection result using postclassification comparison between 2009 and 2013. Change types are agriculture to
initial secondary succession (AGSS1), bare soil to agriculture (BSAG), clean pasture to dirty pasture (CPDP), clean pasture to initial secondary
succession (CPSS1), dirty pasture to clean pasture (DPCP), dirty pasture to initial secondary succession (DPSS1), initial secondary succession
to bare soil (SS1BS), and initial secondary succession to intermediate secondary succession (SS1SS2).
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 391
No Reference biomass; thus, most PF and SS2 are misclassified as clean pasture
change AGSS1 BSAG CPDP CPSS1 DPCP DPSS1 SS1BS SS1SS2
No in Radarsat data (Figure 18.7), leading to impossible transitions in
change 21.3 10.2 27.2 72.4 17.6 26.1 the change detection mapping. Overall, the postclassification com-
AGSS1 parison can only provide the accuracy of CPDP with 77.7% and
BSAG 40.3 BSAG, CPSS1, and SS1BS with accuracies of 40%47.7%. Other
CPDP 77.7 41.2 20.4 land-cover change trajectories were not detected.
CPSS1 40.0 24.9 26.7
18.3.4.3 Analysis of change Detection Results
Classified
DPCP
Using Direct classification
DPSS1
The change detection result using SVM-based direct classifi-
SS1BS 47.7
cation is illustrated in Figure 18.10, which highlights the eight
SS1SS2 35.9 19.4
change classes. Compared with the change detection result using
Nonclassified 32.8 46.6 31.1 postclassification comparison (see Figure 18.8), the SVM-based
Nonobserved 24.9 32.4 22.1 11.9 14.3 52.3 26.5 direct classification provided much better spatial distribution
patterns of the change detection results. The accuracy assess-
ment results illustrated in Figure 18.11 indicate the reliability
FIg u r e 18.9 Confusion matrix for land-cover change map from
of direct classification. Based on the error matrix, the overall
postclassification comparison between 2009 and 2013.
accuracy of 87.8% and kappa coefficient of 0.80 were obtained.
error matrix, the calculated partial kappa coefficient is only 0.28. The change trajectories of AGSS1, CPDP, SS1BS, and SS1
One can notice a great number of unclassified pixels, which corre- SS2 have high accuracy with over 91%, but the change trajecto-
sponds to a calculated 22.3% (mostly impossible type), in the area ries of BSAG and CPSS1 have relatively poor accuracy with
where the majority are nonchange classes. This is due to the fact only 63% and 56%, respectively, because of the confusion with
that Radarsat C-band data saturate very quickly with increasing nonchange classes.
724000 726000 728000 730000 732000 734000
9656000
9656000
9654000
9654000
9652000
9652000
9650000
9650000
724000 726000 728000 730000 732000 734000
No change CPDP DPSS1 N
AGSS1 CPSS1 SS1BS Meters
BSAG DPCP SS1SS2 0 1500 3000
FIg u r e 18.10 Land-use/land-cover change distribution using support vector machinebased direct classification between 2009 and 2013.
Change types are agriculture to initial secondary succession (AGSS1), bare soil to agriculture (BSAG), clean pasture to dirty pasture (CPDP),
clean pasture to initial secondary succession (CPSS1), dirty pasture to clean pasture (DPCP), dirty pasture to initial secondary succession (DP
SS1), initial secondary succession to bare soil (SS1BS), and initial secondary succession to intermediate secondary succession (SS1SS2).
2016 by Taylor & Francis Group, LLC
392 Remotely Sensed Data Characterization, Classification, and Accuracies
Reference Although direct classification provides good change detection
No change AGSS1 BSAG CPDP CPSS1 DPCP DPSS1 SS1BS SS1SS2
accuracy, we needed to prove that the change and nonchange
No change 90.1 37.0 42.4 18.1 classes obtained in the change detection process were indeed
AGSS1 95.6
the existing classes in the region. Thus, to show the representa-
tiveness of this methodology, the 2009 and 2013 land-use/land-
BSAG 63.0
cover maps were retrieved from the change map. Each pixel of
CPDP 99.5 the land-use/land-cover map for 2009 is related to the initial
Classified
CPSS1 56.0 condition in the change process, and each pixel of the land-use/
DPCP 81.9
land-cover map for 2013 contains information related to the
late stage. Their accuracies are evaluated using independent test
DPSS1 84.8
samples (see Table 18.5), and the results for these classifications
SS1BS 100
are shown in Figure 18.12. Comparing this information with the
SS1SS2 91.7 Radarsat classification result as shown in Figure 18.6, the result
in Figure 18.12 overcomes the problems in Radarsat data and
provides a much improved classification result.
FIg u r e 18.11 Confusion matrix for land-cover change result using
The confusion matrixes for the 2009 and 2013 classification
direct classification between 2009 and 2013.
results with nine land-use/land-cover classes are illustrated in
The good change detection result using SVM-based direct Figure 18.13. The kappa coefficients of 0.79 for the 2009 map
classification can be used to infer the roles of human activi- and 0.78 for the 2013 map are calculated, implying that the 16
ties in influencing land-use/land-cover change. For example, classes of change/nonchange used in the change detection pro-
the conversion from BS to AG or from SS1 to BS (preparation cess described earlier were able to generate consistent classifica-
soils) shows the interference of human activities in the region. tions for 2009 and 2013. It is also interesting to observe that the
However, the conversions from CP to DP or SS1, from DP to SS1, change map has a higher kappa coefficient than the land-use/
or from AG to SS1 imply poor management of the land, resulting land-cover map for each year, showing the importance of object
in land degradation. correlation in time, which improves the overall accuracy.
724000 726000 728000 730000 732000 734000
9656000
9656000
9654000
9654000
9652000
9652000
9650000
9650000
724000 726000 728000 730000 732000 734000
(a)
FIg u r e 18.12 Land-use/land-cover map in 2009 (a), which are inferred from the change detection results using support vector machinebased
direct classification. (Continued )
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 393
724000 726000 728000 730000 732000 734000
9656000
9656000
9654000
9654000
9652000
9652000
9650000
9650000
724000 726000 728000 730000 732000 734000
(b)
N
AG CP PF SS1 WAT
Meters
BS DP IS SS2 0 1550 3100
FIg u r e 18.12 (c ontinued ) Land-use/land-cover map in 2013 (b), which are inferred from the change detection results using support vector
machinebased direct classification.
Reference 2009
Reference 2013
AG BS CP DP IS PF SS1 SS2 WT
CP DP IS PF SS1 SS2 WT
AG 71.4 13.7 AG 75.1
BS 28.6 92.8 BS 12.3 97.5
86.6 17.2 10.0 CP 65.0 11.2
CP
DP
Classified 2013
DP 13.3 51.2 14.9 88.8 33.6
Classified 2009
IS
IS 89.2 85.9
PF
PF 91.5 20.6 70.8
SS1 61.4 SS1 12.5 19.5 57.2
SS2 14.1 20.6 79.4 SS2 28.1 93.3
WT WT 98.4
92.1
(a) (b)
FIg u r e 18.13 Confusion matrices for land-cover maps in 2009 (a) and 2013 (b) developed from the results from support vector machinebased
direct classification.
2016 by Taylor & Francis Group, LLC
394 Remotely Sensed Data Characterization, Classification, and Accuracies
18.3.5 Summary of the case Study Thischapterprovides a new exploration for using optical sen-
sor EO-1 ALI and Radarsat data for land-cover change detec-
In the moist tropical regions, collection of the same optical sen- tion in the Brazilian Amazon using direct classification with
sor data at different dates is difficult due to cloud cover. Use the assistance of sound field survey data. More research is
of different sensor data, especially radar data, is needed but is needed to explore the techniques to effectively integrate mul-
challenging. Traditional techniques such as image differenc- tisensor data and multitemporal scales to provide accurate
ing and principal component analysis cannot be used for the change detection results.
change detection based on optical and radar data. Although
the postclassification comparison approach can be used, the
poor land-cover classification using pure radar data makes the Acknowledgments
change detection results very poor. The SVM-based direct clas-
sification used in this research provides a much better change The authors acknowledge financial support from the National
detection result with overall accuracy of 87.8%. One problem Council for Scientific and Technological Development (CNPq-
of the postclassification comparison technique is the feasibil- Brazil, grants 307666/2011-5, 401528/2012-0, and 211268/2013-
ity matrix dependency on the time span considered, which will 5) and facility support from the Center for Global Change and
make the whole process more complex. The direct classification Earth Observations at Michigan State University, United States,
approach can lead to better results because it is supervised not and the National Institute for Space Research (INPEBrazil).
only with change samples but also with nonchange samples. The LVD also thanks the Canadian Space Agency for providing
main drawback of this approach is the difficulty in obtaining a Radarsat-2 data.
sufficient number of reference samples for change classes from
the same region in different years to generate change regions of
interest. Another problem is that some important land-cover References
transitions may be missed without a sound field survey. This
Akiwowo, A. and Eftekhari, M. 2013. Feature-based detection
research provides a preliminary result of using optical sensor
using Bayesian data fusion. International Journal of Image
and radar data for change detection without exhaustive exami-
and Data Fusion, 4(4), 308323.
nation of different image-processing methods for extraction of
Alberga, V. 2009. Similarity measures of remotely sensed multi-
optimal variables and different algorithms for identifying the
sensor images for change detection applications. Remote
best change detection result. More research is needed to exam-
Sensing, 1, 122143.
ine how to integrate multisensor data to implement land-cover
Asner, G.P. 2001. Cloud cover in Landsat observations of the
change detection.
Brazilian Amazon. International Journal of Remote Sensing,
22, 38553862.
18.4 conclusions Baker, C., Lawrence, R., Montagne, C., and Patten, D. 2007. Mapping
wetlands and riparian areas using Landsat ETM+ imagery
Change detection is a comprehensive procedure that requires and decision-tree-based models. Wetlands, 26, 465474.
careful consideration for each step, as illustrated in Figure Barnsley, M.J., Moller-Jensen, L., and Barr, S.L. 2001. Inferring
18.1. When the study area and users needs are defined, selec- urban land use by spatial and structural pattern recogni-
tion of suitable variables and corresponding algorithms tion. In: Donnay, J.P., Barnsley, M.J., and Longley, P.A.,
becomes the most important aspect. Therefore, much research eds. Remote Sensing and Urban Analysis. Taylor & Francis
has concentrated on these topics for a long time. Overall, per- Group, London, U.K., pp. 115144.
pixel-based change detection techniques have been the most Bhagat, V.S. 2012. Use of remote sensing techniques for robust
common approaches. As high-spatial-resolution images are digital change detection of land: A review. Recent Patents on
now easily available, object-based methods are gaining atten- Space Technology, 2, 123144.
tion. On the other hand, the importance of rapidly updating Biging, G.S., Chrisman, N.R., Colby, D.R., Congalton, R.G.,
the land-cover dynamic change at national and global scales Dobson, J.E., Ferguson, R.L., Goodchild, M.F., Jensen,
leads to increasing application of coarse-spatial-resolution J.R., and Mach, T.H. 1999. Accuracy assessment of remote
satellite imagery such as MODIS. Subpixel-based methods are sensing-detected change detection. In: Khorram, S., ed.
needed to solve the mixed-pixel problem, but new techniques Monograph Series. American Society for Photogrammetry
are required to extract the fractional land-cover change in a and Remote Sensing (ASPRS), Bethesda, MD, p. 65.
large area. In order to make full use of the merits of different Borghys, D., Shimoni, M., and Perneel, C. 2007. Change detection
variables and algorithms, hybrid methods that can effectively in urban scenes by fusion of SAR and hyperspectral data.
integrate these features will become a new research trend in Proceedings of SPIE, the International Society for Optical
the near future. Another important research topic is the inte- Engineering, 6749, 2738.
gration of multisensor data for change detection. This is espe- Camps-Valls, G. and Bruzzone, L. 2009. Kernel Methods in Remote
cially valuable in moist tropical regions due to cloud cover. Sensing Image Processing. John Wiley & Sons, Hoboken, NJ.
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 395
Camps-Valls, G., Gmez-Chova, L., Muoz-Mar, J., Rojolvarez, Guild, L.S., Cohen, W.B., and Kauffman, J.B. 2004. Detection
J.L., and Martnez-Ramn, M. 2008. Kernel-based frame- of deforestation and land conversion in Rondonia, Brazil
work for multi-temporal and multi-source remote sensing using change detection techniques. International Journal of
data classification and change detection. IEEE Transactions Remote Sensing, 25(4), 731750.
on Geoscience and Remote Sensing, 46(6), 18221835. Houhoulis, P.F. and Michener, W.K. 2000. Detecting wetland
Chander, G., Markham, B.L., and Helder, D.L. 2009. Summary change: A rule-based approach using NWI and SPOT-XS
of current radiometric calibration coefficients for Landsat data. Photogrammetric Engineering & Remote Sensing, 66,
MSS, TM, ETM+, and EO-1 ALI sensors. Remote Sensing of 205211.
Environment, 113, 893903. Huang, C., Goward, S.N., Masek, J.G., Thomas, N., Zhu, Z., and
Chen, G., Hay, G.J., Carvalho, L.M.T., and Wulder, M.A. 2012. Vogelmann, J.E. 2010. An automated approach for recon-
Object-based change detection. International Journal of structing recent forest disturbance history using dense
Remote Sensing, 33, 44344457. Landsat time series stacks. Remote Sensing of Environment,
Collins, J.B. and Woodcock, C.E. 1996. An assessment of sev- 114, 183198.
eral linear change detection techniques for mapping forest Huang, W., Sun, G., Dubayah, R., Cook, B., Montesano, P., Ni,
mortality using multi-temporal Landsat TM data. Remote W., and Zhang, Z. 2013. Mapping biomass change after for-
Sensing of Environment, 56, 6677. est disturbance: Applying LiDAR footprint-derived mod-
Congalton, R.G. 1991. A review of assessing the accuracy of els at key map scales. Remote Sensing of Environment, 134,
classification of remotely sensed data. Remote Sensing of 319332.
Environment, 37(1), 3546. Hussain, M., Chen, D., Cheng, A., Wei, H., and Stanley, D. 2013.
Congalton, R.G. and Green, K. 2008. Assessing the Accuracy of Change detection from remotely sensed images: From
Remotely Sensed Data: Principles and Practice, 2nd edn. pixel-based to object-based approaches. ISPRS Journal of
CRC Press, Taylor & Francis Group, Boca Raton, FL, p. 183. Photogrammetry and Remote Sensing, 80, 91106.
Coppin, P., Jonckheere, I., Nackaerts, K., Muys, B., and Lambin, E. Ichii, K., Maruyama, M., and Yamaguchi, Y. 2003. Multi-temporal
2004. Digital change detection methods in ecosystem moni- analysis of deforestation in Rondnia state in Brazil using
toring: A review. International Journal of Remote Sensing, Landsat MSS, TM, ETM+ and NOAA AVHRR imagery and
25, 15651596. its relationship to changes in the local hydrological envi-
Coppin, P., Nackaerts, K., Queen, L., and Brewer, K. 2001. ronment. International Journal of Remote Sensing, 24(22),
Operational monitoring of green biomass change for for- 44674479.
est management. Photogrammetric Engineering & Remote Jensen, J.R. 2005. Introductory Digital Image Processing: A Remote
Sensing, 67, 603611. Sensing Perspective, 3rd edn. Prentice Hall, Upper Saddle
Deng, J.S., Wang, K., Deng, Y.H., and Qi, G.J. 2008. PCAbased River, NJ, p. 526.
landuse change detection and analysis using multi-tempo- Jensen, J.R. 2007. Remote Sensing of Environment: An Earth
ral and multi-sensor satellite data. International Journal of Resources Perspective. Prentice Hall, Upper Saddle River, NJ.
Remote Sensing, 29, 48234838. Kennedy, R.E., Cohen, W.B., and Schroeder, T.A. 2007.
Descle, B., Bogaert, P., and Defourny, P. 2006. Forest change Trajectory-based change detection for automated charac-
detection by statistical object-based method. Remote terization of forest disturbance dynamics. Remote Sensing of
Sensing of Environment, 102, 111. Environment, 110, 370386.
Du, P., Liu, S., Gamba, P., Tan, K., and Xia, J. 2012. Fusion of dif- Lambin, E.F. and Strahler, A.H. 1994. Change-vector analysis:
ference images for change detection over urban areas. IEEE A tool to detect and categorize land-cover change pro-
Journal of Selected Topics in Applied Earth Observation and cesses using high temporal-resolution satellite data. Remote
Remote Sensing, 5, 10761086. Sensing of Environment, 48, 231244.
Du, P., Liu, S., Xia, J., and Zhao, Y. 2013. Information fusion tech- Li, D. 2010. Remotely sensed images and GIS data fusion for auto-
niques for change detection from multi-temporal remote matic change detection. International Journal of Image and
sensing images. Information Fusion, 14, 1927. Data Fusion, 1(1), 99108.
Foody, G.M. 2002. Status of land cover classification accu- Li, G., Lu, D., Moran, E., Dutra, L., and Batistella, M. 2012a.
racy assessment. Remote Sensing of Environment, 80(1), A comparative analysis of ALOS PALSAR L-band and
185201. RADARSAT-2 C-band data for land-cover classification in
Foody, G.M. 2010. Assessing the accuracy of land cover change a tropical moist region. ISPRS Journal of Photogrammetry
with imperfect ground reference data. Remote Sensing of and Remote Sensing, 70, 2638.
Environment, 114, 22712285. Li, G., Lu, D., Moran, E., and SantAnna, S.J.S. December 14,
Groen, T.A., Fanta, H.G., Hinkov, G., Velichkov, I., Van Duren,I., 2012b. A comparative analysis of classification algorithms
and Zlatanov, T. 2012. Tree line change detection using and multiple sensor data for land use/land cover classifica-
historical hexagon mapping camera imagery and Google tion in the Brazilian Amazon. Journal of Applied Remote
earth data. GIScience & Remote Sensing, 49(6), 933943. Sensing, 6(1), 061706.
2016 by Taylor & Francis Group, LLC
396 Remotely Sensed Data Characterization, Classification, and Accuracies
Li, X. and Yeh, A.G.O. 1998. Principal component analysis of Negri, R.G., Dutra, L.V., and SantAnna, S.J.S. 2012. Support
stacked multi-temporal images for monitoring of rapid vector machine and Bhattacharyya kernel function for
urban expansion in the Pearl River Delta. International region based classification. IEEE International Geoscience
Journal of Remote Sensing, 19(8), 15011518. and Remote Sensing Symposium (IGARSS), Munich,
Lillestrand, R.L. 1972. Techniques for change detection. IEEE Germany, pp. 54225425.
Transactions Computers, C-21, 654659. Negri, R.G., Dutra, L.V., and SantAnna, S.J.S. 2014. An innovative
Lo, C.P. and Shipman, R.L. 1990. A GIS approach to land-use support vector machine based method for contextual image
change dynamics detection. Photogrammetric Engineering classification. ISPRS Journal of Photogrammetry and Remote
& Remote Sensing, 56, 14831491. Sensing, 87, 241248.
Lu, D., Batistella, M., and Moran, E. 2008a. Integration of Landsat Negri, R.G., SantAnna, S.J.S., and Dutra, L.V. 2011. Semi-supervised
TM and SPOT HRG images for vegetation change detection remote sensing image classification methods assess-
in the Brazilian Amazon. Photogrammetric Engineering & ment. IEEE International Geoscience and Remote Sensing
Remote Sensing, 74, 421430. Symposium (IGARSS), Vancouver, Canada, pp. 29392942.
Lu, D., Ge, H., He, S., Xu, A., Zhou, G., and Du, H. 2008b. Negri, R.G., SantAnna, S.J.S., and Dutra, L.V. 2013. A new con-
Pixel-based Minnaert correction method for reduc- textual version of Support Vector Machine based on hyper-
ing topographic effects on the Landsat 7 ETM+ image. plane translation. IEEE International Geoscience and Remote
Photogrammetric Engineering & Remote Sensing, 74(11), Sensing Symposium (IGARSS), Melbourne, Australia,
13431350. pp.31163119.
Lu, D., Hetrick, S., Moran, E., and Li, G. 2012. Application of time Olofsson, P., Foody, G.M., Stehman, S.V., and Woodcock, C.E.
series Landsat images to examining land use/cover dynamic 2013. Making better use of accuracy data in land change
change. Photogrammetric Engineering & Remote Sensing, studies: Estimating accuracy and area and quantifying
78(7), 747755. uncertainty using stratified estimation. Remote Sensing of
Lu, D., Li, G., and Moran, E. 2014. Current situation and needs of Environment, 129, 122131.
change detection techniques. International Journal of Image Pereira, L.O., Freitas, C.C., SantAnna, S.J.S., Lu, D., and Moran,
and Data Fusion, 5(1), 1338. E.F. 2013. Optical and radar data integration for land use
Lu, D., Li, G., Moran, E., Dutra, L., and Batistella, M. 2011. A and land cover mapping in the Brazilian Amazon. GIScience
comparison of multisensor integration methods for land- & Remote Sensing, 50(3), 301321.
cover classification in the Brazilian Amazon. GIScience & Perumal, K. and Bhaskaran, R. 2009. SVM-based effective land
Remote Sensing, 48(3), 345370. use classification system for multispectral remote sens-
Lu, D., Li, G., Moran, E., and Hetrick, S. 2013. Spatiotemporal ing images. International Journal of Computer Science and
analysis of land-use and land-cover change in the Brazilian Information Security, 6(2), 97105.
Amazon. International Journal of Remote Sensing, 34(16), Petit, C.C. and Lambin, E.F. 2001. Integration of multi-source
59535978. remote sensing data for land cover change detection.
Lu, D., Mausel, P., Batistella, M., and Moran, E. 2005. Land- International Journal of Geographical Information Science,
cover binary change detection methods for use in the 15(8), 785803.
moist tropical region of the Amazon: A comparative study. Reiche, J., Souza, C.M., Hoekman, D.H., Verbesselt, J., Persaud, H.,
International Journal of Remote Sensing, 26(1), 101114. and Herold, M. 2013. Feature level fusion of multi-temporal
Lu, D., Mausel, P., Brondizio, E., and Moran, E. 2004. Change ALOS PALSAR and Landsat data for mapping and monitor-
detection techniques. International Journal of Remote ing of tropical deforestation and forest degradation. IEEE
Sensing, 25, 23652401. Journal of Selected Topics in Applied Earth Observations and
Lu, D. and Weng, Q. 2007. A survey of image classification methods Remote Sensing, 6, 21592173.
and techniques for improving classification performance. Serra, P., Pons, X., and Sauri, D. 2003. Post-classification change
International Journal of Remote Sensing, 28(5), 823870. detection with data from different sensors: Some accuracy
Mas, J.F. 1999. Monitoring land-cover change: A comparison considerations. International Journal of Remote Sensing, 24,
of change detection techniques. International Journal of 33113340.
Remote Sensing, 20, 139152. Shi, W. and Hao, M. 2013. Analysis of spatial distribution pat-
Morisette, J.T. and Khorram, S. 2000. Accuracy assessment curves tern of change-detection error caused by misregistration.
for satellite-based change detection. Photogrammetric International Journal of Remote Sensing, 34(19), 68836897.
Engineering & Remote Sensing, 66(7), 875880. Singh, A. 1989. Digital change detection techniques using
Mountrakis, G., Im, J., and Ogole, C. 2011. Support vector remotely-sensed data. International Journal of Remote
machines in remote sensing: A review. ISPRS Journal of Sensing, 10, 9891003.
Photogrammetry and Remote Sensing, 66(3), 247259. Stow, D.A. and Chen, D.M. 2002. Sensitivity of multitemporal
Mubea, K. and Menz, G. 2012. Monitoring land-use change in NOAA AVHRR data of an urbanizing region to land-use/
Nakuru (Kenya) using multi-sensor satellite data. Advances land-cover change and misregistration. Remote Sensing of
in Remote Sensing, 1, 7484. Environment, 80, 297307.
2016 by Taylor & Francis Group, LLC
Change Detection Techniques Using Multisensor Data 397
Theodoridis, S. and Koutrombas, K. 2008. Pattern Recognition. Vicente-Serrano, S.M., Prez-Cabello, F., and Lasanta, T. 2008.
Academic Press, Burlington, MA, p. 984. Assessment of radiometric correction techniques in ana-
Thomas, N.E., Huang, C., Goward, S.N., Powell, S., Rishmawi, lyzing vegetation variability and change using time series
K., Schleeweis, K., and Hinds, A. 2011. Validation of North of Landsat images. Remote Sensing of Environment, 112,
American forest disturbance dynamics derived from Landsat 39163934.
time series stacks. Remote Sensing of Environment, 115, 119132. Volpi, M., Tuia, D., Bovolo, F., Kanevski, M., and Bruzzone, L.
Tian, J., Reinartz, P., dngelo, P., and Ehlers, M. 2013. Region- 2013. Supervised change detection in VHR images using
based automatic building and forest change detection on contextual information and support vector machines.
Cartosat-1 stereo imagery. ISPRS Journal of Photogrammetry International Journal of Applied Earth Observation and
and Remote Sensing, 79, 226239. Geoinformation, 20, 7785.
Treitz, P. and Rogan, J. 2004. Remote sensing for mapping and Walter, V. 2004. Object-based classification of remote sensing
monitoring land-cover and land-use change: An introduc- data for change detection. ISPRS Journal of Photogrammetry
tion. Progress in Planning, 61, 269279. and Remote Sensing, 58(34), 225238.
Tso, B. and Mather, P.M. 2009. Classification Methods for Webb, A.R. 2002. Statistical Pattern Recognition. John Wiley &
Remotely Sensed Data. Taylor & Francis Group, London, Sons, New York, p. 514.
U.K., p. 356. Wilson, E.H. and Sader, S.A. 2002. Detection of forest type using
Van Oort, P.A.J. 2007. Interpreting the change detection error multiple dates of Landsat TM imagery. Remote Sensing of
matrix. Remote Sensing of Environment, 108(1), 18. Environment, 80, 385396.
Vapnik, V., Golowich, S.E., and Smola, A. 1996. Support vector Yeh, A.G. and Li, X. 2001. Measurement and monitoring of
method for function approximation, regression estima- urban sprawl in a rapidly growing region using entropy.
tion, and signal processing. In: M.C. Mozer, M.I. Jordan, Photogrammetric Engineering & Remote Sensing, 67, 8390.
and T. Petsche, eds., Advances in Neural Information Zeng, Y., Zhang, J., van Genderen, J.L., and Zhang, Y. 2010.
Processing Systems 9. [S.l.] MIT Press, Cambridge, MA, Image fusion for land cover change detection. International
pp.281287. Journal of Image and Data Fusion, 1(2), 193215.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
VII
Integrating Geographic
Information Systems
(GIS) and Remote
Sensing in Spatial
Modeling Framework
for Decision Support
19 g eoprocessing, Workflows, and Provenance Jason A. Tullis, Jackson D. Cothren, David P. Lanter, Xuan Shi,
W. Fredrick Limp, Rachel F. Linck, Sean G. Young, and Tareefa S. Alsumaiti ....................................................................401
Acronyms and Definitions Introduction Historical Context Why Provenance in Remote Sensing
Workflows Selected Recent and Proposed Provenance-Aware Systems Conclusions and Research
Implications References
20 Toward Democratization of g eographic Information: g IS, r emote Sensing, and g NSS Applications
ine veryday Life Gaurav Sinha, Barry J. Kronenfeld, and Jeffrey C. Brunskill .................................................................423
Acronyms and Definitions Rethinking Geographic Information and Technologies in the Twenty-First
Century Toward Democratization of Geo-Information Technologies Discovering Remote Sensing, GIS,
and GNSS Applications in Our Daily Lives Democratization of Geo-Information: Circumscribing Issues and
Challenges Conclusion References
21 Frontiers of g IScience e volution, State of the Art, and Future Pathways May Yuan.................................................445
Acronyms and Definitions Introduction Evolution State of the Art Future Pathways and Concluding
Remarks References
22 Object-Based r egionalization for Policy-Oriented Partitioning of Space Stefan Lang, Stefan Kienberger,
Michael Hagenlocher, and Lena Pernkopf ................................................................................................................................457
Acronyms and Definitions Mapping Latent Phenomena Domain-Specific Regionalization Case
Studies Conclusion References
399
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
19
Geoprocessing, Workflows,
and Provenance
Jason A. Tullis
University of Arkansas
Jackson D. Cothren
University of Arkansas
Acronyms and Definitions ................................................................................................................ 401
David P. Lanter 19.1 Introduction ............................................................................................................................ 401
CDM Smith Working Definitions
Xuan Shi 19.2 Historical Context .................................................................................................................. 404
Digital Provenance in Remote Sensing and Geospatial Workflows Specifications and
University of Arkansas
International Standards for Implementation of Shared Provenance-Aware Remote Sensing
W. Fredrick Limp Workflows
University of Arkansas 19.3 Why Provenance in Remote Sensing Workflows ...............................................................412
Remote Sensing Questions That Only Provenance Can Answer Provenance and Trust in the
Rachel F. Linck Remote Sensing Process
University of Arkansas 19.4 Selected Recent and Proposed Provenance-Aware Systems .............................................415
General Approaches Earth System Science Workbench and
Sean G. Young ES3 MODAPS and OMIDAPS Karma Data Quality Provenance
University of Iowa
System VisTrails UV-CDAT GeoPWProv
Tareefa S. Alsumaiti 19.5 Conclusions and Research Implications ..............................................................................418
United Arab Emirates University References.............................................................................................................................................418
Acronyms and Definitions SOC Service-oriented computing
SOI Service-oriented integration
ACSM American Congress of Surveying and Mapping SQL Structured query language
API Application programming interface W3C World Wide Web Consortium
CDAT Climate data analysis tool WPS Web processing service
CI Cyberinfrastructure WSDL Web services description language
DBMS Database management system XML Extensible markup language
FGDC Federal Geographic Data Committee XSEDE eXtreme Science and Engineering Development
HMM Hidden Markov model Environment
ISO International Standards Organization
LIDAR Light detection and ranging 19.1 introduction
LULC Land use/land cover
MODAPS MODIS adaptive data processing system Integrated remote sensing and GIS-assisted problem solv-
NCDCDS National Committee for Digital Cartographic ing now supports a remarkable array of domains (e.g., food
Data Standards and agricultural security, climate change, forest management,
NSDI National Spatial Data Infrastructure heritage preservation, and urban and regional planning) and is
OGC Open Geospatial Consortium being configured in a great variety of technical means. Given
OpenMI Open modeling interface the sheerquantity of innovations reported in journals and books
PROV-DM PROV data model (including the Remote Sensing Handbook), any one expert
REST Representational state transfer may be keenly aware of only a fraction of the detailed remote
SOAP Simple object access protocol sensing and related geospatial methods available to address a
401
2016 by Taylor & Francis Group, LLC
402 Remotely Sensed Data Characterization, Classification, and Accuracies
given problem statement. Regardless of the remote sensing an increasingly data-intensive geo-cyberinfrastructure (Di etal.
application under study or review, some reliance (whether 2013a) renders comprehensive remote sensingassisted geo-
implied or reported) is always made upon the geoprocesses spatial workflow interpretations, comparisons, and knowledge
and workflows associated with any geospatial artifacts pro- transfers ever more difficult by experts alone. Furthermore,
duced. In the context of a specific geospatial decision support depending on the geospatial laboratory setting and the capabil-
artifact (e.g., a map of predicted crop yield in kg/ha), a record ities of a given research team, the actual digital methods linked
of the specific geoprocesses may be termed geospatial prove- to published materials may overlap significantly with previ-
nance (or lineage; see Section 19.1.1). This chapter explores how ously reported work, may offer similar results using a more or
remote sensingassisted geoprocessing and related GIS work- less computationally efficient means of problem solving, and/
flows have been or may be combined with digital provenance or may be idiosyncratic to individual skills and experience.
information in order to augment scientific reproducibility, In an integrated geoprocessing, workflow, and provenance
comparison, trust, or to otherwise improve remote sensing cycle, expert refinement of remote sensingassisted decision
assisted decision support. support knowledge may be augmented by software agents
Increasingly of interest in computer systems, digital prov- capable of automated exchange and recognition of innovation
enance has relatively early geospatial origins that date back to at (Figure 19.1).
least the 1980s (e.g., Chrisman 1983), with a definite resurgence Over the past 25years, various prototype forms of geospatial
around 2009 (e.g., Yue etal. 2010a). The early and expanded geo- provenance have been implemented in shared workflow envi-
spatial interest and connection to provenance are driven in large ronments, including those specialized for high-performance
part by the question of methodological innovation. For remote capabilities. In spite of the potential of these prototypes, single
sensing and GIS integration to best improve the quality of deci- user/workstation geoprocessing and workflow design continue
sion making tools across a range of applications and domains, it to be a dominant tradition with many active options (e.g., from
seems reasonable that, if possible, such innovation must first be Hexagon Geospatial, Exelis Visual Information Solutions, and
machine recognizable. Unfortunately, many geospatial decision ESRI). There is therefore a discrepancy between futuristic col-
support tools lack suitable means to even replicate their findings, laborative goals and the actual state of the art of remote sensing
and innovation reported is naturally bracketed by complex ques- assisted software. There are also variations in how provenance
tions of accuracy, fitness for use, and a variety of other quali- itself is defined, whether specifically in a remote sensing or
tative and quantitative metrics related to reliability and trust. geospatial-related forum, or more broadly in computer systems.
So, while there is broad conceptual agreement that machine- It therefore seems reasonable to report progress in terms of what
interpretable source and process history records are vital and the actual computational environments entail and which defini-
may even be scientifically transformative in the modern era, tions are implied.
questions remain unanswered on how provenance information
may simultaneously benefit multiple domains (including the
19.1.1 Working Definitions
geospatial domain), and what mechanisms for its digital capture
and exchange will most successfully convey those benefits. Though commonly understood in a broad remote sensing and
There are at least two good reasons to believe that even partial geospatial computation parlance, Wade and Sommer (2006)
success toward machine-interpretable geospatial process history define geoprocessing in the context of the many tools available
records will be rewarded. First, correct expert interpretation in one software platform (ESRIs ArcGIS) with an emphasis
of the full scope of relevant methods, procedures, algorithms, on input GIS datasets, operations performed, and associ-
and expert knowledge is subject to entropy and constitutes an ated outputs. More generically, its root, process, implies an
increasingly complex, even daunting companion to the twenty- instance of a computer program execution, and this is natu-
first-century big [geospatial] data (Hey etal. 2009). Second, as rally compatible with a geospatial/remote sensor data process-
remote sensing and other geospatial techniques are commu- ing software context. Of course, identical geospatial computer
nicated in the scientific literature, there is a well-known con- programs operating on identical input datasets may produce
tinuing expectation and scholarly requirement that previously different results as a function of additional configuration
published studies are carefully acknowledged for their relevant parameters. For example, raster-based geoprocessing tools in
achievements and/or limitations. Failure to increasingly harness ESRIs ArcGIS 10 platform allow for an environment setting
machine power on these two fronts (but to continue interpreta- called Snap Raster. This setting allows the user to specify the
tions by experts alone) is probably not a viable long-term option. spatial grid on which computations are made. In practice, use
In a related example from computer systems, Buneman (2013) of this parameter allows pixels in an output raster layer to be
notes that the underappreciated machine-managed provenance exactly aligned with another raster having the same cell size.
in software version control systems has helped prevent a total To a novice, the resulting subpixel geometric shift may seem
disaster in software engineering. inconsequential at the overview scale. However, remote sens-
It is clear that absent the kinds of methodological analyses ing experts know that when geoprocessing tools are chained
enabled in part through exchange of provenance information, together into a workflow (in the present context, a repeatable
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 403
Design
Ancillary, in situ, and remote sensor data
Geoprocesses, workflows, and metadata
Capture
Workflow output
Provenance
Innovate Store
Online access
Optimize
Specifications
Simplify
and standards
Geoprocessing,
Remote workflows, and
provenance cycle Software
sensing
agent
expert
Replicate Analyze
Open source scripting Execution time
Identical/comparable Comparison of
results workflows and
Curate geoprocesses
Deduplication
Recognize innovation
Conclude
Online replicable spatial decision support
Standardized provenance and metadata
FIg u r e 19.1 Integrated geoprocessing, workflows, and provenance may be conceptualized as a positive developmental cycle used to refine
remote sensing knowledge before decision support is communicated. Highlighted aspects of this cycle suggest a capacity of remote sensing experts,
in conjunction with software agents, to cooperatively capture, store, analyze, curate, replicate, and innovate remote sensingassisted decision sup-
port methods. (Artist image of WorldView-3, Courtesy of DigitalGlobe, 2014.)
sequence of geoprocesses of interest to a person or group), (e.g., archaeology, computer science, forestry, and geology) with
environment settings like Snap Raster can affect the logic of a usually overlapping definitions.
decision support conclusion. Computational definitions of provenance are more numer-
Provenance traces back to 1294 in Old French as a deriva- ous than in other domains, largely because of (1) the difference
tive of the Latin provenire, and while Merriam-Webster (2014) between concepts of digital records and actual digital records,
emphasizes provenance as a concept (e.g., ownership history of and (2) the variation in software environment such as a data-
a painting), Oxford University Press (2014) highlights the record base management system (DBMS) versus file-based process-
of such provenance (Moreau 2010). In the art domain where the ing (Moreau 2010). Understanding provenance within DBMS
term is very well established, provenance entails an artifacts queries requires more computationally detailed observations
complete ownership history, but ideally will also include artistic, than understanding provenance at a more generalized work-
social, and political influences upon the work from its creation flow level (where one step in the workflow may entail multiple
to the present day. There is an established research process for database queries). Various traditions further influence how
obtaining an artifacts trusted provenance, and the information provenance is viewed, for example, whether it is conflated with
is highly valued, particularly to authenticate real versus fraud- metadata or trust, two closely related but distinct concepts
ulent works (IFAR 2013; Yeide et al. 2001). As a related term, (Gil etal. 2010). Given the infrastructural importance of the web
provenance is now increasingly used in a broad range of fields in remote sensingassisted decision support, the following W3C
2016 by Taylor & Francis Group, LLC
404 Remotely Sensed Data Characterization, Classification, and Accuracies
Provenance Incubator Groups working definition of provenance lineage information in quality records would be subject to
(in a web resource context) carries significant weight: entropy or fragmentation as a result of continuous GIS main-
tenance. He described reliability diagrams (for intelligence and
Provenance of a resource is a record that describes enti- other reliability-sensitive applications) embedded with lineage-
ties and processes involved in producing and delivering or related geometry and attributes (e.g., polygons identifying
otherwise influencing that resource. Provenance provides specific aerial photographic sources) and recommended them
a critical foundation for assessing authenticity, enabling to be incorporated in typical GIS design. While not typically
trust, and allowing reproducibility. Provenance assertions portrayed as lineage or provenance today, this type of lineage-
are a form of contextual metadata and can themselves related geodata, such as DigitalGlobe image collection footprints
become important records with their own provenance. accessible in Google Earth, is extremely useful for visualization
purposes and may resist digital entropy due to established geo-
g il etal. (2010)
data interoperability.
Beyond the challenges presented by digital records of lineages
It should be noted that while provenance and lineage are here
for multiple geodata versions, Langran and Chrismans (1988)
used interchangeably, one can argue that there are subtle differ-
emphasis on multitemporal GIS highlighted additional record
ences in their meanings. Process history seems to fit more easily
complexity that would be required. Nyergess (1987) discussion
with the many definitions attributed to provenance, and lineage
on geodata exchange implied that quality metadata (including
implies a kind of genealogy or data pedigree record relative to a
lineage information) could eventually facilitate geoprocessing
remote sensingassisted decision support artifact. While these
design (workflows) with the two being mutually dependent.
semantic differences are not a point of the present focus, each
Others including Grady (1988) reasoned that lineage need not
word will appear in its historical context (beginning with lin-
only support records of data quality but could in turn be used
eage). Also, a number of surveys have been conducted on prove-
to record societal mandates (e.g., legislative drivers of geodata
nance including some with a geoprocessing and workflow flavor.
development) in the lineage information. While the existence
For example, Yue and He (2009) provide a review covering vari-
of these additional complexities and potential requirements for
ous aspects of geospatial provenance. For a broader perspective,
geospatial lineage/provenance did not thwart attempts to forge
Bose and Frew (2005) provide a review covering provenance
ahead with possible software solutions, they pointed to signifi-
in geospatial as well as other domains. More recently, Di etal.
cant challenges.
(2013b) provide an overview of geoscience data provenance.
19.2.1 Digital Provenance in Remote
19.2 Historical context Sensing and Geospatial Workflows
The earliest work in geospatial lineage was spurred in the Over the last few decades and especially in the last 5years, there
United States through the formation of the National Committee has been significant attention given to understanding lineage/
for Digital Cartographic Data Standards (NCDCDS) by the provenance in computer systems, and a variety of formalisms
American Congress of Surveying and Mapping in 1982 (Bossler have been developed to understand their role in scientific work-
etal. 2010). In 1988, chaired by Dr. Harold Moellering from Ohio flows (e.g., Bose and Frew 2005; Buneman and Davidson 2010;
State University, the NCDCDS proposed five fundamental com- Hey etal. 2009; Simmhan etal. 2005). In the following text, we
ponents of a geospatial data quality report, including (1) lineage, highlight pioneering digital advances with geospatial lineage
(2) positional accuracy, (3) attribute accuracy, (4) logical consis- (circa 1990s) and more recent geo-cyberinfrastructure advances
tency, and (5) completeness. The NCDCDS described lineage in in provenance (circa 2000s to present).
detail, which they presented as the first quality component. Less
than a third of their description for lineage follows (Moellering 19.2.1.1 Pioneering Work in Geospatial Lineage
etal. 1988, p. 132): As Chrisman (1986) suggested, evaluation and judgment of fit-
ness of use must be the responsibility of the user, not the pro-
The lineage section of a quality report shall include a ducer. To carry out this responsibility, the user must be presented
description of the source material from which the data were with much more information to permit an informed decision
derived, and the methods of derivation, including all trans- (p. 352). Moellering et al. (1988) later emphasized producers
formations involved in producing the final digital files. The obligation to first document and update the lineage of their data
description shall include the dates of the source material in order to trace all the work (whether analog or digital) from
original source materials through the intermediate processes to
As geospatial workflows began to transition from analog to final digital output. It became obvious that both GIS software
digital environments, it became clear that lineage-implied geo- and international standards would be needed to facilitate the
processes would need to be tracked from their origins, through development and the maintenance of such records.
revisions to the data, and finally to the output (Moore 1983). An early version of ESRIs ARC/INFO Geographic Information
Chrisman (1983) noted that unfortunately over its lifetime, System featured a LIBRARIAN module capable of capturing
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 405
and querying some aspects of geospatial lineage. Using the
modules CATALOG command, a database administrator could
retrieve information on map production status as well as review Source
time stamps and coordinates of recent map updates (Aronson I
and Morehouse 1983). In the mid-1980s, the U.S. Geological Derived
Survey (USGS) began development of a GIS-linked automated n
cartographic workflow system called Mark II with partial lin- m
eage capabilities. An important part of Mark IIs design was its Or (i.e., is either one,
capacity to track the location (e.g., network address) of datas- the other, or both)
ets and their progress from curated archive toward final map
products (Anderson and Callahan 1990; Guptill 1987). While it
Transformed Intermediate Product
was envisioned this system would play a key role in fulfilling the
National Mapping Programs mission through 2000, the agency
focus transitioned by the mid-1990s toward GIS data develop-
ment including the National Map. The first reported develop-
ment of a system to specifically and directly address geospatial
FIg u r e 19.2 Relationship among source and derived datasets,
lineage was David Lanters Geolineus project commenced in the where each instance of the latter may be either an intermediate or prod-
late 1980s as part of his doctoral research at the University of uct dataset, or both.
South Carolinas Department of Geography (Lanter 1989). As
the prototype pioneering work in geospatial lineage/provenance, and provided output datasets child links connecting them back
this is reviewed in detail with added explanation. to their input datasets (Who am I the child of?). Each parent-
Lanter invented a method and means to capture, structure, and-child relationship was defined as an ordered pair of input
and process geospatial lineage to determine and communicate and output datasets. Lanters parent relationship identified the
the meaning and integrity of the contents of a GIS database derived output given a source or derived input dataset, while
(Lanter 1993a). His metadata and processing algorithms track his child relationship would identify a derived or source dataset
and document remotely sensed and other geodata sources and when given an output dataset.
analytic transformations applied to them to derive new datasets. Child links connecting output datasets to their inputs
In addition to differentiating between source and derived data- enabled automatic deduction of which datasets within an ana-
sets, Lanter further distinguished intermediate and product- lytic database are sources and which are derived (Lanter 1993b).
derived datasets. More concisely, let Derived datasets are connected to their inputs by child links,
while sources lack such links. Lanter defined his child opera-
= =
Datasets {Dataset i :i source, derived}, tor to take a derived dataset, access its child links, and identify
inputs used to create it. His Ancestors algorithm applied the
child operator and by a recursive function traced the child links
Dataset derived = {Dataset derived.k : k = intermediate, product}.
to identify datasets used to create a derived dataset, including
any sources in the geoprocessing application. Lanter defined
Source datasets can be the results of insitu sampling and data the parent operator to take a source or derived dataset as input,
collection, remote sensing, or ancillary data (e.g., digitization and access and traverse its parent links to identify all the out-
of maps, or thematic data resulting from digital processing of puts derived from it. His Descendants function recursively
remotely sensed data). Initially, only source datasets are avail- traced parent links and identified all datasets derived from a
able for geoprocessing and transformation into a derived dataset source or other derived input dataset used within a geoprocess-
(Figure 19.2; n 1, m = 0). Later, new datasets can be generated ing application.
exclusively from derived datasets (n = 0, m 1) using spatial Classification of datasets into source, intermediate, and prod-
analysis transformations such as reclassification, distance mea- uct paved the way to structuring additional lineage metadata
surement (buffering), connectivity, neighborhood characteriza- attributes. Lanter used the artificial intelligence frame data
tion, and summary calculations. Alternatively, new datasets can structure to organize knowledge about the metadata properties
be derived from inputs that include sources, derived, or both of source, intermediate, and product dataset types. Each source
(n + m > 1) using multi-input transformations such as arithme- dataset was provided a frame for storing source properties such
tic, statistical, and logical overlays, as well as drainage network as its name, feature type(s), date(s), responsible agency, scale,
and viewshed determinations. projection, and accuracy attributes. He provided each derived
Lanter classified datasets into source, intermediate, and dataset with a frame for storing detailed metadata elements
product types (Figure 19.2), and related them to one another as about where it is physically stored, the command applied to
inputs and outputs of each data processing step of an analytical derive it, the commands parameters, who derived it, and other
application. He gave input datasets parent links pointing to out- aspects of its derivation. Lanter saw products as derived datasets
put datasets they were used to create (Who am I the parent of?) that were provided an additional frame for metadata detailing
2016 by Taylor & Francis Group, LLC
406 Remotely Sensed Data Characterization, Classification, and Accuracies
the analysis goal the dataset was intended to meet, intended Lanter adapted the Ancestors function to respond to lineage que-
audience/users of the dataset, when it was released, etc. ries and to report on data sources and the sequence of process-
More formally, each Dataseti (i.e., source, intermediate, or ing (i.e., data lineage) applied to sources and intermediates to
product) was provided an ordered list of metadata properties, Aj, derive a target dataset (Lanter 1991). He integrated the Ancestors
such that Aj = {Aj1,Aj2,,Ajk = f(i)}. Specifically, function with a rule-based processor that checked the inputs
of each user-entered GIS command, determined their related
Dataset source Asource = {Name, Features, Data, Scale, Projection, sources, and evaluated their metadata to detect and warn users
Agency , Accuracy ,}, when they were entering commands that would otherwise com-
bine datasets of incompatible properties such as projections,
Dataset intermediate Aintermediate = {Name, Command, scales, and dates (Lanter 1989). Lanter subsequently modified
User, Date,},
Parameters,U the Descendants function to automatically generate and run
GIS scripts and propagate new source data to update dependent
and intermediates and products (Lanter 1992a).
Dataset product Aproduct = {Goal, Audience, Release Date, 19.2.1.1.1 Geolineus
Intended Use,}. Lanter and Essinger designed the lineage diagram, an icon-
based flowchart graphical user interface (GUI), to enable users
Given w {source, intermediate, product}, m a metadata prop- direct interaction with lineage metadata to understand, modify,
erty of w, and awm a value of Awm then a datasetw = (aw1,aw2,,awl). and maintain their analytical applications and ESRIs ARC/
Lanters lineage metadata structure represented datasets as INFOs spatial data contents (Essinger and Lanter 1992; Lanter
nodes coupled with source, command, and product properties, and Essinger 1991), and implemented it in Geolineus (Lanter
and connected them with parent and child links (Figure 19.3). 1992b)the first lineage-enabled geospatial workflow system.
Source Source Source
Source Source Source
properties properties properties
Parent Child
Parent Parent
Child Child
Derived Derived
intermediate intermediate
Command Command
properties properties
Parent Child
Derived
intermediate
Command
properties Parent
Child
Child
Parent
Derived
product
Command
properties
Product
properties
FIg u r e 19.3 Lineage represented as structured metadata consisting of parent and child links connecting source, intermediate, and product
datasets. While each source possesses a frame containing metadata properties, frames for derived datasets detail the GIS command used in its
creation. In addition, derived product datasets possess a frame describing analytic goals, release date, and users.
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 407
Geolineus enabled users of ESRIs ARC/INFO and GRID (for identities of the input and output datasets and the command and
image processing) to capture, create, save, exchange, analyze, its parameters, and pass the command off to the geospatial pro-
and reuse lineage metadata to maintain their GIS databases. cessing software running in the background. Geolineus moni-
Geolineuss user interface included a lineage data flow diagram tored the processing and feedback messages returned from the
within one panel, coupled with another panel containing its own geospatial processing software and presented them to the user
command line processor in place of the command line proces- within its own command line window in real time to provide the
sor of ARC/INFO and GRID. As users added source datasets, user with the illusion that they were interacting directly with the
they were presented with a form to document them, after which geospatial software. The system detected whether the processing
they were displayed along the top of the data flow diagram, each successfully completed and, if so, the input/output relationships
with a square icon with a bar at its top. Symbols within the icons and command information would be stored within its metadata-
would identify if the dataset contained points, lines, polygons, base and the data flow diagram dynamically updated with a new
raster grids, and/or value attribute tables. Icons further down icon for the output dataset connected by dotted arrows (labeled
the flowchart represent datasets derived with geospatial analy- with the command) emanating from its data sources. When the
sis operations such as CLASSIFY, BUFFER, and INTERSECT. final data product was reached, the user could click on its icon
Geolineus would create icons and arrows connecting them to the and fill in the displayed product form to document the analytical
flowchart automatically as these commands were used. Icons at goal it represented (e.g., wells at risk from nearby leaking pipes)
the bottom of the flowchart signifying products, that is, derived and who should be contacted if it was updated or changed.
datasets that represent the final step in the geospatial applica- Geolineus also monitored each dataset in the diagram to
tion, each included a bar along its bottom edge (Figure 19.4). determine if it was edited or replaced. If a source or derived data-
Written in Common LISP, Geolineus used multiprocessing set was found to be modified, its icon would turn yellow in the
capabilities of UNIX to run the geospatial processing software diagram. If the dataset needed its topology rebuilt in response
as a background job while providing its own command line to an edit, the polygon or line feature symbol within the icon
window to the user. As the user would enter a command trans- would turn red. If a derived dataset was potentially out of date
forming one or more spatial datasets to derive a new one (e.g., because one of the sources it was derived from was edited, its
classify, union, and intersect), Geolineus would parse, extract the icon would turn orange. Users could click on a source icon to
Geolineus GUI
Visualization of lineage diagrams
Frames for dataset and command metadata
Command propagation or Ripple
Lineage and workflow execution are linked
Allow easy source and intermediate updates
FIg u r e 19.4 Examples of Geolineus interactive lineage diagram GUI. The left screen shot illustrates linkage of a command frame to the
BUFWELLS dataset highlighted in black; clicking on the Ripple button at the bottom of the command frame propagated changed buffer command
parameters throughout the workflow. The right screen shot illustrates the LANDUSE2 datasets source frame, and commands applied to that
source to derive the COV1 and COV2 datasets.
2016 by Taylor & Francis Group, LLC
408 Remotely Sensed Data Characterization, Classification, and Accuracies
view metadata about what it represented, when it was created, created a lineage diagram reflecting the commands contained in
where it camefrom, and cause a propagation (ripple) of its data the log file of a targeted workspace. While the log files contained
through sequences of commands updating intermediate and the name of the dataset and the file system path indicating where
product datasets originally derived and created from it. Users the dataset was stored, they did not include other source meta-
could also click on a derived dataset to rerun the commands nec- data (i.e., thematic feature type, date, agency, scale, projection
essary to pull new, updated, and modified source data through the accuracy, etc.) necessary for achieving a clear understanding of
flowcharts processing logic and update the derived geodata (Lanter contents and qualities of each source. To resolve this, analysts
1994b). Geolineus enabled users to save, exchange, and import lin- and auditors working with Geolineus clicked on the source icons
eage metadata in ASCII file format to meet the Federal Geographic within the lineage diagram, brought up source frames, and filled
Data Committees (FGDCs) Content Standard for Digital in missing source metadata if available.
Geospatial Metadata, document exchanged datasets, accompany Lanter (1994a) formulated metadata comparison functions
source datasets, provide logic for use within other instances of the that enabled him to automatically determine if two spatial ana-
software to reconstitute a derived geospatial database, and serve as lytic datasets were equivalent and if two geospatial datasets were
reusable analytic application logic templates to snap to replacement similar. These were implemented within Geolineus to identify
source datasets associated with different study areas. common and unique geospatial data processing conducted in
Lanter and Veregin (1991, 1992) modified the lineage meta- and among multiple GIS workspaces (Lanter 1994b). His search
data to store error measures and demonstrated new algorithms for datasets common to different lineage metadata representa-
for mathematically modeling how error measures of data sources tions began with a determination of source equivalence. Source
are transformed and combined through a sequence of spatial datasets were considered equivalent when their source meta-
analysis functions to determine the quality of a derived spatial data properties were found to have equivalent values, assuming
analytic product dataset. They added properties to the source these properties are sufficient to uniquely identify their contents
frame for storing user-entered measures of data source error, and qualities. This enabled the detection of equivalent and pos-
and properties to the command frame for storing derived error sibly redundant source datasets that are stored in different file
measures for each derived dataset. Geolineuss Ancestors and system locations but contain equivalent content. Source data
Descendants functions were modified, enabling them to access equivalence was implemented in Geolineuss Merge func-
error properties of input datasets, select and apply an appropriate tion, which enabled users to analyze log files of data processing
error propagation function to derive, store and present the error applications run in different workspaces and produce a single
measure of the derived geospatial dataset as the user typed in unified lineage diagram illustrating their common and unique
their spatial analysis commands. Lanter (1993b) followed this by data sources (Figure 19.5).
modifying the lineage metadata and Ancestors and Descendants In turn, Lanter considered derived datasets equivalent when
functions to use commercial costs of data storage and central pro- (1) their input datasets were equivalent and (2) when transfor-
cessing time to calculate and compare the relative costs of stor- mations applied to compute them from their inputs were found
ing versus using lineage metadata to re-derive intermediate and to be equivalent. Derived data equivalence was implemented in
product datasets when needed. The results enabled Geolineus to Geolineuss Condense function. Condense enabled Geolineuss
determine an optimal spatial database configuration and choose users to detect the lineage representations of redundant process-
which datasets to delete and re-create when needed. Veregin and ing and resulting copies of derived data stored under different
Lanter (1995) modified the metadata frames and Ancestors and names or in different file system locations, remove the redundant
Descendants functions to demonstrated lineage metadata-based data, and consolidate the transformational logic applied in their
error propagation techniques for identifying the best data source derivation within the unified metadata and lineage diagram.
to improve based on cost value per product quality improvement Lanter and Surbey (1994) put Geolineuss capabilities to work
achieved. Geolineus was programmed to systematically vary the in the first enterprise GIS database and geoprocessing quality
error value of each source, iteratively applying mathematical audit. They systematically evaluated the geospatial data sources,
error propagation functions and determining its effect on prod- products, and geoprocessing applied to derive 40 GIS data prod-
uct quality. Comparing slopes of lines graphing source error ver- ucts, developed within 14 projects, for eight departments of a
sus resulting product enables determination of relative impact large southwestern electric utility. Lanter and Surbey identi-
each data source has on data product quality. fied 54 data sources among the 806 raster (GRID) and vector
To help analysts and auditors understand undocumented (ARC/INFO) GIS datasets produced for the electric utilitys
preexisting analytically derived GIS datasets, Lanter provided decision makers. They interviewed the departments GIS spe-
Geolineus with capabilities to extract lineage metadata and cre- cialists, filled in as much missing source metadata that could
ate a lineage diagram from ARC/INFO log files. Similar to the be recalled and confirmed, and noted findings about what was
history list the UNIX operating system recorded user commands unknown about the source data. In addition to assessing ade-
into, the ARC/INFO GIS copied user-entered GIS commands quacy of source data documentation, Lanter and Surbey ana-
into log files, which it stored and maintained within the operating lyzed the resulting lineage diagrams they created and measured
system file system directories or workspaces. Geolineuss Create the complexity of spatial analysis logic employed within the 14
from log option automatically extracted lineage metadata and GIS application projects.
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 409
enable interoperability, data recovery, process history records,
and data version monitoring in commercial GIS (Alonso and
Hagen 1997). Geo-Opera was based on a modular architecture
composed of interface, process, and database modules. It used
its own process scripting language and was based on the OPERA
distributed operating system that allowed for data distribution
and process scheduling within a local area network. In Geo-
Opera, geodata first had to be registered before being utilized,
thus mitigating the common problem (that persists today) of
lack of source metadata.
19.2.1.2 expansion of Limited Provenance in
commercial and Public Geoprocessing
As commercial and public (i.e., free and open-source) GIS appli-
cations rapidly matured and grew in analytical power, it became
necessary to provide a way for users to build and track workflows
involving interactions among many complex and varied geopro-
cessing operations. At least two approaches to create and man-
FIg ur e 19.5 Geolineus GUI illustrating the results of the Merge
age workflows have emergedgraphical block programming
function unifying two lineage diagrams at their common source LOTS
dataset, and Condense function which removed redundant process-
and integrated database style querying. The first is essentially
ing and derived datasets unifying them at their common intermediate a visual interface to programming, while the second approach
ONELOT dataset. The red mark on the yellow LOTS dataset indicated appeals to users trained in database management. While both
an edit and need for polygon topology repair, and the orange color in enable at least some form of provenance, enterprise database
derived datasets reflected the need for changes to be propagated using systems can provide record-level transaction management,
the Ripple function to update the OUTPUT and FINALCOV products. which, at least in detail, is beyond the scope of this chapter.
By far the most common approach, due in large part no doubt
Lanter (1994a) extended his dataset equivalence tests and for- to its ease of use and graphic nature, is graphic block program-
mulated a set of source and derived data similarity tests in order ming approach. Commercial GIS and remote sensing applica-
to detect patterns of data usage and derivations within work- tions such as ESRIs ArcGIS, Hexagon Geospatials ERDAS
flows. He coupled these with a geospatial data taxonomy (e.g., IMAGINE, and PCIs Geomatica expose their complex process-
Anderson etal. 1976) and a GIS command language taxonomy ing tools in this way (e.g., Figure 19.6). The free and open-source
(e.g., Giordano etal. 1994) to generalize analytic logic employed GRASS GIS also provides a visual programming environment
within prior applications of GIS and find common data analy- for both vector and raster operations. Boundless (formerly
sis patterns. Lanter presented a suite of lineage-based metadata OpenGeo) is developing a visual programming environment for
analysis methods for detecting and communicating common- QGIS, an open-source GIS. All of these environments capture
alities and differences among particularly useful spatial analy- and store some degree of provenance including in some cases
sis applications, with the intent of improving geographers basic important environmental settings that can significantly affect
understandings of spatial analytic reasoning and to provide a geoprocessing results. It is important to note that visual pro-
method and means to answer fundamental geographic ques- gramming interfaces can normally be bypassed by skilled users
tions including the following: familiar with application programming interfaces (APIs) or
scripting languages integrated with GIS.
Are there a finite number of spatial relationships studied
A less common approach is incorporated almost exclusively
within and among different GIS applications areas? If so,
in enterprise databases that have integrated spatial operators
what are they?
and native spatial data objects. With this level of integration,
Within particular applications areas, are certain spatial
spatial operators become just another type of operation exposed
relationships stressed more than others? If so, what are they?
through (often extended) structured query language (SQL) inter-
Are common patterns of analytic logic used to build up
faces. At a minimum, the SQL commands used to manipulate
certain complex spatial relationships? If so, what are they?
spatial data objects are recorded and may be inspected in a vari-
Are certain spatial relationships consistently sought at dif-
ety of graphical environments. As of mid-2014, most spatially
ferent spatial, thematic, and temporal scales?
enabled databases have extensive vector operators but limited
raster or image operators of particular interest in remote sensing
19.2.1.1.2 Geo-Opera workflows. However, technologies such as the Oracle Spatial and
Also incorporating geospatial lineage into its design in the Graph option for Oracle Database 12c now enable image algebra
1990s, Geo-Opera was developed as a prototype geoprocessing in addition to other remote sensingoriented capabilities such as
support or geospatial workflow management system that would LIDAR data processing.
2016 by Taylor & Francis Group, LLC
410 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 19.6 Geospatial scientists interact with the ASA Hazard Map (Tullis etal. 2012), a remote sensingassisted silviculture assessment
spatial decision support system, and its five downloadable ArcGIS 10 ModelBuilder workflows using a collaborative multitouch display. Each
yellow rectangle represents an ArcGIS tool (e.g., for estimating incoming solar radiation using a LIDAR-derived DEM) and, together with inputs,
outputs, and other parameters (colored ovals), constitutes a geoprocess. After execution, geoprocesses are marked with shadows that may be
cleared only by resetting or changing geoprocess parameters including geoprocessing environment settings. User interaction with shaded geo-
processes effectively provides access to workflow-level provenance information for the most recent execution and facilitates dependent geoprocess
updates after any modifications are made.
19.2.1.3 interest in Provenance as a component can be deployed for remote procedure calls. Furthermore, with
in Geo-cyberinfrastructure the advancement of telecommunication infrastructure and tech-
Cyberinfrastructure (CI) is a concept that has been exten- nology, wireless networking has been providing another approach
sively used since Atkins et al. (2003) Revolutionizing Science for data sharing and network computing, while varieties of sensor
and Engineering through Cyberinfrastructure: Report of the networks can be connected through wireless networks.
National Science Foundation Blue-Ribbon Advisory Panel on Today, different computing networks can be linked together.
Cyberinfrastructure. As a common infrastructure for scientific Supercomputers on the XSEDE can be accessed through a web
data and computing, a variety of components and topics are portal, while wireless sensor networks can be accessed on the
involved in CI construction, including hardware, software, net- internet. Such a huge but heterogeneous CI increases the diffi-
work, data, and most importantly people. The development of culty and complexity for geoprocessing, workflows, and prov-
CI can be traced back to the construction of the TeraGrid infra- enance research (Wang et al. 2008). In 2007, the National Science
structure in the 1990s that was replaced by eXtreme Science Foundation (NSF) released the DataNet program that would sup-
and Engineering Development Environment (XSEDE) in 2012. port comprehensive data curation research over the CI, and NSFs
By linking supercomputers through high-speed networks, Data Infrastructure Building Blocks program will support devel-
TeraGrid and XSEDE have provided a powerful computing envi- opment and implementation of technologies addressing a subset
ronment and capability to support petascale to exascale scien- of elements of the data preservation and access lifecycle, includ-
tific computation. ing acquisition; documentation; security and integrity; storage;
In a broader and general domain, the internet can be regarded access, analysis and dissemination; migration; and deaccession,
as the CI since all computers can be linked together through the as well as cybersecurity challenges and solutions in data acquisi-
network. When varieties of data and databases can be hosted tion, access, analysis, and sharing, such as data privacy, confiden-
and connected on the internet, data processing and analytics tiality, and protection from loss or corruption (NSF 2014), which
can be conducted through service-oriented computing (SOC). are all topics relevant to the themes in provenance.
In early 2000, web service technology was proposed to be the
solution for software interoperability. In this vision of interop-
19.2.2 Specifications and international Standards
erable software engineering and integration, a service is an API
for implementation of Shared Provenance-
defined in Web Services Description Language, while communi-
Aware Remote Sensing Workflows
cation between the service provider and the service requester is
based on the Simple Object Access Protocol (SOAP). Meanwhile, Since the Moellering etal. (1988) proposal identifying geospatial
Representational State Transfer (REST) services are based on lineage as the first component in a data quality report, a variety
HTTP protocol using its GET/POST methods for mashup online of provenance-related standards have been developed including
resources (Fielding 2000). Both SOAP- and REST-based services those at the international level. The most current standard in use
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 411
is the International Standards Organizations ISO 19115-2, which recommendations (starting with the term PROV) to guide the
has been endorsed by the Federal Geographic Data Committee provenance interchange on the web. Specifically, the current
(FGDC; ISO 2009). PROV data model for provenance (PROV-DM; Moreau and
Missier 2013) defines a core data model for provenance for
19.2.2.1 Metadata interchange Standards building representations of the entities, people and processes
In the United States, the FGDC has been coordinating the involved in producing a piece of data or thing in the world (Gil
development of the National Spatial Data Infrastructure by and Miles 2013).
developing policies and standards for sharing geographic data. To illustrate PROV-DM in a remote sensing and geoprocess-
The Content Standard for Digital Geospatial Metadata defines ing context, the provenance of a 20012006 canopy change
common geospatial metadata about identification information, layer incorporated in the ASA Hazard Map (Jones etal. 2014;
spatial reference, status information, metadata reference infor- Tullis etal. 2012) can be represented using PROV-DM struc-
mation, source information, processing history information, tures. This may be encoded (Figure 19.7; Table 19.1) as agents
distribution information, entity/attribute information, and con- (e.g., a specific version of PCI Geomatica as a software agent),
tact information of the geodata creator. entities (e.g., a Landsat image clipped to a forest boundary),
Partially based on the FGDCs 1994 metadata standards, activities (e.g., ATCOR2 atmospheric correction based on
the ISO Technical Committee (TC) 211 published ISO 19115 specific calibration and other parameters), and relationships
Metadata Standard, covering a conceptual framework and (e.g., wasInfluencedBy to represent the influence of Wang etal.
implementation approach for geospatial metadata generation. (2007) on the change detection methodology). It is impor-
ISO/TC 211 suggests that metadata structure and encoding tant to note that PROV-DM is extensible such that subtypes
are implemented based on the Standard Generalized Markup of agents, entities, activities, and relationships can be identi-
Language that has the same format as the Extensible Markup fied as needed for domain-specific applications (Moreau and
Language (XML). The XML-based ISO metadata standard has Missier 2013).
exemplified the advantage in implementation covering a vari- In the geospatial domain, the efforts of the Open Geospatial
ety of elements in standard definition. ISO 19115 Metadata Consortium (OGC) initially (late 1990s and early 2000s) focused
Standards contain a data provenance component in defining the on the development of specifications that encouraged geo-
data quality within the metadata. Unfortunately, while Gil etal. spatial data interoperability such as the OGC Simple Features
(2010) defined provenance in part as a form of contextual meta- Specification. While not directly related to provenance, this
data, their emphasis on the clear distinction between prov- effort has led to common ontologies and semantic structures that
enance and traditional metadata is not reflected in metadata are foundational to the integration of geoprocessing, workflows,
interchange standards for provenance. For instance, geodata and provenance. In the 2000s, the OGCs attention shifted to
cardinality between a land use land cover (LULC) map and its web processing and interoperability of various web services. The
metadata is one to one; in contrast, geodata cardinality between OpenGIS Web Processing Service (WPS) specification (Schut
an LULC map and its provenance is potentially one to many, 2007) has a lineage element in defining the request message to
thus leading to much duplicate information in a provenance as execute the spatial operation. In case lineage is defined as true,
metadata paradigm. the response message from WPS will contain a copy of input
parameter values specified in the service request definition. The
19.2.2.2 Provenance-Specific (non-Metadata) OGC also developed the Sensor Web Enablement standard, in
interchange Standards which the OpenGIS Sensor Model Language (Botts and Robin
Provenance-specific (non-metadata) standards have been devel- 2007) has one specific element that documents the observation
oped at different levels and in a variety of domains. ISO 8000 lineage to describe how an observation is obtained. Elements of
has a series of standards that address data quality. ISO 8000-110 a number of earth observation process specifications, such as the
specifies requirements that can be checked by computer for the Catalogue Services Standard 2.0 Extension Package for ebRIM
exchange, between organizations and systems, of master data Application Profile: Earth Observation Products (Houbie and
that consists of characteristic data. It provides requirements Bigagli 2010), the Sensor Observation Service Interface Standard
for data quality, independent of syntax. ISO 8000-120 speci- (Brring etal. 2012), and others, increasingly have provenance-
fies requirements for capture and exchange of data provenance related components as key elements. The more recent devel-
information and supplements the requirements of ISO 8000- opments in WaterML and the Open Modeling Interface have
110. ISO 8000-120 includes a conceptual data model for data increasingly emphasized provenance components.
provenance where a given provenance_event records the prov-
enance for exactly one property_value_assignment, and every The purpose of the Open Modeling Interface (OpenMI)
property_value_assignment has its provenance recorded by one is to enable the runtime exchange of data between pro-
or many provenance_event objects. cess simulation models and also between models and
In order to trace the changing information and the prov- other modeling tools such as databases and analytical and
enance of data (and by implication geodata) over the web, visualization applications. Its creation has been driven
W3C has recently published a series of documents and by the need to understand how processes interact and to
2016 by Taylor & Francis Group, LLC
412 Remotely Sensed Data Characterization, Classification, and Accuracies
PROV-DM-structured view
of ASA Hazard Map
Wang et al.
wasGeneratedBy
(2007)
wasAssociatedWith Correct
ATCOR2
Landsat wasInfluencedBy
reflectance
PCI used
used wasAssociatedWith ASA Hazard Map
Detect
NDWI Canopy change
Change wasGeneratedBy
wasAssociatedWith wasGeneratedBy used
Landsat clipped
wasAssociatedWith
Write Web
program
wasGeneratedBy ESRI wasAssociatedWith
CAST
wasAssociatedWith
used
wasAssociatedWith Google Earth API
wasAttributedTo
Clip
Google
used
used Selected structures wasInfluencedBy
of PROV-DM
Forest area
Landsat Archive wasAttributedTo Entity
wasAttributedTo used wasGeneratedBy
wasAttributedTo TM September 2006 Agent
wasAssociatedWith Activity
USGS USFS Arrows point from future to past
ETM + September 2001
FIg u r e 19.7 Selected provenance of the ASA Hazard Map (Jones etal. 2014; Tullis etal. 2012) structured according to W3Cs PROV Data Model
(PROV-DM; Gil and Miles 2013; Moreau and Missier 2013; Table 19.1). Arrows (relationships) point from future to past, first from the online ASA
Hazard Map to its 20012006 canopy change layer, then to various agents, entities, and activities involved in the canopy change layers creation.
Some entities (e.g., Landsat Archive) represent PROV-DM collections of entities (e.g., individual Landsat images available from USGS), and many
potential PROV-DM details are not shown.
predict the likely outcomes of those interactions under a major role in this community, which, as of 2014, has several
given conditions. A key design aim has been to bring NSF-funded research and implementation grants pertaining to
about interoperability between independently developed provenance records in geoprocessing.
modeling components, where those components may
originate from any discipline or supplier. The ultimate
aim is to transform integrated modeling into an opera- 19.3 Why Provenance in Remote
tional tool accessible to all and so open up the potential Sensing Workflows
opportunities created by integrated modeling for innova-
tion and wealth creation. As Buneman (2013) argues, a change of attitude is in order
regarding the role for provenance across a range of com-
Vanecek and Moore (2014, p. ix, emphasis added) puter system-supported domains and activities, including (by
implication), remote sensing workflows. He makes the compari-
It is likely that future OGC efforts will increasingly focus on son between scientific activities where it is considered obvious that
provenance. The OGC is a major participant in EarthCube such information should be recorded and other domains where
(2014). In 2011, NSFs Cyberinfrastructure and Geosciences there is little or no awareness of process history or its value. He
Divisions established the EarthCube community to promote concludes that we should worry less about what provenance is and
geosciences data discovery and interoperability. The OGC plays concentrate more on what we can do with it once we have it (p. 11).
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 413
TABLe 1 9.1 Characteristics of PROV-DM Structures Including Core Types and Selected Relationships (Moreau and Missier 2013), Each with
an Example Provided from the Provenance of the ASA Hazard Map (Jones etal. 2014; Tullis etal. 2012; Figure 19.7)
PROV-DM
Structure Interpretive Highlights Example from ASA Hazard Map Provenance
Core types
Agent Need not be a person but could also represent an Center for Advanced Spatial Technologies (CAST) agent (organization) at
organization or even a specific software process University of Arkansas
Entity May be physical, digital, or conceptual Landsat 5 TM entity (satellite image) collected on September 15, 2006, over Ozark
National Forest
Activity Involves entities and requires some time to ESRI ArcGIS 10 for Desktop Extract by Mask activity (software tool) used to clip
complete the Landsat 5 TM imagery to the bounds of the study area), together with
environment settings (e.g., Snap Raster)
Selected relationships
wasGeneratedBy Can represent creation of only new entities Clipped Landsat 5 TM image that has been corrected for atmospheric attenuation
(that did not already exist) was generated by running the ATCOR2 algorithm
Used Only implies that usage has begun (but not that A GIS model for detecting oak-hickory forest decline or growth used a clipped and
it is completed) atmospherically corrected Landsat ETM+ image collected September 25, 2001
wasAttributedTo Links an entity to an agent without any The Google Earth API (used to write a web program to generate the ASA Hazard
understanding of activities involved Map) was attributed to Google
wasAssociatedWith Links an activity to an agent The ATCOR2 algorithm used to correct Landsat TM and ETM+ imagery for
atmospheric attenuation was associated with PCI Geomatics through their
Geomatica 10 platform
wasInfluencedBy At a minimum, suggests some form of influence The 20012006 oak-hickory forest canopy change data produced for the ASA
between entities, activities, and/or agents; Hazard Map was influenced by Wang etal. (2007), who used statistical thresholds
however, highly specific influence may be of change in Landsat-derived normalized difference water index (NDWI) over
captured time to detect oak canopy changes in the Mark Twain National Forest
19.3.1 Remote Sensing Questions that In addition to these, several questions could be asked trying
only Provenance can Answer to identify the source of errors or anomalies in the data. For
example, one might wonder at what point in the geoprocessing
For volumes that contain primarily raw or unprocessed geodata (e.g., did a specific region get assigned null values and why? Using
imagery telemetered directly from a satellite sensor), provenance (as provenance data, it should be possible to analyze two similar
used in the present context) may not offer much over traditional data products and compare their processing history to see how
metadata. However, when looking at geodata products resulting and why they differ (Bose and Frew 2005; Lanter 1994a). The
from complex geoprocessing workflows, there is much valuable opportunity to better understand and manage the complexity
information that metadata is ill-equipped to capture and store. of spatial scale in remote sensingassisted workflows is a further
There is sometimes confusion concerning what provenance justification for provenance-enabled geoprocessing (Tullis and
offers in terms of valuable information to an end user over the Defibaugh y Chvez 2009). Finally, provenance-aware systems
far more common and better supported (in terms of software could be used to enable and support temporal GIS analyses,
integration) metadata. One way to structure such a discussion is which require detailed history of a datasets change over time to
to look at some of the questions data users might ask that can be properly function (Langran 1988).
only reasonably answered using (at least in part) detailed prov- The value of provenance tracking and visualization was
enance information. For instance, one might ask the following demonstrated in a study conducted at the Regional Geospatial
regarding a remote sensingderived product: Service Center at the University of Texas, El Paso (Del Rio and
1. What was the processing time necessary to create this da Silva 2007). In this study, conducted as part of NSFs GEON
product, and what system configuration was implemented Cyberinfrastructure project, web services were built to per-
(including disk, processor, and RAM information)? form geoprocessing tasks (filtering, gridding, and contouring)
2. In what exact order were processing steps taken, and what required to create a contoured gravity map from a raw gravity
precise parameters were used during each intermedi- dataset. Del Rio and da Silva generated multiple contour maps
ate step? Was the process completely automated, or were with incorrect parameters (e.g., a grid size parameter larger than
manual steps (such as onscreen digitization) included in important anomalies in the gravity field), and participants in the
the workflow? study were asked to evaluate each contour map with and with-
3. What datasets, both source and derived, were used to create out provenance information. Without provenance information,
this product, and how did each contribute to the product? subject matter experts were able to detect errors in only 50% of
4. How were errors expressed and propagated during the the cases and to explain cause in only 25% of the cases. Nonsubject
products creation? Is the result statistically significant? matter experts fared much worse (11% and 11%). However, when
2016 by Taylor & Francis Group, LLC
414 Remotely Sensed Data Characterization, Classification, and Accuracies
LIDAR Structural Old growth portal with
Discrete-return analytics statistics provenance query
LIDAR
Provenance
store and
virtual data Cluster
catalog analysis and
Ortho- Image validation
imagery stack
Image
segmentation Old growth
modeling
Queryable
WorldView-3 geo-
database
GIS/GPS-
Radio metric MODIS assisted
and Correlation
time series analysis and reference data
geometric
Landsat 8 correction feature collection
OLI selection
Input/output
Point geodata Geoprocesses
Database query
Image geodata Database
Multitemporal image geodata Principal end product
FIg u r e 19.8 A provenance-enabled workflow for extracting old-growth bald cypress (Taxodium distichum) forest quality and biophysical
parameters from airborne LIDAR and orbital multispectral imagery. Such a framework could enable (a) detailed methodological transparency
related to old-growth inventory maps published in the web portal, (b) detailed transparency of the accuracy assessment, and (c) autogenerated
scripts for the replication of the detailed workflow despite a complicated blend of commercial and open-source workstation-level software, includ-
ing cyber-enabled high-performance geoprocessing tools.
provenance information was provided and visualized using Del attribution is streamlined, and the interpretability of geodata
Rio and da Silvas (2007) ProbeIt! (a provenance visualization products is enhanced (Simmhan etal. 2005). However, any one
tool), the subject matter experts were able to detect and explain of these advantages by itself is not as critical as establishing trust
all the errors. A more impressive result, though, is that prov- in the remote sensing process, which is clearly not immune to
enance improved the ability of the nonsubject matter experts to being oversold (Jensen 2007). Generally speaking, statistical
detect and explain errors by a factor of 7 (78%). validity is a key to establishing trust in scientific processes and
It can be argued that extending a traditional remote sensing has been extensively developed in remote sensing workflows
workflow to include provenance information offers a number (Congalton 2010). Simply maintaining an audit trail that dem-
of immediate advantages. For example, a web portal display- onstrates sound choices in statistical processing parameters and
ing a detailed map of old growth forest in the southeastern methods is one way to help ensure that trust is not erroneously
United States could include provenance query that enables (1) called into question.
detailed methodological transparency, (2) detailed transpar- In addition to maintaining an audit trail, statistical tech-
ency of the accuracy assessment, and (3) autogenerated scripts niques can be applied to a given workflow to determine trust,
for replication of the detailed workflow even though it includes and recent work has focused on assigning a measure of trust
a complicated blend of commercial and open-source software, by using a workflows provenance. While not demonstrated
including cyber-enabled high-performance geoprocessing tools in a remote sensing context, a nonstationary hidden Markov
(Figure19.8). model can be used to provide a measure of this trust (Naseri and
Ludwig 2013). Provenance can also be useful in helping to deter-
mine workflow trust for data on the web. When handling linked
19.3.2 Provenance and trust in the
data provenance, data authentication may serve as an estimation
Remote Sensing Process
of data trustworthiness. Uniform resource identifiers and digital
When provenance of sources, intermediates, and products is signatures can be used as a measure of authentication (Hart and
captured and maintained throughout the digital remote sens- Dolbar 2013).
ing process life cycle, data quality improves, an audit trail is Traditional indicators of trust may also be used in conjunction
available for reviewers and users, replication is straightforward, with provenance. Recommendation, authority, believability,
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 415
Determining trust in remote sensing workflows
Provenance Quality Usability
Sensor configurations Standards, User specifications
Algorithms specifications Spatial
Software ISO Spectral
environment OGC Temporal
Execution time FGDC Radiometric
Geodata structures Completeness Geodata
Accuracy Cost
Sample scheme Accessibility
In situ reference
Accuracy report
Remote Software
sensing agent
expert
Trust
Workflow
Clear sensor-to- contributors Replicable spatial
artifact record Comparable methods decision support
FIg u r e 19.9 Provenance, quality, and usability can be used by remote sensing experts to make a subjective decision on workflow trust
(Gamble and Goble 2011; Jensen 2005; Malaverri etal. 2012); a sample of their characteristics is shown. As geoprocessing, workflows, and prov-
enance are integrated, software agents can objectively influence replicable spatial decision support. (Artist image of WorldView-3, Courtesy of
DigitalGlobe, 2014.)
reputation, and objectivity can all serve as indicators of trust the workflow level (Tan 2007) and can thus be used to facilitate
(Gamble and Goble 2011). In addition to provenance, a work- scientific reproducibility. As Henzen etal. (2013) point out, the
flows quality and usability should also be evaluated when quality of provenance communication is also very important
determining trust. As geoprocessing, workflows, and prove- even when presented in a text format. A number of recent and
nance are integrated, software agents can objectively influence proposed provenance-aware approaches and systems related to
replicable spatial decision support (Figure 19.9). Until more remote sensing (Table 19.2) have addressed these and other issues.
quantitative techniques are developed for measuring trust of
geographic workflows using provenance, measures of quality
19.4.1 General Approaches
and usability used in conjunction with subjective trust indica-
tors should be examined before making a decision to trust a 19.4.1.1 inversion
workflow and its lineage. Inversion was developed for fine-grain data lineage and prov-
enance in database transaction and transformation (Buneman
19.4 Selected Recent and Proposed etal. 2001; Cui etal. 2000; Woodruff and Stonebraker 1997).
Provenance-Aware Systems Database queries or processing functions that generate a view,
table, or new data product can be registered in a database sys-
Many provenance-aware systems have largely been concerned tem (or provenance store). Registered database transforma-
with provenance capture, and this capability is critical for syn- tions can be inverted so as to trace the lineage between the data
ergistic geoprocessing, workflows, and provenance of interest product and the sources that derive the product. For example,
in remote sensing applications (Figure 19.1). The characteristics when a view is created or updated, the inversion method can
of the captured provenance information can greatly influence help identify the source tables from which the view is gener-
how it may benefit the remote sensing process. Of particular ated. Although inversion can be applied in data provenance
significance to geospatial applications is the level of provenance for geospatial data, not all functions are invertible. However, a
detail or granularity. Fine-grain provenance is obtained at a data weak or general inversion could be substituted to approximate
level and can even deal with individual pixels (Woodruff and the provenance by returning a fraction (or a projection) of the
Stonebraker 1997), whereas coarse-grain provenance represents desired provenance. Examples of inversion can be found in some
2016 by Taylor & Francis Group, LLC
416 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 1 9.2 Characteristics of Selected Provenance-Aware Systems Reported in Remote Sensing and Other Geodata Applications
Geodata Application Successes Limitations or Future Work References
Earth System Science Workbench and ES3
Track processing of a laboratorys raw a. Automates geodata provenance capture from Predates recent geodata Frew (2004), Frew and
satellite imagery (e.g., AVHRR) into running processes interoperability standards Bose (2001), Frew
higher level products b. Stores provenance in both XML documents and in and specifications and Slaughter (2008)
asearchable online store
MODAPS and OMIDAPS
Manage MODIS and other NASA a. Automates version tracking of geodata processing Identifies science Tilmes and Fleig
satellite imagery and its provenance algorithms communitys lack of (2008)
b. Reduces geodata storage via on-demand processing appreciation for
based on a virtual archive provenance information
Karma
Capture provenance for Japans a. Modularizes architecture to facilitate web service Requires additional geodata Conover etal. (2013),
AMSR-E passive microwave interoperability interoperability standards Plale etal. (2011),
radiometer on Aqua b. Is compatible with open provenance model (OPM) to facilitate geodata Simmhan etal. (2008)
and ISO 19115-2 metadata standards (scientific) reproducibility
Data Quality Provenance System
Assess quality of agricultural mapping a. Assigns geodata quality index based on provenance Needs to address geodata Malaverri etal. (2012)
based on SPOT satellite imagery information quality dependencies on
b. Is compatible with OPM and FGDC metadata provenance granularity
standards
VisTrails
Model habitat suitability using a. Provides Python-based open-source provenance and Designed to be domain Freire etal. (2012),
WorldView-3 and LIDAR-derived workflow management generic, VisTrails may Talbert (2012)
forest structure b. Allows key focus on provenance in rapidly changing have a steep learning
workflows (e.g., during remote sensing process curve
development)
UV-CDAT
Analyze large-scale remote sensing a. Built on top of VisTrails with an extensible Future work could adapt Santos etal. (2012)
derived climate data modularized architecture that supports high- UV-CDAT successes in
performance workflows climate change for other
b. First end-to-end provenance-enabled tool for geodata application areas
large-scale climate research
GeoPWProv
Visualize and navigate city planning a. Allows geodata provenance to be visualized and Future work could support Sun etal. (2013)
geodata (e.g., LIDAR-derived elevation explored in a map environment geoprocessing replication
data) provenance via a map b. Provides for several geospatial levels of provenance
query (e.g., via a single polygon versus a larger dataset)
spatial databases including Oracle Database, Microsoft SQL been explored in geospatial applications (Del Rio and da Silva
Server, IBM DB2, and Boundless PostGIS. One early approach 2007; Yue etal. 2010a,b, 2011), though feasible and convincing
was implemented for vector operations in Intergraphs (now approaches for provenance in SOC/SOI need further exploration
Hexagons) Geomedia product. and investigation.
19.4.1.2 Service chaining 19.4.1.3 Virtual Data catalog Service
In a vision of SOC and service-oriented integration (SOI), dif- A virtual data catalog (VDC) is a provenance approach to trace
ferent web services can be found, composed and invoked to the derivation route of data product and virtual data gener-
accomplish certain tasks. The sequence of service discovery, ated in the workflow in order to enable scientists to reproduce
composition, and execution looks like a chain, while alterna- a data product and validate the quality of the workflow and
tively, the composition processes can be constructed through related simulations. The intermediate data may be generated
different approaches, such as service orchestration or chore- within a workflow but may not exist physically in a database
ography that could be applied in enabling business processes or computer system (e.g., due to storage limitations). For this
and transactions on the web. Service composition and chaining reason, such data are called virtual data because it is the
could represent a workflow in which scientific computation can representation and manipulation of data that does not exist,
be implemented through the SOC/SOI approach. Capturing the being defined only by computational procedures (Foster etal.
provenance information within service-oriented workflows has 2002, 2003).
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 417
Within virtual data systems, such as Chimera, which is a System (AMSRE) flown on the Aqua satellite. One of the big-
virtual data grid managing the derivation and analysis of data gest benefits of Karma is its modular architecture, which simpli-
objects, Virtual Data Language (VDL) is developed to define the fies interoperability with Java and other web services. Karmas
workflow, while VDC is a service in the virtual data systems. architecture for this application consists of an application layer,
The latter is defined and implemented based on the virtual data web service layer, core service layer, and a database layer (Plale
schema (VDS). The VDS defines the relevant data objects and et al. 2011). The inclusion of open provenance model (OPM)
relationships, and VDC can be queried by VDL to construct specifications and XML makes its interoperability extend fur-
data derivation procedures from which derived data and output ther (Moreau etal. 2011). Conover etal. (2013) also made use of
can be recomputed (Foster etal. 2002, 2003; Glavic and Dittrich Karma to retrofit a legacy system for provenance capture. They
2007; Simmhan etal. 2005). chose the NASA Science Investigatorled Processing System
(SIPS) for the AMSR-E sensor on the Aqua satellite. Their system
uses a two-tiered approach that captures provenance for individ-
19.4.2 earth System Science
ual data files as well as collections or series, both automatically
Workbench and eS3
and via manual entry using the ISO 19115-2 lineage metadata
The Earth Systems Science Workbench (ESSW) was an early standard. Query and display are handled with a database-driven
attempt at automated provenance management and storage. It web interface called the Provenance Browser.
was a nonintrusive system that made use of Perl scripting tech-
niques and Java to store data provenance as XML documents
19.4.5 Data Quality Provenance System
(Frew and Bose 2001). It contained a registry for provenance
and a server for making the information searchable on the web. Taking into account a sources trustworthiness and the datas
ESSW was followed up by the Earth System Science Server (ES3), age, Malaverri et al. (2012) created a provenance system that
which allowed for more flexibility in client-side implementation, allows a quality index to be assigned. This approach is based on
but used essentially the same structure as the ESSW (Frew 2004). a combination of the OPM and FGDC geospatial metadata stan-
ES3, unlike many other systems, automatically captures prov- dards. Criteria considered in the quality index include granular-
enance from running processes. It can also create provenance ity, accuracy of attribute descriptions, completeness of the data,
graphs in XML that can then be visualized using third-party a logical measure of the data, and spatial positional accuracy.
tools like yEd (Frew and Slaughter 2008). Although measures of trust can be very subjective in nature,
in this case requiring a domain experts input, this approach
is somewhat unique in that it attempts to quantify data quality
19.4.3 MoDAPS and oMiDAPS
(Malaverri etal. 2012).
The MODIS Adaptive Data Processing System (MODAPS) and
OMI Data Processing System (OMIDAPS) were designed for
19.4.6 Vistrails
use by NASA to manage satellite imagery and provenance from
MODIS sensors on the Terra and Aqua satellites, and the OMI VisTrails is a free and open-source scientific workflow and
sensor on Aura respectively (Tilmes and Fleig 2008). Both sys- provenance management system (Freire etal. 2012). Written in
tems are operational and use a scripting process to track changes Python/Qt and designed to be integrated with existing work-
in versions of geodata processing algorithms. Using this tech- flow systems, VisTrails has been used in a number of research
nique, there is no need to store workflow iterations because applications ranging from climate (including the UV-CDAT
enough information is retained that previous versions of the described later) to ecology and biomedical research. Talbert
data can be re-created. Further, these systems periodically are (2012) created software based on VisTrails to capture the details
tasked with reprocessing past data using the most up-to-date of habitat suitability and species distribution modeling. One
algorithms to maintain a consistent and improved series of data of the major advantages of VisTrails is that as an open-source
products. MODAPS in particular makes use of these features project built in Python, it is interoperable, easily customizable,
to maintain a virtual archive with provenance information that and benefits from a large community of developers contribut-
persists after a geodata product is deleted, allowing the system to ing code. A key focus of VisTrails is rapidly changing workflows.
re-create data products on demand rather than keeping exten- The information contained in how workflows are developed (and
sive archives. Data re-creation as implemented in these systems change over time) may provide highly valuable insight into the
is unique and is something that could be useful in other geospa- creative and development aspects of the remote sensing process.
tial provenance systems.
19.4.7 UV-cDAt
19.4.4 Karma
Climate Data Analysis Tools (CDATs) are cutting-edge domain-
Plale et al. (2011) make use of the Karma system designed specific tools for the climate research community, but they are
by Simmhan et al. (2008) to collect provenance data for the ill-equipped to handle very large geodata and provenance infor-
Advanced Microwave Scanning RadiometerEarth Observing mation. The UV-CDAT is a relatively new provenance system for
2016 by Taylor & Francis Group, LLC
418 Remotely Sensed Data Characterization, Classification, and Accuracies
handling large amounts of climate-based data (Santos etal. 2012). traditional metadata in large part because it encompasses
The UV-CDAT uses a highly extensible modular design and process history. Regardless of definitions, the application of
makes use of a Visualization Control System and Visualization provenance benefits in remote sensingassisted decision sup-
Toolkit (VTK)/ParaView infrastructure, which allows for high- port workflows cannot be realized without development and
performance parallel-streaming data analysis and visualization. demonstration of collaborative software architectures including
Its loosely coupled modular design allows for integration with those in a geo-cyberinfrastructure. Provenance has and will be
third-party tools such as R and MATLAB for both analysis of increasing interest to and a focus of organizations that create
and visualization. The UV-CDAT is unique in that it is the first and encourage international specifications and standards (e.g.,
end-to-end application for provenance-enabled analysis and ISO, W3C, and OGC). As these organizations formulate proce-
visualization for large-scale climate research. It has already been dures for the specification of provenance, we will see software
distributed and is widely used by scientists throughout the cli- developers add this capability to their products in a far more
mate change field. complete implementation than is currently the case. Even before
the emerging international standards begin to mature, research
is critically needed to demonstrate and fully understand prac-
19.4.8 GeoPWProv
tical benefits that user-friendly and integrated geoprocessing,
GeoPWProv is a provenance system specializing in displaying workflows, and provenance can offer. With additional research
geospatial provenance as an easily accessible interactive map and development, geospatial provenance has a high potential to
layer. GeoPWProv has the capability to capture provenance benefit quality, trust, and innovation related to remote sensing
at the feature, dataset, service, or knowledge level (Sun et al. assisted spatial decision support.
2013). Comparisons can be made between entities in each level
or between various levels. In addition to displaying provenance
References
as a map layer, GeoPWProv supports displaying provenance
in a workflow or in the more traditional text-based format. Alonso, G. and C. Hagen. 1997. Geo-opera: Workflow concepts
Implementation on the client side through use of a browser for spatial processes. In Advances in Spatial Databases,
and Open Layers allows for ease of use. GeoPWProvs display Springer, Berlin, Germany, pp. 238258.
of provenance in different formats and at different levels allows Anderson, J.R., E.E. Hardy, J.T. Roach, and R.E. Witmer. 1976. A
for a customizable user experience when evaluating a workflow. Land Use and Land Cover Classification System for Use with
Remote Sensor Data. Geological Survey Professional Paper
19.5 conclusions and Research 964. Washington, DC: U.S. Government Printing Office.
implications Anderson, K.E. and G.M. Callahan. 1990. The modernization
program of the U.S. Geological Surveys National Mapping
Integrated geoprocessing, workflows, and provenance may be Division. Cartography and Geographic Information Systems
conceptualized as a positive developmental cycle that enables 17(3): 243248.
experts and software agents to capture, store, analyze, curate, Aronson, P. and S. Morehouse. 1983. The ARC/INFO map library:
replicate, and innovate remote sensing methods. Such integra- A design for a digital geographic database. In Auto-Carto Six;
tion is increasingly understood as a key to high-quality, repli- Proceedings of the Sixth International Symposium on Automated
cable remote sensingassisted spatial decision support. In early Cartography, Vol. 1, Ottawa/Hull, Canada, pp. 372382.
discussions in the 1980s, it soon became clear that provenance Atkins, D.E., K.K. Droegemeier, S.I. Feldman, H. Garcia-Molina,
(or lineage) in particular is a fundamental element in under- M.L. Klein, D.G. Messerschmitt, P. Messina, J.P. Ostriker,
standing earth observation-related and other geodata quality and M.H. Wright. 2003. Revolutionizing science and
(Moellering etal. 1988). As commercial GIS accelerated during engineering through cyberinfrastructure: Report of the
the early 1990s, the Geolineus project (Lanter 1992b) demon- National Science Foundation Blue-Ribbon Advisory Panel
strated how software dedicated to lineage/provenance capture, on Cyberinfrastructure. National Science Foundation:
management, and visualization can enable such gains as replica- Washington, DC. https://arizona.openrepository.com/ari-
ble geospatial workflows, automated workflow comparison, data zona/handle/10150/106224. Accessed June 29, 2014.
quality modeling, data update management, and increased shar- Bose, R. and J. Frew. 2005. Lineage retrieval for scientific data
ing of expert knowledge of geodata creation. Now with increas- processing: A survey. ACM Computing Surveys 37(1): 128.
ingly heightened awareness of provenance in computer systems Bossler, J.D., J.B. Campbell, R.B. McMaster, and C. Rizos, eds.
(Bose and Frew 2005; Ikeda and Widom 2009; Simmhan etal. 2010. Manual of Geospatial Science and Technology, 2nd
2005; Yue etal. 2010a), there has been a maturing appreciation of edn. Boca Raton, FL: CRC Press.
the need to computationally address provenance capture, man- Botts, M. and A. Robin, eds. 2007. OpenGIS Sensor Model
agement, and exchange in an increasingly big data scenario. Language (SensorML) Implementation Specification. Open
While definitions of geodata provenance have varied, it is Geospatial Consortium. http://portal.opengeospatial.org/
quite arguably distinct from and offers unique benefits over files/?artifact_id=21273. Accessed June 24, 2014.
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 419
Brring, A., C. Stasch, and J. Echterhoff, eds. 2012. OGC Sensor Foster, I., J. Vckler, M. Wilde, and Y. Zhao. 2002. Chimera:
Observation Service Interface Standard. Open Geospatial A virtual data system for representing, querying, and
Consortium. automating data derivation. In Proceedings of the 14th
Buneman, P., S. Khanna, and W.-C. Tan. 2001. Why and where: International Conference on Scientific and Statistical
A characterization of data provenance. In International Database Management, Los Alamitos, CA IEEE, pp. 3746.
Conference on Database Theory (ICDT), pp. 316330. http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=
Buneman, P. and S.B. Davidson. 2010. Data provenanceThe 1029704.
foundation of data quality. Carnegie Mellon University Foster, I., J. Vckler, M. Wilde, and Y. Zhao. 2003. The virtual data
Software Engineering Institute, Pittsburgh, PA. http://www. grid: A new model and architecture for data-intensive col-
sei.cmu.edu/measurement/research/upload/Davidson.pdf. laboration. In Proceedings of the First Biennial Conference on
Accessed June 30, 2014. Innovative Data Systems Research (CIDR), Vol. 3. Asilomar,
Buneman, P. 2013. The Providence of Provenance. In Big CA: Citeseer, p. 12.
Data, edited by G. Gottlob, G. Grasso, D. Olteanu, and Freire, J., D. Koop, E. Santos, C. Scheidegger, C. Silva, and H.T.
C.Schallhart, 7968:712. Berlin, Germany: Springer. http:// Vo. 2012. VisTrails. In Brown, A. and Wilson, G., eds.
link.springer.com/10.1007/978-3-642-39467-6_3. The Architecture of Open Source Applications: Elegance,
Chrisman, N.R. 1983. The role of quality information in the Evolution, and a Few Fearless Hacks, Vol. I. aosabook.org.
long-term functioning of a geographic information system. http://aosabook.org/en/vistrails.html.
Automated Cartography, 6: 302312. Frew, J. 2004. Earth system science server (ES3): Local infrastruc-
Chrisman, N.R. 1986. Obtaining information on quality of ture for earth science product management. In Proceedings
digital data. In AutoCarto Proceedings of the International of the Fourth Earth Science Technology Conference, Palo
Symposium on Computer-Assisted Cartography, Vol. 1. Alto, CA. http://esto.gsfc.nasa.gov/conferences/estc2004/
London, U.K.: Cartography and Geographic Information papers/a4p3.pdf. Accessed March 1, 2014.
Society, pp. 350358. Frew, J. and P. Slaughter. 2008. ES3: A demonstration of trans-
Congalton, R. 2010. Remote sensing: An overview. GIScience & parent provenance for scientific computation. In J. Freire,
Remote Sensing 47(4): 443459. D. Koop, and L. Moreau, eds., Provenance and Annotation
Conover, H., R. Ramachandran, B. Beaumont, A. Kulkarni, M. of Data and Processes. Lecture Notes in Computer Science,
McEniry, K. Regner, and S. Graves. 2013. Introducing prov- Vol. 5272. Berlin, Germany: Springer, pp. 200207. http://
enance capture into a legacy data system. IEEE Transactions link.springer.com/chapter/10.1007/ 978-3-540-89965-
on Geoscience and Remote Sensing 51(11): 50985014. 5_21. Accessed June 26, 2014.
Cui, Y., J. Widom, and J.L. Wiener. 2000. Tracing the lineage of Frew, J. and R. Bose. 2001. Earth system science workbench: A
view data in a warehousing environment. ACM Transactions data management infrastructure for earth science products.
on Database Systems 25(2): 179227. In Proceedings of the International Conference on Scientific
Del Rio, N. and P.P. da Silva. 2007. Probe-It! Visualization sup- and Statistical Database Management. Los Alamitos, CA:
port for provenance. In Advances in Visual Computing. IEEE Computer Society, pp. 180189.
Springer, Berlin, Germany, pp. 732741. http://link. Gamble, M. and C. Goble. 2011. Quality, trust, and utility of scientific
springer.com/chapter/ 10.1007/978-3-540-76856-2_72. data on the web: Towards a joint model. In Proceedings of the
Accessed June 24, 2014. Third International Web Science Conference, Vol. 15. ACM.
Di, L., P. Yue, H.K. Ramapriyan, and R.L. King. 2013b. Geoscience New York, NY. http://dl.acm.org/citation.cfm?id=2527048.
data provenance: An overview. IEEE Transactions on Accessed June 24, 2014.
Geoscience and Remote Sensing 51(11): 50655072. Gil, Y., J. Cheney, P. Groth, O. Hartig, S. Miles, L. Moreau, and
Di, L., Y. Shao, and L. Kang. 2013a. Implementation of geospatial P.P. da Silva, eds. 2010. The foundations for provenance on
data provenance in a web service workflow environment with the web. Foundations and Trends in Web Science 2(23):
ISO 19115 and ISO 1911-2 lineage model. IEEE Transactions 99241.
on Geoscience and Remote Sensing 51(11): 50825089. Gil, Y. and S. Miles, eds. 2013. PROV Model Primer. W3C. http://
EarthCube. 2014. EarthCube: Transforming geosciences research. www.w3.org/TR/prov-primer/.
http://earthcube.org/. Accessed June 30, 2014. Giordano, A., H. Veregin, E. Borak, and D.P. Lanter. 1994. A con-
Essinger, R. and D.P. Lanter. 1992. User-centered software design ceptual model of GIS-based spatial analysis. Cartographica:
in GIS: Designing an icon-based flowchart that reveals the The International Journal for Geographic Information and
structure of ARC/INFO data graphically. In Proceedings of Geovisualization 31(4): 4457.
the 12th Annual ESRI User Conference, Palm Springs, CA. Glavic, B. and K.R. Dittrich. 2007. Data provenance: A cat-
Fielding, R.T. 2000. Architectural Styles and the Design of Network- egorization of existing approaches. In Proceedings of the
Based Software Architectures. Irvine, CA: University of 12th GI Conference on Database Systems in Business,
California. http://www.ics.uci.edu/~fielding/pubs/dissertation/ Technology, and Web (BTW), Vol. 7, Aachen, Germany,
top.htm. Accessed June 29, 2014. pp. 227241.
2016 by Taylor & Francis Group, LLC
420 Remotely Sensed Data Characterization, Classification, and Accuracies
Grady, R.K. 1988. The lineage of data in land and geographic Lanter, D.P. 1992a. Propagating updates by identifying data
information systems. In Proceedings of GIS/LIS88 American dependencies in spatial analytic applications. In Proceedings
Congress on Surveying and Mapping: Data Lineage in Land of the 12th Annual ESRI User Conference, Palm Springs, CA.
and Geographic Information Systems, Vol. 2, San Antonio, Lanter, D.P. 1992b. GEOLINEUS: Data Management and
TX, pp. 722730. Flowcharting for ARC/INFO, 92-2. Santa Barbara,
Guptill, S.C. 1987. Techniques for managing digital cartographic CA: National Center for Geographic Information &
data. In Proceedings of the 13th International Cartographic Analysis. http://www.ncgia.ucsb.edu/Publications/tech-
Conference, Morelia, Mexico, Vol. 4(16), pp. 221226. reports/91/91-6.pdf. Accessed June 24, 2014.
Hart, G. and C. Dolbar. 2013. Linked Data: A Geographic Lanter, D.P. 1993a. Method and means for lineage tracing of a spa-
Perspective, 1st edn. CRC Press , London, U.K. tial information processing and database system. Patent
Henzen, C., S. Mas, and L. Bernard. 2013. Provenance informa- No. 5,193,185. United States Department of Commerce
tion in geodata infrastructures. In Geographic Information Patent and Trademark Office.
Science at the Heart of Europe, III. Lecture Notes in Lanter, D.P. 1993b. A lineage meta-database approach toward
Geoinformation and Cartography. Springer International spatial analytic database optimization. Cartography and
Publishing, New York, NY, pp. 133151. Geographic Information Systems 20(2): 112121.
Hey, T., S. Tansley, and K. Tolle, eds. 2009. The Fourth Paradigm: Lanter, D.P. 1994a. Comparison of spatial analytic applications of
Data-Intensive Scientific Discovery. Redmond, WA: GIS. In W. K. Michener, J. W. Brunt, and S. G. Stafford, eds.,
Microsoft Research. Environmental Information Management and Analysis:
Houbie, F. and L. Bigagli. 2010. OGC catalogue services standard Ecosystem to Global Scales. CRC Press , London, U.K.
2.0 extension package for ebRIM application profile: Earth Lanter, D.P. 1994b. A lineage metadata approach to removing redun-
observation products. http://portal.opengeospatial.org/ dancy and propagating updates in a GIS database. Cartography
files/?artifact_id=35528. and Geographic Information Systems 21(2): 9198.
IFAR. 2013. Provenance Guide. International Foundation for Art Lanter, D.P. and C. Surbey. 1994. Metadata analysis of GIS data
Research. New York, NY, http://www.ifar.org/provenance_ processing: A case study. In T.C. Waugh and R.G. Healey,
guide.php. Accessed September 9, 2014. eds., Advances in GIS Research: Proceedings of the Sixth
Ikeda, R. and J. Widom. 2009. Data lineage: A survey. Technical International Symposium on Spatial Data Handling. London,
Report. Stanford University InfoLab, Stanford, CA, http:// U.K.: Taylor & Francis Ltd., pp. 314324.
ilpubs.stanford.edu:8090/918/. Accessed September 13, 2013. Lanter, D.P. and H. Veregin. 1991. A lineage information
ISO 19115-2:2009(E). 2009. Geographic information program for exploring error propagation in GIS appli-
MetadataPart 2: Extensions for imagery and gridded data. cations. In Proceedings of the 15th Conference of the
International Organization for Standardization, Geneva, International Cartographic Association, Bournemouth,
Switzerland. U.K., pp. 468472.
Jensen, J.R. 2005. In K.C. Clarke, ed., Introductory Digital Image Lanter, D.P. and H. Veregin. 1992. A research paradigm for
Processing: A Remote Sensing Perspective, 3rd edn. Prentice propagating error in layer-based GIS. Photogrammetic
Hall Series in Geographic Information Science. Upper Engineering and Remote Sensing 58(6): 825833.
Saddle River, NJ: Prentice Hall. Lanter, D.P. and R. Essinger. 1991. User-Centered Graphical User
Jensen, J.R. 2007. In K.C. Clarke, eds. Remote Sensing of the Interface Design for GIS. Santa Barbara, CA: National Center
Environment: An Earth Resource Perspective, 2nd edn. for Geographic Information & Analysis, pp. 9196. http://
Prentice Hall Series in Geographic Information Science. www.ncgia.ucsb.edu/Publications/tech-reports/91/91-6.
Upper Saddle River, NJ: Prentice Hall. pdf. Accessed June 24, 2014.
Jones, J.S., J.A. Tullis, L.J. Haavik, J.M. Guldin, and F.M. Stephen. Malaverri, J.E.G., C.B. Medeiros, and R.C. Lamparelli. 2012. A
2014. Monitoring oak-hickory forest change during an provenance approach to assess quality of geospatial data.
unprecedented red oak borer outbreak in the Ozark In 27th Symposium on Applied Computing. Riva del Garda
Mountains: 1990 to 2006. Journal of Applied Remote Sensing (Trento), Italy: ACM.
8(1): 113. Merriam-Webster. 2014. Merriam-Webster Online. http://www.
Langran, G. and N.R. Chrisman. 1988. A framework for temporal merriam-webster.com/.
geographic information. Cartographica 25(3): 114. Moellering, H., L. Fritz, D. Franklin, R.W. Marx, J.E. Dobson, D.
Langran, Gail. 1988. Temporal GIS Design Tradeoffs. In Edson, J. Dangermond etal. 1988. The proposed standard
Proceedings of GIS/LIS 88, 89099. San Antonio, TX: for digital cartographic data. The American Cartographer
American Congress on Surveying and Mapping. 15(1): 9140.
Lanter, D.P. 1989. Techniques and Method of Spatial Database Moore, H. 1983. The impact of computer technology in the map-
Lineage Tracing. Columbia, SC: University of South Carolina. ping environment. In Proceedings of the Sixth International
Lanter, D.P. 1991. Design of a lineage-based meta-data base for Symposium on Automated Cartography, Vol. 1. Ottawa/Hull,
GIS. Cartography and Geographic Information Systems Ontario/Quebec, Canada: Cartography and Geographic
18(4): 255261. Information Society, pp. 6068.
2016 by Taylor & Francis Group, LLC
Geoprocessing, Workflows, and Provenance 421
Moreau, L., B. Clifford, J. Freire, J. Futrelle, Y. Gil, P. Groth, N. Tilmes, C. and A.J. Fleig. 2008. Provenance tracking in an
Kwasnikowska etal. 2011. The open provenance model core earth science data processing system. In J. Freire and
specification (v1.1). Future Generation Computer Systems D. Koop, eds., Provenance and Annotation of Data and
27(6): 743756. Processes. Lecture Notes in Computer Science, Vol. 5272.
Moreau, L. 2010. The foundations for provenance on the web. Berlin, Germany: Springer-Verlag, pp. 221228. http://
Foundations and Trends in Web Science 2(23): 99241. ebiquity.umbc.edu/_file_directory_/papers/445.pdf.
Moreau, L. and P. Missier, eds. 2013. PROV-DM: The PROV Accessed September 19, 2014.
data model. http://www.w3.org/TR/2013/REC-prov-dm- Tullis, J.A., F.M. Stephen, J.M. Guldin, J.S. Jones, J. Wilson, P.D.
20130430/#section-example-two. Smith, T. Sexton et al. 2012. Applied silvicultural assess-
Naseri, M. and S.A. Ludwig. 2013. Evaluating workflow trust ment (ASA) Hazard Map. University of Arkansas Forest
using hidden Markov modeling and provenance data. In Q. Entomologys Applied Silvicultural Assessment, Fayetteville,
Liu, Q. Bai, S. Giugni, D. Williamson, and J. Taylor, eds., AR, http://asa.cast.uark.edu/hazmap/. Accessed August 30,
Data Provenance and Data Management in eScience. Studies 2014.
in Computational Intelligence, Vol. 426. Berlin, Germany: Tullis, J.A. and J.M. Defibaugh y Chvez. 2009. Scale management
Springer-Verlag, pp. 3558. and remote sensor synergy in forest monitoring. Geography
NSF. 2014. Data Infrastructure Building Blocks (DIBBs). http:// Compass 3(1): 154170.
www.nsf.gov/pubs/2014/nsf14530/nsf14530.htm. Accessed Vanecek, S. and R. Moore. 2014. OGC open modelling interface
June 30, 2014. standard, Version 2.0. https://portal.opengeospatial.org/
Nyerges, T. November 1987. GIS research needs identified dur- files/?artifact_id=59022. Accessed June 26, 2014.
ing a cartographic standards process: Spatial data exchange. Veregin, H. and D.P. Lanter. 1995. Data-quality enhancement
International Geographic Information Systems Symposium: techniques in layer-based geographic information systems.
The Research Agenda 1: 319330. Computers Environment and Urban Systems 19(1): 2336.
Oxford University Press. 2014. Oxford English Dictionary. http:// Wade, Tasha, and Shelly Sommer. 2006. A to Z GIS: An Illustrated
www.oed.com/. Dictionary of Geographic Information Systems. Redlands,
Plale, B., B. Cao, C. Herath, and Y. Sun. 2011. Data provenance CA: Esri Press.
for preservation of digital geoscience data. In A.K. Sinha, Wang, C., Z. Lu, and T.L. Haithcoat. 2007. Using Landsat images to
D. Arctur, I. Jackson, and L.C. Gundersen, eds., Societal detect oak decline in the Mark Twain National Forest, Ozark
Challenges and Geoinformatics. Geological Society of America Highlands. Forest Ecology and Management 240: 7078.
Special Paper 482. Boulder, CO: Geological Society of Wang, S., A. Padmanabhan, J.D. Myers, W. Tang, and Y. Liu.
America, pp. 125137. 2008. Towards Provenance-Aware Geographic Information
Santos, E., D. Koop, T. Maxwell, C. Doutriaux, T. Ellqvist, G. Systems. Irvine, CA: ACM. http://acmgis08.cs.umn.edu/
Potter, J. Freire, D. Williams, and C.T. Silva. 2012. Designing papers.html#posterpapers.
a provenance-based climate data analysis application. In P. Woodruff, A. and M. Stonebraker. 1997. Supporting fine-
Groth and J. Frew, eds., Provenance and Annotation of Data grained lineage in a database visualization environment. In
and Processes. Lecture Notes in Computer Science, Vol. W.A. Gray and P.-. Larson, eds., Proceedings of the 13th
7525. Santa Barbara, CA, pp. 214219. International Conference on Data Engineering, Birmingham,
Schut, P., ed. 2007. OpenGIS Web Processing Service. Open U.K., pp. 91102.
Geospatial Consortium. Yeide, N.H., K. Akinsha, and A.L. Walsh. 2001. The AAM Guide
Simmhan, Y.L., B. Plale, and D. Gannon. 2005. A survey of data to Provenance Research. Washington, DC: American
provenance in e-science. SIGMOD Record 34(3): 3136. Association of Museums.
Simmhan, Y.L., B. Plale, and D. Gannon. 2008. Karma2: Yue, P., J. Gong, and L. Di. 2010a. Augmenting geospatial data
Provenance management for data-driven workflows. provenance through metadata tracking in geospatial service
International Journal of Web Services Research 5(2): 122. chaining. Computers & Geosciences 36: 270281.
Sun, Z., P. Yue, L. Hu, J. Gong, L. Zhang, and X. Lu. 2013. Yue, P., J. Gong, L. Di, L. He, and Y. Wei. 2010b. Semantic prove-
GeoPWProv: Interleaving map and faceted meta- nance registration and discovery using geospatial catalogue
data for provenance visualization and navigation. IEEE service. Proceedings of the Second International Workshop
Transactions on Geoscience and Remote Sensing 51(11): on the role of Semantic Web in Provenance Management
51315136. (SWPM 2010), Shanghai, China.
Talbert, C. 2012. Software for Assisted Habitat Modeling Package Yue, P., Y. Wei, L. Di, L. He, J. Gong, and L. Zhang. 2011. Sharing
for VisTrails (SAHM: VisTrails) v. 1. Fort Collins, CO: USGS geospatial provenance in a service-oriented environ-
Fort Collins Science Center. https://www.fort.usgs.gov/ ment. Computers, Environment and Urban Systems 35(4):
products/23403. Accessed August 31, 2014. 333343.
Tan, W.-C. 2007. Provenance in databases: Past, current, and Yue, P. and L. He. 2009. Geospatial data provenance in cyber-
future. Bulletin of the IEEE Computer Society Technical infrastructure. In Proceedings of the 17th International
Committee on Data Engineering 30(4): 312. Conference on Geoinformatics, Fairfax, VA.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
20
Toward Democratization of
Geographic Information:
GIS, Remote Sensing, and GNSS
Applications in Everyday Life
Acronyms and Definitions ................................................................................................................ 423
20.1 Rethinking Geographic Information and Technologies in the Twenty-First Century .... 423
20.2 Toward Democratization of Geo-Information Technologies .......................................... 425
Enlightened Phase (Twentieth Century): State-Sponsored Geo-Information
Technologies Transition Phase (Twentieth Century): Governance and Scholarship for
Society Contemporary Phase (Twenty-First Century): The Vision of Democratization
Gaurav Sinha 20.3 Discovering Remote Sensing, GIS, and GNSS Applications in Our Daily Lives .......... 430
Ohio University Location-Based Services Disaster Relief and Emergency Management Community
Building Participatory Sensing
Barry J. Kronenfeld 20.4 Democratization of Geo-Information: Circumscribing Issues and Challenges........... 436
Eastern Illinois University
Quality of Information and Services Privacy and Confidentiality Empowerment and
Equity
Jeffrey C. Brunskill
Bloomsburg University 20.5 Conclusion ............................................................................................................................... 441
of Pennsylvania References............................................................................................................................................ 441
Acronyms and Definitions 20.1 Rethinking Geographic
FGDC Federal Geographic Data Committee
information and technologies
Geo-ICT Geospatial information and communication in the twenty-First century
technology
There are very few academic domains that remain unaffected
GIS Geographic information system
by the reach of geo-information technologies. Geography and
GIScience Geographic Information Science
other closely affiliated environmental science disciplines rely
GNSS Global Navigation Satellite System
heavily on such technologies for their empirical work, but
GPS Global Positioning System
environmental engineering, social science, humanities, health
HOT Humanitarian OpenStreetMap Team
science, and the business communities have also found geo-
LBS Location-based services
technologies and the geographic perspective crucial to many
LKCCAP Local Knowledge and Climate Change Adaptation
of their disciplinary pursuits. There was a time, until very
Project
recently, when a statement such as this would suffice to describe
NCGIA National Center for Geographic Information and
the geo-information revolution. Yet, today, the growing influ-
Analysis
ence of geographic/geospatial theory, information, and tech-
OSM OpenStreetMap
nology goes far beyond the academe and the workplaceit
PGIS Participatory GIS
has become the story of our everyday lives. There is no better
PPGIS Public participation GIS
example of this than the life story of a poor Indian boy who lost
SDI Spatial data infrastructure
his family in 1986, ending up on a wrong train that took him
UAV Unmanned aerial vehicle
all the way across India to the bustling city of Calcutta, where
VGI Volunteered geographic information
he had to beg for a living, but from where ultimately he was
423
2016 by Taylor & Francis Group, LLC
424 Remotely Sensed Data Characterization, Classification, and Accuracies
transported far away to Australia through adoption by a loving functional aspects. The following define each briefly in the con-
couple, the Brierleys. Saroo Brierley ended up in much more text of this chapter:
comfortable settings than he would have probably been enti-
Remote sensing is the science of collecting, processing, and
tled in the impoverished settings of his earlier life, but he never
interpreting information of the Earth from aircraft or space-
gave up the urge to find his lost family. He had some memories
craft equipped with instruments for sensing signals emitted
of his first neighborhood, and that it was within a 14 h train
or reflected by the surface (or atmosphere) of the Earth.
ride to Calcutta, but there was no way to use that informa-
A GIS is an epistemology institutionalized through soft-
tion gainfullyuntil Google Earth was released and satellite
ware and social practices for processing, managing, ana-
imagery and maps made it possible to visually search places
lyzing, modeling, visualizing, and communicating about
anywhere remotely using a computer. Saroo drew a circle to
geospatial datasets of various types.
locate all places within 14h train ride of Calcutta (~1200km),
A GNSS is a suite of satellites in orbit around the Earth
and while it seemed like finding the proverbial needle in a hay-
transmitting location and timing data that may be used
stack, his persistence paid off when he recognizes landmarks
by receivers worldwide to determine location on or near
in Khandwa, Madhya Pradesh, several hundred miles from
the Earth.
Kolkata (formerly Calcutta, when Saroo got lost). Saroo even-
tually went to India and tracked down his biological family, These technologies have been constantly evolving, and lately
25years after he was separatedwith mostly Google Earth to through synergistic combinations, they have revolutionized every
thank. The story of his life is now a bestseller book A Long Way aspect of geographic information creation, analysis, and visualiza-
Home (Brierley, 2014) and a great ode to the power of personal tion in all the major sectors of the modern economy: scientific,
spatial memories and geospatial technologiesespecially commercial, educational, and governmental. Their collective con-
high-resolution satellite imagerybased mapping services that tribution to the world economy has grown significantly over time to
anybody can use. several tens of billions of dollars (Dasgupta, 2013). Jerome Dobson,
Google Earths success and popularity is symbolic of many a pioneer who helped establish the field of Geographic Information
different transitions that signify today geographic information Science (GIScience), envisions remote sensing, GIS, GNSS, and
creation and consumption is no longer under the tight con- related technologies, collectively as a macroscope for viewing
trol of institutions, experts, and the powerful few. Their grips large-scale phenomena in finer detail (similar to the microscope
have been loosened incrementally and sometimes disruptively that magnifies truly small-scale phenomena or the telescope that
by certain developments that finally have given individuals, magnifies distant, apparently small phenomena) (Dobson, 2011).
novices, and the traditionally disenfranchised a strong say Over the last decade, the development of newer forms of geospa-
in how places, events, and perceptions are mapped and com- tial information and communication technologies (Geo-ICTs)
municated. We now live in a world where the average person and Web 2.0 services has played a key role in widening the scope
is increasingly aware of the everyday benefits and drawbacks of the geospatial macroscope by making the collection, process-
for geo-information and associated technologies. This includes ing, and sharing of geographically referenced information a part of
powerful tools that support navigation and exploration, com- our everyday lives. We afford ourselves the functionalities of these
munication, social interaction, public participation, security, technologies not just in professional contexts but also for managing
and the like. Our increasing reliance on mobile devices (e.g., our personal information. In doing so, we uncover another impor-
phones, tablets, and notebooks) is an important catalyst for tant contextthe fascinating relationship between technology and
many of these developments. Such developments also give rise society. As the German philosopher Martin Heidegger observed,
to increased concerns over the quality and quantity of the spa- the essence of technology is by no means anything technical
tially oriented data we collect and maintaining our privacy (Heidegger, 1982: 4, as quoted in Crampton, 2010: 6); technology,
as personal communication devices, web services, electronic science, and society do not evolve independent of one another. As
cards, and surveillance cameras (in developed countries) create such, questions of society have become increasingly important as
digital shadows (Klinkenberg, 2007) of our everyday lives. In a geotechnologies have become more personalized.
short time, we have gone from representing places and features This chapter complements other chapters of this book with
to populating our geospatial databases with our doppelgangers, a sociotechnical discussion of how remote sensing, GIS, and
whose controls, ironically, lie in hands other than our own and GNSS are codependent and will continue to strengthen the
who will continue to roam within large databases, long after we foundations of our newly geo-enabled societies. We avoid the
cease to exist ourselves. deep technical trenches to keep the reader focused on the mac-
The start of this geo-information revolution goes back to the roscopic trends, since technical discussions of these technolo-
1960s, when three distinct clusters of major geo-information gies abound elsewhere. The material presented here is intended
technologies started to form the expansive geo-information to help readers recognize these technologies at work in ordi-
landscape of today: remote sensing, geographic information nary life situationsnot just in scientific and professional
system (GIS), and Global Navigation Satellite System (GNSS) settings (which is already covered extensively in other chap-
technologies. At their core, both remote sensing and GNSSs ters of this book). With this in mind, the remainder of this
are geospatial data collection systems, while GISs have multiple chapter is organized as follows. In Section 20.2, we review the
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 425
historical development of remote sensing, GIS, and GNSS tech- intellectual debates about how and if GIS can matter to societal
nologies as a context for discussing the democratization of the development, recognition of the importance of geographic infor-
geospatial domain in the twenty-first century. In Section 20.3, mation and technologies by governments, development of Internet
we explore several application domains to showcase the diverse technologies, and emergence of the open-source movement. This
ways in which these technologies are being integrated into our short period was quite critical in setting the stage for the defining
everyday lives. In Section20.4, we provide a critical analysis of developments of the ongoing contemporary phase, especially those
the technical challenges in trying to make geo-information and related to the democratization of geographic information.
related technologies easily accessible to the public, as well as the In this section, we present a representative, but not exhaus-
political, organizational, social, and ethical issues that must be tive, chronology of developments during the enlightened, tran-
considered in an increasingly geospatially enabled society. We sition, and contemporary phases to provide a basic background
conclude with a brief summary of this chapter in Section 20.5. for reasoning about the events that have fostered the publics
awareness and understanding of geo-information. We do not
20.2 toward Democratization of discuss the historical phase hereafter, since our concern is
mostly with later geo-information technologies and events that
Geo-information technologies have primarily shaped the twenty-first century democratization
For centuries, paper maps were the state-of-the-art technology for of geographic information.
representing, analyzing, and communicating geographic infor-
mation. While they still play a significant role in many spheres of
20.2.1 enlightened Phase(twentiethcentury):
our personal and professional lives today, paper maps are being
State-Sponsored Geo-information
replaced by services on mobile phones, virtual globes, Google
technologies
Glass, and other Geo-ICTs. The history and development of spatial
representations from prehistoric maps to modern spatially enabled The enlightened phase started in earnest with the massive
mobile devices are characterized by Goodchild etal. (2007) accord- increase in surveying and mapping needs during the two World
ing to three stages of development, namely, historic, enlightened, Wars and with the rise of the scientific method in the early to
and contemporary. During the historic phase, the longest of the middle twentieth century. Large-scale investments in geo-infor-
three phases, map production was uncoordinated, incomplete, and mation technologies by governments (as detailed in Table 20.1)
undertaken by both public and private entities for various human led to the development of national mapping agencies (both mili-
needs. The enlightened phase began toward the beginning of the tary and civil) with large budgets. The agencies were typically
twentieth century with the advent of state-sponsored mapping charged with the task of producing high-accuracy maps of the
initiatives and evolved into the contemporary phase starting in physical and social state of nations at several scales. This fostered
the last decade of the twentieth century. This last phase is ongoing the development of several new technologies including aerial
and characterized by increasing emphasis on cooperation and data photogrammetry, remote sensing with satellites (e.g., Television
sharing, as well as movement toward greater access to the modes Infrared Observation Satellite [TIROS], Landsat), automated
of constructing and using geographic information. In this chapter, cartography, GIS, and the first GNSSthe U.S. Navstar GPS.
we extend Goodchild etal.s (2007) original model with a transition The technological developments and emerging benefits to soci-
phase to clearly distinguish the intervening period between the ety and science were unquestionable. They were, however, not
enlightened and contemporary phases. This period started with equally distributed among nations as most of the developments
TABLe 2 0.1 Important Geo-Information Technology Developments during the Enlightened Phase
Year Enlightened Phase Events (19601989)
1960 First nonmilitary satellite, TIROS-1, launched by the United States for space-based meteorological observations.
1964 Canada Geographic Information System project launched.
1967 U.S. Census Bureau develops Dual Independent Map Encoding vector topological data model.
1969 ESRI, the leading GIS software and services company today, is founded.
1972 Landsat 1, the first civilian Earth-monitoring satellite, is launched.
1978 Map Overlay and Statistical System, the first full-fledged interactive vector and raster GIS, deployed at many U.S. Federal agencies.
1982 ARC/INFO released by ESRI, to become most popular commercial GIS.
1983 United States announces GPS will be available for civilian sector when operational.
1985 GPS becomes operational.
1985 Graphical Resources Analysis Support System software, a raster focused, open-source GIS, is released by U.S. Army Corps.
1986 The Australian Land Information Council is established.
1987 Handling Geographic Information report (Chorley Report) released by the United Kingdom.
1987 National Center for Geographic Information and Analysis (NCGIA) established.
1989 U.S. Census Bureau releases Topologically Integrated Geographic Encoding and Referencing products into the public domain.
2016 by Taylor & Francis Group, LLC
426 Remotely Sensed Data Characterization, Classification, and Accuracies
occurred in the United States, followed by the United Kingdom GIS was imperialistic and used to subjugate local and indigenous
and other European countries, and Australia. The United States perspectives of space, place, and culture in favor of the majoritys
led the world in topographic mapping, remote sensing, GIS, and perspectives. GIS and its practitioners were accused of helping
GNSS. In addition to government-sponsored initiatives, signifi- strengthen hegemonic narratives serving the powerful elite, who
cant contributions were made by academics and by the commer- spared no thought for marginalization of those on the less for-
cial sector with the creation of popular cartographic, GIS, and tunate side of the economic, social, and digital divide (Pickles,
remote-sensing software. The development of commercial soft- 1994; Sheppard, 1995; Harley, 2002). Others criticized specifi-
ware, in particular, facilitated collaborations between govern- cally the military origins and historical complicity of remote
ment and private entities and led to the creation of a special class sensing, GIS and GNSS in military, colonial, racist and discrimi-
of geospatial professionals. At the time, the average person, how- natory practices (Crampton, 2010: 7) and sought to expose the
ever, was still dependent on various types of maps produced by the role of such technologies as weapons of intelligence and war
professionals to understand their own spaces. It is for this reason (Smith, 1992; Cloud, 2002; Monmonier, 2002). These social cri-
that this era, particularly from the late 1980s onward, has been tiques were ultimately channelized, and a set of critical research
criticized for creating a hegemonic system that perpetuated the issues were identified at the Friday Harbor National Center for
grip of powerful agencies and trained professionals on geographic Geographic Information and Analysis meeting in 1996 (Harris
information. This critique marks the beginning of the transition and Weiner, 1996). The results from this and some other similar
phase, during which the importance of reconceiving mapping engagements provided the foundation for a broad GIS and soci-
technologies as products of and for society became evident. ety research agenda that has thrived since the late 1990s.
A particularly long-lasting, tangible outcome of the GIS and
society debates was the rise of grassroots mapping projects led by
20.2.2 transition Phase (twentieth century):
communities and facilitated by GIS and mapping experts. These
Governance and Scholarship for Society
projects, which may take on many forms, have been called par-
During the transition phase (Table 20.2), some social theo- ticipatory or community mapping projects, community integrated
rists began to argue that the seductive power of technologies GISs, public participation GISs (PPGIS), or participatory GISs
had blinded many users from the more complex sociopolitical (PGIS), depending on the goals, participation, mapping process,
realities of technological development (Chrisman, 1987; Taylor, and technologies used (Craig et al., 2002; Sieber, 2006; Elwood,
1990). The epistemological interpretation of GISs (and related 2011; Brown, 2012). These methods of geographic information cre-
technological developments) was considered quite intellec- ation and/or use are context and issue driven rather than technology
tually impoverished since the positivist, and absolute space- driven (Dunn, 2007). Over the last 20years, hundreds of such map-
focused view of conventional GIS, was in denial of the social ping efforts have been undertaken worldwide. The projects, rang-
roots and impacts of GIS and the social constructivist nature ing from basic sketch mapping to sophisticated online map-based
of geographic information (Warf and Sui, 2010). This led to the surveys, have been undertaken to help communities broaden their
so-called GIS wars, instigated by social theorists criticism that spatial, environmental, social, and political perspectives, address
TABLe 2 0.2 Important Geo-Information Technology-Related Developments during the Transition Phase That Paved the Way
for Contemporary Processes of Democratization of Geo-Information
Year Transition Phase Events (19891998)
1989 Association for Geographic Information lobbying/advisory group established in United Kingdom.
1990 U.S. Federal Geographic Data Committee is established.
1990 GIS wars erupt between social theorists and GIS theorists/practitioners.
1991 Miniature GPS receiver technology becomes widely available.
1992 The term Geographic Information Science is coined (Goodchild, 1992).
1992 U.S. Remote Sensing Act of 1992 opens skies to commercial satellites.
1993 The NCGIA Geographic Information and Society meeting is held at Friday Harbor, Maine.
1993 European Umbrella Organisation for Geographic Information is established.
1994 U.S. President Bill Clinton signs Executive Order 12906 Coordinating Geographic Data Acquisition and Access: The National Spatial
Data Infrastructure.
1994 OpenGIS Consortium (now Open Geospatial Consortium) is formed.
1995 The U.S. Navstar GPS attains full operational capability.
1995 The Predator unmanned aerial vehicle becomes operational.
1996 MapQuest and MultiMap, the earliest interactive web-mapping services based on government-collected public domain data, are
released in the United States and United Kingdom.
1996 NCGIA Initiative 19: GIS and society workshop is organized.
1996 Public participation GIS is defined at an NCGIA workshop at the University of Maine.
1996 U.S. Federal Communications Commission requires wireless carriers to determine and transmit the location of callers who dial 9-1-1.
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 427
land ownership disputes, conserve and manage natural resources, than 100 countries (Masser, 2011). SDIs are still largely one-way
document and protect cultural heritage, revitalize neighborhoods, vehicles for sharing government data and still have to evolve
and challenge existing narratives of their community. The proj- to become two-way vehicles, allowing the public to also con-
ects are fertile grounds for the convergence of cartography, GISs, tribute to the SDI (Budhathoki et al., 2008). Despite this limi-
GNSSs, and remote sensing. Topographic maps, aerial photos, tation, the existence of these first-generation SDIs made many
satellite imagery, GNSS collected waypoints and tracks, and GIS- contemporary phase developments possible. During the transi-
based demographic and environmental data may all contribute tion phase, several private/public groups such as Association for
to a groups mapping and decision making. This is evidenced by Geographic Information and European Umbrella Organisation
the PPgis.net electronic forum and the Integrated Approaches to for Geographic Information (EUROGI) were also established to
Participatory Development (IAPAD) website,* which are world- further the interests of geographic information communities.
wide resources for those seeking to develop PGIS/PPGIS/partici-
patory mapping projects, especially in developing countries.
20.2.3 contemporary Phase (twenty-First
The development of the GIS and society research agenda
century): theVision of
coincided with the development of a new multidisciplinary field
Democratization
called GIScience, a field that explores the conceptual founda-
tions, design, and the application contexts of geospatial tech- The contemporary phase began during the first decade of the
nologies (Goodchild, 1992). While the definition and scope of twenty-first century (see Table 20.3 for a list of important events).
the field has continued to evolve (Goodchild, 2012; Blaschke and Unlike the enlightened phase, in which geo-information technolo-
Merschdorf, 2014), the one thing that remains constant is the idea gies were managed by large institutions in a fundamentally top
that GIScience research proceeds in at least three dimensions: down hierarchy, the contemporary phase is defined by a much more
computer, individual, and society (Longley etal., 2010). The issues democratized approach to geo-information. Contemporary uses of
of interest to GIScience specialists are diverse, pertaining not just GISs and related technologies have become more context sensitive
to the computational contexts of geographic data use but also to and issue driven, with some (but not nearly enough) recognition to
the cognitive, behavioral, social, legal, and ethical factors that gov- issues of power, commodification, and surveillance (Warf and Sui,
ern the creation, dissemination, and consumption of geographic 2010). This democratized framework is only partly the result of the
information. Arguably, society still remains the weakest dimen- government SDIs, academic critiques and research agendas, and
sion of GIScience, but compared to the time of origination of the grassroots participatory mapping initiativesa major push came
GIS and society debates, there is much better understanding today from the rise of several private sector companies that invested heav-
of the two-way relationship between geo-information technologies ily in online and mobile mapping technologies. By the start of the
and society (Warf and Sui, 2010). This is supported by the success new century, government agencies across the world were searching
of the PPGIS/PGIS movement and the more recent establishment for new ways to provide geospatial services in a time of reduced
of critical GIS as an important area of research within GIScience budgets, Geo-ICTs that dropped the entry barrier for geographic
(Schuurman, 1999; OSullivan, 2006; Wilson and Poore, 2009). information collection and sharing, and an intellectual shift from
During the transition phase, several key technological develop- mapping static places to representing dynamic activities of people
ments occurred to lay the groundwork for broader access and so- across places over time (Goodchild etal., 2007). In the contempo-
called democratization of Geo-ICTs in the current contemporary rary phase, governments still play a major role but increasingly as
phase. These included the replacement of older, more cumber- facilitators (e.g., maintaining national SDIs). They are also more
some GPS devices with miniature receivers, operational comple- inclusive, involving citizens in topographic and resource mapping
tion of the GPS satellite system, development of unmanned aerial initiatives, environmental monitoring, disaster preparation, and
vehicles (UAVs), and, of course, the popularization of the World health and emergency services. The development of inclusive, bot-
Wide Web. The development of open-source GIS, which started tomup (individual and collaborative) processes for creating and
with Geographic Resources Analysis Support System (GRASS) in sharing geographic information is made possible by integrating
1985, progressed enough to justify establishment of a consortium the functionality of traditional geospatial technologies with newer
of open-source GIS software systems. Meanwhile, the commer- Geo-ICTs (Goodchild, 2007; Sui etal., 2013). The development of
cial sector began to see the potential market in interactive web- technologies like OpenStreetMap (OSM), Microsoft Virtual Earth,
mapping services (WMSs) such as MapQuest and MultiMap. Google Earth, and Google Glass are prime examples.
As a final point of interest, the transition phase should also At the end of the twentieth century, a seminal event occurred
be recognized for the establishment of government-run agen- in 1998, when U.S. Vice President Al Gore presented an inspiring
cies, national spatial data infrastructures (SDIs), and related vision called Digital Earth as a multiresolution, three-dimen-
organizations (e.g., U.S. Federal Geographic Data Committee sional representation of the planet, into which we can embed vast
[FGDC]), which were designed to coordinate the collection and quantities of geo-referenced data (Gore, 1998). Gore proposed the
sharing of geographic data between government agencies and Digital Earth as a comprehensive information system composed of
with the public. Today, SDIs have been implemented in more many distributed components, together providing access to all his-
torical and current information about the entire planet (including
* www.iapad.org. the activities of its inhabitants), and supported by modeling tools
2016 by Taylor & Francis Group, LLC
428 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 2 0.3 Important Geo-Information Technology-Related Developments That Have Proven to Be Central to the Ongoing
Democratization of Geographic Information in the Contemporary Phase
Year Contemporary Phase Events (1998Present)
1998 U.S. Vice President Gore presents the Digital Earth vision.
1998 Microsoft launches TerraServer in partnership with U.S. Geological Survey (USGS).
1998 NASA starts the Digital Earth Initiative and creates the web-mapping system data-sharing standard.
1999 The first U.S. commercial satellite, IKONOS, is successfully launched with a very high panchromatic spatial resolution of about 0.82 m.
2000 Selective availability, which degraded Navstar GPS signals for civilians, is switched off.
2001 Keyholes EarthViewer and GeoFusions GeoPlayer virtual globes are released.
2001 The infrastructure for spatial information in Europe is launched by the European Union.
2001 Wikipedia, a crowdsourced encyclopedia, is launched.
2002 USGS launches The National Map web service.
2002 Friendster, a social networking website, goes online.
2003 U.S. e-government data access initiative Geospatial One-Stop goes online.
2003 Urban and Regional Information Systems Association approves Code of Ethics for geospatial industry professionals.
2004 The National Imagery and Mapping Agency is reoriented as National Geospatial-Intelligence Agency.
2004 OpenStreetMap launched as first crowdsourced, public domain street mapping database.
2004 Facebook and Flickr are launched.
2005 The Google Earth, Google Maps, and Microsoft Virtual Earth geobrowsers are released.
2005 Global Earth Observation System of Systems 10-year implementation plan adopted by intergovernmental Group on Earth Observations.
2006 Twitter is launched.
2006 The term neogeography is proposed to describe creation and use of geospatial information by nonexperts.
2007 Concepts of citizens as censors and volunteered geographic information (VGI) are promoted and gain a foothold (Goodchild, 2007).
2007 The National Science Foundation promotes a cyberinfrastructure to foster collaborative research and data sharing.
2007 Apple releases the first iPhone and Google releases the Android mobile operating system.
2008 Ushahidi VGI platform launched to track violence during Kenya post-election crisis.
2010 Google announces its autonomous car project.
2013 Google Glass is launched for testing purposes.
2014 The U.S. government relaxes restrictions on satellite imagery to allow image resolutions below 50cm for commercial purposes.
for predicting future conditions. Gore envisioned the resource as set of technological services and practices serving as a collective
a publicly available one-stop virtual environment in which any- geographic knowledge organization and information retrieval
bodychild or adultcould explore information about the Earth system (Craglia etal., 2008; Grossner etal., 2008). In this regard,
effortlessly (e.g., on a magic carpet that could fly through space and the virtual globes of today can be imagined as small private digital
also back in time). In many ways, Gores presentation of his vision Earths that offer access to the global Digital Earth system.
can be seen as the start of the contemporary phase. It encour- The popularity of WMSs and virtual globes established geo-
aged existing initiatives and also stimulated new initiatives in the browsing, that is, the search for geographic information via a map
government and the private sector where. For example, the U.S. interface (Lemmens, 2011) as a near ubiquitous phenomenon,
National Aeronautics and Space Administration (NASA) initiated with new users, services, and regions still being added regularly.
the Digital Earth Initiative in 1998 and, among other things, cre- Geobrowsing is an intrinsically geospatial endeavor that is so intui-
ated the current WMS standard, crucial for seamless sharing of tive that even young school kids can engage in itbecause of intel-
geographic data and services on the web (Grossner et al., 2008). ligent obfuscation of certain peculiarities of geographic data (e.g.,
NASAs World Wind was an early virtual globe inspired by the scale and projection). The value of geobrowsing lies in how it brings
Digital Earth vision. Today, Googles Google Earth, Microsofts mapping down to the level of the everyday user and encourages
Bing, Environmental Systems Research Institute (ESRIs) ArcGIS people to engage with geographic information. For example, geo-
Explorer, and SkyGlobes TerraExplorer are the popular virtual browsers can be used to explore places, assist with everyday navi-
globe platforms.* While NASA does not maintain special units to gation and way-finding tasks, visualize data, and create new maps
support this vision, the vision is now promoted by the International and datasets. They highlight the manner in which our everyday
Society for Digital Earth, spearheaded by China. The Digital Earth activities are tied to space. With the growing popularity of geo-
vision today has evolved significantly since Gores speech, and browsing applications on mobile devices (Google Maps was the
today, it is understood more pragmatically as a globally distributed most frequently downloaded smartphone application in 2014), it
seems reasonable to suggest that there is a strong desire in people
* See http://en.wikipedia.org/wiki/Virtual_globe for a complete list of vir-
tual globes. www.businessinsider.comgoogle-smartphone-app-popularity-2013
http://www.digitalearth-isde.org. 9#infographic (Accessed July 17, 2014).
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 429
Google Maps Google Maps
Bing Maps OpenStreetMap
FIg u r e 20.1 As the high-resolution satellite image indicates, the slum in Kibera is a heavily populated part of Nairobi. While OSM map cap-
tures substantial detail of the structures and roads collected from community-mapping efforts, Google Maps and Bing Maps do not map most
structures in Kibera since they are deemed illegal by local authorities. (Screenshot from www.tools.geofabrik.de/mc.)
from all walks of life and professions to directly engage and manip- knowledge of place-names. Wikimapia is a comparable database
ulate geographic information rather than passively absorb it from that actually contains many more user-generated entries for place-
static maps. Clearly, in the twenty-first century, we have moved on names than any official list of place-names. These databases are
from merely looking at maps to participating and interacting with important repositories of socially generated geographic informa-
geobrowsers. tion and, in some countries, may be the only digital maps avail-
Fueled partly by the success of the virtual globes, WMSs (i.e., able to the public for economic or political reasons (see Figure 20.1
geobrowsers), and easy-to-use GNSS units, the contemporary for an example). Further, even if official or commercial maps are
phase has witnessed an explosion of user-generated geographic available, crowdsourced maps offer diversity by recording users
informationa phenomenon called neogeography. Neogeography perspectives of space. As Monmonier (1996) states, multiple maps
refers to the practice of using mapping techniques and tools for can be made for the same place, with the same data, for the same
and by nonexpert individuals or communities (Turner, 2006). situation. The diverse nature and versatility of these datasets have
Its applications are generally not analytical or formal but mostly been invaluable in efforts to support humanitarian relief fol-
descriptive and visual. The rise of neogeography lies in a synergy lowing natural disasters and to empower communities through
that has developed between various technologies (e.g., Geo-ICTs, grassroots mapping (as will be discussed in the next section).
GNSS-enabled mobile devices, and Web 2.0) and social data col- Over the last 15years, many of the barriers to democratiza-
lection practices (e.g., crowdsourcing and volunteered geographic tion of geographic information have disappeared or at least
information) that allow everyday users to create mash-ups of map become less relevant. This is evidenced, to a certain degree,
services and individualized content (Goodchild, 2007, 2009; Sui by the Digital Earth initiative and by neogeography efforts to
etal., 2013; Wilson and Graham, 2013). Evidence of this synergy design collaborative, bottomup processes for creating and shar-
can be seen in the popularity of crowdsourcing repositories like ing geographic information. Yet, despite this progress, efforts
OSM* and Wikimapia. OSM is a free-to-use global map database to democratize geographic information still have a long way
of geographic features built by individual volunteers relying on to goneogeography has limited connection to the academic
a mix of resources including satellite imagery, GNSS tracks, and domain of geography or GIScience (Goodchild, 2009), and its
claims of democratization and making geographic information
* www.openstreetmap.org. available to anyone, anywhere, and anytime (Turner, 2006) have
www.wikimapia.org. been shown to be premature and shallow (Haklay, 2013). Still,
2016 by Taylor & Francis Group, LLC
430 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 20.2 The USGS National Map provides free topographic mapping and data-downloading services for the entire United States.
(Screenshot from www.viewer.nationalmap.gov.)
the developments that define the contemporary phase suggest datasets are also used extensively in the rectification, classification,
that the traditional technologies of remote sensing, GISs, and mapping, and dissemination of remotely sensed imagery. The inte-
GNSSs can be used to empower people and help them view the gration of these technologies is also evident; in these ways, data are
world from a perspective other than a traditional topdown, often seamlessly distributed to professional users. For example, the
institutional frameworks of the recent past. U.S. Federal Governments Geospatial One-Stop project provides
a single portal to geographic information. One of its services, The
National Map* (see Figure20.2) web service, supports both inter-
20.3 Discovering Remote Sensing, active online visualization and free downloading of a wide array of
GiS, and GnSS Applications historical and current datasets including satellite imagery, eleva-
in our Daily Lives tion, land cover, transportation, hydrography, boundaries, and
geographic names for the United States.
In this twenty-first century contemporary age, the functionalities In addition to the more standard methods for integrating
offered by remote sensing, GISs, and GNSSs are finally becom- remote sensing, GISs, and GNSSs, outlined earlier, a variety of new
ing increasingly integrated into our daily lives. This integration is increasingly democratized methods have evolved during the con-
largely the result of new Geo-ICTs, as well as an increasing pub- temporary phase of geospatial development. For example, PPGIS
lic awareness of the value of location as a search parameter when projects have taken advantage of low-cost, high-accuracy GNSSs
seeking information about the world. Over three decades worth of to map local communities and develop representations of features
research exists on the conceptual and technical aspects of efforts that are not evident on existing maps or from satellite imagery
to integrate these technologies (Gao, 2002; Mesev, 2007; Merchant alone. In communities where current and/or high-resolution imag-
and Narumalani, 2009). GNSSs are used to georeference aerial ery is not available, amateur aerial photographs have been devel-
photos and satellite images and are a core component of high-res- oped with makeshift aerial cameras consisting of GNSS-enabled
olution light detection and ranging (LIDAR) remote sensing of the digital cameras attached to helium-filled balloons. The aerial cam-
Earth. Both remotely sensed data and GNSS data can be imported eras have been used to map small neighborhoods at a cost of less
and overlaid with other data layers in a GIS environment. In a GIS, than U.S. 35 per hectare (Seang etal., 2008). Perhaps, the most
these layers can be used to create and analyze a variety of built (e.g., obvious example for owners of mobile Geo-ICTs arises in the con-
buildings, roads) and environmental (e.g., elevation, hydrography, text of geobrowsing in which high-resolution satellite images and
land use/land cover) datasets. While it is more common to incorpo-
rate remotely sensed and GNSS-derived data in a GIS, GIS-derived * www.NationalMap.gov.
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 431
aerial photos, made available by a service provider (e.g., Google start of LBSs can be traced to a 1996 U.S. Federal Communications
or Microsoft), are used as base maps for displaying thematic data Commission mandate that set a minimum accuracy for determin-
layers. These integrations are often referred to as map mash-ups, ing the location of an emergency E911 call by a wireless device.
or simply mash-ups. Similar combinations of imagery and data Although this policy was oriented toward emergency response,
layers presented in 3D virtual globe software (e.g., Google Earth, resulting improvements in location accuracy have fostered
NASA World Wind) are also used to create virtual fly-through innovation in the marketplace by catalyzing the development of
animations. In addition to government-run geospatial data clear- mobile mapping programs and LBSs. As a particularly interesting
inghouses, public/private clearinghouses like OSM have also been and futuristic example, consider the recent initiative by Google
developed, which allow everyday users to contribute and use geo- and others to develop automated cars (see Figure20.3). In 2010,
spatial data in a community-driven setting. The ability to both edit Google announced that it was working on a project to develop
and download the entire OSM database locally is a tremendous self-driving cars (Thrun, 2010). Not surprisingly, the project
benefit that no other map service provides. leader was also the coinventor of Googles Street View mapping
The integration of geo-information data and functionalities technology. Other automotive companies and universities are also
provide the backbone for many of the services and tools that are developing prototypes and testing them on roads. Google is ahead
developed by government agencies, nonprofit agencies, and busi- of most though, since it has already conducted more than half a
nesses to help people make everyday decisions (Yang etal., 2010; million miles of pilot testing. Driverless cars will be legal on the
Lemmens, 2011). Such services include but are not limited to emer- road from January 2015 in the United Kingdom and already so in
gency response, disaster response and planning, natural resources the U.S. states of California, Florida, and Nevada, with legislation
management, environmental and public health services, weather pending in many others (Weiner and Smith, 2014). Autonomous
forecasting, floodplain mapping, precision farming, agricultural cars are stellar examples of Geo-ICT technology and services inte-
services, community planning, property and utility mapping, gration. Such cars now depend on LIDAR and/or 360 cameras
crime monitoring and analysis, operations research, traffic moni- and computer vision technology, and currently data from driving
toring, real estate services, and even K-12 education. To explore patterns of cars driven by human beings, to create real-time high-
this topic in greater detail, we present four broad contexts in which resolution 3D models of a cars surroundings. It is estimated that
Geo-ICTs are interwoven in peoples everyday lives: location-based they may need to sense about 1 GB of data every second to work
services (LBSs), disaster relief and emergency management, par- optimally today! The GNSS-derived location and LIDAR model
ticipatory mapping, and participatory sensing of our everyday also need to be combined with information from a GIS database
physical and social environments. In presenting these examples, of static infrastructure (e.g., telephone poles, crosswalks, and traf-
particular attention is paid to the manner in which these applica- fic lights) to identify all types of static and moving objects (e.g.,
tions facilitate democratization of geographic information. pedestrians, cyclists) to plot a safe path through space (Fisher,
2013). Similar to cars, lightweight UAVs or drones may be found
at the disposal of any citizen in the future for collecting remote-
20.3.1 Location-Based Services
sensing data about any area of interest for business, environmen-
The term location-based service, or LBS, generally refers to soft- tal, recreational, and other purposes. Apart from such futuristic
ware applications that use location as a basis for providing infor- vehicles, smartphones and the recently launched Google Glass are
mation or performing a service for a user (Zipf and Jst, 2012). The also great platforms for LBS integration.
(a) (b) (c)
FIg ur e 20.3 (a) A car retrofitted by Google for collecting 360 panoramic street view photos. (Photograph distributed by Wikipedia author,
Kowloonese, under a CC-BY 2.0 license. http://en.wikipedia.org/wiki/Google_Street_View#/media/File:GoogleStreetViewCar_Subaru_Impreza_
at_Google_Campus.JPG.) (b) A car retrofitted by Google to be tested as part of its self-driving car fleet. (Photograph by Jurvetson, S., modified
by Mariordo, distributed under a CC-BY 2.0 license. http://commons.wikimedia.org/wiki/File:Googles_Lexus_RX_450h_Self-Driving_Car.jpg.)
(c)Googles prototype of a fully automated self-driving car with numerous sensors but no manual traditional controls (steering wheel and accelerator
and brake pedals). (Modified image available from Google.com at http://googleblog.blogspot.com/2014/05/just-press-go-designing-self-driving.html.)
2016 by Taylor & Francis Group, LLC
432 Remotely Sensed Data Characterization, Classification, and Accuracies
LBSs today represent such an important category of business With such a diverse list of activities, it should be obvious
that it is being fiercely competed over by information technol- that the commercial sector is agog with LBS innovations and
ogy, telecommunication, mobile phone, and increasingly other improvements. LBSs almost always combine data and function-
types of companies. Today, LBSs are designed to help people alities from GIS, mobile cartography, and GNSS. For example,
make decisions in the contexts of an incredibly wide variety of Google Maps provides several types of map services ranging
personal and professional matters (Raper etal., 2007) including from locating and describing landmarks, routing directions for
but not limited to the following: different modes of transport (driving, walking, biking, and pub-
lic transport), traffic patterns, and traffic alerts. In Figure20.4, a
Navigation: Turn-by-turn navigation; public transit; traf-
photomontage derived from several screenshots depicts exam-
fic and road condition updates; roadside emergency assis-
ples of these LBSs as they would look on a single-map view. Bing
tance; user-centered route selection; fuel consumption;
Maps, MapQuest, and other mapping sites also provide simi-
autonomous car; vehicle to vehicle communication
lar services. Apart from such map or imagery-based LBSs, more
Retail/Business services: Retail advertising; store locators;
specialized map-based LBSs include weather advisory services,
shopping aisle information services; retail store/mall maps;
field surveying by professionals or by untrained volunteers, and
real estate services; credit/bank card fraud prevention
disaster relief work after earthquakes, fires, floods, and storms.
Recreation: Online and real-world games; sky gazing; geo-
LBSs that post real-time alerts about environmental conditions,
social networking; mobile place guides; location-based
manage drone delivery services, and support augmented reality
augmented reality; travel/tourism services
Geo-ICTs such as Google Glass should be quite popular soon.
Societal services: Location-based warnings and alerts; seek-
The reach of LBSs into both mundane and critical decision-mak-
ing/providing emergency help; disabled people mobility;
ing processes will also continue to get deeper, with technological
toll collection; weather services; environmental services;
developments such as 3D mobile cartography, ubiquitous posi-
agricultural services; participatory community planning
tioning (Mannings, 2008), and ubiquitous computing/Internet
Mapping: Crowdsourcing points of interest; monitoring
of Things (Weiser, 1991) making mobility patterns of people, not
environmental conditions; infrastructure maintenance;
just their locations, the focus of LBSs. The implications of such
participatory/citizen sensing; mapping and monitoring
deep LBS integration are obviously both exciting and unsettling
personal health; documenting geocoded events; mapping
at the same time. There are many issues and challenges that will
disease outbreaks; finding or tracking people/animals/
come to the forefront regarding service reliability; information
objects; mash-up services
overload; energy use, humanmachine interaction; personal
FIgu r e 20.4 Examples of various map-based services available through Google Maps (www.maps.google.com). Note that this image is a com-
posite of several computer screenshots, since these services cannot be simultaneously viewed on the same map view (as of August 2014).
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 433
privacy; provision of social, environmental, and health services; In a different context, Goodchild and Glennon (2010) discuss
perception and experience of places; social networking; and cog- the case of volunteered geographic information (VGI) being use-
nitive and behavioral modifications. ful in making quick real-time decisions during four wildfires near
Santa Barbara between 2007 and 2009. The volunteers used GNSS-
20.3.2 Disaster Relief and emergency enabled cameras and phones to report information as text, pho-
Management tos, and videos over the Internet. For one wildfire, some citizens
were able to access and interpret comparatively fine temporal and
On November 8, 2013, Super Typhoon Haiyan swept through the
spatial resolution imagery from the Moderate Resolution Imaging
Eastern Visayas region of the Philippines, killing 6000 people, dis-
Spectroradiometer satellite sensor, and, using services such as
placing 1.9 million and impacting over 9 million (REACH etal.,
Google Maps, created several maps that were often more up to date
2014). Even before the typhoon hit, a Humanitarian OpenStreetMap
than official maps. For another fire, volunteer map sites were set
Team (HOT) was created in anticipation to map infrastructure and
up to readily synthesize the volunteer and official information as
damage in the affected region. Harnessing the power of global vol-
it became available. This real-time lifesaving participatory sensing
unteers to quickly evaluate postdisaster satellite imagery, the HOT
of an endangered environment bears testimony to the power of
performed damage assessment and facilitated prioritization of
democratization of geo-information and placing remote sensing,
scarce resources. Maps created from the effort were visible to the
GISs, GNSSs, and other Geo-ICTs in peoples homes.
public within days of the disaster (Buchanan etal., 2013).
Until only a few years ago, it was the norm that government
The Haiyan HOT was just the latest in a long list of geospa-
and transgovernmental agencies (e.g., United Nations) would
tial crowdsourcing efforts for disaster relief. This growth can
lead disaster relief, with NGOs playing a supporting role. Emrich
be traced back at least to Hurricane Katrina in 2005, when a
et al. (2010) provide an excellent review of the history of use of
computer programmer in Austin, Texas, created a website that
remote sensing and GISs in preparing or responding to emer-
allowed users to post descriptions of local conditions onto a map
gencies. Remote sensing and GISs have been used in all phases
(Singel, 2005). In 2007, a local radio station contributed to cov-
of emergency management for multiple decades now. While the
erage of the San Diego wildfires in California by maintaining
early focus was on understanding the physical processes, at the
a dynamic web map of user-reported conditions (Zook et al.,
beginning of the twenty-first century, geo-information technolo-
2010). In 2009, OSM was first used for humanitarian purposes
gies were being used in all phases of emergency management
in the Philippines in response to Tropical Storm Ondoy. While
though still primarily to support the topdown hazard research,
crisis mapping* as a concept was beginning to take hold in 2009
analysis, and disaster response command structure (Emrich etal.,
(Soden and Palen, 2014), it came to the fore with the formation of
2010; Zook etal., 2010). It is only since 2010, due to technological
HOT following the January 12, 2010, earthquake in Haiti.
advances, supporting services, and changing mindsets, that virtu-
In the days that followed the Haiti earthquake, relief workers
ally every major disaster has stimulated VGI/crowdsourcing map-
posted information on locations of trapped survivors, damaged
ping. These have included the Queensland and Australian floods
buildings, and triage centers; a global network of volunteers trans-
in 20102011; the earthquake in Christchurch, New Zealand, in
lated information from Haitian Creole, geotagged it, and placed
2011; the earthquake and tsunami in northern Japan in 2011; the
it into online mapping services (created based on remote-sensing
2013 Super Typhoon Haiyan in the Philippines; the 2013 torna-
imagery) for other relief workers to use. The effort was made pos-
does in Illinois, United States; and even the search for the missing
sible by the maturation of communities such as OSM, Ushahidi,
Malaysian Airlines aircraft MH-370 in 2014, which enlisted mil-
Geocommons, and CrisisCamp Haiti (Pool, 2010), the coopera-
lions of volunteers scanning imagery for evidence of the missing
tion of private companies (e.g., Google, Microsoft, and GeoEye),
aircraft. The trend is likely to become more popular as adoption of
government agencies, nongovernmental organizations (NGOs),
mobile devices continues in all parts of the world.
and Global Earth Observation Catastrophe Assessment Network
While the crowd and volunteer cannot be a replacement for
(GEO-CAN)a unique voluntary network of more than 600
domain experts, these successful relief efforts still clearly dem-
experts from 23 countries. These entities relied heavily on infor-
onstrate two things. The first is that crowdsourcing, aided by
mation produced from the convergence of GIS software and data,
appropriate online and field Geo-ICTs, can often provide critical
GNSS field data, several sources of aerial and satellite imagery,
geospatial information much faster than traditional methods.
and expert-created information products (Duda and Jones, 2011;
The second is that crowdsourcing is highly suited to the large
Van Aardt etal., 2011). Without any centralized command, these
and sudden demand created by natural and man-made disasters.
disparate actors and resources converged within hours, and the
Local volunteer groups and government agencies can also use
collective, organic systems that were created in the short time
crowdsourcing to gather crucial information that may be needed
easily surpassed what could ever be achieved by traditional top
by both officials and communities to plan ahead of time for
down, centralized government efforts.
disasters in areas especially prone to storms, floods, landslides,
fires, and disease outbreaks. On the other hand, this transi-
* www.crisismappers.net. tion from traditional command-and-control structures to dis-
www.ushahidi.com. tributed crowdsourcing efforts is not without its own problems
www.geocommons.com. and raises numerous technological, organizational, and ethical
2016 by Taylor & Francis Group, LLC
434 Remotely Sensed Data Characterization, Classification, and Accuracies
issues that will need to be addressed in the coming years. Some It should be kept in mind that OSM facility and infrastruc-
of these issues are touched upon in Section 20.4. ture mapping projects are only one kind of approach to PGIS.
Another approach is to pair trained geospatial professionals with
20.3.3 community Building local communities to help them record their spatial histories and
The story of crowdsourcing in Haiti after the earthquake could explore and communicate about community problems through
easily have dissipated as the disaster started slipping away from maps. Currently, one of the authors of this chapter (Sinha) is part
the media spotlight. Fortunately, the HOT that formed initially of an international collaborative research project called the Local
to respond to the earthquake dug its heels and worked for about Knowledge and Climate Change Adaptation Project (LKKCAP),
a year and half after the earthquake to help people claim owner- under the auspices of which several community mapping proj-
ship of their broken and impoverished communities (Soden and ects were organized in Mwanga District, Tanzania. LKCCAP
Palen, 2014). The Haitian HOT is recognized today as remarkable was funded by the U.S. National Science Foundation to explore
for several reasons: it established HOT as a serious techno-human- the relationship between local knowledge, institutions, and cli-
itarian group, changed the perception of OSM and its contributors, mate change adaptation practices embedded in rural livelihoods
became a quintessential example for champions of neogeography, of Tanzania. The community-mapping projects were designed to
and blurred the boundary between crowdsourcing, VGI, and PGIS let village residents explore social, political, and environmental
through this long-term collaborative mapping initiative. Above all, dimensions of livelihood adaptation to climate change. The proj-
HOT inspired Haitians to reclaim legal, cultural, and emotional ect involved sketch mapping (Figure20.5) and GPS field map-
ownership of the land and codify their local knowledge through ping, with collected data being digitized into GIS databases and
geospatial data and maps, thus becoming more resilient against represented on poster-sized maps to help the community hold
future threats. Undoubtedly, the project will be long cited as the discussions with district officials and aid agencies.
quintessential example of how combining geo-information tech- We have focused on OSM and rural mapping projects to high-
nologies can produce extremely valuable geo-information for the light that geo-information technologies are currently being used
massesby the masses. by nonprofessionals, even in the poorest of communities. PGIS/
The mapping success of the Haitian HOT can be traced back PPGIS and collaborative mapping are, obviously, equally relevant
to another iconic OSM-led PGIS/collaborative mapping effort and popular in more prospering communities as well. Community
the Map Kibera project.* Kibera, an informal settlement of about concerns about neighborhood planning, natural resource conser-
170,000 people in Nairobi, Kenya, has captured worldwide attention vation, climate change, flood mapping, environmental pollution,
in recent years as supposedly the largest slum in Africa. Despite such water scarcity, economic revitalization, infrastructure develop-
population, and a tremendous presence of NGOs, it is still invisible ment, social discrimination, substance abuse, indigenous cul-
on most maps of the Nairobi, appearing only as a large blank space ture preservation, and many similar issues become the starting
or an uninhabited forest. Similar to most informal settlements, point for the PPGIS/PGIS projects. Special online collaborative
Kibera developed on occupied public lands, and its illegality pre- spatial decision support systems have been developed for many
vents it from being officially mapped or becoming eligible to receive studies (Jankowski, 2011). Various social networking platforms
government services. Its residents had little awareness of its geog- and online community planning forums, such as PlaceSpeak,
raphy or their collective sociopolitical capital until November 2009 MindMixer, and MetroQuest, may be used in combination with
when OSM contributors Mikel Maron and Erica Hagen started OSM, Google Earth, Google Map Maker, and other Geo-ICTs to
the Map Kibera project, which helped Kiberans create a commu- engage communities, planners, and governments in more devel-
nity information system to empower themselves (Hagen, 2011). oped countries (and also some urban communities in developing
The two project leaders slowly incentivized and built a mapping countries). As Geo-ICTs, such as augmented reality, indoor posi-
team of youth by teaching them GNSS mapping, videography, and tioning and mapping, and mobile ground remote-sensing devices,
journalism skillsrather than paying them as NGOs had become become more common in developed societies, communities will
accustomed to doing. The achievement of OSMs Map Kibera proj- be able to generate 3D virtual and augmented reality models of
ect can be easily shown by simply comparing satellite imagery for their neighborhoodsresulting in seamless boundaries between
the area with map views from Google Maps, Bing Maps, and OSM. virtual and real-world experiences of places.
As can be seen in Figure 20.1, Google and Bing map the area with
only a few main streets. The emptiness in these officially sanctioned 20.3.4 Participatory Sensing
maps stands in stark contrast to the dense settlement visible from Participatory sensing, a mapping process in which citizens and
the satellite imagery and OSMs map, which is being used by local community groups engage in sensing and documenting their daily
residents to conceptualize their neighborhoods, demand services activities, has become more common in recent years. The scope of
from the government, and fight many social problems. Based on participatory sensing ranges from individuals making personal
the success of the Map Kibera project, mapping efforts have now
spread to other slums in Nairobi and elsewhere in Africa and Asia. www.tzclimadapt.ohio.edu.
www.placespeak.com.
www.mindmixer.com.
* www.mapkibera.org. www.metroquest.com.
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 435
FIg ur e 20.5 A participatory sketch mapping session in Kirya village located in the Mwanga District of Tanzania for helping villagers map
water infrastructure and ecological resources.
observations to the combination of data from hundreds, or even 2010). Figure 20.6 maps the results from one such NoiseTube project
thousands, of individuals that reveals patterns across an entire city using Google Earth satellite imagery as a base map, with a pie chart
(Goldman et al., 2009). Participatory sensing leads to incredibly also providing the breakdown for the variously tagged sources of
rich models of environments and spaces that people define through noise pollution by the participants. A different implementation
their activities (Sagl et al., 2014). Participatory sensing is a good was seen for the Common Scents project of the SENSEable City
example of citizen science, since the projects are generally designed Laboratory at the Massachusetts Institute of Technology, United
for a specific purpose, often by researchers interested in learning States, which equipped Copenhagen citizens bikes with location
about a phenomenon. It is this characteristic that distinguishes it aware environmental sensors to develop a mobile sensor network.
from agenda-free VGI or politically motivated PPGIS/PGIS proj- The projects goal was to gather real-time, fine-grained, air-quality
ects. The most common type of collaborative sensing and mapping data to allow citizens and local officials to assess environmental
involves the use of citizens mobile devices, which act as nodes of a conditions (Resch etal., 2009). Another project, called the Personal
sensor network to collectively measure any number of interesting Environmental Impact Report, used location records from mobile
geocoded environmental parameters such as traffic characteristics, phones to calculate estimates of an individuals impact and expo-
air quality, noise levels, luminosity, temperature, humidity, and sure to different environments (e.g., carbon monoxide emissions
radiation (Burke etal., 2006). Participatory sensing projects such as from cars) (Mun etal., 2009). A similar approach based on mobile
these have been fostered by advances in mobile phone technology, phone mobility analysis also helped identify and characterize varia-
increased access to remote-sensing imagery and relevant spatial tions in human activity in urban environments (Sagl etal., 2014).
datasets, LBSs, and the increasing social appeal of crowdsourcing. In all such projects, citizens are given the ability to pool their data
When participatory-sensed data are integrated via mash-ups, or to reveal impacts and exposures in numbers that could be useful to
displayed on virtual globes, they can be used to assess environmen- both the individual and city planners.
tal variables at multiple spatial scales or in relation to other geo- Participatory sensing also includes projects for monitoring
graphic datasets (e.g., air quality data). For example, as part of the of environments, mapping live traffic, and collecting data and
NoiseTube* project, citizens in Paris used mobile phones to mea- responding to environmental hazards real time, particularly
sure their personal exposure to noise in their everyday environment those that need to be monitored real time (e.g., storms, forest
and created collective noise maps by sharing their geolocalized and fires, floods, volcanic eruptions) and might require evacuation
annotated measurements with the community (Maisonneuve etal., or help from emergency responders within a short period of time
after a critical event. The informal network of storm chasers/
* www.noisetube.net. spotters, operational meteorologists, and television media
2016 by Taylor & Francis Group, LLC
436 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 20.6 Map of noise pollution made from participatory sending data collection by the NoiseTube project in Paris, France. Image repro-
duced with permission from the NoiseTube project. (www.noisetube.net.)
outlets is a classic example of volunteers and experts collabo- casual monitoring of personal fitness to developing early warn-
rating monitor weather threats and to help people make deci- ing systems for high-risk segments of the population. The current
sions in their daily life (Palmer and Kraushaar, 2013). Another trend clearly seems to suggest that such information, contextu-
project, undertaken by more than 100 volunteers trained by the alized with other forms of static and real-time geo-information
New Jersey statewide environmental agency, involved the col- about the local environment, will provide an important basis for
lection of ground truth verification data for a database of 13,000 personalized, governmental/institutional, and environmental
ephemeral vernal pools (breeding grounds for many endangered health decision making in the future.
or threatened species) that had been derived from satellite imag-
ery (Tulloch 2008). A web-based mapping system was set up to
collect the verification data, which must be verified by the state 20.4 Democratization of
biologists to meet legal requirements. Although the project is far Geo-information: circumscribing
from complete due to lack of budget and inaccessibility of many issues and challenges
of the pools, the projects of this nature highlight the potential of
using citizens to collect important cultural and scientific data. From the start of the geo-information revolution in the 1960s
Thus, environmental collaborative monitoring networks have remote sensing, GIS and GNSS technologies have evolved to
been proposed (Gouveia and Fonseca, 2008) as a framework play an increasingly significant role in our everyday lives. The
for combining traditional environmental monitoring networks resoundingly clear message in the previous discussions of LBSs,
with the bottomup, participatory, and open-source movement. disaster relief, community mapping, and participatory sensing
Personal mobile health monitoring has also become an is that people now have incredibly diverse opportunities to col-
extremely popular activity/sport for many people in urban cen- lect and use geographic information to improve their lives. These
ters. This includes the use of personal fitness devices for track- changes are remarkable and have undoubtedly saved countless
ing physiological parameters such as heart rate, perspiration, lives and empowered large new segments of the global population.
steps or distance traveled, and velocity. Many devices are also At the same time, it is important to recognize that technological
geo-enabled, thereby allowing users to map their physiological revolutions have a tendency to outpace critical theories and insti-
parameters in relation to the places they visit during the day. The tutional structures designed to assure that their use furthers the
benefits of geo-enabled personal health monitoring range from social and cultural goals of society at large (Chrisman, 1987).
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 437
The democratization of Geo-ICTs is also affecting the manner in
which data are analyzed and translated into actionable informa-
OpenStreetMap
tion. In this section, we examine some impediments to democ-
data in the
ratization and also remind readers that uncritical submission to Phillipines
the easy appeal of trendy Geo-ICTs is regressive and problematic.
20.4.1 Quality of information and Services
As the task of geospatial data collection has been transferred from
Manila
geospatial specialists to the general public, many concerns related
to quality (accuracy, completeness, geographic diversity, lack of Buildings digitized
metadata) of the data and metadata have naturally arisen. The lack by OSM volunteers
one dot = 2500 buildings
of obligatory commitment of free volunteers or the crowd is a
further problem since interest in projects can easily dampen or die,
leaving many data collection and mapping efforts incomplete. To Tacloban
objectively assess quality of such volunteered and crowdsourced Track of
Typhoon Haiyan
data, a couple of studies explicitly compared OSM data to authorita-
tive Ordnance Survey data in England (Haklay, 2010) and German
Amtliche Topographisch-Kartographische Informationssystem Population density
(ATKIS) data for Munich (Fan et al., 2014). The errors were not 0 32,000
large (less than 20 m offset for roads in England and 4 m offset and persons/km2
simplification of building footprints in Munich). One way to judge
the quality of VGI is the number of peers who have reviewed and
edited the content, a principle known as Linuss law (Elwood etal.,
0 200
2011). Large numbers of peers can yield even higher quality data km
than obtained by the traditional authoritative sources. This is often
evident during emergencies, when fast decisions must be made real FIg u r e 20.7 Map of buildings digitized by OpenStreetMap volun-
time, so that the risks associated with volunteered information are teers in the Philippines until end of July, 2014. Typhoon Haiyan resulted
often outweighed by the benefits, as was found in the case of the in a dramatic increase in volunteer activity in the affected areas, whereas
wildfires near Santa Barbara between 2007 and 2009 (Goodchild buildings in other areas, except around Manila, remain sparsely mapped.
and Glennon, 2010). Similarly, McDougall (2012) examined three
disasters in Australia, New Zealand, and Japan, respectively, and reference imagery, lack of expertise and visual/spatial interpreta-
concluded that crowdsourced mapping provided a unique bottom tion skills, imagery georectification errors, and overly simplistic
up perspective unattainable through the conventional emergency instructions for volunteers. Figure 20.8 shows examples of building
service command-and-control structure or implementation. digitization errors by volunteers for Typhoon Haiyan HOT.
Moreover, postdisaster mapping not just is useful for guiding It would be nave to expect untrained volunteers with no back-
immediate development efforts but also creates much needed ground in geospatial technology and science to realize or rectify
information about places with important long-term use value. For such errors. Champions of neogeography must understand that
example, as shown in Figure 20.7, Typhoon Haiyan resulted in a a lot of expertise is needed to engage in deep geographic think-
dramatic increase in volunteer mapping activity in affected areas, ing and geospatial analysis, quite contrary to the scope of neo-
which now have a much higher density of mapped buildings than geography (Goodchild, 2009). Moreover, the need to process data
other places with comparable population densities. Otherwise, fast and reliably along with lack of incentives for deep theoretical
only Manila, the political and cultural capital, has a high density of analysis means that many technologies are being deployed in non-
mapped buildings. professional settings in ways that lead to lowering or relaxation of
Yet, volunteered and crowdsourced datasets do have a lot of quality and consistency. This is occurring not only at the level of
quality issues, which, if ignored, lead to faulty decisions. For individual projects but also within the context of institutional and
example, ground truthing of building damage assessment done by professional norms. A prime example of this is the de facto estab-
volunteers looking at satellite imagery after Typhoon Haiyan in the lishment of Web Mercator as the standard coordinate system for
Philippines found overall classification accuracy to be merely 36%, web mapping in all geobrowsers. Established for reasons related
primarily due to overestimation of building damage (REACH etal., to server efficiency, Web Mercator is inappropriate for many (per-
2014). For the 2011 Christchurch earthquake, a similar analysis haps most) cartographic and quantitative applications (Battersby
found large errors of omission (56% in downtown, 86% in residen- etal., 2014). Another example is the normalized practice of rely-
tial areas) but virtually no errors of commission (3% in downtown, ing on users to uncover database errors, enabling a faster product
0% in residential areas) (Foulser-Piggot et al., 2013). Reasons for release cycle. The botched launch of Apples iOS6 Maps in 2012,
such errors are related to low spatial resolution, lack of predisaster for which Apple CEO Tim Cook had to offer the first open apology
2016 by Taylor & Francis Group, LLC
438 Remotely Sensed Data Characterization, Classification, and Accuracies
FIgu r e 20.8 Screenshot from OSM showing substantial offsets for some buildings digitized by volunteers for Typhoon Haiyan HOT. While
the buildings on the right align closely with the satellite image, the three buildings on the left are considerably offset, because volunteers likely had
access to different images at different times. (Screenshot from http://www.openstreetmap.org, created July 12, 2014.)
in the history of the company,* demonstrates that this practice can attacked for being an extension of the state for monitoring and sur-
also disenfranchise users who expect high-quality spatial data. veillance (Pickles, 1991; Smith, 1992). The concept has now evolved
In many cases, the appeal of spatial data is leading to misap- to include a much wider spectrum of Geo-ICTs, and the surveil-
propriation of technologies that are not designed for geospatial lance is not just being done by governments but also powerful cor-
accuracy. For example, location inference methods used by social porates, and that too with our implicit acquiescence. The exaflood
networking services, for which location is just a secondary param- of big data (Sui etal., 2013) that results from massive data collection
eter, often rely on coarse methods of positioning, and using such efforts is simultaneously a source of exciting new possibilities and a
inaccurate locations for applications where precise positions are threat to our privacy, and thus, our freedoms. Data mining makes
needed creates unanticipated problems. Worse, many LBSs are it possible to derive deep information on individuals behaviors
not designed properly for efficient decision making, especially in virtual and physical space (Sui and Goodchild, 2011). High-
for real-time decisions while on the move (Raubal, 2011). One has resolution imagery today allows us to explore our world in fasci-
only to think of the many limitations of automated navigation sys- nating detail and simultaneously empowers us to counter secretive
tems that people routinely experience and have to address based government strategies (Perkins and Dodge, 2009).
on their own contextual geographic knowledge. Such problems Yet, much of this big data are owned by either corporations or
may be ignored in many contexts, but in emergency response situ- government intelligence agencies (Sui and Goodchild, 2011)with
ations, inaccurate locations can put lives in danger. Resolutions of a nexus between the two also possible. Online search is currently
such errors are as much about technology improvement as explor- only possible through commercial search engine platforms, with no
ing how human cognitive systems function and differ across use practical open-source or nonprofit alternative. Companies benefit
contexts and for individuals based on gender, culture, native lan- from free user-generated content and by monitoring and mining
guage, age, and many other specific factors. user behavior through advertising and sale of such data to third
parties. Should we not be quite concerned that the online infor-
20.4.2 Privacy and confidentiality mation we receive is filtered through commercial search engines
and that the digital shadows (Klinkenberg, 2007) we create are not
The constant demand for geospatial data and increasing use of
in our control? As the benefits derived from joining social media
linked socialspatial data raises concerns about the ability to pro-
networks and other Geo-ICTs have increased, so has the willing-
tect confidentiality and privacy (NRC, 2007). Geosurveillance as a
ness of individuals to share geospatial data about themselves and
concept has been around from the early 1990s when GIS was first
their family and friends, without realizing the implications of their
choices. When such information is combined across space, time,
* www.apple.com/letter-from-tim-cook-on-maps. and the multiple thematic dimensions of our lives, the analysis
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 439
reveals hidden and surprising patterns about our individual and reexamined, since it involves manipulation of content based on
collective behaviors. Why is that the data miners are not obligated user location. For example, while Googles automated car initia-
to share such information they find with the very people that unsus- tive has the potential to offer great benefits to society by preventing
pectingly helped generate such information? accidents, reducing fuel usage and carbon emissions, minimizing
It is also important to distinguish the concept of data protection traffic congestion, and saving commuting times, the company is
from that of individual privacy, the latter being much more nebu- not motivated by altruistic reasons, but inspired by the companys
lous and culturally dependent. In this new age of the geoweb and long-term goals to collect and eventually monetize information
Geo-ICTs, the very notion of privacy needs is being reconceptu- about its customers. Clearly, Google needs and wants to be at the
alized to reflect the experienced impacts of the use of such tech- forefront regarding the design and sale of automated driving tech-
nologies (Elwood and Leszczynski, 2011). Asignificant problem in nology and information systems, the collection and sale of data
this era of instant multimedia communication and proliferation of on driving habits and navigation patterns, and the ability to pro-
social networks is that individuals do not realize they may be reveal- vide LBSs to drivers. How Google will monetize such information
ing unwelcome, unauthorized, or even incriminating information remains to be seen, but one thing is certainit will create many
about others inadvertently. Moreover, impacts might be several controversies. Googles Street View technology raised tremendous
years away in the future, when information from multiple sources is privacy concerns in several contexts, and the company has been
combined to reveal new patterns. Crowdsourcing is another devel- perennially engaged in legal battles and paid fines for alleged inva-
opment that presents similar problems of lack of awareness of par- sion of privacy in several countries.* Its latest technology, Google
ticipants. Despite the bottom-up and open nature of crowdsourced Glass, is already generating similar concerns and polarizing early
mapping efforts, it is not clear that participants realize that they adopters from the skeptics.
may be collecting unauthorized information or helping to build No matter what the technology, the fundamental ethical ques-
databases that may be used by organizations for profit motives, tion for all LBSs is quite simple: how can ethical guidelines be con-
rather than public good. Similarly, in the context of PGIS, com- structed to allow for legitimate manipulation of content while
munity trust needs to be respected and nurtured and it should be excluding more subversive purposes? There is rarely a context in
transparent what kind of data are being collected and shared, about which an LBS technology cannot be subverted to collect informa-
whom, and if required permissions have been acquired legitimately. tion against the will of the user, and the more accurate it gets, the
This has not always been the case, due tobreaches of confidentiality more harm that can be inflicted. Several years ago, Dobson and
or insensitive protocols for data collection and sharing. Fisher (2003) coined the term geoslavery to alert us to this kind
In response to the deluge of technologies and information, of misuse of geo-information technologies to spy on peopleby
an interesting new approach to privacy is that of sousveillance government, corporations, and individuals aided by powerful
(inverse surveillance), first proposed in Mann etal. (2003) as the technologies. Several questions arise, therefore, in the quest of
philosophy and practice of wearing computing devices visibly to democratization and freedom from geoslavery. How do we prevent
surveil the surveillers reflectivelyas a form of protest against ourselves from becoming geoslaves and fight the commoditization
state and company surveillance measurements. Pervasive sous- of our lives? What kind of in-built controls and preventative mea-
veillance would mean we are all being constantly monitored by sures can be provided to users? Can we through our own choices
everybody around us, which in itself raises many moral, legal, send strong messages about what we can accept and what must
and ethical problems. Using phones today or Google Glass in the change? What legislations need to be passed to rein in companies?
future, and many participatory sensing activities, can be inter- Originally, the concept of geosurveillance was associated
preted as forms of sousveillance. In contrast, proponents of the only with governments, whose data collection powers far out-
postprivacy movement believe that since technology makes it reach that of corporations. While government may offer recourse
impossible to control ownership of information, society should against corporations, there may be none against the government
practically abandon the notion of privacy in favor of a completely itself, especially in nondemocratic countries. Even in the United
transparent society, where everybody knows whats going on most States, one of the better-functioning democracies, there is cur-
of the time (Brin, 1999; Jarvis, 2011). Social networking giant rently great concern about military and intelligence uses of the
Facebooks founder and CEO Mark Zuckerberg has even made technology, against citizens and foreign nationals. One only has
the claim that people have accepted that privacy is no longer a to follow the Edward Snowden leaks of the National Security
social norm (The Economist, 2013). Perhaps, it is under such Agencys documents to realize how much power and techni-
assumptions that Facebook (and researchers employed at Cornell cal ability is at the disposal of the intelligence community.(The
University) had intentionally manipulated Facebook news feeds Guardian, 2013). Research of the U.S. intelligence economy is
to investigate their influence on users emotional states (Meyer, enough to justify all concerns. The U.S. government is the larg-
2014). The study drew considerable indignation for Facebook, but est consumer of geospatial data and Geo-ICTs in the world, with
the company is hardly unique in its designs to profit from manip- most of that budget allocated for intelligence gatheringand,
ulating peoples behaviorsany marketing or lobbying campaign not surprisingly, satellite imagery is the largest component of
with biased information can be accused of such infringements. intelligence budget (Crampton et al., 2014). There also exists a
If the manipulation of users for experimental purposes strikes
many as unethical, the very premise of LBSs may need to be * www.en.wikipedia.org/wiki/Google_Street_View_privacy_concerns.
2016 by Taylor & Francis Group, LLC
440 Remotely Sensed Data Characterization, Classification, and Accuracies
massive contracting nexus of more than 50,000 contractor and places (often in the context of a disaster) from high-reso-
companies, universities, and nonprofits, who have received a lution satellite imagery. Volunteers are often recruited for such
staggering 3.75 trillion dollars between 2000 and 2012 from campaigns through feel-good messages, without providing
the Department of Defense and intelligence agencies (Crampton concrete information regarding how the data will be used, who
etal., 2014). These figures make one thing cleargeo-informa- owns the data, or whether it will be made available to the volun-
tion technologies and data are being developed and used in the teers themselves. At the very least, more transparency is needed
name of national security in diverse ways by diverse entities, to clarify how their data are being used. Google similarly owns
while the public is largely unaware of what the data may or may all maps and data created by users. Similar to the concerns asso-
not be used for and who may have access to it. ciated with the dangers to individuals arising from data mining
of their personal information from variety of sources, VGI and
20.4.3 empowerment and equity crowdsourced information can be easily used without permis-
sion for enhancing profit margins or, worse, for unethical or ille-
Aided by digital mapping technologies and social networking prac-
gal purposesby corporations, governments, and individuals.
tices of Web 2.0, the neogeography movement has been critical
Linus law also implies, unfortunately, that volunteered and
in advancing the democratization of geographic information and
crowdsourced information about accessible, popular, or popu-
broadened the ontological scope of the formal systems for repre-
lated places and readily observable phenomena will be more
senting and analyzing geographic information. Neogeographical
accurate than information about remote places and not so read-
knowledge can be interpreted as arising from contextual personal
ily observable phenomena. A recent study (Graham etal., 2014)
and communal interpretive interactions of space and place and thus
analyzed the geography of Wikipedia articles, only to discover
intrinsically about the culture of everyday life (Warf and Sui, 2010).
that there is disproportionate amount of information about North
Despite all its successes, neogeography is no panacea to the digital
America, much of Europe, and heavily populated parts of Asia,
divide between the technological elite and the much larger group
and not much about the rest of the world. A large part of the geo-
of untrained and uncritical laborers who are not empowered by the
graphic variation was statistically explained by three variables:
use of the technology (Dodge and Kitchin, 2013; Haklay, 2013). The
population, availability of broadband Internet, and number of
free labor from neogeography is opportunistic capital for many
edits originating from a country. Despite Wikipedia being one
entrepreneurial companies who benefit directly from such user-
of the ten most popular websites in the world, its content reflects
generated content through advertising and sale of user-generated
existing and possibly also creates new geographies of the digital
information (including users online behavior), with the users
divide. Other sources of biases in peoples perception of places and
receiving no share of such profits. However, without these laborers,
how they act on such perceptions include lack of cultural diversity,
we still would be limited to the sparse and barren maps of Kibera,
lack of geographic education, limited exposure to places, and dis-
when the reality, as shown in Figure 20.1, is quite different. This
proportionate media attention. For example, Typhoon Haiyan
reveals the true meaning of democratization, which is the empow-
related mapping efforts exhibited a clear media effect since OSM
erment of people who would otherwise be underrepresented.
volunteers disproportionately mapped damage in Tacloban City,
Through individual and coordinated organic acts of mapping,
which was the focus of media coverage (REACH etal., 2014), but
neogeographers have put on the map the stories of people who live
ignored other affected places. Since OSM maps were used to guide
far on the unprivileged side of the digital dividethose who can-
relief efforts, this may have had consequences as to how much
not even read or have access to the basic amenities, those displaced
aid was received by different locales. This geographic unevenness
due to war, or those persecuted by their governments. Motivated
in OSM efforts can be observed more generally at the national
by curiosity, purpose, and humanitarian instinct, neogeographers
scale as well. As shown in Figure 20.7, buildings in Manila, being
have challenged academic researchers to engage in a place-centric
the well-known capital of Philippines, and in areas (especially
GIS grounded in qualitative human discourse rather than stay lim-
Tacloban) affected by Typhoon Haiyan have been digitized at a
ited to the conventional spatial perspective (Elwood etal., 2013).
disproportionately higher rate compared to other areas with com-
Despite neogeographers having been a prime force behind
parable population density but lacking special appeal.
many recent developments, their efforts may fail to serve the
Yet another form of mapping bias creeps in due to the profit
original purpose or contribute toward purposes they did not
motive and involvement of corporates, which engage in public
sanction explicitly. Corporations such as Google and Microsoft
service often for self-serving public relations purposes. Are recent
have retained copyright on volunteer-collected spatial data using
campaigns sponsored by DigitalGlobe through their Tomnod
their platforms. This has impeded data sharing during humani-
platform to locate the missing airplane MH370, map damage after
tarian operations such as the 2011 Haiti earthquake response
the 2013 Illinois tornado outbreak, and identify invasive species
(Zook etal., 2010). Disasters and other nonurgent crowdsourc-
in Hawaii examples of corporate good deeds or media attention
ing efforts may even be seen as opportunities for data collection
grabbing? Though both perspectives contain some truth, the big-
for private gain. DigitalGlobe manages the Tomnod* online
ger picture is that as corporations begin to see public relations
platform to crowdsource identification of interesting objects
benefits of sponsoring VGI efforts for good causes, they might
coopt (perhaps unwittingly) the power to decide which causes
* http://www.digitalearth-isde.org. are worth sponsoring. There is little doubt that the charitable
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 441
activities of large corporations such as Google and DigitalGlobe, to commercial search engines, and our online behavior is being
as well as nonprofit organizations such as OSM, have contributed harvested for profit purposes. We do benefit from such services
immensely to recent disaster relief efforts and other causes. While through more targeted advertising and more relevant information
these initiatives should continue to be supported, guidelines retrieval, but the downsides are often hidden from us. Even the
should also be developed from within the geospatial community relationships of large corporations with colleges and universities
to ensure that volunteers derive tangible benefits from participat- need to be monitored to limit unethical influences, even if sub-
ing in VGI projects and that societal values such as equity, political tly, on teaching, research and administrative decisions of grants,
freedom, information sharing, etc., be supported in such efforts. scholarships, endowments, hiring, and free/discounted software
made available by companies. Governmental and private geosur-
20.5 conclusion veillance may be unavoidable in practice, simply because of geo-
political complexities and ubiquity of monitoring devices, but that
The goal of this chapter has been to diligently cover the brief his- is no excuse for different branches of the government not exercis-
tory of democratization of geographic information and through ing valid checks on each other, and other violators. Furthermore,
some selective examples highlight the immense impact it has on if knowledge-privileged professionals and intellectuals become
our ability to not just function better but completely change the accessible to the populace outside their exalted circles, many insid-
course of our lives. There is no better example of this than Saroo ious impacts of Geo-ICTs can be avoided. Ultimately, it falls on
Brierley, whose discovery of his family using Google Earth serves all individuals and communities to be vigilant about developments
as a shining example of Digital Earth technologies as life-altering that violate or can help protect our basic freedoms.
solutions and, less radically, as technologies for creating vicarious At the same time, critical analyses that merely pander to inse-
place experiences and expanding our geographic understanding. curities and fan the politics of hope and fear need to be coun-
Saroo found his family because of these developments, but think tered intelligently and collectively. As Monmonier (2002) and
of a future where there exist advanced geo-information retrieval Klinkenberg (2007) have reasoned, the reasonable approach is to
programs through which Saroo could express and search auto- stay critical, but only to negate the evils of technologiesand to
matically places resembling mental map of his neighborhood. ensure that all strata of society are involved in the path to socially
Saroo would not have to spend months browsing imagery to find positive technologies. There has undoubtedly been considerable
his family, in such a future. Indeed, the democratization of geo- progress in geospatial technology, its use, and its discourse in just
information not just is about peoples involvement, and accessibil- a few decades, which became the inspiration and provided the
ity to information and tools, but must also depends on innovations content for this chapter on democratization of geographic infor-
in computational reasoning, humancomputer interfaces (Google mation. What remains to be seen now is whether the democratiza-
Earths success was primarily due to its ease of use factor), and tion process flourishes in practice and transforms our lives or if it
growth of public information infrastructures. While universities remains a mere cause clbre in the academic community.
and government settings are better suited for more scientific and
socially responsible research, commercial entities have played an References
important role through technological innovations and expanding
the reach of Geo-ICTs to the common citizen. Battersby, S. E., Finn, M. P., Usery, E. L., and Yamamoto, K. H.
To come back to Heideggers quote from earlier in the chapter, (2014). Implications of Web Mercator and Its Use in Online
technological problems are not technicaland technical strate- Mapping. Cartographica: The International Journal for
gies cannot alone help us resolve the conflicting demands for data Geographic Information and Geovisualization, 49(2), 85101.
access, data quality, and confidentiality (NRC, 2007). Clearly, Blaschke, T. and Merschdorf, H. (2014). Geographic information
technology does not operate in a vacuum and cannot be viewed science as a multidisciplinary and multiparadigmatic field.
as devoid of social, political, and economic contexts. However, if Cartography and Geographic Information Science, 41(3),
there is to be true democratization of geographic information and 196213.
many of its supporting technologies, citizens must step up to nur- Brierley, S. (2014). A Long Way Home: A Memoir. New York:
ture this fledgling democracy. Neogeography, VGI, crowdsourc- Putnam Adult.
ing, citizen science, Web 2.0, and many other related neologisms Brin, D. (1999). The Transparent Society: Will Technology Force
capture many such citizen-led empowering forms of technological us to Choose between Privacy and Freedom? Reading, MA:
progress. Unfortunately, these new practices have also created new Basic Books.
problems of data quality, devaluation of established knowledge or Brown, G. (2012). Public participation GIS (PPGIS) for regional
capital production systems, exploitation of volunteers, and biasing and environmental planning: Reflections on a decade of
discourses in favor of popular and easily accessible information. empirical research. URISA Journal, 24(2), 718.
As much as we need technologies to promote our development Buchanan, L., Fairfield, H., Parlapiano, A., Peanha, S., Wallace,
as individuals and societies, we must also never lose sight of how T.,Watkins, D., and Yourish, K. (2013). Mapping the destruc-
they are being used by those who control thembe it individu- tion of Typhoon Haiyan. Retrieved from http://www.nytimes.
als, communities, corporations, institutions, or governments. As com/interactive/2013/11/11/world/asia/typhoon-haiyan-
mentioned earlier, information retrieval is practically hostage map.html. Accessed July 14, 2014.
2016 by Taylor & Francis Group, LLC
442 Remotely Sensed Data Characterization, Classification, and Accuracies
Budhathoki, N., Bruce, B., and Nedovi-Budi, Z. (2008). Elwood, S. (2011). Participatory approaches in GIS and society
Reconceptualizing the role of the user of spatial data infra- research: Foundations, practices, and future directions.
structure. GeoJournal, 72(3), 149160. In T. Nyerges, H. Couclelis, and R. B. McMaster (eds.),
Burke, J. A., Estrin, D., Hansen, M., Parker, A., Ramanathan, N., The SAGE Handbook of GIS and Society (pp. 381399).
Reddy, S., and Srivastava, M. B. (2006). Participatory sens- Thousand Oaks, CA: SAGE Publications Ltd.
ing. Boulder, CO: Center for Embedded Network Sensing. Elwood, S., Goodchild, M. F., and Sui, D. (2013). Prospects for
Retrieved from http://escholarship.org/uc/item/19h777qd. VGI research and the emerging fourth paradigm. In D. Sui,
Accessed July 14, 2014. S.Elwood, and M. Goodchild (eds.), Crowdsourcing Geographic
Chrisman, N. (1987). Design of geographic information sys- Knowledge (pp. 361375). Dordrecht, the Netherlands:
tems based on social and cultural goals. Photogrammetric Springer.
Engineering and Remote Sensing, 53(10), 13671370. Elwood, S., Goodchild, M. F., and Sui, D. Z. (2011). Researching
Cloud , J. (2002). American Cartographic Transformations dur- volunteered geographic information: Spatial data, geo-
ing the Cold War. Cartography and Geographic Information graphic research, and new social practice. Annals of the
Science, 29(3), 261282. Association of American Geographers, 102(3), 571590.
Craglia, M., Goodchild, M. F., Annoni, A., Camara, G., Gould, Elwood, S. and Leszczynski, A. (2011). Privacy, reconsidered: New
M., Kuhn, W. et al. (2008). Next-generation digital representations, data practices, and the geoweb. Geoforum,
earth: A position paper from the Vespucci initiative for 42(1), 615.
the advancement of geographic information science. Fan, H., Zipf, A., Fu, Q., and Neis, P. (2014). Quality assess-
International Journal of Spatial Data Infrastructures ment for building footprints data on OpenStreetMap.
Research, 3, 146167. International Journal of Geographical Information Science,
Craig, W., Harris, T., and Weiner, D. (2002): Introduction. In 28(4), 700719.
W. Craig, T. Harris, and D. Weiner (eds.), Community Fisher, A. (2013, September 18). Inside Googles quest to popular-
Participation and Geographic Information Systems, London: ize self-driving cars. Popular Science. Retrieved from http://
Taylor & Francis, 116. www.popsci.com/cars/article/2013-09/google-self-driving-
Crampton, J. W. (2010). Mapping: A Critical Introduction to car. Accessed July 14, 2014.
Cartography and GIS: A Critical Introduction to GIS and Foulser-Piggot, R., Spence, R., and Brown, D. (2013). The use
Cartography. Malden, MA: Wiley-Blackwell. of remote sensing for building damage assessment fol-
Crampton, J. W., Roberts, S. M., and Poorthuis, A. (2014). lowing 22nd February 2011 Christchurch Earthquake:
The new political economy of geographical intelligence. The GEOCAN study and its validation. Cambridge, U.K.:
Annals of the Association of American Geographers, Cambridge Architectural Research, Ltd. Retrieved from
104(1), 196214. http://www.willisresearchnetwork.com/assets/templates/
Dasgupta, A. (May, 2013). Economic Value of Geospatial Data: wrn/files/GEOCAN%20Christchurch%20Report.pdf.
the great enabler. Geospatial World. Retrieved from Accessed July 14, 2014.
http://geospatialworld.net/paper/business/ArticleView. Gao, J. (2002). Integration of GPS with remote sensing and GIS:
aspx?aid=30534. Accessed July 14, 2014. Reality and prospect. Photogrammetric Engineering &
Dobson, J. (2011). Through the macroscope: Geographys view Remote Sensing, 68(5), 447453.
of the world. ArcNews, Winter 2011. Retrieved from http:// Goldman, J., Shilton, K., Burke, J., Estrin, D., Hansen, M.,
www.esri.com/news/arcnews/winter1112articles/through- Ramanathan, N. et al. (2009). Participatory Sensing: A
the-macroscope-geographys-view-of-the-world.html. citizen-powered approach to illuminating the patterns
Accessed July 14, 2014. that shape our world. Washington, DC: Woodrow Wilson
Dobson, J. and Fisher, P. (2003). Geoslavery. IEEE Technology and International Center for Scholars. Retrieved from http://
Society Magazine, Spring 2003, 4752. wilsoncenter.org/topics/docs/participatory_sensing.pdf.
Dodge, M. and Kitchin, R. (2013). Crowdsourced cartography: Accessed July 14, 2014.
Mapping experience and knowledge. Environment and Goodchild, M. F. (1992). Geographical information science.
Planning A, 45(1), 1936. International Journal of Geographic Information Systems,
Duda, K. A. and Jones, B. K. (2011). USGS remote sens- 6(1), 3145.
ing coordination for the 2010 Haiti Earthquake. Goodchild, M. F. (2007). Citizens as sensors: The world of volun-
Photogrammetric Engineering and Remote Sensing, teered geography. GeoJournal, 69(4), 211221.
77(9), 899907. Goodchild, M. F. (2009). NeoGeography and the nature of geographic
Dunn, C. (2007). Participatory GIS - a peoples GIS? Progress in expertise. Journal of Location Based Services, 3(2), 8296.
Human Geography, 31(5), 616637. Goodchild, M. F. (2012). GIScience in the 21st century. In W. Shi,
Emrich et al. (2011). GIS and Emergency Management. In M. F. Goodchild, B. Lees, and Y. Leung (eds.), Advances
The SAGE Handbook of GIS and Society (pp. 321343). in Geo-Spatial Information Science (pp. 310). Leiden, the
Thousand Oaks, CA: SAGE Publications Ltd. Netherlands: CRC Press.
2016 by Taylor & Francis Group, LLC
Toward Democratization of Geographic Information 443
Goodchild, M. F., Fu, P., and Rich, P. (2007). Sharing geographic Maisonneuve, N., Stevens, M., and Ochab, B. (2010). Participatory
information: An assessment of the geospatial one-stop. Annals noise pollution monitoring using mobile phones.
of the Association of American Geographers, 97(2), 250266. Information Polity, 15(1), 5171.
Goodchild, M. F. and Glennon, J. A. (2010). Crowdsourcing geo- Mann, S., Nolan, J., and Wellman, B. (2003). Sousveillance:
graphic information for disaster response: A research fron- Inventing and using wearable computing devices for data
tier. International Journal of Digital Earth, 3(3), 231241. collection in surveillance environments. Surveillance &
Gore, A. (1998). The digital earth: Understanding our planet in the Society, 1(3), 331355.
21st century. Los Angeles, CA: California Science Center. Mannings, R. (2008). Ubiquitous Positioning. New York: Artech
Retrieved from http://www.isde5.org/al_gore_speech.htm. House.
Accessed July 14, 2014. Masser, I. (2011). Emerging frameworks in the information
Gouveia, C. and Fonseca, A. (2008). New approaches to environ- age: The spatial data infrastructure (SDI) phenomenon.
mental monitoring: The use of ICT to explore volunteered In T. Nyerges, H. Couclelis, and R. B. McMaster (eds.),
geographic information. GeoJournal, 72(34), 185197. The SAGE Handbook of GIS and Society (pp. 271286).
Graham, M., Benie, H., Straumann, R. K., and Medhat, A. Thousand Oaks, CA: SAGE Publications Ltd.
(2014). Uneven geographies of user-generated information: McDougall (2012). An Assessment of the Contribution of
Patterns of increasing informational poverty. Annals of the Volunteered Geographic Information During Recent
Association of American Geographers, 104(4), 746764. Natural Disasters. In Spatially enabling government, indus-
Grossner, K. E., Goodchild, M. F., and Clarke, K. C. (2008). Defining try and citizens: research and development perspectives
a digital earth system. Transactions in GIS, 12(1), 145160. (pp. 201214), GSDI Association Press, Needham, MA.
Hagen, E. (2011). Mapping change: Community information Merchant, J. W. and Narumalani, S. (2009). Integrating remote sens-
empowerment in Kibera (Innovations Case Narrative: Map ing and geographic information systems. In T. A. Warner, M.
Kibera). Innovations: Technology, Governance, Globalization, D. Nellis, and G. M. Foody (eds.), The SAGE Handbook of
6(1), 6994. Remote Sensing. London, U.K.: SAGE Publications Ltd.
Haklay, M. (2010). How good is volunteered geographical infor- Mesev, V. (2007). Integration of GIS and Remote Sensing. Hoboken,
mation? A comparative study of OpenStreetMap and NJ: Wiley.
Ordnance Survey datasets. Environment and Planning B: Meyer, R. (2014, June 28). Everything we know about Facebooks
Planning and Design, 37(4), 682703. secret mood manipulation experiment. The Atlantic.
Haklay, M. (2013). Neogeography and the delusion of democrati- Retrieved from http://www.theatlantic.com/technology/
sation. Environment and Planning A, 45(1), 5569. archive/2014/06/everything-we-know-about-facebooks-
Harley, J. B. (2002). The New Nature of Maps: Essays in the History of secret-mood-manipulation-experiment/373648/. Accessed
Cartography. Baltimore, MD: Johns Hopkins University Press. July 14, 2014.
Harris, T. M. and Weiner, D. (1996). GIS and Society: The Social Monmonier, M. (1996). How to Lie with Maps (2nd edn.). Chicago,
Implications of How People, Space, and Environment are IL: University of Chicago Press.
Represented in GIS. Scientific Report for the Initiative 19 Monmonier, M. (2002). Spying with Maps: Surveillance
Specialist Meeting. (Scientific Report for the Initiative 19 Technologies and the Future of Privacy. Chicago, IL:
Specialist Meeting No. #96-7). Koinonia Retreat Center, University of Chicago Press.
South Haven, MN: National Center for Geographic Mun, M., Reddy, S., Shilton, K., Yau, N., Burke, J., Estrin, D. etal.
Information and Analysis (NCGIA). (2009). PEIR, the personal environmental impact report,
Heidegger, M. (1982). The Question Concerning Technology, and Other as a platform for participatory sensing systems research. In
Essays. (W. Lovitt, Trans.). New York: Harper Torchbooks. Proceedings of the Seventh International Conference onMobile
Jankowski, P. (2011). Designing public participation geographic Systems, Applications, and Services (pp. 5568). New York:
information systems. In T. Nyerges, H. Couclelis, and R. B. ACM.
McMaster (eds.), The SAGE Handbook of GIS and Society. National Research Council (NRC) (2007). Putting People on the
Thousand Oaks, CA: SAGE Publications Ltd, pp. 417421. Map: Protecting Confidentiality with Linked Social-Spatial
Jarvis, J. (2011). Public Parts: How Sharing in the Digital Age Improves Data. Washington, DC: The National Academies Press.
the Way We Work and Live. New York: Simon & Schuster. Nouwt, S. (2008). Reasonable expectations of geo-privacy?
Klinkenberg, B. (2007). Geospatial technologies and the geogra- SCRIPTed, 5(2), 375403.
phies of hope and fear. Annals of the Association of American OSullivan, D. (2006). Geographical information science: Critical
Geographers, 97(2), 350360. GIS. Progress in Human Geography, 30(6), 783791.
Lemmens, M. (2011). Geo-information: Technologies, Applications Palmer, M. H. and Kraushaar, S. (2013). Volunteered geographic
and the Environment. New York: Springer. information, actor-network theory, and severe-storm
Longley, P. A., Goodchild, M., Maguire, D. J., and Rhind, D. W. reports. In D. Sui, S. Elwood, and M. Goodchild (eds.),
(2010). Geographic Information Systems and Science (3rd edn.). Crowdsourcing Geographic Knowledge (pp. 287306).
Hoboken, NJ: Wiley. Dordrecht, the Netherlands: Springer.
2016 by Taylor & Francis Group, LLC
444 Remotely Sensed Data Characterization, Classification, and Accuracies
Perkins, C. and Dodge, M. (2009). Satellite imagery and the spec- Sui, D. and Goodchild, M. (2011). The convergence of GIS and
tacle of secret spaces. Geoforum, 40(4), 546560. social media: Challenges for GIScience. International Journal
Pickles, J. (1991). Geography, GIS, and the surveillant society. Papers of Geographical Information Science, 25(11), 17371748.
and Proceedings of Applied Geography Conferences, 4, 8091. Sui, D., Goodchild, M., and Elwood, S. (2013). Volunteered
Pickles, J. (1994). Ground Truth: The Social Implications of Geographic Geographic information, the exaflood, and the growing
Information Systems. New York: The Guilford Press. digital divide. In D. Sui, S. Elwood, and M. Goodchild
Pool, B. (2010). Crisis Camp Haiti: Techno-types volunteer their (eds.), Crowdsourcing Geographic Knowledge (pp. 112).
computer skills to aid quake victims. Los Angeles Times. Dordrecht, the Netherlands: Springer.
Retrieved from http://articles.latimes.com/2010/jan/16/world/ Taylor, P. (1990). Editorial comment: GKS. Political Geography
la-fg-haiti-crisiscamp17-2010jan17. Accessed July14, 2014. Quarterly, 9, 211212.
Raper, J., Gartner, G., Karimi, H., and Rizos, C. (2007). The Economist. (2013, November 16). The peoples panopticon. The
Applications of location-based services: A selected review. Economist. Retrieved from http://www.economist.com/news/
Journal of Location Based Services, 1(2), 89111. briefing/21589863-it-getting-ever-easier-record-anything-or-
Raubal, M. (2011). Cogito Ergo Mobilis sum. In T. Nyerges, H. everything-you-see-opens. Accessed July 14, 2014.
Couclelis, and R. B. McMaster (eds.), The SAGE Handbook The Guardian. (2013, June 8). The NSA files. Retrieved
of GIS and Society (pp. 159173). Thousand Oaks, CA: from http://www.theguardian.com/world/the-nsa-files.
SAGE Publications Ltd. Accessed July 14, 2014.
REACH, American Red Cross, and USAID. (2014). Groundtruthing Thrun, S. (2010). What were driving at. Google Official Blog.
OpenStreetMap building damage assessment, Haiyan Commercial. Retrieved from http://googleblog.blogspot.
Typhoon, The Philippines, final assessment report. Retrieved com/2010/10/what-were-driving-at.html. Accessed July14,
from http://www.reach-initiative.org. Accessed July 14, 2014. 2014.
Resch, B., Mittlboeck, M., Lipson, S., Welsh, M., Bers, J., Britter, Tulloch, D. L. (2008). Is VGI participation? From vernal pools to
R., and Ratti, C. (2009). Urban sensing revisitedCommon video games. Geojournal 72(3), 161171.
scents: Towards standardised geo-sensor networks for pub- Turner, A. (2006). Introduction to Neogeography. Sebastopol, CA:
lic health monitoring in the city. In Proceedings of the 11th OReilly Media.
International Conference on Computers in Urban Planning Van Aardt, J. A. N., McKeown, D., Faulring, J., Raqueo, N.,
and Urban ManagementCUPUM2009, Hong Kong, Casterline, M., Renschler, C. et al. (2011). Geospatial
China, June 1618, 2009. disaster response during the Haiti earthquake: A case
Sagl, G., Delmelle, E., and Delmelle, E. (2014). Mapping collective study spanning airborne deployment, data collection,
human activity in an urban environment based on mobile transfer, processing, and dissemination. Photogrammetric
phone data. Cartography and Geographic Information Engineering & Remote Sensing, 77(9), 943952.
Science, 41(3), 272285. Warf, B. and Sui, D. (2010). From GIS to neogeography:
Schuurman, N. (1999). Critical GIS: Theorizing an emerging Ontological implications and theories of truth. Annals of
discipline. Cartographica: The International Journal for GIS, 16(4), 197209.
Geographic Information and Geovisualization, 36(4), 1108. Weiner, G. and Smith, B. W. (2014). Automated driving:
Seang, T. P., Mund, J.-P., and Symann, R. (2008). Low cost Legislative and regulatory action. The Center for Internet
amateur aerial pictures with balloon and digital camera. and Society. Retrieved from http://cyberlaw.stanford.edu.
MethodFinder. Retrieved from http://methodfinder.net. Accessed July14, 2014.
Accessed July 14, 2014. Weiser, M. (1991). The computer for the 21st century. The
Sheppard, E. (1995). GIS and society: Toward a research agenda. Scientific American, 256(3), 94104.
Cartography and Geographic Information Systems, 22(1), 516. Wilson, M. W. and Poore, B. S. (2009). Theory, practice, and
Sieber, R. (2006). Public Participation Geographic Information history in critical GIS: Reports on an AAG panel session.
Systems: A Literature Review and Framework. Annals of the Cartographica: The International Journal for Geographic
Association of American Geographers, 96(3), 491507. Information and Geovisualization, 44(1), 516.
Singel, R. (2005). A disaster map Wiki is born. WIRED Magazine. Wilson, M.W., and Graham M. (2013). Guest Editorial.
Retrieved from http://archive.wired.com/software/coolapps/ Environment and Planning A, 45(1), 39.
news/2005/09/68743. Accessed July 14, 2014. Yang, C., Wong, D., Miao, Q., and Yang, R. (eds.). (2010). Advanced
Smith, N. (1992). History and philosophy of geography: Real wars, Geoinformation Science. Boca Raton, FL: CRC Press.
theory wars. Progress in Human Geography, 16(2), 257271. Zipf, A. and Jst, M. M. (2012). Location-based services. In
Soden, R. and Palen L. (2014). From Crowdsourced Mapping W. Kresse and D. M. Danko (eds.), Springer Handbook of
to Community Mapping: The Post-Earthquake Work of Geographic Information. New York: Springer, pp. 417421.
OpenStreetMap Haiti. In R. Chiara, L. Ciolfi, D. Martin, Zook, M., Graham, M., Shelton, T., and Gorman, S. (2010).
and B. Coneien (eds.), COOP 2014: Proceedings of the Volunteered geographic information and crowdsourc-
11th International Conference on the Design of Cooperative ing disaster relief: A case study of the Haitian earthquake.
Systems (pp. 311326), 2730 May 2014, Nice, France. World Medical & Health Policy, 2(2), 733.
2016 by Taylor & Francis Group, LLC
21
Frontiers of GIScience:
Evolution, State of the Art,
and Future Pathways
Acronyms and Definitions ................................................................................................................ 445
21.1 Introduction ............................................................................................................................ 445
21.2 Evolution .................................................................................................................................. 446
21.3 State of the Art ........................................................................................................................ 448
Abstraction Algorithms Assimilation
May Yuan 21.4 Future Pathways and Concluding Remarks ........................................................................452
University of Texas at Dallas References............................................................................................................................................ 453
Acronyms and Definitions Canada Department of Agriculture (Tomlinson, 1962). Yet,
when GIS is broadly defined as a system that deals with geo-
ABM Agent-based modeling graphic information, it can be traced far back into the time
ACM Association for Computing Machinery when humans started recording and sharing knowledge about
CA Cellular automata the environment. Before computer-based GIS technology,
COSIT Conference of Spatial Information Theory oral traditions, and maps were primary means to communi-
ESRI Environmental Systems and Research Institute cate geographic information. Nowadays, GIS technologies are
FGDC Federal Geographic Data Committee diverse and thriving in mapping, spatial analysis and model-
GIS Geographic information systems ing, location-based services, cyber geographical applications,
GIScience Geographic information science and spatial crowdsourcing. GIS technologies are now important
GWR Geographically weighted regression research tools for research and operations in environmental sci-
IaaS Infrastructure as a Service ences, biological and agricultural sciences, public health, urban
IEWAT Information everywhere, any time planning, and economic, political, and social studies. GIScience
LISA Local indicator of spatial autocorrelation serves the conceptual, theoretical, and computational founda-
NCGIA National Center for Geographic Information tions for these technologies.
and Analysis Dating to ~6600 BC, the mural found at the Neolithic site of
NSDI National spatial data infrastructure Catalhyk is considered the worlds oldest map (Schmitt etal.,
NSF National Science Foundation (US) 2014). The early adoption of maps is of no surprise. Maps are
PaaS Platform as a Service intuitive and effective ways to give directions, express spatial
SaaS Software as a Service arrangements of features, and plan spatial activities. Humans
SIGSPATIAL Special Interest Group on Spatial Information made maps long before they invented writing. Likewise,
UCGIS University Consortium for Geographic Tomlinsons GIS was motivated by the use of computers to
Information Science automate map analysis and production. To date, map libraries
VGI Volunteered geographic information continue the important role of curating and providing access
to atlases, aerial photographs, and spatial data in digital forms,
21.1 introduction while massive and diverse geographic information is also widely
available from government agencies, businesses, organizations,
Geographic information science (GIScience) is the science that and communities.
underlies geographic information systems (GIS) technology. While mapping is essential, a GIS also consists of tools to pro-
Roger Tomlinson introduced GIS in his report on computer cess geospatial data, manage geospatial databases, integrate data
mapping and analysis to the National Land Inventory in the through georeferencing and compatible attribute definitions,
445
2016 by Taylor & Francis Group, LLC
446 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 2 1.1 GIScience Synonyms
Term Primary Communities Special Emphases
Spatial science (or geospatial science to stress Geography, statistics, and other domains Mapping, spatial statistics, spatial analysis, spatial modeling
the space of interest at a geographic scale) applying mapping and spatial methods in
their fields, such as spatial social science
Spatial information science Computer science and management Spatial computing, spatial database management, and
information science visualization and communication of spatial information
Geoinformatics Geoscience, computer science, geostatistics Information computing and management for geoscience data,
most commonly with earth science data
Geomatics Geodetics, survey engineering, geophysics, Geoid, datums, land surveys, mapping
earth science
Note: Generally, these terms are synonyms of GIScience, but they are popular in different communities with some variations in areas of emphases.
analyze embedded spatial patterns, model geographic phenom- 1. A phase of pioneers from the early 1960s to 1975. Key lead-
ena and processes, and render data and findings in multiple ers included Howard Fisher of the Harvard Laboratory for
ways. The technology was initially developed out of application Computer Graphics, Roger Tomlinson of the Canadian
needs, and its conceptual and computational frameworks were Geographic Information System, and David Bickmore
fragmented across solutions. GIScience research contributes to at the Experimental Cartographical Unit in the United
developing fundamental frameworks for GIS technologies and Kingdom.
takes the technological challenges to improve our understand- 2. A phase of national drivers from 1973 to early 1980s. Key
ing of geographic information, processes for geographic knowl- agencies included Canadas Department of Agriculture,
edge building and communication, and spatial decision support. the United States Bureau of the Census, and the Ordnance
This chapter aims to highlight the past, present, and future Survey in Great Britain. In the United States, GIS tech-
of GIScience research. As a field of interdisciplinary and multi- nology attracted great interest from many federal agencies
disciplinary research, GIScience enjoys outstanding advances in such as the Department of Defense, Central Intelligence
both breadth and depth as evidenced by the multitude of names Agency, US Forest Service, Fish and Wildlife Service,
associated with the discipline, such as geospatial science, spa- and Department of Housing and Urban Development, as
tial science, spatial information science, geoinformatics, and well as state and local governments including California,
geomatics (Table 21.1). Consequently, it is challenging to cap- Maryland, Minnesota, New York, and others.
ture the full scope of research development in the field. What 3. A phase of commercial dominance from early to late 1980s,
follows reflects the authors perspectives on the evolution, state most noticeably the Environmental Systems and Research
of the art, and future pathways of GIScience. Since the chapter Institute (ESRI, now Esri) and Integraph. The compa-
is focused on GIScience, the discussions here emphasize the nies not only developed GIS software packages but also
key intellectual development of spatial concepts, theories, and designed and implemented GIS projects for government
computational approaches. GIS applications are not GIScience agencies. These GIS packages were adopted in college
research and, therefore, are beyond the scope of this chapter. courses, and to date, they remain the primary tools for
The next section elaborates on the evolution of GIS technologies learning GIS and doing GIS projects. In 1988, the United
to GIScience: from the early emphases on the transitions from States National Science Foundation (NSF) awarded a
technological advances in mapping, spatial database building, grant to establish the National Center for Geographic
and inventory and planning applications to scientific inquiries Information and Analysis (NCGIA) with the University
into the nature of geographic information, spatial computing, of California at Santa Barbara, State University of New
and geographical understanding. Section 21.3 highlights the York at Buffalo, and University of Maine. The NSF grant
active GIScience research directions in cognition, representa- provided $10M dollars for 8 years of NCGIA leadership
tion, integration, and computation. The chapter concludes with that transformed GIS to GIScience and resulted in last-
promising pathways for future GIScience development. ing impacts in education and research in the United States
and around the world.
21.2 evolution 4. A phase of user dominance since early 1990s with the rise
of desktop GIS that emphasized ease of use and promoted
Computer-based GIS technology revolutionized the processes of wide adoption of GIS technology beyond research univer-
recording and disseminating geographic information and invoked sities, large government organizations, and big compa-
new possibilities to represent, analyze, and compute geography. nies. In 1994, US Executive Order 12906 established the
Since its conception, the term GIS was often referenced exclusively Federal Geographic Data Committee (FGDC) as the exec-
to computer-based GIS. Coppock and Rhind (1991) character- utive branch leadership to develop the National Spatial
ized the early development of computerized GIS into four general Data Infrastructure (NSDI) marked the first multiagency
phases from 1960 to 1990: nation-wide efforts to coordinate GIS data management
2016 by Taylor & Francis Group, LLC
Frontiers of GIScience 447
and access. The expanded availability of free GIS data generalization (Buttenfield and McMaster, 1991), cartographic
stimulated many geospatial research and business oppor- label placement (Marks and Shieber, 1991), and interactive digi-
tunities and popularized GIS technology in a wide range tal atlases production (MacEachren, 1998). Landmark spatial
of domain applications. studies led to new methods that account for local variations and
local processes, such as map algebra (Tomlin, 1994), local indica-
In a short period of 30years (1960s1990s), GIS started with a tor of spatial autocorrelation (LISA) (Anselin, 1995), geographi-
few visionaries who sought ways to use computers for mapping cally weighted regression (GWR) (Brunsdon et al., 1998), and
and analyzing geographic data and then grew to a generation of geoalgebra (Takeyama and Couclelis, 1997). Furthermore, spa-
researchers and professionals that brought GIS into mainstream tial modeling advanced new approaches to simulate hydrologi-
college curricula, government functions, and business opera- cal processes (Olivera and Maidment, 1999) and urban systems
tions. With this growth, research efforts went beyond mapping (Couclelis, 1997, Batty, 2007) by leveraging dynamic methods
and spatial data handling. Researchers ventured into the unique from other fields, such as distributed modeling, cellular autom-
complexity of geographic information and ensuing challenges in ata (CA), and agent-based modeling (ABM).
acquiring and using spatial data to understand geographic pro- Moreover, arguments were made that foundations of GIScience
cesses and make spatial predictions. The International Journal of should tie closely to information science (Mark, 2003). Information
Geographical Information Systems (IJGIS) was launched in 1987 science studies the means and processes of information transmis-
and was recognized as the primary academic journal in the field sion among humans and/or computers. Syntactic form, semantic
(Caron etal., 2008). Goodchild published a landmark paper in content, and contextual relevance are key elements in determin-
IJGIS, entitled Geographic information science (Goodchild, ing the value and optimal means of information flows from trans-
1992). The paper highlighted scientific problems unique to geo- mitters to receivers (Worboys, 2003). However, any judgment
graphical data and established the topical content for GIScience. about value and optimality of the key elements must rely on a
Since then, many organizations and journals adopted the term common understanding of the domain between transmitters and
GIScience over GIS. Efforts of the academic community, with receivers. Geographic ontologies became an important subject
most participants from Geography, established the University in GIScience research (Agarwal, 2005), and research on spatial
Consortium for Geographic Information Science (UCGIS) in ontologies and representation along with other issues related to
1994 and, through community efforts, defined GIScience as the the nature of geographic information was prominent in NCGIA
development and use of theories, methods, technology, and data research initiatives and UCGIS research challenges. Fundamental
for understanding geographic processes, relationships, and pat- GIScience research has been promoted through the Conference
terns. The transformation of geographic data into useful informa- on Spatial Information Theory (COSIT) starting in 1993 and
tion is central to geographic information science (UCGIS, 2002, International Conference on GIScience (GIScience), which began
Mark, 2003). It is important to note that GIScience research is not in 2000. Since then, the two conferences have been held in alter-
about using GIS technologies to solve scientific problems. This is nate years and locations between Europe and North America. In
similar to statistics and mathematics; applications of statistical or addition, the Auto-Carto International Symposium on Automatic
mathematical methods to solve a biological problem contribute Cartography and International Symposium on Spatial Data
to biological science, not the sciences of statistics or mathematics. Handling both have a long history as primary academic venues
The early development of GIScience can be attributed to the in GIScience. Computer scientists interested in spatial database
NCGIAs leadership in a series of initiatives as well as the UCGIS and information started the Annual Association for Computing
community efforts to identify and articulate research challenges. Machinery (ACM) Workshop on Advances in Geographic
In 1997, the International Journal of Geographical Information Information Systems in 1993. They successfully expanded the
Systems was renamed International Journal of Geographical annual workshop to Annual ACM-GIS International Symposium
Information Science, marking its second decade of publication in 1998 and furthermore established the ACM Special Interest
(Fisher, 2006). However, the tendency to use GIScience as a syn- Group on Spatial Information (SIGSPATIAL) as the catalyst for
onym for GIS was quite common in early 2000 (Mark, 2003) and research on spatially related information among computer scien-
remains rather persistent today. Many programs offer GIScience tists (Samet etal., 2008).
courses with the same instructional materials for GIS, and many These pioneer efforts established a strong foundation for
do not differentiate GIScience research from research using GIS. GIScience. Research has migrated from GIS-enabling com-
Nevertheless, leading journals (such as IJGIS and Geoinformatics) puterization of geographic data processing and mapping to
and conferences in GIScience (such as GIScience and ACM- GIScience inquiries into the essence of geographic informa-
SIGSPATIAL) emphasize papers with contributions to concep- tion and epistemology. Goodchild (2014) highlighted research
tual, theoretical, and computational innovations. and institutional accomplishments in the 20 years of prog-
Foundational work in cartography, spatial statistics, and spa- ress since the introduction of GIScience in 1994. On mea-
tial modeling has significantly contributed to the development surements, research foci shifted from spatial errors in the
of GIS, and these continue to be important subjects in GIScience 1980s to spatial uncertainty in the 1990s. On representation,
research today. Computer cartography made notable prog- research advanced from vector/raster in the 1980s and objects/
ress in line generalization (Douglas and Peucker, 1973), map fields in the 1990s to complex object-fields and field-objects in
2016 by Taylor & Francis Group, LLC
448 Remotely Sensed Data Characterization, Classification, and Accuracies
Roger Tomlinson of the Canadian Geographic Information system, Canada
Howard Fisher of the Harvard Laboratory for Computer Graphics, United States
1960s1975 David Bickmore of the Experimental Cartographical Unit, United Kingdom
Pioneers
Canadian Department of Agriculture
United States Bureau of Census, and other federal and state governments
1973-early 1980s The Ordnance Survey, Great Britain
National Drivers
Environmental Systems and Research Institute (now Esri)
Integraph
Early to late 1980s
MapInfo
Commercial Dominance
Desktop GIS (ArcView, MapInfo, AtlasGIS, IDRISI, etc.) with strong emphases on user interface design and
ease of use
1990s The Federal Geographic Data Committee (FGDC) efforts on the National Spatial Data Infrastructure (NSDI)
User dominance
National Science Foundation funded the National Center for Geographic Information and Analysis (NCGIA)
Journals, organizations, and conferences adopted names like GIScience, spatial science, geospatial science,
Early mid 1990s2010 geoinformatics, geomatics, etc.
Academic Establishment Academic degrees and certification in GIScience; residence and online programs.
Web services, geospatial cyberinfrastructure, mobile computing
Volunteered geographic information, crowdsourcing, and spatial big data
2010- to date Open source platforms, social coding, and internet of things
Internet, Open and
Mobile
FIg u r e 2 1.1 A summary of the evolution of GI Systems and GIScience.
2000s. GIScience research on analysis progressed from spatial Reality
autocorrelation in the 1970s to spatial heterogeneity in the 1990s.
These efforts also built a strong foundation that support GIScience
research and development into the mainstream of information Uncertainty
technology. The GIScience evolution is summarized in Figure 21.1. Visualization
Over the years, GIScience research frontiers were articu- Spatial cognition
lated through 21 NCGIA research initiatives UCGIS research Spatial simulation
priorities and Computing Community Consortium Spatial Web service
Computing Visioning. Goodchild (2014) listed some of the top- Spatial modeling HPC
The human
ics that resulted from discussions in GIScience communities Spatial analysis
Cloud
IoT
in several venues and summarized in a conceptual framework Algorithm
for GIScience that connects the dimensions of human, soci- Privacy NSDI Data Cyberinfrastructure
ety, and computer. Expanded upon his conceptual framework, Society Computing
Figure 21.2 incorporates major developments in cyberinfra-
structure and computing that have transformed the interplays FIg u r e 21.2 A conceptual framework for research themes in
among the human, society, and computers as well as how we GIScience, whereas HPC stands for high performance computing, and
perceive and understand human, physical, biological, and many IoT is the Internet of Things (Modified from Goodchild, M. F. Journal
of Spatial Information Science, (1), 3, 2014.)
other dimensions of reality. There is no shortage of research
challenges in GIScience. This becomes evident with a quick
search on Google Scholar, which results in more than 5000 21.3 State of the Art
publications on the subject. Some fundamental topics remain
outstanding and are likely to persist at the core of GIScience GIScience continues to evolve with an increasing attention to
research, such as geographic ontologies, space-time representa- what is local rather than global, individual rather than aggre-
tion, spatial algorithms, spatial cognition, geovisualization, and gated, collaborative rather than authoritative, culturally aware
spatial decision support. rather universal, open rather than exclusive, and mobile rather
2016 by Taylor & Francis Group, LLC
Frontiers of GIScience 449
than desktop. Moreover, models and methods are being devel- As a result, geographic categorization and reasoning may vary
oped to represent and visualize multidimensional and multime- from person to person or place to place. For example, cognitive
dia data. Leveraged by the internet, new GIS platforms are being geo-ontologies recognize that people may see things differently,
realized on the World Wide Web, with cyberinfrastructure, and and their conceptualizations may vary due to environmental,
in the cloud. All these developments have profound influences cultural, or linguistic differences (Mark etal., 2011, Wellen and
on what is summarized here as 3 As: abstraction, algorithms, and Sieber, 2013, Turk and Stea, 2014). Such differences have pro-
assimilation throughout GIScience epistemology. found implications for information sharing and integration,
spatial data infrastructure building, spatial decision support,
and many other issues that deal with the usefulness of GIS tech-
21.3.1 Abstraction
nologies and intrinsic technological biases.
Abstraction takes place at multiple levels in GIScience research. Information sharing and integration was the initial motiva-
It is concerned with how we conceptualize geographic worlds tor of ontological research through the rise of the Semantic Web
and spatial problems and subsequently, how we represent, that extends the World Wide Web for people to share and reuse
compute, and communicate all the relevant concepts and find- data beyond application boundaries. Ontological approaches
ings. As spatial data are from different sources, integration can are now commonly used to define specifications of geographic
be challenging at each level of abstraction. Perhaps, the most abstractions in a problem domain (Jung etal., 2013, Ujang and
popular abstraction used in GIScience is the so-called data lay- Rahman, 2013), achieve semantic consistency for data integra-
ers (Figure 21.3). While the data-layer abstraction is intuitive, tion and complex query support (Wiegand, 2012), and assure
GIScience research examines issues in cognition, ontology, and interoperability across systems and over web services (Shi and
statistics (e.g., sampling) for better abstraction of reality. Nellis, 2013). Different frameworks have been proposed for geo-
Across all the levels of abstraction, cognitive research helps ontologies. Frank (2001) argued for a tiered ontology to assure
understand how people learn and organize geographic knowl- consistency constraints based on how different kinds of things
edge. Such cognitive understanding can improve GIS usability are conceptualized and from where they are abstracted. Tier 0
and communication. Montello (2009) summarized five main ontologies are for human-independent reality where natural laws
areas of cognitive research in GIScience since 1992: human fac- prevail regardless of human observers. Tier 1 ontologies are for
tors of GIS, geovisualization (including spatialization), navi- observations of physical world with measurements and statistics.
gation systems, cognitive geo-ontologies, spatial thinking and Tier 2 ontologies are for objects with properties that can be used
memory, and cognitive aspects of geographic education. Much to identify individuals and determine categorical memberships
of the cognitive research confirms the complexity of geographic with necessary and sufficient properties. Tier 3 ontologies are
information and knowledge due to indeterminacy, vagueness, for social reality that is subject to social, cultural, and linguis-
and the interdependency of individuals and geographic context. tic contexts. Finally, Tier 4 ontologies are for subjective knowl-
edge, which may be incomplete or partial, used by individuals or
institutions for reasoning or decision making. Couclelis (2010)
articulated the need for geographic information constructs as
Land cover the core of ontologies in GIScience. Her framework centers on
an ontological hierarchy to connect intentionality and relevant
Structures
information. There are seven levels of semantic resolution in the
Boundaries
hierarchy. In the order of low to high levels, the semantic levels
of resolution include existence, observables, similarities, simple
Hydrography objects, composite objects, function, and purpose. She introduced
the idea of semantic contraction to generalize semantic rich-
Geographic names ness from higher, more complex levels to a lower level of simpler
semantics, and object of discourse to represent entities as com-
Transportation
posites of geographic information constructs at the higher levels
Elevation
of the hierarchy. This ontological research expanded our under-
standing of semantic granularity (Fonseca etal., 2002) and spa-
Orthoimagery tial tasks (Wiegand and Garca, 2007) and laid the foundation
for building theories of geographic information.
In addition to ontologies, abstraction also accounts for
means by which geographic information can be effectively
acquired, analyzed, and communicated. Traditionally, geog-
raphy is abstracted in forms of data from field surveys, maps,
imagery, tables, graphs, and text. Advances have opened new
FIg ur e 21.3 An example of geographic abstractions: the National means to acquire geographic information with new kinds of geo-
Map from the United States Geological Survey. graphic abstraction. For example, data from dynamic geosensor
2016 by Taylor & Francis Group, LLC
450 Remotely Sensed Data Characterization, Classification, and Accuracies
networks (Llaves and Kuhn, 2014), tweets (Tsou etal., 2013), geo- and the programming environmental allow qualitative methods
tagged photos (Samet etal., 2013), and information from various that are commonly used by sociologists and humanities schol-
social media (Croitoru etal., 2013; Jiang and Miao, 2014) offer ars, like coding, triangulating source materials, and content
real-time or near real-time environmental and social abstrac- analysis in recursive and iterative forms to produce knowledge,
tions that enable detection of events and activities as they unfold. such as Geo-Narrative (Kwan and Ding, 2008).
As the geographic world captured by these data is transitory and VGI is only one data source available from the Web. There
ephemeral, so is the ensuing geographic abstraction. Volunteered are many crowdsourced systems (Yuen et al., 2011). For geo-
geographic information (VGI), crowdsourced geographic infor- spatial data, crowdsourced systems usually provide web map
mation (Goodchild, 2007; Goodchild and Glennon, 2010), services or web feature services that support map mash-ups by
and ambient geospatial information (Stefanidis et al., 2013) which geospatial data from remote servers can be visually over-
commonly condense information entries to point locations. lapped in a browser on a client site. As discussed in the abstrac-
Consequently, geographic abstraction is generally reduced to tion section, the Semantic Web transforms web content to data
individuals and collections of points. Spatial synthesis would be as Web 2.0. Various social media facilitate crowdsourcing and
more appropriate than analysis to decipher these data. provide ambient geographic information that can be exploited
Besides geosensor and social media data, multimedia data to recognize social pulses (Croitoru etal., 2013) or validate envi-
incorporate video, audio, virtual reality, and augmented real- ronmental conditions (See et al., 2013). Crowdsourced data are
ity to represent geography (Camara and Raper, 1999). Videos either directly requested by a project web service such as the Did
may be interviews, documentary films, or animation of tem- you feel it? web portal by the United States Geological Surveys
poral information. Audios may be oral stories, narration by (USGS) Earthquake Hazards Program* or harvested from social
a native speaker, testimonies, songs, or animal sounds. Both media feeds via application programming interfaces (APIs), such
video and audio enrich abstractions of geographic reality by as OpenStreetMap API. Heipke (2010) provided a good introduc-
enriching the context of spatial abstraction. Virtual reality and tion on crowdsourcing geospatial data with highlights of success-
augmented reality, usually with 3D visualization, supplement ful projects, the basic technologies and comments on data quality.
spatial abstraction with videos, audios, photographs, digital Since VGI and crowdsourced data lack statistical sampling
documents, and labels in a dynamic context-aware immersive schemes and are collected from various sites, researchers need
environment. Granularity of geographic abstraction becomes to develop customized algorithms for data preprocessing, map-
finer or coarser, depending on the users location and view. ping, and analysis. Of great challenge is the fact that these
Virtual geographic environment (Lin et al., 2013) leverages data violate most, if not all, sampling assumptions based on
virtual reality and multidimensional GIS to provide a digital which conventional statistical methods are founded. Location
platform for geographic experiments through collaborative information associated with VGI, crowdsourced data, and
visualization and simulation. Collaboration requires shared data from web crawling may be explicitly tagged through GPS
geographic abstraction of both declarative knowledge and readings as latitude, longitude, or other x, y coordinate pairs.
procedural knowledge as the basis for communication and Alternatively, location may be implicitly noted in forms of place
integration, which in turn rests on cognitive and ontological names or addresses. Addresses can be matched through geoc-
compatibility. oding against street network databases. For place names, top-
onym resolution and gazetteer matching will be necessary for
georeferencing (Adelfio and Samet, 2013). More generally, con-
21.3.2 Algorithms
ceptual and computational frameworks are being developed to
Algorithms are step-by-step procedures for calculations. Here, transform text to a rich geospatial data source (Vasardani etal.,
algorithms are broadly defined as approaches to data processing, 2013; Yuan etal., 2014). While several studies showed that VGI
analysis, modeling, and simulation. As geographic abstraction and crowdsourced data are timely and at times more represen-
shifts emphases to semantics, the development of spatial algo- tative of geographic reality than authoritative data (Goodchild
rithms also attempts to reveal local meanings and individual and Glennon, 2010), most VGI and geospatial crowdsourced
behaviors in space and time. projects remain primitive and do not go beyond visualization,
The rise of critical GIS (OSullivan, 2006; Schuurman, 2006) animation, and frequency graphing (Batty etal., 2010). Because
reflects the needs to engage social critiques in GIS-based geo- crowdsourced data collection does not follow any statistical
graphic knowledge production in terms of basic concepts, rep- sampling methods, they cannot be applied to established sta-
resentation, participation, and social implications. VGI and web tistical models. Sentiment analysis of postings and messages
map services partially address the needs by empowering ordi- is often based on keywords without reference to the content.
nary citizens to create geographic data and participate in geo- Adetailed view of crowdsourcing can be found in Chapter 26 of
graphic knowledge production. Many critical GIS researchers this book. Chapter 26 also provides many examples of crowd-
also echo the criticisms of positivist biases in GIS and advocate sourcing in spatial sciences.
for qualitative GIS (Cope and Elwood, 2009) to address the needs
to incorporate contextual details and interpretations of the * http://earthquake.usgs.gov/earthquakes/dyfi.
described situation and processes. Broadened GIS methodology http://wiki.openstreetmap.org/wiki/API_v0.6.
2016 by Taylor & Francis Group, LLC
Frontiers of GIScience 451
Besides VGI and crowdsourced data, GIScience researchers 21.3.3 Assimilation
are active in cyberinfrastructure research and cloud comput-
ing. CyberGIS integrates GIS and spatial analysis and model- Assimilation is defined in the Oxford English Dictionary as the
ing into cyberinfrastructure that provides high-performance action of becoming conformed to or conformity with. In this
(terra grid) computing and large-scale data repositories (Wang chapter, assimilation is broadened to processes that bring indi-
et al., 2013). It transforms GIS from an isolated platform to a viduals contributions to the commons for a greater good or to
cybernetwork of supercomputers, virtual organizations, and reveal a bigger picture. With the definition, assimilation efforts
massive shared data resources. The fundamental differences in in GIScience have flourished through Open Source GIS, Social
computing platforms require new algorithms for data process- Coding, Open GIS, and Spatial Turns.
ing, management, analysis, and modeling, and much has been Open source GIS, like GRASS, Quantum GIS, and PostGIS,
implemented as middleware. While also taking the advantage were developed by individuals through community efforts
of internet information technologies, cloud computing lever- (Neteler and Mitasova, 2008) and have gained significant
ages four types of services: Infrastructure as a Service (IaaS), momentum since 2005. To date, there are more than 350 free
Platform as a Service (PaaS), Software as a Service (SaaS), and and open source GIS software packages available.* Along with
Data as a Service (DaaS), with open source resources Hadoop the free software are free data and documents to serve as a
and MapReduce to offer elastic, distributed, and on-demand foundation for building a learning society in which source
computing facilities. code, algorithms, and models can be tested and continuously
The ideas of spatial cloud computing encompass not only improved upon. Steiniger and Bocher (2009) reviewed 10 free
utilization of existing cloud services for intensive spatial com- and open source GIS software packages and argued for the use of
puting but also the development of data and tool services for open source practices and software in research for transparency,
geospatial applications that are made accessible over the web testability, and adaptability to other projects. Many diverse open
(Yang etal., 2011). Cloud computing provides elastic, advanced source GIS communities thrived in 2012 (Steiniger and Hunter,
resources for experimenting with ideas, application develop- 2013). Assimilation of individuals contributions for tool devel-
ment, and app distribution. Location-aware or spatially enabled opment and code improvement in an open source environment
apps are widely available to map property values, routes, crime collectively results in richer and better GIS resources for all.
incidents, restaurants, and gas stations, for example. Cloud com- Social coding follows a similar idea of collaboration, but instead
puting opens GIS workflows to tightly connect to resources on of working toward a package, social coding can be any project
the web and transforms GIS into a service that frees users from or program codes initiated by individuals rather than a com-
desktop computers to anywhere and any platform with internet munity. Perhaps, the most popular social coding site is Github
access. CloudGIS emerges as a promising platform for large- where people can freely copy and modify codes to assimilate into
scale geospatial computing and opens many opportunities for other projects. There are many geospatial projects on Github,
mobile GIS, geosensors, and spatial big data (Bhat et al., 2011; such as CartoDB, GeoNode, Spatial4J, OSGeo, and geopython.
Shekhar etal., 2012; Fan etal., 2013). It is noteworthy that Esri is also active at Github with a suite of
Even before the introduction of big data, many spatial data are open source projects.
big and grow exponentially, especially imagery data and sensor Another large-scale collaborative assimilation is the R proj-
data. The popularity of location-aware devices and geosensors ect, an open source environment for statistical computing and
has motivated algorithm development for trajectory analysis, graphics built upon the R language developed by Ihaka and
geometrically (Li etal., 2011) or semantically (Yan etal., 2013), Gentleman (1996). R users can study the source code to under-
among other methods for movement modeling (Long and stand the underlying statistical procedures and assimilate their
Nelson, 2012). Intensive observation updates of location-aware new modules with existing R methods, which facilitates advances
and sensor data further challenge hardware and software capac- in methodological research and opportunities to submit pro-
ity. CudaGIS is an example of GIS design with GPUs to provide posed models from publications for testing and reuse. Over the
parallel data processing capabilities (Zhang and You, 2012). GPU last 15years, R has gained strong volunteer support in building
algorithms are being developed to enable rapid urban simula- various extensions, including packages for spatial statistics, for
tion (Ma etal., 2008), lidar data processing (Sugumaran etal., example, SpatStat (Baddeley and Turner, 2005), GeoXp (Laurent
2011), viewshed analysis, and other data-intensive computation etal., 2009), and Spacetime (Pebesma, 2012). The call for integra-
(Steinbach and Hemmerling, 2012). Some of the parallel and tion of GIS and spatial data analysis (Goodchild etal., 1992) was
GPU algorithms have been implemented in open source GRASS originally intended to add more spatial analysis capabilities to
GIS for fast spatial computing and rendering (Osterman, 2012). a GIS. Instead, much success has been realized by assimilating
With on-demand IaaS, GPU-based cloud computing has shown spatial data and methods into R statistics. Currently, R consists
to be effective for intelligent transportation management (Wang
and Shen, 2011). A detailed view of cloud computing can be
* http://freegis.org.
found in Chapter 27 of this book. Chapter 27 also gives several http://github.com.
examples of where and how cloud computing is used in spatial One may use R as a GIS (http://pakillo.github.io/R-GIS-tutorial/. Accessed
science. March 22, 2014).
2016 by Taylor & Francis Group, LLC
452 Remotely Sensed Data Characterization, Classification, and Accuracies
of a large suite of spatial modules covering raster analysis, inter- research and development in computing and information sci-
polation and geostatistics, spatiotemporal simulation models, ence (Hazas et al., 2004). Programming libraries are being
spatial autocorrelation, spatial econometrics, spatial structure developed to improve the integration of GIS and remote sens-
models, spatial Bayesian models, spatiotemporal cluster analy- ing (Karssenberg etal., 2007; Bunting etal., 2014). Besides map-
sis, and various mapping and graphing tools (Bivand et al., ping and visualization, these spatial turns not only provide new
2013). In addition, efforts are being made to apply R directly to analytical innovations and leveraged space as a problem fram-
GRASS GIS database files (Bivand, 2000) or port R scripts to ing and reasoning framework but also invoked new perspec-
Quantum GIS (Solymosi etal., 2010). Similarly, spatial analysis tives to improve understanding in natural sciences (Rosenberg
functions are being assimilated into the Python programming and Anderson, 2011), social sciences (Raubal et al., 2014), and
environment, most notably PySAL module (Rey and Anselin, humanities (Bodenhamer, 2013). It is important to make clear
2010), and many spatial functions has been refactored to sup- that these assimilating efforts are developing new spatially inte-
port parallelization (Rey etal., 2013). Free spatial data analysis grated thinking and methodologies, not just applying existing
packages such as GeoDa, although not open source nor extend- GIS technologies in domain sciences.
able, offer a graphic user interface and tight coupling of GIS and
exploratory spatial analysis tools (Anselin etal., 2006). 21.4 Future Pathways and
Assimilation of GIScience into other disciplines led to excit- concluding Remarks
ing new approaches, such as spatial ecology, spatial epidemi-
ology, spatial history, spatial humanities, and spatial social As spatial abstraction, algorithms, and assimilation continue
sciences. In addition to spatial analysis and modeling, a suite evolving, GIScience thrives for multiperspective, distributed,
of geospatial online data processing, information services, and collaborative research across people, platforms, and domain
and computational methods popularizes web mapping and sciences. CyberGIS and Cloud GIS foster high-performance
applications. Figure 21.4 illustrates an example of web appli- and ubiquitous spatial computing. Both technologies not only
cations for spatial ecology research that assimilates species, accelerate spatial data processing but transform the ways of
ecological, and environmental data in the Gulf of Mexico doing GIScience and developing GIS applications. Wright (2012)
(Simons et al., 2013). Location-awareness is now common in sketched a post-GISystems world where GIS is subsumed into a
FIg u r e 21.4 An example of spatial ecology research in the Gulf of Mexico. Species, ecology, and environmental data for 4092 sites were
extracted from 449 references to support meta-analysis of fish habitat and trophic dynamics.
2016 by Taylor & Francis Group, LLC
Frontiers of GIScience 453
broader framework known simply as the web, divorced from the develop geospatial solutions. Spatial analysis, spatial data min-
desktop in a new paradigm (p. 2197). The future of GIScience will ing, mapping, geovisualization, and visual analytics remain
manifest itself in the grand scheme of computational, environ- critical to geospatial data exploration, information understand-
mental, and social sciences. While time and themes are common ing, and knowledge discovery. Moreover, the pathway that will
axes along which disciplines build knowledge, GIScience distin- revolutionize GIScience is heading to the direction in which the
guishes itself with the emphasis of using space as the first-order common mode of GIScience practices is not confined to conven-
principle to acquire, organize, and compute information as well tional research groups but involves scientists, practitioners, and
as discover and share knowledge. The distinction was already citizens in a collaborative social cloud environment. It will be an
apparent in early GIS development and initial discussions on IEWAT world of the people, by the people, and for the people.
GIScience (Goodchild, 1992; Mark, 2003). It will be even more
prominent in the big data era when data from location-aware References
devices continue to grow exponentially in volume and complex-
ity, and spatial contextualization and integration are becoming Adelfio, M. D. and H. Samet (2013). GeoWhiz: Toponym reso-
more effective to sensing making and prediction. lution using common categories. Proceedings of the 21st
The emphasis of space (e.g., spatiality, location, and situation) ACM SIGSPATIAL International Conference on Advances in
will continue to be the focus in pathways for future GIScience Geographic Information Systems. ACM.
development in a world where we have access to needed infor- Agarwal, P. (2005). Ontological considerations in GIScience.
mation everywhere, any time, that is, an IEWAT world, enabled International Journal of Geographical Information Science
through onlineoffline integration, the Internet of Things, cloud- 19(5): 501536.
mobile computing, collaborative information seeking and knowl- Anselin, L. (1995). Local indicators of spatial associationLISA.
edge building, and integrative cyberphysicalsocial systems. Geographical Analysis 27(2): 93115.
Clearly, these are also hot topics in the broader computational, Anselin, L., I. Syabri, and Y. Kho (2006). GeoDa: An introduction
environmental, and social sciences. In other words, the future to spatial data analysis. Geographical Analysis 38(1): 522.
pathways of GIScience are intimately intertwined with those of Baddeley, A. and R. Turner (2005). Spatstat: An R package for ana-
computational, environmental, and social sciences, and further- lyzing spatial point patterns. Journal of Statistical Software
more, GIScience should contribute substantially through under- 12(6): 142.
standing of space and use of space to achieve the vision of an Batty, M. (2007). Cities and Complexity: Understanding Cities
IEWAT world. with Cellular Automata, Agent-Based Models, and Fractals.
Recent developments in GIScience have built strong founda- Cambridge, MA: The MIT Press.
tions in all the three areas of spatial abstraction, algorithms, and Batty, M., A. Hudson-Smith, R. Milton, and A. Crooks (2010).
assimilation, as discussed in Section 21.3. The pathways forward Map mashups, Web 2.0 and the GIS revolution. Annals of
for an IEWAT world would extend the three areas into a mul- GIS 16(1): 113.
tiverse of a truly diverse, distributed, and collaborative nature. Bhat, M. A., R. M. Shah, and B. Ahmad (2011). Cloud comput-
Every location, every person, and everything is becoming a data ing: A solution to Geographical Information Systems (GIS).
producer. Data are from everywhere and anytime with differ- International Journal on Computer Science & Engineering
ent ontological notions. Algorithms are being developed, coded, 3(2): 594600.
modified, and forked by many over the web. Information is being Bivand, R. S. (2000). Using the R statistical data analysis language
analyzed and synthesized dynamically and continuously to on GRASS 5.0 GIS database files. Computers & Geosciences
reflect real-time and near real-time situations in the environment 26(9): 10431052.
and our society. Online and offline computational platforms Bivand, R. S., E. Pebesma, and V. Gmez-Rubio (2013). Applied
are being transitioned seamlessly to maximize the efficiency Spatial Data Analysis with R. New York: Springer.
of mobile computing anywhere and anytime. Fully integrative Bodenhamer, D. J. (2013). Beyond GIS: Geospatial technologies
cyberphysicalsocial systems inform us of the past, present, and the future of history. In History and GIS, von Lunen, A.
and future of what things/people are, where they reside, how they and Travis, C. (eds.). New York: Springer, pp. 113.
work, how they may evolve, where we should go, and what we Brunsdon, C., S. Fotheringham, and M. Charlton (1998).
should do. Geographically weighted regression. Journal of the Royal
To date, a GIS is no longer confined in a computer system Statistical Society: Series D (The Statistician) 47(3): 431443.
or as a software package. GIS is immersed into the greater web Bunting, P., D. Clewley, R. M. Lucas, and S. Gillingham (2014).
computing environment and heading to an IEWAT world of The Remote Sensing and GIS Software Library (RSGISLib).
truly ubiquitous spatial computing. Ontological and cogni- Computers & Geosciences 62: 216226.
tive understandings of geospatial categorization and reason- Buttenfield, B. P. and R. B. McMaster (1991). Map Generalization:
ing are essential to properly conceptualize geospatial problems Making Rules for Knowledge Representation. New York:
and realize geospatial abstractions to connect reality and GIS Longman Scientific & Technical.
databases. Spatial programming, web programming, and sta- Camara, A. S. and J. Raper (1999). Spatial Multimedia and Virtual
tistical programming are essential skills to analyze data and Reality. CRC Press.
2016 by Taylor & Francis Group, LLC
454 Remotely Sensed Data Characterization, Classification, and Accuracies
Caron, C., S. Roche, D. Goyer, and A. Jaton (2008). GIScience Ihaka, R. and R. Gentleman (1996). R: A language for data analy-
journals ranking and evaluation: An international Delphi sis and graphics. Journal of Computational and Graphical
study. Transactions in GIS 12(3): 293321. Statistics 5(3): 299314.
Cope, M. and S. Elwood (2009). Qualitative GIS: A Mixed Methods Jiang, B. and Y. Miao (2014). The evolution of natural cities from
Approach. Sage. the perspective of location-based social media. arXiv pre-
Coppock, J. T. and D. W. Rhind (1991). The history of GIS. print arXiv:1401.6756.
Geographical Information Systems: Principles and Jung, C.-T., C.-H. Sun, and M. Yuan (2013). An ontology-enabled
Applications 1(1): 2143. framework for a geospatial problem-solving environment.
Couclelis, H. (1997). From cellular automata to urban models: Computers, Environment and Urban Systems 38: 4557.
New principles for model development and implementa- Karssenberg, D., K. de Jong, and J. Van Der Kwast (2007).
tion. Environment and Planning B 24: 165174. Modelling landscape dynamics with Python. International
Couclelis, H. (2010). Ontologies of geographic information. Journal of Geographical Information Science 21(5): 483495.
International Journal of Geographical Information Science Kwan, M.-P. and G. Ding (2008). Geo-narrative: Extending
24(12): 17851809. geographic information systems for narrative analysis in
Croitoru, A., A. Crooks, J. Radzikowski, and A. Stefanidis (2013). qualitative and mixed-method research. The Professional
Geosocial gauge: A system prototype for knowledge discov- Geographer 60(4): 443465.
ery from social media. International Journal of Geographical Laurent, T., A. Ruiz-Gazen, and C. Thomas-Agnan (2009).
Information Science 27(12): 24832508. GeoXp: An R package for exploratory spatial data analy-
Douglas, D. H. and T. K. Peucker (1973). Algorithms for the reduc- sis. TSE Working Paper Series 99-099. Toulouse, France:
tion of the number of points required to represent a digi- Toulouse School of Economics.
tized line or its caricature. Cartographica: The International Li, Z., J. Han, M. Ji, L.-A. Tang, Y. Yu, B. Ding, J.-G. Lee, and
Journal for Geographic Information and Geovisualization R.Kays (2011). MoveMine: Mining moving object data for
10(2): 112122. discovery of animal movement patterns. ACM Transactions
Fan, X., S. Wu, Y. Ren, and F. Deng (2013). An approach to on Intelligent System and Technology 2(4): 132.
providing cloud GIS services based on scalable clus- Lin, H., M. Chen, and G. Lu (2013). Virtual geographic environ-
ter. 21st International Conference on Geoinformatics ment: A workspace for computer-aided geographic experi-
(GEOINFORMATICS). IEEE. ments. Annals of the Association of American Geographers
Fisher, P. F. (2006). Classics from IJGIS: Introduction Twenty Years 103(3): 465482.
of IJGIS. CRC Press, pp. 16. Llaves, A. and W. Kuhn (2014). An event abstraction layer for
Fonseca, F., M. Egenhofer, C. Davis, and G. Cmara (2002). the integration of geosensor data. International Journal of
Semantic granularity in ontology-driven geographic infor- Geographical Information Science (ahead-of-print): 122.
mation systems. Annals of Mathematics and Artificial Long, J. A. and T. A. Nelson (2012). A review of quantita-
Intelligence 36(12): 121151. tive methods for movement data. International Journal of
Frank, A. U. (2001). Tiers of ontology and consistency constraints Geographical Information Science 127.
in geographical information systems. International Journal Ma, C., Y. Qi, Y. Chen, Y. Han, and G. Chen (2008). VR-GIS: An
of Geographical Information Science 15(7): 667678. integrated platform of VR navigation and GIS analysis for
Goodchild, M. F. (1992). Geographical information science. city/region simulation. Proceedings of the Seventh ACM
International Journal of Geographical Information Systems SIGGRAPH International Conference on Virtual-Reality
6(1): 3145. Continuum and Its Applications in Industry. ACM.
Goodchild, M. F. (2007). Citizens as sensors: The world of volun- MacEachren, A. M. (1998). Cartography, GIS and the world wide
teered geography. GeoJournal 69(4): 211221. web. Progress in Human Geography 22: 575585.
Goodchild, M. F. (2014). Twenty years of progress: GIScience in Mark, D. M. (2003). Geographic information science: Defining
2010. Journal of Spatial Information Science (1): 320. the field. Foundations of Geographic Information Science
Goodchild, M. F. and J. A. Glennon (2010). Crowdsourcing geo- 318.
graphic information for disaster response: A research fron- Mark, D. M., A. G. Turk, N. Burenhult, and D. Stea (2011).
tier. International Journal of Digital Earth 3(3): 231241. Landscape in Language: Transdisciplinary Perspectives. John
Goodchild, M. F., R. Haining, and S. Wise (1992). Integrating Benjamins Publishing.
GIS and spatial data analysis: Problems and possibilities. Marks, J. and S. M. Shieber (1991). The Computational Complexity
International Journal of Geographical Information Systems of Cartographic Label Placement. Citeseer.
6(5): 407423. Montello, D. R. (2009). Cognitive research in GIScience: Recent
Hazas, M., J. Scott, and J. Krumm (2004). Location-aware com- achievements and future prospects. Geography Compass
puting comes of age. Computer 37(2): 9597. 3(5): 18241840.
Heipke, C. (2010). Crowdsourcing geospatial data. ISPRS Journal Neteler, M. and H. Mitasova (2008). Open Source Software and
of Photogrammetry and Remote Sensing 65(6): 550557. GIS. Springer.
2016 by Taylor & Francis Group, LLC
Frontiers of GIScience 455
Olivera, F. and D. Maidment (1999). Geographic Information interaction database for management and modeling research
Systems (GIS)based spatially distributed model for runoff in the Gulf of Mexico large marine ecosystem. Bulletin of
routing. Water Resources Research 35(4): 11551164. Marine Science 89(1): 135160.
Osterman, A. (2012). Implementation of the r. cuda. los module Solymosi, N., S. E. Wagner, . MartiAgts, and A. Allepuz
in the open source grass gis by using parallel computation (2010). maps2WinBUGS: A QGIS plugin to facilitate data
on the nvidia cuda graphic cards. Elektrotehniski Vestnik processing for Bayesian spatial modeling. Ecography 33(6):
79(12): 1924. 10931096.
OSullivan, D. (2006). Geographical information science: Critical Stefanidis, A., A. Crooks, and J. Radzikowski (2013). Harvesting
GIS. Progress in Human Geography 30(6): 783. ambient geospatial information from social media feeds.
Pebesma, E. (2012). Spacetime: Spatio-temporal data in r. Journal GeoJournal 78(2): 319338.
of Statistical Software 51(7): 130. Steinbach, M. and R. Hemmerling (2012). Accelerating batch
Raubal, M., G. Jacquez, J. Wilson, and W. Kuhn (2014). processing of spatial raster analysis using GPU. Computers
Synthesizing population, health, and place. Journal of & Geosciences 45: 212220.
Spatial Information Science (7): 103108. Steiniger, S. and E. Bocher (2009). An overview on current free and
Rey, S. J. and L. Anselin (2010). PySAL: A Python library of spa- open source desktop GIS developments. International Journal
tial analytical methods. In Handbook of Applied Spatial of Geographical Information Science 23(10): 13451370.
Analysis, Springer, pp. 175193. Steiniger, S. and A. J. Hunter (2013). The 2012 free and open
Rey, S. J., L. Anselin, R. Pahle, X. Kang, and P. Stephens (2013). source GIS software mapA guide to facilitate research,
Parallel optimal choropleth map classification in PySAL. development, and adoption. Computers, Environment and
International Journal of Geographical Information Science Urban Systems 39: 136150.
27(5): 10231039. Sugumaran, R., D. Oryspayev, and P. Gray (2011). GPU-based
Rosenberg, M. S. and C. D. Anderson (2011). PASSaGE: Pattern cloud performance for LiDAR data processing. Proceedings
analysis, spatial statistics and geographic exegesis. Version of the Second International Conference on Computing for
2. Methods in Ecology and Evolution 2(3): 229232. Geospatial Research and Applications. ACM.
Samet, H., M. D. Adelfio, B. C. Fruin, M. D. Lieberman, and Takeyama, M. and H. Couclelis (1997). Map dynamics:
J. Sankaranarayanan (2013). PhotoStand: A map query Integrating cellular automata and GIS through Geo-
interface for a database of news photos. Proceedings of the Algebra. International Journal of Geographical Information
VLDB Endowment 6(12): 13501353. Science 11(1): 7391.
Samet, H., W. G. Aref, C.-T. Lu, and M. Schneider (2008). Tomlin, C. D. (1994). Map algebra: One perspective. Landscape
Proposal to ACM for the establishment of SIGSPATIAL, and Urban Planning 30(1): 312.
ACM-SIGSPATIAL. Tomlinson, R. (1962). An introduction to the use of electronic
Schmitt, A. K., M. Danik, E. Aydar, E. en, . Ulusoy, and O. M. computers in the storage, compilation and assessment of
Lovera (2014). Identifying the volcanic eruption depicted natural and economic data for the evaluation of marginal
in a neolithic painting at atalhyk, Central Anatolia, lands. The National Land Capability Inventory Seminar.
Turkey. PloS One 9(1): e84711. Ottawa, Ontario, Canada: The Agricultural Rehabilitation
Schuurman, N. (2006). Formalization matters: Critical GIS and and Development Admininstration of the Canada
ontology research. Annals of the Association of American Department of Agriculture, p. 11.
Geographers 96(4): 726739. Tsou, M.-H., J.-A. Yang, D. Lusher, S. Han, B. Spitzberg, J. M.
See, L., A. Comber, C. Salk, S. Fritz, M. van der Velde, C. Perger, Gawron, D. Gupta, and L. An (2013). Mapping social
C. Schill, I. McCallum, F. Kraxner, and M. Obersteiner activities and concepts with social media (Twitter) and
(2013). Comparing the quality of crowdsourced data web search engines (Yahoo and Bing): A case study in 2012
contributed by expert and non-experts. PloS One 8(7): US Presidential Election. Cartography and Geographic
e69958. Information Science 40(4): 337348.
Shekhar, S., V. Gunturi, M. R. Evans, and K. Yang (2012). Spatial Turk, A. and D. Stea (2014). David Marks contribution to ethno-
big-data challenges intersecting mobility and cloud com- physiography research. International Journal of Geographical
puting. Proceedings of the Eleventh ACM International Information Science (ahead-of-print): 118.
Workshop on Data Engineering for Wireless and Mobile UCGIS. (2002). UCGIS bylaws. Retrieved March 6, 2014, from
Access. ACM. http://ucgis.org/basic-page/laws.
Shi, X. and M. D. Nellis (2013). Semantic web and service com- Ujang, U. and A. A. Rahman (2013). Temporal three-dimensional
putation in GIScience applications: A perspective and pro- ontology for geographical information science (GIS)A
spective. Geocarto International (ahead-of-print): 118. review. Journal of Geographic Information System 5(3).
Simons, J. D., M. Yuan, C. Carollo, M. Vega-Cendejas, T. Shirley, Vasardani, M., S. Timpf, S. Winter, and M. Tomko (2013). From
M. L. Palomares, P. Roopnarine, L. Gerardo Abarca Arenas, descriptions to depictions: A conceptual framework. In
A. Ibanez, and J. Holmes (2013). Building a fisheries trophic Spatial Information Theory. Springer, pp. 299319.
2016 by Taylor & Francis Group, LLC
456 Remotely Sensed Data Characterization, Classification, and Accuracies
Wang, K. and Z. Shen (2011). Artificial societies and GPU-based Yan, Z., D. Chakraborty, C. Parent, S. Spaccapietra, and K. Aberer
cloud computing for intelligent transportation manage- (2013). Semantic trajectories: Mobility data computation
ment. IEEE Intelligent Systems 26(4): 2228. and annotation. ACM Transactions on Intelligent Systems
Wang, S., L. Anselin, B. Bhaduri, C. Crosby, M. F. Goodchild, and Technology (TIST) 4(3): 49.
Y.Liu, and T. L. Nyerges (2013). CyberGIS software: A syn- Yang, C., M. Goodchild, Q. Huang, D. Nebert, R. Raskin, Y. Xu,
thetic review and integration roadmap. International Journal M. Bambacus, and D. Fay (2011). Spatial cloud computing:
of Geographical Information Science 27(11): 21222145. How can the geospatial sciences use and help shape cloud
Wellen, C. C. and R. Sieber (2013). Toward an inclusive seman- computing? International Journal of Digital Earth 4(4):
tic interoperability: The case of Cree hydrographic features. 305329.
International Journal of Geographical Information Science Yuan, M., J. McIntosh, and G. De Lozier (2014). GIS as a narrative
27(1): 168191. generation platform. In Spatial Narrative and Deep Mapping,
Wiegand, N. (2012). Ontology for the Engineering of Geospatial Bodenhamer, J. C. D. and Harris, T (eds.). Indianopolis, IN:
Systems: Geographic Information Science. Springer, pp. 270283. Indiana University Purdue University Indianapolis Press.
Wiegand, N. and C. Garca (2007). A taskbased ontology Yuen, M.-C., I. King, and K.-S. Leung (2011). A survey of crowd-
approach to automate geospatial data retrieval. Transactions sourcing systems. Privacy, security, risk and trust (pas-
in GIS 11(3): 355376. sat). 2011 IEEE Third International Conference on Social
Worboys, M. F. (2003). Communicating geographic information in Computing (socialcom). IEEE.
context. Foundations of Geographic Information Science 3345. Zhang, J. and S. You (2012). CudaGIS: Report on the design
Wright, D. J. (2012). Theory and application in a post-GISystems and realization of a massive data parallel GIS on GPUs.
world. International Journal of Geographical Information Proceedings of the Third ACM SIGSPATIAL International
Science 26(12): 21972209. Workshop on GeoStreaming. ACM.
2016 by Taylor & Francis Group, LLC
22
Object-Based Regionalization
for Policy-Oriented
Partitioning of Space
Acronyms and Definitions ................................................................................................................ 457
22.1 Mapping Latent Phenomena ................................................................................................. 457
Stefan Lang Ambitious Policy Targets Require a Systemic View Approaching Complex Phenomena with
University of Salzburg a Spatial Focus GeonsTerminology and Conceptual Background
22.2 Domain-Specific Regionalization ........................................................................................ 461
Stefan Kienberger Principles of Regionalization Integrated Geons Monitoring Geons How To Validate
University of Salzburg Geons? In Depth: Systemic Areal Units
22.3 Case Studies ............................................................................................................................. 468
Michael Hagenlocher Socioeconomic Vulnerability to Hazards Social Vulnerability To Malaria Landscape
University of Salzburg
Sensitivity Climate Change Susceptibility (Cumulative CC Impact)
Lena Pernkopf 22.4 Conclusion ................................................................................................................................475
University of Salzburg References.............................................................................................................................................475
Acronyms and Definitions OFA Object fate analysis
PCA Principal component analysis
ASEAN Association of Southeast Asian Nations R Red
AVHRR Advanced very-high-resolution radiometer RBC Recognition-by-component
CCCI Cumulative climate change impact RBD River basin district(s)
CRU Climate Research Unit SD Spectral distance
DEM Digital elevation model SHP Shape Index
DFO Dartmouth Flood Observatory TIN Triangulated irregular network
EAC East African Community UNEP United Nations Environment Programme
EC European Commission VBD Vector-borne disease
ECOWAS Economic Community of West African States VGI Volunteered geographic information
ESP Estimate scale parameter VHI Vegetation Health Index
EU European Union WHO World Health Organization
GEOBIA Geographical object-based Image analysis
GIS Geographic Information System(s) 22.1 Mapping Latent Phenomena
IPCC Intergovernmental Panel on Climate Change
LV Local variance Latent, that is, not directly observable, spatial phenomena can be
MAUP Modifiable areal unit problem operationalized and thus mapped by integrating multiple sets of
MODIS Moderate-resolution imaging spectroradiometer indicators. Spatial representation of such complex phenomena,
MOVE Methods for the improvement of vulnerability in order to be useful to decision makers in different domains,
assessment in Europe needs to be likewise efficient and robust. Drawing from the
NIR Near infrared experiences in object-based image analysis (OBIA) and the spa-
NOAA National Oceanic and Atmospheric Administration tial integration of multidimensional image data, regionalization
OBA Object-based analysis techniques can be applied to map, analyze, and represent these
OBIA Object-based image analysis phenomena in a spatial explicit way.
457
2016 by Taylor & Francis Group, LLC
458 Remotely Sensed Data Characterization, Classification, and Accuracies
1 Vulnerability
Conceptual to specific hazard
level (e.g., floods)
1 Policy-relevant phenomenon
Abstract (latent construct)
Function of ...
Operationalization
2 Sensitivity 2 Adaptive capacity
2 (a) Domains 3 Susceptibility Function of ...
indicator A
3 Susceptibility 2 Social capacity 2 Resilience
(b) Subdomains indicator B
3 Skills indicators 3 Economic
surplus/economic
3 Technologies alternatives
3 Susceptibility indicators
indicators
indicator n
Singular compartments of Cultural/social/
3 Information 3
information (layers) political constraints
indicators
indicators
3 Indicators 3 Governance
Concrete (proxies) indicators
(a) (b)
FIg u r e 22.1 Operationalizationbreaking down latent phenomena to measurable indicators. (a) From abstract, conceptual level to measurable
indicators. (b) Example of operationalizing vulnerability to a specific hazard (e.g., a flood, an earthquake) and its conceptualization via domains,
subdomains, and sets of indicators.
22.1.1 Ambitious Policy targets When reaching down to the bottom so to speak, we approach
Require a Systemic View the problem by separate aspects, that is, analytically, and
likely not able to see the full picture. Measuring and monitor-
The complexity of global challenges faced today entails a holis- ing every contributing aspect separately is fundamental, but
tic, systemic viewpoint, in order to better understand their more important, though also more challenging, is to capture
causeeffect relationships and mutual dependencies. Assessing their cumulative effect (Hagenlocher et al. 2014b). Target-
and monitoring the status of sustainable development policies oriented strategic commitments increasingly adopt an inte-
in a comprehensive and holistic sense relies on integrative, syn- grated assessment approach (Risbey etal. 1996, Jakeman and
thesizing techniques (Fiksel 2006, Stahl etal. 2011). The global Letcher 2003). On implementation level, policies and direc-
warming debate, for example, entails an integrated view on the tives on national, regional (EU, ECOWAS, ASEAN, etc.), and
interrelationships among the environmental sphere and the global level try to accommodate holistic concepts and inte-
social sphere (including cultural, institutional, and economic grated assessments, influenced by systems thinking (Capra
aspects) in integrated assessment models (Parson 1994, Weyant and Luisi 2014). At the same time, they strive for pragmatic
etal. 1996). These spheres are complex phenomena themselves ways to communicate complexity in a simplified manner.
with several subspheres to be considered: for example, in the Comprehensive but integrative, these concepts assume that
ecological sphere, one may investigate the behavior of ecologi- through systemic behavior, new qualities emerge, which to
cal sensitivity to changing climatic conditions; in the socio- capture is both a challenge and a chance. Vulnerability, resil-
cultural sphere, one may study social vulnerability to natural ience, mitigation, etc. can be considered as systemic proper-
hazards. The conceptual operationalization aims at breaking ties that require integrative, synthesizing mapping techniques
down the complexity of these systems into graspable, and ulti- (Capra and Luisi 2014). In other words, spatial analysis tech-
mately measurable, compartments. For example, the concept of niques that are capable to integrate compartments of informa-
vulnerability as a latent phenomenon has been operationalized tion, and overcome the classical, analytical concept of single
by Kienberger et al. (2009)* in several stages of domains and geospatial data layers are sought (Lang etal. 2014b). A suitable
subdomains, until the level of measurable indicators is reached integrative geographic technique for this is regionalization,
(see Figure 22.1). also known as spatial classification (Wise etal. 2001), that is
also the methodological core element of the geon approach, is
* Recently the concept has been refined, as discussed in Section 22.3. presented in this chapter.
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 459
22.1.2 Approaching complex Phenomena units (Kienberger et al. 2009), sensitivity units for strategic
with a Spatial Focus environmental assessment (Pernkopf and Lang 2011), and
hotspot analysis for climate change adaptation (Hagenlocher
Abstract or complex phenomena that cannot be observed etal. 2014b). The framework builds on a workflow to region-
directly are termed latent variables, or factors (Byrne 1998). alize geodata (i.e., imagery or other continuous geospatial
Byrne concludes if latent variables are not observed directly, data layers), resulting in a set of geons. These are units that
[] they cannot be measured directly (p. 4). A way out of this synthetically aggregate domain-specific information with a
dilemma is to operationally define the latent variable of inter- uniform response regarding the complex phenomenon under
est in terms of behavior believed to represent it (Byrne 1998) concern. A geon set represents the spatial explicit distribution
(p. 4). The underlying construct is constituted by the direct mea- of this aggregated information. Table 22.1 and Section 22.3
surement of the observed (or manifest) variable. Manifest vari- contain examples to illustrate this. The term geon was initially
ables are presumed to represent the underlying construct and introduced in cognitive psychology by Biederman (1987) in
serve as indicators of it (Byrne 1998). When using such indica- his theory of recognition-by-components (RBC). It is based
tors, a conceptual bridge is built between the measurable and on the concept of volumetric primitives, defined as geons
the latent part of the phenomenon, a fact that requires careful (geometric ions), and the hypothesis that cognitive objects
investigation. can be decomposed into basic shapes or components. Geons
Complex phenomena are, for example, vulnerability to haz- or gaeons (Peuquet 2002) in Biedermans view are basic volu-
ards, adaptability to climate change, landscape sensitivity to metric bodies such as cubes, spheres, cylinders, and wedges.
human impact, and quality of life, among others. All these can- Conformity with the original geon concept is discussed by
not be directly measured with a specific measuring device and Lang et al. (2008) with regard to (1) the role of generaliza-
hardly reduced to a single indicator. A strategy is to approach tion for the definition and strength of a geon (though: scale
such complex properties by aggregating manifest variables or dependent); (2) the significance of the spatial organization of
indicators into an index or composite indicator, a procedure that the elements, which leads to emergent properties and specific
works on multidimensional attribute spaces (Nardo etal. 2005). qualities; (3) the possibility of recovering objects in the pres-
However, such procedures neglect the true spatial distribution of ence of occlusions (i.e., data errors, measure failures, lack of
such properties when using a priori geographies (such as neigh- data, mismatch of data due to bad referencing). So the con-
borhood or district boundaries) that obscure or bias the actual cept presented here is related, but not identical to Biedermans
spatial distribution. idea. The term geon has been proposed by Lang (2008) to
Implementing policies on regional, national, or local level widen the scope of the original concept and adapt it to the
that address such complex phenomena ideally adopts a synthetic domain of GIScience. Lang etal. (2014b) proposed the follow-
view on multidomain and multisource geospatial datasets. The ing redefinition:
fine granularity of recent geographical datasets enables a multi-
purpose, effective usage for the public sector in fulfilling public A geon (derived from Greek g () = land, earth and the
tasks (Wise etal. 2001), which are more and more interrelated suffix -on = something being) is a type of region, semi-
and integrated. The unitization of space may ideally follow the automatically delineated with expert knowledge incorpo-
topic-related functional characteristics of the phenomenon rated, scaled and of uniform response to a phenomenon
addressed, both in terms of scale and thematic discontinuities. under space-related policy concern. The aim of generating
In other words, boundaries should be set where they reflect a sig- geons is to map policy-relevant spatial phenomena in an
nificant change in behavior of this phenomenon. An example is adaptive and expert-validated manner, commensurate to
the European Water Framework Directive (2000/60/EC), where the respective scale of intervention.
so-called river basin districts (RBDs) transgress national bound-
aries and represent catchments rather than existing administra- The geon approach aims at to delineate regions, but does not
tive units. Generalizing from this example, we argue that the relate to administrative regions with predefined, normative
spatial variability of the actual phenomenon being addressed boundaries. Instead of the commonly used term analytical
requires a more flexible spatial unitization within the geographi- regions (Duque and Suriach 2006), we prefer to use the term
cal policy scope. synthetic [sensu Sui (2011)] regions to emphasize the integra-
tive character of geons. Thus, geon stands for a systems-theory-
driven, scale-dependent earth-object. Policy relevant units are
22.1.3 Geonsterminology and
based on advanced geodata integration, expert knowledge, and
conceptual Background
user validation. Lang et al. (2014b) distinguish between two
For providing such policy-related spatial units, a methodolog- types of geons, composite and integrated geons. We focus on
ical framework has been developed (Lang et al. 2014b). This the latter in the present chapter (see Section 22.2). Geonalytics
has been tested in different policy domains such as functional are built upon a comprehensive pool of techniques, tools, and
biotope complexes for regional development planning (Tiede methods for (1) geon generation (i.e., transformation of con-
etal. 2010), disaster risk reduction, for example, vulnerability tinuous geoinformation into discrete objects by algorithms for
2016 by Taylor & Francis Group, LLC
460 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 2 2.1 Operationalization of Latent Phenomena in a Certain Policy Scope and Geographical Scale Using a Set of Spatialized Indicators
(Dimensionality of Indicator Space = Number of Indicators, Modeling Scale = Minimum Resolution)
Latent Phenomenon Geographic Scale (Modeling Dimensionality of # of Geons
(Policy Scope) Scale) Domains Indicator Space (Geon Size)a
Vulnerability units/climate Salzach river catchment Sensitivity, adaptive capacity 52 1462 [36/4.0/3.8]
change context (Austria) (1km2) (socioeconomic dimension)
Vulnerability units/disaster Salzach river catchment (1km2) Exposure, susceptibility, lack of 16 181 [147/32.6/28.6]
risk reduction context resilience (social dimension)
Economic dimension 6 300 [118/19.7/18.5]
Environmental dimension 21 314 [376/18.8/38.5]
Physical dimension 22 248 [147/23.8/22.1]
Vulnerability units/disaster District of Bzi, Mozambqiue Susceptibility, adaptive 11 307 [108/23.4/18.7]
risk reduction context (1km2) capacity (social dimension)
Economic dimension 6 225 [127/32.0/23.9]
Environmental dimension 5 391 [120/18.4/33.9]
Physical dimension 4 213 [801/33.8/105.9]
Sensitivity units/strategic Marchfeld region, Ability of the ecosystem to 9 151 [46.3/5.8/6.5]
impact assessment context Austria (250 m) resist to external disturbances
Ecological significance of the 6 121 [55.3/7.2/9.5]
affected areas
Societal significance 6 139 [40.8/6.3/6.8]
Climate change hot spots/ Sahel and western Africa n/a 4 2,283 [645/63.5/42.1]
climate change adaptation (subcontinental) (~16km2)
context
The delineated geons are characterized by number (#) as well as size in km (max/mean/std-dev).
interpolation, segmentation, and regionalization); (2) analyzing pragmatic reasons, we can represent complex phenomena in a
spatial arrangements and characterize form and spatial orga- more appropriate way.
nization, investigate spatial emergent properties, and issues of So just like image segmentation is used to represent tangible
scale; (3) monitoring of modifications and changes and evalua- real-world objects in multidimensional feature spaces, latent
tion of the status of geons. phenomena can be represented by regionalizing multiple sets of
The key issue is spatial regionalization of a complex spatial spatialized indicators. Spatialized in this context means: indica-
reality. This reality is represented exhaustively and in a range of tors are represented spatially explicit as continuous fields. What
interdependent, inherent scales. The strategy of the approach is considered continuous is scale depending though. Usually this
is per se geographically motivated. Geographic location is a implies a sampling of the indicator using any kind of tessellation
key to an integrative assessment of complex and multidimen- into small units, be it regular (grid cells, hexagons, etc.) or irreg-
sional phenomena that have a spatial component. With the ular (TINs, enumeration units, etc.). These again may source
ever-increasing maturity of Geographic Information Systems from data collected on finer level (irregular sensor measurement
(GIS) in its broadest sense, the integrative power of space has grids, socioeconomic data collected on household level, etc.) and
been boosted over recent years, also across disciplines, on then interpolated over the area. Often simply the term spatial is
conceptual, technological, and methodological levels (Lang used for simplification, but strictly speaking spatial indicators
etal. 2014a). Integrated spatial analysis methods support the measure spatial properties (size, distribution, proximity, etc.) of
shift from a more mechanistic, analytical view, to a more sys- spatial units (Bock etal. 2005).
temic one. This includes a change in perspective from a focus Since we are looking at latent phenomena that involve spa-
on objects to relationships, from (quantitative) measuring to tial (and temporal) variability, we need spatialized indicators
(qualitative) mapping (Capra and Luisi 2014). Undoubtedly, presumed to represent the components of such phenomena in
the ubiquity in spatial data repositories (both public and pri- a spatially explicit way. The number of such indicators varies
vate, cf. big data) and a cross-cultural technology prolifera- depending on the phenomenon studied (Table 22.1). Ecosystem
tion (e.g., through smart phones) has recently prepared the integrity may be simply approximated by a vegetation index
ground for spatializing all kinds of societal or physical phe- relating the red (R) and near-infrared (NIR) bands of a remotely
nomena. But there is a gap between the factual capacity of spa- sensed image for indicating tree species diversity and vital-
tial analysis and decision support tools and their actual usage. ity within a deciduous forest. But even if standardization and
Usually, socioeconomic indicators are integrated on pre- image calibration is applied and the results are technically
defined administrative or regular grid cells (Nardo etal. 2005); robust, most probably, many aspects are missing in this strat-
the full potential of geographical synthesis is underexploited egy, as many other aspects may contribute to the integrity of the
(Lang etal. 2014a). While this is often due to conventional or forest ecosystem, like groundwater regime, timber usage and
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 461
other anthropogenic impacts, climatic conditions, as so on. As 3. to minimize the effects of outliers or inaccuracies in the
pointed out earlier using the example of societal vulnerability data and to facilitate the visualization and interpretation
to flood hazards, multidimensional problems require concep- of information in maps
tual models to operationalize them, a process that is often expert 4. to limit the sensitivity due to data fidelity by aggregating
based (Kienberger etal. 2013a). For pragmatic and performance the original units (e.g., pixels) into larger zones (Blaschke
reasons, the number of indicators should not be too high either, and Strobl 2001, Wise etal. 2001)
so we may aim at a commensurate number of indicators to best
Regionalization is based on the principle of spatial autocor-
represent the phenomenon in a methodologically sound way
relation, assuming that neighboring areas tend to have similar
(Moldan and Dahl 2007).
properties or uniform behavior (Tobler 1970). Region-building
We try to approximate as much as possible the phenomenon
assumes that such uniform behavior exists as long as transi-
under concern with an integrated set of proxies, to statistically
tions occur, leading to a different behavior along a certain
and heuristically turn its latency into aggregates exhibiting gra-
gradient or boundary. While, due to the principle of spatial
dients and other spatial behavior that we are able to map. In
autocorrelation, the internal structure of continuous spatial
adding a spatial component to the integration of indicators, we
data enables an empirical construction of regions, there is no
arrive at the level of spatial composite indicators (Hagenlocher
a priori fixed set of regions to be built. As with classification
et al. 2014b) or metaindicators (Lang et al. 2008). Lang (2008)
in general, the aggregation of data is to some degree arbitrary
uses the term conditioned information to underline that this
(Johnston 1968), and the areal units to be built by spatial aggre-
process entails the creation of new geographies as a flexible, yet
gation can be done at different scales and in different (though
statistically robust and (user) validated unitization of space. The
equally plausible) ways (Wise et al. 2001). As with statistical
aim is to map complex phenomena to better understand their
data in general, the problem of aggregation effects in partic-
spatial variability and dynamics over time.
ular applies to spatial studies (modifiable areal unit problem
[MAUP]). As Openshaw (1984) points out there are no stan-
22.2 Domain-Specific Regionalization dards or international conventions for spatial aggregation and
it is subject to the whims and fancies [] of whoever is doing
Integrated geons are delineated in a semiautomated way incor- the aggregation (p. 3). We will return to this problem and see
porating expert knowledge by adhering to statistical robustness how we (partly) cope with it.
and scale optimization. Thus, we are able to map and monitor
units of uniform response to the phenomenon under concern,
commensurate to the scale of policy intervention measures and 22.2.2 integrated Geons
stable in their aggregation.
Experiences in analyzing high-fidelity, multispectral imag-
ery using geographic object-based image analysis, (GE-)
22.2.1 Principles of Regionalization OBIA (Hay and Castilla 2008, Blaschke 2010), can be trans-
ferred to address complex spatial phenomena. The transfer of
Within spatial science, the term regionalization implies both a
OBIA techniques for the analysis of nonimage data has been
topdown (i.e., disaggregating) and a bottomup (i.e., aggregat-
discussed for univariate phenomena such as a digital eleva-
ing) notion. Disaggregating a larger whole into smaller regions
tion model (DEM) by Drgut and Eisank (2011). Kienberger
is often associated with political will, for which the European
et al. (2009) used object-based analysis (OBA) to regionalize
Statistical Office uses the term normative regions (Eurostat 2006).
an n-dimensional indicator space for assessing socioeconomic
Scientific regionalization usually follows a more bottomup strat-
vulnerability to flood hazards. Nonimage data in this context
egy applying routines to group neighboring subunits, that is,
are gridded geospatial data provided by interpolation of point
small geographical units, pixels, raster cells, etc., into larger ones
samples over space, or based on spatially disaggregated indica-
(Figure 22.2). While some regionalization techniques do imply a
tors. Principles and conceptual findings from social sciences
topdown component (e.g., the split-and-merge algorithm), most
with technical achievements from OBIA are bridged, to gener-
regionalization methods are implemented in a bottomup fashion,
ate integrated geons (Lang etal. 2014b). Next to the challenge
performing any kind of spatially constrained aggregation method
of incorporating expert knowledge and adhering to statistical
(Duque and Suriach 2006). Such aggregation is done in a way that
robustness and scale optimization, an issue remains in assign-
the resulting analytical regions are conveniently related to the
ing nominal categories or even labels to the generated units
phenomena under examination (Duque and Suriach 2006, p. 2).
and characterize their evolution over time.
The main objectives of regionalization (Berry 1967) are sum-
Just like image segmentation is used to represent tangible real-
marized by Lang etal. (2014b):
world objects, latent phenomena can be represented by regional-
1. to organize, visualize, and synthesize the information izing multiple indicators (Lang etal. 2014a). The indicators may
contained in multivariate spatial data (Long etal. 2010) be mapped as singular layers in a GIS and evaluated separately.
2. to reduce data dimensionality (Ng and Han 2002) while A common strategy is to approach such complex properties by
minimizing information loss (Nardo etal. 2005) aggregating variables/indicators toward an index or composite
2016 by Taylor & Francis Group, LLC
462 Remotely Sensed Data Characterization, Classification, and Accuracies
e.g., ecosystem integrity
Similarity in location
Indicator 2
Similarity in indicator space spatially contiguous
Pixelxyz(t) = (Vi1, Vi2, Vi3)
Indicator 1
e.g., housing
3
or on
c at cati
di u n-Dimensional indicator[feature] space
In , ed
.
e.g
FIg u r e 22.2 Regionalizationsimilarity in attribute and real space. Next to similarity in feature space, the classification of spatial data is con-
trolled by location. Spatially constraint classification is called regionalization.
indicator, a procedure that works on multidimensional attribute Conceptual
spaces (Nardo etal. 2005). However, such procedures neglect the level
true spatial distribution of such properties when using a priori Abstract
geographies (such as neighborhood or district boundaries) that
obscure or bias the actual spatial distribution.
Our strategy is to extend multivariate clustering from attribu- Geons
tive to real space, thereby reaching a flexible portioning of real
Regionalization
space, (re-)composing new geographies. Just like multispectral
Indexing
imagery can be clustered into homogenous objects by applying 4 Spatial composite
Aggregation indicator
segmentation techniques, we represent complex spatial phe-
nomena by regionalizing multidimensional attribute/indicator
spaces. This is different to overlaying or intersecting of geospa-
tial layers in treating the spatial dimension as a constraint, but
flexible in terms of scale. Regionalization techniques that utilize
3 Indicators
rule-bases with flexible strategies for geodata integration under Concrete (proxies)
consideration of and expert knowledge follow the methodologi-
cal framework of OBIA. Nonspatial Spatial explicit
By applying segmentation routines, we go one step further Spatialization
toward spatial explicitness (see Figure 22.3). Classical aggrega- FIg ur e 22.3 Regionalization vs. indexing. Note that this is a contin-
tion that uses a raster overlay with a local operator performed uation of Figure 22.1a. Indexing, for example, aggregating on a per-cell
on regular grid cells or given reference units does not change basis does not generate new spatial information. Instead, regionaliza-
the level of spatial explicitness, while regionalization does. Here, tion provides additional spatial-explicit information by its generation
spatial explicitness is not meant in terms of more spatial detail, of (new) spatial units.
but spatial synthesis. Regionalization does not increase the spa-
tial detail, but introduces additional spatial characteristics in the or social vulnerability to malaria (Kienberger and Hagenlocher
aggregation process, and thereby spatializes it. 2014). Currently available methods to model integrated geons
An initial workflow to delineate geons was developed by build on four major stages (see Table 22.2). These stages serve as
Kienberger etal. (2009) and applied to assess place-based vul- a framework and can be adapted accordingly.
nerability in a climate change context. Since then, the approach In a first stage, the identification and definition of associated
has been refined and transferred to various application domains concepts of the phenomena under investigation is carried out.
(see Section 22.3), including, for example, hot spot analysis of Subsequently this is followed in a second stage by the identifi-
cumulative climate change impacts (Hagenlocher etal. 2014b), cation of indicators and data collection and related preprocess-
landscape sensitivity in the context of environmental impact ing. In the third and core stage, the multidimensional indicator
assessment for infrastructure projects (Pernkopf and Lang 2011), framework is integrated through regionalization, which can
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 463
TABLe 2 2.2 Stages of the Geon Workflow and Relevant Aspects to Be Considered
# Stage #.n Key Steps Involved Issues to Be Considered
1 Definition of conceptual framework 1.1 Conceptual operationalization Soundness
Practicability
Expert opinion
Scientific communities/schools, etc.
1.2 Domains and dimensions
1.3 Identification of indicators Literature/expert knowledge
Selection criteria: salience, credibility, legitimacy
Data availability
2 Data preprocessing 2.1 Data acquisition Scale: global, national, etc.
Data availability: public domain, authority mandate, commercial
Resolution: admin units, grid
2.2 Preprocessing Resampling to continuous grids
Data transformation
Data imputation and outlier treatment
Normalization
2.3 Multivariate analysis Multicollinearity analysis
PCA
2.4 Sensitivity analysis Evaluating the influence of indicator choice, normalization method,
aggregating method, and weighing
3 Regionalization 3.1 Indicator weighting Equal weights
Expert weighting
Statistical weights
3.2 Unit delineation (Multi)scale assessment and segmentation
Composite index
4 Visualization/geonalytics 4.1 Mapping Legend and intervals
4.2 Explorative analysis Shape and size variation, diversity
4.3 Monitoring E.g., using WebGIS solutions
Source: Kienberger, S. etal., Nat. Hazards Earth Syst. Sci., 9, 767, 2009; Kienberger, S. and Hagenlocher, M., International Journal of Health Geographics,
13, 29, 2014; Lang et al., Cartography and Geographic Information Science 41(3), 214, 2014b.
be built on different weightings of the indicators. This includes Once a preliminary indicator set is identified, it is envisaged
sensitivity analysis. In a final and fourth stage, the results are in the second stage to establish a quantitative, valid, and repre-
appropriately communicated and visualized to the intended sentative indicator framework. This includes the collection and
user group. quality assurance of different datasets to populate the indicators,
In the following, these four steps are described in detail: latent, as well as different statistical preprocessing routines. Such pre-
multidimensional phenomena are concept driven and need to be processing routines are well known in the composite indicator
defined accordingly. The first stage has a strong link to different community (see, e.g., Nardo etal. 2005), and can be applied to
theories originating from various disciplines such as sociology, derive a statistically sound indicator framework. This includes
ecology, or geographyto name a few. Such concepts and frame- the transformation of the indicators to continuous grids, the
works may also change over time and within different schools transformation from absolute into relative measures, the iden-
of thought. Therefore, it is important to define an appropriate tification and treatment of outliers, missing data and multicol-
framework depending on the context of the study. Associated linearities in the data, as well as the application of normalization
to the selection of appropriate theories and useful frameworks, techniques to render the different indicator values comparable
a first set of possible indicators has to be identified. Irrespective (e.g., minmax normalization, z-score standardization, etc.).
of the availability of data, one should consider how the differ- Alternatively, value function approaches (Beinat 1997) can be
ent dimensions of the chosen theoretical framework could be applied, where values are normalized on expert or empirically
characterized and measured. It is likewise important to have an defined relationships.
in-depth understanding of the underlying framework that com- Once a final indicator framework is set up, the grid-based indi-
prises causal relationships and logical dependencies between the cator data are integrated through regionalization to derive homog-
single domains and indicators relevant for operationalizing the enous regions of the investigated phenomena. Geons are delineated
phenomena. using the multiresolution segmentation algorithm of Baatz and
2016 by Taylor & Francis Group, LLC
464 Remotely Sensed Data Characterization, Classification, and Accuracies
Schpe (2000) as implemented in the eCognition software envi- To optimize the degree of homogeneity between two neighbor-
ronment (Trimble Geospatial). This is a region-based, local mutual ing pixels or objects, the specific heterogeneity is minimized at
best fitting approach that merges image segments according to the every merge. The current degree of fitting (hdiff ) is characterized
gradient of degree of fitting (Baatz and Schpe 2000). It allows for by the change in heterogeneity (Equation 22.3):
controlling two complementary criteria of similarity of neighbor-
ing segments: likeness in color and form. The latter refers to the SD1 + SD2
hdff = hmin (22.3)
influence of space in the regionalization process. Spatial objects 2
can be generated that are rather compact or have rather smooth
outlines. A scale-factor enables user-driven control of appropriate By weighting heterogeneity with object size, the requirement
scaled representations. Here we shortly reflect the algorithm as dis- of producing objects of similar size can be accomplished. Form
cussed earlier by Kienberger etal. (2009): the difference between homogeneity is considered by relating actual boundary length
adjacent objects (Kienberger etal. 2009) is expressed by the spec- (perimeter) to the perimeter of the most compact form of the
tral distance (SD) of two pixels or objects p1, p2 in a feature space: same size. This ideal form is a circle, the deviation of which can
be expressed by the shape index:
n
SD = (p p ) ,
d =1
1 2
2
(22.1)
SHP =
p
, (22.4)
2 *s
or noted as vector difference for a three-dimensional feature where
space as p equals the perimeter
s equals the size of an object
SD = (v1 v 2 );
Identifying an appropriate scale factor has been subject to much
d11 d21 debate, the common approach is to use trial and error. To over-
come this subjectivity in object generation, a routine suggests
with v1 = d12 and v 2 = d22 . (22.2)
d13 d23 initial parameters for the multiresolution segmentation (Drgut
et al. 2014). The so-called estimation of scale parameter (ESP)
FIg u r e 2 2.4 Graphical user interface of the ESP II tool.
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 465
tool is implemented as a generic tool for eCognition software the output can be apportioned, qualitatively or quantitatively, to
(Figure 22.4). different sources of variation in the assumptions. A combination
The tool utilizes local variance (LV) graphs (Woodcock and of both practices, uncertainty and sensitivity analysis, provides
Strahler 1987) to detect scale domains in geospatial data (Dragut insights into the robustness of the modeling results. Testing the
et al. 2010). It segments continuous datasets iteratively apply- robustness of geons confronts us with the challenge of changing
ing a multiresolution segmentation algorithm, from smaller to geometries in combination with changing index values.
coarser scale in a small, constant increment. The mean LV value The generated units, though considered plausible, do not
is computed for each image level that was created as the ratio directly correspond with observable real-world objects. The valid-
between the sum of the LVs for each layer (LV1 LVn) and the ity of the spatial extent and the precategorical nature of such
number of layers (n) used in the image segmentation. When a units can be indirectly assessed by collecting evidence data on the
scale level records an LV value that is equal to or lower than the prevalence of the phenomenons impact. Examples are damage
previous value, the iteration ends, and the objects segmented in scenarios or loss estimation scenarios in the case of vulnerability
the previous level are retained (Drgut etal. 2014). to natural hazards or, in the case of disease vulnerability, to assess
According to different indicator weights, the algorithm also the spatial distribution of the disease combined with patient data
considers their relative contribution to the resulting composite on socioeconomic background and demographic setting.
value. To assign a quantitative value to each of the identified In a final stage, integrated geons are visualized. To avoid the
units or regions, a (weighted) vector magnitude is calculated, categorization and classification of different class ranges, a con-
which measures the vector distance of each unit within its tinuous color scheme is applied. Each unit is visualized based
n-dimensional indicator (feature) space (see Equation 22.1). on its composite indicator value, and can be visually compared
Alternative aggregation approaches, such as weighted sum or to the other regions. To allow an interactive exploration of the
geometric mean, are also available, while the vector magnitude different regions, and its underlying indicators, integrated geons
emphasizes variability and underlines changes context. Finally, are best visualized through interactive, web-based mapping
the values of the resulting spatial composite indicator are nor- tools, for example, Kienberger etal. (2013b).
malized to a defined classification scheme (e.g., from zero to one)
to ease its interpretation. Additionally for each unit, the con-
22.2.3 Monitoring Geons
tribution of the underlying indicators to each integrated geon
through different metrics (mean values, standard deviation, etc.) Once geons are generated in agreement to the policy realm, they
can be assessed. can be used to monitor the spatiotemporal dynamics of the
So far, we discussed how geons are generated by integrating considered (latent) phenomenon over time.
a set of indicators and dimensions. All of the modeling stages Understanding the status quo of a complex spatial phenome-
discussed introduce uncertainty in the results, since there is non is one thing, observing its behavior over time, via the dynam-
a range of plausible alternatives at each stage. If another con- ics of geons as space-time entities (Peuquet 2002), is another.
ceptual framework with a different set of indicators was used, While the decision for particular intervention may be based on
or the experts had weighted the relative importance differ- analyzing the situation as is, assessing the comparison to one or
ently, other units, and thus policy messages, would have been several previous ones helps well understand its relative meaning.
generated. The impact of the normalization method and the Also, the impact of such intervention requires a repetitive analy-
regionalization parameters should not be neglected either. For sis of altering conditions. A multitemporal geon set may provide
illustrative purposes, we discuss the impact of the final choice support in the reassessment and correction of intervention mea-
of indicators and indicator weights on the modeling outputs in sures. The delineation of geons enables comparable updates as a
Section 22.3. basic requirement for regular and consistent monitoring, here:
We recommend a combination of data-driven (statistical) against a specific policy background (Lang etal. 2008). Working
and normative (expert-based) approaches to generate concept- in a spatial-explicit object-based environment allows for char-
related fiat objects of this kind and to deliver a consolidated out- acterizing the shapes of single geons on individual level, as well
put of domain-relevance. This especially applies to the choice of as their spatial arrangement and distribution on collective level.
indicators and the setting of weights (Decancq and Lugo 2012). The spatial variability of the phenomenon under concern may be
In all cases, the modeling of integrated geons should be comple- a key to better understand the overall systemic behavior of it. In
mented with robustness tests to determine to what degree the addition, assessing the spatiotemporal dynamics and the evolu-
results are driven by specific indicators or weights. Different tion of a geon set over time allows an evaluation of the underly-
approaches to uncertainty and sensitivity analysis have been ing trend. Here, as in any change analysis, we need to separate
proposed by the composite indicators community (Nardo etal. noise or slight modifications from real changes. Dealing with
2005, Saisana et al. 2005, Saltelli et al. 2008). Whereas uncer- aggregated units rather than single cells helps in overcoming the
tainty analysis quantifies the overall uncertainty in the output problem of data fidelity. Originating from OBIA, the concept
as a result of the uncertainties in the model input, sensitivity called object fate analysis (OFA) (Schpfer etal. 2008, Hernando
analysis can be used to evaluate how changes in each individual et al. 2012) can be used to characterize the behavior of geons
input parameter affect the final output and how the variation in over time, and their possible transitions to other stages. When
2016 by Taylor & Francis Group, LLC
466 Remotely Sensed Data Characterization, Classification, and Accuracies
State 1
of geon
1 1
Fair state range
(0.40.9)
0.5
Intervention line (<0.4)
0 2
Fair ... Critical! Fair... 2
(a) Time (b)
FIg ur e 22.5 (a) Monitoring integrated geons supports a scenario of threshold-based intervention planning. Fair state range: natural dynamic of a
geon state through time. Green: geon in fair/good state, no action required, red: below fair-state line, action to be considered, dark red/dashed: inter-
vention required. (b) Two geons (representing vulnerability to floods) in different states below fair state. The dark red tone (1) would indicate required
measures. See Figure 22.7 for larger extent and additional context. (From Lang, S. etal., South-Eastern Eur. J. Earth Observ. Geomatics, 3(2s), 625, 2014.)
monitoring the state of geons over time, we may encounter a cer- While there are an infinite number of possible combinations or
tain threshold beyond which we would consider it favorable. In groupings, regionalization techniques aim to algorithmically or
other words, no policy intervention is required. But as soon as heuristically group neighboring units according to any kind of
its state deteriorates and drops beneath that threshold, certain similarity criterion. Still, merging neighboring pixels assigned to
countermeasures have to be taken. We may call this allowable the same category (class) may depend on the classification rou-
range the fair-state range (see Figure 22.5). tine and/or a given size constraint applied. But there is an inher-
ent reason why these units are generated and they bear any kind
22.2.2.4 MAUP and Scale Dependency of homogenous internal behavior, in particular when compared
Until today, many assessments of latent spatial phenomena are to neighboring units (basic principle of delineation or demarca-
carried out using administrative boundaries or continuous tion). In the case of integrated geons, the mapping of a spatial
grids as the final reporting unit. As policies and interventions composite indicator is embedded in a conceptual framework;
are often spatially targeted at the administrative level, results its methodological realization, provided the basic principles of
reported on administrative units match the scale of current multivariate data analysis are obeyed, does provide intersubjec-
policy interventions. Policy and decision makers on the ground tively relevant units. Boundaries are generated where the latent
have a long tradition in using them for analysis, benchmark- phenomenon under concern changes its behavior. In contrast to
ing, reporting, and planning purposes. However, as pointed administrative units and grids, geons provide a representation
out by Openshaw (1984), the results and their interpretation of the real spatial distribution of the respective phenomenon
have to be treated with caution as they are biased by the fact (Hagenlocher et al. 2014a). Although geons do not necessarily
that these units are modifiable, artificial areal units. This holds match the scale of current policy interventions, thus calling for a
also true for continuous grids, or pixels, whichlike adminis- paradigm shift in intervention planning, they provide a powerful
trative unitsdo not have an intrinsic geographical meaning means for planning as their internal heterogeneity is minimized
(Hagenlocher etal. 2014a). (Lang etal. 2014b).
The MAUP concept has been developed for addressing the Integrated geons can be delineated at various scale levels, rang-
drawbacks of mapping any measurement, model outcome, or ing from local to national, continental or global. It is challenging
statistical value on existing geographies (units, e.g., census or to find a commensurate scale for finding the respective gradient
other administrative units). In relation to that, ecological fal- to represent the phenomenon best (Marceau 1999). Next to the
lacy problematizes the fact that all individuals (or subunits) in nature of the phenomenon, this is also influenced by the avail-
a given spatial unit are treated collectively, how close (spatially ability of data that are appropriate to be integrated in a regional-
or property wise) they may be to the reported phenomenon ization approach, the level of spatial detail that is required, and
(Openshaw 1984). MAUP can also be linked to the observation the scope of the analysis (Hagenlocher etal. 2014b). The three dif-
that smaller units (e.g., pixels in an image, cells in a raster repre- ferent kinds of scale mentioned by Wu and Li (2006) are likewise
sentation, etc.) can be aggregated to larger units quite arbitrarily. important when modeling integrated level geons: the intrinsic
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 467
scale, the modeling/observational scale, and the policy scale. As by value ranges of the computed metaindicator As conceptual
we postulate that integrated level geons are policy relevant, we links are still missing, currently, we cannot move from ordinal
have to consider the relevant scale level at which policies occur. scaled ranks (low, medium, high) to nominal categories, just as
For instance, if a district is responsible for disaster risk reduc- we transform continuous elevation data into landform classes
tion activities, decision makers require relevant information on (Drgut and Eisank 2011).
(sub-)district level in a meaningful and disaggregated manner. We argue that the farther we move from bona fide objects, the
However, this has to be in line with the intrinsic scale level of trickierand more domain subjectivevalidation gets. Lang
the certain phenomena investigated (e.g., where flood occurs; et al. (2014b) have characterized integrated geons as concept-
vulnerability at a local level is characterized different than one at related fiat entities, which cannot be assessed by established
the global one). The intrinsic scale level may be difficult to iden- accuracy assessment methods. The term object validity (Lang
tify in an objective manner as it can also rely on our perception; etal. 2010) reflects the limited power of binary assessments in
however, certain assumptions need to be made. This scale level judging thematic accuracy of a given object label. Object validity
needs to bridge toward the observational and modeling scale. It is should ensure a purpose-oriented judgment whether the prod-
essential that the observational scale (e.g., resolution of raw data) uct meets the users demand.
and the final modeling scale match, and are valid to establish the
bridge between the intrinsic and final policy scale level. These
22.2.5 in Depth: Systemic Areal Units
considerations need to be taken into account, when identifying
indicators (do they reflect the intrinsic sale at the district level Systems thinking has been widely referred to in spatial science
appropriately?), the associated input data and modeling domain literature as dealing with scaled representations of continuous
(e.g., resolution of raster-based data), and finally, the policy scale data, for example, in multiscale image analysis (Hay etal. 2001)
level, for which the results will provide decision makers with rel- and in addressing areal phenomena over several scales, such
evant information. Therefore, such a consistency among the dif- as the scaling ladder concept (Wu 1999) in landscape ecology.
ferent kinds of scales needs to be maintained in order to provide Originally conceptualized in a nonspatial or not explicitly spa-
valid results, while developers of spatial composite indicators tial context, systems theory (Bertalanffy 1969) deals with the
should be aware of the implications of spatial scales for analyz- hierarchical organizations of concrete systems, which on each
ing and monitoring latent phenomena (Hagenlocher etal. 2014a). level exhibit systemic, that is, emergent properties (in fact one
may consider it vice versa, a level is constituted where emergent
properties become obvious). Koestler (1967) coined the term
22.2.4 How to Validate Geons?
holon to underline the nested behavior of systemic units. The
User validation is meant in the sense of object validity (Lang geon concept as a holistic regionalization approach intends to
et al. 2010) that comprises both policy relevance and expert- abstract from a specific application domain by explicitly address-
proven functional relevance. From a data model point of view, ing the level of the generated units while not limiting our view
the generated geons are areal (i.e., polygonal) objects. The vec- on a specific thematic topic or application. The common denom-
tor data model suggests crispness of the generated boundaries inator is to generate units that are not a priori, but demanded
and soundness of the assigned label. If units are not a priori, can and thusonce providedof immanent importance for policy-
they be objective, that is, intersubjectively acceptable? Johnston related action. Geons show uniform response regarding the
stated in 1968 that all approaches to classification are actu- spatial phenomenon under concern, and are (ideally) expert
ally subjective, cited in Hancock (1993), so the question arises: validated regarding practical usability and relevance. They are
do such units have some validity beyond a narrow thematic carriers of integrated spatial information and, we hypothesize,
domain? When delineating integrated geons, we aggregate val- exhibit emergent properties of systemic areal units.
ues of the underlying indicators and their weighted contribution Terminologically, geo-on suggests synonymy with geo-atom
in a multidimensional variable space. At the same time, we deal (Goodchild etal. 2007) or even more radical, the Elementary_
with spatial constraints that check the similarity of neighbor- geoParticle (Voudouris 2010). But geons do not claim atomic-
ing cells. After that we visualize the resulting composite values: ity (Masolo and Vieu 1999) as something undividable. In fact,
depending on the number of intervals (decile, quantile, etc.), the a geon is not the smallest possible unit in space and/or time as
geon-scape will appear differently, while the delineated units as topons or chronons (Couclelis 2010), but the smallest valid one
such remain. Particularly, when aiming at deriving hot spots in a certain context (as small as necessary, as big as possible)
of these complex phenomena, thresholds need to be defined, (Lang etal. 2014b). In generating an aggregate that has its own
which indicate whether or not an object will be marked as a hot meaning and stability, the composition of a geon will start with,
spot. Defining these thresholds adds a certain level of subjectiv- but go beyond, a purpose-driven exercise. In the composite indi-
ity to the final results, as a priori thresholds do not exist and cator community, it is widely acknowledged that integrating
must be defined by either making use of expert knowledge or single indicators leads to systemic effects (emergent properties)
a needs-driven approach (e.g., specific number of hot spots in (Nardo etal. 2005). In other words, the variance of a systemic
an area). A challenge is to assign nominal categories (labels) to whole is much lower than the sum of the variances of its com-
the generated units, which are at this stage only characterized ponents. However, raster cells (with composite values) that can
2016 by Taylor & Francis Group, LLC
468 Remotely Sensed Data Characterization, Classification, and Accuracies
represent an Elementary_geoParticle would not qualify as a aggregation); it is important to make the approach trans-
geon, due to its technically fixed spatial definition. A geon is a parent and to conduct a sensitivity analysis; for geons, this
spatial object with systemic stability features such as minimized is still an area under development.
inner variance and gradients toward the outside through vec- In some stages, subjective judgment has to be made, which
tor encoding. Regionalization implies some loss of information affects the result (selection of indicators, weighting of
(Hancock 1993), redundant detail that acts as noise in the con- indicators, regionalization parameter, etc.).
text under concern. Transferability across geographic/policy-scales (local to
In congruence with systems thinking (Laszlo 1972), we argue regional) as well as across domains (disaster risk reduc-
that spatial systemic units also bear a dual character in terms tion, climate change adaptation, public health, strategic
of self-assertive tendencies (whole-ness) and self-integrative planning), see Figure 22.6.
ones (part-ness) (Lang etal. 2004). This nested behavior can be The geon approach works with multidomain, multisource
applied to systemic areal units as well, as a geographical corre- datasets and independent from the number of indicators;
spondence to holons (Wu 1999). Geons that are composed from however, indicators and indicator weights must be chosen
spatial elementary units show such systemic properties in bear- carefully to best reflect the idiosyncrasies of the phenom-
ing certain functional qualities, which are crucial to be treated enon that is addressed.
as a conceptual whole. Geons show a uniform response to a cer- Validation of latent phenomena is tricky (how to validate
tain phenomenon to be approximated by applying spatial classi- the immeasurable?), there are no real observations to cali-
fication of an integrated set of spatialized indicators (Kienberger brate the model.
etal. 2009).
22.3 case Studies 22.3.1 Socioeconomic Vulnerability to Hazards
22.3.1.1 Latent Phenomenon Addressed
Here we present several examples how the geon approach was
utilized in different application domains including disaster risk Currently much emphasis is put on conceptualizing and map-
reduction, public health, climate change adaptation, and envi- ping vulnerabilities in the context of disaster risk reduction and
ronmental impact assessment. climate change adaptation. Here we illustrate an example of
The case studies section will use practical examples to illus- how to perform an integrated, holistic assessment of social, eco-
trate the following summary statements: nomic, and environmental vulnerability to flood hazards. The
study has been performed in the Salzach river basin, Austria,
Conceptual frameworks build the basis for operational- and discussed in greater detail by Kienberger et al. (2014) in
izing the latent phenomenon addressed; they guide the the context of flood hazards at the catchment scale level for the
assessment and foster the reproducibility of results as well social, economic, and environmental domains. The Salzach
as the transferability of the approach to different scales or catchment is characterized through its alpine upstream area
regions. and highly dynamic river valleys with the city of Salzburg in the
Several plausible methods exist for constructing spatial downstream area. Recent severe floods have occurred in 2002,
composite indicators (e.g., normalization, weighting, 2005, and 2013, affecting large parts of the river catchment. The
Data rich environment
Landscape process Socioeconomic
dimension dimension
Data scarce environment
FIg u r e 2 2.6 Geon-scapesexemplary illustration considering different policy domains and scales, dimensions and data conditions.
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 469
majority of the population lives in the city of Salzburg (approx. vulnerability have been delineated that share a commonality with
150,000 inhabitants) and its urban fringe. The area is highly regard to their underlying indicator values as well as a spatial con-
dynamic with regard to its economic and settlement develop- straint. A vulnerability index was calculated through a weighted
ments, putting high pressure on adequate spatial planning poli- vector magnitude (the length of the vector for each region) in the
cies in an area with a limited amount of suitable land. multidimensional indicator space. Final index values were then
To achieve the objective to protect people against floods, normalized within a scale range of 01, and visualized by applying
Austria is currently implementing flood hazard mappings, a centile classification with exponential kernel algorithm.
improvement of river regulations, and technical flood protec-
tion measures along with sophisticated early warning systems. 22.3.1.3 Results
Despite these measures, floods are not prevented per se. They The results allow the identification of spatially explicit regions
allow for enough time to prepare, warn, and evacuate people and with different levels of vulnerability (hot spots) independent
minimize economic damage and loss of human lives based on from administrative boundaries. Furthermore, the results pro-
specific flood protection activities. vide decision makers with place-specific options for targeting
disaster risk reduction interventions that aim to reduce vulner-
22.3.1.2 indicators and Datasets ability and ultimately the risk of impacts from floods. The most
The study aimed to model homogenous vulnerability regions for vulnerable regions in the three dimensions are located along the
the social economic and environmental dimensions. As such, Salzach river and its tributaries. However, the most vulnerable
homogenous regions in terms of their degree as well as their region in all the dimensions is the city of Salzburg and its sur-
inherent characteristic of vulnerability should facilitate the roundings. These results are due to the density of the built-up
improved identification of place-specific intervention measures area, a big concentration of historic buildings, and widespread
for disaster risk reduction. infrastructure. Therefore, the highest vulnerability degree
An expert-based approach was selected by considering the fact with regard to the social dimension is concentrated in the city
that vulnerability is not directly measurable, due to its complex of Salzburg as one of the largest settlements located along the
dimension and social construction. In order to model the spatial Salzach river. The economic dimension has its hot spot also in
distribution of a complex phenomenon, established methodolo- this urban region due to the presence of employment sources,
gies such as multicriteria decision analysis, Delphi exercises, and and the city is an important node in the transport network in the
regionalization approaches to derive homogenous vulnerability country. The environmental dimension shows the highest degree
regions were integrated. At the initial stage of the workflow, the of vulnerability in the stretch of the Salzach river from the city to
concept of vulnerability is addressed, which in this case builds on the north, associated to river fragmentation (Figure22.7).
the adapted MOVE risk and vulnerability framework. An essen-
tial step is the identification of indicators and available datasets, 22.3.2 Social Vulnerability to Malaria
the weighting of the different indicators, and its aggregation to
subindices. As the final list of appropriate indicators identified 22.3.2.1 Latent Phenomenon Addressed
from expert choice, literature, and community perspective, five Despite the global recession in malaria cases over the past
key experts were asked to weigh the indicators through budget decades, malaria remains the most prevalent vector-borne dis-
allocation methods. Data availability has been an advantage in ease (VBD). Caused by the bite of an infected Anopheles mos-
the Salzach catchment, as important key datasets are available in quito, malaria resulted in approximately 207 million infections
the spatial data infrastructure established at the Government of and has causes approximately 627,000 malaria-attributable
Salzburg, as well as the availability of detailed raster-based census deaths in 2012 (WHO 2013), primarily among African children.
data for Austria, which specifically helped to apply the regional- Thus, in addition to modeling transmission potentials, it is of
ization approach. The data used range from infrastructure data to utmost importance for the planning of (preventive) interven-
different socioeconomic parameters such as the size of companies, tions to assess prevailing levels of malaria vulnerability in a spa-
means of subsistence, age, and workforce in economy sectors, ori- tially explicit manner (Hagenlocher et al. 2014a). Based on an
gin, and education level of the population. They originated from adapted version of the MOVE risk and vulnerability framework,
the census survey in 2001 and are not only provided on the basis relative levels of social vulnerability to malaria were modeled
of different administrative units but also for a standardized grid for the East African Community (EAC) region using the geon
with different available spatial resolutions. To allow comparabil- approach (Kienberger and Hagenlocher 2014).
ity, data were normalized through linear minmax normaliza-
tion within an 8 bit scale range (values with 0255). The modeling 22.3.2.2 indicators and Datasets
scale level was based on the standardized grid with a regular cell Based on the outcomes of a systematic review of literature, the con-
size of 1000 m. In the next step, the modeling of homogenous sultation of domain experts, and data availability, (1) a preliminary
vulnerability regions for each dimension (as integrated geons) was set of 15 biological and disease-related (e.g., immunity, age, preg-
implemented through the application of a multiresolution segmen- nancy, etc.), (2) socioeconomic (e.g., socioeconomic status, poverty,
tation/regionalization approach in combination with a smooth- nutritional status, education, etc.), as well as (3) accessibility-related
ing algorithm (polynomial approximation). Therefore, regions of indicators (e.g., access to health services, etc.) were identified. Data
2016 by Taylor & Francis Group, LLC
470 Remotely Sensed Data Characterization, Classification, and Accuracies
Access to
13E Access to Access to
roads; 4.41
Social vulnerability first hospitals;
index responders; Age 0.72
Centile classification 0.15 structure;
10.15 Employees
High [1] Level of in economy
histogram of in dex values (gray) education; sector; 9.83
Centile class breaks (blue) and
23.64
Population
origin;
Means of 28.81
subsistence;
1 22.14
Low [0] 1 Early
warning
system;
0 2.5 5 10
kilometers Age Employees
2 structure; in economy
0.92 sector; 6.20
Access to
hospitals;
18.40
Access to Population
2 Access to roads; 6.55 origin;
first 26.62
responders;
0.99
Level of
Means of
education; Early
subsistence;
24.35 warning
15.83
system;
0.14
FIg u r e 22.7 Geon-scape showing social vulnerability to flood hazards in the Salzach river catchment. Geons score from low to high in range
between 0 and 1. See legend. Single geons can be explored in terms of the contributing indicators and their respective contribution to the overall
score (see pie charts 1 and 2).
for these indicators was acquired from multiple sources, includ- index. The vulnerability index revealed a high overall robust-
ing remote sensing data (e.g., land use/land cover information, ness with regard to the final choice of vulnerability indicators,
etc.), survey data (e.g., Demographic and Health Survey data), with the exception of the immunity indicator that has a marked
and volunteered geographic information (VGI) as provided by impact on the composite vulnerability index (Kienberger and
OpenStreetMap. After statistical preprocessing, as described in Hagenlocher 2014).
Section 22.2, one of the 15 indicators was omitted to reduce exist- Figure 22.10 compares the modeling outputs based on expert
ing multicollinearities in the data. weights (Figure 22.10, panel 1) and equal weights (Figure 22.10,
panel 2), revealing several interesting findings regarding the
22.3.2.3 Results geometry of the delineated integrated geons and the vulnerabil-
Figure 22.8 shows the spatial distribution of social vulnerability to ity index. First, the relevance of the immunity indicator, which
malaria for the EAC region. In the map, areas of high vulnerability was ranked one of the most important indicators by the expert,
are displayed in red, while areas of low vulnerability are displayed is clearly visible when comparing both maps. The expert-based
in blue, indicating high levels of malaria vulnerability in the high- map clearly shows that vulnerability is high in highland areas
land areas where immunity of the population is currently low. where immunity to malaria is generally low. As the impact of the
A sensitivity analysis was carried out to assess the impact immunity indicator is lowered when assigning equal weights to all
of (1) the individual indicators on the vulnerability index and indicators, the marked impact of immunity on the vulnerability
(2)the choice of the weighting scheme (expert weights vs. equal index, and thus the clear distinction between high levels of vul-
weights) on the size and shape of the geons as well as on the final nerability in the highlands and levels of vulnerability in the low-
vulnerability index for each geon. The influence of the single lands, becomes less distinct when using equal weights. Second, it
indicators on the vulnerability index was evaluated by discard- becomes obvious that the geometry of geons (i.e., their size and
ing one indicator at a time, while keeping all other settings, shape) is clearly influenced by the weights assigned to the individ-
including the geometry of the geons, constant. Figure22.9 shows ual indicators. At the same time, it is interesting to see that both
the influence of the vulnerability indicators on the vulnerability approaches clearly demarcate urban centers that differ from
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 471
3000E 3500E 4000E their vicinity in terms of both socioeconomic and demographic
characteristics, such as Kampala, Uganda, or Nairobi, Kenya.
22.3.3 Landscape Sensitivity
22.3.3.1 Latent Phenomenon Addressed
000
Major infrastructure plans such as the construction of a new
road require an integrative assessment of their likely impacts
on the environment. In addition to environmental factors, also
social and economic factors should be considered to minimize
potentially negative effects. The degree of sensitivity of a par-
ticular landscape determines how it is able to accommodate
500S
developments. Landscape sensitivity varies both spatially and
temporally, and can only be gauged through use of indicators.
Pernkopf and Lang (2011) have presented a spatial composite
indicator that combines the different aspects of sensitivity in an
integrative approach instead of taking them separately, which
1000S
is a standard practice in environmental impact assessment. The
Social approach was tested in the Marchfeld region, an area of 1000km2
vulnerability
in the East of Austria, Europe, where a highway connection was
High [1]
planned between two major cities (Figure 22.11).
Low [0] 0 125 250 500
km
National boundaries 22.3.3.2 indicators and Datasets
FIgu r e 22.8 Social vulnerability to malaria in East Africa. Pie-
Based on a review of environmental reports for different devel-
charts show the contribution of the single vulnerability indicators for opment plans, a set of 21 indicators was selected to model the
each geon. (From Kienberger, S. and Hagenlocher, M., International sensitivity of the landscape. Natural as well as socioeconomic
Journal of Health Geographics, 13, 29, 2014.) factors were considered in three domains of sensitivity: (1) ability
1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
0.3 0.98 0.99 0.98 1.00
Modified composite indicatorreference index
0.95
0.1 0.90
Correlation coefficient
0.1 0.80
0.3 0.70
0.62
0.5 0.60
Minimum/maximum
Correlation coefficient
0.7 0.50
s ted
/d ing
# nter to
re ing
en
s o
ita to
Imght p
su on
De (%) er
ity
un )
tio nc
ad e t
i ee
m (%
on c
ce e
ur ave age
Tr ing f
SD iv
m2
)
(k t
s
ar o
om
ay
sp ce
h
n im
rs nfe
ity ic
ild nt
es ti
ra de
U le l
ro nc
t n sl
io ig
ls
be en
pr ula
re
ns nfl
D n
ho an
ch stu
ba l t
at /h
fw
las not
n
pe -i
a
< 2 op
ild om
ist
ist
pe
p
de Co
IV
uc ary
of
Po
#o
on pe
n
D
d
ch W
fH
r n di
ed d
of
n
#o
de ild
co
et
#
un Ch
Se
Discarded indicator
FIgu r e 22.9 Influence of the single vulnerability indicators on the vulnerability index. The higher the interquartile range (IQR), the higher the
impact of the respective indicator on the final index. (From Kienberger, S. and Hagenlocher, M., International Journal of Health Geographics, 13, 29, 2014.)
2016 by Taylor & Francis Group, LLC
472 Remotely Sensed Data Characterization, Classification, and Accuracies
3000E 3500E 4000E 3000E 3500E 4000E
500N
(1) (2)
500N
Kampala, N
N Uganda
000
000
Kampala,
Uganda
Mount Kenya
500S
500S
National
Park
Nairobi,
Kenya
1000S
Mount Kenya
Social
1000S
National
vulnerability Pak
High [1]
Low [0]
Nairobi,
0 125 250 500 Kenya 0 125 250 500
National boundaries km km
FIg u r e 22.10 Evaluating the impact of the choice of the weighting scheme (panel 1: expert weights; panel 2: equal weights) on the modeling
outputs.
Policy relevance Landscape sensitivity
3 pillars
SEA directive
2001/42/EC Coping capacity
Ecological significance
Assessment of the effects of Societal significance
certain plans and programs
on the environment
Aggregation Sensitivity units
Regionalisation techniques are
applied to derive SensUs in a
spatially explicit manner.
FIg u r e 2 2.11 Using regionalization techniques to model landscape sensitivity in an integrated way.
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 473
to resist to external disturbances, (2) ecological significance and The influence of single indicators on the sensitivity index is
(3) societal significance. Each of them consists of 510 indicators shown in Figure 22.12 (panel 2) using the example of agricultural
(e.g., land use, degree of landscape fragmentation, conservation land use. The intensity of land use is considered as one indicator
status, and population density), which are weighted by experts of the societal significance of the landscape. Discarding this indi-
according to their relative importance. cator has an effect on the resulting geon geometry as well as on
the final sensitivity index values. In most areas, modeling outputs
22.3.3.3 Results are relatively robust to changes in the indicator set, because agri-
Figure 22.12 (panel 1) shows the spatial variability of landscape cultural land use is only one of 21 equally weighted indicators.
sensitivity to the road infrastructure project in the Marchfeld Only areas with an intensive land use (wine-growing regions and
region. The resulting sensitivity units are discrete spatial units irrigated areas) show changes in the size and shape of the sensi-
with similar characteristics in terms of their sensitivity to the tivity units as well as different index values, which are typically
impact. The results provide aggregated information to deci- lower when the socioeconomic value of agriculture is not consid-
sion makers while preserving the detail provided by single ered in the sensitivity of the landscape to a road infrastructure
indicators. Some areas, even within protected landscapes, are project. In some areas, with an extensive agricultural land use but
more sensitive than other areas, which is indicated by geon other important sensitivity aspects, the index also increases when
boundaries that are independent from predefined administra- agricultural land use is not included in the indicator set.
tive units. It shall, therefore, be possible to indicate particular
landscape areas, which would be suitable or less sensitive for 22.3.4 climate change Susceptibility
the construction of a road and to avoid an undesirable devel- (cumulative cc impact)
opment. The results certainly cannot substitute an in-depth
analysis of the impact effects, but support a more integrated 22.3.4.1 Latent Phenomenon Addressed
and transparent decision-making in environmental impact The recently published IPCC WGII contribution to the
assessment. Fifth Assessment Report (AR5) highlights that the risk of
16300E 16400E 16500E 1700E 16300E 16400E 16500E 1700E
48300N
(1) (2)
48300N
N N
48200N
48200N
48100N
48100N
Landscape sensitivity Landscape sensitivity
High [1] High [1]
0 3 6 12 0 3 6 12
Low [0] km Low [0] km
16300E 16400E 16500E 16300E 16400E 16500E
FIg u r e 22.12 Evaluating the impact of individual indicators on the modeling outputs (panel 1: use of full indicators set; panel 2: agricultural
land use indicator was discarded) on the modeling outputs.
2016 by Taylor & Francis Group, LLC
474 Remotely Sensed Data Characterization, Classification, and Accuracies
150W 100W 50W 00 50E 100E 150E 200E
Areas most affected by changes in climate
250N
250N
Algeria Libya
N
200N
200N
Mauritania
Mali
Niger
150N
150N
Senegal
Chad
The
Gambia
Burkina Faso
Guinea- Guinea Benin
100N
100N
Bissau
Sierra d'lvoire
Togo
Cote Hotspot intensity Pie charts (statistics)
Leone Nigeria Hotspot intensity* Share of integrated indicators
Ghana per hotspot
Liberia High: [1.0]
Precipitation
Low: [0.0]
Cameroon Temperature
50N
50N
Cape Verde *Areas of significant
intensity (>0.65) have Drought
Atlantic Ocean
Equatorial been defined as
hotspots Flood
Prava Guinea
150W 100W 50W 00 50E 100E 150E 200E
FIg u r e 22.13 CCCI in the Sahel and western Africa based on the aggregation of a set of four climate-/hazard-related indicators (temperature,
precipitation, drought, and flooding). Hotspots are displayed in red color. (From Hagenlocher etal., 2014b.)
climate-related impacts results from the interaction of climate- NESDIS-STAR vegetation health index (VHI) dataset, which is
related hazards with the vulnerability of exposed human and based on measurements of the advanced very-high-resolution
natural systems (Field etal. 2014). Based on this concept, an inte- radiometer (AVHHR) onboard the NOAA satellite, while data
grative, spatial explicit assessment of cumulative climate change on flood frequency was acquired from the Dartmouth Flood
impact (CCCI) was carried out using the Sahel and Western Observatory (DFO). They provide an active archive of large flood
Africa as study region. The focus was placed on identifying, events that is updated based on remote sensing information (e.g.,
mapping, and evaluating hotspots of climate change impact to MODIS, Landsat, etc.) as well as various other sources (news,
provide conditioned information for targeted climate change governmental reports, etc.). In line with the IPCC definition of
adaptation measures (Hagenlocher 2013, Hagenlocher et al. climate, which was defined as a period of 30years, the observa-
2014b). tion period was set to the past 2436years (depending on data
availability).
22.3.4.2 indicators and Datasets
To provide information on the hazards component of the IPCC 22.3.4.3 Results
risk framework, a set of four climate-related datasets was iden- Following a statistical data preprocessing routine as described
tified in collaboration with domain experts from the United in Section 22.2, the four singular climate-related indicators were
Nations Environment Programme (UNEP), including long-term mapped and visualized. In addition to mapping singular trends
average seasonal (1) precipitation and (2) temperature trends, as in the four indicators, 1233 units showing homogeneous regions
well as frequency of extreme events, such as (3) drought occur- of CCCI were delineated using the geon concept. Based on the
rences, and (4) major flood events over the past decades. The statistical analysis of the resulting geon-scape, a set of 19 cli-
analysis was based on time-series of freely available continuous mate change hotspots was identified. These represent areas most
datasets, including remote sensing data. Data on the two essen- affected by changes in climate conditions (precipitation, temper-
tial climate variables, monthly mean temperature and precipita- ature) and related extreme events (drought and flooding) over
tion, was acquired from the Climatic Research Unit (CRU) using the past decades (Hagenlocher etal. 2014b). Next to the location
their time series (TS 3.0) datasets. These comprise 1224 monthly and size of the hotspots, the proportional contribution of each
grids of observed climate for the period from 1901 to 2006, of the four climate-related indicators was analyzed and mapped
as reported by more than 4000 weather stations around the by means of a pie-chart for each of the hotspots by decomposing
globe. Data on drought frequency was derived from the NOAA each hotspot into its underlying indicators (Figure 22.13).
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 475
22.4 conclusion Drgut, L., O. Csillik, C. Eisank, and D. Tiede. 2014. Automated
parameterisation for multi-scale image segmentation on
In this chapter, we have presented the geon concept as a con- multiple layers. ISPRS Journal of Photogrammetry 88:
ceptual approach and methodological toolset to operationalize 119127.
spatial latent phenomena. We demonstrated that geons are an Drgut, L. and C. Eisank. 2011. Object representations at multiple
appropriate means to present geographical information more scales from digital elevation models. Geomorphology 129:
effectively and thus support efficient policy implementation 183189.
or intervention planning We showed how geons can be built Dragut, L., D. Tiede, and S. R. Levick. 2010. ESPA tool to esti-
through integrating dedicated geospatial datasets applying mate scale parameter for multiresolution image segmen-
regionalization techniques. We discussed aspects that influ- tation of remotely sensed data. International Journal of
ence the robustness of geon delineation, also in relation to Geographical Information Science 24(6): 859871.
MAUP. This generic concept reduces complexity by applying Duque, J. C. and R. R. Suriach. 2006. Supervised regionaliza-
a systemic perspective while maintaining the level of integra- tion methodsA survey. Research Institute of Applied
tion at the specific level of intervention. By this we treat com- Economics Series Title: Working papers, San Diego. 8: 31.
plex phenomena, such as vulnerability, sensitivity, or climate Eurostat. 2006. Regions in the European UnionNomenclature
change impact as a spatial phenomenon emerging from the of Territorial Units for Statistics NUTS 2010/EU-27.
specific arrangement of units generated by an expert-based Luxembourg, Publications Office of the European Union.
partitioning of space. Field, C. B. et al. 2014. Part A: Global and Sectoral Aspects.
Contribution of Working Group II to the Fifth Assessment
References Report of the Intergovernmental Panel on Climate Change.
Cambridge, U.K., Cambridge University Press.
Baatz, M. and A. Schpe. 2000. Multiresolution Segmentation: An Fiksel, J. 2006. Sustainability and resilience: Toward a systems
Optimization Approach for High Quality Multi-Scale Image approach. Sustainability: Science, Practice, & Policy 2(2):
Segmentation. Salzburg, Austria, Wichmann Verlag. 1421.
Beinat, E. 1997. Value Functions for Environmental Management. Goodchild, M. J., M. Yuan, and T. J. Cova. 2007. Towards a general
Dordrecht, the Netherlands, Kluwer Academic Publishers. theory of geographic representation in GIS. International
Berry, B. J. L. 1967. Grouping and regionalizing: An approach to Journal of Geographical Information Science 21(3): 239260.
the problem using multivariate analysis. In Quantitative Hagenlocher, M. 2013. Identifying and evaluating hotspots of
Geography. W. L. Garrison and D. F. Marble (eds.), climate change in the Sahel and Western Africa. In From
Northwestern Studies in Geography Evanston, pp. 219251. Social Vulnerability to Resilience: Measuring Progress toward
Bertalanffy, L. V. 1969. General System TheoryFoundations, Disaster Risk Reduction. S. L. Cutter and C. Corendea
Developments, Applications. New York, George Braziller. (eds.). Bonn, Germany, Publication Series of UNU-EHS,
Biederman, I. 1987. Recognition-by-components: A theory of pp. 93107.
human image understanding. Psychological Review 94(2): Hagenlocher, M., S. Kienberger, S. Lang, and T. Blaschke. 2014a.
115147. Implications of spatial scales and reporting units for the
Blaschke, T. 2010. Object based image analysis for remote sens- spatial modelling of vulnerability to vector-borne diseases.
ing. ISPRS International Journal of Photogrammetry and In GI_Forum 2014: Geospatial Innovation for Society. R.
Remote Sensing 65(1): 216. Vogler, A. Car, J. Strobl, and G. Griesebner (eds.). Berlin,
Blaschke, T. and J. Strobl. 2001. Whats wrong with pixels? Some Germany, Wichmann Verlag, pp. 197206.
recent developments interfacing remote sensing and GIS. Hagenlocher, M., S. Lang, D. Hlbling, D. Tiede, and S. Kienberger.
Zeitschrift fur Geoinformationssysteme 14(6): 1217. 2014b. Modeling hotspots of climate change in the Sahel
Bock, M. et al. 2005. Spatial indicators for nature conserva- using object-based regionalization of multi-dimensional
tion from European to local scale. Ecological Indicators 5: gridded datasets. Journal of Selected Topics in Applied Earth
322338. Observations and Remote Sensing 7 (1): 229234.
Byrne, B. M. 1998. Structural Equation Modeling with LISREL, Hancock, J. R. 1993. Multivariate regionalization: An approach
PRELIS, and SIMPLIS. Mahwah, NJ, Lawrence Erlbaum using interactive statistical visualization. AutoCarto XI.
Associates, Inc. Minneapolis, MN, pp. 218227.
Capra, F. and P. L. Luisi. 2014. The Systems View of Life: A Unifying Hay, G. J. and G. Castilla. 2008. Geographic object-based image
Vision. New York, Cambridge University Press. analysis (GEOBIA): A new name for a new discipline.
Couclelis, H. 2010. Ontologies of geographic information. In Object-Based Image Analysis: Spatial Concepts for
International Journal of Geographical Information Science Knowledge-driven Remote Sensing Applications. T. Blaschke,
24(12): 17851809. S. Lang, and G. J. Hay (eds.). Berlin, Germany, Springer.
Decancq, K. and M. A. Lugo. 2012. Weights in multidimensional Hay, G. J., D. J. Marceau, P. Dub, and A. Buchard. 2001. A mul-
indices of wellbeing: An overview. Econometric Reviews tiscale framework for landscape analysis: Object-specific
32(1): 734. analysis and upscaling. Landscape Ecology 16(6): 471490.
2016 by Taylor & Francis Group, LLC
476 Remotely Sensed Data Characterization, Classification, and Accuracies
Hernando, A., D. Tiede, F. Albrecht, and S. Lang. 2012. Spatial Lang, S., P. Zeil, S. Kienberger, and D. Tiede. 2008. GeonsPolicy-
and thematic assessment of object-based forest stand relevant geo-objects for monitoring high-level indicators.
delineation using an OFA-matrix. International Journal A. Car and J. Strobl (eds.). GI Forum Salzburg: Geospatial
of Applied Earth Observation and Geoinformation 19: Crossroads @ GI_Forum08, Heidelberg, Germany, pp.
214225. 180186.
Jakeman, A. J. and R. A. Letcher. 2003. Integrated assessment and Laszlo, E. 1972. The Systems View of the World. New York, George
modelling: Features, principles and examples for catchment Braziller.
management. Environmental Modelling & Software 18(6): Long, J., T. Nelson, and M. Wulder. 2010. Regionalization of land-
491501. scape pattern indices using multivariate cluster analysis.
Johnston, R. J. 1968. Choice in classificationThe subjectivity of Environmental Management 46: 134142.
objective methods. Annals of the Association of American Marceau, D. J. 1999. The scale issue in the social and natu-
Geographers 58: 575589. ral sciences. Canadian Journal of Remote Sensing 25:
Kienberger, S., T. Blaschke, and R. Z. Zaidi. 2013a. A framework 347356.
for spatio-temporal scales and concepts from different dis- Masolo, C. and L. Vieu. 1999. Atomicity vs. Infinite Divisibility of
ciplines: The vulnerability cube. Natural Hazards. Space. London, U.K., Springer, pp. 235250.
Kienberger, S., D. Contreras, and P. Zeil. 2014. Spatial and Moldan, B. and A. L. Dahl. 2007. Challenges to sustainability indi-
holistic assessment of social, economic, and environ- cators. In Sustainability IndicatorsA Scientific Assessment.
mental vulnerability to floodsLessons from the Salzach T. Hk, B. Moldan, and A. L. Dahl (eds.). Washington, DC,
River Basin, Austria. In Assessment of Vulnerability to Island Press, pp. 126.
Natural Hazards: A European Perspective. J. Birkmann, Nardo, M., M. Saisana, A. Saltelli, S. Tarantola, A. Hoffmann, and
D. Alexander, and S. Kienberger (eds.). San Diego, CA, E. Giovannini. 2005. Handbook on Constructing Composite
Elsevier, pp. 5374. IndicatorsMethodology and User Guide. OECD Working
Kienberger, S., M. Hagenlocher, E. Delmelle, and I. Casas. 2013b. Papers. Organisation for Economic Co-Operation and
A WebGIS tool for visualizing and exploring socioeco- Development, Paris, France.
nomic vulnerability to dengue fever in Cali, Colombia. Ng, R. T. and J. Han. 2002. CLARANS: A method for cluster-
Geospatial Health 8(1): 313316. ing objects for spatial data mining. IEEE Transactions on
Kienberger, S. and M. Hagenlocher. 2014. Spatial-explicit mod- Knowledge and Data Engineering 14: 10031016.
eling of social vulnerability to malaria in East Africa. Openshaw, S. 1984. The Modifiable Areal Unit Problem. Norwich,
International Journal of Health Geographics, 13: 29. U.K.
Kienberger, S., S. Lang, and P. Zeil. 2009. Spatial vulnerability Parson, E. A. 1994. Searching for integrated assessment: A pre-
unitsExpert-based spatial modeling of socio-economic liminary investigation of methods and projects in the inte-
vulnerability in the Salzach catchment, Austria. Natural grated assessment of global climatic change. Third Meeting of
Hazards and Earth System Sciences 9: 767778. the CIESIN Harvard Commission on Global Environmental
Koestler, A. 1967. The Ghost in the Machine. London, U.K., Change Information Policy, Washington, DC.
Hutchinson. Pernkopf, L. and S. Lang. 2011. Spatial meta-indicators:
Lang, S. 2008. Object-based image analysis for remote sensing Assessing landscape sensitivity in the context of SEA.
applications: Modeling realityDealing with complex- RegioResources Dresden. A Cross-Disciplinary Dialogue on
ity. In Object-Based Image AnalysisSpatial Concepts Sustainable Development of Regional Resources. Dresden,
for Knowledge-Driven Remote Sensing Applications. Germany.
T.Blaschke, S. Lang, and G. J. Hay (eds.). Berlin, Germany, Peuquet, D. J. 2002. Representations of Space and Time. New York,
Springer, pp.328. The Guilford Press.
Lang, S., F. Albrecht, S. Kienberger, and D. Tiede. 2010. Object Risbey, J., M. Kandlikar, and A. Patwardhan. 1996. Assessing inte-
validity for operational tasks in a policy context. Journal for grated assessments. Climatic Change 34, 369395.
Spatial Science 55(1): 922. Saisana, M., A. Saltelli, and S. Tarantola. 2005. Uncertainty and
Lang, S., C. Burnett, and T. Blaschke. 2004. Multi-scale object- sensitivity analysis techniques as tools for the quality assess-
based image analysis: A key to the hierarchical organisation ment of composite indicators. Journal of the Royal Statistical
of landscapes. Ekologia Supplement 23: 19. Society 168(2): 307323.
Lang, S., S. Kienberger, L. Pernkopf, and M. Hagenlocher. 2014a. Saltelli, A. etal. 2008. Global Sensitivity Analysis: The Primer. John
Objectbased multiindicator representation of complex Wiley & Sons, West Sussex.
spatial phenomena. South-Eastern European Journal of Schpfer, E., S. Lang, and F. Albrecht. 2008. Object-fate analysis
Earth Observation and Geomatics 3(2s): 625628. Spatial relationships for the assessment of object transition
Lang, S., S. Kienberger, D. Tiede, M. Hagenlocher, and L. and correspondence. In Object-Based Image Analysis
Pernkopf. 2014b. Geonsdomain-specific regionalization Spatial Concepts for Knowledge-Driven Remote Sensing
of space. Cartography and Geographic Information Science Applications. T. Blaschke, S. Lang, and G. J. Hay (eds.).
41(3): 214226. Berlin, Germany, Springer, pp. 785801.
2016 by Taylor & Francis Group, LLC
Object-Based Regionalization for Policy-Oriented Partitioning of Space 477
Stahl, C., A. Cimorelli, C. Mazzarella and B. Jenkins. 2011. Contribution of Working Group III to the Second Assessment
Toward sustainability: A case study demonstrating transdis- Report of the Intergovernmental Panel on Climate Change.
ciplinary learning through the selection and use of indica- P. Bruce, H. Lee, and E. F. Haites (eds.). Cambridge, U.K.,
tors in a decision-making process. Integrated Environmental Cambridge University Press.
Assessment and Management 7(3): 483498. WHO. 2013. World Malaria Report 2013.
Tiede, D., S. Lang, F. Albrecht, and D. Hlbling. 2010. Object-based Wise, S., R. Haining, and J. Ma. 2001. Providing spatial statisti-
class modeling for cadastre constrained delineation of geo- cal data analysis functionality for the GIS user: The SAGE
objects. Photogrammetric Engineering & Remote Sensing project. International Journal of Geographical Information
76 (2): 193202. Science 3(1): 239254.
Tobler, W. 1970. A computer movie simulating urban growth Woodcock, C. E. and A. H. Strahler. 1987. The factor of scale
in the Detroit region. Economic Geography 46(2): in remote sensing. Remote Sensing of Environment 21(3):
234240. 311332.
Voudouris, V. 2010. Towards a unifying formalisation of geo- Wu, J. 1999. Hierarchy and scaling: Extrapolating information
graphic representation: The object-field model with uncer- along a scaling ladder. Canadian Journal of Remote Sensing
tainty and semantics. International Journal of Geographical 25(4): 367380.
Information Science 24(12): 18111828. Wu, J. and H. Li. 2006. Concepts of scale and scaling. In Scaling and
Weyant, J. P. etal. 1996. Integrated assessment of climate change: Uncertainty Analysis in Ecology: Methods and Applications.
An overview and comparison of approaches and results. J. Wu, K. B. Jones, H. Li, and O. L. Loucks (eds.). Berlin,
In Climate Change 1995: Economic and Social Dimensions. Germany, Springer.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
VIII
Global Navigation
Satellite Systems
(GNSS) Remote Sensing
23 g lobal Navigation Satellite Systems Theory and Practice: e volution, State of the Art, and Future
Pathways Mohinder S. Grewal .................................................................................................................................................. 481
Acronyms and Definitions Introduction GPS Global Orbiting Navigation Satellite
System Galileo Compass(BeiDou-2) Conclusions References
24 g lobal Navigation Satellite System r eflectometry for Ocean and Land Applications Kegen Yu, Chris Rizos,
and Andrew Dempster .................................................................................................................................................................493
Acronyms and Definitions Introduction Satellite Missions Related to GNSS Remote Sensing Ocean
Observation Land Applications Challenging Issues and Future Directions Conclusions References
25 g lobal Navigation Satellite Systems for Wide Array of Terrestrial Applications D. Myszor, O. Antemijczuk,
M. Grygierek, M. Wierzchanowski, and K.A. Cyran...............................................................................................................509
Introduction SBAS: Satellite-Based Augmentation System GNSS Errors Consumer Grade GNSS Positioning
Precision Low-Cost Receivers GNSS Applications Summary Acknowledgments References
479
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
23
Global Navigation Satellite
Systems Theory and Practice:
Evolution, State of the Art,
and Future Pathways
Acronyms and Definitions .................................................................................................................481
23.1 Introduction ............................................................................................................................ 482
23.2 GPS ............................................................................................................................................ 482
GPS Orbits GPS Signals Selective Availability Modernization of GPS Elements of
the Modernized GPS GPS Satellite Blocks GPS III
23.3 Global Orbiting Navigation Satellite System ...................................................................... 488
GLONASS Orbits GLONASS Signals Next-Generation GLONASS
23.4 Galileo ...................................................................................................................................... 490
Constellation and Levels of Services Navigation Data and Signals Updates
23.5 Compass (BeiDou-2) ...............................................................................................................491
Compass Satellites Frequency
Mohinder S. Grewal 23.6 Conclusions...............................................................................................................................491
California State University References.............................................................................................................................................491
Acronyms and Definitions HPNS High-precision navigation signal
I/NAV Integrity navigation
ARNS Aeronautical Radio Navigation Service ICD Interface control document
BOC Binary offset carrier IGSO Inclined geosynchronous satellite orbit
BPSK Binary phase-shift keying IOV In-orbit validation
BW Bandwidth L2C Civil signal
C&P Coarse align and precision M code Military code
C/NAV Commercial navigation MEOs Medium earth orbits
CBOC Combined BOC MMT Multipath mitigation technology
CDMA Code division multiple access OS Open service
CNAV Civil navigation P code Precise code
CS Commercial service PPS Precise positioning service
DGNSS Differential Global Navigation Satellite System PRN Pseudorandom noise
DOD Department of Defense PY code Precision encrypted code
ESA European Space Agency RHCP Right hand circular polarization
EU European Union RNSS Radio navigation satellite services band
F/NAV Free navigation SA Selective availability
FAA Federal Aviation Administration SAR Search and rescue
FDMA Frequency division multiple access SBAS Space-Based Augmentation System
FOC Final operational capability SNR Signal-to-noise ratio
GEO Geostationary Earth orbit SoL Safety of life
GLONASS Global Orbiting Navigation Satellite System SPNS Standard precision navigation signal
GPS Global positioning system SPS Standard positioning service
HOW Hand over word U.S. United States
481
2016 by Taylor & Francis Group, LLC
482 Remotely Sensed Data Characterization, Classification, and Accuracies
23.1 introduction is a Differential Global Navigation Satellite System (DGNSS).
DGNSSs reduce the errors in GNSS derived positions by using
Global navigation satellite system (GNSS) has become almost a additional data from a reference GNSS receiver at a known loca-
household term. GNSS applications track trains, guide planes, tion. The most common form of DGNSS involves determining the
and make the world a smaller, more functional planet, as shown combined effects of navigation message ephemeris and satellite
in Figure 23.1. clock errors (including the effects of propagation) at a reference
There are currently four Global Navigation Satellite Systems station and transmitting corrections and integrity in real time to a
(GNSSs) operating or being developed. The first one is the global users receiver (Grewal 2013) (Table 23.1), (Logsdon 1992).
positioning system (GPS) developed by the United States (U.S.)
Department of Defense (DOD) under its Navstar program. The 23.2 GPS
United States launched its first GPS satellite in 1978, and the
system was declared with a final operational capability (FOC) 23.2.1 GPS orbits
in April 1995. The second GNSS configuration developed is the
The fully operational GPS includes 31 or more active satellites
Global Orbiting Navigation Satellite System (GLONASS), placed
approximately uniformly dispersed around 6 circular orbits with
in orbit by the Soviet Union and now maintained by the Russian
4 or more satellites each. The orbits are inclined at an angle of
Republic. The first satellite was launched in 1982, and GLONASS
55 relative to the equator and are separated from each other by
was declared an operational system on September 1993. The
multiples of 60 right ascension. The orbits are nongeostationary
European Union (EU) and European Space Agency (ESA) are
and approximately circular, with radii of 26,560km and orbital
developing the Galileo system, the third GNSS. The first two
periods of one-half sidereal day (11.967h). Theoretically, three
operational satellites were launched in October 2011, preceded
or more GPS satellites will always be visible from most points on
by two Galileo in-orbit validation elements (GIOVE A and B)
the Earths surface, and four or more GPS satellites can be used
launched in 2005 and 2008. Chinas Compass, or BeiDou (Big
to determine an observers position anywhere on the Earths
Dipper), is the fourth GNSS, initially developed in 20002003.
surface 24h/day.
In December 2012, the BeiDou Navigation Satellite System pro-
vided fully operational regional service.
23.2.2 GPS Signals
A summary of the signal characteristics of the four GNSSs are
given in Figure 23.2. Detailed GPS and GLONASS descriptions Each GPS satellite carries a cesium and/or rubidium atomic
are given in Sections 23.2 and 23.3, respectively. Sections 23.4 clock to provide timing information for the signals transmit-
and 23.5 describe Galileo and BeiDou-2, respectively. Section ted by the satellites. Internal clock correction is provided for
23.6 gives conclusions, followed by references. each satellite clock. Each GPS satellite transmits two spread
To improve the accuracy and integrity of the GNSSs, vari- spectrum, L-band carrier signalsan L1 signal with carrier fre-
ous countries are developing space-based augmentation system quency f1 = 1575.42MHz and an L2 signal with carrier frequency
(SBAS). SBAS uses geostationary Earth orbit (GEO) satellites f2= 1227.6 MHz. These two frequencies are integral multiples,
to relay corrections and integrity information to users. SBAS f1 = 1540f0 and f2 = 1200f0, of a base frequency f0 = 1.023 MHz.
Navigation GIS
Aircraft (Geographical Information System)
Ships, submarines Municipalities
Missiles, smart bombs Water and power
Automobiles Fleet Management
Hiking Geology
Large industry
Agriculture
Fertilizing Locators
Children
Weed spraying
Animals
Crop management
Stolen vehicles
Surveying Parolees
911
Mining
Military personnel
Leisure Communications
Golf Internet
Geocaching Cellular phones
Time
FIg u r e 2 3.1 GNSS applications make the world a smaller more functional planet.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems Theory and Practice 483
~33% BW Precise positioning service: Precise positioning service
(PPS) is the full-accuracy (within 5 m), single-receiver
GPS positioning service provided to the United States
~11% BW ~3% BW and its allied military organizations and other selected
Compass agencies. This formal, proprietary service includes
access to the unencrypted P code and the removal of
any SA effects.
Galileo Standard positioning service without SA: Standard posi-
tioning service (SPS) provides GPS single-receiver
(standalone) positioning service to any user on a con-
GLONASS tinuous, worldwide basis. SPS is intended to provide
access only to the C/A code and the L1 carrier.
Standard positioning service with SA: The horizontal-
GPS
position accuracy, as degraded by SA, advertised as 100
m when in use, the vertical-position accuracy as 156 m,
and time accuracy as 334 nsall at the 95% probability
1100 1200 1300 1400 1500 1600 1700 level. SPS also guarantees the user-specified levels of
Frequency (MHz) coverage, availability, and reliability.
FIgu r e 23.2 Comparison of frequency and bandwidth (BW) for
various GNSSs. 23.2.4 Modernization of GPS
The L1 signal from each satellite uses binary phase-shift keying Since GPS was declared with an FOC in April 1995, applications
(BPSK), modulated by two pseudorandom noise (PRN) codes in and the use of GPS have evolved rapidly, especially in the civil
phase quadrature, designated as the coarse acquisition or C/A sector. See Figure 23.1 for a sampling of applications. As a result,
code and precise or P code. The L2 signal from each satellite is radically improved levels of performance have been reached in
BPSK modulated by only the P code. A brief description of the positioning, navigation, and time transfer. However, the avail-
nature of these PRN codes follows (Rockwell 1991). ability of GPS has also spawned new and demanding applica-
tions that reveal certain shortcomings of the present system.
Therefore, since the mid-1990s, numerous governmental and
23.2.3 Selective Availability
civilian committees have investigated these shortcomings and
Prior to May 1, 2000, the U.S. Department of Defense deliber- requirements for a more modernized GPS.
ately derated the accuracy of GPS for nonauthorized (i.e., non- The modernization of GPS is a difficult and complex task that
U.S. military) users, using a combination of methods. This requires trade-offs in many areas. Major issues include spectrum
selective availability (SA) included pseudorandom time dither- needs and availability, military and civil performance, signal
ing and truncation of the transmitted ephemerides. SA degraded integrity and availability, financing and cost containment, and
the navigation solution by 100 m horizontally and 156 m verti- potential competing interests by other GNSSs and developing
cally. The initial satellite configuration used SA with pseudoran- countries. Major decisions have been made for the incorporation
dom dithering of the onboard time reference (Janky 1997) only, of new civil frequencies, and new civil and military signals that
but this was discontinued on May 1, 2000. will enhance the overall performance of the GPS.
TABLe 2 3.1 Comparison of Various GNSSs
GPS GLONASS Galileo Compass
Number of satellites 31 24 30 35
Orbit radius (km) 26,600 25,510 29,600 19,100
Number of orbit planes when fully developed (global) 6 3 3 Now regional
Orbit plane inclination () 55.5 64.8 54 55
Architecture CDMA FDMA CDMA CDMA
SA No (SA discontinued in 2000) No No No
Service Free Free Encrypted Fee
Codes Coarse align and precision (C&P) same for C, P (different codes C&Psame C&P same
all PRN for each PRN) for all PRN for all PRN
Carrier (MHz) 1575.42 1602 1575.42 1561.098
PRN (C/A) code length (chips) 1023 511 4096 2046
Data bit rate (ms) 20 20 4 20
2016 by Taylor & Francis Group, LLC
484 Remotely Sensed Data Characterization, Classification, and Accuracies
Normalized psd (dB/Hz)
L2 C
L5 L2 P(Y) L1 P(Y) L1 C/A
L2 M L1 M
1176.45 1227.60 1575.42 f (MHz)
BPSK(1) BPSK(10) BOC(10, 5) BPSK(1) BPSK(10)
FIg u r e 2 3.3 A modernized GPS signal spectrum shown along with the legacy GPS signals.
GPS IIF and GPS III are being designed under various con- has a substantial amount of uncontrolled signals in it.)
tracts (Raytheon, Lockheed Martin). These will have a new L2 In applications involving public safety, the integrity of
civil signal and new L5 signal modulated by a new code struc- the current system can be improved with a robust dual-
ture. These frequencies will improve the ambiguity resolution, frequency capability where both GPS signals are within
ionospheric calculation, and C/A code positioning accuracy. the ARNS bands. This is particularly true in aviation
landing systems that demand the presence of an adequate
23.2.4.1 Areas to Benefit from Modernization number of high-integrity satellite signals and functional
cross-checks during precision approaches.
The areas that could benefit from a modernized GPS are the
5. Improvement is needed in multipath mitigation capability:
following:
Multipath remains a dominant source of GPS positioning
1. Robust dual-frequency ionosphere correction capability for error and cannot be removed by differential techniques.
civil users: Since only the encrypted P(Y) code appears on Although certain mitigation techniques, such as mul-
the L2 frequency, civil users have lacked a robust dual- tipath mitigation technology (MMT), approach theo-
frequency ionosphere. Civil users had to rely on semicode- retical performance limits for in-receiver processing, the
less tracking of the GPS L2 signal, which is not as robust required processing adds to receiver costs. In contrast,
as access to a full strength unencrypted signal. While effective multipath rejection could be made available to all
civil users could employ a differential technique, this adds receivers by using new GPS signal designs.
complexity to an ionosphere free user solution. 6. Military requirements in a jamming environment: The fea-
2. A better civil code: While the GPS C/A code is a good, ture of SA was suspended at 8 pm EDT on May 1, 2000.
simple spreading code, a better civil code would provide SA was the degradation in the autonomous positioning
better correlation performance. Rather than just turning performance of GPS, which was a concern in many civil
on the C/A code on the L2 frequency, a more advanced applications requiring the full accuracy of which GPS is
spreading code would provide robust ranging and iono- capable. If the GPS C/A code was ever interfered with, the
sphere error predictions. accuracies that it could be afforded would not be pres-
3. Ability to resolve ambiguities in phase measurements needs ent. Because the P(Y) code has an extremely long period
improvement: High-accuracy differential positioning at (7days), it is difficult to acquire unless some knowledge
the centimeter level by civil users requires rapid and reli- of the code timing is known. P(Y) timing information
able resolution of ambiguities in phase measurements. is supplied by the GPS Hand Over Word (HOW) at the
Ambiguity resolution with single-frequency (L1) receiv- beginning of every subframe. However, to read the HOW,
ers generally requires a sufficient length of time for the the C/A code must first be acquired to gain access to the
satellite geometry to change significantly. Performance is navigation message. Unfortunately, the C/A code is rela-
improved with robust dual-frequency receivers. However, tively susceptible to jamming, which would seriously
the effective signal-to-noise ratio (SNR) of the legacy P(Y) impair the ability of a military receiver to acquire the P(Y)
encrypted signal is dramatically reduced, because the code. Direct P(Y) acquisition techniques are possible, but
encrypted P code cannot be despread by the civil user. these techniques still require information about the satel-
4. Dual-frequency navigation signals in the ARNS band: The lite position, user position, and clock errors to be success-
Aeronautical Radio Navigation Service (ARNS) band of ful. Furthermore, an interference on the C/A coded signal,
frequencies are federally protected and can be used for would also effect the P(Y) coded signal in the same fre-
safety-of-life applications. The GPS L1 band is an ARNS quency band. It would be far better if direct acquisition of
band but the GPS L2 band is not. (The L2 band is in the a high-performance code were possible, without the need
Radio Navigation Satellite Services Band [RNSS] that to first acquire the C/A code.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems Theory and Practice 485
7. Additional power received from the satellite would also 3. The additional L2C signal will improve robustness in acqui-
help military users operate more effectively. While same sition and tracking: The new spreading code identified for
advantages can be gained in the user equipment, hav- civil (C) users will provide more robust acquisition and
ing an increased power from the satellite could provided tracking performance.
added value.
Originally, modernization efforts considered turning on the
8. Compatibility and operability with other GNSSs: With the
C/A code at the L2 carrier frequency (1227.60 MHz) to pro-
advances in other GNSSs by other nations, the require-
vide the civilian community with a robust ionosphere correc-
ment exists for international cooperation in the develop-
tion capability as well as additional flexibility and robustness.
ment of new GNSS signals to ensure they do not interfere
However, later in the planning process, it was realized that addi-
with each other and potentially provide an interoperable
tional advantages could be obtained by replacing the planned
combined GNSS service.
L2 C/A signal with a new L2 civil signal (L2C). The decision was
made to use this new signal, and its structure was made public
early in 2001. Both the L2C and the new military M code signal
23.2.5 elements of the Modernized GPS
(to be described) appear on the L2 carrier orthogonal to the
Figure 23.3 illustrates the modernized GPS signal spectrum. The current P(Y).
legacy GPS signals (L1 C/A and P(Y), and L2 P(Y)) and the addi- Like the C/A code, the C code is a PRN code that runs at
tional modernized signals (L2C, L5, and GPS L1 and L2 M code) a 1.023 10 6 cps (chips per second) rate. However, it is gen-
are illustrated. The L1C signal is not shown. erated by 2:1 time-division multiplexing of two independent
A modernized GPS signal spectrum is shown along with the subcodes, each having half the chipping rate, namely, 511.5
legacy GPS signals. 103 cps. Each of these subcodes is made available to the receiver
The bandwidths potentially available for the modern- by demultiplexing. These two subcodes have different periods
ized GPS signals are up to 24MHz, but the compatibility and before they repeat. The first subcode, the code-moderate (CM),
power levels relative to the other codes need to be considered. has a moderate length of 10,230 chips, a 20 ms period. The
Furthermore, assuming equal received power and filtered moderate length of this code permits relatively easy acquisi-
bandwidth, the ranging performance (with or without mul- tion of the signal although the 2:1 multiplexing results in a
tipath) on a GPS signal, the performance is highly dependent 3dB acquisition and data demodulation loss. The second sub-
upon the signals spectral shape (or equivalently, the shape of code, the code-long (CL), has a length of 707,250 chips, a 1.5 s
the autocorrelation function). In this sense, the L1 C/A coded period, and is data-free. The CM and CL codes are combined
and L2 civil signals are somewhat equivalent in scope, as are to provide the C code at the 1.023 Mcps rate. Navigation data
the P(Y) and L5 civil signals (albeit with very different charac- can be modulated on the C code. Provisions call for no data,
teristics). As we will see, the military M-coded signal and other legacy navigation data at a 50 bps rate, or new civil navigation
GNSS codes are different, because they use different subcarrier (CNAV) data at a 25 bps; the CNAV data at a 25 bps rate would
frequencies and chipping rates. These different subcarriers, in be encoded using a rate 1/2 convolutional encoding technique,
essence, add an aspect known as frequency division multiplex- to produce a 50 sps (symbols per second) data bit stream that
ing to the GPS spectrum. could then be modulated onto the L2C signal. With no data,
The major elements of these modernized signals are discussed the coherent processing time can be increased substantially,
in the following sections. thereby permitting better code and carrier tracking perfor-
mance, especially at low SNR. The relatively long CL code
23.2.5.1 L2 civil Signal length also generates smaller correlation sidelobes as com-
pared to the C/A code. Details on the L2 civil signal are given
This new civil signal has a new code structure that has some per-
by Fontana etal. (2001).
formance advantages over the legacy C/A code. The L2 civil signal
The existing C/A code at the L1 frequency will be retained for
(L2C) signal offers civilian users the following improvements:
legacy purposes.
1. Robust dual-frequency ionosphere error correction: The
dispersive delay characteristic of the ionosphere propor- 23.2.5.2 L5 Signal
tional to 1/f 2 can be estimated much more accurately with Although the use of the L1 and L2C signals can satisfy most
this new, full strength signal on the L2 frequency. Thus, civil users, there are concerns that the L2 frequency band may
civil users can choose to use a semicodeless P(Y) L2 and be subject to unacceptable levels of interference for applications
C/A L1, or a new L2C and C/A L1 technique to estimate involving public safety, such as aviation. The potential for inter-
the ionosphere. ference arises, because the International Telecommunications
2. Carrier phase ambiguity resolution will be significantly Union (ITU) has authorized the L2 band on a coprimary basis
improved: The accessibility of the full strength L1 and L2 with radiolocation services, such as high-power radars. As a
signals provides wide-lane measurement combinations result of Federal Aviation Administration (FAA) requests, the
having ambiguities that are much easier to resolve. Department of Transportation and Department of Defense have
2016 by Taylor & Francis Group, LLC
486 Remotely Sensed Data Characterization, Classification, and Accuracies
called for a new civil GPS frequency, called L5 at 1176.45MHz in GPS modernization for the L5 signal calls for a completely
the ARNS of 9601215MHz. To gain maximum performance, new civil signal format (i.e., L5 code) at a carrier frequency
the L5 spread spectrum codes were selected to have a higher of 1176.45 MHz (i.e., L5 carrier). The L5 signal is defined in
chipping rate and longer period than the C/A codes to allow for a quadrature scene where the total signal power is divided
better accuracy measurements. Additionally, the L5 signal has equally between in-phase (I) and quadrature (Q) compo-
two signal components in phase quadrature, one of which will nents. Each component is modulated with a different but
not carry data modulation. The L5 signal will provide the follow- synchronized 10,230 chip direct sequence L5 code transmit-
ing system improvements: ted at 10.23 10 6 cps (chips per second), the same rate as the
P(Y) code, but with a 1 millisecond (ms) period, the same as
1. Ranging accuracy will improve: Pseudorange errors due
the C/A code period. The I channel is modulated with a 100
to random noise will be reduced below levels obtainable
sps data stream, which is obtained by applying rate 1/2, con-
with the C/A codes, due to the larger bandwidth of the
straint length 7, forward error correction (FEC) convolutional
proposed codes. As a consequence, both code-based posi-
coding to a 50 bps navigation data message that contains a
tioning accuracy and phase ambiguity resolution perfor-
24-bit cyclic redundancy check (CRC). The Q channel is
mance will improve.
unmodulated by navigation data. However, both channels are
2. Errors due to multipath will be reduced: The larger band-
further modulated by NeumanHoffman (NH) synchroniza-
width of the new codes will sharpen the peak of the code
tion codes, which provide additional spectral spreading of
autocorrelation function, thereby reducing the shift in the
narrowband interference, improve bit and symbol synchroni-
peak due to multipath signal components. The eventual
zation, and also improve crosscorrelation properties between
multipath mitigation will depend upon the final receiver
signals from different GPS satellites. The L5 signal is shown in
design and delay of the multipath.
Figure 23.4 illustrating the modernized GPS (and legacy GPS)
3. Carrier phase tracking will improve: Weak-signal phase
signal spectrum.
tracking performance of GPS receivers is severely limited
Compared to the C/A code, the 10-times larger chip count
by the necessity of using a Costas (or equivalent-type)
of the I and Q channel civil L5 codes provides lower auto-
PLL to remove carrier phase reversals of the data modula-
correlation sidelobes, and the 10 times higher chipping rate
tion. Such loops rapidly degrade below a certain threshold
substantially improves ranging accuracy, provides better
(about 2530 dB Hz), because truly coherent integration
interference protection, and substantially reduces multipath
of the carrier phase is limited to the 20 ms data bit length.
errors at longer path separations (i.e., long delay multipath).
In contrast, the data-free quadrature component of the
Additionally, these codes were selected to reduce, as much as
L5 signal will permit coherent integration of the carrier
possible, the crosscorrelation between satellite signals. The
for arbitrarily long periods, which will permit better phase
absence of data modulation on the Q channel permits lon-
tracking accuracy and lower tracking thresholds.
ger coherent processing intervals in code and carrier track-
4. Weak-signal code acquisition and tracking will be
ing loops, with full-cycle carrier tracking in the latter. As a
enhanced: The data-free component of the L5 signal
result, the tracking capability and phase ambiguity resolution
will also permit new levels of positioning capability
become more robust.
with very weak signals. Acquisition will be improved,
Further details on the civil L5 signal can be found in refer-
because fully coherent integration times longer than
ences given within parenthesis (Hegarty and Van Dierendonck
20 ms will be possible. Code tracking will also improve
1999, Van Dierendonck and Spilker 1999, Spilker and Van
by virtue of better carrier phase tracking for the purpose
Dierendonck 2001, GPS Directorate 2012).
of code rate aiding.
5. The L5 signal will further support rapid and reliable carrier
phase ambiguity resolution: The L5 signal is a full strength,
GPS evolution
high chipping rate code that will provide high-quality
code and carrier phase measurements. These can be used GPS III (under dev.)
to support various code and carrier combinations for high
GPS IIF (under dev. by Boeing)
accuracy carrier phase ambiguity resolution techniques.
6. The codes will be better isolated from each other: The longer GPS IIR (19972004)
length of the L5 codes will reduce the size of crosscorre-
lation between codes from different satellites, thus mini- GPS IIA (19901997)
mizing the probability of locking onto the wrong code
during acquisition, even at the increased power levels of GPS II (19861990)
the modernized signals.
GPS I (19781985)
7. Advances Navigation Messaging: The L5 signal structure
has a new civil navigation (CNAV) messaging structure
that will allow for increase data integrity. FIg u r e 2 3.4 GPS evolution.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems Theory and Practice 487
23.2.5.3 M code 23.2.5.4 L1c Signal
The new military (M) codes will also be transmitted on both the A proposed new L1 civil (L1C) signal is planned for the next
L1 and L2 carrier frequencies. These M codes are based on a new generation of GPS SVs. Although the current C/A code is planned
family of split-spectrum GNSS codes for military and new GPS to remain on the L1 frequency (1575.42 MHz), the additional
civil signals (Spilker etal. 1998). The M codes will provide the L1C signal will add a higher-performance civil signal at the L1
following advantages to military users: frequency with potential interoperability with other GNSSs.
Like the L5 civil signal, the planned L1C signal will have a data-
1. Direct acquisition of the M codes will be possible: The
free quadrature component.
design of these codes will eliminate the need to first
The original L1C signal structure considered a pure BOC(1, 1)
acquire the L1 C/A code with its relatively high vulner-
signal but evolved into a more complex signal that multiplexed
ability to jamming.
two BOC signals. Additional complexities involved the desire
2. Better ranging accuracy will result: As can be seen in
to have the L1C signal interoperable with other GNSS signals,
Figure 23.3, the M codes have significantly more energy
as well as, potential intellectual property issues. The L1C signal
near the edges of the bands, with a relatively small amount
contains a dataless (i.e., pilot) and data signal component trans-
of energy near the band center. Since most of the C/A
mitted in quadrature. The L1C signal for GPS that emerged from
code power is near the band center, potential interference
development is based upon a time-multiplexed BOC (TMBOC)
between the codes is mitigated. The effective bandwidth
modulation technique that synchronously time-multiplexed the
of the M codes is much larger than that of the P(Y) codes,
BOC(1, 1) and BOC(6, 1) spreading codes for the pilot compo-
which concentrate most of their power near the L1 or L2
nent (designated as L1Cp), and a BOC(1, 1) modulated signal
carrier. Because of the modulated subcarrier, the autocor-
(designated as L1CD). For both, the BOC codes are generated
relation function of the M codes has not just one peak, but
synchronously at rate of 1.023 MHz and based the Legendre
several peaks spaced one subcarrier period apart, with the
sequence called Weil code. These codes have a period of 10 ms,
largest at the center. The modulated subcarrier will cause
so 10,230 chips are within one period. Additionally, there is an
the central peak to be significantly sharpened, signifi-
overlay code that is encoded onto the L1Cp pilot channel. One
cantly reducing pseudorange measurement error.
bit of the overlay code has a duration, and is synchronized to the
3. Error due to multipath will be reduced: The sharp central
10 ms period of the BOC code generators. The overlay code rate
peak of the M code autocorrelation function is less suscep-
is 100 bps and has 1800bits in an 18 s period.
tible to shifting in the presence of multipath correlation
To generate the TMBOC signal for the L1Cp channel,
function components.
the BOC(1, 1) and BOC(6, 1) spreading sequences are time-
The M coded signals will be transmitted on the L1 and L2 car- multiplexed. With 33 symbols of a BOC(1, 1) sequence, 4 sym-
riers, with the capability of using different codes on the two fre- bols are replaced with BOC(6, 1) chips. These occur at symbols 0,
quencies. The M codes are known as binary offset carrier (BOC) 4, 6, and 29. Thus, with a 75% of the power planned for distribu-
encoded signals where the notation of BOC(fsx, fcx) is used where tion in the pilot signal, there will be 1/11th of the power in the
fsx represents the subcarrier multiplier, and fcx represents the code BOC(6, 1) component and 10/11th of the power in the BOC(1, 1)
rate multiplier, with respect to a nominal code rate of 1.023MHz. component of the carrier.
The M code is a BOC(10, 5) code in which a 5.115 Mcps chipping The L1C signal also has a new navigation message structure, des-
sequence modulates a 10.23MHz square wave subcarrier. Each ignated as CNAV-2, with three different subframe formats defined.
spreading chip subtends exactly two cycles of the subcarrier, with Subframe 1 contains GPS time information (i.e., time of interval
the rising edge of the first subcarrier cycle coincident with initia- [TOI]). Subframe 2 contains ephemeris and clock correction data.
tion of the spreading chip. The spectrum of the BOC(10, 5) code Subframe 3 is commutated over various pages and provides less
has considerably more relative power near the edges of the signal time-sensitive data such almanac, Universal Time, Coordinated
bandwidth than any of the C/A, P(Y), L2C, and L5 coded sig- (UTC), ionosphere, and can be expended in the future.
nals. As a consequence, the M coded signal has minimal spectral The split-spectrum nature of the L1C BOC(1, 1) encoded
overlap with the other GPS transmitted signals, which permits signal will provide some frequency isolation from the L1 C/A
transmission at higher power levels without mutual interference. encoded signal. Each spreading chip subtends exactly one cycle
The resulting spectrum has two lobes, one on each side of the of the subcarrier, with the rising edge of the first subcarrier
band center, thereby producing the split-spectrum code. The M half-cycle coincident with initiation of the spreading chip. The
code signals are illustrated in Figure 23.10. The M code signal MBOC codes provide a larger RMS bandwidth compared to
is transmitted in the same quadrature channel as the C/A code pure BOC(1, 1).
(i.e., with the same carrier phase), that is, in phase quadrature For many years now, cooperation at the international level has
with the P(Y) code. The M codes are encrypted and unavailable been ongoing to enable the L1C signal to be interoperable with
to unauthorized users. The nominal received power level is 158 other GNSSs. A combined interoperable signal would allow a
dBW at Earth. Additional details on the BOC(10, 5) code can be user to ubiquitously use navigation signals from different GNSS
found in a paper by Barker etal. (2000). with known and specified performances attributes.
2016 by Taylor & Francis Group, LLC
488 Remotely Sensed Data Characterization, Classification, and Accuracies
Additional details on the L1C signal can be found in Spilker 23.3 Global orbiting navigation
etal. 1998, Barker etal. 2000, Spilker and Van Dierendonck 2001,
Issler etal. 2004, Betz etal. 2006, and GPS Directorate 2011.
Satellite System
A second configuration for global positioning is the GLONASS,
23.2.6 GPS Satellite Blocks placed in orbit by the former Soviet Union, and now main-
tained by the Russian Republic. GLONASS is the Russian
The families of satellites launched prior to recent modernization GNSS. The GLONASS has similar operational requirements to
efforts are referred to as Block I (19781985), Block II (1989 GPS with some key differences in its configuration and signal
1990), and Block IIA (19901997); all of these satellites transmit structure. Like GPS, GLONASS is an all-weather, 24 h satel-
the legacy GPS signals (i.e., L1 C/A and P(Y) and L2 P(Y)). (The lite-based satellite navigation system that has a space, control,
United States Naval Observatory has an up-to-date listing of all and user Segment. The first GLONASS satellite was launched
of the GPS satellites in use today (USNO 2012).) in 1982, and the GLONASS declared an operational system on
In 1997 a new family, the Block IIR satellites, began to replace September 24, 1993.
the older Block II/IIA family. The Block IIR satellites have several
improvements, including reprogrammable processors enabling
problem fixes and upgrades in flight. Eight Block IIR satellites
23.3.1 GLonASS orbits
were modernized (designated as Block IIR-M) to include the new The GLONASS satellite constellation is designed to operate with
military M code signals on both the L1 and L2 frequencies, as 24 satellites in 3 orbital planes at 19,100 km altitude (whereas
well as the new L2C signal on L2. The first modernized Block IIR GPS uses 6 planes at 20,180 km altitude). GLONASS calls for
was launched in September 2005. 8SVs equally spaced in each plane. The GLONASS orbital period
To help secure the L5 frequency utilization, one of the Block is 11h 15min, which is slightly shorter than the 11h 56min for
IIR-M satellites (GPS IIR-20(M)), SV49, was outfitted with a spe- a GPS satellite. Because some areas of Russia are located at high
cial L5 payload and launched on March 24, 2009. This particu- latitudes, the orbital inclination of 64.8 is used as opposed to
lar satellite had hardware configuration issues relating to the L5 the inclination of 55 used for GPS.
payload installation and is transmitting a degraded signal. Since GLONASS has 24 satellites, distributed approximately uni-
that time, the navigation signals have been set unhealthy in the formly in 3 orbital planes (as opposed to 6 for GPS) of 8 satel-
broadcast navigation message. lites each. Each orbital plane has a nominal inclination of 64.8
The Block IIF (i.e., follow-on) family was the next generation relative to the equator, and the three orbital planes are separated
of GPS satellites, retaining all the capabilities of the previous from each other by multiples of 120 right ascension. GLONASS
blocks, but with many improvements, including an extended orbits have smaller radii than GPS orbits, about 25,510km, and
design life of 12 years, faster processors with more memory, a satellite period of revolution of approximately 8/17 of a sidereal
and the inclusion of the new L5 signal on a third L5 frequency day. A GLONASS satellite and a GPS satellite will complete 17
(1176.45MHz). The first Block IIF satellite was launched in May and 16 revolutions, respectively, around the Earth every 8days.
2010. Each GLONASS satellite transmits its own ephemeris and
system almanac data. Via the GLONASS ground control seg-
ment, each GLONASS satellite transmits its position, velocity,
23.2.7 GPS iii
and lunar/solar acceleration effects in an ECEF coordinate frame
The next block of GPS satellites planned is designated as the (as opposed to GPS, that encodes SV positions using Keplerian
Block III family, which is still under development. The GPS III orbital parameters). GLONASS ECEF coordinates are refer-
block of satellites and associated GPS ground control segment enced to the PZ-90.02 datum and time reference linked to their
components will represent a major advancement in capabilities National Reference of Coordinated Universal Time UTC(SU)
for military and civil users. GPS III is planned to include all leg- (Soviet Union, now Russia).
acy, and modernized GPS signal components, including the new
L1C and L5 signals and add specified signal integrity. The added
23.3.2 GLonASS Signals
signal integrity planned for in GPS III may be able to satisfy
some of the aviation requirements (FAA 2008). Improvements The GLONASS system uses frequency-division multiplex-
for military include two high-power spot beams for the L1 and ing of independent satellite signals. Its two carrier signals
L2 military M code signals, providing 20 dB higher received corresponding to L1 and L2 have frequencies f1 = (1.602 +
power over the earlier M code signals. However, in the fully 9k/16) GHz and f2 = (1.246 + 7k/16) GHz, where k = 0, 1, 2, ,
modernized Block III satellites, the M coded signal components 23 is the satellite number. These frequencies lie in two bands
are planned to be radiated as physically distinct signals from a at 1.5971.617 GHz (L1) and 12401260 GHz (L2). The L1 code
separate antenna on the same satellite. This is done in order to is modulated by a C/A code (chip rate = 0.511MHz) and by a
enable optional transmission of a spot beam for greater antijam P code (chip rate = 5.11MHz). The L2 code is presently modu-
resistance within a selected local region on the Earth. Figure lated only by the P code. The GLONASS satellites also trans-
23.4 shows the evolution of GPS. mit navigational data at a rate of 50 baud. Because the satellite
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems Theory and Practice 489
frequencies are distinguishable from each other, the P code in 2003. These GLONASS-M satellites had improved frequency
and the C/A code are the same for each satellite. The meth- plans and the accessible signals that are known today. The new gen-
ods for receiving and analyzing GLONASS signals are similar eration of GLONASS satellites is designated as GLONASS-K satel-
to the methods used for GPS signals (Janky 1997). GLONASS lites and is considered a major modernization effort by the Russian
does not use any form of SA. government. These GLONASS-K satellites plan to transmit the
legacy GLONASS FDMA signals as well as a new CDMA format.
23.3.2.1 Frequency Division Multiple Access Signals Additional details of the GLONASS signal structure and
GLONASS uses multiple frequencies in the L-band and has GLONASS-M satellite capabilities can be found in GLONASS
used frequencies separated by a substantial distance for iono- (2008).
sphere mitigation (i.e., L1 and L2), but these are slightly different
than the GPS L1 and L2 frequencies. One significant difference
23.3.3 next-Generation GLonASS
between GLONASS and GPS is that GLONASS has historically
used a frequency division multiple access (FDMA) architecture The satellite for the next generation of GLONASS-K was
as opposed to the CDMA approach used by GPS. launched on February 26, 2011, and continues to undergo flight
tests. Twenty-four satellites consisting of four GLONASS-M
23.3.2.2 carrier components satellites are all transmitting healthy signals. Recently, Russian
The GLONASS uses two L-band frequencies, L1 and L2, as scientists have proposed a new code-division multiple-access
defined in the following. The channel numbers for GLONASS signal format to be broadcast on a new GLONASS L3 signal.
signal operation are for Once implemented across the modernizing GLONASS constel-
lation, this will facilitate interoperability with, and eventually
f K 1 = f 01 + Kf 1 interchangeability among, other GNSS signals. The flexible mes-
sage format permits relatively easy upgrades in the navigation
message (GPS World 2013).
f K 2 = f 02 + Kf 2
23.3.3.1 code Division Multiple
where Access Modernization
K = channel number (7 K + 6) One of the issues with an FDMA GNSS structure is the inter-
f01 = 1602M; f1 = 562.5 kHz channel (i.e., interfrequency) biases that can arise within the
f02 = 1246M; f2 = 437.5 kHz FDMA GNSS receiver. If not properly addressed in the receiver
design, these interchannel biases can be a significant error
23.3.2.3 Spreading codes and Modulation source in the user solutions. These error sources arise, because
With the GLONASS signals isolated in frequency, an optimum the various navigation signals pass through the components
maximal-length (m-sequence) can be used as the spreading within the receiver at slightly different frequencies. The group
code. GLONASS utilizes two such codes, one standard precision delay thought these components are noncommon, at the differ-
navigation signal (SPNS) at a 0.511 Mcps rate that repeats every ent frequencies, and produced different delays on the various
1 ms and a second high-precision navigation signal (HPNS) navigation signals, coming from different satellites. This inter-
at a 5.11 Mcps rate, that repeats every 1 s. Similar to GPS, the frequency bias is substantially reduced (on a comparative basis)
GLONASS signals utilized BPSK modulation and are transmit- with CDMA-based navigation systems, because all of the sig-
ted out of a right hand circularly polarized antenna. nals are transmitted at the same frequency. (The relatively small
amount of Doppler received from the various CDMA navigation
23.3.2.4 navigation Data Format signals is minor when considering the group delay.)
The format of the GLONASS navigation data is similar to the GPS An additional consideration with an FDMA GNSS signal
navigation data format, with different names and content. The structure is the amount of frequency bandwidth that is required
GLONASS navigation data format is organized as a Superframe to support the FDMA architecture. CDMA architecture typically
that is made up of frames, where frames are made up of strings. has all the signals transmitted at the same carrier frequency, for
ASuperframe has a duration of 150 s and is made up of five frames, more efficient utilization of a given bandwidth.
so each frame lasts 30 s. Each frame is made up of 15 strings, where GLONASS has established several separate versions of its
a string has a duration of 2 s. GLONASS encodes satellite ephem- GLONASS-K satellites. The first GLONASS-K1 satellite was
eris data as immediate data and almanac data as nonimmedi- launched on February 26, 2011, which carried the first GLONASS
ate data. There is a time mark in the GLONASS navigation data CDMA signal structure and has been successfully tracked on
(last 0.3 s of a string) that is an encode PRN sequence. Earth (Septentrio 2011). The GLONASS-K1 satellite transmits
a CDMA signal at a designated L3 frequency of 1202.025MHz
23.3.2.5 Satellite Families (test signal), as well as the legacy GLONASS FDMA signals at
While the first series of GLONASS satellites were launched from L1 and L2. The CDMA signal from the GLONASS-K1 satellite
1982 to 2003, the GLONASS-M satellites were launched beginning is considered a test signal. The follow-on generation of satellites
2016 by Taylor & Francis Group, LLC
490 Remotely Sensed Data Characterization, Classification, and Accuracies
is designated as the GLONASS-K2 satellites (Revnivykh 2011). data format, and what service the signal is intended to support.
Afull constellation of legacy and new CDMA signals are planned The European Union has published the Galileo Open Service
for these GLONASS-K2 satellites including plans to transmit its (OS) Interface Control Document that contains significant detail
CDMA signal on or near the GPS L1 and L2 frequency bands. on the Galileo signals in space (EU etal. 2010). All of the Galileo
The GLONASS KM satellites are in the research phase and plans signals transmit two orthogonal signal components, where
call for the transmission of the legacy GLONASS FDMA signals, the in-phase component transmits the navigation data and the
the CDMA signals introduced in the GLONASS-K2, and a new quadrature component is dataless (i.e., a pilot). These two com-
CDMA signal on the GPS L5 frequency. ponents have a power sharing so that the dataless channel can be
used to aid the receiver in acquisition and tracking of the signal.
23.4 Galileo All of the individual Galileo GNSS signal components utilize
phase-shift keying modulation and right hand circular polariza-
Galileo is a GNSS being developed by the European Union and tion (RHCP) for the navigation signals.
the ESA. Like GPS and GLONASS, it is an all-weather, 24h sat- To support the various services for Galileo, three different
ellite-based navigation system being designed to provide various navigation formats are planned for implementation: (1) A free
services. The program has had three development phases: (1) def- navigation (F/NAV) format to support OS in for the E1 and E5a
inition (completed), (2) development of on-orbit validation, and signals, (2) the integrity navigation (I/NAV) format is planned
(3) launch of operational satellite, including additional develop- to support SoL services for the L1 and E5a signals, and (3) a
ment (ESA 2012). The first Galileo in-orbit validation element commercial navigation (C/NAV) format is to support CS.
(GIOVE) satellite, designated as GIOVE-A, was launched in Galileo E1 frequency (same as GPS L1) at 1574.42MHz will
December 28, 2005, followed by the GIOVE-B on April 27, 2008. have a split-spectrum type signal around the center frequency.
The next two Galileo operational satellites were launched on This signal is planned to interoperate with the GPS L1C signal;
October 21, 2011, to provide additional validation of the Galileo. however, discussions continue on the technical, political, and
intellectual property aspects with its implementation. Despite
23.4.1 constellation and Levels of Services these challenges, the Galileo E1 signal is planned to be a com-
The full constellation of Galileo is planned to have 30 satellites in bined BOC (CBOC) signal that is based upon two BOC signals
medium earth orbits (MEOs) with an orbital radius of 23,222km (a BOC(1, 1) and BOC(6, 1) basis component) in phase quadra-
(similar to the GPS orbital radius of 20,180km). The inclination ture. The E1 in-phase component has I/NAV data encoded on it,
angle of the orbital plane is 56 (GPS is 55), with three orbital and the quadrature phase has no data (i.e., pilot).
planes (GPS has 6). This constellation will thus have 10 satellites The Galileo E6 signal is planned to support CS at a center fre-
in each orbital plane. quency of 1278.750MHz, with no offset carrier, and a spreading
Various services are planned for Galileo, including an open code at a rate of 5.115MHz (5 1.023MHz). The E6 signal has
service (OS), commercial service (CS), public regulated, a safety two signal components in quadrature, where the C/NAV mes-
of life (SoL), and a search and rescue (SAR) service. These ser- sage format is encoded on the in-phase component, and no data
vices will be supported with different signal structures and is on the quadrature component.
encoding formats tailored to support the particular service. The E5 signal is a unique GNSS signal that has an overall
center frequency of 1191.795 MHz, with two areas of maxi-
23.4.2 navigation Data and Signals mum power at 1176.450 and 1207.140MHz. The wideband E5
signal is generated by a modulation technique called AltBOC.
Table23.2 lists some of the key parameters for the Galileo that The generation of the E5 signal is such that it is actually com-
will be discussed in this section. The table lists the Galileo sig- posed of two Galileo signals that can be received and processed
nals, frequencies, identifiable signal component, its navigation separately, or combined by the user. The first of these two
signals within the composite E5 signal is the E5a, centered at
TABLe 2 3.2 Key Galileo Signals and Parameters 1176.450MHz (same as the GPS L5). The E5a signal has again
two signal components, transmitted in quadrature, where one
Signal Frequency (MHz) Component Date Service
has data (in-phase) and other does not (quadrature). The F/NAV
E1 1575.420 E1-B I/NAV OS/CS/SOL data is encoded onto the in-phase E5a channel at a 50 sps rate.
E1-C Pilot The second of these two signals within the composite E5 signal
E6 1278.750 E6-B C/NAV CS is the E5b, centered at 1207.140MHz. The E5b signal has two
E6-C Pilot signal components, transmitted in quadrature, where one has
E5 1191.795 data (in-phase) and other does not (quadrature). The in-phase
E5a 1176.450 E5a-I F/NAV OS
channel has the I/NAV message format to support OS, CS, and
E5a-Q Pilot
SOL applications. The data encoded within the I/NAV format
E5b 1207.140 E5b-I I/NAV OS/CS/SOL
will contain important integrity information necessary to sup-
E5b-Q Pilot
port SOL applications.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems Theory and Practice 491
23.4.3 Updates 23.5.2 Frequency
The first two Galileo in-orbit validation (IOV) satellites were The nominal carrier frequency is 1561.098MHz with B1 signal
launched in October 2011 and began broadcasting in December is currently a quadrature phase-shift key (QPSK) in modulation.
2011. All Galileo signals were activated on December 17, 2011, Compass now has 10 BeiDou-2 satellites operating in its constel-
simultaneously for the first time across the European GNSS lation (BeiDou 2011).
systems three spectral bands known as E1 (15591592MHz), E5
(11641215MHz), and E6 (12601300MHz). The remaining two
IOVs were launched in 2012. Current updates are available at
23.6 conclusions
ESA website (European Space Agency 2014). Future developments will provide enhanced safety and services
for users. Most future users are anticipated to rely on multicon-
stellation receivers, that is, that will receive and process GPS,
23.5 compass (BeiDou-2) GLONASS, Galileo, and Compass signals, complementing and
The BeiDou Navigation Satellite System is being developed, enhancing each other. A GNSS data integrity channel (GIC) will
starting with regional services, and later expanding to global ser- be provided in the next generation of satellites; for example, in
vices. Phase I was established in 2000, and Phase II provided for GPS, the GPS IIF and GPS III will provide an integrity channel.
areas in China and its surrounding areas by 2012. Also known In addition, next generations will include airborne monitoring
as Compass and BeiDou-2, the Chinese BDS started operations by using redundant measurements to provide signal integrity.
in December 2012 and has 14 active satellites in service over New satellites to be built and launched will have more frequencies,
the Asia-Pacific region available to general users. When fully higher power, and larger bandwidths. These improvements will
deployed by 2020, BDS (Phase III) is expected to comprise a total reduce the errors such as multipath and atmospheric (ionospheric
of 35 satellites offering complete coverage around the globe (GPS and tropospheric). The support for various services, including
World Staff 2013). open service (OS), commercial service (CS), public regulated ser-
vice (PRS), SoL service, and SAR, will be supported with different
signal structures and encoding formats, tailored to each service.
23.5.1 compass Satellites Global navigation satellite systems will continue to make the
Compass is the Chinese developed GNSS, where Compass is an world a smaller, more fully functional planet. Each of the exist-
English translation for BeiDou. The Chinese Government per- ing systems will individually improve and seek engagement with
formed initial development on the Compass in the 20002003 its neighboring systems to provide a seamless overlay of safe and
timeframe with reference to BeiDou (BD)-1. BeiDou consists of reliable service around the globe.
14 satellites, including 5 GEO satellites and 9 non-geostationary
earth orbit (Non-GEO) satellites. The Non-GEO satellites References
include four medium earth orbit (MEO) and five inclined geo-
synchronous satellite orbit (IGSO) satellites. Barker, B. C., J. W. Betz, J. E. Clark, J. T. Correia, J. T. Gillis, S.
The Compass space segment is called a Dippler Constellation Lazar, K. A. Rehborn, and J. R. Straton. Overview of the
that plans for an eventual 27 MEO, 5 GEOs, and 3 Inclined GEOs GPS M Code Signal. January 2000. Proceedings of the 2000
(IGSO) satellites. The GEOs are planned for longitude locations National Technical Meeting of The Institute of Navigation,
at: 58.75E, 80E, 110.5E, 140E, and 160E. The system uses its Anaheim, CA, pp. 542549.
own datum, China Geodetic Coordinate System 2000 (CGCS BeiDou Satellite Navigation System News Center. 2014. Article
2000) and time reference BeiDou Time (BDT) System, relatable on first anniversary of BeiDou navigation satellite system
to UTC. Of the BD-2 SVs, the first MEO Compass M-1 satellite providing full operational regional service. Available at
was launched on April 14, 2007. The GEOs have been launched http://en.beidou.gov.cn/ (accessed February 2014).
with an orbital radius of 42,164.17km. The first IGSO SV was Betz, J. W., M. A. Blanco, C. R. Cahn, P. A. Dafesh, C. J. Hegarty,
launched on July 31, 2010, with an inclination angle of 55 (same K. W. Hudnut, V. Kasemsri et al. September 2006.
as GPS MEOs). In December 2012, the BeiDou Navigation Description of the L1C signal. Proceedings of the 19th
Satellite System provided fully operational regional service International Technical Meeting of the Satellite Division of
(BeiDou 2014). The Institute of Navigation (ION GPS 2006), Fort Worth,
The current Interface Control Document (ICD) specifies a B1 TX, pp. 20802091.
GNSS signal at a center frequency of 1561.098MHz. The archi- China Satellite Navigation Office. December 2011. BeiDou
tecture is based on a CDMA approach. The spreading codes used Navigation Satellite System, Signal In Space, Interface
are Gold codes, based on 11 stages of delay, running at 2.046 Control Document (Test Version), December 2011, avail-
Mcps, so the code will repeat after 1 ms. Navigation data on the able at http://www.beidou.gov.cn/attach/2011/12/27/2011
MEO and IGSO SVs is at a rate of 50 bps, with a secondary code 12273f3be6124f7d4c7bac428a36cc1d1363.pdf, visited July
rate of 1kbps (China 2011). 15, 2012.
2016 by Taylor & Francis Group, LLC
492 Remotely Sensed Data Characterization, Classification, and Accuracies
European Space Agency. 2014. About the European GNSS Grewal, M. S., A. P. Andrews, and C. G. Bartone. 2013. Global
Evolution Programme. http://www.esa.int/Our_Activities/ Navigation Satellite Systems, Inertial Navigation, and
Navigation/GNSS_Evolution/About_the_European_ Integration. John Wiley & Sons, New York.
GNSS_Evolution_Programme (accessed February 10, 2014). Hegarty, C. and A. J. Van Dierendonck. 1999. Civil GPS/WAAS
European Union, European Commission, Satellite Navigation, signal design and interference environment at 1176.45
Galileo Open Service. September 2010. Signal-in-space MHz: Results of RTCA SC159 WG1 activities. Proceedings
interface control document, OS SIS ICD, Issue 1.1. Available of the 12th International Technical Meeting of the Satellite
at http://ec.europa.eu/enterprise/policies/satnav/galileo/ Division of The Institute of Navigation (ION GPS 1999),
files/galileo-os-sis-icd-issue1-revision1_en.pdf (accessed September, Nashville, TN, pp. 17271736.
February 10, 2014). Issler, J. L., L. Ries, J. M. Bourgeade, L. Lestarquit, and C.
ESA, Galileo, 2012, available at http://www.esa/NA/galileo.html, Macabiau. 2004. Probabilistic Approach of Frequency
visited July 15, 2012. Diversity as Interference Mitigation Means. Proceedings
FAA, GNSS Evolutionary Architecture Study, Phase IPanel of the 17th International Technical Meeting of the Satellite
Report, February 2008, available at http://www.faa.gov/ Division of The Institute of Navigation (ION GPS 2004),
about/office_org/headquarters_office/ato/service_units/ September, Long Beach, CA, pp. 21362145.
techops/navservices/gnss/library/documents/media/GEAS_ Janky, J. M. May 3, 1997. Clandestine location reporting by a
PhaseI_report_FINAL_15Feb08.pdf, visited July 14, 2012. missing vehicle. U.S. Patent 5629693.
Fontana, R. D., W. Cheung, P. M. Novak, and T. A. Stansell. 2001. Logsdon, T. 1992. The NAVSTAR Global Positioning System. Van
The new L2 civil signal. Proceedings of the 14th International Nostrand Reinhold, New York, pp. 190.
Technical Meeting of the Satellite Division of The Institute of Revnivykh, S. 2011. GLONASS Status and Modernization.
Navigation (ION GPS 2001), September, Salt Lake City, UT, Proceedings of the 24th International Technical Meeting of
pp. 617631. the Satellite Division of The Institute of Navigation (ION GPS
Global Navigation Satellite system (GLONASS). 2008. Interface 2011), September, Portland, OR, pp. 839854.
Control Document L1, L2, Version 5.1. Russian Institute of Rockwell International Corporation. July 3, 1991. GPS inter-
Space Device Engineering, Moscow, Russia. Available at face control Document ICD-GPS-200. Satellite Systems
www.glonass-ianc.rsa.ru/en/ (accessed February 10, 2014). Division, Revision B.
GPS Directorate, Systems Engineering & Integration Interface Septentrio. April 12, 2011. Septentrios AsteRx3 receiver tracks
Specification. 2011. Navstar GPS space segment/user seg- first GLONASS CDMA signal on L3. Available at http://
ment L1C interface, IS-GPS-800B.pdf. Available at http:// www.insidegnss.com/node/2563 (accessed February 10,
www.navcen.uscg.gov/pdf/gps/IS-GPS-800B.pdf (accessed 2014).
February 10, 2014). Spilker, J. J, E. H. Martin, and B. W. Parkinson. 1998. A family
GPS Directorate, Systems Engineering & Integration Interface of split spectrum GPS civil signals. Proceedings of the 11th
Specification, IS-GPS-705. 2012. Navstar GPS space seg- International Technical Meeting of the Satellite Division of
ment/user segment L5 interface, IS-GPS-705B. Available The Institute of Navigation (ION GPS 1998), September,
at http://www.navcen.uscg.gov/pdf/gps/IS-GPS-705B.pdf Nashville, TN, pp. 19051914.
(accessed February 10, 2014). Spilker, J. J. and A. J. Van Dierendonck. 2001. Proposed new
GPS World. 2013. CSR location platforms go live with Chinas L5 civil GPS codes. Navigation, Journal of the Institute of
Beidou 2 tracking. Available at http://gpsworld.com/csr- Navigation, 48(3), 135144.
location-platforms-go-live-with-chinas-beidou-2-tracking/ United States Naval Observatory (USNO), GPS Operational
(accessed February 10, 2014). Satellites (Block II/IIA/IIR/IIR-M/II-F), 2012, ftp://tycho.
GPS World Staff. November 2013. New structure for GLONASS usno.navy.mil/pub/gps/gpsb2.txt, visited July 14, 2012.
nav message. Available at http://gpsworld.com/new-structure- Van Dierendonck, A. J. and J. J. Spilker, Jr. 1999. Proposed civil
for-glonass-nav-message/ (accessed February 10, 2014). GPS signal at 1176.45 MHz: In-phase/quadrature codes
Grewal, M. S. 2012. Space-based augmentation for global at 10.23 MHz chip rate. Proceedings of the 55th Annual
navigation satellite systems. IEEE Transactions on Ultrasonics, Meeting of The Institute of Navigation, June, Cambridge,
Ferroelectrics, and Frequency Control, 59(3), 497504. MA, pp. 761770.
2016 by Taylor & Francis Group, LLC
24
Global Navigation Satellite
System Reflectometry for
Ocean and Land Applications
Acronyms and Definitions ................................................................................................................ 493
24.1 Introduction ............................................................................................................................ 493
24.2 Satellite Missions Related to GNSS Remote Sensing......................................................... 494
Kegen Yu 24.3 Ocean Observation ................................................................................................................. 495
Wuhan University Sea Surface Wind Speed Estimation Sea Surface Altimetry
24.4 Land Applications................................................................................................................... 503
Chris Rizos Soil Moisture Forest Change Detection
The University of New South Wales
24.5 Challenging Issues and Future Directions ......................................................................... 505
Andrew Dempster 24.6 Conclusions.............................................................................................................................. 506
The University of New South Wales References............................................................................................................................................ 506
Acronyms and Definitions LNA Low-noise amplifier
NASA National Aeronautics and Space Administration
BRCS Bistatic radar cross section PARIS Passive Reflectometry and Interferometry System
CHAMP Challenging minisatellite payload PDF Probability density function
COSMIC Constellation observing system for meteorology, PNT Positioning, navigation, and timing
ionosphere, and climate PRN Pseudo random noise
CYGNSS Cyclone Global Navigation Satellite System RHCP Right-hand circularly polarized
DMC Disaster monitoring constellation SAR Synthetic aperture radar
ESA European Space Agency SSP Specular scattering point
GEROS-ISS GNSS reflectometry, radio occultation and scat- SSH Sea surface height
terometry onboard the International Space Station STD Standard deviation
GGOS Global geodetic observing system SWH Significant wave height
GNSS Global Navigation Satellite System SWP Significant wave period
GNSS-R GNSS reflectometry TPL Total path length
GNSS-RO GNSS radio occultation UNSW University of New South Wales
GPS Global positioning system
GRACE-A Gravity recovery and climate experiment 24.1 introduction
IAG International Association of Geodesy
IERS International Earth Rotation and Reference Investigating the use of reflected Global Navigation Satellite
Systems Service System (GNSS) signals for remotely sensing the earths surface
IGS International GNSS Service was initiated about two decades ago (Martn-Neira 1993). Remote
ILRS International Laser Ranging Service sensing based on processing and analyzing reflected GNSS sig-
IMU Inertial measurement unit nals is commonly termed GNSS reflectometry (GNSS-R). When
ITRF International Terrestrial Reference Frame using the GNSS-R technique to build a remote sensing system,
IVS International VLBI Service for Geodesy and only the receiver needs to be designed and manufactured. The
Astrometry receiver platform (static or mobile; land-based, aircraft, or sat-
LEO Low earth orbit ellite) needs to be selected based on the specific application. In
LHCP Left-hand circularly polarized the case of an aircraft or satellite platform, the direct signal is
493
2016 by Taylor & Francis Group, LLC
494 Remotely Sensed Data Characterization, Classification, and Accuracies
received via a zenith-looking right-hand circularly polarized 24.2 Satellite Missions Related
(RHCP) antenna, while the reflected signal is received through
a nadir-looking left-hand circularly polarized (LHCP) antenna.
to GnSS Remote Sensing
The reason for such an antenna selection is that the GNSS signals Useful reviews of radio occultation missions can be found
are designed as RHCP; however, when reflected over a ground in Anthes (2011) and Jin et al. (2011). The concept was first
surface, they are changed to be LHCP. In the case of a land-based proved by the GPS/MET (GPS/Meteorology) satellite mission
platform, either two antennas are used to receive the direct and in 19951997. That satellite took few measurements but was
reflected signals separately or a single antenna is used to capture followed by the more productive CHAllenging Minisatellite
both the two signals. Payload (CHAMP) and Satellite de Aplicaciones Cientificas-C
The GNSS signals are always available, globally, and the sig- (SAC-C) satellites. CHAMP provided 8years of radio occul-
nal structures are typically well known, except for those dedi- tation data consisting of around 440,000 measurements
cated to military use. The L-band GNSS signals are sensitive from February 2001 to October 2008 (Heise et al. 2014).
to ground surface parameters so that they can be utilized for Those missions led to the launch in 2006 of a six-satellite
remote sensing purposes. Recently, there has been an increase constellation FORMOSAT-3 (Formosa Satellite Mission #3)/
in such investigations by academia and research institutions, COSMIC (Constellation Observing System for Meteorology,
partly because this innovative use of the GNSS signals has Ionosphere, and Climate), which provided 15002000 sound-
many potential applications. In particular, space agencies such ings per day, resulting in GPS-RO becoming an operational
as NASA and ESA have already funded, or are going to fund, a data source for weather prediction and ionospheric moni-
number of projects/missions that focus on the applications of toring. Other missions that provided significant quantities
the GNSS-R. The Cyclone Global Navigation Satellite System of RO data are GRACE-A (Gravity Recovery and Climate
(CYGNSS) project is just one example (http://aoss-research. Experiment), METOP-A (METeorological Operations), C/
engin.umich.edu/missions/cygnss/), which aims to develop a NOFS (Communications/Navigation Outage Forecasting
system using a constellation of eight microsatellites to improve System), TerraSAR-X, and TanDEM-X. Table24.2 (taken from
hurricane forecasting especially with regard to the storm inten- the COSMIC website: http://www.cosmic.ucar.edu/) shows
sity. Another example is the ESAs passive reflectometry and the contributions made by these and other missions to both
interferometry system (PARIS) project (http://www.esa.int/ atmospheric and ionospheric sounding. With the COSMIC
Our_Activities/Technology/PARIS). PARIS can be used as a constellation degrading, as it has reached its design life, a new
passive radar altimeter. Different from current radar altimeters, FORMOSAT-7/COSMIC-2 constellation is being constructed
PARIS would measure multiple samples from different tracks that has multifrequency, multi-GNSS receivers on board 12
and rapidly form images of mid-sized (mesoscale) phenomena satellites to be launched6 into low inclination orbits in 2015
such as ocean currents or tsunamis. Geophysical parameters and 6 into high inclination orbits in 2018 (http://www.cosmic.
that can be measured using a GNSS-based reflectometry system ucar.edu/cosmic2/).
include those (but not limited to) listed in Table24.1 (Garrison Far fewer satellites have been able to perform the GNSS-R.
etal. 2002, Gleason etal. 2005, Gleason 2006, Font etal. 2010, The first was UK-DMC from 2003 to 2011 (Unwin et al.
Yu etal. 2012a,b). 2003). Others that have been proposed are Techdemosat-1
This chapter focuses on the GNSS-R for ocean applications due for launch Q1 2014 (http://www.sstl.co.uk/Missions/
including sea surface wind speed estimation and sea surface TechDemoSat-1) with a custom reflectometry payload, PARIS, a
altimetry and for land applications including soil moisture
retrieval and forest change detection. In addition, past and
planned satellite missions associated with the GNSS-R as well TABLe 2 4.2 Contributions to Atmospheric and Ionospheric
Sounding by Mission, from the COSMIC Website
as the GNSS radio occultation (GNSS-RO) are summarized.
Finally, some challenging issues related to the GNSS remote Mission Total Atm Occs Total Ion Occs
sensing products and services are addressed. CHAMP 399,968 303,291
CNOFS 120,588 0
COSMIC 4,039,311 3,707,966
TABLe 2 4.1 Examples of Ocean and Land Applications of GNSS-R GPSMET 5,002 0
Ocean Applications Land Applications GPSMETSA 4,666 0
GRACE 273,013 132,817
Ocean wind speed and direction estimation Soil moisture retrieval
METOPA 993,084 0
Sea surface altimetry Biomass density estimation
SACC 353,808 0
Tropical cyclone intensity estimation Forest change detection
TSX 276,549 0
Sea surface salinity estimation Land surface classification
Total 6,465,989 4,144,074
Sea wave height characteristic estimation Snow depth estimation
Source: Anthes, R.A., Atmos. Measure. Techn., 4, 1077, 2011, last updated:
Oil slick detection Vegetation height estimation
December 7, 23:25:02 MST 2013.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite System Reflectometry for Ocean and Land Applications 495
Direct signal
for positioning
LEO satellite
Refracted signal
Reflected signal for occultation
for reflectometry
GNSS satellite
FIg u r e 2 4.1 Illustration of the geometry of the direct, reflected, and refracted GNSS signals and their use for different applications.
dedicated mission for GNSS-R (Martin-Neira etal. 2009, 2011), 24.3.1.1 Sea Wave Spectrum
CYGNSS, a constellation of eight microsatellites, GEROS-ISS Sea surface undulation is a complex process, and sea wave
(GNSS REflectometry, Radio Occultation and Scatterometry heights change randomly in time and space. Sea surface rough-
onboard International Space Station) (http://www.ice.csic.es/ ness can be described by a number of parameters including
en/view_project.php?PID=155), and Microsat-1 (Camps et al. significant wave height (SWH) and significant wave period
2011). Several of the authors are also from the University of (SWP). The SWH is defined as the average height of the one-
New South Wales (UNSW), which is preparing a CubeSat to third highest waves, and the SWP is defined as average period of
carry both GNSS occultation and reflectometry experiments the waves used to calculate the SWH. Alternatively, wave height
as part of the QB50 mission (http://www.acser.unsw.edu.au/ spectrum and wave direction spectrum can be used to describe
QB50/index.html). Figure24.1 illustrates that a launched LEO the surface roughness. Among the wave height spectral models,
satellite can receive the direct, reflected, and refracted GNSS the PiersonMoskowitz model, the JONSWAP model, and the
signals to enable positioning, reflectometry, and occultation Elfouhaily model are widely studied (Pierson and Moscowitz
applications. 1964, Elfouhaily etal. 1997).
The Elfouhaily model that describes the wind-driven wave
height spectrum is defined as
24.3 ocean observation
This section discusses two specific GNSS-R-based ocean applica- 4
W (, ) = (B () + Bh ())(1 + ()cos(2( 0 )) (24.1)
tions: sea surface wind speed estimation and sea surface height 2
(SSH) estimation.
where
is the wave number
24.3.1 Sea Surface Wind Speed estimation is the azimuth angle
Wind speed retrieval using the GNSS-R is mainly based on the 0 is the wind direction
theoretical model of the received signal power reflected over
sea surface. Since the wind speed is estimated by observing the The long-wave curvature spectrum B () is defined as
wind-driven sea surface wave slope characteristics, the estima-
tion is reliable only when the wind and wave interaction has 3 103U 10 5 p 2
reached the steady state after the wind has continuously blown B () = exp 1 exp
c 10 p 4
the surface in one direction for at least say half an hour. The
(24.2)
wind speed estimation accuracy can be better than 2 m/s based
on processing the data observed from several GNSS-R airborne where U10 is the wind speed at a height of 10 m above the sea. Note
experimental campaigns. Some details of the technique are that the wind speed at a height of 19.5 m above the sea is related
described as follows. to U10 by U19.5 = 1.026U10, showing little difference between wind
2016 by Taylor & Francis Group, LLC
496 Remotely Sensed Data Characterization, Classification, and Accuracies
speeds within the vicinity of these heights. is the inverse wave where the wave number cutoff * can be calculated according to
age, which is equal to 0.84 for a well-developed sea (driven by
wind). p is the wave number of the dominant waves defined as 2
* = (24.9)
3
g 2
p = 2
(24.3)
U 10 where is the wavelength (0.1904 m for the GPS L1 signal). In
Garrison etal. (2002), the wave number cutoff is modified as
where g is gravity. The other three parameters in (24.2) are
defined as 2 sin()
* = (24.10)
3
g (1 + ( / m )2 )
c = , m = 370
where is the incidence angle (complementary to the eleva-
tion angle of the satellite). Some details about how to simulate
1.7, 0.83 < < 1
= (24.4) this model and the associated simulation codes can be found in
1.7 + 6 log(), 1< < 5 Gleason and Gebre-Egziabher (2009).
2
1
= exp 2 1 , = 0.08(1 + 4 3 ) 24.3.1.2 Sea Surface Scattering
2 p
As the GNSS signals arrive at the sea surface, some of the signal
energy is absorbed by the seawater, while the other energy is
The short-wave curvature spectrum Bh() in the Elfouhaily reflected. The reflected signals that travel toward the receiver
model is defined as will be captured by the nadir-looking antenna. Let the posi-
tions of the transmitter (on the GNSS satellite) and the receiver
(on an aircraft, LEO satellite, or land-based) be (xt,yt,zt) and
c m m 1
2
Bh () = exp 1 (24.5) (xr,y r,zr) respectively. Also define the position of the scattering
2c 4 m point on the sea surface as (x s,ys,z s). Then, the distance from
the transmitter through the scattering point to the receiver is
where cm = 0.23,and the parameter m is determined by given by
2 uf dtsr (x s , y s , z s ) = (x t x s )2 + (y t y s )2 + (z t z s )2
10 1 + ln , u f < cm
cm + (x r x s )2 + (y r y s )2 + (z r z s )2 (24.11)
m = (24.6)
2 uf
10 1 + 3 ln c m , u f cm
The specular scattering point (SSP) is the scattering point
(xSSP,ySSP,zSSP) on the surface where the distance dtsr is mini-
where uf is the friction velocity that can be iteratively computed by mal. With respect to the signal reflected at the SSP, the signals
reflected at other scattering points arrive at the receiver with a
1 delay given by
10 6 1 0.48u 3f
u f = 0.4U 10 ln
, b (u f ) = 0.11 14 10 u f +
b(u f ) gU 10 dtsr (x s , y s , z s )
= (24.12)
c c
(24.7)
where
The initial value may be chosen as 103 (0.81 + 0.065U 10 )U 10 .
c is the speed of light
The mean square slopes of the surface in the upwind direction
c = dtsr(xSSP,ySSP,zSSP)/c
and in the cross-wind direction are then respectively calculated by
Given the transmitter and receiver positions and the delay , the
*
scattering points define an ellipse on the surface. That is, at each
mss x =
cos W (, )dd,
2 2
specific delay, the signals reflected on the ellipse will arrive at the
0 receiver at the same time, supposing that they travel toward the
* nadir-looking antenna. Figure 24.2 shows an example of the iso-
mss y =
sin W (, )dd (24.8)
2 2
delay when the satellite elevation angle is 63. The velocity vector
0 of the aircraft is (21.166946, 52.149224, 2.527502 m/s) and the
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite System Reflectometry for Ocean and Land Applications 497
Delay (chip) Doppler frequency (Hz)
100 900
15 15
90
10 10 1000
80
5 70 5
1100
Width (km)
Width (km)
60 0
0
50 1200
5 5
40
1300
30 10
10
20 15 1400
15
10
20
20 20 15 10 5 0 5 10 15
20 15 10 5 0 5 10 15
Length (km)
Length (km)
FIg ur e 24.3 Example of an iso-Doppler map for the case where the
FIg ur e 24.2 Example of an iso-delay map for the case where the
receiver is carried by an aircraft at a speed of 202.8km/h and an altitude
receiver is carried by an aircraft at a speed of 202.8km/h and an altitude
of 508 m.
of 508 m.
receiver position is (33.693117, 151.2745950, 508.0884 m) in code acquisition and tracking. The carrier frequency of the IF
the WGS84 system. The satellite position is (52.3194, 162.2910, signals is then compensated for, and the resulting baseband
1.9903e+007 m) and its velocity vector is (131.75275, 2727.07856, signal is correlated with a replica of the PRN code related
573.77554 m/s). to a specific satellite. At the central Doppler frequency, the
Due to the relative movement between the transmitter and cross-correlation with a sequence of code phases produces
the receiver, Doppler frequencies are produced, resulting in a delay waveform (correlation power versus code phase).
the increase or decrease in the signal carrier frequency. Let the
Adelay-Doppler waveform is produced when both a sequence
velocity vectors of the transmitter and the receiver be Vt and Vr of Doppler frequencies and a sequence of code phases are
respectively. The Doppler frequency is determined by considered.
Theoretically, the signal power with respect to a code phase
(Vt m Vr n ) and a Doppler frequency can be computed by Zavorotny and
fD = (24.13)
Voronovich (2000):
where
Ti 2 2Pt Gt G r 0
m and n are the unit vectors of the incident wave and the
reflected wave respectively
Y (, f D ) =
(4)2 R R
A
2
ts
2
sr
2 ( c ) sinc 2 (( f D f c )Ti )dA
denotes the vector dot product (24.14)
For a given Doppler frequency, Equation 24.13 represents a
where
hyperbola. That is, the signals reflected on such a hyperbola will
Ti is the coherent integration time
have the same Doppler frequency. Figure 24.3 shows an example
Pt is the transmit power
of the iso-Doppler map under the same circumstance as the iso-
is the atmospheric attenuation
Doppler map depicted in Figure 24.2. The intersections of the
Gt and Gr are the antenna gains of the transmitting and
iso-delay lines and the iso-Doppler lines form a network of grids
receiving antennas, respectively
that will be used to determine the power of the reflected signals
Rts and Rsr are the distance from the transmitter to the scat-
arriving at the receiver for each pair of relative propagation delay
tering point and the distance from the scattering point to
and Doppler frequency.
the receiver, respectively
(c) is the triangle correlation function of the PRN code
24.3.1.3 Reflected Signal Power with the delay of the replica code
The reflected signals received via the nadir-looking antenna c is the delay of the received code
are first down-converted to IF signals. The code phase off- sinc() is the sinc function representing the attenuation
set or delay, and the Doppler frequency associated with the caused by Doppler misalignment with fc the Doppler fre-
satellite of interest can be estimated based on processing the quency of the received signal
direct signal received via the zenith-looking antenna through f D is the Doppler frequency of the replica signal
2016 by Taylor & Francis Group, LLC
498 Remotely Sensed Data Characterization, Classification, and Accuracies
A is the effective scattering surface area, and the bistatic radar different flight heights were tested: 0.5, 2, 5, and 10km. In the
cross section 0 can be calculated according to case of 0.5km altitude, the six curves are nearly identical, and
the spread versus time is rather small. Since the waveforms are
insensitive to the wind variation when the flight altitude is less
| |2 q
0 = 4 p (24.15) than 0.5km, it would be inappropriate to use the trailing edge to
(q z q ) qz perform any sea state or wind speed retrieval. As the flight alti-
tude increases, the signal spread increases, and the waveforms
where distinguish from each other better. Thus, more accurate param-
is the polarization-dependent Fresnel reflection coefficient eter estimates would be expected.
q is the scattering unit vector that bisects the incident vector
24.3.1.4 example
and the reflection vector
q and q z are the horizontal and vertical components of q Figure 24.5 shows the five theoretical delay waveforms corre-
respectively sponding to five different wind speeds (3, 4, 5, 6, and 7 m/s) and
p() is the probability density function of the surface slope, the measured delay waveform associated with a specific satel-
which may be simply assumed as omnidirectional lite. The data were collected during an airborne experiment at
Gaussian distribution an altitude of about 3km. The measured waveform is produced
through coherent integration of 1 ms IF signals and then nonco-
When performing the double integration in (24.14), the size of herent integration of 1000 such 1 ms waveforms. It can be seen
the effective scattering area should be appropriately selected. As that the measured waveform has a good match with the theoreti-
the flight height increases, the scattering area increases accord- cal waveform of wind speed 4 m/s, which is a good estimate of
ingly. Nevertheless, it is not necessary to make the area dimen- the real wind speed, which is about 4.3 m/s.
sions too large so as to reduce computational complexity. The This is just an illustrative example to show how the wind
contribution of the reflected signals beyond the effective scat- speed is estimated. In practice, a mathematical approach will
tering area will be negligible due to the limited antenna beam be employed to automatically produce estimation solutions.
width and the fact that the power is inversely proportional to the Regarding this model-matching approach, a certain number
squared distance between the GNSS satellite and the scattering of theoretical waveforms are produced, and a cost function is
point and between the scattering point and the receiver. defined. The theoretical waveform with the minimal cost func-
Figure 24.4 shows the decibel delay waveforms based on the tion is then selected, and the corresponding wind speed is the
theoretical models studied earlier. Six curves correspond to estimate of the real wind speed. The cost function can be defined
the six different wind speeds (4, 7, 10, 13, 16, and 19 m/s). Four as the sum of the squared difference between the theoretical and
Flight height = 0.5 km Flight height = 2 km
0 0
Normalized power (dB)
Normalized power (dB)
10 10
20 20
19 m/s
4 m/s 19 m/s 4 m/s
30 30
0 2 4 6 8 0 2 4 6 8
Flight height = 5 km Flight height = 10 km
0 0
Normalized power (dB)
19 m/s
10 10
4 m/s 19 m/s
4 m/s
20 20
30 30
0 2 4 6 8 0 2 4 6 8
A/C code phase (chip) A/C code phase (chip)
FIg ur e 24.4 Normalized correlation powers of the simulated reflected signals using the Elfouhaily wave elevation model with four different
flight heights and five different wind speeds (4, 7, 10, 13, 16, and 19 m/s).
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite System Reflectometry for Ocean and Land Applications 499
Mean speed estimates (m/s)
0 2.5 m/s
5
3 m/s
1 3.5 m/s 4.8
Normalised correlation power (dB)
4 m/s
2 4.5 m/s 4.6
Measured 4.4
3
4.2
4 0 50 100 150 200 250 300 350
5 0.9
Error STD (m/s)
6 0.8
0.7
7
0.6
8
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0.5
0 50 100 150 200 250 300 350
Time (C/A code chip)
Wind direction (deg)
FIg ur e 24.5 Wind speed estimation through matching the mea-
FIg ur e 24.6 Mean wind speed estimate versus assumed wind direc-
sured delay waveform with a number of theoretical waveforms.
tions using the delay waveform matching method.
measured waveforms, that is, least-squares fitting. However,
the smallest STDs may be treated as the actual wind direction.
this method requires the alignment of the two waveforms so
However, the estimation accuracy is rather low, and there exists
that the difference between the two waveforms is minimized.
an ambiguity of 180. More accurate wind direction estimation
Alternatively, a slope-based method proposed in Zavorotny
may be realized by using the properties of the delay Doppler map
and Voronovich (2000) can be employed. However, as observed
as reported in Valencia etal. (2014).
in Figure 24.5, it is rather difficult to distinguish the slopes of
neighboring curves related to different wind speeds. The area-
based waveform fitting method may be a more suitable option to 24.3.2 Sea Surface Altimetry
estimate the wind speed with the following multistep procedure
24.3.2.1 Sea Surface Height calculation
(Yu etal. 2012c):
As shown in Figure 24.7, SSH is calculated relative to the surface
1. Interpolating both the measured and theoretical wave- of the theoretical earth ellipsoid in the WGS84 system, which
forms without changing the original data; when using has zero altitude. The SSH at a specific sea surface point is the
MATLAB, the library function INTERP can be distance from the point to the WGS84 earth ellipsoid surface,
directly used to perform the interpolation.
2. Selecting the cutoff correlation power to retain the wave-
form above certain power lever so that the slope of the
trailing edge does not change abruptly.
3. Calculating the areas of the interpolated waveforms Receiver
above the cutoff power and calculating the area difference
between the measured waveform and each of the theoreti-
cal waveforms.
4. Selecting the theoretical waveform that produces the min-
From GNSS satellite
imum area difference and taking the corresponding wind
speed as the estimate.
h
Figure 24.6 shows the average of wind speed estimates asso-
1 2
ciated with eight satellites when 16 different wind directions Rough sea surface
are assumed. One observation is that although the mean wind
speed estimate varies with the assumed directions, the varia- Actual MSL S
tion is only about 0.53 m/s. That is, in this case, the variation is
not significant. Also shown is the standard deviation (STD) of
the wind speed estimation versus the assumed wind directions. WGS84 surface with zero altitude
The STD ranges from 0.52 to 0.9 m/s. The true wind direction
is about 110, so this STD plot provides some useful informa- FIg ur e 24.7 Geometry of the receiver, WGS84 earth ellipsoid sur-
tion about the wind direction. That is, the wind direction with face, rough sea surface, actual MS, and direct and reflected signal paths.
2016 by Taylor & Francis Group, LLC
500 Remotely Sensed Data Characterization, Classification, and Accuracies
Initial SSH guess
discussed later. Then, as indicated in Figure 24.7, the calculated
TPL by (24.16) is compared with the measured TPL by (24.18)
to determine whether the tentative SSH should be increased or
SPP calculation decreased. The process is terminated once the difference between
Case 1:
the two TPLs is smaller than the predefined threshold.
RtSr RtSr th To reduce the computational complexity, a simpler technique
Measurement SPP-based
based TPL TPL Case 2: may be used. For instance, if RtSr > R tSr , the tentative surface
calculation calculation RtSr RtSr th height is increased by a relatively larger increment such as 40 m.
Case 3: At the next iteration of the outer loop, if RtSr < R tSr , the increment
TPL comparison |RtSr RtSr |< th is decreased by half of the previous increment. In this way, the
process will quickly converge to the steady state. Note that the
Decreasing
specular reflection must satisfy Snells law, that is, the two angles
Case 1 Case 2 Increasing
tentative SSH tentative SSH (1 and 2 in Figure 24.7) between the incoming wave and the
Case 3
reflected wave, separated by the surface normal, must be equal
or the difference is extremely small. Thus, the results should be
Final SSH estimate tested to see if this law is satisfied.
From (24.16), the partial derivatives with respect to the coor-
FIg ur e 24.8 Flowchart for two-loop iterative SSH calculation. th is dinates of the specular point can be determined as
a small positive number.
RtSr uS x r uS x t
and the mean SSH is the average of a large number of these dis- = + , uS {x SSP , y SSP , z SSP } (24.20)
uS RrS RtS
tances over an area of interest.
Figure 24.8 shows the flowchart of how to calculate the SSH which can be rewritten in a vector form as
using a two-loop iterative method. Specifically, for a given ten-
tative SSH, since both the GNSS satellite position (xt,yt,zt) and
R S T S
the receiver position estimates (x r , y r , z r ) are known, the SSP dS = + (24.21)
RrS RtS
(xSSP , ySSP , zSSP ) on the tentative sea surface can be readily deter-
mined. The SSP estimation is realized by minimizing the total
where S , R, and T are the position vectors of the SSP, the receiver,
path length (TPL) from the satellite through the SSP and to the and the transmitter, respectively. Equation 24.21 is used to gen-
receiver, which is given by erate an iterative solution to the minimum path length. That is,
at time instant n + 1, the SSP position is updated according to
RtSr = RtS + RSr (24.16)
Sn +1 = Sn + dS (24.22)
where
where is a constant that typically should be set as a larger value
RtS = (x t xSSP )2 + (y t ySSP )2 + (z t zSSP )2
as the flight altitude increases. The initial guess of the SSP can be
(24.17)
simply the projection of the receiver position onto the surface.
RSr = (xSSP x r )2 + (ySSP y r )2 + (zSSP z r )2
At each iteration, a constraint must be applied to restrain the
SSP on the surface that is meters above or below the WGS84
The TPL can also be estimated using the propagation time (Rtr) ellipsoid surface, which has a zero altitude if is a positive or
of the direct signal and the relative delay (rd) of the reflected negative number. That is, the SSP position is scaled according to
signal, that is,
Sn +1
R tSr = R tr + c rd (24.18) Sn +1 = (rS + ) (24.23)
| Sn +1 |
where
where the radius of the earth at the specular point is calculated by
c is the speed of light
rd is the estimated relative delay
Rtr is the estimate of Rtr calculated by 1 eWGS
2 z
rS = aWGS 84 84
, S = arcsin S (24.24)
1 eWGS 84 (cos S )
2 2 |S |
R tr = (x t x r )2 + (y t y r )2 + (z t zr )2 (24.19)
where
where the satellite position is assumed error free, while the receiver eWGS84 = 0.08181919084262
position is an estimate. How to estimate the relative delay will be aWGS84 = 6,378,137 m
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite System Reflectometry for Ocean and Land Applications 501
Since the altitude of the WGS84 earths surface is zero, the Therefore, the measured relative delay should be calibrated by
WGS84 altitude of the SSP is equal to . Clearly, the altitude of subtracting the offset from itself. The lengths of the two cables
a single SSP cannot be treated as the estimate of the mean SSH. can be readily measured in advance. In case where an LNA is
However, a reasonable estimate of the mean surface height will used to amplify the signal such as captured by the nadir-looking
be produced through the generation and subsequent processing antenna, the path length from D to R will be the sum of the two
of the altitude estimates of many SSPs over a period of time. cable lengths plus the distance between the two connection
points of the LNA. The distance from U to C and that from D
24.3.2.2 calibration to C can be manually determined in advance. The positions of
Since the zenith-looking and nadir-looking antennas, the receiver, points U and C are estimated by the receiver and frame trans-
and the reference point are not in the same position, it is necessary formation; thus, distances AC and AU can be readily calculated.
to calibrate the relative delay measurements to remove the effect Calculation of the distances SC and SD requires a knowledge of
of these position differences. Note that the reference point may be the SSH, which is unknown in advance. However, initial infor-
set at the center of the inertial measurement unit (IMU) provided mation about the SSH or previous SSH estimation results can
that such a device is used. Figure 24.9 illustrates the configuration be exploited. The uncertainty in the SSH estimate will affect the
of the devices. The two antennas are connected to the receiver via calculation of both SC and SD in a very similar way, so that a
two cables whose lengths are LCR and LCD. The actual measure- small SSH error will have a negligible impact. Thus, distance CV
ment of the relative delay of the reflected signal is given by can be estimated, and distance SC can be calculated using eleva-
tion angle. Calculation of distance SD requires its orientation to
Lmeasured = AS + SD + LDR (AU + LUR ) + (24.25) determine the angle SDC. In the case where the nadir-looking
antenna is fixed directly beneath the reference point, the dis-
where is the measurement error. On the other hand, using the tance can be simply calculated by
reference point position, the relative delay is calculated as
SD = SC 2 + CD 2 2 SC CD cos(90 ) (24.28)
Lcalculated = AS + SC AC (24.26)
Once the distances and the cable lengths in (24.27) are known,
As described earlier, the SSH is estimated by comparing the the relative delay offset can be readily determined. For instance,
measured and calculated relative delays. However, typically, suppose that points U and D are directly above and below point
there is an offset between the calculated and measured delays C respectively; AC = 20,000km; satellite elevation angle is 50;
even in the absence of measurement error. That is, UC = 0.8 m; CD = 0.4 m; CV = 300 m; LUR = 1 m; and LDR = 0.6m.
Then, the unknown distances are obtained from some simple cal-
Loffset = Lmeasured Lcalculated = (SC SD ) + (AU AC ) + (LUR LDR ) culations: AU = 19,999.999387km, SD = 391.32 m, SC = 391.62m.
(24.27) As a result, the relative relay offset is calculated to be 0.094 m.
where the measurement error is ignored. The SSH estimation error caused by this relative delay offset can
be approximated as
Loffset
= 6.1 cm (24.29)
A 2 sin
U Zenith-looking
GNSS satellite
antenna
Thus, when the configuration of the devices is arranged properly
C Reference point and the cable lengths are selected based on similar analysis, the
Receiver (IMU)
R SSH error caused by the device configuration will not be large.
However, to achieve accurate altimetry, such an error must be
D Nadir-looking
antenna
compensated for.
24.3.2.3 Relative Delay estimation Methods
The SSH estimation accuracy is largely dependent on the per-
formance of estimation of the TPL of the reflected signal, from
the transmitter through the SSP on the sea surface and to the
receiver. The TPL is equal to the sum of the path length of the
Specular point S V direct signal and the relative delay of the reflected signal. Since
the satellite position is known, the measurement error associ-
FIg ur e 24.9 Geometric relationship between the devices (receiver, ated with the path length of the direct signal comes from the
antennas, and inertial measurement unit [IMU]) mounted on a LEO receiver position estimation error. When the receiver is given,
satellite or an aircraft. such an error is typically not reducible, although smoothing may
2016 by Taylor & Francis Group, LLC
502 Remotely Sensed Data Characterization, Classification, and Accuracies
improve the accuracy marginally. Thus, it is important to use a code chip width (i.e., half of the correlogram triangle width).
GNSS receiver that can achieve satisfactory position estimation Then, the following result exists.
accuracy. When an accurate receiver is used, the relative delay The peak correlation power location of the combined mul-
estimation error would be dominant. Therefore, it is vital to tipath signals will shift from the peak correlation power location
reduce this error. The relative delay is estimated by determining of the first path signal provided that
the code phase of the direct signal and that of the reflected signal
J
based on the measured correlation waveform.
There are two methods associated with the delay waveform P > P
j =2
j 1 (24.30)
based method. The first one makes use of the clean code (C/A
code), while the second one utilizes the interferometry tech- where Pj is the peak correlation power of the jth path signal.
nique. The clean code method deals with the direct signal and Since both the first path signal and signals of other paths
the reflected signal separately and uses only the C/A code to are reflected signals, the signal power of the second path and a
generate the delay waveform. In the interferometry technique, number of following paths can be significant with respect to the
on the other hand, both signals are processed together. That first path. Thus, intuitively, (24.30) would always be valid with
is, the direct signal observed from a zenith-looking antenna is a rough sea surface. The relationship in (24.30) can be proved
cross-correlated against the reflected one captured by a nadir- mathematically as in Yu etal. (2014).
looking antenna. High-gain antennas are needed to achieve In the case where the P(Y) code or the military M-code is
separability among different satellites. The interferometry tech- employed, the peak power location shift can be much smaller
nique is intended to exploit either the P(Y) code or military so that the shift may be ignored. However, when using the C/A
M-code to achieve an accuracy gain at the cost of high gain and code, the shift needs to be taken into account in the presence of a
directional antennas. In addition, a high-bandwidth front end/ rough surface. In the absence of information of accurate surface
receiver is required. wave statistics, the power ratiobased approach to be discussed
In the case where the zenith-looking antenna is high above next may be employed.
the ground, especially when the receiver is mounted on a satel-
lite or on an aircraft, the code phase of the direct GNSS signal 24.3.2.4 Power RatioBased Sea Surface
can be readily estimated by determining the location of the peak Height estimation: An example
power of the correlation waveform. On the other hand, it may The power ratio is defined as the ratio of the correlation power
not be easy to obtain an accurate estimate of the desired code at the desired code phase over the peak correlation power of the
phase of the reflected signal forwarded from a rough sea sur- reflected signal. The desired code phase corresponds to the peak
face. The main reason is that the location of the peak power of power location when the surface is perfectly smooth. Figure24.10
the reflected signal would not be the desired code phase of the is an illustration of the delay waveform of the reflected signal in
reflected signal since the peak power location is shifted due to the presence of a rough surface. C(m) is the peak power of the
rich multipath propagation. Clearly, using the peak power loca-
tion to calculate the relative delay would produce a large bias
error. The time shift or offset would depend on a number of fac- C(m)
tors including the surface roughness and the receiver altitude. C(1)
The peak power location of the delay waveform derivative can
be used as the desired code phase of the reflected signal, but the
estimate would be biased (Rius etal. 2010). It is observed that the
desired code phase of the reflected signal is somewhere between
the peak power location of the delay waveform and that of the
waveform derivative. However, the exact location of the code
phase is typically unknown.
For clarity, it is desirable to explain why the peak power
location could shift when a flat sea surface is replaced with a
rough sea surface. In the presence of a perfect smooth sea sur-
face, the signal will be reflected only at the SSP and then travels
to the receiver. In the presence of a rough sea surface, besides
the first path signal, there will be multipath signals arriving 1 m
at the receiver. Consider the idealized case where there is no
noise or error so that the correlation diagram of each path is FIg ur e 24.10 Illustration of code phases/delays and related corre-
an isosceles triangle. Suppose that there are J multipath signals lation power (delay waveform) of reflected signal in the presence of a
whose delays relative to the first path are less than the GNSS rough surface.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite System Reflectometry for Ocean and Land Applications 503
reflected signal received via the down-looking antenna where m
is the time point at which the peak power occurs, while C(1) is 1.6
the correlation power at the time point 1 where the reflected
signal peak power occurs when the surface is perfectly smooth. 1.5
Since C(m) can be measured, the time parameter m can be esti-
SSH error STD (m)
mated. On the other hand, neither C(1) nor 1 can be simply
1.4
measured. The power ratio is simply defined as
C (1) 1.3
= (24.31)
C (m )
1.2
Clearly, in the case of a perfectly smooth sea surface, the power
ratio is equal to unity. Otherwise, it is less than 1. Given a power 1.1
ratio and the measured peak power, the power at the desired
code phase (1) can be calculated using (24.31), and then 1 can 0.93 0.94 0.95 0.96 0.97 0.98 0.99 1
be determined using the measured delay waveform. As a con- Mean power ratio
sequence, the desired delay of the reflected signal relative to the
direct signal can be obtained. The relative delay is then used to FIg ur e 24.11 Cost function (SSH error STD) versus mean power
calculate the SSH using the method described earlier. A sequence ratio for mean SSH estimation. Data related to four satellites are used.
of SSH estimates is produced based on a time series of measured
TABLe 2 4.3 PRN Numbers, Elevation, and Azimuth Angles of Four
and smoothed delay waveforms. In the case of low-altitude air- Satellites
borne altimetry, the mean SSH could be approximately the same
Satellite
over the surface specular reflection tracks related to several GNSS
(PRN#) 22 18 6 21
satellites with the largest elevation angles. Then, sequences of
SSH estimates associated with these GNSS satellites can be jointly Elevation () 62.6762.99 58.1457.41 50.8950.92 48.5747.95
processed to estimate the desired power ratio and the mean SSH. Azimuth () 238.19240.29 148.82149.94 266.29268.22 78.8779.93
One method is to define a cost function of the power ratio as
() = () (24.32) 24.4 Land Applications
As mentioned earlier in Table 24.2, GNSS-R technique can be
where is the STD of all the sequences of SSH estimates. N
used to retrieve a range of land parameters. This section briefly
power ratio values correspond to N cost function values. The
describes how this technique can be employed to estimate soil
power ratio and the sequence of the SSH estimates or the mean
moisture and detect forest change.
SSH are selected if the corresponding cost function value is the
minimum. This criterion selection comes from the consider-
ation that using the desired power ratio would yield estimates
24.4.1 Soil Moisture
that have minimum variations. Figure 24.11 shows an example
of the cost function versus the power ratio, which is basically Knowledge of soil moisture content is critical for drought
a convex function. In this example, the mean SSH estimate and irrigation management, so as to increase crop yields and
is 23.36 m in the WGS84 system, while the mean SSH mea- to gain a better understanding of natural processes linked
sured by a Lidar device is 23.44 m. Measurements related to to the water, energy, and carbon cycles. Soil moisture can be
four satellites as listed in Table 24.3 were used to produce the estimated by analyzing surface scattered signals transmit-
mean SSH estimate. The average of the mean SSH errors and ted and received by active radar sensors or natural surface
the root-mean-square mean SSH errors are both 8 cm. This emission detected by microwave radiometers. Alternatively,
sea surface altimetry method is suited only for the case where GNSS-R signals can be employed to estimate soil moisture.
the receiver is aircraft-borne. In the case of the satellite-borne Since the soil dielectric constant and ref lectivity depend
receiver, the theoretical model of the received reflected signal on soil moisture, the variation in the ref lected GNSS signal
power is required to obtain the offset of the peak power loca- power or signal-to-noise ratio indicates changes in soil mois-
tion of the reflected signal in the presence of a rough sea sur- ture (Masters etal. 2004). Using an interferometric method,
face. However, when the sea surface is calm or the GPS P(Y) the power associated with the coherent sum of the direct and
code is used, the offset can be negligible, so that measurements ref lected GNSS signals can be used to estimate soil mois-
associated with a single GNSS satellite can be used to retrieve ture. After continuously measuring the total power for a few
the mean SSH. hours, the location of the power notch over satellite elevation
2016 by Taylor & Francis Group, LLC
504 Remotely Sensed Data Characterization, Classification, and Accuracies
can be observed, and the soil moisture may be estimated the refraction indices of ice, water, rock, and air, and treats the
(Rodriguez-Alvarez etal. 2009). The total power received can transition moisture value as an adjustable parameter. Thedielec-
be described as tric constant of the soil can be estimated by measuring the
surface reflectivity. The model is quite general since it was devel-
oped by considering many different types of soils. GNSS-based
F ()| 1 + R exp( j)|2 (24.33)
soil moisture estimation is complicated by a number of issues
including surface roughness, vegetation canopy, and variation
where in the percentage of individual soil components. To achieve reli-
is a scaling factor able soil moisture estimation, these issues must be taken into
F() is the antenna pattern/gain account through processes such as modeling and compensation.
is the incidence angle Currently, the accuracy of GNSS-R-based soil moisture estima-
R is the reflectivity coefficient tion is generally not as good as that of active sensors and pas-
is the phase difference between the incident and reflected sive microwave radiometers. Most of the GNSS-R experiments
signals conducted for soil moisture retrieval are ground based, and only
a few are aircraft-borne. In the case of a satellite-borne receiver,
The reflectivity coefficient may be described by a three-layer it is still a challenge to infer useful information from GNSS-R
reflectivity model (air and two soil layers) as measurements.
4 s ri ,i +1 + ri +1,i + 2 exp( + j 2)
2
24.4.2 Forest change Detection
= exp (24.34)
1 + ri2+1,i + 2 exp( + j 2)
Forest change detection is another possible application of
GNSS remote sensing. Forest change can provide useful infor-
where mation about global climate change impacts and carbon stor-
s is the soil surface roughness age, and knowledge of forest change is vital for effective forest
is the signal wavelength (for GPS L1 signal, = 19cm) management. Received signal strength of the reflected GNSS
ri,i + 1 and ri + 1,i + 2 are the Fresnel coefficients of the first and signals can be employed to distinguish forest conditions from
second soil layers each other, since different surface cover has different reflectiv-
is the surface roughness correction factor ity. A number of signal strength ranges may be defined to be
is a phase term associated with a group of surfaces such as lake/river water,
typical dense forest, and cleared area due to logging (Yu
may be simply set to be unity only when using the power et al. 2013). Figure 24.12 shows four ground specular reflec-
notch position for soil moisture estimation, while it can be tion tracks of the reflected signals from four GNSS satellites
adjusted when using the pattern or shape of the total power. It received by a receiver on an aircraft. The tracks are colorized
is an advantage to use existing GPS receivers, installed primar- by the reflected signal power. The low signal power corre-
ily for geophysical and geodetic applications, to estimate soil sponds to the dense forest areas, while the high signal power
moisture (Zavorotny et al. 2010). These receivers, if exploited occurs over three areas marked by A1 (cleared area), A2 (par-
effectively, could provide a global network for soil moisture tially cleared area), and A3 (lake water). Accordingly, a sur-
monitoring. In addition to the interferometric method, the face can be classified by determining within which range its
reflectometry technique also can be employed. For instance, measured strength falls. In addition to signal strength, other
the empirical dielectric modelbased method in Wang and signal characteristics such as those related to the observed
Schmugge (1980) may be applied to GNSS-based soil moisture correlation waveform may be used to enhance surface change
retrieval. In this method, the transition moisture parameter detection or surface classification. The major issue related to
(t) is defined as surface change detection is the dependence of reflectivity on
several factors such as soil moisture and surface roughness.
Forest change may be quantified by evaluating forest biomass,
t = 0.165 + 0.49wt (24.35)
which depends on the volume of both living and dead trees
(leaves, branches, and trunks). The received signal power can
where wt is the wilting point moisture that is a function of the be used to determine the scattering coefficient, which, based
percentage of the clay and sand contents. Depending on whether on simulation (Ferrazzoli etal. 2011), should be a function of
the soil moisture is greater or less than the transition moisture, a the biomass. However, it is a challenging problem to derive a
specific formula can be used to calculate the soil moisture. This formula that accurately describes the relationship between the
model employs the mixing of either the dielectric constants or scattering coefficient and the biomass.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite System Reflectometry for Ocean and Land Applications 505
10
8
A1
Lake Burragorang 6
A2
4
2
A3
0
FIg ur e 24.12 Ground specular reflection tracks associated with four GPS satellites, colorized by reflected signal power. The picture was gener-
ated by Google Earth and GPS Visualizer.
24.5 challenging issues and their operational and research systems, and is eager for more.
However, GNSS-R is still very much a novelty technology as far
Future Directions as the geosciences community is concerned.
GNSS is a global space-based positioning, navigation, and timing The International Association of Geodesy (IAG) has been very
(PNT) capability. To come close to matching the extraordinary successful in launching technique-specific services, see http://
success of GNSS-PNT, there are many technological challenges www.iag-aig.org. These services aid other geoscientists in their
for GNSS remote sensing, especially in the case of the GNSS-R research as well as support important applications in the wider
technique. Satellite-borne GNSS remote sensing must demon- community. Examples of the latter are the contribution of the
strate its value as a reliable, high-quality sensing technology. International GNSS Service (IGS) to precise positioning (http://
With the launch of multiple LEO satellites with GNSS-RO- and igs.org), and the International Earth Rotation and Reference
GNSS-R-capable instrumentation, it seems we are at last close Systems Service to the International Terrestrial Reference Frame
to a renaissance. But are more LEO missions and better algo- (http://itrf.ensg.ign.fr). The IGS does generate troposphere and
rithms for geophysical parameter extraction sufficient? In the ionosphere parameter products (http://www.igs.org/components/
case of GNSS-RO, the answer is yes, because the meteorologi- prods.html); however, they are based on observations made from
cal/climate community now assimilates GNSS-RO products into the global ground-based GNSS tracking network (Figure 24.13),
2013 December 16 16:48:42
FIg ur e 24.13 One of the reasons for the success of IAG services are the permanent geodetic observing networks such as the IGSs global GNSS
tracking network, providing accurate, reliable data products for science and society.
2016 by Taylor & Francis Group, LLC
506 Remotely Sensed Data Characterization, Classification, and Accuracies
not from satellite platforms. There are no geodetic services produc- Elfouhaily, T., B. Chapron, K. Katsaros, D. Vandemark. 1997.
ing GNSS remote sensing products on a continuous synoptic basis. A unified directional spectrum for long and short
From the IAGs perspective, the challenging issues are related to wind-driven waves. Journal of Geophysical Research
operationalizing GNSS remote sensing, in all its forms, so that the 102(C7):1578115796.
remotely sensed geophysical parameters (atmospheric, oceanic, Ferrazzoli, P., L. Guerriero, N. Pierdicca, R. Rahmoune. 2011.
wind, soil moisture, biomass, etc.) are included in the suite of geo- Forest biomass monitoring with GNSS-R: Theoretical simu-
detic outputs of the Global Geodetic Observing System (http:// lations. Advances in Space Research 47:18231832.
www.ggos.org). Font, J., A. Camps, A. Borges, M. Martin-Neira, J. Boutin, N. Reul,
The nontechnical challenges therefore include the evolution Y.H. Kerr, A. Hahne, S. Mecklenburg. 2010. SMOS: The
of current GNSS remote sensing science missions to operational challenging sea surface salinity measurement from space.
services by increasing the number of LEO satellites equipped Proceedings of the IEEE 98(5):649655.
with GNSS receivers to satisfy coverage requirements in space Garrison, J.L., A. Komjathy, V.U. Zavorotny, S.J. Katzberg. 2002.
and time; standardizing data formats, instrumentation, and Wind speed measurements using forward scattered GPS
calibration, preprocessing, and analysis procedures; and estab- signals. IEEE Transactions on Geoscience and Remote
lishing a coordinating agency or authority so as to ensure con- Sensing 40(1):5065.
tinuous, high-quality product generation and dissemination. GEROS-GNSS REflectometry, Radio Occultation and Scatterometry
The IAG has considerable experience with geometry technique Onboard International Space Station (GEROS-ISS), http://www.
based services (such as the IGS, ILRS, IVS and IDS) (http://www. ice.csic.es/en/view_project.php?PID=155, viewed December 9,
iers.org/nn_10880/IERS/EN/Organization/TechniqueCentres/ 2013.
TC.html?__nnn=true); however, it has not yet established any Gleason, S.T. September 2006. Land and ice sensing from low
geodetic imaging services based on technologies such as syn- Earth orbit using GNSS bistatic radar. Proceedings of the
thetic aperture radar, satellite radar altimetry, lidar, or GNSS International Technical Meeting of the Satellite Division of
remote sensing (GNSS-RO, GNSS-R). One of the future chal- the Institute of Navigation (ION GNSS), Fort Worth, TX,
lenges is to address this shortcoming, so that GNSS remote sens- pp. 25232530.
ing can be recognized as a geodetic technique that is making Gleason, S.T., D. Gebre-Egziabher. 2009. GNSS Applications and
critical contributions to science and society. Methods. Artech House, Boston, MA.
Gleason, S.T., S. Hodgart, Y. Sun, C. Gommenginger, S.
24.6 conclusions Mackin, M. Adjrad, M. Unwin. 2005. Detection and
processing of bistatically reflected GPS signals from
This chapter studied GNSS-based remote sensing with a focus low Earth orbit for the purpose of ocean remote sens-
on GNSS-R for both land and ocean applications. In particular, ing. IEEE Transactions on Geoscience and Remote Sensing
details about GNSS-R-based sea surface wind speed estimation 43(6):12291241.
and sea surface altimetry were presented. The wind speed estima- Global Geodetic Observing System (GGOS), http://www.ggos.
tion accuracy can be around 1 m/s, and the mean SSH estimation org, viewed December 11, 2013.
can be of the order of decimeters. In addition to the advantage Heise, S., J. Wickert, C. Arras, G. Beyerle, A. Faber, G. Michalak,
of low cost, the GNSS-R technique can measure the geophysical T. Schmidt, F. Zus. 2014. Reprocessing and application of
parameters over a wide area. The major drawback of the tech- GPS radio occultation data from CHAMP and GRACE. In:
nique may be that the accuracy can be affected considerably F. Flechtner, N. Sneeuw, W.-D. Schuh, eds. Observation of
by modeling errors. As mentioned in the chapter, a number of the System Earth from SpaceCHAMP, GRACE, GOCE and
dedicated satellite missions associated with GNSS remote sens- Future Missions, GEOTECHNOLOGIEN Science Report
ing have been accomplished, several missions are scheduled, and No. 20, Springer, Heidelberg, pp. 6371.
more missions are expected to be funded. It is envisaged that as Jin, S., G.P. Feng, S. Gleason. 2011. Remote sensing using GNSS
a cost-effective technique, GNSS-based remote sensing will play signals: Current status and future directions. Advances in
a significant role in a wide range of remote sensing applications. Space Research 47(10):16451653.
Martn-Neira, M. 1993. A passive reflectometry and interferom-
References etry system (PARIS): Application to ocean altimetry. ESA
Journal 17(14):331355.
Anthes, R.A. 2011. Exploring Earths atmosphere with radio occul- Martin-Neira, M., S. DAddio, C. Buck, N. Floury, R. Prieto-
tation: Contributions to weather, climate and space weather. Cerdeira. 2009. The PARIS in-orbit demonstrator.
Atmospheric Measurement Techniques 4:10771103. Proceedings of IEEE IGARSS, Vol. 2, Cape Town, South
Camps, A., J.F. Marchan, E. Valencia, I. Ramos, X. Bosch-Lluis, N. Africa, pp. II-322II-325.
Rodriguez, H. Park etal. July 2011. PAU instrument aboard Martin-Neira, M., S. DAddio, C. Buck, N. Floury, R. Prieto-
INTA Microsat-1: A GNSS-R demonstration mission for Cerdeira. 2011. The PARIS ocean altimeter in-orbit dem-
sea state correction in L-band radiometry. Proceedings of onstrator. IEEE Transactions on Geoscience and Remote
IGARSS, Vancouver, Canada, pp. 41264129. Sensing 49(6):22092237.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite System Reflectometry for Ocean and Land Applications 507
Masters, D., P. Axelrad, S.J. Katzberg. 2004. Initial results of land- Yu, K., C. Rizos, A. Dempster, September 2012a. Error analysis of
reflected GPS bistatic radar measurements in SMEX02. sea surface wind speed estimation based on GNSS airborne
Remote Sensing of Environment 92:507520. experiment. Proceedings of ION GNSS, Nashville, TN.
Pierson, W.J., L. Moscowitz. 1964. A proposed spectral form for Yu, K., C. Rizos, A. Dempster. October 2012b. Performance of
fully developed wind seas based on the similarity theory GNSS-based altimetry using airborne experimental data.
of S. A. Kitaigorodskii. Journal of Geophysical Research Proceedings of Workshop on Reflectometry Using GNSS and
69(24):51815190. Other Signals of Opportunity, West Lafayette, IN.
Rius, A., E. Cardellach, M. Martn-Neira. 2010. Altimetric analysis Yu, K., C. Rizos, A.G. Dempster. July 2012c. Sea surface wind
of the sea-surface GPS-reflected signals. IEEE Transactions speed estimation based on GNSS signal measurements.
on Geoscience and Remote Sensing 48(4):21192127. Proceedings of International Geoscience and Remote
Rodriguez-Alvarez, N., X. Bosch-Lluis, A. Camps, M. Vall- Sensing Symposium (IGARSS), Munich, Germany, pp.
Llossera, E. Valencia, J.F. Marchan-Hernandez, I. Ramos- 25872590.
Perez. 2009. Soil moisture retrieval using GNSS-R Yu, K., C. Rizos, A. Dempster. 2014. GNSS-based model-free sea
techniques: Experimental results over a bare soil field. surface height estimation in unknown sea state scenarios.
IEEE Transactions on Geoscience and Remote Sensing IEEE Journal of Selected Topics in Applied Earth Observations
47(11):36163625. and Remote Sensing, 7(5):14241435.
Unwin, M.J., S. Gleason, M. Brennan. September 2003. The space Yu, K., C. Rizos, A.G. Dempster. July 2013. Forest change detec-
GPS reflectometry experiment on the UK disaster monitor- tion using GNSS signal strength measurements. Proceedings
ing constellation satellite. Proceedings of ION-GPS/GNSS, of International Geoscience and Remote Sensing Symposium
Portland, OR. (IGARSS), Melbourne, Victoria, Australia, pp. 10031006.
Valencia, E., V.U. Zavorotny, D.M. Akos, A. Camps. 2014. Using Zavorotny, V.U., K.M. Larson, J.J. Braun, E.E. Small, E.D.
DDM asymmetry metrics for wind direction retrieval from Gutmann, A.L. Bilich. 2010. A physical model for GPS mul-
GPS ocean-scattered signals in airborne experiments. tipath caused by land reflection: Toward bare soil moisture
IEEE Transactions on Geoscience and Remote Sensing, retrievals. IEEE Journal of Selected Topics in Applied Earth
52(7):39243936. Observations and Remote Sensing 3(1):100110.
Wang, J.R., T.J. Schmugge. 1980. An empirical model for the com- Zavorotny, V.U., A.G. Voronovich. 2000. Scattering of GPS sig-
plex dielectric permittivity of soils as a function of water nals from the ocean with wind remote sensing applica-
content. IEEE Transactions on Geoscience and Remote tion. IEEE Transactions on Geoscience and Remote Sensing
Sensing 18(4):288295. 38(2):951964.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
25
Global Navigation Satellite
Systems for Wide Array of
Terrestrial Applications
D. Myszor 25.1 Introduction ............................................................................................................................ 509
Contemporary Global Navigation Satellite Systems
Silesian University of
Technology and LG Nexera 25.2 SBAS: Satellite-Based Augmentation System ......................................................................511
Business Solutions AG 25.3 GNSS Errors..............................................................................................................................512
Troposphere and Ionosphere Errors Mitigation Techniques
O. Antemijczuk 25.4 Consumer Grade GNSS Positioning Precision ...................................................................517
Silesian University of Technology
25.5 Low-Cost Receivers .................................................................................................................518
M. Grygierek 25.6 GNSS Applications ..................................................................................................................521
LG Nexera Business Solutions AG Consumer Applications Industry Applications Transportation Applications Surveying
Applications Accident and Disaster Recovery Applications Health Researches
M. Wierzchanowski and Monitoring Applications Tracking Applications Unmanned Vehicles
Silesian University of Applications Time Synchronization Applications Military Applications Aeronautic
Technology and LG Nexera Applications
Business Solutions AG
25.7 Summary .................................................................................................................................. 528
K.A. Cyran Acknowledgments .............................................................................................................................. 528
Silesian University of Technology References............................................................................................................................................ 528
25.1 introduction be determined (calculated value is called pseudorange, because
various errors can influence obtained results) (Kaplan etal. 2005,
Global navigation satellite systems (GNSS) provide positioning ser- Grewal etal. 2001, Xu 2007). If signals from four satellites are avail-
vices across the globe (Hofmann-Wellenhof etal. 2008). Utilization able, then based on obtained pseudoranges, and ephemeris of satel-
of these systems gained huge popularity because of many advan- lites, receiver is able to calculate coordinates in three-dimensional
tages that they provide such as constant availability (although rare space (Gleason and Gebre-Egziabher 2009). It is worthy to mention
cases of downtimes are known), free of charge service, and good that contemporary receivers are able to utilize many visible naviga-
positioning accuracy (Gleason and Gebre-Egziabher 2009, 2010, tion satellites at the same time, and combined data obtained from
Jeffrey 2010). In general, GNSS might be internally split into three various GNSS in order to increase the precision of obtained coor-
segments: space, control, and users (Prasad and Ruggieri 2005, dinates. System requires high time precision and synchronization
Groves 2008). Space segment is comprised of satellites, which are of time between satellites, even small errors in time synchroniza-
constantly orbiting around the globe. Orbits of individual satellites, tion lead to significant errors at the level of the Earths coordinates
within particular systems (constellation), are arranged in a way, calculations. In order to achieve these tasks, satellites are equipped
which allows for observation of at least four satellites, from almost with several clocks (i.e., a battery of cesium and rubidium atomic
any point of Earth surface, at the same time (Petrovski and Tsujii clocks) that are regularly synchronized by control segment.
2012). Satellites send ceaselessly information with time stamp, pre- Control segment is composed of installations located at the
cise orbit coordinates (ephemeris), description of satellites constel- surface of the Earth. These installations form network of sta-
lation (almanac), error corrections as well as health state indicating tions scattered around the globe. Following facilities are usually
their operational state (Ward etal. 2005a). GNSS receiver obtains possessed by particular GNSS: master control station as well as
navigation signals and utilizes its internal clock in order to deter- backup control station, set of command sites with control anten-
mine delay between time of navigation message transmission and nas, and a battery of monitoring stations. The master control
time of message acquisition by the receiver antenna. As a result, station receives data from monitoring stations; thus, it is able
pseudorange between transmitting satellite and receiver might to control proper functioning of navigation system. In general,
509
2016 by Taylor & Francis Group, LLC
510 Remotely Sensed Data Characterization, Classification, and Accuracies
control segment is responsible for constant monitoring of the TABLe 2 5.1 GPS Constellation
navigation satellites states, supervision of orbital plane drift, Total Number of Satellites in GPS Constellation (June 3, 2014) 32
sustaining of constellation geometry, taking satellites out of ser-
Operational 30
vice for maintenance tasks such as orbit modification or software
In commissioning phase 1
update, and resolution of various anomalies and malfunctions in
In maintenance 1
satellites function. It also regularly communicates with satellites
in order to synchronize their atomic onboard clocks and adjust
in order to improve signal availability, especially on areas that
the ephemeris of satellite orbital model. Every satellite, belong-
are characterized by obstruction reach environment, three slots
ing to GNSS constellation, has a designated lifetime; therefore,
were expanded that can host two satellites each (fore and aft
control segment is responsible for disposal of retired satellites,
location within expanded slot are defined). Therefore, 27 naviga-
launching of new ones on early orbits as well as introduction of
tion satellites could be accommodated in baseline constellation.
these satellites into navigation constellation.
Constellation can possess additional spare satellites; however,
User segment consists of individual, institutional (government,
they do not occupy predefined slots (see Table25.1). Devices are
commercial as well as scientific), and military users, which are in
located in the Middle Earth Orbit at an altitude of 20,200 km
the possession of navigation signal receivers. Receivers might be
and orbit the Earth approximately every 12h. Arrangement of
characterized by various levels of accuracy, qualities of internal
satellites ensures that from almost every point of Earth, at least
clocks, multichannel abilities (calculation of coordinates based on
four satellites are visible at the same time and their elevation
many satellites), and possibility of utilization of different GNSS.
angles are greater than 15 relative to the ground (Kim et al.
1998). All satellites utilize the same frequency and code division
25.1.1 contemporary Global navigation
multiple access (CDMA) technique. Importantly equipment and
Satellite Systems
constellation configuration are continuously improved in order
Global positioning systemnavigation satellite time and rang- to increase accuracy (see Table 25.4 later in this chapter.).
ing (GPSNAVSTAR). The system is operated and controlled Russian Global Navigation Satellite System (GLONASS). The
by the U.S. Department of Defense. Originally, it was created system is operated by the Coordination Scientific Information
solely for military purposes; however, later, it was opened for Centre of the Ministry of Defence of the Russian Federation.
civilian applications. With time, improvements were intro- Similar to GPS, GLONASS was originally created for mili-
duced, which allow for achievement of positioning accuracy tary purposes and later, it was opened for civilian application.
close to those obtained with military signals (Harte and Levitan Currently, satellites transmit two types of signals: open one
2007). Currently open signals are transmitted on two frequen- (for civilian purposes) and the encrypted one (for military pur-
cies; therefore, there is a possibility of autonomous mitigation poses). Importantly signals for civilian purposes are broadcasted
of ionosphere errors when proper receiver is utilized (Xu 2007). in two bands; thus, ionosphere effects can be mitigated if two
Baseline constellation is composed of 24 slots (Figure 25.1). band receivers are applied. Baseline constellation is composed
There are six equally spaced orbital planes, each orbit possess of 24 satellites, organized in 3 orbital planes; each orbit pos-
four slots. Initially each slot could possess one satellite; however, sess 8 satellites (Figure 25.2). Fully operational state requires
PRN 27
GLO 12
PRN 9 PRN 26 PRN 28 GLO 04 GLO 20
PRN 22 PRN 15 GLO 03 GLO 21
PRN 18
GLO 11
PRN 11 GLO 13
PRN 19 PRN 17 GLO 19
PRN 24 GLO 05
PRN 14
PRN 8
PRN 3 GLO 02
GLO 10
PRN 6 PRN 32 GLO 06 GLO 22
PRN 21 GLO 18
GLO 14
PRN 12 PRN 20 PRN 7
GLO 09
PRN 31 PRN 16 GLO 15 GLO 07
PRN 30 PRN 4 GLO 17 GLO 01 GLO 23
PRN 5 GLO 16
PRN 23
PRN 13 GLO 08
PRN 29 PRN 10
PRN 2 GLO 24
FIg u r e 2 5.1 GPS constellation. FIg u r e 2 5.2 GLONASS constellation.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 511
TABLe 2 5.2 GLONASS Constellation requires 27 satellites, additional 3 satellites (called active spare)
Total Number of Satellites in GLONASS Constellation (June 3, 2014) 29 will be maintained (1 on each orbit) for the purpose of replace-
ment of malfunctioning devices. Such number of satellites will
Operational 24
allow to increase the accuracy level in city canyons; in addition,
In commissioning phase 0
orbital planes in the constellation are designated to obtain better
In maintenance 0
accuracy, at high latitudes, than GPS-based solution; thus, the
Spares 4
accuracy of Northern Europe coverage will improve. Current
In flight tests phase 1
plans assume that fully operational state should be achieved by
2019. Satellites are located in the Middle Earth Orbit at altitude
21satellites, additional 3 satellites (called active spares) are main-
of 23,222km and orbit the Earth approximately every 15h. The
tained for the purpose of replacement of malfunctioning devices
system is operated by the civilian organization European Space
(see Table 25.2). Orbital planes are designed in order to obtain
Agency (ESA). GALILEO will provide open signal for all users
higher accuracy than GPS in high altitudes. Devices are located
(as in GPS, two frequencies are available); however, there will
in the Middle Earth Orbit at an altitude of 19,100km and orbit
be possibility to purchase access to encrypted signal, which will
the Earth approximately every 11h. Satellites utilize the 15-chan-
increase the resistance to spoofing and jamming. Interestingly,
nel frequency division multiple access technique (Harper 2009).
additional encrypted service will be provided for government
GALILEO is the European Unions response to military-
agencies. There will be also a possibility of picking up emergency
controlled GNSS systems for which civil accuracy can be degen-
signals transmitted by beacons installed aboard ships, aircrafts,
erated or even disabled in case of conflicts. Such an action could
or carried by individuals. Furthermore, system will be able to
introduce huge disturbances in many fields of industry and
inform users about signal degradation, such ability is especially
human lives; therefore, the decision was made to create civil con-
important in applications where guaranteed precision is impor-
trolled system that will allow obtaining of precise positioning
tant, for example, plane navigation (Tables 25.3 and 25.4).
across the globe. Planned baseline constellation is composed of
30 satellites organized in 3 orbital planes, each orbit possesses 10
satellites (Figure 25.3). System, in order to be fully operational,
25.2 SBAS: Satellite-Based
GAL 11 Augmentation System
GAL 01 GAL 02
GAL 12 Precision of calculated coordinates based solely on GNSS is
GAL 22 insufficient in many applications; therefore, additional SBASs
GAL 21
were created that can improve the accuracy of obtained results.
GAL 10 GAL 08 Among them the most popular are Wide Area Augmentation
GAL 23
System (WAAS), European Geostationary Navigation Overlay
GAL 13 Service (EGNOS), and Multifunctional Satellite Augmentation
GAL 20
GAL 03 System (MSAS). WAAS was created by United States; it is des-
ignated for GPS only and it is available on the area of North
GAL 07
GAL 24 America. EGNOS, on the other hand, was the first complex proj-
ect in the field of satellite navigation systems, created in coopera-
GAL 14
GAL 09 GAL 15
tion between EU members; it was established in 2005 and the
full operational state was achieved in 2009. The purpose of the
GAL 06 system was to create augmentation for GPS and GLONASS over
GAL 19
GAL 04 the entire Europe (ESA SP-1303 2006). Finally, MSAS was cre-
GAL 17
GAL 16 ated by Japanese in order to deliver corrections to GPS signals
GAL 05
over its area.
GAL 18 In order to ensure the highest positioning accuracy, these
systems constantly monitor satellites constellations with the
FIg u r e 2 5.3 GALILEO constellation. network of ground monitoring stations, calculate errors, and
TABLe 2 5.3 GNSS Summary
No. of Satellites (Base Accuracy (Perfect Conditions,
GNSS Constellation, without Spares) Operating Country Civilian Applications Military Applications without Supporting Systems)
GPS 27 USA True True Up to 7 m
(Allan 1997)
GLONASS 21 Russian Federation True True Up to 15 m
(Miller 2000)
GALILEO 27 European Union True True Constellation is not fully operational
2016 by Taylor & Francis Group, LLC
512 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 2 5.4 Historical Accuracy of GPS System impact on accuracy of obtained results; dilution of precision (DOP)
Year 1993 1995 2000 2014 is a geometric quantity, which describes this relation (Hewlett-
Packard 1996). Another issue appears when change in satellites
Horizontal accuracy (m) ~100 to ~300 ~100 ~20 ~7.8
configuration occurs: jump in assessed location can be obtained
and accuracy might be compromised (Sukkarieh etal. 1999).
provide correction data to end users in real time. As a result, Earth tides, which are small cyclical ground movements caused
significant reduction of positioning errors is obtained, for exam- by gravity forces of other celestial bodies (mainly Sun and Moon),
ple, in case of GPS, positioning accuracy can be increased from introduce issues when a very precise conversion of coordinates to
approximately 10 m when it is calculated without correction the location on the map is required. Similar issues are introduced
services (one frequency receiver) (DoD 2008) to approximately by tectonic movements, which cause constant displacement of
2 m when EGNOS service is utilized. Interestingly aforemen- reference sites on Earths surface in relation to reference points
tioned augmentation systems are compatible, they were created set at the level of reference frame, for example, World Geodetic
to increase positioning precision mainly in aviation applica- System 1984 (WGS84) (Harper 2009) or its high accuracy version
tions; thus, correction signals are broadcasted through satellites International Terrestrial Reference System (ITRS). When pre-
(Figure 25.4). It implies that corrections might be hard to obtain cise measurements are required, magnitude of this error cannot
in areas surrounded by high obstacles. In such a case the Internet be underestimated. In some areas, this phenomenon can intro-
might be utilized for correction delivery (SISNeT technology in duce displacement at rate up to 120mm/year, for example, Great
EGNOS, Zinkiewicz etal. 2010, and ongoing work over EGNOS Britain is in constant move with reference to WGS84 and dis-
Data Access System [EDAS]) (Table 25.5). placement is at the level of 25mm/year. In order to mitigate this
effect, frames of reference that are tied to the ground and fixed in
25.3 GnSS errors time are introduced. In Europe, the most popular one is European
Terrestrial Reference System 1989 (ETRS89), which can be easily
Utilization of satellite navigation systems for measuring of the converted to ITRS by simple transformation of coordinates that
object position is exposed to a variety of errors that affect the are published by International Earth Rotation Service (IERS). Also
accuracy of designated coordinates (Grewal etal. 2001). Some of Earths rotator parameters are not constant; therefore, additional
the factors that can lead to incorrect results are satellite position source of errors is introduced. As in the previous case, influence of
errors. They could be caused by various factors such as solar radia- this type of errors can be mitigated by application of corrections,
tion pressure and solar flares, electromagnetic forces (Silva etal. which could be based on data obtained from IERS. It is worthy
2002), and relativistic effects. In addition, gravity force of Earth, to mention that all types of conducted calculations, for example,
Sun, Moon, and other celestial bodies as well as oceanic and ter- conversions between different reference systems are also prone to
restrial tidal waves influence the satellites positions, thus degrad- precision errors. It happens because of floating point number rep-
ing obtained coordinates. Noteworthy is the fact that geometry of resentation in devices memory, which usually results in necessity
observed satellites in relation to the receiver antenna also has an of numerical rounding when extensive calculations are done.
PRN 134 PRN 137
PRN 129
PRN 122
PRN 135
PRN 127
PRN 138
PRN 131
PRN 126
PRN 124
PRN 136
PRN 120
FIg u r e 2 5.4 EGNOSS constellation.
TABLe 2 5.5 GPS Accuracy When Various Augmentation Systems Are Utilized
GPS WASS NDGPS GDGPS AGPS EGNOS MSAS
L1 = ~7.8 m ~3 m ~1015cm ~10cm ~5 m ~3 m ~0.85.3 m
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 513
TABLe 2 5.6 Typical Errors, Their Magnitudes, and Methods of Mitigation
Typical Magnitude
Type of Error of an Error (m) Methods of Error Mitigation
Time base error 3 Utilization of signals from many satellites
Inaccuracy in positioning of satellite on the orbit 2 Utilization of correction services providing exact location of satellites on the orbit
Satellite orbit geometry errors 2.5 Utilization of correction services that provide precise orbit information
Ionosphere influence 5 Utilization of differential corrections/dual-frequency receivers/models of influence
of ionosphere
Troposphere influence 0.5 Utilization of differential corrections/dual-frequency receivers/models of influence
of troposphere
Propagation of electromagnetic waves 1 Utilization of proper algorithms
Distortions of propagation of signals transmitted by satel- navigation signal. The purpose of this action is to mislead the
lites in Earths atmosphere, which are mainly caused by iono- receiving device so that it determines its coordinates based on
sphere and troposphere refraction, are another cause of errors fake signal. Techniques applied in order to fulfill this task might
(MacGougan 2006). Signal can also be disturbed by high volt- include utilization of simulator of particular GNSS that is con-
age lines and other transmitters (e.g., various GNSS signal nected to the physical antenna, or controlled retransmission of
jammers). In general, all objects that are blocking direct visual signals transmitted by original navigation satellites. It is worthy
contact between receiver and transmitter (such as trees leaves, to mention that especially civil receivers are prone to this issue.
buildings, cliff faces) are also potential source of signal degra- For selected applications (military based), GPS and GLONASS
dation or even its unavailability. When signal is weakened, spe- provide additional encrypted services, which reduce vulner-
cial software algorithms might be applied, which could improve ability for spoofing and jamming. GALILEO also would provide
the ability to obtain the position (Ziedan 2006). In addition, encrypted navigation services for selected companies and gov-
receivers surrounding environment might introduce the phe- ernment agencies such as police, search and rescue operators,
nomenon of multipath signal propagation (Ward etal. 2005b): humanitarian aid, fire brigades, health services, defense, coast-
it happens when the antenna receives the same signal reflected guard, border controls, customs and civil protection units as
from various obstacles (i.e., building walls), then multiple copies well as critical infrastructure and networks.
of the same signal reach receiver antenna at various times, it can In order to mitigate atmospheric errors, corrections pro-
introduce the effect of jumping of the coordinates determined vided by SBAS might be utilized (approximate accuracy 2 m);
by GNSS receiver. Although software incorporated in receivers moreover, there is a possibility of utilization of professional
often tries to detect such a situation and to limit influence of two frequency (in case of GPS or GALILEO) or two band (for
this phenomenon, however, it still might have an effect on the GLONASS) receivers, which can autonomously rule out this
accuracy of obtained results. The requirement of visual contact kind of errors (approximate accuracy 1 m) (Table 25.6).
between receiver and transmitter antennas rules out the pos-
sibility of GNSS employment for navigation within buildings. 25.3.1 troposphere and ionosphere
In another area, errors of transmitting and receiving devices errors Mitigation techniques
are located. They could be caused by various factors such as
During the course of Research on the EGNOS/Galileo in
instability of internal clocks frequency (most GNSS receivers
Aviation and Terrestrial Multi-sensors Mobility Applications for
utilize quartz-based generators, which are easily influenced by
Emergency Prevention and Handling (EGALITE) project, our
temperature fluctuations), internal noises within transmitting
team created application GPS 3D Viewer, which designates coor-
and receiving devices, and variability of antennas phase centers.
dinates based on RAW GPS data. Various algorithms calculating
Likewise all electronic devices, navigation satellites are prone
the influence of troposphere and ionosphere on acquired posi-
to hardware as well as software errors and malfunctions, which
tions were included. Implemented model of troposphere signal
could influence acquired positioning level accuracy or even
propagation delay is described by the following equation:
completely disabled services in given areas. In some rare cases,
particular GNSS might broadcast wrong information (e.g., 11h
of GLONASS failure in April of 2014), thus rendering the pro- TCi = (dhyd + dwet ) m(Eli ) (25.1)
vided services unusable.
For military controlled system such as GPS and GLONASS, where values of [dhyd,dwet] are calculated based on the receiver
there is a possibility of temporal degradation of provided ser- antenna height above sea level and are estimated by five meteo-
vice in the selected areas. Also malicious third parties might jam rological parameters: pressure (P [mbar]), temperature (T [K]),
the navigation signal on selected areas although such a behavior the vapor pressure of water (e [mbar]), change of temperature
is restricted to an extent in some countries, for example, in the ( [K/m]), and change in the rate of evaporation ( [dimension-
United States. In addition, there is a possibility of spoofing of less]). Value of every meteorological parameter is calculated for
2016 by Taylor & Francis Group, LLC
514 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 2 5.7 Seasonal Values for GPS Receivers Troposphere
106 k 2Rd e
Corrections Z wet = (25.6)
g m ( + 1) Rd T
P T e
Latitude () (mbar) (K) (mbar) (K/m)
15 or less 0.00 0.00 0.00 0.00e3 0.00 where
30 3.75 7.00 8.85 0.25e3 0.33 k1 = 77.604 K/mbar
45 2.75 11.00 7.24 0.32e3 0.46 k2 = 382,000 K 2/mbar
60 1.75 15.00 5.36 0.81e3 0.74 Rd = 287.054 J/kg/K
75 or more 0.50 14.50 3.39 0.62e3 0.30 gm = 9.784 m/s2
[dhyd,dwet] are calculated with
receiver latitude () and day in astronomical year (D) (which
starts on January 1); in addition, average seasonal changes (pre- H
g /Rd
sented in Table 25.7) are taken into account. Values of particular dhyd = 1 Z hyd (25.7)
T
meteorological parameters, denoted for simplification as (), are
calculated with the following formula:
(( g m ( +1))/Rd )1
H
2(D D min ) dwet = 1 Z wet (25.8)
(, D ) = 0 () () cos (25.2) T
365.25
where Parameter (g) is equal to 9.80665 m/s2, height of receiver (H) is
Dmin= 28 for northern latitudes measured in [m] above the sea level.
Dmin = 211 for southern latitudes Troposphere correction function for the satellite elevation
[0, ] represent mean seasonal fluctuation of values of (m (Ei)) is calculated using following equation:
parameters for latitude of receiver antenna
1.001
When latitude is lower than || 15 or higher than || 75, m(Ei ) = (25.9)
0.002001 + sin 2 (ELi )
values of [0] as well as [] are taken directly from Table 25.8.
For latitudes in the range 15 < || < 75 values of [0, ] are
calculated using the following equations: The important factor is: function is incorrect for elevations less
than 5. Finally ith satellite troposphere delay error is equal to
i
0 () = 0 (i ) + [0 (i +1) 0 (i )] (25.3)
i +1 i
2j tropo = (0.12 m(E ))2 (25.10)
i
() = (i ) + [(i +1) (i )] (25.4)
i +1 i and vertical troposphere error is TVE = 0.12 m.
The influence of Earths environment on positioning
Seasonal values: accuracy was presented on 3D graphs. Application respon-
Mean values: sible for visualization, GPS3D Viewer, gets RAW data from
Delays [Zhyd,Zwet], for the level of the sea are calculated with fol- Septentrio PolaRx-3 receiver and calculates antenna position
lowing formulas: with scheduled frequency. Data presented in this work were
sampled with frequency equal to 1 Hz in an experiment last-
106 k1Rd P ing 24h (Cyran etal. 2011). Gathered samples are presented
Z hyd = (25.5)
gm on 3D graph with violet points (see Figures 25.5 through
25.8). Obtained results show clearly that in horizontal plane,
GNSS-based coordinates have deviation at the level of 12 m;
TABLe 25.8 Mean Values for GPS Receivers Troposphere Corrections however, accuracy in the vertical plane is worst and, depend-
Latitude () P0 (mbar) T0 (K) e0 (mbar) 0 (K/m) 0 ing on the conditions, could be a few meters. Consecutive
figures show influence of taking into account the effect of
15 or less 1013.25 299.65 26.31 6.30e3 2.77
ionosphere and troposphere on positioning accuracy. Figure
30 1017.25 294.15 21.79 6.05e3 3.15
25.6 presents antenna position when troposphere and iono-
45 1015.75 283.15 11.66 5.58e3 2.57
sphere corrections are turned off, while Figures 25.7 through
60 1011.75 272.15 6.78 5.39e3 1.81
25.8, present the influence of these corrections on achieved
75 or more 1013.00 263.65 4.11 4.53e3 1.55
accuracy.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 515
FIg u r e 2 5.5 Influence of Earths environment on positioning accuracy for troposphere corrections OFF, Ionosphere corrections OFF.
FIg u r e 2 5.6 Influence of Earths environment on positioning accuracy for troposphere corrections OFF, ionosphere corrections ON.
2016 by Taylor & Francis Group, LLC
516 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 2 5.7 Influence of Earths environment on positioning accuracy for troposphere corrections ON, ionosphere corrections OFF.
FIg u r e 2 5.8 Influence of Earths environment on positioning accuracy for troposphere corrections ON, ionosphere corrections ON.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 517
Horizontal deviation from reference
m
5
20
4
3
10 2
Delta-North (m)
1
0 0
1
2
10
3
4
20
5
5 4 3 2 1 0 1 2 3 4 5
20 10 0 10 20 m Delta-East (m)
FIgu r e 25.9 Graph of coordinates obtained from Samsung Galaxy FIg u r e 25.10 Graph of coordinates obtained from PolaRx-3 receiver
III internal GNSS receiver, when GPS was utilized (24h period). when GPS + SBAS + two frequencies were utilized (24h period).
25.4 consumer Grade GnSS
4m
Positioning Precision
Typical mobile devices (mobile phones and tablets) available 3m
in the market are equipped with multichannel GNSS receivers,
which allow for acquisition of the signal from many navigation 2m
systems at the same time (GPS, GLONASS, GALILEO); more-
over, usually, acquisition of augmentation signal is also enabled. 1m
List of typical receivers is presented in Table 25.9.
Samsung Galaxy SIII, as a representative of typical mobile
device, was utilized in order to present GNSS precision issues.
Special application was prepared and installed in the device.
Task of this application was to log co-ordinates of the device
to internal database. Outcomes of the experiment were com-
pared with results obtained from Septentrio PolaRx-3 receiver,
which is widely utilized in aeronautics. During the experiment,
receivers were lying stationary in the same location, in order
to provide the same conditions for both devices. Positions of
receivers were logged for 24h, with a frequency of 1Hz. Based
on positioning graphs, estimation of errors was performed; FIg u r e 25.11 Graph of coordinates obtained from Samsung
according to results obtained (see Figure 25.9), Samsung XPERIA S internal GPS + GLONASS receivers, when GPS and
Galaxy SIII is characterized by mean error at the level of 15 m GLONASS were active (24h period).
TABLe 2 5.9 Typical Receiver Chips Utilized in Mobile Devices while PolaRx-3 is characterized by mean error at the level of 2 m
Manufacturer Chipset Channel Number
(Figure 25.10). There are many reasons for such a huge difference
in precision. The receiver built in Samsung Galaxy SIII obtains
SiRF (CSR) SIRFstarIII 20
positioning data based on one frequency only. In addition, it has
SiRF (CSR) SIRFstarIV 40
small antenna and lower sensitivity, whereas PolaRx-3 is a dual-
MediaTek MTK3318 51
frequency receiver (GPS L1 and L2 frequency) and it is equipped
MediaTek MTK3339 66
with lightweight high precision geodetic dual-frequency antenna
MaxLinear Mxl800sm 12
PolaNt.
2016 by Taylor & Francis Group, LLC
518 Remotely Sensed Data Characterization, Classification, and Accuracies
There is a possibility of limitation of errors when mobile 25.5 Low-cost Receivers
device is able to utilize more than one GNSS such as Sony
XPERIA S, which is equipped with dual system receiver (GPS Nowadays, cheap GNSS receivers, with parameters similar to
and GLONASS). Under the same conditions, this device was professional chips utilized by aeronautics, are available in the
able to achieve precision at the level of 3 m when both GPS market. Authors of this chapter conducted studies, in which
and GLONASS data were taken into account (Figure 25.11). GNSS receivers, produced by U-Blox company, were employed.
Interestingly when only GLONASS was utilized, the positioning To test applicability of these chips, custom device has been devel-
precision was at the level of 18 m (Figure 25.12). oped, schematic is presented in Figure 25.13. GNSS receiver was
based on the low-cost U-Blox LEA-6S chip.
During precision tests, receivers were stationary and data
were gathered for 24h. Results presented in Figure 25.14 show
that LEA-6S exhibits the precision of 2.5 m when GPS + SBAS
reference signal was utilized, it is close to the precision obtained
by PolaRx-3 Septentrio receiver. In order to present precision
that can be obtained at the level of different GNSS, NEO7
2m receiver, which can determine position with GPS, GLONASS,
GALILEO, and QZSS, was utilized. Result of survey, of two
5m
navigation systems (GPS and GLONASS), for experiments that
took 24h and frequency refresh rate, was equal to 1Hz; this is
8m presented in Figures 25.15 and 25.16 (note that scale of axis is
different).
10 m Importantly, the NEO7 chip is able to work in the PPP mode
(precise point positioning), a method that offers enhanced posi-
12 m tioning precision (see Figure 25.17) by utilization of the car-
rier phase measurements to smooth calculated pseudoranges
between receiver and satellites antennas. The algorithm needs
15 m
continuous carrier phase measurements to be able to smooth
the pseudo-range measurements effectively. Additionally, iono-
FIg u r e 25.12 Graph of coordinates obtained from Samsung sphere corrections like those received from SBAS or from GPS
XPERIA S internal GLONASS receiver when only GLONASS was active are required. Unfortunately, technology works correctly when
(24h period). un-obscured sky view is available.
GND GND GND
P$AGND P$AGND
BTM-222
C5 P$1 P$38
P$1_GND P$38_GND
P$2 P$2_PVCC P$37_RF-10-ANTENNA P$37
4.7uF P$3 P$36
U$4
P$3_AlO(0) P$36_PlO(10)
P$4 P$4_AlO(1) P$35_PlO(11) P$35
P$5 P$34
P$5_PlO(0) P$34_SPl-MISO P$33
P$6
+3V3
U3 RF RF
P$6_PlO(1) P$33_SPl-CLK P$32
P$7 P$7_PlO(2) P$32_SPl-CSB GND GND
15 14 R5 P$8 P$31
16 GND4 GND3 13 P$8_PlO(3) P$31_SPl-MOSI P$30
RF_IN GND2 P$9 P$9_PlO(4) P$30_UART-CTS
17 GND5 RESERVED 12 P$10 P$10_GND P$29
18 11 LED2 330 Ohm P$29_GND P$28
10 Ohm VCC_RF V_BCKP P$11 P$11PlO(5) P$28_UART-RTS
19 10 R6 P$12 P$27 GND
20 V_ANT RESET_N P$12_PlO(6) P$27_UART-TX P$26
+3V3
AADET_N 9 P$13 P$26UART-RX
R1 CFG_COM1/NC GND P$13_PlO(7)
BAT1
P$14 P$14_PlO(8) P$25_PCM-CLK P$25
+3V3
21 8 LED3 330 Ohm P$15 P$24
VCC_OUT
4.7uF
GND 22 RESERVED/FWD P$15_PlO(9) P$24_PCM-OUT
7 P$16 P$23
P$BGND
RESERVED/SPI_SCS1_N GND1 R4 P$16_RESET P$23_PCM-IN
C4
23 RESERVED/SPI_SCK VCC 6 P$17 P$17_VCC P$22_PCM-SYNC P$22
24 NC 5 P$18 P$21
25 VDDUSB 4 LED4 P$18_GND P$21_USB-DN P$20
P$19
LED1
USB_DM RXD1 GND GND P$19_GND P$20_USB-DP
26 3
27 USB_DP TXD1 2
R2
P$BGND
EXTINT0/SPEED SCL2/SPI_MISO 1 P$1
28 TIMEPULSE SDA2/SPI_MOSI GND
330 Ohm
UBLOX-LEA-6
GND
GND
+3V3
U$2
4.7 uF
R3
C1 1 VIN_2 5
2 PWRGD
DPDT VIN 8
BAT-1 3 VOUT
6 ~SHDN C3
BATTERY 7 CDELAY
C2 SENSE 1 uF
BAT-2 4
1000pF GND
GND
S1 MCP1725-3302E/SN
GND GND
GND
FIg u r e 2 5.13 LEA-6S GPS receiver schematic.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 519
min + dev avg dev
N
max
dev
W E
2.5 m 5.0 m avg
+ dev
min
50.449000* 19.027046*
FIg u r e 2 5.14 Graph of coordinates obtained from LEA-6S (GNSS) receiver (24h period).
min + dev avg dev
N
max
dev
W 5m E
1.0 m avg
+ dev
min
FIg u r e 25.15 Graph of coordinates obtained from NEO-7 (GNSS) receiver when GLONASS was active. Sampling frequency 1 Hz
(24h period).
2016 by Taylor & Francis Group, LLC
520 Remotely Sensed Data Characterization, Classification, and Accuracies
min + dev avg dev max
N
max
dev
W E
2.5 m 5.0 m avg
+ dev
min
S
50.449059* 19.027004*
FIg u r e 2 5.16 Graph of coordinates obtained from NEO-7 (GNSS) receiver when GPS was active. Sampling frequency 1Hz (24h period).
min + dev avg dev max
N
max
W E
1m 2m + deg
min
S
50.449056* 19.027006*
FIg u r e 2 5.17 Graph of coordinates obtained from NEO-7 working in GPS+SBAS mode when PPP was activated (24h period).
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 521
Fleet management
Fleet system
Client stations
FIg u r e 2 5.18 Fleet management system.
25.6 GnSS Applications activities trackers that allow for the visualization of people
movement and creation of various statistics, automatic photo-
GNSS play important role in various fields (see Table 25.10). taggers that add coordinates to photos, objects trackers (Hasan
Basically, two main types of applications of these services can etal. 2009) (e.g., cars, phones) for determination of location of
be distinguished: civilian and military. However ,within these stolen or lost goods, friends trackers for searching of acquain-
fields, many common categories can be distinguished. tances in the area, and many others. However, the most pop-
ular seems to be various navigation applications, which can
automatically calculate the best route to the destination loca-
25.6.1 consumer Applications
tion, and present it on the screen as 2D/3D map or in live view
GNSS-based devices and applications play important role in mode (Huang et al. 2012). What is more: if the driver misin-
everyday life of millions of peoples. The ability of precise deter- terprets directions provided by the system, these applications
mination of location allows for the creation of various applica- can instantly work out new route. It hugely reduces time of trip
tions such as interactive maps with the ability to localize the preparation. Furthermore, by utilization of proper algorithms
closest services in the area (such as restaurants, hotels, etc.), that are able to determine the best routes, limitation of the dis-
automatic location aware reminders (Hariharan et al. 2005), tance that have to be traveled could be achieved. Another use-
ful feature of navigation systems is their ability of utilization of
TABLe 2 5.10 Application of GNSS
database of speed limits (Bargeton et al. 2010). It could allow
Categories Applications for fuel saving, reduction of air pollution, and increase safety
Consumer Precise determination of user location on roads.
Various navigation applications In combination with other services such as wireless internet
Localization of the closest services access, navigation application can check current traffic statis-
Automatic, location aware, reminders tics (e.g., existence of traffic jams on the route, road works) and
Activity trackers modify the route on the fly in order to circumvent problematic
Automatic photo taggers areas (Bar-Gera 2007). If such data are collected from many
Industry Determination of precise location of users simultaneously, then they can be utilized for traffic pattern
equipment and employees recognition and planning of traffic issues solutions. However,
Location of potential clients (transportation) in such a case, privacy standards should be maintained, so that
Monitoring of tourist flow (e.g., ski resorts) users that provide the data can keep anonymity. It might be hard
Planning of actions and processes in some cases, because even if the name of the user could not be
Optimization of various processes included in the package sent to the data collecting center, the
Archeology Surveying of archaeological sites user can be identified by many different factors, for example, the
Site map creation place of living or place of work. Therefore, various methods of
Geodesy and cartography Preparation and update of maps deeper anonymization of data were prepared: for example, it can
Onsite measurements obtained by simulation of system of induction loops embedded
Analysis of geological and tectonic movements in roads; special location are designated and data are collected
Military Co-ordination of military actions only when vehicles drive over such a virtual marker.
Soldiers localization Application designated for wearable devices can also benefit
Homing missiles from the ability of current location obtainment. These applica-
Detailed map creation tions can learn human behaviors and predict future user loca-
Rescue operations tions. It will give ability of automatic accommodation to user
Aviation Primary navigation system or support for environment, for example, in order to increase productivity
ground-based navigation systems
(Ashbrook and Starner 2003).
2016 by Taylor & Francis Group, LLC
522 Remotely Sensed Data Characterization, Classification, and Accuracies
25.6.2 industry Applications skiers flow and accommodate tracks accordingly; also safety
might be improved. Tracking can be done through utilization
GNSS-based systems are widely utilized in various workplaces of custom devices or with tourist mobile devices equipped in
and industrial applications. Employers can hugely benefit from special application. In order to facilitate sightseeing and survey-
utilization of GNSS technology, especially when it is combined ing of interesting localizations, automatic guide system based
with wireless means of communication. Ability of precise loca- on GNSS services can be prepared; it can be combined with
tion of equipment as well as employees allows for simplification aforementioned statistic gathering system so that data can be
of control processes and reallocation of resources (e.g., for soil collected. It is worthy to mention that interesting researches are
excavation process, in order to control movement of hauling also done in the field of geotagged photos utilization for the pur-
units, loaders, and excavators; Ahn etal. 2011); it can also reduce pose of extraction of travel patterns (Zheng etal. 2012).
significantly costs. Although GNSS proved to be a useful tool in company envi-
Many fields can profit from utilization of this kind of sys- ronment, there are some legal issues, depending on the coun-
tems, especially those in which employees are working outside try or region in which such methods are applied, that should be
of the company headquarters, such as transportation munici- taken into account at the level of planning and implementation
pal services (street sweeping, snowploughs, garbage collection, of GNSS-based solutions. In particular, the issues of workers
road mending), sales departments, emergency services, agri- privacy should be taken into account. Also workers co-ordinates
culture, etc. GNSS-based systems are successfully utilized in should not be logged in their private time (after work or during
the area of construction works for the purpose of localization brakes) and special zones that are excluded from logging should
of land and marine construction sites, for example, in offshore be established (Inks and Loe 2005, Towns and Cobb 2012).
oil exploration.
Application of these systems combined with logging of
25.6.3 transportation Applications
obtained data allows for creation of reports and introduction of
various optimizations. Employers gain control, because there Public transportation benefits hugely from a combination of
is a possibility of validation of realization of task by employ- GNSS with wireless communication abilities. Means of trans-
ees. Routes can be checked and all deviations from the planned portation equipped in such devices can constantly communicate
path, or omission of planned locations, determined. Current their location to the control centre and then information can be
and historical positions of chosen unit can be obtained, detec- posted at bus/tram stops as well as at web page. It can also be
tion of entering or leaving of selected area might be performed distributed to dedicated mobile applications utilized by passen-
(so-called geofencing), and various geoalerts might be set. On gers. Importantly, such system can be integrated with in-vehicle
the other hand, application of GNSS technology can support information systems; thus, information about next stop can be
workers, that is, when salary that was paid is insufficient for the presented to the passengers, without drivers action. Further inte-
time of work, then readouts obtained from GNSS-based system gration, with city traffic control systems, is often implemented,
can confirm or reject legitimacy of workers complaint. In many so traffic lights are controlled in a way to give priority for public
cases, utilization of these systems allowed the detection of law transportation vehicles (Bain 2011). When actual vehicle posi-
violations such as stealing of companys possession, or improper tion is combined with historical data and information about
habits such as utilization of company resources for personal pur- current state of traffic flow, predicted time of arrival of bus or
poses. It is worthy to mention that abidance of driving laws can tram at the line stops can be calculated. As a result, not only
easily be determined, it can reduce the amount of accidents and public transport controllers can rapidly react at emerging situa-
improve safety. Self-employed persons, such as taxi drivers, can tions and, for example, dispatch additional resources or redirect
also benefit from utilization of GNSS services. Special GNSS- routes of particular lines, but also passengers can dynamically
based dispatching systems, which allow for easy location of modify their travel plans and utilize different lines in order
passengers and determination of temporal hot spots, allows for to reach their destinations faster. Application of such systems
optimization of routes, reduction of passenger await time, and could reduce frustration of the end users in case of delays; in
increase of drivers incomes (Hou and Chia 2011). In addition, addition, it supports the idea of utilization of public transporta-
data possessed from systems located in taxis are often utilized tion, thus counteracting formation of traffic jams and decrease
by scientists in order to conduct various analyses and researches, pollution level. When additional sensors are incorporated, for
for example, automatic land-use classification (Pan etal. 2013), example, automatic passengers counters, there is a possibil-
traffic flow patterns, assessment of road quality, etc. ity of creation of detailed passenger load statistics for all lines
Tourist resorts can automatically acquire information about segments. Implementation of such system, which ceaselessly
locations that are visited by guests at various times across the collects data from various means of public transportation, can
day, tourist moving patterns as well as typical activities in which improve process of schedules modification (in order to obtain
tourists are involved (Skov-Petersen et al. 2012). It allows for better time coverage of some areas of the city), determination
introduction of modifications in various areas such as offered of the number of required vehicles, optimization of routes on
services, resort layouts, travel recommendations presented to particular lines, and modification of timetables, so they accom-
the tourists, etc. (Zheng et al. 2009). Ski resorts can monitor modate real arrival times.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 523
It is worthy to mention that such systems installed in school the relation between given point and all other points that are
buses, through automatic notifications about delays, can hugely localized on analyzed site. In some cases, accuracy provided by
facilitate lives of parents and increase their trust in the transpor- pure GNSS solution is insufficient; therefore, in order to increase
tation system. precision, additional techniques, that can increase accuracy
of measurement to centimeters, are often employed. The most
popular one is utilization of dedicated GNSS-based reference
25.6.4 Surveying Applications
stations (ifsuch are available on selected area) and application
GNSS positioning systems play important role in the field of of differential corrections (Cosentino etal. 2005) or Real Time
geodesy, archaeology, landscape surveys, and map creation Kinematics (Wei etal. 2010). Deployment of stations is usually
(Ainsworth and Thomason 2003, Bargeton et al. 2010, Bussios done at the national level. For every reference station, very pre-
etal.). It is a great tool for mapping large areas or difficult-to- cise coordinates are determined. There are two types of stations:
access locations (Bargeton et al. 2010). GNSS receivers can be active ones are equipped in stationary GPS receivers, operate all
carried by surveyor or mounted in the car and automatically the time, and usually share data through dedicated web sites.
gather coordinate data. The main advantage of the utilization of They are usually scattered around the area: for example, in
GNSS technology is the analysis of obtained data can be done on Great Britain there were 32 active stations based on GPS (data
site or raw data, obtained from the receiver, can be recorded and for December 2002), any point of Great Britain should be within
further processed in offline mode. GNSS-based measurements 100km from the closest active stations; in addition, urban areas
can be done relatively quickly when compared to the utilization are usually covered by more than one active station. If higher
of theodolite, tape, or electromagnetic distance measurement level of accuracy is required, passive stations might be employed.
methods (Figure 25.19). This kind of station is usually realized as a ground maker, which
Utilization of GNSS technology in the field of archaeology has does not possess its own GNSS receiver; therefore, user, in order
a long history; it was successfully applied to small-scale projects to utilize this station, has to provide its own equipment that is
such as mapping of single archaeological objects (stones/walls located at passive station site for the time of survey conduction.
etc.) as well as large ones. Application of this technology sig- In Great Britain, there were around 900 such markers; any point
nificantly improves the accuracy of obtained result, it limits the of land should be within 2030 km from the closest marker.
utilization of other geodetic techniques such as tapes measure- Data for differential correction can also be obtained from SBAS.
ments, and speeds up the process of coordinate generation. As As mentioned earlier, Earths continents are in constant
a result, the reduction in the number of personnel experienced move: GNSS in combination with fixed reference stations, on the
in geodesy measurements involved in the realization of these surface of the ground, are great tool for the measurement of such
kinds of projects is possible. The utilization of GNSS technology relocations; level of oceans can also be monitored (Yang and Lo
perfectly fulfills the requirements of the surveying processes: 2000). In general, GNSS are valuable tool for measurement of
this requires determination of the location of the site in rela- object deformation (e.g., damns), and in engineering survey
tion to the map, followed by the determination of the location applications (Frei etal. 1993). Furthermore, scientists investigate
of all points of interest on the site level and determination of ways to utilize sensitivity of GNSS-based devices for measure-
ment of various disturbances in atmosphere. Such systems can
be applied in meteorology for determination of the amount of
vaporized water in the atmosphere, and current researches point
out that they can also be applied to volcanic ash detection; there-
fore, improvement of the early volcanic ash alerting system in
affected areas might be improved (Aranzulla et al. 2013). As a
result, flight safety and comfort of population inhabiting these
areas might be improved.
25.6.5 Accident and Disaster
Recovery Applications
Studies conducted for Germany point out that in 2002, there
were more than 45 private cars per 100 inhabitants and on aver-
age, every vehicle was characterized by annual travel distance at
the level of 12,600km (Kalinowska and Hartmut 2006). These
results show the significance of mobility in contemporary soci-
Known point Unknown point ety. Unfortunately, mobility requirement entails exposure of
road users to various harmful events. Statistical data gathered
FIg u r e 25.19 Utilization of dedicated GNSS-based reference for UE reveal that every year, over 1.5 million of people met with
station. accidents; to make matter worse, as a consequence of occurrence
2016 by Taylor & Francis Group, LLC
524 Remotely Sensed Data Characterization, Classification, and Accuracies
of such events, more than 30,000 people annually lost their lives. specificity equal to 1 was created. In the case of accident occur-
It is a important social and economic issue. Researches for UE, rence, position based on GNSS signal is determined, then coor-
done in 2000, estimated such costs at the level of 160 billion of dinates, time of event as well as user ID are sent to the emergency
euro per annum (Petersen 2000). center through GSM network. The application of GNSS technol-
Outcomes of researches performed point out that in many ogy allows for quick and precise location of accident site, thus
cases, quicker delivery of medical help could reduce the num- contributing to a significant decrease in the emergency service
ber of casualties and decrease the time of postaccident trauma waiting time. In addition, time that is sent to the emergency
recovery. There are many issues when it comes to emergency centre is based on GNSS readouts; it provides precise time syn-
services notification: often accident victims or witnesses can- chronization between various devices: as a result, there is a pos-
not give precise location of event occurrence, it is usually espe- sibility of determination of groups of related events and accurate
cially hard in rural areas: the stated location might be incorrect. reconstruction of roads accidents in which many vehicles are
Another issue is that, in many cases, victims of accident lost involved is facilitated.
their consciousness; however, there might be no witnesses to call In order to automatically determine car environment (for
for help, such a situation is common in uninhabited areas espe- activating or deactivating the application), the NFC module is
cially for accidents that occur during night hours. Up to 52% utilized. An assumption is made that car should be equipped in
of fatalities could be avoided if victims were located earlier by NFC tag. When device detects the presence of proper NFC card,
medical personnel (5%), were transported to the hospital (12%) it automatically switches to active mode; in other cases, inactive
or advanced trauma centre (32%) soon (Henriksson 2001). The mode is chosen. If users device does not possess NFC capabil-
EU estimates that the introduction of efficient system of rapid ity, there is a possibility of manually choosing the proper mode.
emergency notification, which will decrease the amount of res- Interestingly, GNSS receivers also can be utilized in this area
cue time and speed up the process of medicine delivery, reduces of application. Some creators of similar systems assumed that
severity of injures: up to 2500 lives can be saved and 26 billion of mobile devices must move with speed higher than some empiri-
euro can be saved per annum. cally determined threshold value, and for the purpose of speed
The aforementioned analysis leads to the proposals of sys- determination, GNSS readouts were employed. However, such a
tems, which are responsible for automated detection of accidents solution consumes a lot of power, because vehicle position has
and notification of emergency center. One such system was to be continuously refreshed; in addition, this system will not
developed during the course of research on EGNOS/Galileo in detect accident when the car is stationary on the crossroad or in
Aviation and Terrestrial Multisensors Mobility Applications for the traffic jam (Thompson etal. 2010, White etal. 2011).
Emergency Prevention and Handling (EGALITE) project. One of In more platform-dependent solution, there is a possibility
the aims of EGALITE project was the creation of efficient system of integration of mobile accidentdetecting application with
of emergency handling and prevention base on mobile devices. air bags deployment controller; creation of additional hardware
Contemporary mobile devices (e.g., smart-phones, tablets) have module, which contains high precision accelerometers, is per-
feature-rich items equipped in many sensors (i.e., accelerom- manently attached to the cars body and communicates with
eters, gyros, magnetometer) and receivers (i.e., GNSS receiver) mobile device through Bluetooth technology (Matthews and
they are easily accessible and widely available. Moreover, rules Adetiba 2011). Such an approach can further reduce amount of
introduced by U.S. Federal Communication Commission (FCC) power required and increase detection precision.
state that in United States, every mobile carrier must be able to As mentioned earlier, important issue of utilization of GNSS
determine 911 mobile caller longitude and latitude with a cer- receiver, built into mobile devices, is limited battery power.
tain accuracy (Yang and Lo 2000). However, precise localization Although battery power consumption of specialized GNSS
based solely on transmitting tower was, in the time of introduc- receivers in perfect conditions can be as low as 10 mW (e.g., Sony
tion of this rule, insufficient and it required costly investments CXD5600GF), typical power utilization in modern device is
in development of infrastructure. It was one of driving forces higher at the level above 140 mW, while operating GNSS receiv-
to equip mobile phones with GPS receivers, which were able ers are responsible for significant fraction of battery consump-
to introduce a better positioning precision. Vast popularity of tion. Traffic accidentdetecting application can be combined
smart-phones allows for the utilization of implemented system with other services such as navigation or traffic jam avoidance
by a sufficient fraction of the population without incurring any system; in that case, constant utilization of GNSS receiver is jus-
additional costs. In proposed systems, readouts gathered from tified. However, end users can consider continuous employment
accelerometer built in mobile device are analyzed by artificial of GNSS receiver only for the purpose of accident-detecting
intelligence based on feed-forward artificial neural networks. application as a waste of energy/resources. Researchers try to
Artificial intelligence was trained with real data emerging from determine approaches that can decrease utilization of power
test car crashes and nonaccidents car runs. Test cases encom- through application of various time patterns of coordinates
pass various types of cars, speeds, and types of roads. As a result, obtainment. Theoretically GNSS receiver can be turned off all
system that is able to achieve high accuracy of accident detec- the time and utilized only in case of accident detection; how-
tion (at the level of test set composed of 18,266,234 samples from ever, the issue here is the time required for the first fix obtain-
accident and nonaccidents runs): sensitivity equal to 0.97 and ment. After activation of the receiver, time of determination of
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 525
location can span from single second up to a few minutes. This
process is shorter (single seconds) when receiver possesses cur-
rent information about almanac, ephemerides, and has synchro-
nized clock with satellite time; however, it extends when such
information is not available, and has to be downloaded directly
from satellites. For example, for GPS-based system, collection
of almanac data takes the longest time (at least 12.5min in fos-
ter conditions, longer if GNSS signal is degraded by obstacles); Health researches and
however, data are valid for several months. If ephemerides have monitoring applications
to be updated (which are valid up to 46h; Dorsey etal. 2005),
the process can take more than 30 s. When only time synchro-
nization is required, the process of fixing takes a few seconds.
Obtaining the first fix can be sped up if receiver is able to utilize
wireless services that provide almanac, ephemerides, accurate
time, and satellite status, for example, assisted GNSS (A-GNSS)
(Harper 2009, Van Diggelen 2009). Importantly, then the posi-
tion can be calculated within seconds even under poor satellite FIgu r e 25.20 Application of GNSS for health researches and moni-
signal conditions. In addition, utilization of such services can toring applications.
improve positioning accuracy, although GNSS-based readouts
conditions, on humans health state, and combination of health
without additional corrections should be sufficient for the pur-
hazards with locations (Thierry etal. 2013) (Figure 25.20).
pose of emergency site location.
Many studies concerning contextual effects influence on
Noteworthy is the fact that GNSS receivers in combination
health is focused on human residential area (Thierry et al.
with other sensors (such as pulse oximetry sensor) can sup-
2013); however, people spend a vast amount of time outside their
port other types of emergency situations such as heart failure
homes. Therefore, the ability of easy collection of co-ordinate
(Hashmi etal. 2005).
time series and combination of these data with information
GNSS play important role in disaster rescue and recovery
about users activities opens new possibility to conduct research.
actions support. Emergency Centre personnel need fresh data
However, some big issues with raw GNSS data are a vast amount
about locations of rescue resources in order to make proper deci-
of information that should be processed and lack of broader
sions in a timely manner (Ameri et al. 2009). Therefore, every
context. Therefore, subject of location activity is introduced and
rescue unit should be equipped with GNSS receiver that has the
specialized algorithms are employed in order to detect activity
ability to automatically send correct location to the Emergency
place location, based on GNSS readouts (e.g., time distant clus-
Center. In addition, precise interactive maps of the location,
ter detection, kernel based, etc.). Then data might be analyzed
combined with automatic localization, allow for easy navigation
from a wider perspective and influence of series of consecutive
in the areas in which characteristic landmarks were hindered
activities locations on each other might be analyzed. Such an
by disasters, for example, flood (Van Westen 2000); it can also
approach might be useful for various researches, for example,
facilitate introduction of additional support units that are not
in the determination of causality in epidemiological studies
familiar with terrain (e.g., rescue units from other regions/coun-
(Thierry etal. 2013). In addition, the determination of life-space
tries). Pilot programs conducted in Italy showed that if medi-
can be the base for automatic estimation of older adults activity
cal personnel are supported by management application, that is,
level and their health status (Wan and Lin 2013).
utilizing GNSS readouts, then the response time can be reduced
from 16 to 9min (on average) (Bellini et al. 2013). Furthermore,
unmanned autonomic carriers can be sent to the disaster site in
25.6.7 tracking Applications
order to gather visual data from chosen areas. Navigation of car-
riers and activation of the camera might be based on the loca- Other area of GNSS-based tracking is utilized by researchers to
tion obtained from GNSS receiver. GNSS-based system can track behavior of animals (Zagami etal. 1998). Various animals
also support early warning of population about disasters such as can be equipped with devices that transmit their locations. Such
earthquakes and volcanic eruptions. researches have many practical aspects, for example, leading to
better understanding of causes of road collisions with animals
or allow for the determination of animals grazing areas (Rutter
25.6.6 Health Researches and
etal. 1997). GNSS collars are often coupled with various sensors
Monitoring Applications
(e.g., neck movement sensor), in order to improve abilities of
Combination of GNSS with mobile devices, which are car- automatic determination of animal activity. Issue in this area
ried ceaselessly by users, opens up new possibilities for various of application is battery lifetime. Tracking devices should be
location services. Interesting application of such services is the small and light, they should be virtually unnoticeable for the
trial of assessment of contextual effects, such as environmental tracked animal; on the other hand, battery life time depends
2016 by Taylor & Francis Group, LLC
526 Remotely Sensed Data Characterization, Classification, and Accuracies
SMS GPRS
FIg u r e 2 5.21 GNSS are widely utilized in tracking applications.
on its size and weight. When it comes to farm animals, replace- supported by SBAS, are processed at the same time. It helps to
ment of power source is relatively easy; however, often replace- improve precision; however, still there is a possibility of obtained
ment of batteries in devices designed for wild animals is not coordinates getting affected by high errors or event complete
possible, because it introduces stress and might cause temporal loss of GNSS signal in tunnels. Therefore, advanced statistic
modification of animal behavior. Therefore, devices utilized techniques are applied in order to determine whether coordi-
for this purpose are characterized by low frequency of coordi- nates obtained with GNSS should be accepted and utilized. In
nates obtained and limited communication with central server order to be able to determine the state of an object between suc-
(Gottardi etal. 2010) (Figure 25.21). cessive accepted GNSS readouts, obtain better insight into state
When it comes to person localization, GNSS-based devices of controlled process as well as increase system state sampling
are perfect for systems that are responsible for children tracking. frequency and to be able to direct the process despite the fact
Parents can utilize such devices in order to check whether their that some GNSS readouts might be rejected, Internal Navigation
descendants are in the proper location, for example, school/ Systems (INS) can be utilized. They are composed of additional
home. In combination with correct algorithms, deviations from sensors that give information about state of controlled object; as
daily routines can be determined and caretakers can be auto- a result, when proper calculations are done, location of an object
matically informed. Similar systems can be utilized in case of can be deduced. This group includes accelerometers, gyros, mag-
persons with dementia or similar illness, that influence their netometers, etc. Therefore, there is a possibility of determination
ability of returning to the place of living (Landau et al. 2011). of speed, and angle, of controlled object movement; thus, the cal-
Also prison system could reduce costs, because commitment culation of its coordinates based on previously determined posi-
of minor crimes might be punished with home detention and tion, and results obtained from employed sensors, is possible.
tracking of convicts can be done with GNSS collars. Integrated battery of such supporting devices is referred to as
Monitored person should be aware about carried GNSS internal measurement unit (IMU). Main advantage of these sen-
receiver; however, there is ongoing discussion whether caretak- sors is high refresh rate (typical 100Hz; however, it can be more
ers can utilize such devices, without the knowledge or agree- in case of utilization of industrial level devices); in addition,
ment of tracked person. most of them is jamming and deception resistant (Sukkarieh
etal. 1998, Zhong etal. 2008). However, a major problem of IMU
application is fast accumulation of readout errors that, with time,
25.6.8 Unmanned Vehicles Applications
leads to significant degradation of calculated parameters; there-
GNSS are utilized in various areas of autonomous and semiauton- fore, periodical calibration is required. In consequence system
omous systems. Very promising field is application of this tech- controlling process cannot be solely based on IMU readouts.
nology in autonomous land vehicle applications, for example, for Combination of IMU with GNSS is a great way of obtaining high
open-cast mining, agriculture applications (Heraud and Lange IMU frequency refresh rate with periodical calibration provided
2009, Prakash et al. 2012), cargo, and human transportation by GNSS. Usually at the heart of such systems Kalman filters
(Manabu etal. 2006, Bevly and Cobb 2010, Ozguner etal. 2011) are applied (Bullock etal. 2005, Groves 2008), which works in
(especially in extreme conditions; Nagashima etal. 2013). Also two stages: predictionresponsible for constant determination
domain of autonomous robots is developing quickly. Precise con- of coordinates, velocity as well as attitude based on IMU read-
trol process needs nonstop monitoring of the state of an object; outs, and estimationresponsible for determination of errors of
however, refresh rate of typical GNSS receiver varies between 1 aforementioned values and correction of their values when valid
and 10 MHz; in addition, precision of particular GNSS is also GNSS readout is available. In terrestrial applications, in order
limited, usually to a few meters. Furthermore, GNSS receivers to minimize errors introduced by IMU, there is a possibility of
are subject of errors, especially common are multipath errors live determination of values of its internal sensors biases (that
and signal loss caused by occlusion of navigation signal by vari- are influenced by various factors, for example, temperature)
ous obstacles, for example, high buildings in city centers. There which is done every time when the controlled object is station-
are many problems to solve, because consequences of errors in ary. Then obtained results are taken into account in prediction
autonomous system can be tragic. In order to overcome preci- stage. In addition detection of stationary state triggers reset of
sion issues, often signals from many satellite navigation systems, internally tracked velocity.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 527
Another way of improving of GNSS positioning accuracy sensors (e.g., acoustic ones) (Tian etal. 2010). Message encryp-
is utilization of magnetic markers incorporated in road lanes tion could also require precise time synchronization (Bahder
(Hernandez and Kuo 2003) or application of radio-frequency and Golding 2004).
identification (RFID)-assisted localization in combination with Utilization of GNSS services allows for achieving of time
sensor network (Lee et al. 2009). Such systems are especially accuracy at the level of nanoseconds (HP 1996). A huge advan-
useful in emergency prevention and unmanned vehicles appli- tage of solutions based on GNSS is the low cost; as mentioned
cations, that is, collision avoidance for vehicular ad hoc network. earlier, access to GNSS signal is free of charge; therefore, only
In such a system, precise location and ability of determination maintenance of receiver infrastructure is required.
of ranges between cars are very important. Unfortunately, sys-
tems solely based on GNSS, even if supported by SBAS correc-
25.6.10 Military Applications
tions, suffer because of precision issue that could be caused by
local conditions. It happens because SBAS systems provide cor- GNSS-based system plays important roles in military applica-
rections, which are valid for large areas; therefore, they cannot tions. It might be confirmed by the fact that GPS and GLONASS
perfectly fit to all locations, because they are characterized by systems are controlled by army. Among many applications,
various environmental factors (e.g., trees, city canyons, etc.). these systems can be utilized by precise targeting units (Brown
RFID-assisted concept assumes that roadsides of tracks, on et al. 1999), support troops reallocation, and be incorporated
which system is utilized, should be equipped with RFID tags that into vehicles and soldiers navigation devices. It can hugely
allow the calculation of GNSS corrections. Then vehicles could improve cooperation between troops and facilitates activities in
improve position precision obtained from GPS with RFID read- unfamiliar terrains. GNSS is utilized by the army not only dur-
outs and broadcast corrections through wireless radio to neigh- ing military operations but also in civilian supporting activities,
borhood cars. As a result, units that are not equipped in RFID for example, during disaster recovery. Interestingly UE develops
readers, or are located on inner lane, outside of range of RFID system that will be controlled by civilians and focused mainly on
tags, can also benefit from the system. Such an approach allows civilian applications; however, services for various government
obtaining corrections fitted to local conditions, and enables agencies also will be provided (Figure 25.22).
precise determination of relation between moving objects; how-
ever, it is worthy to point out that it does not solve the issue of
25.6.11 Aeronautic Applications
localization when GNSS signal is not available in city canyons or
tunnels. Although additional equipment is required, RFID tags Contemporary GNSS became popular tool in the area of avionic
are already utilized on many roads, for example, for the purpose navigation (Clarke 1998). Utilization of these systems is regu-
of automatic tool collection; therefore. such systems might be lated by regional legislations (Andrade 2001). In general, when
based on already existing infrastructure. pure GPS or GLONASS signals are utilized, GNSS-based naviga-
tion equipment that is incorporated in plane is treated as supple-
mentary system. It happens because there is a lack of monitoring
25.6.9 time Synchronization Applications
of positioning precision within signal broadcasted by navigation
GNSS services require precise time synchronization between sat- satellites; it implies that information about not complying with
ellites and receivers clocks. Therefore, other applications of such predetermined accuracy levels, determined solely on readouts
systems emerged. Precise time and frequency synchronization, from these systems, is missing. In addition, precision of obtained
among different sites on the globe, are important in many areas coordinates in vertical plane is worse than those obtained in hor-
(Hewlett Packard Company 1996). High-frequency trading uti- izontal planes. Avionics applications require high precision and
lizes precise time in order to time-stamp transactions (Korreng reliability of GNSS-based positioning in all spatial planes. As
2010); mobile telecommunication requires precise time synchro- mentioned before, in order to overcome these issues, and allow
nization between base stations so that they can efficiently share aircraft navigation to be based primarily on GNSS, SBASs were
common radio spectrum (Sterzbach 1997, Chunping etal. 2003). deployed: for example, in Europe, ESA together with air traffic
Precise time is important in all kinds of sensor networks, which control organizations created EGNOS. These systems provide
collect data from the environment, for example, seismographic
researches need time synchronization between sensors scat-
tered around the globe, power companies utilize precise time for
efficient energy distribution and pining down issues of distur-
bances in the network (Fan et al. 2005), physics can precisely
measure various values (e.g., one-way light speed) (Gift 2010),
observational astronomy needs time synchronization between
data-collecting sites. Structural health monitoring (Kim et al.
2012) such as in pipeline transportation can precisely localize
malfunction site through utilization of synchronized sensors
and measurement of error signal time detection by individual FIg u r e 2 5.22 Various kinds of military equipment utilize GNSS.
2016 by Taylor & Francis Group, LLC
528 Remotely Sensed Data Characterization, Classification, and Accuracies
25.7 Summary
Utilization of GNSS in various terrestrial applications is char-
acterized by increasing popularity. It is visible in a number of
aspects of human lives, various scientific researches are based on
data gathered with GNSS; many companies rely on the proper
functionality of these systems. The ability of rapid determina-
tion of precise location across the globe is widely utilized in mili-
tary and disaster recovery applications, even artists are trying
to use GNSS data as a form of artistic expression (Lauriault and
FIg u r e 25.23 GNSS allow for determination of precise coordinates Wood 2009).
of planes. The most important factors which allowed GNSS for such a
popularity are availability of navigation signals (free of costs,
information about GNSS failures and at the same time increase everywhere on Earth), good positioning precision (especially
positioning accuracy. Various organizations (e.g., Institute of when supported by various augmentation systems), and high
Computer Science in Silesian University of Technology) are reliability of service (i.e., rare cases of downtimes of GPS system
involved in constant monitoring of performance correctness of functionality). In addition, after achievement of fully operational
GPS, EGNOS, and GLONASS systems. Positioning accuracy, state by GALILEO system, which will be fully controlled by civil-
accessibility, and reliability of data provided by GNSS satel- ian institutions, new applications of these systems are expected.
lites are collected and sent to EGNOS Data Collection Network
(EDCN) that is controlled by EUROCONTROL, the organiza- Acknowledgments
tion that is responsible for ceaseless monitoring of accuracy
of functionality of GNSS for the purpose of air navigation in The research leading to these results has received funding from
European airspace (Figure 25.23). the PEOPLE Programme (Marie Curie Actions) of the European
The huge advantages of these systems are reliability and cost Unions Seventh Framework Programme FP7/20072013/
reduction of infrastructure. Therefore, development works under REA grant agreement no 285462. We also thank Marcin
are conducted that should result in increase of involvement of Paszkuta for support in the figures creation.
GNSS in this field. Application of GNSS improves work of air
traffic control, ability of precise location of airplanes leads to References
avoidance of collision and optimization of routes for reduction
of travel time and fuel consumption. Promising area of GNSS- Ahn, S.M., C. Park, J. Kang, 2011, Application of GPS fleet tracking
based system incorporation involves support (and in the future and stochastic simulation to a lean soil excavation practice,
automation) of procedure of aircraft approach to landing so- Proceedings of the 28th ISARC, Seoul, Korea, pp.335336.
called localizer performance with vertical guidance (LPV) pro- Ainsworth, S., B. Thomason, 2003, Where on Earth are we? The
cedures. It facilitates the optimization of landing approaches Global Positioning System (GPS) in archaeological field
paths and reduces the amount of financial resources that have survey, Technical Paper, English Heritage Publishing,
to be devoted for airfield infrastructure (e.g., it can rule out Swindon, England.
necessity of investment in instrument landing system [ILS]), at Allan, D.W., 1997, The Science of Timekeeping, Hewlett Packard
the same time airplanes can approach runways even in condi- Application Note 1289.
tions of low visibility. Implementation of LPV procedures has Ameri, B., D. Meger, K. Powert, 2009, UAS applications: Disaster
many advantages; some of them are: & emergency management, ASPRS Annual Conference,
Baltimore, MD.
Can be utilized on thousands of airfield across the world
Andrade, A.A.L., 2001, The Global Navigation Satellite System:
(more than 1800)
Navigating into the New Millennium, Ashgate, Surrey, U.K.,
Approach path eliminates intermediate phases (dive and
ISBN 9780754618256.
drive)
Aranzulla, M., F. Cannav, S. Scollo, G. Puglisi, G. Imm, 2013,
Approach path is independent of airfield and airplane
Volcanic ash detection by GPS signal, GPS Solutions, 17(4),
barometrical equipment
485497.
Reduction of influence of low temperatures and influence
Ashbrook, D., T. Starner, 2003, Using GPS to learn significant
of errors in flight instruments settings
locations and predict movement across, Personal and
Therefore, LPV procedures increase safety of air traffic. In order Ubiquitous Computing, 7(5), 275286.
to perform LPV approach, equipment that meets the specific Bahder, T.B., W.M. Golding, 2004, Clock synchronization based
requirements determined by current legislations, such as dual- on second-order quantum coherence of entangled photons,
frequency GPS receiver compatible with SBAS, as well as special AIP Conference Proceedings, Glasgow, U.K., Vol. 734, July
GPS receiver antenna, is required. 2529, p. 395.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 529
Bain, R.S., 2001, The use of GPS-based automatic vehicle location Fan, R., I. Chakraborty, N. Lynch, 2005, Clock synchroniza-
technologies for bus transit: State of the practice in the USA tion for wireless networks principles of distributed sys-
and lessons for elsewhere, European Transport Conference, tems, Lecture Notes in Computer Science, Higashino, T.
September 1012, Homerton College, Cambridge. Ed., Principles of Distributed Systems Lecture Notes in
Bar-Gera, H., 2007, Evaluation of a cellular phone-based system for Computer Science Volume 3544, Springer, New York,
measurements of traffic speeds and travel times: A case study Vol.3544, pp. 400414.
from Israel, Transportation Research Part C, 15, 380391. Frei, E., A. Ryf, R. Scherrer, 1993, Use of the global positioning
Bargeton, A., F. Moutarde, F. Nashashibi, A.S. Puthon, 2010, system in dam deformation and engineering surveys, SPN 2.
Joint interpretation of on-board vision and static GPS car- Gift, S.J.G., 2010, One-way light speed measurement using the
tography for determination of correct speed limit, CoRR, synchronized clocks of the global positioning system (GPS),
abs/1010.3867. Physics Essays, 23(2), 271.
Bellini, P., S. Boncinelli, F. Grossi, M. Mangini, P. Nesi, L. Sequi, Gleason, S., D. Gebre-Egziabher, 2009, GNSS Applications and
2013, Mobile Emergency, and Emergency Support System Methods, Artech House, London, U.K., ISBN 9781596933309.
for Hospitals in Mobile Devices: Pilot Study, Journal of Gleason, S., D. Gebre-Egziabher, 2010, Global Navigation Satellite
medical internet research protocols, 2(1):e19 Systems, McGraw-Hill Education (India) Pvt Limited,
Bevly, D.M., S. Cobb, 2010, GNSS for Vehicle Control, Artech Noida, India, ISBN 9780070700291.
House, London, U.K., ISBN 9781596933026. Gottardi, E., F. Tua, B. Cargnelutti, M.L. Maublanc, J.M. Angibault,
Brown, A.K., G. Zhang, D. Reynolds, 1999, Precision target- S. Said, H. Verheyden, 2010, Use of GPS activity sensors to
ing using GPS/internal-aided sensors, ION 55th Annual measure active and inactive behaviours of European roe
Meeting, June 2730, Cambridge, MA, pp. 689695. deer, Mammalia, 74(4), 355362.
Bullock, J.B., M. Foss, G.J. Geier, M. King, 2005, Integration of GPS Grewal, M.S., L.R. Weill, A.P. Andrews, 2001, Global Positioning
with other sensors and network assistance, Understanding Systems, Inertial Navigation, and Integration, John Wiley &
GPS: Principles and Applications, Kaplan, E.D., Hegarty, C.J., Sons, New York, ISBN 978-0-47135-032-3.
eds., Artech House, London, U.K., ISBN 9781580538954. Groves, P.D., 2008, Principles of GNSS, Inertial, and Multi-Sensor
Bussios, N., Y. Tsolakichy, M. Tsakiri-Strati, O. Goergoula, 2004, Integrated Navigation Systems, Artech House, London, U.K.,
Integrated high resolution satellite image, GPS and carto- ISBN 9781580532556.
graphic data in urban studies, Municipality of Thessaloniki, Hariharan, R., J. Krumm, E. Horvitz, 2005, Web-enhanced GPS,
International Society for Photogrammetry and Remote First International Conference on Location and Context
Sensing 2004, Istanbul, Turkey. July 1223, pp. 215221. Awareness LoCA05, May 1213, Bavaria, Germany,
Chunping, L., R. Shuangchen, G. Xiangdong, 2003, The pro- pp. 95104.
grammable logic implementation of GPS/GLONASS clock Harper, N., 2009, Server-Side GPS and Assisted-GPS in Java, Artech
synchronization, ASIC03, Fifth International Conference, House, Incorporated, London, U.K., ISBN 9781607839866.
Vol. 2, Beijing, China, October 2124, pp. 732735. Harte, L., B. Levitan, 2007, GPS Quick Course: Technology,
Clarke, B., 1998, Aviators Guide to GPS, McGraw-Hill, New York, Systems and Operation, Althos, Fuquay Varina, NC, ISBN
ISBN 9780070094932. 9781932813708.
Cosentino, R.J., D.W. Diggle, M.U. de Haag, C.J. Hegarty, D. Hasan, K.S., M. Rahman, A.L. Haque, M.A. Rahman, T. Rahman,
Milbert, J. Nagle, 2005, Understanding GPS: Principles M.M. Rasheed, 2009, Cost effective GPS-GPRS based
and Applications, Kaplan, E.D., Hegarty, C.J., eds., Artech object tracking system, Proceedings of the International
House, London, U.K., ISBN 9781580538954. MultiConference of Engineers and Computer Scientists,
Cyran, K.A., D. Sokoowska, A. Zazula, B. Szady, O. Antemijczuk, March 1820, Hong Kong, China, Vol. 1, pp. 1820.
2011, Data gathering and 3D-visualization at OLEG mul- Hashmi, N., D. Myung, M. Gaynor, S. Moulton, 2005, A sensor-
ticonstellation station in EDCN system, Proceedings of based web service-enabled emergency medical response
the 21st International Conference on Systems Engineering, system, Workshop on End-to-End, Sense-and-Respond
August 1618, Las Vegas, NV, pp. 221226. Systems, Applications, and Services, EESR 05, Seattle, WA,
DoD, 2008, Global Positioning System Standard Positioning pp. 2529.
Service Performance Standard, 4th edn., U.S. Department of Henriksson, E., M. Ostrm, A. Eriksson, 2001, Preventability of
Defense, Washington, DC. vehicle-related fatalities. Accident analysis and prevention,
Dorsey, A.J., W.A. Marquis, P.M. Fyfe, E.D. Kaplan, L.F. Wiederholt, 33(4), 467475.
2005, GPS system segments, Understanding GPS: Principles Heraud, J.A., A.F. Lange, 2009, Agricultural automatic vehicle guid-
and Applications, Kaplan, E.D., Hegarty, C.J., eds., Artech ance from horses to GPS, Agricultural Equipment Technology
House, London, U.K., ISBN 9781580538954. Conference, February 912, Louisville, KY, pp. 167.
ESA SP-1303, The European EGNOS Project, 2006, EGNOS The Hernandez, J.I., C.Y. Kuo, 2003, Steering control of automated
European Geostationary Navigation Overlay SystemA vehicles using absolute positioning GPS and magnetic
Cornerstone of Galileo, ESA Publications Division, markers, IEEE Transactions on Vehicular Technology, 52(1),
Noordwijk, the Netherlands, ISBN: 92-9092-453-5. 150161
2016 by Taylor & Francis Group, LLC
530 Remotely Sensed Data Characterization, Classification, and Accuracies
Hewlett Packard Company, 1996, GPS and precision timing Matthews, V.M., E. Adetiba, 2011, Vehicle accident alert and loca-
applications, HP Application Note, No 1272. tor (VAAL), International Journal of Electrical & Computer
Hofman-Wellenhof, B., H. Lichtenegger, E. Wasle, 2008, GNSS Sciences IJECS-IJENS, 11(02), 3538.
Global Navigation Satellite Systems: GPS, GLONASS, Galileo, Miller, K.M., 2000, A review of GLONASS, The Hydrographic
and More, Springer, New York, ISBN 978-3-211-73012-6. Journal, 98, 1521.
Hou, S.T., F. Chia, 2011, Making senses of technology: A triple Nagashima, K., K. Yamada, A. Tadano, 2013, Driverless Antarctic
contextual perspective of GPS use in the taxi industry, tractor system, Hitachi Review, 62(3), 229230.
service sciences (IJCSS), International Joint Conference on Ozguner, U., T. Acarman, K.A. Redmill, 2011, Autonomous Ground
Service Sciences, Taipei, Taiwan, May 2527. Vehicles, Artech House, London, U.K., ISBN 9781608071937.
Huang, J.Y., C.H. Tsai, S.T. Huang, 2012, The next generation Pan, G., G. Qi, Z. Wu, D. Zhang, 2013, Land-use classification
of GPs navigation systems, Communications of the ACM, using taxi GPS traces intelligent transportation systems,
55(3), 8493. IEEE Transactions, 14(1), 113123.
Inks, S.A., T.W. Loe, 2005, The ethical perceptions of salespeo- Petersen E.H., 2000, Draft report on the Commission commu-
ple and sales managers concerning the use of GPS track- nication to the Council, the European Parliament, the
ing systems to monitor salesperson activity, Marketing Economic and Social Committee and the Committee of the
Management Journal, 15(1), 108. Regions on the Priorities in EU road safetyProgress report
Jeffrey, C., 2010, An Introduction to GNSS: GPS, GLONASS, and ranking of actions ((COM(2000) 125C5 0248/2000
Galileo and Other Global Navigation Satellite Systems, 2000/2136(COS)), Committee on RegionalPolicy, Transport
1st edn., NovAtel Inc., Calgary, Alberta, Canada, ISBN and Tourism.
978-0-9813754-0-3. Petrovski, I.G., T. Tsujii, 2012, Digital Satellite Navigation and
Kalinowska, D., K. Hartmut, 2006, Motor vehicle use and travel Geophysics: A Practical Guide with GNSS Signal Simulator
behaviour in Germany: Determinants of car mileage, DIW- and Receiver Laboratory, Cambridge University Press,
Diskussionspapiere, No. 602. Cambridge, U.K., ISBN 9780521760546.
Kaplan, E.D., J.L. Leva, D. Milbert, M.S. Pavloff, 2005, Prakash, N.R., D. Kumar, K. Nandan, 2012, An autonomous
Fundamentals of satellite navigation, Understanding GPS: vehicle for farming using GPS, International Journal
Principles and Applications, Kaplan, E.D., Hegarty, C.J., eds., of Electronics and Computer Science Engineering, 1(3),
Artech House, London, U.K., ISBN 9781580538954. 16951700.
Kim, R., T. Nagayama, H. Jo, B.F. Spencer, 2012, Preliminary study Prasad, R., M. Ruggieri, 2005, Applied Satellite Navigation Using
of low-cost GPS receivers for time synchronization of wire- GPS, GALILEO, and Augmentation Systems, Artech House,
less sensors, Sensors and Smart Structures Technologies for London, U.K., ISBN 9781580538145.
CivilMechanical and Aerospace Systems, 83451A. Rutter, S.M., N.A. Beresford, G. Roberts, 1997, Use of GPS to
Kim I., C. Park, G. Jee, J. G. Lee, 1998, GPS positioning using identify the grazing areas of hill sheep, Computers and
virtual pseudorange, Control Engineering Practice, 6(1), Electronics in Agriculture, 17, 177188.
2535. Silva, J.M., R.G. Olsen, 2002, Use of global positioning sys-
Korreng, M.D., 2010, UTC time transfer for high frequency trad- tem (GPS) receivers under power-line conductors, Power
ing using IS-95 CDMA base station transmissions and Delivery IEEE Transactions, 17(4), 938944.
IEEE-1588 precision time protocol, 42nd Annual Precise Skov-Petersen, H., R. Rupf, D. Kchli, B. Snizek, 2012, Revealing
Time and Time Interval (PTTI) Meeting, November 1518, recreational behaviour and preferences from GPS record-
Reston, VA, pp. 359368. ings, MMV6Stockholm 2012, Session 4ERecent
Landau, R., G.K. Auslander, S. Werner, N. Shoval, J. Heinik, 2011, advances in visitor monitoring: GPS tracking and GIS
Who should make the decision on the use of GPS for people technology.
with dementia? Aging Mental Health, 15(1), 7884. Sterzbach, B., 1997, GPS-based clock synchronization in a mobile,
Lauriault, T.P., J. Wood, 2009, GPS tracingsPersonal cartog- Real-Time Systems, 12(1), 6375.
raphies, art & cartography, special issue, The Cartographic Sukkarieh, S., E.M. Nebot, H.F. Durrant-Whyte, 1998, Achieving
Journal, 46(4), 360365. integrity in an INS/GPS navigation loop for autonomous
Lee, E.K., S. Yang, S.Y. Oh, M. Gerla, 2009, RF-GPS: RFID assisted land vehicle applications, Robotics and Automation, IEEE
localization in VANETs, mobile ad hoc and sensor systems, International Conference, May 20, Leuven, Belgium, Vol. 4,
MASS 09, IEEE Sixth International Conference, October May 1620, pp. 34373442.
1215, Macao, China, pp. 621626. Sukkarieh, S., E.M. Nebot, H.F. Durrant-Whyte, 1999, A high
MacGougan, G.D., 2006, A Short Summary of Tropospheric integrity IMU/GPS navigation loop for autonomous land
Math Models By Glenn D. MacGougan (October 11, 2006). vehicle applications, IEEE Transactions on Robotics and
Manabu, O., H. Naohisa, F. Takehiko, S. Hiroshi, 2006, The appli- Automation, 15(3), 572578.
cation of RTK-GPS and steer-by-wire technology to the Thierry, B., B. Chaix, Y. Kestens, 2013, Detecting activity loca-
automatic driving of vehicles and an evaluation of driver tions from raw GPS data: A novel kernel-based algorithm,
behavior, IATSS RESEARCH, 30(2), 2938. International Journal of Health Geographics, 12, 14.
2016 by Taylor & Francis Group, LLC
Global Navigation Satellite Systems for Wide Array of Terrestrial Applications 531
Thompson, C., J. White, B. Dougherty, A. Albright, C.D. Schmidt, Yang, M., C.F. Lo, 2000, Real-time kinematic GPS positioning for
2010, Using smartphones to detect car accidents and pro- centimeter level ocean surface monitoring, Proceedings of
vide situational awareness to emergency responders, National Science Council ROC(A), 24(1), 7985.
Lecture Notes of the Institute for Computer Sciences, Social Zagami J.M., S.A. Pari, J.J. Bussgang, K.D. Melillo, 1998, Providing
Informatics and Telecommunications Engineering, 48, 2942. universal location services using a wireless E911 location
Tian, J., J. China, H. Zhao, 2010, Pipeline damage locating based network, IEEE Communications Magazine, 36(4), 6671.
on GPS fiducial clock intelligent control and automation Zheng, Y., L. Zhang, X. Xie, W.Y. Ma, 2009, Mining interesting
(WCICA), Eighth World Congress, July 79, Jinan, China, locations and travel sequences from GPS trajectories, 18th
pp. 41614164. International Conference on World Wide Web, Madrid,
Towns, D.M., L.M. Cobb, 2012, Notes on: GPS technology; Spain, pp. 791800.
employee monitoring enters a new era, Labor Law Journal, Zheng, Y.T., Z.J. Zha, T.S. Chua, 2012, Mining travel patterns from
63(3), 203. geotagged photos, ACM Transactions on Intelligent Systems
Van Diggelen, F., 2009, A-GPS: Assisted GPS, GNSS, and SBAS, and Technology TIST, 3(3), 56:156:18.
Artech House, London, U.K., ISBN-13 978-1-59693-374-3. Zhong, P. etal., 2008, Adaptive wavelet transform based on cross-
Van Westen, C., 2000, Remote sensing for natural disaster manage- validation method and its application to GPS multipath
ment, International Archives of Photogrammetry and Remote mitigation, GPS Solution, 12, 109117.
Sensing, Vol. XXXIII, Part B7, Amsterdam, the Netherlands. Ziedan, N.I., 2006, GNSS Receivers for Weak Signals, Artech
Wan, N., G. Lin, 2013, Life-space characterization from cellular House, Incorporated, London, U.K., ISBN 9781596930520.
telephone collected GPS data Computers, Environment and Zinkiewicz, D., B. Buszke, M. Houdek, F. Toran-Marti, 2010,
Urban Systems, 39, 6370. SISNeT as a source of EGNOS information: Overview of
Ward, P.W., J.W. Betz, C.J. Hegarty, 2005a, GPS satellite signal char- functionalities and applications, Fifth ESA Workshop on
acteristics, Understanding GPS: Principles and Applications, Satellite Navigation Technologies and European Workshop on
Kaplan, E.D., Hegarty, C.J., eds., Artech House, London, GNSS Signals and Signal Processing (NAVITEC), December
U.K., ISBN 9781580538954. 810, Noordwijk, the Netherlands, pp. 17.
Ward, P.W., J.W. Betz, C.J. Hegarty, 2005b, Interference, mul- Figures which were included from external sources, or created
tipath and scintillation, Understanding GPS: Principles with external applications (agreement of external authors were
and Applications, Kaplan, E.D., Hegarty, C.J., eds., Artech obtained)
House, London, U.K., ISBN 9781580538954. SBAS simulatorhttp://www.iguassu.cz/sbas-sim/by:
Wei, W., S. Baosheng, Z. Kefa, Z. Xia, 2010, The use of GPS for IGUAU Software Systems, Evropsk 120, Dejvice, 160 00 Praha
the measurement of details in different areas, Electronic 6, Czech Republic
Commerce and Security (ISECS), Third International Figure 25.1, Figure 25.2, Figure 25.3, Figure 25.4
Symposium, July 2931, Guangzhou, China, pp. 5962. U-Center software by UBLOX http://www.u-blox.com by:
White, J., C. Thompson, H. Turner, B. Dougherty, D.C. Schmidt, u-blox AG, Zrcherstrasse 68, 8800 Thalwil, Switzerland
2011, Automatic traffic accident detection and notification Figure 25.14
with smartphones, Mobile Networks and Applications, 16(3), PolaRx Control Software http://www.septentrio.com/by:
285303. Septentrio nv, Greenhill Campus, Interleuvenlaan 15G, 3001
Xu, G., 2007, GPS: Theory, Algorithms and Applications, 2nd edn., Leuven, Belgium
Springer, New York, ISBN 978-3-540-72714-9. Figure 25.10
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
IX
Crowdsourcing and
Remote Sensing Data
26 Crowdsourcing and r emote Sensing: Combining Two Views of Planet e arth Fabio DellAcqua ............................ 535
Acronyms and Definitions Introduction What Is Crowdsourcing From Web 1.0 to Web 2.0 and Social
Media Citizen Sensor Current Implementations Conclusions Acknowledgments References
533
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
26
Crowdsourcing and
Remote Sensing:
Combining Two Views
of Planet Earth
Acronyms and Definitions .................................................................................................................535
26.1 Introduction .............................................................................................................................535
26.2 What Is Crowdsourcing ..........................................................................................................536
26.3 From Web 1.0 to Web 2.0 and Social Media ........................................................................536
26.4 Citizen Sensor...........................................................................................................................538
26.5 Current Implementations ...................................................................................................... 541
Seismic Risk: The Did You Feel It Service Seismic (and Multi)Risk: Exposure
Mapping Land Cover Updating Research (and Operability) in Progress
26.6 Conclusions.............................................................................................................................. 547
Fabio DellAcqua Acknowledgments .............................................................................................................................. 547
Universit degli Studi di Pavia References............................................................................................................................................ 547
Acronyms and Definitions UGC User-Generated Content
U.S./USA United States of America
CA California USGS United States Geological Survey
CIIM Community internet intensity map VGI Volunteered Geographic Information
COST European Cooperation in Science and Technology WebGIS Web-based Geographic Information System
DO Direct Observation ZIP Zone Improvement Plan (mailing codes)
DYFI Did You Feel It service
ENV Environment theme of the Seventh Framework 26.1 introduction
Programme of the EU
EU European Union Although the term crowdsourcing, according to the official
FP 7 Seventh Framework Programme records, was used for the first time in 2006 in a Wired jour-
GED Global Exposure Database of GEM nal paper by Howe (2006), the concept of on-the-fly recruit-
GEM Global Earthquake Model ing of a generally large set of contributors to solving a specific
GEO Group on Earth Observations problem dates back by centuries. For instance, it is well known
GEOSS Global Earth Observation System of Systems that in 1714, the British government offered a money prize (the
GIS Geographic Information System Longitude Prize) to anyone who could devise a reliable method
IDCT GEM inventory data capture tools to compute the longitude a ship was located at (Sobel, 1998); this
LANDSAT Land Remote Sensing Satellite Program was an example of entrusting a nonorganizedyet having unex-
MMI Modified Mercalli Intensity scale for earthquakes plored intelligent capabilitiespool of people with the task of
OECD Organisation for Economic Co-operation and solving a specific, information-retrieval problem. Probably less
Development widely known is the fact that Toyota, the Japanese car manu-
OSM OpenStreetMap facturer, started a competition in 1936 to get its current logo,
SDI Spatial Data Infrastructures chosen among 27,000 candidate entries (DesignCrowd, 2010).
SMS Short Message Service Several examples of such calls can be found across history, but
TCP/IP Transmission Control Protocol/Internet Protocol it was not until the advent of the World Wide Web, and especially
535
2016 by Taylor & Francis Group, LLC
536 Remotely Sensed Data Characterization, Classification, and Accuracies
the Web 2.0, that crowdsourcing found the optimal environ- 4. Wisdom of the crowd: Merging contributions by individu-
ment for being widely applied in practice. The Internet, indeed, als, based on the idea that collective ideas are better that
allows recruitment of contributors from virtually any place on individual ideas, that is, asking many people increases
the Earth, easy sharing of resources, and immediate transmission chances of finding the right solution to a problem
of results. This is why the Internet will have an important role in 5. Microwork: Splitting a big task into microtasks entrusted to
this book chapter. The Internet itself is based on the shared con- several human agents (see Amazon Mechanical Turk, 2005)
tribution of the infrastructures that physically compose it, yet the
As we will see in the following, crowdsourcing as a support to
term crowdsourcing refers to applications requiring intelligence
Earth observation mainly falls into the categories of crowd-
or otherwise capacity of a human operator to solve problems that
voting and wisdom of the crowd, although new specific concepts
machines are still unable to deal with. An example is the so-called
like the Citizen Sensor (Sheth, 2009) need to be considered for a
wiki philosophy, marked in 2001 by the birth of the free ency-
complete picture.
clopedia named Wikipedia (2001) that is maintained by millions
Crowdsourcing can thus be conceived as a production model, or
of volunteers, and later extended to tens of similarly conceived
problem-solving model, exploiting a sort of distributed intelligence
projects. Although Wikipedia writers work for free, this is not
or, as we will see for our topic of interest, distributed data collection.
necessarily a rule for all crowdsourced work: in 2005, for exam-
In its most classical conception, opinions regarding the solving of
ple, the Amazon Mechanical Turk was launched, featuring paid,
a given problem or the development of a project is sought from a
online crowdsourcing services. The name of the service was cho-
group of individualsa crowdnot necessarily defined a priori,
sen in reference with the famous mechanical Turkish chess player
which contribute, individually and/or through mutual interac-
built in 1770 by Wolfgang Von Kempelen (17341804) to shock
tions, to finding or defining a solution to the proposed problem,
the Austrian empress Maria Theresa: faking a mechanical chess
possibly subject to the proposers approval. The solution typically
player, it actually contained a disguised human operatorsome-
ends up belonging to the company or organization that initially
how like the modern realization of crowdsourcing, hiding human
proposed the problem, while the individuals that contributed to
intelligence behind an electronic interface and infrastructure.
the solution are compensated for their work through small money
This chapter will describe and analyze how this concept is
payments, prizes, or even simply moral or intellectual satisfaction.
exploited in the field of remote sensing and Earth observation,
The advantages of such a model with respect to traditional
where crowdsourcing and the Citizen Sensor may represent an
commercial models are obvious. First of all, the solution may
effective way to fill in the information gaps left behind by special-
be found at a reasonable costor even for zero, in terms of raw
ized remote sensing operation using conventional infrastructure
labor cost, if the remuneration consists of personal satisfaction
like space-borne and air-borne sensors. The chapter is organized
as in the case of Wikipedia. Then, the pool of considered, pos-
as follows: Section 26.2 introduces the concept of crowdsourc-
sible solutionsor, more relevant for our remote sensing case, of
ing, Section 26.3 discusses the concepts of user-generated con-
collectable informationis much wider than any single solver or
tent (UGC), Web 2.0 and social media. Section 26.4 gets closer to
information collector can reach.
the aim of the chapter by presenting the Citizen Sensor and vol-
On the other hand, the process of crowdsourcing oftenraises con-
unteered geographic information (VGI). Section 26.5 presents
cerns about the quality of the information produced (Allahbakhsh
some examples of implementations of the crowdsourcing con-
etal., 2013), as contributors might have different levels of skills and
cept in a geospatial and remote sensing environment. Section
expertise, possibly insufficient for performing certain taskseven
26.6 closes the chapter while drawing some conclusions.
simple data collection in many cases requires a minimum degree
of knowledge about the observed targets; contributors might also
26.2 What is crowdsourcing have various and even biased interests and incentives; even mali-
cious activities can be set up to bias the output of a crowdsourcing
As already mentioned, the term crowdsourcing was coined by operation. The usual countermeasures act on the sides of assessing
Howe (2006) as a neologism generated on the model of outsourc- workers profiles and reputation (Jsang etal., 2007; Alfaro etal.,
ing while implying the meaning of sourcing to the crowd, that is, 2011) and of smartly designing both the task (Dow etal., 2012) and
entrusting an external provider for the provision of a service, in the compensation schemes (Scekic etal., 2012).
which the external provider happens to be a group of people not
determined a priori rather than a specific, single agent.
After Howes invention of the term crowdsourcing, several
26.3 From Web 1.0 to Web 2.0
derived terms have been born. Some of the most notable ones are: and Social Media
1. Crowd-voting: Collecting opinions (possibly forced to fit The Internet as we came to know it when it started becoming
into a predefined set of options) on a given topic or product popular in the second half of the 1990s was later termed Web
2. Crowd-funding: Collective funding of a project by many 1.0. This term, which was not there at the time, was coined a
small investors/donors posteriori to point out the difference with the new features that
3. Creative crowdsourcing: Coworking on a collective were then emerging within the net, or, better, with the new kind
artwork of Internet that was developing.
2016 by Taylor & Francis Group, LLC
Crowdsourcing and Remote Sensing 537
The term Web 2.0 was indeed coined not too far away in time the one side, and (many more) information consumers on the
(OReilly, 2007), highlighting a trend that emerged since the other side. Search engines try and facilitate the access to the right
beginning of twentieth century but only recently consolidated. information among a flood of less relevantor irrelevant at all-
Web 2.0 does not mean a new revision of the World Wide Web: information. Figure 26.1 depicts the concept of Web 1.0.
it is still based on TCP/IP protocols, and all the elements that The Web 2.0 is instead based on a high level of interaction
characterized the Web since its beginning, like hypertext and between the website and the user, which frequently is, at the
linking among contents. Web 2.0 is rather the arriving point of a same time, also a generator of the information contained in the
seamless evolution in terms of approach to online information. websites, as highlighted in the scheme in Figure 26.2. Web 2.0
The Web 1.0 is indeed based on static websites, where the allows users to publish, that is, make available to anyone request-
user mainly accesses the information (or data) generated by a ing to access the website, some self-generated pieces of informa-
limited set of other agents. Here, a pretty clear distinction is tion. This is the so-called UGC (OECD, 2007), which entails
found between (comparatively few) information contributors on creative effort and nonprofessional, nonstandard labor.
many
Content
consumers
Web 1.0
Discussion email
HTML
group
pages
Hosting HTML
service
Few
File
Content creators server
Internet
Slow
connections
Media
files
Stores
Software and
distribution Snail mail
FIg u r e 2 6.1 Schematic concept of Web 1.0. (From http://en.wikiversity.org/wiki/File:Web_1.0_elements.png.)
Many
Web Content
2.0 Discussion sharers
group
email
Website
hosting
service
HTML
Few
Facilitators Internet Fast
Wiki Text
of content connections
sharing
Blog
Audio
Video
Media file sharing
FIg u r e 2 6.2 Schematic concept of Web 2.0. (From http://en.wikiversity.org/wiki/File:Web_2.0_elements.png.)
2016 by Taylor & Francis Group, LLC
538 Remotely Sensed Data Characterization, Classification, and Accuracies
The means through which this can be done are various and collected information is conveyed to a single, virtual place where it
include the so-called social networks, photo and video-sharing becomes usable, including by those who created it.
websites, blogs, forums, wikis, and other web services. The Among the most notable examples of VGI collectors, we find
information can be made completely public, or restrictions can Wikimapia (2007) and OpenStreetMap (2004) websites. The for-
be placed on who can access what, through, for example, regis- mer encourages participants to post comments about georefer-
tration and/or linking among specific individuals or groups. enced locations; the latter is an international effort to create a
By uploading material on such sites, commenting material free source of map data through volunteers efforts.
posted by others, and interacting with other individuals, the users In Figure 26.3, a comparison between the Google Maps ver-
generate a large amount of information, which is highly dynamic in sion and the OpenStreetMap (OSM) version of a map on the
that it is continuously amended, topped, and made to evolve includ- same area highlights the amount of information that the Citizen
ing through closed-loop feedback. The phenomenon was spontane- Sensor is willing to contribute to a public VGI repository. Note
ous, still fostered and sped up by the availability of user interfaces, that this version of the OSM map does not display the 3D build-
software, and online storage tools, perfectly integrated in the Web, ing information here, which is however frequently contributed
which is increasingly becoming a platform on which it is possible to (Over etal., 2010; Goetz and Zipf, 2012).
utilize different applications including remote GIS (WebGIS) ones. It is however to be noted that the comparison represented in
Not only the huge amount of shared information, especially Figure 26.3 refers to a small-sized town in a developed coun-
its evolution and the possibility of live interaction, build up a sort try, where a large fraction of the population has access to elec-
of collective intelligence (Levy, 1999) or global brain (Heylighen, tronic navigation devices and fast Internet connections. One
1997). This also casts a spotlight on Web 2.0 as a possible tool may wonder what is the real coverage to expect when the sight is
(the tool?) to practically implement crowdsourcing. These details expanded to a global level, including areas where such favorable
are discussed later in this chapter. conditions are not found at all.
The social media, as defined by Kaplan and Haenlein (2010), The level of coverage offered by a crowdsourced mapping ser-
are the expression of massive UGC generation in the framework vice like OSM is an interesting issue, but it is difficult to assess
of the Web 2.0 as an enabling technology. Kaplan and Haenlein for various reasons:
(2010) define different types of social media, but the most rel-
1. No reference data ensuring full coverage of the entire
evant for our purposes are the following:
globe is available, against which to measure OSMs cov-
1. Collaborative projects: Wikis, social bookmarking. eragethis could actually be solved by subsampling, but
2. Content communities: Sharing of content among users. there are other issues.
3. Social networking sites: Much less interesting as the con- 2. No standard definition of items to be included in a map is
tents are not public by default. provided, so a definite response on what is actually mapped
versus what should have been mapped cannot be given. For
In the following, we will see how the social media helps in build-
instance, an unpaved road could be considered unimport-
ing the so-called Citizen Sensor Networks.
ant if the focus is on residential area accessibility, while if
the map is to be used in an agricultural context, the same
26.4 citizen Sensor road becomes important; still on the unpaved road, in a
developing country, it may be as important as a paved one
Today, the Earth supports about 7 billion people; every year those
in a residential neighborhood.
people, according to statistics (IDC, 2013), purchase millions of
3. Assessing a global coverage of an ever-changing set in
smartphones with positioning, picture-taking, video- and sound-
diverse environments across the globe like that of the
recording capabilities; the figure for 2012 was 712 million units
man-made structures that cover the solid Earth surface is
(IDC, 2013), reporting a 44.1% increase with respect to 2011. It is
in itself a tricky task. Buildings appear so well delineated
not difficult to imagine that in a few years practically every inhab-
and separated from each other in Figure 26.3; other sort
ited area of the Earth will see the presence of at least one device with
of buildings may not be so easy to tell from each other in a
capabilities of acquiring environmental datain a wide senseand
context such as that of, for example, informal settlements
of transmitting it to the cyberspace. By the way, with a person using
in urban areas with social problems.
it, which thus happens to fit the definition of Citizen Sensor (Sheth,
2009); whether in an organized or informal way, a distributed set of The statements above naturally do not mean that nothing can be
Citizen Sensors forms a Citizen Sensor Network, that is, a distrib- concluded on the coverage and we just have to hope that the area we
uted network of potential, mobile, connected sensing instruments. are interested in is covered with all the information that we need.
The concept of Citizen Sensor, however, finds its natural place, Local studies provide encouraging results; (Neis et al., 2012; De
and gains even more sense, in the framework of the VGI concept Leeuw et al. (2012)), yet on limited areas; at a global level, qualitative
(Goodchild, 2007), that is, a special case of the more general Web assessments can be made. A recent, interesting example is the OSM
phenomenon of UGC. VGI is defined as the widespread engagement node density map published by Raifer (2014). As can be seen in
of large numbers of private citizensin the creation of geographic Figure26.4, the color-coded density correlates well with the pop-
information and it can be seen as the vehicle through which sparsely ulation distribution depicted in Figure 26.5. Low coverage rates
2016 by Taylor & Francis Group, LLC
Crowdsourcing and Remote Sensing 539
(a)
(b)
FIg u r e 2 6.3 The city centre of Pavia, Italy, as mapped on the Google Map website (a) and the OpenStreetMap website (b).
2016 by Taylor & Francis Group, LLC
540 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 2 6.4 OSM node density 2014. (From Martin Raifer, CC-BYsource data, OpenStreetMap contributors, ODbL.)
The World: Population Density, 2000 GPW [v3]
Robinson Projection
Based on 2.5 arc-minute resolution data
Gridded Population of the World
Persons per km2
0
14
524
25249
250999
1000+
This document is licensed under a
cc BY
Creative Commons 3.0 Attribution License
http://creativecommons.org/licenses/by/3.0/
FIg u r e 26.5 Global population density according to CIESIN as available from NASA SEDAC. (From Center for International Earth Science
Information Network (CIESIN), Columbia University; Centro Internacional de Agricultura Tropical (CIAT), Gridded Population of the World
(GPW), Version 3, CIESIN, Columbia University, Palisades, NY, Copyright 2005, available at: http://sedac.ciesin.columbia.edu/gpw.)
2016 by Taylor & Francis Group, LLC
Crowdsourcing and Remote Sensing 541
are reported in noninhabited (or nearly so) areas such as Saharan The DYFI program has recorded an increasing success.
Africa, in-land Greenland, or the Australian Outback. All in all, Currently, the total input questionnaire count largely exceeds
the maps suggest that where one expects to find settlements and 1 million units, and the service has been scaled up to world-
thus roads, they are actually found on OSM. wide coverage, allowing global contributors to feed their
observation through the dedicated outside US section of the
website. In Figure 26.7, the spatial distribution of questionnaires
26.5 current implementations filed between 1991 and 2012 is translated into a map, provid-
ing a compelling view of the seismic activity across the North
In this section, we will see how the concepts illustrated so far find
American continent, most interesting for purposes of seismicity
their application in the real world. Various ways of using VGI
on a continental scale.
mechanisms are illustrated here in different fields of applica-
A few words are in order on the accuracy of DYFI data, because
tions, including disaster management and land cover mapping.
they make an interesting case of how crowdsourcing can be use-
ful not only for qualitative but also for quantitative purposes. As
26.5.1 Seismic Risk: the DidYou
Atkinson and Wald (2007) stated, DYFI data make up in quantity
Feel it Service
what they may lack in quality. The authors focused on the 2004
Although crowdsourcing is potentially very useful to assess dam- Parkfield, CA, earthquake, comparing MMI data collected through
age to buildings and structures in the postdisaster phase (Kerle, the DYFI system across the affected ZIP codes with the measured
2011), the earliest examples in the field of seismic risk focus on earthquake ground motions from local instruments, as cataloged
mapping the phenomenon itself rather than its consequences. by the USGSs ShakeMap (2004) Web-based database, and esti-
The Did You Feel It (DYFI; Wald etal., 2011) program was mated using consolidated models. The two data series match very
developed by the United States Geological Survey (USGS) in the well, not only in terms of general MMI vs. distance trend, but also
early 1990s, long before the concept of crowdsourcing came to be on more subtle features such as a flattening of attenuation in the dis-
formalized by Howe (2006), and can be considered the first nota- tance range from 70 to 150km. The uncertainty bars of DYFI data
ble example of crowdsourcing applied to natural hazard man- appeared a bit wider than those associated with the model-based
agement. DYFI consists of a website (DYFI, 1991) onto which estimation, but this is a reasonable price to be paid in exchange for
individuals (Citizen Sensors) can report their time-stamped, an incredibly dense network of virtual sensors.
georeferenced observations regarding a seismic phenomenon
that they have experienced. The idea underpinning DYFI is to
26.5.2 Seismic (and Multi)Risk:
best exploit the public engagement (an informal Citizen Sensor
exposure Mapping
Network) to generate a map of the event which be useful, in case
of a relevant event, to get a prompt input to the emergency proce- The global earthquake model (GEM, 2006) is an international
dures, but also to build a historical record of minor quakes to be forum where organizations and people come together (virtu-
used as an input to earthquake hazard models. The map is called ally, but also physically) to develop, use, and share tools and
community internet intensity map (CIIM), and it is effectively a resources for transparent assessment of earthquake risk. The
consolidated output of a crowdsourcing mechanism. GEMs collaborative efforts are expected to build a heightened
A short online questionnaire is proposed to the candidate public understanding and awareness of seismic risk, leading
contributor, including three different sections on context, expe- to increased earthquake resilience worldwide. One of the most
rience, and effects. The questionnaire was designed to generate important products of GEMs efforts is OpenQuake (Silva etal.,
reports fitting the modified Mercalli intensity (MMI) through 2013), a suite of open-source software allowing the GEM com-
suitable mapping of responses into MMI grades (Dengler and munity to use the data, best practices, and applications collab-
Moley, 1994; Dengler and Dewey, 1998). oratively being developed. The principle scheme of OpenQuake
Every filled questionnaire will thus result in an estimated MMI is illustrated in Figure 26.8.
grade, linked to the microregion to which the geolocation of the The platform relies on a geographical database containing the
contributor corresponds. Microregions are defined conventionally exposure data, that is, information on how human beings and
to correspond with U.S. zip codes as a compromise between statisti- goods, at stake in case of disasters, are distributed globally, at dif-
cal significance of the aggregates and spatial resolution. An average ferent levels of spatial detail. This is called the global exposure
value of the linked, estimated MMI grades is computed on every database (GED), and its structure is designed to contain informa-
microregion, and attached to it. Visually, the average is turned into tion on buildings and people from the country level all the way
a color through a conventional color palette ranging from white down to individual buildings. The first version of GED contains
(nonperceived) to dark red (extreme). Within a few minutes from aggregate information on population and the number/built area/
the earthquake occurrence, a CIIM starts building on the website, reconstruction cost of residential and nonresidential buildings
often collecting thousands of contributions for clearly felt quakes at a 1km resolution. Detailed datasets on single buildings will be
(from grade IV upward). Together with the CIIM map, a graph available for a selected number of areas and are increasing over
depicting how the average intensity decays with distance is also time, thanks to crowdsourcing mechanisms. New data are indeed
constructed; examples are shown in Figure 26.6. produced through a set of tools termed the GEM inventory data
2016 by Taylor & Francis Group, LLC
542 Remotely Sensed Data Characterization, Classification, and Accuracies
USGS Community Internet Intensity Map
Central California
December 22, 2003 11:15:56 PST 35.7058N 121.1017W M6.5 Depth: 7 km ID: nc40148755
Canta rosa Laguna
38N Vajo 38N
San Francisco
Fremont
San jose Merced
37N 37N
Santa Cruz
Fresno
Salines
Visalra
36N 36N
Delano
Bakerstield
35N 35N
Santa Maria
Palmdale
Santa Clanta
34N Los Angeles 34N
Long Beach
Murrieta
33N 21,970 responses in 1,066 ZIP codes (Max CDI = VII) 50 km Oceanside 33N
Encmitas
122W 120W 118W
Intensity I IIIII IV V VI VII VIII IX X+
Shaking Not felt Weak Light Moderate Strong Very strong Severe Violent Extreme
Damage None None None Very light Light Moderate Moderate/Heavy Heavy V. Heavy
(a) Processed: Friday September 7 19:06:57 2012
Distance vs. intensity plot (ID: nc40148755) Responses vs. time plot (ID: nc40148755)
10 20,000
All reported data
AtkinsonWald (2007) Calif.
9 AtkinsonWald (2007) Calif.
1 STDEV intensity each distance bin
8 Median intensity each distance bin 16,000
Number of responses
7
12,000
6
Intensity
5
8,000
4
3 4,000
2
Processed Friday Processed Friday
1 September 7 19:07:13 2012 0 September 7 19:07:13 2012
5 10 20 50 100 200 500 1000 0 1 2 3 4 5 6 7 8
(b) Hypocentral distance (km) (c) Time since earthquake (h)
FIg u r e 26.6 DYFI products for the for the Central California event on December 22nd, 2003. (a) Color-coded community Internet intensity
map. (b) Estimated intensity vs. distance plotting of aggregated responses. (c) Cumulated responses vs. time elapsed. All images from the DYFI
website. (From USGS, Reston, VA.)
2016 by Taylor & Francis Group, LLC
Crowdsourcing and Remote Sensing 543
0 50 100 200 300 400 500
Miles N
Did You Feel It?
Legend
did you feel it?
(intensity)
II
III
IV
V
VI
VII
VIII
IX
Signification event
0 50 100
0 50 100 200 300 400 500 Miles
Miles
0 50 100 200 400
Miles
FIg u r e 2 6.7 Distribution of questionnaires filed between 1991 and 2012 on the DYFI system (from the USGS DYFI web site).
Scientific and Platform Local/
regional
stakeholder Data, models, experts
community
Local data, results, tools Local data
models, results View, explore, capture
download for further use
Global Results Data and results
projects and
working groups
Global best practice
and datasets
Engine Modelling
Carry out hazard and tools
physical risk calculations
Develop hazard, exposure and
physical vulnerability models,
develop socio-economic variability indicators
Local/ Hazard, exposure and
regional physical vulnerability models
experts
Regional and
local models
FIg u r e 2 6.8 The OpenQuake system. (From Global Earthquake Model [GEM] Foundation, Pavia, Italy.)
2016 by Taylor & Francis Group, LLC
544 Remotely Sensed Data Characterization, Classification, and Accuracies
capture tools (IDCT). The IDCT and the accompanying user Tool to develop homogeneous exposure datasets
protocols are meant to enable users to collect and modify build- QuantumGIS plug-in to extract building footprints from
ing exposure information, which can be input into the GED. The satellite imagery
exposure information can be generated from remote sensing and Paper forms for building inventory capture
field observations through the following tools:
Figure 26.9 depicts the operation principle for GEMIDCT.
Building data capture application for Android phone or Figure 26.10 represents instead one output of a collaborative
tablet digitalization work of building footprints overlaid on a nadir
Windows tool for field data collection and management satellite image of LAquila, Italy.
PC Mobile device User collects
building info:
a. Basic inventory
b. Detailed inventory
c. Consequences
d. Internal
(of building)
Local data Local data with photos,
GPS
User syncs position using field
User defines building data to mobile tools or paper forms
footprints using A device
GIS/RS tools
User produces Fields tools
B homogeneous areas
using GIS/RS tools
GIS/RS Tools Sync and
SIDD aggregation User syncs
data from
mobile
device
User downloads from
OpenQuake platform
GIS RS tools User uploads data
SIDD from field tools to
Sync Aggregation tool User downloads OQ platform via PC
Field tools building info and sync tool and/or
PDF forms uses SIDD to fills in forms and
uploads photos via
produce mapping online prizm tool
User uploads scheme and
mapping schemes exposure data
and exposure data to
OQ platform
Openquake platform
ECD APIs for
Inventory, ECD, GED Prizm UI
GED
Physical
Mapping vulnerability
Library
schemes models
User requests
scientific review of
contributed data
FIg u r e 2 6.9 The operation model for GEM-IDCT. (From Global Earthquake Model [GEM] Foundation, Pavia, Italy.)
2016 by Taylor & Francis Group, LLC
Crowdsourcing and Remote Sensing 545
N
100 m
FIg u r e 26.10 Results of a collaborative building footprint digitization on LAquila, Italy. Yellow lines define boundaries of single buildings.
North up. (From Global Earthquake Model [GEM] Foundation, authors reworking, Pavia, Italy.)
Such data can be integrated with field observation through even in ideal conditions (Congalton, 1991). Also in this case,
the direct observation (DO) tools running on mobile devices; crowdsourcing can offer some help, with the Geo-Wiki Project
such tools have been made available even on the Google Play (Fritz etal., 2009). This project aggregates a global network of
(2014) repository, fully implementing the concept of crowd- volunteers willing to help improve the quality of global land
sourcing. Again, accuracy assessment is made difficult by the cover maps. Each single contributor is requested to review hot-
scarce availability of actionable ground truth data. Still-to-be- spot maps of global land cover and determine, based on what
published reports (Iannelli, 2014) on remote sensing IDCT tools they actually see in Google Earth and their local knowledge, if
convey figures of around 70% accuracy in footprint delineation the land cover maps are correct or incorrect at the spot under
in the cases of minimal human intervention. (Figure 26.10 refers review. Georeferenced ground pictures can be uploaded and
to manual digitization, which is a different story with accuracy support the proposed land cover class. Each input is recorded
reaching above 90%.) in a database, along with (possible) uploaded photos, to be used
The case of GEM is particularly interesting, because it real- in the future for the creation of a new and improved global
izes the crowdsourcing concept at two different stages of its life land cover map. Figure 26.11 shows the interface of Geo-Wiki,
cycle: during the definition of risk assessment practices and dur- where the familiar Google Earth Hi-Res mosaic is overlaid with
ing collection of hard data that will be used to feed the defined a much spatially coarser land cover map from the GlobCover
practices. (2005) dataset.
The first implementation is through GEM Nexus, that is, the A note on accuracy should also be made here. Accuracy
GEM collaboration platform where experts and professionals assessment of crowdsourced land cover data encounters the
from around the globe find an opportunity of working collec- same issues that have been pointed out in Chapter 4 for com-
tively on state-of-the-art global earthquake risk assessment. pleteness assessment of crowdsourced maps. The issues are made
The second implementation is through the IDCT set of tools even worse by the higher variability of land cover with respect
enabling a pool of contributors to generate and feed data into the to, for example, settlement location or road geometry. However,
system from insitu observation or processing of satellite images. assessments have been attempted, but obviously on datasets
whose size is very small compared to the global world coverage.
Foody etal. (2013) report a kappa coefficient of around 0.7 for a
26.5.3 Land cover Updating
10-class Geo-Wiki classification experiment based on labeling
Land cover mapping from multispectral spaceborne data dates of about 300 samples carried out by 65 volunteers, among which
back to the beginnings of the LANDSAT satellite series in 1972 the 10 most productive were selected for the final assessment.
(Anderson etal., 1976). It became soon evident that land cover Similar figures in terms of accuracy (60%70%) are reported in
classification from remotely sensed data is far from perfect, another Geo-Wiki experiment reported in See etal. (2013) where
2016 by Taylor & Francis Group, LLC
546 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 26.11 The Geo-Wiki Interface after connecting as a guest user. The GlobCover land cover map is shown overlaid on the Google Earth
image of an area (Pavia, Italy, in the example) and the user can report on disagreement among the recorded and the actual land cover class by click-
ing on the appropriate radio button on the left bar. (From http://www.geo-wiki.org/.)
it is also pointed out that the difference between expert and non- to gain expertise in crowdsourcing/citizen science techniques
expert contributors is narrower than one may expect. Accuracy combined with spatial data infrastructures (SDI), which is still a
figures are thus not so high, still, even the combination of dis- comparatively new field of development in spatial data science.
agreeing inputs can bring valuable information (Foody et al., The CITI-SENSE project (CITISENSE, 2012) is another EU
2013). The situation is probably set to improve as more and more FP7 ENV project, started in October 2012, aiming at the devel-
remotely sensed data becomes publicly available as it happened opment of a system for community-based environmental moni-
for the LANDSAT repository in 2008. toring and information in urban areas using Earth observation
and crowdsourcing.
The Citizens Observatory for Coast and Ocean Optical
26.5.4 Research (and operability) in Progress
Monitoring (CITCLOPS, 2012) project is also funded by the
As one can imagine, the last word is not said yet on VGI, let alone EU FP7 under the ENV theme. It uses volunteer inputs from
Web 2.0. While still trying to understand whether the upcoming mobile phones on individuals on the spot to help monitoring
Web 3.0 (Hendler, 2009) is finally the Semantic Web envisaged the color and transparency of natural water, that are impor-
by Berners-Lee etal. (2001), on the side of geospatial collabora- tant indicators of, for example, the ecosystem health status.
tive work, not only cooperation actions have been established An Android app is distributed from the project website, and
under the COST hat, but also projects are running to improve the job requested is really participatory, in that it is not just
the effectiveness and productivity of VGI systems. about taking a picture of the water surface from ones mobile
Examples of cooperation actions are Mapping and the phone and sending it to the repository; it entails interaction
Citizen Sensor (COST, 2012a) and ENERGIC (COST, 2012b). including, for example, selecting the best matching color from
One example of project is the Citizen Observatory Web a range of proposed standard water colors, feeding additional
(COBWEB, 2012) project funded under theme Environment environmental parameters, and paying off the contributor by
of the European Unions Seventh Framework Programme (FP 7). giving specific (though obviously automatic) feedback on the
Its main aim is to create a testbed environment that will enable acquired sample. Another FP7 ENV project, WeSenseIt (2012),
citizens living within biosphere reserves to collect environ- is also about Citizen Sensing of water, but for purposes of flood
mental data using mobile devices. The goal is twofold: on the risk assessment and related capacity building.
one side, to build a repository of environmental data useful for Apart from research projects, the state of the art includes an
driving policies and helping decision makers; on the other side, open source platform like Ushahidi (2007), which enables the
2016 by Taylor & Francis Group, LLC
Crowdsourcing and Remote Sensing 547
easy deployment of crowdsourced interactive mapping applica- weighing all the implied risks of data pollution, misunderstand-
tions with Web forms, e-mail, short message service (SMS), and ing, misinterpretationis still a big step up.
Twitter (2006) support. Ushahidi can be freely downloaded and This is not however the only power of crowdsourcing that
deployed on ones own server or used as an online service hosted comes from the active participation of intelligent humans in
by the platform providers itself, that is, Crowdmap. Ushahidi a task assigned to them. People are social beings and as such
can be accessed on smartphones and tablets through suitable are also oriented toward sharing information and helping one
mobile apps. Commercial exploitation of crowdsourcing of sat- another; the crowdsourcing capitalizes on this orientation,
ellite and geospatial data is also there, with companies offering allowing to gather more information (both in terms of quantity
services based on crowdprocessing of satellite multispectral and depth), from many more sensors and much more quickly.
images (Tomnod 2010).
The attention level of institutions to the crowdsourcing and
Acknowledgments
Citizen Sensor theme is generally high. Recently, a COST action
named Mapping and the Citizen Sensor TD1202 (2012) has The author especially thanks two persons for their useful inputs:
been approved by the European Science Foundation, and is Silvio DellAcqua and Pietro Demattei. Silvio DellAcqua is the
expected to increase knowledge and practicability of crowd- editor of Laputa (2012), a website devoted to the publication of
sourcing and VGI provision. The intergovernmental Group on original research on historical and geographical weird facts.
Earth Observations (GEO, 2005), coordinating efforts to build
a Global Earth Observation System of Systems, or GEOSS, in References
its 20122015 WorkPlan, makes several references to fostering
crowdsourcing and citizen science observation as additional Alfaro, L.D. etal. 2011. Reputation systems for open collabora-
sources of data and information. tion. Commun. ACM, 54(8), 8187.
Allahbakhsh, M., Benatallah, B., Ignjatovic, A., Motahari-
Nezhad, H.R., Bertino, E., Dustdar, S. MarchApril 2013.
26.6 conclusions Quality control in crowdsourcing systems: Issues and direc-
tions. IEEE Internet Comput., 17(2), 7681. doi: 10.1109/
It is a painful historical fact that, after the enthusiasm in the MIC.2013.20.
1980s and 1990s for the apparently endless reach of informa- Amazon Mechanical Turk. 2005. Amazon Mechanical Turk:
tion gathering through airborne or space-based remote sensing, Artificial Artificial Intelligence. Online. Available at:
a grim disappointment followed in the scientists and user com- https://www.mturk.com/.
munity at large after realizing that remotely sensed data fell short Anderson, J.R., Hardy, E.E., Roach, J.T., Witmer, R.E. 1976. A
of sustaining the wide range of applications that had been envis- Land Use and Land Cover Classification System for Use
aged at the beginning. The foundations of detecting any type of with Remote Sensor Data. Government Printing Office,
material on the Earth surface by observing it from above had Washington, DC, U.S. Geological Survey, Professional
been laid (Clark and Roush, 1984), but translating these prin- Paper 964.
ciples into something really operational turned out to be tougher Atkinson, G.M., Wald, D.J. May/June 2007. Did You Feel It?
than expected. A part of the problem was the inherent difficulty intensity data: A surprisingly good measure of earthquake
in translating noisy measures of faraway physical quantities into ground motion. Seismol. Res. Lett., 78, 362368. doi:
estimates that could be relied upon. Yet, another part was instead 10.1785/gssrl.78.3.362.
connected with the limitations in coverage (information on point Berners-Lee, T., Hendler, J., Lassila, O. 2001. The semantic web.
x at time y cannot be acquired, because no sensor is there at that Scient. Am., 284(5), 3543.
time) and visibility of observed features (information on feature CITCLOPS. 2012. The Citizens observatory for coast and ocean
z is not accessible, simply, because it is not visible from above). optical monitoring project web site. Online. Available at:
Both these latter issues can in principle be addressed through http://www.citclops.eu/.
integration of crowdsourcing into the production chain of geo- CITISENSE. 2012. The Citizens observatory community for
spatial information. Crowdsourcing, seen as an immersive, dis- improving quality of live in cities (CITI-SENSE) project
tributed network of countless insitu sensors, carries a potential to web site. Online. Available at: http://www.citi-sense.eu/.
Clark, R.N., Roush, T.L. July 10, 1984. Reflectance spectroscopy:
Substantially increase the a priori likelihood of being there
Quantitative analysis techniques for remote sensing appli-
when it matters
cations. J. Geophys. Res.: Solid Earth (19782012), 89(B7),
Gain accessibility to features that are invisible to the aver-
63296340.
age, faraway nadir or quasinadir sensor, for example,
COBWEB. 2012. The Citizen Observatory Web (COBWEB) proj-
small details, or reflectance of a vertical surface
ect web site. Online. Available at: http://cobwebproject.eu/.
In addition to the aforementioned points, the introduction of Congalton, R.G. July 1991. A review of assessing the accuracy
human beings in the loop scales up the observation paradigm of classifications of remotely sensed data. Remote Sens.
from purely quantitative to semanticeven after including and Environ., 37(1), 3546.
2016 by Taylor & Francis Group, LLC
548 Remotely Sensed Data Characterization, Classification, and Accuracies
COST. 2012a. European Cooperation in Science and Technology. Google Play. 2014. IDCT Direct Observation Survey app.
ICT action: Mapping and the Citizen Sensor. Action web site: Available on the Internet at: https://play.google.com/store/
http://www.cost.eu/domains_actions/ict/Actions/TD1202. apps/details?id=org.globalquakemodel.org.idctdo. Accessed
COST. 2012b. European Cooperation in Science and Technology. March 24, 2015.
Action: European Network Exploring Research into Hendler, J. January 2009. Web 3.0 emerging. Computer, 42(1),
Geospatial Information Crowdsourcing: Software and meth- 111113. doi: 10.1109/MC.2009.30.
odologies for harnessing geographic information from the Heylighen, F. 1997. Towards a global brain. In: Der Sinn der
crowd (ENERGIC). Action web site: http://www.cost.eu/ Sinne, U. Brandes, and C. Neumann, eds. Integrating
COST_Actions/ict/Actions/IC1203. Accessed March 24, Individuals into the World-Wide Electronic Network. Steidl
2015. Verlag, Gttingen, Germany, 1997.
De Leeuw, J., Said, M., Ortegah, L., Nagda, S., Georgiadou, Y., Howe, J. 2006. The rise of crowdsourcing. Wired Magazine
DeBlois, M. 2011. An assessment of the accuracy of vol- Issue 14.06, June 2006. Available at: http://archive.wired.
unteered road map production in Western Kenya. Remote com/wired/archive/14.06/crowds.html. Accessed March 24,
Sens., 3, 247256. 2015
Dengler, L.A., Dewey, J.W. 1998. An intensity survey of house- Iannelli, G.C. 2014. Processing and fusion of multiresolution
holds affected by the Northridge, California, earthquake of spaceborne Earth Observation data for assessing expo-
January 17, 1994. Bull. Seismol. Soc. Am., 88, 441462. sure and vulnerability to natural disasters risk. PhD
Dengler, L.A., Moley, K. 1994. Toward a quantitative, rapid Thesis of Gianni Cristian Iannelli. To be published on the
response estimation of intensities. Seismol. Res. Lett., 65, 48. Faculty repository at http://www-3.unipv.it/dottIEIE/
DesignCrowd. 2010. 5 Famous Logo ContestsToyota, Google, index.php?pag=italiano/tesi_dottorato.html. Accessed
Wikipedia & More!. Online. Available at: http://blog. March 24, 2015.
designcrowd.com/article/218/5-famous-logo-contests IDC. 2013. International Data Corporation (IDC) worldwide
toyota-google-wikipediamore. quarterly mobile phone tracker. Online. Available at: http://
Dow, S.P., Kulkarni, A., Klemmer, S.R., Hartmann, B. 2012. www.idc.com/tracker/showproductinfo.jsp?prod_id=37.
Shepherding the crowd yields better work. Proceedings of the Jsang, A., Ismail, R., Boyd, C. 2007. A survey of trust and repu-
2012 ACM Conference on Computer Supported Cooperative tation systems for online service provision. Decision Supp.
Work (CSCW12), ACM, 1115 Feb 2012, Seattle, WA. Syst., 43(2), 618644.
pp.10131022. Kaplan, A.M., Haenlein, M. 2010. Users of the world, unite! The
DYFI. 1991. Did You Feel It earthquake hazards program. Online. challenges and opportunities of social media. Business
Available at: http://earthquake.usgs.gov/earthquakes/dyfi/ Horiz., 53, 5968.
Accessed March 24, 2015. Kerle, N. 2011. Remote sensing based post-disaster damage map-
Foody, G.M., See, L., Fritz, S., Van der Velde, M., Perger, C., Schill, pingReady for a collaborative approach? IEEE Earthzine.
C., Boyd, D.S. 2013. Assessing the accuracy of volunteered Posted on March 23, 2011 in Articles, Disaster Management
geographic information arising from multiple contributors Theme, Earth Observation.
to an internet based collaborative project. Trans. GIS, 17(6), Levy, P. 1999. Collective Intelligence: Mankinds Emerging World in
847860. Cyberspace. Basic Books, New York. ISBN: 9780738202617.
Fritz, S., McCallum, I., Schill, C., Perger, C., Grillmayer, R., Laputa, 2012. Nekutima Geografio. Owner and Editor: Silvio
Achard, F., Kraxner, F., Obersteiner, M. 2009. Geo-Wiki. DellAcqua. Online at: http://www.laputa.it/. English ver-
Org: The use of crowdsourcing to improve global land sion coming soon. Accessed March 24, 2015.
cover. Remote Sens., 1(3), 345354. doi: 10.3390/rs1030345. Neis, P., Zielstra, D., Zipf, A. 2012. The street network evolution
Open access. of crowdsourced maps: OpenStreetMap in Germany 2007
GEM. 2006. The global earthquake model. Available at: http:// 2011. Future Internet, 4, 121.
www.globalquakemodel.org/. OECD. 2007. Participative Web and User-Created Content:
GEO. 2005. The group on earth observation web site. Online. Web 2.0, Wikis, and Social Networking. Organisation for
Available at: http://www.earthobservations.org/. Economic Co-operation and Development, Paris, France.
GlobCover. 2005. The European Space Agency GlobCover portal. Openstreetmap. 2004. Collaborative mapping web site. Online.
Online. Available at: http://due.esrin.esa.int/globcover/. Available at: http://www.openstreetmap.org/.
Goetz, M., Zipf, A. 2012. Towards defining a framework for the OReilly, T. 2007. What is Web 2.0: Design patterns and business
automatic derivation of 3D CityGML models from volun- models for the next generation of software. J. Commun.
teered geographic information. Int. J. 3-D Inform. Model., 1, Strat., 65, 17.
496507. Over, M., Schilling, A., Neubauer, S., Zipf, A. 2010. Generating
Goodchild, M.F. 2007. Citizens as sensors: The world of volun- web-based 3D city models from OpenStreetMap: The cur-
teered geography. GeoJournal, 69, 211221. doi: 10.1007/ rent situation in Germany. Comput. Environ. Urban Syst.,
s10708-007-9111-y. 34, 496507.
2016 by Taylor & Francis Group, LLC
Crowdsourcing and Remote Sensing 549
Raifer, M. 2014. OSM node density 2014. Map available at: http:// Sobel, D. 1998. Longitude: The True Story of a Lone Genius Who
tyrasd.github.io/osm-node-density/; explanation available at: Solved the Greatest Scientific Problem of His Time. Fourth
http://www.openstreetmap.org/user/tyr_asd/diary/22363. Estate Ltd., London, U.K., p. 6. ISBN: 1-85702-571-7.
Accessed on August 14, 2014; Re-accessed March 24, 2015. TD1202. 2012. Information and Communication Technologies
Scekic, O., Truong, H., Dustdar, S. 2012. Modeling rewards and (ICT) Action TD1202 Mapping and the citizen sensor.
incentive mechanisms for social BPM. In: A. Barros etal., Available at: http://www.cost.eu/domains_actions/ict/
eds. Short paper, 10th International Conference on Business Actions/TD1202.
Process Management (BPM2012), September 36, 2012, Tomnod. 2010. Originally a start-up, now owned by DigitalGlobe.
Springer, Tallinn, Estonia, pp. 150155. Online at: http://www.tomnod.com. Accessed March 24,
See, L., Comber, A., Salk, C., Fritz, S., van der Velde, M., Perger, 2015.
C., Schill, C., McCallum, I., Kraxner, F., Obersteiner, M. Twitter 2006. The twitter web site. Online. Available at: http://
2013. Comparing the quality of crowdsourced data contrib- twitter.com/.
uted by expert and non-experts. PLoS ONE, 8(7), e69958. Ushahidi. 2007. The Ushaihidi Project web site. Online. Available
doi: 10.1371/journal.pone.0069958. at: http://ushahidi.com/.
ShakeMap. 2004. ShakeMap Archive, M 6.0 event on September Wald, D.J., Quitoriano, V., Worden, B., Hopper, M., Dewey, J.W.
28th, 2004, 17:15:24 UTC. Addressed maps available at: 2011. USGS Did You Feel It? Internet-based macroseis-
http://earthquake.usgs.gov/eqcenter/shakemap/nc/shake/ mic intensity maps. Ann. Geophys., 54(6), 688707. doi:
51147892/; service description available at: http://earthquake. 10.4401/ag-5354.
usgs.gov/earthquakes/shakemap/. WeSenseIt. 2012. WeSenseIT: Citizen Observatory of Water
Sheth, A. JulyAugust 2009. Citizen sensing, social signals, and project web site. Online. Available at: http://www.wesen-
enriching human experience. IEEE Internet Comput., 13(4), seit.eu/.
8792, doi: 10.1109/MIC.2009.77 Wikimapia. 2007. Collaborative mapping web siteNot a part
Silva, V., Crowley, H., Pagani, M., Monelli, D., Pinho, R. 2013. of non-profit Wikimedia foundation. Online. Available at:
Development of the OpenQuake engine, the Global Earthquake http://wikimapia.org/.
Models open-source software for seismic risk assessment. Nat. Wikipedia. 2001. The free on-line encyclopedia. Online. Available
Hazards, 72(3), 14091427. doi: 10.1007/s11069-013-0618-x. at: http://www.wikipedia.org/.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
X
Cloud Computing
and Remote Sensing
27 Processing r emote-Sensing Data in Cloud Computing e nvironments Ramanathan Sugumaran, James W.
Hegeman, Vivek B. Sardeshmukh, and Marc P. Armstrong ................................................................................................... 553
Acronyms and Definitions Introduction Introduction to Cloud Computing Cloud-Computing-Based Remote-
Sensing-Related Applications Conclusions Acknowledgments References
551
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
27
Processing Remote-Sensing
Data in Cloud Computing
Environments
Acronyms and Definitions .................................................................................................................553
Ramanathan Sugumaran
The University of Iowa
27.1 Introduction .............................................................................................................................553
and John Deere
Remote Sensing and Big Data Big-Data Processing Challenges
27.2 Introduction to Cloud Computing .......................................................................................555
James W. Hegeman Definitions Cloud Paradigms Cloud Service Models Advantages and Limitations of
The University of Iowa Cloud Computing
27.3 Cloud-Computing-Based Remote-Sensing-Related Applications ...................................557
Vivek B. Sardeshmukh A Case Study: Cloud-Based LiDAR Processing System
The University of Iowa
27.4 Conclusions.............................................................................................................................. 560
Marc P. Armstrong Acknowledgments .............................................................................................................................. 560
The University of Iowa References............................................................................................................................................ 560
Acronyms and Definitions 27.1 introduction
AWS Amazon Web Services 27.1.1 Remote Sensing and Big Data
CLiPS Cloud-based LiDAR processing system
During the past four decades, scientific communities around
CPU Central processing unit
the world have regularly accumulated massive collections of
EC2 Elastic Compute Cloud
remotely sensed data from ground, aerial, and satellite platforms.
EOSDIS Earth Observing System Data and Information System
In the United States, these collections include the U.S. Geological
GIS Geographic information system
Surveys (USGS) 37-year record of Landsat satellite images
GPGPU General-purpose computing on graphics processing
(comprising petabytes of data) (USGS, 2011); the NASA Earth
units
Observing System Data and Information System, having multi-
GPU Graphics processing unit
ple data centers and more than 7.5 petabytes of archived imagery
HPC High-performance computing
(Hyspeed Computing, 2013); and the current NASA systems that
HTC High-throughput computing
record approximately 5 TB of remote-sensing-related data per
IaaS Infrastructure as a service
day (Vatsavai et al., 2012). In addition, new data-capture tech-
MODIS Moderate-Resolution Imaging Spectroradiometer
nologies such as LiDAR are used routinely to produce multiple
OCC Open Cloud Consortium
petabytes of 3D remotely sensed data representing topographic
OGC Open Geospatial Consortium
information (Sugumaran et al., 2011). These technologies have
PaaS Platform as a service
galvanized changes in the way remotely sensed data are collected,
Saas Software as a service
managed, and analyzed. On the sensor side, great progress has
TIN Triangulated irregular network
been made in optical, microwave, and hyperspectral remote
UAV Unpiloted aerial vehicles
sensing with (1) spatial resolutions extending from kilometers
USGS U.S. Geological Survey
to submeters, (2) temporal resolutions ranging from weeks to
VPN Virtual private network
30 min, (3) spectral resolutions ranging from single bands to
WMS Web Map Service
553
2016 by Taylor & Francis Group, LLC
554 Remotely Sensed Data Characterization, Classification, and Accuracies
Sources/types Big data Challenges
Collect
Ground, aerial, Volume
satellite, and UAV Manage
Store
Optical, microwave,
hyperspectral, LiDAR RS as Archive
Veracity Velocity
Big data
Analyse
Large achieve, real- Visualize
time, Variety Distribute
FIg u r e 2 7.1 Remote sensing: big-data sources and challenges.
hundreds of bands, and (4) radiometric resolutions ranging architectural advances, the analysis of big data presents new
from 8 to 16bits. The platform side has also seen rapid develop- challenges to both cluster-infrastructure software and parallel-
ment during the past three decades. Satellite and aerial platforms application design, and it requires the development of new com-
have continued to mature and are producing large quantities of putational methods. These methods and several articles about
remote-sensing data. Moreover, sensors deployed on unpiloted the importance of HPC in remote sensing are featured in spe-
aerial vehicles (UAVs) have recently begun to produce massive cial journal issues, books, and conferences devoted to this topic
quantities of very-high-resolution data. (Plaza and Chang, 2007, 2008; Lee etal., 2011; Prasad, 2013).
The technological nexus of continuously increasing spatial, Graphics processing units (GPUs) have been widely used (in
temporal, spectral, and radiometric resolutions of inexpensive GPGPU applications) to address remote-sensing problems (Chang
sensors, on a range of platforms, along with internet data acces- etal., 2011; Christophe etal., 2011; Song etal., 2011; Yang etal.,
sibility is creating a flood of remote-sensing data that can easily 2011). Oryspayev et al. (2012) developed an approach to LiDAR
be included in what is commonly referred to as big data. This processing that used data-mining algorithms coupled with par-
term refers to datasets that have grown sufficiently large that allel computing technology. A specific comparison was made
they have become difficult to store, manage, share, and analyze between the use of multiple central processing units (CPUs) (Intel
using conventional software tools (White, 2012). Big data are Xeon Nehalem chipsets) and GPUs (Intel i7 Core CPUs using the
often thought to span four dimensions: volume (data quantity), NVIDIA Tesla s1070 GPU cards). The experimental results dem-
velocity (real-time processing), variety (source multiplicity), and onstrated that the GPU option was up to 35 times faster than the
veracity (data accuracy) (IBM, 2012). Operating hand in glove CPU option. In a similar vein, distributed parallel approaches have
with Moores law, the growth of big data is largely a consequence also been developed. Haifang (2003) implemented various algo-
of advances in acquisition technology and increases in storage rithms using a heterogeneous grid computing environment, and
capacity. Figure 27.1 summarizes the overall sources and chal- Liu (2010) analyzed the efficiency improved by grid computing for
lenges presented by big remote-sensing data. the maximum likelihood classification method. Yue etal. (2010)
used cluster computing to solve remote-sensing-image fusion,
filtering, and segmentation. Commodity-cluster-based parallel
27.1.2 Big-Data Processing challenges
processing of various multispectral and hyperspectral imagery
As the pace of imaging technology has continued to advance, has also been used by various authors (e.g., Plaza etal., 2006).
the provision of affordable technology for dealing with issues While HPC environments such as clusters, grid computing,
such as storing, processing, managing, archiving, disseminating, and supercomputers (Simmhan and Ramakrishnan, 2010) can be
and analyzing large volumes of remote-sensing information has used, these platforms require significant investments in equip-
lagged. One major challenge is related to the computational power ment and maintenance (Ostermann etal., 2010) and individual
required to process these massive data sources. Traditionally, researchers and many government agencies do not have routine
desktop computers with single or multiple cores have been used access to these resources. In addition to data volume, the variety
to process remote-sensing data for small areas. In contrast, large- and update rate of datasets often exceed the capacity of com-
or macroscale remote-sensing applications may require high- monly used computing and database technologies (Wang etal.,
performance computing (HPC) technologies; general-purpose 2009; Yang etal., 2011; Shekhar etal., 2012). As a result of these
computing on graphics processing units (GPGPU); and parallel, limitations, users have begun to search for less-expensive solu-
cluster, and distributed-computing approaches are gaining broad tions for the development of large-scale data-intensive remote-
acceptance (Plaza et al., 2006; Gonzlez et al., 2009; Simmhan sensing applications. Cloud computing provides a potential
and Ramakrishnan, 2010; Shekhar et al., 2012). Given these solution to this challenge due to its scalability advantages in data
2016 by Taylor & Francis Group, LLC
Processing Remote-Sensing Data in Cloud Computing Environments 555
storage and processing and its relatively low cost as compared to by the cluster are usually assumed to be tightly connected by
user-owned, high-power compute clusters (Kumar etal., 2013). a low-latency, private network. This generally implies physi-
The goal of this chapter is to provide a short review of remote- cal locality of the system itself. The concept of cluster comput-
sensing applications using cloud-computing environments, as ing was a precursor to todays notion of HPC (Lee etal., 2011).
well as a case study that provides greater implementation detail. Cluster computing differs from cloud computing in its empha-
An introduction to cloud computing is provided in Section sisa compute cluster is often a localized system dedicated to
27.2, and then in Section 27.3, various applications, including a one particular problem (or class of problems) at a time.
detailed case study, are described to illustrate the advantages of Grid computing can also be thought of as a subset of cloud
cloud-computing environments. computing having a slightly different emphasis (Buyya et al.,
2009). A compute grid is a distributed system, often encompass-
ing machines physically separated by a large distance, together
27.2 introduction to cloud computing with a job scheduler that allows a user to easily and simultane-
ously run a certain small set of programs on many different data
27.2.1 Definitions
inputsthe action of a users program on one such input consti-
Cloud computing is a vague term, as nebulous as its eponym. The tutes a task, or job. Grid computing is thus the use of a parallel/
NIST Definition of Cloud Computing (Mell and Grance, 2011) distributed system for solving a large problem or accomplishing
defines cloud computing as a model for enabling ubiquitous, con- a large collection of tasks that would take too long to complete
venient, on-demand network access to a shared pool of configu- on a single machine. In grid computing, the problem to be solved
rable computing resources. In research, cloud computing is a can generally be broken down into many nearly identical tasks
popular idiom for distributed computing, encompassing the same that can run concurrently and independently on the nodes of the
fundamental conceptsmultiplicity, parallelism, and fault toler- system (see, e.g., Wang and Armstrong, 2003). Once complete,
ance. Distributed computing has come to the forefront as an area the solutions to, or results of, these tasks are aggregated.
of research over the past two decades as dataset sizes have out- The type of computational problem-solving approach exem-
stripped traditional sequential processing power, even of modern plified by grid computing is also known as high-throughput
high-performance processors, and as the bottlenecks of large-scale computing (HTC), in which similar computations are done inde-
computation have moved outside the CPU (e.g., to storage I/O). pendently by a (large) number of processors, and the network
To employ distributed or cloud computing means to leverage the interconnect is used primarily for data distribution and results
additional hardware and computing throughput available from aggregation. Another, distinct, distributed computational para-
large networks of machines. Because of the challenges inherent in digm is HPC, which emphasizes slightly different aspects of the
optimizing resource scale and utilization for a problem, two cen- system. Whereas HTC emphasizes the problem division, HPC
tral focuses of cloud computing in practice have been (1) the elas- refers to the use of many high-power servers connected by a fast
ticity of resource provisioning and (2) abstraction layers capable (usually >10 Gbps) network, under any algorithmic paradigm.
of simplifying these challenges for users. Indeed, as exemplified in An HPC algorithm may also qualify as HTC, or it may require
an eScience Institute document (2012), it is these two qualities of much more intermediate communication between processes in
cloud computingelasticity and abstractionthat are often most order to accomplish the goal. In other words, in HPC, individual
important from the end users perspective. Thus, while research processes/processors may need to communicate very rarely, or
with cloud computing generally focuses on the distributed scal- very often, during the intermediate stages of a computation. The
ability of the cloud, the characteristic feature of cloud computing, concepts of HTC and HPC thus focus on slightly different quali-
in practice, is flexibility. There are several related terms associated ties of a whole system and are neither identical nor mutually
with cloud computing as explained below. exclusive. In Figure 27.2, an example HTC workflow is depicted.
Cluster computing is a similar, but more limited, model of par- The problem input is divided into many segments, each of
allel/distributed computing, in which the machines comprised which is sent to a node of the system; depending on processing
T1,1 T1,2 T1,3
T2,1 T2,2 T2,3
T3,1 T3,2 T3,3
Input Results
T49,1 T49,2 T49,3
T50,1 T50,2 T50,3
FIg u r e 2 7.2 A typical high-throughput computing workflow model.
2016 by Taylor & Francis Group, LLC
556 Remotely Sensed Data Characterization, Classification, and Accuracies
capacities, a single node may receive multiple segments. Several A private cloud is just thata private network of computers
tasks (in Figure 27.2, three) may be required to complete the pro- owned by the using entity, usually segregated from the public
cessing of each segment. Thus, in Figure 27.2, any two tasks Ti, j internet. Examples of private clouds include any compute clus-
with the same first (i) index are distinct but operate (in series) ters to which access is limited, such as the research clusters com-
on the same segment of the input, whereas any two tasks with monly operated by and within large universities. Private cloud
the same second (j) index are identical but operate on different computing makes sense under many scenarios: (1) For research
portions of the input. After processing of each segment, the indi- and development, it may be necessary to have more control
vidual results are merged in some manner to form the output. over the system than is afforded by the typical public cloud.
Unlike HTC, there is no canonical diagram for HPC, since HPC Oftentimes, complete control of all aspects of a system can only
is defined more by the power of the computational resources and be achieved when the system is private. (2) When it is financially
admits many algorithmic paradigms. feasible to own a system whose response time would be sufficient
Presently, there are several large-scale commercial options for the users, and if the users have a sufficient quantity of compu-
for cloud computing, and many larger institutions have their tational tasks that the system would be highly utilized, it could
own distributed systems that users may use as a cloud. make sense to use a private cloudthus cutting out the middle
Amazons Elastic Compute Cloud (EC2) was the first public man from the services.
cloud to be available to large numbers of users over the Internet. Finally, the term hybrid cloud is sometimes used to refer to an
Subsequently, Google and Microsoft, as well as several other aggregate system of which some portion is owned, or fully con-
vendors, have also begun to offer large-scale public cloud ser- trolled, by the user and some portion is available over a public
vices. Since the distributed nature of a cloud is opposite of the network. A hybrid cloud can make sense when different parts
traditional mainframe server, most cloud-computing systems in of the cloud are being used for distinctly different tasks and a
practice run some version of the GNU/Linux operating system. different cloud-computing model would be ideal for each. For
However, Microsoft Windows Server is also an option for cloud example, an institution with critical data it needs to process may
systems, and indeed the availability of geographic information opt to store the data permanently on a smaller, private cloud,
system (GIS)-specific software packages makes Windows Server where members have full control over data security, but send
relevant in this domain. some portion of the data to a public cloud, on demand, for pro-
cessing. While a private cloud offers more control to its owners/
users, a public cloud can be easier to use and can usually provide
27.2.2 cloud Paradigms
greater computational power, with little or no overhead cost.
While a variety of common terms have arisenpublic cloud, As such, a hybrid cloud model may make sense in practice for
private cloud, and hybrid cloudthese models of computing are many businesses and smaller, short-to-medium-term research
fundamentally the same, differing only in ancillary issues such endeavors.
as security and usability.
A public cloud is a cloud system available to the public (or
27.2.3 cloud Service Models
some subset of the public) over the World Wide Web. A public
cloud may even rely on Internet infrastructure for internal Infrastructure as a service: The fundamental concept of out-
network connectivity. In practice, compute access to most pub- sourcing and commercialization of compute time is referred to
lic clouds is available for rent to the general public. Amazons as infrastructure as a service (IaaS). In this approach, a cloud ser-
EC2 is the quintessential public cloudAmazon Web Services vice provider (for instance, Amazon) owns and operates a col-
(AWS) was an early leader in providing computing power as a lection of networked servers available for rent. At a base level,
commercial service. Others in the commercial sector have fol- the cloud service provider provisions rental machines (often vir-
lowed suit, such as Google with their Google Cloud Platform. tual machines) with a clients desired operating system, as well
The emergence of public clouds is a good example of an economy as related facilities and tools. For instance, a client might rent
of scalepowerful servers and high-performance networks can compute time from a cloud service provider and request that
be expensive and require specialized expertise to administer the rental machines run Fedora Linux. The physical machines,
and support. Many institutions possess problems for which together with the network, Linux kernel, and Fedora distribu-
cloud computing is [a part of] an ideal solution, but only the tion constitute the infrastructure, and this system in its entirety
largest have a sufficient quantity of such problems that it makes is the product provided to the client.
financial sense to own and manage their own servers. Because Platform as a service: The concept of a platform as a service
of the expense, cloud-computing systems of scale are often only (PaaS) lies on top of IaaS. For many clients, a compute infra-
financially practical under high loadsmaller organizations structure alone is not sufficient for their goalsthere is a large
without a sufficient volume of computational problems may gulf between the presence of computing infrastructure and
find themselves only utilizing a private system, say, 50% of the the desired end result. For end users who may have a higher-
time. In this scenario, the costly overhead looms even larger level abstraction of their computational task(s), an intermedi-
compared to the costbenefit analysis of problems solved versus ate platforma computational service that provides more than
power consumption. just an operating system and network and is managed at that
2016 by Taylor & Francis Group, LLC
Processing Remote-Sensing Data in Cloud Computing Environments 557
higher levelmay be appropriate. In PaaS, a software platform is design must take the distributed/parallel nature of cloud com-
provisioned by the cloud service provider and can be composed puting into accountthe power of cloud computing lies not
of one or more different software layers. One common example in any single machine, but in the capacities of the system as a
is Hadoop, by Apache. Hadoop is an open-source implemen- whole. Problems of an HTC nature are natural candidates for
tation of Googles MapReduce framework that provides the solving in the cloud because of the intrinsically scalable aspects
MapReduce computational model along with an underlying of HTC solutions. More generally, for all problem paradigms,
distributed file system. In addition, there are a variety of addi- the communication requirements of the problem play a role in
tional layers compatible with Hadoop that can exist on top of selecting the appropriate service model when seeking a cloud-
the HadoopMapReduce platforma common such layer is the computing solutionif substantial communication is required
data warehouse Hive, also developed by Apache. For more infor- between compute nodes in a particular algorithm, a lower-level
mation on the MapReduce computing framework, see Dean and service model (IaaS) may be necessary. For example, it may be
Ghemawat (2004). difficult to find an appropriate, off-the-shelf SaaS-level software
Software as a service: At a higher level than PaaS is software implementation for an algorithmic approach that requires com-
as a service (SaaS), in which the cloud service provider furnishes pute nodes to communicate heavily. In contrast, in HTC, the
a complete software package designed for a particular domain. focus of the software developer can be narrowed to the proper
The end user need only perform configuration-level tasks on the provisioning of resources; network bandwidth is often a bottle-
system before it is ready for use. This model of computational neck only during the dissemination of problem input and gath-
services is well suited for organizations that must perform ubiq- ering of output. Thus, an analysis akin to Amdahls law plays
uitous tasks, such as vehicle location tracking or legal docu- an important role in determining the practicality of a particu-
ment preparation, and that have no desire for, or commitment lar scale of cloud computing for running a specific algorithm or
to, computational research and development. SaaS is becoming solving a particular problem.
increasingly popular as an operational business model. Finally, on issues of cloud security and liability: it should be
emphasized that, in part because cloud computing is not an
established commodity, many contractual aspects (e.g., various
27.2.4 Advantages and Limitations
liabilities) are left to the user and provider to agree upon.
of cloud computing
On a practical level, the cloud-computing paradigm has many
advantages: access to HPC systems; pay-as-you-go payment 27.3 cloud-computing-Based Remote-
schemes, with few overhead costs; on-demand provision- Sensing-Related Applications
ing of resources; highly scalable/elastic compute and storage
resources; and automated data reliability (McEvoy and Schulze, This section provides a short summary of remote-sensing appli-
2008; Watson et al., 2008; Cui et al., 2010; Huang et al., 2010; cations that use cloud-computing environments as well as a
Rezgui etal., 2013; Yang and Huang, 2013). Note that this last more detailed case study. HPC frameworks (e.g., supercomput-
improvement, high data reliability, is different from data secu- ers at various research organizations) can potentially be widely
rityone of the major challenges in cloud computing. By its very adapted for remote-sensing applications (Parulekar etal., 1994;
nature of high accessibility/availability, the most scalable type Simmahan and Ramakrishnan, 2010; Lee et al., 2011; Plaza,
of cloud computingthe use of a public cloudis inherently 2011a, b). The main limitations associated with the use of HPC
less secure than a computing model based on private control are as follows: (1) HPC resources are not readily available for
of the entire system. The security challenges of using a public large user communities and (2) HPC resources are expensive to
cloud are manyfold: depending on privacy requirements, it may acquire and maintain (Ostermann etal., 2010).
not be acceptable that data reside on the cloud service providers Krishnan et al. (2010) evaluated a MapReduce approach for
machines; users are reliant on the internal security controls of LiDAR gridding on a small private cluster consisting of around
the service provider to prevent unauthorized data access, and 810 commodity computers. They investigated the effects of sev-
the public network infrastructure that must be traversed for data eral parameters, including grid resolution and dataset size, on
and compute access may be compromised. More elaborate (and performance. For their software implementation using Hadoop,
from a performance perspective, costly) security and encryption the authors experimented with Hadoop-specific factors, such as
measures must commonly be taken when using a public cloud, the number of reducers allocated for a problem and the inherent
such as the use of a virtual private network. concurrency therein. In their particular study, using quad-core
On a fundamental level, the advantages of cloud computing machines with 8 GB of main memory and connected by gigabit
lie in the distributed paradigm of the hardware systems and the Ethernet, they found that doubling the size of their Hadoop clus-
prospect of lowered barriers to access through the commoditiza- ter from four to eight nodes had little effect on their experimen-
tion of computing power itself. The cloud-computing paradigm tal runtimes. They thus showed that their solution to the task
provides flexibility, scalability, robustness, and, when managed was not strictly compute bound. On the other hand, the authors
by a dedicated provider, ease of use. In order to make best use did note a substantial degradation in performance for their
of these new computational opportunities, however, software single-node, nondistributed algorithm control (implemented in
2016 by Taylor & Francis Group, LLC
558 Remotely Sensed Data Characterization, Classification, and Accuracies
C++) when the problem size grew larger than could fit in main spatial data. As case studies, SpatialHadoop has three spatial
memory. Krishnan et al. (2010) concluded that Hadoop could operations, range queries, k-nearest-neighbor queries, and
be a useful framework for algorithms that process large-scale spatial join.
spatial data (roughly 150 million points), but that certain serial Cary etal. (2009) studied the performance of the MapReduce
elements, such as output generation, could still be rate limiting, framework for bulk construction of R-trees and aerial image
especially on commodity hardware. Their work also (1) moti- quality computation on both vector and raster data. They
vates the study of systems with larger memories (this stems from deployed their MapReduce implementations using the Hadoop
their experience with HPC resources) and (2) demonstrated that framework on the Google and IBM clouds. The authors pre-
the task of designing optimal systems for the processing of mas- sented results that demonstrate the scalability of MapReduce
sive spatial data is complex. and the effect of parallelism on the quality of the results. This
Project Matsu is an open-source project for processing paper also studied various metrics to compare the performance
satellite imaginary using a community cloud (Bennett and of their implemented algorithms including execution time, cor-
Grossman, 2012). This project, a collaboration between NASA rectness, and tile quality. Their results indicate that the appro-
and the Open Cloud Consortium (OCC), has been developed to priate application of MapReduce could dramatically improve
process data from NASAs EO-1 satellite and to develop open- task completion times and also provide close to linear scalabil-
source technology for public cloud-based processing of satellite ity. This study motivates further investigation of the MapReduce
imagery. Most computations were completed using a Hadoop framework for other spatial-data-handling problems.
framework running on 9 compute nodes with 54 compute cores Li etal. (2010) studied the integration of data from ground-based
and 352GB of RAM. The stated goal of this project (Project sensors with the Moderate-Resolution Imaging Spectroradiometer
Matsu) is (1) to use an open-source cloud-based infrastructure satellite data using the Windows Azure cloud platform. Specifically,
to make high-quality satellite image data accessible through the authors provide a novel approach to reproject input data
an Open Geospatial Consortium-compliant (OGC-compliant) into timeframe- and resolution-aligned geographically format-
Web Map Service (WMS), (2) to develop an open-source cloud- ted data and also develop a novel reduction technique to derive
based analytic framework for analyzing individual images and important new environmental data through the integration of
collections of images, and (3) to generalize this framework to satellite and ground-based data. Slightly modified Windows
manage and analyze other types of spatialtemporal data. This Azure abstractions and APIs were used to accomplish the repro-
project also features an on-demand cloud-based disaster assess- jection and reduction steps. They suggest that cloud computing
ment capability through satellite image comparisons. The image has a great potential for efficiently processing satellite data. It
comparisons are done via a MapReduce job using a Hadoop- should be noted that their current framework doesnt fit into the
streaming interface. As an example, this project hosts a website MapReduce framework since it uses Azures general queue-based
that provides real-time information about flood prediction and task model.
assessment in Namibia. The final data are served to end users Berriman et al. (2010) compared various toolkits to create
using a standard OGC WMS and Web Coverage Processing image mosaics and to manage their provenance using both the
Service tools. Amazon EC2 cloud and the Abe high-performance cluster at
Oryspayev etal. (2012) studied LiDAR data reduction algo- NCSA, UIUC. They conducted a series of experiments to study
rithms that were implemented using the GPGPU and multi- performance and costs associated with different types of tasks
core CPU architectures available on the AWS EC2. This paper (I/O bound, CPU bound, and memory bound) in these two envi-
tests the veracity of a vertex-decimation algorithm for reducing ronments. Their experiments show that for I/O-bound applica-
LiDAR data size/density and analyzes the performance of this tions, the most expensive resources are not necessarily the most
approach on multicore CPU and GPU technologies, to better cost-effective, that data transfer costs can exceed the processing
understand processing time and efficiency. The paper docu- costs for I/O-bound applications on Amazon EC2, and that the
ments the performance of various GPGPU and multicore CPU resources offered by Amazon EC2 are generally less powerful
machines including Tesla family GPUs and the Intels multicore than those available in the Abe high-performance cluster and
i-CPU series for the data reduction problem using large-scale consequently do not offer the same levels of performance. They
LiDAR data. The study raises several questions about imple- concluded from their results that cloud computing offers a pow-
mentation of spatial-data processing algorithms on GPGPU erful and cost-effective new resource for compute and memory
machines, such as how to reduce overhead during the initial- intensive remote-sensing applications.
ization of devices and how to optimize algorithms to minimize
data transfer between CPUs/GPUs.
27.3.1 A case Study: cloud-Based
Eldawy and Mokbel (2013) developed an open-source
LiDAR Processing System
framework, SpatialHadoop, that extends Hadoop by provid-
ing native support for spatial data. As an extension of Hadoop, To more completely illustrate an approach to cloud comput-
the framework operates similarlyprograms are written in ing of remote-sensing information, in this section we describe
terms of map and reduce functions, though the system is opti- a LiDAR application. In this cloud-based LiDAR processing
mized to exploit underlying properties and characteristics of system (CLiPS) project, we use a statewide (Iowa) LiDAR data
2016 by Taylor & Francis Group, LLC
Processing Remote-Sensing Data in Cloud Computing Environments 559
Client Server Amazon web services
EC2 instances
LiDAR and tiles
vector data
Web Web-GIS
browser Large, X, 2X
Open layers
GIS server
Buckets
Web
server Simple storage service
FIg u r e 2 7.3 Overall architecture developed for cloud-based LiDAR processing system.
repository (Iowa DNR, 2009) in which data are distributed to the management services, such as monitoring and analysis, and the
public as a collection of 34,000 tiles, each covering 4km2; this third tier is dedicated to data and file services. The client-side
comprises roughly 7 TB of data. In order to process this massive interface was developed using Flex and the server-side uses cus-
data, CLiPS was designed (Figure 27.3) as a web portal imple- tom-built tools, constructed from open-source products.
mented using Adobes Flex framework along with ESRIs ArcGIS Figure 27.4 shows the user interface developed for this study.
API for Flex (ESRI, 2012; Sugumaran etal., 2014), OpenLayers, The interactive user interface requires a user to, first, select a
and the Amazon EC2 cloud environment. CLiPS use a three- region of interest on a map and then select AWS credentials and
tier clientserver model (Figure 27.3). The top tier supports user computing resources as well as a location to which results will
interaction with the system, the second tier provides process be sent for downloading (this is typically an e-mail address). The
FIg u r e 2 7.4 Cloud-based LiDAR processing system user interface for the state of Iowa, United States.
2016 by Taylor & Francis Group, LLC
560 Remotely Sensed Data Characterization, Classification, and Accuracies
system was tested by creating a triangulated irregular network Acknowledgments
(TIN) from a LIDAR point cloud using 18 dataset and proces-
sor configurations: three terrain data (flat, undulating, urban), This research was conducted with support from an Amazon
two tile sizes (9 and 25), and three Amazon EC2 processing con- Research Grant and USGSAmericaView projects.
figurations (large, Xlarge, and double Xlarge). The undulating
terrain dataset took more time than the other terrain types for References
5 5 tile groups, while the urban terrain was the most compu-
tationally intensive for the 3 3 tile groups used in this study Bennett, C. and Grossman, R. 2012. OCC Project Matsu: An
(Sugumaran etal., 2014). The results clearly show that as com- open-source project for cloud-based processing of satellite
puter power increases, processing times decrease for all three imagery to support the earth sciences. http://matsu.open-
types of LiDAR terrain data. The various combinations in our sciencedatacloud.org (accessed on March 4, 2014).
evaluations showed that even with up to 25 tiles with varying Berriman, G. B., Deelman, E., Groth, P., and Juve, G. 2010. The appli-
processing configuration types, each request required less than cation of cloud computing to the creation of image mosaics
an hour and cost less than a dollar for data processing (e.g., TIN and management of their provenance. In SPIE Astronomical
creation). Moreover, the cost of uploading data from our server Telescopes+ Instrumentation, International Society for Optics
and data storage on the cloud was less than 50 dollars. Thus, the and Photonics, Bellingham WA, Vol. 7740F.
overall cost for our test using the Amazon cloud was less than Buyya, R., Yeo, C. S., Venugopal, S., Broberg, J., and Brandic, I.
$100, an amount that is affordable by most users. 2009. Cloud computing and emerging IT platforms: Vision,
hype, and reality for delivering computing as the 5th utility.
27.4 conclusions Future Generation Computer Systems, 25, 599616.
Cary, A., Sun, Z., Hristidis, V., and Rishe, N. 2009. Experiences
It is abundantly clear that as a consequence of technological and on processing spatial data with MapReduce. In Scientific
sensor licensing improvements, the spatial, spectral, tempo- and Statistical Database Management, Springer, Berlin,
ral, and radiometric resolution of remote-sensing imagery will Germany, pp. 302319.
continue to increase. This translates into massive quantities of Chang, C. C., Chang, Y. L., Huang, M. Y., and Huang, B. 2011.
data that must be processed to glean meaningful information Accelerating regular LDPC code decoders on GPUs. IEEE
that can be used in a variety of decision support and visualiza- Journal of Selected Topics in Applied Earth Observations and
tion applications. Such data quantities quickly overwhelm the Remote Sensing (JSTARS), 4(3), 653659.
capabilities of even the most powerful desktop systems available Christophe, E., Michel, J., and Inglada, J. 2011. Remote sens-
now and in the foreseeable future. As a consequence, researchers ing processing: From multicore to GPU. IEEE Journal of
continue to explore cost-effective, yet powerful, computing envi- Selected Topics in Applied Earth Observations and Remote
ronments that can be harnessed for remote-sensing applications. Sensing (JSTARS), 4(3), 643652.
Dedicated HPC systems are expensive to acquire and maintain Cui, D., Wu, Y., and Zhang, Q. 2010. Massive spatial data
and have relatively short half-lives. This significantly diminishes processing model based on cloud computing model. In
this alternative as a practical solution. Instead, the availability Computational Science and Optimization CSO, 2010 Third
of high-speed communication technologies now makes the use International Joint Conference on Computational Science and
of distributed, pay-as-you-go resources an attractive choice for Optimization, May 2831, Anhui, China, Vol. 2, pp. 347350.
researchers and government agencies, as well as users in the pri- Dean, J., and Ghemawat, S. 2004. MapReduce: Simplified data pro-
vate sector. One term that is used to describe these distributed cessing on large clusters. In Proceedings of the Sixth Conference
resources is cloud computing. The cloud provides new patterns on Symposium on Operating Systems Design & Implementation,
for deploying remote-sensing-data processing and provides December 68, San Francisco, CA, Vol. 6, 13pp.
easy, inexpensive access to servers, elastic scalability, managed Eldawy, A., and Mokbel, M. F. 2013. A demonstration of
infrastructure, and low complexity of deployment. SpatialHadoop: An efficient MapReduce framework for
Despite these significant advantages, the use of cloud comput- spatial data. Proceedings of the VLDB Endowment, 6(12),
ing does have some limitations. First, since data are distributed 12301233.
across networks to distributed, but unknown, locations, security eScience Institute, University of Washington. 2012. Understanding
can become problematic. Thus, public cloud resources cannot cloud computing for research and teaching. http://escience.
be used exclusively for applications that require the processing washington.edu/get-help-now/understanding-cloud-com-
of many types of individual-level information. This limitation, puting-research-and-teaching (accessed on March 4, 2014).
however, is not normally significant for remote-sensing applica- ESRI. 2012. ArcGIS server. http://www.esri.com/software/arcgis/
tions. Secondly, different types of spatial algorithms may require arcgisserver (accessed on March 23, 2012).
considerable amounts of interprocessor communication, par- Gonzlez, J. F., Rodrguez, M. C., and Nistal, M. L. 2009. Enhancing
ticularly in big-data applications, and certain elements of cloud reusability in learning management systems through the inte-
infrastructure may therefore induce processing latencies that gration of third-party tools. In 39th IEEE Frontiers in Education
may be unacceptable to users. Conference, FIE09, October 1821, San Antonio, TX, pp. 16.
2016 by Taylor & Francis Group, LLC
Processing Remote-Sensing Data in Cloud Computing Environments 561
Haifang, Z. 2003. Study and implementation of parallel algo- Ostermann, S., Iosup, A., Yigitbasi, N., Prodan, R., Fahringer, T.,
rithms for remote sensing image processing. PhD the- and Epema, D. 2010. A performance analysis of EC2 cloud
sis, National University of Defence Technology, Chasha, computing services for scientific computing. In Cloud
China. Computing, Springer, Berlin, Germany, pp.115131.
Huang, Q., Yang, C., Nebert, D., Liu, K., and Wu, H. 2010. Cloud Parulekar, R. etal. 1994. High performance computing for land
computing for geosciences: Deployment of GEOSS clear- cover dynamics. In Proceedings of the 12th IAPR International
inghouse on Amazons EC2. In HPDGIS 10: Proceedings Conference on Pattern Recognition, 1994, Vol. 3Conference
of the ACM SIGSPATIAL International Workshop on High C: Signal Processing, October 913, Jerusalem, Israel, IEEE,
Performance and Distributed Geographic Information New York, 1994.
Systems, November 35, San Jose, CA, pp. 3538. Plaza, A., and Chang, C.-I. 2007. High Performance Computing in
Hyspeed Computing. 2013. Big data and remote sensingWhere Remote Sensing, CRC Press, Boca Raton, FL.
does all this imagery fit into the picture? http://hyspeedblog. Plaza, A., and Chang, C.-I. 2008. Special issue on high performance
wordpress.com/2013/03/22/big-data-and-remote-sensing- computing for hyperspectral imaging. International Journal of
where-does-all-this-imagery-fit-into-the-picture (accessed High Performance Computing Applications, 22(4), 363365.
on March 2014). Plaza, A., Du, Q., Chang, Y.-L., and King, R. L. 2011a. High perfor-
IBM. 2012. Bringing big data to the enterprise. http://www-01. mance computing for hyperspectral remote sensing. IEEE
ibm.com/software/data/bigdata (accessed on April 4, 2013). Journal of Selected Topics in Applied Earth Observations and
Iowa DNR. 2009. State of Iowa. http://www.iowadnr.gov/map- Remote Sensing (JSTARS), 4(3), 528544.
ping/lidar/index.html, retrieved April 29, 2009 (accessed Plaza, A., Plaza, J., Paz, A., and Sanchez, S. 2011b. Parallel hyper-
on April 4, 2013). spectral image and signal processing. IEEE Signal Processing
Krishnan, S., Bary, C., and Crosby, C. 2010. Evaluation of Magazine, 28(3), 119126.
MapReduce for gridding LIDAR data. In 2010 IEEE Second Plaza, A., Valencia, D., Plaza, J., and Martinez, P. 2006. Commodity
International Conference on Cloud Computing Technology cluster-based parallel processing of hyperspectral imagery.
and Science, CloudCom, November 30December 3, Journal of Parallel and Distributed Computing, 66(3), 345358.
Indianapolis, IN, pp. 3340. Prasad. 2013. Special issue on High performance computing in
Kumar, N., Lester, D., Marchetti, A., Hammann, G., and Longmont, remote sensing. Remote Sensing.
A. 2013. Demystifying cloud computing for remote sens- Project Matsu, http://matsu.opensciencedatacloud.org/. Accessed
ing application. http://eijournal.com/newsite/wp-content/ on March 4, 2014.
uploads/2013/06/cloudcomputing.pdf. Accessed on March 4, Rezgui, A., Malik, Z., and Yang, C. 2013. High-resolution spatial
2014. interpolation on cloud platforms. In Proceedings of the 28th
Lee, C. A., Gasster, S. D., Plaza, A., Chang, C. I., and Huang, B. Annual ACM Symposium on Applied Computing, March
(2011). Recent developments in high performance comput- 1822, Coimbra, Portugal, pp. 377382.
ing for remote sensingA review. IEEE Journal of Selected Shekhar, S., Gunturi, V., Evans, M. R., and Yang, K. 2012. Spatial
Topics in Applied Earth Observations and Remote Sensing, 4.3, big-data challenges intersecting mobility and cloud comput-
508527. ing. In Proceedings of the 11th ACM International Workshop
Li, J., Humphrey, M., Agarwal, D., Jackson, K., Ingen, C., and Ryu, on Data Engineering for Wireless and Mobile Access, May
Y. 2010. eScience in the cloud: A MODIS satellite data 2024, Scottsdale, AZ, pp. 16.
reprojection and reduction pipeline in the windows azure Simmhan, Y., and Ramakrishnan, L. 2010. Comparison of
platform. In IEEE International Symposium on Parallel & resource platform selection approaches for scientific
Distributed Processing (IPDPS), April 1923, Atlanta, GA, workflows. In Proceedings of the 19th ACM International
pp. 110. Symposium on High Performance Distributed Computing,
Liu, T. et al. 2010. Remote sensing image classification tech- June 2125, Chicago, IL, pp. 445450.
niques based on the maximum likelihood method. FuJian Song, C., Li, Y., and Huang, B. 2011. A GPU-accelerated wavelet
Computer, (001), 78. decompression system with SPIHT and Reed-Solomon decod-
McEvoy, G. V., and Schulze, B. 2008. Using clouds to address grid ing for satellite images. IEEE Journal of Selected Topics in Applied
limitations. In Proceedings of the 6th International Workshop Earth Observations and Remote Sensing (JSTARS), 4(3), 683690.
on Middleware for Grid Computing, December 15, Leuven, Sugumaran, R., Burnett, J., and Armstrong, M. P. 2014. Using a cloud
Belgium. computing environment to process large 3D spatial datasets.
Mell, P. and Grance, T. 2011. The NIST definition of cloud com- In H. Karimi, ed., Big Data: Techniques and Technologies in
puting. National Institute of Standards and Technology, Geoinformatics, CRC Press, Boca Raton, FL, pp. 5365.
Special Publication 800-145. Sugumaran, R., Oryspayev, D., and Gray, P. 2011. GPU-based
Oryspayev, D., Sugumaran, R., DeGroote, J., and Gray, P. 2012. cloud performance for LiDAR data processing. In COM.
LiDAR data reduction using vertex decimation and pro- Geo 2011: Second International Conference and Exhibition
cessing with GPGPU and multicore CPU technology. on Computing for Geospatial Research and Applications,
Computers & Geosciences, 43, 118125. May 2325, Washington, DC.
2016 by Taylor & Francis Group, LLC
562 Remotely Sensed Data Characterization, Classification, and Accuracies
USGS. 2011. Landsat archive. http://landsat.usgs.gov (accessed on Watson, P., Lord, P., Gibson, F., Periorellis, P., and Pitsilis, G. 2008.
April 7, 2013). Cloud computing for e-Science with CARMEN. In Second
Vatsavai, R. R., Ganguly, A., Chandola, V., Stefanidis, A., Klasky, Iberian Grid Infrastructure Conference Proceedings, May
S., and Shekhar, S. 2012. Spatiotemporal data mining in 1214, Porto, Portugal, pp. 314.
the era of big spatial data: Algorithms and applications. White, T. 2012. Hadoop: The Definitive Guide, OReilly Media,
Proceedings of the First ACM SIGSPATIAL International Inc., Sebastopol, CA.
Workshop on Analytics for Big Geospatial Data, November Yang, C. and Huang, Q. 2013. Spatial Cloud Computing: A
79, Redondo Beach, CA, pp. 110. Practical Approach, CRC Press, Boca Raton, FL.
Wang, S., and Armstrong, M. P. 2003. A quadtree approach to Yang, H., Du, Q., and Chen, G. 2011. Unsupervised hyperspec-
domain decomposition for spatial interpolation in grid com- tral band selection using graphics processing units. IEEE
puting environments. Parallel Computing, 29(10): 14811504. Journal of Selected Topics in Applied Earth Observations and
Wang, Y., Wang, S., and Zhou, D. 2009. Retrieving and index- Remote Sensing (JSTARS), 4(3), 660668.
ing spatial data in the cloud computing environment. In Yue, P., Gong, J., Di, L., Yuan, J., Sun, L., Sun, Z., and Wang, Q.
Proceedings of the First International Conference on Cloud 2010. GeoPW: Laying blocks for the geospatial processing
Computing, December 14, Beijing, China, Lecture Notes web. Transactions in GIS, 14(6), 755772.
in Computer Sciences, Vol. 5931, pp. 322331.
2016 by Taylor & Francis Group, LLC
XI
Google Earth for
Remote Sensing
28 g oogle e arth for r emote Sensing John E. Bailey ................................................................................................................565
Acronyms and Definitions Introduction Google Earth: Free and Accessible High-Resolution
Imagery Evolution of Google Earth in Everyday Life and in Science Google Earth for Telling Geospatial Stories
of Societal Importance Strengths and Limitations of Google Earth Leveraging Google Earth Imagery and
Databases Next-Generation Google Earth Conclusions Acknowledgments References
563
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
28
Google Earth for Remote Sensing
Acronyms and Definitions ................................................................................................................ 565
28.1 Introduction ............................................................................................................................ 565
28.2 Google Earth: Free and Accessible High-Resolution Imagery ........................................ 566
Google Earth Virtual Globe Google Earths First Major Application:
HurricaneKatrina Google Earths Numerous Application Possibilities
28.3 Evolution of Google Earth in Everyday Life and in Science ............................................ 569
Keyhole Markup Language for Easy to Create Geospatial Visualizations Early Adopters:
Bloggers and Journalists Early Science Application: Volcanology Evolution in a Multitude
of Science Applications
28.4 Google Earth for Telling Geospatial Stories of Societal Importance ..............................570
28.5 Strengths and Limitations of Google Earth ........................................................................571
28.6 Leveraging Google Earth Imagery and Databases .............................................................574
28.7 Next-Generation Google Earth .............................................................................................574
28.8 Conclusions.............................................................................................................................. 577
John E. Bailey Acknowledgments .............................................................................................................................. 577
University of Alaska Fairbanks References............................................................................................................................................ 577
Acronyms and Definitions 28.1 introduction
API Application programming interface A decade has now passed since the release of Google Earth (GE)
ASTER Advanced Spaceborne Thermal Emission and (Google Inc. 2005). It is hard to fully measure the impact it has
Reflection Radiometer had on all areas of society, as it opened possibilities in the world
AVHRR Advanced Very High Resolution Radiometer of remote sensing to everyone from the youngest student look-
CPU Central processing units ing at satellite images for the first time to an experienced pro-
DEM Digital elevation model fessor who was now able to access imagery at a scale and speed
EE (Google) Earth Engine never before possible. In 2004 the technology that was to later
GE Google Earth become GE was called EarthViewer (Figure 28.1) and was the
GEC Google Earth Community proprietary property of Keyhole Inc. It was one of several virtual
GUI Graphical user interface globes (Bailey 2010), computer applications creating a 3D model
KML Keyhole Markup Language of Earth or other planets that evolved in the early 2000s as tech-
LiDAR Light Detection and Ranging nology caught up with ideas (Bailey and Chen 2011). For many
MODIS Moderate Resolution Imaging Spectroradiometer they were seen as realization of a vision that had been laid out
NAIL Neighbors Against Irresponsible Logging in different forms by many from futurists (Fuller 1962) to aca-
NGA National Geospatial-Intelligence Agency demics (Bailey etal. 2012), to artists (Eames and Eames 1977) to
NGO Non governmental organization science fiction writers (Stephenson 1992). However, many credit
NOAA National Oceanic and Atmospheric Al Gores 1998 address to the California Science Center (Gore
Administration 1998) as succinctly verbalizing the concept of a Digital Earth,
OGC Open Geospatial Consortium of which GE ultimately became the most popular actualization.
SPOT Systme Pour lObservation de la Terre GE is currently available for a range of operating systems and
SRTM Shuttle Radar Topographic Mapping Mission platforms (Google Inc. 2014a). These can primarily be differenti-
WYSIWYG What you see is what you get ated into three categories: (1) desktop, (2) mobile, and (3) web
XML Extensible Markup Language (Figure 28.2). The desktop version is a downloadable binary that is
currently supported on PC, Mac, and Linux systems. The mobile
version is available as Android and iOS Apps but has reduced
565
2016 by Taylor & Francis Group, LLC
566 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 2 8.1 Screenshot of Keyholes EarthViewer 3D showing the Port of Tokyo near Hamazakibashi Junction.
functionality relative to the desktop version. GE is found on web- public. It will conclude by exploring the steps in the evolution of
pages in two different forms: as a WebGL rendering option in Google Geo as an analytical remote sensing toolset (Section 28.5)
Google Maps or as a web plug-in with a JavaScript application and will consider what Googles acquisition of their own satellite
programming interface (API). Google announced the release of company might mean for remote sensing using GE.
the desktop version of GE on June 28, 2005, as a free 3D mapping
and local search technology (Google Inc. 2005). The capability 28.2 Google earth: Free and Accessible
of importing GIS data was mentioned but the emphasis was on
GE as a searchable 3D map rather than as a visualization tool.
High-Resolution imagery
Content creation was referenced as easy creation and sharing of GE has integrated high and very high spatial resolution imagery
annotations among users (Google Inc. 2005). However, as sci- seamlessly for the entire world. The concept has many science
entists and remote sensing professionals began to explore with and practical applications.
GE, they realized that its greatest potential involved the use of
Keyhole Markup Language (KML), the code supported by the
28.2.1 Google earth Virtual Globe
application to visualize placemakers, lines, polygons, overlays,
and other types of content generated by the user. The core of the GE technology is the rendering of a 3D globe com-
This chapter explores how GE has become the go-to application prised of a combination of terrain data overlain by satellite imagery
for free and accessible high-resolution satellite imagery (Section and aerial photography. The base terrain data are a digital elevation
28.2) and how that imagery can be augmented with illustrations model (DEM) collected by the shuttle radar topographic mapping
(Section 28.3) and used for geospatial storytelling (Section 28.4), mission (SRTM) supplemented by other datasets for high latitudes
by both the professional remote sensing community and the and mountainous regions that requiring higher-resolution data
2016 by Taylor & Francis Group, LLC
Google Earth for Remote Sensing 567
(a)
(b)
(d)
(c)
FIg u r e 28.2 Screenshots of Google Earth on different platforms: (a) current binary application (GE 7.1.2.2041), (b) mobile Google Earth on the
Nexus 7 Android tablet, (c) earth globe in Google Maps, and (d) Google Earth plug-in.
to show the true topography. There are many areas, for example, broadband Internet connectivity, created use cases for large vol-
Western United States, The Swiss Alps, and Canary Islands, where umes of high-resolution data. Many have joked that the first thing
high-resolution light detection and ranging (LiDAR) datasets have new users do with GE is look at their house, but this in itself was an
been acquired and supplant the SRTM imagery. The original base evolutionary step in the world of remote sensing. For the first time
imagery for GE was Landsat 30 m multispectral data, pan sharp- hundreds of millions of people were presented with a view of what
ened using Landsats 15 m panchromatic imagery. In some rural their home or town looks like from space. The concept that satel-
areas, this remains the only available dataset. However, for many lites can be used to observe from Earth moved from the realm of
regions and most urban centers, very high spatial resolution imag- science fiction and something scientists do, to any everyday view-
ery (VHRI; submeter to 10 m resolution) has been acquired from point for anyone with a computer and an Internet connection.
DigitalGlobe, Systme Pour lObservation de la Terre (SPOT)
Image, and other commercial satellite providers.
28.2.2 Google earths First Major
The modern era of commercially available high-resolution sat-
Application: Hurricane Katrina
ellite data began with the launches of GeoEyes Ikonos (1999) and
DigitalGlobes Quickbird (2001) satellites. For the first time almost An opportunity for GE to display its potential occurred just
global coverage of Earth at centimeters to meters resolution was weeks after the Google announcement of the reworked Keyhole
available to all who could afford it. At the same time the devel- technology (Google Inc. 2005). On August 29, 2005, Hurricane
opment of Keyholes EarthViewer 3D and other virtual globes, Katrina made landfall in Louisiana, causing devastation across
technology now possible thanks to the gaming industries needs the region (Knabb etal. 2005). Nowhere was the effect felt more
for high-performance graphics cards and the global expansion of than in the city of New Orleans. Much of the city lies below
2016 by Taylor & Francis Group, LLC
568 Remotely Sensed Data Characterization, Classification, and Accuracies
sea level and the high rainfall produced by the hurricane led to The National Geospatial-Intelligence Agency (NGA) later
protective riverbank levees breaking, and the city was flooded. recognized Googles contribution with a Hurricane Katrina
Thousands of people were trapped in their houses and as the Recognition Award, but more importantly it demonstrated the
floodwaters rose, people escaped to the roofs of their houses but facility of the application to be more than just a tool for looking
often without any form of communication. at your house.
In order to find those stranded, plan logistics, and work
out access routes, the National Oceanic and Atmospheric
28.2.3 Google earths numerous
Administration (NOAA) captured thousands of high-resolution
Application Possibilities
aerial images. This information was invaluable for disaster
response efforts, with more than 5 million photos down- Despite the obvious benefits for efforts such as crisis
loaded from the NOAA website each day during the first week response, the ability of every user to become armchair
(Nourbakhsh etal. 2006). However, processing and navigating remote sensing analysts also raised concerns as conspiracy
this information in a useful manner was a colossal effort and theorists now had evidence to back up their claims of discov-
one for which GE was to provide the backbone. The GE engi- ering Atlantis or a hidden alien base (Geens 2009; Google
neers worked around the clock to serve the images through Inc. 2009). But for many communities, the positives far out-
the GE client application as KML overlays. These geolocated, weighed any negative connotations as these tools opened up
stitched together layers of data were searchable and easily acces- exploration of the whole planet for professionals and curious
sible by federal agencies, disaster response groups, and the gen- public alike. In particular, it has developed as an observation
eral public. For example, this imagery was used to identify intact tool for structural geologists, geomorphologists, volcanolo-
churches in flood-free suburbs of New Orleans that could used gists, and other scientists who require easy access to detailed
as outlet centers for aid donations (Nourbakhsh et al. 2006). imagery. In these cases GE has enabled studies at a scale that,
(a) (b) N
Surficial deposits
Units for mapping in
Google Earth 9
1
Gray-brown unit 2
8
Pale brown unit with 100 m
speckled texture
7
(c) N
White unit
6
Units with Ground
Truth Added
5 2
Nonresistant, fissile 4 15 m
limestone largely
mantled by lag deposit of
dark, desert-varnished (d) N
chert pebbles
3
5
4
Pale buff limestone 3
with large chert shell 6
concretions
7
2 8
9
Massive white 1
limestone 400 m
FIg ur e 28.3 Tewksbury et al. (2012) used Google Earth to study structural geology in remote areas in Egypt. Using Google Earth imagery, they
established a stratigraphy consisting of one subunit in the El-Rufuf Formation (Unit 1) and eight subunits in the Drunka Formation (Units 2-9).
2016 by Taylor & Francis Group, LLC
Google Earth for Remote Sensing 569
while technically possible, was not practically obtainable without is also annotated in a way that is syntactically distinguishable
extensive time-consuming and well-funded efforts (Figure 28.3). from regular text (Figure 28.4). Through KML, GE becomes
the canvas on which users can display geospatial information
28.3 evolution of Google earth in (Bailey 2010; Chen and Bailey 2011; Whitmeyer etal. 2012).
everyday Life and in Science
28.3.2 early Adopters: Bloggers and Journalists
28.3.1 Keyhole Markup Language for easy
Among early adoption users were bloggers and journalists who
to create Geospatial Visualizations
grasped the potential for bringing their stories to life (Lubick 2005;
While many different fields of Earth science have used GEs Butler 2006a). However, where this broke new ground was rather
imagery as a tool for landscape studies, it is another compo- than GE just being another way to great colorful illustrations, it
nent of Keyholes technology that has made GE revolutionary gave the journalists themselves the power to great dynamic repre-
as a visualization and scientific tool. KML is a computer code sentation of the data. This is what Declan Bulter, a Nature reporter
that can be generated by users to create and share visualiza- and early champion for GE abilities, did with data points show-
tions in GE (Wernecke 2009). Originally developed by Keyhole ing the spread of avian flu (Butler 2006b,c; Figure28.5). Although
Inc., in 2008 it became an international standard of the Open wide acceptance was slower at first, the scientific community also
Geospatial Consortium (OGC). KML is an Extensible Markup embraced the possibilities offered by GE for both research and
Language (XML), a computer code that defines a set of rules education (Grambling 2007; Simonite 2007; Bailey and Chen
for encoding documents in a format that is human readable but 2011; Yu and Gong 2012).
FIg u r e 28.4 A simple example of a keyhole markup language (KML) file. Different components of the text have been colored (in this example) for
the purposes of identification. The grey text is namespace information, which references the rules of KML file types. The black text creates a container
for KML features. The blue text creates a placemark at the given coordinates. The red text defines the default fly-to viewpoint for this placemark.
2016 by Taylor & Francis Group, LLC
570 Remotely Sensed Data Characterization, Classification, and Accuracies
H5N1 Human cases
Bird outbreaks
Period Number
JanApr 2005 of bird deaths
SepDec 2005 < 10
MayAug 2005 < 100
JanApr 2005 < 1,000
SepDec 2004 < 10,000
MayAug 2004 < 100,000
JanApr 2004
>100,000
SepDec 2003
Data SIO, NOAA, U.S. Navy, NGA, GEBCO
Image Landsat
FIg u r e 2 8.5 Visualizing the spread of the N5H1 virus (aka avian flu) using Google Earth and KML (Butler, 2006b,c).
28.3.3 early Science Application: Volcanology how GE has become integral to Earth observation activities could
be cited from a multitude of disciplines from structural geology
One of the first fields to demonstrate GEs capabilities to dynam- (Lisle 2006; De Paor 2008; De Paor and Whitmeyer 2011) or cryo-
ically visualize Earth observations was volcanology (Figure spheric studies (Ballagh etal. 2007, 2011; Gergely etal. 2008), to
28.6). The Smithsonian Global Volcanology Program worked land use assessment (Sheppard and Cizek 2008; van Lammeren
with the GE team to develop a volcano layer (Venzke et al. et al. 2010) or water resources management (Silberbauer and
2006; Figure 28.6a), which became one of the most frequently Geldenhuys 2008; USGS 2014), and everything in between.
accessed datasets in the application (Rebecca Moore, personal Ultimately all these fields (and others) are using GE because they
communication). Meanwhile researchers in the Alaska Volcano want to use Earth imagery to tell a story.
Observatory remote sensing group developed the use of KML
for modeling volcanic ash clouds (Webley etal. 2009a,b; Figure
28.6b) and observing real-time thermal satellite imagery (Bailey
28.4 Google earth for telling Geospatial
and Dehn 2006; Figure 28.6c and d).
Stories of Societal importance
From simple fly-throughs of the 3D landscape (Lang etal. 2012)
28.3.4 evolution in a Multitude
or KML-enhanced tours (Rueger and Beck 2012; Treves and Bailey
ofScience Applications
2012) to complex interactive experiences using the Earth API
The use of GE for volcanology is just one illustration of how it (Dordevic and Wild 2012), an obvious development of GE was
has become an important remote sensing tool. Examples showing for use as a platform for Virtual Field trips. Teachers embraced
2016 by Taylor & Francis Group, LLC
Google Earth for Remote Sensing 571
(a)
(b)
(c) (d)
FIg u r e 28.6 Examples of early uses of Google Earth by volcanologists: (a) Smithsonian GVPs Google Earth layer of Holocene volcanoes,
(b) 3D visualization of an ash plume that erupted from Cleveland Volcano in the Aleutian Islands, (c) locations of thermal hotspots shown using
KML derived from thermal IR images geolocated in GE, and (d) ash cloud density identified using AVHRR thermal IR imagery and visualized as
translucent overlays in GE.
this ability (Lamb and Johnson 2010) as way of taking students to While not the original intent, the success of NAIL led to
places they otherwise couldnt hope to go. However, nonprofits and conservation and nonprofit groups contacting Moore for help
conservation saw an opportunity to add a narrative to those tours with telling their stories. NGOs such as the Sierra Club, the
and explorations. The catalyst for this development was a citizen Jane Goodall Institute, and Appalachian Voices began to use
action groups response to a logging plan for their neighborhood GE as an educational platform for their messages (Google Earth
in Los Gatos, CA. The Neighbors Against Irresponsible Logging Outreach 2014; Figure 28.8). The interest in GE as an outreach
(NAIL) was a nonprofit organization formed in 2005 by Rebecca tool led Rebecca Moore to form the GE Outreach team, which
Moore, a software engineer who had recently joined the GE team. focuses on working with nonprofit and educational groups to
After receiving a low-quality black and white leaflet in her mail use Google Geo for Good.
outlining a companys logging plan for the Santa Cruz forest she
lived in, Moore used GE to create powerful visualizations of the 28.5 Strengths and Limitations
true scale of the plan that helped rally her neighbors into action of Google earth
(Figure 28.7). The press picked up on these visualizations and the
subsequent coverage led to the proposed logging plan being ruled The strengths in the choice of GE applications over applications
ineligible by the California Department of Forestry (Google Earth using other remote sensing data or imagery are due to three pri-
Outreach 2014). mary reasons:
2016 by Taylor & Francis Group, LLC
572 Remotely Sensed Data Characterization, Classification, and Accuracies
Lands of San Jose water Company
(a) Notice of Intent to Harvest Timber & Request for Information on Domestic Water Supplies
Area found on Los Gatos, Laurel, and Castle Rock Ridge, California USGS 7.5 Quadrangles
Miles
0 0.4 0.8 1.6 2.4 3.2
N
Legend
Nonindustrial Timber W E
Management Plan Area
Helicopter Landings S
(b)
FIg u r e 28.7 Information on proposed logging given to a neighborhood in Los Gatos, CA: (a) the original map circulated to neighbors by the
logging company and (b) a Google Earth visualization of the same data created by a Googler who lived in the neighborhood.
2016 by Taylor & Francis Group, LLC
Google Earth for Remote Sensing 573
1. Cost: Google Earth, and the more importantly access to the 3. Usability: The user-friendly interface and associated what
imagery, is free. you see is what you get (WYSIWYG) graphical user inter-
2. Accessibility: The process of continually making imagery faces (GUIs) allows anyone to explore imagery and create
and terrain data available at a global scale, for a variety annotations using KML.
of platforms and operating systems, requires considerable
However, some have viewed GE as a fun way to look at imagery,
computing infrastructure. Googles resources allow them
rather than an actual remote sensing or GIS platform. They cer-
to do this at a level few others are capable of.
tainly do not consider it a real remote sensing or GIS (Avraam
(a)
(b)
FIgu r e 28.8 Examples of NGOs uses of Google Earth: (a) The Sierra Club visualized the impacts of sea level rise on Vancouver, Canada
(Sheppard and Cizek 2008). (b) The Jane Goodall Institute created a geolocated blog about chimpanzees activity (Jane Goodall Institute
2010). (Continued)
2016 by Taylor & Francis Group, LLC
574 Remotely Sensed Data Characterization, Classification, and Accuracies
(c)
FIg u r e 28.8 (c ontinued ) Examples of NGOs uses of Google Earth: (c) Appalachian Voices used Google Earth imagery and KML overlays to
show mountaintop removal due to open-pit mining (Appalachian Voices 2014).
2009), though others disagree (Turner 2008). For others the idea An example of the power of EE is the Timelapse project,
that GE was a type of GIS for the masses (Bader and Glennon which leverages the archive of Landsat data, a program man-
2007) was acceptable as it became the go-to technology for pre- aged by the USGS that has been acquiring images of the Earths
sentations and illustrations using Earth imagery. surface since 1972 (see Chapter 1). Using EE a global mosaic
The primary argument of the not remote sensing or GIS crowd was constructed using 29years of Landsat satellite data, cover-
is that GE is a qualitative, not quantitative, tool. It is an excel- ing the globe from 1984 to 2012. Each frame of the Timelapse
lent platform for viewing imagery, but little to no analysis is pos- map is constructed from a year of Landsat imagery, creating an
sible directly within the application. But the success of GEand annual 1.7 terapixel snapshot of the Earth at 30 m resolution. In
Google Mapsled to continuous expansion of Googles develop- total the project processed 2,068,467 scenes, a total of 909 TB
ment of Geo tools and the development of an application that of data. This required 2 million CPU (central processing unit)
now meets the needs of remote sensing analysis. hours and processing was distributed over 66,000 CPUs allow-
ing the mosaics to be computed in 1.5 days. Through a com-
28.6 Leveraging Google earth bination of image stacking and pixel matching, the composite
images were created free from clouds and without the data loss
imagery and Databases created by the scan line corrector failure on Landsat 7. River
Google Earth Engine (EE) (Google Inc. 2014) is a cloud com- system changes (Figure 28.9), urban growth, receding glaciers,
puting platform for hosting and processing satellite imagery and and an expansion of mining operations are a few examples of
other Earth observation data. It leverages Googles large distrib- Earth surface changes that are clearly highlighted by Timelapse
uted server capabilities to analyze imagery. (Klugger and Walsh 2013).
Despite the name, EE does not use the GE technology (the
interface displays its output using the Maps API), but it does use 28.7 next-Generation Google earth
the same imagery and GIS databases created to support Earth,
and it was developed to meet the needs of the users using that This ability to map on a global scale is powerful enough by itself,
imagery to observe changes on our planets surface. EE brings given the availability of Landsat, moderate resolution imag-
together 40 years of satellite imagery and makes it available to ing spectroradiometer (MODIS), advanced spaceborne ther-
detect changes, map trends, and quantify differences on the mal emission and reflection radiometer (ASTER), and other
Earths surface. Applications include detecting deforestation, clas- large archives of other land observing satellites. However, the
sifying land cover, estimating forest biomass and carbon, and future offers even more potential with the recent acquisition by
mapping the worlds roadless areas (Hansen etal. 2013; Giri etal. Google of Skybox Imaging, a company building and launching
2014; Patel etal. in press). its own satellites to acquire high-resolution satellite imagery and
2016 by Taylor & Francis Group, LLC
Google Earth for Remote Sensing 575
ucallpa ucallpa
Map data Terms of use Map data Terms of use
2 km 2 km
2 ml 2 ml
1985 2000
Slow 1984 2012 Slow 1984 2000 2012
ucallpa ucallpa
Map data Terms of use Map data Terms of use
2 km 2 km
2 ml 2 ml
1990 2005
Slow 1984 1990 2012 Slow 1984 2005 2012
ucallpa ucallpa
Map data Terms of use Map data Terms of use
2 km 2 km
2 ml 2 ml
1995 2010
Slow 1984 1995 2012 Slow 1984 2010 2012
FIgu r e 28.9 Timelapse, powered by Google Earth Engine, uses 29 years of Landsat imagery to show the meandering changes of the Ucayali
River, near Pucallpa, Peru.
2016 by Taylor & Francis Group, LLC
576 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) (e)
(b) (f)
(c) (g)
(d) (h)
FIg u r e 28.10 Imagery and video from SkySat-1 and -2. (a to g) An example of how these satellites can be used to see the growth and decline of
a city in the Black Rock Desert, Nevada, during the week of Burning Man. (h) A still from Skybox satellite video captured over the Burj Khalifa,
Dubai. Note the airplane at the bottom right.
2016 by Taylor & Francis Group, LLC
Google Earth for Remote Sensing 577
TABLe 2 8.1 Technical Specifications for Sensors on SkySat-1 and SkySat-2 Satellites
Spatial Resolution
Image Band (at Nadir) (m) Temporal Resolution Swath Width/FoV File Format
Blue (450515nm) 2 Repeat coverage in 8km 16-bit GeoTIFF
3days to 1 week
Green (515595nm) 2 Repeat coverage in 8km 16-bit GeoTIFF
3days to 1 week
Red (605695nm) 2 Repeat coverage in 8km 16-bit GeoTIFF
3days to 1 week
Near-IR (740900nm) 2 Repeat coverage in 8km 16-bit GeoTIFF
3days to 1 week
Panchromatic 0.9 Repeat coverage in 8km 16-bit GeoTIFF
(450900nm)images 3days to 1 week
Panchromatic 1.1 Up to 90 s at 30 2km by 1.1km MPEG-4
(450900nm)video frames per second
high-definition video (Figure 28.10). Skyboxs innovative meth- References
ods allow them to build smaller and cheaper satellites (<$50 mil-
lion, Truong 2013), while still maintaining performance and an Appalachian Voices. 2014. Mountaintop removal in Google Earth.
ability to capture high-resolution imagery on par with capabili- http://ilovemountains.org/google_earth_tutorial (accessed
ties of DigitalGlobes and other high-resolution (<5 m) satellites September 26, 2014).
(Table 28.1). The combination of Skybox-acquired imagery, the Avraam, M. 2009. Google Maps and GIS. Intersecting Space
processing power of EE, and GE as an interface to view and and Time Through GIS Endeavors, http://michalisavraam.
explore offers exciting possibilities for the future of remote org/2009/10/google-maps-and-gis, October 2, 2009 (accessed
sensing. September 27, 2014).
Bader, J. and A. Glennon. 2007. Virtual Globes: GIS for the
28.8 conclusions Masses? AAG Annual Meeting, Paper Session 4510, April
1721, San Francisco, CA.
As GE turns a decade old, it seems appropriate to reflect on Bailey, J.E. 2010. Entry for Virtual Globe. In Warf, B. ed.,
the role it has played in terrestrial remote sensing. While its Encyclopedia of Geog, Sage Publications, Los Angeles, CA,
analytical potential is still being unlocked, the role it has 3528p.
played in democratizing satellite imagery cannot be denied. Bailey, J.E. and A. Chen. 2011. The role of Virtual Globes in geo-
The idea that anyone, anywhere can take out their phone and science. Comp Geosci 37(1):12.
in a matter of seconds see a relatively recent view from space Bailey, J.E. and J. Dehn. 2006. Volcano monitoring using Google
of their current location seemed like the stuff of science fic- Earth. Geoinformatics 2006Abstracts, USGS Scientific
tion as recently as the end of the last millennium. Yet now Investigations Report 2006-5201, p. 25., May 1012,
GE is an accepted everyday tool for scientists, educators, gov- Reston, VA.
ernment, and the general public alike and has fundamentally Bailey, J.E., S.J. Whitmeyer, and D.G. De Paor. 2012. Introduction:
changed how we view our planet (and house). The application of Google Geo Tools to geoscience educa-
tion and research. Geol Soc Am Special Paper 492:viixix.
Acknowledgments Ballagh, L.M., M.A. Parsons, and R. Swick. 2007. Visualising
cryospheric images in a virtual environment: present chal-
True appreciation of the evolution of Google Earth involves lenges and future implications. Polar Record 43(4):305310.
acknowledging countless Keyhole and Google software engi- doi: http://dx.doi.org/10.1017/S0032247407006523.
neers and product and program managers, who developed and Ballagh, L.M., B.H. Raup, R.E. Duerr, S.J.S. Khalsa, C. Helm,
continue to transform the technology. However, the role of D. Flower, and A. Gupte. 2011. Representing scientific
researchers, teachers, bloggers, and citizen scientists cannot be data sets in KML: Methods and challenges. Comp Geosci
understated either. It is fair to say that GE would not have become 37(1):5764. doi:10.1016/j.cageo.2010.05.004.
what it has without the active Google Earth Community (GECa Butler, D. 2006a. The Web-Wide World. Nature 439:776778.
user forum formerly known as the Keyhole BBS) and blogging Butler, D. 2006b. Avian flu maps in Google Earth. http://declan-
community. GE has impacted the lives of countless groups and butler.info/blog/?p=16 (accessed September 22, 2014).
individuals, including this author who discovered it while in Butler, D. 2006c. The spread of avian flu with time; new maps
postdoctoral researcher and ended up on a career path that ulti- exploiting Google Earths time series function. http://declan-
mately saw him become part of the Google Earth Outreach team. butler.info/blog/?p=58 (accessed September 22, 2014)
2016 by Taylor & Francis Group, LLC
578 Remotely Sensed Data Characterization, Classification, and Accuracies
De Paor, D.G. 2008. Enhanced visualization of seismic focal Hansen, M.C., P.V. Potapov, R. Moore, M. Hancher, S.A. Turubanova,
mechanisms and centroid moment tensors using solid A. Tyukavina, D. Thau et al. 2013. High-resolution global maps
models, surface bump-outs, and Google Earth. J Virtual of 21st-century forest cover change. Science 342:850853.
Explorer, 29:100. Jane Goodall Institute. 2010. Video Highlights Partnership with
De Paor, D.G. and S.J. Whitmeyer 2011. Geological and Amazon Tribe, Google Outreach. http://www.janegoodall.
geophysical modeling on virtual globes using KML, org/media/news/google-earth-video-tour-highlights-jgi-
COLLADA, and Javascript. Comp Geosci 37(1):100110. partnership-amazon-tribe (accessed September 22, 2014).
doi:10.1016/j.cageo.2010.05.003. Klugger, J. and B. Walsh. 2013. Timelapse, TIME. http://world.
Dordevic, M.M. and S.C. Wild. 2012. Avatars and multi-student time.com/timelapse/ (accessed September 5, 2014).
interactions in Google Earthbased virtual field experi- Knabb, R.D, J.R. Rhome, and D.P. Brown. 2005. Hurricane Katrina:
ences. In Whitmeyer, S.J., J.E. Bailey, D.G. De Paor, and August 2330, 2005 (Tropical Cyclone Report). US National
T.Ornduff, eds., Google Earth and Virtual Visualizations in Oceanic and Atmospheric Administrations National Weather
Geoscience Education and Research. Geol Soc of Am Special Service (accessed September 30, 2014).
Paper 492:315321. doi:10.1130/2012.2492(22). Lamb, A. and L. Johnson 2010. Virtual expeditions: Google Earth,
Eames, C. and R. Eames 1977. Powers of Ten: A Film Dealing with GIS, and geovisualization technologies in teaching and
the Relative Size of Things in the Universe and the Effect of learning. Teacher Librarian 37(3):8185.
Adding Another Zero. IBM. Lang, N.P., K.T. Lang, and B.M. Camodeca. 2012. A geology-
Fuller, B. 1962. Education Automation, Freeing the Scholar to focused virtual field trip to Tenerife, Spain. In Whitmeyer,
Return to His Studies. Southern Illinois University Press, S.J., J.E. Bailey, D.G. De Paor, and T. Ornduff, eds., Google
Carbondale and Edwardsville, Feffer & Simons Inc., Earth and Virtual Visualizations in Geoscience Education
London, U.K. and Research. Geol Soc of Am Special Paper 492:323334.
Geens, S. 2009. Media stupidity Watch: No, Its not Atlantis, Ogle doi:10.1130/2012.2492(23).
Earth Blog. http://ogleearth.com/2009/02/media-stupidity- Lisle, R.J. 2006. Google Earth: A new geological resource. Geology
watch-no-its-not-atlantis/ (accessed September 23, 2014). 22(1):2932
Gergely, K.L., T.M. Haran, and B. Billingsley. 2008. Virtual Globe Lubick, N. 2005. Spinning around the globe online. Geotimes,
Visualizations of Cryospheric Data at the National Snow and December 2005.
Ice Data Center. Eos Trans. AGU, 89(53), Fall Meet. Suppl., National Oceanic and Atmospheric Association. 2005. Hurricane
Abstract IN41A-1119, December 1519, San Francisco, CA. Katrina Images. http://ngs.woc.noaa.gov/katrina/ (accessed
Giri, C., J. Long, S. Abbas, R.M. Murali, F.M. Qamer, and September 18, 2014).
D. Thau, D. 2014. Current status and dynamics of man- Nourbakhsh I., R. Sargent, A. Wright, K. Cramer, B. McClendon, and
grove forests of South Asia. J Environ Management M. Jones. 2006. Mapping disaster zones. Nature 439:787788,
doi:10.1016/j.jenvman.2014.01.020. doi :10.1038/439787a.
Google Earth Outreach. 2014. Get inspired by organizations who Patel, N.N., Angiuli, E., Gamba, P., Gaughan, A., Lisini, G.,
have used Google mapping tools for good, http://www. Stevens, F.R., Tatem, A.J., and Trianni, G. Multitemporal
google.com/earth/outreach/stories/index.html (accessed settlement and population mapping from Landsat using
September 26, 2014). Google Earth Engine. J. Appl. Geophys.
Google Inc. 2005. Google Launches Free 3D Mapping and Rueger, B.F. and E.N. Beck. 2012. Benedict Arnolds march to
Search Product, Google Blog, http://googlepress.blogspot. Quebec in 1775: An historical characterization using Google
com/2005/06/google-launches-free-3d-mapping-and_28. Earth. In Whitmeyer, S.J., J.E. Bailey, D.G. De Paor, and
html, (accessed September 21, 2014). T.Ornduff, eds., Google Earth and Virtual Visualizations in
Google Inc. 2009. Atlantis? No, it Atlant-isnt, Google Blog. http:// Geoscience Education and Research. Geol Soc of Am Special
googleblog.blogspot.com/2009/02/atlantis-no-it-atlant- Paper 492:347354. doi:10.1130/2012.2492(25).
isnt.html (accessed September 27, 2014). Sheppard, R.J. and P. Cizek. 2008. The ethics of Google Earth:
Google Inc. 2014a. Google Earth. http://www.google.com/earth Crossing thresholds from spatial data to landscape visualisa-
(accessed September 30, 2014) tion. J Environ Management 90(6):21022117. doi:10.1016/j.
Google Inc. 2014b. Google Earth Engine. https://earthengine. jenvman.2007.09.012.
google.org (accessed September 30, 2014) Silberbauer, M.J. and W. Geldenhuys. 2008. Using Keyhole
Gore, A. 1998. The Digital Earth: Understanding our planet in Markup Language to create a spatial interface to South
the 21st Century. Speech given at the California Science African water resource data through Google Earth.
Center, Los Angeles, CA, on January 31, 1998, http://www. In FOSS4G 2008 Free and Open Source Software for
digitalearth.gov/VP19980131.html (accessed September Geospatial, OSGeo, GISSA.
30, 2014). Simonite, T. 2007. Virtual Earths let researchers mash up
Gramling, C. February 2007. Google PlanetWith virtual globes, data. New Scientist, www.newscientist.com/article/dn11773
Earth scientists see a new world. Geotimes, p. 3840. (accessed September 15, 2014)
2016 by Taylor & Francis Group, LLC
Google Earth for Remote Sensing 579
Stephenson, N. 1992. Snow Crash. Bantam Dell, Random House van Lammeren R., J. Houtkamp, S. Colijn, M. Hilferink, and
Inc., New York, 470p. A. Bouwman. 2010. Affective appraisal of 3D land use
Tewksbury, B.J., A.A.K. Dokmak, E.A. Tarabees, and A.S. visualization. Comp Environ Urban Syst 34:465475.
Mansour. 2012. Google Earth and geologic research in Venzke, E., L. Siebert, J.F. Luhr. 2006. Smithsonian volcano
remote regions of the developing world: An example data on Google Earth. Eos Trans. AGU, 87(52), Fall
from the Western Desert of Egypt. In Whitmeyer, S.J., Meet. Suppl., Abstract N43A-0900, December 1115, San
J.E. Bailey, D.G. De Paor, and T. Ornduff, eds., Google Francisco, CA.
Earth and Virtual Visualizations in Geoscience Education Webley, P.W., J. Dehn, J. Lovick, K.G. Dean, J.E. Bailey, and L.
and Research. Geol Soc of Am Special Paper 492:2336, Valcic. 2009a. Near-real-time volcanic ash cloud detec-
doi:10.1130/2012.2492(02). tion: Experiences from the Alaska volcano observatory.
Treves, R. and J.E. Bailey. 2012. Best practices on how to design In Mastin L., and P. Webley. eds. Volcanic Ash Clouds, J.
Google Earth tours for education. In Whitmeyer, S.J., Volcanol. Geotherm. Res. Special Issue 186:7990, doi:
J.E. Bailey, D.G. De Paor, and T. Ornduff, eds., Google 10.1016/j.jvolgeores.2009.02.010
Earth and Virtual Visualizations in Geoscience Education Webley P.W., K. Dean, J.E. Bailey, J. Dehn, and R. Peterson.
and Research. Geol Soc of Am Special Paper 492:383394. 2009b. Automated forecasting of volcanic ash dispersion
doi:10.1130/2012.2492(28). utilizing Virtual Globes. Natural Hazards 51(2):345361.
Truong A. 2013. Proof that cheaper satellites still can take doi:10.1007/s11069-008-9246-2.
incredibly detailed photos of Earth, Fast Company, http:// Wernecke, J. 2009. The KML Handbook: Geographic Visualization
www.fastcompany.com/3023325/fast-feed/proof-that- for the Web. Addison-Wesley, Upper Saddle River, NJ, 368p.
cheaper-satellites-still-can-take-incredibly-detailed- Whitmeyer, S.J., J.E. Bailey, D.G. De Paor, and T. Ornduff. 2012.
photos-of-earth, (accessed September 6, 2014) Google earth and virtual visualizations in geoscience edu-
Turner, A. 2008. Is GoogleMaps GIS? High Earth Orbit, http:// cation and research. Geol Soc of Am Special Paper 492, 468p.
highearthorbit.com/is-googlemaps-gis/, June 12, 2008 Yu, L. and P. Gong. 2012. Google Earth as a virtual globe tool for
(accessed September 17, 2014). Earth science applications at the global scale: progress and
USGS. 2014. WaterWatch, http://waterwatch.usgs.gov/kml.html, perspectives. Int J Rem Sens 33(12):39663986. doi:10.1080/
(accessed September 22, 2014). 01431161.2011.636081
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
XII
Accuracies, Errors,
and Uncertainties
of Remote Sensing
Derived Products
29 Assessing Positional and Thematic Accuracies of Maps g enerated from r emotely
SensedData RussellG.Congalton ...........................................................................................................................................583
Acronyms and Definitions Introduction Assessing Map Accuracy Positional Map Accuracy
Assessment Thematic Map Accuracy Assessment Conclusions References
581
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
29
Assessing Positional and Thematic
Accuracies of Maps Generated
from Remotely Sensed Data
Acronyms and Definitions ................................................................................................................ 583
29.1 Introduction ............................................................................................................................ 583
29.2 Assessing Map Accuracy ....................................................................................................... 583
29.3 Positional Map Accuracy Assessment ................................................................................. 585
Initial Considerations Collecting Reference Data Computing Descriptive Statistics
29.4 Thematic Map Accuracy Assessment .................................................................................. 589
Initial Considerations Collecting Reference Data Computing Descriptive
Statistics Basic Analysis Techniques Advanced Analysis Techniques
Russell G. Congalton 29.5 Conclusions.............................................................................................................................. 600
University of New Hampshire References............................................................................................................................................ 600
Acronyms and Definitions be carefully considered from the beginning of the project as well
as several statistical and practical methodologies that must be
ASPRS American Society for Photogrammetry and Remote balanced to achieve a successful and valid assessment. Therefore,
Sensing instead of a few simple steps, this chapter presents these consid-
AVHRR Advanced very-high-resolution radiometer erations and methodologies in a flowchart (Figure 29.1) to help
CMAS Circular Map Accuracy Standards the reader begin to see all the components that must be thought
FEMA Federal Emergency Management Agency out and planned for to conduct a valid accuracy assessment. The
FGDC Federal Geographic Data Committee rest of this chapter deals with each of the parts of this flowchart.
GPS Global positioning system
KHAT Estimate of the Kappa statistic 29.2 Assessing Map Accuracy
LiDAR Light detection and ranging
MAS Map Accuracy Standards Today, assessing the accuracy of a map generated from remotely
MMU Minimum mapping unit sensed imagery is a routine component of most mapping projects.
MODIS Moderate resolution imaging spectroradiometer However, this was not always the case. By the end of World War
MSS Multispectral scanner II, the use of aerial photography and photo interpretation was a
NMAS National Map Accuracy Standards well-established means of learning about the earths resources.
NSSDA National Standard for Spatial Data Accuracy Maps generated from aerial photos for such uses as agricultural
OBIA Object-based image analysis monitoring, forest inventory, geologic exploration, and many
RMSE Root mean square error others became commonplace. A key component of every photo
interpretation project included the necessary field visits to train
29.1 introduction the interpreter to recognize the objects of interest on the ground
and on the aerial photos. It was generally recognized that the
This chapter is devoted to assessing the accuracy of maps created human interpreter drew lines on the photo (i.e., created poly-
from remotely sensed data. In the ideal handbook, it would be gons) around areas that seemed to be distinct from each other
great to list the 10 steps that the reader must follow in order to (i.e., had more variation between polygons than within a poly-
conduct such an accuracy assessment. Unfortunately, map accu- gon) and then did their best to appropriately label the polygons.
racy assessment does not follow such simple procedures. Instead, Those areas that were difficult to label were checked in the field
there are a considerable number of important factors that must during another field visit. Very little thought was given to any
583
2016 by Taylor & Francis Group, LLC
584 Remotely Sensed Data Characterization, Classification, and Accuracies
Assessing map accuracy
Positional Thematic
accuracy accuracy
Initial Classification
Sources of error scheme
considerations
Collecting
reference data
Sampling Other
considerations
Independence
Source
Sampling Sample Sample
scheme size unit
When
Spatial
autocorrelation Consistency
Computing
descriptive
statistics
Basic
analysis
techniques
FIg u r e 2 9.1 Flowchart of accuracy assessment considerations.
type of quantitative evaluation or accuracy assessment of the changed the way that we looked at creating maps from remotely
resulting photo-interpreted maps. sensed imagery. Instead of using the human being as the inter-
A notable exception to this lack of interest in photo interpretation preter, a computer could now be used to manipulate the digital
accuracies occurred in the early 1950s. Some researchers wishing values in the imagery and produce a map. Since the computer,
to promote the field of photo interpretation as a science recognized and more specifically some mathematical/statistical algorithm
the need to evaluate the accuracy of their work and published a implemented by the computer, was now the entity producing the
number of papers on the topic (Sammi, 1950; Katz, 1952; Colwell, map from the digital remotely sensed imagery, questions about
1955; Young, 1955). A panel discussion entitled, Reliability of how good the map was soon followed.
Measured Values was held at the 18th Annual Meeting of the Actually, in any new technology, there is an initial exuberance
American Society of Photogrammetry in 1952. At this discussion, of use of that technology without much thought to the quality of
Mr. Amrom Katz (1952), the panel chair, made a strong case for the what is being produced. This is quite natural as the excitement of
use of statistics in photogrammetry. The results of all these papers the new technology dominates and many new uses of the technol-
and discussions culminated with a paper by Young and Stoeckler ogy are investigated. However, as the technology begins to mature,
(1956). In this paper, these authors proposed techniques for a quan- questions about quality and limitations are inevitable. Such was the
titative evaluation of photo interpretation, including the use of an case for digital image processing. During the first 510years after
error matrix to compare field and photo classifications, and went the launch of Landsat, many exaggerated claims about the quality
on to present a discussion of the boundary error problem when and detail that could be produced from the imagery prevailed such
labeling polygons. Unfortunately, nothing really became of these that there was a great overselling of the technology. Fortunately, by
forward-thinking ideas for about the next 30years. the early 1980s, a number of researchers began to question some of
In 1972, the first Landsat was launched thus strongly pro- these claims and started the development of methods for assessing
moting the field of digital image processing. That first Landsat the accuracy of maps derived from remotely sensed data.
multispectral scanner had only four bands and sensed in only It must be noted at this point that map accuracy really has
three wavelengths (green, red, and two NIR), but it completely two components: positional accuracy and thematic accuracy.
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 585
Positional accuracy deals with the accuracy of the location of from the U.S. Bureau of the Budget in 1947. NMAS is a very
map features and measures how far a spatial feature on a map is simple and straightforward standard that states the following:
from its true or reference location on the ground (Bolstad, 2005).
For horizontal accuracy, not more than 10% of the points
Thematic accuracy deals with the labels or attributes of the fea-
tested may be in error by more than 1/30th of an inch
tures of a map and measures whether the mapped feature labels
(at map scale) for maps larger than 1:20,000 scale, or by
are different from the true or reference feature label. This chapter
more than 1/50th of an inch for maps of 1:20,000 scale or
deals with both types of accuracy, and the flowchart in Figure 29.1
smaller, and
is appropriate for both types also.
For vertical accuracy, not more than 10% of the elevation
tested may be in error by more than one half the contour
interval.
29.3 Positional Map Accuracy
Assessment While easy to apply, NMAS provides no information about
determining statistical bounds around the error, but rather just
In order to make a map, you must know where you are. uses the percentile method for accepting or rejecting the map as
Therefore, it is critical in any mapping project that the deter- accurate. Therefore, NMAS is not commonly used today.
mination of the exact same location on both the image or The next step in developing methods for positional accuracy
map and the reference data (often the ground) be assessed assessment resulted from the work of Greenwalt and Schultz
(see Figure 29.2). If this correspondence is not attained, then (1962 and 1968). Their report (The Principles of Error Theory
being in the wrong location may result in a thematic error. and Cartographic Applications, Greenwalt and Schultz 1962 and
For example, it is possible to be in the right place and misla- 1968) proposed equations by which to estimate the maximum
bel (incorrectly measure, observe, or label) the attribute. This error interval for a given probability. They computed a one-
error would be a thematic error. However, it is also possible dimensional map accuracy standard (MAS) used for elevation/
to correctly label the attribute, but be in the wrong place. vertical (z) data assessment and a two-dimensional circular
This locational error could then also lead to a thematic error. map accuracy standard (CMAS) for horizontal (x andy) data
Positional error/accuracy and thematic error/accuracy are assessment. Greenwalt and Schultz assumed that map errors
not independent of each, and it is critical that every possible are normally distributed and used the equations for MAS to
effort be made to control both of them in order to effectively estimate the interval around the mean vertical error and the
assess the accuracy of any map. equations for CMAS to estimate the interval around the mean
The history of assessing the positional accuracy of maps horizontal error within which 90% of the error should occur.
begins with the National Map Accuracy Standards (NMAS) These equations can also be used to estimate the distribution of
True point location
Roads
FIg u r e 29.2 Example of positional accuracy showing that the road intersections as indicated by the yellow dots (reference points) and the roads
are not in the same location.
2016 by Taylor & Francis Group, LLC
586 Remotely Sensed Data Characterization, Classification, and Accuracies
errors at probabilities other than 90% and offers more flexibility two considerations are intimately linked together. Historically,
than the previous NMAS. guidelines such as NMAS and even NSSDA did not specify any-
In 1989, the American Society for Photogrammetry and thing about where the samples should be taken to conduct the
Remote Sensing (ASPRS) released the ASPRS Interim Accuracy assessment. However, with the advent of LiDAR data, it was
Standards for Large-Scale Maps (ASPRS, 1989). Instead of stipu- quickly recognized that land cover or vegetation type at the loca-
lating that no more than 10% of the errors may exceed a given tion the samples were collected significantly impacts the accu-
maximum value (the NMAS approach), the ASPRS standard racy of the results. Therefore, vegetation or land cover type is
computes a mean error from a set of samples and then provides now considered an important source of error in positional accu-
a threshold value that this mean error cannot exceed. This stan- racy assessment and is mitigated by taking a minimum num-
dard is expressed in ground units and not map units as NMAS ber of samples (usually 20) in each of the major vegetation or
had used. The document also reviews and confirms the work of land cover types regardless of the size of the area (FEMA, 2003;
Greenwalt and Schultz, but does not make any recommenda- ASPRS, 2004). As previously mentioned, the exact vegetation/
tions about the use of these equations. land cover classes (i.e., the classification scheme) varies between
The National Standard for Spatial Data Accuracy (NSSDA) the FEMA and ASPRS guidelines, but both agree that samples
was released by the U.S. Federal Geographic Data Committee must be acquired in a variety of cover types in order to have a
(FGDC) in 1998. This standard represents the currently approved valid assessment.
method for assessing positional accuracy and establishes some
much needed guidelines for measuring, analyzing, and report-
29.3.2 collecting Reference Data
ing positional accuracy for maps as well as other geo-referenced
imagery (FGDC, 1998). These guidelines have been widely There are many factors that must be considered when collecting
adopted by those in the federal, state, and local governments and the reference data used to assess the positional accuracy of the
also in the private sector. The full version of the NSSDA stan- map (see Figure 29.1). If the reference data are not collected prop-
dards (FGDC, 1998) can be downloaded at http://www.fgdc. erly, then the entire assessment may be invalid. Many of these con-
gov/standards/projects/FGDC-standards-projects/accuracy/ siderations involve sampling while others deal with issues such as
part3/chapter3. independence, source, timing of collection, and consistency.
NSSDA does not use a maximum allowable error as the deci- Independence is a key component of the assessment process.
sion point if a map has acceptable positional error at any scale. The data used must be independent from any data used in the
Instead, it recommends determining an allowable error thresh- registration of the map or other spatial data. This fact is com-
old as needed such that accuracy is then reported in ground dis- monly understood in statistical analysis, but there seems to
tances at the 95% confidence level. This 95% level is stricter than still be many examples in positional accuracy where statements
previous standards that tended to use the 90% level. NSSDA about the goodness of the position are still expressed using the
incorporates the work of Greenwalt and Schultz and used these data that were employed to register the map to the ground. These
equations to determine accuracy as a maximum threshold error data are clearly not an independent data set and represent a most
at a given or specified probability. However, there are issues with optimistic and invalid estimate of the actual positional accuracy.
the calculations of the NSSDA as there is a mistake in the calcu- Obviously, the source of the reference data must be more
lations that will be discussed later in this section. accurate than the map that is being assessed. Sometimes, a map
Most recently, three new documents/guidelines have been of larger scale is sufficient. In other situations, a survey using
developed that deal with positional accuracy, but especially with GPS or other equipment is required. NSSDA recommends data
regard to LiDAR data. These documents include (1) Guidelines be of the highest accuracy that is feasible and practical (FGDC,
and Specifications for Flood Hazard Mapping Partners (FEMA, 1998). Others have suggested that the positional reference data
2003), which specifies that there must be 20 samples taken in each be from one to three times more accurate than the anticipated
of the major vegetation cover types (at least three cover types); accuracy of the map being tested (e.g., Ager, 2004; ASPRS, 2004;
(2) the ASPRS Guidelines for Reporting Vertical Accuracy of NDEP, 2004).
LiDAR Data (ASPRS, 2004), which confirms the FEMA stan- When and who collects the reference data can also be impor-
dard, but uses a slightly different vegetation cover type classi- tant although typically these factors are more important in
fication; and (3) Guidelines for Digital Elevation Data from the thematic accuracy assessment. It is important that major changes
National Digital Elevation Program (NDEP, 2004), which agrees have not taken place due to earthquakes or other natural phe-
with the various accuracy measures specified in the ASPRS nomena that could alter the positions of objects (again, this is a
(2004) guidelines. rare situation). Also, the collection of the data must be consis-
tent between multiple collectors. In other words, if more than
one individual is collecting the data, it is important to establish
29.3.1 initial considerations
objective procedures to ensure that the data are collecting simi-
Important initial considerations when assessing positional larly by all collectors.
accuracy include both sources of error and appropriate classifi- Since the assessment of the positional accuracy of a map is
cation scheme. In fact, in positional accuracy assessment, these performed not at every place on the map, but rather at a series
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 587
of points, it is critical that the sampling be valid to achieve an the map into quadrants and forcing a minimum of 20% of the
appropriate assessment. Sampling involves determining the samples into each quadrant. This method also minimizes spatial
number of samples, the sample unit (identification), and the autocorrelation in the sampling by setting a minimum spacing
sampling scheme or distribution. Again, failure to properly plan between sample points such that no two points can be closer
for these factors will result in an invalid assessment. The NSSDA together than d/10 where d is the diagonal dimension of the map.
(FGDC, 1998), the current standard, states that a minimum of This method seems to be a very effective way to ensure that the
20 samples must be used in the assessment. Some of the new sampling considerations for positional accuracy assessment are
guidelines that require selecting samples from a number of veg- not ignored.
etation/land cover classes require 20 samples per class, while
some have increased that number to 30 per class (ASPRS, 2004;
NDEP, 2004). From a statistical standpoint, taking 30 samples 29.3.3 computing Descriptive Statistics
is often recommended. Unfortunately, many assessments have As previously discussed, assessing positional accuracy is per-
been conducted with either nonindependent samples or too few formed using a sample of data to estimate the agreement (i.e., fit)
samples resulting in statistically invalid results. between the map or other geospatial data and a reference data
The sample unit in positional accuracy assessment is actually set that is assumed to be correct. This analysis is done through
a point. In fact, it is critical that these points be well defined and the computation of a number of statistics. The key statistic that
represent a feature for which the horizontal position is known is computed in positional accuracy assessment is the root mean
to a high degree of accuracy and position with respect to the square error (RMSE). The RMSE is simply the square root of the
geodetic datum (FGDC, 1998). Any ambiguity about where a mean of squared differences between the samples on the map
point is located will disqualify it as a sample point for use in the and those same samples on the reference data. The differences
positional accuracy assessment. between the map and reference sample points are squared to
Finally, the sampling scheme determines how the samples are ensure positive values since the simple arithmetic difference
collected throughout the map (i.e., the sample distribution). It is between the map and reference sample points can be either neg-
important that the samples are distributed throughout the map ative or positive.
so that the entire map is assessed. The full range of variation in As alluded to previously, positional accuracy can be measured
the map should be considered including topography, vegetation/ in the vertical dimension (z) or in the horizontal dimension,
land cover, and important features. Figure 29.3 demonstrates a which involves both x and y. The analysis is similar in each case,
stratified approach proposed by ASPRS (1989) that guarantees but it is little more complicated for the horizontal dimension.
that the samples are distributed throughout the map by dividing The equation for RMSE for vertical accuracy is
(e )
n
>20% >20% 2
vi
RMSEv = i
(29.1)
n
d where
10
e vi = v ri v mi (29.2)
vri equals the reference elevation at the ith sample point
vmi equals the map elevation at the ith sample point
n is the number of samples
The equation for RMSE for horizontal accuracy is
d
e
n
2
hi
RMSEh = i
(29.3)
n
where
>20% >20% e hi2 = (x ri x mi )2 + (y ri y mi )2 (29.4)
xri and yri are the reference coordinates
FIg ur e 29.3 Sampling scheme showing samples distributed through- xmi and ymi are the map coordinates for the ith sample point
out the map. n is the number of samples
2016 by Taylor & Francis Group, LLC
588 Remotely Sensed Data Characterization, Classification, and Accuracies
The NSSDA states that positional accuracy be reported at the required statistics for assessing the positional accuracy of any
95% level defined as 95% of the locations in the data set will map or geospatial data set.
have an error with respect to the reference position that is equal It should be noted here that there has been significant
to or less than the computed statistic (FGDC, 1998). The equa- confusion in the mapping community over some of the
tion for computing the NSSDA for vertical accuracy is given in computations used in assessing positional accuracy. This
the guidelines as confusion results because the term RMSE as used by map-
ping professionals differs from the term RMSE used in sta-
NSSDA Vertical Accuracyv = 1.96(RMSEv) (29.5) tistics. In addition, the term standard error is unfortunately
used to depict different parameters in different professions.
The equation for computing the NSSDA for horizontal accuracy is While most statistics textbooks define the standard error as
a little more complicated because of the two dimensions (x and y). the square root of the variance of the population of means, X ,
It is possible to have a different distribution of errors in the x many mapping texts define the standard error as the square
direction than in the y direction causing the errors around the root of the variance of the population signified by .
position to be oblong in shape rather than circular as one would Statisticians call this term the standard deviation. As a result,
expect if the errors were equally distributed. Most use the sim- some of the equations that are correct in the work by
plified equation ignoring the distribution of errors and compute Greenwalt and Schultz have been misapplied or at least mis-
NSSDA for horizontal accuracy as interpreted by the FGDC for use in developing the NSSDA.
While it is not possible in this chapter to provide a full expla-
NSSDA Horizontal Accuracy = 1.7308 * RMSEh (29.6) nation of this confusion and the statistical ramifications that
have resulted from this issue, a full discussion is provided
Table 29.1 shows the computation of NSSDA and RMSE for a in Congalton and Green (2009). Currently, the NSSDA is the
small horizontal data set. Careful study of this table shows why accepted standard, and most positional accuracy assessments
the differences between the reference position and map position report this statistic along with the RMSE. Newer guidelines
are squared to eliminate the positive and negative values. These as reviewed previously have suggested refinements including
computations can easily be executed in an Excel spreadsheet sampling in the major vegetation/land cover types as well as
or other software package to quickly compute the currently other ways to assess positional accuracy.
TABLe 2 9.1 Example of Computing Positional Accuracy (RMSE and NSSDA)
Sample No. x (ref) x (map) x diff (x diff)2 y (ref) y (map) y diff (y diff)2 (x diff)2 + (y diff)2
1 36.5 37.2 0.7 0.49 842.5 843.7 1.2 1.44 1.93
2 12.0 11.1 0.9 0.81 900.0 896.2 3.8 14.44 15.25
3 66.8 66.7 0.1 0.01 1010.1 1009.1 1.0 1.00 1.01
4 4.1 4.2 0.1 0.01 786.5 782.4 4.1 16.81 16.82
5 9.1 8.5 0.6 0.36 655.8 658.2 2.4 5.76 6.12
6 77.0 79.0 2.0 4.00 676.0 672.1 3.9 15.21 19.21
7 112.1 112.1 0.0 0.00 655.1 666.2 11.1 123.21 123.21
8 99.0 99.8 0.8 0.64 688.1 689.2 1.1 1.21 1.85
9 55.5 54.9 0.6 0.36 744.3 737.1 7.2 51.84 52.20
10 102.1 110.5 8.4 70.56 810.7 811.2 0.5 0.25 70.81
11 6.8 6.7 0.1 0.01 845.6 847.2 1.6 2.56 2.57
12 200.5 198.2 2.3 5.29 902.1 902.4 0.3 0.09 5.38
13 252.1 250.9 1.2 1.44 937.4 939.2 1.8 3.24 4.68
14 260.8 262.0 1.2 1.44 945.5 940.1 5.4 29.16 30.60
15 300.9 299.5 1.4 1.96 960.8 962.0 1.2 1.44 3.40
16 266.8 267.0 1.8 3.24 971.2 971.1 0.1 0.01 3.25
17 142.1 135.8 6.3 39.69 1000.3 999.1 1.2 1.44 41.13
18 96.7 96.5 0.2 0.04 1006.8 1008.2 1.4 1.96 2.00
19 42.0 36.9 5.1 26.01 1101.3 1110.0 8.7 75.69 101.70
20 12.2 15.1 2.9 8.41 999.9 1002.0 2.1 4.41 12.82
Sum 515.94
Ave. 25.78
RMSE 5.08
NSSDA 8.79
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 589
29.4 thematic Map Accuracy Assessment Reference data
F OV D W
The history of assessing the thematic accuracy of maps derived
from remotely sensed data is shorter than that of positional accu- F 55 11 4 3 Land cover categories
racy. Except for a brief effort to quantitatively assess the accu- F = Forest
Map OV 6 51 5 8
racy of photo interpretation in the 1950s, the history of thematic OV = Other vegetation
accuracy assessment began shortly after the advent of digital D 0 8 55 9 D = Developed
imagery (i.e., the launch of the first Landsat). Researchers, nota- W = Water
bly Hord and Brooner (1976), van Genderen and Lock (1977), W 4 7 3 62
and Ginevan (1979), proposed criteria and basic techniques for
testing overall thematic map accuracy. In the early 1980s, more FIg u r e 2 9.4 An example error matrix.
in-depth studies were conducted and new techniques proposed
(Rosenfield et al., 1982; Congalton et al., 1983; Aronoff, 1985). contingency table in statistics. The process of creating an
Finally, from the late 1980s up to the present time, a great deal error matrix is demonstrated in Figure 29.5. An error matrix
of work has been conducted on thematic accuracy assessment. is a square array of numbers set out in rows and columns that
As we develop increasingly high spatial and spectral resolu- expresses the number of sample units assigned to a particular
tion sensors, more complex classification algorithms including category (i.e., land cover class) in one classification as compared
object-based image analysis (OBIA), and faster computer hard- to the number of sample units assigned to a particular category
ware with easier-to-use software, it is vital that our methods for (i.e., land cover class) in another classification. Typically, one of
effectively assessing thematic maps continue to improve as well. the classifications is considered to be correct (or at least of sig-
The history of thematic accuracy assessment has progressed nificantly higher accuracy than the other) and is called the refer-
through a number of stages as it developed and matured. As ence data. Unfortunately, it has become common practice to call
already discussed, the initial excitement of any new technology these reference data the ground truth. While the reference data
typically results in a time of intensive use of the technology with- should have high accuracy, there is no guarantee that it is 100%
out much, if any, attention paid to quality of the results. This stage correct and using the term ground truth is not appropriate. This
quickly moved into the next stage in which the map looks good. author hopes that the geospatial community will abandon this
There are a number of ways to qualitatively assess that a map term for ones that better represent the actual data such as refer-
looks good including visual comparison of the thematic map to ence data or ground data or field data or the like. The columns
the original imagery (Congalton, 2001) or perhaps to other image of the error matrix typically represent the reference data, while
sources (e.g., Google Earth or the like). The map could be shown the rows indicate the map derived from remotely sensed data or
to those especially familiar with the area or the map could be visu- other geospatial data set. It is possible to switch the headings for
ally compared to other thematic maps of the area. Each of these the row and columns so the analyst must take care to notice how
methods could convince the analyst that the map looks good. It the axes of the matrix are labeled before any further analysis is
is important that a thematic map look good. If the analyst is not undertaken.
convinced the map is of sufficient quality, there really is no sense
in conducting a quantitative accuracy assessment. Therefore, that
29.4.1 initial considerations
it looks good is a necessary but not a sufficient characteristic of
any thematic map. Unfortunately, some are still stuck in the it There are two very important initial considerations that must be
looks good stage of thematic map accuracy assessment and never taken into account before beginning a thematic accuracy assess-
go any further to effectively assess the map. ment. These are classification scheme and sources of error.
Quickly following the it looks good stage of thematic map assess-
ment was the stage called non-site-specific assessment. In this stage, 29.4.1.1 classification Scheme
some effort was made to obtain quantitative estimates of the accu- Thematic maps are an abstraction of the real world and as such use
racy of the thematic map. However, these estimates represent results some type of classification scheme to group together and simplify
of the map as a whole but do not say anything about the accuracy reality. Determining which classification scheme to use or devel-
at any specific location on the map. Instead, estimates such as the oping your own must be completed at the beginning of the map-
total hectares of forest on the map are compared to some refer- ping project. Failure to select an efficient and effective scheme from
ence estimate of the total forest hectares in that same area. While as early in the project as possible dooms the project to significant
non-site-specific assessments do begin to introduce some quantita- inefficiencies and potential failure. Whether the thematic map is
tive assessment into the process, the results are left lacking in that of land cover or vegetation or soil types or some other thematic
nothing can be said about any specific location on the map. Therefore, information, it is critical that the scheme represents the informa-
this method rapidly gave way to the current stage of thematic map tion that is necessary for the end user. However, it is also impor-
accuracy assessment called site-specific accuracy assessment. tant that this information be discernible with the remotely sensed
Site-specific thematic accuracy assessment is represented data selected for the mapping. Therefore, selecting the appropri-
in the form of an error matrix (Figure 29.4), also called a ate scheme at the beginning of the project facilitates choosing the
2016 by Taylor & Francis Group, LLC
590 Remotely Sensed Data Characterization, Classification, and Accuracies
Reference data
Accuracy assessment data work sheet
Map classification of Reference data
Sample site name from land cover
land cover sample site
sample site
1 Town reserve SB C
2 Yatsovitch preserve C SB
3 Browns forest C C
4 Sunshine farm
AG AG
5 State park SB C
6 Plown grow farm AG AG
7 Greens woods D D
Compile error matrix
A. Reference data: Data collected
from sample sites.
B. Map classification: Classification from A
satellite imagery. Reference data
C. Class titles: land cover classes identified C Row
D C AG SB
from map classification and reference data. Totals
D
1
D. Major diagonal boxes: Number of times that B D
MAP Classification
map classification agrees with the C
2
reference data for a particular class. E
E. Row totals: For each row, sum of all tally AG 2
marks in that row. H
SB 2
F. Column totals: For each column, sum of
all tally marks in that column. Column 1 3 2 1 7
Totals
G. Tot al data samples: Number of data samples.
G
H. Tally marks: Tallied data, that is, |||| F
Accuracy assessment
Overall accuracy indicates how well the map identifies the land cover type on the ground.
Producers accuracy indicates how many times a land cover type on the ground was identified as the
land cover type on the map. It expresses how well the map producer identified a land cover type on the
map from the satellite imagery data.
Users accuracy indicates how many times a land cover type on the map is really the land cover type on
the ground. It expresses how well a person using the map will find that land cover type on the ground.
FIg ur e 29.5 A diagram showing how an error matrix is generated from a series of samples. (From Landcover Protocols of the GLOBE Teachers
Manual, www.globe.gov.)
suitable imagery and allows reference data to be collected for use definitions to clearly eliminate any overlap between classes so that
in assessing the accuracy of the resulting thematic map. every area on the map or ground falls into one and only one class
Any classification scheme selected should have the following in the scheme. (3) Totally exhaustive: The scheme must be totally
four characteristics: (1) Definition: The scheme must contain rules exhaustive. In other words, every area on the map or ground must
or definitions that explicitly define each of the map classes. It is fall into one of the classes in the classification scheme. An effec-
not sufficient to simply list the map classes and assume that every- tive means of ensuring that your scheme is totally exhaustive is
one agrees on exactly what each class is. For example, a forest may to include an other class. However, if a large portion of the the-
be defined as trees greater than 5 m tall. Using this definition, a matic map ends up in the other class, then it may be appropriate
grouping of trees that are only 3 m tall would not be defined as a to rethink the classification scheme as perhaps some important
forest. (2) Mutually exclusive: The scheme must be mutually exclu- information on the map has been neglected. (4) Hierarchical:
sive. In other words, every effort must be made using the class Finally, it is useful if the classification scheme is hierarchical. That
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 591
is having multiple levels of detail (a hierarchy) such that the map area to be mapped on digital data, and therefore, consideration
can be generated and especially assessed for accuracy at differ- of an mmu even on digital mapping projects is important.
ent levels of detail. For example, the forest class may be further
divided in conifer and deciduous classes that add more detail. The 29.4.1.2 Sources of error
conifer class may be further divided into a pine class and a non- Creating thematic maps from remotely sensed imagery is possi-
pine class. As we will see shortly, it may not be possible to assess ble because there is a strong linkage between what can be sensed
the accuracy of the map at the most detailed level because of costs on the imagery and what is actually happening on the ground.
of collecting the reference data. Therefore, it may be possible with However, this correlation is not perfect, and every situation
a hierarchical classification scheme to map to one level of the hier- where this correlation breaks down is a source of potential error
archy but to assess the accuracy to a lesser level of detail. between the map and the ground. Lunetta etal. (1991) presented
In determining the appropriate classification scheme for a the- a discussion of these sources of error, which included errors
matic mapping project, the analyst must also consider the con- from image acquisition, errors from data processing, errors that
cept of minimum mapping unit (mmu). The mmu is the smallest occur within the analysis including data conversion, errors that
area that is uniquely delineated on a thematic map. This concept can occur in the accuracy assessment process itself, and finally
is widely known and used in photo interpretation where the maps error in decision making and implementation.
are created from aerial photographs of a given scale. However, While it is clear that error can enter a thematic mapping proj-
the concept has not been as widely adopted in digital imagery as ect from a great variety of sources, little has been done to evalu-
the imagery can be rendered at any scale, and pixel size seems to ate these errors and prioritize methods for dealing with them.
be more of the limiting factor. As will be seen in the discussion Congalton and Brennan (1999) and Congalton (2009) proposed
of sample unit, the pixel should not be selected as the minimum an error budget analysis method that not only clearly lists the
TABLe 2 9.2 Error Budget Analysis for a Remote Sensing Project
Error Source Error Contribution Potential Implementation Difficulty Implementation Priority Error Assessment Technique
Systematic
Sensor Low 5 21 Instrumentation and analysis
Natural
Atmosphere Medium 4 20 Instrumentation and analysis
Preprocessing
Geometric registration Low 2 Positional accuracy assessment
Image masking Medium 3 7 Single date error matrix
Derivative data
Band ratios Low 1 18 Data exploration
NDVI Low 1 18 Data exploration
PCA Low 1 Data exploration
TCA Low 1 8 Data exploration
Other ancillary data Medium 3 9 Data exploration
Classification
Classification scheme Medium 2 14 Single date error matrix
Training data collection Medium 3 15 Single date error matrix
Classification algorithm Medium 3 11 Single date error matrix
Post-processing
Data conversion High 2 12 Single date error matrix
Accuracy assessment
Sample unit Low 2 13 Single date error matrix
Sample size Low 2 16 Single date error matrix
Sampling scheme Medium 3 17 Single date error matrix
Spatial autocorrelation Low 1 3 GeoStatistical analysis
Positional accuracy High 2 2 RMSE/NSSDA
Final product
Decision making Medium 2 1 Sensitivity analysis
Implementation Medium 2 6 Sensitivity analysis
Note: Error Contribution Potential, ranked from low to medium to high; Implementation Difficulty, ranked from 1: not very difficult to 5: extremely
difficult; Implementation Priority, ranked from 1 to n showing order in which to implement improvements.
2016 by Taylor & Francis Group, LLC
592 Remotely Sensed Data Characterization, Classification, and Accuracies
sources of error in any mapping project, but also documents a used for reference data must be part of the accuracy assessment
method for evaluating the error contribution, implementation report.
difficulty, and implementation priority for each error. Table 29.2
shows an example of this error budgeting. 29.4.2.3 When Should the Reference Data Be collected
Timing for collecting reference data for assessing the accuracy of
thematic maps clearly depends on the type of information being
29.4.2 collecting Reference Data
mapped. Some information is just more timely than others. For
There are many factors that must be considered when collecting example, if mapping agricultural crops, it is very important that
the reference data used to assess the thematic accuracy of the the reference data be collected as near to the day the imagery
map. Collection of reference data is more complicated for the- was collected as possible (in some cases, same day collection is
matic accuracy assessment than positional accuracy assessment, important). In other examples where change occurs on a much
and therefore, there is even more risk that the entire assessment slower time scale, collection of the reference data within a year
may be invalid if the reference data are poorly collected. Many or 2 or 5 may be sufficient. Clearly, if a change occurs between
of the factors that must be considered involve sampling, while the date of the remotely sensed image capture and the date of the
others deal with issues such as independence, source, timing of reference data collection, then the reference data will be of little
collection, and consistency. or no use in effectively assessing the accuracy of the map.
29.4.2.4 consistency in Reference Data collection
29.4.2.1 independence
A good classification scheme consisting of labels with clear rules
It is absolutely critical that the reference data used to assess
or definitions goes a long way to ensuring consistency in refer-
the thematic map accuracy be independent of any other data
ence data collection. A guidebook documenting the procedures
used for training or any other purpose during the mapping
used to collect the data is also important. Finally, a field form,
project. This independence can be achieved in two ways. First,
whether on paper or using computer input, completes the pro-
the collection of the reference data can be done at a separate
cess. Having these three components facilitates training of all
time and/or by different personnel from the collection of the
field personnel whether it is a single collector or a large group
training data. While effective, this method is inefficient. A sec-
being disbursed throughout the study area. Objectivity in the
ond approach involves collecting the training and reference
data collection process is key, and having a good reference data
data simultaneously and then splitting the data into two sets
collection form will help enforce this. Regardless of the classifi-
whereby the reference data set is put aside and not viewed by
cation scheme, the form will have certain common components
any project personnel until it is time to conduct the accuracy
such as the names of the collectors, date of collection, location
assessment.
(GPS) information, a key to the map classes (dichotomous keys
are excellent), the actual map class determined by the collection,
29.4.2.2 Sources of Reference Data and a place to describe any issues, special findings, or problems
The source of the reference data in any thematic map accuracy that occurred at the collection site.
assessment is a function of the complexity of the classification
scheme (i.e., level of map detail) and the budget of the project. 29.4.2.5 Sampling
It is fortuitous if existing maps (e.g., USDA Cropland data layer) As in positional accuracy assessment, thematic accuracy assess-
or ground data (e.g., U.S. Forest Service Inventory Data) can ment relies on a statistical valid sample of reference data to
be used as the reference data. However, differences in classifi- compare to the map to compute the required statistics. It is not
cation scheme and most often size of the sample unit typically possible to assess the entire map (a total enumeration) because
limit using existing reference data, no matter how tempting of time, money, and effort constraints. Therefore, sampling
they are. Often, the existing data are too old or the sampling (a subset of the map) is used to facilitate the accuracy assessment.
plot size (sample unit) is too small to be used as reference data. Sampling for collecting reference data for assessing the accu-
Therefore, more commonly, the reference data are newly col- racy of thematic maps derived from remotely sensed data can
lected information assumed to be more accurate than the map be divided into three components: the sample unit, the sample
that is being assessed. Aerial photo interpretation has often size, and the sampling scheme including spatial autocorrelation.
been used to assess thematic maps generated from moderate Each of these is very important to ensure proper and valid refer-
resolution digital imagery, ground/field visits are used to assess ence data collection.
maps made from high-resolution airborne imagery, and manual
image interpretation has been used to assess maps created from 29.4.2.5.1 SampleUnit
automated classification techniques. No matter what source is Historically, a single pixel has often been used as the sample unit
selected as the reference data, it is imperative that the data are for assessing thematic map accuracy. Unfortunately, a single
effectively and appropriately collected so as to provide a viable pixel is a very poor choice for the sample unit because a pixel is
data set from which to validly assess the accuracy of the the- arbitrary, difficult to locate, and typically smaller than the mini-
maticmap. Therefore, explanation and justification of the data mum mapping unit defined by the project. Even with all the latest
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 593
advances in GPS, terrain correction, and geometric registration, sample unit for collecting the reference data. There are a few new
all reference data sample units have some positional errors. Mid- issues that arise when a polygon is used for accuracy assessment.
resolution sensors such as Landsat consider positional errors of First, the polygon may be much larger that a 3 3 pixel cluster,
one half a pixel (1015 m) to be reasonable. Higher spatial resolu- and therefore, more effort may need to go into accurately label-
tion sensors have smaller pixel sizes but increased positional errors ing that polygon on the reference data (MacLean et al., 2013).
if expressed in terms of pixels (i.e., a sensor with 1 m pixels could Second, the polygons are no longer of the same area as was the
have a positional error of 1015mthe same amount as Landsat, case for a 3 3 pixel cluster. Therefore, a more area-based error
but what is half of a Landsat pixel represents a 1015pixel error matrix may be appropriate instead of just tallying the sample
with this sensor). If a Landsat pixel is registered to the ground units (Maclean and Congalton, 2012, 2013). This topic will be
with 1015 m accuracy and the GPS unit used to locate the center covered more in the section on advanced analysis techniques
of the pixel on the ground has an accuracy of 1015 m, then it is later in this chapter.
totally impossible to use that single pixel as the sample unit for
assessing the thematic accuracy of the map because the pixel sim- 29.4.2.5.2 SampleSize
ply cannot be located accurately. The problem is only exacerbated Collecting enough samples to obtain a statistically valid reference
more with increased spatial resolution imagery. data set is perhaps the most challenging component of assessing
Therefore, the appropriate sample units are either a cluster of the accuracy of a thematic map. In almost every situation, there
pixels or a polygon. Considering positional accuracy limitations, is a balance between what is required to be statistically valid
a cluster of pixels, typically 3 3 pixels for moderate resolution and what is affordable within the given budget. Early on, many
imagery, has been a good choice. A homogeneous cluster of pix- researchers, including Hord and Brooner (1976), van Genderen
els minimizes positional error problems while maintaining a and Lock (1977), Hay (1979), Ginevan (1979), Rosenfield et al.
constant-sized sample unit. If assessing the accuracy of a the- (1982), and Congalton (1988b), published equations and guide-
matic map made from course resolution imagery such as MODIS lines for choosing the appropriate sample size. Initial efforts used
or AVHRR, it is unlikely to find many homogeneous single pix- the binomial equation to estimate the number of samples required
els let alone a 3 3 cluster. In this case, it may be necessary to to report the overall accuracy of a map. However, sampling to
document the proportions of the sample unit in each land cover complete an error matrix is not a binomial situation, but rather a
class and set rules for labeling the reference data accordingly. If multinomial one in which there is one correct answer and n 1
assessing the accuracy of a thematic map made from fine-resolu- wrong answers (where n is the number of thematic classes) for
tion imagery such as GeoEye or Digital Globe data, it is impor- each sample (Congalton, 1988b; Congalton and Green, 2009).
tant to select appropriate sampling units that account for the Tortora (1978) presents the equations needed to compute the
positional error. For example, if the imagery has 1 m pixels, then required sample size using the multinomial sampling approach.
a 3 3 cluster will not compensate for positional error. Instead, The number of samples depends on the number of thematic
a sample unit of perhaps 20 20 pixels is needed to compensate classes and the allowable error and the desired confidence level.
for a 10 m registration error. Congalton (1988b) determined, using an extensive series of
It is important to remember that if a cluster of pixels is selected Monte Carlo simulations, that most maps could be assessed using
as the sample unit, then the cluster (sample unit) represents a 50 samples per thematic map class (maps less than about 500,000
single sample and is tallied as such in the error matrix. There ha and 12 or less thematic map classes). More complex maps with
are far too many examples in the literature where an analyst has more than 12 thematic map classes and covering larger areas
selected a cluster of pixels and then considered each pixel in the should use 75100 samples per thematic map class. Assessing the
cluster as a separate sample unit. Collecting reference data in this thematic accuracy of global and/or continental maps requires
way is incorrect and results in an invalid assessment. It should partitioning the area into some type of ecological zones with the
also be noted that typically the clusters selected for sampling are requisite number of samples per zone. Therefore, assessing these
homogeneous (this may not be possible with coarse-resolution maps requires some minimum number of samples per zone times
imagery). Selecting homogenous clusters has the great advan- the number of zones.
tage of minimizing positional error, but may result in a biased It should be noted that practical considerations are a key com-
or inflated assessment as borders between classes (i.e., heteroge- ponent for determining sample size. There is a trade-off between
neous areas) are avoided, and therefore, these edge errors may the budget and the number of samples needed. There is a point
not be assessed. Finally, remember that while a 3 3 pixel cluster that is reached when the samples per thematic class are under 30
should work well for moderate resolution imagery, much larger per class that the assessment loses statistical validity completely,
clusters (in terms of number of pixels, not meters) must be col- and it may be better not to even attempt the accuracy assessment.
lected for higher spatial resolution imagery in order to account
for positional error. 29.4.2.5.3 SamplingScheme
Thematic maps generated from aerial photo interpretation A number of sampling schemes (how to collect the samples)
and more recently from OBIA are vector maps where polygons have been suggested for collecting the reference data including
are delineated on the imagery instead of the typical rasters (i.e., simple random sampling, systematic sampling, stratified sam-
pixels). In these situations, the polygon should be used as the pling, stratified random sampling, and cluster sampling. While
2016 by Taylor & Francis Group, LLC
594 Remotely Sensed Data Characterization, Classification, and Accuracies
random sampling has very nice statistical properties, it is often If the sand and water were correctly mapped, but the other
impractical to implement because of the costs of getting to every classes were incorrect all the time, we would still determine that
random location and because thematic classes consisting of this map was 95% correct based on this sampling scheme. This
small areas are left under-sampled. Systematic sampling ensures is a statistically valid and yet completely misleading result of our
that samples are distributed over the entire study area and is random sampling strategy. Therefore, selection of the appro-
very effective if higher resolution imagery is the source used to priate sampling scheme is as important as selecting the correct
obtain the reference data. Systematic sampling has similar prob- sample unit and number of samples.
lems of access when collecting ground reference data. Cluster A final concept that influences the choice of sampling scheme
sampling, that is, collecting a number of sample units within and how the samples are selected is called spatial autocorrelation.
close proximity to each other, offers certain efficiencies, but care Spatial autocorrelation occurs when the presence, absence, or
must be taken to not take too many samples too close together degree of a certain characteristic affects the presence, absence, or
because of spatial autocorrelation (samples are not independent degree of that same characteristic in neighboring units (Cliff and
of each other). Therefore, most reference data collection efforts Ord, 1973). In other words, the concept involves independence
use stratified sampling in order to ensure some minimum num- between neighboring samples. If an error is made at a certain
ber of samples in each stratum (thematic map class). location and spatial autocorrelation exists, then it is more likely
Many accuracy assessments use the rule expressed earlier and that the same error will occur in nearby locations. Congalton
include 50 or so samples in each thematic map class. In some (1988a) and Pugh and Congalton (2001) have demonstrated that
situations, it may be appropriate to have more samples in certain spatial autocorrelation is an important consideration in the col-
thematic classes than in others. These situations include when lection of reference data, and therefore, samples should be taken
some classes are more important or when some classes occupy so as to maximize the separation between samples as practical.
more of the map area than the others. Care is necessary here to Additionally, if cluster sampling is necessary, the number of
ensure some minimum number of samples in each thematic class clusters taken at a location should be as few as possible, and they
so as to be able to determine the accuracy of every thematic map should be as far apart as possible.
class. For example, imagine an area that is 95% sand and water Given the complexities, considerations, and choices that are
(see Figure 29.6) and the 5% divided among commercial, indus- involved in collecting proper reference data to assess the the-
trial, and residential areas. If a stratified sampling approach was matic accuracy of a map generated from remotely sensed data,
not employed where a minimum number of samples were taken it is absolutely required that a report be produced to document
from each map class and instead a random sample of 200 sam- the process. Simply reporting an error matrix or, even worse, an
ples were selected, it is possible to achieve a high map accuracy overall accuracy measure without describing the choices made
and yet poorly map 3 of the 5 map classes. Probability tells us and the justification for these choices results in an incomplete
that if 200 samples were randomly selected from a map that was assessment. Such a document is necessary for the map user to
95% sand and water, then 95% or 190 of these samples would thoroughly understand the accuracy assessment process.
fall in the sand and water. Only 10 samples would fall within
the other 3 map classes (commercial, industrial, and residential).
29.4.3 computing Descriptive Statistics
Assuming that the error matrix has been created using a valid
reference data collection approach as described earlier, the
matrix is then the starting point for a number of descriptive sta-
tistics including overall accuracy, producers accuracy, and users
accuracy. Figure 29.7 contains the same error matrix as was pre-
sented in Figure 29.4, but now shows the computation of some
descriptive statistics. Remember that an error matrix is a very
effective way to represent map accuracy because the individual
accuracies of each category are easily discerned along with both
the errors of inclusion (commission errors) and the errors of
exclusion (omission errors) present in the map. Just like a coin
has two sides (heads and tails), map errors have two components
(omission error and commission error). A commission error
can be defined as including an area into a thematic class when it
Water Sand Commercial Residential Industrial doesnt belong to that class, while an omission error is excluding
1:250,000 map sheet that area from the thematic class in which it truly does belong.
Each and every error is an omission from the correct thematic
FIg u r e 29.6 A map showing an area of 95% sand and water with a map class and a commission to a wrong thematic map class.
small portion (5%) of the area commercial, residential, and industrial Figure 29.7 shows that six areas (sample units) were mapped
land cover types. as other vegetation, while the reference data show that they were
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 595
Reference data that 55 samples were called forest on the map that were actu-
Row
F OV D W total ally forest, but in addition, the map called 11 samples forest that
F 55 4 3 73 Land cover categories
were actually other vegetation, 4 samples forest that were actu-
11
F = Forest
ally developed, and 3 samples forest that were actually water. The
OV 6 51 5 8 70
Map OV= Other vegetation map, therefore, called 73 samples forest, but only 55 are actually
D 0 8 55 9 72 D = Developed forest. There was a commission error of 18 samples into the forest
W 4 7 3 62 76 W= Water that were not forest. The forest users accuracy is then computed
by dividing the major diagonal value for the forest class by the
Column 65 77 67 82 291 Overall accuracy = total number of samples mapped as forest, 55/73= 75%. In evalu-
total (55 + 51 + 55 + 62 )/291= ating the accuracy of an individual map class, it is important to
123/291= 77%
consider both the producers and the users accuracies.
Producers accuracy Users accuracy Overall, producers, and users accuracies can also be
F = 55/65 = 85% F = 55/73 = 75% explained with the following equations. Begin with n samples
OV = 51/77 = 66% OV = 51/70 = 73% that are distributed into k2 cells where each sample is assigned
D = 55/67 = 82% D = 55/72 = 76% to one of k thematic map classes in the map (usually the rows)
W = 62/82 = 75% W = 62/76 = 82% and, independently, to one of the same k thematic map classes in
the reference data set (usually the columns). Then, let nij denote
FIg u r e 29.7 Error matrix showing the calculations for overall,
the number of samples mapped into thematic map class i (i = 1,
producers, and users accuracies. Note: (1) As the map producer,
2,, k) in the map and thematic map class j (j = 1, 2,, k) in the
85% of the forest sample units (in the reference data) are cor-
reference data set.
rectly classified as forest on themap. (2) As a map user, 75% of the
Let
sample units classified as forest on the map are indeed forests.
k
actually forest. In other words, six areas were omitted from the ni + = n
j =1
ij (29.7)
forest class and committed to the incorrect other vegetation
class. In addition to showing the omission and commission
be the number of samples classified into thematic map class i
errors, the error matrix in Figure 29.7 also shows the compu-
in the map, and
tation of three other accuracy measures: (1) overall accuracy,
(2) producers accuracy, and (3) users accuracy (Story and k
Congalton, 1986). Overall accuracy is computed by summing up
the major diagonal of the error matrix (i.e., the correctly mapped
n+ j = n
i =1
ij (29.8)
sample units) and dividing by the total number of sample units
in the matrix. While this statistic is perhaps the most commonly be the number of samples classified into thematic map class j
reported measure of thematic map accuracy, this value alone is in the reference data set.
not sufficient. It is vital that the error matrix be reported so that Then, the overall accuracy between map and the reference
other measures of accuracy can be computed and that the errors data can then be computed as follows:
in the map can be clearly seen and understood.
k
Overall accuracy represents the accuracy of the entire map. nii
Overall accuracy = i =1
(29.9)
However, in addition, we may want to know the accuracy of an n
individual thematic map class. Computation of individual class
accuracies is a little more complicated than overall accuracy and Producers accuracy can be computed by
requires two measures: producers accuracy and users accuracy
(Story and Congalton, 1986). The producer of the map may want n jj
Producers accuracy j = (29.10)
to know how well they mapped a certain thematic map class, n+ j
the producers accuracy. This value is computed by dividing the
value from the major diagonal (the agreement) for that class by and the users accuracy can be computed by
the total number of samples in that map class as indicated by the
nii
sum of the reference data for that class. Looking at Figure 29.7 Users accuracy i = (29.11)
shows that the map producer called 55 areas forest, while the ref- ni +
erence data indicate that there were a total of 65 forest areas. Ten
areas were omitted from the forest, and of these, six were com-
29.4.4 Basic Analysis techniques
mitted to other vegetation and four were committed to water. So,
55/65 samples were correctly called forest for a forest producers After computing the descriptive statistics from the error matrix,
accuracy of 85%. However, this is only half the story. If you now some additional analysis techniques can also be implemented
observe the users perspective of the map, you see once again tolearn even more about the results of the accuracy assessment.
2016 by Taylor & Francis Group, LLC
596 Remotely Sensed Data Characterization, Classification, and Accuracies
Reference data Reference data Row
F OV D W F OV D W total
D 0.7908 0.1187 0.0579 0.0334 F 72 9 5 1 87
Land cover categories
Map OV 7 83 6 4 100 F = Forest
Map OV 0.1194 0.6850 0.0912 0.1045 OV = Other vegetation
D 2 4 69 6 81
D = developed
W 5 8 1 74 88 W = water
D 0.0079 0.0975 0.7935 0.1008
Column Overall accuracy =
86 104 81 85 356
W 0.0819 0.0988 0.0575 0.7613 total (72 + 83 + 69 + 74)/356 =
298/356 = 84%
3.0306 Producers accuracy Users accuracy
F = 72/86 = 84% F = 72/87 = 83%
Land cover categories
Normalized accuracy = OV = 83/104 = 80% OV = 83/100 = 83%
F = Forest
0.7908 + 0.6850 + 0.7935 + 0.7613 = D = 69/81 = 85% D = 69/81 = 85%
OV = Other vegetation
3.0306/4.0 = 76% W = 74/85 = 87% W = 74/88 = 84%
D = developed
W = water
FIg u r e 29.9 An error matrix from the same study area as Figure 29.7,
but the analyst selected more sampling units for accuracy assessment.
FIg u r e 29.8 Error matrix showing the results of the Margfit analy-
sis in Figure 29.7. Reference data
F OV D W
The first of these techniques is called Margfit and is used to
normalize the error matrix to remove the effect of sample size D 0.8265 0.0909 0.0641 0.0180
for comparison with other error matrices. Margfit uses a stan-
dard iterative proportional fitting algorithm that normalizes
OV 0.0843 0.7872 0.0747 0.0532
the matrix by forcing each row and column total (i.e., the mar- Map
ginal) to sum to 1. The result is a normalized matrix in which
differences in sample sizes used to generate different matrices D 0.0297 0.0449 0.8446 0.0813
are eliminated and the individual cell values in the matrix are
directly comparable with another matrix. Figure 29.8 shows the W 0.0595 0.0771 0.0166 0.8475
results of the Margfit analysis on Figure 29.7. Notice that each
value in the matrix is now a percentage of 1, and multiplying
each value by 100 would result in the accuracy for each cell. The 3.3058
major diagonal values represent the accuracy of each thematic
Land cover categories
map class and are a combination of the individual producers and
F = Forest Normalized accuracy =
users accuracies for each class. For example, the accuracy of the 0.8265 + 0.7872 + 0.8446 + 0..8475 =
OV = Other vegetation
forest class is shown as 0.7908 or 78%. Finally, as in computing 3..3058/4.0 = 83%
D = Developed
overall accuracy, the normalized accuracy of the error matrix
W = Water
can be computed by summing the values in the major diagonal
and dividing by 4 (the number of map classes and also the sum FIg u r e 29.10 Error matrix showing the results of the Margfit analy-
of all values in the matrix). The normalized accuracy of Figure sis in Figure 29.9.
29.8 is 76%, while the overall accuracy is 77% (see Figure29.7).
In order to fully appreciate the usefulness of the Margfit anal- Because the original error matrices (Figures 29.7 and 29.9)
ysis, it is necessary to have a second error matrix to compare to have been normalized (Figures 29.8 and 29.10), the values in the
the first. Figure 29.9 presents an error matrix for the same study normalized error matrices can be directly compared. For exam-
area as was assessed in Figure 29.7. However, the analyst collected ple, comparing the 55 correct sample units in Figure 29.7 for the
more accuracy assessment sample units from which to assess forest class with the 72 correct sample units in Figure 29.9 for
the quality of the map. Overall, producers, and users accura- the forest class would not have much meaning since the matrices
cies were computed as described and shown in Figure 29.9. Note were generated using different total sample sizes (291 and 356,
that 356 sample units were used for this assessment, while only respectively). However, with Margfit, the matrices are normalized
291sample units were selected for the assessment in Figure 29.7. so that these values are now directly comparable with Figure 29.8
The overall accuracy of the error matrix in Figure 29.9 is 84%. having a forest accuracy of 79% and Figure 29.10 having a forest
The results of the Margfit analysis on Figure 29.9 are shown in accuracy of 83%. It seems that the analyst who created the map
Figure 29.10. Again, the individual cell values represent thematic assessed using the error matrix in Figure 29.9 used a method that
class accuracies, and the normalized accuracy is 83%. produced better forest class accuracy. This normalization process
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 597
produces a single accuracy value for each of the map classes, While the Kappa statistic (KHAT) has been used extensively
which is a combination of the producers and users accuracies for as a measure of accuracy, a number of researchers have pointed
that class. In this way, a single number rather than two separate out the shortcomings of using it as such (Zhenkui and Redmond,
values can be used to represent a class accuracy. 1995; Naesset, 1996; Pontius and Millones, 2011). However, the
A second technique that has often been used in accuracy major contribution of the Kappa analysis is not as a measure
assessment is called Kappa (Cohen, 1960). The results of per- of accuracy, but rather as a technique used to determine if one
forming a Kappa analysis is a KHAT statistic (an estimate of error matrix is statistically significant from another. Such a test
Kappa), which like overall accuracy and normalized accuracy using a matrix derived from one classification approach could
can be used as another measure of agreement or accuracy. The then determine if that matrix is statistically significantly differ-
KHAT statistic is computed as ent than a matrix derived from a second classification approach.
In other words, is technique A statistically significantly bet-
ter than technique B. Actually, two tests are possible with the
r r
N x ii (x i + x +1)
K = i =1 i =1
(29.12) Kappa analysis. The first is to test if a single error matrix is sig-
r
N2 (x i + x +1) nificantly better than random. In other words, is the map better
i =1
than a random assignment of pixels (hopefully it is)? The KHAT
where statistic is asymptotically normally distributed, and therefore,
r is the number of rows in the matrix confidence intervals can be calculated using the approximate
xii is the number of observations in row i and column i large sample variance (Congalton and Green, 1999). The test
xi+ and x+i are the marginal totals of row i and column i, to determine if a classification is better than random uses the
respectively Ztest as follows:
N is the total number of observations (Bishop etal., 1975)
KHAT
Z= (29.13)
The equations necessary for computing the variance of the KHAT var ( KHAT )
statistic and the standard normal deviate can be found in many
publications including Congalton et al. (1983), Rosenfield and
Fitzpatrick-Lins (1986), Hudson and Ramm (1987), Congalton where Z is standardized and normally distributed.
(1991), and Congalton and Green (1999, 2009). Table 29.3 shows The test to determine if two error matrices are significantly
three accuracy measures that can be computed easily from the different from one another uses the following Z test:
error matrix. These measures include overall accuracy computed
from the original matrix, normalized accuracy computed from KHAT1 KHAT2
the normalized matrix, and kappa. Which measure or measures Z= (29.14)
var(KHAT1) + var(KHAT
2)
to use for expressing thematic map accuracy is open for discus-
sion. Each brings a different amount of information into the
calculation. Overall accuracy simply sums the major diagonal where
and divides by the total number of sampling units. Normalized KHAT1 represents the KHAT statistic from one error matrix
accuracy directly incorporates the off-diagonal elements of the KHAT2 is the statistic from the other matrix
matrix (i.e., the errors) through the normalization (i.e., itera-
tive proportional fitting) process. Kappa indirectly incorporates Running the Kappa test of significance on the matrix in
the errors by using the sums of the row and column totals in Figure 29.7 produces a Z value of 20.8. At the 95% confidence
the computation of the statistic. Therefore, each measure rep- level, at Z value above 1.96 is considered significant so it can
resents different information and should be evaluated accord- be concluded that the process that went into making this map
ingly. Landis and Koch (1977) proposed certain range values is significantly better than a random distribution of thematic
of the Kappa statistic to represent different levels of agreement. labels to the pixels. The same can be said for the matrix in
A value greater than 80% represents high agreement, while a Figure 29.9 with a Z value of 29.9. However, the most valuable
range between 60% and 80% represents moderate agreement question is whether these two matrices are statistically signifi-
and values below 40% represent weak agreement. Presenting the cantly different than the other. The resulting test statistic for this
full error matrix allows the analyst to compute any or all of these Kappa analysis is 2.2 again indicating statistical significance.
three measures of thematic accuracy. Therefore, the matrices are significantly different and look-
ing at the accuracy values from Table 29.3 shows that matrix 2
TABLe 2 9.3 Three Measures of Thematic Map Accuracy (Figure 29.9) is the better of the two. If matrix 2 (i.e., the map
Matrix #1 Matrix #2 assessed in Figure 29.9) was produced with a different classi-
fication algorithm than matrix 1, then it can be said that the
Overall accuracy (%) 77 84
algorithm is better. This ability to determine if one matrix, and
Normalized accuracy (%) 76 83
therefore map, is statistically significantly better than another is
Kappa (%) 69 78
the real power of the Kappa analysis.
2016 by Taylor & Francis Group, LLC
598 Remotely Sensed Data Characterization, Classification, and Accuracies
29.4.5 Advanced Analysis techniques methods employed to collect the training and reference data,
some acknowledgment of the potential fuzziness in the map is
In addition to the basic analysis techniques for thematic map justified.
accuracy assessment described earlier, there are two other more Green and Congalton (2004) and more fully described in
advanced techniques that must be seriously considered depend- Congalton and Green (2009) have proposed the modification
ing on the methodologies used to make the map. These tech- of the standard error matrix to incorporate fuzziness into the
niques include fuzzy accuracy assessment and area-based error assessment process. In this case, the matrix consists of the cor-
matrix generation. rect values, the acceptable values, and the unacceptable values.
Use of the fuzzy error matrix is extremely powerful as it incorpo-
29.4.5.1 Fuzzy Accuracy Assessment rates all the benefits of the standard error matrix while allowing
One of the four characteristics of a good classification scheme is for situations where classification scheme breaks are artificial
that it is mutually exclusive. That is, any area on the map should points along a continuum and/or where interpreter variability is
fall into one and only one map class. While this is a valid goal for difficult to control. The best way to understand the fuzzy error
any scheme, in reality, there are always issues with achieving this matrix is to provide an example as shown in Figure 29.11.
goal. Gopal and Woodcock (1994) proposed the use of fuzzy set A quick examination of Figure 29.11 reveals that it is the same
theory in thematic accuracy assessment to recognize the possi- error matrix as in Figure 29.7 with the exception that there are
bility that ambiguity exists in determining the appropriate map now two values in the off-diagonal elements of the matrix. The
class. They further state that it is rare to find the situation where values in the major diagonal still represent the best determi-
one map class is exactly right and all the other map classes are nation of the correct classification. For example, there are 55
equally wrong. sample units that the map said were the forest class and the refer-
There are many places that fuzziness can enter into a the- ence data agreed. However, the off-diagonal elements now have
matic map assessment. The classification scheme itself is a large two values: the first value represents acceptable matches while
source as many map classes are more a continuum that has been the second value represents the errors. Looking at Figure 29.11
grouped into a single class. For example, mapping hardwood shows that the map labeled an area as forest when it was really
forest vs. conifer forest is easy if the forest is 100% hardwoods other vegetation a total of 11 times. However, four times this
or 100% conifer. However, as the percentages of each get closer label is considered acceptable because the area, when visited to
to 50%, it becomes harder to decide which type of forest it is. collect the reference data, was with actually young trees that had
Human interpretation also is an important source of variation. not yet grown tall enough to meet the definition of forest, which
In many cases, training data used in the classification process stated that the trees must be 3 m tall to be considered a forest.
and accuracy assessment data used in the validation process are The other seven times, the other vegetation was actually other
generated from visual interpretation of high-resolution imag- vegetation (i.e., not trees), and the map is in error. Fuzziness
ery. Even visiting field sample sites on the ground is not without comes into play in the matrix only where appropriate. It is not an
variation as one observer may determine the boundary between excuse to accept errors, but rather a mechanism to account for
classes is in one place and another observer put it in another. variability and is especially helpful with the reference data. Note
Therefore, depending on the classification scheme used and the that no confusion between the forest class and the developed
Users accuracies
Reference data Deterministic Deterministic Fuzzy Fuzzy
F OV D W totals accuracies totals accuracies
F 55 4, 7 0,4 0,3 55/73 75% 59/73 81%
OV 2,4 51 0,5 2,6 51/70 73% 55/70 79%
Map
D 0,0 0,8 55 0,9 55/72 76% 55/72 76%
W 1,3 0,7 0,3 62 62/76 82% 63/76 83%
Producers accuracies
Deterministic Overall accuracies
55/65 51/77 55/67 62/82
totals Deterministic 223/291 = 77%
Deterministic 85% 66% 82% 75% Fuzzy 232/291 = 80%
accuracies
Fuzzy totals 58/65 55/77 55/67 64/82
Fuzzy
89% 71% 82% 78%
accuracies
Land cover categories F = Forest OV= Other vegetation D= developed W= water
FIg u r e 2 9.11 Deterministic and fuzzy error matrix example.
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 599
class is considered acceptable. Instead, the first value in the off- work equally well if the collector was in the field making mea-
diagonal is zero, and the second value shows the errors (4). So, surements or observations from the center of the sampling unit
what about the other vegetation/water confusion? Again, look- or if they were interpreting some imagery to label the reference
ing at the fuzzy error matrix shows that twice the map said the data. Clearly, making measurements in the field to label a certain
sample unit was other vegetation when the reference data said sample unit should result in less fuzziness (i.e., fewer acceptable
that it was water. The fuzziness here can be explained in that labels included) than reference data collected by the interpreta-
there were lily pads and other green vegetation growing on the tion of some medium-scale imagery. Regardless of the reference
surface of a pond. The pond is clearly water, but it is acceptable in data collection approach, producing a fuzzy error matrix with
this case to call it other vegetation because of what was floating its counterpart, the deterministic error matrix does provide an
on the surface. effective way of compensating for fuzziness in the classification
This new error matrix (Figure 29.11) is really two error matri- process while also giving the information contained within the
ces in one. It is the original error matrix (now called the deter- traditional error matrix.
ministic error matrix) and it is also a fuzzy error matrix. The
difference is simply how one handles the two values in the off- 29.4.5.2 Area-Based error Matrix Generation
diagonals. In the case of the deterministic error matrix, only the Historically, most land cover/vegetation/thematic maps created
values in the cells of the major diagonal are considered correct. from digital imagery have analyzed the individual image pixels.
Therefore, the deterministic overall accuracy, the deterministic More recently, a new method called object-based image analysis
row and column totals, and the deterministic producers and (OBIA) has become increasingly popular. In OBIA, pixels are
users accuracies are identical in every way to the calculations grouped together into segments or objects and classified together
shown in Figure 29.7. However, the fuzzy error matrix calcula- instead of as individual pixels. This approach more closely
tions are a little bit different as they include the acceptable val- mimics human interpretation and incorporates significant addi-
ues (represented by the first value in the off-diagonal cells) along tional information about the object than was possible for the
with the correct values (represented by the cells on the major individual pixel (Blaschke, 2010). With a pixel-based map, it is
diagonal). Therefore, the fuzzy producers accuracy for the forest very appropriate to take a cluster of pixels as the sample unit for
class is 58/65 = 89% where the 58 is equal to the 55 correct values the reference data collection used in the error matrix generation.
plus 2 acceptables from the other vegetation class and 1 accept- As previously described in this chapter, for medium-resolution
able from the water class. The fuzzy overall accuracy is 232/291= imagery such as Landsat, a 3 3 cluster of pixels is used as a
80%, which includes summing the values on the major diago- single reference sample unit. Therefore, since all the sample units
nal plus all the acceptables (first values in the off-diagonal cells). are exactly the same size, the error matrix is generated by simply
That is adding 55 + 51 + 55 + 62 + 2 + 1 + 4 + 2 = 232. tallying each sample unit as either correct (on the major diagonal
In some cases, the difference between the deterministic error of the error matrix) or incorrect (in the appropriate off-diagonal
matrix and the fuzzy error matrix can be quite small. This situ- cell). However, for assessing the accuracy of maps created using
ation would occur when the classification scheme is rather gen- OBIA, objects of various sizes have been created to replace the
eral, and there is little variation between thematic map classes. equal-area pixels. It is possible then to use these objects as the
In other situations, the differences in the matrices can be quite reference sampling unit (i.e., polygons). If this is done, the sam-
large especially if the classification scheme is quite complex with pling units are no longer equal areas and therefore cannot simply
great variability among the classes. Another situation in which be tallied in the error matrix (Radoux etal., 2010; MacLean and
the fuzzy error matrix can be very important is when the only Congalton, 2012). Instead, the actual area of each of the refer-
reference data available are difficult to interpret causing great ence sampling units must be incorporated into the error matrix,
variation in the reference data. For example, using medium- and the matrix becomes an area-based error matrix. In the tradi-
scale aerial photographs (i.e., 1:25,000) as the reference data tional error matrix, each reference sample unit has equal weight
source for assessing a map generated from Landsat Thematic since each has the same area. In the area-based error matrix,
Mapper imagery using a classification scheme that has over 20 each reference sample unit is weighted by the area of the sample.
land cover types. In this case and because only the photos could Creating this area-based error matrix is no more difficult than
be used without any ground visits, the interpreters would likely the tally method used to create the traditional matrix accept
select a best class label and then perhaps one to three acceptable that the areas are input into the matrix instead of just a tally.
labels for each of the reference data sample units. MacLean and Congalton (2012) present a comparison of these
An effective way to implement this fuzzy accuracy assess- methods. Radoux etal. (2010) not only document these methods,
ment method is through the use of a reference data collection but also show that some sampling efficiencies can be gained by
form. All the thematic map classes can be listed on the form, knowing the sizes of all the objects in the map and using this
and the reference data collector can then indicate which map information to reduce the number of sample units needed to cre-
class they believe is the correct one. The collector would also ate the matrix.
have the option then of indicating if any of the other map classes It should be noted that area-weighted error matrices are pos-
would be an acceptable label for that particular sampling unit sible even with pixel-based classifications if different-sized clus-
and also which would be not acceptable. This approach would ters of pixels are selected for the reference sampling units. Also,
2016 by Taylor & Francis Group, LLC
600 Remotely Sensed Data Characterization, Classification, and Accuracies
it is possible to use a cluster-based reference sampling unit for Cliff, A.D. and J.K. Ord. 1973. Spatial Autocorrelation. Pion
OBIA-based maps. However, the unequal area of the objects Limited, London, U.K., 178pp.
in an OBIA-based map more readily leads one to conduct an Cohen, J. 1960. A coefficient of agreement for nominal scales.
area-based accuracy assessment and, therefore, this technique is Educational and Psychological Measurement, 20(1), 3740.
described in this part of the chapter. Colwell, R. N. 1955. The PI picture in 1955. Photogrammetric
Engineering, 21(5), 720724.
Congalton, R. G. 1988a. Using spatial autocorrelation analysis
29.5 conclusions to explore errors in maps generated from remotely sensed
data. Photogrammetric Engineering and Remote Sensing,
This chapter has presented a discussion of the many consider-
54(5), 587592.
ations necessary to conduct a valid accuracy assessment. Both
Congalton, R. G. 1988b. A comparison of sampling schemes used
positional error and thematic error must be carefully consid-
in generating error matrices for assessing the accuracy of
ered as an error if one can result in an error in the other. While
maps generated from remotely sensed data. Photogrammetric
there is no definitive step-by-step method for conducting a
Engineering and Remote Sensing, 54(5), 593600.
valid assessment, it is clear that the assessment must be care-
Congalton, R. 1991. A review of assessing the accuracy of clas-
fully thought out and planned from the very beginning of the
sifications of remotely sensed data. Remote Sensing of
mapping project. All decisions regarding how the assessment is
Environment, 37, 3546.
conducted must be thoroughly documented so that the process
Congalton, R. 2001. Accuracy assessment and validation of remotely
can be easily understood and reproduced by anyone interested
sensed and other spatial information. The International
in the map.
Journal of Wildland Fire, 10, 321328.
Accuracy assessment is an essential component of any map-
Congalton, R. 2009. Accuracy and error analysis of global and
ping project. However, accuracy assessment is also an extremely
local maps: Lessons learned and future considerations.
expensive component. Therefore, every effort must be made to
In: Remote Sensing of Global Croplands for Food Security,
balance statistical validity with what is practically attainable.
P. Thenkabail, J. Lyon, H. Turral, and C. Biradar (eds.).
Improper assessments can cost the entire budget for the proj-
CRC/Taylor & Francis, Boca Raton, FL, pp. 441458.
ect. Poorly designed assessments can turn out to be statisti-
Congalton, R. and M. Brennan. 1999. Error in remotely sensed
cally invalid causing the assessment to not be worth the effort.
data analysis: Evaluation and reduction. Proceedings of
However, with care, proper consideration, and wise planning,
the Sixty Fifth Annual Meeting of the American Society
accuracy assessment can be an extremely satisfying part of the
of Photogrammetry and Remote Sensing. Portland, OR,
project that not only shows the accuracy of the map, but also
pp.729732 (CD-ROM).
indicates where the map can be improved in the future.
Congalton, R. and K. Green. 1999. Assessing the Accuracy of
Remotely Sensed Data: Principles and Practices. CRC/Lewis
References Press, Boca Raton, FL, 137pp.
Congalton, R. and K. Green. 2009. Assessing the Accuracy of
Ager, T. March, 2004. An Analysis of Metric Accuracy Definitions Remotely Sensed Data: Principles and Practices, 2nd edn.
and Methods of Computation. Unpublished memo prepared CRC/Taylor & Francis, Boca Raton, FL, 183pp.
for the National Geospatial-Intelligence Agency. InnoVision, Congalton, R. G., R. G. Oderwald, and R. A. Mead. 1983. Assessing
Washington, DC. Landsat classification accuracy using discrete multivariate
Aronoff, S. 1985. The minimum accuracy value as an index of statistical techniques. Photogrammetric Engineering and
classification accuracy. Photogrammetric Engineering and Remote Sensing, 49(12), 16711678.
Remote Sensing, 51(1), 99111. Federal Emergency Management Agency (FEMA). 2003. Guidelines
ASPRS. 1989. ASPRS interim accuracy standards for large-scale and Specifications for Flood Hazard Mapping Partners. http://
maps. Photogrammetric Engineering and Remote Sensing, www.fema.gov/fhm/dl_cgs.shtm. Accessed 3/25/15.
54(7), 10381041. Federal Geographic Data Committee, Subcommittee for Base
ASPRS. May 24, 2004. ASPRS Guidelines, Vertical Accuracy Cartographic Data. 1998. Geospatial Positioning Accuracy
Reporting for LiDAR Data. American Society for Standards. Part 3: National Standard for Spatial Data
Photogrammetry and Remote Sensing, Bethesda, MD. Accuracy. FGDC-STD-007.3-1998, Federal Geographic Data
Bishop, Y., S. Fienberg, and P. Holland. 1975. Discrete Multivariate Committee, Washington, DC, 24pp.
Analysis: Theory and Practice. MIT Press, Cambridge, MA, Ginevan, M. E. 1979. Testing land-use map accuracy: Another
575pp. look. Photogrammetric Engineering and Remote Sensing,
Blaschke, T. 2010. Object based image analysis for remote sens- 45(10), 13711377.
ing. ISPRS Journal of Photogrammetry and Remote Sensing, Gopal, S. and C. Woodcock. 1994. Theory and methods for
65, 216. accuracy assessment of thematic maps using fuzzy sets.
Bolstad, P. 2005. GIS Fundamentals, 2nd edn. Eider Press, White Photogrammetric Engineering and Remote Sensing, 60(2),
Bear Lake, MN, 543pp. 181188.
2016 by Taylor & Francis Group, LLC
Assessing Positional and Thematic Accuracies of Maps Generated from Remotely Sensed Data 601
Green, K. and R. Congalton. 2004. An error matrix approach to NDEP. May 10, 2004. Guidelines for Digital Elevation Data. Vertion
fuzzy accuracy assessment: The NIMA Geocover project. A 1.0. National Digital Elevation Program, Washington, DC.
peer-reviewed chapter. In: Remote Sensing and GIS Accuracy Pontius, G. and M. Millones. 2011. Death to kappa: Birth of quan-
Assessment, R. S. Lunetta and J. G. Lyon (eds.). CRC Press, tity disagreement and allocation disagreement for accuracy
Boca Raton, FL, 304pp. assessment. International Journal of Remote Sensing, 32(15),
Greenwalt, C. and S. Melvin. 1962 and 1968. Principles of Error 44074429.
Theory and Cartographic Applications, ACIC Technical Pugh, S. and R. Congalton. 2001. Applying spatial autocorrela-
Report Number 96. United States Air Force, Aeronautical tion analysis to evaluate error in New England forest cover
Chart and Information Center, St. Louis, MO, 60pp. Plus type maps derived from Landsat Thematic Mapper Data.
appendices. This report is cited in the ASPRS standards as Photogrammetric Engineering and Remote Sensing, 67(5),
ACIC, 1962. 613620.
Hay, A. M. 1979. Sampling designs to test land-use map accuracy. Radoux, J., R. Bogaert, D. Fasbender, and P. Defourny. 2010.
Photogrammetric Engineering and Remote Sensing, 45(4), Thematic accuracy assessment of geographic object-based
529533. image classification. International Journal of Geographical
Hord, R. M. and W. Brooner. 1976. Land-use map accuracy Information Science, 25(6), 895911.
criteria. Photogrammetric Engineering and Remote Sensing, Rosenfield, G. H. and K. Fitzpatrick-Lins. 1986. A coefficient of
42(5), 671677. agreement as a measure of thematic classification accuracy.
Hudson, W. and C. Ramm. 1987. Correct formulation of the kappa Photogrammetric Engineering and Remote Sensing, 52(2),
coefficient of agreement. Photogrammetric Engineering and 223227.
Remote Sensing, 53(4), 421422. Rosenfield, G. H., K. Fitzpatrick-Lins, and H. Ling. 1982.
Katz, A. H. 1952. Photogrammetry needs statistics. Photogrammetric Sampling for thematic map accuracy testing. Photogrammetric
Engineering, 18(3), 536542. Engineering and Remote Sensing, 48(1), 131137.
Landis, J. and G. Koch. 1977. The measurement of observer agree- Sammi, J. C. 1950. The application of statistics to photogramme-
ment for categorical data. Biometrics, 33, 159174. try. Photogrammetric Engineering, 16(5), 681685.
Lunetta, R., R. Congalton, L. Fenstermaker, J. Jensen, K. McGwire, Story, M. and Congalton, R. 1986. Accuracy assessment: A users
and L. Tinney. 1991. Remote sensing and geographic infor- perspective. Photogrammetric Engineering and Remote
mation system data integration: Error sources and research Sensing, 52(3), 397399.
issues. Photogrammetric Engineering and Remote Sensing, Tortora, R. 1978. A note on sample size estimation for multino-
57(6), 677687. mial populations. The American Statistician, 32(3), 100102.
MacLean, M., M. Campbell, D. Maynard, M. Ducey, and R. United States Bureau of the Budget. 1941/1947. National Map
Congalton. 2013. Requirements for labeling forest poly- Accuracy Standards. Washington, DC. http://nationalmap.
gons in an object-based image analysis classification. gov/standards/pdf/NMAS647.PDF. Accessed March 25,
International Journal of Remote Sensing, 34(7), 25312547. 2015.
MacLean, M. and R. Congalton. 2012. Map accuracy assessment van Genderen, J. L. and B. F. Lock. 1977. Testing land use map
issues when using an object-oriented approach. Proceedings of accuracy. Photogrammetric Engineering and Remote Sensing,
the Annual Meeting of the American Society of Photogrammetry 43(9), 11351137.
and Remote Sensing. Sacramento, CA, 5pp. (CD-ROM). Young, H. E. 1955. The need for quantitative evaluation of the
MacLean, M. and R. Congalton. 2013. Applicability of multi-date photo interpretation system. Photogrammetric Engineering,
land cover mapping using Landsat 5 TM imagery in the 21(5), 712714.
Northeastern US. Photogrammetric Engineering and Remote Young, H. E. and E. G. Stoeckler. 1956. Quantitative evalua-
Sensing, 79(4), 359368. tion of photo interpretation mapping. Photogrammetric
Naesset, E. 1996. Conditional tau coefficient for assessment of Engineering, 22(1), 137143.
producers accuracy of classified remotely sensed data. Zhenkui, M. and R. Redmond. 1995. Tau coefficients for accu-
ISPRS Journal of Photogrammetry and Remote Sensing, racy assessment of remote sensing data. Photogrammetric
51(2), 9198. Engineering and Remote Sensing, 61(4), 435439.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
XIII
Space Law and
Remote Sensing
30 r emote Sensing Law: An Overview of Its Development and Its Trajectory in the g lobal Context P.J. Blount ....605
Acronyms and Definitions Introduction Four Pillars of Remote Sensing Law Conclusion References
603
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
30
Remote Sensing Law:
An Overview of Its
Development and Its Trajectory
in the Global Context
Acronyms and Definitions ................................................................................................................ 605
30.1 Introduction ............................................................................................................................ 605
30.2 Four Pillars of Remote Sensing Law .................................................................................... 606
Space Treaty Regime Disarmament and Verification Law Remote Sensing
Principles Geospatial Law
P.J. Blount
University of Mississippi 30.3 Conclusion ................................................................................................................................617
School of Law References.............................................................................................................................................617
Acronyms and Definitions each emerged, though it will not attempt to identify the spe-
cific temporal locations due to the fluidity and subjectivity
CD Conference on Disarmament of such assignments. The chapter begins with a brief exami-
EO Earth observation nation of relevant international space law as developed in the
GNSS Global Navigation Satellite System space treaty regime. Next, it will turn to remote sensing law as
INPE National Institute for Space Research (Brazil) developed in relation to military uses with a particular focus
NTM National Technical Means on disarmament treaty verification. Then, it will address inter-
UNCOPUOS United Nations Committee on the Peaceful national remote sensing law applicable to civil remote sens-
Uses of Outer Space ing satellites as articulated in the United Nations Principles
UNGA United Nations General Assembly on Remote Sensing (UNGA Res. 41/65, 1986). Finally, it will
argue that remote sensing law is increasingly being subsumed
30.1 introduction into the developing field of geospatial law, which is resulting
from the processes of commercialization, globalization, and
The law relating to remote sensing is a complex mix of interna- technological convergence. The purpose of this exercise will
tional and national laws, regulations, policies, and agreements. be to trace a trend of increasing adoption of domestic law
This body of remote sensing law has numerous facets that regu- regimes that intersect with international governance systems
late a number of uses of remote sensing satellites, related tech- (Gabrynowicz, 2007, 7). These laws are in a constant state of
nology, and collected data. This chapter will serve as a survey of flux as a result of rapid advancements in technology and differ-
the laws and regulations that governs remote sensing activities ent perceptions on how we manage space data gathering that is
across a variety of applications and a variety of legal spaces. agreeable to all.
This overview of remote sensing law will assert that there The thrust of this chapter will be to address laws and regula-
are essentially four bodies of law that govern remote sensing tions specific to remote sensing activities. Many of the provisions
and define its regulatory regime. These are not distinct bod- of the Outer Space Treaty apply to all space activities including
ies; they overlap and intersect with each other. The typology is remote sensing activities, and reference will be made to these
not intended to obscure these interactions but instead to serve provisions as a whole, but in this chapter, we will endeavor to
as an explanatory framework. This chapter will address each focus our attention on legal provisions that are specific to remote
of these bodies of law chronologically in the order in which sensing activities. The goal will be to highlight the regulations
605
2016 by Taylor & Francis Group, LLC
606 Remotely Sensed Data Characterization, Classification, and Accuracies
that are most relevant to remote sensing activities and the end 30.2.1 Space treaty Regime
uses of remotely sensed data. Finally, it should be noted that
while this survey of remote sensing law seeks to give a broad The core body of international space law is found in four multi-
overview of the law, it is not by any means comprehensive, and lateral treaties (Treaty on Principles Governing the Activities of
the readers attention is drawn to the footnotes for further read- States, 1967; Agreement on the Rescue of Astronauts, 1968;
ings and elaboration. Convention on International Liability, 1972; Convention on
Registration, 1976; on the space-law-making process, see
Dembling and Arons, 1967; Marchisio, 2005). These treaties set
30.2 Four Pillars of Remote Sensing Law the parameters for lawful conduct by states when engaging in
Remote sensing law is a multifaceted legal regime that draws outer space activities. There are several introductions to these
on a number of regulatory systems. This chapter will attempt core treaties; therefore, this chapter will refrain from a full evalua-
to address the four core regulatory systems, namely, traditional tion of these and instead focus on provisions of special concern to
space law, disarmament law, international remote sensing law, remote sensing satellites.*
and geospatial law. These regimes should be understood as sepa- Article I of the Outer Space Treaty grants all states free access
rate regulatory spaces that interact with each other in a variety of to outer space, without interference by other states. By coupling
ways. Indeed, depending on the issue being addressed, they can freedom of access with the principle of nonappropriation found
reorient themselves to the others; it would be folly to suggest that in Article II, which places space outside the sovereign territory
they are always oriented in a specified way. These regimes shift of all states, the negotiators of the Outer Space Treaty enshrined
from discrete and separate to hierarchical to overlapping as the the principle of freedom of overflight by space objects over other
context of their application changes. The best description would states territories (Vsquez, 1965; Jahku, 2003, 7377). This is of
be that their relationship is discursive in nature. It is system of critical importance to remote sensing activities, since it allows
regulatory feedback loops across different governance spaces (see states to observe other states from the nonsovereign vantage
Figure 30.1). For example, international law creates obligations point of space. Since no state has ever protested the overflight of
for states, but state implementation of these obligations often its territory by other states space object, overflight is a custom-
leads to influencing the content of the obligation (Blount, 2012). ary norm that preexists the Outer Space Treaty (Jahku, 2003,
This chapter will proceed by addressing these varying regimes 7374).
as discrete, but will endeavor to highlight the areas wherein we Article VI of the Outer Space Treaty places a duty on states
can observe their interactions. It should not be assumed that at to authorize and supervise nongovernmental actors and
any given moment, one regime maintains primacy over another gives states international responsibility for nongovernmental
or that the delineations among them are clear by any means. actors. This means that many states have adopted full regulatory
Instead, this serves as a framework for understanding the overall regimes in order to manage the risk of commercial actors. These
governance system for remote sensing technologies. regimes, which take the form of licensing regimes, are meant
to ensure government oversight over nongovernmental actors.
A primary concern with many of these licensing regimes is to
ensure that the nongovernmental satellite operators comply with
minimum standards of conduct in its activities, so that the state
avoids situations of international responsibility. These regimes
emphasize cradle to grave operations plans that are meant
Space law to reduce liability issues as well mitigate space debris. These
licenses are often attached to the act of launching. Some states,
however, have licensing regimes specifically for remote sensing
activities that ensure compliance with national security and
global data policies. These regimes will be addressed in Section
Disarmament Remote International RS 30.2.3.1.
law sensing law
law
* For a variety of approaches, see Diederiks-Verschoor (1999), Lyall and
Larsen (2013), Reynolds and Merges (1997), and Lachs (1972, 2010).
For instance, one of the primary goals of the Canadian licensing regime
is to regulate the operation of the satellite itself (Gillon, 2008, 2627).
Geospatial law See also, Mann (2008, 6971).
For instance, the Chinese regime is based on launch licenses (Chinese
Law, 2007).
Canada, Remote Sensing Space Systems Act, Gillon, Regulating
Remote Sensing, 26 (noting a second concern of the Canadian licensing
regime is the distribution of raw data and remote sensing products pro-
FIg u r e 30.1 Remote sensing law is the confluence of multiple legal duced by satellites); Germany, Satellite Data Security Act (2008); and
regimes. the United States, Licensing of Private Remote Sensing Systems (2013).
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 607
There are numerous information-sharing clauses in the peaceful activities when they deploy and use defensive technolo-
Outer Space Treaty, which read together lead to an overall duty gies in a nonaggressive way; this is supported by the fact that
to warn other states of situations of impending disaster. Most until the launch of Landsat-1, remote sensing was exclusively
relevant is Article IXs requirement that states engaging in developed and used for military purposes (Jahku, 2003, 70).
space activities be guided by the principles of cooperation and This use of remote sensing was critical in the negotiation of
mutual assistance.* While the information-sharing provisions disarmament and arms control agreements between the United
are not directly related to remote sensing, they build strong evi- States and the Soviet Union. [F]reedom of space was perhaps
dence that there is a customary norm of information sharing the principal aim of U.S. space policy throughout the first ten
in a wide variety of contexts as a way to increase cooperation years of the space age (Petras, 2005, 86). The technology allowed
in the use space for the benefit of all mankind. This norm is the two states to overcome the problem of verification. The
further articulated in the Remote Sensing Principles (Section Soviets rejected verification by on-site visits or overflights, and
30.2.3, infra). the United States rejected any agreement without reliable verifi-
Finally, reference should be made to the liability provi- cation measures (Vsquez, 1965, 163164). Once it was clear that
sion found in Article VII of the Outer Space Treaty and fur- both sides were using satellites for remote sensing of the others
ther articulated in the Liability Convention. Whether damage territory, they were able to agree on satellite technology as the
caused by remote sensing data might represent damage caused core method of verification. While some states have chal-
by [a states] space object is a contested interpretation of the lenged the military and intelligence remote sensing regimes as
treaty. However, this author would argue that it is an increas- nonpeaceful, the international acceptance of the legitimacy of
ingly irrelevant debate. While parsing the language of the treaty intelligence gathering from outer space was effectively secured
is an excellent exercise, state practice is a better touchstone for with the enshrinement of space-based photoreconnaissance
determining the practicality of the argument. State reluctance to in the Cold War strategic arms control regime (Petras, 2005,
apply the Liability Convention in clear cases such as Cosmos 954 8990). This was made possible by a legal structure, articulated
and the Cosmos-Iridium collision points to the likelihood that in the Outer Space Treaty and embodied in customary interna-
damage as a result of data presents a weak case for the applica- tional law, that ensured a nearly complete freedom of access and
tion of the Liability Convention (see generally, Listner, 2012). In use outside of the earths atmosphere. As a result, there was no
this case, state practice indicates that such damage would most way either state could object to the use of satellites to sense its
likely sit outside the scope of damage defined in the Outer Space territory, because such activity was consistent with international
Treaty and the Liability Convention, though this is admittedly law (Petras, 2005, 87). These technologies were indicated in dis-
an argument based less on strict legal analysis and more on a armament agreements under the term National Technical Means
constructivist view of norm creation and definition, in which (NTM),** a phrase that would remain officially undefined until
practice identifies the content of the norm. its meaning was revealed in a speech by president Carter (1978).
The use of satellites as the central method of verification has
implications for space security, since treaties that use NTM
30.2.2 Disarmament and Verification Law
often require that parties to the treaty not interfere with the
The space race and the resulting international law of outer space others NTM (Diederiks-Verschoor, 1999, 85; Petras, 2005, 91).
are a product of the Cold War; the laws early development was Arguably, this helped to solidify space security by creating a
linked to both technological diplomacy and military uses (see de facto drawdown of space weaponry development, due to the
Gabrynowicz, 2004). The launch systems represented ballis- risk that any interference with space objects could interfere with
tic missile technology; the communications applications could NTM and thereby offset the strategic stability these agreements
support the emerging concept of a globally projected military; provided (Harrison, 9). It can be argued that this limitation is
and remote sensing could open up intelligence secrets about the becoming less relevant as space is populated by an increasingly
other sides weapons. While one of the core guiding principles diverse set of actors.
and central legal obligations in space law is the use of space for While there has been scholarship on expanding the verifi-
peaceful purposes, it is well established that this allows states to cation possibilities of remote sensing satellites to other areas,
use space technologies in accordance with peaceful as set out in little movement has been made. France made a proposal at
the UN Charter as a de minimis threshold (Petras, 2005, 8890; the Conference on Disarmament (CD) in 1978 for an inter-
Blount, 2009). As a result, states are considered to be engaging in national satellite body for monitoring disarmament treaties
* Outer Space Treaty, Art. IX; see also Articles III, V, VIII, X, XI, and XII. For a history of such a system, see generally, Richelson (1999).
Ito notes that the Outer Space Treaty and the Liability Convention are ** For example, SALT I & SALT II, Diederiks-Verschoor (1999, 8485);
silent as to the types of damage associated with satellite remote sensing ABM Treaty, Petras (2005, 90); CFE Treaty, Petras (2005, 9293); and
(Ito, 2008, 51). Treaty between the United States of America and the Russian Federation
The author has developed and applied this type of analysis to other areas on Measures for the Further Reduction and Limitation of Strategic
of space law. See generally, Blount (2009, 2011). Offensive Arms (signed April 10, 2010) Art. X(b) and (c).
Legally these opportunities were facilitated by a regime of complete Importantly, Petras notes that NTM includes a wider range of technolo-
freedom (John etal., 2011). gies than just remote sensing satellites (Petras, 2005, 91).
2016 by Taylor & Francis Group, LLC
608 Remotely Sensed Data Characterization, Classification, and Accuracies
(Diederiks-Verschoor, 1999, 84; Achilleas, 2008, 3). This mea- becomes an epistemic unit informing that norms interpreta-
sure was rejected by the CD, and there has been no progress in tion. In other words, when states think they are complying
the CD on such a system. It is unlikely, in the near term, under with a norm, their actions construct the meaning and content
current political conditions, that there is sufficient political will of that norm. Since states may disagree as to the exact content
for such a system to be implemented. There is, however, a great of a norms meaning, customary norms are often better defined
deal of potential for remote sensing technologies to play a role in along a spectrum of acceptable state practice. In this analysis,
the international lawmaking toolkit for its verification and mon- the key question is not if states are complying, but instead, how
itoring capabilities in a variety of areas outside disarmament states are complying. This is because [c]ustom as a source of
such as human rights and the environment (see, for example, international law, leads to the recognition of the legality of the
Onoda, 2005; Alvarado, 2010). existing practice if there is general consent to the observable
rule of conduct (Vereschetin and Danilenko, 1985, 30).
The following inquiry will address the Remote Sensing
30.2.3 Remote Sensing Principles
Principles in two different groupings. First, it will identify prin-
The core of international legal principles dealing with remote ciples that are substantively based in preexisting binding legal
sensing activities is contained in UN General Assembly (UNGA) principles. It will then address the legal innovations of the UN
Resolution 41/65, Principles Relating to Remote Sensing of the Principles, which articulated new content into the international
Earth from Outer Space. These principles are the first and foun- law governing remote sensing. These innovations will be evalu-
dational source of policy guidance for remote sensing activities ated by examining the spectrum of state practice to illustrate
(Uhlir etal., 2009, 209). acceptable state conduct that defines the content of these norms.
The Principles are contained in a UNGA resolution, and as
such, The Principles themselves are not binding as legal prin- 30.2.3.1 Principles Based on Preexisting
ciples.* However, to the extent that they represent customary Legal obligations
international law, these principles are binding. The existence of Numerous of The Principles appeal to preexisting legal obliga-
customary international law is evinced by two criteria: opinio tions that can be found in general international law or in the lex
juris and state practice. Opinio juris represents a subjective belief specialis of international space law. Essentially, these principles
by states that they are legally bound by a rule. In the case of makes explicit the application of more general legal standards
the Remote Sensing Principles, opinio juris is found in the con- explicit that arguably would apply regardless of the acknowledg-
sensus process of negotiation at the United Nations Committee ment of such in The Principles. There are three different group-
on the Peaceful Uses of Outer Space (UNCOPUOS) and by the ings of these: those that implement general international law,
unanimous adoption by the General Assembly. In light of the those that implement international space law, and those that
fact that there are no denunciations of The Principles, that states seek to increase international cooperation.
often appeal to The Principles in their domestic laws, and that
states make claims of compliance with The Principles, a suffi- 30.2.3.1.1 GeneralInternationalLaw:PrinciplesIII,IV,andXV
cient level of acceptance seems to have been achieved to consti-
Principle III makes the application of general international
tute opinio juris.
law to remote sensing activities explicit.** This principle makes
The second evidentiary parameter that a rule constitutes a
specific reference to three treaty instruments: the UN Charter,
binding customary norm requires that states indeed follow the
the Outer Space Treaty, and International Telecommunications
rule they believe to be binding. State practice, essentially, helps
Union Instruments. This is explicit acknowledgment that these
define the content of the norm. If states are found to be claiming
treaties represent preexisting law and that the principles should
compliance and are doing so in good faith, then their compliance
not be read in such a way as to conflict with those instruments.
Principle III also acknowledges that other general international
* On the legal status of the UN Principles, see generally, Terekov (1997,
100101) and Kopal (1988, 1420).
See UNGA Meeting Record no. A/41/PV.95 (1986). Some commenta- So, for instance, nondiscriminatory access is being more narrowly con-
tors argue that this rises to the level of instant custom. For instance, strued as a norm due to state practice that has expanded national secu-
Jackson (1997, 873). Additionally, Hosenball notes that the language rity exceptions over data (Gabrynowicz, 2007, 11).
evinces that states made a commitment to live by these Principles For instance, while all jurisdictions recognize a crime of murder, the
(Gabrynowicz, 2002, 35). elements of the crime changes as jurisdictional borders are crossed. As a
For instance, the UN Principles are the guiding principle for the result, we can say that the prohibition against murder is a general prin-
practice of Earth remote sensing and was, accordingly, considered in ciple, but that the specific content of that principle exists in a spectrum
the preparation of the German satellite data law (Schmidt-Tedd and of meaning. Customary norms suffer from a similar lack of clarity and
Kroyman, 2008, 105). For the United States and Japan as evidence of require a spectrum analysis (if youll allow me the pun). Furthermore,
custom, see Gabrynowicz (2002, 42). The nondiscriminatory access this type of analysis is particularly apt in high-technology cases where
principle has repeatedly been incorporated in the U.S. law and in inter- there is unequal distribution of access coupled with wide acceptance of
national agreements (Gabrynowicz, 2007, 6). Jahku (2003, 8688) notes negotiated rules.
that there is a divide on the issue, but that Principle XII at least creates a ** Compare to the Outer Space Treaty, Article III.
legal obligation. UN Principles, III. The Outer Space Treaty is also referenced in Principle IV.
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 609
law principles apply, so, for instance, the law of armed conflict, observation data could be used to expose the violations that give
private international law principles, and environmental law rise to a legitimate international action. In this capacity, remote
principles could all be applied to space in the proper context. sensing as a global technology can continue to play a role in
While this acknowledgment is important, it is hardly necessary, the way in which the international community constructs the
since The Principles are embodied in a document defined under idea of sovereignty. Indeed, there are current efforts to leverage
international law, and the form reifies the notion that its content remote sensing data to expose gross human rights violations
exists within those parameters. such as Sudan (Satellite Sentinel Project, 2014), Darfur (Eyes on
While referencing a preexisting right to the freedom of use of Darfur, 2014), and North Korea.
outer space, Principle IV specifically notes the that remote sens- Finally, Principle XV states that disputes will be settled in
ing activities shall be accordance with the norm of the peaceful settlement of disputes,
which is previously enshrined in Article 1 of the UN Charter.**
conducted on the basis of respect for the principle of full
and permanent sovereignty of all States and peoples over 30.2.3.1.2 GeneralSpaceLaw:PrinciplesIXandXIV
their own wealth and natural resources, with due regard to
As noted earlier, the Remote Sensing Principles acknowledge the
the rights and interests, in accordance with international
applicability of the Outer Space Treaty to remote sensing activi-
law, of other States and entities under their jurisdiction.
ties, and The Principles reference the treaty throughout. Two
Such activities shall not be conducted in a manner detri-
provisions, though, are concerned explicitly with general space
mental to the legitimate rights and interests of the sensed
law provisions.
State.*
Principle IX affirms the registration obligation found in the
Registration Convention as well as the Article IX obligation to
This principle seeks to reaffirm the freedom data collection
provide information on request of affected states. It lowers the
and distribution, while at the same time ensuring respect of
threshold for requesting such information from the Article IX
the principle of full and permanent sovereignty of all States and
standard of harmful interference to affected states. However,
peoples over their own wealth and natural resources (Achilleas,
the scope of the obligation is likely the same in practice, where
2008, 2). The content of this norm is consistent with the prin-
relevant information will likely be readily available for civil sys-
ciples of nonintervention and sovereign equality that sit at the
tems and scarce for military systems (see Section 30.2.3.2, infra).
heart of the international legal system. The formulation in
Principle XIV incorporates Article VI of the Outer Space
The Principles reflects the concern of developing nations that
Treaty, which requires supervision and authorization of remote
developed nations would use satellite technology to exploit the
sensing activities, which has already been discussed earlier. It
resources of the developing world and advance economic impe-
also acknowledges that in no way does the Article VI obliga-
rialism (Sekhula, 2011, 233). However, as addressed earlier, the
tion preclude the application of the law of state responsibility
right (Ling, 2010, 436437) to overflight had long been estab-
to remote sensing activities (see generally, International Law
lished, and such protestations have never adversely affected
Commission, 2001). Though not explicitly stated in the Outer
operational progress (Diederiks-Verschoor, 1999, 83). Once
Space Treaty, the customary law of state responsibility is cer-
deployed, technology is difficult to withdraw, and as a result, the
tainly applicable to states as they engage in space activities, so
developing nations settled for a compromise in the requirements
this can hardly be seen as an innovation.
for data distribution (see Section 30.2.3.2). This principle can-
not be read as extending or elaborating on the concept of sover-
eignty as already established in international law and, in no way, 30.2.3.1.3 I nternationalCooperation:Principles
requires a state to seek consent or authorization from a state it II,V,VI,VII,VIII,andXIII
intends on sensing (Jahku, 2003, 79). A large portion of international space law is dedicated to pro-
It is worth noting that the intersection of remote sensing moting the soft obligation of international cooperation. It is soft
and sovereignty could become more relevant in the future if because the obligation is always defined by terms to be nego-
the principle of responsibility to protect gains traction as a jus- tiated by the states engaging in the cooperation. Additionally,
tification for military intervention for gross human rights vio- while it can be said that there is an obligation to engage in inter-
lation (UNGA Res. 60/1, 2005). In such scenarios, the earth national cooperation, there is no corresponding affirmative right
* UN Principles, IV. United Nations High Commission on Human RIghts (2014) (relying on
UN Charter, Art. 1 and 2. satellite data as evidence).
Such contentions reflect current critiques of neoliberalism, for instance, ** UN Charter, Art. 1.
see Harvey (2005) (Information technology is the privileged technol- Outer Space Treaty, Art. VII; Registration Convention, Art. II; and
ogy of neoliberalism). UNGA Res. 62/102 (2007). See generally, Schrogl and Hedman (2008).
R2P is an emerging norm that supports broader ability of the inter- It should not be assumed that all states necessarily comply with regis-
national community to intervene when states are committing gross tration requirements, for a remote sensing specific case, see Mayence
human rights violations against their own citizens, such as genocide. Id. (2008, 9294).
para. 138140. Outer Space Treaty, Art. IX.
2016 by Taylor & Francis Group, LLC
610 Remotely Sensed Data Characterization, Classification, and Accuracies
to international cooperation. The following provisions are con- program that the United States implemented, which allowed any
tinuations of this cooperation framework. state to build and receive the raw data from the Landsat system.
Principle II states that remote sensing should be done for the Principles V, VI, VII, VIII, and XIII all incorporate the general
benefit of all nations and taking into particular consideration obligation for international cooperation in space activities and
the needs of the developing countries.* This principle is an can arguably be read more as a list of possible cooperative activi-
extension of Article I of the Outer Space Treaty that states that ties. Principle V makes a general appeal for cooperation, which
space activities shall be carried out for the benefit and in the echoes the principle of cooperation found in the Outer Space
interests of all countries, irrespective of their degree of economic Treatys Article IX. Principle VI puts forth a specific example of
or scientific development. The Remote Sensing Principles add a type of cooperation in the form of data receiving and process-
an additional obligation to take into consideration the special ing facilities within the framework of regional agreements,** and
needs of developing countries. At best though, this obligation Principle VII encourages states to exchange technical assistance
is weak. The Outer Space Law regime puts strong emphasis on on mutually agreed terms. Principle VIII requires the United
cooperative activities in space, but the terms and extent of the Nations to promote international cooperation. Article XIII gives
cooperation are, by design, entirely up to the states involved. sensed states a right to request consultations to seek cooperative
As a result, the addition of the additional clause on develop- opportunities with the sensing state. Each of these principles
ing countries represents the extension of an obligation that is sets forth cooperation in relatively soft terms under which states
more akin to an international policy statement. However, it is can pursue international cooperation as they see fit.
an important statement, since [i]n developing nations these
technologies have proven essential for developing public policies 30.2.3.2 Legal innovations
on issues such as deforestation assessment and management, The following principles represent legal innovations, the content
urban planning, agricultural production, and environmental of which is defined by state practice. These principles added new
assessment (Ferreira and Camara, 2008). substance to the legal framework governing state remote sensing
That is not to say that states willfully ignore this obligation. activities and as such represent legal innovations.
On the contrary, there is abundant evidence that states actively
engage in such cooperation. The need to extend an extra obli- 30.2.3.2.1 Definitions:PrincipleI
gation to developing countries, which are burdened by tech- The Principles begin with a definitions section that serves the
nological lag, has been adopted as a set of principles in UNGA scope of the application of the principles. The principles begin
Resolution 51/122, The Declaration on International Cooperation by defining remote sensing as the sensing of the Earths surface
in the Exploration and Use of Outer Space for the Benefit and from space by making use of the properties of electromagnetic
in the Interest of All States, Taking into Particular Account the waves emitted, reflected or diffracted by the sensed objects, for
Needs of Developing Countries (UNGA Res. 51/122, 1996; see the purpose of improving natural resources management, land
also, Carpanelli and Cohen, 2012). This resolution represents use and the protection of the environment. This is a twofold
international consensus that developing countries should have
access to space technologies, but the principles enunciated give For instance, Iran maintains a receiving station for Landsat data
states broad leeway to determine the extent to which they will (Tarikhi, 2008).
Outer Space Treaty, Art. IX; Diederiks-Vershoor (1999, 82) (Principle V
engage in such cooperation. This certainly makes the terms of
strengthens the other provisions of the Space Treatys Article IX).
this principle very aspirational. ** An example of such cooperation would be the Landsat Ground Station
Despite the aspirational nature, such principles of conduct Operators Working Group, which attempt[s] to address how data
have led to an expansion of space technologies to developing policies should be formulated (Graham and Gabrynowicz, 2002a,
nations, and remote sensing is no longer a technology, which 294295). On ground station policy and operations, see generally,
only superpowers can enjoy; it is now getting diffused among Gabrynowicz (1997, 229).
It should be noted that this consultation request is different in scope
states around the globe, including emerging and developing than the consultations afforded by Article IX of the Outer Space Treaty.
ones (Kuriyama, 2010). Specifically, in the field of remote sens- The Outer Space Treatys consultation provision is triggered by harm-
ing, states have sought to assist developing nations through their ful interference, whereas Principle XIII is triggered by remote sensing
data policies, which will be addressed in depth later (Section activities (as defined in the principles) and an opportunity for coopera-
30.2.3.2.3). Japan, for instance, seeks to make a safe, secure, and tion between the two states. While arguably, this expansion could have
been included in the following innovations section, it is included here
affluent society through the establishment of an effective space-
due to both its softness as an obligation and its strong connection to the
based data use policy (Aoki, 2010, 337). Another mode of assis- use of space for international cooperation. Additionally, tracking com-
tance is allowing direct access to the satellite data via ground pliance and use of this principle is difficult due to the informal nature of
stations. A great example of this is the highly successful Landsat consultations.
UN Principles, I(a). This definition is legal in nature and is lacking as
a technical definition. Vasquez and Lara suggestas a definition to
* UN Principles, II. help bridge the gap between law and policy makers and scientiststhat
Outer Space Treaty, Art. I. remote sensing is a way of obtaining information about an object by
Diederiks-Verschoor (1999, 83) (noting a trend toward increased analyzing the data acquired through a device that is not in physical con-
cooperation). tact with said object (Vsquez and Lara, 2010, 367).
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 611
definition. It incorporates both a technical and intent parameters. Principle X finds support in the general principles of envi-
The first element describes in a technical sense what remote sens- ronmental and human rights law (Carpanelli and Force, 2011).
ing activities will be governed by The Principles (so, for instance, It first requires that states use remote sensing to promote the
GNSS is excluded). The second element, though, limits the scope protection of the environment, which is a fairly broad, yet soft,
of application by making the principles applicable only to activi- requirement. Then, it requires states to share information
ties of a specific nature, which is the political dimension of the that is capable of averting any phenomena harmful to the Earths
definition. Specifically, this definition purposely places military natural environment. The second obligation is surprisingly
remote sensing beyond the pale of The Principles applicability.* narrow in application. It requires only that these data should be
Next, The Principles define three types of information, making shared when the information is capable of averting the envi-
distinction among primary data, processed data, and analyzed ronmental harm. Post-incident information as well as cumula-
information. These definitions are consistent with technical tive data that might give indications as to environmental trends
usage. Primary data are the raw data that are acquired by the is not included in this sharing regime. While compliance with
remote sensors. Processed data are value-added data that have this provision is difficult to monitor, it seems that the narrow-
been processed into a usable form, and analyzed information ness is likely unimportant. With the amount of available data
is the knowledge garnered when remote sensing data are inter- and expanding access, states are increasingly able to get the
preted along with other inputs (Diederiks-Verschoor, 1999). information they need to manage the domestic environmental
Finally, Principle I defines remote sensing activities as the concerns.
operation of remote sensing space systems, primary data collec- Indeed, state practice indicates that states are actively engag-
tion and storage stations, and activities in processing, interpret- ing in environmental monitoring through remote sensing
ing and disseminating the processed data. The significance of and increasingly making that data openly available. Japan has
this definition is its expansiveness in including not only ground- adopted the policy that remote sensing satellites should be used
link stations, but also the data analysis and dissemination. This as The Guardian of the Environment (Aoki, 2010, 345, 348).
holistic definition is important as it scopes these activities suf- Korea uses remote sensing for environmental protection pur-
ficiently broad to allow The Principles to articulate obligations poses (Lee, 2010, 419, 423424), and China uses satellites for
in regard to data access. environmental monitoring (Ling, 2010, 439, 451452). An EU
directive mandates that all information held by public authori-
30.2.3.2.2 E
nvironmentalandDisaster ties relating to imminent threats to human health or the envi-
Provisions:PrinciplesXandXI ronment is immediately disseminated to the public likely to be
Principles X and XI are environmental provisions. Principle X affected (EU Directive 2003/4/EC, 2003; see Harris, 2008, 37).
states that remote sensing should be used to protect the earths State practice will continue to evolve in relation to this principle
environment and that data collected that can help to avert envi- as the international community seeks to deal with global envi-
ronmental degradation. This acknowledges that remote sensing ronmental problems. In particular, as the climate change debate
data can be used to assess and locate damage, monitor the pro- intensifies, remote sensing technologies will likely be critical in
gression and effect of corrective measures, verify the application both engineering and monitoring responses to climate change.
of environmental treaties and assist in the response to man-made These technologies will gather the evidence that will underlay
and natural disasters (Carpanelli and Force, 2011). Article XI on the political debates surrounding the existence of, scope of, and
the other hand seeks to encourage states to exchange data in the responses to the problem. Additionally, it is likely that envi-
wake of natural disasters as a way to alleviate suffering occurring ronmental treaties negotiated in the future will likely rely on
in affected areas. Both of these are data-sharing provisions and satellite technology for verification purposes (see, for example,
reflect core values found in Article IX of the Outer Space Treaty. Onoda, 2005; Vsquez and Lara, 2010, 371377; Froehlich, 2011,
These principles support a general duty to inform other states of 221222).
environmental damage (Carpanelli and Force, 2011, 33).** Principle XI is worded similarly to Principle X and also has
a two-pronged obligation. First, it invokes a broad humanitar-
* See Graham and Gabrynowicz (2002a, 1819) (quoting Neil Hosenball ian obligation to promote the use of remote sensing technology
as stating that the UN Principles do not apply to domestic military to protect humans from natural disasters.*** The second prong is
systems and that the issue was never raised during the negotiations), that states should promptly share processed data and analyzed
Achilleas (2008, 2), Diederiks-Verschoor (1999, 86).
information with states that have been or will soon beaffectedby
Remote Sensing Principles, I.
For more on the different definitions of data, see Jahku (2003, 6667).
Remote Sensing Principles, I(e). Remote Sensing Principles, X.
Outer Space Treaty, Art. IX. It should be noted that meteorological data are considered a public
** Carpanelli and Force (2011, 34) argue that this particular provision is good used for the benefit of all (Jahku, 2003, 84).
part of a dynamic process of customary international law development For example, see Japans Greenhouse Gases Observation Satellite
and that the extent to which it represents custom is still being defined. (GOSAT) (Aoki, 2010, 338340).
Onoda (2005, 341) also argues that there is general obligation of inter- Global cooperation in environmental data is on the rise (Oprong and
national environmental law for States to cooperate in promotion of Rwehumbiza, 2011, 252).
global Earth observation to protect the environment. *** Remote Sensing Principles, XI.
2016 by Taylor & Francis Group, LLC
612 Remotely Sensed Data Characterization, Classification, and Accuracies
a natural disaster. This is a much broader secondary obligation principles are making data available for scientific, social, and
than that articulated under Principle X and is certainly geared economic benefit and restricting access to some data for national
toward the immediacy of the threat to life that disasters cause. security reasons, and that [d]ifferences occur in variables
Principle XI has seen a great deal of state action. Several inter- (Gabrynowicz, 2007). Significantly, data provisions (both the
national systems have been established to facilitate the sharing nondiscriminatory access mandates and the national security
of satellite data for disaster response, including the Disasters exceptions) are found in the licenses and agreements between
Charter (Charter on Cooperation, 2000), UN-SPIDER,* and states and private or commercial actors.
Asia Sentinel (Sentinel Asia, 2014). One of the most promi- The policy of nondiscriminatory access was driven by Cold
nent of these initiatives is The Charter on Space and Disaster War foreign policy and finds its roots in the U.S. national
Cooperation, the purpose of which is to provide EO data at policy, which had the goal of influencing allies and non-
times of natural or technological disasters to all authorised aligned Nations by demonstrating technological superiority
users free of charge (Beets, 2010; Harris, 2013, 47). This agree- and encouraging them to use the data (Gabrynowicz, 2007,
ment was established in 2000 and includes a broad range of par- 5).** The U.S. law and policy grants nondiscriminatory access to
ticipants beyond Nation-States to enable pragmatic responses to all data from government-funded civil satellites and provides
a disaster (Uhlir etal., 2009, 211). China has launched satellites that commercial satellite systems can use reasonable commer-
for disaster management and is involved in both UN-SPIDER cial terms and conditions (Jahku, 2003, 8889). It does give
and Asia Sentinel (Ling, 2010, 439, 453). While there have been the governments of sensed states increased rights to data from
a number of successful programs, Ito argues that specific short- commercial actors, but that access is still on reasonable terms
comings in the principles themselves inhibit effective use of and conditions (Jahku, 2003, 89; 15 CFR 960.11(b)(10); 15
remote sensing technologies for disaster warning, response, and CFR 960.12). Nondiscriminatory access policies include Japans
management (see generally, Ito, 2008). GOSAT data policy (Aoki, 2010, 339); Brazils National Institute
for Space Researchs (INPE) policy, which is to give out free
30.2.3.2.3 NondiscriminatoryAccess:PrincipleXII on the Internet all remote sensing data received by INPE, the
The most significant legal innovation in The Principles is resulting maps, and the software for image processing and GIS
nondiscriminatory access. Principle XII requires that sensed (Ferreira and Camara, 2008, 15); Landsat data, which are avail-
states be given primary data and the processed data concern- able free on the Internet; and similarly, GeoScience Australias
ing the territory under its jurisdiction on a non-discrimina- policy to make data free on the Internet (Gabrynowicz, 2007,
tory basis and on reasonable cost terms. Additionally, sensed 16). Korea and Japan both have graduated system wherein
states are given a right of access, on the same terms, to ana- data for different types of users are priced at different points
lyzed data that any state engaged in remote sensing activities (Gabrynowicz, 2007, 1819; Lee, 2010, 429430). India grants
may have. There are two keys to this obligation: nondiscrimi- nondiscriminatory access for data above a certain resolution
natory and reasonable cost terms. Nondiscriminatory has (Gabrynowicz, 2007, 20). The Canadian law provides that the
a common interpretation that the sensing States have an government of a sensed state has a right to obtain data from
obligation to provide the data to the sensed States under the a commercial actor on reasonable terms. The French civil-
same conditions as other States that wish to access the data ian remote sensing policy is based on the promotion of a
(Ito, 2008, 4950). Reasonable cost basis, on the other hand, space imagery global market where data could be acquired on
is ambiguous and open to different interpretations, and a nondiscriminatory basis (Achilleas, 2008, 2). The German
it in no way serves as a general guideline for price settings Act on Satellite Data Security favourable to commercial dis-
(Ito, 2008, 50; see also, Gabrynowicz, 2002, 31; Oprong and semination, create[s] de facto a wide database accessible to all
Rwehumbiza, 2011, 257). Principle XII is often seen as a com-
promise between the developed and developing nations in the There is at least one exception to this rule found in Israels ImageSat,
negotiating process, wherein data access is given in exchange which openly promotes exclusivity and secrecy (Gabrynowicz, 2007,
for freedom to sense (Harris, 2008, 45; Schmidt-Tedd and 21). For more on ImageSat, see Crook (2007).
Kroyman, 2008, 105). For instance, France in the SPOT Image commercial policy (Achilleas,
Nondiscriminatory access has been implemented by a 2008, 5). Gabrynowicz (2007, 710) notes that policies are still more
critical mass of states, and [g]enerally speaking, all national common than formal legal regulation.
** For more on law and policy connected to Landsat, see generally Graham
data policies and laws contain the same fundamental prin-
and Gabrynowicz (2002a) and Gabrynowicz (2005).
ciples (Gabrynowicz, 2007, 3). Gabrynowicz notes that these For an extensive, though dated, overview of data availability under U.S.
law, see Gabrynowicz (1999). Landsat data are now available free on the
* UNOOSA (2014) (UN-SPIDER is a United Nations programme, with Internet (NASA Landsat Science, 2014).
the following mission statement: Ensure that all countries and inter- USGS, Landsat, http://landsat.usgs.gov/ (accessed May 15, 2014).
national and regional organizations have access to and develop the Remote Sensing Space Systems Act (S.C. 2005, c. 45), 8(4)(c). See also
capacity to use all types of space-based information to support the full Mann (2008, 78).
disaster management cycle). Achilleas (2008) further notes that in order to secure the market from
These critiques are echoed by Jackson (1997, 872873). a legal point of view, France has supported the adoption of the Remote
Remote Sensing Principles, XII. Sensing Principles.
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 613
third persons on a non-discriminatory basis (Schmidt-Tedd mental satellite companys collection or distribution of data
and Kroyman, 2008, 105).* Other examples include Argentina, (Petras, 2005, 95).**
Malaysia, South Africa, and Thailand (Gabrynowizc, 2007, The current trend is for states to restrict data on a transac-
ixvii). tional basis in which the specifics of each request are exam-
In addition to actions at the state level, the Group on Earth ined (Gabrynowicz, 2007, 3). One of the most developed of these
Observations has developed data sharing principles that endorse regimes is in the German satellite data law, which implements
the full and open access to data and Information (Uhlir a two phase procedure: the sensitivity check undertaken by
etal., 2009, 220). This is based on an underlying philosophy that the data provider and the granting of the permit by the respon-
these data are a public good, and data should be distributed as sibility authority (Schmidt-Tedd and Kroyman, 2008, 109).
such (Uhlir etal., 2009, 220). The European Union has adopted Under this regime, data dissemination is undertaken only after
a directive to ensure[] that environmental information is sys- that data have been cleared through a process that determines
tematically available and disseminated to the public (Harris, any potential endangerment of security (Schmidt-Tedd and
2008, 37), as well as the INSPIRE Directive, which promote[s] Kroyman, 2008). Frances 20032008 Military Program Law
unrestricted access to spatial data (Doldirina, 2013, 303). China seeks to [ensure] that certain civil programs comply with
and Brazil distribute CBERS data free of charge to many users defense requirements (Achilleas, 2008, 3). While CBERS has
and have established a specific program to distribute these data some open access, Zhao argues that remote sensing data are
to African states (Ferreira and Camara, 2008, 15; Ling, 2010, strictly controlled by the Chinese government on the grounds
446447). of protecting state secrets (Zhao, 2010, 563). Additionally,
Full and open access is limited by national security concerns, Italy, India, Israel, Japan, and Korea have all implemented
but [d]ata denial is the exception, not the rule (Gabrynowicz, national security exceptions (Gabrynowicz, 2007, 1112; Aoki,
2007, 11; Uhlir etal., 2009, 230231). Not only do The Principles 2010, 338; Lee, 2010, 431432).
not apply to military activities, but also they do not apply to Shutter control provisions are found in a handful of state
civil or commercial imagery with national security implica- regulations, but likely exists de facto in many nations. Canada
tions. As a result, when a state determines that a civil or com- (Gillon, 2008, 2930; Mann, 2008, 8081) and the United States
mercial remote sensing satellite has taken imagery that affects (15 CFR 960, 2013) have shutter control provisions that they can
national security concerns, that state does not have to distribute use to interrupt a satellite operator from distributing or collect-
that data on a nondiscriminatory basis. Proper analysis of such ing data that may harm national security. Canadas legal regime
provisions begs the question of where on the spectrum is the was adopted explicitly to ensure that remote sensing providers
point passed that required data access moves into the rea- would comply with national security issues (Gillon, 2008, 25).
sonable national security constraints (Gabrynowicz, 2007, 22). The regime allows the government to interrupt normal service
National security exceptions are also often implemented via the of a satellite in cases of the most serious of national security,
licensing process that states have implemented in accordance national defence, and foreign policy/international obligations
with Principle XIV and Article VI of the Outer Space Treaty.
National security restrictions come in two different categories:
transactional restrictions on the distribution of data and the use
of shutter control, which are regulations that allow a govern-
** Gabrynowicz (2007, 13) notes [a]ll current and pending national legis-
ment to restrict, for national security reasons, a nongovern- lation and policy provide for some sort of shutter control, which cor-
rectly classifies all national security exceptions to nondiscriminatory
access as forms of shutter control. This is a classification made on the
* Nondiscriminatory in this context means that it is impossible for a outcome. The distinction made in this paper is at the point of govern-
customer to prevent a third person from accessing data about a specific ment intervention. The transactional controls are made at the data level,
region (Schmidt-Tedd and Kroyman, 2008). whereas shutter control, though including distribution limitations, also
Harris (2008) notes that the term reasonable cost is similar to, but less gives a state authority over collection activities as well. This division is
definite than cost of fulfilling a user request. meant to be explicatory, and it hardly represents two distinct categories.
Jahku (2003, 8990) argues that such laws and policies are inconsistent This sensitivity check takes into account all aspects of the transfer
with Principle XII. However, it should be noted that these exceptions including the identity of the customer (Schmidt-Tedd and Kroyman,
have become prevalent enough that they can be considered part of the 2008).
acceptable state practice under the norm. This is especially so in light According to Achilleas, though France lacks a specific remote sensing
of the prevalence of national security exceptions across a wide range law governmental control is imposed on the SPOT Image commercial
of international law regimes. See also Sekhula (2011, 229) (arguing that policy (Achilleas, 2008, 5). The state can restrict data when hostile
the nondiscriminatory access rule to have proved unworkable in prac- entities might use data representing protected and sensitive French
tice and [to] require new interpretation consistent with contemporary areas , location of French troops abroad , or the location of allied
practice) troops abroad (Achilleas, 2008b, 6).
Gabrynowicz (2007, 13) (regardless of actual practice, no Nation or Kuriyama delineates between a U.S. approach to data distribution and a
data supplier wants to appear to denounce the nondiscriminatory access Chinese approach (Kuriyama, 2010, 569).
policy). Mann adds three additional public interest: justifications for the law,
But see Jahku (2003, 8384) (noting a contractual agreement between the environment, public health, and safety of persons and prop-
India and Space Imaging restricting imagery distributed in India). erty (Mann, 2008, 69).
2016 by Taylor & Francis Group, LLC
614 Remotely Sensed Data Characterization, Classification, and Accuracies
concerns (Gillon, 2008, 2930).* The U.S. provisions require (Graham and Gabrynowicz, 2002a, 291). For instance, Frances
licensees to agree to [s]pecific limitations on operational perfor- SPOT Image is a commercial data distributor of public data
mance, including, but not limited to, limitations on data collec- (Jackson, 1997, 859), and India has made data from government
tion and dissemination in their licenses on a case-by-case basis remote sensing satellites commercially available (Gabrynowicz,
(15 CFR 960.11, 2013). 2007, 6). Commercialization creates legal issues because the
UN principles were agreed [to] at a time when the commer-
cialization of remote sensing activities were still not envisaged
30.2.4 Geospatial Law
(John et al., 2011, 261). Commercialization is not a completed
Remote sensing data, primarily as a result of the Internet, has process (and will most likely remain that way indefinitely), and
become increasingly available to individuals and more promi- an important lesson of its limitations can be found in the failed
nent in society. Products such as Google Earth and extensive Landsat commercialization in the United States (Gabrynowicz,
media usage of satellite imagery have brought remote sensing 2005, 5059; Gabrynowicz, 2007, 56; and, generally, Graham
data into mainstream consciousness. Along with other tech- and Gabrynowicz, 2002a).
nological advances, such as GPS, an array of geo-location and Next is the process of globalization, the modern version of
mapping applications are increasingly available to individuals which has cultural and conceptual roots in the space age and
globally through devices like smartphones and tablets, and geo- the reimagination of the world as a global space. Held argues
technologies are being embedded directly into these devices. that globalization is the widening, deepening and speeding up
This means that an array of legal issues have arisen, primar- of worldwide interconnectedness in all aspects of contemporary
ily in the domestic law context, which situates remote sensing social life, from the cultural to the criminal, the financial to the
activities within a broader complex of geo-location technologies. spiritual. It leads to the combining of the local and the global
This has given rise to the area of geospatial law, which is loosely to create distant proximities (Ferguson and Mansbach, 2012).
defined as the body of law that governs the collection, use, and Essentially, globalization changes the space that society exists in,
distribution of geospatial information. and importantly, satellite imagery, digital maps, and associated
This new trend is directly related to three phenomena: com- information have transformed our ability for understanding
mercialization, globalization, and technological convergence. the forces that shape geographical space (Ferreira and Camara,
The first is commercialization. This process occurred in a num- 2008). As such, globalization is an information-driven pro-
ber of countries and drove the development of the domestic cess, and one of the effects of globalization has been increased
law in those countries. Canada adopted The Remote Sensing access to global technologies.** This is certainly true in the field
Space Systems Act of 2005 after advances in satellite remote of Remote Sensing with numerous states now engaging in
sensing technology in the private sector started to drive the remote sensing activities including new entrants such as Nigeria,
development of commercial space systems (Gillon, 2008, 19). Algeria, Morocco, Egypt, Kenya, South Africa, Colombia, and
Similarly, Germanys data security law was triggered by the Turkey (Jahku, 2003, 6869; Gabrynowicz, 2007, 7; Oprong and
development of publicprivate partnerships and commer- Rwehumbiza, 2011, 255). More importantly, though, is the fact
cial use of data (Schmidt-Tedd and Kroyman, 2008, 100102). that remote sensing is a global technology, and as such, it facili-
Commercialization does not just refer to the operation of the tates the conceptualization of the world as a global space.
satellite; it also cuts across issues such as whether or not data Finally, the process of technological convergence has changed
should be distributed by the public sector or the private sector the way in which we interact with this type of data. Technological
convergence is the process through which technologies that
* Shutter control restrictions in Canada can apply to spatial, temporal, or
resolution elements of the remote sensing activities (Gillon, 2008, 30).
The regime also gives the government priority access to data in cases of This author ascribes to the view that globalization is a process that can
emergency response , in support of requests for aid of a civil power, or retract and recede. For a fuller account of different theories of globaliza-
in support of Canadian Forces, which moves such requests to the front tion and its meaning, see generally, Ferguson and Mansbach (2012).
of the order queue (Gillon, 2008, 3031; Mann, 2008, 81). Ferguson and Mansbach (2012, 21) note that one of the features of glo-
Gabrynowicz notes that [t]he distinction between public and pri- balization is the nonexclusiveness of territoriality and note that outer
vate in the remote sensing space segment is disappearing worldwide. space is one of the environments that have allowed territoriality to be
What constitutes commercial operations varies among nations reimagined as nonexclusive. For a more in-depth argument that the
(Gabrynowicz, 2007, 3, 1516). While these definitional elements are space age is an important historical factor in modern globalization, see
important, the purpose here is to acknowledge the general trend as Blount and Fussell (2014).
opposed to engaging with the competing definitions. In the context of David Held referenced in Ferguson and Mansbach (2012, 17).
this paper, commercialization represents the processes along this spec- ** Ferguson and Mansbach (2012, 22, 167) (Globalization advocates also
trum that result in an increased number of private, nongovernmental reject critics concerns about modern technology More information
actors in the field. Additionally, Gabrynowicz notes that there is a rising creates an informed citizenry And, although a digital divide exists
trend of new government entities organized like private corporations new technologies are reaching more and more people and accelerating
(Gabrynowicz, 2007, 1618). See generally, Jackson (1997) (arguing that economic development in poor countries.). In relation to space tech-
the UN Principles Regime is inadequate to cope with increased com- nologies, see generally Gabrynowicz (2005, 30).
mercialization). For an early view of the remote sensing industry, see For example, Gabrynowicz argues that Landsat is a national program
generally, Graham and Gabrynowicz (2002b). with an inherently global function (Gabrynowicz, 2005, 47).
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 615
were once significantly distinct (e.g. a map and a phone) are involved. This section will not attempt to give a full account of
integrated and become indistinguishable. Current convergence geospatial law, which is a broad category encompassing a vari-
trends are connected to the way in which information has been ety of legal issues; instead, it will attempt to focus on prominent
redefined as data in a technical sense. This is not to imply that issues that are directly related to remote sensing technologies.
information is data, but instead that, technologically, they are
indistinguishable. For example, when a user downloads a remote 30.2.4.1 Privacy and Security
sensing image, s/he receives information when s/he look at the The 9/11 terrorist attacks shattered the notion that globalization
image. However, as that image was transferred over the Internet, was necessarily a sign of positive progress. Not only did global-
it was simply data broken down into manageable packets, and ization enable global public goods and mass communication,
the application on the end users device reassembled it into but it also enabled global threats like terrorism. In this context,
information.* This means that data are now highly integrable privacy has become an increasingly important issue. This results
(depending on creativity and innovation in applications) and from both of the faces of globalization. The increasing access
highly portable (depending on bandwidth). This can cause an to and use of remote sensing imagery by the general popula-
array of issues since integrated data packages pull from a variety tion have generated significant privacy issues based on global
of sources. For example, mapping software on mobile devices connectivity and content coupled with global threats. The UN
may use geo-location information provided by the device, map Principles offer only fairly minimal guidance on the topic of
databases, remote sensing data, and crowd-sourced information, privacy of individual persons and entities, and as a result, pri-
among other sources. When this information becomes data, it vacy protections are consequently to be found only at national
becomes indistinguishable as separate sets of knowledge and level (von der Dunk, 2013, 257), but at the national level, [p]
becomes a single information point for the user. This is one of rivacy is a broad area of law, and relevant legislation is unclear in
the features that has allowed ICT technologies [to reduce] geog- concept and application (Cho, 2013, 261).
raphy and physical distance in politics, economics, and war In the wake of the 9/11 attack, there were sweeping organi-
render[ing] territory less important (Ferguson and Mansbach, zational changes in the U.S. defense and intelligence organi-
2012, 112). While territory has become less important, location zations, which g[ave] rise to myriad potential statutory and
has become increasingly important, meaning that [g]lobal regulatory impediments (Petras, 2005, 81). This creates specific
politics no longer reflects neat, exclusive territorial boxes, issue about the use of U.S. military satellites for use of surveil-
creating a need to rethink and design a wide variety of maps lance of the U.S. citizens, which raises particularly vexing and
that take into account the many forms of political activity controversial issues (Petras, 2005, 81).** While the freedom of
(Ferguson and Mansbach, 2012, 112). Remote sensing data have use of remote sensing satellites is established at the international
become part of technological convergence, and there is a trend law level, these technologies are often treated differently under
toward increas[ing] the use of data (Gabrynowicz, 2007, 14). domestic law (Petras, 2005, 94). In the American context, this
Additionally, convergence has made the Internet the core tool represents a cultural value that has been embedded in the law
for improving transparency, and geospatial technologies play an (Petras, 2005, 102). The Reconstruction era Posse Commitatus
important role in transparency (Gabrynowicz, 2007, 11). Act restricts the use of the military in law enforcement, and the
These three processes have significantly changed the place values found in the Posse Commitatus Act have been furthered
of remote sensing data in society. This naturally results in legal formalized through laws, regulations, policies, and procedures.
tensions as the law is recalibrated to cope with emerging tech- The application of Posse Commitatus and its progeny to remote
nologies. Commercialization challenges remote sensing law by sensing technologies is less clear in practice, though there are
introducing new types of stakeholders that do not fit into the clear exceptions (Petras, 2005, 110113). However, the core value
international law regime. Globalization challenges the regime of separating these institutions from domestic law enforcement
because the overlap of authority and governance among sub- can be found in the law and policies governing remote sensing
national, national, transnational, and international polities is technologies.
fraught with difficulties (Ferguson and Mansbach, 2012, 134).
And technological convergence challenges the regime by reori-
enting how and why remote sensing data are used. The result- Petras notes specifically that 9/11 caused the U.S. national security
ing complexity situates remote sensing law within the bounds of regime to turn inward (Petras, 2005, 8283).
** The securitization phenomenon will be addressed in the American con-
geospatial law, at least as far as data and information issues are
text since it is the locus of this trend.
While this is a value, it should be noted it is not a constitutional value,
* This is the result of the network design of the Internet (Clark and though it reflects core American Values. Congress has the ability to
Landau, 2010, 27). This design is critical to the functionality of the waive the prohibition in certain circumstances, which in a securitiza-
Internet as we know it. tion context means that political will can shift this value. See Petras
Location is identity (Clark and Landau, 2010, 25). (2005, 104105) (noting the Congressional waiver for military partici-
As an example of convergence specific to remote sensing, India has seen pation in antidrug trafficking activities near the U.S. Borders).
moves to regulate web-based image suppliers (Gabrynowicz, 2007, 21). For a common line of cases addressing the Posse Commitatus Act, see
Ferreira and Camara (2008, 17) link the Internet to a change in how U.S. v. Yunis (1991), Wrynn v. U.S. (1961), U.S. v. Red Feather (1975),
remote sensing data are used. Bissonette v. Haig (1985), U.S. v. Rasheed (1992), U.S. v. Kahn (1994).
2016 by Taylor & Francis Group, LLC
616 Remotely Sensed Data Characterization, Classification, and Accuracies
The U.S. law requires defense collection of information on marijuana (Gunasekara, 2010, 124126). The Court determined
U.S. persons to be gathered using the least intrusive means, that the technology that was used was not available to the gen-
which requires the use of a four-step methodology (Petras, 2005, eral public, and as a result, it violated a reasonable expectation of
97). Petras points out that satellite data present an issue since privacy (Gunasekara, 2010). Of course, remote sensing technol-
it is unclear whether that data represent a publicly available ogy is now arguably available to the general public as a result of
source of information, in light of commercial imagery available technological convergence (Petras, 2005, 99; Gunasekara, 2010,
on the Internet (the first tier of the methodology), or whether 135136). A final and more recent case is Jones v. United States,
such information retrieval requires a warrant or approval by which addressed the use of a GPS device on a car. The opinion
the Attorney General (the fourth tier of the methodology) was split, but one of the concurring opinions posited that such
(Petras, 2005, 9798). National Geospatial-Intelligence Agency searches created a mosaic of the life of an individual and that
procedures require that imagery gathered by domestic satellites such mosaics breach a reasonable expectation of privacy (U.S. v.
must be for a foreign intelligence mission and only if com- Jones, 2012). If such a standard were to be adopted by the Court
mercial imagery is unsuitable or unavailable (Petras, 2005, in the future, it would certainly have implications for uses of
100, 108109). Any other military request must be for autho- remote sensing data in law enforcement.
rized activities of the armed forces (Petras, 2005, 100101). In general, states have a variety of approaches to privacy
Consistent with this theme, there are further procedures to issues. Korea, in Article 17 of its Space Development Promotion
limit the use of such technologies in the American context. The Act, has legislated that the Government shall endeavor to avoid
collection cannot be focused on a specific U.S. person and any invasion of privacy during the course of utilizing satellite
must be in connection with activities that are not domestic in information (Lee, 2010, 425). Lee notes that privacy is a fun-
nature (Petras, 2005, 101).* damental right provided for in the Korean Constitution. Legal
Since satellite technology is not entirely out of the picture regimes create a variety of contexts in which privacy must be
for law enforcement, the restrictions on government observa- considered; for example, remote sensing data issues have been
tion of citizens are primarily contained in search and seizure identified in the Health Insurance Portability and Accounting
law, and as remote sensing technology continues to become Act in the United States (Secunda, 2004, 251). Other states do
a viable law enforcement tool, issues of proper use by govern- not have fully formed privacy protections. Zhao notes that Hong
ments will become more important (Gunasekara, 2010). In the Kong has no comprehensive data protection law (Zhao, 2010,
United States, this is governed by a Constitutional restriction 551). In the Canadian remote sensing regime, there are no pro-
found in the Fourth Amendment, which forbids the government visions dealing with privacy, and no privacy conditions have
from engaging in unreasonable searches without a proper war- been incorporated in the first remote sensing satellite system
rant. The standards involved come from a line of cases that hold license; therefore, privacy concerns must be addressed under
that searches that breach a reasonable expectation of privacy Canadas Constitution and Privacy Act (Mann, 2008, 8587).
violate Constitutional protections (Gunasekara, 2010, 118119). Additionally, commercial use of data has caused numerous con-
However, courts have been challenged by technologies that make cerns. In particular, the Google suite of geo-technologies has
observations of individuals by remotely sensing them or their come under extensive fire for privacy violations (Foresman, 2011;
property. While the Supreme Court has never ruled directly on Google Faces Lawsuits over Gmail, 2013), and as content provid-
the use of satellite remote sensing data, there is a line of cases ers increasingly incorporate geospatial data, privacy breaches
that can be traced, which gives some indication of how satellites are becoming more common (e.g., Chen, 2011; Frizell, 2014).
might be treated under the U.S. law. This creates cross-jurisdictional problems since content provid-
Several cases address the use of aerial surveillance and hold ers operate in a multiplicity of jurisdictions at any given time.
that the government has fairly wide discretion to use aerial sur-
veillance without a warrant. Of particular note, though, is the 30.2.4.2 Use of Remote Sensing Data as evidence
dictum in Dow Chemical in which the Court posits that cer- Data derived from remote sensing activities has become increasingly
tain technologies not available to the general public, like satel- important in judicial proceedings; this is related to both the integra-
lite technology, might violate the Fourth Amendment (Petras, tion of geo-location technologies into everyday lives and large-scale
2005, 99). This theme is taken up in the Kyllo v. United States problems such as environmental damage. The ability to introduce
case. In this case, police used a thermal imager to determine this type of data is governed by rules of evidence for the particular
whether grow lights were being used inside a residence to grow jurisdiction meaning that geospatial and remote sensing data must
* Domestic activities are those that take place within the United States
that do not involve a significant connection with a foreign power, orga-
nization, or person (Petras, 2005). Petras specifically argues that inter-
national terrorist organizations operations within the borders of the
United States would sit outside this definition (Petras, 2005, 110). Mann (2008, 87) argues that in Canada [p]rivacy rights are adequately
U.S. Constitution. 4th Amendment. protected. It should also be noted that the Canadian Supreme Court in
Gunasekara identifies Ciraolo, Dow Chemical, and Riley (Gunasekara, a case very similar to the United States Kyllo case addressed earlier split
2010, 120124). from the U.S. rational (Mann, 2008, 8687; Gunasekara, 2010, 130134).
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 617
be fit into a preexisting evidentiary categories.* One of the core issues certain degree of human intellectual intervention is required
has been authentication of remote sensing evidence in courts due to for information to be copyright protected (Ito, 2008, 55). The
its digital nature (Shipman, 2013, 359377). This problem has been nature of remote sensing data means that [c]opyright may not
addressed through technical means such as digital signatures (Croi be the most appropriate regime to protect EO data (Doldirina,
etal., 2013, 379398) and complete authentication procedures, as in 2013, 298).
the case of USGS/EROS data (Rychlak etal., 2007, 212218). State regulations vary significantly. Goh notes a Singaporean
Korea (Lee, 2010, 423424), Queensland, Australia case in which a data provider was held to violate copyright
(Goulevitch, 2013), Singapore (Escolar, 2013, 108110), and protections by distributing maps created with unlicensed data
the United States (Rychlak et al., 2007, 206210; Dighe et al., (Escolar, 2013, 111112). The European Union protects intellec-
2013) use satellite technology for investigating and prosecuting tual property rights in databases to secure remuneration to the
breaches of environmental laws. Satellite data have also found maker of the database, which Harris argues could have impli-
its way into evidence in a variety of international Institutions cations for remote sensing technologies (Harris, 2008, 3738).
(see generally Williams, 2013, 195216). The International Court He notes that problems could arise when database protections
of Justice has used satellite data in numerous cases, which often are involved with public data (Harris, 2008, 38). Nigeria explic-
involve boundary disputes (see generally Froehlich, 2011, 222 itly grants intellectual property rights to data producers via its
225). The Permanent Court of Arbitration and the International National Geoinformation Policy (John etal., 2011, 262; see also
Criminal Court have both made use of earth observation data as Birisbe, 2006, 241245), and JAXA maintains copyright to data
evidence. The use of these data in both civil and criminal courts it creates in Japan (Gabrynowicz, 2007, 18). The Korean regime
will increase across a number of areas as data become increas- can protect remote sensing imagery via copyright or database
ingly embedded into society. protections (Lee, 2010, 426428). Additionally, Germany has
denied copyright of Landsat data, while France has allowed it
30.2.4.3 torts (MejiaKaiser, 2006).
Misuse of geospatial data could result in torts. Navigation sys-
tems providers have been repeatedly sued for incorrect data 30.3 conclusion
resulting in incorrect directions (e.g., Hadhazy, 2010; Gannes,
2012). A U.S. court held a map provider liable for incorrect data Remote sensing will remain an important technology in the fore-
that it purchased from a third party (Brocklesby v. U.S., 1985), seeable future. Based on the international principles examined
and Nigerian law can hold GIS professionals legally account- in this chapter, remote sensing is regulated within a stable inter-
able for the accuracy and reliability of information stored in national legal framework that is permissive in nature. Emerging
their databases, sold, or issued to the public (John etal., 2011, challenges will come from the increasing use of remote sensing
263). Ito states that [d]ata suppliers bear liability risks in cases data in a data-driven society. National regulations will need to
where the population is affected by disasters or aid workers are balance among competing interests of national security, individ-
injured as a result of inappropriate instructions (Ito, 2008, 57). ual liberty, and commercial growth in an increasingly complex
While Ito is arguing for a more complete international legal technological infrastructure.
regime, data torts will likely continue to be handled ad hoc at the
domestic judicial level. References
30.2.4.4 intellectual Property Achilleas, P., French remote sensing law, The Journal of Space Law,
The UN Principles are completely silent as to intellectual 34 (2008) 1.
property rights in data generators (Ito, 2008, 50). This is likely Agreement on the Rescue of Astronauts, the Return of Astronauts
because, at the time of the negotiations, these activities had not and the Return of Objects Launched into Outer Space (the
yet seen a sufficient amount of commercial or private activity to Rescue Agreement) (entered into force on December 3,
warrant addressing the issue. Commercialization has resulted in 1968).
a variety of users making intellectual property rights a growing Alvarado, S.C., The analysis of existing international space coop-
concern for remote sensing data providers and users. Generally, eration initiatives for UNESCOs World Heritage Sites, IAC-
current practice is that the ownership of data stays with data 10.E3.1A.14 (2010).
generators and the copyright is claimed by the majority of data Aoki, S., Japanese law and regulations concerning remote sensing
generators, but it is not clear to what extent these claims can activities, The Journal of Space Law, 36 (2010) 335.
be maintained for raw data especially in jurisdictions where a
* For analysis of these rules in different jurisdictions, see, England and For an overview of copyright implications for EO data, see Doldirina
Wales, Germany, the Netherlands, Belgium, Mosteshar (2013, 147175); (2013, 293310). For an example of such a regime, see Zhao on Hong
Nigeria, John etal. (2011, 262263); Singapore, Escolar (2013, 98108); Kong, Zhao (2010, 552554).
United States, Dighe etal. (2013, 8190) and Wright (2013, 313320). Cromer presents a full analysis of the conflicts between the copyright
PCA, Macauley (2013, 224225); ICC, Macauley (2013, 226238). regime and the space law regime Cromer (2006).
2016 by Taylor & Francis Group, LLC
618 Remotely Sensed Data Characterization, Classification, and Accuracies
Beets, J., The international charter on space and major disasters Croi, W., F.-M. Foeteler, and H. Linke, Introducing digital sig-
and international disaster law: The need for collaboration natures and time-stamps in the EO data processing chain,
and coordination, Air & Space Lawyer, 22/4 (2010) 1215. in R. Purdy and D. Leung (eds.), Evidence from Earth
Birisbe, T., Outer space activities and intellectual property protec- Observation Satellites, Leiden, the Netherlands: Martinus
tion in Nigeria, The Journal of Space Law, 32 (2006) 229. Nijhoff Publishers (2013), p. 379.
Bissonette v. Haig, 776 F.2d 1384 (1985). Cromer, J.D., How on earth terrestrial laws can protect geospatial
Blount, P.J., Limits on space weapons: Incorporating the law of data, The Journal of Space Law, 32 (2006) 253.
war into the Corpus Juris Spatialis, in Proceedings of the Crook, J., Corporate Sovereign symbiosis: Wilson v. ImageSat
51st Colloquium on the Law of Outer Space (AIAA 2009). International, Shareholders actions, and the dualistic
Blount, P.J., Developments in space security and their legal impli- nature of state-owned corporations, The Journal of Space
cations, Law/Technology, 44/2 (2011) 18. Law, 33 (2007) 411.
Blount, P.J., Renovating space: The future of international space Dembling, P.G. and D.M. Arons, The evolution of the outer space
law, Denver Journal of International Law and Policy 40 treaty, Journal of Air Law & Commerce, 33 (1967) 419456.
(2012) 515686. Diederiks-Verschoor, I.H.Ph., An Introduction to Space Law, The
Blount, P.J. and J. Fussell, Musical counter narratives: Space, Hague, the Netherlands: Kluwer (1999).
skepticism, and religion in American music, in AIAA Dighe, K., T. Mikolop, R.W. Mushal, and D. OConnell, The use of
52nd Aerospace Sciences Meeting (January 2014), National satellite imagery in environmental crimes prosecutions in the
Harbor, MD. United States: A developing area, in R. Purdy and D. Leung
Brocklesby v. U.S. 767 F.2d 1288 (9th Cir. 1985). (eds.), Evidence from Earth Observation Satellites, Leiden,
Carpanelli, E. and B. Cohen, A legal assessment of the 1996 decla- the Netherlands: Martinus Nijhoff Publishers (2013), p. 65.
ration on space benefits on the occasion of its fifteenth anni- Doldirina, C., The impact of copyright protection and public
versary, The Journal of Space Law, 38 (2012) 1. sector information regulations on the availability of remote
Carpanelli, E. and M.K. Force, The protection of the earth natural sensing data, in R. Purdy and D. Leung (eds.), Evidence
environment through space activities: A general overview of from Earth Observation Satellites, Leiden, the Netherlands:
some legal issues, in Proceedings of the International Institute of Martinus Nijhoff Publishers (2013), p. 293.
Space Law, The Hague, the Netherlands: Eleven (2011), p. 31. Escolar, G.G., The use of EO data as evidence in the courts of
Carter, J., Remarks at the congressional space medal of honor awards Singapore, in R. Purdy and D. Leung (eds.), Evidence from
ceremony, Kennedy space Center, FL, Weekly Compilation of Earth Observation Satellites, Leiden, the Netherlands:
Presidential Documents, 14(40) (October 8, 1978) 1684. Martinus Nijhoff Publishers (2013), p. 93.
CBERS Data Policy, The Journal of Space Law, 31 (2005) 281. EU Directive 2003/4/EC, On public access to environmental
Charter on Cooperation to Achieve the Coordinated Use of Space information and repealing Council Directive 90/313/EEC
Facilities in the Event of Natural or Technological Disasters (January 28, 2003).
Rev. 3 (April 25, 2000), http://www.disasterscharter.org/ Eyes on Darfur, http://www.eyesondarfur.org/ (last visited
web/charter/charter. Accessed February 22, 2014. February 22, 2014).
Chen, B.X., iPhone tracks your every move, and theres a map for Ferguson, Y.H. and R.W. Mansbach, Globalization: The Return of
that, WIred (April 20, 2011), http://www.wired.com/gadget- Borders to a Borderless World?, London, U.K.: Routledge (2012).
lab/2011/04/iphone-tracks/. Accessed February 22, 2014. Ferreira, H.S. and G. Camara, Current status and recent develop-
Chinese Law: Registration, launching and licensing space objects, ments in Brazilian remote sensing law, The Journal of Space
The Journal of Space Law, 33 (2007) 437. Law, 34 (2008) 11.
Cho, G., Privacy and EO: An overview of legal issues, in R. Purdy Filho, J.M. and A.F. dos Santos, ChineseBrazilian accord on dis-
and D. Leung (eds.), Evidence from Earth Observation tribution of CBERS products, The Journal of Space Law, 31
Satellites, Leiden, the Netherlands: Martinus Nijhoff (2005) 271.
Publishers (2013), p. 259. Foresman, S., Google faces $50 million lawsuit over Android
Clark, D.C. and S. Landau, Untangling attribution, in Committee location tracking, Ars Technica (April 30, 2011), http://
on Deterring Cyberattacks, Proceedings of a Workshop on arstechnica.com/tech-policy/2011/04/google-faces-50-
Deterring Cyberattacks: Informing Strategies and Developing million-lawsuit-over-android-location-tracking/. Accessed
Options for U.S. Policy, Washington, DC: National February 22, 2014.
Academies Press (2010), p. 25. Frizell, S., Tinder security flaw exposed users locations, Time
Convention on International Liability for Damage Caused by (February 19, 2014), http://techland.time.com/2014/02/19/
Space Objects (the Liability Convention) (entered into tinder-app-user-location-security-flaw/. Accessed February
force on September 1, 1972). 22, 2014.
Convention on Registration of Objects Launched into Outer Froehlich, A., Space related data: From justice to development,
Space (the Registration Convention) (entered into force in Proceedings of the International Institute of Space Law,
on September 15, 1976). The Hague, the Netherlands: Eleven (2011), p. 221.
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 619
Gabrynowicz, J.I., Earth observation: The view from the ground Harris, R., Science, policy, and evidence in EO, in R. Purdy and
up, Space Policy (August 1997), 229. D. Leung (eds.), Evidence from Earth Observation Satellites,
Gabrynowicz, J.I., Defining data availability for commercial Leiden, the Netherlands: Martinus Nijhoff Publishers
remote sensing systems under United States federal law, (2013), p. 43.
Annals of Air and Space Law, 23 (1999) 93. Harrison, R.G., Space and Verification, Vol. 1: Policy Implications,
Gabrynowicz, J.I. (ed.), The UN Principles Relating to Remote USAFA, CO: Eisenhower Center for Space and Defense
Sensing of the Earth from Space: A Legislative History Studies (2011).
Interviews of Members of the United States Delegation, Harvey, D., A Brief History of Neoliberalism, Oxford, U.K.: Oxford
University, MS: National Center for Remote Sensing and University Press (2005).
Space Law (2002). International Law Commission, Draft articles on Responsibility
Gabrynowicz, J.I., Space law: Its cold war origins and challenges of States for Internationally Wrongful Acts, with commen-
in the era of globalization, Suffolk University Law Review, taries (2001).
37 (2004) 1043. Ito, A., Improvement to the legal regime for the effective use of
Gabrynowicz, J.I., The perils of Landsat from grassroots to glo- satellite remote sensing data for disaster management and
balization: A comprehensive review of US remote sensing protection of the environment, The Journal of Space Law,
law with a few thoughts for the future, Chicago Journal of 34 (2008) 45.
International Law, 6/1 (2005) 45. Jackson, S.M., Cultural lag and the international law of remote
Gabrynowicz, J.I., The International space treaty regime in the era sensing, Brooklyn Journal of International Law, 23 (1997) 853.
of globalization, Ad Astra (Fall 2005), 30. Jahku, R., International law governing the acquisition and dis-
Gabrynowicz, J.I., The Land Remote Sensing Laws and Policies of semination of satellite imagery, Journal of Space Law, 29
National Governments: A Global Survey, University, MS: (2003) 65.
National Center for Remote Sensing, Air, and Space Law John, O.N., E. Ezekiel, and S.O. Mohammed, Legal regime of
(2007). remote sensing and geographic information systems in
Gannes, L., GPS App Strava sued over cyclists death, All Things Nigeria, in Proceedings of the International Institute of Space
D (June 19, 2012), http://allthingsd.com/20120619/ Law, The Hague, the Netherlands: Eleven (2011), p. 260.
gps-app-strava-sued-over-cyclists-death/. Kopal, V., The role of United Nations declarations of principles
Gillon, T., Regulating remote sensing space systems in Canada in the progressive development of space law, The Journal of
New Legislation for a New Era, The Journal of Space Law, Space Law, 16 (1988) 5.
34 (2008) 19. Kuriyama, I., Environmental monitoring cooperation paves the
Google Faces Lawsuits over Gmail, Street View privacy, CBC way for common rules on remote sensing, The Journal of
News (October 2, 2013), http://www.cbc.ca/news/busi- Space Law, 36 (2010) 567.
ness/google-faces-lawsuits-over-gmail-street-view-pri- Lachs, M., The Law of Outer Space: An Experience in Contemporary
vacy-1.1876594. Law-Making, Leiden, the Netherlands: Martinus Nijhoff
Goulevitch, B., Ten years of using earth observation data in sup- (1972, 2010).
port of Queenslands vegetation management framework, Lee, J.G., Remote sensing issues as they relate to Korea, The
in R. Purdy and D. Leung (eds.), Evidence from Earth Journal of Space Law, 36 (2010) 415.
Observation Satellites, Leiden, the Netherlands: Martinus Licensing of Private Remote Sensing Systems, 15 CFR 960 (2013).
Nijhoff Publishers (2013), p. 113. Ling, Y., Remote sensing data distribution and application in
Graham, J.F. and J.I. Gabrynowicz (eds.), Landsat 7: Past Present the environmental protection, disaster prevention, and
and Future, University, MS: National Remote Sensing and urban planning in China, The Journal of Space Law, 36
Space Law Center (2002a). (2010) 435.
Graham, J.F. and J.I. Gabrynowicz, The Remote Sensing Industry: Listner, M., Iridium 33 and Cosmos 2251 three years later:
A CEO Forum, University, MS: National Remote Sensing Where are we now? The Space Review (February 13, 2012),
and Space Law Center (2002b). http://www.thespacereview.com/article/2023/1. Accessed
Gunasekara, S.G., The March of science: Fourth Amendment February 22, 2014.
implications on remote sensing in criminal law, The Journal Lyall, F. and P.B Larsen, Space Law: A Treatise, Aldershot, U.K.:
of Space Law, 36 (2010) 115. Ashgate (2013).
Hadhazy, A., Bad directions from Google maps lead to lawsuit, Macauley, E.D., The use of EO technologies in court by the
TechNewsDaily (June 3, 2010), http://www.technewsdaily. office of the prosecutor of the international criminal court,
com/556-bad-directions-from-google-maps-lead-to-law- in R. Purdy and D. Leung (eds.), Evidence from Earth
suit.html. Accessed February 22, 2014. Observation Satellites, Leiden, the Netherlands: Martinus
Harris, R., Current status and recent development in UK and Nijhoff Publishers (2013), p. 217.
European remote sensing law and policy, The Journal of Mann, B., First license issued under Canadas remote sensing sat-
Space Law, 34 (2008) 33. ellite legislation, The Journal of Space Law, 34 (2008) 67.
2016 by Taylor & Francis Group, LLC
620 Remotely Sensed Data Characterization, Classification, and Accuracies
Marchisio, S., The evolutionary stages of the legal subcommittee of Sentinel Asia, Sentinel Asia: Disaster management support sys-
the United Nations Committee on the peaceful uses of outer tem in the Asia-Pacific Region, http://www.aprsaf.org/
space (COPUOS), The Journal of Space Law, 31 (2005) 219. initiatives/sentinel_asia/pdf/Sentinel-Asia.pdf (last visited
Mayence, J.-F., Belgian legal framework for earth observation February 22, 2014).
activities, The Journal of Space Law, 34 (2008) 89. Shipman, A., Authentification of images, in R. Purdy and D.
MejiaKaiser, M., Copyright claims for Meteosat and Landsat Leung (eds.), Evidence from Earth Observation Satellites,
images under court challenge, The Journal of Space Law, 32 Leiden, the Netherlands: Martinus Nijhoff Publishers
(2006) 293. (2013), p. 359.
Mosteshar, S., EO in the European Union: Legal considerations, Tarikhi, P., Mahdasht satellite receiving station verging into a
in R. Purdy and D. Leung (eds.), Evidence from Earth space center, Res Communis (October 13, 2008), http://
Observation Satellites, Leiden, the Netherlands: Martinus rescommunis.olemiss.edu/2008/10/13/guest-blogger-par-
Nijhoff Publishers (2013), p. 147. viz-tarikhi-mahdasht-satellite-receiving-station-verging-
NASA Landsat Science, The numbers behind Landsat, http:// into-a-space-center/. Accessed February 22, 2014.
landsat.gsfc.nasa.gov/?page_id=9 (last visited February 22, Terekov, A., UN general assembly resolutions and outer space
2014). law, in Proceedings of the International Institute of Space Law
Onoda, M., Satellite Earth Observation As Systematic (1997), p. 97.
Observation in multilateral environmental treaties, The Treaty between the United States of America and the Russian
Journal of Space Law, 31 (2005) 339. Federation on Measures for the Further Reduction and
Oprong, A.A. and V. Rwehumbiza, A glance at the earth observa- Limitation of Strategic Offensive Arms (signed April 10,
tion policies and regulations and the impact on developing 2010).
countries: Focusing on the African continent, in Proceedings Treaty on Principles Governing the Activities of States in the
of the International Institute of Space Law (2011), p. 251. Exploration and Use of Outer Space, including the Moon
Petras, C.M., Eyes on freedomA view of the law governing and Other Celestial Bodies (the Outer Space Treaty)
military uses of satellite reconnaissance in U.S. Homeland (entered into force on October 10, 1967).
defense, The Journal of Space Law, 31 (2005) 81. Uhlir, P.F., R.S. Chen, J.I. Gabrynowicz, and K. Janssen, Toward
Remote Sensing Space Systems Act (S.C. 2005, c. 45). implementation of the global earth observation system of
Reynolds, G.H. and R.P. Merges, Outer Space: Problems of Law systems data sharing principles, The Journal of Space Law,
and Policy, Boulder, CO: Westview Press (1997). 35 (2009) 201.
Richelson, J.T., Americas Space Sentinels: DSP Satellites and UN Charter (1945).
National Security, Lawrence, KA: University Press of Kansas UNGA Meeting Record no. A/41/PV.95 (December 3, 1986).
(1999). UNGA Res. 41/65, Principles relating to remote sensing of the
Rychlak, R.J., J.I. Gabrynowicz, and R. Crowsey, Legal certification earth from outer space (December 3, 1986).
of digital data: The earth resources observation and science UNGA Res. 51/122, The Declaration on International
data center project, The Journal of Space Law, 33 (2007) 195. Cooperation in the Exploration and Use of Outer Space
Satellite Data Security Act, The Journal of Space Law, 34 (2008) for the Benefit and in the Interest of All States, Taking into
115. Particular Account the Needs of Developing Countries
Satellite Sentinel Project, http://www.satsentinel.org/ (last visited (December 13, 1996).
February 22, 2014). UNGA Res. 60/1, 2005 World Summit Outcome (October 24, 2005).
Schmidt-Tedd, B. and M. Kroyman, Current status and recent UNGA Res. 62/102, Recommendations on Enhancing the
developments in German remote sensing law, The Journal Practice of States and International Intergovernmental
of Space Law, 34 (2008) 97. Organizations in Registering Space Objects (December 17,
Schrogl, K.-U. and N. Hedman, The U.N. General 2007).
Assembly Resolution 62/102 of 17 December 2007 on UNOOSA, About UN-SPIDER, http://www.unoosa.org/oosa/en/
Recommendations on Enhancing the Practice of States unspider/index.html (last visited February 22, 2014).
and International Intergovernmental Organizations in United Nations High Commission on Human RIghts, Report
Registering Space Objects, The Journal of Space Law, 34 of the commission of inquiry on human rights in the
(2008) 141. Democratic Peoples Republic of Korea, A/HRC/25/63
Secunda, P.M., A mosquito in the ointment: Adverse HIPPAA (February 17, 2014).
implications for health-related remote sensing research U.S. Constitution.
and a Reasonable Solution, The Journal of Space Law, 30 U.S. v. Jones, No. 10-1259, Concurring Opinion of Justice
(2004) 251. Sotomayor (January 23, 2012).
Sekhula, P.P., The right to remote sense data: Impact of mul- U.S. v. Kahn, 35 F.3d 426 (1994).
tilateral cooperation on international space law [sic], in U.S. v. Rasheed, 802 F.Supp. 312 (1992).
Proceedings of the International Institute of Space Law, The U.S. v. Red Feather, 392 F.Supp. (1975)
Hague, the Netherlands: Eleven (2011), p. 228. U.S. v. Yunis, 924 F.2d 1086 (1991).
2016 by Taylor & Francis Group, LLC
Remote Sensing Law 621
Vsquez, F.R. and S.C. Lara, What lawyers need to know about Williams, M., Satellite evidence in international institutions,
science to effectively make and address laws for remote in R. Purdy and D. Leung (eds.), Evidence from Earth
sensing and environmental monitoring, The Journal of Observation Satellites, Leiden, the Netherlands: Martinus
Space Law, 36 (2010) 365. Nijhoff Publishers (2013), p. 195.
Vsquez, M.S., Cosmic International Law, Detroit, MI: Wayne Wright, M., The use of remote sensing evidence at trial in the
State University Press (1965), p. 164. United StatesOne State Court Judges observations,
Vereschetin, V.S. and G.M. Danilenko, Custom as a source of in R. Purdy and D. Leung (eds.), Evidence from Earth
international law of outer space, The Journal of Space Law, Observation Satellites, Leiden, the Netherlands: Martinus
13 (1985) 22. Nijhoff Publishers (2013), p. 313.
von der Dunk, F.G., Outer space law principles and privacy, Wrynn v. U.S., 200 F.Supp. 457 (1961).
in R. Purdy and D. Leung (eds.), Evidence from Earth Zhao, Y., Regulation of remote sensing activities in Hong Kong:
Observation Satellites, Leiden, the Netherlands: Martinus Privacy, access, security, copyright and the case of Google,
Nijhoff Publishers (2013), p. 243. The Journal of Space Law, 36 (2010) 547.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
XIV
Summary
31 r emote Sensing Data Characterization, Classification, and Accuracies: Advances of the Last 50Years
anda Vision for the Future Prasad S. Thenkabail ...............................................................................................................625
Acronyms and Definitions Remote-Sensing Satellites and Sensors Fundamentals of Remote Sensing for Terrestrial
Applications Overview of Satellite Image Radiometry Post-launch Radiometric Calibration of Optical Satellite
Sensors Normalization of Remotely Sensed Data Satellite data degradations and their impacts on high level
products Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced Vegetation Indices Toward
Standardization of Vegetation Indices Digital Image Processing: Methods and Techniques Urban Image
Classification Methods and Approaches Image Classification Methods in Land Cover and Land Use and Cropland
Studies Hyperspectral Image Processing Methods and Approaches Thermal Infrared Remote Sensing: Principles
and Theoretical Background Object Based Image Analysis (OBIA): Evolution and State of the Art Geospatial
Data Integration in OBIA and Implications on Accuracy and Validity Image Segmentation Algorithms for Land
Categorization LiDAR Data Processing and Applications Change Detection Geoprocessing, Workflows, and
Provenance Toward Democratization of Geographic Information GIScience Object-Based Regionalization for
Policy-Oriented Partitioning of Space Global Navigation Satellite Systems GNSS Reflectometry for Ocean and Land
Applications GNSS and Wide Array of Applications Crowdsourcing in Remote Sensing and Spatial Technologies
to Study Planet Earth Cloud Computing and Remote-Sensing Data Google Earth for Earth Sciences Map
Accuracies Remote-Sensing Law or Space Law Acknowledgments References
623
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
31
Remote Sensing
DataCharacterization,
Classification,and Accuracies:
Advances of the Last 50Years
and a Vision for the Future
Acronyms and Definitions ................................................................................................................ 626
31.1 Remote-Sensing Satellites and Sensors ............................................................................... 626
31.2 Fundamentals of Remote Sensing for Terrestrial Applications .......................................629
31.3 Overview of Satellite Image Radiometry .............................................................................629
31.4 Post-launch Radiometric Calibration of Optical Satellite Sensors ..................................632
31.5 Normalization of Remotely Sensed Data .............................................................................632
31.6 Satellite data degradations and their impacts on high level products ............................ 634
31.7 Inter- and Intrasensor Spectral Compatibility and Calibration of the Enhanced
Vegetation Indices .................................................................................................................. 634
31.8 Toward Standardization of Vegetation Indices...................................................................635
31.9 Digital Image Processing: Methods and Techniques .........................................................637
31.10 Urban Image Classification Methods and Approaches .................................................... 638
31.11 Image Classification Methods in Land Cover and Land Use and Cropland Studies... 639
31.12 Hyperspectral Image Processing Methods and Approaches ........................................... 640
31.13 Thermal Infrared Remote Sensing: Principles and Theoretical Background ............... 642
31.14 Object Based Image Analysis (OBIA): Evolution and State of the Art ........................... 642
31.15 Geospatial Data Integration in OBIA and Implications on Accuracy and Validity .... 643
31.16 Image Segmentation Algorithms for Land Categorization.............................................. 643
31.17 LiDAR Data Processing and Applications .......................................................................... 644
31.18 Change Detection ................................................................................................................... 647
31.19 Geoprocessing, Workflows, and Provenance ..................................................................... 648
31.20 Toward Democratization of Geographic Information ..................................................... 648
31.21 GIScience.................................................................................................................................. 650
31.22 Object-Based Regionalization for Policy-Oriented Partitioning of Space .....................652
31.23 Global Navigation Satellite Systems......................................................................................653
31.24 GNSS Reflectometry for Ocean and Land Applications ...................................................653
31.25 GNSS and Wide Array of Applications ................................................................................653
31.26 Crowdsourcing in Remote Sensing and Spatial Technologies to Study Planet Earth.... 655
31.27 Cloud Computing and Remote-Sensing Data .....................................................................655
31.28 Google Earth for Earth Sciences ...........................................................................................657
31.29 Map Accuracies ....................................................................................................................... 658
Prasad S. Thenkabail 31.30 Remote-Sensing Law or Space Law .......................................................................................659
United States Geological Acknowledgments .............................................................................................................................. 660
Survey (USGS) References............................................................................................................................................ 660
625
2016 by Taylor & Francis Group, LLC
626 Remotely Sensed Data Characterization, Classification, and Accuracies
Acronyms and Definitions SMTs Spectral Matching Techniques
SST Sea Surface Temperature
ALI Advanced Land Imager SODAR Sonic Detection and Ranging
ALOS Advanced Land Observing Satellite SONAR Sound Navigation and Ranging
ANN Artificial Neural Networks SPOT Systme Pour lObservation de la Terre
ASPRS American Society for Photogrammetry and SRTM Shuttle Radar Topographic Mission
Remote Sensing SVM Support Vector Machines
AVHRR Advanced Very High Resolution Radiometer TIN Triangular Irregular Network
CDMA Code Division Multiple Access TIR Thermal Infrared
CEOS Committee on Earth Observation Satellites TM Thematic Mapper
CHRIS Compact High Resolution Imaging Spectrometer TOA Top of Atmosphere
CYGNSS Cyclone Global Navigation Satellite System UAS Unmanned Aircraft Systems
DEM Digital Elevation Model UAV Unmanned Aerial Vehicles
DESDynI Deformation, Ecosystem Structure and Dynamics VGI Volunteered Geographic Vegetation
of Ice VHRI Very High Resolution Imagery
DSM Digital Surface Model VI Vegetation Index
EO Earth Observation
ERTS Earth Resources Technology Sateelites This chapter provides a summary of each of the 30 chapters
ESRI Environmental Systems Research Institute in Remotely Sensed Data Characterization, Classification, and
ETM+ Enhanced Thematic Mapper+ Accuracies of the Remote-Sensing Handbook. The topics cov-
EVI Enhanced Vegetation Index ered in the chapters of Remotely Sensed Data Characterization,
FGDC Federal Geographic Data Committee Classification, and Accuracies include: (a) satellites and sen-
GEOBIA Geographic Object-based Image Analysis sors; (b) remote-sensing fundamentals; (c) data normaliza-
GIMMS Global Inventory Modeling and Mapping Studies tion, harmonization, and standardization; (d) vegetation
GIS Geographic Information System indices (VIs) and their within and across sensor calibration;
GLONASS Global Orbiting Navigation Satellite System (e) image classification methods and approaches; (f) change
GNSS Global Navigation Satellite Systems detection; (g) integrating remote sensing with other spa-
GOES Geostationary Operational Environmental tial data; (h) Global Navigation Satellite Systems (GNSS); (i)
Satellite crowdsourcing; (j) cloud computing; (k) Google Earth remote
GEOSS Global Earth Observing System of Systems sensing; (l) accuracy assessments; and (m) remote-sensing law.
GLAS Geoscience Laser Altimeter System Under each of the aforementioned broad topics, there are one
GNSS Global Navigation Satellite System or more chapters. For example, there are nine chapters under
GPS Global Positioning System image classification methods and approaches. In a nutshell,
ICESat Instrument aboard the Ice, Cloud, and land these chapters provide a complete and comprehensive over-
Elevation view of these critical topics, capture the advances over the last
ISODATA Iterative Self-organizing Data Analysis Technique 50years, and provide a vision for further development in the
LiDAR Light Detection and Ranging years ahead. By reading this summary chapter, a reader can
LST Land Surface Temperature have a quick understanding of what is in each of the chapters of
MODIS Moderate Resolution Imaging Spectroradiometer this book, see how the chapters interconnect and intermingle,
NDVI Normalized Difference Vegetation Index and get an overview on the importance of various chapters
NOAA National Oceanic and Atmospheric in developing complete and comprehensive knowledge of the
Administration remote-sensing data characterization, classification, and accu-
NPOESS National Polar-orbiting Operational racies. These chapters, together, capture the advances of the
Environmental Satellite System last 50years and provide a vision for the future.
NPP NPOESS Preparatory Project
OBIA Object Oriented Image Analysis
OMI Ozone Monitoring Instrument 31.1 Remote-Sensing Satellites
PALSAR Phased Array type L-band Synthetic Aperture
and Sensors
Radar
PROBA Project for On Board Autonomy Today, remote-sensing satellites and sensors are ubiquitous.
RS Remote Sensing Spatial data from these sensors are used to understand, query,
RADAR Radio Detection and Ranging measure, model, monitor, store, and map a wide array of
SAR Synthetic Aperture Radar land, water, and atmospheric features and resources. Sensors
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 627
continuously and repeatedly collect data in various spectral, Array type L-band Synthetic Aperture Radar (PALSAR),
spatial, radiometric, and temporal resolutions. Indeed, for many Japanese Earth Resources Satellite (JERS) Synthetic Aperture
applications, remote sensing has become indispensable and Radar(SAR)). Less frequently collected data come from hyper-
often the most important data source. Its strengths are obvi- spectral (e.g., Earth Observation [EO]-1 Hyperion, Compact
ous: repeated coverage of the planet at different scales; ability to High Resolution Imaging Spectrometer (CHRIS) Project for On
collect data in wide array of spatial, spectral, radiometric, and Board Autonomy (PROBA)). These are either Sun synchronous
temporal resolutions; collecting consistent data without human or geostationary, and collected data are wide array of spatial,
subjectivity; and ability to collect data from inaccessible parts of spectral, radiometric, and temporal resolutions that allow rou-
the world without any political boundary restrictions. However, tine, consistent, and repeated study of the planets land, water,
the complexities of data collection from satellites and sensors and atmosphere. Chapter 1 captures a wide range of these sat-
are many. They vary due to orbital distances, Sun angles, view ellites and sensors detailing their characteristics. The United
angles, spectral bands, bandwidths, spatial resolutions, sensor States, France, India, China, Germany, the United Kingdom,
platforms, type of sensors (e.g., active or passive) (e.g., Figure Brazil, and several others (see Chapter 1) now routinely launch,
31.1), radiometry, data download links, ground stations, calibra- operate, and share remote-sensing data. With the 1992 Land
tion of data collected, normalization of data collected, geometric Remote Sensing Policy Act of United States, which permitted
registration, data delivery platforms, and a host of other factors. private companies to enter the satellite imaging business, pri-
As a scientist involved in the study of planet Earth, apart from vate companies like WorldView Imaging Corporation partici-
the full understanding of the remotely sensed (RS) data, it is pated in launching and operating commercial Earth-observing
equally important to understand the process involved in design, satellites, and very recently, it has picked up tremendously.
launch, and operation of satellites and sensors. Especially, given These initiatives have helped in obtaining very high spatial and
many countries operate their own satellites and sensors, and spectral resolution satellite imageries. Increasingly, UASs are
there are many private players. also being used for carrying sensors of various kinds.
Chapter 1 by Dr. Sudhanshu Panda etal. is focused on pro- In immediate years ahead, as we enter the second decade of
viding a comprehensive overview of wide array of satellites and the twenty-first century, we are likely to have numerous remote-
sensors operated by various governments and private entities. sensing satellites and sensors from government and several
The idea of the chapter is to provide readers with a full under- private operators. There is also likely greater role for UASs and
standing of the variety of remote-sensing data characteristics UAVs. Compared to conventional remote sensing that maps and
as well as their sources. The chapter provides a progressive analyzes the Earth, water, and atmospheric phenomena, impor-
development of remote sensing from the early days of airborne tance of RAdio Detection And Ranging (RADAR), and SOund
remote sensing, mainly developed, perfected, and used during NAvigation Ranging (SONAR), Light Detection and Ranging
the World Wars I and II, to the more modern hyperspectral, (LiDAR), remote-sensing technology is gaining ground in the
light detection and ranging (LiDAR), and Unmanned Aircraft twenty-first century due to their unconventional application of
Systems (UAS) sensors. The early era of remote sensing was real-world problem solving like accurate ground topography
mainly for military usage. Subsequently, owing to its strengths mapping and tree heigh measurement, or other Earth object
in mapping and survey, civilian applications of remote sensing identification and measurements, accurate weather forecast-
grew mainly through airborne, ground-based, and platform- ing (e.g., Next-Generation Radar (NEXRAD or Nexrad) and
mounted sensors. With the launch of Sputnik and the National Doppler SOnic Detection And Ranging (SODAR)), and deep-sea
Oceanic and Atmospheric Administrations Advanced Very- exploration. The private enterprise is also dwelling into newer,
High-Resolution Radiometer (NOAA-AVHRR), spaceborne more revolutionary, approaches and techniques of capturing
remote sensing began its initial steps in late 1950s and early remote-sensing data. These initiatives come from companies
1960s. With the launch of Earth resources technology sat- such as the Planet Labs Inc., the Skybox Imaging Inc. (recently
ellites (ERTS-1 later renamed Landsat) in the year 1972, purchased by Google), and the Boeing company recently receiv-
remote sensing of the Earth system science truly evolved. ing its first commercial order for the 502 Phoenix small satellite
More recently or currently, data on planet Earth are rou- from HySpecIQ of Washington, DC. The satellites planned by
tinely gathered by numerous satellite sensors that include very these initiatives will allow for gathering spectral signatures, or
high spatial resolution (e.g., IKONOS, QuickBird, GeoEye, submeter data on a continuous basis, or even videos, and many
RapidEye) and multispectral moderate to coarse resolution other innovative ways.
(e.g., Landsat, Sentinel, Resourcesat, Moderate-Resolution Further, the chapter also provides the details of the principles
Imaging Spectroradiometer [MODIS] in Terra/Aqua, AVHRR, of electromagnetic spectrum in remote sensing. It also describes
Geostationary Operational Environmental Satellite (GOES), the future of remote sensing, that is, transformation of com-
The National Polar-orbiting Operational Environmental puter/lab imaging spectrometry technology to field-imaging
Satellite System (NPOESS) Preparatory Project or NPP, spectrometry to determine crop, vegetation, fruit, and other
Radarsat, Advanced Land Observing Satellite (ALOS) Phased physical object quality.
2016 by Taylor & Francis Group, LLC
2016 by Taylor & Francis Group, LLC
628
8 km, AVHRR 1 km, AVHRR 57 m, MSS 60 m, ETM+
Remotely Sensed Data Characterization, Classification, and Accuracies
1 km
30 m, ETM+ 4 m, IKONOS 1 m, IKONOS Single pixel
0.6 0.6
Sudan Sudan 0.45
ETM + NDVI = 0.8694*IK ONOS NDVI 0.1908 savanna ETM-NCS
Reffecture factor (no units)
savanna ETM + NDVI = 0 .852 * IKONOS NDVI 0 .1943
R2 = 0.68 2
ETM--DS
ETM + NDVI (4 m resampled Pixel)
R =0.67 ETM-HF
ETM + NDVI (30 m actnal pixel)
IKONOS-XCS
0.4 IKONOS-DS
Derived 0.4 Derived 0.3 IKONOS-HF
savanna savanna
0.2
Humid Humid 0.15
forests 0.2 forests
0 All three All three 0
0 0.2 0.4 0.6 0 ecoregions ecoregions
7
0
nd
nd
nd
nd
nd
nd
0.2 Linear(all
ba
ba
ba
ba
ba
ba
Linear(all 0
0 0.2 0.4 0.6 three
three Wave bands (#)
ecoregions) ecoregions)
0.4 0.2
IK ONOS NDVI (30 m resampled pixel) IKONOS NDVI (4m actual pixel)
FIg u r e 31.1 The top two rows of images depict the impact of spatial resolution on information content. An area in National Oceanic and Atmospheric Administrations Advanced Very-
High-Resolution Radiometer (NOAA-AVHRR) 8km resolution (top left) is geo-linked to images acquired over the same areas from various other resolutions: 1km NOAA-AVHRR, 57 m
Landsat MSS, Landsat ETM+ thermal image of 60 m, Landsat ETM+ multispectral image of 30 m, IKONOS multispectral of 4 m, IKONOS panchromatic of 1 m. The leftmost bottom row plot
shows the intersensor relationships between the Landsat ETM+ 30 m normalized difference vegetation index (NDVI) and IKONOS NDVI (resampled to 30 m) derived from images acquired
about the same dates. In the last row, the left and the center plots show the intersensor relationships between the Landsat ETM+ NDVI (resampled to 4 m) and IKONOS 4 m NDVI derived
from images acquired about the same dates. The line chart shows the normalized reflectance derived from various sensors that collected data over various African sites.
Remote Sensing DataCharacterization, Classification,and Accuracies 629
31.2 Fundamentals of Remote Sensing Developments and trends mentioned earlier require overlapping
missions and well- (inter-) calibrated sensors. To fulfill these
for terrestrial Applications requirements, international cooperation and data sharing will
Early remote sensing was primarily based on images and data gath- be increasingly important. Joint commitments between inter-
ered from aircrafts. This had limited scope, given the complexities of governmental, international, and regional organizations there-
covering large areas and the cost factor. Thereby, land remote sens- fore are a powerful means of stabilizing new space initiatives
ing from space commenced in earnest once the first Landsat was and hence data availability and continuity. Even though great
launched in 1972. Even though NOAA AVHRR and Sputnik pre- advances have been made as a result of use of data from multiple
ceded Landsat and the data from these satellites were used for land sensors using advanced methods, techniques, and modern com-
applications, it is with the launch of Landsat that the real momen- puting power, there is still a long way to go in reducing uncertain-
tum in terrestrial applications of spaceborne remote sensing began. ties of various terrestrial themes studied using remote sensing.
Oppelt etal. in Chapter 2 provide a brief overview on the remote-
sensing platforms as well as the electromagnetic spectrum from 31.3 overview of Satellite
where these data are captured. Limitations of early visual interpreta- image Radiometry
tions are pointed out. Then, the chapter provides a focus on typical
applications at local, regional, and global levels as well as specialized A primary goal of remote-sensing data acquired from the wide
thematic applications. What is clear in this chapter is the following: array of EO satellites is to establish protocols and mechanisms
for delivering calibrated and normalized data over the entire
1. Specific applications (e.g., urban, vegetation, water, terrain) planet. This allows analyses of data over space and time with
have specific data needs (e.g., thermal, normalized difference minimal and\or known uncertainties.
vegetation index [NDVI], near-infrared, microwave). For In the early years, remote-sensing applications of EO satellites
example, one cannot study species using coarse spatial and/ were often done using raw digital counts (Level-0 data) without
or coarse spectral resolutions. Similarly, a particular sensor physical units. Great advances have been made over the years.
performs better than another in certain applications, for Currently, remote-sensing data are analyzed routinely after cali-
example, radar sensors are suitable to derive digital elevation bration to their physical units (e.g., radiance expressed in W m2
models. So knowledge of sensor performance for specific ter- sr1m1) using appropriate radiometric, geometric, spectral, and
restrial applications is required. atmospheric corrections, as explored in detail in Chapters 3 through
2. Multisensor approaches (e.g., combine Landsat with 7 of this book. These calibrations lead to data being converted to
Sentinel or optical with microwave sensors) become cru-
1. At-sensor or top-of-atmosphere (TOA) radiance and/or
cial for improved accuracies in mapping, modeling, and
reflectance (Level-1)
monitoring complex terrestrial themes.
2. Surface radiance or reflectance (Level-2)
3. For a value adding beyond the sole physical measurement
of remote-sensing data, multidisciplinary approaches using Currently, almost all EO data are provided in one of these
remote-sensing data in combination with other data sources units, typically scaled for digital processing. Yet calibrations
became increasingly important. Figure 31.2 presents an remain challenging and complex given that satellite sensors are
example for such an approach to monitor coastal vegetation launched by various space agencies of different nations as well as
combining different analysis techniques, biological mapping, private initiatives, which have widely varying geometric, radio-
hydrography, and a geographic information system. metric, and spectral performance characteristics. Chapter 3 by
4. Best results are obtained when studies incorporate various Teillet provides a comprehensive summary of the wide array of
dimensions (e.g., spectral, spatial, temporal, radiometric) radiometric factors involved in EO data. However, not all these
of RS data. parameters are used in developing standard products, produced
5. Data policy is changing; free and open data policy sup- through various algorithms, due to difficulty and/or uncertainty
ports existing trends to provide remote-sensing data (e.g., in obtaining some or many of these parameters.
Landsat, MODIS) and standardized products (e.g., global In addition, intersensor calibrations (cf. Chapter 4) further
land cover maps) free of charge. facilitate the use of data from multiple sensors from multiple
6. Global networks have formed to develop robust methodol- sources.
ogies and services for decision support tools and commu- In order to overcome these difficulties, the Committee on
nity portals to meet specific needs of stakeholders, scientists, Earth Observation Satellites (CEOS; http://www.ceos.org/),
and user communities (e.g., Global Earth Observation under the umbrella of the Group on Earth Observations (GEO;
System of Systems [GEOSS], Global Forest Watch). http://www.earthobservations.org/index.shtml), established an
7. Continuous and comparable data are of high importance international calibration/validation (Cal/Val) working group
to develop standardized measures, especially for global (http://calvalportal.ceos.org/), where many of these issues are
monitoring. To increase acceptance of standardized prod- discussed and implementation recommendations are provided.
ucts and to ensure their widespread application, standards The ultimate goal of these efforts is to create an effective Global
for accuracy assessment will have to be set. Earth Observation System of Systems (GEOSS) that will deliver
2016 by Taylor & Francis Group, LLC
630 Remotely Sensed Data Characterization, Classification, and Accuracies
750'0"E 752'0"E 754'0"E 756'0"E
5414'0"N
5414'0"N
0 0.5 1 2
Macroalgae
habitats
km
5413'0"N
5413'0"N
5412'0"N
5412'0"N
Legend
Helgoland
Shallow eu- and sublitoral
Fucales
Laminariales shallow
5411'0"N
Green algae
5411'0"N
Deep sublitoral
Laminariales deep
Bathymetry
>=0 m
02 m
25 m
5410'0"N
510 m
5410'0"N
Water depth (m) based
on sea chart level zero
(lowest astronomical tide)
750'0"E 752'0"E 754'0"E 756'0"E
FIg u r e 31.2 Map of macroalgae habitats in the coastal waters of Helgoland (North Sea, Germany); the map was generated in the scope of an
interdisciplinary research project of geographers, biologists, and remote-sensing experts. The data sources include airborne hyperspectral data
(AISAeagle), floristic mappings, diving mappings, and bathymetry information from the German Federal Maritime and Hydrographic Agency.
The remote-sensing data have been analyzed separately for algae cover in the eu- and sublitoral (the methods are described in Oppelt etal. (2012)
[eulitoral] and Uhl etal. (2013) [sublitoral]). Since analysis of optical remote-sensing data is limited by visibility depth (approximately 5 m dur-
ing airborne data acquisition), the analysis is valid up to a depth of ~4 m. All data have been combined in a GIS, which enables validation of
remote-sensing results and mappings. The resulting user accuracy is 97.5% (number of validation points = 45), and GIS analysis also revealed
that large sublitoral brown algae (kelp) cover 248.31 ha of the area with water depth 5 m. (From Uhl, F. et al., KelpMapDevelopment of
an EnMAP approach to monitor sublitoral marine macrophytes [KelpMapEntwicklung eines EnMAP Verfahrens zur Bestimmung von sub-
litoralen marinen Makrophyten], Final report of research project FKZ: 50EE1020, funded by the German Federal Ministry of Economy and
Technology (BMWi), 36pp., 2014.)
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 631
EO data that are well calibrated, reliable, and consistent and have 1. Space agencies and commercial data providers do not yet
comprehensive coverage over space and time. provide calibrated data
Current approaches to data delivery by many space agencies 2. Calibration coefficients provided are incomplete (e.g., do
and commercial data providers are already moving in the direc- not allow for all the factors listed in Chapter 3)
tion of surface reflectance products, such as for MODIS (http:// 3. Gaps in intersensor calibration exist (e.g., the Systme Pour
modis.gsfc.nasa.gov/data/dataprod/dataproducts.php?MOD_ lObservation de la Terre [SPOT] High Resolution Visible
NUMBER=09) and Landsat 8 (http://earthexplorer.usgs.gov/). (HRV) series of satellite data are not adequately intercali-
The approaches and methods are discussed in detail in Chapters brated with the Landsat series of satellites; Goward etal.,
3 through 7. Some difficulties yet to be overcome include when 2012, Price, 1987) (Figure 31.3)
L7 ETM+ and IRS-P6 AWiFS TOA reflectance, Band 2 L7 ETM+ and IRS-P6 AWiFS TOA reflectance, Band 3
0.5 0.5
AWiFS- BD Quad AWiFS- BD Quad
y = 1.1469x 0.0145 y = 1.1693x 0.0037
11 line
0.4 R2 = 0.9855 11 line R2 = 0.9801
P39R3537 0.4
P39R3537
L7 ETM+ TOA reflectance
L7 ETM+ TOA reflectance
P39R32 P39R32
P41R36 P41R36
0.3 0.3
P36R3539 P36R3539
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
IRS-P6 AWiFS TOA reflectance IRS-P6 AWiFS TOA reflectance
L7 ETM+ and IRS-P6 AWiFS TOA reflectance, Band 4 L7 ETM+ and IRS-P6 AWiFS TOA reflectance, Band 5
0.5 0.5
AWiFS- BD Quad AWiFS-BD Quad
y = 1.1304x 0.0015 y = 1.0311x 0.0125
1x1 line
0.4 11 line R2 = 0.973 0.4 R2 = 0.9414
P39R3537 P39R3537
L7 ETM+ TOA reflectance
L7 ETM+ TOA reflectance
P39R32 P39R32
P41R36 P41R36
0.3 0.3
P36R3539 P36R3539
0.2 0.2
0.1 0.1
0.0 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
IRS-P6 AWiFS TOA reflectance IRS-P6 AWiFS TOA reflectance
FIg ur e 31.3 Comparison of top-of-atmosphere reflectance measurements from Landsat 7 ETM+ and AWiFS. Each data point on these plots
represents an ensemble average of all pixels in a defined ROI for a given day and spectral band. (From Goward, S.N. etal., Remote Sens. Environ.,
123, 41, August 2012, ISSN: 0034-4257, http://dx.doi.org/10.1016/j.rse.2012.03.002. Accessed JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
632 Remotely Sensed Data Characterization, Classification, and Accuracies
31.4 Post-launch Radiometric 31.5 normalization of Remotely
calibration of optical Sensed Data
Satellite Sensors Modern remote sensing involves using data/imagery from mul-
The scientific integrity of satellite sensor data can only be main- tiple dates and from multiple sensors. The only way such data can
tained if the sensors are well calibrated. Once a sensor is launched be used in scientific investigations is through appropriate cali-
into orbit, postlaunch calibrations need to come into effect and be bration and normalization. Sensors, once launched, go through
reiterated for the entire life of the sensor, which is often from a few degradation over time. They also operate in different seasons and
years to even decades. Chapter 4 by Teillet and Chander provides climates. Various sensor characteristics such as push broom (or
an exhaustive state-of-the-art review of postlaunch radiometric along track) versus whisk broom (or across track), solar zenith
calibration of optical sensors. The standard and well-established angles, field of view, and angle of the sensor also add to the need
approach adopted by various agencies and researchers for post- for standardization and normalization of the sensors. In Chapter
launch calibration is to use built-for-purpose onboard systems 5, Dr. Rudiger Gens and Dr. Jordi Cristbal Rossell address these
initially and adjust calibration coefficients over time using suit- issues and provide a clear approach to normalization of sensor
able reference targets vicariously. The international community characteristics. They recommend
has adopted and recommends the use of approximately 20 such 1. Absolute normalization based on radiative transfer mod-
targets, selected with specific criteria in mind (cf. Chapter 4). els that account for atmospheric and sensor characteristics
Currently, the best practice is for optical sensors to be cali- 2. Relative normalization where one or more images are cor-
brated vicariously using rected to a standard master image
1. Eight CEOS-endorsed reference instrumented standard They illustrate these two normalization methods separately for
sites based in the deserts and playas of the world optical, thermal, and radar imagery.
2. Six CEOS reference pseudoinvariant calibration sites Instruments are always calibrated prelaunch. Then, vicari-
located in various parts of the Sahara desert ous calibration takes place-throughout the life of the sensor on
The multigeneration Landsat series of sensors from 1972 to the board either using an onboard calibration reference unit or by
present are recalibrated postlaunch using vicarious targets and focusing the sensor to time-invariant sites like the full moon.
are known to have uncertainties of within 5%10% across sen- Factors that need to be considered during calibration include
sors and spectral bands. The calibration efforts in the contexts (Figure 31.5)
of GEO, GEOSS, and CEOS are important and are discussed in 1. Satellites
Chapter 4 by Teillet and Chander. Well-calibrated sensors allow a. Height of acquisition (e.g., 500, 700, 36,000km above
multisensor data continuity, facilitating long-term studies of Earth)
Earth and environment (Figure 31.4). b. Orbital parameters
r2 = 0.88 2011 r2 = 0.7
0.36 2004
0.28 2002 2001
2010 2005
GIMMS-3g NDVI
2008
0.34
GIMMS-3g NDVI
0.26 2006
2000 2009
2011 2003 2007
2000 2007
2001
0.24 2003 0.32
2002 2005
2006
2009
0.22
2008 0.30 2010
2004
0.20
0.2 0.22 0.24 0.26 0.28 0.30
(a) MOD13C2 NDVI (b) MOD13C2 NDVI
FIg u r e 31.4 Linear regression between third-generation Global Inventory Modeling and Mapping Studies 8km and Moderate-Resolution
Imaging Spectroradiometer Terra normalized difference vegetation index (MOD13C2) at 5.6km resolution over (a) the Gourma (Mali) region
and (b) the Fakara (Niger) region. (From Dardel, C. et al., Remote Sens. Environ., 140, 350, January 2014, ISSN: 0034-4257, http://dx.doi.
org/10.1016/j.rse.2013.09.011. Accessed JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 633
Normalizations for atmospheric effects
factors to consider
Sun
Satellite
Radiance (W m2 sr1 m1) @
TOA = Radiance leaving the
Ground * Transmission
Energy off target factor + path radiance.
Reflectance =
Energy from the source Top of atmosphere (TOA)
Solar zenith angle Zone of
trouble for
Path radiance RS date
acquisition!
Reflectance
Target @ Ground Note: Transmission factor assumed 1 expect in 6S model.
Also in arid and semi-arid regions, it is anyway nearly 1.
FIg ur e 31.5 An overview of factors involved in normalization of remotely sensed data. This generic approach can be applied across any satellite
sensor system.
2. Sensors The data are first normalized to TOA reflectance using the fol-
a. Radiometry lowing common approach, irrespective of the sensors involved:
b. Bandwidth
Energy off Target
c. Optics/design Reflectance ( % ) =
d. Degradation over time Energy from theSource
e. Nadir, off-nadir viewing
3. Solar flux or irradiance
=
(
Radiance W m 2 sr 1 m 1 ) 100
a. Function of wavelength (
Irradiance W m sr m 2 1 1
)
4. Sun
a. Sun elevation at time of acquisition
5. SunEarth Ld 2
Reflectance (%) =
a. Distance between Earth and Sun ESUN cos s
6. Stratosphere or atmosphere
a. Ozone, water vapor, haze, aerosol where
b. Path radiance TOA reflectance (%) is the at-satellite exoatmospheric reflectance
7. Surface of Earth L is the radiance (W m2 sr1 m1)
a. Topography d (dimensionless) is the Earth-to-Sun distance in astronomic
8. Seasons units at the acquisition date (see Markham and Barker, 1987)
a. EarthSun distance ESUN is the irradiance (W m2 sr1 m1) or solar flux
(Neckel and Labs, 1984)
Normalization steps involve three major steps:
s (degrees) is the solar zenith angle
Digital number (unit less) to radiance (W m2 sr1m1)
Radiance to top of the atmosphere reflectance (also referred to Note: s is the solar zenith angle in degrees (i.e., 90 minus the
as at-satellite exoatmospheric (or apparent or) reflectance (%) Sun elevation or Sun angle when the scene was recorded as given
TOA reflectance to surface reflectance (%) in the image header file).
2016 by Taylor & Francis Group, LLC
634 Remotely Sensed Data Characterization, Classification, and Accuracies
The aforementioned data are then converted to ground 2. There are systematic differences between MODIS Terra
reflectance through atmospheric correction as described in and Aqua C5derived information at certain places. This
Chapter 5. Atmospheric corrections are part of normalization lead to recommendation that for interannual variability
and eliminate or reduce path radiance resulting from haze (e.g., studies, it is better to use MODIS Aqua data rather than
thin clouds, dust, harmattan haze, aerosols, ozone, and water MODIS Terra data before new release with improved cali-
vapor). Atmospheric correction methods include (a) dark object bration stability is available.
subtraction technique (Chavez, 1988); (b) improved dark object 3. Annual degradation of Landsat Enhanced Thematic
subtraction technique (Chavez, 1989); (c) radiometric normal- Mapper Plus (ETM+) are less than 0.4% and are well
ization technique, bright and dark object regression (Elvidge characterized.
etal., 1995); and (d) 6S model (Vermote etal., 2002). As pointed 4. Landsat Thematic Mapper (TM) showed trends in sensor
out by Dr. Rudiger Gens and Dr. Jordi Cristbal Rossell in response over time. So new method that involved desert
Chapter 5, the main difficulty in atmospheric radiosondes are sites and cross calibration with ETM+ was preferred for
usually not available at the time of satellite pass, and data from better understanding Landsat TM calibration.
a single atmospheric radiosonde or aerosol robotic network are
inadequate for large areas covered by single swath (e.g., AVHRR, These inferences, drawn from Chapter 6, only depict some of
MODIS, Landsat) of most of the satellites. Even when the swath the issues. These issues clearly indicate the complexities as
are not big (e.g., IKONOS, QuickBird, GeoEye), local weather well the critical need for well-established calibration of vari-
data may not be representative since these networks are often ous orbiting sensors. Images are also normalized taking time-
quite away from the images. A classic paper on intersensor cali- invariant sites into consideration (e.g., Figure 31.6). In such a
bration involving multiple sensors is provided by Chander etal. case, an image is considered reference, and all other images of
(2009). So most uncertainty in atmospheric correction of imag- the same area are normalized to the reference image by tak-
ery is due to lack of detailed, reliable input data for themodels. ing time-invariant sites. For example, in Figure 31.6, Wei etal.
(2014) selected the features that emit temporally stable, and
invariant nighttime lights that are used as pseudoinvariant fea-
31.6 Satellite data degradations and their tures (PIFs) for the Defense Meteorological Satellite Program
Operational Linescan System (DMSP-OLS) image normaliza-
impacts on high level products tion. The fully developed and stable urban districts, where no
One of the greatest challenges for terrestrial remote sensing significant further development and changes occur, satisfy the
has always been to ensure well-calibrated data from within and stable nighttime light requirement and can be used as PIFs
between sensors throughout its data acquisition. All sensors are (Wei etal., 2014).
precalibrated before launch and continue to be calibrated dur-
ing their life span in space vicariously and/or on board. Chapter 31.7 inter- and intrasensor Spectral
6 by Dongdong Wang discusses three issues affecting stability of compatibility and calibration of
calibration: the enhanced Vegetation indices
1. Radiometric calibration with onboard calibration systems Long-term consistent global studies from satellite sensors are
2. Radiometric calibrations without onboard systems only feasible if there is within and between sensor calibration
3. Data degradation due to orbit drift of Sun-synchronous leading to clear, well-understood relationships between these
satellites sensors. Establishing relationships is required for both within
There are very many factors influencing sensor degradation a family of sensors such as Landsat series and AVHRR series
such as orbital drift, solar zenith angle, and sensor degradation and across family of sensors such as Landsat versus MODIS ver-
in space due to adverse conditions, clouds, haze, shadows, and sus AVHRR versus IKONOS. This is because of a host of issues
surface changes even on so-called time-invariant sites. Some of such as (a) sensor degradation over time; (b) their characteris-
these key factors are summarized in Table 6.1. What is impor- tic differences in spatial (e.g., Figure 31.1), spectral, radiomet-
tant for a terrestrial scientist to ensure is that the measurements ric, and temporal resolutions; (c) atmospheric conditions under
made by satellites are reliable and temporally stable and, if there which data are acquired; and (d) different processing algorithms
are uncertainties, those are well characterized. This important applied over time.
and fundamental first step in using satellite remote-sensing data Chapter 7 by Dr. Tomoaki Miura et al. provides the lat-
in Earth sciences is well illustrated in this chapter. For example, est understanding of intersensor spectral compatibility
the chapter summarized some significant factors of different across multiple sensors, including visible infrared imaging
sensors such as the following: radiometer suite, MODIS, second-generation global imager,
VEGETATION (VGT), AVHRR/3, AVHRR/2, and ETM+, for
1. Difference in reflectances from early AVHRR sensor two well-known vegetation indices (VIs): enhanced vegeta-
derived with different versions of calibration coefficients tion index (EVI) and EVI2. EVI uses three bands (red, near-
can be as large as 20%. infrared, and blue), whereas EVI2 does not use the blue band.
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 635
60 60
F162006
40 40
F152002
20 20
0 0
0 20 40 60 0 20 40 60
F142000 F142000
60
60
F162009
40
40
F182010
20
20
0
0 20 40 60 20 40
0 60
F142000 F142000
FIg u r e 31.6 Scatter plots of the digital numbers (DN) values of the target years against the DN values of the reference year (2000) for pixels
within desaturated pseudoinvariant features for the U.S. Defense Meteorological Satellite Program Operational Linescan System satellites. The
solid line stands for the trend line of regression equation. (From Wei, Y. etal., Landscape Urban Plan., 128, 1, August 2014, ISSN: 0169-2046,
http://dx.doi.org/10.1016/j.landurbplan.2014.04.015. Accessed JuneOctober, 2014.)
It is clear from this chapter that correlation between the same or more confounding environmental factors. Over the years,
families of sensors (e.g., AVHRR/2 vs. AVHRR/3) is typically many satellites and sensors have been launched, and their char-
high and can be reasonably well established. However, corre- acteristics can differ in many ways: spectral band placement
lation between dissimilar sensors (e.g., MODIS vs. AVHRR) (e.g., differences between broad and narrow bands and their
can be poor. Nevertheless, it is feasible to establish intersensor positions within a given spectral range), viewing characteris-
calibrations (e.g., Figure 31.7a) that will help studies beyond tics (e.g., nadir or off-nadir viewing), conditions under which
a particular sensors life span (e.g., Figure 31.7b). Chapter 7 data are acquired (e.g., Sun angle, atmospheric conditions,
results introduce potential methodologies that allow for the EarthSun distance), sensor degradation over time, process-
derivation of robust intersensor relationships. Even though ing methods, and many others (see Chapter 8 by Michael D.
Chapter 7 uses EVI and EVI2 to illustrate these intersensor Steven, Timothy J. Malthus, and Frdric Baret). Consistency
relationships, other indices like NDVI can also be used with in the integration of data from different sources and continu-
equal effect. ity in long-term studies of vegetation change on planet Earth
requires us to standardize VIs to adjust for intersystem dif-
31.8 toward Standardization ferences such as, say, Landsat, with MODIS or SPOT or IRS,
of Vegetation indices or SPOT VGT with AVHRR Global Inventory Modeling and
Mapping Studies (GIMMS) as well as intrasystem changes such
Monitoring tools such as the NDVI have been part of remote- as changes of the Landsat sensor from 1972 to the present day
sensing studies ever since pioneering work reported by Tucker (e.g., Figure 31.8).
(1979). Simple reflectance from a fixed object will vary over Such standardization will involve normalization for various
time, even if the object has not changed, due to changes in factors (e.g., as mentioned earlier), deriving VIs, and building
the measurement conditions, so that, for example, two plants relationships that will allow for consistent and stable monitoring
of identical type, biomass, and health will reflect differently if of vegetation over time and space. For example, VIs of certain
they are observed under different conditions. A number of such time-invariant areas (e.g., certain areas of Sahara desert) should
factors influence reflectivity making it difficult to conduct sci- remain the same over time, and extraneous factors such as view
entific studies over space and time using raw RS data. This diffi- angle, atmosphere, and solar elevation should have minimal
culty is overcome by using a VI: either a simple ratio derived by influence on measurements at such sites. Therefore, measure-
dividing reflectivity in the near-infrared waveband by reflec- ments over these invariant sites remain the same over time, and
tivity in the red waveband, or its equivalent, the NDVI, or a one can use such time-invariant sites to normalize and correct
more complex index involving adjustment coefficients for one for system variations.
2016 by Taylor & Francis Group, LLC
636 Remotely Sensed Data Characterization, Classification, and Accuracies
0.7
y = 0.7633 x0.0483
R2 = 0.7793
0.6
0.5
0.4
0.3
AVHRR
0.2
0.1
0.1
0.2
0.05 0.05 0.15 0.25 0.35 0.45 0.55 0.65 0.75 0.85
February 2000 March 2000 April 2000 May 2000 June 2000 July 2000 August 2000 September 2000
October 2000 November 2000 December 2000 January 2001 February 2001 March 2001 April 2001 May 2001
June 2001 July 2001 August 2001 September 2001 Year-20002001 Year-20002001
(a)
0.30
NDVI Dev
AVHRR NDVI original AVHRR NDVI
simulated from
MODIS
0.20
0.10
0.00
3
98
5
y- 2
ar 92
0
82
ar 84
5
ar 86
ar 91
ar 93
4
8
y- 9
1
nu 990
03
7
99
99
99
00
00
99
99
y-
00
98
nu 199
98
98
nu 98
9
19
nu -19
nu 19
20
ar
Ja y-1
Ja y-1
Ja ry-1
Ja y-2
Ja ry-2
Ja y-1
Ja y-1
Ja y-1
Ja y-2
Ja y-1
Ja y-1
Ja y-1
Ja y-1
Ja ry-1
1
Ja y-1
y-
Ja ry-
nu
y
y
ar
ar
ar
ar
ar
ar
ar
ar
ar
ar
ar
ar
a
a
a
Ja
a
nu
nu
nu
nu
nu
nu
nu
nu
nu
nu
nu
nu
nu
nu
nu
nu
0.10
Ja
Ja
Ja
Ja
Ja
0.20
Time (months)
Rajkot
0.30
(b)
FIg u r e 31.7 (a) Intersensor relationships between Moderate-Resolution Imaging Spectroradiometer (MODIS) 500 m derived normalized
difference vegetation index (NDVI) versus Advanced Very-High-Resolution Radiometer 8km NDVI. Relationship was developed taking data
from various years to ensure robustness. (b) The intersensor relationship developed in (a), enabled continuation of drought studies using NDVI
deviation from long-term mean (NDVIdev). From 1982 to 2000, the study was conducted using AVHRR 8km data. Beyond the year 2000, MODIS
500 m data were used. (From Thenkabail, P.S. etal., IWMI Research report # 85. Pp. 25. IWMI, Colombo, Sri Lanka, 2004. http://www.iwmi.cgiar.org/
Publications/IWMI_Research_Reports/PDF/pub085/RR85.pdf. Accessed JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 637
1.0 1.0 1.0
0.8 0.8 0.8
MODIST-NASA
MODISA-NASA
SPOT-VGT
0.6 0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
GIMMS GIMMS GIMMS
1.0 1.0 1.0
0.8 0.8 0.8
MODIST+A-BOKU
MODIST-BOKU
0.6
eMODIST
0.6 0.6
0.4 0.4 0.4
0.2 0.2 0.2
0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
GIMMS GIMMS GIMMS
FIg u r e 31.8 Example of empirical scatterplots showing intercalibration between seasonally averaged and regionally aggregated normal-
ized difference vegetation indices (NDVIs) from various systems over Kenya. Scatterplots showing intercalibration between NDVI* (season-
ally averaged NDVI and aggregated per division) derived from Global Inventory Monitoring and Modeling System (x-axis) against NDVI*
from each of the other NDVI products such as the Systme Pour lObservation de la TerreVEGETATION and various Moderate-Resolution
Imaging Spectroradiometer data. Each plot contains a total of 1512 data points (84 divisions 9years 2 seasons/year). (From Vrieling, A.
etal., Int. J. Appl. Earth Observ. Geoinform., 28, 238, May 2014, ISSN: 0303-2434, http://dx.doi.org/10.1016/j.jag.2013.12.010. Accessed June
October, 2014.)
In Chapter 8, Drs. Steven, Malthus, and Baret start with how enhancement, (4) registration and reprojections, (5) mosaicking
VIs are computed, what factors influence variation in VIs, and and megafile creation, (6) analysis, (7) classifications, and (8)
VI intercalibration approaches. The chapter then provides equa- accuracy assessments. Fundamental methods and approaches
tions for cross-sensor intercalibration, based on precise simu- (see Jensen, 1996; Lillesand etal., 2008; Richards and Xiuping,
lations of system spectral responses from a common dataset of 2006) have remained firm over the years and are a must for any-
field observations. Then, the chapter lays out certain universal one wanting to master the remote-sensing data-processing meth-
principles for VI standardizations. The principles include that ods and techniques. In recent years, numerous advances have
the standardization should be (a) applicable to VIs from all sys- been made in digital image processing (Nagamalai etal., 2011)
tems, (b) fully traceable as standards evolve and develop over that include data fusion, object-based image analysis (OBIA) or
time, and (c) modular. Finally, the chapter outlines specific pro- geographic OBIA (GEOBIA), support vector machines (SVMs),
posals for VI standards and discusses the precision of the adjust- decision tress, spectral unmixing, artificial neural networks
ments required. (ANN), phenological matrices, Fourier transforms, and spectral
matching techniques (SMTs). In data composition, multidate,
31.9 Digital image Processing: multispectral data are time composited to long time periods into
Methods and techniques single megafile data cubes, akin to hyperspectral data cubes.
This has facilitated applying hyperspectral data analysis tech-
Chapter 9 by Dr. Sunil Narumalani and Paul Merani provides niques like SMTs for multidate multispectral data composited
a step-by-step approach to understanding the fundamental over long time periods. Such data enable analysis of time series
methods and techniques involved in digital image process- phenological matrices. All aspects of image analysis (Table 31.1)
ing. These steps are summarized in Table 31.1 and are broadly have become efficient as a result of speed in computing power as
grouped into image: (1) quality assessment, (2) preprocessing, (3) well as smarter and better algorithms.
2016 by Taylor & Francis Group, LLC
638 Remotely Sensed Data Characterization, Classification, and Accuracies
TABLe 3 1.1 Standard Digital Image Processing Steps 31.10 Urban image classification
1. Image quality evaluation Methods and Approaches
Histograms evaluation (e.g., frequency distribution)
Basic statistical evaluation (e.g., mean, median, mode) Chapter 10, by Dr. Soe W. Myint etal., discussed four distinct
Visual evaluation (e.g., for cloud, haze) methods of image classification for urban mapping. These four
Digital number (brightness value) evaluation (e.g., in various bands methods, discussed in Chapter 10, are summarized as follows:
indication conditions)
1. Per-pixel spectrally based methods (hard classifications,
2. Image preprocessing
e.g., Figure 31.9a) are used in all types of classification
Radiometric correction (digital numbers to at-sensor reflectance)
(e.g., land use/land cover, urban, or forests categories)
Atmospheric correction (deriving ground reflectance) using algorithms such as maximum likelihood, iterative
Geometric correction (corrections for various geometric distortions) self-organizing data analysis technique (ISODATA), and
3. Image enhancement minimum distance to mean. In per-pixel classifiers, each
Contrast enhancement (e.g., histogram stretch) pixel falls into a class or a group and forms minimum map-
Edge enhancement (e.g., filtering) ping unit. It is noted that the per-pixel classifiers have sig-
Color enhancement (e.g., true and false color) nificant limitations in detecting and accurately mapping
Differencing (e.g., from one date with another) individual urban features, but they can produce simple
Ratioing (e.g., vegetation indices) and convenient thematic maps with reasonable accuracies
Transformations (e.g., PCA, vegetation indices) (often in realm of 70%) in producing few urban classes.
4. Image registration and reprojections 2. Subpixel methods (soft classifications, e.g., Figure 31.9b)
Image to image (e.g., all images of a area to a master reference image) quantify fractions of multiple classes within a pixel. Due
Image to map (e.g., image to a reference map or GPS points) to complexity of landscape, the presence of multiple
Image to coordinate systems (e.g., assigning projections, datum) classes within a pixel is common. Methods used to resolve
5. Image mosaicking, fusion, and megafile creations subpixel composition of pixels include linear methods
Mosaicking in required coordinate systems (e.g., entire country, (e.g., linear spectral mixture analysis, regression trees,
continent, or globe) and regression analysis) and nonlinear methods (e.g.,
Data fusion (e.g., multispectral with panchromatic) ANN). Success of linear spectral mixture analysis will
Time compositing (e.g., monthly maximum value composites) depend on the purity of collecting endmember (preferably
Megafile data cubes (e.g., single file from multiple images of same year through exact ground knowledge). Typically, the number
over several years) of endmembers does not exceed the number of bands.
6. Image analysis However, multiple endmember spectral mixture analysis
Thresholding (e.g., NDVI thresholds) can overcome this. Nevertheless, it is complex to identify
Edge detection (e.g., to highlight lineaments) all the possible endmembers, and the many unidentified
Texture analysis (e.g., to segment an image into distinct objects) endmembers lead to uncertainties.
Change analysis (e.g., drought year vs. non-drought year) 3. With the advent of very-high-resolution (submeter to 5 m)
7. Image classifications imagery around year 2000, OBIA or GEOBIA (Blaschke,
Supervised classification (e.g., maximum likelihood, minimum distance
2010) became a standard approach to discern numer-
to mean) ous urban features like individual buildings, pools, roads,
Unsupervised classification (e.g., ISOCLASS clustering) sidewalks, urban trees, and golf courses. This involves use
Decision trees (e.g., rule base) of spectral and textural characteristics of the image data
Object based image analysis (e.g., segment an image into objects and to first segment the image into distinct groups and then
then classify) apply classification algorithms separately on individual
Spectral unmixing (e.g., linear spectral unmixing) segments to identify detailed urban classes accurately. Yet
Support vector machines computing challenges of large number of segments that
Artificial neural networks need to be carried out based on various scale parameters in
Fourier transforms object-oriented algorithms (e.g., Definiens or eCognition)
Spectral matching techniques (e.g., spectral similarity value, spectral
remains challenging. Nevertheless, numerous studies that
correlation similarity) have shown an increase of anywhere between 10%and 20%
8. Image accuracy assessment increase in accuracies (reaching the realm of 90% accuracy)
Error matrices (e.g., remote sensing versus reference data, leading to using OBIA or GEOBIA have been reported when com-
users producers and overall accuracies) pared with pixel-based approaches (e.g., Figure 31.9).
Fuzzy classification accuracies (e.g., accuracy measures in terms such as 4. Some geospatial methods such as spatial co-occurrence
absolutely correct to absolutely wrong) matrix, spatial autocorrelation, and fractal and lacunar-
Comparison of statistics with other reference data (e.g., areas computed ity approaches capture the local texture similarities into
from remote sensing with national statistics) distinct group of pixels, before a classification algorithm
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 639
0 0.5 1 2 0 0.5 1 2
km Class names Grassland km Class names Grassland
Eucalypt open forest Melaleuca riparian forest Eucalypt open forest Melaleuca riparian forest
Burnt Eucalypt woodland Mixed closed forest Burnt Eucalypt woodland Mixed closed forest
N Burnt Eucalypt open forest Mixed woodland N Burnt Eucalypt open forest Mixed woodland
Eucalypt woodland Open woodland Eucalypt woodland Open woodland
Eucalypt woodland-rocky Eucalypt woodland-rocky
outcrops outcrops
(a) (b)
FIg u r e 31.9 Resultant images of the (a) object-based and (b) pixel-based classifications. (From Whiteside, T.G. etal., Int. J. Appl. Earth Observ.
Geoinform., 13(6), 884, December 2011, ISSN: 0303-2434, http://dx.doi.org/10.1016/j.jag.2011.06.008. Accessed JuneOctober, 2014.)
is applied to improve classification accuracy. This can be croplands, and other land themes. The classification methods
done by local moving windows (e.g., 2 2, 3 3, 5 5). are broadly classified into two categories:
In contrast to the aforementioned advanced geospatial 1. Parametric
approaches, a new spatial frequency-based algorithm 2. Nonparametric
called wavelets characterizes spatial features in different
directions at multiple scales. A window-based approach Parametric classification algorithms assume a known (often
employing wavelet theory and dyadic decomposition pro- normal) statistical distribution of data. Parametric clas-
cedures to measure spatial arrangements of features at sification methods include minimum distance to mean,
multiple scales has been determined to be more effective Gaussian maximum likelihood, Mahalanobis distance, and
than the aforementioned advanced geospatial approaches parallelepiped.
in directly identifying urban classes (Xu et al., 2013). While Nonparametric methods are not limited by any statistical
this approach outperformed other spatial approaches, it is distribution assumption of data and are based on class statis-
limited by a finite decomposition procedure that requires tics that include mean vectors and covariance matrices. This is
a large geographic area (or a large window size) to perform specifically suitable in conditions where there is great hetero-
a classification. An overcomplete wavelet analysis using an geneity in landscape and variances between classes are high.
infinite-scale decomposition procedure is more powerful Nonparametric classification methods include classification and
in measuring the spatial complexity of geographic features regression trees, ANN, k-nearest neighbor, and SVMs.
and identifying detailed urban classes more effectively. These classification algorithms are applied to satellite data
through either supervised or unsupervised logic. In a super-
vised approach, classes are trained through available knowledge
31.11 image classification Methods (e.g., ground data, accurate secondary maps, expert knowledge).
in Land cover and Land Use Widely used supervised methods include Gaussian maximum
likelihood and minimum distance to mean. Decision tree algo-
and cropland Studies
rithms, random forest (RF), SVMs, ANN, and nearest neighbor
Chapter 11 by Dr. Mutlu zdoan discussed methods of classifi- classifier are also supervised classifications but fall into the non-
cation of land cover and land use (e.g., Figure 31.10), vegetation, parametric category. Gaussian maximum likelihood, k-means
2016 by Taylor & Francis Group, LLC
640 Remotely Sensed Data Characterization, Classification, and Accuracies
Water Greenhouses
Tropical crops Conifers
Bare soils Poplar Grove
Herb. Dry Urban
Lig. Irrig. DEM
(a) (b) Herb. Irrig. Value
Oak Grove High: 3480
Grasslands
N 0 10 20 40 60 80 100 Olive Grove Low: 0
km
Shrublands
FIg u r e 31.10 Land cover change maps between classification trees (CT) and random forest (RF) for Granada Province: (a) RF and
(b) CT. (From Rodriguez-Galiano, V.F. et al., ISPRS J. Photogram. Remote Sens., 67, 93, January 2012, ISSN: 0924-2716, http://dx.doi.
org/10.1016/j.isprsjprs.2011.11.002. Accessed JuneOctober, 2014.)
nearest neighbor, and SVMs, on the other hand, are widely used 31.12 Hyperspectral image Processing
supervised parametric methods.
In unsupervised classification, the algorithm assigns classes
Methods and Approaches
based on spectral characteristics (e.g., mean, standard deviation, Hyperspectral data are rich in information content but bring
matrices such as covariance, and correlations) of a pixel. Classes with it many challenges of data handling and processing. In
are then identified based on available knowledge (e.g., ground Chapter 12, Dr. Jun Li and Dr. Antonio Plaza explore some cur-
data, reference data), which will then turn the information classes rent trends in hyperspectral image processing. The classification
generated by the algorithm into land cover classes. Unsupervised approaches discussed in the chapter include the following:
classification is also known as clustering and includes ISODATA
clustering, k-means clustering, and NarendraGoldberg cluster- 1. Supervised classification consisting discriminant analysis
ing. Unsupervised classification approach is recommended when techniques. In these discriminant classifiers, discriminant
a priori knowledge of the landscape (e.g., ground data) is unavail- functions such as nearest neighbor, decision trees, linear
able, limited, or inadequate. k-Means is an unsupervised non- functions, and nonlinear functions are applied.
parametric classification algorithm. Note that there are times that 2. Kernel methods for supervised classification including the
unsupervised classification is used as a first step, before applying well-known SVM classifier.
supervised classification. This is known as a hybrid classification. 3. Spectralspatial classification approaches, which provide
These image classification methods can be performed based significantly improved classification results compared to
on (a) per pixel, (b) subpixel, (c) per field, and (d) per object. These spectral-based classifiers. Spatial features are extracted
can also be soft classification (e.g., mixed classes such as vegeta- through advanced morphological techniques such as
tion that may include croplands, rangelands, forests) meaning morphological profiles and Markov random fields as
that the class boundaries can intermingle to some extent or hard described in the chapter.
classifications (e.g., croplands, soils, rangelands) when the class 4. Probabilistic classification methods such as the multino-
boundaries are distinct without any intermingling. mial logistic regression.
Chapter 11 also discusses other classification approaches that The main problem with the supervised classifiers is addressing
are becoming more popular. For example, object-oriented clas- the huge data dimensionality of hyperspectral data in compari-
sifications or a combination of object-oriented segmentation fol- son with the limited availability of training samples. Specifically,
lowed by pixel-based classification is propagated by some (e.g., supervised approaches require sample data from pure locations
Chapters 14 through 16). both for classification and validation. The number of samples
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 641
required increases with data dimensionality, and as a result, the commonly used SVM classifier with (1) several discriminant
challenge of using a large number of training samples to address classifiers, (2) composite SVM obtained using summation ker-
the high dimensionality of hyperspectral data is costly and often nel, (3) combination of the morphological EMP for feature
prohibitive. This issue, known as Hughes phenomenon, is hard extraction followed by SVM for classification (EMP/SVM),
to address and is often the bottleneck in the effective use of (4) pixel-wise SVM classifier with the morphological watershed,
supervised classification approaches. So, often, the uncertainty (5) SVM classifier combined with the segmentation result pro-
in classification results from the supervised hyperspectral image vided by the unsupervised recursive hierarchical segmentation
analysis is high. (RHSEG) algorithm (Tilton et al.), and (6) a subspace-based
The concept of semisupervised classification, as discussed by multinomial logistic regression classifier followed by spatial
Dr. Jun Li and Dr. Antonio Plaza, is to use not only the labeled postprocessing using a multilevel logistic prior (MLRsubMLL).
samples for training but also unlabeled data. Semisupervised For this purpose, they use nine land use classes and show how
learning is performed, for example, through transductive (or SVM performs relative to other methods. In general, the per-
machine learning) SVMs, among other strategies discussed in formance of classification increases progressively as we move
the chapter. However, the unlabeled data may not be confident from step 1 to step 6. The chapter does not discuss hyperspectral
enough and hence can also lead to uncertainties. image analysis using unsupervised classification, even though
The aforementioned difficulties related with the Hughes phe- RHSEG is discussed briefly.
nomenon in supervised classification methods can be effectively Some of the well-known hyperspectral data dimensionality
overcome through kernel methods such as the SVM. This is also reduction and classification methods are summarized in
discussed extensively in Chapter 7. Authors use a well-known Figure31.11 by Damodaran and Nidamanuri (2014). They con-
and widely used hyperspectral dataset, collected by the reflected cluded that the choice of dimensionality reduction method
optics spectrographic imaging system over the University of and classifier significantly influences the classification results
Pavia, Italy, to provide a series of results comparing the most obtained in hyperspectral image analysis.
Image
Dimensionality reduction methods
(PCA, ICA, MNF, DWT-DR, OBS)
Classifiers
(NED, SAM, SSM, MF, ACE, OSP, TCIMF)
Classified Classified Classified Classified Classified
images images images images images
obtained by obtained by obtained by obtained by obtained by SVM classification
PCA ICA MNF DWT-DR OBS
Assessment of classification
results
Identification of suitable classifiers Identification of optimal
for each dimensionality reduction classifiers and dimensionality
method reduction method for each classes
based on overall accuracy based on per-class accuracy
Comparison and validation
FIg u r e 31.11 Flowchart showing hyperspectral image data reduction and classification methods. Dimensionality reduction methods include
principal component analysis, independent component analysis, minimum noise fraction, discrete wavelet transformbased dimensionality reduc-
tion, and optimal band selection. Multiple classifier system includes normalized Euclidian distance, spectral angle mapper, spectral similarity mea-
sure, matched filter, adaptive coherence estimation, orthogonal subspace projection, and target constraint minimum filter. (From Damodaran, B.B.
and Nidamanuri, R.R., Adv. Space Res., 53(12), 1720, 2014, ISSN: 0273-1177, http://dx.doi.org/10.1016/j.asr.2013.11.027. Accessed JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
642 Remotely Sensed Data Characterization, Classification, and Accuracies
31.13 thermal infrared Remote One major limitation of spaceborne thermal data is their
relatively low spatial resolution. Whereas optical and multi-
Sensing: Principles and spectral data in the visible and near infrared domains can
theoretical Background be collected with resolution of better than 1 m pixel size,
spaceborne thermal sensor data are collected at between 60
Different authors define the wavelength range of thermal infra-
m and 1km spatial resolution. Whereas Landsat 7 ETM+ has
red (TIR) domain differently, but most sensor measurements
a thermal band acquiring the data at 60 m, the novel Landsat
are undertaken in the 814 m wavelength range. In gen-
8 (formerly LDCM or Landsat Data Continuity Mission) only
eral, TIR is the field of remote sensing, which utilizes emitted
offers 100 m spatial resolution in the thermal bands, and sen-
radiation in the wavelength domain, where our planet has its
sors such as AVHRR or MODIS only allow for the monitoring
emission maximum. A large example of TIR data is its inde-
and mapping at 1 km resolution. In the TIR domain, band-
pendence of an illumination source: TIR sensitive sensors can
widths and spatial resolution usually need to be wider and
collect TIR data during the day as well as during the night.
lower to ensure that a proper amount of incoming energy is
The principles and theoretical background of thermal data and
collected. Furthermore, higher-resolution TIR sensors are
thermal data analyses are presented in Chapter 13 by Kuenzer
very costly to build. However, as demonstrated in Chapter 13
etal. This chapter not only provides a comprehensive overview
by Kuenzer etal., the number of nations launching satellites
of the theoretical background including all the laws applicable
including TIR sensors is growing rapidly. We can therefore
for the TIR domain but also elucidates numerous application
be confident that remote sensing in the TIR domain will con-
examples.
tinue to strive catering to unique applications as well supple-
TIR data usage is exemplarily demonstrated for the analy-
menting\complimenting optical remote sensing (e.g., Figure
ses, the general derivation of land and sea surface temperature
31.12).
(LST), the derivation of urban heat islands, and the detection
of hot spots, including thermal anomalies resulting from for-
est fires, coal fires, or gas flaring activities. Also, examples of 31.14 object Based image
sea surface temperature (SST) products and thermal water Analysis (oBiA): evolution
pollution are presented. The authors furthermore explain the and State of the Art
valuable principle of diurnal temperature changes (T), which
makes use of the fact that different surface types exhibit a dif- OBIA or GEOBIAthe latter is used when emphasizing scales
ferent thermal behavior over the day. Some heat up faster than and applications in remote sensing and geographic information
others in the morning, some cool down slower than others, and scienceevolved with the advent of very high-spatial-resolution
the extent of the diurnal temperature change (T) contains satellite imagery (submeter to 5 m) around the year 2000.
valuable clues about the physical characteristics of a surface. However, the underlying concepts of OBIA have been around
For example, water bodies warm up very slowly during the day far longer as enumerated in Chapter 14 by Dr. Thomas Blaschke
and cool down slowly during the nightoverall, their diurnal etal. OBIA involves segmenting an image into distinct objects
temperature change (T) of water is very low and usually lies and then performing the classification on distinct objects sepa-
within the range of only few degrees Celsius. T is much more rately by utilizing the spatial, spectral, radiometric, and tempo-
accentuated for, for example, a dark asphalt surface, which ral characteristics of image data and eventually auxiliary data.
heats up very fast during a sunny day and also cools down OBIA addresses object properties such as shape, size, pattern,
relatively fast after dawn. Here, diurnal temperature change tone, texture, shadows, and association (see Blaschke, 2010;
is much more accentuated and can reach several tenths of Blaschke etal., 2014). An overview of GEOBIA approach to clas-
degrees Celsius. This principle can be exploited when diurnal sification is shown in Figure 31.13 (Ke et al., 2010) in classify-
data are available, for example, daytime and nighttime data of ing forest species using very-high-spatial-resolution QuickBird
the Landsat thermal band or diurnal data of the MODIS sensor multispectral imagery and LiDAR data. QuickBird imagery
thermal bands, and can support the TIR-based mapping of dif- is used for spectral-based segmentation and LiDAR data used
ferent surface types. Furthermore, statements on the moisture for LiDAR-based segmentations (such as height and intensity
content of the surface can be derived as, for example, a wet soil Figure 31.13).
will exhibit a lower diurnal temperature change (T) than a Chapter 14 discusses the evolution of OBIA, concepts of image
dry soil. segmentation and fusion, classification and synthesis, and vari-
However, TIR data can also be used for many other applica- ous applications of GEOBIA. As Chapter 14 shows us, there is
tions, which could not be presented within the chapter. Airborne substantial increase in accuracies using OBIA as opposed to tra-
TIR data can support the monitoring of irrigated areas and agri- ditional per-pixel classification results. For example, GEOBIA
cultural monitoring of crop water stress, can support with the is now increasingly used for various integration tasks includ-
detection of technical accidents and leakages (e.g., pipe bursts, ing new technologies such as mobile applications (e.g., locating
nuclear accidents), and are also frequently employed in medical where you make a call), geo-intelligence, and volunteered geo-
imaging. graphic information (VGI).
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 643
39.1C 25.7C
39
38 25
37
36 24
35
23
34
33
22
32
31
21
30
29
20
28
27 19
26.7C 19.0C
(a) (b)
FIg u r e 31.12 Thermal images gathered using an airborne ThermaCAM P40 (FLIR systems AB, Danderyd, Sweden) infrared digital imager
measuring emitted energy in the range of 7.513 m for an experiment site located near Horsham, Australia (3644S, 14206E; elevation 133 m).
The study areas comprised 48 plots (18 m 12 m) planted to wheat arranged in a randomized block design with three replications for (a) October
2004 and (b) October 2005. Rainfed or irrigated blocks are indicated. Numbers denote kg/ha N applied to each plot. Lighter shades correspond
to higher temperatures. However, readers should note that 90% of Chapter 14 (Section 13.13) focuses on spaceborne. Readers should note thermal
airborne thermal infrared data are really rare and rarely used. But this is a good application example that one can easily replicate with spaceborne
sensors as well. (From Tilling, A.K. etal., Field Crops Res., 104(13), 77, OctoberDecember 2007, ISSN: 0378-4290, http://dx.doi.org/10.1016/j.
fcr.2007.03.023. Accessed JuneOctober, 2014.)
31.15 Geospatial Data integration accurately through such means as color, texture, shape, size, scale,
or form. Recent advances in OBIA have resulted in increasing accu-
in oBiA and implications racies in the objects delineated such as the building features and
on Accuracy and Validity their areas (e.g., Figure 31.14). Beyond that, OBIA enables to assess
the way how objects are delineated, in terms of appropriate scale,
OBIA is now extensively supporting geographic information needs.
complexity, shape, and the fitness to existing geospatial datasets.
The type of information derived from OBIA is wide ranging and
could include such features as forest species, buildings, road net-
works, and farm field boundaries. Availability of very-high-spatial- 31.16 image Segmentation Algorithms
resolution imagery (submeter to 5 m) from various satellite sensors for Land categorization
(e.g., IKONOS, QuickBird, GeoEye), LiDAR, and UAVs has made
deriving objects over large areas feasible. However, remote-sensing Early remote sensing heavily depended on land categorization
data are acquired in a number of platforms, in number of resolu- and classification algorithms that are purely spectral based. These
tions (spatial, spectral, radiometric, and temporal). It is handled by methods include simple thresholding (e.g., using NDVI or band
different users in different way (e.g., different atmospheric correc- reflectivity), single or multiband classifications using supervised or
tion models, with and without atmospheric correction). Methods unsupervised approaches as k-means algorithms, maximum likeli-
and approaches in analyzing the data are often different. hood, and minimum distance to mean. But, over the years, remote-
In Chapter 15, Dr. Stefan Lang and Dr. Dirk Tiede discuss sensing scientists have been looking to improve our understanding
strategies and approaches of geospatial data integration in OBIA. of the land categorization with improved accuracies and reduced
The chapter introduces various concepts of object validation that uncertainties. One of the powerful approaches has been to include
include multistage validation, object fate analysis, object-based image segmentation in the analysis. Chapter 16 by Dr. Tilton etal.
accuracy assessment, object-based change detection, and object provides a comprehensive overview of various segmentation algo-
linking. Object identification means able to delineate objects rithms and illustrates their strengths and limitations. The chapter
2016 by Taylor & Francis Group, LLC
644 Remotely Sensed Data Characterization, Classification, and Accuracies
QuickBird Multispectral (MS) LIDAR dataset
image (four bands)
Image segmentation at different
scales using: Image segmentation at different
Image segmentation at different
QuickBird MS bands + DEM, scales using:
scales using:
DSM, height, intensity DSM, DSM, height, intensity
QuickBird MS bands
Collect features of training point features including:
Spectral features from QuickBird multispectral bands (SP)
Topographic features from LIDAR-derived DEM, slope, aspect and CTl (TOPO)
Height features from LIDAR DSM and DEM (HT)
LIDAR intensity features (INT)
Decision tree classification using training point metrics
Spectral features only (SP)
Spectral features + topographic features (SPTOPO)
Spectral features + topographic features + height features (SPTOPOHT)
Spectral features + topographic features + height features + LIDAR intensity features
(SPTOPOHTINT)
Topographic features + height features (TOPOHT)
Topographic features + height features + LIDAR intensity features (TOPOHTINT)
Accuracy assessment using test points
FIg u r e 31.13 An object-based image analysis protocol for forest species classification through synergistic use of QuickBird multispectral
imagery and LiDAR data. (From Ke, Y. etal., Remote Sens. Environ., 114(6), 1141, 2010, ISSN: 0034-4257, http://dx.doi.org/10.1016/j.rse.2010.01.002.
Accessed JuneOctober, 2014.)
also summarizes in a table various commercial and noncommercial etal., 2010; e.g., Figure 31.15). Chapter 16 concludes that the existing
segmentation software packages available and discusses on where and evolving wide range of image segmentation approaches provide
and how they have been used. Chapter 16 shows that segmentation a rich menu for the users to choose the method, which is the best
algorithms involve using spectral, textural, spatial, and other sec- adapted for each particular application.
ondary (e.g., elevation, slope) information of the landscape obtained
from remote sensing to segment the land into unique units or
segments (e.g., uplands from lowlands or agricultural lands from 31.17 LiDAR Data Processing
natural vegetation) and then classify them separately. This reduces and Applications
the complexities within each segment and also enables shape-based
as well as contextual modeling of segments that are not possible LiDAR data are collected as point clouds with each point hav-
using pixel-based approaches. Furthermore, when classification ing 3D coordinates: horizontal (x, y) and vertical (z). LiDAR data
algorithms are applied separately on each segment, we have more are acquired with a laser scanning system in very high point rate.
meaningful and accurate classes. In addition, we can also get more For each point measurement, a laser pulse is transmitted and the
unique classes. For example, lowland vegetation is often distinct reflected energy is caught. The return signal can be fully recorded
from upland vegetation. When we classify the image that included as a waveform or interpreted to be a single or multiple returns to
both upland and lowland segments, we will be often not able to form point clouds. The features embedded in LiDAR point clouds
distinguish upland and lowland vegetation with great degree of include location (x, y, z), intensity, echo, and waveform. The data
accuracy due to spectral limitations of data. But, when we look at can be used to model, map, and study various Earth surface features
them separately, we will be able to categorize upland vegetation and such as tress, buildings, and terrain in 3D. There are four broad
lowland vegetation with greater degree of certainty. There are many categories of LiDAR systems: (1) terrestrial or ground-based laser
other segmentation algorithms evolving, for instance, region-based scanning with footprint of 110cm and point accuracy of 15cm,
image segmentation algorithm based on k-means clustering (Wang (2) airborne laser scanning (ALS) with footprint of 550cm and
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 645
point accuracy of 10100cm, (3) mobile laser scanning with foot-
print of 110 cm and point accuracy of 130 cm, and (4) space-
borne LiDAR with footprint of 30100 m and vertical accuracy of
25cm10 m (e.g., GLAS, DESDYnI, ICESatII). Waveform LiDAR is
recent advance in ALS technology. The ability to capture 3D struc-
ture of plant and objects is a major advantage of LiDAR data, unlike
other remote sensing. In 3D modeling, it has big advantages com-
pared to photogrammetry or land surveying due to its relatively fast
acquisitions and automated processing. However, LiDAR data often
have small coverage of data collection and large processing volume
of data and are affected by cloud and haze.
Chapter 17 by Dr. Yi-Hsing Tseng and others focuses on LiDAR
data processing and applications. The LiDAR data processing
involves noise filtering, processing 3D (x, y, z) points, independent
variables such as laser use time, scanner position, and scanner
orientation. The LiDAR data are often delivered in the American
Society for Photogrammetry and Remote Sensing (ASPRS) laser
(LAS) format. Data quality is required to meet accuracies of ASPRS
large-scale mapping standard and/or the national spatial data
0 200 400 accuracy standard. The chapter extensively discusses LiDAR data:
Extraction m
(a) 1. Quality assessment and control (e.g., system components,
error budgets, quality assessment and control)
2. Management (e.g., storage of LiDAR point cloud, spatial
indexing, hierarchical representation)
3. Point cloud feature extraction (e.g., spatial feature in
LiDAR data, methods of extraction of spatial features that
include extraction of line and plane features)
4. Full-waveform airborne processing (e.g., shape of received
waveform, echo detection, feature extraction)
The chapter enumerates various applications of LiDAR data such
as in the following:
1. Forestry studies: Structure, diameter at breast height,
crown size, basal area, tree height, aboveground biomass
(e.g., Kronseder etal., 2012; Figure 31.16). LiDAR ability
to penetrate vegetation and look through it and capture
information is a huge benefit that is infeasible using other
remote-sensing techniques.
2. digital elevation model (DEM), digital terrain model
(DTM), digital surface model (DSM), and triangular
irregular network (TIN) generation.
Reference 3. City modeling (e.g., building detection/construction, road
(b) extraction).
4. Structures and objects (e.g., dams, bridges).
FIg u r e 31.14 Assessing accuracy of geographic object-based image 5. Agriculture (e.g., farm topography, plant height).
analysis (GEOBIA) extracted building features and their areas (a) rela- 6. Hydrology and geomorphology (e.g., stream network,
tive to reference data (b) for Lisbon, Portugal. Data used for GEOBIA slopes derived off DEM).
include QuickBird pan sharpened 0.6 m, QuickBird multispectral 7. Land cover classification.
2.4 m, QuickBird imagederived normalized difference vegetation 8. Flood mapping and assessment.
index 2.4 m, normalized digital surface model (nDSM) 1 m, and 9. Disasters (e.g., earthquakes).
reference map of bridges. GEOBIA extracted 317 of the 330 features
found in reference data. (From Freire, S. et al., ISPRS J. Photogram. LiDAR data are often used by integrating with other remote-
Remote Sens., 90, 1, 2014, ISSN: 0924-2716, http://dx.doi.org/10.1016/ sensing data to achieve higher accuracies and discern features
j.isprsjprs.2013.12.009. Accessed JuneOctober, 2014.) that is otherwise infeasible.
2016 by Taylor & Francis Group, LLC
646 Remotely Sensed Data Characterization, Classification, and Accuracies
(a) Segmentation level 10 (RISA) (d) Segmentation level 10 (Definiens)
(b) Segmentation level 40 (RISA) (e) Segmentation level 15 (Definiens)
(c) Segmentation level 80 (RISA) (f) Segmentation level 20 (Definiens)
FIg u r e 31.15 Multiscale image segmentation comparison using the proposed algorithm and the Definiens algorithm. The left three images
(a, b, and c) were segmented using region-based image segmentation algorithm based on k-means clustering (RISA) with the default parameter
values and segmentation scales of 10, 40, and 80, respectively. The right three images (d, e, and f) were segmented using the Definiens software with
the color criterion of 0.9, shape criterion of 0.1 (compactness 0.2 and smoothness 0.8), and segmentation scales of 10, 15, and 20, respectively. (From
Wang, Z., et al., Environ. Modell. Softw., 25(10), 1149, October 2010, ISSN: 1364-8152, http://dx.doi.org/10.1016/j.envsoft.2010.03.019. Accessed
JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 647
LDFUnlogged
Absolute height (m) LDFUnlogged
3045
3,000 Mean = 25.67
2530
Std. dev. = 7.91
2025
n = 25,379
1020
Frequency
0.5 10
2,000
1,000
0
0 10 20 30 40 50
LiDAR height (m)
PSFLittle Logging 2,000 PSFLittle Logging
Absolute height (m) Mean = 15.95
3045 Std. dev. = 6.07
2530 1,500 n = 8,588
2025
Frequency
1020
0.510 1,000
500
0
0 10 20 30 40 50
LiDAR height (m)
FIg u r e 31.16 Distribution of LiDAR point heights within 1 ha plots: comparison of lowland dipterocarp forest and mixed peat swamp forest.
Mean tree height, its standard deviation (std. dev.), and the total number of points higher than 0.5 m (n) are given in the histograms. (From
Kronseder, K. et al., Int. J. Appl. Earth Observ. Geoinform., 18, 37, August 2012, http://dx.doi.org/10.1016/j.jag.2012.01.010, ISSN: 0303-2434.
Accessed JuneOctober, 2014.)
31.18 change Detection 3. Specific changes: Deforestation, urbanization
4. Continuous change: Forest disturbance
Remote sensing by its inherent characteristics such as repeat
Chapter 18 presents a complex case of performing land cover/
coverage of any location- and consistency of data offers the
land use change detection using multisensor data. They use
best data type for change detection anywhere on planet Earth.
for 1 year radar remote-sensing data from Radarsat and for
Therefore, remote sensing, particularly spaceborne remote sens-
another date use optical remote-sensing data from Earth
ing, is widely used for change detection of a wide range of factors
Observing-1 Advanced Land Imager (EO-1 ALI). This is not
such as the land cover (physical footprint), land use (anthropo-
an ideal situation to get best results since using characteristi-
genic), forests, agriculture, glaciers, and biomass. Daniela Anjos,
cally different data sources for change detection leads to higher
Dengsheng Lu, and others in Chapter 18 provide an outline of
degree of uncertainties. Nevertheless, using distinctly differ-
change detection methods, reason out why remote sensing is
ent remote-sensing data is necessary, especially in areas of
central, identify various approaches for change detection, and
high cloud cover such as Amazon or Congo rainforests. The
present an example of land cover/land use change detection
ability of radar to penetrate the clouds becomes useful. So, in
in the Brazilian Amazon. There are four broad approaches to
this regard, the case study presented for change detection in
change detection studies:
Chapter 18 is important. Further, Chapter 18 also shows us the
utility of using the direct classification approach of SVMs as
1. Conversion: Detailed land cover change trajectories
a preferred advanced method for change detection. Whereas
2. Binary change: Change versus no change
2016 by Taylor & Francis Group, LLC
648 Remotely Sensed Data Characterization, Classification, and Accuracies
N 31.19 Geoprocessing, Workflows,
and Provenance
Today remote-sensing data go through complex chains of events
from acquisition, curation, preprocessing, and applying analyti-
cal methods and approaches to derive and synthesize information,
leading to postprocessing and dissemination to users (e.g., Figure
31.18). Different laboratories or individuals may apply different
models, workflows, inferences, and processing chains toward
achieving the same goal. For example, a forest change map may be
produced using at-sensor surface reflectance from similar but dis-
tinct sensor systems (e.g., Landsat 5 TM vs. Landsat 8 operational
land imager). Another forest change map may be produced using a
combination of optical, radar, and LiDAR data. Yet another study
may also use the same combination of input data to produce a for-
est change map, but using a different analytical change detection
algorithm. Unless a user has a full understanding on how the maps
are produced, it is difficult to put real value to them or to even rep-
licate another scientists remote-sensing workflow. This is where
careful geospatial provenance (or lineage) becomes very valuable
and effective. Provenance in this case entails keeping an exact his-
torical and machine-readable record of the geospatial processes
whereby a remote-sensing product originated and was developed.
In Chapter 19, Dr. Jason A. Tullis et al. provide a histori-
cal context of geospatial provenance with a focus on David P.
Lanters (1992b) Geolineus project in the early 1990s; show us how
remote-sensing product quality, reproducibility, and trust depend
on provenance; and provide a number of examples of recent
provenance-aware remote-sensing geoprocessing systems. They
discuss current international standards, specifications, and rec-
ommendations pertaining to implementation of shared remote-
sensing provenance information. These include, for example,
International Standards Organizations (ISO) 19115-2 metadata
standard, endorsed by the Federal Geographic Data Committee
(FGDC) (ISO, 2009), and World Wide Web Consortiums PROV
Data Model (Moreau and Missier, 2013) and accompanying rec-
ommendations of keen interest in the provenance research com-
munity. They identify questions pertaining to remote-sensing
0 2.5 5 10 Kilometers
data products that only provenance can help answer such as the
exact processing steps applied to develop a remote-sensing prod-
FIg u r e 31.17 Detailed excerpts from the classifications (right col- uct, processing order, parameters, execution time, and how errors
umn) and their respective areas in false color composite images in the
were expressed and propagated. Finally, they discuss provenance-
left. The six rows show the following areas in descending order: Beijing
aware systems applied to remote sensing such as MODIS Adaptive
1990, Beijing 2010, Shanghai 1990, Shanghai 2010, Shenzhen 1990, and
Shenzhen 2010. Images used are Landsat TM and HJ-1A/B. Images were Data Processing System and OMI Data Processing System
studied using random forest classifier, tassel cap transformations, urban designed for use by NASA to manage satellite imagery and prov-
indices, and landscape matrices. (From Haas, J. and Ban, Y., Int. J. Appl. enance from MODIS sensors on the Terra and Aqua satellites and
Earth Observ. Geoinform., 30, 42, August 2014, ISSN: 0303-2434, http:// the OMI sensor on Aura, respectively (Tilmes and Fleig, 2008).
dx.doi.org/10.1016/j.jag.2013.12.012. Accessed JuneOctober, 2014.)
simple and direct change detection approaches such as the dif- 31.20 toward Democratization of
ference imaging (e.g., NDVI differences) can be used, methods Geographic information
such as SVMs allow us to use the power of multiband data and
from multiple sensors leading to relatively greater accuracies. Chapter 20 by Dr. Gaurav Sinha et al. discusses the democra-
Many other methods such as RF classifiers (e.g., Figure 31.17; tization of geo-information and geospatial information com-
Haas and Ban, 2014) can be used in change detection studies. munication technologies (Geo-ICTs). Early use of geographic
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 649
Users
User management
Geoprocessing portal
Geoprocessing
Interactive
workbench
Data, service, and
geoprocessing model model discovery and
development Geoprocessing
retrieval
modeling and
Geospatial standards
application
Security and privacy
Model registration Peer-review and
collaboration platform
Catalogue service Inference engine Workflow engine
Data, service,
and model
publishing
Geospatial
manager
Visualization Model/workflow Knowledge
service warehouse data, service,
service
and model
management
Sensor service Geospatial data Geoprocessing platform
service service
Geoprocessing
Geospatial Geoprocessing resources
Sensors
data facilities
FIgu r e 31.18 Framework for the Geoprocessing Web. (From Yue, P. etal., Trans. GIS, 14(6), 755, 2010; Zhao, P. etal., Comput. Geosci., 47,
3, October 2012, ISSN: 0098-3004, http://dx.doi.org/10.1016/j.cageo.2012.04.021. Accessed JuneOctober, 2014.)
information technologies, much like computing, was an elitist accessibility of Geo-ICTs, neogeography (or new geography),
intellectual activity where general public had hardly any par- participatory GIS, crowdsourcing, VGI, and social networking
ticipation. The information was processed, analyzed, and dis- have further integrated geo-information into peoples lives. For
seminated by very few highly specialized experts knowledgeable example, there exist several crowdsourcing initiatives, especially
in remote sensing, geographic information system (GIS), global for managing disasters like earthquakes, floods, and tsunamis.
positioning systems (GPS)/GNSS, and spatial modeling. This One such example are flood databases, such as the International
topdown approach was criticized as a hegemonic system per- Disaster Database (EM-DAT), ReliefWeb (launched by the
petuating the grip of powerful agencies and trained profession- United Nations Office for the Coordination of Humanitarian
als on geographic information as viewed by several thinkers. Affairs), the International Flood Network, and the Global Active
This has been discussed in detail in Chapter 20. This was also a Archive of Large Flood Events (created by the Dartmouth Flood
period where GIS, remote sensing, and GNSS were mainly tools Observatory) (Wan et al., 2014; Figure 31.19). Another global
of well-funded national and international specialized institutes initiative called Geo-wiki (Fritz et al., 2012) helps improve
and military establishments. global land cover assessment and mapping. Readers can expand
Chapter 20 discusses how geo-information has become on these ideas after reading Chapter 20. Long back, Prof. Duane
ubiquitous, linking web mapping, wireless delivery, crowd- Marble, one of the pioneers of geographic information science,
sourcing, and other digital technologies through expert col- predicted that all information will be geospatial, meaning
laboration and user participation into a seamless platform for that the information will be tied to precise geographic location.
powerful delivery of data and knowledge. One of the earliest Today, an enormous amount of information is captured location
influences of this democratization and merger of technologies specific and delivered with location specificity. However, this
was U.S. Vice President Al Gores Digital Earth vision. As a ubiquitous nature of geo-information, and the ease with which
result, access to digital geo-information through virtual globes it can now be captured, stored, shared, and sold, has also raised
(e.g., Google Earth, Microsoft Bing, ESRI ArcGIS Explorer) is many questions about individual rights and freedom. There
now integral to billions of peoples daily lives. In addition to the also arise issues related to the quality of information generated
2016 by Taylor & Francis Group, LLC
650 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 31.19 The map visualization of global flood cyber infrastructure. The top and bottom maps are color-coded by severity and fatalities,
respectively. (From Wan, Z. etal., Environ. Modell. Softw., 58, 86, August 2014, ISSN: 1364-8152, http://dx.doi.org/10.1016/j.envsoft.2014.04.007.
Accessed JuneOctober, 2014.)
and processed by nonprofessionals. For example, when data are pioneers (see Chapter 21) started the development of GIS technol-
crowdsourced, experts as well as nonexperts share data. Thus, ogy in 1960s and 1970s. The maturing of GIS technology has led
quality of such data can vary widely depending on the source, to a wide adoption of mapping and spatial analysis in domain
which could even be detrimental to the progress of precise geo- applications with extensive data from multiple sources (e.g., GIS,
information that is reliable and trusted. GPS, remote sensing) and sophisticated algorithms (e.g., image-
processing algorithms, map algebra, spatial econometrics) for
31.21 GiScience spatial decision support. GIScience grounds GIS technologies with
theories of spatial representation, spatial relationships, and a range
GIScience is the science of geographic information and analy- of computational methods that transform the underlying concep-
sis that substrates the development of GIS and a host of related tualization and problem-solving approaches. Today, GIScience
technologies (cartography, geodesy, surveying, photogrammetry, research expands into data uncertainty, spacetime analytics, and
GPS, remote sensing, spatial modeling) for solving problems (e.g., social implications on web mapping, crowdsourcing, and cloud
detecting a landfill site, Figure 31.20) related to a wide spectrum of computing as well as mobile delivery. In Chapter 21, Dr. May Yuan
themes such as agriculture, water, forestry, geography, geology, geo- presents an outline and history of GIScience and highlights its core
physics, oceanography, ecology, environmental science, and social epistemology: abstraction, algorithms, and assimilation. Also see
sciences. GIScience is contributing spatial thinking and spatial Chapter 26 (crowdsourcing), Chapter 27 (cloud computing), and
computing that propels many disciplines to take a spatial turn, Chapter 28 (web mapping) and how GIScience is intricately linked
such as spatial ecology, spatial epidemiology, spatial social sciences, to these techniques, especially going forward as discussed by Dr.
and spatial humanities. Soon after the advent of computers, early Yuan in this chapter.
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 651
To identify the most ideal place for a municipal
Aim landfill site in Bo, Southern Sierra Leone
Identification of criteria
Geo-environmental Socio-political Economic
Water bodiesperencial streams Urban Distance from
intermittent streams Airfield/helipad generation point
Ground water level Wind direction Distance from major
Landform/Topography (Aspect) roads
Slope Community Land price
Geology acceptance
Soil
Land cover
Data acqaisition
Topographic,
Heads-up on screen
Secondary data Geological, soil and
digitization
land use map
Data types Database
GPS survey of
Primary data water wells to
obtain ground Thematic maps
water level Ground water level map layer Soil map layer
Geological map layer DEM map layer
Slope map layer Aspect map layer
Residential/Built up map layer Land use map layer
Surface water map layer Road network map layer
Data analysis
Criteria map layers
Factor maps Constraint map
GEO-environmental Economic factors Socio-political map GEO-environmental Socio economic
factors layers constraints constraints
Map standardized using fuzzy logic Final constraint
map
Standardized
Standardized Standardized
GEO-environmental
Economic factors social factors
factors
Factor and Factor and Factor and
ordered weights ordered weights ordered weights
WLC/OWA WLC/OWA
WLC/OWA
Final geo- Final socio-
Final economic political
environmental
suitability map suitability map
suitability map
Factor and Factor and Key
WLC/OWA WLCWeighted linear combination
ordered weights ordered weights
OWAOrdered weighted averaging
Landfill suitability map
FIg u r e 31.20 Spatial model for the identification of a potential landfill site by integrating spatial data from various sources (e.g., remote sensing,
GIS, GPS). (From Gbanie, S.P. etal., Appl. Geogr., 36, 3, January 2013, ISSN: 0143-6228, http://dx.doi.org/10.1016/j.apgeog.2012.06.013. Accessed
JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
652 Remotely Sensed Data Characterization, Classification, and Accuracies
31.22 object-Based Regionalization for etal., 2013). These issues are discussed in Chapter 22 by Dr.
Stefan Lang etal., introducing the geon approach as a strategy
Policy-oriented Partitioning of Space to integrate multiple sets of geospatial data.
One of the biggest challenges facing policy makers and oth- Chapter 22 begins with definitions of latent, complex, and
ers using remote sensing and other geospatial dataderived multidimensional phenomena and the related concept of
information is to ensure its integrity, robustness, and reliabil- geons and sets the platform for its spatial representation that
ity. Inherently, remote-sensing data are consistent and robust is robust and useful to decision makers in different domains.
(with proper calibration and normalization). But methods The authors discuss the principles of regionalization, how
and approaches in analyzing the data are often different. geons are used in domain-specific regionalization, and OBIA
Further, in order to derive new information or enhance exist- in image segmentation and regionalizations. They take four
ing geospatial information, remote-sensing data are used as case studies (socioeconomic vulnerability of hazards, social
a data source to be integrated with existing geospatial data vulnerability to malaria, landscape sensitivity, and climate
in different ways. Resultant products are often dissimilar, change susceptibility) to show readers how latent phenom-
depending on the algorithms used, or the weights assigned enon is addressed, what indicators and datasets are used, and
to each of the information source (e.g., Figure 31.21; Branger the results.
Land use N
Synthesis 2.5 m
Urban
Mixed/Agriculture
Mixed/Forest
Rural/Agriculture
Rural/Forest
0 2 4 km
Land use N Land use N
Spot 2.5 m Quickbird 2.44 m
Urban Urban
Mixed/Agriculture Mixed/Agriculture
Mixed/Forest Mixed/Forest
Rural/Agriculture Rural/Agriculture
Rural/Forest Rural/Forest
0 2 4 km 0 2 4 km
N N
Land use Land use
Spot 10 m Ortho 0.5 m
Urban Urban
Mixed/Agriculture Mixed/Agriculture
Mixed/Forest Mixed/Forest
Rural/Agriculture Rural/Agriculture
Rural/Forest Rural/Forest
0 2 4 km 0 2 4 km
FIg u r e 31.21 Model hydrological response units and reaches and land use classification for Synthesis 2008, Spot 2008 (2.5 and 10 m), QuickBird
2008, and Ortho 2008 land use maps. (From Branger, F. etal., J. Hydrol., 505, 312, November 15, 2013, ISSN: 0022-1694, http://dx.doi.org/10.1016/
j.jhydrol.2013.09.055. Accessed JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 653
KIRUNA_ESOC
0
ESTEC
IENG
USNO MIZU_GFZ
0 WUHAN_GFZ
KOUROU_ESOC
MALINDI_ESOC
TAHITI_ESOC
30 NEW-NORCIA_ESOC
LA PLATA_GFZ
DUNEDIN_GFZ
60
TROLL
GIOVE_TRACKING_NETWORK
150 120 90 60 30 0 30 60 90 120 150
FIgu r e 31.22 Galileo System Test Bed, Version 2 (GSTB-V2) station network. (From Dow, J.M. etal., Adv. Space Res., 39(10), 1545, 2007, ISSN:
0273-1177, http://dx.doi.org/10.1016/j.asr.2007.04.064. Accessed JuneOctober, 2014.)
31.23 Global navigation theoretical concepts of GNSS-R. NASA and ESA are launching
series of satellites focused on GNSS-R such as the Cyclone Global
Satellite Systems Navigation Satellite System (CYGNSS). Even though the use of
In the modern-day world, GNSSs have become ubiquitous. They GNSS-R data in remote sensing is still a novelty, with increasing
guide us from place A to place B; pinpoint any target in the world number of low-Earth-orbit satellites equipped with GNSS receiv-
for gathering information of various natures; control air, road, and ers, the situation is going to change soon. The chapter highlights
sea traffic; precisely locate an agricultural crop or forest species, or the key land applications of GNSS-R such as soil moisture and
a water body as tiny as a well; and numerous other applications. forest change detection. In many other applications like that for
Currently, there are five distinct operational GNSS: (1) GPS con- tsunamis (e.g., Figure 31.23; Stosius etal., 2011), GNSS-R is very
trolled by the United States; (2)Global Orbiting Navigation Satellite powerful. Similarly, it highlights the key ocean applications of
System (GLONASS) controlled by Russia; (3) Galileo controlled by GNSS-R such as sea surface undulation and sea surface height.
Europe (e.g., Galileo System Test Bed, Version 2 [GSTB-V2, Figure Just as the UAVs are becoming increasingly important in land
31.22]); (4) BeiDou (Compass) controlled by China, and 5. Indian remote sensing, GNSS-R is also becoming very important. They
Regional Navigational Satellite System (IRNSS). Typically, GNSSs add to conventional remote sensing and bring in unique capa-
have upto to 36 satellites, of which a minimum of four satellites bilities that only GNSS-R can provide such as CYGNSS.
always accessible for signals from any part of the world. These
systems can provide positional accuracy at any location on Earth 31.25 GnSS and Wide Array
to within a few centimeters or even millimeters. Chapter 23 by
Grewal discusses in detail GNSS characteristics, including sig-
of Applications
nal spectrums, differences between civil and military signals (for GNSS include satellite constellations (typically, 2030) from
GPS), and the nature and characteristics of the different signals, for NAVSTAR GPS of the United States, GLONASS of Russia,
example, frequency division multiple access, code division multi- Galileo of European Union, BeiDou navigation systems of
ple access (CDMA), and time division multiple access. China, and IRNSS of India that provide instant, freely available
positiondataat high accuracies (few centimeters to few meters)
31.24 GnSS Reflectometry for ocean for any place in the world (see Figure 31.24; Jin etal., 2011). As
and Land Applications a result of global international GNSS service constellation, a
dense network of GNSS satellites and sites are available. The first
Chapter 24 by Kegen Yu et al. discusses the increasing interest and the most reliable GNSS acquires position data and altitude
and usage of GNSS reflectometry (GNSS-R) in ocean and land (x,y,z) through a baseline constellation of 24 satellites positioned
applications. The chapter provides a solid foundation on the about 20,000 km from Earth. To get accurate position for any
2016 by Taylor & Francis Group, LLC
654 Remotely Sensed Data Characterization, Classification, and Accuracies
25 2
1.5
20 1
0.5
Sea surface
15 0
0.5
10 1
1.5
Latitude
5 2
2500
2000
1500
0
950
700
Topography
450
5 200
150
100
10 50
10
5
2
15 0
70 75 80 85 90 95 100 105 110
[m]
Longitude
FIg u r e 31.23 Simulation of the Sumatra tsunami 1h after the earthquake. Ground tracks (black) show Global Navigation Satellite System
reflectometry detections within 1min of observation. The ellipse encloses all detections made up to this moment and show the current tsunami
expansion. (From Stosius, S. etal., Adv. Space Res., 47(5), 843, March 2011, ISSN: 0273-1177, http://dx.doi.org/10.1016/j.asr.2010.09.022. Accessed
JuneOctober, 2014.)
)
ASS How does GNSS
MP
SS CO remote sensing work?
GN LILEO
, G A
ASS
ON
S , GL
(GP
Re
fra
cte
ds
ign
als
Leo GNSS receiver
Refle
cted
signa
ls
90
60
30
30
60 IGS sites
90
180 120 60 0 60 120 180
FIg ur e 31.24 Remote sensing using future denser Global Navigation Satellite System (GNSS) network and more satellite constellations. The left
lower corner shows the global International GNSS Service (IGS) sites distribution. (From Jin, S. etal., Adv. Space Res., 47(10), 1645, May 17, 2011,
ISSN: 0273-1177, http://dx.doi.org/10.1016/j.asr.2011.01.036. Accessed JuneOctober, 2014.)
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 655
location on Earths surface, one needs data from at least three satel- ground-based data extensively across the planet timely, rou-
lites. Typically, at least four satellites are available from any spot on tinely, and cost-effectively is extremely difficult if not infeasible.
Earth. Of the 24 GPS satellite constellation at any given time, 21 are This is overcome to significant extent through crowdsourcing. In
in operation continuously, and 3 are used as spare to replace any Chapter 26, Dr. Fabio DellAcqua illustrates some of the major
failing satellite. The signals (e.g., CDMA) transmitted by them are advances made in crowdsourcing such as the following:
either for military (encrypted) or civilian (open).
1. OpenStreetMap, which is widely used.
Dr. Myszor etal. in Chapter 25 discuss various applications of
2. Did you feel it service, a pioneering earthquake intensity
GNSS. These are broadly categorized into civilian, military, and
mapping based on crowdsourced information gathered
aeronautic. The applications discussed include the following:
from people on the ground. This leads to Community
Internet Intensity Maps.
1. Consumer applications such as the interactive maps to
3. Geo-wiki for land cover mapping.
locate any place or location of interest (e.g., house, restau-
rant, airports, bus stations). These are ubiquitous today in There are, of course, numerous other applications of crowdsourcing
our daily life, in cars and smart phones. for spatial information (e.g., Figure 31.25; Heipke, 2010). This can
2. Industrial applications such as in agriculture (e.g., detect- be, for example, used for conducting election pool surveys, product
ing location of farms, crops grown in farms), stores, and surveys, and even mapping field boundaries or host of other ser-
city services (e.g., location of severs, pipelines). vices. Crowdsourcing assumes that there are sufficient numbers of
3. Transportation applications such as route from point volunteers who will provide such data either for free (personal sat-
A to B. isfaction) or for a fee. Chapter 26 discusses the evolution of crowd-
4. Surveying applications such as mapping large areas and sourcing and provides many practical examples of crowdsourcing.
isolated areas. In an increasingly sophisticated, Internet-dominated world
5. Disaster applications where disaster maps that provide of today, crowdsourcing is both attractive and desirable.
information such as people injured, property lost, and Nevertheless, one needs to be cautious of many aspects of crowd-
damage areas are instantly mapped. GNSS data are used in sourced data. It is not scientifically sound data for many appli-
wide array of disasters like earthquakes, floods, droughts, cations. For example, when millions of users feed data on where
famine, and fire. agricultural croplands are or say something about their produc-
6. Health applications (e.g., monitoring and managing dis- tivity situation (e.g., drought or stress condition), uncertainties
ease spread). in such data are likely to be high. Even when such data are col-
7. Tracking applications (e.g., animal grazing paths, wildlife lected by experts, there are differences as a result of the factors
tracking). such as the depth of ones understanding and definition issues.
8. Military applications (e.g., locating and monitoring troop So, when untrained volunteers provide such data, the likelihood
movements, military installations). of these uncertainties increases. Yet, besides statistical approaches
9. Aeronautic applications (e.g., tracking aircraft movements). to data cleaning, it is expected that the technology itself will
develop increasingly smart solutions to reduce these uncertainties
Chapter 25 also provides the GNSS satellite characteristics as
and make the crowdsourced data more accurate and reliable. For
well as errors in GNSS and ways and means to control them and
example, when one takes a photo of a particular crop and uploads,
keep the errors to a minimum.
the technology may identify the crop type automatically along
with its precise location, removing the uncertainties that may
31.26 crowdsourcing in Remote creep from a volunteer input. Given these facts, crowdsourcing
Sensing and Spatial technologies will become an important and integrated component of the future
to Study Planet earth remote sensing and spatial data analysis practices.
Remote sensing made it possible to gather data from any spot
on our planet Earth: remotely, consistently, repeatedly, and in 31.27 cloud computing and
various levels of detail. Nevertheless, the need to be there on the Remote-Sensing Data
spot physically to decipher or verify the actual conditions on
the ground remains. For example, satellite remote sensing can Cloud computing can generate and involve a massive amount
help establish important information such as location, size, and of data, especially in GeoSciences and specifically when remote
orientation of buildings in a residential area through processing sensing is involved. Today, data are gathered in increasing quanti-
of multispectral, high-resolution images. Still, the most impor- ties by both public and private entities, and in every dimension:
tant proxies for building vulnerability to natural disasters, such spatial, spectral, radiometric, and temporal, covering the entire
as presence of steel frames or seismic retrofitting, are not set to planet. The challenges of processing these gigabytes, terabytes,
become visible in spaceborne data anytime soon. Or even when and petabytes of big data over and over again and converting
remote sensing establishes these facts, there is a need to verify, this data into meaningful information for the study of our planet
to be absolutely sure or nearly so. Naturally, gathering such are real today. How this can be done is presented and discussed in
2016 by Taylor & Francis Group, LLC
656 Remotely Sensed Data Characterization, Classification, and Accuracies
Scale = 1:2441
Nederland Nederland Nederland
Nederland Nederland Nederland
Permalink
4.99077, 52.17341
Scale = 1:2441
Nederland Nederland
Missing road
detected here
Nederland
Permalink
4.99081, 52.17341
FIg u r e 31.25 Example of a result from passive mapping. Upper, existing map data; lower, superimposition with data from global positioning
systems tracks derived from cars. The missing road in the center is clearly visible; however, the individual tracks need to be aggregated into an
attributed graph structure for use in routing applications (Tele Atlas). (From Heipke, C., ISPRS J. Photogram. Remote Sens., 65(6), 550, November
2010, ISSN: 0924-2716, http://dx.doi.org/10.1016/j.isprsjprs.2010.06.005. Accessed JuneOctober, 2014.)
Chapter 27 on cloud computing and remote-sensing data by Dr. etal., 2014). In spite of these advances, cloud computing is still in
Sugumaran etal. They define cloud computing and contrast it with its nascent stages as far as remote-sensing data processing is con-
cluster, grid, high-performance computing, and high-throughput cerned. Some of the biggest challenges are
computing. The chapter discusses cloud-computing paradigms
such as the public cloud, private cloud, and hybrid cloud. Cloud- 1. Ability to process very high-resolution spatial (e.g., a few
computing services such as Amazons elastic compute cloud allow meters), spectral (e.g., hundreds of bands), radiometric
researchers to use paradigms such as infrastructure as a service, (e.g., 16 bit or higher), and temporal (e.g., daily) images of
platform as a service, and software as a service and make possible the world
the processing of massive amounts of data for nominal costs. They 2. Development of image-processing algorithms in the cloud
illustrate the ability to process one large LiDAR dataset (~6 TB) environment
covering the entire state of Iowa. An open system architecture for 3. Data security
geospatial cloud computing is shown in Figure 31.26 (Evangelidis 4. Storage and backup of big data over long time periods
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 657
Mobile clients Thick clients Thin clients 4. Providing the power of information to common person
who may not have any idea of remote sensing or geoscience
Client layer
5. Enabling people to tell stories or inform changes through
images
Google Earth involves use of geocoded imagery of any various
resolutions along with the 3D elevation derived from Shuttle
Internet
Radar Topographic Mission (SRTM) and other (e.g., LiDAR)
high-resolution data. The baseline imagery for the Google Earth
Web server is Landsat 30 m imagery, but for large swaths of the world, very
Catalog services Data services
Processing high-spatial-resolution imagery (VHRI; submeter to 5 m) from
services
providers such as DigitalGlobe are also available. The system
CSW WMS WFS WCS WPS
is supported and revolutionized by Keyhole Markup Language
Catalog server Data server Processing server
(KML).
Application layer
Further, the ease of use and the power of data and informa-
tion available in Google Earth for anyplace and anywhere and
to anyone truly democratized remote sensing and GIS sciences.
Chapter 28 also shows us how several scientific applications
Processes and applications of societal importance of Google Earth soon
became apparent. They show us example of volcanoes of the
world mapped by the Smithsonian. Others like Thenkabail
etal. (2009a,b) have extensively made use of VHRI as ground
data to identify and label croplands of the world as well as
assess cropland accuracies. These days, VHRIs in Google
Data layer
Earth are extensively used by remote-sensing scientists as data
File system Database
for developing or testing their algorithms in certain applica-
tions. Nongovernmental organizations and bloggers have
Data providers Metadata used Google Earth to illustrate and demonstrate issues of
environmental and societal importance that often influence
FIg u r e 31.26 Open system architecture for Geospatial Cloud public opinion. Crowdsourcing of information gathered from
Computing. (From Evangelidis, K. etal., Comput. Geosci., 63, 116, February remote-sensing imagery is depicted on Google Earth during
2014, ISSN: 0098-3004, http://dx.doi.org/10.1016/j.cageo.2013.10.007. such events as earthquakes, floods, hurricanes, and tsunamis
Accessed JuneOctober, 2014.) (e.g., Figure 31.27).
Even though Google Earth has tremendous strengths, it
Nevertheless, in due time, these challenges will be overcome cannot be used for all scientific studies that use remote sensing.
and allow ubiquitous processing of big remote-sensing data in This is because the imagery available is used only visually,
various cloud-computing environments. without the underlying values and metadata available, which
are required for quantifying variables and studying changes.
For example, Google Earth imagery cannot be used for quan-
31.28 Google earth for earth Sciences tifying drought, deriving drought indices, and determining
drought conditions over time. However, in order to overcome
In Chapter 28, Dr. John E. Bailey outlines how Google Earth, these limitations, Google Earth Engine (EE) has been intro-
launched in the year 2005, brought complex remote-sensing duced. EE stores real images. For example, the entire Landsat
technology to the doors of common person. Based on the imagery archive (from 1972 to present) is currently available in
philosophy of Digital Earth, a concept popularized by for- EE. This imagery can be used for scientific studies by building
mer Vice President of the United States Mr. Al Gore in 1998, algorithms directly in the EE platform to analyze the imag-
Google Earth enabled some powerful possibilities. These ery. This also allows for time-lapse analysis. EE offers a cloud-
included computing platform to quickly and instantly analyze images
1. Visualization of the entire Earth from your desktop, lap- at global scale even at 30 m spatial or higher resolutions. This
top, or mobile seamlessly platform is still being developed, but already some interesting
2. Ability to zoom in and zoom out of any part of the planet global applications are becoming available (e.g., forest cover
and visualize a number of details such as cities, forests, and wetland changes using Landsat mentioned in Chapter 28).
agricultural lands, houses, trees, and networks (e.g., The recent purchase of Skybox Imaging by Google also sets the
roads, rivers) stage for more frequent and innovative acquisition of satellite
3. Precisely locate annotations imagery for Google EE applications.
2016 by Taylor & Francis Group, LLC
658 Remotely Sensed Data Characterization, Classification, and Accuracies
FIg u r e 31.27 Tsunami field survey data available in Google Earth .kml format since 2004 and plate tectonics in the Indian and southwest
Pacific Oceans. A collection of the Google Earth (http://www.google.com/earth/index.html) .kml files of tsunami measurement data from field
surveys after the 2004 event including the 2004 Indian Ocean, 2006 Java, 2007 Solomon, and 2010 Chile are summarized (Fujima, 2011; Mori and
Takahashi, 2012) and shown with plate tectonics (USGS, 2012) in this figure. (From Suppasri, A. etal., Int. J. Disaster Risk Reduct., 1, 62, October
2012, ISSN: 2212-4209, http://dx.doi.org/10.1016/j.ijdrr.2012.05.003. Accessed JuneOctober, 2014.)
31.29 Map Accuracies 3. Reference data: Typically, reference data are collected
through ground visits. A rule of thumb is to collect 50
Chapter 29 by Dr. Russell G. Congalton provides an in-depth samples for each land cover category when total num-
guideline on assessing thematic and positional accuracies of bers of classes are 12 or less. When classes are more than
maps. When RS data are processed, classified, and information 12, the recommendation is to gather data from 75 or 100
deciphered through digital image processing and interpretations samples per class. However, the number of samples can
(or through manual interpretations as in the past), a thematic map go up for large areas. Reference data are collected using
is produced. Examples of thematic maps include products such well-designed sampling strategies (e.g., random or strati-
as land cover types, species types, biomass categories, and leaf fied random). Further homogeneous areas are selected for
area index. Scientific use of any thematic map requires assessing collecting ground data. For example, collecting data from a
its thematic as well as positional accuracies. Thematic accuracy single 30 m pixel could lead to errors due to misregistration
addresses how well classes in thematic map correspond to what is (locational error). Therefore, it is better to select homoge-
observed or measured in the field or ground. Positional accuracy neous areas for a land cover class such that a 3 3 cluster of
refers to quantifiable differences between two map locations or a 30 m pixels are chosen to reduce the positional error.
map location and a location on ground. Dr. Congalton addresses
Dr. Congalton then provides a series of error matrices to
these issues systematically and comprehensively in Chapter 29,
evaluate map accuracies. These error matrices compute the
and these are summarized as follows:
following:
1. Error sources: Errors in thematic maps can occur as a 4. Three basic accuracies: An error matrix computes three
result of any number of causessystematic (e.g., sensor basic accuraciesproducers, users, and overall. Accuracies
calibration errors), natural (e.g., atmospheric), preprocess- are computed by comparing the thematic map classes with
ing (e.g., geometric and radiometric correction), deriva- reference data. Producers accuracy refers to percentage of
tive (e.g., NDVI), classification schemes and methods (e.g., class X in reference map that is correctly classified as class
digital image processing), class definitions, reference data X in a thematic map. A users accuracy is the percentage of
collection design (e.g., sample design, position inaccura- the pixels classified as a class X in a thematic map is indeed
cies), and user interpretation of classes and class labeling. class X. Overall accuracy is the percentage of pixels cor-
2. Classification: Thematic maps are produced by digital rectly classified from all classes to total number of pixels
image analysis. The four characteristics of classification from all classes. If sample sizes vary, so does the producers,
are identified as (a) complete definition, (b) mutually users, and overall accuracies. The larger the sample size,
exclusive, (c) totally exhaustive, and (d) hierarchical. the more robust these accuracies are.
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 659
Fuzzy component of many other technological delivery systems (e.g.,
Boolean Google Earth, web maps, and mobile delivery of location specific
Remote sensing Field survey
data). This means that the legal principles governing remote-
sensing technologies are increasingly being developed in the
Bare
context of the broader field of geospatial law. This is especially
so in light of the increasing number of EO satellites launched
and operated by various nations (e.g., Figure 31.29). Dr. P.J.
Blount in Chapter 30 not only provides an excellent discussion
Grazing land
on these matters but refers to an extensive set of international
and national treaties, regulations, agreements, and policies that
when combined form the complex web of regulations governing
these activities.
Broadly, the four pillars of remote-sensing law are
Urban
1. International space treaty regime
2. International remote-sensing law for military era or disar-
mament and verification law
Vegetation
3. International remote-sensing principles for civilian
purposes
4. National geospatial law principles
First, space treaty regime defines laws pertaining to
Woodland
1. Free access to outer space for any nation
2. Observation and collection of data for any part of the
world, cutting across sovereignty
3. Governments to authorize and supervise nongovern-
FIg u r e 31.28 The data collected in the field and classified from
remotely sensed data. In the Boolean maps, solid symbols () indicate mental or other players (e.g., commercial) satellites
where both field and RS agree, hollow circles () where only the RS indi- Second, disarmament and verification law was for military pur-
cates the class, and crosses (+) where only field survey indicates the class. poses involving control of ballistic missiles, freedom to monitor
In the maps of fuzzy classes, the size of the plot characters indicates the
military activities, and intelligence gathering.
degree of membership to the class. (From Comber, A., et al., Remote
Third, remote sensing for civilian purposes is guided by the
Sens. Environ., 127, 237, December 2012, ISSN: 0034-4257, http://dx.doi.
org/10.1016/j.rse.2012.09.005. Accessed JuneOctober, 2014.) United Nations General Assembly Resolution 41/65 and the
United Nations Committee on the Peaceful Uses of Outer Space,
which outlines the core legal principles dealing with remote-
5. Normalized accuracies: Margfit normalized the error
sensing activities for civilian purposes. These principles can be
matrix by removing the effect of sample size. Kappa
divided into three groups:
KHAT statistic is another measure of accuracy and is
effectively used as a way to test if one map is statistically 1. Preexisting binding legal principles
better than another. 2. Innovations involving international law governing remote
6. Fuzzy accuracy assessment: Instead of deterministic accu- sensing
racy assessment, like the ones mentioned earlier, fuzzy 3. Promoting the soft obligation of international cooperation
classification provides measure of accuracy that consid-
The centerpiece of the international law principles are the
ers possibilities such as correct, mostly correct, somewhat
freedom of data collection and distribution, while at the same
correct, mostly incorrect, somewhat incorrect, and incor-
time ensuring respect of the principle of full and permanent
rect (e.g., Figure 31.28; Comber etal., 2012).
sovereignty of all states and peoples over their own wealth and
natural resources. Through these measures, a large number
31.30 Remote-Sensing Law or Space Law of cooperation of internal remote-sensing data collection and
access is determined for human welfare. These efforts are fur-
Remote-sensing law is part of international space law as well as ther solidified through efforts through GEO and GEOSS where
the quickly developing field of geospatial law. The early evolution web-enabled (free) remote-sensing data sharing and distribution
of space law was based primarily on the military and strategic is encouraged for a wide range for peaceful applications (e.g., the
requirement of nations. However, wide and ubiquitous use of nine societal beneficial areas such as forests, agriculture, water,
remote sensing has resulted in broad set of legal principles gov- biodiversity, disasters).
erning remote sensing for civilian uses. The evolution of tech- Fourth, remote-sensing law is increasingly perceived to be
nology has contributed to remote sensing becoming an essential part and parcel of geospatial law. This as a result of the use of
2016 by Taylor & Francis Group, LLC
660 Remotely Sensed Data Characterization, Classification, and Accuracies
70
25 60
63
20 50
17
40
30
20
10
FIg u r e 31.29 Map showing the total number of near-polar orbiting, land imaging civilian satellites launched by (or on behalf of) different geo-
graphical regions between July 23, 1972, and December 31, 2013. The legend to the right of the map shows the number according to seven groups: 0,
15, 610, 1120, 2140, 4160, and 6070. Note that no region falls into the 4160 category. The numbers of launches made by the top four individual
countries (India, China, Russia, and the United States) are specifically citednote that collectively Europe has launched 30. (From Belward, A.S. and
Skien, J.O., ISPRS J. Photogram. Remote Sens., 2014, ISSN: 0924-2716, http://dx.doi.org/10.1016/j.isprsjprs.2014.03.009, available online: April 28, 2014.)
many interconnected technologies such as remote sensing, GIS, image analysis: A new paradigm in remote sensing and geo-
GPS, Internet delivery of data, mobile delivery of data, and num- graphic information science. ISPRS International Journal of
ber of other technological and commercial convergence of spatial Photogrammetry and Remote Sensing, 87(1), 180191.
and nonspatial data and their delivery systems. Remote sensing Branger, F., Kermadi, S., Jacqueminet, C., Michel, K., Labbas,
has the ability to gather information at great detail, repeatedly, M., Krause, P., Kralisch, S., Braud, I. November 15, 2013.
and over any area of the globe. This means that remote-sensing Assessment of the influence of land use data on the water
activities and data have become implicated in a wide variety of balance components of a peri-urban catchment using a
legal fields such as evidence, torts, and privacy, among others. distributed modelling approach. Journal of Hydrology, 505,
These issues are most often domestic in nature and operate out- 312325, ISSN: 0022-1694, http://dx.doi.org/10.1016/j.
side the international regimes jhydrol.2013.09.055.
Chander, G., Markham, B.L., Helder, D.L. 2009. Summary of cur-
Acknowledgments rent radiometric calibration coefficients for Landsat MSS,
TM, ETM+, and EO-1 ALI sensors. Remote Sensing of
I thank the lead authors and coauthors of each of the chapters Environment, 113, 893903.
for providing their insights and edits of my chapter summaries. Chavez, P.S. 1988. An improved dark-object subtraction tech-
nique for atmospheric scattering correction of multispec-
References tral data. Remote Sensing of Environment, 24, 459479.
Chavez, P.S. 1989. Radiometric calibration of Landsat thematic
Belward, A.S., Skien, J.O. 2014. Who launched what, when mapper multispectral images. Photogrammetric Engineering
and why; trends in global land-cover observation capac- and Remote Sensing, 55, 12851294.
ity from civilian earth observation satellites. ISPRS Journal Comber, A., Fisher, P., Brunsdon, C., Khmag, A. December 2012.
of Photogrammetry and Remote Sensing, ISSN: 0924-2716, Spatial analysis of remote sensing image classification accu-
http://dx.doi.org/10.1016/j.isprsjprs.2014.03.009, available racy. Remote Sensing of Environment, 127, 237246, ISSN:
online: April 28, 2014. 0034-4257, http://dx.doi.org/10.1016/j.rse.2012.09.005.
Blaschke, T. 2010. Object-based image analysis for remote sens- Damodaran, B.B., Nidamanuri, R.R. 2014. Assessment of the
ing. ISPRS International Journal of Photogrammetry and impact of dimensionality reduction methods on informa-
Remote Sensing, 65(1), 216. tion classes and classifiers for hyperspectral image clas-
Blaschke, T., Hay, G.J., Kelly, M., Lang, S., Hofmann, P., sification by multiple classifier system. Advances in Space
Addink,E., Feitosa, R., van der Meer, F., van der Werff, H., Research, 53(12), 17201734, ISSN: 0273-1177, http://
Van Coillie, F., Tiede, D. 2014. Geographic object-based dx.doi.org/10.1016/j.asr.2013.11.027.
2016 by Taylor & Francis Group, LLC
Remote Sensing DataCharacterization, Classification,and Accuracies 661
Dardel, C., Kergoat, L., Hiernaux, P., Mougin, E., Grippa, M., ISO, 2009. ISO 19115-2:2009 Geographic Information-Metadata-
Tucker, C.J. January 2014. Re-greening Sahel: 30 years of Part 2: Extensions for imagery and gridded data Workbook
remote sensing data and field observations (Mali, Niger). (2.99 MB) - Guide to Implementing ISO 19115-2:2009(E),
Remote Sensing of Environment, 140, 350364, ISSN: 0034- the North American Profile (NAP), and ISO 19110 Feature
4257, http://dx.doi.org/10.1016/j.rse.2013.09.011. Catalogue.
Dow, J.M., Neilan, R.E., Weber, R., Gendt, G. 2007. Galileo Jensen, J.R. 1996. Introductory Digital Image Processing: A Remote
and the IGS: Taking advantage of multiple GNSS con- Sensing Perspective, 3rd edn. Prentice Hall, p. 318, ISBN:
stellations. Advances in Space Research, 39(10), 1545 0-13-145361-0. Publisher: Prentice Hall, NJ.
1551, ISSN: 0273-1177, http://dx.doi.org/10.1016/j. Jin, S., Feng, G.P., Gleason, S. May 17, 2011. Remote sensing
asr.2007.04.064. using GNSS signals: Current status and future directions.
Elvidge, C.D., Yuan, D., Weerackoon, R.D., Lunetta, R.S. 1995. Advances in Space Research, 47(10), 16451653, ISSN: 0273-
Relative radiometric normalization of Landsat multispec- 1177, http://dx.doi.org/10.1016/j.asr.2011.01.036.
tral scanner (MSS) data using an automatic scattergram Ke, Y., Quackenbush, L.J., Im, J. 2010. Synergistic use of
controlled regression. Photogrammetric Engineering and QuickBird multispectral imagery and LIDAR data for
Remote Sensing, 61, 12551260. object-based forest species classification. Remote Sensing of
Evangelidis, K., Ntouros, K., Makridis, N., Papatheodorou, C. Environment, 114(6), 11411154, ISSN: 0034-4257, http://
February 2014. Geospatial services in the cloud. Computers dx.doi.org/10.1016/j.rse.2010.01.002.
& Geosciences, 63, 116122, ISSN: 0098-3004, http://dx.doi. Kronseder, K., Ballhorn, U., Bhm, V., Siegert, F. August 2012.
org/10.1016/j.cageo.2013.10.007. Above ground biomass estimation across forest types
Freire, S., Santos, T., Navarro, A., Soares, F., Silva, J.D., Afonso, at different degradation levels in Central Kalimantan
N., Fonseca, A., Tenedrio, J. 2014. Introducing map- using LiDAR data. International Journal of Applied Earth
ping standards in the quality assessment of buildings Observation and Geoinformation, 18, 3748, ISSN: 0303-
extracted from very high resolution satellite imagery. 2434, http://dx.doi.org/10.1016/j.jag.2012.01.010.
ISPRS Journal of Photogrammetry and Remote Sensing, Lillesand, T.M., Kiefer, R.W., Chipman, J.W. 2008. Remote Sensing
90, 19, ISSN: 0924-2716, http://dx.doi.org/10.1016/j. and Image Interpretation, 6th edn., p. 768, ISBN: 978-0-470-
isprsjprs.2013.12.009. 46555-4. Publisher: Wiley.
Fritz, S., McCallum, I., Schill, C., Perger, C., See, L., Markham, B.L., Barker, J.L. 1987. Radiometric properties of
Schepaschenko, D., van der Velde, M., Kraxner, F., U.S. Processed Landsat MSS data. Remote Sensing of the
Obersteiner, M. May 2012. Geo-Wiki: An online platform Environment, 22, 3971.
for improving global land cover. Environmental Modelling Luc Moreau; Paolo Missier; eds. PROV-N: The Provenance
& Software, 31, 110123, ISSN: 1364-8152, http://dx.doi. Notation. 30 April 2013, W3C Recommendation. URL:
org/10.1016/j.envsoft.2011.11.015. http://www.w3.org/TR/2013/REC-prov-n-20130430/.
Fujima, K. 2011. Tsunami measurement data compiled by IUGG Mori, N., Takahashi, T. 2012. The 2011 Tohoku Earthquake
tsunami commission, http://www.nda.ac.jp/cc/users/ Tsunami Joint Survey Group Nationwide post event sur-
fujima/TMD/index.html, accessed December 2, 2011. vey and analysis of the 2011 Tohoku Earthquake Tsunami.
Gbanie, S.P., Tengbe, P.B., Momoh, J.S., Medo, J., Kabba, V.T.S. Coastal Engineering Journal, 54, 1250001.
January 2013. Modelling landfill location using Geographic Nagamalai, D., Renaulat, E., Dhanuskodi, M. 2011. Advances
Information Systems (GIS) and Multi-Criteria Decision in Digital Image Processing and Information Technology:
Analysis (MCDA): Case study Bo, Southern Sierra Leone. First International Conference on Digital Image Processing
Applied Geography, 36, 312, ISSN: 0143-6228, http:// and Pattern Recognition, Tirunelveli, India. Series:
dx.doi.org/10.1016/j.apgeog.2012.06.013. Communications in Computer and Information Science.
Goward, S.N., Ghander, G., Pagnutti, M., Marx, A., Ryan, R., Paperback. Springer, Tirunelveli, Tamil Nadu, India p. 478,
Thomas, N., Tetrault, R. August 2012. Complementarity ISBN-10: 3642240542.
of ResourceSat-1 AWiFS and Landsat TM/ETM+ sensors. Neckel and D. Labs. 1984. The solar irradiance between 3300 and
Remote Sensing of Environment, 123, 4156, ISSN: 0034- 12500a. Sol. Phys., 90:205258.
4257, http://dx.doi.org/10.1016/j.rse.2012.03.002. Oppelt, N., Schulze, F., Bartsch, I., Doernhoefer, K., Eisenhardt, I.
Haas, J., Ban, Y. August 2014. Urban growth and environmen- 2012. Hyperspectral classification approaches for intertidal
tal impacts in Jing-Jin-Ji, the Yangtze, River Delta and the macroalgae habitat mapping: A case study in Helgoland. Optical
Pearl River Delta. International Journal of Applied Earth Engineering, 51, 111703, doi: 10/1117/1.OE.51.11.111703.
Observation and Geoinformation, 30, 4255, ISSN: 0303- Price, J.C. 1987. Calibration of satellite radiometers and the
2434, http://dx.doi.org/10.1016/j.jag.2013.12.012. comparison of vegetation indices. Remote Sensing of the
Heipke, C. November 2010. Crowdsourcing geospatial data. ISPRS Environment, 21, 1527.
Journal of Photogrammetry and Remote Sensing, 65(6), Richards, J.A., Xiuping, J. 2006. Remote Sensing Digital Image
550557, ISSN: 0924-2716, http://dx.doi.org/10.1016/ Analysis, Vol. XXV, 4th edn. Springer-Verlag, p. 439
j.isprsjprs.2010.06.005. Publisher: Berlin: Springer.
2016 by Taylor & Francis Group, LLC
662 Remotely Sensed Data Characterization, Classification, and Accuracies
Rodriguez-Galiano, V.F., Ghimire, B., Rogan, J., Chica-Olmo, Uhl, F, Oppelt, N., Bartsch, I. 2013. Mapping marine macroalgae
M., Rigol-Sanchez, J.P. January 2012. An assessment of in case 2 waters using CHRIS PROBA. Proceedings of the
the effectiveness of a random forest classifier for land- ESA Living Planet Symposium, September 913, Edinburgh,
cover classification. ISPRS Journal of Photogrammetry and U.K., ESA Special Proceedings SP-722 (CD-ROM).
Remote Sensing, 67, 93104, ISSN: 0924-2716, http://dx.doi. Uhl, F., Oppelt, N., Bartsch, I., Geisler, T., Heege, T., Nehring, F. 2014.
org/10.1016/j.isprsjprs.2011.11.002. KelpMapDevelopment of an EnMAP approach to moni-
Stosius, S., Beyerle, G., Hoechner, A., Wickert, J., Lauterjung, J. tor sublitoral marine macrophytes (KelpMapEntwicklung
March 1, 2011. The impact on tsunami detection from space eines EnMAP Verfahrens zur Bestimmung von sublitoralen
using GNSS-reflectometry when combining GPS with marinen Makrophyten). Final report of research project
GLONASS and Galileo. Advances in Space Research, 47(5), FKZ: 50EE1020, funded by the German Federal Ministry of
843853, ISSN: 0273-1177, http://dx.doi.org/10.1016/j. Economy and Technology (BMWi), 36pp.
asr.2010.09.022. Vermote, E.F., El Saleous, N.Z., and Justice, C.O. 2002. Atmospheric
Suppasri, A., Futami, T., Tabuchi, S., Imamura, F. October correction of MODIS data in the visible to middle infrared:
2012. Mapping of historical tsunamis in the Indian and First results. Remote Sensing of Environment, 83(12), 97111.
Southwest Pacific Oceans. International Journal of Disaster Vrieling, A., Meroni, M., Shee, A., Mude, A.G., Woodard, J.,
Risk Reduction, 1, 6271, ISSN: 2212-4209, http://dx.doi. (Kees) de Bie, C.A.J.M., Rembold, F. May 2014. Historical
org/10.1016/j.ijdrr.2012.05.003. extension of operational NDVI products for livestock
Thenkabail, P., Lyon, G.J., Turral, H., Biradar, C.M. 2009a. Remote insurance in Kenya. International Journal of Applied Earth
Sensing of Global Croplands for Food Security. Boca Raton, Observation and Geoinformation, 28, 238251, ISSN: 0303-
FL: CRC Press, Taylor & Francis Group, p. 556 (48 in color). 2434, http://dx.doi.org/10.1016/j.jag.2013.12.010.
Published in June 2009. Wan, Z., Hong, Y., Khan, S., Gourley, J., Flamig, Z., Kirschbaum,
Thenkabail, P.S., Biradar, C.M., Noojipady, P., Dheeravath, V., Li, D., Tang, G. August 2014. A cloud-based global flood
Y.J., Velpuri, M., Gumma, M. etal. July 20, 2009b. Global disaster community cyber-infrastructure: Development
irrigated area map (GIAM), derived from remote sensing, and demonstration. Environmental Modelling & Software,
for the end of the last millennium. International Journal of 58, 8694, ISSN: 1364-8152, http://dx.doi.org/10.1016/j.
Remote Sensing, 30(14), 36793733. envsoft.2014.04.007.
Thenkabail, P.S., Gamage, N., and Smakhin, V. 2004. The use of Wang, Z., Jensen, J.R., Im, J. October 2010. An automatic region-
remote sensing data for drought assessment and monitoring based image segmentation algorithm for remote sensing
in south west Asia. IWMI Research report # 85. Pp. 25. IWMI, applications. Environmental Modelling & Software, 25(10),
Colombo, Sri Lanka. http://www.iwmi.cgiar.org/Publications/ 11491165, ISSN: 1364-8152, http://dx.doi.org/10.1016/j.
IWMI_Research_Reports/PDF/pub085/RR85.pdf. envsoft.2010.03.019.
Thenkabail, P.S., Smith, R.B., De-Pauw, E. 2002. Evaluation of Wei, Y., Liu, H., Song, W., Yu, B., Xiu, C. August 2014. Normalization
narrowband and broadband vegetation indices for deter- of time series DMSP-OLS nighttime light images for urban
mining optimal hyperspectral wavebands for agricultural growth analysis with Pseudo Invariant Features. Landscape
crop characterization. Photogrammetric Engineering and and Urban Planning, 128, 113, ISSN: 0169-2046, http://
Remote Sensing 68(6), 607621. dx.doi.org/10.1016/j.landurbplan.2014.04.015.
Tilling, A.K., OLeary, G.J., Ferwerda, J.G., Jones, S.D., Fitzgerald, Whiteside, T.G., Boggs, G.S., Maier, S.W. December 2011.
G.J., Rodriguez, D., Belford, R. OctoberDecember 2007. Comparing object-based and pixel-based classifications for
Remote sensing of nitrogen and water stress in wheat. Field mapping savannas. International Journal of Applied Earth
Crops Research, 104(13), 7785, ISSN: 0378-4290, http:// Observation and Geoinformation, 13(6), 884893, ISSN:
dx.doi.org/10.1016/j.fcr.2007.03.023. 0303-2434, http://dx.doi.org/10.1016/j.jag.2011.06.008.
Tilmes, C., Fleig, A.J. 2008. Provenance tracking in an earth sci- Xu, H., Huang, S., Zhang, T. October 15, 2013. Built-up land map-
ence data processing system. In: J. Freire, D. Koop (eds.), ping capabilities of the ASTER and Landsat ETM+ sensors
Provenance and Annotation of Data and Processes. Berlin, in coastal areas of southeastern China. Advances in Space
Germany: Springer-Verlag, pp. 221228. http://ebiquity. Research, 52(8), 14371449, ISSN: 0273-1177, http://dx.doi.
umbc.edu/_file_directory_/papers/445.pdf. org/10.1016/j.asr.2013.07.026.
Tucker, C.J. 1979. Red and photographic infrared linear combi- Yue, P., Gong, J., Di, L., Yuan, J., Sun, L., Sun, Z., Wang, Q. 2010.
nations for monitoring vegetation. Remote Sensing of the GeoPW: Laying blocks for geospatial processing web.
Environment, 8, 127150. Transactions in GIS, 14(6), 755772.
U.S. Geological Survey (USGS). 2012. Earths Tectonic Plates Zhao, P., Foerster, T., Yue, P. October 2012. The geoprocessing
USGS, http://earthquake.usgs.gov/regional/nca//kml/ web. Computers & Geosciences, 47, 312, ISSN: 0098-3004,
Earths_Tectonic_Plates.kmz, accessed May 25, 2012. http://dx.doi.org/10.1016/j.cageo.2012.04.021.
2016 by Taylor & Francis Group, LLC
Index
A ARNS band, see Aeronautical Radio algorithms, selection
Navigation Service (ARNS) band categories, 380381
Accuracy assessment Artificial neural networks (ANNs), 213, image-processing techniques, 380381
deterministic error matrix, 599 236238, 637, 639 per-pixel-based approaches, 381
high-resolution imaging, 598 Asia Sentinel, 612 subpixel-based and object-based
human interpretation, 598 ASPRS, see American Society for approaches, 381382
off-diagonal elements, 598 Photogrammetry and Remote binary change, 380, 647
producers and users accuracies, 599 Sensing (ASPRS) continuous change, 647
reference data, 598 ASTER, see Advanced space borne thermal conversion, 647
standard error matrix, 598 emission and reflection radiometer definition, 378
Advanced Earth observing satellite (ADEOS), (ASTER) factors, 646
65, 113 Automated image classification algorithms forest change, 377
Advanced land observing satellite (ALOS), nonparametric tools land cover/use, 647
48, 66, 71, 7677, 136, 626 ANNs, 236238 land-use/land-cover change, concepts
Advanced space borne thermal emission and ART framework, 236 of,378
reflection radiometer (ASTER), 10, classification results for SVM, 238 on planet Earth, 646
66, 68, 75, 138, 574 CT classifier, 236 remote sensing variables, selection,
Advanced very high resolution radiometer LIBSVM implementation, 236237 378380
(AVHRR), 64, 93, 143, 178, 235, MODIS time-series data, 238 RF classifiers, 647
261, 268, 570571, 633 SVMs, 236 scale issues on design, impact
Aerial photography parametric tools coarse-spatial-resolution images,
digital data processing, 63 KMC algorithm, 237 382383
high resolution imaging, 566 MD procedure, 237 high-spatial-resolution images, 382
NAIP imagery, 26 MLC procedure, 236237 medium-spatial-resolution images, 382
NAPP and NHAP, 26 AVHRR, see Advanced very high resolution using multisensor data, 384
photogrammetry/aerial stereo radiometer (AVHRR) using optical and radar data (see Land
photography, 26 cover)
reference data collection, 592 Charge-coupled device (CCD), 7
remote sensing systems, 6
B
China/Brazil Earth Resources Satellite
satellite sensor platforms, 49 Ballistic missile technology, 607 (CBERS), 63
soft-copy photogrammetry, 26 BeiDou-2, see Compass (BeiDou-2) Circular map accuracy standard (CMAS), 585
stereoscopy, 26 Bidirectional reflectance distribution CITI-SENSE project (CITISENSE, 2012), 546
Aeronautical Radio Navigation Service function (BRDF), 135, 144, 178 Citizen Observatory Web (COBWEB, 2012)
(ARNS) band, 484, 486 Big equation, 94 project, 546
Airborne laser scanning (ALS), 344, 346347 Bing Maps and MapQuest, 432 Citizen Sensor Networks
Airborne LiDAR systems, 344345 Brazils National Institute for Space definition, 538
ALOS, see Advanced land observing satellite Researchs (INPE) policy, 612 global population density, 540, 542
(ALOS) BRDF, see Bidirectional reflectance Google Maps version vs. OSM version,
ALS, see Airborne laser scanning (ALS) distribution function (BRDF) 538539
Amazon Web Services (AWS), 556 Brightness value (BV), 12, 15, 198 OSM node density, 540, 542
American Society for Photogrammetry and VGI, 538
Remote Sensing (ASPRS), 586 Citizens Observatory for Coast and Ocean
c
Amtliche Topographisch-Kartographische Optical Monitoring (CITCLOPS)
Informationssystem (ATKIS) Calibration, definition, 89 project, 546547
data,437 Calibration-validation (cal-val), 89 Civil navigation (CNAV) messaging
ANNs, see Artificial neural networks (ANNs) Canadas Constitution and Privacy Act, 616 structure, 486
Apparent thermal inertia (ATI), 270 Challenging minisatellite payload Clean code (C/A code) method, 502
Application programming interface (CHAMP),494 Climate Absolute Radiance and Refractivity
(API),566 Change detection techniques Observatory (CLARREO), 98, 119
663
2016 by Taylor & Francis Group, LLC
664 Index
Climate data analysis tools (CDATs), 417 Cosmos 954, 607 Data processing and applications, LiDAR
Cloud computing Cosmos-Iridium collision, 607 agriculture, 646
aerial platforms, 554 Crowdsourcing categories, 644
big-data advantages, 536 city modeling, 646
computational power, 554 applications, 654 data quality, 644645
database technologies, 554 attractive and desirable, 655 DEM, 366367, 646
GPUs, 554 building vulnerability to natural disasters, 646
HPC, 554 disasters,654 distribution, 646
LiDAR, 554 concept of on-the-fly recruiting, 535 DSM, 366367, 646
maximum likelihood, 554 countermeasures act, 536 DTM, 646
cluster computing, 555 creative, 536 flood mapping and assessment, 646
commoditization, 557 crowd-funding, 536 forestry studies, 646
costbenefit analysis, 556 crowd-voting, 536 full-waveform airborne processing, 645
definition, 656 distributed data collection, 536 hydrology and geomorphology, 646
elasticity and abstraction, 555 distributed intelligence, 536 independent variables, 644
GIS, 556 Geo-wiki for land cover mapping, 654 land cover classification, 646
grid computing, 555 Internet, 536 laser scanning system, 644
high data reliability, 557 microwork, 536 management, 645
HPC, 556 model of outsourcing, 536 noise filtering, 644
HTC, 555 OpenStreetMap, 654 point clouds, 644645 (see also Point
hybrid clouds, 556 passive mapping, 654655 cloud data, LiDAR)
internet data accessibility, 554 production/problem-solving processing 3D points, 644
Landsat satellites, 553 models,536 quality assessment and control,
NASA systems, 553 untrained volunteers, 655 348350,645
open system architecture, 656 Web 2.0, 535536 structures and objects, 646
private clouds, 556 wiki philosophy, 536 3D coordinates, 644
public and private entities, 656 wisdom of the crowd, 536 TIN, 559560, 646
public clouds, 556 CudaGIS, 451 waveform, 360362, 644
and RS data, 656 CyberGIS, 451 Data protection, concept of, 439
services, 656 Cyberinfrastructure (CI), 410 Data Quality Provenance System, 417
software implementation, 557 Cyclone global navigation satellite system Data standardization, satellite image
topographic information, 553 (CYGNSS) project, 494, 651 radiometry
UAVs, 554 calibration, definition, 89
Cloud service models, 556557 calibration-validation, 89
D
Cold War foreign policy, 612 GEOSS, 89
Color infrared (CIR) photographs, 63 Data mining, 438439 QA4EO, 89
Committee on Earth Observation Satellites Data normalization, remote sensing, 133 and quality assurance, 88
(CEOS), 135, 630 atmosphere reflectance, 633 satellite data requirements, 8889
Common Scents project, 435 atmospheric correction methods, 633 space-based observations of Earth, 88
Community building digital number, 631 WGCV, 89
Haitian HOT, 434 factors Defense Meteorological Satellite Program
LKKCAP, 434 satellites, 630 Operational Linescan System
OSMs Map Kibera project, 434 seasons, 631 (DMSP-OLS), 634
PPGIS/PGIS projects, 434435 sensors, 631 Degree of dissimilarity, 332
Community internet intensity map (CIIM), solar flux/irradiance, 631 DEM, see Digital elevation model (DEM)
542543 stratosphere/atmosphere, 631 Democratization of geo-information
Compass (BeiDou-2) Sun, 631 community building, 434435
frequency, 491 SunEarth, 631 disaster relief and emergency
satellites, 491 surface of Earth, 631 management, 433434
Complementary metal-oxide semiconductor imagery, types, 134 empowerment and equity
(CMOS), 7 optical and thermal remote sensing, 134 DigitalGlobe, 440441
Computation, thematic maps passive remote sensing, 134 geography of Wikipedia, 440, 649
commission errors, 594 radiometric (see Radiometric Linus law, 440
error matrix, 594 normalization) neogeography movement, 440
forest sample units, 594595 radiometric cross-calibration, 140 Typhoon Haiyan-related mapping,
overall accuracy, 595 radiometric distortion, sources, 441, 649
producers accuracy, 595 135136 GISs and GNSSs, 430431, 648
users accuracy, 595 remote sensing data, 134135 LBS (see Location-based services (LBS))
Conference on Disarmament (CD), 607608 SAR data, 133135 map mash-ups or mash-ups, 431
Conference on Spatial Information Theory sensor characteristics, 630 National Map, 430
(COSIT), 447 TOA reflectance, 633 new Geo-ICTs, 430431, 648
Convolution kernel, 207208 wavelengths, electromagnetic participatory sensing, 435436
Cosine topographic correction model, 135 spectrum,134 privacy and confidentiality, 438440
2016 by Taylor & Francis Group, LLC
Index 665
quality of information and services, classification Disaster relief and emergency management
437438 supervised techniques, 212 Haiti earthquake, 433
remote sensing, 430431 unsupervised techniques, 212213 HOT, 433
technologies see Geo-information (GI) data fusion, 637 Hurricane Katrina, 433
technologies decision tress, 637 remote sensing and GISs, 433
in twenty-first century EME, 198 VGI/crowdsourcing mapping, 433434
GIS, 424425 enhancement Disasters Charter, 612
GIScience, 424 composite generation, 200, 203 Doppler radar, 6
Geo-ICTs and Web 2.0 services, 424 density slicing, 199, 202 DSM, see Digital surface model (DSM)
GNSS, 424425 spatial and spectral profiles, 199, 202 DYFI service, see Did You Feel It (DYFI)
Google Earth, 424 techniques/effects/application service
high-resolution satellite imagery, 424 examples, 199, 201
A Long Way Home, 423424 Fourier transforms, 637
e
remote sensing, 424425 future trends, 215216
Derivative of morphological profile (DMP), GEOBIA, 637 Earth observation (EO) satellites, 63, 628, 630
329, 332 OBIA, 637 Earth Observing-1 Advanced Land Imager
Did You Feel It (DYFI) service PCA, 207 (EO-1 ALI), 647
accuracy, 543 phenological matrices, 637 Earth Observing System (EOS) MODIS
CIIM, 542543 preprocessing (see Image preprocessing) whiskbroom sensors, 111
distribution of questionnaires filed, quality assessment Earth resources technology satellites (ERTS),
542543 correlation coefficient, 198199 63, 626
earthquake hazard models, 542 histogram, 199200 Earth Systems Science Workbench
MMI, 543 statistical information, 198 (ESSW),417
products, Central California event, simple ratio, 208, 211 eCognition software, 464
541,543 SMTs, 637 EGALITE project, 513, 524
public engagement, 542 spatial filtering Electromagnetic energy (EME), 8, 198
questionnaire, 543 convolution kernel, 207208 Electromagnetic radiation (EMR), 5
USGS, 542 Fourier transform (FT) on Landsat plane of polarization, 89
USGSs ShakeMap, 543 TM data, 208 transmission direction and amplitude, 8
website, 542 low- and high-frequency images, wavelength/frequency, 89
Differential interferometric SAR (D-InSAR) 207208 Electromagnetic spectrum (EMS), 5
data acquisition, 6667 low/high pass filters, 208210 EME, 8
Digital data processing spectral unmixing, 637 EMR, elements, 89
advent of satellite-borne instruments, 63 standards, 637 remote sensing systems, 910
aerial photography, 63 SVMs, 637 Elementary_geoParticle, 467468
CIR photographs, 63 VI (see Vegetation indices (VI)) Elfouhaily model, 495496
ENVISAT, 66 Digital numbers (DNs), 135 EMR, see Electromagnetic radiation (EMR)
ERS satellite, 66 Digital provenance and geospatial workflows EMS, see Electromagnetic spectrum (EMS)
ERTS, 63 in commercial and public Enhanced thematic mapper plus (ETM+),
fAPAR, 66 geoprocessing,409 145146
Landsat mission timeline, 63 component in Enhanced vegetation index (EVI), 176,
MODIS/MERIS, 66 geo-cyberinfrastructure,410 212213, 634
MSS, 63 in geospatial lineage aerosol contaminations, 156
NOAA, 6364 Geolineus, 406409 climatevegetation interactions, 155156
Sputnik (first satellite into space), 63 Geo-Opera, 409 disadvantages, 156
Terra, 66 shared provenance-aware remote sensing and EVI2
TIROS-1, 63 workflows atmospheric correction, simulation
Digital Earth, 427428 metadata interchange standards, 411 of, 161
Digital elevation model (DEM), 95, 99, provenance-specific (nonmetadata) AVHRR, intersensor compatibility,
345346, 365366, 566 interchange standards, 411412 165166
airborne InSAR, 74 Digital signatures, 617 Hyperion data and preprocessing,
ASTER stereopair images, 75 Digital surface model (DSM), 74, 345, 365366 160161
DLR, 75 Digital terrain model (DTM), 74 intersensor, 162163
DSM, 7475 Direct georeferencing (DG), 344 intersensor calibration, 166169
DTM, 74 Disarmament and verification law intrasensor, 163, 165
SRTM mission, 74 ballistic missile technology, 607 multisensor compatibility studies, 157,
TerraSAR-X and TanDEM-X, 75 CD in 1978, 607608 159160
DigitalGlobe, 440441 disarmament and arms control sensors and application fields, 158159
Digital image processing, 584 agreements, 607 sRMPD values, 165
active/passive sensors, 198 Landsat-1, 607 statistical difference analysis, 161162
ANN, 637 NTM, 607 LVI, 157
band ratio, 209 satellite technology, 607 NDVI, 156
brightness value, 198 UN Charter, 607 OLI, 156
2016 by Taylor & Francis Group, LLC
666 Index
PROBA-V, 156 G commercial dominance, phase of, 446
sensors and application fields, 158 computer cartography, 447
Suomi NPP VIIRS, 156 Galileo, 511 COSIT, 447
constellation and levels of services, 490 and GI systems, 448
vegetation productivity, 156
description, 490 IJGIS, 447
Environmental Systems Research Institute
GIOVE, 490 national drivers, phase of, 446
(ESRI), 5
navigation data and signals pioneers, phase of, 446
ENVISAT ASAR, 66
E1 and E6 signal, 490 research themes, 448
Equator crossing time (ECT), 147148 E1 frequency, 490
Error budget analysis, 591 spatial errors, 447
E5 signal, 490 UCGIS definition, 447
Estimation of scale parameter (ESP), Galileo OS Interface Control user dominance, phase of, 446447
464465 Document, 490 GIS technology, 445
European Geostationary Navigation Overlay navigation formats, 490 landfill site, 649651
Service (EGNOS), 511 signals and parameters, 490 map libraries, 445
European Remote Sensing (ERS) satellite, 66 Galileo Open Service (OS) Interface Control research expands, 651
European Space Agency (ESA) SAR imagery, Document, 490 spatial thinking and computing, 651
34, 36 GE, see Google Earth (GE) synonyms, 446
EVI, see Enhanced vegetation index (EVI) GEM, see Global earthquake model (GEM) Geographic information system (GIS), 219,
Experimental results in segmentation of GEOBIA developments 277, 424425, 445, 556
Landsat 1 (ERTS-1) imagery, 279 emergence Geo-ICTs, 424, 439
Extended morphological profiles (EMPs), 250 business-as-usual pixel-based Geo-information (GI) technologies
Extreme Science and Engineering algorithms, 279 contemporary phase (twenty-first
Development Environment eCognition software, 279 century)
(XSEDE), 410 Fractal Net Evolution, 279 cooperation and data sharing, 425
high-resolution multiband Digital Earth, 427428
imagery,279 geobrowsing, 428429
F Landsat imagery, 280 GISs and technologies, 427
QuickBird satellite image, 280 national SDIs, 427
Fast Line-of-Sight Atmospheric Analysis
satellite imagery, 279 neogeography, 429
of Spectral Hypercubes
establishment (20052010) OSM and Wikimapia, 429
(FLAASH),203
accuracy, 280 USGS National Map, 429430
Federal Emergency Management Agency
applications, 280 vision of democratization, 425,
(FEMA), 585 data fusion, 280281 427430
Federal Geographic Data Committee (FGDC) software, 281282 democratization (see Democratization of
(ISO, 2009), 648 Geobrowsing, 428429 geo-information)
Fluorescence Imaging Spectrometer Geographic information science enlightened phase (twentieth century)
(FLORIS), 77 (GIScience),424 commercial software, development
Forest change detection, GNSS-R, 504505 abstraction of, 426
Forward-looking infrared (FLIR), 6 cognitive research, 449 national mapping agencies,
Fourier transform (FT) on Landsat TM data layers, 449 development of, 425
data,208 geographic information, 449450 state-sponsored mapping initiatives,
Fraction of absorbed photosynthetically information sharing and advent of, 425
active radiation (fAPAR), 66 integration,449 technological developments, 425426
French civilian remote sensing policy, 612 virtual reality and multidimensional transition phase (twentieth century)
French Satellite Pour lObservation de la GIS, 450 community integrated GISs, 426
Terre (SPOT) satellites, 112 algorithms, 453 democratization of Geo-ICTs, 427
Full aperture solar calibrator (FASC), 112 cloud computing leverages, 451 GIS and society research agenda,
Full-waveform airborne LiDAR crowdsourcing, 450 development of, 427
applications, 365 CudaGIS, 451 GIS wars, 426
CyberGIS, 451 governance and scholarship for
data record, 362
Geo-Narrative, 450 society, 425427
data storage capacity and processing,
introduction of big data, 451 NSDIs, 427
361362
rise of critical GIS, 450 online map-based surveys, 427
definition, 345
VGI, 450 participatory/community mapping
echo detection, 363364 assimilation projects, 426
shape of received waveform, 363 definition, 451 PGIS, 426
waveform feature extraction GIS software packages, 451 PPGIS, 426
amplitude, 364 R project, 451452 Geolineus
backscatter coefficient, 365 spatial analysis and modeling, 452 Ancestors and Descendants functions, 408
backscatter cross section, 364365 CyberGIS and Cloud GIS, 452453 Condense function, 408409
Gaussian waveform, 365 development, 649, 651 geospatial data taxonomy, 409
pulse width, 364 emphasis of space, 453 GUI, 406407
waveform LiDAR systems, 360361 evolution lineage metadata, 408
2016 by Taylor & Francis Group, LLC
Index 667
metadata comparison functions, 408 Geospatial information and communication Global Climate Observing System
multiprocessing capabilities, 407 technologies (Geo-ICTs), 424 (GCOS),115
source/derived dataset, 407408 Geospatial law Global Earth Observation System of Systems
Geons commercialization, 614615 (GEOSS), 76, 89, 110, 630
aggregation approaches, 465 definition, 614 Global earthquake model (GEM)
a priori geographies, 462 globalization, 614615 collaborative building footprint
conditioned information, 461 intellectual property rights, 617 digitization, 545546
DEM, 461 privacy and security crowdsourcing concept, 546
ESP, 464465 Article 17 of Space Development direct observation (DO) tools, 546
image segmentation, 460 Promotion Act, 616 GED, 543544
integrated spatial analysis methods, 460 fundamental right, 616 IDCT, 544, 546
local variance (LV) graphs, 465 Google suite of geo-technologies, 616 Nexus, 546
monitoring methodology, 616 OpenQuake system, 543
OFA, 465466 9/11 terrorist attacks, 615 operation principle, GEM-IDCT, 544, 546
spatial variability, 465 Posse Commitatus Act, 615 seismic risk, 543
threshold-based intervention remote sensing technology, 616 Global exposure database (GED), 543544
planning, 466 satellite technology, 616 Global instrumented and automated network
multidimensional attribute/indicator surveillance, 616 of test sites (GIANTS), 115
spaces, 462 thermal imager, 616 Global navigation satellite system (GNSS), 45,
normalization method, 465 U.S. law, 616 424425, 651, 653
OBA, 461 U.S. military satellites, 615 applications
object validity, 467 remote sensing data, 616617 accident and disaster recovery
place-based vulnerability, 462 technological convergence, 614615 applications, 523525, 652653
regionalization vs. indexing, 462 torts, 617 aeronautic applications, 527528, 653
scale levels Geospatial lineage consumer applications, 521, 652
intrinsic scale, 467 classification of datasets, 405406 health applications, 525, 653
modeling/observational scale, 467 ESRIs ARC/INFO Geographic industrial applications, 522, 652
policy scale, 467 Information System, 404405 military applications, 527, 653
societal vulnerability, 461 Geolineus project, 405 surveying applications, 523, 652
spatial indicators, 460 Lanter classified datasets, 405 time synchronization applications, 527
spatial regionalization, 460 Lanters lineage metadata structure, 406 tracking applications, 525526, 653
stages Mark II with partial lineage transportation applications,
data preprocessing, 462463 capabilities,405 522523,652
regionalization, 462464 source datasets, 405 unmanned vehicles applications,
theories and frameworks, 462463 Geospatial methods 526527
visualization, 464 classification procedures and assisted GNSS (A-GNSS), 525
terminology, 459460 characteristics, 221 BeiDou Navigation Satellite System, 491
theory of recognition-by-components environmental indices, calculations, characteristics, 651
(RBC), 459 219220 consumer grade GNSS positioning
uncertainty and sensitivity analysis, 465 image characterization and modeling precision
Geo-Opera, 409 system, 226 GPS and GLONASS data, 518
Geoprocessing image spatial co-occurrence matrix/ PolaRx-3 receiver, 517
definitions, 402404 GLCM, 226 receiver chips, mobile devices, 517
developmental cycle, 402403 isarithm method, 226 Samsung Galaxy SIII, 517
digital (see Digital provenance and lacunarity approaches, 226 contemporary
geospatial workflows) Morans I and Gearys C spatial GALILEO, 511
forest change map, 647 autocorrelation measures, 226 GPS-NAVSTAR, 510511
framework, 647648 multiresolution analysis, 226 Russian GLONASS, 510511
geo-cyberinfrastructure, 402 overview, 220 disaster rescue and recovery, 525
geospatial provenance, 401402 variograms to measurements, 226 errors
machine-interpretable geospatial wavelets, texture analysis, 226227 distortions of propagation of
process,402 Geospatial object-based image analysis signals,512
provenance in remote sensing workflows (GEOBIA), 224 Earth tides, 512
(see Provenance) GEOSS, see Global Earth Observation System European Terrestrial Reference System
research implications, 418 of Systems (GEOSS) 1989 (ETRS89), 512
Snap Raster, 402403 Geostationary Coastal and Air Pollution GALILEO, encrypted navigation
GeoPWProv, 418 Events (GEO-CAPE), 77 services, 512
GeoScience Australias policy, 612 Geosurveillance, 438, 440 GPS and GLONASS, temporal
Geoslavery, 439440 Geo-wiki, 545546, 649, 654 degradation, 512
Geospatial data integration, OBIA GIS, see Geographic information system (GIS) IERS, 512
assessing accuracy, GEOBIA, 643644 GIScience, see Geographic information ITRS, 512
object identification, 643 science (GIScience) satellite position errors, 512
object validation, 642643 GIS wars, 426 satellites configuration, 512
2016 by Taylor & Francis Group, LLC
668 Index
Galileo, 490491 military requirements, jamming avian flu virus, 569570
GLONASS (see Global Orbiting environment, 484 binary application, 565, 567
Navigation Satellite System multipath mitigation capability, 484 cost and accessibility, 573
(GLONASS)) phase measurements, 484 desktop version, 565
GPS (see Global positioning system (GPS)) robust dual-frequency ionosphere EarthViewer, 565
GPS satellite constellation, 652 correction capability, 484 EE, 657
GSTB-V2 station network, 651, 653 orbits, 482 geocoded imagery, 656
low-cost receivers satellite blocks, 488 geospatial, 566
LEA-6S receiver, 518519 selective availability, 483 geospatial visualizations, 569571
NEO-7 chip, 518520 signals GIS data, 566
NFC card, 524 BPSK, 483 GUIs, 573
operational, 651 cesium and/or rubidium atomic high resolution imaging, 566569
receivers, 524 clock,482 Holocene volcanoes, 570571
reflectometry, 651, 653 (see also GNSS internal clock correction, 482 information, 571572
reflectometry (GNSS-R)) space segment, 47 KML, 566, 569, 571, 573
SBAS (see Satellite-based augmentation trilateration technique, 45 leveraging, 574
system (SBAS)) unit, 344 mobile version, 565
segments Global positioning systemnavigation NGO, 571, 573
control, 509510 satellite time and ranging (GPS nongovernmental organizations and
space, 509 NAVSTAR), 510 bloggers, 657
user, 510 Global Space-Based Inter-Calibration System operating systems, 565
troposphere and ionosphere errors (GSICS), 110 philosophy of, 656
mitigation techniques GLONASS, see Global Orbiting Navigation satellites, 577
correction function, 514 Satellite System (GLONASS) Skybox imaging, 574, 576577
GNSS, see Global navigation satellite system terrestrial remote sensing, 577
EGALITE project, 513
(GNSS) time-lapse, 574575
positioning accuracy, 514516
GNSS radio occultation (GNSS-RO) Tsunami field survey data, 657
signal propagation delay, 513514
atmospheric and ionospheric sounding by VHRI, 656657
world as functional planet, 482
mission, 494 volcanology, 570
Global Orbiting Navigation Satellite System
CHAMP, 494 water resources management, 570
(GLONASS)
COSMIC website, 494 Google Glass, 427, 432433
CDMA modernization, 489490
GPS/MET satellite mission, 494 Google Maps, 432
Galileo, 490491
LEO satellites, 495, 505506 Googles Street View mapping technology,
orbits, 488
meteorological/climate community, 505 431, 439
signals
SAC-C satellites, 494 GPS, see Global positioning system (GPS)
carrier components, 489
TechDemoSat-1, 494495 GPS/MET (GPS/Meteorology) satellite
FDMA, 489
UK-DMC, 494 mission, 494
frequency-division multiplexing, 488 GNSS reflectometry (GNSS-R) Graphical user interfaces (GUIs),
navigational data transmission, 488 CYGNSS project, 494 406407,573
navigation data format, 488 designed and manufactured, 493 Graphics processing units (GPUs), 554
satellite families, 489 geodetic technique, 506 Gray-level co-occurrence matrix
spreading codes and modulation, 488 geophysical parameters, 494 (GLCM),226
Global positioning system (GPS); see also geosciences community, 505 Gross primary productivity (GPP)
Global navigation satellite system land applications, 651 estimation,157
(GNSS) forest change detection, 504505 Ground sample distance (GSD), 12
data collection, segments, 4546 soil moisture, 503504 Group on Earth Observations (GEO), 547, 630
elements, modernized GPS land-based platform, 494 GSTB-V2 station network, 651, 653
L1C signal, 487488 L-band, 494 The Guardian of the Environment, 611
L2C signal, 485 LHCP antenna, 494
L5 signal, 485486 NASA and ESA, 651
M codes, 487 H
ocean applications, 651
evolution, 486 surface altimetry (see Sea surface Health Insurance Portability and Accounting
GPS.gov, 46 altimetry) Act, 616
GPS III, 488 surface wind speed estimation (see Sea Hierarchical step-wise optimization (HSWO)
high-accuracy GPS, 47 surface wind speed estimation) approach, 322323, 336337
image classification procedures, 47 PARIS project, 494 High performance computing (HPC), 556
location-based information, 47 receiver platform, 493494 High-precision navigation signal (HPNS), 489
modernization RHCP antenna, 494 High resolution imaging
better civil code, 484 satellite mission, 494495 aerial photography, 566
compatibility and operability, 485 space agencies, 494 armchair remote sensing, 568
dual-frequency navigation signals, Google Earth (GE), 424, 427, 656657 broadband Internet connectivity, 567
ARNS band, 484 API, 566 DEM, 566
increased power, 485 AVHRR, 570571 disaster response, 568
2016 by Taylor & Francis Group, LLC
Index 669
LiDAR, 567 satellite sensors, data characteristics, commercial and noncommercial
NGA, 568 1518 segmentation software
NOAA, 568 spatial resolution packages,643
satellite data, 567 angular resolution of sensor, 15, 18 D8- and SEGEN-based classification, 336
SRTM, 566 brightness value, 12, 15 image partitions, 317
stratigraphy, 568569 classes, 1819 image segmentation quality evaluation,
High-throughput computing (HTC), 555 dual spatial resolutionbased image 335338
Hough transform, 356 acquisition, 18 kappa coefficient (), 338
HSWO approach, see Hierarchical step-wise GSD, 12 k-means algorithms, 644
optimization (HSWO) approach IRS-1C image, 19 MSF segmentation, 336337
Humanitarian OpenStreetMap Team panchromatic bands of satellites, 19 overall accuracy (OA), 338
(HOT),433 sensor specifications, 18 Potts model, 338
Human vs. machine vision spectral resolution PV region-based classification
cerebral reaction, 301 EMS, wavelength intervals, 19 approach,336
gestalt approach, 302 Hyperion sensor, 20 quantitative segmentation quality
multiobject recognition, 302 MODIS sensor, 19 evaluation
pattern recognition, 301 spectral reflectance curves, 20 empirical evaluation function, 336
visual information processing, levels, 301 U.S. Landsat, 19 entropy-based evaluation
Hyperspectral image processing temporal resolution, 12, 2022 function,336
classification approaches Image classification methods, land cover and fitness function, 336
semisupervised classification, 251252 land use fragmentation and area-fit-index, 335
spectralspatial classification, 250251 advantages, 231232 geometric features circularity and
subspace-based approaches, 251252 automated (see Automated image shape index, 335
supervised classification, 249 classification algorithms) SVM classifier, 336
cloud and cloud shadow masking, weighted disorder function, 336
contributions, 255256
239240 RISA, 644645
data reduction and classification methods,
CT, 639
640641 ROSIS, 336
definitions, 232
HySens project, 252253 spatially based segmentation approaches
future development, 241
kernel methods, 640 graph-based algorithms, 332334
historical perspective, 233
probabilistic classification methods, 640 morphological algorithms, 327332
large area mapping, 240241
semisupervised approaches, 248249 MRF-based algorithms, 335
manual/visual interpretation, satellite
semisupervised classification, 640 region-growing algorithms, 321322
data, 233234
spatial and spectral information, texture-based algorithms, 325326
nonparametric methods, 639
247248,255 spectrally based segmentation approaches
parametric classification algorithms, 639
spectralspatial classification clustering-based algorithms, 318319
pixel vs. object based classifications, 239
approaches,640 support vector machines, 319320
remote sensing for map making,
subspace-based classifier, 255256 thresholding-based algorithms, 318
advantages and limitations, 232
supervised classification, 247248, 640 SVMs classification, 318, 337
RF, 639
SVM classifier, 640 supervised/unsupervised logic, 639 upland and lowland vegetation, 644
combination of the morphological uncertainty assessment, 241 Image-to-image registration process, 205
EMP, 254 Image preprocessing Image-to-map rectification process
contributions, 256 geometric correction, 204206 digital orthophotoquads, 205
discriminant classifiers, 253 external and internal, 204 digital planimetric maps, 205
output of pixel-wise, 254255 image-to-image registration GPS instruments, 205
RHSE, 255 process,205 hard-copy planimetric maps, 205
summation kernel, 253 image-to-map rectification process, intensity interpolation, 205
Hyperspectral remote sensing 204205 resampling
AVIRIS, 32 nonsystematic (random) error, 204 bilinear interpolation, 205206
high-spatial-resolution system, 27, 32 systematic (predictable) error, 204 cubic convolution, 205206
imaging spectrometer, 27 image restoration, 201 nearest neighbor algorithm, 205206
lab/computer imaging technology, 32 internal and external errors, 201, 203 spatial interpolation algorithms, 205
multispectral satellite imaging, 27 radiometric correction, 203204 well-identified GCPs, 204205
absolute, 203 Image understanding system (IUS)
multi-date image normalization color and form, 299300
i
techniques, 204 LCCS, 300
Image characteristics relative, 203204 multiresolution segmentation, 301
basic matrix configuration, 12, 15 systematic striping/banding, 204 regionalization, 301
radiometric resolution, 12, 20 topographic correction methods, 204 SIAM software, 301
definition, 20 Image segmentation algorithms, land spatial features, 300
levels of energy quantization, 20 categorization visual accuracy assessment, 300
temporal image resolution, average accuracy (AA), 338 Inclined geosynchronous satellite orbit
advantages,22 categories, 317 (IGSO) satellites, 491
2016 by Taylor & Francis Group, LLC
670 Index
Indian Remote Sensing Satellite (IRS-1C) K cloud and cloud shadow masking,
image, 19 239240
Indian Remote Sensing (IRS) satellite Karma, 417 CT, 639
program, 63 Kernel trick, 249250 definitions, 232
Inertial measurement unit (IMU), 344, 501 Keyhole Markup Language (KML), 566, 569, historical perspective, 233
Infrared sensors, 6 571, 573, 656 large area mapping, 240241
Inland water ecosystems Kirchhoffs law manual/visual interpretation of
freshwater resources, 7273 blackbodys reflectance, 266 satellite data, 233234
Landsat MSS, 73 emissivity coefficient, 265268 map making, advantages and
radiance to kinetic temperature, limitations, 232
multispectral sensors, 73
conversion, 267 nonparametric methods, 639
phytoplanktons, 73
thermal imagery, 268 parametric classification
water quality parameters, 74
wavelength region, emissivity of surfaces, algorithms,639
In-orbit validation (IOV) satellites, 491
266267 pixel vs. object based
Intercalibration approaches, VI
K-means clustering (KMC), 235 classifications,239
conversion coefficients, 182183
RF, 639
cross-sensor conversion, validation
supervised/unsupervised logic, 639
airborne spectral measurements, 185 L
uncertainty assessment, 241
AVHRR and MODIS, NDVI Laser scanning, see Light detection and
Land cover
values,184 ranging (LiDAR)
change detection, optical and radar data
ETM+ and QuickBird NDVI, LBS, see Location-based services (LBS)
case study, 384, 394
intercomparisons, 184185 L1 civil (L1C) signal, 487488
data collection and preprocessing, 386
Hyperion hyperspectral image data, L2 civil signal (L2C)
direct classification, 391393
184185 image preprocessing, 384 acquisition and tracking, 485
image-based comparisons, 184 post-classification comparison, advantages, 485
intersensor conversions, 185 387388, 390391 carrier phase ambiguity resolution, 485
NDVI in China, 186187 study area, 385386 robust dual-frequency ionosphere error
quadratic correction factors, 184 support vector machine, 386390 correction, 485
VGT and GIMMS NDVI, annual definition, 232 Leaf area index (LAI) estimation, 157
differences, 186 mapping Least squares fitting algorithm, 353355
database of high-resolution spectra, 181 accuracy, 546 Left-hand circularly polarized (LHCP)
NDVI values, 182 classification, 546 antenna, 494
OrbView systems, 182, 184 georeferenced ground pictures, 546 Legal innovations
spectral band responses, 181182 Geo-Wiki interface, 545546 definitions, 610611
Interferometric synthetic aperture radar environmental and disaster provisions,
Land cover classification system (LCCS), 300
(InSAR), 74 611612
Landsat-7 Enhanced Thematic Mapper Plus
Internal calibrators (ICs), 111112 nondiscriminatory access, 612614
(L7 ETM+) imagery, 115117
Internal measurement unit (IMU), 527 LEO satellite, 495
FASC, 112
Internal Navigation Systems (INS), 527 LiDAR, see Light detection and ranging
PASC, 112
International Association of Geodesy (IAG), (LiDAR)
Landsat 8 Operational Land Imager
505506 Light Amplification by Stimulated Emission
(OLI),156
International Criminal Court, 617 of Radiation (LASER) altimetry
Landsat Thematic Mapper, 91, 157, 633
imaging
International Earth Rotation Service Landscape Interpretation Support Tool
CryoSat-2 interferometric radar
(IERS),512 (LIST), 309310 altimeter,35
International GNSS Service (IGS), 505 Land surface temperature (LST), 137138, Jason-2 OSTM, 35, 38
The International Journal of Geographical 260261, 264, 640 and radar altimeters, 3435
Information Systems (IJGIS), 447 Land use; see also Image classification Light detection and ranging (LiDAR), 6, 24,
International Terrestrial Reference System methods, land cover and land use 344, 586, 626, 628
(ITRS), 512 definition, 232 aerial LiDAR data, 39
Internet Intensity Maps, 654 and forest monitoring client-side interface, 559
Intersensor calibrations, 630 accuracy assessment, 70 data, 224, 559
Iterated conditional modes (ICM), 334335 generation of reference data sets, 70 database, 352
Iterative self-organizing data analysis ground forest biomass, 71 data management, 352
(ISODATA), 215 interferometric coherence values, 71 DEM and DSM generation, 3941
IUS, see Image understanding system (IUS) LULC, 6970 data processing, 366
microwave applications, 71 quality assessment and control,
J remote sensing based forest 366367
monitoring, 71 full-waveform airborne, 360362
Japanese Earth Resources Satellite (JERS), 626 image classification methods interactive user interface, 559
Japans GOSAT data policy, 612 advantages, 231232 invention of LASER, 38
Joint Polar Satellite System (JPSS), 49 automated (see Automated image by law enforcement, 41
JONSWAP model, 495 classification algorithms) mapping natural resources, 40
2016 by Taylor & Francis Group, LLC
Index 671
nDSM, 39 normalized accuracies, 658 Multinomial logistic regression followed
OBIA, 41 overall accuracy, 658 by Markov random fields
point cloud (see Point cloud data, LiDAR) photo interpretation project, 583 (MLRsubMRF) method, 251
quality assessment (QA), 348349 positional accuracy, 657 Multipath mitigation technology (MMT), 484
quality control (QC), 349350 positional map, 585588 Multiple endmember spectral mixture
satellite-based LiDAR data acquisition, 39 producers accuracy, 658 analysis (MESMA), 223224
terrestrial (see Terrestrial LiDAR data reference data, 658 Multisensor data, terrestrial applications
processing) remotely sensing, 583 land use and forest monitoring, global
three-dimensional city modeling (see RS data, 657 level, 6972
Three-dimensional city modeling, thematic map accuracy, 589 accuracy assessment, 70
LiDAR data) thematic maps, 657 generation of reference data sets, 70
TIN, 559560 users accuracy, 658 ground forest biomass, 71
vehicle-borne (see Vehicle-borne LiDAR Margfit analysis, 596 interferometric coherence values, 71
data) Marine Optical BuoY (MOBY), 114 LULC, 6970
Linear Vegetation Index (LVI), 157 Markov random fields (MRFs)-based microwave applications, 71
Localizer performance with vertical guidance algorithms remote sensing based forest
(LPV) procedures, 528 graph cuts, 335 monitoring, 71
Local Knowledge and Climate Change ICM, 334335 precision farming, local level
Adaptation Project (LKKCAP), 434 land categorization applications, 334 fluorescence measurements and
Location-based services (LBS) notion of Markov chain, 334 thermography, 69
autonomous cars, 431432 Potts model, 334 high spatial resolution, 68
Bing Maps and MapQuest, 432 simulated annealing optimization, 335 imaging spectroscopy, 6869
development of mobile mapping spatial smoothing, 334335 management of agricultural
programs, 431 Mash-up, see Map mash-ups or mash-ups production systems, 68
Google Glass, 432433 MAUP concept, 466 multispectral and hyperspectral
Google Maps, 432 Maximum likelihood classification sensors, 68
Googles Street View mapping (MLC),235 polarimetric decomposition
technology,431 Medium-resolution imaging spectrometer techniques, 69
mapping, 432 (MERIS), 178 satellite instruments, 68
navigation, 432 Meteosat Second GenerationSpinning from space to species, 69
recreation, 432 Enhanced Visible and Infrared Multispectral scanner system (MSS), 63
retail/business services, 432 Imager (MSG SEVIRI), 268
self-driving cars, 431 Microsatellites
n
societal services, 432 UAS and UAV sensor platforms, 4950
Logarithmic discriminant analysis UAV eBee, 4950 NASAs Earth Observation System Data and
(LogDA),249 Microsoft Virtual Earth, 427 Information System (EOSDIS),
A Long Way Home, 423424 Microwave remote sensing 1011
L5 signal active and passive forms, 4142, 45 National Agriculture Imagery Program
carrier phase ambiguity resolution, 486 sensorbased satellites, 4244 (NAIP) imagery, 24
carrier phase tracking, 486 soil moisture content, 42 National Geospatial-Intelligence Agency
codes, 486 Microwave SAR, 6 (NGA), 568, 616
errors reduction, 486 Military (M) codes, 487 National High Altitude Program (NHAP), 26
GPS modernization, 486 Minimum distance (MD), 235 National Map, 430
navigation messaging, 486 Minimum spanning forest (MSF), 332333 National Map Accuracy Standards
ranging accuracy, 486 Mobile mapping programs, development (NMAS),585
weak-signal code acquisition and of,431 National mapping agencies, development
tracking, 486 Mobile phone mobility analysis, 436 of,425
LST, see Land surface temperature (LST) Mobile terrestrial LiDAR systems, 344 National Oceanic and Atmospheric
Mobile TLS, 345347 Administration (NOAA), 6364,
MODAPS and OMIDAPS, 417 93, 113114, 143, 268, 568, 626
M
Moderate-resolution atmospheric National Polar-orbiting Operational
Map making, remote sensing transmission (MODTRAN), 180 Environmental Satellite System
advantages, 232 Moderate resolution imaging (NPOESS) Preparatory Project, 626
disadvantages, 232 spectroradiometer (MODIS), 95, National spatial data infrastructures
for land-cover studies, 232 143, 178, 574, 630 (NSDIs),427
Map mash-ups or mash-ups, 431 Modified Soil-Adjusted Vegetation Index National Standard for Spatial Data Accuracy
Maps accuracy assessment (MSAVI), 176 (NSSDA), 586, 588
boundary error problem, 584 Modulation Transfer Function (MTF) National Technical Means (NTM), 607
classification, 658 effect,239 NDVI, see Normalized difference vegetation
digital image processing, 584 MRFs-based algorithms, see Markov random index (NDVI)
earths resources, 583 fields (MRFs)-based algorithms Neogeography movement, 429, 440
error sources, 657658 Multifunctional Satellite Augmentation Networked online mapping of atmospheric
fuzzy accuracy assessment, 658 System (MSAS), 511 data (NOMAD) transfers, 94
2016 by Taylor & Francis Group, LLC
672 Index
Next-Generation Radar (NEXRAD) satellites, spatial properties and integrated assessment approach, 458
34, 36 relationships,296 spatial analysis techniques, 458
NOAA, see National Oceanic and spatial units (digital geographic social vulnerability to malaria (case study)
Atmospheric Administration features), 297 indicators and datasets, 469470
(NOAA) WFD, 297 latent phenomenon addressed, 469
NoiseTube project, 435436 GIS, 277 results, 470472
Nondiscriminatory access, Principle XII high resolution, notion of, 278 socioeconomic vulnerability to hazards
data provisions, 612 image segmentation, 278279 (case study)
data sharing principles, 613 LiDAR data, 642643 indicators and datasets, 469
full and open access, 613 needs and driving forces, 279 latent phenomenon addressed,
implementation, 612 nomenclature and sound 468470
national security restrictions, 613 methodologies,288 results, 469
policy, 612613 ongoing developments, influences Object fate
primary data and processed data, 612 changing workplace, 285286 class-constrained object-by-object
and reasonable cost, 612 GIScience and remote sensing, comparison, 311
shutter control provisions, 613614 284285 co-registration errors, 310311
transactional restrictions, 613 usage, 285286 information update
Non governmental organization (NGO), per-pixel classification, 642 costs and quality, 308309
571,573 QuickBird imagery, 642643 differentiated boundary treatment,
Normalized difference vegetation index remote sensing, hardware/software, 277 308309
(NDVI), 66, 138, 144, 148150, 176, segmentation techniques, 279, 282283 GIS-ready information, 308
199, 211213, 239, 318 validity (see Object validity) image-based biotope complex
Normalized digital surface model (nDSM), Object-based methods delineation, 309
3940 advantages, 225 planning association, 308
classification and characteristics, 221 VHRI, 308
decision rules, 224 VHSR data, 307
o Definiens/eCognition, 225 OFA (see Object fate analysis (OFA))
environmental indices, calculations, rapid information extraction
Object-based analysis (OBA), 461
219220 context-based automated
Object-based change detection (OBCD),
GEOBIA, 224 approach,312
306307
hierarchical image segmentation, 225226 Google/DigitalGlobe, 312
Object based image analysis (OBIA), 457, 593
high-spatial-resolution remote-sensing high-spatial resolution satellite
addresses object properties, 642
data, 223224 imagery, 312
civilian geo-intelligence, 288289
LiDAR data, 224 QuickBird data, 312
classification, 283, 642
maximum likelihood, 225 VHR images, 311
complex geo-intelligence tasks, 282
overview, 220 VHSR imagery, 312
conceptual foundations, 278
QuickBird, 225 from static to dynamic change concepts
consolidation (since around 2010), 282
spectral noise, 223 categories of geometric changes, 307
enriched information stratified random sampling approach, 225 multitemporal imagery, 307
decision makers and policy Object-based regionalization OBCD, 306307
implementation, 298 abstract/complex phenomena segmentation goodness, 307
generalization, 299 a priori geographies, 459 spatial disagreement/error, 307
GEOBIA, 298 latent variables/factors, 459 Object fate analysis (OFA), 465466
information conditioning, 298 river basin districts (RBDs), 459 LIST, 309310
issue of scale, 298299 climate change susceptibility (case study) OBIA classification, 310
policy scope, 298 indicators and datasets, 474 object relationships, 310
spatial representation, 299 latent phenomenon addressed, segmentation algorithms/visual analysis,
epistemological and ontological 473474 309310
challenges, 289 results, 474 Object validity
fate (see Object fate) domain-specific regionalization (see class modeling
GEOBIA (see GEOBIA developments) Regionalization, domain-specific) manual interpretation/image
geospatial data integration, 642644 geon approach (see Geons) analysis,304
geospatial information landscape sensitivity (case study) multiresolution segmentation
biotope complexes, 297 indicators and datasets, 471473 algorithm, 302
class modeling, 296 latent phenomenon addressed, 471 region-based segmentation algorithms,
earth observation (EO), 297 results, 473 303304
multiscale option, 297 OBIA, 457 supervised regionalization technique,
object-oriented data model, 297 policy-oriented partitioning of space, 302303
regionalization techniques, 296 651652 concept of binary accuracy
salt-and-pepper effect, 296 policy targets assessment,299
SIIs, 297 concept of vulnerability, 458 definition, 304
spatially constraint class conceptual operationalization, 458 human vs. machine vision, 301302
modeling,297 global warming, 458 IUS, 299301
2016 by Taylor & Francis Group, LLC
Index 673
multiple-stage validation Participatory sensing NSSDA, 586, 588
four-stage validation procedure, 306 Common Scents project, 435 reference data, 586587
land use/cover units, semiautomated ground truth verification data, 436 RMSE, 587
classification, 305306 mash-ups/displayed on virtual globes, 435 road intersections, 585
object delineation and classification mobile phone mobility analysis, 436 standard error, 588
appropriateness/validity, 304 NoiseTube project, 435436 vegetation/land cover, 586, 588
identification, 304 Personal Environmental Impact Report, Positioning and orientation systems
image analysis, 304305 435436 (POS),344
label verification, 305 personal mobile health monitoring, 436 Post-launch sensor radiometric calibration
machine-based knowledge PPGIS/PGIS projects, 435 methodologies, 630, 632
representation, 304 web-based mapping system, 436 GEOSS and GSICS, 110, 112113
perceptual classification, 304 Passive reflectometry and interferometry L1L5 MSS sensors, 112
procedural knowledge, 304 system (PARIS) project, 494 L8 OLI radiometric calibration
recognition, 304 PCA, see Principal component analysis (PCA) system,112
semantic classification and naming, 304 Permanent Court of Arbitration, 617 onboard radiometric calibration, 119
structural knowledge, 304 Per-pixel methods EOS MODIS whiskbroom sensors, 111
Ocean Surface Topography Mission (OSTM), classification procedures and FASC, 112
35, 38 characteristics, 221 internal calibrators, 111112
OFA, see Object fate analysis (OFA) class training process, 222 L7 ETM+, 112
Onboard calibration systems environmental indices, calculations, MSS sensors, 111
AVHRR visible bands onboard NOAA, 145 219220 partial-aperture calibrators, 111
device GIS models, 221 Thematic Mapper sensors, 111
bidirectionality of SD reflectance, 146 hard classifications, 221 prelaunch calibration methodologies, 111
ETM+, 145146 levels of suitability/scale of pushbroom scanning technology, 112
Landsat 5/TM, 146 appropriateness, 221 QA4EO, 111
MODIS, 146147 maximum likelihood algorithm, 221222 relative and absolute calibration, 110
spaceborne sensors, 145 microurban remote sensing, 222 SRCA, 111
vicarious calibration, 145 overview, 220 vicarious calibration methodologies, 110,
Open Cloud Consortium (OCC), 558 Personal Environmental Impact Report, 112113
Open Geospatial Consortium (OGC), 411, 558 435436 Postprivacy movement, 439
Open Modeling Interface (OpenMI), 411412 Personal mobile health monitoring, 436 Potts model, 338
OpenStreetMap (OSM), 427, 429 Phased Array type L-band Synthetic Aperture Power ratio
crowdsourced mapping service, 538 Radar (PALSAR), 626 code phases/delays and correlation power,
and Google Maps, 538539 PiersonMoskowitz model, 495 502503
Map Kibera project, 434 Plancks law vs. cost function, 503
node density map, 538, 540 blackbody radiation curves, 261262, 265 definition, 502503
Optical remote sensing formula, 262 satellite-borne receiver, 503
across-track/whiskbroom scanner, 33 Point cloud data, LiDAR smooth sea surface, 503
AVIRIS image, 27, 32, 34 city area and forest area, 345346 SSH estimates, 503
electro-optical sensors, 27 hierarchical representation, 351 Preexisting legal obligations, remote sensing
hyperspectral imaging systems, 26 organization and generalization principles
image acquisition process, 27, 34 hierarchical representation, 351 international cooperation, Principles II, V,
multispectral imaging system, 26 spatial indexing, 350351 VI, VII, VIII and XIII, 609610
multispectral scanner, 33 spatial features international law, Principle III, IV and
optical sensorbased remote sensing boundary features, 353 XV, 608609
satellites, 2732 fitting features, 352353 space law, Principles IX and XIV, 609
panchromatic imaging system, 26 Hough transform, 356 Principal component analysis (PCA)
pushbroom system, 27 intersection features, 353 components, 207
super-spectral imaging system, 26 least squares fitting algorithm, Landsat Thematic Mapper bands 2 and
Optical satellite sensors, 630, 632 353355 3, 207
calibration, 630 RANSAC, 356 Privacy and confidentiality, geo-information
international community, 630 segmentation algorithm, 355 concept of data protection, 439
linear regression, 630, 632 spatial indexing, 350351 crowdsourcing, 439
Optimized SAVI (OSAVI), 176177 storage, 350 data mining, 438439
OSM, see OpenStreetMap (OSM) Policy-oriented partitioning of space, 651652 Geo-ICTs, 439
Outer Space Treaty, 606607 Positional map accuracy geoslavery, 439440
CMAS, 585 geosurveillance, 438, 440
error theory principles, 585 Googles Street View technology, 439
P
error thresholds, 586 LBS technology, 439
Partial-aperture solar calibrator (PASC), 112 FEMA, 585 postprivacy movement, 439
Participatory GISs (PGIS), 426 LiDAR Data, 586 social networking, 439
Participatory/community mapping mapping community, 588 sousveillance (inverse surveillance), 439
projects,426 NMAS, 585 Provenance
2016 by Taylor & Francis Group, LLC
674 Index
aware systems atmospheric influence, 135 Relative delay estimation methods
Data Quality Provenance System, 417 sensor calibration and degradation, 135 C/A code method, 502
Earth system science workbench and topographic influence GNSS receiver, 502
ES3, 417 on optical data, 135 interferometry technique, 502
GeoPWProv, 418 on SAR data, 135136 P(Y) code/military M-code, 502
inversion, 415416 Radiometric normalization peak power location, 502
Karma, 417 absolute radiometric correction, 133 smooth and rough sea surfaces, 502
MODAPS and OMIDAPS, 417 AERONET network, 137 SSH estimation accuracy, 501
service chaining, 416 data atmospheric measurements/ zenith-looking antenna, 502
UV-CDAT, 417418 reanalysis data, 137 Remote-sensing (RS) data
VDC service, 416417 Landsat-5 TM data, 138139 active/passive, 57, 9
VisTrails, 417 NDVI, 138 aerial photographs, 6
definition, 403404 SAR calibration, 139 applications
digital (see Digital provenance and sensor-specific calibration commodity hardware, 558
geospatial workflows) parameters,136 HPC, 557
in remote sensing workflows surface reflectance/LST, 137138 inherent concurrency, 557
provenance and trust, 414415 thermal infrared, 137 I/O-bound applications, 558
questions, 413414 data types, 136 LiDAR, 557560
Pseudoinvariant calibration sites (PICS), see definition, 133 MapReduce implementations, 558
Pseudoinvariant features (PIFs) relative radiometric correction, 133 Moderate-Resolution Imaging, 558
Pseudoinvariant features (PIFs), 112, 634 optical and thermal data, 139 OCC, 558
Public participation GISs (PPGIS), 426 relative correction of SAR data, OGC, 558
139140 spatialtemporal data, 558
Radiometric resolution, 20 WMS, 558
Q
RANSAC, 356 and cloud computing, 656
Quadratic discriminant analysis (QDA), 249 Recursive hierarchical segmentation data acquisition, elements, 6
Quality assessment/control (QA/QC), LiDAR (RHSE),255 data analysis, 6
data Reference data collection data collection and processing, 67
DEM and DSM generation, 366367 aerial photo interpretation, 592 ESRI, 5
error budget ASPRS, 587 factors affecting, 6
direct georeferencing errors, 348 consistency, 592 FLIR/FLIR, 6
laser scanner errors, 348 independence, 586, 592 mapping sciences, 56
external, 349350 NSSDA, 586 photogrammetry, 6
internal, 349 sampling, 587, 592594 platforms (see RS platforms)
management activities, 348349 spatial autocorrelation, 587, 592 principles
system components, 347348 statistical analysis, 586 innovations (see legal innovations)
Quality assurance and stability reference thematic accuracy, 586 opinio juris, 608
(QUASAR) monitoring, 115 timing, 592 preexisting legal obligations, 608610
Quality Assurance Framework for Earth vegetation/land cover, 587 state practice, 608
Observation (QA4EO), 89, 111 Reflected signal power UNGA, 608
QuickBird imagery, 642643 code phase and Doppler frequency, reflected/transmitted EMR, 5
497498 and sensor characteristics (see Sensor
correlation power vs. code phase, 497 characteristics)
R
decibel delay waveforms, 498 thermal sensors, 5
Radarsat1 and 2, 66 delay-Doppler waveform, 497 UAVs, 6
Radio detection and ranging (RADAR), Elfouhaily wave elevation model, 498 Right-hand circularly polarized (RHCP)
6,628 IF signals, 497 antenna, 494
altimetry imaging, 3438 nadir-looking antenna, 497 Root mean square error (RMSE), 587
Doppler radar, 32 normalized correlation powers, 498 R project, 451452
ESA SAR imagery, 34, 36 scattering area, 498 RS platforms
far lightning strikes, 32 Reflective optics system imaging aerial imaging, 26
interferometric SAR, 32 spectrometer (ROSIS), 336 electromagnetic spectrum, principles, 89
NEXRAD satellites, 34, 36 Regionalization, domain-specific global positioning system, 4548
RADARSAT, 34 integrated geons (see Geons) hyperspectral imaging sensors, 26
remote sensing, 3234 MAUP and scale dependency, 466467 hyperspectral (see Hyperspectral remote
sensorbased satellites, 34, 37 monitoring geons, 465466 sensing)
TerraSAR-X sensor, 34, 36 object validity, 467 LASER imaging (see Light Amplification
weather and climate analysis, 34 principles, 461 by Stimulated Emission of
Radio-frequency identification (RFID), 527 spatially constraint classification, 462 Radiation (LASER) altimetry
Radiometric calibration systemic areal units imaging)
with onboard calibration systems, 633 Elementary_geoParticle, 467468 LiDAR, 26, 3841
without onboard systems, 633 geon concept, 467 microsatellites, 4951
Radiometric distortion, sources scaling ladder concept, 467 microwave (see Microwave remote sensing)
2016 by Taylor & Francis Group, LLC
Index 675
new satellite sensor platforms, 4849 sensor degradation, effects, 144 SBAS, see Satellite-based augmentation
optical remote sensing, 2527 SPOT VEGETATION, 143 system (SBAS)
RADAR and SODAR remote sensing, time-invariant sites, 633634 Sea surface altimetry
3234 vicarious calibration, 144 calibration, 501
sensor type, 25 Satellite de Aplicaciones Cientificas-C (SAC- power ratio, 502503
SODAR remote sensing, 4445 C) satellites, 494 relative delay estimation methods,
SONAR remote sensing, 26, 4445 Satellite image radiometry 501502
Russian Global Navigation Satellite System, atmospheric correction, 9394 SSH calculation, 499501
510511 CEOS, 630 Sea surface height (SSH)
CEOS QA4EO effort, 9798 computational complexity, 500
S CLARREO, 98 flowchart, two-loop iterative method, 500
data standardization, 8889 power ratio, 502503
Sampling, reference data collection EO satellites, 8889, 628, 630 specular point, 500
biased assessment, 593 GEO, 630 total path length (TPL), 500
cluster sampling, 593 geometric effects WGS84, 499501
coarse-resolution, 593 adjacency effects, 9596 Sea surface scattering
error matrix, 593 atmospheric refraction, 95 delay frequency, 497
Landsat pixel, 593 illumination and viewing Doppler frequency, 497
midresolution sensors, 593 geometries,95 iso-delay map, 496497
minimum mapping unit, 592 pixel, 95 iso-Doppler map, 497
multinomial sampling, 593 position of the Sun, 9697 SSP, 496
OBIA, 593 reflectance anisotropy, 95 Sea surface temperature (SST), 260261,
positional accuracy, 593 topographic effects, 96 271272, 640
random sampling strategy, 594 GEOSS, 630 Sea surface wind speed estimation
remote sensing, 592 information extraction, data flow accuracy, 495
spatial autocorrelation, 592, 594 options,90 average, 499
systematic sampling, 594 cost function, 498499
Landsat 7 ETM+ and AWiFS, 630631
SAR, see Synthetic aperture radar (SAR)
MODIS, 630 cutoff correlation power, 499
SAR imaging from 2D to 3D
normalization approaches, 97 interpolating, 499
D-InSAR data acquisition, 6667
photon pathways, 8990 mathematical approach, 498
satellite sensors, 66
postlaunch methodologies, 88 measured delay waveform, 498499
Satellite-based augmentation system (SBAS)
processing considerations, 97 model-matching approach, 498
EGNOS, 511
sensor radiometric calibration, 9093 reflected signal power, 497498
GPS, positioning accuracy, 511512
in solar-reflective optical domain, 88 sea surface scattering, 496497
MSAS, 511
spectral characterization, 97 sea wave spectrum, 495496
WAAS, 511
spectral reflectance, 89 slope-based method, 499
Satellite-based Earth observation sensor
success of QA4EO, 98 slope characteristics, 495
systems, 110
surface reflectance retrieval, 9394 standard deviation (STD), 499
Satellite data degradations
TRUTHS mission concept, 98 theoretical waveform, 498499
AVHRR NDVI datasets, 144
AVHRR sensor, 633 Satellite Pour lObservation de la Terre Sea-Viewing Wide Field-of-View Sensor
BRDF, 144 (SPOT) VEGETATION, 143 (SeaWIFS), 178
Earth observation satellites, 143 Satellites and sensors, remote-sensing Sea wave spectrum
ETM+, 633 data characteristics, 626 Elfouhaily model, 495496
impacts on high-level products data collection, 626 SWH, 495
AVHRR orbit drift, 149150 military usage, 626 SWP, 495
MODIS/Terra degradation, 148149 spatial data, 626 Segmentation algorithm
impacts on satellite products, 144 spatial resolution, 626627 clustering algorithms, 355
Landsat TM, 633 UASs and UAVs, 626 coplanarity, 355
MODIS Aqua data, 633 Satellite sensor platforms LiDAR point clouds, 355
MODIS Terra data, 633 ALOS 2, 48 profile segmentation algorithm, 355
orbit drift of Sun-synchronous DigitalGlobe, 49 region growing algorithms, 355
satellites,633 GPM Core Observatory satellite, 48 split-and-merge algorithms, 355
ECT, 147148 Hodoyoshi 3 and 4, 48 Selective availability (SA), 483
NOAA heritage satellites, 147 IRNSS 1B, 48 Self-driving cars, 431
SZA, 147148 JPSS, 49 Sensor characteristics
radiometric calibration, 143144, 633 oblique aerial photograph, 49 aerial platforms, 9
radiometric degradation older programs, 49 categorization of satellites, 2223
with onboard calibration device, Sentinel-1A, 48 Earth-observing programs, countries,
145147 SkySat 2, 48 2223
without onboard calibration TSAT, 48 image characteristics
systems,145 UKube, 48 basic matrix configuration, 12, 15
scatter plots of digital numbers, 633634 Velox 1, 48 radiometric resolution, 12, 20
2016 by Taylor & Francis Group, LLC
676 Index
satellite sensors, data characteristics, Soil and Water Assessment Tool (SWAT), 24 heterogeneity, color and shape,
1518 Soil calibration factor, 176 324325
spatial resolution, 12, 15, 1819 Soil moisture, GNSS-R HSWO approach, 322
spectral resolution, 1920 advantages, 504 hybrid of spectral clustering, 323
temporal resolution, 12, 2022 dielectric constant, 504 logical predicate segmentation,
optical sensors, 11 disadvantages, 504 321322
passive and active sensors estimation, 503504 marker-based HSeg algorithm, 325
classification, 12 interferometric method, 503504 mean squared error contribution,
imaging and nonimaging sensors, 10 power notch position, 504 323324
NASAs EOSDIS, 1011 satellite-borne receiver, 504 SAM criterion, 323324
nonscanning and scanning types, 10 three-layer reflectivity model, 504 seeded region growing, 323
sensor characteristics, 1112 total power received, 504 SEGEN, 324
types, 1315 transition moisture parameter, 504 segmentation hierarchy, 322
principles of electromagnetic spectrum, Solar-radiation-based calibration (SRBC), 91 selection of single segmentation, 325
89 SONAR, see Sound navigation ranging split-and-merge region growing, 323
remote sensors or sensors, 6 (SONAR) remote sensing land categorization
satellite platforms, 910 Sonic detection and ranging (SODAR) remote applications, 321322
sensing technologies, types sensing, 6, 3234, 4446 SAR imagery data, 322
advantage of microwave remote Sound navigation ranging (SONAR), 6, 44, texture-based algorithms
sensing, 24 4647, 628 agglomerative hierarchical clustering
biophysical variables, analysis, 2425 Sousveillance (inverse surveillance), 439 algorithm, 326
Earth-observing satellites, 24 Spaceborne LiDAR systems, 344 generic approaches, 326
geosynchronous orbits, 24 Space law image segmentation, 326
irrigation and fertilizer civilian purposes, 659 man-made structures, 326
applications,24 remote sensing image analysis,
confluence of multiple legal regimes, 606
Landsat data, 24 325326
disarmament and verification law,
near-polar-orbiting satellites, 24 unsupervised image segmentation, 325
607608
satellite Aryabhata, 24 watershed algorithms
free access to outer space, 659
Sun synchronous orbits, 24 concept of dynamics, 327328
GEO, 659
SWAT, 24 gradient functions, 327
GEOSS, 659
types of satellite orbits, 24 immersion process, 327
international space treaty regime, 658
Sensor radiometric calibration manual selection/selection of
military and strategic requirement of
converting at-sensor radiance to at-sensor pixels,328
nations, 658
reflectance, 92 multiscale watershed segmentation,
national geospatial law principles, 659
converting digital counts to at-sensor 328329
nongovernmental satellites, 659
radiance, 9192 preprocessing/postprocessing
observation and collection of data, 659
dynamic range, 91 methods, 327
Outer Space Treaty, 605 Spatial resolution, 12, 15, 1819
Landsat-5 Thematic Mapper, 91
principles, 658 Spectral Angle Mapper (SAM) criterion,
methodologies and instrumentation,
9091 and regulatory regime, 605 323324
optical imaging sensors, 91 space treaty regime, 606607 Spectral band difference effects (SBDEs), 113
SRBC, 91 Space probe sensors, 6 Spectrally based segmentation approaches
vicarious calibration, 9293 Space treaty regime, 606607 clustering-based algorithms
vicarious methods, 91 Spatial autocorrelation, 594 Gaussian mixture model, 318
Service chaining, 416 Spatial information infrastructures (SIIs), 297 pattern recognition, 318
Service-oriented computing (SOC), 410 Spatially based segmentation approaches pixel-based image analysis techniques,
Shuttle radar topographic mapping mission graph-based algorithms 318319
(SRTM), 66, 566, 656 normalized cuts, 333334 postprocessing techniques, 319
SIAM (Satellite Image Automatic Mapper) optimal spanning forests, 332333 spatial smoothing, 319
software, 301 morphological profiles split-and-merge procedure, 319
Significant wave height (SWH), 495 advantage of reconstruction filters, 329 support vector machines, 319320
Significant wave period (SWP), 495 DMP, 329, 332 thresholding-based algorithms, 318
Simple Object Access Protocol (SOAP), 410 granulometry, 329 Spectral resolution, 1920
Social media non-Euclidean geodesic metric, 329 Spectralspatial classification
collaborative projects, 538 segmentation hierarchy, 330 feature extraction stage, 250
content communities, 538 spectral homogeneity, 331 kernel-based classifiers, 250251
definition, 538 structuring element (SE), 328331 morphological profiles, 250
social networking sites, 538 MRFs-based algorithms (see Markov MRF-based spatial postprocessing, 251
Social networks, 538 random fields (MRFs)-based Spectrometry, 6
SODAR remote sensing, see Sonic detection algorithms) Spectro-radiometric calibration assembly
and ranging (SODAR) remote region-growing algorithms (SRCA), 111
sensing dissimilarity criterion, 323324 Specular scattering point (SSP), 496
Soil-Adjusted Vegetation Index (SAVI), 176 eCognition Developer, 324325 SPOT-1, 63
2016 by Taylor & Francis Group, LLC
Index 677
Sputnik 1, 143 Sentinel-2 satellites, 77 diurnal and time series
SSH, see Sea surface height (SSH) soil moisture measurement, 76 acquisition data and time, 271
SST, see Sea surface temperature (SST) U.S. National Space Policy of 2010, 76 ATI, 270
SSURGO soil data, 24 evolution BRDF, 270
State-sponsored mapping initiatives, advent digital data processing, 6366 concept of T, 269270
of, 425 environmental parameters, 66 daytime data, 268
StefanBoltzmann laws, 262263 SAR imaging from 2D to 3D and higher-resolution data, 270271
Stereogram, 26 beyond, 6667 LST imagery, 270
Subpixel methods global networks, 628 MODIS TIR data, 268269
challenges, 223 monitor coastal vegetation, 628629 nighttime data, 268
classification procedures and multisensor approaches, 628 sand dune, temperature variation,
characteristics, 221 physical measurement, 628 270271
environmental indices, calculations, primary spectral regions, 62 sea surface temperature, 271272
219220 subdivisions of active microwave temperature curve, 269
linear spectral mixture analysis, 223 sensory,63 thermal inertia, 269270
overview, 220 thematic applications water and dry soil/rock, temperature
regression analysis and regression inland water, 7274 variation, 269270
trees,223 urban applications, 7273 diurnal temperature changes (T),
scale of urban analysis, 222223 using multisensor data 640641
soft classifications, 222 land use and forest monitoring, 6972 domain, bandwidths and spatial
Sun synchronous/geostationary, 626 precision farming: local level, 6869 resolution, 641
Suomi National Polar-orbiting Partnership from space to species, 69 electromagnetic radiation, 260
VIIRS, 156 Terrestrial laser scanning (TLS), 345347 ground data collection/validation,
Support vector machine (SVM), 249 Terrestrial LiDAR data processing 272,274
binary classification problem, 319320 data processing strategy, 367 high-resolution data Google Earth,
classification, 318, 337 point cloud organization and 272273
Gaussian radial basis function kernel, 320 segmentation, 368
Landsat 7 ETM+, 641
hyperspectral images, 319 registration and georeferencing, 367368
Landsat-5 Thematic Mapper, 260261, 266
pixel-based supervised classification, 320 visualization system, 368
laws
quadratic optimization problem, 320 measurement mechanism, 344
Kirchhoff and the role of emissivity,
SVM, see Support vector machine (SVM) properties, 367
265268
Synthetic aperture radar (SAR), 63, 133, Thematic map accuracy
Planck, 261262
322,626 computation, 594595
StefanBoltzmann laws, 262263
Systme Probatoire dObservation de la Terre decision making, 591
Wien, 263265
satellite (SPOT-VEG), 178 digital imaging, 591
LST, 260261, 264, 640
error matrix, 589590, 599600
satellite sensors, 261, 263
fuzzy accuracy assessment, 598599
t sensitive sensors, 640
implementation priority, 589, 592
spaceborne thermal data, 641
TechDemoSat-1, 494495 Kappa statistics (KHAT), 597
Television Infrared Observation Satellite Margfit analysis, 596 SST, 260261, 271272, 640
(TIROS-1), 63 minimum mapping unit (mmu), 591 temperature, 259
Temporal resolution, 2022 mutually exclusive, 589590 thermal camera imagery, 261262
Terrain correction, 135136 non-site-specific assessment, 589 thermal infrared domain, 261
Terrestrial applications, remote sensing normalized matrix, 596597 wavelength domain, 259260, 265
continuous and comparable data, 628 OBIA, 589, 600 Three-band model (TbM), 212213
data availability photo interpretation, 589, 591 Three-dimensional city modeling, LiDAR
and comparability, 76 reference data collection, 592594 data, 345
and continuity, 7778 remote sensing project, 591 building extraction
data policy, 628 sample variance, 597 attribution, 359
commercial, 75 site-specific, 589 detection, 358359
databuy, 76 spectral resolution sensors, 589 reconstruction, 359
web-enabled, 75 test statistic, 597 object reconstruction strategies, 357358
digital elevation models, 7475 users accuracies, 596 properties, city area, 356357
electromagnetic spectrum, 628 vegetation/land cover, 589 road extraction, 359362
enhanced technology Thematic Mapper (TM) sensors, 111 3D laser scanners, 344
Copernicus program, 77 Thermal infrared (TIR) remote sensing Time-invariant sites, 633
FLEX, 7677 airborne ThermaCAM P40, 640, 642 TIR remote sensing, see Thermal infrared
FLORIS, 77 data analyses on land (TIR) remote sensing
hyperspectral GEO-CAPE, 77 SST analyses, 271 Traceable Radiometry Underpinning
Landsat 8, 76 thermographic cameras, TIR Terrestrial and Helio Studies
Sentinel-3, 77 spectroscopy, 272 (TRUTHS) mission, 98, 119
Sentinel-1A in 2014, 77 data usage, 640 Transductive support vector machines
Sentinel-1B, 77 definition, 640 (TSVMs), 252
2016 by Taylor & Francis Group, LLC
678 Index
Transformed Soil-Adjusted Vegetation Index principles, 187, 636637 or cross-calibration, 110
(TSAVI), 176 sources of variation QUASAR monitoring, 115
Triangulated irregular network (TIN), atmospheric effects, 179180 radiometric calibration
559560 directional effects, 180 coefficients,117
Typhoon Haiyan-related mapping, 441 pixel-size effects, 180181 remote sensing calibration, 115
soil background, 179 site selection, criteria for, 113
spectral band effects, 181 surface measurement methodologies,
U
standardization, 636 113115
United Nations Committee on the alternatives, 179 U.K.s NPL, 118
Peaceful Uses of Outer Space benefits, 178 WSNM, 113
(UNCOPUOS), 608 BRDF, 178 Virtual data catalog (VDC) service, 416417
United Nations General Assembly data acquisition problem, 178 Visible Infrared Imaging Radiometer Suite
(UNGA),605 formula, 178179 (VIIRS), 156
United States Television Infrared limits, 188 VisTrails, 417
Observation Satellite Program proposals, 187188 Volunteered geographic information (VGI),
(TIROS)-1, 143 virtual constellation, 178 433434
University Consortium for Geographic three-band model, 212213 definition, 538
Information Science (UCGIS), 447 uses DYFI service (see Did You Feel It (DYFI)
Unmanned aerial vehicles (UAVs), 6, 651 EVI, 176 service)
Unmanned aircraft systems (UAS) laboratory-measured reflectance framework, 538
sensors,626 spectrum, 176 GEM (see Global earthquake model
UN-SPIDER, 612 MSAVI, 176 (GEM))
Urban image classification NDVI, 176 land cover mapping, 545546
geospatial methods, 638 OSAVI, 176177 research and operability, 546547
GIS data sources and methods, 219 physiological and anatomical seismic (and multi) risk, 543544, 546
object-based methods, 638 characteristics, 176 Voxel-based marked neighborhood searching
per-pixel spectrally based methods, SAVI, 176 (VMNS) method, 369
637638 soil calibration factor, 176
remote-sensing methods TSAVI, 176
W
geospatial methods, 219221, 226227 wide dynamic range vegetation index,
object-based methods, 219221, 212213 Water Framework Directive (WFD), 297
223225 VEGETATION sensor, 178 Web 1.0, 536537
per-pixel methods, 219222 Vehicle-borne LiDAR data, 345 Web 2.0, 424, 537
subpixel methods, 219223, 637638 auxiliary data structures, 369 Web-based mapping system, 436
USGS National Map, 429430 nonroad surface points, 369 Web-Enabled Landsat Data (WELD), 240
U.S. national policy, 612 on road, 368369 Web Map Service (WMS), 558
UV-CDAT, 417418 3D point cloud, 369 White Sands, New Mexico (WSNM), 113
VMNS method, 369 Wide Area Augmentation System
Very-high-resolution radiometer (VHRR), (WAAS),511
V
6364 Wide dynamic range vegetation index
Variable length records (VLRs), 350 Very high-spatial-resolution imagery (VHRI), (WDRVI), 212213
Vegetation biophysical variables (VBVs), 209 308, 656657 Wien, law of
Vegetation indices (VI), 199, 634, 636637 VGI, see Volunteered geographic blackbody radiation curves, 263, 265
applications information(VGI) Landsat-5 Thematic Mapper data,
direct estimation of fAPAR, 178 VI, see Vegetation indices (VI) 263, 266
LAI, 177 Vicarious calibration methodologies shortwave infrared (SWIR) band,
measures of canopy density, 177 error analyses, 112 263, 265
sensitivity of OSAVI, 177178 PIFs, 112 Wiens displacement law, 263
BRDF modeling, 188189 SBDEs, 113 Wikimapia, 429
cross-sensor effects, 189 via terrestrial reference standard sites Wikipedia, geography of, 440
empirical scatterplots, 636 CEOS-endorsed reference standard Working Group for Calibration and
EVI, 634 sites, 115, 117118 Validation (WGCV), 89
EVI 1 and 2, 212213, 634 CEOS/WGCV/IVOS, 115117 World Wide Web Consortiums PROV Data
intercalibration approaches, 636 Earth-monitoring systems, 115 Model, 648
intersystem differences, 636 GIANTS concept, 115, 118
land cover dependence, 190 imaging the Moon, 115
Z
normalized difference vegetation index, Landsat-7 ETM+ imagery, 115117
211213 ocean color satellite sensors, MOBY, 114 Zenith-looking antenna, 497, 502
2016 by Taylor & Francis Group, LLC
You might also like
- (PDF Download) Remote Sensing Digital Image Analysis Sixth Edition John Alan Richards Fulll Chapter100% (6)(PDF Download) Remote Sensing Digital Image Analysis Sixth Edition John Alan Richards Fulll Chapter64 pages
- Total Ionizing Dose Effects On Current Sense Amplifiers100% (1)Total Ionizing Dose Effects On Current Sense Amplifiers4 pages
- Networked Control Systems: Using Matlab/SimulinkNo ratings yetNetworked Control Systems: Using Matlab/Simulink45 pages
- Data Manual: TMS320F28335, TMS320F28334, TMS320F28332 TMS320F28235, TMS320F28234, TMS320F28232No ratings yetData Manual: TMS320F28335, TMS320F28334, TMS320F28332 TMS320F28235, TMS320F28234, TMS320F28232199 pages
- DLNA Guidelines June 2016 - Part 1-1 Architectures and ProtocolsNo ratings yetDLNA Guidelines June 2016 - Part 1-1 Architectures and Protocols797 pages
- Examining The Efficacy of FHSS and OFDM Protocol For Drone CommNo ratings yetExamining The Efficacy of FHSS and OFDM Protocol For Drone Comm60 pages
- Instant Download Silicon Photonics For High-Performance Computing and Beyond 1st Edition Gabriela Nicolescu (Editor) PDF All Chapter100% (1)Instant Download Silicon Photonics For High-Performance Computing and Beyond 1st Edition Gabriela Nicolescu (Editor) PDF All Chapter69 pages
- (2002 Sams) Wireless LANs - 2nd EditionNo ratings yet(2002 Sams) Wireless LANs - 2nd Edition360 pages
- Certificateless Signcryption For Wireless Sensor NetworksNo ratings yetCertificateless Signcryption For Wireless Sensor Networks160 pages
- Ubeda - Control PID para Brazo RobóticoNo ratings yetUbeda - Control PID para Brazo Robótico50 pages
- Get Handbook on Satellite Communications HSC 3rd Edition International Telecommunications Union PDF ebook with Full Chapters NowNo ratings yetGet Handbook on Satellite Communications HSC 3rd Edition International Telecommunications Union PDF ebook with Full Chapters Now84 pages
- Transcede 2xxx Hardware Data Sheet (PN 842xx-DSH-001-A) PDFNo ratings yetTranscede 2xxx Hardware Data Sheet (PN 842xx-DSH-001-A) PDF257 pages
- 5287 Full Ducument Rotary Position SensorsNo ratings yet5287 Full Ducument Rotary Position Sensors66 pages
- A Complex Variable Approach To The Analysis of Linear Multivariable Feedback Sy PDFNo ratings yetA Complex Variable Approach To The Analysis of Linear Multivariable Feedback Sy PDF178 pages
- Multifunction Radar Simulator (MFRSIM) : Defence R&D Canada - OttawaNo ratings yetMultifunction Radar Simulator (MFRSIM) : Defence R&D Canada - Ottawa55 pages
- DynamicProfileofSwitched ModeConverter ModelingAnalysisandControl (2009)No ratings yetDynamicProfileofSwitched ModeConverter ModelingAnalysisandControl (2009)362 pages
- How To Become An HLA Guru in Ashort TimeNo ratings yetHow To Become An HLA Guru in Ashort Time39 pages
- Comprehensive System Identification of Ducted Fan UAVsNo ratings yetComprehensive System Identification of Ducted Fan UAVs174 pages
- [Ebooks PDF] download Silicon Photonics for High-Performance Computing and Beyond 1st Edition Mahdi Nikdast full chapters100% (3)[Ebooks PDF] download Silicon Photonics for High-Performance Computing and Beyond 1st Edition Mahdi Nikdast full chapters37 pages
- Image Denoising With The Non-Local Means AlgorithmNo ratings yetImage Denoising With The Non-Local Means Algorithm9 pages
- Performance Analysis of Wired, Wireless and Optical Network Using NS2No ratings yetPerformance Analysis of Wired, Wireless and Optical Network Using NS27 pages
- NASA technology Roadmaps TA08 SensorsObservatories_finalNo ratings yetNASA technology Roadmaps TA08 SensorsObservatories_final115 pages
- Flexible Bioelectronics With Power Autonomous Sensing and Data AnalyticsNo ratings yetFlexible Bioelectronics With Power Autonomous Sensing and Data Analytics187 pages
- Conversion of Earth-Centered Earth-Fixed Coordinates To Geodetic CoordinatesNo ratings yetConversion of Earth-Centered Earth-Fixed Coordinates To Geodetic Coordinates6 pages
- Small Radar - Sea.clutter - Scattering.the.k.distribution - And.radar - Performance.radar - Sonar..navigationNo ratings yetSmall Radar - Sea.clutter - Scattering.the.k.distribution - And.radar - Performance.radar - Sonar..navigation586 pages
- Download full (Ebook) Advanced Mathematical Techniques in Engineering Sciences by Davim, J. Paulo; Ram, Mangey ISBN 9781315148977, 9781351371896, 1315148978, 1351371894 ebook all chapters100% (4)Download full (Ebook) Advanced Mathematical Techniques in Engineering Sciences by Davim, J. Paulo; Ram, Mangey ISBN 9781315148977, 9781351371896, 1315148978, 1351371894 ebook all chapters57 pages
- Modern Inertial and Satellite Navigation SystemsNo ratings yetModern Inertial and Satellite Navigation Systems119 pages
- Window on the past: The photographic ethnohistory of the Northern and Kaigani HaidaFrom EverandWindow on the past: The photographic ethnohistory of the Northern and Kaigani HaidaNo ratings yet
- Prasad S T Land Resources Monitoring Modeling and Mapping Wi PDFNo ratings yetPrasad S T Land Resources Monitoring Modeling and Mapping Wi PDF858 pages
- Remote Sensing Handbook. Volume III, Remote Sensing of Water Resources, Disasters, and Urban Studies (PDFDrive)No ratings yetRemote Sensing Handbook. Volume III, Remote Sensing of Water Resources, Disasters, and Urban Studies (PDFDrive)679 pages
- Prasad S. Thenkabail - Remote Sensing Handbook - Three Volume Set - Land Resources Monitoring, Modeling, and Mapping With Remote SensingNo ratings yetPrasad S. Thenkabail - Remote Sensing Handbook - Three Volume Set - Land Resources Monitoring, Modeling, and Mapping With Remote Sensing858 pages
- Nanoscale Spectroscopy with Applications 1st Edition Sarhan M. Musa instant download100% (1)Nanoscale Spectroscopy with Applications 1st Edition Sarhan M. Musa instant download53 pages
- Instant ebooks textbook (Ebook) Principles of Modeling Uncertainties in Spatial Data and Spatial Analyses by Wenzhong Shi ISBN 9781420059274, 1420059270 download all chapters100% (4)Instant ebooks textbook (Ebook) Principles of Modeling Uncertainties in Spatial Data and Spatial Analyses by Wenzhong Shi ISBN 9781420059274, 1420059270 download all chapters81 pages
- Get Advances in Environmental Remote Sensing Sensors Algorithms and Applications Remote Sensing Applications Series 1st Edition Qihao Weng free all chapters100% (15)Get Advances in Environmental Remote Sensing Sensors Algorithms and Applications Remote Sensing Applications Series 1st Edition Qihao Weng free all chapters60 pages
- Challenges and Opportunities in Carbon Capture, Utilization and Storage: A Process Systems Engineering PerspectiveNo ratings yetChallenges and Opportunities in Carbon Capture, Utilization and Storage: A Process Systems Engineering Perspective26 pages
- Journal of Catalysis - My International Research Paper - 2005 AD.No ratings yetJournal of Catalysis - My International Research Paper - 2005 AD.11 pages
- H. R. Byers, H. E. Landsberg, H. Wexler, B. Haurwitz, A. F. Spilhaus, H. C. Willett, H. G. Houghton (Auth.), Thomas F. Malone (Eds.)-Compendium of Meteorology_ Prepared Under the Direction of the CommNo ratings yetH. R. Byers, H. E. Landsberg, H. Wexler, B. Haurwitz, A. F. Spilhaus, H. C. Willett, H. G. Houghton (Auth.), Thomas F. Malone (Eds.)-Compendium of Meteorology_ Prepared Under the Direction of the Comm1,304 pages
- IFU - Efficia CM Series Rel A.01 - English100% (1)IFU - Efficia CM Series Rel A.01 - English444 pages
- Electrochemical Reduction of CO To Oxalic Acid: V.S. Boor Thesis Defense 09-04-2020No ratings yetElectrochemical Reduction of CO To Oxalic Acid: V.S. Boor Thesis Defense 09-04-202036 pages
- Polyoxometalate-Based Catalysts For CO2 ConversionNo ratings yetPolyoxometalate-Based Catalysts For CO2 Conversion26 pages
- 3.3 Yield and Atom Economy of Chemical Reactions QPNo ratings yet3.3 Yield and Atom Economy of Chemical Reactions QP23 pages
- Article of Global Warming: Submitted By: Mikko Gabrielle OrprecioNo ratings yetArticle of Global Warming: Submitted By: Mikko Gabrielle Orprecio7 pages
- Multiple Choice Test On 1.1 Introduction To The Particulate Nature of Matter and Chemical ChangeNo ratings yetMultiple Choice Test On 1.1 Introduction To The Particulate Nature of Matter and Chemical Change4 pages
- AQA GCSE Chemistry of The Atmosphere Multiple Choice Questions Set 3100% (1)AQA GCSE Chemistry of The Atmosphere Multiple Choice Questions Set 32 pages
- (PDF Download) Remote Sensing Digital Image Analysis Sixth Edition John Alan Richards Fulll Chapter(PDF Download) Remote Sensing Digital Image Analysis Sixth Edition John Alan Richards Fulll Chapter
- Total Ionizing Dose Effects On Current Sense AmplifiersTotal Ionizing Dose Effects On Current Sense Amplifiers
- Data Manual: TMS320F28335, TMS320F28334, TMS320F28332 TMS320F28235, TMS320F28234, TMS320F28232Data Manual: TMS320F28335, TMS320F28334, TMS320F28332 TMS320F28235, TMS320F28234, TMS320F28232
- DLNA Guidelines June 2016 - Part 1-1 Architectures and ProtocolsDLNA Guidelines June 2016 - Part 1-1 Architectures and Protocols
- Examining The Efficacy of FHSS and OFDM Protocol For Drone CommExamining The Efficacy of FHSS and OFDM Protocol For Drone Comm
- Instant Download Silicon Photonics For High-Performance Computing and Beyond 1st Edition Gabriela Nicolescu (Editor) PDF All ChapterInstant Download Silicon Photonics For High-Performance Computing and Beyond 1st Edition Gabriela Nicolescu (Editor) PDF All Chapter
- Certificateless Signcryption For Wireless Sensor NetworksCertificateless Signcryption For Wireless Sensor Networks
- Get Handbook on Satellite Communications HSC 3rd Edition International Telecommunications Union PDF ebook with Full Chapters NowGet Handbook on Satellite Communications HSC 3rd Edition International Telecommunications Union PDF ebook with Full Chapters Now
- Transcede 2xxx Hardware Data Sheet (PN 842xx-DSH-001-A) PDFTranscede 2xxx Hardware Data Sheet (PN 842xx-DSH-001-A) PDF
- A Complex Variable Approach To The Analysis of Linear Multivariable Feedback Sy PDFA Complex Variable Approach To The Analysis of Linear Multivariable Feedback Sy PDF
- Multifunction Radar Simulator (MFRSIM) : Defence R&D Canada - OttawaMultifunction Radar Simulator (MFRSIM) : Defence R&D Canada - Ottawa
- DynamicProfileofSwitched ModeConverter ModelingAnalysisandControl (2009)DynamicProfileofSwitched ModeConverter ModelingAnalysisandControl (2009)
- Comprehensive System Identification of Ducted Fan UAVsComprehensive System Identification of Ducted Fan UAVs
- [Ebooks PDF] download Silicon Photonics for High-Performance Computing and Beyond 1st Edition Mahdi Nikdast full chapters[Ebooks PDF] download Silicon Photonics for High-Performance Computing and Beyond 1st Edition Mahdi Nikdast full chapters
- Image Denoising With The Non-Local Means AlgorithmImage Denoising With The Non-Local Means Algorithm
- Performance Analysis of Wired, Wireless and Optical Network Using NS2Performance Analysis of Wired, Wireless and Optical Network Using NS2
- NASA technology Roadmaps TA08 SensorsObservatories_finalNASA technology Roadmaps TA08 SensorsObservatories_final
- Flexible Bioelectronics With Power Autonomous Sensing and Data AnalyticsFlexible Bioelectronics With Power Autonomous Sensing and Data Analytics
- Conversion of Earth-Centered Earth-Fixed Coordinates To Geodetic CoordinatesConversion of Earth-Centered Earth-Fixed Coordinates To Geodetic Coordinates
- Small Radar - Sea.clutter - Scattering.the.k.distribution - And.radar - Performance.radar - Sonar..navigationSmall Radar - Sea.clutter - Scattering.the.k.distribution - And.radar - Performance.radar - Sonar..navigation
- Download full (Ebook) Advanced Mathematical Techniques in Engineering Sciences by Davim, J. Paulo; Ram, Mangey ISBN 9781315148977, 9781351371896, 1315148978, 1351371894 ebook all chaptersDownload full (Ebook) Advanced Mathematical Techniques in Engineering Sciences by Davim, J. Paulo; Ram, Mangey ISBN 9781315148977, 9781351371896, 1315148978, 1351371894 ebook all chapters
- Understanding Checksums and Cyclic Redundancy ChecksFrom EverandUnderstanding Checksums and Cyclic Redundancy Checks
- Window on the past: The photographic ethnohistory of the Northern and Kaigani HaidaFrom EverandWindow on the past: The photographic ethnohistory of the Northern and Kaigani Haida
- Prasad S T Land Resources Monitoring Modeling and Mapping Wi PDFPrasad S T Land Resources Monitoring Modeling and Mapping Wi PDF
- Remote Sensing Handbook. Volume III, Remote Sensing of Water Resources, Disasters, and Urban Studies (PDFDrive)Remote Sensing Handbook. Volume III, Remote Sensing of Water Resources, Disasters, and Urban Studies (PDFDrive)
- Prasad S. Thenkabail - Remote Sensing Handbook - Three Volume Set - Land Resources Monitoring, Modeling, and Mapping With Remote SensingPrasad S. Thenkabail - Remote Sensing Handbook - Three Volume Set - Land Resources Monitoring, Modeling, and Mapping With Remote Sensing
- Nanoscale Spectroscopy with Applications 1st Edition Sarhan M. Musa instant downloadNanoscale Spectroscopy with Applications 1st Edition Sarhan M. Musa instant download
- Instant ebooks textbook (Ebook) Principles of Modeling Uncertainties in Spatial Data and Spatial Analyses by Wenzhong Shi ISBN 9781420059274, 1420059270 download all chaptersInstant ebooks textbook (Ebook) Principles of Modeling Uncertainties in Spatial Data and Spatial Analyses by Wenzhong Shi ISBN 9781420059274, 1420059270 download all chapters
- Get Advances in Environmental Remote Sensing Sensors Algorithms and Applications Remote Sensing Applications Series 1st Edition Qihao Weng free all chaptersGet Advances in Environmental Remote Sensing Sensors Algorithms and Applications Remote Sensing Applications Series 1st Edition Qihao Weng free all chapters
- Challenges and Opportunities in Carbon Capture, Utilization and Storage: A Process Systems Engineering PerspectiveChallenges and Opportunities in Carbon Capture, Utilization and Storage: A Process Systems Engineering Perspective
- Journal of Catalysis - My International Research Paper - 2005 AD.Journal of Catalysis - My International Research Paper - 2005 AD.
- H. R. Byers, H. E. Landsberg, H. Wexler, B. Haurwitz, A. F. Spilhaus, H. C. Willett, H. G. Houghton (Auth.), Thomas F. Malone (Eds.)-Compendium of Meteorology_ Prepared Under the Direction of the CommH. R. Byers, H. E. Landsberg, H. Wexler, B. Haurwitz, A. F. Spilhaus, H. C. Willett, H. G. Houghton (Auth.), Thomas F. Malone (Eds.)-Compendium of Meteorology_ Prepared Under the Direction of the Comm
- Electrochemical Reduction of CO To Oxalic Acid: V.S. Boor Thesis Defense 09-04-2020Electrochemical Reduction of CO To Oxalic Acid: V.S. Boor Thesis Defense 09-04-2020
- Polyoxometalate-Based Catalysts For CO2 ConversionPolyoxometalate-Based Catalysts For CO2 Conversion
- 3.3 Yield and Atom Economy of Chemical Reactions QP3.3 Yield and Atom Economy of Chemical Reactions QP
- Article of Global Warming: Submitted By: Mikko Gabrielle OrprecioArticle of Global Warming: Submitted By: Mikko Gabrielle Orprecio
- Multiple Choice Test On 1.1 Introduction To The Particulate Nature of Matter and Chemical ChangeMultiple Choice Test On 1.1 Introduction To The Particulate Nature of Matter and Chemical Change
- AQA GCSE Chemistry of The Atmosphere Multiple Choice Questions Set 3AQA GCSE Chemistry of The Atmosphere Multiple Choice Questions Set 3