THE EVALUATION OF CORONA AND
IKONOS SATELLITE IMAGERY FOR
ARCHAEOLOGICAL APPLICATIONS IN A
SEMI-ARID ENVIRONMENT.
Volume 1 of 1
Thesis submitted for the degree of
Doctor of Philosophy
at the University of Durham
By
Anthony Richard Beck BA (Newcastle upon Tyne)
Department of Archaeology
University of Durham
2004
THE EVALUATION OF CORONA AND
IKONOS SATELLITE IMAGERY FOR
ARCHAEOLOGICAL APPLICATIONS IN A
SEMI-ARID ENVIRONMENT.
Volume 1 of 1
Thesis submitted for the degree of
Doctor of Philosophy
at the University of Durham
By
Anthony Richard Beck BA (Newcastle upon Tyne)
Department of Archaeology
University of Durham
2004
THE EVALUATION OF CORONA AND
IKONOS SATELLITE IMAGERY FOR
ARCHAEOLOGICAL APPLICATIONS IN A
SEMI-ARID ENVIRONMENT.
Volume 1 of 1
Thesis submitted for the degree of
Doctor of Philosophy
at the University of Durham
By
Anthony Richard Beck BA (Newcastle upon Tyne)
Department of Archaeology
University of Durham
2004
ABSTRACT
‘The evaluation of Corona and Ikonos satellite imagery for archaeological
applications in a semi-arid environment’ by Anthony Richard Beck
Archaeologists have been aware of the potential of satellite imagery as a tool almost since the
first Earth remote sensing satellite. Initially sensors such as Landsat had a ground resolution
which was too coarse for thorough archaeological prospection although the imagery was
used for geo-archaeological and enviro-archaeological analyses. In the intervening years the
spatial and spectral resolution of these sensing devices has improved. In recent years two
important occurrences enhanced the archaeological applicability of imagery from satellite
platforms: The declassification of high resolution photography by the American and Russian
governments and the deregulation of commercial remote sensing systems allowing the
collection of sub metre resolution imagery.
This thesis aims to evaluate the archaeological application of three potentially important
resources: Corona space photography and Ikonos panchromatic and multispectral imagery.
These resources are evaluated in conjunction with Landsat Thematic Mapper (TM) imagery
over a 600 square km study area in the semi-arid environment around Homs, Syria. The
archaeological resource in this area is poorly understood, mapped and documented.
The images are evaluated for their ability to create thematic layers and to locate
archaeological residues in different environmental zones. Further consideration is given to
the physical factors that allow archaeological residues to be identified and how satellite
imagery and modern technology may impact on Cultural Resource Management.
This research demonstrates that modern high resolution and historic satellite imagery can be
important tools for archaeologists studying in semi-arid environments. The imagery has
allowed a representative range of archaeological features and landscape themes to be
identified. The research shows that the use of satellite imagery can have significant impact on
the design of the archaeological survey in the middle-east and perhaps in other environments.
Key words: archaeology, Corona, CRM, GIS, Ikonos, Landsat, landscape, prospection,
remote sensing, satellite, semi-arid, soil, Syria
TABLE OF CONTENTS
TABLE OF CONTENTS .................................................................................................................................. I
LIST OF FIGURES ........................................................................................................................................ VI
LIST OF TABLES .........................................................................................................................................XII
ACKNOWLEDGMENTS........................................................................................................................... XIII
GLOSSARY....................................................................................................................................................XV
SECTION 1 INTRODUCTION.......................................................................................................................1
CHAPTER 1 RESEARCH INTRODUCTION AND SUMMARY.............................................................2
1.1
INTRODUCTION .................................................................................................................................2
1.2
THESIS STRUCTURE ...........................................................................................................................3
1.3
RESEARCH SUMMARY .......................................................................................................................3
1.3.1 Traditional approaches to landscape survey ...................................................................................4
1.3.2 Settlement and landscape development in the Homs region............................................................6
1.3.3 Introduction to the data sources .......................................................................................................6
1.3.4 Remote sensing..................................................................................................................................8
1.3.5 Image interpretation .........................................................................................................................9
1.3.6 Archaeological data........................................................................................................................11
1.3.7 Remote sensing and archaeology ...................................................................................................13
1.4
METHOD SUMMARY ........................................................................................................................14
1.4.1 Aims and objectives for the evaluation of satellite imagery ..........................................................17
1.5
OTHER RESEARCH INTO THE ARCHAEOLOGICAL APPLICATION OF HIGH RESOLUTION SATELLITE
IMAGERY ........................................................................................................................................................18
CHAPTER 2 CONCEPTS OF REMOTE SENSING FOR ARCHAEOLOGY .....................................26
2.1
REMOTE SENSING – INTRODUCTION AND DEFINITION ....................................................................26
2.1.1 Electromagnetic energy ..................................................................................................................27
2.1.2 Interactions with the atmosphere....................................................................................................32
2.1.3 Interactions with an object .............................................................................................................36
2.1.4 Discussion .......................................................................................................................................36
2.2
THE COMPONENTS OF REMOTE SENSING SYSTEMS .........................................................................38
2.2.1 Hardware: Platforms......................................................................................................................38
2.2.2 Hardware: Sensor systems .............................................................................................................40
2.2.3 Hardware: Sensor characteristics..................................................................................................40
2.2.4 Software: Image processing systems ..............................................................................................46
2.2.5 Processing: Image pre-processing .................................................................................................47
2.2.6 Processing: Image display and visualisation.................................................................................52
2.2.7 Processing: Image classification....................................................................................................62
2.3
IMAGE INTEGRATION AND PROBLEMS OF SCALE ............................................................................68
2.4
ARCHAEOLOGICAL INTERPRETATION .............................................................................................69
CHAPTER 3 CONCEPTS OF LANDSCAPE FOR ARCHAEOLOGY .................................................73
3.1
LANDSCAPE ARCHAEOLOGY: INTRODUCTION AND DEFINITION .....................................................73
3.2
THE COMPONENTS OF ARCHAEOLOGICAL LANDSCAPES .................................................................74
3.2.1 Natural landscapes .........................................................................................................................75
3.2.2 Landscapes and space ....................................................................................................................76
3.2.3 Landscapes and scale .....................................................................................................................77
3.2.4 Cultural landscapes ........................................................................................................................77
3.2.5 Discussion .......................................................................................................................................78
3.3
TRANSFORMATION OF ARCHAEOLOGICAL LANDSCAPES ................................................................79
3.3.1 Cultural transforms.........................................................................................................................79
3.3.2 Natural transforms..........................................................................................................................81
3.3.3 Discussion .......................................................................................................................................83
3.4
PRACTICE AND ARCHAEOLOGICAL LANDSCAPES ...........................................................................85
3.4.1 Survey objectives.............................................................................................................................85
3.4.2 Desk based assessment ...................................................................................................................87
3.4.3 Ground reconnaissance ..................................................................................................................90
3.4.4 Summary..........................................................................................................................................94
3.5
LANDSCAPE MODELLING ................................................................................................................95
3.5.1 Modelling archaeological landscapes............................................................................................96
3.5.2 ‘Sites’ as interpretative units ........................................................................................................101
3.6
DISCUSSION ..................................................................................................................................102
3.6.1 Research implications of landscape archaeology........................................................................103
SECTION 2: METHODOLOGY AND ANALYSIS.................................................................................105
CHAPTER 4 SETTLEMENT AND LANDSCAPE DEVELOPMENT IN THE HOMS REGION,
SYRIA: AIMS, SETTING, RESEARCH QUESTIONS AND FIELDWORK (1999-2003) ................106
4.1
THE SHR PROJECT ........................................................................................................................106
4.2
ENVIRONMENTAL ZONES, LAND USE AND ARCHAEOLOGICAL SUMMARY ....................................109
4.2.1 The marl zone (units 1 and 4) .......................................................................................................109
4.2.2 The basalt zone (units 3 and 7).....................................................................................................112
4.2.3 The alluvial zone (units 2, 5 and 8) ..............................................................................................113
4.2.4 Lake Qatina (unit 8)......................................................................................................................115
4.3
SURVEY METHODOLOGY ..............................................................................................................117
4.3.1 SHR project DBA ..........................................................................................................................117
4.3.2 Ground observation ......................................................................................................................118
4.4
FIELDWORK PROGRAMME.............................................................................................................120
4.5
SHR RESEARCH GOALS .................................................................................................................123
CHAPTER 5 IMAGE PREPARATION.....................................................................................................125
5.1
DATA SOURCES.............................................................................................................................125
5.1.1 Landsat..........................................................................................................................................125
5.1.2 Ikonos ............................................................................................................................................131
5.1.3 Corona...........................................................................................................................................134
5.1.4 Ancillary data sets.........................................................................................................................138
5.2
IMAGE SELECTION .........................................................................................................................141
5.2.1 Atmospheric ramifications............................................................................................................143
5.2.2 Land management ramifications ..................................................................................................144
5.2.3 Field observations.........................................................................................................................146
5.2.4 Determining acquisition times......................................................................................................146
5.2.5 Data sets acquired ........................................................................................................................149
5.3
IMAGE PRE-PROCESSING ...............................................................................................................153
5.3.1 Digitising Corona .........................................................................................................................153
5.3.2 Atmospheric calibration ...............................................................................................................156
5.3.3 Correction of topographic effects.................................................................................................157
5.4
GEO-RECTIFICATION .....................................................................................................................158
5.4.1 Co-registration..............................................................................................................................163
5.5
IMAGE RESCALING ........................................................................................................................165
5.6
IMAGE FUSION ...............................................................................................................................166
5.6.1 Pan-sharpening.............................................................................................................................167
5.7
DISCUSSION ..................................................................................................................................170
CHAPTER 6 SATELLITE IMAGERY FOR THEMATIC INFORMATION EXTRACTION........171
6.1
LANDSCAPE THEMES.....................................................................................................................171
6.2
LAND COVER MAPPING .................................................................................................................172
6.2.1 Land cover classification systems.................................................................................................173
6.2.2 Land cover mapping methodology ...............................................................................................175
6.2.3 Vector digitising methodology......................................................................................................177
6.3
HYDROLOGY NETWORKS ..............................................................................................................179
6.4
COMMUNICATION NETWORKS ......................................................................................................184
ii
6.5
LAND COVER PARCELS..................................................................................................................184
6.6
GEOLOGY/SOIL MAPPING ..............................................................................................................188
6.7
DIGITAL TERRAIN MODELLING .....................................................................................................192
6.7.1 DTM from contour data................................................................................................................193
6.7.2 DTM generation from remotely sensed data................................................................................194
6.7.3 DTM evaluation ............................................................................................................................197
6.8
DISCUSSION ..................................................................................................................................197
CHAPTER 7 SATELLITE IMAGERY AS A PROSPECTION TOOL................................................199
7.1
ARCHAEOLOGICAL PROSPECTION .................................................................................................199
7.2
IMPACT OF THE ENVIRONMENTAL ZONES ON PROSPECTION .........................................................201
7.3
METHODOLOGICAL BACKGROUND ...............................................................................................202
7.3.1 Qualitative methodologies ............................................................................................................203
7.3.2 Quantitative methodologies ..........................................................................................................204
7.3.3 Digitising methodology.................................................................................................................205
7.4
PROSPECTION IN THE BASALT ZONE .............................................................................................210
7.4.1 Image selection in the basalt zone ................................................................................................214
7.4.2 Qualitative techniques in the basalt zone.....................................................................................214
7.4.3 Quantitative techniques in the basalt zone...................................................................................216
7.4.4 Case study in the basalt zone........................................................................................................216
7.4.5 Evaluation of sensors in the basalt zone ......................................................................................220
7.4.6 Russian aerial photographs in the basalt zone ............................................................................224
7.5
PROSPECTION IN THE MARL ZONE .................................................................................................226
7.5.1 Image selection in the marl zone ..................................................................................................231
7.5.2 Spectral nature of the residues in the marl zone ..........................................................................233
7.5.3 Qualitative techniques in the marl zone .......................................................................................238
7.5.4 Quantitative techniques in the marl zone .....................................................................................250
7.5.5 Evaluation of sensors in the marl zone.........................................................................................253
7.6
PROSPECTION IN THE ALLUVIAL ZONE (AND FLOODPLAIN)..........................................................256
7.7
DISCUSSION ..................................................................................................................................259
CHAPTER 8 ANALYSIS OF MARL SOILS ............................................................................................261
8.1
INTRODUCTION .............................................................................................................................261
8.2
PEDOLOGY ....................................................................................................................................263
8.2.1 Soil biochemistry...........................................................................................................................265
8.2.2 Organic matter..............................................................................................................................265
8.2.3 Soil particle size ............................................................................................................................266
8.3
SOIL SPECTRAL REFLECTANCE......................................................................................................268
8.3.1 Organic matter response ..............................................................................................................270
8.3.2 Moisture response.........................................................................................................................271
8.3.3 Iron and Iron-Oxide response ......................................................................................................272
8.3.4 Particle size response ...................................................................................................................272
8.3.5 Theoretical reasons for changes in reflectance ...........................................................................273
8.4
SOIL TEXTURE ANALYSIS ..............................................................................................................274
8.4.1 The 2001 soil sampling programme.............................................................................................276
8.4.2 The 2003 soil sampling programme.............................................................................................293
8.4.3 Interpretation of the analysis........................................................................................................296
8.5
DISCUSSION ..................................................................................................................................304
8.5.1 Non-organic imports.....................................................................................................................305
8.5.2 Organic matter imports ................................................................................................................306
8.5.3 Reflectance implications of the soil analysis: summary and conclusions ...................................307
8.5.4 Recommendations for field collection and analysis.....................................................................308
CHAPTER 9 SATELLITE IMAGERY AS A CRM TOOL....................................................................309
9.1
INTRODUCTION .............................................................................................................................309
9.2
SATELLITE IMAGERY AS A CONTEXTUAL BACKDROP ...................................................................309
9.2.1 The comparison of satellite imagery against digital map bases..................................................310
9.2.2 Satellite imagery as a presentation and visualisation tool ..........................................................313
iii
9.3
SATELLITE IMAGERY AND MOBILE APPLICATIONS .......................................................................315
9.3.1 Site navigation...............................................................................................................................316
9.3.2 Spatial data collection ..................................................................................................................317
9.4
TIME CHANGE ANALYSIS ..............................................................................................................317
9.4.1 Data preparation...........................................................................................................................318
9.4.2 Impact of changes over time .........................................................................................................319
9.4.3 Site monitoring..............................................................................................................................326
9.5
DISCUSSION ..................................................................................................................................328
SECTION 3 SUMMARY, RECOMMENDATIONS AND CONCLUSIONS.......................................330
CHAPTER 10 DISCUSSION OF THE RESULTS ...................................................................................331
10.1
ARCHAEOLOGICAL PROSPECTION SUMMARY ...............................................................................331
10.1.1 Positive evidence.........................................................................................................................337
10.1.2 Negative evidence .......................................................................................................................337
10.1.3 Masked evidence .........................................................................................................................338
10.1.4 Image interpretation key.............................................................................................................342
10.1.5 Quantitative summary.................................................................................................................342
10.1.6 Effects of resolution on archaeological detection......................................................................343
10.2
THEMATIC EXTRACTION SUMMARY .............................................................................................347
10.3
CRM APPLICATION SUMMARY .....................................................................................................347
10.3.1 Impact of landscape change on archaeological residues and site monitoring .........................347
10.4
LIMITATIONS OF ARCHAEOLOGICAL INTERPRETATION FROM SATELLITE IMAGERY ....................348
10.4.1 Crop mark identification.............................................................................................................348
10.5
RECOMMENDATIONS ....................................................................................................................349
10.5.1 Issues of implementation.............................................................................................................351
10.5.2 Impact of the recommendations on the application area...........................................................355
10.6
FUTURE SYSTEMS..........................................................................................................................358
CHAPTER 11 CONCLUSIONS ..................................................................................................................360
11.1
GENERAL DISCUSSION ..................................................................................................................360
11.2
SATELLITE IMAGERY AS A COMPLEMENT TO LANDSCAPE ARCHAEOLOGICAL SURVEY ...............361
11.2.1 Satellite imagery comparison against aerial photography........................................................363
11.2.2 Satellite imagery comparison against other landscape survey techniques ...............................364
11.3
SOME LIMITATIONS OF ARCHAEOLOGICAL REMOTE SENSING ......................................................365
11.4
CLASSIFICATION, CLASSIFICATION, CLASSIFICATION ...................................................................366
11.5
SUMMARY .....................................................................................................................................367
APPENDIX I : THE ARCHAEOLOGICAL DATA MODEL AND ENVIRONMENT .....................369
I.1
INTRODUCTION ..................................................................................................................................369
I.1.1 Project requirements of the data model ........................................................................................371
I.1.2 Model brief.....................................................................................................................................374
I.2 MODEL ENVIRONMENT ......................................................................................................................376
I.2.1 Software environment ....................................................................................................................376
I.2.2 Hardware environment..................................................................................................................381
I.3 SHR PROJECT DATA MODEL ..............................................................................................................382
I.3.1 Data flowline..................................................................................................................................382
I.3.2 A-spatial model..............................................................................................................................383
I.3.3 Spatial model .................................................................................................................................389
I.3.4 Data validation ..............................................................................................................................391
I.3.5 Audit trail .......................................................................................................................................393
I.3.6 Data generalisation .......................................................................................................................393
I.4 DOCUMENTATION ..............................................................................................................................395
I.4.1 Metadata ........................................................................................................................................396
I.5 ARCHIVING AND RE-USE ....................................................................................................................397
I.6 FUTURE REQUIREMENTS ....................................................................................................................399
I.7 USING THE DATA MODEL AND RECORDING SYSTEM ..........................................................................400
iv
APPENDIX II : COULTER SAMPLE PROCESSING METHODOLOGY ........................................401
APPENDIX III PARTICLE SIZE ANALYSIS RESULTS FROM SITES 97, 218, 221, 238, 271, 279,
339, 478, 496 AND 508...................................................................................................................................402
III.1
III.2
III.3
III.4
III.5
III.6
III.7
III.8
III.9
III.10
PARTICLE SIZE ANALYSIS AT SITE 97............................................................................................402
PARTICLE SIZE ANALYSIS AT SITE 218..........................................................................................405
PARTICLE SIZE ANALYSIS AT SITE 221..........................................................................................405
PARTICLE SIZE ANALYSIS AT SITE 238..........................................................................................408
PARTICLE SIZE ANALYSIS AT SITE 271..........................................................................................408
PARTICLE SIZE ANALYSIS AT SITE 279..........................................................................................411
PARTICLE SIZE ANALYSIS AT SITE 339..........................................................................................411
PARTICLE SIZE ANALYSIS AT SITE 478..........................................................................................414
PARTICLE SIZE ANALYSIS AT SITE 496..........................................................................................414
PARTICLE SIZE ANALYSIS AT SITE 508..........................................................................................417
REFERENCES CITED.................................................................................................................................418
v
LIST OF FIGURES
Number
Page
FIGURE 1 EXAMPLES OF THE ARCHAEOLOGICAL RESIDUES IN THE APPLICATION AREA......................................5
FIGURE 2 COMPARISON OF THE 3 PRIMARY SATELLITE SENSORS USED IN THE RESEARCH. .................................7
FIGURE 3 OVERVIEW OF REMOTELY SENSED IMAGE ACQUISITION (AFTER CAMPBELL 2002).............................8
FIGURE 4 SCHEMATIC SEQUENCE FOR DIGITAL ANALYSIS (AFTER CAMPBELL 2002). ......................................10
FIGURE 5 A SCHEMATIC REPRESENTATION OF CATEGORIES OF INFORMATION ABOUT THE REAL WORLD (AFTER
CLARK 1987 P. 57)....................................................................................................................................16
FIGURE 6 SCHEMATIC ARCHAEOLOGICAL INTERPRETATION PROCESS..............................................................17
FIGURE 7 SOIL RELATED ARCHAEOLOGICAL APPLICATIONS AT DIFFERENT PORTIONS OF THE EM SPECTRUM
(AFTER LUCAS 2001 P. 156)......................................................................................................................26
FIGURE 8 DIAGRAMMATIC REPRESENTATION OF A PHOTON. .............................................................................27
FIGURE 9 THE ELECTROMAGNETIC SPECTRUM AND ATMOSPHERIC ABSORPTION CURVE..................................28
FIGURE 10 ELECTROMAGNETIC ENERGY INTERACTIONS WITH A TARGET. ........................................................29
FIGURE 11 THE SPECTRAL DISTRIBUTION OF ENERGY EMITTED BY A BLACKBODY AS A FUNCTION OF ITS
TEMPERATURE (AFTER LILLESAND AND KIEFER 1999)............................................................................30
FIGURE 12 DIURNAL TEMPERATURE VARIATIONS (AFTER LILLESAND AND KIEFER 1999)...............................31
FIGURE 13 SOLAR RADIATION INTERACTIONS WITH THE ATMOSPHERE FOR SHORT WAVELENGTHS (AFTER
CAMPBELL 2002)......................................................................................................................................32
FIGURE 14 FIVE TYPES OF RADIATIVE INTERACTION WITH THE ATMOSPHERE AND HOW THEY IMPACT THE
INSTANTANEOUS FIELD OF VIEW (IFOV: AFTER TSO AND MATHER 2001 P. 15)....................................33
FIGURE 15 CHANGES IN REFLECTED, DIFFUSE, SCATTERED AND OBSERVED RADIATION OVER WAVELENGTH
(AFTER CAMPBELL 2002). ........................................................................................................................34
FIGURE 16 REFRACTION: HOW THE PATH OF RADIATION IS AFFECTED BY CHANGES IN THE DENSITY OF THE
MEDIUM.....................................................................................................................................................34
FIGURE 17 ATMOSPHERIC PARTICULATES AND THEIR SCATTERING EFFECTS (AFTER CAMPBELL 2002). .........35
FIGURE 18 SPECULAR AND PERFECTLY DIFFUSE (LAMBERTIAN) REFLECTANCE...............................................36
FIGURE 19 REMOTE SENSING FROM DIFFERENT PLATFORMS. ............................................................................37
FIGURE 20 SUN-SYNCHRONOUS ORBIT...............................................................................................................39
FIGURE 21 COMMON DIGITAL IMAGING SYSTEMS..............................................................................................40
FIGURE 22 CONTRASTING SPECTRAL RESOLUTIONS OF AVIRIS, LANDSAT AND PANCHROMATIC IMAGERY.
NOTE HOW CLOSELY THE 4 BANDS IN LANDSAT FOLLOW THE 100 AVIRIS BANDS. ..............................42
FIGURE 23 DECREASING SPATIAL RESOLUTION..................................................................................................42
FIGURE 24 THE CREATION, RECORDING AND ANALYSIS OF MIXED AND PURE PIXELS. ......................................43
FIGURE 25 THE BENEFITS OF INCREASED RADIOMETRIC RESOLUTION. THE 8 BIT IMAGERY IS OVEREXPOSED
WHEREAS STRUCTURES ARE IDENTIFIABLE IN THE 11 BIT IMAGERY (IMAGE COURTESY OF
DIGITALGLOBE AND DR. AMR AL-AZM).................................................................................................44
FIGURE 26 DECREASING RADIOMETRIC RESOLUTION (NOTE THIS IS NOT JUST IMPROVING CONTRAST)...........45
FIGURE 27 TEMPORAL RESOLUTION: LOOKING AT CHANGES OVER TIME. THE NEGATIVE CORONA IMAGE OF
1970 IS COMPARED TO A POSITIVE IKONOS IMAGE FROM 2002. THE MAJOR CHANGES ARE NOTED ON THE
IKONOS IMAGE. .........................................................................................................................................46
FIGURE 28 ROLL, PITCH AND YAW EFFECTS ON THE NADIR POINT.....................................................................47
FIGURE 29 COMMON PIXEL RESAMPLING TECHNIQUES......................................................................................48
FIGURE 30 EXAMPLES OF CAMERA LENS AND OBLIQUE DISTORTIONS (AFTER SCOLLAR 1990 P. 83; TENG 1997
P. 82). ........................................................................................................................................................49
FIGURE 31 UNIT GRID DISTORTIONS OF THE CORONA KH-4B PANORAMIC CAMERA (AFTER GALIATSATOS IN
PREP P. 93).................................................................................................................................................50
FIGURE 32 IMAGE DISPLACEMENT DUE TO VARIATIONS IN RELIEF (AFTER DIAL AND GRODECKI 2003). .........51
FIGURE 33 CORRECTION FOR RELIEF DISPLACEMENT IN A MONOSCOPIC IMAGE BY ORTHORECTIFICATION
(AFTER DIAL AND GRODECKI 2003).........................................................................................................52
FIGURE 34 LANDSAT BANDS 1, 2, 3, 4, 5, 6 AND 7 IN A LAYER STACK. ..............................................................53
FIGURE 35 INCREASED MAGNIFICATION ON 15 METER LANDSAT PANCHROMATIC IMAGERY...........................55
FIGURE 36 COMPARISON OF A TRUE AND FALSE COLOUR COMPOSITE MADE UP FROM THE VISUAL BANDS. ....56
FIGURE 37 HISTOGRAM OF LANDSAT TM BAND 4. ...........................................................................................56
vi
FIGURE 38 HISTOGRAM EQUALISATION OF LANDSAT TM BAND 4. ...................................................................57
FIGURE 39 6 COMPONENTS OF A PCA ON LANDSAT TM SCENE (EXCLUDING BAND 6). ...................................60
FIGURE 40 A MOVING ‘SHARPENING’ 3X3 KERNEL. ...........................................................................................61
FIGURE 41 THE CONCEPT OF CLASSIFICATION (AFTER TSO AND MATHER 2001 P. 4)........................................62
FIGURE 42 FEATURE SPACE REPRESENTATION (AFTER TSO AND MATHER 2001 P. 57). ....................................63
FIGURE 43 SPECTRAL SIGNATURES OF HAEMATITE AND FIR TREE....................................................................64
FIGURE 44 IMPROVING CLASSIFICATION ACCURACY BY REMOVING PIXEL MIXING EFFECTS THROUGH
INCREASING SPATIAL RESOLUTION. ..........................................................................................................65
FIGURE 45 IKONOS IMAGERY WITH ARCHAEOLOGICAL SITES OUTLINED IN WHITE (NOTICE THE INCREASE IN
REFLECTANCE AT THE ‘SITES’). ................................................................................................................66
FIGURE 46 UNSUPERVISED CLASSIFICATION OF A LANDSAT TM SCENE. ..........................................................67
FIGURE 47 AERIAL IDENTIFICATION: CROP, SHADOW AND SOIL MARKS (AFTER GREENE 1990). .....................69
FIGURE 48 CROPMARKS OBSERVED OVER A SEVEN YEAR PERIOD FROM AERIAL SORTIES FROM THE YORK
OFFICE OF ENGLISH HERITAGE. ................................................................................................................71
FIGURE 49 SCHEMATIC OF INTERACTIONS IN A HUMAN ECOSYSTEM HIGHLIGHTING THE RELATIONSHIPS
BETWEEN THE CULTURAL AND NON-CULTURAL ENVIRONMENT (MODIFIED FROM CLARKE 1978 P. 133;
MODIFIED FROM WATERS 1992 P. 5) ........................................................................................................76
FIGURE 50 FORMATION AND DEFORMATION PROCESS IN THE MARL. ................................................................84
FIGURE 51 NATURAL AND CULTURAL FORMATION AND DEFORMATION PROCESSES ON TELL NEBI MEND......84
FIGURE 52 HYPOTHETICAL SITE DISTRIBUTION AS PERTURBATIONS IN A DISTRIBUTION OF ARTEFACTS. THE
2
SITE LEVEL IS CALCULATED AT A DENSITY OF 2.5 ARTEFACTS PER M .....................................................97
FIGURE 53 EXAMPLE OF ASSEMBLAGE OVERLAP TO ERRONEOUSLY PRODUCE A SITE (AFTER BANNING 2002 P.
19).............................................................................................................................................................99
FIGURE 54 LOCATION MAP OF THE STUDY AREA..............................................................................................107
FIGURE 55 THE ENVIRONMENTAL ZONES IN THE APPLICATION AREA (LABELLED BY UNIT AND THEIR
RESPECTIVE AREAS) ON A 4,3,2 LANDSAT IMAGE (28TH OCTOBER 2000).............................................108
FIGURE 56 ELEMENTS OF THE MARL LANDSCAPE ............................................................................................109
FIGURE 57 ARCHAEOLOGICAL RESIDUES IN THE MARL....................................................................................110
FIGURE 58 ELEMENTS OF THE BASALT LANDSCAPE. ........................................................................................111
FIGURE 59 ARCHAEOLOGICAL RESIDUES IN THE BASALT. ...............................................................................112
FIGURE 60 ARCHAEOLOGICAL RESIDUES IN THE ALLUVIUM............................................................................112
FIGURE 61 ELEMENTS OF THE ALLUVIAL LANDSCAPE. ....................................................................................113
FIGURE 62 ELEMENTS OF THE LANDSCAPE AROUND LAKE QATINA................................................................114
FIGURE 63 ARCHAEOLOGICAL RESIDUES ERODED AT LAKE QATINA. .............................................................115
FIGURE 64 TELL SITES IN THE LACUSTRINE DEPOSITS AT THE SOUTH WESTERN END OF LAKE QATINA (SCALE
1:20,000).................................................................................................................................................116
FIGURE 65 SUMMARY OF FIELDWORK FROM 1999 TO 2003.............................................................................122
FIGURE 66 RELATIVE SPECTRAL RESPONSE CURVES FOR LANDSAT 7, 5 AND 4...............................................127
FIGURE 67 DIAGRAM OF THE VISIBLE AND IR REGION OF THE EM SPECTRUM AND LANDSAT TM BANDS.
GASES RESPONSIBLE FOR ATMOSPHERIC ABSORPTION ARE INDICATED (AFTER SABINS 1997 P. 5).......128
FIGURE 68 RELATIVE SPECTRAL RESPONSE CURVE FOR THE IKONOS MULTISPECTRAL AND PANCHROMATIC
BANDS (COURTESY SPACE IMAGING). ....................................................................................................131
FIGURE 69 THE LOCATION OF SPACE IMAGING REGIONAL AFFILIATES AND THEIR DIRECT SPHERES OF
INFLUENCE (COURTESY SPACE IMAGING). .............................................................................................133
FIGURE 70 CORONA MODULE PHOTOGRAPHED AT THE AIR AND SPACE MUSEUM (WASHINGTON D.C., USA).
NOTE THE FILM SPOOLS AND STEREO PANORAMIC CAMERAS. COURTESY OF KEITH CHALLIS. ............136
FIGURE 71 SPECTRAL RESPONSE OF THE KODAK EK 3404 FILM (AFTER KODAK 2003). ................................137
FIGURE 72 DIGITISED VECTOR BASEMAP. NOTE THE POOR QUALITY CONTOUR DATA IN THE NORTHERN
SEGMENT OF THE NORTHERN STUDY AREA.............................................................................................140
FIGURE 73 COMPARISON OF THE RUSSIAN AERIAL PHOTOGRAPHY, IKONOS AND CORONA IMAGERY (SCALE
1:5,000)...................................................................................................................................................142
FIGURE 74 SENSOR COMPARISON OVER DIFFERENT FEATURE TYPES...............................................................143
FIGURE 75 RAMS AND RESERVOIR IN THE BASALT...........................................................................................144
FIGURE 76 ELEVATION AND ANNUAL PRECIPITATION (AFTER HIRATA ET AL. 2001 P. 509) IN SYRIA. ............145
FIGURE 77 MONTHLY RAINFALL AVERAGE FOR THE HOMS REGION (DATA KINDLY SUPPLIED BY ICARDA).
TOTAL RAINFALL IS 480MM FOR FAO AND 442.6MM FOR GRUZGIPROVODKHOZ. ...............................148
FIGURE 78 COMPARISON OF THE SPATIAL CLARITY AND GEOMETRIC ACCURACY OF RAW IKONOS
PANCHROMATIC IMAGERY FROM DIFFERENT DATES AND ACQUISITION ORIENTATIONS........................151
vii
FIGURE 79 COMPARISON OF THE DIFFERENT RECTIFIED CORONA MISSIONS. ..................................................152
FIGURE 80 CORONA IMAGERY SCANNED AT DIFFERENT RESOLUTIONS ...........................................................154
FIGURE 81 SUBTRACTING A CONSTANT FROM A BAND IS EQUIVALENT TO TRANSLATING THE ORIGIN AND HAS
NO EFFECT ON THE VARIANCE-COVARIANCE MATRIX. HENCE DARK OBJECT SUBTRACTION HAS NO
EFFECT ON CLASSIFICATION RESULTS (AFTER SONG ET AL. 2000 P. 232). ..............................................155
FIGURE 82 ISOMETRIC VIEW OF THE APPLICATION AREA AND CONTOUR LINES...............................................158
FIGURE 83 RAW GPS READINGS OVERLYING RAW IKONOS DATA (AFTER BECK ET AL. IN PRESS). .................160
FIGURE 84 COMPARISON OF CORONA RECTIFICATION USING GCPS DERIVED FROM GPS AND IKONOS (AFTER
BECK ET AL. IN PRESS).............................................................................................................................162
FIGURE 85 THE LOCATION OF TWO GCPS, RAW IKONOS AND CORRECTED IKONOS........................................163
FIGURE 86 GEO-REFERENCING ERRORS. THE IKONOS IMAGE ON THE RIGHT IS OFFSET FROM THE LEFT IMAGE
BY 7.3M EAST AND 9.9M NORTH. ...........................................................................................................164
FIGURE 87 COMPARISON OF RESCALING TECHNIQUES. ALTHOUGH THE 4,3,2 FCC LOOK SIMILAR THE
HISTOGRAM OF THE IMAGE USING SD RESCALING IS SIGNIFICANTLY ALTERED FROM THE ORIGINAL...165
FIGURE 88 THE RESCALE PARAMETERS USED IN ERDAS IMAGINE TO CONVERT 11 BIT IKONOS TO 8 BIT. ......166
FIGURE 89 IKONOS PAN AND MS IMAGERY WITH A PCT AND SFIM RESOLUTION MERGE DERIVATIVE.........168
FIGURE 90 LAND USE DIGITISING SCHEMA (AFTER CAMPBELL 2002 P. 559)...................................................176
FIGURE 91 DIGITISING TOPOLOGICALLY INTACT POLYGONS AND NETWORKS FOR GIS ANALYSIS.................178
FIGURE 92 HYDROLOGY NETWORK IMAGE INTERPRETATION KEY. .................................................................180
FIGURE 93 THE DIGITISED HYDROLOGICAL AND COMMUNICATION NETWORKS..............................................182
FIGURE 94 COMMUNICATION NETWORK IMAGE INTERPRETATION KEY...........................................................183
FIGURE 95 THE DIGITISED LAND COVER PARCELS FROM THE IKONOS 2002 IMAGERY. ...................................186
FIGURE 96 LAND COVER PARCELS IMAGE INTERPRETATION KEY. ...................................................................187
FIGURE 97 THE UNSUPERVISED CLASSIFICATION ALGORITHM FOR GEOLOGY/SOIL CLASSIFICATION .............189
FIGURE 98 THE SOIL/GEOLOGY/URBAN CLASSIFICATION .................................................................................191
FIGURE 99 THE EFFECTS OF INCREASED RAINFALL. NOTE THE LEFT HAND IMAGES ARE ALL SUMMER IMAGES
WHERE THERE HAS BEEN NORMAL RAINFALL (HENCE THE IRRIGATION CHANNELS ARE EMPTY). .........192
FIGURE 100 DEM, ASPECT AND SLOPE DERIVED FROM CONTOUR DATA.........................................................194
FIGURE 101 GRAPH HIGHLIGHTING THE RELATIONSHIP BETWEEN PHOTOGRAMMETRIC DERIVED DTM
HEIGHTS FROM CORONA AND DGPS CHECK HEIGHTS (AFTER GALIATSATOS IN PREP).........................196
FIGURE 102 CHANGES IN CROP COVER AT SITE 339 AT DIFFERENT TIMES OF THE YEAR. ................................202
FIGURE 103 ‘SITE' 358 IN THE BASALT LANDSCAPE AND ITS SURFACE COLLECTION SUBSIDIARIES................206
FIGURE 104 DIGITISED CAIRNS AND WALLS IN THE BASALT AREA (SCALE 1:11,000). ....................................207
FIGURE 105 SCHEMATIC EXAMPLE OF THE A-SPATIAL AND SPATIAL LINKAGES FOR THE ARCHAEOLOGICAL
DATA. ......................................................................................................................................................208
FIGURE 106 EXAMPLE OF THE HIERARCHICAL STRUCTURING SYSTEM AT SITE 173. .......................................209
FIGURE 107 THE BASALT LANDSCAPE. MODERN DESTRUCTION IS CONTRASTED BETWEEN THE IKONOS AND
CORONA IMAGERY (IN ITALICS). POINTS A, B, C AND D IN THE IKONOS AND CORONA IMAGES REFER TO
THE LOCATIONS OF THE PHOTOGRAPHS (BUILDING FOUNDATIONS, OLIVE-PRESS, WALL, ROAD AND
BIRKA RESPECTIVELY). ...........................................................................................................................211
FIGURE 108 COMPARISON OF THE DIFFERENT SPATIAL RESOLUTIONS OF THE IMAGERY AND THEIR EFFECTS ON
IDENTIFICATION IN THE BASALT (SCALE 1:5,000). .................................................................................212
FIGURE 109 COMPARISON OF THE DIFFERENT SPATIAL RESOLUTIONS OF THE IMAGERY AND THEIR EFFECTS ON
IDENTIFICATION IN THE BASALT (SCALE 1:1,000). .................................................................................213
FIGURE 110 KERNEL FILTERS APPLIED IN THE BASALT ZONE (SCALE 1:5,000). IT SHOULD BE NOTED THAT AS
DEMONSTRATED IN FIGURE 108 AND FIGURE 109 THAT INTERPRETATIVE CLARITY IS RELATED TO
IMAGE SCALE. .........................................................................................................................................215
FIGURE 111 HINTERLAND ZONES AT THE SE OF KRAD AD-DÂSINIYA (DIGITISED WALLS ARE DERIVED FROM
IKONOS IMAGERY). THE PARALLEL WALLS REPRESENT ACCESS TRACKS. .............................................217
FIGURE 112 COMPARISON OF THE OBSERVABLE DETAIL BETWEEN CORONA AND IKONOS IMAGERY AND THEIR
RESULTANT DIGITISED INTERPRETATIONS (SCALE 1:2,000). THE LACK OF PARALLEL WALLS ON THE
CORONA COULD BE DUE TO ITS LOWER RESOLUTION. ALTERNATIVELY THE TRACKS COULD HAVE BEEN
CREATED TO ALLOW TRACTOR ACCESS. .................................................................................................218
FIGURE 113 COMPARISON OF THE CORONA AND IKONOS IMAGERY IN A BULLDOZED AREA OF THE BASALT
(SCALE 1:3000). EVEN THOUGH BULLDOZED THE IKONOS RESOLUTION MERGE IMAGE EXHIBITS SOME
OF THE RESIDUES SEEN ON THE CORONA IMAGE. ...................................................................................219
FIGURE 114 COMPARISON OF DIGITISING USING THE CORONA AND IKONOS IMAGERY AS BACKDROPS (SCALE
1:3,000)...................................................................................................................................................221
viii
FIGURE 115 THE EXTENT OF THE RUSSIAN AERIAL PHOTOGRAPHS AND CORONA IMAGERY IN THE NORTHERN
APPLICATION AREA. ................................................................................................................................223
FIGURE 116 COMPARISON OF THE RUSSIAN AERIAL PHOTOGRAPHY, CORONA AND IKONOS IMAGERY. ........224
FIGURE 117 COMPARISON OF THE RUSSIAN AERIAL PHOTOGRAPHY, IKONOS AND CORONA IMAGERY OVER AN
UNKNOWN SITE AND A COMPARATIVE EVALUATION OVER A DIFFERENT AREA USING THE MISSION 1108
CORONA..................................................................................................................................................225
FIGURE 118 THE MARL LANDSCAPE AND ‘SITES’. MODERN DEVELOPMENT IS CONTRASTED BETWEEN THE
IKONOS AND CORONA IMAGERY (IN ITALICS). .......................................................................................227
FIGURE 119 COMPARISON OF THE DIFFERENT SPATIAL RESOLUTIONS OF THE IMAGERY AND THEIR EFFECTS ON
IDENTIFICATION IN THE MARL (SCALE 1:12,500). SITES ARE HIGHLIGHTED IN IMAGE A.......................228
FIGURE 120 COMPARISON OF THE DIFFERENT SPATIAL RESOLUTIONS OF THE IMAGERY AND THEIR EFFECTS ON
IDENTIFICATION IN THE MARL (SCALE 1:5,000)......................................................................................229
FIGURE 121 DIFFERENT VISUALISATIONS OF RAW AND PROCESSED LANDSAT IMAGERY OVERLAID BY ‘SITES’.
YELLOW SITES ARE FLAT, BLACK SITES ARE TELLS................................................................................230
FIGURE 122 SPECTRAL PROFILE OF ON AND OFF-SITE POINT LOCATIONS DERIVED FROM IKONOS MS IMAGERY.
................................................................................................................................................................232
FIGURE 123 THE DISTRIBUTION OF ARCHAEOLOGICAL RESIDUES WHERE TRANSECT SPECTRAL PROFILES AND
SURFACE PROFILES WERE CONDUCTED...................................................................................................234
FIGURE 124 TRANSECT SPECTRAL PROFILE AND SURFACE PROFILE OF SITES 97-256. THE LINE AND
RECTANGLE IN THE LOCATION REPRESENT THE TRANSECT AND SURFACE PROFILE RESPECTIVELY. .....235
FIGURE 125 TRANSECT SPECTRAL PROFILE AND SURFACE PROFILE OF SITES 259-339. THE LINE AND
RECTANGLE IN THE LOCATION REPRESENT THE TRANSECT AND SURFACE PROFILE RESPECTIVELY. .....236
FIGURE 126 TRANSECT SPECTRAL PROFILE AND SURFACE PROFILE OF SITES 454-602. THE LINE AND
RECTANGLE IN THE LOCATION REPRESENT THE TRANSECT AND SURFACE PROFILE RESPECTIVELY. .....237
FIGURE 127 COMPARISON OF ZONAL AND NORMAL IKONOS MS IMAGERY (3,2,1 FCC), WITH A CLOSE UP OF
SITE 238. .................................................................................................................................................240
FIGURE 128 STATISTICAL DISTRIBUTION OF THE RED BAND AFTER 200M AVERAGING KERNEL.....................242
FIGURE 129 COMPARISON OF 200M KERNEL AVERAGE AND NORMAL IKONOS MS IMAGERY (3,2,1 FCC), WITH
A CLOSE UP OF CLUSTER OF SITES. ..........................................................................................................243
FIGURE 130 TRANSECT SPECTRAL PROFILES FOR 200M MEAN CORONA 1111 AND IKONOS MS IMAGERY AT
SITES 256, 279 AND 339. NOTE THE USE OF NEGATIVES FOR CORONA RESULTING IN A LOWER DN
VALUE FOR SITES.....................................................................................................................................244
FIGURE 131 COMPARISON OF THE 200 AND 400M RADIUS AVERAGING KERNEL ON AN IKONOS MS IMAGE. .247
FIGURE 132 COMPONENTS OF TASSELLED CAP AND PCA ANALYSIS IN THE MARL. .......................................249
FIGURE 133 CLASSIFICATION OF THE MARL AND THICK MARL ZONES (RESIDUES IN BLUE AND GREEN AND
SITES OUTLINED IN WHITE). ....................................................................................................................250
FIGURE 134 IKONOS MS 200M MEAN IMAGE SEGMENTATION (100 BLOCK SIZE, 60 SPECTRAL THRESHOLD AND
250 REGION SIZE) ....................................................................................................................................251
FIGURE 135 STRIP FIELDS OFF WADI AL-RABAYA ARE EASIER TO IDENTIFY IN THE WINTER (MISSION 1108) AS
OPPOSED TO THE SPRING (MISSION 1111) IMAGE. ALSO NOTE THE DIFFERENT ILLUMINATION OF THE
FOOTHILLS IN THE SE..............................................................................................................................254
FIGURE 136 PROBLEMS OF CLOUD AND CROP COVER. .....................................................................................255
FIGURE 137 COMPARISON OF CORONA AND IKONOS IN THE ALLUVIAL ZONE.................................................257
FIGURE 138 EVIDENCE FROM SITE 494 INCLUDING A TENUOUS LINEAR FEATURE IN THE FOUNDATIONS OF A
RESERVOIR. .............................................................................................................................................258
FIGURE 139 IKONOS IMAGE (VERTICAL VIEW) AND PHOTOGRAPH (HORIZONTAL VIEW) OF A TELL (SITE 191).
NOTE THE INCREASED AND DECREASED REFLECTANCE OF THE SE AND NW PART OF THE RAMPARTS
DUE TO DIFFERENTIAL ILLUMINATION FROM THE SUN AND SITE TOPOGRAPHY.....................................261
FIGURE 140 SOIL PROFILE (AFTER IRONS ET AL. 1989 P. 77). ...........................................................................264
FIGURE 141 SOIL DESCRIPTION MATRIX (AFTER ASRAR 1989 P. 75) AND ARCHAEOLOGICAL FIELD SOIL
DESCRIPTION CHART (AFTER MIDDLETON 2000). ..................................................................................267
FIGURE 142 BIDIRECTIONAL REFLECTANCE OF SOILS (AFTER ASRAR 1989 P. 89). .........................................269
FIGURE 143 SOIL REFLECTANCE FOR A SILT LOAM SOIL WITH VARYING MOISTURE CONTENT (AFTER ASRAR
1989 P. 90) ..............................................................................................................................................271
FIGURE 144 LOCATION OF SITES WHERE SOIL SAMPLES WERE TAKEN BY WILKINSON. ..................................277
FIGURE 145 LOCATIONS OF SOIL SAMPLES COLLECTED BY WILKINSON OVER SITE 259 AND RESULTS OF
PARTICLE SIZE ANALYSIS (GENERALISED TO THE BRITISH SOIL SURVEY CLASSIFICATION)..................278
ix
FIGURE 146 AVERAGED SPECTRAL CURVE AND SIMULATED IKONOS MS READINGS FROM THE SPECTRORADIOMETER READINGS OF THE SOIL SAMPLES FROM SITE 259. ............................................................279
FIGURE 147 ANALYSIS OF THE SOIL SAMPLES FROM SITE 259 AND THE DN VALUE FROM THE CORONA AND
IKONOS MS SATELLITE SENSORS FOR THE CORRESPONDING GPS POINT...............................................281
FIGURE 148 ANALYSIS OF THE SOIL SAMPLES FROM SITE 279 AND THE DN VALUE FROM THE CORONA AND
IKONOS MS SATELLITE SENSORS FOR THE CORRESPONDING GPS POINT...............................................282
FIGURE 149 LOCATIONS OF SOIL SAMPLES COLLECTED BY WILKINSON OVER SITE 279 AND RESULTS OF
PARTICLE SIZE ANALYSIS (GENERALISED TO THE BRITISH SOIL SURVEY CLASSIFICATION) ON THESE
SAMPLES..................................................................................................................................................283
FIGURE 150 AVERAGED SPECTRAL CURVE AND SIMULATED IKONOS MS READINGS FROM THE SPECTRORADIOMETER READINGS OF THE SOIL SAMPLES FROM SITE 259. ............................................................284
FIGURE 151 LOCATIONS OF SOIL SAMPLES COLLECTED BY WILKINSON OVER SITE 339 AND RESULTS OF
PARTICLE SIZE ANALYSIS (GENERALISED TO THE BRITISH SOIL SURVEY CLASSIFICATION) ON THESE
SAMPLES..................................................................................................................................................286
FIGURE 152 AVERAGED SPECTRAL CURVE AND SIMULATED IKONOS MS READINGS FROM THE SPECTRORADIOMETER READINGS OF THE SOIL SAMPLES FROM SITE 339. ............................................................287
FIGURE 153 ANALYSIS OF THE SOIL SAMPLES FROM SITE 339 AND THE DN VALUE FROM THE CORONA AND
IKONOS MS SATELLITE SENSORS FOR THE CORRESPONDING GPS POINT...............................................288
FIGURE 154 ANALYSIS OF THE SOIL SAMPLES FROM SITE 602 AND THE DN VALUE FROM THE CORONA AND
IKONOS MS SATELLITE SENSORS FOR THE CORRESPONDING GPS POINT...............................................289
FIGURE 155 LOCATIONS OF SOIL SAMPLES COLLECTED BY WILKINSON OVER SITE 602 AND RESULTS OF
PARTICLE SIZE ANALYSIS (GENERALISED TO THE BRITISH SOIL SURVEY CLASSIFICATION) ON THESE
SAMPLES..................................................................................................................................................290
FIGURE 156 AVERAGED SPECTRAL CURVE AND SIMULATED IKONOS MS READINGS FROM THE SPECTRORADIOMETER READINGS OF THE SOIL SAMPLES FROM SITE 602. ............................................................291
FIGURE 157 COMBINED SPECTRAL CURVE FROM SITES 259 AND 339 BY SAMPLE LOCATION..........................292
FIGURE 158 COMPARISON OF MARL SOIL CURVE (FROM SPECTRO-RADIOMETRY) AGAINST STANDARD
SAMPLES OF HEMATITE AND GOETHITE. .................................................................................................294
FIGURE 159 LOCATION OF SITES SAMPLED IN THE 2003 SEASON.....................................................................295
FIGURE 160 COMPARISON OF THE DIFFERENT COULTER 'RUNS' ON SELECTED 2001 AND 2003 SAMPLES AT
SITES 279 AND 339..................................................................................................................................297
FIGURE 161 MUD-BRICK STRUCTURES IN VARYING DEGREES OF COLLAPSE. ..................................................299
FIGURE 162 GRAPH OF PARTICLE SIZE DISTRIBUTIONS FROM MUD-BRICK AND SOIL. THE SAMPLE NUMBERS
RELATE TO TABLE 21..............................................................................................................................300
FIGURE 163 COMPARISON OF CURVE PROFILES FOR A NUMBER OF ATTRIBUTES FOR SITE 339. ......................303
FIGURE 164 VARIATIONS IN ORGANIC MATTER CONTENT ACROSS SITES 279, 259, 602 AND 339. ..................304
FIGURE 165 IMAGE INTERPRETATION KEY FOR THE SYRIAN TOPOGRAPHIC MAP. ...........................................311
FIGURE 166 ISOMETRIC VISUALISATION (WITH 15 TIME EXAGGERATION) OF DIFFERENT IMAGERY DRAPED
OVER A TERRAIN MODEL.........................................................................................................................314
FIGURE 167 PROBLEMS OF SITE LOCATION (SITE 308).....................................................................................315
FIGURE 168 COMPARISON OF CORRECTED AND UNCORRECTED IKONOS IMAGERY WITH GPS OVERLAY .......316
FIGURE 169 COMPARISON OF OCTOBER LANDSAT SCENES FROM 1987 AND 2000. NOTE THE INCREASE IN
VEGETATION IN THE 2000 SCENE (DENOTED BY THE RED COLOURATION) AND SHRINKAGE OF LAKE
QATINA DUE TO IRRIGATION...................................................................................................................321
FIGURE 170 DEPTHS OF THE WATER TABLE IN TEL HADYA (1983 – 1997 (RODRIGUEZ ET AL. 1999 P. 9)).....322
FIGURE 171 URBAN EXPANSION IN THE MARL ZONE........................................................................................323
FIGURE 172 SEASONAL EFFECTS ON THE SATELLITE IMAGERY........................................................................324
FIGURE 173 COMPARISON OF CORONA MISSIONS OF THE SAME AREAS IN THE BASALT AND MARL ZONES. NOTE
THE MUCH BETTER RELATIVE IMAGE QUALITY OF THE WINTER SCENE..................................................325
FIGURE 174 MONITORING OF TELL (SITE 173) ON THE LAKE EDGE..................................................................326
FIGURE 175 COMPARISON OF SITE EXTENTS OVER TIME FOR SITES 251, 319, 339, 471 AND 477....................327
FIGURE 176 RESIDUE IMAGE INTERPRETATION KEY: POSITIVE FEATURES IN THE MARL. ................................339
FIGURE 177 RESIDUE IMAGE INTERPRETATION KEY: NEGATIVE AND MASKING FEATURES IN THE MARL. ......340
FIGURE 178 RESIDUE IMAGE INTERPRETATION KEY: BASALT ZONE. ...............................................................341
FIGURE 179 REVISED SEQUENCE FOR THE INCORPORATION OF SATELLITE IMAGERY INTO AN
ARCHAEOLOGICAL LANDSCAPE PROJECT. ..............................................................................................357
FIGURE 180 SCHEMATIC OF AN ARCHAEOLOGICAL DATA MODEL (AFTER MARTIN 1991 P. 55) . NOTE ALL
ARROWS PERFORM SOME TRANSFORMATION. ........................................................................................369
x
FIGURE 181 HERMENEUTIC LOOP OF ARCHAEOLOGICAL ENQUIRY..................................................................370
FIGURE 182 CONCEPTUAL MODELLING SCHEMA. ............................................................................................375
FIGURE 183 PROJECT DIRECTORY STRUCTURE................................................................................................379
FIGURE 184 HARDWARE SCHEMA FOR THE PROJECT DATA MODEL. ................................................................381
FIGURE 185 SCHEMATIC DATA FLOWLINE........................................................................................................382
FIGURE 186 DATABASE SCHEMA......................................................................................................................383
FIGURE 187 PDADB.MDB ENTITY RELATIONSHIP MODEL. . NOTE THE FIELDS IN BOLD TEXT ARE THE PRIMARY
(OR COMPOUND PRIMARY) KEYS. ...........................................................................................................385
FIGURE 188 DSKTOPDB.MDB ENTITY RELATIONSHIP MODEL. NOTE THE FIELDS IN BOLD TEXT ARE THE
PRIMARY (OR COMPOUND PRIMARY) KEYS.............................................................................................386
FIGURE 189 THE MAIN SWITCHBOARD FOR FORMS.MDB. ................................................................................386
FIGURE 190 FULLY INTEGRATED DATABASE ENTITY RELATIONSHIP MODEL. .................................................387
FIGURE 191 THE METADATA SWITCHBOARD FOR FORMS.MDB........................................................................388
FIGURE 192 AN EXAMPLE OF A SITE FORM (AFTER CLICKING ADD/VIEW DATA ON THE MAIN SWITCHBOARD).
NOTE THE TABS TO ACCESS OTHER DATA ABOUT THE SITE....................................................................388
FIGURE 193 COMPARISON OF VECTOR VERSUS HIGH AND MEDIUM RESOLUTION RASTER REPRESENTATIONS
(COURTESY NERC, GETMAPPING AND ENGLISH NATURE). .................................................................389
FIGURE 194 DATA SCHEMA FOR MULTI SCALAR DATA INTEGRATION .............................................................394
FIGURE 195 EXAMPLE OF DUBLIN CORE RESOURCE DISCOVERY METADATA MAINTAINED IN TABLE METAT.
................................................................................................................................................................395
FIGURE 196 EXAMPLE OF TABLE LEVEL METADATA STORED IN TABLE METATDB........................................396
FIGURE 197 EXAMPLE OF BIBLIOGRAPHIC METADATA HELD IN TABLE DOCUT. .............................................397
FIGURE 198 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 97 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................403
FIGURE 199 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 218 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................404
FIGURE 200 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 221 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................406
FIGURE 201 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 238 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................407
FIGURE 202 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 271 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................409
FIGURE 203 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 279 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................410
FIGURE 204 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 339 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................412
FIGURE 205 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 478 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................413
FIGURE 206 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 496 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................415
FIGURE 207 LOCATIONS OF SOIL SAMPLES COLLECTED OVER SITE 508 AND RESULTS OF PARTICLE SIZE
ANALYSIS ON THESE SAMPLES. ...............................................................................................................416
xi
LIST OF TABLES
Number
Page
TABLE 1 RADIOMETRIC ENHANCEMENT TECHNIQUES .......................................................................................58
TABLE 2 SPECTRAL ENHANCEMENT TECHNIQUES ..............................................................................................59
TABLE 3 EXAMPLES OF DIFFERENT KERNEL FILTERS .........................................................................................62
TABLE 4 FIELDWORK SUMMARY. .....................................................................................................................120
TABLE 5 LANDSAT TM TECHNICAL SPECIFICATION (AFTER TOWNSHEND ET AL. 1988)..................................126
TABLE 6 TM AND ETM+ SPECTRAL BAND WIDTHS AND SPATIAL RESOLUTION..............................................127
TABLE 7 LANDSAT TM INTERPRETATIVE SENSOR CHARACTERISTICS (AFTER SABINS 1997 P. 74; CAMPBELL
2002 P. 173). ...........................................................................................................................................129
TABLE 8 USER DEFINABLE PROCESSING PARAMETERS FOR LANDSAT IMAGERY.............................................130
TABLE 9 IKONOS TECHNICAL SPECIFICATION (AFTER SPACE IMAGING 2003 P. 1)...........................................132
TABLE 10 IKONOS PRODUCT LEVELS (AFTER SPACE IMAGING 2003 P. 2: CE90 = CIRCULAR ERROR AT 90%
CONFIDENCE, RMS = ROOT MEAN SQUARE ERROR AND NMAS = US NATIONAL MAP ACCURACY
STANDARDS). NOTE THE COSTINGS ARE BASED UPON THE MAY 2003 PRICING LEVELS AND THE PRICING
FOR GEO REPRESENTS BESPOKE AND ARCHIVE IMAGERY. .....................................................................132
TABLE 11 USER DEFINABLE PROCESSING PARAMETERS FOR IKONOS IMAGERY..............................................134
TABLE 12 CORONA KEY-HOLE CAMERA MISSION CHARACTERISTICS (AFTER GALIATSATOS IN PREP). .........135
TABLE 13 SUMMARY OF HISTORIC AND MODERN MAPPING USED IN THE RESEARCH. .....................................138
TABLE 14 LAYERS AND SUMMARY INFORMATION FOR THE DIGITISING OF THE MAP-BASE .............................139
TABLE 15 CORONA ACQUISITION SUMMARY....................................................................................................153
TABLE 16 CORONA CO-REGISTRATION RMSE ACCURACY..............................................................................164
TABLE 17 COMBINED USGS AND SHR LAND USE CODES. ..............................................................................174
TABLE 18 PCA (SOUTHERN MARL ONLY) AND TASSELLED CAP COMPONENT MATRIX AND PERCENTAGE
VARIATION FOR IKONOS MS (THE TASSELED CAP STATISTICS ARE AFTER HORNE 2003)......................248
TABLE 19 COMPARISON OF ON AND OFF SITE SOIL COLOURS WHEN WET AND DRY. NOTE THIS REQUIRES THE
COLOUR CHIPS DESCRIBED BY MIDDLETON (2000)................................................................................270
TABLE 20 BRITISH SOIL SURVEY PARTICLE SIZE CLASSIFICATION ..................................................................275
TABLE 21 PARTICLE SIZE ANALYSIS OF MUD BRICK AND SOIL. NON SHR SAMPLES ARE TAKEN FROM
SAUVAGE (1998 P. 19, TABLES 2 AND 3)................................................................................................298
TABLE 22 PIVOT TABLE DISPLAYING A SUMMARY OF THE AVERAGE PERCENTAGE CHANGE OF SOIL FRACTIONS
BETWEEN OFF-SITE AND SITE. .................................................................................................................301
TABLE 23 VARIATIONS IN SOIL TEXTURE CALCULATED FROM THE AVERAGE OF THE COULTER RESULTS FOR
EACH LOCATION......................................................................................................................................302
TABLE 24 SUMMARY OF SITES IN THE WHOLE APPLICATION AREA..................................................................332
TABLE 25 SUMMARY OF EVIDENCE FOR INDETERMINATE (POTENTIAL) SITES IN THE SOUTHERN MARL.........332
TABLE 26 SUMMARY OF EVIDENCE FOR NON-SITES IN THE SOUTHERN MARL. ................................................333
TABLE 27 SUMMARY OF EVIDENCE FOR TELL SITES IN THE SOUTHERN MARL.................................................333
TABLE 28 SUMMARY OF EVIDENCE FOR SCATTER SITES IN THE SOUTHERN MARL. .........................................334
TABLE 29 SUMMARY OF RESULTS IN THE SOUTHERN MARL BY METHOD. .......................................................335
TABLE 30 SENSOR SUMMARY BY ZONE. ...........................................................................................................349
TABLE 31 TYPES OF DIGITAL RESOURCE .........................................................................................................378
TABLE 32 ACTIVE AND ARCHIVE FILE FORMATS ..............................................................................................378
TABLE 33 DEFINITION OF THE FEATURE DATA SETS AND THEIR CLASSES WITHIN SHR_GEOBASE.MDB. ......390
TABLE 34 GENUS LOOKUP TABLE WITH ‘VALUE ADDED’ INFORMATION (LGENUS IS THE PRIMARY KEY):
COURTESY OF DR. PHILIP PIPER. ............................................................................................................391
xii
ACKNOWLEDGMENTS
Any work such as this is impossible without the support of a whole range of people and
organisations. I would like to thank them for their support throughout the duration of this
research. Initially I’d like to thank Mr. Fraser Brown for helping me with my first faltering
steps on the road to archaeology. Dr. Philip Piper and Mr. Tim Allen were great companions
on this road and taught me much. To my wife Maria: words are not enough to express my
gratitude, all I can do is provide a paltry thanks. At least you’ll be happy that my sentence
structure has improved. To my supervisors Drs. Graham Philip and Danny Donoghue: this
work would have been impossible without their unstinting support and friendship (even
during my ranting phases). Furthermore, thanks to the students and staff of the Departments
of Archaeology and Geography at the University of Durham (particularly Dr. Kay McManus,
Dr. Paul Newson, Nikolaous Galiatsatos and Terry Harrison).
I am indebted to the Council for British Research in the Levant and the Department of
Archaeology, University of Durham for their financial, logistical and moral support.
Furthermore, the advice and companionship of many Middle Eastern archaeologists,
particularly Drs. Bill Finlayson, Alex Wasse and Amr al-Azm, was indispensable. I am also
indebted for all the hard work by those at the Director General of Antiquities and Museums
(Syria) particularly Dr. Michel Maqudassi, Dr. Mamoun Abdulkareem and Eng. Farid Jabour.
Further thanks are due to Dr. Abdulkareem for providing the Russian aerial photographs and
his permission to use them in this research. I would especially like to thank Eng. Maryam
Bshesh not only for her assistance during the fieldwork but her great friendship. ICARDA
very kindly provided articles on Syrian soils and agriculture and data on regional rainfall. Dr.
Keith Wilkinson of King Alfred’s College, Winchester also provided invaluable help in the
interpretation of Syrian soils. Keith Challis of York Archaeological Trust provided me with a
wealth of information on Corona. Dr. Bob Bewley, Peter Horne, Dave MacLeod and all
others at the Aerial division of English Heritage who provide me with support and data. Dr.
Meredith Williams (Department of Geomatics, University of Newcastle-upon-Tyne)
provided me with invaluable information concerning the nature of the projection of the
Syrian mapping. Valerie Hood and Mary Foussier at EuriSy who provided me with the
opportunity of seeing things with my eyes wide open. Final thanks go to NERC (grant award
GT0499TS53): without whom none of this research would have been possible. Dr. Stuart
xiii
White’s support was particularly welcome when there were problems in the acquisition of the
Ikonos satellite imagery.
I would like to express my appreciation to all the local people of the Homs region for their
cooperation and hospitality. Particular thanks go to the Bshesh family, Hannan, Hassan,
Salam Wakim, Abdullah Khan and the irrepressible Mimo. Just so you all know I’ll always be
happy to be Homsy.
This thesis is dedicated to the people and the cultural heritage of Syria, my friends and my
family. Salam.
This thesis was prepared in Microsoft Word (XP) with bibliographic referencing supplied by
Endnote (v. 5.0.2). Supporting information was prepared in Microsoft Access (XP),
Microsoft Excel (XP), Adobe Illustrator (v. 9), Adobe Photoshop (v. 6.5), ArcGIS 8.3 and
AutoCAD MAP 2000.
The copyright of this thesis text and other original content rests solely with the author. No
quotation from it should be published without the authors’ prior written consent and
information derived from it should be duly acknowledged.
xiv
GLOSSARY
Active Sensor
ADMS
ADS
AHDS
AHRB
AIS
Albedo
AP
ArcCatalog
ArcGIS
ArcMAP
ArcPAD
Attitude
AutoCAD
AVHRR
Band
Breaklines
CAD
CAM
Cluster
Co-register
Corona
CRM
DEM
DGAM
DGPS
DSM
DTM
Dublin Core
Dynamic Range
EH
Sensor that supplies its own energy source
Archaeological Data Management System (see AIS)
Archaeological Data Service
Arts and Humanities Data Service
Arts and Humanities Research Board
Archaeological Information System (see ADMS)
The percentage of incoming radiation that is reflected by a natural
surface such as the ground, ice, snow, water, clouds, or particulates in
the atmosphere.
Aerial Photography
Data management component of ArcGIS software suite
Scalable system of software for geographic data creation,
management, integration, analysis and dissemination; framework for
articulating inter-site, intra-site and landscape relationships using both
spatial and attribute data. Released by ESRI.
Visualisation component of ArcGIS software suite.
Mobile GIS application
The angular orientation of a remote sensing system with respect to a
geographical reference system. The orientation of the sensor along
with information about the accuracy and precision with which this
orientation is known.
A Computer Aided Design software package.
Advanced Very High Resolution Radiometer
A portion of the electromagnetic spectrum recorded by a sensor
Lines that delineate a break of slope.
Computer Aided Design
Computer Aided Mapping
A homogeneous group of units based upon how ‘alike’ an object is
top its neighbours. ‘Likeness’ is usually determined by the association,
similarity, or distance among the measurement patterns associated
with the units.
Placing two or more images into the same co-ordinate scheme
American 'Spy' satellite program. The results are now available
commercially
Cultural Resource Management
Digital Elevation Model. A raster method of creating a DTM
Directorate General of Antiquities and Museums
Differential Global Positioning Systems
Digital Surface Model
Digital Terrain Model. A computerised representation of the Earth’s
surface
MetaData standard
The range between the maximum and minimum amount of input
radiant energy that an instrument can measure.
English Heritage
xv
ENVI
Ephemeris
Image Processing Software
A table of predicted satellite orbital locations for specific time
intervals. The ephemeris data help to characterize the conditions
under which remotely sensed data are collected and are commonly
used to correct the sensor data prior to analysis.
Erdas IMAGINE
Image Processing Software
EROS
Earth Resources Observations Systems: The EROS program was
established in the early 1970s, under the Department of the Interior's
U.S. Geological Survey, to receive, process, and distribute data from
United States Landsat satellite sensors and from airborne sensors
ERSC
Environmental Remote Sensing Centre
ES
Electromagnetic Spectrum
ESRI
Environmental Systems Research Institute. Purveyors of GIS
software systems ArcVIEW, Arc/Info and ArcGIS.
ETM+
Enhanced Thematic Mapper +: a sensor on the Landsat program
Extrapolate
To infer (values of a variable in an unobserved interval) from values
within an already observed interval.
Fiducial Marks
A set of four marks located in the corners or edge-centred, or both,
of a photographic image. These marks are exposed within the camera
onto the original film and are used to define the frame of reference
for spatial measurements on aerial photographs.
Fusion
Combining the spatial or spectral characteristics of two or more
sensors
GCP
Ground Control Point
Geodatabase
A GIS data format (designed by ESRI) that stores both spatial and
attribute data within a single RDBMS architecture
Georegistered
An image that has been geographically referenced or rectified to an
Earth model, usually to a map projection. Sometimes referred to as
geocoded or geometric registration.
GeoTIFF
A TIFF file with geo-referencing information contained within the
header.
GIS
Geographical Information System. A mechanism of linking
geographic entities to associated database information
GPS
Global Positioning Systems
Ground Observation Testing desk based hypothesis by field observation
Ground Truthing
Testing desk based hypothesis by field observation (superseded by
Ground Observation)
GRS
Ground Receiving Station
Handspring
PDA running the PALM Operating System
Heads Up Digitising A method of creating a vector drawing using a raster backdrop rather
than a digitising tablet
HRS
Homs Regional Survey
HSSF
Hellenic State Scholarship Fund
Hyperspectral
Imagery with multiple bands. Normally greater than 20
IFA
Institute of Field Archaeologists
IFOV
Instantaneous Field-of-View
Ikonos
High resolution commercial satellite program managed by
SpaceImaging
xvi
Interpolate
IP
iPAQ
Isopleth
JADIS
KH
KVR
Landsat
Lookup Table
MetaData
MIR
MSS
Multispectral
Nadir
NASA
NDVI
NERC
NIR
Orbit
Orbital period
Ortho-correction
Panchromatic
Passive Sensor
PCI
PCMCIA card
PDA
Pixel
Primary Key
Radiance
To insert a value between known values by using a procedure or
algorithm specifically related to the known values.
Image Processing
PDA running Pocket PC Operating System
A line on a map connecting points at which a given variable has a
specified constant value.
Jordan Archaeological Database
Key Hole
Russian 'Spy' satellite program. The results are now available
commercially
Commercial satellite program
Data dictionary used in RDBMS
Data ABOUT data. For example creation date, creator, copyright etc.
Middle Infra Red
MultiSpectral Scanner: a sensor on the Landsat program
Imagery with multiple bands. Normally less than 20
Point on the ground vertically beneath the centre of a remote sensing
platform.
National Aeronautics Space Administration
Normalized Difference Vegetation Index: NDVI is computed by
calculating the ratio of the VI (vegetation index, i.e., the difference
between Channel 2 and 1) and the sum of Channels 2 and 1. Thus
NDVI = (channel 2 - channel 1) / (channel 2 + channel 1).
Natural Environment Research Council
Near Infra Red
The path of a body acted upon by the force of gravity.
The time it takes a satellite to complete one revolution (orbit) around
the Earth. The orbital period of Landsat 7 is about 1.5 hours.
Correction applied to satellite imagery to account for terrain-induced
distortion.
Sensor sensitive to all or most of the visible spectrum, between 0.4
and 0.7 micrometers.
Sensor that records reflected or emitted radiation
Image Processing Software
Personal Computer Memory Card International Association card.
Credit card hardware used in laptops to add functionality (Modem,
GPS, hard disk, etc.)
Personal Digital Assistant. Handheld computer. For examples see
Handspring and iPAQ
An abbreviation of picture element. The minimum size area on the
ground detectable by a remote sensing device. The size varies
depending on the type of sensor.
Mechanism to uniquely identify records in a RDBMS table. A primary
key can be based on a single field or multiple fields (referred to as
compound primary key)
Measure of the energy radiated by an object. In general, radiance is a
function of viewing angle and spectral wavelength and is expressed as
energy per solid angle.
xvii
Raster
RDBMS
Reflectance
RMSE
SA
SHR
SMR
Spectral signature
Spin-2
SWIR
Theme
TIFF
TIN
TM
Unique Identifier
USGS
UTM
Vector
WGS-84
The generic term for a grid based spatial data storage mechanism. E.g.
GeoTIFF
Relational Database Management System
Reflectance is the fraction of the total radiant flux incident upon a
surface that is reflected and that varies according to the wavelength
distribution of the incident radiation.
Root Mean Square Error: The RMSE statistic is used to describe
accuracy encompassing both random and systematic errors. The
square of the difference between a true test point and an interpolated
test point divided by the total number of test points.
Selective Availability: The reduction in the accuracy of GPS
controlled by the American Military
Settlement and Landscape Development in the Homs Region, Syria
Sites and Monuments Record
The unique spectral response that a feature displays over a range of
wavelengths
Commercial satellite program
Short Wave Infra Red
Group of related geographical data
Tagged Image File Format, a common raster file format also referred
to as TIF
Triangular Irregular Network. A vector method of creating a DTM
Thematic Mapper: a sensor on the Landsat program
GIS link between Spatial and A-spatial databases
United States Geological Survey
Universal Tranverse Mercator. Worldwide projection system
The generic term for a point, line and polygon spatial data storage
mechanism. For example AutoCAD .dwg or ESRI .shp files.
World Geodetic System 1984: GPS reference ellipsoid
xviii
SECTION 1 INTRODUCTION
1
CHAPTER 1 RESEARCH INTRODUCTION AND SUMMARY
1.1 Introduction
This PhD research project originated from fieldwork undertaken in 1996 by Dr. Graham
Philip (Department of Archaeology, University of Durham) and Mr. Stephen Holmes,
(Department of Archaeology, University of Edinburgh) a doctoral student studying ‘The
application of Remote Sensing and GIS to the Location of Prehistoric Settlement in part of
Anatolia’. At the time, Dr. Philip was evaluating a potential project in an area of known
archaeological and historical significance around Homs, Syria. Although a few large sites had
been excavated very little was known about the chronological and spatial distribution of less
obvious settlements and other archaeological residues across the landscape. In the semi-arid
environment of the application area Dr. Philip recognised that satellite applications had the
potential to, amongst other things, pinpoint areas of archaeological activity. After
conversations with Dr. Daniel Donoghue (Department of Geography, University of
Durham), who informed Dr. Phillip of the declassified Corona programme and the proposed
Ikonos satellite, a research proposal was submitted to the Natural Environmental Research
Council (NERC) to study the archaeological applications of high resolution satellite imagery
within this semi-arid environment.
This application was successful and secured a post-graduate student, Anthony Beck (NERC
grant award GT0499TS53), to undertake a three year research PhD. NERC also generously
agreed to cover the acquisition costs of Ikonos imagery within the grant application. In
addition another post-graduate student, Nikolaous Galiatsatos, approached Dr. Donoghue
with an application to undertake research into satellite applications and archaeology. This
student had already secured funding towards a doctoral research programme through the
Hellenic State Scholarship Foundation (speciality T1327.06, contract 368 (NG)). Galiatsatos’
thesis is provisionally entitled ‘Assessment of satellite imagery in landscape archaeology
applications: case study from Orontes valley, Syria’. Hence, two PhD students were engaged
to research different aspects of satellite imagery applications to archaeology, both based in
the University of Durham: Beck primarily affiliated to the Department of Archaeology and
Galiatsatos primarily affiliated to the Department of Geography.
2
1.2 Thesis structure
This thesis has been structured to incorporate stand alone sections. This approach has been
used due to the different conceptual and methodological approaches employed by landscape
archaeologists and remote sensing specialists. Therefore, archaeologists who are familiar with
the concepts of archaeological landscapes but are unfamiliar with remote sensing can review
the chapters appropriate to their research. Broadly the thesis has been sub-divided into three
sections:
1. Introduction (Chapters 1-3): These chapters introduce physical, methodological
and interpretative processes in remote sensing and landscape archaeology.
2. Methodology and Analysis (Chapters 4-9): These chapters outline the physical
environment, methodology and analysis conducted on the satellite imagery.
Chapter 8 specifically examines what physical variations make archaeological
residues visible in satellite imagery.
3. Summary, recommendations and conclusions (Chapters 10-11): These chapters
evaluate the methodology in relation to other archaeological techniques and
provide recommendations for future research.
Each section addresses archaeological and remote sensing aspects of the data. This
culminates in a breakdown of the cost benefit analysis of the data sets and recommendations
for future projects that may want to adapt the methodologies employed. Further pertinent,
but not directly relevant, material is contained in the appendix. Overall this is a science based
programme of research which is evaluating new archaeological data sources. Hence, the
outcomes are predominantly methodological and although the imagery is interpreted for its
archaeological content there is little archaeological interpretation.
1.3 Research summary
The use of satellite imagery for archaeological investigations seems to be increasing. The
spatial, spectral, radiometric and temporal resolutions of satellite imagery have developed to
such an extent that satellite imagery shares many of the physical properties of aerial imagery
at a vastly reduced acquisition cost. However, it is not clear that the full capabilities of the
data are being exploited. This research will attempt to address some of the potentials and
methodological issues surrounding the use of satellite imagery within archaeology with
particular reference to the semi-arid landscape around Homs, Syria. Corona and Ikonos high
3
resolution satellite image data have been evaluated for a range of archaeological activities
including cultural resource management, data visualisation and thematic data extraction. The
main thrust of the research has been the application of these data to site prospection and
their impact on landscape survey. A suite of medium resolution Landsat imagery was also
incorporated into the research in order to evaluate the effect of larger spatial and spectral
resolution on the information that can be obtained.
1.3.1 Traditional approaches to landscape survey
Many landscape surveys initially start with a Desk Based Assessment (DBA). A DBA will
normally look at data derived from the Sites and Monuments Record (SMR, or its equivalent),
aerial photography, geophysics and historical document searches. This information is
normally synthesised into a document that outlines the archaeological potential within the
application area and provides a framework for subsequent fieldwork programmes and
analysis. This archaeological information can be augmented with additional data sets. These
secondary data sets tend to include topographic and environmental data (such as soils,
hydrology and elevation). However, there are many countries where the archaeological
resource is not managed in this way.
Syria is one example of a country, typical of many parts of the developing world, where there
is no systematic regional database of archaeological remains, no ready access to topographic
mapping at scales greater than 1:25,000, nor is aerial-photography available (Donoghue et al.
2000; Philip et al. 2002a). Thematic data (for example soil and land use maps) in this region
are normally highly generalised. Furthermore, due to military restrictions, the national map
projection remains secret and the opportunity to conduct any aerial photographic survey is
limited.
When faced with this situation most projects would embark on an archaeological survey. The
tradition in Mediterranean environments is to employ a systematic surface survey strategy (Cherry
et al. 1991; Alcock et al. 1994; Bintliff 2000; Banning 2002). These techniques are extremely
well developed, but have the disadvantage of being expensive and time consuming. Given the
consolidation of funding within a few large projects, a re-evaluation of satellite applications as
a complementary survey strategy is timely.
4
Figure 1 Examples of the archaeological residues in the application
area.
5
1.3.2 Settlement and landscape development in the Homs region
In this data poor environment the project brief is to improve the understanding of Settlement
and Landscape Development in the Homs Region (the SHR project). The project team is required to
rapidly evaluate the archaeological resource and to create a variety of integrated thematic data
sets to provide an interpretative context. The vast majority of these data sets were and still
are unavailable from traditional sources. Satellite imagery with different resolutions has
therefore been evaluated to determine their efficacy in bridging this significant data gap. This
imagery has different spatial, spectral, radiometric and temporal resolution. The research
postulates that individual satellite imagery or the integration of imagery with different
resolving characteristics can significantly enhance the understanding of the archaeological
resource.
The SHR project application area is particularly suitable for this evaluation as it contains
three distinct environmental zones: basaltic plateau, alluvium and marl. Each of these zones
employs distinctive urban and rural land management strategies. Furthermore, the
archaeological record has been subjected to different formation and de-formation events in
each zone. In the basaltic zone the archaeology takes the form of abandoned villages, isolated
villas, roads, tombs and agricultural systems all preserved as a palimpsest of stone walls and
concentrations of rubble. By contrast, the marl zone contains a few mounded tell sites and
many ploughed out artefact scatters (see Figure 1).
1.3.3 Introduction to the data sources
Landsat has already proved useful in providing landscape environmental data, and has been
applied to map soil, vegetation and geology for archaeological purposes (Cox 1992; Gaffney
et al. 1996; Ostir et al. 1999; Rothaus and De Morett 1999). However, Landsat’s 30m cell
resolution limits its application for site prospection (Allan and Richards 1983).
The last decade has, however, seen the declassification of high-resolution (sub 2-3 metre)
panchromatic military photography such as the American Corona and Russian KVR missions
(Comfort 1997; Day et al. 1998; Fowler 2001; Campbell 2002). Furthermore, this period has
seen the deregulation of high spatial resolution commercial sensors that have recently
resulted in the Ikonos and Quickbird satellites which provide geo-corrected panchromatic (at
1 and 0.61m respectively) and 4 band multispectral imagery (at 4 and 2.44m respectively).
6
Figure 2 Comparison of the 3 primary satellite sensors used in the
research.
This research uses a combination of Corona, Ikonos and Landsat imagery (see Figure 2),
although other data sources may be integrated in the future for evaluation purposes. All of
these data sets are well documented (Colwell et al. 1983; Day et al. 1998; Space Imaging 2003;
USGS 2003e).
7
1.3.4 Remote sensing
Remote sensing systems rely on collecting energy that is either emitted by or reflected from
an object under study. The energy source for all passive remote sensing of the Earth is the
Sun (even for emitted radiation where the Sun’s energy is transformed and re-emitted).
Active sensors, such as RADAR and LiDAR, provide their own (artificial) source of energy.
Figure 3 Overview of remotely sensed image acquisition (after
Campbell 2002).
The Sun, with a temperature of 6000 Kelvin, emits a spectrum of radiation which is
transmitted through space without undergoing major attenuation (see Figure 3). This
radiation then passes through the Earth’s atmosphere prior to interaction with an object.
After interaction energy travels through another portion of the Earth’s atmosphere to reach
the sensor. During this process the energy may be altered in intensity and wavelength
(attenuated) by particles and gases within the atmosphere itself (Nunnally 1973).
8
Every object with a temperature above absolute zero (0˚Kelvin) emits electromagnetic
radiation. Objects also reflect radiation that has been emitted by other objects. An object’s
ability to reflect radiation is dependent upon its physical, chemical and surface characteristics.
The radiation that is not reflected by the object is transmitted or absorbed and then reradiated as
thermal (emitted) energy. Remote sensing systems record this reflected and/or emitted
radiation.
The underlying premise of remote sensing is that interpreters can extract information about
objects and features on the Earth’s surface by studying the radiation measured by a sensor
system. Interpreting remotely sensed imagery depends upon being able to differentiate
features of interest from reflected or emitted energy on the basis of variations in the signal
strength and spectral response. The basis for interpretation of multispectral images is the spectral
signature (see Figure 22) i.e. the unique spectral response that a feature displays over a range of
wavelengths (Campbell 2002 p. 15). However, every sensor has limitations: most limitations
revolve around the resolving characteristics, or resolution, of the sensor. The most important
axes of resolution are spatial, spectral, radiometric and temporal. For sensor platforms there
is normally a negative relationship between a sensor’s spatial and spectral resolution (i.e. the
higher the spatial resolution the lower the spectral resolution and vice-versa).
1.3.5 Image interpretation
Once data has been collected it is then interpreted using a combination of manual and digital
procedures. Figure 4 suggests a schematic for the flowline of image processing.
Pre-processing is where the data is prepared for subsequent analysis. Analogue data (non-digital
information such as photographs) will require digitising using an appropriate scanning
methodology. Radiometric pre-processing is designed to compensate for errors introduced by
defective sensors, atmospheric attenuation, system noise, and variations in illumination and
scan angle. Geo-metric pre-processing is designed to place the raw image in a known
geographic co-ordinate system and projection. This is to compensate for errors introduced by
variations in orientation of the platform and the curvature of the Earth.
Feature Extraction is the process of reducing data complexity by selecting only the appropriate
data sources for the problem in hand. Therefore, the degree of feature extraction is
dependent upon the interpreters’ knowledge of which bands are appropriate for the objects
9
under study. The reduction in data complexity reduces the number of variables that can be
examined and hence the time and costs of interpretation (Campbell 2002 p. 119). It is not
essential to discard any information at this stage.
Figure 4 Schematic sequence for digital analysis (after Campbell
2002).
For manual interpretation the image can undergo a variety of enhancements that allow the
interpreter to extract more information from the image visually. These enhancement
techniques can drastically disrupt the structure of the original or post-processed data; hence,
enhanced images are rarely, if ever, used for any statistical classification.
Decision and Classification by either manual or computerised methods is the assignment of
specific information classes based upon their visual or numerical appearance in the image.
10
Computerised classification is the quantitative extraction of information using one of two
main techniques: supervised and unsupervised classification techniques.
Each classification should have an associated level of confidence. It is beneficial to subdivide
this confidence rating into three areas: Detection is the determination of the presence of a feature
(note that the concomitant absence of a feature does not necessarily mean that it is absent),
for example, vegetation. Recognition is the further characterisation of the feature into a class,
category or genus, for example, pulse crop. Identification places the feature into a specific class,
category or sub-genus, for example, pea (Campbell 2002 p. 124). This hierarchically
structured knowledge base is an intrinsic element of many classification programmes, for
example, the CORINE international land cover map derived from Landsat (Gerard 2002).
The final stage of all processing is that of accuracy assessment. This should occur for all image
interpretations and normally requires a continuing programme of ground observation.
Accuracy assessment also includes post-classification evaluation. Both supervised and
unsupervised classifications produce confusion matrices which statistically define the degree of
uncertainty when placing a pixel into a category. Confused classes should be highlighted
during this process.
The final results of this procedure is a series of classified, synthetic and attribute information
which is commonly presented within maps or reports. The preparation, articulation and
analysis of these final data sets are augmented by the analytical functionality of Geographical
Information Systems (GIS).
1.3.6 Archaeological data
Archaeological data are some of the most complex and diverse information sets studied
within any discipline (Ryan 1991). Data of archaeological relevance encompasses virtually all
areas of nature and culture. Furthermore, archaeological data is inherently spatial: all past
actions happened somewhere in space (Aldenderfer 1998).
The increased use of computerised storage and analysis systems such as Relational DataBase
Management Systems (RDBMS), Computer Aided Design (CAD), GIS, Image Processing
(IP) systems and more traditional ‘office’ applications means that the vast majority of
archaeological data is now stored in a digital format. A survey of users’ needs (Condron et al.
1999) highlights the increased production of and demand for digital data by archaeologists.
11
This use of digital data has important ramifications: more sophisticated forms of analysis can
be entertained and raw and synthetic data from multiple archaeological projects can be
relatively easily integrated and re-analysed by other researchers.
The ability to perform multi-criteria and other complex analyses upon spatial data can only
realistically occur in a computerised environment. Indeed it can be argued that the
reconstruction of past environments and society within a spatial context can only be fully
articulated and analysed using GIS and associated software.
Although GIS allows the integration and analysis of multiple data sets collected at different
scales, archaeological GIS applications are still relatively immature. There are standardised
GIS ready archaeological data sets (such as the Sites and Monuments Record (SMR)), though
these are normally transpositions of synthetic paper catalogues. Thus, the complexity of the
archaeological record is not articulated through any of these commonly available data sets.
This lack of data has limited the way in which archaeological studies are analysed within
computerised systems, as the collection of ancillary information becomes very time
consuming or difficult. Therefore, each landscape study sets its own criteria and goals for its
analysis and collects its own regional information accordingly.
These analyses can vary drastically in their methods of execution which are, in part, defined
by research goals. Six main approaches can be recognised:
1. Archaeological prospection.
2. Archaeological evaluation/data collection.
3. Thematic analyses.
4. Holistic analyses.
5. Archaeological prediction.
6. Archaeological interpretation.
Each approach endeavours to understand aspects of past human ecology or to manage such
landscape resources (Scudder et al. 1996). Models of the organisation of human systems are
used to understand the mechanisms behind mobility, the placement of archaeological
activities in space, and discard strategies. Human organisation at a systems level responds not
simply to the unique placement of specific resources at a single time and place, but also to the
12
regional, spatial and temporal patterning of all resources – that is, to the organisation of the
ecosystem as a whole (Ebert 1989). GIS can be used to articulate these complex and subtle
relationships.
1.3.7 Remote sensing and archaeology
Aerial photography is the most widely used and oldest form of remote sensing in
archaeology. Archaeologists are expert at re-using information from other organisations and
have made great use of vertical and stereo aerial photographic surveys conducted by national
mapping agencies and the military. Historically, bespoke archaeological remote sensing has
been based on low altitude aerial survey using handheld cameras with films sensitive to the
optical and near infra-red (Wilson 2000). The techniques were introduced in the first half of
the 20th century by such pioneers as Antoine Poidebard (in the Middle East), Charles
Lindbergh (in the United States) and Osbert Crawford and Major George Allen (in the
United Kingdom). The improvement in aircraft and camera technology after the second
World War led to changes in technique spearheaded by Derrick Riley, Irwin Scollar, Keith St.
Joseph and David Wilson amongst others (Crawford 1953; Parrington 1983; Bewley 2000;
Wilson 2000; Donoghue 2001).
Archaeologists were quick to spot the potential of Earth observation satellites in the 1970’s.
Lyons and Avery (1977) rapidly produced ‘Remote sensing: a handbook for archaeologists
and cultural resource managers’ which discussed satellite imagery. Lyons became the series
editor for the influential ‘Cultural resources remote sensing’ book and supplements (Lyons
and Mathien 1980). These supplements provided best practice methodology for the analysis
of satellite imagery and the expanding airborne multi and hyperspectral imaging systems.
Furthermore, this series introduced more formal remote sensing quantitative and qualitative
analytical techniques.
Until satellite imagery has much higher spatial resolution than is currently available its role
must be limited in archaeological studies. However, there is one area in which the satellite
perspective can provide very useful information, namely in the provision of a preliminary
overview of a large study area where detailed study or sampling is proposed.
(Allan and Richards 1983 p. 4)
13
The above quote encompases how many archaeologists have viewed satellite applications
within archaeological research. In general archaeologists have only considered satellite
imagery useful for enviro-archaeological applications (see for example Cox 1992). However,
it is timely to re-evaluate this stance in light of the high spatial resolution imagery that is
available from modern sensors and declassified sources.
Most archaeologists are familiar with aerial photography and its interpretative requirements.
However, most applications focus primarily on the identification of archaeological residues in
a European context. The majority of sites are identified through variations in crop vigour
influenced by subsurface archaeological features. It has yet to be demonstrated if this form of
identification is actually appropriate in Middle Eastern environments. Hence, wholesale
incorporation of aerial interpretation techniques in this area may be unwise.
Unfortunately most archaeologists are unfamiliar with the more formal remote sensing
interpretation techniques described by Lyons. Although over twenty years old this series of
handbooks outlines many of the techniques used in this research. The difference between the
European and American traditions is possibly explained by access to data. In the USA access
to satellite imagery and aerial multi and hyperspectral scanning systems was much easier than
in Europe. Furthermore, the USA consists of a number of starkly contrasting environmental
zones that do not occur within any single European country.
1.4 Method summary
Classification is one of the major technical goals of archaeology. Archaeological entities from
artefacts through to landscapes are classified in order to generalise the complexity of
archaeological data so that synthetic analysis is simplified.
From a remote sensing basis archaeological classification of imagery requires the use of
different analytical techniques. Archaeological features vary in permanence and construction.
They are commonly constructed using locally available building material or by creating
‘negative’ features in the landscape which are subsequently filled in by local material. Where
archaeological features are transitory their deformation sometimes results in subtle changes to
the underlying soil matrix which can be identified as a localised change in soil colour or crop
vigour. However, some archaeological residues are more permanent in nature (such as
upstanding monuments) and vestiges of past anthropogenic action can be easily identified.
14
Where this does occur many different phases of archaeological occupation can be
superimposed on one another to create a palimpsest. Disentangling this information can also
be extremely complex. Hence, the nature of archaeological residue creation means that the
identification and interpretation of archaeological phenomena is potentially very difficult.
Unlike other Earth observation techniques there is no standard archaeological ‘spectral
signature’ to aid classification.
Archaeological remote sensing techniques, from any platform, are based on the measurement
of contrast between the physical properties of materials that constitute discrete activity loci
and those of their environment. Most remote sensing techniques aimed at archaeological
prospection involve identifying contrasts in the surface or near surface attributes of soils,
topography and the vegetation canopy. One of the benefits of remote sensing imagery as a
resource is that it has a large synchronic footprint. The co-registered bands are multiple layers
of numeric information that have spatial and spectral structure. Some of this structure relates
to archaeological phenomena. Interpretation is the process of extracting the pertinent
elements of the data structure. The creative act of ‘interpretation’ itself requires that the
interpreter has an understanding of the data and its structures so that there is more self
awareness of the processes in play during the act of archaeological ‘discovery’ (Aldenderfer
1987 p. 92).
This ability to identify or detect archaeological residues is based on the conceptual premise
that a portion of the data available in the present comprises aspects of past human behaviour
which can be isolated and studied. A problem can be addressed by subdividing aspects of the
real world that are relevant to a particular problem (referred to as the problem domain or
Total Data Structure (see Figure 5: from a remote sensing perspective the problem domain
exists in n dimensional feature space)). The Relevant Data Structure is a subset of the problem
domain that is relevant to the phenomena of interest. The Expected Data Structure is a subset of
the problem domain that is expected to be relevant to the phenomena of interest. The expected
data structure may correspond well, poorly or not at all with the relevant data structure.
These data structures are articulated through deductive or inductive reasoning and are
extrapolated from remotely sensed data through quantitative or qualitative techniques. The
relationship between these categories of information progressively isolates portions of a
relevant data structure and its relationship with the real world (Carr 1985 cited in; Clark 1987
p. 58).
15
Figure 5 A schematic representation of categories of information
about the real world (after Clark 1987 p. 57)
The simplicity of this conceptual model belies a great deal of complexity and, in reality, it is
difficult to ascertain how to define each component and, once defined, how to access its
accuracy. More recent theoretical stances (for example reflexivity: Hodder 1999; Lucas 2001)
demand a more iterative and pluralistic model to address this problem.
Figure 6 outlines a schema for the collection, analysis and interpretation of satellite imagery.
The ‘image interpretation and classification’ stage is where the relevant data is extracted from
the total data. However, this is an iterative approach such that detection and interpretative
procedures are part of a feedback loop and can be continually refined in light of new
information.
However, there are a number of formal stages through which the data will be processed:
•
Data pre-processing (Chapter 5).
•
Thematic data extraction (Chapter 6).
•
Archaeological prospection (Chapters 7 and 8).
•
Imagery as a Cultural Resource Management (CRM) tool (Chapter 9).
•
Archaeological evaluation (Chapter 10).
16
These processing stages will be used to create a range of archaeological resources that can be
used to address different archaeological problems as described in section 1.3.6. The primary
focus will be to address practical issues such as archaeological prospection and thematic data
extraction which will build capacity and add value to the overall programme of enquiry.
These are the core issues which should be addressed by the research programme. These
datasets can also be used to address some of the other mechanisms of articulation (for
example virtual reality). Furthermore, satellite imagery itself can provide novel frameworks
for re-addressing methodology, particularly holistic and predictive analyses.
Figure 6 Schematic Archaeological Interpretation process.
1.4.1 Aims and objectives for the evaluation of satellite imagery
The primary thrust of the research is to address issues of archaeological detection and
interpretation and to evaluate satellite imagery against other landscape archaeological data
sources. Furthermore, as there is an approximate thrity year gap between the collection of the
Corona imagery and the Ikonos imagery critical CRM issues relating to landscape change can
also be addressed. In comparison to mapping, satellite imagery can be viewed as a more
17
objective data resource (i.e. it has not been subject to generalisation or synthesis beyond the
limitations of the sensor itself). This could offer alternative mechanisms for visualising and
classifying archaeological landscapes. In the same vein satellite imagery could be used to
reproduce landscape thematic data to contextualise the archaeological resource. Satellite
imagery may even be able to improve on traditional thematic resources on the basis of
content, scale and utility. As new data sources are employed in this research a best-practice
methodological overview is essential for future practitioners.
Given the above the thesis formally proposes to address the following aims and objective:
•
To evaluate whether satellite imagery can detect previously unrecognised
archaeological residues in different environmental zones. If residues are
recognised to attempt to understand the physical processes which allow their
detection.
•
The level of interpretation that is possible following residue detection.
•
To establish how the landscape has changed over time (multi-temporal analysis).
•
To identify if satellite imagery allows alternative mechanisms of visualising and
analysing archaeological landscapes.
•
The comparison of high resolution satellite imagery with other desk based
assessment tools.
•
The impact of high-resolution imagery for landscape survey in arid environments.
•
Establishing the strengths and weaknesses of satellite imagery for archaeological
research.
•
The ability to produce other archaeologically pertinent thematic information.
•
Methodological best practice in employing satellite imagery.
1.5 Other research into the archaeological application of high resolution satellite
imagery
This research will evaluate the utility of high resolution satellite imagery for archaeological
enquiry in the semi-arid environment around Homs, Syria. However, this work has not been
conducted in isolation. Other researchers have also examined the implications of satellite
18
imagery for archaeological applications. This section places this research into the broader
worldwide analysis of high resolution satellite archaeological applications.
Gaffney et al. (1996) summarises archaeological GIS applications involving satellite imagery
just prior to the introduction and declassification of hi-resolution satellite sensors. Hence, for
the purposes of this research it acts as a benchmark summary. They employed Landsat
imagery to define a range physiographic and soil zones in a region around Hvar, Yugoslavia.
Daels and Al Saadi (1990) applied both Landsat TM and SPOT sensors for a
geoarchaeological investigation of relict channels in Mesopotamia. They also identified the
utility of the multispectral nature of Landsat particularly, band 5. A less rigorous and
potentially spurious Landsat prospection scheme was undertaken by Mumford and Parcak
(2002) in Sinai.
In general medium spatial resolution satellite sensors, predominantly Landsat, have only been
employed to classify the landscape into different themes as a precursor to other
archaeological analyses. Although most of these researchers dismiss the utility of satellite
imagery as a prospection tool due to the perceived problem of limited spatial resolution, this
research has demonstrated that satellite imagery can be a significant tool for archaeological
prospection. However, the resolving characteristics of the sensor and the environmental
conditions do dictate what residues will be detected. Sever and Wagner (1991) were fortunate
enough to analyse airborne sensors that simulated satellite sensors but had a spatial resolution
of between 5 and 10 metres. Thematic Mapper Simulator (TMS: with similar spectral
characteristics to Landsat) and day and night-time Thermal Infrared Multispectral Simulator
(TIMS: with narrow spectral bandwidths in the thermal) detected a number of surface and
subsurface phenomena such as prehistoric walls, buildings, agricultural fields and roadways.
The application area was the semi-arid environment at Chaco Canyon in north-western New
Mexico, USA. TIMS imagery was found to be a superior interpretative medium than the
TMS imagery. By exploiting the thermal inertia properties between the day and night-time
images, the thermal channels were analysed independently or as false colour composites to
identify roadways, structures and other archaeological residues. The authors also took a
distinctly statistical stance on interpretation and developed a number of different filtering
techniques to improve interpretation. Finally, they recognised that the analytical methodology
could be employed in other semi-arid environments throughout the world, including North
Africa and the Middle East. This highlights the utility of non-visual wavelengths, particularly
19
thermal, which have yet to be fully evaluated for archaeological purposes (although see BenDor et al. (1999b) and Donoghue (2001) for localised examples) and that it is the spatial
resolution, and not the spectral resolution, of sensors such as Landsat that limit their
archaeological application.
Fowler and El-Baz have produced a number of articles that have raised the awareness of high
and medium resolution satellite resources within the archaeological community. These have
tended to focus on such high profile landscapes as Danebury, Giza, the Great Wall of China
and Stonehenge (Fowler 1993; 1994; 1996; El-Baz 1997; Fowler 2001; 2002) employing
sensors such as Landsat, SPOT, Corona, KVR and Ikonos. El-Baz (1997) discusses a range
of sensors (including RADAR) used in different projects throughout the world outlining the
potential of satellite remote sensing in different environments. Furthermore, this article
discusses a range of complementary ground-based remote sensing techniques. Fowler’s work
is generally in environments where the archaeological resource is well understood and these
enquiries have added little new archaeological information. However, they have been very
useful in providing examples of an alternative viewpoint from which to contextualise a
landscape. Although Fowler does identify the scale-dependent benefits of different sensors.
This is exemplified in his paper concerning the landscape environs of Figsbury Ring,
Wiltshire (Fowler 2002).
Of more import has been the work of Comfort (1997; 1999; 2000), Kennedy (1998),
Kouchoukos (2001), Stone (1995; 2003), Wilkinson et al. (2001) and Ur (2002; 2003) with
Corona imagery. These studies have focused on archaeological residue detection (Comfort,
Kennedy and Ur), geoarchaeological investigations (Stone and Wilkinson et al.) and
human/environmental interactions (Kouchoukos). Each project has tended to use Corona in
conjunction with other satellite imagery (mainly Landsat, SPOT and KVR) and in general
SPOT has been employed as the geo-referencing medium (see Ur 2003 for a methodological
discussion). Comfort’s reports outlined possibly the first large-scale archaeological survey to
employ Corona photography. Although the analytical techniques employed were not
sophisticated, the results demonstrated the utility of the recently declassified resource
(particularly for road and aqueduct networks). Kennedy, Kouchoukos and Ur each made the
significant point that Corona could provide a useful substitute to aerial photography where it
is difficult to access and that the antiquity of the imagery is important. Stone (2003) applied
Corona photography and SPOT imagery to locate palaeochannels in the unstable Tigris
20
floodplain and any associated settlement activity. Wilkinson et al. (2001) employed Corona for
geoarchaeological and landscape studies around Tell Brak, Syria. Once again the antiquity of
the photography meant that identification of some features was easier than with present day
imagery. The authors identified possible sites, hollow ways, wadis and the extent of the
alluvial fan surrounding Tell Brak. At the tell itself they were able to identify different postdeposition alluvial fans and colluvial slopes.
Ur employed Corona imagery on a comparative scale to this research project, although in
Ur’s research area there are no structural remains as observed in the basalt zone. He also
recognised the limitations of Landsat and SPOT imagery for the large scale identification of
archaeological residues and in response applied high resolution Corona photography to study
ancient road networks in Northern Mesopotamia. Ur (2003 p. 105) also recognised that
anthropogenic modification of the soils at archaeological sites led to changes in reflectance
and accords this to improved drainage (and hence reduced moisture content) at these sites.
However, he has yet to undertake physical characterisation of these soils to demonstrate this
hypothesis (as discussed in Chapter 8). In summary, although different methodological
techniques are employed in Ur’s research the overall implications of the Corona photography
for archaeological interpretation are similar.
Hence, Corona has been employed by a number of different research projects in a number of
different, but mainly semi-arid, application areas. Less archaeological work has been
conducted on the use of high resolution commercial satellite sensors. Campana and
Francovich (2003) have integrated high resolution satellite remote sensing techniques into the
high quality GIS driven CRM system of Tuscany. The Department of Medieval Archaeology
at the University of Siena, Italy has been conducting landscape surveys and analysis for over
25 years. In addition, over the past 10 years they have been employing innovative GIS
techniques. The result is an exemplary archive within which archaeological data from any
scale can be fully integrated and articulated.
Like Ur (2002; 2003) they recognised the benefits of large area synoptic collection and the
multispectral capacity that satellite sensors offered, but also recognised that the
predominantly low spatial resolution of the sensors limited their archaeological utility. In
early 2001 they first integrated Ikonos high resolution satellite imagery into their data model,
quickly followed by Quickbird imagery. In contrast to this research project they were able to
21
evaluate these data in a well studied landscape. This provided benchmarks (mature
methodological and theoretical hypotheses) against which the archaeological interpretation of
the imagery could be evaluated. They used a number of different visual and statistical
techniques for image interpretation including false colour composites, stretching, PCA and
Tasselled Cap. Analysis of the imagery identified a significant number (84) of potential
archaeological features which were not observed on any other dataset. Ground observation
was conducted on a 40% sample; 59% of this sample proved to be archaeological. For their
environment they concluded that the NIR band proved the most useful. The positive results
of this study are particularly interesting as the landscape has been intensively studied with a
range of different archaeological techniques and therefore it would be expected that the
imagery would only locate previously identified residues. However, until recently bespoke
archaeological aerial photography has been restricted by the Italian military.
Little work has been conducted on satellite imagery and its application in recording and
monitoring field-systems. However, Romano and Tolba (1996) identified centuriation
patterns in Corinthia, Greece using SPOT imagery. However, although they were able to
identify the larger frameworks, they were not able to locate smaller divisions.
Although Brivio et al. (2000) did not explicitly employ high resolution satellite imagery, they
did define an integrated analytical model which explicitly examined aspects of scale changes
in archaeological interpretation. This is another area where satellite imagery will help in
archaeological enquiries to determine empirical scale thresholds through which different
prospection and interpretation mechanisms are required. Such analyses can be very fine
grained. Buck et al. (2003) examined the applicability of distinguishing pottery and obsidian
artefacts from background soils using spectral signatures. High resolution spectro-radiometry
data was collected in-situ and compared with laboratory samples. Further research can be
conducted into the scale implications of generalising these studies. Understanding the scale
implications of these different research projects will elucidate a range of archaeological
problems.
Clark et al. (1998 p. 1475), although employing slightly lower resolution sensors, do highlight
the impact that satellite images have on fieldwork design. They specifically discuss the fact
that satellite imagery can be used to determine which areas are more likely to contain
22
archaeological residues. They agree that this has a significant positive cost implications for
archaeological fieldwork.
In contrast to these other studies this research was established with the formal aim of
evaluating high resolution satellite sensors. The application area was chosen because it
comprised of different environmental zones which provided a more representative context
for evaluation. Furthermore, although this research is framed within the broader goals of the
SHR project this component had strictly defined methodological aims. As such the research
has allowed the full potential of the sensors to be explored.
Like Comfort (1997; 1999; 2000), Kennedy (1998), Kouchoukos (2001), Stone (2003),
Wilkinson et al. (2001) and Ur (2002; 2003) this research has further demonstrated that
Corona imagery is a significant tool for residue prospection. In addition this research has
compared the utility of Corona against the modern Ikonos satellite imagery and found this to
be an equally useful resource (more so in the basalt zone). Rather than just visually analysing
the raw data a number of quantitative techniques were produced which improved the
likelihood of detecting residues, particularly in the marl zone. This was augmented with the
laboratory analysis of soil samples from the marl. Even though the results of these analyses
were not conclusive they provided a promising platform with which to understand the
formation and deformation processes which produce reflectance increases which make sites
visible. Furthermore, the work in the basalt zone has extended the scope of these sensors for
recording fieldsystems, which was previously unrecognised. The ability to record such
systems with a high degree of spatial accuracy will be of value to archaeologists facing similar
problems.
The ability to conduct time change analysis on the Corona and Ikonos imagery provided a
number of critical CRM insights. Not only was it possible to identify which residues had been
destroyed in the intervening period, it was also possible to determine what caused their
destruction. This information is essential in order to frame a realistic management strategy
for the archaeological resource.
The integration of imagery with different spectral and spatial resolutions allowed the creation
of spatially and a-spatially accurate thematic layers which provide context for archaeological
interpretations. It is likely that the integration of multi-resolution data collected at different
23
scales for highly accurate thematic analysis will become the norm in remote sensing
applications. Irrespective of the multi-scalar integration this thematic information was not
previously available to the SHR project.
Due to the different environmental conditions it is difficult to compare the results of this
research with the work conducted by Campana and Francovich (2003) although it is
interesting to note that Ikonos imagery produced positive results in a European and
Mediterranean context (although see below). Campana and Francovich also conducted their
research in a well studied environment and were therefore able to provide benchmark figures
concerning the utility of high resolution satellite imagery for landscape prospection.
Unfortunately, this was not possible within the application area and hence the levels of
interpretation are still limited and are likely to remain so until further ground observation is
conducted.
However, this lack of information has allowed the SHR project to methodologically and
theoretically review its ground survey procedures from first principles. In this context both
the Corona and Ikonos satellite imagery has proven invaluable in providing a structure for
ground survey. This has saved a considerable amount of time and money and has meant that
after only five field seasons that SHR project has a mature grasp of the archaeological
problem within the application area.
In well researched environments, such as most of Europe, it is unlikely that high resolution
sensors will provide much benefit. In general these countries have mature CRM systems and
access to a range of aerial imagery. However, the larger footprint of satellite sensors may
make them useful for some enviro-archaeological applications (see for example Cox 1992),
although in these instances increased spectral, rather than spatial, resolution is likely to be
important. Mosaiced high spatial resolution aerial imagery, such as GetMapping, is also
becoming available, further reducing the likelihood that satellite imagery would be employed.
However, even in Europe some archaeologists have had difficulty in accessing and
employing aerial imagery, such as those in Italy (where until recently aerial photography was
subject to restrictions (Jones 2000 p. 53)). In these environments satellite imagery still has
utility as a prospection tools (see for example Campana and Francovich 2003).
24
However, the majority of the applications discussed in this section have been conducted in
semi-arid or arid environments. It appears that archaeological satellite applications are ideally
suited to these environments where many of the residues exist as upstanding architectural
monuments or as soil marks. Corona, Ikonos and other high resolution commercial sensors
will continue to have a huge impact on site and landscape studies in these areas.
25
CHAPTER 2 CONCEPTS OF REMOTE SENSING FOR ARCHAEOLOGY
2.1 Remote sensing – Introduction and definition
This chapter is a selective introduction of remote sensing concepts and techniques which are
likely to be of relevance for understanding the analytical techniques employed in the research.
Figure 7 Soil related archaeological applications at different
portions of the EM spectrum (after Lucas 2001 p. 156).
Remote sensing has been generically defined as:
….. the acquisition of information about an object without being in physical contact with it.
(Elachi 1987)
However, a more stringent definition is required in order to discuss the elucidation of
archaeologically related information on the Earth’s surface from space, which is the thrust of
this research. A more appropriate definition comes from the United Nations in their annex
‘Principles Relating to Remote Sensing of the Earth from Space’ (United Nations 1985):
26
The term Remote Sensing means the sensing of the Earth's surface from space by making
use of the properties of electromagnetic waves emitted, reflected or diffracted by the sensed
objects, for the purpose of improving natural resources management, land use and the
protection of the environment.
Visible light of wavelengths from 0.4 to 0.7 µm (micrometers) is a small section of the EM
energy spectrum. The vast majority of archaeological remote sensing applications have relied
on this small portion of the spectrum (see Figure 7). Extending beyond these wavelengths
allow archaeologists to explore potentially significant information. Archaeologists do employ
non-visual wavelengths: x-rays of archaeological objects, thin section analysis and geophysical
prospection (Sever 1988; Lucas 2001 pp. 154-156). However, there has been little research
into how archaeological sites respond to different wavelengths.
Figure 8 Diagrammatic representation of a photon.
2.1.1 Electromagnetic energy
The underlying basis for most remote sensing systems is that of measuring the varying energy
levels of a photon. A photon travels as an electromagnetic (EM) wave having two
27
components, oscillating as sine waves at right angles, one consisting of a varying electric field,
the other a varying magnetic field. Both have the same amplitudes (strengths) which reach
their maxima and minima at the same time (see Figure 8). Variations in photon energies are
tied to the wavelength (or its inverse, frequency). EM radiation that varies from high to low
energy levels comprises the electromagnetic spectrum (see Figure 9). Radiation from specific parts
of the EM spectrum contain photons of different wavelengths whose energy levels fall within
a discrete range of values (Nunnally 1973). When any target material is excited by internal
processes or by interaction with incoming EM radiation, it will emit photons of varying
wavelengths whose radiometric quantities differ in a way that is diagnostic of the material.
The plot of variation of power with wavelength gives rise to a specific pattern or curve that is
the spectral signature for the object being sensed (see Figure 22 and Figure 43).
Figure 9 The electromagnetic spectrum and atmospheric
absorption curve.
The majority of EM remote sensing instruments, including the human visual cortex, passively
monitor the Earth, detecting the reflected Sun’s energy at selected wavelength groupings (bands)
from different elevations (platforms). The images produced depend directly on the efficiency
with which the ground and vegetation reflect the measured wavelengths, how the atmosphere
affects (attenuates) this signal (see Figure 10) and the resolution of the sensing devices. The
28
wavelength determines some physical properties. For example, there is an inverse
relationship between wavelength and the degree of atmospheric scattering. For example, the
shortest visible wavelength (blue) is affected by atmospheric particulates and scattered
(referred to as Rayleigh scattering). Most wavelengths only interact with the surface microns of
the Earth, which for the vast majority of the Earth’s surface is composed of vegetation, soil
or water. Wavelength penetration is dependent upon the characteristics of both the material
and the wavelength. For example shorter wavelengths have increased penetration in water
and longer wavelengths have increased penetration in dry sand.
Figure 10 Electromagnetic energy interactions with a target.
However, not all EM radiation is reflected: the Earth also emits its own radiation. A body
emits radiation as a function of its temperature. The peak of the Sun’s emitted energy is in
the visible energy band (0.55 µm), whilst the Earth itself at an ambient temperature of 300K
radiates energy in the mid infrared range (3-50 µm see Figure 11). Therefore, in contrast to
the visible and near to mid infrared (0.4-5.0 µm), where the reflection characteristics
determine the structure of an image, thermal infrared sensors principally detect the intensity
of emitted thermal energy (Shell 2000).
29
Figure 11 The spectral distribution of energy emitted by a
blackbody as a function of its temperature (after Lillesand and
Kiefer 1999).
During the day the Sun’s contribution to thermal imaging is not negligible. All materials
absorb the Sun’s energy to differing extents, depending on their absorption/reflection
characteristics at these wavelengths. This will change their temperature as a function of their
thermal conductivity and thermal capacity characteristics. The topographical effect of ground
slope and aspect on the angle of incidence of the Sunlight is also a significant factor in the
Sun’s ability to heat the ground, and thermal shadowing occurs in daytime thermal imaging.
At night, in the absence of the Sun’s influence, objects cool at a rate determined by the same
factors influenced by the local environment. This leads to a situation where, for example, dry
soil and rocks heat up more rapidly during the day than water, and cool more rapidly at night
(see Figure 12). The diurnal variation in temperature leads to a phenomenon where,
sometime after both dawn and Sunset, respectively heating and cooling objects transiently
30
have the same radiant temperature, and cannot be distinguished in the thermal infrared image
(Shell 2000).
Figure 12 Diurnal temperature variations (after Lillesand and
Kiefer 1999).
Sensors can be established to analyse different portions (bands) of the EM spectrum with
differing degrees of resolution (Kruckman 1987; Holden et al. 2002). These bands can be
combined to display what the eye would see in the red, green and blue portions of the
spectrum (as with colour aerial photography), but also other spectral bands that the eye
cannot discriminate (see Figure 34).
2.1.1.1 Visible region
The visible region of the EM spectrum (0.4-0.7 µm) is nominally delimited on the basis of
human vision. This region coincides with an atmospheric window (see Figure 9) which makes
the atmosphere almost transparent and with the peak emmittance of the Sun (see Figure 11).
31
Figure 13 Solar radiation interactions with the atmosphere for
short wavelengths (after Campbell 2002).
2.1.2 Interactions with the atmosphere
All radiation collected by remote sensors passes through the Earth’s atmosphere, to varying
degrees. However, radiation that reaches satellite sensors must pass through the entire
atmosphere at least once which substantially attenuates the radiation (see Figure 14 and
Figure 15). Atmospheric effects can be divided into three categories: scattering, absorption and
refraction. At the shorter wavelengths approximately 90% of the incoming energy is affected
(see Figure 13).
2.1.2.1 Scattering
Scattering, as the name suggests, is the scattering of energy as it interacts with particles or
atmospheric molecules. The amount of scattering is dependent upon the ratio of the particle
size to the wavelength: the higher this ratio value the less the likelihood that scattering will
occur. Radiation is scattered towards space, the Earth and, importantly, the sensor.
32
Figure 14 Five types of radiative interaction with the atmosphere
and how they impact the Instantaneous Field of View (IFOV:
after Tso and Mather 2001 p. 15)
Scattering is predominantly wavelength dependent. When there is a very low particle size to
wavelength ratio Rayleigh scattering occurs. Tiny particles and some of the smaller molecules
(such as N2 and O2) affect radiation with shorter wavelengths. Its effects start to become
negligible in the NIR. When the particle size to wavelength ratio approaches 1 Mie scattering
occurs (i.e. particulate diameter and wavelength are approximately the same: see Figure 15
and Figure 17). When particulate size is much larger than the wavelength then Nonselective
scattering occurs.
33
Scattering affects sensor readings in a variety of ways. Many interpreters do not consider the
blue and ultraviolet regions as useful, when collected from a satellite platform, due to the
large amount of Rayleigh scattering. Furthermore, the preponderance of forward scattering
effects can reduce spatial detail by scattering radiation from adjacent pixels into the
‘observed’ pixel (see Figure 14).
Figure 15 Changes in reflected, diffuse, scattered and observed
radiation over wavelength (after Campbell 2002).
2.1.2.2 Refraction
Refraction (see Figure 16) is the change of direction caused by radiation striking a transparent
material with a different density. The atmosphere is composed of different ‘layers’
characterised by variations in clarity, humidity and temperature. Each of these variables
affects the density of the layer and in turn the amount of attenuation that occurs.
Figure 16 Refraction: How the path of radiation is affected by
changes in the density of the medium.
34
2.1.2.3 Absorption
Atmospheric absorption is the prevention or severe attenuation of energy transmission
through the atmosphere. Energy which is absorbed by the atmosphere is re-radiated at longer
wavelengths. Absorption is mainly caused by three gases: Ozone (O3), carbon dioxide (CO2)
and water vapour (H2O). The parts of the spectrum through which EM energy passes
relatively unhindered are called atmospheric windows. Obviously, satellite systems focussed on
remote sensing of the Earth’s surface have sensor systems that are configured to coincide
with an atmospheric window (see Figure 9).
Figure 17 Atmospheric particulates and their scattering effects
(after Campbell 2002).
35
2.1.3 Interactions with an object
As electromagnetic energy reaches the object on the Earth’s surface it must be reflected,
absorbed or transmitted. The proportions of each process are dependent upon the surface
characteristics of the object, the wavelength of the energy and the angle of illumination.
2.1.3.1 Reflection
Energy is reflected when it interacts with a non-transparent surface. The type of reflection
that occurs is dependent upon the relationship of the wavelength of the energy and the
relative roughness of the surface on the object. If, for example, the surface of the object
appears smooth at a certain wavelength (i.e. the irregularities on the surface are much smaller
than the wavelength) then specular reflection occurs. For visible wavelengths specular
reflection occurs on the surface of mirrors and water (Campbell 2002 p. 42). Conversely, if
the surface of the object appears rough at a certain wavelength then diffuse reflection occurs.
A perfectly diffuse surface is known as a Lambertian surface (see Figure 18). Most surfaces
behave in between these two extremes. It is important to note that the view angle of the
sensor will effect the amount of radiation that is received. Archaeologists have exploited nonlambertian reflectance in the use of oblique photography of crop marks.
Figure 18 Specular and perfectly diffuse (Lambertian) reflectance.
2.1.4 Discussion
A satellite sensor records the intensity of EM radiation over portions of the EM spectrum.
However, as explained above, the values recorded by the sensor do not reflect the actual
values observed at the object. On the path from the object to the sensor radiation is removed
by scattering, refraction and absorption. Furthermore, radiation is also added by scattering
and refraction from adjacent objects.
36
Figure 19 Remote sensing from different platforms.
37
2.2 The components of remote sensing systems
2.2.1 Hardware: Platforms
The distance between a sensor and the object under study is important not only for its
relationship to spatial resolution. The closer the sensor the lower the impact of attenuation
by the ‘masking’ medium be this soil, vegetation or the atmosphere. All remotely sensed
imagery is attenuated by the medium through which it passes. The effect of this attenuation is
a function of the distance between the sensor, the object and the energy source, the
characteristics of the medium and the characteristics of the specific wavelength.
The choice of platform also has a significant impact on the ability to geo-register the imagery.
Geo-registering, in this context, provides locational information to a digital file. The process
of geo-registration will be dealt with in more detail in later chapters. Suffice it to say that it is
more desirable to obtain imagery which is either pre-registered or easy to geo-register. This is
particularly relevant to airborne multispectral imagery where the pitch, tilt and yaw of the
aircraft makes registration a particularly time-consuming process.
Archaeologists use four platforms for remote sensing (Figure 19):
Ground Level: Traditional terrestrial geophysics and handheld photography. Imagery is
normally located through terrestrial survey techniques, if at all.
Near Ground Level: Remote photography from a kite, blimp or tower. Normally
located through terrestrial survey techniques, if at all.
Aerial: Bespoke oblique archaeological imagery or vertical landscape survey. Resolution
from this platform is variable and dependent upon the elevation and resolving
characteristics of the sensor. The majority of sensors are based upon cameras
employing different film types, although archaeologists are researching the
application of airborne multispectral systems (Donoghue 1999). Correcting scanned
airborne imagery used to be a very time-consuming process (Teng 1997 p. 76).
However, integrated Differential Global Positioning Systems (DGPS) and inertial
navigation systems have improved the ease and cost of geo-referencing.
Satellite: Normally multispectral electromagnetic scanners. Spatial resolution is
dependent upon the sensor set-up and varies between less than 1m to many
hundreds of metres. The footprint of a satellite image is much larger than other RS
38
platforms, consequently increasing the land area that can be explored efficiently.
Satellite imagery suffers from relatively severe atmospheric effects. Most imagery is
automatically geo-located by referring to the orbital (ephemeris) and sensor
characteristics at the time of data capture.
Figure 20 Sun-synchronous orbit.
Satellites used for archaeological purposes require relatively high spatial resolution. High
spatial resolution satellites tend to follow a low Sun-synchronous orbit (see Figure 20). As the
vast majority contain passive sensors, Earth related information is normally only collected
during daylight. However, when the satellite is orbiting the dark side of the Earth sensors will
pick up emitted and artificial radiation primarily from the Earth’s emitted thermal energy (this
has been used to great effect for mapping sea surface temperature and plotting geo-thermal
anomalies) and artificial light sources (i.e. major conurbations).
39
2.2.2 Hardware: Sensor systems
Most analogue camera systems collect on a frame by frame basis. Each frame has a relatively
large synoptic footprint. Although there are some analogue camera scanning systems, such as
Corona KH4-B, the majority have been replaced by digital scanning systems (see Figure 21).
Digital systems use a variety of techniques to collect information mainly employing pushbroom,
whiskbroom or area array configurations. These systems collect data along the line of flight with
a certain degree of pixel overlap. Consequently, the stability and attitude of the collection
platform plays an important role in how easily the data can be corrected and geo-referenced.
Figure 21 Common digital imaging systems.
2.2.3 Hardware: Sensor characteristics
Due to the complexities of sensor systems only the aspects pertinent to this research will be
highlighted. For a broader discussion see the standard remote sensing texts (Elachi 1987;
40
Cracknell and Hayes 1991; Jensen 1996; Sabins 1997; Lillesand and Kiefer 1999; Mather
1999; Campbell 2002). Sensors fall into two main categories: passive and active. Passive sensors
are the most common form of sensor, and record naturally occurring electromagnetic
radiation that is reflected or emitted from the Earth. Active sensors (such as RADAR and
LiDAR) bathe the terrain in artificial energy and then record the amount of radiant flux
scattered back to the sensor. No active sensors will be evaluated in this research.
Other, non-instrumental, aspects of sensors are related to their ability to resolve data. Each
remote sensing system has four major axes of resolution:
1. Spatial resolution.
2. Spectral resolution.
3. Radiometric resolution.
4. Temporal resolution.
2.2.3.1 Spectral resolution
Spectral resolution (see Figure 22) refers to the dimensions and number of specific
wavelengths for which a sensor is sensitive. Black and white (or more correctly grey scale)
imagery is normally sensitive to a broad spectral range normally in the visible and NIR
wavelengths. This is often referred to as panchromatic imagery. In comparison, the
multispectral scanner of Landsat Thematic Mapper contains 7 bands covering discrete
wavelength ranges over different parts of the spectrum. For example between 0.5 and 0.7 µm
(the visible wavelengths) the Landsat TM sensor has 3 bands that broadly relate to the
blue/green, green and red parts of the spectrum.
Hyperspectral scanners can collect data in many very narrow band passes. For example, the
AVIRIS sensor can collect approximately 224 bands between 0.4 and 2.5 µm at 10 nm
intervals. Therefore, between 0.5 and 0.7 µm AVIRIS collects data in 20 bands. Thus,
spectral resolution can be seen to increase from a ‘broad band’ panchromatic to the very
narrow bands of hyperspectral. Increasing spectral resolution at the appropriate areas of the
electromagnetic spectrum may help to improve image interpretation and classification.
However, this is dependent upon how easy it is to discriminate between the different spectral
signatures of the components in the image.
41
Figure 22 Contrasting spectral resolutions of AVIRIS, Landsat and
Panchromatic imagery. Note how closely the 4 bands in Landsat
follow the 100 AVIRIS bands.
Figure 23 Decreasing spatial resolution.
42
2.2.3.2 Spatial resolution
Spatial resolution (see Figure 23) is dependent upon the resolving power of the sensor, the
distance from the object and the size of the object to be identified. Fortunately, many
systems now represent this relationship as simply the ground dimension in metres for each
pixel. As a general rule, when using the same sensor system, spatial resolution will decrease as
distance from the object increases (a negative relationship). This is particularly important for
aerial platforms where aircraft elevation can change dramatically. However, the current
generation of commercial satellite sensors (such as Ikonos) have very powerful resolving
characteristics (approximately 1m ground resolution) which are comparable to many aerial
surveys. A useful rule of thumb is that in order to detect a feature the spatial resolution of the
sensor should be one half of the feature’s smallest dimension (Jensen 2000).
Figure 24 The creation, recording and analysis of mixed and pure
pixels.
An important function related to pixel size is that of mixed pixels. In reality it is rare for a
single pixel to contain homogenous information (see Figure 24). Even areas which look
43
homogenous to the eye contain a mixture of information. For example, a Landsat pixel
collected over a field containing wheat will also contain information about the soil and
weeds. This means that the spectral values of any particular pixel are really composite values
for each of the objects present. However, if the reflectance curves of the objects are known
and their scattering properties are identical, then the pixel’s reflectance can be mathematically
decomposed into the reflectance of its constituent objects.
2.2.3.3 Radiometric resolution
Radiometric resolution (see Figure 26) determines the sensitivity of the sensor to differences
in received signal strength. The data are normally quantised into bits (power of 2). For
example, Landsat Thematic Mapper (TM) data is quantised as 8 bits (8 to the power of 2: 256
different values or Digital Numbers (DNs)). Ikonos data is quantised in 11 bits (2048 DN
values). Consequently this can have a significant impact on the ability to measure and
discriminate objects. For example, in 8 bit imagery bright areas may be overexposed and dark
areas in shadow whereas with 11 bit data it is possible to distinguish objects within these
bright and dark areas (see Figure 25). Some researchers do not use Landsat imagery in arid
and semi-arid environments as they consider it to be too overexposed (Stone 2003).
Figure 25 The benefits of increased radiometric resolution. The 8
bit imagery is overexposed whereas structures are identifiable in
the 11 bit imagery (image courtesy of DigitalGlobe and Dr. Amr
Al-Azm).
Sensors with high radiometric resolution require image manipulation to appreciate the
increase in data quality (as exemplified by Figure 25 and Figure 26). The Ikonos multispectral
imagery has the following approximate standard deviations for each band: Blue (30DN),
44
green and red (60DN) and near infra-red (110DN). This means that, with the exception of
the near infra-red band, all bands could have been recorded in 8 bit and maintained at least
two standard deviations from the mean. So why is the Ikonos imagery 11 bit rather than 8
bit? In reality sensors are configured to record the full range of values expected across the
whole Earth. Although a sensor has the potential to record in 8 bit within each scene it only
requires a subset of its full range. Therefore, to ensure full 8 bit resolution a sensor must have
the ability to record a much higher range of values.
Figure 26 Decreasing radiometric resolution (Note this is not just
improving contrast).
2.2.3.4 Temporal resolution
Temporal resolution (see Figure 27) refers to how often a sensor system records a particular
area. For all platforms except satellite this value is likely to be infrequent. All satellites, except
45
those in geo-stationary orbits, have a temporal resolution dependent upon their orbital
characteristics, their ability to record off-nadir (see Figure 28) and the number of satellites
containing the same or similar sensor system. For example, a single SPOT satellite has a
repeat rate of 26 days at-nadir and 5 – 10 days for off-nadir. Furthermore, the current
constellation of 3 satellites with similar sensor characteristics (SPOT 2, 4 and 5) has a much
lower repeat rate as each satellite is placed in a complementary orbit. Temporal resolution is
particularly important for undertaking time change analysis. The choice of sensor is
dependent upon the rate of change of the object under study.
2.2.4 Software: Image processing systems
The vast majority of remotely sensed data are now captured in raster digital format. These
digital images can be processed with computerised image processing software. Raster
processing systems vary greatly in functionality and price, ranging from a simple imageediting suite such as Adobe Photoshop to complex hyperspectral image processing systems
such as ENVI, PCI or Erdas IMAGINE.
Figure 27 Temporal Resolution: Looking at changes over time.
The negative Corona image of 1970 is compared to a positive
Ikonos image from 2002. The major changes are noted on the
Ikonos image.
Image enhancement is the process of making an image more interpretable for a particular
application. A dedicated image-processing package has the capability to perform varied digital
manipulation. The most important are:
•
Image pre-processing.
•
Histogram manipulations.
46
•
Band ratioing and other Boolean algebraic techniques.
•
Multivariate analyses (including Principal Components Analysis).
•
Kernel filtering.
•
Image classification.
•
Colour composite combinations from different raw or processed bands.
Landsat Thematic Mapper (TM) imagery is primarily used for illustrative purposes
throughout this section, although the techniques described can be used for data from any
platform (ground, near ground, aerial and satellite) and any image depth (the image
processing term for radiometric resolution).
Figure 28 Roll, pitch and yaw effects on the nadir point.
2.2.5 Processing: Image pre-processing
The data from aerial or orbiting sensors are initially uncorrected for radiometric and
geometric discrepancies; they are considered ‘raw’ (if one wanted to apply bespoke correction
one can purchase imagery in this state). However, most users prefer to have errors corrected
by the supplier. The subject of correction is tied to the procedures called pre-processing or
image restoration.
47
Pre-processing is an important stage where models are used to artificially correct for problems
that attenuate the EM signal between the object and the sensor (in the case of active sensors
this also includes the passage of energy from the sensor to the object: as discussed in 2.1).
Pre-processing can be subdivided into radiometric and geometric corrections.
Geometric correction includes correcting for skew (the effect owing to rotation of the Earth)
and platform movements (roll and pitch) that cause the straight-down line of sight (nadir) to
deviate from the vertical (off-nadir, see Figure 28). The pixels acquired off nadir are
progressively elongated depending on the look angle and the natural curvature of the surface
of the Earth. Some projection compensation is needed: normally through registering the
image to a known coordinate system. The most commonly applied projection is Universal
Transverse Mercator (UTM). Once the various corrections are made, the result is usually a
shift in position of any given pixel into its new projection system, such that it does not
necessarily have the same DN values that it had in the original (distorted) position (Shennan
and Donoghue 1992). A new set of values can be calculated by re-sampling, a mathematical
process involving interpolation of values using an algorithm such as Nearest-Neighbour,
Bilinear, or Cubic Convolution (see Figure 29). These re-sampling techniques give an
estimation of the new brightness value (as a DN) that is closer to the new condition (A).
Figure 29 Common pixel resampling techniques.
In the Nearest Neighbour technique, the transformed pixel takes the value of the closest
pixel in the pre-shifted array. In the Bilinear Interpolation approach, the average of the DNs
for the 4 pixels surrounding the transformed output pixel is used. The Cubic Convolution
technique averages the 16 closest input pixels.
48
Corrections can be made that affect undue radiometric variations, i.e. changes in the
measured radiances owing to a variety of factors. These are divided into natural and
instrumental variations. One natural condition relates to temporal and spatial differences in
atmospheric conditions. For example, presence of atmospheric water vapour influences the
radiance, such that the reflectance from the same object varies over space because of
differences in atmospheric conditions. Another correction takes into account the changes in
Sun angle (seasonal elevation; time of day) as reflection values from the same object vary
according to the angle and intensity of incoming radiation.
Instrument corrections involve variations in detector response and electronic anomalies.
Most common are systematic differences in one or more detectors. This can induce such
effects as variable line darkening (one detector may produce a line that is brighter or darker
than its neighbours), line drop out (a fluctuation may cause all or part of a line to be missing),
and random noise (speckling). Procedures are available to apply computer-generated
corrections for any of these based upon localised assumptions, improving overall image
quality. Furthermore, sensors are periodically calibrated against objects on the ground with
known reflectance characteristics.
Figure 30 Examples of camera lens and oblique distortions (after
Scollar 1990 p. 83; Teng 1997 p. 82).
49
2.2.5.1 A short note on image geometry
Many satellite platforms now allow programmable view angles which can result in scenes
taken from multiple view angles. Platforms that can collect multiple view angles provide a
number of advantages including revisit frequency, stereoscopy, sun-spot avoidance and
angular reflectance analysis.
Figure 31 Unit grid distortions of the Corona KH-4B panoramic
camera (after Galiatsatos in prep p. 93).
Variable view angles are essential for the generation of stereo pairs which use differences in
relative image orientation to deduce depth. Hotspots effects can be reduced with non-vertical
view angles. Hotspots occur when the instananeous field of view, sensor and Sun are in a
direct line resulting in a strongly illuminated shadow free area. So called sunspots are where
the Sun’s energy is relected directly into the sensor by a specular reflector (i.e. water) resulting
in glare. This effect is view angle dependant. Galiatsatos (in prep) recognised that variations
in view angle altered the archaeological interpretability of the Corona KH-4B fore and aft
50
cameras for the same area in Syria. Changes in feature contrast is a function of both
directional reflectance properties and the absolute reflectance of the ground features. View
angle can be considered important as different features can produce different responses
under different view angles (Barnsley et al. 1997). Hence, an increased understanding in view
angle effects could be increasingly important for archaeological interpretation in the future,
however, it is not a major focus of this work.
Oblique images have a number of important geometric properties, the understanding of
which are germaine to their interpretation. All collection platforms introduce geometric
distortions. A number of these are inherent to the device itself. For example, pin and barrel
distortions in camera systems are dependent on the lenses employed and commonly increase
away from the nadir point (see Figure 30). A schematic diagram of the panoramic distortions
in Corona can be seen in Figure 31. Oblique imagery causes variation in scale. The more the
principal point moves away from the nadir the greater the scale variation in the image (as
demonstrated in Figure 30). This is further exacerbated when there are variations in terrain
which can result in displacement in the imagery.
Figure 32 Image displacement due to variations in relief (after Dial
and Grodecki 2003).
Many satellite images can be projected to a map projection at a constant height. Terrain
distortions are not corrected. Hence, objects above the reference elevation are displaced away
from the sensor while objects below the reference elevation are displaced towards the sensor
(see Figure 32). These terrain displacements affect the spatial accuracy of the imagery.
To mitigate against increasingly complex geometric errors a range of rectification techniques
have been developed (ERDAS 1999). The majority involve one or more of the following: an
51
increase in the number of ground control points, sensor construction information or a
Digital Terrain Model (DTM) and are either monoscopic or stereoscopic in approach.
Orthorectification is a means of correcting monoscopic images for relief displacement by
using a DEM. The DEM is collected from an alternative source (such as a basemap). As
illustrated in Figure 33 the DEM is used to adjust terrain displacement so that the resultant
image has correct planimetric coordinates.
Figure 33 Correction for relief displacement in a monoscopic
image by orthorectification (after Dial and Grodecki 2003).
Stereoscopic techniques also adjust for terrain displacement. However, stereo techniques
have the added benefit that they use the input stereo pair to create the DEM. The DEM
generated from the stereo pair produces better accuracy as the DEM is at a comparable scale
to the input image and it reflects the height differences at the time of image collection. For
example, if contours from a basemap are used they will not take into account buildings and
the landscape may have changed. Further, stereoscopic techniques can allow digitisation to
occur directly in 3d as opposed to 2d in monoscopic imagery.
Due to the minimum amount of terrain distortion in the application area neither stereo nor
orthrectificationwas necessary for the interpretative techniques employed.
2.2.6 Processing: Image display and visualisation
Image processing software considers remotely sensed data as a stack of spatially co-registered
layers over n bands, where n refers to the number of discrete bands collected by the sensor
(see Figure 34). Multichannel analysis is based on the premise that each pixel in each layer is
perfectly (or near-perfectly) co-registered (Sever 1988). Thus, any pixel in a scene can be
52
considered as a stack, each layer corresponding to one wavelength. Image processing systems
allow raster imagery to be manipulated in a variety of ways, by comparing and combining the
different data ‘layers’ without disrupting the basic structure of the image. The results of
mathematical operations can be used to create a new ‘layer’ in the stack. Hence, the integrity
of the original data sources is not compromised. These layers can also be used in the
production of a variety of false colour composites for human visualisation (Harris et al. 1999).
Figure 34 Landsat bands 1, 2, 3, 4, 5, 6 and 7 in a layer stack.
For visual analysis bands are commonly viewed as colour scales or colour composites.
Individual bands are commonly viewed as grey scale (which is a ramped colour scale between
white and black). The combination of three bands substituted for red, green and blue can
produce a colour composite image. For raw TM imagery this provides 120 (6x5x4) possible
colour combinations without band duplication (Landsat TM Band 6 is rarely used due to its
53
relatively low spatial resolution) and even more new layers are created through quantitative
techniques. Projecting band 3 as red, band 2 as green, and band 1 as blue yields a result close
to what humans perceive as ‘true colour’. All other types of RGB combination are referred to
as false colour composites (for example the Landsat 4, 3, 2 band false colour composite in
Figure 34).
It should be noted that each image (or image stack) consists of purely digital data. Although
the visualisation approaches employed mimic how the human brain handles visual imagery,
(hence a reliance on systems that provide variations in tone, texture, colour, pattern, shape
and size) other mathematical systems can be employed to extract meaning from imagery that
at first sight are not as visually intuitive as RGB or greyscale visualisation.
2.2.6.1 Human perception
Human perception of imagery is a complex procedure. People instinctively interpret qualities
such as tone, texture, colour, pattern, shape and size. Tone is a measure of the relative
amount of light reflected by an object on the ground and is fundamental to all other
recognition elements except colour. On a black and white photograph, tones range from
black to white and the spacing of variation in tone determines the texture of a portion of the
image. Texture can be used, for example, to separate disturbed soils from undisturbed
surfaces, to discriminate differences in vegetation cover that might signal the presence of
buried materials, or to discern areas of disturbed topography. Pattern refers to the
arrangement of features seen in an image (Ebert 1984 pp. 313-315; Teng 1997).
Pattern, shape and size are relative cultural interpretation flags. The spatial resolution of the
image is important for visual interpretation; this is a scale-dependent parameter. Some parts
of the scene are identifiable at all resolutions, whilst others only become apparent with
increased detail. Increasing magnification will eventually result in the component pixels
interfering with image clarity. However prior to this point smaller resolution features can
become clearer (see Figure 35).
54
Figure 35 Increased magnification on 15 meter Landsat
panchromatic imagery.
The human eye can discriminate between 20 to 30 shades of grey under normal viewing
situations (see Figure 26). Under the same conditions, it can discriminate a much larger
number of colours. A Landsat sensor can gather up to 256 shades of grey (8 bit) for each
band in its detection array: literally thousands of pieces of grey scale information are available
for analysis. The same numeric data from a scanner system can be combined to produce
millions of colours. An understanding of the perception of colour is, however, one of the
most important aspects for image interpretation. The human brain is very adept at
interpreting the visual spectrum when each band corresponds with its own colour (a Landsat
3, 2, 1, true colour composite). However, projecting Green as Red, Blue as Green and Red as
Blue gives an unfamiliar image (a Landsat 2, 1, 3 false colour composite, see Figure 36). The
same electromagnetic bands are used although they are presented in a different way.
Although this is a simple example, as the image structure is still recognisable, this is an
important concept for multispectral image interpretation. Colours produced in a false colour
composite correspond to variations in an objects reflectance for the projected wavelengths.
55
Figure 36 Comparison of a true and false colour composite made
up from the visual bands.
2.2.6.2 Histograms (radiometric enhancement)
Information about the possible range of Digital Number (DN) values within each band is
commonly represented in a histogram. For example, a histogram of a single band of data is a
representation of how the EM radiation energy at the collected wavelength is distributed in
two dimensions: the x-axis is the integer radiometric value and the y-axis is the frequency.
Figure 37 Histogram of Landsat TM Band 4.
56
Its shape indicates the contrast and homogeneity (or modality) of a scene. For example, a scene
with an homogenous surface with a low contrast will produce a histogram with a single sharp
peak. Conversely, images containing several distinct types of surface cover may show
multiple peaks. For example, Figure 37 is a multimodal example of Landsat TM Band 4 from
North Yorkshire. A = Water, B = Land. There is not enough spectral difference in Band 4 of
this scene to distinguish between the broad categories of urban and rural landuse. The
distribution of brightness values, in the lower quartiles, indicates the data is low in contrast as
highlighted by the summary band statistics.
Figure 38 Histogram equalisation of Landsat TM band 4.
The distribution of the histogram can be mathematically manipulated in a variety of ways to
improve image visualisation (and hence interpretation) or to focus in on certain areas in the
57
distribution. More complex analyses (such as Principal Components) use multiple histograms
from different bands (producing multi-dimensional distributions) to examine degrees of
correlation.
2.2.6.2.1 Histogram manipulation
Histogram manipulation techniques are used to increase or decrease image contrast across
the histogram. These manipulations can occur across the whole range of values or only in
specific areas. Contrast stretching expands a measured range of values (DNs) in an image to a
larger range to improve the contrast of the image and its component parts. The simple
algorithms in contrast stretching are very important for image visualisation and hence human
interpretation. An image rarely uses the full range of radiometric values available to it; in fact
they are normally clustered over a small range of values. Thus, an unstretched image can
appear to have low contrast to the human eye (Figure 38).
These enhanced images, as single bands or in false colour composites (see Figure 34), can be
interpreted with greater confidence as the increased contrast displays information which
would otherwise have been hidden. Common stretching methods are described in Table 1.
Stretch
Piecewise Stretch
Gaussian Stretch
Ramp Stretch
Logarithmic Stretch
Exponential Stretch
Histogram Equalising
Use
A linear stretch involving two or more lines of differing gradient to
enhance contrast in specific areas of the histogram
Transforms the original histogram to a gaussian curve (or normal
distribution) with the arithmetic mean, median and mode at 127 (for an
8 bit image)
A ramp stretch forces the cumulative frequency curve into a straight
line. This has the effect of increasing the contrast in the most
populated regions of the histogram
A logarithmic stretch expands the contrast of the dark component
while still maintaining contrast in the light range
Is the inverse of the logarithmic stretch
Applies the greatest contrast to the image by reducing the contrast in
very light or dark areas (at the tails of a normal distribution)
Table 1 Radiometric enhancement techniques
2.2.6.3 Spectral enhancements
Spectral enhancement techniques require more than 1 band of data. They can be used to:
•
Compress bands of data that are similar.
58
•
Extract new bands of data that are more interpretable to the eye.
•
Apply mathematical transformations and algorithms.
Table 2 outlines some common spectral enhancements:
2.2.6.3.1 Ratioing
Ratioing multispectral channels consists of dividing the radiance value in one channel by the
corresponding radiance value in another channel. Amongst other things ratioing is used to
improve the identification of materials. For example in Figure 22 the spectral response of
vegetation is at a maximum in band 4 and minimum in bands 1, 2 and 3. However the
geological spectra have a similar correlation to the vegetation wavelength in band 1, less in
band 2 and very little in band 3. In this instance the ratio of band 3 over band 4 should give a
low (c. 0.25) ratio for vegetation and a high (c. 1) ratio for geology (Rothaus and De Morett
1999).
Spectral Enhancement
Principal Components Analysis
Inverse Principal Components
Decorrelation Stretch
Tasseled Cap
RGB to IHS
IHS to RGB
Indices (ratio)
Description
Statistically compresses redundant data values into fewer
bands, which are often more interpretable than the source
data.
Performs an inverse Principal Components Analysis
Applies a contrast strech to the principal components of an
image
Rotates the data structure axes to optimise data viewing for
vegetational studies
Transofrms red, green, blue values to intensity, hue,
saturation values
Transforms intensity, hue, saturation values to red, green,
blue values
Performs band ratios commonly used in vegetation and
mineral studies
Table 2 Spectral enhancement techniques
An important consideration when using ratioing is how it will affect image storage. All optical
sensor systems normally collect integer values. However, rationing an image will produce non
integer values which will need to be stored at a higher level of precision. This will
dramatically increase file size, particularly if incorporated into a layer stack.
59
Figure 39 6 Components of a PCA on Landsat TM scene
(excluding band 6).
2.2.6.3.2 Principal components analysis (multivariate spectral enhancement)
Multispectral imagery bands are often highly correlated (i.e. they are visually and numerically
similar (see Figure 34)). Principal Components Analysis (PCA) is a tool used to remove this
redundancy and has proven invaluable in the analysis of multispectral data. A PCA
transformation can create new images that may be more interpretable than the original data
(Jensen 2000). It can also be used to reduce the number of bands used to describe the whole
image (used as part of the feature extraction process (see Chapter 5)). PCA uses multivariate
correlation statistics as its input data. It then calculates what components within the scene
determine the greatest and least variation. For example, Figure 39 shows the correlation for a
Landsat scene. Nearly 99% of the variance within the image can be described in the first four
components. Principal component 1 is usually a brightness component. This has high
loadings for bands 3, 4 and 5 suggesting this to be a near infrared dominated component.
Component 2 has high values for bands 3 and 4 suggesting a vegetation component.
60
Multivariate statistics characterise the underlying factors that define a scene by combining the
different bands statistically. A correlation matrix is calculated based upon the statistical
distributions of the bands and expressed as a ratio. All values are expressed between ±1. +1
indicates a perfect positive relationship. Conversely –1 indicates a perfect negative
relationship. A high correlation suggests there is a substantial amount of redundancy in the
image among the correlated bands. Conversely, a low correlation indicates no relationship
and uniqueness in the data. This technique is particularly useful for extracting key
information from hyperspectral data.
2.2.6.4 Kernel filtering (spatial enhancement)
Spatial enhancement techniques modify pixel values based on the values of surrounding
pixels. This technique uses a kernel: a moving matrix normally of dimensions 3x3, 5x5 or
7x7. This matrix mathematically alters the central pixel based upon the values of its
surrounding pixels. These filtering techniques are used with great effect in image editing
software, such as Photoshop. They are usually used within image processing applications to
enhance local variation (the extent is determined by the size of the kernel and pixels, see
Figure 40).
Figure 40 A moving ‘sharpening’ 3x3 kernel.
Table 3 describes a number of commonly used kernel filters.
61
Spatial Enhancement
Description
Convolution
Uses a matrix to average a small set of pixels across an image
Non-directional Edge
Averages the results from two orthogonal 1st derivative edge detectors
Texture
Defines texture as a quantitative characteristic in an image
Adaptive Filter
Varies the contrast stretch for each pixel depending upon the DN values in
the surrounding window
Statistical Filter
Averages pixels within a moving window that fall within a statistical range
Resolution Merge (Fusion)
Merges imagery of different spatial resolutions
Table 3 Examples of different kernel filters
2.2.7 Processing: Image classification
Image classification is the process of sorting pixels into a finite number of individual classes
or categories of information based upon their spectral characteristics (see Figure 42). If a
pixel satisfies a certain set of criteria for a class then it is assigned to that class (see Figure 41).
However, pixel mixing can significantly compromise the accuracy of the classification
procedure (see Figure 24).
Figure 41 The concept of classification (after Tso and Mather
2001 p. 4).
Increasing resolution does not always improve classification accuracy. Obviously
classification using six bands of Landsat imagery (excluding the thermal band) will produce
62
more accurate results than classification based upon a single band. More important is that the
bands are located over discriminatory areas of the object’s spectral signature. Researchers
who use hyperspectral imagery often find improved classification accuracy, but the
computational cost is expensive. To reduce these computational requirements irrelevant
bands should be discarded during the image extraction stage.
Figure 42 Feature space representation (after Tso and Mather 2001
p. 57).
From an intuitive basis one would assume that improving spatial resolution would improve
classification accuracy (see Figure 44). However, increasing spatial resolution means that
63
smaller objects can be identified thus increasing image heterogeneity which can decrease
classification accuracy. In essence increasing spatial resolution increases image complexity
requiring more complex classifiers for accurate analysis.
Figure 43 Spectral signatures of Haematite and Fir tree.
2.2.7.1 Spectral signatures
The ability to analyse and interpret remotely sensed imagery is due, in part, to the unique
spectral responses of objects during interaction with the EM spectrum. For any given
material, the amount of solar radiation that is reflected, absorbed or transmitted varies with
wavelength. This important property of matter makes it possible to identify different
substances or classes and separate them from their spectral signature. However, this signature
can fluctuate depending upon local environmental conditions, vegetation health and physical
properties (such as particle size which affects the percentage of energy reflected), although
the basic characteristics of the curve remain the same (Figure 43). Accurate identification
occurs by comparing standard reflectance responses against observed responses. Thus, the
64
resolution of the sensor becomes very important. A high spectral resolution (hyperspectral)
sensor will provide an abundance of sampling intervals allowing accurate identification of the
material (and for vegetation its current state of health). The lower spectral resolution Landsat
TM sensor only collects data over seven broad wavelengths. However, these wavelengths are
placed in the best locations to detect signature variation for geological and vegetation
applications.
Figure 44 Improving classification accuracy by removing pixel
mixing effects through increasing spatial resolution.
However, in practice, identification is rarely as simple as this. Laboratory spectra rarely match
with field measurements, as in reality each pixel will contain mixed responses for a variety of
65
different material types and is attenuated by the atmosphere. This is exacerbated by
archaeological features which provide a subtle variation to the background soil or vegetation
response (see Figure 45).
There are two main classification techniques:
•
Supervised classification.
•
Unsupervised classification.
2.2.7.2 Supervised classification
Supervised classification extrapolates information from a few ‘known’ areas (referred to as
training areas) derived from ancillary data (normally from ground-observation) to classify the whole
image. These areas are chosen carefully, as they need to fully represent the objects of interest
throughout the whole image for the best possible classification. The classification procedure
then evaluates the image on a pixel by pixel basis and assigns each pixel to a category defined
by its ‘goodness of fit’ to the training areas.
Figure 45 Ikonos imagery with archaeological sites outlined in
white (notice the increase in reflectance at the ‘sites’).
66
The success of supervised classifications is based upon the accuracy of the training areas and
the ability to discriminate among different classes of training area. If there is any feature
space overlap of the classes then there is likely to be ‘confusion’ during the classification
process. As already discussed in reality there should be confusion as most pixels contain a
mixture of objects. Pixel unmixing techniques and fuzzy classification systems are designed to
cope with these ambiguities.
2.2.7.3 Unsupervised classification
Unsupervised classification techniques assign pixels to categories based upon ‘natural’
groupings of data within the image (see Figure 46). The interpreters’ task is then to assign
these ‘natural’ clusters to a category. User input for unsupervised classification is limited to
the classification technique and the number of categories to be assigned. However, ‘cropping’
an image (reducing the image footprint) can significantly change unsupervised classifications
as the global statistics across the image are changed.
Figure 46 Unsupervised classification of a Landsat TM scene.
2.2.7.4 Image segmentation
Image segmentation is much like unsupervised classification. However, image segmentation
also utilises a spatial component. This technique groups neighbouring pixels which have the
same or similar spectral properties. This means that rogue classifications (i.e. lone pixel)
which commonly occur on boundary classes in supervised and unsupervised classifications
can be avoided. This technique can be particularly beneficial where objects in the scene do
67
cluster into discreet areas. There are a number of image segmentation techniques that require
different user input parameters, the most common of which are pixel-based region growing
algorithms (Beauchemin and Thomson 1997; Kartikeyan et al. 1998)
2.3 Image integration and problems of scale
Earth focused remote sensing applications collect synchronic and diachronic data about the
surface of the Earth. In this context, scale is related to resolution and most closely to spatial
resolution (or more accurately the instantaneous field of view). Spatial resolution refers to the
ability of a sensor to record and display spatial detail that can be distinguished from its
surroundings. The paradox of scale, in relation to remote sensing, is that by increasing
resolution the object becomes more detailed i.e. the closer we look at the world the more
detail we see (Goodchild and Quattrochi 1997 p. 1). This paradox was examined by
Mandelbrot (1967; cited in Pecknold et al. 1997 p. 364) when measuring the length of the
British coastline and led to the conclusions that measurements of an object depend upon the
scale at which it was examined. This led to the definition of a new field of research: Fractal
geometry.
These problems of representation are observed when changing map scales: a 1:25,000 map is
much more complex and detailed than the same area mapped at 1:250,000. However, both
representations are correct. The 1:250,000 map is generalised, at the cartographer’s discretion,
to improve interpretability. Indeed it is argued, justifiably, that the generalisation process
‘adds’ information as larger scale processes become more obvious. However, the process of
generalisation adds uncertainty to the data set (Goodchild and Quattrochi 1997 p. 4). Hence,
remotely sensed imagery are a much more flexible resource. If users have the tools and
techniques to operate with multiscalar data then they can aggregate or disaggregate in ways
that suit their own decision making and presentation process.
The Corona and Ikonos imagery have a spatial resolution approximately an order of
magnitude less than Landsat. The scale implications in the movement from medium
resolution to high resolution imagery could highlight a number of important issues for
archaeological detection and spatial patterning (Lang 1992).
68
2.4 Archaeological interpretation
Pictures, in the form of aerial photography and, more recently, multispectral imagery have
played a significant role in the development of archaeology and its management (referred to
as Cultural Resource Management (CRM)). These pictures convey information pertaining to
the size, positions and relationships of objects. Archaeologists trained in the collection and
interpretation of, predominantly, aerial photography possess a high level of proficiency in
deriving archaeological information from such images, even though they appear visually
complex. The human brain and visual systems are extremely good at extracting information
from imagery, particularly when it is derived from visual wavelengths. A large component of
this interpretation relies on the acquired knowledge of the interpreter (Bewley 2000; Bewley
and Raczkowski 2000; Wilson 2000). However, when images are formed using data from
outside the visual components of the electromagnetic spectrum our experience is not
adequate to interpret this data without understanding the fundamentals of how EM energy
interacted with the objects in the image.
Figure 47 Aerial Identification: Crop, shadow and soil marks (after
Greene 1990).
Aerial photography is a long-established means of assessing archaeological data in a
landscape context. Aerial reconnaissance may highlight crop marks, soil marks, parch marks
or shadow sites, each of which are formed by different natural and cultural processes and
each of which may only be recognised under specific environmental circumstances (see
Figure 47). This is one of the major difficulties with this form of evidence for landscape
survey. The visibility of any site depends not only upon the type of photography used
(oblique or vertical), but also upon environmental conditions and the type and level of
natural light available at the moment that the photograph was taken. For example, parch
69
marks may only be visible in the driest summer, at least in the visual wavelengths. Therefore,
an aerial photographic survey cannot be considered comprehensive. In European contexts
this problem is resolved by analysing archived photography taken over many years
encompassing a range of environmental conditions (Bewley 2000). However, where sites are
already known to exist, aerial photography can be used to highlight plan form or structure
that is essentially invisible on the ground.
However, the probability of spotting archaeological sites in existing vertical aerial
photographs depends very much on whether or not the imagery was acquired at a propitious
time of day and year or not. When aerial photographs are made specifically for archaeology,
the probability of detection can be raised by many orders of magnitude
(Scollar 1990 p. 26)
Aerial photographs intended for mapping purposes are normally taken with the camera lens
axis vertical using a highly accurate camera system. Conversely, hand-held aerial photographs
are normally oblique to illustrate vegetation pattern or shadow (Ebert 1984 p. 307). It is
essential to make a distinction between the application of vertical and oblique images as the
different perspectives provide very different results and interpretations.
Vertical or near vertical imagery, collected with the appropriate camera system, have very
good geometric properties. These allow the images to be used in quantitative programmes for
reconnaissance, desktop mapping or contour creation (from stereo pairs). The high
geometric accuracy allows many individual images to be accurately mosaiced together. Hence, a
whole application area can be viewed as a single pseudo image. However, obtaining bespoke
vertical imagery (particularly aerial imagery) can be an expensive process. When using existing
imagery (satellite or aerial) not collected expressly for archaeological purposes, the selection
process is much more restricted, as the observer has little control over the appearance of the
image. Furthermore, optimal conditions for archaeological discovery occur within a few
relatively small daily and seasonal time frames (see Figure 48) and it is unlikely that vertical
cover will occur during these timeframes (Scollar 1990 p. 28).
70
Figure 48 Cropmarks observed over a seven year period from
aerial sorties from the York office of English Heritage.
However, this premise is resolution and wavelength dependent. The majority of aerial
photographic applications have not focussed solely on the detection of archaeological
phenomena but rather at recognition or interpretation. Hence, the applications are focused at the
feature rather than site level. For example, Scollar (1990 pp. 37-46) discusses sites detected as
soil marks with reference to the response of negative features rather than the soil ‘halo’ that
surrounds them. However, if one focuses solely on the detection of archaeological activity at
the site level then the use of large-scale vertical photography (either from satellite imagery or
mosaiced aerial photography) is appropriate. Therefore, the window of opportunity where
imagery is pertinent for archaeological discovery expands, particularly for soil mark sites.
However, in well studied, European, contexts vertical imagery provides fewer new
discoveries.
Although vertical photographs are geometrically accurate, many archaeologists prefer to trade
geometric precision against archaeological interpretability by using oblique images. Oblique
images (especially those collected during optimum conditions) can highlight particularly
subtle changes in topography that are not immediately visible in vertical photography. This
trade off between precision and interpretability is of most import to archaeologists when the
71
photographic reconnaissance is part of a Cultural Resource Management (CRM) programme
(such as the National Mapping Programme: Bewley 1993). In an area with a known
archaeological background the geometric errors are acceptable, as the increased
archaeological interpretation is of more import (Scollar 1990 p. 28).
However, recent satellite platforms can allow collection on demand to specific days. This is
of particular import in areas where aerial photographic evidence is minimal and flights are
restricted. Although current satellite resolution does not equate with the spatial resolution of
a handheld SLR camera used at low elevation, they are able to determine areas of
archaeological activity, particularly those evident as soil marks. As demonstrated in Figure 48
soil mark evidence accounts for a small proportion of sites identified. However, this may be
linked with the desire to provide a high level of interpretation by archaeological aerial
photographers and interpreters. Soil marks do not produce the crisp, easily interpretable
imagery seen in crop mark evidence, thus soil marks may be overlooked in favour of more
responsive sites. Non soil mark sites are normally identified by the geometric properties
which distinguish archaeological features from the surrounding natural landscape. This is
further exacerbated by intensive agricultural regimes where the landscape is almost
permanently under crop. A similar point is made by Donoghue (2001 p. 558).
Given this information satellite imagery is likely to be an important resource for the detection
of archaeological residues in a Middle Eastern context for the following reasons:
•
The archaeological resource has not been studied in as much detail in the Middle
East as in Europe. Hence, vertical imagery is still likely to reveal many previously
undetected sites.
•
Archaeological residues do not arise from the same architectural traditions as in
Europe. Given the reduction in negative residues (fewer ditches, postholes and
palisades) European aerial photographic techniques may not be as responsive in
this environment.
•
Traditional land management techniques between temperate and semi-arid
environments are very different. For a large proportion of the year much of the
Earth’s surface in semi-arid environments will comprise of soil rather than crop
cover.
72
CHAPTER 3 CONCEPTS OF LANDSCAPE FOR ARCHAEOLOGY
3.1 Landscape archaeology: Introduction and definition
The landscape is a palimpsest on to which each generation inscribes its own impressions and
removes some of the marks of earlier generations. Constructions of one age are often overlain,
modified or erased by the work of another. The present patchwork of settlement …….. has
evolved as a result of thousands of years of human endeavour, producing a landscape which
possesses not only a beauty associated with long and slow development, but an inexhaustible
store of information about the many kinds of human activities in the past.
(Aston and Rowley 1974 p. 14)
Our cultural heritage lies scattered all around us. It is an intrinsic element of the landscapes
which we inhabit and is transformed over successive generations of inhabitation by both
cultural (anthropogenic) and natural processes. Thus, the present landscape represents the
manifestations of past and present decision making by individuals, groups and institutions in
society and the long term variable effects of natural processes. Understanding the interrelationships between cultural and natural systems and how they are transformed is necessary
to re-construct a theoretical understanding of archaeological landscape processes and how
they impact interpretation. Such an understanding can be gained by modelling the present
landscape and incorporating data from the historical, archaeological and environmental
record. One of the main difficulties of archaeological interpretation is to extend cultural
values to these models of physical residues and processes.
We suggest that a landscape approach is relevant to archaeology’s goal to explain
humanity’s past through its ability to facilitate the recognition and evaluation of the
dynamic, interdependent relationships that people maintain with the physical, social and
cultural dimensions of their environments across space and over time.
(Anschuetz et al. 2001 p. 159)
73
Landscapes have always played a fundamental role in archaeological enquiries. Initially they
provided a backdrop onto which archaeological data and interpretations were plotted and
evaluated. More recently, archaeologists have shifted their attention from single sites to
problems of regional change and variation. Employing landscape theories requires
archaeologists to reconsider how they evaluate the archaeological problem. For example,
non-site and off-site landscape approaches arose to combat the limitations of forcing
important archaeological residues into systems that consider the ‘site’ as the primary analytical
unit (Knapp and Ashmore 1999).
Archaeologists are placing increased importance on its multi-disciplinary nature. Therefore,
by inference, multiple researchers can produce different, but complementary, data sets for the
interpretative melting pot. A landscape approach provides a framework whereby these data
can be used collectively to form a more comprehensive understanding of the past.
3.2 The components of archaeological landscapes
…. what was once theorised as a passive backdrop or forcible determinant of culture is now
seen as an active and far more complex entity in relation to human lives.
(Knapp and Ashmore 1999 p. 2)
All landscape approaches address the fundamental nature of the relationship(s) between
people and the spaces they occupy (Anschuetz et al. 2001 p. 158). Researchers employ a
variety of classified variables to describe natural (e.g. ecological, geomorphological and
hydrological) and cultural (e.g. organisational, ritual, technological and ideological) features of
the landscape. Different interpretative or theoretical models use different variables in their
analysis.
The systemic frameworks of analysis are complicated by changes in cultural systems. It is
assumed that the adaptive strategies of each community leave separate but distinct traces in
the archaeological record and, furthermore, that these communities are constrained by their
local and regional ecology and topography (Rossignol 1992). For example, the framework of
analysis and theory underpinning hunter-gatherer communities is different to that of
sedentary communities and how and what they exploit are defined by their regional ecology.
74
3.2.1 Natural landscapes
Natural landscapes, in an archaeological context, encompass all elements of the landscape
which are crudely believed to be non-cultural. Hence the lithosphere, biosphere, hydrosphere
and atmosphere are all aspects of the natural landscape. However, culture can impact on each
of these systems so natural landscapes are, therefore, not completely independent from
cultural systems (Clarke 1978; Butzer 1982; Waters 1992).
The natural landscape cannot be viewed as a passive entity. It guided movement, offered
resources and formed backdrops for the human presence. Furthermore, the form of the
landscape in each period itself influenced the processes of inhabitation (Barrett 1995). Hence,
many archaeologists consider it important to model aspects of the natural environment to
improve their understanding of the archaeological context. From the point of view of
landscape survey it is difficult to model both the biosphere and atmosphere at the local or
regional level through time without significant supporting material from environmentalists.
However, some changes in both the lithosphere and hydrosphere are evident on the surface
of the Earth in the form of floodplains, terraces, relict channels and other topographic
features. Furthermore, although drift geology may have changed significantly since the
Pleistocene the solid geology has not. Hence, it is theoretically possible to retrogressively
model aspects of past natural landscapes.
3.2.1.1 Ecology
Ecology is the holistic study of living organisms and their environment. Ecology has many
sub-disciplines including those of landscape, historical and human ecology (Winterhalder
1994). Human ecology, particularly the contextual archaeology of Butzer (1982), applies
specifically to how human agents interact with other agents and ecological systems (see
Figure 49). Human-environment interaction thus covers all relationships between people and
the environment, whilst also encompassing the reciprocal influences of human behaviour on
the ecosystem and how this maintains stability or causes change. The use of human
ecological analysis has led to the deductive process of settlement ecology where models are
used to predict where human agents will settle in a landscape based upon ecological
indicators and assumptions about what they will exploit. Ebert (1988) discusses the
applications of satellite techniques for ecosystemic analysis by looking at variations in
environmental diversity.
75
Figure 49 Schematic of interactions in a human ecosystem
highlighting the relationships between the cultural and noncultural environment (modified from Clarke 1978 p. 133; modified
from Waters 1992 p. 5)
3.2.2 Landscapes and space
All archaeological sites were once areas of human activity that took place within a landscape
context. Archaeologists attempt to describe and understand the nature of these activities
through their relationships with the landscape and each other. This is to observe and
understand the delineation of the formal variability, temporal loci and spatial loci of sites,
activity areas and communication networks (Golledge and Stimson 1987 pp. 5-6).
3.2.2.1 Distribution and settlement patterns
If a sampling strategy has been employed that will identify spatial patterning within a
temporal framework (see section 3.4.3) then the next task is to understand why the patterning
occurred as it did. Patterning in the landscape is seen as a reflection of the ways that people
respond to the agents of ecology, economy or society. Archaeologists should not only
76
attempt to understand these agents but also the interaction between them over space, time
and within the context of society (Knapp 1997 p. 14). Ethno-archaeological sources predict
that spatial patterning within a landscape is rarely the product of actors within a single
cultural system. Rather, there are spheres of multiple interactions between localised actors
through the aforementioned agents. Furthermore, these systems operate at a variety of
differing spatial and temporal scales (Wandsnider 1992b). Hence, the analysis of settlement
patterns requires the integration of different landscape data collected at different scales.
3.2.3 Landscapes and scale
The impact of scale on spatial analyses has been recognised by the social and natural sciences
for more that forty years (Marceau 1999). In archaeological research there is, however, no
real consensus on what spatial scales are appropriate for specific archaeological analyses
beyond the broad terminology of ‘micro’, ‘meso’ and ‘macro’. These terms, with differing
emphasis in their interpretations, are used to broadly differentiate patterning from the local to
the landscape level (Allen 2000 p. 101). However, each of these terms tends to be used
during interpretative stages and is thus a product of a theoretical stance. Conversely many
data sets are now being collected with a more thorough understanding of how they should be
integrated with other data (for examples see Shennan 1985; Francovich et al. 2000). Hence,
archaeology has formed both empirical and synthetic appreciation of scalar interactions.
Traditionally, scale has been utilised as a convenient tool to partition the landscape (i.e. site –
micro, inter-site – meso and regional – macro) in order to help facilitate the production of
syntheses or narratives. More recently research has focussed on the relationships between
scales of synthesis, data scales and data resolutions (Crumley and Marquardt 1987; Rossignol
and Wandsnider 1992; Ramenofsky and Steffen 1998; Allen 2000).
3.2.4 Cultural landscapes
The cultural landscape is fashioned from a landscape by a culture group. Culture is the
agent, the natural area is the medium, the cultural landscape is the result.
(Sauer 1938 p. 46; cited in Anschuetz et al. 2001 p. 164)
Understanding the inhabited landscape requires more than the description of the form of
that landscape’s organisation. Inhabitation concerns perception, practice, and experience. It
77
addresses the history of the formation of the communities who moved through and who
worked the land in its different aspects. The context, which included all the material aspects
of the contemporary landscape, natural as well as cultural, framed human actions and
provided foci for them. Hence, the physical context was built up and changed; elements
endured and embodied different values and meanings for those who moved through it at
different times. For example in the Homs area, the seasonal ‘rams’ (depressions which collect
water) have a different value to farmers and pastoralists.
If the actions of actors within a landscape are studied with regard to the physical, material
and cognitive way in which people manage their natural and cultural environment, then social
organisation may be extrapolated in terms of the material and ideological landscapes in which
people acted (Barrett 1995; Knapp 1997).
Sauer’s definition of cultural landscapes is important in this context as it highlights the
transformative power of culture upon the landscape itself. Although culture can be identified
as a mechanism for landscape transformation, archaeologists consider landscapes as
constructs that embed information on the structure and organization of societies. Therefore,
the physical landscapes themselves offer insights into social and communal structures and
practices. The difficulty is how to access past cultural actions from data embedded in the
present landscape. The difficulties in applying many of these concepts become self-evident
when reading, as a landscape archaeologist, Chatwin’s (1987) fictional(?) traveller’s narrative
of Australian Aborigines and their relationships with their landscapes. Multiple
interpretations are particularly evident when he is exploring the conflicting cultural tensions
within and between native and non-native Australians.
3.2.5 Discussion
The literature shows that by using the same data sources a variety of different landscape
interpretations could be constructed depending upon the theoretical, analytical and
interpretive approach taken. Fisher and Thurston (1999 p. 631) noted that some of the most
productive landscape research employs different theoretical models to juxtapose different
research goals, hence producing complementary interpretations with multiple interpretative
strands (Hodder 1993; 1999). Within the context of this research it is essential that many of
the themes and interpretations fulfil these diverse theoretical, analytical and interpretative
criteria. Satellite imagery is an important resource for the evaluation and interpretation of the
78
cultural and natural components of archaeological landscapes. However, it is only when
satellite-derived themes are integrated with other information sources structured by a
theoretical research agenda that their full potential will be realised.
3.3 Transformation of archaeological landscapes
Regrettably, neither the historic record nor the archaeological record gives up its secrets about
the past easily. Each must be handled with great care by the investigator seeking to infer
past behaviors, for the evidence that survives has been changed in many ways by a variety of
processes. To make justifiable inferences the investigator must consider and take into account
the factors that have introduced variability into the historical and archaeological record. The
factors that create the historic and archaeological record are known as formation processes.
(Schiffer 1987 p. 7)
The study of formation and deformation targets the interaction between cultural and natural
processes, which thus determine the preservation of the archaeological record (Butzer 1982;
Schiffer 1987; Waters 1992). This leads to an understanding of the dynamics of the formation
process that in turn influence how the archaeological record is perceived and understood
(Rossignol 1992).
As Wandsnider states (1992a p. 96) ‘we understand that if the cultural deposition rate is faster
than the natural deposition rate a palimpsest deposit will result….Conversely, if the natural
deposition rate exceeds the rate of cultural deposition, then burial results’. However, these
rates of natural and cultural aggradation and degradation are not homogenous; rather, they
are spatially discrete and temporally variable. Furthermore, their causation may be traced to
other anthropogenic actions in the region.
3.3.1 Cultural transforms
Cultural transforms involve the deliberate or accidental activities of human beings as they
conduct their daily activities. Over time, a variety of factors change the way that societies
interact and structure their environment. Settlements are abandoned as, for example, regional
resources are exhausted, internal political organisations change or are supplanted. Such
actions can leave ‘snapshots’ of past systems (as is observed in the so-called ‘Dead Cities’
79
near Aleppo in Northern Syria). However, these examples are rare; the majority of
archaeological residues are transformed by the more pervasive acts of inhabitation over time.
Each successive generation alters the fabric of the archaeological record: structures can be
dismantled and the material re-used. The hinterland can undergo various re-structuring
regimes as systems of land tenure develop or new systems are imposed (such as Roman land
cadastration or the English enclosure awards). The hinterland itself has been subject to a
range of destructive (and preserving) agricultural activity for millennia. At a finer scale
artefact reuse modifies the original functional context and can subsequently cause erroneous
interpretation. For example, broken pottery is not just ‘discarded’ locally but can be re-used
during land fertilisation (night soiling), leading to a potential misinterpretation of a pottery
scatter as a site. The scale and impact of the resultant transforms can vary greatly.
Extensive cultural modification has occurred over the past century. Social and technological
changes have significantly impacted on the fabric of urban and rural landscapes. Urban
expansion and the use of more destructive agricultural techniques (deep ploughing, boundary
re-organisation and increased drainage), have caused immense destruction of archaeological
residues. Although these recent modifications have, arguably, destroyed more archaeological
residues than at any other time they are still part of a larger scale process of change. The
archaeological resource has been subject to a range of cultural modifications (for example
ploughing, settlement abandonment, land tenure re-organisations and re-use of structural
material) for as long as human activity has occurred. It is the responsibility of archaeologists
and cultural resource managers to ensure that appropriate mitigation or preservation systems
are in place prior to any destruction.
Although specific will be discussed in greater detail later in the research cultural transforms in
the study area include:
•
Past obliteration of early sites through changes in landscape management and
agricultural practices (i.e. roman cadastration).
•
Mudbrick quarrying (shallow scoops associated with tells).
•
Modern agricultural practices (bulldozers, ploughing and irrigation).
•
The modern transport of archaeological soil horizons for topsoil.
•
Cairn creation and clearances in the basalt.
80
3.3.2 Natural transforms
The archaeological record affects and is affected by its immediate surroundings: its
environment or matrix (Stein 1992; Schiffer 1996). The immediate environmental matrix
determines the medium of preservation, and the attributes of the material determine if the
material will survive within this matrix. Thus, understanding the immediate matrix leads to
increased appreciation of material bias governed by artefact-environment interactions. There
are three main agents of deterioration:
Chemical Agents: Chemical agents are pervasive in both systemic and archaeological
contexts. The atmosphere and the soil contain water and oxygen, which are sufficient
to propagate many chemical reactions, the most common of which are oxidation and
reduction. Irradiation of materials by Sunlight induces photochemical degradation,
particularly within organic compounds. Furthermore, increase in heat increases
reaction rates. Acids and Bases react chemically with a variety of material types and
occur naturally both in the atmosphere (for example, CO2 reacts with water to form
the weak Carbonic Acid H2CO3) and within soils.
Physical Agents: Physical agents of deterioration are ubiquitous. Volcanoes, hurricanes,
earthquakes, landslides, floods and other natural disasters affect artefacts and
structures over large geographical extents. Water is a particularly pervasive
deteriorating agent: streams, rivers and the sea erode the physical matrix, reworking
and abrading any archaeological material, drainage leads to erosion and moisture is
the normal medium of decay. Wind can erode the most labile elements of a soil
structure which causes formation patterns to combine and collapse while the redepositing material may obscure some other residues. Temperature changes have a
significant impact on soil expansion and contraction providing fissures through
which residues can fall. Even the action of gravity on a steep slope can displace or
bury archaeological residues.
Biological Agents: Living organisms are the principal agents of long term biological
decay. Bacteria and fungi initiate the process of decay or rot and can survive within a
variety of extreme environmental conditions. Animals ranging from insects to
mammals can cause a variety of pre and post-depositional disturbance. Plant roots
and burrowing animals cause particular damage to the local matrix, disturbing artefact
sequences.
81
The rate at which these processes operate is dependent upon local conditions. As the local
matrix is modified the conditions change to promote or inhibit particular agents (Schiffer
1987). These effects are variable over different environmental, ecological and temporal scales.
For example, a criticism of using ‘soil-type’ as a variable within any historical modelling
exercise is that the current soil-type may not be representative of any past conditions as the
interaction of biological, physical and chemical agents in conjunction with anthropogenic
actions have changed the nature of the soil. Furthermore, there is a more philosophical issue
on the nature of classification: do modern soil science classifications have any relationship
with how archaeological agents classified soil (if, indeed, they did).
The micro-environmental conditions determine the differential preservation of the
archaeological record. However, further modification and masking can occur through meso
and macro environmental effects. The primary areas of interaction are:
1. Site mesoenvironment: the topographic setting and landforms of the area utilised
for subsistence.
2. Site modification: pre and post-depositional disturbance through the actions of
running water, frost, expansion and contraction, deflation, animals (including
humans), plants and gravity.
3. Site destruction and dispersal: through the same agents.
Geoarchaeology is the study of the effect of these natural processes (Scudder et al. 1996).
One aim of geoarchaeology is to understand how natural processes have transformed the
archaeological record and how this impacts upon interpretation. For example, a multi-period
pottery scatter makes much more sense when one is informed that significant deflation has
occurred in the landscape. Alternatively, an area that is considered to be sterile during surface
survey does not actually provide negative evidence when one is informed that it is part of an
aggrading floodplain.
Many anthropogenic actions can have a significant impact upon the natural environment.
Prehistoric woodland clearances have been demonstrated to cause massive soil erosion
events that substantially affect the hydrological system and the topography of a regional river
catchment (Macklin and Needham 1992; Macklin et al. 1992). Thus, the inter-relationships of
natural and cultural processes can modify the landscape in many complex ways.
82
3.3.3 Discussion
Landscapes are the manifestation of tensions between Cultural and Natural (C and N)
transformations working at various spatial and temporal scales and with varying degrees of
positive or negative interaction. The development of landscapes is complex, and it is difficult
to extract the specific impact of either cultural or natural events on this development (Layton
and Ucko 1999 p. 2). However, the relationships between natural and anthropogenic agents,
although complex, have been hypothetically modelled (see particularly the works of Butzer
1982; Crumley and Marquardt 1987; Waters 1992; Crumley 1994). Cultural formation and
deformation influence the structure of the archaeological record, ranging from larger scale
site and hinterland modifications (or even destruction) down to micro scale discard strategies.
It is also possible for ‘sites’ to go through a cyclical process of birth (inception), life (use),
death (disuse) and then re-birth (re-use) (Langran 1992) creating a palimpsest of
discontinuous occupations that may be difficult to disentangle (see Figure 53). Natural
formation processes in such forms as erosion, mixing and burial interact with cultural
processes and further modify them. Furthermore, the nature of the archaeological material
undergoes long-term degradation based upon physical, chemical and biological agents that is
a function of the localised environmental matrix (Rossignol 1992). The long term
modification and even destruction of archaeological residues has a profound impact on their
interpretation. Settlement patterns become increasingly fragmentary and those from some
periods may have lost so much of their original structure that coherent and valid
interpretation is now impossible (Taylor 1972).
In some environments disentangling and interpreting these natural and cultural formation
and deformation sequences is extremely complex (for example see Figure 50 and Figure 51).
Ebert (1988) compares a number of projects that have determined depositional and postdepositional processes from satellite imagery. The study area (see Chapter 4) has undergone a
range of natural and cultural processes that have occurred over numerous millennia
including; deflation, fluvial burial, modern and past anthropogenic practices (examples
include irrigation, topsoil movement, bulldozing, land reorganisation and mudbrick
extraction pits).
83
Figure 50 Formation and deformation process in the marl.
Figure 51 Natural and cultural formation and deformation
processes on Tell Nebi Mend.
84
3.4 Practice and archaeological landscapes
The vast majority of archaeological residues are buried and essentially invisible to the human
eye. However, traces can be identified through such evidence as upstanding monuments,
relict fieldsystems, clusters of artefacts, chemical and physical residues and variations in
surface relief. Archaeologists employ two main approaches to identify and understand
archaeological residues: Survey (with no or limited destruction) and excavation (destructive).
Each approach uses a variety of different techniques that sample different attributes of a
landscape in progressively increasing detail.
Survey itself comes in many different forms each providing different results. It is often the
first stage of a long-term archaeological project providing an overview of the range and
nature of the archaeological residues. However, survey is not just the preliminary stage for
future intrusive excavations; a well designed survey strategy will address questions that
excavation can never answer. Realistically only regional survey provides the opportunity to
study and interpret related archaeological residues dispersed over space (Banning 2002 p. 1).
In landscape survey a block of land is examined with the purpose of detecting ‘sites’ or loci of past
activity. Background information is collated through a desk-based assessment. The areas are
then examined during site survey where intensive fieldwalking, geophysical prospection and
sub-surface sampling techniques can be used to further characterise (or recognise) the residues
or reveal areas that may have been missed. Survey of this nature is widely used throughout
the world. Many archaeological investigations only use the two broad techniques of landscape
and site survey as they provide the archaeologist with a wealth of data for interpretation.
However, for more detailed understanding (or identification) then the many techniques of
excavation are required (Roskams 2001). For the requirements of this research excavations will
not be discussed in detail.
3.4.1 Survey objectives
The results of archaeological survey depend on the objectives it was designed to achieve. The
survey design or methodology inherently biases any data collected. For example, a methodology
that focuses on identifying areas of past settlement will identify few, if any, temporary or
nomadic sites. Therefore, it is important to record the survey objectives in any metadata
descriptor of the data set. Recording such information will reduce erroneous application of
the survey data by other researchers. For example using the previous case of surveying past
85
settlements, these data should not be used to determine the range of site types within a
landscape as there may be significant bias in the collection methodology.
All surveys do not aim to achieve the same ends. There are many different survey objectives,
but the majority can be generalised into the following groups:
Reconnaissance survey: (Detection) Primarily designed to detect all the positive and
negative archaeological evidence within a study area.
Evaluation survey: (Recognition) To assess the archaeological content of a landscape
using survey techniques that allow either field-prospection, statistical hypothesis
building or the identification of spatial structure to occur.
Landscape research: (Identification) To form theoretical understanding of the
relationships between settlement dynamics, hinterlands and the landscape itself.
Cultural Resource Management (CRM): (Protection) Primarily designed for
management of the available resources. CRM applications are not necessarily distinct
from other survey objectives although they may be conducted as part of a more
general information capture system.
The aim of landscape survey is to detect areas of past human activity. Two principal data
collection techniques are employed:
•
Desk based assessment.
•
Ground reconnaissance.
Desk Based Assessment (DBA) is the collation, integration and analysis of all archaeologically
pertinent material for the study area. Ground reconnaissance is the process of field data
capture. Most ground reconnaissance techniques employ solely surface collection. This is
ideal where post-depositional process actually leave an assemblage on the surface. However,
this is rarely the case and many surface assemblages are the result of modification from
ploughing or exposure by erosion. Only experimental work will help to understand the
relationships and transforms from sub-surface assemblages to surface assemblages. However,
for residues revealed by ploughing there is a strong correlation between surface and
subsurface remains (Lambrick 1977). In some situations sub-surface collection will occur by
test-pitting or shovelling to access buried residues.
86
3.4.2 Desk based assessment
Desk based assessment involves the thorough researching of all existing information. The
Institute of Field Archaeologists (IFA) defines it as:
... a programme of assessment of the known or potential archaeological resource within a
specified area or site on land, inter-tidal zone or underwater. It consists of a collation of
existing written, graphic, photographic and electronic information in order to identify the
likely character, extent, quality and worth of the known or potential archaeological resource
in a local, regional, national or international context as appropriate.
(IFA 2001)
Sources that are normally considered for reference during a DBA are:
•
Regional and national site inventories.
•
Public and private collections of artefacts and ecofacts.
•
Modern and historical mapping.
•
Aerial photography and other remote sensing.
•
Historic documents.
•
Geo-technical information (such as soil maps and borehole data).
Desktop studies allow a broad understanding of the landscape and thus aid the
implementation of the project design, saving valuable time and money.
It is within this framework that satellite imagery plays an important role. Medium spatial and
spectral resolution satellite imagery (such as Landsat) allows the determination of geotechnical information in scales that are more useful for archaeological enquiry than many
available map sources. High spatial resolution satellite imagery can supplant modern mapping
sources by providing a mapping base which has not been subjected to generalisation, and
hence loss of detail. This imagery also shares many of the characteristics of aerial
photographic imagery but with a much larger synoptic footprint.
87
3.4.2.1 Prospection
Prospection techniques optimise the probability of detecting archaeological residues. As
opposed to statistical surveys (see section 3.4.3.1), which use more elaborate and timeconsuming techniques, prospection takes advantage of any available information which will
improve the chances of discovering archaeological material. Historical and geo-technical
documentation are a valuable resource for prospection as they can indicate areas to exclude
from a survey. For example it is a waste of resources to intensively survey for prehistoric
remains on a terrace that was formed in the 12th Century AD. However, rarely do these
sources provide accurate spatial referencing for specific events. To address this issue, data
collated and created during the DBA are modelled from manual interpretation or computer
classification in a geo-referenced format.
Manual interpretation correlates the information layers (and possibly enhances them using
computer techniques) and, after reference to an image interpretation key, is used to describe
and interpret as many potential archaeological residues as possible.
Computer classification employs a variety of different digital classification techniques to
provide archaeological information. The approach used is determined by the methodology
employed by the user. If, for example, the user believes there is an identifiable archaeological
spectral signature, then one might create training areas and classify the information
accordingly. Alternatively, if it is believed that the archaeological residues exhibit a higher or
lower statistical distribution relative to a ‘background’ reading then a variety of regional or
local statistical modelling exercises could be undertaken. It may also be necessary to combine
multiple themes using, for example, multivariate techniques. Whatever the approach
archaeologists tend to refer to these mechanisms of classification as predictive modelling.
More formally they can be called Exploratory Data Analysis.
Whatever technique is used ground observation is required to check the validity of the
detection exercise and to collect more data to further characterise the residues under
investigation. Prospection is used to provide maximum archaeological return for minimum
investment of time and resources and is applied in a structured way that intersects with areas
of high archaeological probability.
88
This technique by its very nature creates biases in the collected data as areas of high
archaeological probability are examined in preference to other areas. Hence, the results of
prospection should only be generalised with care. If the assumptions used in prospection
have not been statistically ground tested then erroneous landscape generalisations could
occur. However, generalisation and interpretation is not the primary goal of prospection;
rather, prospection is used to characterise particular archaeological residues within the
landscape.
3.4.2.2 Predictive modelling
Predictive modelling is based on determining the correlation between known sites and
environmental features in a particular region and projecting the knowledge to
environmentally similar areas (Warren and Asch 2000 p. 6).
Predictive models attempt to propose a:
...simplified set of testable hypotheses based on either behavioural assumptions or on
empirical correlations which at a minimum attempts to predict a loci of past human
activities resulting in the deposition of artefacts or the alteration of the landscape.
(Judge and Sebastian 1988 p. 33)
Archaeological predictive modelling has its basis in the settlement studies carried out by
archaeologists in the 1950s and 1960s. In the mid 1960s locational concepts were borrowed
from geography and provided a supporting body of theory. It was also at this time that
archaeologists began to appreciate the importance of ecological and environmental variables
in understanding settlement variability. Since then these variables have been used to great
effect (Dalla Bona 1994).
Predictive models are either inductively or deductively derived. Inductively derived models
test hypotheses against a database usually using multivariate classifiers. Hence these models
are subject to any bias existing in the database. Deductively derived models begin with
theories predicting human behaviour (for example, all settlements will be close to water).
While deductive models better encompass the range of human behaviour they suffer from
changing interpretations and theoretical viewpoints (Dalla Bona 1994).
89
Those models that focus on the physical environment and its effects on settlement have been
referred to as man-land relationships (Kvamme 1988 p. 332). The examination of sitecatchments, topography, vegetation and other environmental features are major elements of
this approach. On the other hand man-man relationships refer to analyses that assess the
importance of the human or social environment in structuring patterns of settlement. These
analyses focus on such themes as central place theory, rank size theory and population
distributions over the landscape (Banning 2002).
Finally, all predictive modelling exercises can be broken down into three primary stages (after
Dalla Bona 1994):
1. Hypothesis development, organisation and data collection.
2. Initial model development and testing.
3. Continued application of the model and on-going refinement.
Most predictive models stop at the secondary stage. However the tertiary stage can be viewed
as the most important and is ideally a never ending process whereby the predictive
robustness of the model is increased. Ebert (1988) compares a number of American
predictive modelling exercises employing satellite imagery as an analytical layer.
The analysis of landscape information has traditionally occurred in a case by case fashion.
However, GIS applications allow archaeologists the ability to reproduce their methodologies
over a discontiguous area and, furthermore, re-evaluate analyses after more information has
been collected. The use of such a toolkit will allow the continual development of hypotheses
to occur.
3.4.3 Ground reconnaissance
Fieldwork programmes that collect landscape-orientated information are not focused on
excavation techniques; rather they are focused upon large scale sampling strategies that
collect physical and artefactual data and place extant residues into a contextual framework
(Plog et al. 1978; Ammerman 1981). The developmental history of the application areas needs
to be researched, not only to integrate any other archaeological works, but to understand the
extent to which surface material may have been disturbed by natural or cultural
transformations. This information aids the identification of areas where the methodology
employed will not work (for example fieldwalking on an aggrading alluvial floodplain). Many
90
regions contain a variety of landscape types and a single methodology may not be
appropriate. Therefore, flexible methodologies are employed that can respond to the unique
developmental histories of different landscape zones. However, such flexibility has an impact
on the Archaeological Information System (AIS), the associated recording system and the
interpretative frameworks as these data sources will need to be interpreted as a single coordinated information set.
3.4.3.1 Statistical survey
Sampling techniques are used because it would be extremely expensive and time consuming
to apply the same strategy across the whole landscape. Landscape surveys employ various
sampling methodologies during data collection (i.e. random, non-probabilistic, cluster,
multistage and total (Judge et al. 1975; Schiffer et al. 1978; Banning 2002)). These are
predominantly determined by the research goals of the project, the accessibility of sites in the
landscape, the types of analysis which will be performed upon the data and cost. A sample is
a subset of a population; if the sample is representative of the population under study then
the characteristics of the population can be estimated using attributes of the sample. If this
assumption holds for archaeological survey then it is not necessary to detect all archaeological
residues as processes that affect the whole population can be extrapolated from the sample.
The problems of sampling in an archaeological context are clear, primarily concentrating on
the bias of different strategies and the fact that any sample, however well defined, remains
just a sample. Its results may be broadly applicable to the rest of the area, but will never be
entirely so. Furthermore, no sampling technique is total, even the misnomer of ‘total’ survey
rarely samples everything.
In total survey the whole landscape may be evaluated by transects or quadrats, but only a
small sample of each transect is actually surveyed (Banning 2002 p. 167). The intensity of the
survey reveals a level of probability that the survey has adequately characterised the required
residues. Intensity does not just refer to increased coverage but also to re-surveying in
different environmental conditions, times of day and with different crew members (Schiffer et
al. 1978; Shennan 1985 p. 10; Banning 2002 pp. 224 - 225). The results are compounded by
the problem of detectability which is the possibility that an observer may fail to notice a target
even though the survey intersects with the target (Shennan 1997). This could be due to the
target being completely obscured, for example, by alluvium, transparently obscured, for
example, by vegetation or that the observer does not have the skills to recognise the target.
91
The visibility of sites and artefacts can vary widely predominantly due to factors such as
seasonal weather patterns, vegetation cover and land use regimes. Results tend to be more
reliable from long-term projects where the region is covered repeatedly. However, it is
financially untenable to conduct a total survey and be statistically confident that all of the
archaeological residues have been characterised, never mind understood.
Even with these statistical limitations, landscape sampling provides a data resource which not
only complements excavation data but, for certain types of enquiry, transcends it. It is a
particularly effective means of analysing spatial organisation. Hence, a clearly defined
sampling strategy with a robust research design will facilitate landscape interpretations which
are used to develop more refined fieldwork approaches.
Sampling reduces the costs of both fieldwork and analysis. As sampling is selective in its
approach to the landscape, areas that are unsampled remain undisturbed for future
researchers. However, in order to define what form of sampling is appropriate one needs to
understand the general spatial and temporal attributes of the population, which, in the initial
phases of an unsurveyed landscape is difficult.
Unlike prospection, statistical surveys are commonly used to directly infer the following
information (after Banning 2002):
•
Site densities (by type and/or period).
•
Artefact densities (by classification and/or period).
•
Spatial patterning of residues.
•
Material diversity.
This information is then used within other theoretical systems to indirectly estimate
population change, ecological preference, settlement systems and many of the other cultural
products already discussed. Equally importantly the data can be used to refine the hypotheses
and assumptions defined during the prospection phase.
To address the statistical limitations of sampling, surveys are often undertaken at varying
resolutions and intensities. Rapid low resolution surveys can be conducted to infer large scale
patterning, these are then refined by conducting higher intensity surveys around areas of
92
interest. In such a situation it is implicit that the information system can manage the data
effectively to ensure consistency in interpretation and generalisation. Surveys with multiple
goals can thus be realised using different survey techniques.
3.4.3.1.1 Consistency
Consistency within the survey is essential in order to confidently interpret survey results
across the landscape. However, researchers are increasingly combining the results of different
surveys in order to look at archaeological variations at even larger scales (this is an emerging
body of work referred to as meta landscapes (Raczkowski 2000)). It is inevitable that different
techniques will be used by different surveys. Hence the integration of multiple surveys into a
single body may well be significantly undermined by differences in survey technique.
Although ensuring consistency for other future researchers may not be a high priority for the
survey team it is possible to build in levels of consistency for the longer term:
•
The local or national CRM body may insist on core information to facilitate other
enquiries.
•
Document all aspects of the survey strategy and methodology and include them
during data deposition.
•
Where practicable, if inter-project goals have been identified, incorporate them
into the survey methodology.
3.4.3.1.2 Artefacts
Artefacts provide the major chronological frameworks for the archaeological record and
insights into the social, political, artistic and economic backdrop. Differential preservation of
artefacts is dependent upon the local matrix, temperature and material of the object. Rarely
do conditions prevail which are conducive to the retrieval of all the materials originally
present. This differential preservation will normally give large biases in the collection of
artefacts, usually in the favour of inorganic materials.
However, in the context of detection, the mere clustering of artefacts is as important as the
attributes each artefact contains. From the spatial patterning of artefact distributions within
the landscape, sites are ascribed. The types of artefact may indicate the date, duration, form,
93
function and variability of the site. Further, relationships between sites are extrapolated to
produce thematic understanding of past societies.
3.4.3.1.3 Ecofacts
Only recently have archaeologists attempted to place archaeological sites into their
contemporary environment and view the human action as an inherent part of a broader
ecological system. A reliable date for a site allows coarse interpretations of the local climate
by reference to environmental indicators contained within secure deposits (such as sediment
or ice cores). Finer grade questions follow; local flora and fauna can be reconstructed by their
archaeological remains. Some of these plants, insects and animals live in specific
environmental conditions and therefore their presence is a good indication of the local
habitat at that time. Further, analysis gives indications of subsistence and diet, which can be
extrapolated to gain an understanding of how the ecology was exploited.
3.4.3.1.4 Other techniques
With the emphasis on cost and time, excavation is a tool that is rarely used in landscape
studies due to its inherent expense. However, a battery of remote sensing techniques can
detect sub-surface features quickly and cheaply. The results of these geophysical surveys,
particularly on single-period sites, complement surface survey by improving spatial control.
3.4.4 Summary
It is a truism that the survey design and methodology constrain the results that the survey can
be expected to yield. Consequently, surveys are designed to produce the results that are
required for interpretation and analysis. This design process may take many years as most
surveys start with an imperfect knowledge of the problem and the resource. Increased
understanding occurs over successive seasons of fieldwork and interpretation. Hence, the
relationship between goals, data and methodology is hermeneutic.
Commonly, surveys are designed to optimise the discovery of archaeological residues
(prospection), to use statistics to extrapolate attributes of a population from a sample or to
detect spatial patterning. One aspect of this research is to address how high resolution
satellite imagery can impact upon survey design. It should be expected that high spatial
resolution imagery with a large synoptic footprint, within which archaeological residues can
be easily identified, will frame how survey is undertaken. This has occurred for the SHR
94
project where residue identification from satellite imagery is the first stage of a battery of
survey techniques which also includes:
•
Rapid recording and grab samples focussed on diagnostic material for all sites.
•
Gridded sieved collections on a sample of sites
•
Off-site field-walked transects.
•
Artefact fall off by distance from site.
The survey design and methodology provides the archaeologist with data. It is then the
archaeologist’s goal to transform this data into information which corresponds with the
research hypotheses. However, it must be remembered that these data are not intrinsically
meaningful: archaeologists have spent many years defining mechanisms to classify
archaeological residues to facilitate meaningful interpretation. All of these classification
schemas are cultural products of archaeologists rather than true cultural reflections of past
societies.
3.5 Landscape modelling
In order to address questions of organisational change and stability, interpretation of the
archaeological record requires one to make certain inferences regarding:
1. How the material record reflects the role that people and place played in the
organisation of the past.
2. How the pre and post-depositional formation processes have affected the
material record.
3. How these factors vary over space and time.
Given a robust collection strategy and understanding of landscape formation, archaeologists
need to take a theoretically informed ‘leap of faith’ from perceived variations in a regional
archaeological record to interpretations about past cultural systems (Wandsnider 1992a).
These interpretations are constrained by any assumptions implicit in the way the landscape is
analytically modelled.
95
3.5.1 Modelling archaeological landscapes
All landscape applications model their data in unique ways dictated by their survey
methodology, research goals and the established regional classification systems. Although it
would be impossible to explicitly define all the modelling techniques used, it is possible to
outline the theoretical development of the main approaches.
3.5.1.1 The ‘site’ model
This is the oldest model of archaeological residue dispersion and relies on the simple premise
that the archaeological population consists of a set of discrete activity areas (nominally
referred to as sites). Originally ‘site’ was only applied to residues that had a monumental
component such as those containing upstanding architectural remains (of stone, brick or
mud-brick), earthen mounds (tells) or substantial fortifications (Iron Age hillforts).
Subsequently, after methodological and interpretative techniques improved, surface artefact
concentrations were also recognised as sites.
These sites can be classified upon a number of criteria. The following themes are commonly
used:
•
Size.
•
Material diversity.
•
Function.
•
Morphology.
Although simplistic, this model is a convenient interpretative tool in environments where the
archaeological residues are actually discrete.
Associated with the site model is the concept of networks between (i.e. communication) or
surrounding (i.e. hinterland) sites. Early aerial archaeologists were quick to exploit the fact
that many large scale landscape features survived intact or as earthworks. Many such features
could easily be overlooked during ground survey but are obvious on aerial photographs. In
fact these ‘off-site’ traces were being recorded by aerial archaeologists long before the
concept of landscape archaeology existed (Banning 2002 p. 13).
96
3.5.1.2 Sites as perturbations in a distribution
The inclusion of surface residues (particularly artefact scatters) into the site model marked a
subtle shift in the appreciation of site definition. Residue attributes are measured to establish
if they have a higher or lower concentration than a ‘background’ value. This variation is based
on the simple assumption that the background (which can also be referred to as off-site or
non-site, both of which will be defined later) values represent predominantly natural
transforms and site values represent cultural transforms, and that these distributions can be
distinguished mathematically. This assumption can be used in the interpretation of many data
collection techniques including artefact distributions, chemical residues, soil differences and
magnetic or resistance variations. In the hypothetical example illustrated in Figure 52 the
landscape was sampled in 2 metre square units, a ‘site’ being defined as having at least 2.5
artefacts per m2. This technique can be extended across a landscape. Two common methods
employ fixing a set value for the background level (i.e. assuming it is an isotropic plain) or by
calculating a moving average for the background level.
Figure 52 Hypothetical site distribution as perturbations in a
distribution of artefacts. The site level is calculated at a density of
2.5 artefacts per m2.
97
This simplistic analysis can be improved by more robust statistical techniques. For example
the uniform distribution approach assumes that both ‘sites’ and ‘non-sites’ have attributes
that are normally distributed around a mean. The mean value for the ‘sites’ is relatively higher
or lower than that of the ‘non-sites’. This property, therefore, allows sites to be distinguished
(Buck et al. 1996).
These techniques primarily model the landscape as a binary entity (site or non-site). However,
many surveys now postulate that the landscape is a blanket of recognisable archaeological
residues with varying intensities (Cherry et al. 1991; Robinson and Zubrow 1999; Bintliff
2000; Francovich et al. 2000; Wilkinson 2000; 2001) Therefore, the values of a particular
archaeological residue (such as phosphate value or artefact count) vary continuously over the
landscape and can be modelled in much more subtle ways. One approach, referred to as the
‘fried egg’ model, interpolates isopleths of equal attribute values (analogous to contours).
This technique highlights the gradual variations between the boundaries of sites and nonsites. There are many such statistical approaches that allow more refined (and fuzzier)
interpretations. However, a criticism of many of these approaches is that the temporal
classification of sites can be under-represented.
3.5.1.3 Sites as palimpsests
Archaeological residues can reflect prolonged cultural activity. Hence, a settlement site may
represent continued occupation from the Neolithic, as is the case with many tell sites. These
sites may have seen variable cultural systems in their lifespan and may have been disused for
prolonged periods.
Therefore, the clustering of artefacts upon such a site does not represent a single cultural
event but many overlapping cultural sequences that form a palimpsest. It is assumed that data
collection represents the mathematical union of these activities. These data are normally
classified to highlight overlapping distributions, each distribution representing, for example, a
different set of activities or time-frames. The palimpsest model requires attribute
classification and is therefore only used on such data where the attribute collection is
adequate for the modelling goal (for example, pottery assemblages to define spatio-temporal
divisions in an area with a robust pottery sequence).
98
Figure 53 Example of assemblage overlap to erroneously produce
a site (after Banning 2002 p. 19)
99
A criticism of this technique is that cultural, natural and post-depositional processes can
transform the relationships between the original and modern distributions. Hence, these data
may not be representative of any original cultural/temporal variations: i.e. the Neolithic
phases may be obscured by later phases (particularly in an aggrading environment) or modern
ploughing may have destroyed all structure. However, Rossignol and Wandsnider (1992)
amongst others have applied mechanisms to mitigate against some post-deposition processes
during modelling.
The logical extension of the palimpsest approach, as demonstrated by Foley (1981), is that
non-palimpsest approaches would identify an area as a site through artefact density even
though it actually represents the overlap of several distinct cultural assemblages (see Figure
53).
3.5.1.3.1 Systemic contexts
In order to reconstruct past human activity it is important to be aware of the context of an
artefact, feature or site. Its context relates to the analytical unit’s immediate matrix, its date
(or ability to provide a date) and its association with other analytical units. The selection of an
analytical unit, in some respect, dictates the scale of the analysis to be performed (Shennan
1985; Renfrew and Bahn 1991 p. 42; Banning 2002). However, artefacts and features
recovered from an archaeological context may be in different systemic contexts. A systemic
context refers to residues when they are participating in the same behavioural system. Thus,
due to the multi-faceted interpretation and classification of artefacts, features and settlements
each may exist in multiple archaeological and systemic contexts depending upon the
viewpoint of the observer (Schiffer 1987 p. 3). The palimpsest model was one of the first
models that could properly account for changes in systemic context.
3.5.1.4 The off-site (or hinterland) model
The off-site model is a complementary technique to a site model because it focuses on
modelling cultural activity between and around sites. Off-site collection was a reaction in the
late 1970s and early 1980s to the site-centric view of landscape survey. Binford (1982; 1992;
1996) argued that increased understanding of settlements and settlement patterning could be
achieved by understanding how the landscape between sites was exploited and managed. This
led to a re-appraisal of survey techniques that explicitly included collection and interpretation
of off-site residues. In landscapes where persistent settlement evidence is the norm (such as
100
many areas of the Middle East (Wilkinson 2000)), the off-site model is regularly used to
incorporate other aspects of human occupation that occur away from a settlement. These
approaches are primarily focussed on subsistence activities such as agriculture, pastoralism
and other resource procurement and management strategies. Furthermore, off-site modelling
is complementary to more modern landscape theories of perception and phenomenology
(for examples see Barrett 1994; Tilley 1994) which consider that the landscape is much more
complex than categories of sites and empty spaces. The off-site model should not be
confused with the non-site model which uses different assumptions and techniques for
collection and interpretation.
3.5.1.5 The non-site model
The non-site model is a development of the off-site model which applies a distinctly
statistical approach to residue variation within the landscape (Ebert 1991; Dunnel 1992). The
methodology is not designed to detect sites per se, rather, it is designed to focus upon residue
(and attribute) variations. It is postulated that spatial analysis of these variations in
conjunction with ecological parameters (i.e. soil type, hydrology and landform) leads to better
archaeological understanding.
3.5.2 ‘Sites’ as interpretative units
The term ‘site’ has already been referred to extensively, but what is a ‘site’? What constitutes
its existence and what, as an analytical unit, does ‘site’ mean? The notion of ‘site’ is ubiquitous
as an archaeological tool for conceptualising the archaeological record. It is expressed as a
recognised unit, which, in itself, brings credence to the definition, although it is rarely
defined. The problem of site, within the context of approaches to landscape archaeology, lies
in the definition of form, function, duration, variability and spatial boundedness.
Initially sites were any prominent areas in the landscape, as determined by topography or
monumental architecture, of any antiquity. Surveys initially focused on the prospection and
characterisation of these sites. However, as archaeologists became more interested in
reconstructing human impact over the whole landscape, it was recognised that this approach
severely biased any archaeological collection programme. Smaller settlements, transitory
camps, fields and other activity loci can produce clusters of artefacts and other less obvious
residues. Hence it was postulated (at least for artefacts) that the landscape contains a variable,
but continuous, distribution on or near the surface (see section 3.5.1.5). How does the
101
quantitative or qualitative variation of attributes reflect human dynamics and thus upon what
basis can we extrapolate the notion of site (Dunnel 1992)? Surveying the whole landscape to
discover such residues gave the theoretical and terminological paradox of ‘when does a site
become a site’? Off-site and non-site methodologies were developed as an attempt to resolve
this paradox. These methodologies require a significant shift in techniques that can only be
achieved by applying systematic statistical survey, as comparative analysis with other areas is
necessary during interpretation.
So is the term ‘site’ still a useful term in synthetic applications? Amongst others, Binford
(1992) criticises many uses of the term site. He argues that without supporting evidence from
morphological sources and extrapolation into the social (and spatial) dynamics of the
landscape one cannot define a site. In short, simple aggregation of artefacts does not lead to
the discovery of a site. There is a movement away from ‘site’ studies at the landscape level
into a more profound consideration of artefacts and morphological factors. In fact, Binford
has argued that aggregating landscape information within sites, during collection and in
subsequent interpretation, can mask the underlying patterns and processes. These
philosophical problems extend beyond the remits of this research. However, they are
important in relation to the exploitation of remote sensing resources for archaeological
purposes and offer important new avenues for visualising and evaluating landscapes.
3.6 Discussion
The improvement of reconnaissance techniques and the integration of information from
different sources have developed surface survey from a preliminary stage for fieldwork into
an area of inquiry in its own right, forming the framework for Cultural Resource
Management. The integration of information from a number of neighbouring surveys can
lead to a large scale information resource, although the quality, accuracy and consistency of
projects with different goals and methodologies can vary widely.
A well-developed and implemented surface survey methodology can produce a great deal of
archaeological information. However, survey does not give the same quality of information
as excavation. This is particularly important during the assessment of artefacts. Artefacts
provide the chronological, functional and economic frameworks for the survey; yet how can
we be sure that surface traces reflect the same distributions below ground. This argument has
been countered by the assertion that most artefacts were originally surface assemblages and if
102
one can understand the transformation processes that affected artefacts then biases can be
modelled (Camilli and Ebert, 1992). Furthermore, the chronological frameworks of pottery
sequences, for example, are subject to change at both a local and regional level. This causes a
problem when dating, derived from such sequences, enters a site record. How can a revised
pottery sequence be propagated back into the archaeological record so that it reflects the
revised sequence?
Artefacts from single period or shallow sites would be expected to give the most reliable
information. Conversely, multi-period sites, particularly tells, could show little evidence on
the surface of the earliest and deepest levels. In addition, natural transformation agents may
effectively mask an area of archaeological interest (for example, the burial of sites beneath
alluvial or aeolian deposits) or alternatively they may mix a group of stratigraphically distinct
artefacts by a process of deflation. Although this is problematic, if the methodology and
analytical components state the assumptions and biases in survey strategy then interpretation
should not be unduly affected. After interpretation it could be possible to test the hypothesis
by further empirical collection, for example, by test pitting and excavation.
Some archaeologists denigrate the usefulness of archaeological surveys. They point out the
fact that many have failed, and still continue to fail, to detect sites of significance or that
surveys introduce undue bias when estimating a site’s date or relative import. However,
surveys do collect information that is regularly overlooked during excavation and other
higher resolution data collection. Furthermore, most failures are due to poor survey design
and lack of focus for the survey objectives.
3.6.1 Research implications of landscape archaeology
Approaches to landscape archaeology are becoming increasingly multi-disciplinary. The SHR
project (see Chapter 4) is in its early stages and is in a state of imperfect knowledge. The only
traditional DBA resources are modern and historic mapping at different scales. These only
provide a limited indication of the range of archaeological residues in the application with a
significant bias towards monumental remains. Satellite imagery is viewed as a potentially costeffective medium to improve on this state of knowledge. It is hoped that information derived
from the satellite imagery will frame the theoretical and methodological techniques employed
in the short to medium term. These improvements are not just related to the detection and
interpretation
of
archaeological
residues.
103
Allied
colleagues
(geomorphologists,
environmentalists and historical geographers) require a range of thematic information in
order to contextualise their data. Satellite imagery also has the potential to aid in their
research enquiries. The integration of satellite imagery into landscape archaeological
approaches has the potential to bridge the gap between processual and post-processual
theoretical stances by generalizing the digital imagery into a variety of theoretically driven
archaeological themes (Anschuetz et al. 2001 p. 159). Aerial photography and satellite imagery
have not only aided landscape collection by the inherent information they contain, they can
be used as analytical tools. It is not unreasonable to collect satellite imagery for the whole
application area, and through a process of ‘ground-observation’ and image manipulation it is
possible to extend the correlation of the ground-observed (or CRM) data through the whole
application area in conjunction with other thematic data sets (geology, hydrology etc.).
Irrespective of the above, the impact of satellite imagery on landscape archaeology should be
evaluated as to whether it can improve method and analysis or whether it can reduce costs in
general.
104
SECTION 2: METHODOLOGY AND ANALYSIS
105
CHAPTER 4 SETTLEMENT AND LANDSCAPE DEVELOPMENT IN THE
HOMS REGION, SYRIA: AIMS, SETTING, RESEARCH QUESTIONS AND
FIELDWORK (1999-2003)
4.1 The SHR project
Settlement and Landscape Development in the Homs Region, Syria (abbreviated to SHR) is a
multidisciplinary, multi-period regional survey project with both archaeological and palaeoenvironmental dimensions. The project was formally initiated in 1999 and is jointly directed
by Dr. Graham Philip (University of Durham, UK) and Drs. Michelle Maqudassi and
Mamoun Abdulkareem (Directorate General of Antiquities and Museums, Damascus, Syria).
This chapter draws upon the results of several project publications (Donoghue et al. 2000;
Beck 2002; Philip et al. 2002a; Philip et al. 2002b; Beck 2003; Bridgland et al. 2003; Philip et al.
in press), unpublished project documentation and various discussions with colleagues both in
Durham and in Syria. As such this chapter outlines many of the overarching agendas of the
SHR project of which this research forms a part.
The project area itself consists of two distinct study areas (see Figure 54): one located to the
north-west of Homs covering some 195 km2 and the other located south-west of Homs
covering some 385 km2. Each of these areas contains elements of the three principal
environmental zones characteristic of this region of the Orontes valley: marl, alluvium and
basalt plateau (see Figure 55). In part these contrasting zones were deliberately chosen with a
view to evaluate the methods for their applicability in a broad range of environment types.
The study area was designed to be large enough to permit the analysis of overall settlement
structure while remaining small enough to permit representative high intensity survey. The
extents of each environmental zone are referred to as landscape units, and form the basic
analytical units for sampling.
The project aims to understand the ecological context of human activity and the complex
interplay between natural and anthropogenic factors in structuring long term trends in
regional landscape development in western Syria.
106
Figure 54 Location map of the study area.
107
Figure 55 The environmental zones in the application area
(labelled by unit and their respective areas) on a 4,3,2 Landsat
image (28th October 2000).
108
4.2 Environmental zones, land use and archaeological summary
There is little recent literature on the geological and geomorphological characteristics of the
region. Hence, the acquisition of such data is fundamental to the success of the project (see
for example Bridgland et al. 2003). The following summaries of the main environmental
zones is reliant upon a limited range of older publications (Voûte 1955; Ponikarov et al.
1967), the summary statements of Wirth (1971) and Wolfart (1967) and field observations by
the project teams.
Figure 56 Elements of the marl landscape
4.2.1 The marl zone (units 1 and 4)
The area consists of Upper Miocene lacustrine marls, overlain by thin irregular sheets of
Pleistocene pebbles and gravels. These are covered by a layer of reddish brown loam, which
is derived in part from the weathering of the marls, but also through colluviation from the
Eocene limestone hills to the south-east. This soil depth ranges from 0.2 to 0.4 m but can
109
reach up to 0.7 m in wadi beds. These soils are undergoing aeolian deflation. With the
exception of the active river terraces, where extensive irrigation means that crop cover is
virtually continuous throughout the year, agriculture relies on rainfall. The rainfall within the
study area varies between 200mm and 1000mm per annum (see Figure 76). This variation
demonstrates that some of the area is marginal. However, in recent years further crops have
been realised by pumping water from the underlying water table. The area is crossed by
several very shallow wadi systems which run in a north-westerly direction from the course of
the large wadi al-Rabaya that hugs the lower slopes of the Anti-Lebanon range. These are
shown on recent maps as seasonal watercourses, although field investigation indicates that
they are scarcely perceptible in some places, and there is no indication that they are still
actively connected to wadi al-Rabaya. Rather, it appears, that they now remove surface runoff (Bshesh, pers. comm.). Thus, for large parts of the growing season the area can be
considered to be bare soil in a semi-arid environment (Philip et al. 2002a). However, the
interim fieldwork season conducted during April and May 2001, after the rainfall, has
demonstrated that the application area is far from agriculturally marginal. Although a
reducing water table over the past two decades has led to increased cultivation of olive
groves, almond groves and fruit orchards within the rainfed areas (Bshesh, pers. comm.), at
least one harvest of wheat and other crops is possible within the application area. This
significant shift from bare soil to intensive vegetation has a significant impact on the
prospection techniques used.
Figure 57 Archaeological residues in the marl.
110
In Unit 4, the marl in the northern study area, the agricultural practices were significantly
modified after the introduction of concrete irrigation infrastructure in the 1930s. Although
this has impacted on the archaeological residues in the area, it is useful from a CRM
perspective to attempt to quantify this archaeological impact.
Archaeologically this zone is characterised by a number of distinctive tell sites and far less
obvious smaller sites that appear as either low mounds or flat areas (sometimes associated
with a depression) marked by artefact scatters and soil colour differences (see Figure 57).
There are also a number of obvious industrial and architectural fragments (such as columns,
olive presses and rotary mills) distributed through the landscape.
Figure 58 Elements of the basalt landscape.
111
Figure 59 Archaeological residues in the basalt.
4.2.2 The basalt zone (units 3 and 7)
Units 3 and 7 are located on the eastern edge of the Shin plateau, a large expanse of
Pleistocene basalt originating some 40 km west of Homs. This zone consists of large areas of
thin boulder strewn basaltic soils. Soil and water are transported downslope by surface run
off either towards the Orontes or into small internal depressions (rams). Sediment is
deposited during this process creating distinct areas with deeper soils (referred to as ‘well
drained basalt’ in section 6.6). The rams may be inundated in winter and, as some have no
external drainage, can retain water well into the summer. It is assumed that these natural
reservoirs would have been significant for past human populations. While the rainfall is
sufficient for dry-farming of cereals, water for humans and livestock has generally to be
collected by artificial means (see Figure 58).
Figure 60 Archaeological residues in the alluvium.
112
Archaeologically this zone is characterised by structural remains in the form of abandoned
villages, isolated villas, roads, tombs, cairns and agricultural systems. These are all preserved
as a palimpsest of stone walls and concentrations of rubble. In contrast to the other zones
the basalt should potentially provide rapid insights into how hinterlands were managed and
exploited.
4.2.3 The alluvial zone (units 2, 5 and 8)
Units 2 and 8 north of Lake Qatina and the eastern portion of unit 5 encompass the most
recent incision of the Orontes and its active floodplain zone. In units 2 and 8 the Orontes is
at its western extent as it cannot migrate further west due to the basalt blocking further
movement. The terrace formations associated with migration of the Orontes and other
broader scale fluvial events are poorly understood at present. However, Bridgland et al. (2003)
have tentatively suggested that the application area resides within a previously unidentified
Quaternary terrace sequence.
Figure 61 Elements of the alluvial landscape.
113
Due to the proximity of the river these areas are constantly irrigated and under virtually
continuous crop cover. West of the river terraces in zone 5 is an area of land that was initially
identified as alluvial floodplain (see section 7.6 for a re-interpretation). The formation
sequences that created this environment are extremely complex and hence detailed
geomorphological investigations are required prior to any archaeological investigations.
Figure 62 Elements of the landscape around Lake Qatina.
Archaeologically, the alluvial zone is difficult to characterise: the intensive cropping and
complex geomorphology mean that archaeological residues can vary dramatically. However,
most archaeological residues within the active river incision consist of flat sites which are very
difficult to locate under crop. The lacustrine margins between the Orontes and the southwestern end of Lake Qatina contain the same residue range as observed in the marl. On the
114
floodplain to the west of the Orontes in the south archaeological residues appear to be very
sparse with only a handful of low mounds and tells.
4.2.4 Lake Qatina (unit 8)
Lake Qatina is one of the few bodies of long term standing water in western Syria, having
probably originated as a natural depression adjacent to the basalt. This natural feature was
exploited by an ancient dam which raised the water level to c. 497 m. The completion of a
larger dam and associated irrigation system in the late 1930s raised the lake level by a further
2-3 m. However, capping the spring at ‘Ain at-Tannur (to divert the water to Homs), a
reduction in the yearly rainfall average and extensive irrigation has led to the shrinkage of lake
Qatina (see Figure 64 and Figure 169). The lake itself offers great potential for obtaining data
suited to palaeo-environmental reconstruction. Cores in the sediments have yielded
radiocarbon dates that indicate that the sediments in lake Qatina have been accumulating for
at least 4000 years (Philip et al. 2002b p. 14).
The original damming of Lake Qatina has produced many interesting ramifications. From a
geomorphological perspective, erosion at the lake margins has revealed different geological
layers that elucidate the formation sequence in the surrounding environments (see Figure 62).
Archaeologically, the damming has helped preserve sites in the centre of the lake from
further anthropogenic modification. However, there have been obvious erosional
consequences. Those tells on the lake margins have been significantly eroded and in some
case destroyed by the lake (see Figure 63).
Figure 63 Archaeological residues eroded at Lake Qatina.
115
At the western edge of Lake Qatina there is an area of low lying land covered in lacustrine
deposits. After the modern (1930s) dam was built this area is very susceptible to flooding if
the lake is high enough. Figure 64 displays how water level changes in the lake have affected
these residues. When the lake was at peak capacity in the late 1960s none of the tells were
accessible. Even at its lowest level (prior to the winter rain (Corona 1108) most of the tells
are still surrounded by water. However, in the 2002 Ikonos image the lake has fallen to such a
level that all of these tells are now on reclaimed ‘lacustrine’ land and are now subject to
intensive cropping.
Figure 64 Tell sites in the lacustrine deposits at the south western
end of Lake Qatina (scale 1:20,000).
116
4.3 Survey methodology
Philip et al. (2002b pp. 7-14) define the survey methodology for the application area in detail.
Only the specific components of the methodologies which affect this research will be
discussed in depth. The satellite imagery is employed for a number of different purposes
(prospection, archaeological evaluation and thematic analysis). Of particular import is the
relationship that satellite imagery may have in the determination of survey methodologies.
Wherever possible all survey data is collected digitally. Where data is not collected digitally it
is digitised after collection. These data are stored, manipulated and analysed in the project
information system primarily employing relational database and Geographical Information
Systems (GIS) technology (see Appendix I). This information system is pivotal for the rapid
assessment and analysis of the full range of data. Results of these analyses aid the structuring,
programming and management of fieldwork.
As Alcock et al. (1994 p. 138) recommend, the available evidence was assessed to help
provide a framework for field survey. This formed the Desk Based Assessment.
4.3.1 SHR project DBA
As discussed in section 3.4.2 a Desk Based Assessment is an analysis of the known or
potential resources within a study area. The initial DBA was focussed on the topographic and
thematic map sources and is hence discussed first.
4.3.1.1 Non-satellite data sources
Prior to incorporating satellite imagery into the research programme the only available
information sources were topographic and thematic maps. Of these resources the Syrian
1:25,000 mapping proved to be the most useful for any initial archaeological DBA in the
application area. Approximately 90 ‘sites’ were positively identified from the mapping,
referred to as ‘known sites’, from specific legend symbols or identified place names such as
tell or khirbah (ruin). Approximately 80 additional ‘sites’, referred to as ‘potential sites’, were
located through less obvious indicators that might be associated with archaeological residues.
These included nameless contour features and suggestive toponyms with no obvious
associated settlement. In the basalt zone field systems, settlements and clusters of ovoid
structures (tentatively interpreted as animal pens) were mapped. It was only after the
acquisition of the satellite imagery that it became apparent how generalised this mapping was.
117
All of these sites were entered into the GIS and given their own unique site number.
Associated attribute information was entered into the database.
A digital elevation model was created for the application area from digitised contours derived
from the 1:25,000 mapping. This is discussed in greater detail in sections 5.1.4.1 and 6.6.
Thematic maps of the application area are available at a variety of small scales ranging from
1:500,000 to 1:1,000,000. These proved to be inadequate for most purposes apart from the
definition of broad environmental zones. Therefore, satellite imagery is used to create these
themes in more detail. Landsat TM data has a proven track record in the creation of such
mapping.
4.3.1.2 Satellite data sources
Although the topographic and thematic mapping provided an initial framework for
evaluation of the survey methodology, preliminary fieldwork identified numerous ‘flat’
surface concentrations in the marl zone and a palimpsest of field systems and structural
evidence in the basalt that were not apparent on the mapping. After geo-rectifying the
imagery (section 5.4) a variety of prospection techniques were used to locate areas of
‘potential’ archaeological residues (Chapter 7). In summary, the satellite imagery revealed over
400 ‘potential sites’ which were plotted in the GIS. In order to evaluate the accuracy of these
‘potential sites’ they had to be evaluated on the ground.
4.3.2 Ground observation
Various forms of fieldwork were designed to:
•
Ground observe potential sites identified from the satellite imagery – Site visits.
•
Locate sites which were not identified on the satellite imagery, in order to clarify
the weaknesses of satellite prospection – intensive surface collection of sample
areas.
•
Record information about sites visited (extent, morphology, soil and vegetation
cover) in order to elucidate the relationship between the satellite images and the
archaeology visible on the ground.
•
Collect a small sample of diagnostic surface material to allow a provisional
identification of the main periods of activity represented on each site.
118
•
Record information about the different environmental zones to aid thematic map
production.
•
Collect
samples
for
analytical
determination
of
site
visibility
and
geomorphological analysis.
4.3.2.1 Attribute recording
A range of attributes were recorded during site visits, including: extent, morphology, postdepositional erosion, masking, soil colour (both off and on-site), artefacts (types and ranges),
land use, vegetation cover and land setting. The artefacts were washed, labelled and their
attributes entered into the object database. This data was augmented by digital photography
to provide a rapid and cheap aide-memoir of the site. Other, more innovative, techniques
were also employed including 360 degree photography which allows individuals to immerse
themselves in a digital landscape.
These same attributes were recorded for ‘potential sites’ which revealed no unambiguous
archaeological residues. This was in order to gain further understanding of the physical
conditions that actually caused ‘sites’ to be located and to improve the image interpretation
key (see Chapter 8).
4.3.2.2 Spatial recording
There are a variety of ways in which the spatial record can be collected. In the basalt zone
spatial data is normally derived from the Ikonos imagery as it is the most accurate record.
However, where this data is found lacking then additional information observed on the
ground can be added using GPS or by digitising directly into ArcPAD.
In all other zones the spatial component becomes more complex. It is possible to get a
spatial extent for archaeological residues from the Landsat, Ikonos and Corona imagery and
by defining the extent of the artefact residues or soil colour differences on the ground. For
the sake of consistency it was decided to use the extent derived from handheld GPS as the
definitive version. However, analysis of the extents from the other imagery should not be
neglected as it is likely that the spatial extent of the residues has changed over time. This
could give some insight into how modern post-depositional processes have impacted on the
residues.
119
4.4 Fieldwork programme
Six visits were made to the application area to conduct fieldwork (see Table 4 and Figure 65).
The financial and logistical assistance of the CBRL, DGAM and NERC are duly
acknowledged.
The initial fieldwork season, conducted during December 1999 and January 2000, was
focussed on evaluating surface collection procedures at a number of key sites in the southern
marl. Three dimensional CAD models were produced at each site using a total station. This
season also gave an insight into the nature of the archaeological residues across the
environmental zones.
Se a son
1
2
3
4
5
6
Da te
Goa ls
Familiarisation with the project zones, the
evaluation of surface collection procedures
December 1999 - January 2000
and 3 dimensional modelling of some key
sites.
To evaluate the efficacy of satellite
prospection: A 20% , by area, sample was
August - September 2000
rapidly surveyed using a combination of
vehicle and pedestrian survey techniques.
To determine the benefits and problems of
surveying in late spring when crop cover is at
April - May 2001
its peak and to evaluate the 7 x 7 km Ikonos
panchromatic imagery.
To focus on site and off-site ground
observation in all zones. Soil samples were
August - September 2001
taken to attempt to determine the physical
basis for the increased reflectance of
archaeological residues in the marl area.
August - October 2002
To contine the goals of season 4 and to
evaluate the full Ikonos panchromatic and
multi-spectral imagery. GPS co-ordinates
were taken as rectification points to improve
the spatial accuracy of the Ikonos imagery.
August - September 2003
A study season to address methodological
issues of recording and analysis. Further
sites were ground observed in the basalt and
a further group of soil samples were taken in
the marl zone.
Table 4 Fieldwork summary.
The second season, conducted during August and September 2000, was aimed at locating
and identifying residues identified from the satellite imagery. The study area was broken up
120
into 2 x 2 km squares from which a 20% sample (across environmental zones) was selected
for intensive vehicle and pedestrian survey. It was intended that this survey should provide
rapid feedback on which sites the imagery helped to identify, which sites were misidentified
and which sites were missed. Unfortunately, at the time, the geo-referencing of the Corona
imagery was quite poor. To compound matters the Syrian Grid projection was employed
which meant that the hand held GPS was difficult to use (see section 5.4 for a discussion of
the Syrian Grid). Therefore, navigation to the residues became very problematic: so much so
that no residues could be confidently located in the basalt zone. Hence, although traversed,
residues identified in this zone could not be mapped with confidence. In total 138 potential
sites were visited of which 121 were positively identified.
A third season was conducted during April and May 2001. The aim was primarily to evaluate
the landscape under different environmental conditions (i.e. crop). 36 new sites were visited
of which 31 were positively identified. Soil characteristics (colour, compaction and
composition) were also taken at these sites. A GPS map of roads and road intersections was
created to aid the geo-rectification of the Corona imagery. This also provided an opportunity
to evaluate the spatial accuracy of an archive 6 x 7 km panchromatic Ikonos imagery located
in the southern Marl (see section 5.4). This was the first season in which the whole recording
process was digital.
During the fourth season, conducted during August and September 2001, this programme of
ground observation was continued (a further 55 sites). On and off-site soil samples were
taken for physical and chemical analysis in order to investigate what determined the
distinctive reflectance properties associated with archaeological residues in the marl. ‘Off-site’
sample units, referred to as transects, were surveyed to determine the level and range of
‘background material’ and to attempt to locate sites not identified from the imagery.
A fifth season was conducted during August and October 2002. This season had the same
overall goals as the previous season (a further 70 sites and c. 5 sq. km of transects). This was
the first season where the Ikonos panchromatic and multispectral imagery for the full
application area was available. Further rectification points were taken with hand held GPS to
improve the geo-rectification of the Ikonos imagery to facilitate desk-based mapping of the
residues in the basalt.
121
A sixth season was conducted during August and September 2003. This was primarily a study
season to analyse the pottery and flint artefacts and address artefact data recording issues.
Intensive survey was also carried out in the basalt zone to establish the range of evidence and
the accuracy of the desk based interpretation of the archaeological residues (a further 50
sites). Further samples were taken for particle size analysis.
Figure 65 Summary of fieldwork from 1999 to 2003
122
4.5 SHR research goals
The SHR project has been designed to address the following research issues (Philip et al.
2002b):
•
Collection of survey data for inter-regional comparative analysis.
•
Monitoring inter-regional interactions.
•
Region specific themes.
•
Methodological issues.
These research issues are discussed in more depth by Philip et al. (2002b). However, there are
issues that directly impact upon this research programme.
The increased availability of survey data for the Near East has begun to make inter-regional
analysis possible (Wilkinson 2000). However, different surveys vary in technique, quality and
coverage. As highlighted in pan-regional reviews of settlement evidence (Alcock 1994 p. 181)
Western Syria is virtually devoid of any systematic archaeological survey. Where these surveys
have been conducted they have focussed upon the more obvious archaeological residues
(such as tells). This bias, inherent in such archaeological surveys, has been discussed by
Wilkinson (1998). The methodological techniques employed by these surveys would, in all
likelihood, have failed to detect many of the archaeological residues extant in the SHR
project study area. This is particularly pertinent when one considers that Alcock et al. (1994 p.
138) recommend that sampling (and hence survey) should be guided by the known
distribution of archaeological residues and a detailed knowledge of the local landscape types.
Although sensible, this recommendation presupposes that such data is actually available. For
this study area it was not.
The initial fieldwork seasons demonstrated that there was a positive correlation between
areas of lighter soil and archaeological residues in the marl zone and that the basalt zone
contained an extensive, but unmapped, palimpsest of ancient field walls, cairns and
structures. However, given the size of the application area, a high-intensity ground-based
survey and mapping programme would have proven to be prohibitively expensive.
Furthermore, some mechanism was required to rapidly collate the distribution of
archaeological residues and provide a framework for determining the local landscape types as
suggested by Alcock et al. (1994 p. 138). Satellite sensors offer the potential to rapidly image
123
large areas in different bands of the electromagnetic spectrum. Some sensors have a spatial
resolution approaching that of vertical aerial photography. This potentially makes satellite
imagery a more cost-effective resource than aerial photography and the large ground
footprint also makes it a more appropriate medium for large scale metric survey
(Kouchoukos 2001). Hence, it was proposed to examine how satellite imagery can aid
archaeological prospection, what impact it may have on survey methodology and how it can
help in providing a framework for contextualising the archaeological residues within their
environment and landscape.
The application area itself intersects a range of different environmental zones, all located with
easy access to the Homs-Tripoli gap. It covers an area of transition between mobile pastoral
and sedentary agricultural groups. In part these zones will have structured the social,
economic and political dynamics in this landscape. The survey will produce results which will
help elucidate the nature of contemporary human activity within and between these zones.
Hence, it is proposed that satellite imagery can aid archaeological interpretation by providing
thematic contextual backdrops.
In agricultural and industrial terms the Homs region is economically significant. As a
consequence the archaeological resource is under severe threat from development, in
particular residential construction, industrial expansion and intensified agricultural practices.
The frequency of lower visibility ‘flat’ sites within the study area means that these sites are
under even greater threat as they are less likely to be recognised and protected than tell sites.
The documentation of the archaeological residues is intended as the first step in preparing a
cultural resource mitigation and management programme to preserve permanently the
character of the archaeological landscape.
124
CHAPTER 5 IMAGE PREPARATION
5.1 Data Sources
The research uses a combination of Landsat TM, Landsat ETM+, Ikonos and Corona
imagery. Each of these sensors has different spatial and spectral characteristics and
mechanisms of image acquisition. A variety of other information sources have been
integrated into the project, including present day and historic mapping. Prior to integration
into the data model and analysis, each of the data sets needed a certain degree of preprocessing. This is discussed in this chapter.
5.1.1 Landsat
The Landsat (‘Land Satellite’) programme (originally entitled Earth Resources Technology
Satellite) was first publicly proposed by Dr. Robert Alexander in 1964 and the first satellite
was launched in 1972 (Morain 1998). It was the first repetitive worldwide surface focussed
satellite imaging system with a relatively high instantaneous field of view. The Landsat family
of seven satellites (Landsat 6 was destroyed at launch) have not only seen technical changes
in their sensor arrays but also changes in their ownership and management. Throughout
these changes, the United States Geological Societies (USGS) Earth Resources Observation
Systems (EROS) data centre retained primary responsibility for the maintenance of the
Landsat archive (Campbell 2002).
The programme consists of three distinct generations of satellites (at the time of writing
Landsats 5 and 7 are the only working satellites):
1. Landsats 1, 2 and 3: contained a MultiSpectral Scanner (MSS) and Return Beam
Vidicom (RBV).
2. Landsats 4 and 5: contained MSS and Thematic Mapper (TM).
3. Landsat 7: contains Enhanced Thematic Mapper plus (ETM+) with
approximately the same frequencies as TM with increased (60m) resolution for
band 6 and an additional Panchromatic channel (0.52 – 0.90 µm, at 15m
resolution).
125
The consistency of the Landsat data over the past three decades offers opportunities to
compare land cover change over significant periods of time. This archive provides a rich
collection of information about the Earth's surface. Major characteristics and changes to the
surface of the planet can be detected, measured, and analysed. Thus, the effects of
desertification, deforestation, pollution, volcanic activity and other natural and anthropogenic
events can be examined. The information obtained from the historical and current Landsat
data play a key role in analysing local environmental changes through time. Recognition of
the significance of this body of data led to the 1992 Land Remote Sensing Policy Act which
called for continuity beyond Landsat 7 (Morain 1998 p. 44).
For more information on the history, sensor characteristics, pre-processing and analysis of
Landsat imagery please consult the following references: Morain (1998), Townshend et al.
(1988) and USGS (2003c; 2003d; 2003e).
5.1.1.1 Thematic mapper
The TM sensor array achieved orbit in 1982 on the Landsat 4 platform. Table 5 describes the
technical characteristics of the Landsat TM sensor.
Altitude
Equa toria l Crossing
705 km
approx 9:45 am
Fie ld of Vie w
Ima ge ove rla p a t Equa tor
Inclina tion
Numbe r of Ba nds
Orbit cycle
Orbits pe r da y
Progra mm a ble
Qua ntisa tion
Size of im a ge
Spa tia l Re solution
Spe ctra l Range
Ste re o
Sw a th
Te mpora l Re solution (Re pe a t
cycle )
Se nsor Type
15.4°
7.60%
Sun-synchronos
7
233
14.5
Yes
8 bit
185 x 172 km
30-120m
0.45-12.5 µm
No
185 km
16 days
opto-mechanical sensor
Table 5 Landsat TM technical specification (after Townshend et al.
1988)
126
Landsat TM 5 has a similar sensor array to Landsat TM 4, but Landsat 7 has the ETM+
sensor array. This sensor has the same basic characteristics of the TM sensor except that
band 8 (15m panchromatic) has been added, band 6 (thermal) now has a 60m resolution and
it is collected in both high and low gain (see Figure 66 and Table 6).
Se nsor
TM
ETM+
Range of
Ba nd 1
Ba nd 2
Ba nd 3
0.45 - 0.52 0.52 - 0.60 0.63 - 0.69
0.45 - 0.52 0.53 - 0.61 0.63 - 0.69
30m
30m
30m
Me asure : Bandw idth (µ)
Ba nd 4
Ba nd 5
Ba nd 6
Ba nd 7
Band 8
0.76 - 0.90 1.55 - 1.75 10.4 - 12.5 2.08 - 2.35
N/A
0.78 - 0.90 1.55 - 1.75 10.4 - 12.5 2.09 - 2.35
.52 - .90
30m
30m
120/60m
30m
15m
Table 6 TM and ETM+ spectral band widths and spatial
resolution.
Figure 66 Relative spectral response curves for Landsat 7, 5 and 4.
The original choice of the spectral bands sensed by TM was primarily related to the spectral
reflectance of vegetation and the available atmospheric windows (see Figure 67 and Table 7).
Band 7 was the final band to be included due to its ability to discriminate geology and is the
reason why the band sequence is not in numerical order (see Figure 66).
5.1.1.2 Landsat acquisition
Landsat imagery is available through a variety of academic sources (NERC funded
researchers have free access to a variety of remotely sensed imagery including Landsat at the
NERC Earth Observation Data Centre) and third party vendors (such as Infoterra Ltd.:
http://www.infoterra-global.com). The whole Landsat archive is also available from the
United
States
Geological
Survey
(USGS)
127
through
the
EROS
data
centre
(http://edc.usgs.gov/). 19,662 free Landsat images of the majority of the Earth’s surface are
available from the Global Landcover Facility (http://glcf.umiacs.umd.edu/index.shtml).
Figure 67 Diagram of the visible and IR region of the EM
spectrum and Landsat TM bands. Gases responsible for
atmospheric absorption are indicated (after Sabins 1997 p. 5)
128
Band
Wavelength, µm
Lower
Upper
Sensors
Resolution (m)
1
0.45
0.52
16
30
2
0.52
0.60
16
30
3
0.63
0.69
16
30
4
0.76
0.90
16
30
5
1.55
1.75
16
30
6
10.40
12.50
4
120
7
2.08
2.35
16
30
Characteristics
Blue-green. Maximum penetration of water. Useful for
distinguishing soil from vegetation, deciduous from coniferous
plants, bathymetry, study of water laden sediments and
surface properties of snow/ice.
Green. Matches green reflectance peak of vegetation
(chlorphyll). Useful for assessing plant vigour, distinguishing
forest types, soil toxicity and pedology.
Red. Matches a chlorophyll absorption band that is important
for discriminating vegetation types.
Near Reflected IR. Vegetation survey through reflection of
mesophyll layer. Useful for determining biomass content,
healthy vegetation and for mapping shorelines.
Mid Reflected IR. Indicates moisture content of soil and
vegetation. Penetrates thin clouds. Provides good contrast
between vegetation types.
Thermal IR. Night time images are useful for thermal mapping
and estimating soil moisture.
Mid Reflected IR. Coincides with an absorption band caused by
hydroxyl ions in minerals. Useful as a lithological discriminant.
Table 7 Landsat TM interpretative sensor characteristics (after
Sabins 1997 p. 74; Campbell 2002 p. 173).
Landsat 5 imagery can be purchased with up to 3 levels of processing (however, precision
and terrain corrected imagery can only be purchased by approved United States government
and affiliated users):
1. Systematic correction: is both radiometrically and geometrically corrected.
Ephemeris data is used to geometrically correct the imagery.
2. Precision correction: is the same as systematic correction product with improved
geometric accuracy though the incorporation of Ground Control Points (GCPs).
3. Terrain correction: is the same as precision correction product but corrected for
topographic relief using a digital elevation model.
Landsat 5 imagery can be purchased from EROS at $425 per scene (c. 170 x 183 km) for
systematic correction processing, $550 per scene for precision correction processing and
$625 per scene for terrain correction processing. There is a significant discount for bulk
purchases.
Landsat 7 imagery can be purchased with up to 5 levels of processing. Levels 1P and 1T can
only be purchased by USGS approved researchers:
1. Level 0R: is the raw downloaded data with all the metadata required to conduct
radiometric and geometric correction.
129
2. Level 1R: is the Level 0R product radiometrically corrected.
3. Level 1G: is the Level 0R product that is both radiometrically and geometrically
corrected. Ephemeris data is used to geometrically correct the imagery. Residual
positional error is approximately 250 metres (1 sigma).
4. Level 1P: is the Level 1G product with improved geometric accuracy though the
incorporation of GCPs.
5. Level 1T: is the Level 1P product corrected for topographic relief using a digital
elevation model.
Further information about the processing of Landsat imagery can be found at the USGS
(2003d) website. Landsat 7 imagery can be purchased from EROS at $475 per scene for
Level 0R processing, $600 per scene for Level 1R or Level 1G processing, $725 per scene for
Level 1P processing and $800 per scene for Level 1T processing. There is a significant
discount for bulk purchases.
Furthermore, users can define aspects of the processing parameters for Landsat imagery as
outlined in Table 8.
Available options
UTM
Space Oblique Mercator
Albers Equal-Area
other
WGS84
NAD83
NAD27
other
cubic convolution (CC)
nearest neighbor (NN)
other
Map (north up)
Path (satellite; not recommended for UTM
projection)
30 metre (30m/120m)
28.5 meter (28.5m/114m)
other
available for up to 3 scenes
available in 10% increments (north-tosouth only)
Processing Parameters
Map projection
Horizontal datum
Resampling method
Image orientation
Pixel size
Multi-scene
Scene shift
Table 8 User definable processing parameters for Landsat imagery.
130
5.1.2 Ikonos
In 1994 the US approved the development of commercial satellite sensors with a ground
resolution of up to 1 metre. The first of these new generation of high resolution commercial
satellite sensors to attain orbit was the Ikonos satellite owned by Space Imaging. It
successfully achieved orbit on 24th September 1999 (Space Imaging 2003).
Figure 68 Relative spectral response curve for the Ikonos
Multispectral and Panchromatic bands (courtesy Space Imaging).
The Ikonos satellite provides 0.82m resolution imagery in panchromatic mode and 3.26m
resolution imagery in multispectral mode (4 bands: blue, green, red and near infrared with
similar spectral characteristics to bands 1, 2, 3 and 4 of Landsat (see Figure 67)). This spatial
resolution is resampled to 1m and 4m respectively. All the imagery has a radiometric dynamic
range of 11 bits (see Table 9 and Figure 68). The sensor can be pointed off-nadir to an angle
of 60° (i.e. acquisition from forward and reverse positions (Gerlach 2000)) and can hence
collect stereo imagery.
131
681 km
approx 10:30 am
Sun-synchronos
5
14.7
Yes
11 bit
programmable
Nadir: 0.82 - 3.2m
26° off nadir: 1 - 4m
MS Band 1: 0.45-0.52 µm (blue)
MS Band 2: 0.52-0.60 µm (green)
MS Band 3: 0.63-0.69 µm (red)
MS Band 4: 0.76-0.90 µm (near infrared)
Panchromatic: 0.45-0.90 µm
Yes
Nadir: 11.3 km
26° off nadir: 13.8 km
Altitude
Equatorial Crossing
Inclination
Number of Bands
Orbits per day
Programmable
Quantisation
Size of image
Spatial Resolution
Spectral Range
Stereo
Swath
Temporal Resolution
(Repeat cycle)
Cloud Cover
Sensor Type
c. 3 days at 1 m resolution, 40° lattitude
<20%
pushbroom and whiskbroom
Table 9 Ikonos technical specification (after Space Imaging 2003
p. 1).
There are six levels of Ikonos product determined by the level of positional accuracy and preprocessing (see Table 10). Some of these products require GCPs from the client.
Positional Accuracy
CE90 (m) RMS (m)
NMAS
Geo
15
Ortho
Corrected
Target
Mosaiced
Elevation
N/A
N/A
No
60° to 90°
Stereo
option
No
No
Cost (US $) per km²
Pan
MS
Bundle
28/20
28/20
Applications
32/22
Visual and interpretative applications.
Basic mapping projects
Standard
Ortho
50
25
1:100,000
Yes
60° to 90°
No
No
35
35
39
Reference
25.4
11.8
1:50,000
Yes
60° to 90°
Yes
Yes
Bespoke
Bespoke
Bespoke
Pro
10.2
4.8
1:12,000
Yes
66° to 90°
Yes
Yes
Bespoke
Bespoke
Precision
4.1
1.9
1:4,800
Yes
72° to 90°
Yes
Yes
100
100
Precision Plus
2
0.9
1:2,400
Yes
75° to 90°
Yes
Yes
120
n/a
Regional large area mapping and
general GIS applications.
Transportation, infrastructure, utilities
Bespoke
planning and economic development.
High positional accuracy for urban
110
applications.
Detailed urban analysis, cadastral and
n/a
infrastructure mapping.
Table 10 Ikonos product levels (after Space Imaging 2003 p. 2:
CE90 = Circular Error at 90% confidence, RMS = Root Mean
Square error and NMAS = US National Map Accuracy Standards).
Note the costings are based upon the May 2003 pricing levels and
the pricing for Geo represents bespoke and archive imagery.
The Geo Ikonos image is used in this research. It is the cheapest option in the range of
Ikonos products with the lowest level of geometric processing and has not been corrected
for terrain effects (Fraser et al. 2002).
132
Other satellite sensors are also available with comparable resolutions including Quickbird
(DigitalGlobe; 0.61m Pan, 2.44 MS) and EROA-A1 (ImageSat; 1.5m Pan). In 2001, the
approved ground resolution for commercial satellites was further reduced from 1m to 0.5m.
Satellites with improved spatial and spectral sensor systems are expected.
5.1.2.1 Ikonos acquisition
Space Imaging is a commercial, non-research venture. All company costs are recouped
through the sale of the imagery. It is expected that the majority of funds will accrue through
the seven year lifespan of the satellite with a further, diminishing, income stream from the
sale of archive imagery after the satellite is decommissioned. As Ikonos is a commercial
proposition the products come with stringent copyright and licensing terms (as detailed in
Space Imaging 2003). This has implications for the future re-use and archivingof the imagery.
Figure 69 The location of Space Imaging regional affiliates and
their direct spheres of influence (courtesy Space Imaging).
Imagery can be purchased directly from Space Imaging (USA), from one of its regional
affiliates (see Figure 69) or from third party vendors (such as Infoterra Ltd). Space Imaging’s
home web-site (http://www.spaceimaging.com) and those of its regional affiliates have
sophisticated viewers which allow the viewing and ordering of archive imagery on-line. The
133
price of the imagery is detailed in Table 10. This is subject to change and discounts are
offered on a regular basis. Furthermore, products can be ordered as bespoke imagery (by
defining the Area Of Interest (AOI) and the time-frame for collection) or as archive imagery
(imagery is archived 4 months after collection). There is a significant discount of c. 30% for
archived imagery. A minimum order size of 100 square km is necessary for new collections
and 49 square km for archive products.
Every Ikonos image comes with supporting metadata, licence and AOI shape files. However,
the Rational Polynomial Coefficient (RPC) camera model files are only supplied when stereo
imagery is purchased. Users can also define aspects of the processing and collection
parameters as outlined in Table 11.
Available options
UTM
State Plane
Albers Equal-Area
Lambert Conformal Conic
Tranverse Mercator
WGS84
NAD83
NAD27
cubic convolution (CC default)
nearest neighbor (NN)
GeoTIFF
Uncompressed NITF 2.0
8 bit (unknown reclass)
11 bit
On
Off
On
Off
Definable
Processing Parameters
Map projection
Horizontal datum
Resampling method
File Format
Radiometric Resolution
Dynamic Range Adjust
Tonal Balance Mosaic
Off-Nadir angle
Table 11 User definable processing parameters for Ikonos
imagery.
5.1.3 Corona
The Corona programme was endorsed by President Eisenhower in the late 1950s as a system
to improve the intelligence gathering efforts of denied areas (Brugioni 1996). The Corona
satellite system employed a high acuity panoramic camera for the collection of panchromatic
photographic images called Key Hole (KH). Unlike modern satellites, the Corona system was
placed in a decaying Earth orbit for ‘missions’ that lasted from 1 to 19 days. This allowed the
134
mission controllers to change the spatial resolution of the resultant photographic image by
changing the orbital characteristics of the capsule (see Table 12). Furthermore, the orbital
parameters could be adjusted for each mission. Although most launches coincided with an
image acquisition time of between 10:00 to 14:00 (to optimise lighting conditions), some
missions collected imagery at different times of the day (for example mission 1111 at c.
18:30).
There were a total of 134 launches between 1959 and 1972. Of these 134 launches, 102 were
considered successful, acquiring over 800,000 frames of photographs covering some 650 –
750 square nautical miles (Hall 1997). In 1995 President Clinton declassified the Corona
project and released the information to the public as it was no longer considered to be
militarily sensitive (Campbell 2002 p. 197).
Pe riod of ope ra tion
Amount of fra me s
Mission life (da ys)
Low e r Altitude
(e stima te d in km)
Highe r Altitude
(e stima te d in km)
Succe ssful missions
Ta rge ts
Ape rture w idth
Pa n a ngle
Ste re o a ngle
Le ns
KH-1
27/6/5913/9/60
1432
1
KH-2
26/10/6023/10/61
7246
2-3
KH-3
30/8/6113/1/62
9918
1-4
KH-4
27/2/6224/3/64
101743
6-7
KH-4A
24/8/6322/9/69
517688
4-15
KH-4B
15/9/6725/5/72
188526
19
192
252
217
211
180
150
817
704
232
415
n/a
n/a
1
USSR
5.265º
71.16º
3
5
Emphasis on USSR
5.265º
5.265º
71.16º
71.16º
F/5.0 Tessar F/5.0 Tessar
61
61
Foca l le ngth (cm)
40
25
Ground Re solution (ft)
50-100
50-100
Film (lp/mm)
15.3x209 to 15.3x209 to
Nomina l ground
cove ra ge ima ge fra m e 42x579 (km) 42x579 (km)
Nomina l photosca le in 1:275,000 to 1:275,000 to
1:760,000
1:760,000
film
20
49
16
W orld-wide/emphasis on denied areas
5.265º
5.265º
5.265º
71.16º
71.16º
71.16º
30º
30º
30º
F/3.5
F/3.5
F/3.5
F/3.5
Petzval
Petzval
Petzval
Petzval
61
61
61
61
12-25
10-25
9-25
6-25
50-100
50-100
120
160
15.3x209 to 15.3x209 to
13.8x188
17x232 (km)
42x579 (km) 42x579 (km)
(km)
1:275,000 to
1:300,000
1:305,000
1:247,500
1:760,000
Table 12 Corona Key-Hole camera mission characteristics (after
Galiatsatos in prep).
Throughout the lifetime of the Corona project a succession of Key Hole camera systems
were employed (KH-1, KH-2, KH-3, KH-4, KH-4A and KH-4B). Each camera improved
the resolving characteristics of the sensor system, predominantly through increasing image
resolution and decreasing platform vibration (see Table 12).
135
Figure 70 Corona module photographed at the Air and Space
Museum (Washington D.C., USA). Note the film spools and
stereo panoramic cameras. Courtesy of Keith Challis.
The film employed by the Corona mission was also improved during the lifetime of the
system. The KH-1 missions originally employed 3mm thick acetate film. Unfortunately this
film became brittle in a vacuum and had a tendency to break. To overcome this problem
Kodak developed a 2.75mm thick Estar (polyester) film. There is some dispute about the
actual film types and their specific spectral responses although Day et al. (1998) states that the
KH-4B camera employed the Kodak EK-3404 film format (see Figure 71 for its spectral
response).
Further information on the specific technical characteristics of Corona missions are available
in the following references: Campbell (2002), Day et al. (1998), Galiatsatos (in prep) and
McDonald (1997).
136
5.1.3.1 Corona archivation
There is very little information on the quality of storage between image capture in the 1960s
and 1970s and declassification and archivation in the late 1990s. Although it is known that
duplicates were made of the original photographic film, the relationship of the archived
‘original’ to the true ‘original’ is unknown. As discussed by Donoghue (2001 p. 556)
photographic film is a complex material that degrades with time, depending upon how it is
stored.
Figure 71 Spectral response of the Kodak EK 3404 film (after
Kodak 2003).
After declassification, copies of the original photographic film were duplicated by Eastman
Kodak and sent to USGS for metadata digitisation, archivation and re-sale. The original
photographic films (and a duplicate positive copy) were sent for long-term archivation at the
National Archives Record Administration (NARA) in Maryland, USA. NARA also possess a
full set of all the paper catalogues, mission summaries, camera manuals, mission evaluations,
technical review reports and other materials generated by the CIA during the life of Corona.
137
5.1.3.2 Corona acquisition
Corona can be acquired through the USGS website via the sophisticated EarthExplorer web
interface (http://edcsns17.cr.usgs.gov/EarthExplorer/). This interface allows a variety of
textual and spatial searches of multiple image archives including Corona. The interface allows
access to some supporting metadata and provides the user with a preview image. Previewing
Corona imagery is essential as it is the most effective way to determine the amount of cloud
cover on the imagery and if that cover affects the AOI. However, as the imagery is only
crudely located it is advisable to expand the AOI during any image searches.
Corona imagery can be purchased as a black and white print ($14), film positive ($18) and
film negative ($18) exclusive of a small handling charge and postage. The KH-4B media is
2.25” by 29.9” (USGS 2003b; 2003a).
Alternatively, one can duplicate the archived copies of the Corona photography held in
NARA. Although this is a more time consuming process it has been argued by Vick (1999
cited in Galiatsatos in prep) that this will produce a higher quality product. It is advisable to
retrieve all the AOI and metadata information from EarthExplorer before visiting NARA.
5.1.4 Ancillary data sets
In addition to the satellite imagery a number of historic and modern maps were procured.
Historic mapping was available from the Royal Geographical Society (see Table 13). This
mapping was pre-scanned by the Royal Geographic Society in TIF format.
These scans were geo-referenced using the corrected Syrian mapping. The techniques used
are described in section 5.3.3. As the original scans were of unknown quality and are only for
visual and comparative purposes a RMSE of 10 pixels or less was deemed acceptable.
Country of Origin
England
England
England
French
German
Ottoman
Syrian
Sheet Nos.
NI-37-XIII-4a,
Blatt Nr. 27
Scale
1:100,000
1:10,000
1:50,000
1:50,000
1:50,000
1:100,000
1:25,000
Year
1920
1952
1952
1933
1941
1932
1980s
Map_Type
Demographic
Topographic
Topographic
Topographic
Topographic
Topographic
Topographic
Table 13 Summary of historic and modern mapping used in the
research.
138
Photocopies of eight modern (1980s) 1:25,000 Syrian maps of the Homs region were also
available. These photocopies were of poor quality in some places. These maps were scanned
on a Contex A0 greyscale roller scanner in 8 bit and converted into a 1 bit tiff. The existing 1
km grid was used to calibrate the images for scanning and photocopying errors and georeference the images in the Syrian Grid (see section 5.4 for a discussion of the Syrian Grid).
Once geo-referenced they were merged into one image file.
5.1.4.1 Digitising the Syrian mapping
Digitising of the modern Syrian mapping was undertaken in AutoCAD MAP
CADOVERLAY using the conventions described in Table 14. The CADOVERLAY
extension has a line following algorithm that simplifies digitising of complex basemaps. All 3dimensional information had its elevation incorporated during the digitising process so that
they could be used for creating a DTM (see section 6.6). Unfortunately the photocopying
errors are particularly significant at the northern boundary of the Northern application area
(see Figure 72). This also has ramifications for any derived terrain model created using the
digitised contour data. Subsequently contour data was integrated from the 1:50,000 mapping
for these areas. However, this does pose some potential problems as the data is derived from
different scales.
Theme
Archaeology
Archaeology
DTM
DTM
DTM
DTM
Hydrology
Hydrology
Hydrology
Hydrology Network
Hydrology Network
Hydrology Network
Hydrology Network
Mapping
Mapping
Rail Network
Road Network
Road Network
Road Network
Site Polygons
Site Polygons
Study Area
Layer
Fieldboundary
Structureamorphous
Contour
Spotheight
Trigpoint
Contourpoint
Lakeedge
Marsh
Riveredge
Canal
Irrigationchannel
Rivercent
Seasonalwater
Bridge
Unidentifed
Traintrack
A-road
B-road
Track
Sitecent
Sites
Studyarea
Topology
Network/Polygon
Polygon
None
None
None
None
TBC
TBC
TBC
Network
Network
Network
Network
None
None
Network
Network
Network
Network
Point/Polygon
Polygon
None
Description
Field boundaries identified from the mapping
Amorphous structures identified from the mapping
3D Contour lines
3D Spot height
3D Triangulated trig. toints
3D points 'weeded' from the contour lines
Lake edge
Marsh edge
River banks
Network of canals
Network of irrigation channels
Network of rivers
Network of seasonal water courses
Bridges
Unidentified features
Rail network
Network of major roads
Network of minor roads
Network of tracks
Centroid for 'sites' polygon (seed) containing identifier
Boundary for 'sites' polygon
Boundary of study area
Total
Objects
6308
1035
535
470
27
58701
1
2
5
37
117
144
403
2
4
3
608
1437
425
282
282
2
Segments
13144
2003
102014
470
27
58701
1227
491
1929
182
259
1441
4277
6
9
53
2109
10409
3719
282
3643
23
70830
206418
Table 14 Layers and summary information for the digitising of the
map-base
Once the digitising was completed each layer was cleaned and topology was created within
MAP (see Table 14). These files were exported as shapefiles and incorporated into the
project GeoDatabase (SHR_GeoBASE.mdb see Figure 72).
139
Figure 72 Digitised vector basemap. Note the poor quality contour
data in the northern segment of the northern study area.
5.1.4.2 Aerial photography
During the August 2003 fieldwork season Russian vertical photography (taken in 1958) was
made available through Dr. Mamoun Abdulkareem (Head of Museums, DGAM). Five
140
frames intersected with the northern application area. These were scanned using the
resources available in Syria. Each photograph was scanned on a Canoscan GDE 20 flatbed
scanner as an 8 bit greyscale tiff file at 600 dpi (the maximum physical resolution of the
scanner) providing a ground resolution of 1 metre. These images were then geo-corrected
within Erdas Imagine 8.4 (using a first order polynomial with an RMSE error of less than 1
pixel) with GCPs supplied from both the geo referenced Ikonos and Corona (mission 1110,
1970) imagery (discussed in section 5.4). The Corona data was required because the
photographs extended beyond the range of the Ikonos imagery available, therefore control
points had to be established from alternative sources. This was not an ideal geo-referencing
procedure, particularly when one considers the quality of mission 1110 Corona imagery (see
Figure 73). Unfortunately the higher quality mission 1108 Corona imagery did not extend
into this region.
The receipt, scanning and rectification of this imagery took under 4 hours to achieve. Such
speed was only possible because of the pre-registered Ikonos imagery. It is important to
recognise that collection of further ground control for these images would have taken a long
time: the requirements for these photographs are much more rigorous than that of Corona.
A large Corona strip could be coarsely referenced using handheld GPS but it would have a
footprint of many kilometres. However, the individual Russian photographs have a footprint
of some 4 x 4 km, and finding appropriate hard-detail, in the absence of Ikonos and/or
Corona, is difficult. The success of this initial conversion procedure will hopefully result in
the procurement of more imagery under the direction of Dr. Abdulkareem. This
photography is evaluated against the Ikonos and Corona imagery in section 7.4.6.
5.2 Image selection
Image selection is either the collection of new imagery at propriotous times, the selection of
specific image sources from an archive of imagery or the reduction of the corpus of imagery
by the selection of specific sub-sets. Prior to starting the research project Dr. Philip and Dr.
Donoghue had already acquired a batch of satellite imagery. They had decided to focus on
the Corona missions which employed the KH-4B lens as these missions provided both
stereoscopic coverage and the highest spatial resolution. A sub-set of the mission data which
intersected the application area was selected and purchased. This sub-set was based upon
appropriate spatial resolution, good ground cover, a low cloud cover index and different
seasonality so that the imagery could be effectively compared and evaluated (see Figure 74).
141
Mission 1111 was added due to the anomalous collection time of c. 18:30 hours. All missions
(1108, 1110 and 111) had imagery collected from the aft camera and mission 1110 also had
stereo imagery from the forward camera. For consistency the aft camera photography for
each mission has been employed unless stereo analysis has been conducted.
Figure 73 Comparison of the Russian aerial photography, Ikonos
and Corona imagery (scale 1:5,000).
The Landsat sensors have a large archive of imagery. It was initially assumed that the Landsat
imagery would provide important information pertaining to drift geology and vegetation.
Hence images were selected that covered different environmental regimes in the application
area. Two scenes were selected from 1987 (26th May and 1st October respectively) which
display the maximum difference between crop cover in the landscape. May is just prior to
142
harvesting when the crops are at their peak and October is prior to the winter rains, at peak
aridity, where crop cover is minimal (see Figure 102).
The Corona and Landsat imagery provided the benchmark from which to evaluate the most
appropriate timing to collect the Ikonos imagery. Initial analysis of these Landsat and Corona
images gave great insights into the operation of environmental processes and recent
landscape changes in the application area.
Figure 74 Sensor comparison over different feature types.
5.2.1 Atmospheric ramifications
Although it is difficult to compare the Corona imagery on a like-for-like basis as each mission
had different orbital characteristics, it was rapidly deduced that increased atmospheric
particulates in the spring and summer vastly reduce image quality (Donoghue et al. 2000).
Particulates are reduced in the winter months when the wind channelled through the HomsTripoli gap is reduced and the atmosphere is cleansed with the first winter precipitation (see
Figure 173).
143
5.2.2 Land management ramifications
Agricultural practices in the majority of the basalt and marl zones are based upon rain-fed
agriculture. Most rainfall occurs between December and April (see Figure 77). Increased river
levels are observed for a few months after this period as recharged aquifers feed local springs
and snow melt (from the anti-Lebanon range) adds to the surface and subsurface
hydrological regime.
Figure 75 Rams and reservoir in the basalt.
During this period, in the marl zone, the water table temporarily rises and coupled with the
increased surface run-off some of the inactive wadi channels become active again. In the
basalt zone, natural depressions (ram) and purpose built reservoirs hold standing water (birka,
see Figure 75). This hydrological system provides enough water for one main season of
winter crop (between December and May see Figure 76) in all zones. According to Rodriguez
144
et al. (1999) encroaching desertification means that this area may soon be on the margins of
sustainable agriculture. However, in exceptional years two crops may be realised (Bshesh,
pers. comm.). This cropping cycle can be extended by irrigation. The alluvial zones and their
margins are heavily irrigated and hence these areas are under nearly continuous crop cover.
The same is true in the northern marls where, if enough water is in reserve in Lake Qatina,
the French irrigation canals allow multiple cropping regimes. The image selection process
(see section 5.2) determined that imagery collected during the months of September or
October would be most appropriate for archaeological detection. The surface cover at this
time predominantly consists of crop (alluvium), bare soil/scrub (marl) and grazing/scrub
(basalt). However, where extensive irrigation does intersect with the area of interest it may be
more appropriate, although logistically difficult, to collect imagery between different crop
rotations to ensure that the major component of reflectance is soil.
Figure 76 Elevation and annual precipitation (after Hirata et al.
2001 p. 509) in Syria.
145
This general increase in irrigation has extended the amount of time during which bare soil is
masked by vegetation. Concomitantly this decreases the window of opportunity to collect
appropriate imagery that contains predominantly bare soil to prospect for soil marks.
Furthermore, it can be argued, that the use of deeper ploughing techniques damages the
underlying archaeological residues and moves them to the surface (Lambrick 1977). Although
destructive this could make the surface identification of archaeological residues much easier.
5.2.3 Field observations
The first phase of fieldwork (October 1999 to January 2000) was undertaken to refine some
of the methodological field techniques which would be employed by the SHR project in
subsequent field seasons. It also allowed an initial understanding of the landscape under
study and the nature of the archaeological residues. This initial fieldwork led to the following
conclusions:
•
Except for fluvial margins the landscape could be considered as either completely
bare soil or a combination of bare soil and crop throughout the year.
•
Site soil colour in the marl zones was significantly different to off-site soil colour
when dry and similar when wet.
•
Areas of high artefact density had a positive relationship with areas of light soil
colour in the marl.
•
Establishing sites from crop marks would be difficult due to the perceived lack of
negative features (i.e. ‘positive’ mud-brick construction as opposed to ‘negative’
postholes and ditches). The main agricultural season was between October
(seeding) and May (harvesting).
•
The majority of walls in the basalt zone have a width of between 0.5 and 2m.
5.2.4 Determining acquisition times
To determine when would be the most appropriate time of year to purchase Ikonos imagery
the data collected from the fieldwork and through studying the Landsat and Corona imagery
was combined. The rationale was to ensure that the collected imagery provides the maximum
observable information for the phenomena of interest (Teng 1997).
146
For the mapping of small features, such as field walls, image fidelity needs to be high. Hence,
the summer months (May to September) where increased airborne particulates increase
specular reflection and therefore decrease spatial resolution were to be avoided. Furthermore,
collecting lithological information is better after a rainfall as the surfaces under study are
clean of dust. This could improve the reflectance of the basalt walls.
In the marl and alluvial zone it was considered important to collect imagery when there was
little or no crop cover. This is primarily related to the nature of the archaeological residues
and the difficulty in targeting crop marks. European models of crop mark formation are
based upon the theory that crop vigour or crop stress is a function of sub-surface
archaeological residues such as walls and ditches, and their effect upon soil moisture (see
Figure 47). However, it has yet to be demonstrated if this is an appropriate system for the
Homs area of Syria (or the Middle East generally) where the formation and deformation
systems are strikingly different. For example, walls have limited footings and ditches rarely
form part of the archaeological repertoire.
The majority of settlements were built of mud-brick and incorporated a variety of organic
material that did not transform the landscape in the manner observed in Europe. Hence,
European models of crop mark detection would need fully re-evaluating for this
environment. Crop mark formation is a function of the environmental conditions which are
different between Europe and the Middle East. Furthermore, the visible effects of crop
marks have a short time span (see, for example, Figure 48) and the interpretation of crop
mark evidence as a systematic prospection tool is subject to bias (Cowley 2002). Hence, even
if crop marks were susceptible in the application area, the likelihood of intersecting the
period of peak crop mark formation would require extensive determination. Even if these
times were determined it would still be difficult to capture them with current sensors (even
though Ikonos is a bespoke platform the ordering process is time consuming and the exact
time of the collection can not be guaranteed (see section 5.2.5)). Finally, would the sensors
have the spatial resolution to detect crop marks? The 1m resolution Ikonos panchromatic
may display crop marks, but, the NIR band in the 4m multispectral is likely to improve
detection. It is unlikely that linear crop marks can be adequately detected with a 4m sensor.
Using Jensen’s approximation the crop mark would have to be 8m in diameter! However, it
can be hoped that future sensor systems with increased spectral and spatial resolution will
allow the earlier and later identification of differential crop stress and vigour characteristics,
147
allowing the window of opportunity for identification to be extended (Hanson, pers. comm.).
Therefore, as the likelihood of detecting crop marks was very low the months of intense crop
cover (February-May) were avoided.
Homs region monthly rainfall averages
120
100
FAO-Clim2
(unknown dates)
Rainfall (mm)
80
60
Gruzgiprovodkhoz;
Oct 1955 - May
1980
40
20
st
ly
Ju
gu
Au
ay
r il
ne
Ju
M
Ap
ct
O
N
Se
pt
em
be
r
ob
e
ov
r
em
b
D
ec er
em
be
r
Ja
nu
a
ry
Fe
br
ua
ry
M
ar
ch
0
M onth
Figure 77 Monthly rainfall average for the Homs region (data
kindly supplied by ICARDA). Total rainfall is 480mm for FAO
and 442.6mm for Gruzgiprovodkhoz.
Fieldwork had demonstrated that sites could be readily identified from soil colour differences
alone. Representative samples were taken for both on and off-site soils. Amongst other
measurements, the soil colour values were taken in both ambient (dry) and wet conditions.
Colour was calculated by reference to a standard colour matrix reference sheet. Originally the
archaeological soils recording chart produced by Artacorn (Middleton 2000) was used, but it
did not prove to be adequate for recording soils in arid environments. Hence the Munsell
colour charts (Munsell Colour Company 1975) were employed. The results showed that the
colour difference between wet and dry off-site soils is very close. However, there is a large
difference between wet and dry on-site soils (wet soils are significantly browner (Munsell
1981 pp. 55-57)). Furthermore, the on-site wet colour is very close to both the wet and dry
off-site soil colour (possibly highlighting the shared relationships with parent material or
improved drainage on sites). From this information it was deduced that the lower the soil
moisture (or the higher the aridity) then the greater the colour difference between on and off148
site soils and hence the easier it is to discriminate archaeological residues. After talking to
local farmers (Bshesh, pers. comm.) and evaluating the rainfall data provided by ICARDA
(see Figure 77) it was determined that peak aridity would occur between the months of
September to December (this includes a time lag for the evaporation of moisture).
Hence the months of February to September were to be avoided due to inclement crop
cover or environmental factors. The ideal months for acquiring satellite imagery for
archaeological prospection are during the months of November and December. Although
some fields are sometimes under crop at this time the crop is usually immature, the initial
rains would still not unduly affect the overall soil colour but would remove dust from
surfaces in the basalt area.
5.2.5 Data sets acquired
With this in mind, Ikonos imagery for the application area was ordered from Infoterra Ltd. to
be collected between November and December 2000. However, due to communication
problems between the various agencies in the supply chain this collection programme was
not successful. Although a small sample of Ikonos data was retrieved from Space Imaging’s
archival data, the lack of Ikonos imagery necessitated a six month suspension of the research
programme between July and December 2001. Another application for imagery was
requested for November 2001. Unfortunately, this collection period coincided with the
military action in Afghanistan and hence the Ikonos satellite was in heavy demand.
Syria currently is part of one of the "area of high contention" for obvious global political
reasons and, as such, our acceptance of new collections in these regions have been impacted.
We can certainly accept the order but will note that the backlog is such that the order will
not receive any priority for approximately 30 to 45 days and, even at that point, standard
collection times cannot be guaranteed. Also, due to the severity of the situation, we cannot
accommodate rush tasking requests for the same reasons.
(Key 2001)
The Ikonos imagery was finally collected on 26th January 2002 and 3rd February 2002. The
total cost of the imagery was £21,770.26 excluding VAT and delivery (1m and 4m Geo
149
bundle for the Northern area @ 195 sq. km £7,512.63 and 1m and 4m Geo bundle for the
Southern area @ 370 sq. km £14,256.63). The imagery was delivered on 4 CD-ROMs to Dr.
Donoghue on the 20th February 2002. Unfortunately, one set of multispectral data for the
southern area (Order ID: 87554_0010000 MS) was corrupt. All disks were returned to Space
Imaging and replacement data was received by Dr. Donoghue on the 26th April 2002. Due to
an administrative error by Space Imaging the second set of disks did not contain the
Panchromatic or Multispectral imagery for the Northern area. This completed data set was
finally received by Dr. Donoghue on 27th June 2002.
An initial inspection revealed that the quality of the data sets was very good. The supporting
metadata is excellent. Cloud cover is low (the largest obscured image has < 5% cloud). Of
the 2048 DN values possible for 11bit data each scene, on average, has a dynamic range of
approximately 400-600. Image clarity is very good (individual olive stands can be resolved in
panchromatic and inferred from MS). Image geometry is good (within Space Imaging’s
specification and closely correlated with handheld GPS readings) with a maximum
displacement between overlapping images of 25m in the South and 15m in the North. This
good correlation is in part due to the limited relief.
Due to the size of acquisition each application area was collected in two blocks (a western
and eastern block). Unfortunately, the southern blocks did not overlap which necessitated the
collection of a third block of data. This data will be useful for examining the
photogrammetric potential of the geo product. Although the images were not collect in the
stipulated time frame and final delivery was some 7 months after the closure of the collection
window it was recognised that this would not significantly affect the research. A combination
of good radiometric resolution and a less dusty atmosphere (working on the assumption that
the atmosphere has reduced particulate matter after the rains) has provided higher quality
data than would have been expected during our stipulated time frame even though the
environmental characteristics are not ideal (see Figure 78). Some equalisation of surface soil
colour must be expected in the area due to an increase in soil moisture but it is difficult to
quantify such a problem. This is exemplified in Figure 78 where haze effects are visible in the
September image but not in the February images. This haze effect has reduced the effective
spatial resolution of the September imagery. For example, the markings in the open ground
and detail on the rooftops are blurred in the September image.
150
Figure 78 Comparison of the spatial clarity and geometric accuracy
of raw Ikonos panchromatic imagery from different dates and
acquisition orientations.
151
Figure 79 Comparison of the different rectified Corona missions.
152
Three missions of Corona imagery were purchased as detailed in Table 15. Crop cover in
mission 1108 and 1111 imagery should be minimal in most areas as these images were taken
pre and post the growing season. Furthermore, mission 1111 imagery was collected at c.
18:30 hours instead of in the morning, which may exacerbate shadow marks. Mission 1110
was collected during peak crop cover. The imagery quality of this mission was quite poor in
comparison to the other Corona images.
Mission No.
Date
Time
Season
Image fidelity
No. of frames
Stereo
1108
17/12/1969
c. 10:30
pre-crop
Good
3
D203043-5
N
1110
28/05/1970
c. 10:30
pre-harvest
Poor
6
D106007-9
D106013-5
Y
1111
31/6/1970
c. 18:30
post-harvest
Good
2
D135001-2
N
Table 15 Corona acquisition summary.
Two full scenes of Landsat 5 were purchased for the 26th May 1987 and 1st October 1987
with a processing level of systematic correction. Both images were in UTM map projection,
with the WGS84 datum, used a nearest neighbour resampling method and a 28.5m pixel size.
Two full scenes of Landsat 7 were purchased for the 14th January 2000 and the 28th October
2000 with a processing level of 1G. Both images were in UTM map projection, with the
WGS84 datum, used a nearest neighbour resampling method and a 28.5m pixel size.
5.3 Image pre-processing
As discussed in section 2.2.5, image pre-processing is a preparatory step to remove
radiometric and geometric discrepancies in imagery. The Corona imagery also requires
converting from an analogue to a digital medium.
5.3.1 Digitising Corona
In order to convert the Corona photographic negative into a digital image it requires
scanning. The film strips cover a large ground footprint. The first task is to locate and mark
out the extent of the areas of interest on the negatives, so that only these areas are scanned.
153
Figure 80 Corona imagery scanned at different resolutions
As the resultant digital facsimile will be employed in a variety of quantitative processes the
spatial and radiometric fidelity of the scanning process is of paramount importance. For this
reason, desktop scanning solutions were discounted due to their low spatial and radiometric
fidelity (Coburn et al. 2001). The Corona photography has a nominal resolution of 160
lp/mm (see Table 12). Full resolution scanning would require a scanner with 3µm (c. 8000
dpi) resolution which are not available in a desktop range. Other researchers (Ur 2003,
Challis pers. comm.) have used desktop scanners for digitising Corona photographic prints
with adequate results. However, it has yet to be demonstrated if these less rigorous
approaches facilitate a broad range of quantitative techniques (for example DTM creation see
section 6.6).
154
A high resolution photogrammetric scanner (a Vexcel VX4000) was employed to scan the
imagery. This scanner has a high resolution scan head without interpolation, a geometric
accuracy of 1/3 pixel RMSE and a radiometric accuracy (in 8 bit) of 2 DNs RMSE.
For evaluation purposes the same segment was scanned at four different resolutions (22.5,
20, 15 and 7.5 µm (see Figure 80)). The 7.5 µm resolution produced the best results and was
used for all the other Corona negatives. The cost of each scan at 7.5 µm was £18.
Figure 81 Subtracting a constant from a band is equivalent to
translating the origin and has no effect on the variance-covariance
matrix. Hence dark object subtraction has no effect on
classification results (after Song et al. 2000 p. 232).
Galiatsatos (in prep) discusses the radiometric and geometric effects of the scanning process
and the geometric distortions inherent in panoramic cameras in detail.
155
5.3.2 Atmospheric calibration
As discussed in Chapter 2, electromagnetic energy detected by a satellite sensor, particularly
those in the optical region of the spectrum, consist of a mixture of energy reflected from or
emitted by the ground surface. This energy is modified by a variety of mechanisms while
travelling through the atmosphere (see Figure 9, Figure 13, Figure 14, and Figure 15). Hence,
energy recorded at the satellite sensor is not a true measurement of surface reflectance
(Franklin and Giles 1995; Song et al. 2000).
Atmospheric effects can mask the subtle differences in reflectance or emmittance. For optical
satellite sensors absorption is less important. As described in Figure 67 the bands are located
in regions with low absorption characteristics (Song et al. 2000 p. 232). Hence, scattering is
the main effect which, in a scene with an homogenous atmosphere, normally produces a
linear (additive) offset and does not affect either variance or covariance (see Figure 81).
When, then, does one correct for atmospheric effects? Tso and Mather (2001 p. 12) and
Song et al. (2000 p. 232) recommend that if only single date images are used for land cover
identification then it can be assumed that all pixels are equally affected by atmospheric
processes and atmospheric correction need not occur.
However, correction may prove desirable when one intends to conduct time change analysis,
requiring comparison of images taken at different times. There are four main methods of
correction:
1. Dark object subtraction (Histogram adjustment): This method is based upon the
assumption that some objects in a scene should have zero reflection. Hence, any
DN value recorded in these pixels is derived solely from atmospheric scattering.
This value is offset against all values in the band. Shadows are commonly used as
dark objects in the visible bands and deep clear water in the Near Infrared (Tso
and Mather 2001 pp. 13-14). Multiple image normalisation improves this
technique for multi-temporal image sets (Jensen 1996 p. 116). Although dark
object subtraction is easy to apply it provides only an approximation and can fail
in some areas (Liang et al. 2002).
2. Histogram Equalisation: If images are collected at similar times and on similar
dates then many of the environmental factors can be assumed to be the same
156
(solar illumination, crop cover, rainfall etc.). It is further assumed that matching
the histograms of each image will suppress atmospheric effects.
3. Use of a model atmosphere: An assumed atmosphere is calculated using the time
of year, altitude, latitude and longitude of the application area and environmental
variables. This model is used to determine what corrections are required for each
band. However, determining the ambient variables for this technique can be
difficult.
4. Correction using contemporaneous ground readings: If ground reflectance
readings have been taken in the AOI with a spectro-radiometer during the
satellite collection phase then these readings can be used to accurately correct for
atmospheric effects at the sample points. This correction can then be
extrapolated across the whole scene using, for example, the empirical line method
(Karpouzli and Malthus 2003). However, the large spatial resolution of most
satellite imagery means that this technique is rarely used.
Other researchers have attempted to remove atmospheric noise by using multi-temporal
Principal Components Analysis (PCA) with some success (Song et al. 2000).
For time change analysis only comparable sensors will be analysed together. The Landsat
imagery was corrected for atmospheric, haze and solar elevation effects using the Erdas
module produced by Skirvin (2003). This module applies the correction methodology
outlined by Chavez (1996) based on dark object subtraction. Furthermore, this algorithm
produces 6 band data (by removing both the panchromatic (for ETM only) and the thermal
bands). Due to the complex differences in geometric fidelity and spectral content between
the Corona and Ikonos and the lack of atmospheric data for the Corona imagery, histogram
matching was decided as the most appropriate mechanism for the Ikonos and Corona
imagery. However, this process disrupts the structure of the image histogram and will only be
applied when necessary.
5.3.3 Correction of topographic effects
Most remote sensing classifications assume that the terrain under study is flat with
Lambertian reflectance behaviour. However, topographic slope and aspect may introduce
radiometric distortion (Franklin and Giles 1995; Tso and Mather 2001 p. 21). In some
instances the area may be in complete shadow, dramatically affecting the brightness value.
157
The goal of topographic correction is to remove all topographically induced illumination so
that two objects having the same reflectance properties show the same brightness no matter
what their orientation to the Sun’s position. Topographic correction has been shown to
improve some classifications (Jensen 1996 p. 122) and reduce the visual impression of terrain
ruggedness. Most topographic correction procedures require a digital elevation model (DEM)
as a basis for modelling, although others use ratioing procedures (for example Holben and
Justice 1981).
Fortunately, the application area is relatively flat (see Figure 82). With the exception of the
basalt walls and tells, few archaeological residues demonstrated a significant topographic
component. No topographic correction was applied for the majority of analyses. However,
Galiatsatos (in prep) has conducted extensive research into digital topographic modelling in
the area.
Figure 82 Isometric view of the application area and contour lines.
5.4 Geo-rectification
Archaeology is a spatial discipline: every archaeological residue has a spatial component. It is
therefore important that archaeological residues are located somewhere in space.
Rectification is the process of correcting systematic and random errors in imagery.
Rectification procedures can either be spatial or non-spatial. Non-spatial rectification is
commonly used to correct camera lens and scanning aberrations (or other errors in a
collection device). Spatial rectification is used to locate imagery somewhere in space. Image
rectification places a data set with a spatial component into a spatial framework. Amongst
158
other things, this will allow accurate measurements to be taken from the imagery, and
integration with other spatial data sets and spatial data collection devices.
Prior to any rectification or data collection procedure, a projection system needed to be
determined. In most areas with institutionalised CRM bodies, the regional or national
projection system is easily accessible. It is advisable (and in some instances mandatory) that
this projection is used. This will ensure that any results will integrate seamlessly with the
national CRM data and other data sets enabling subsequent data re-use and integration
(Bewley et al. 1999).
Where such a system does not exist, it is advisable to use one of the standard worldwide
referencing systems such as Universal Transverse Mercator (UTM) or Lat/long projections
(both standard worldwide reference projections) and an appropriate datum (if in doubt use
WGS84). All systems must support the projection used. Pre-registered satellite imagery will
normally come in a worldwide referencing system. If the registered imagery needs reprojecting by one of the many available algorithms, then some data loss is likely due to the
pixel resampling technique employed (see Figure 29). However, many current GIS and image
processing systems allow on-the-fly rectification of imagery in different projections.
During the first two fieldwork seasons the only rectification medium available to the project
was the 1:25,000 Syrian mapping (in a projection that was referred to as the Syrian Grid) and
handheld GPS. At the time (late 1999) the use of handheld GPS was discounted as Selective
Availability (SA: the deliberate degradation of GPS accuracy) had only just been removed and
there was no reference material to determine the accuracy of handheld GPS (Rees 2001). The
use of Differential GPS (DGPS) was impossible due to military restrictions. Hence, the
Syrian mapping was the only available resource. This was also the grid system employed by
the Homs regional office of the Directorate General of Antiquities and Museums. Using
GCPs identified from the maps both the Corona and Landsat imagery were corrected to the
Syrian Grid.
When overlying this information with data collected by handheld GPS (in UTM 37N) it was
obvious that the Syrian Grid and UTM, although similar, were not the same. The only
indication the Syrian maps gave for their coordinate system was a reference to a ‘4th spheroid
UTM’. The Syrian Grid required offsetting, rotating and scaling to fit the UTM grid
159
(Galiatsatos in prep). In order to re-project the Syrian Grid into UTM an understanding of
the parameters of the projection system were required. Contacts at the Directorates of
Museums (including the Al-Bassal centre), Agriculture, Remote Sensing and Survey were
approached to see if they could provide any further information on the technical
characteristics of the projection. Unfortunately, for bureaucratic or sensitivity reasons this
information was unavailable. However, it was determined that the majority of mapping
created within the directorates (including the Al-Bassal centre which is in the process of
creating a national archaeological inventory) was projected in UTM.
During the third field season (17th April to 10th May 2001) a comparison (see Figure 83) was
made between GPS readings and the recently received 6 x 7 km archived Ikonos ‘geo’
panchromatic image (collected on 6th September 2000). A very high spatial correlation was
observed between the Ikonos imagery and handheld GPS readings. The excellent geometry
and clarity of the Ikonos panchromatic meant that this imagery could be used to supplant the
Syrian 1:25,000 map series or GPS co-ordinates for rectification purposes. Furthermore, the
accuracy of the handheld GPS appeared to be about 4-5 m in this region (Rees 2001).
Figure 83 Raw GPS readings overlying raw Ikonos data (after Beck
et al. in press).
On the basis that UTM was used by many of the Syrian directorates and it fulfilled many of
the other criteria previously outlined the Syrian Grid was abandoned in favour of UTM.
Similar problems were encountered by Harrower et al. (2002) in their projection and datum
definition. It was subsequently ascertained after conversations with Dr. Meredith Williams
160
(Department of Geomatics, University of Newcastle-upon-Tyne) that the Syrian mapping
was probably based upon Russian military mapping. Russian mapping of this nature employs
the Grauss-Kruger projection.
UTM is more intuitive for in-field work than Lat/Long (units are metres as opposed to
seconds of arc). It is also widely supported (e.g. by Landsat, Ikonos and most GPS systems)
and is already used by Cultural Resource Management databases elsewhere in the region
(Palumbo 1992). World Geodetic System (WGS) 84 is used for the datum.
Beck et al. (in press) discuss a comparative rectification methodology using GPS and the 6 x 7
km archived Ikonos as sources for Ground Control Points (GCPs) to reference the Corona
imagery. They determined that Ikonos imagery provides the best reference source (see Figure
84) due to its ability to provide a substantially greater number of concurrent GCPs. It should
also be noted that locating GCPs by GPS for Corona imagery is very difficult: the landscape
has changed significantly in the intervening 30 years. This situation was also noted by
Altmaier and Kany (2002 p. 227).
Upon receipt of the whole Ikonos data set it was possible to expand the results from the
small scale 6 x 7 km area to the whole application area. After examining the different Ikonos
images and comparing them to GPS readings (see for example Figure 78) it was obvious that
geometric accuracy of the imagery was not consistent. For some reason the ephemeris
characteristics of the September 2000 Ikonos imagery allowed a particularly accurate
rectification. In light of this information the results of the rectification process undertaken by
Beck et al. (in press) were re-evaluated.
The geometric errors associated with the Ikonos imagery still provide a good level of
accuracy for rectification across the application area. Any image rectified using the Ikonos
imagery as a GCP source will have the combined error of the rectification process and the
original error of the Ikonos basemap. The accuracy of the ‘geo’ imagery (> 25m RMSE) is
still appropriate for the mapping and location of archaeological residues in the marl where
absolute accuracy is not as important. However, in the basaltic landscape some discrete
elements are less than 10 m in width. Hence, the accuracy of the Ikonos imagery is too coarse
to enable accurate desk based mapping in this zone. In such instances, field checking based
upon GPS navigation of the digitised segment would require one to make a choice between a
161
number of wall segments. Fraser et al. (2002) had improved the accuracy of the Ikonos geoproduct to sub-meter levels by using Differential GPS to re-geo-correct the imagery. They
noted that the internal geometries of the Ikonos imagery were very accurate and hence only a
few GCPs were required for the correction. They found that a configuration of 3 good GCPs
gave similar accuracies to 6 or 8. It was recommended that the centre of roundabouts
provided some of the best GCPs.
Figure 84 Comparison of Corona rectification using GCPS derived
from GPS and Ikonos (after Beck et al. in press).
Applying the methodology of Fraser et al. (2002), 15 GCPs were taken with hand-held GPS
throughout the application areas. At each piece of detail an average of 100 GPS readings was
taken to produce 1 point. This was repeated so each location had 2 GCP points (in retrospect
it would have been more appropriate to take 200 individual readings and then perform a
more rigorous averaging technique). A photograph of each GCP point was taken to aid the
subsequent rectification procedure (see Figure 85). Three GCPs were used to correct each of
the five Ikonos pan images using Erdas Imagine with a first polynomial nearest neighbour
rectification. The panchromatic images were used first as the higher spatial resolution enabled
162
a more accurate location of the tie point. These same GCPs and tie points were used to
correct the 5 Ikonos MS images.
The re-geo-correction of the Ikonos imagery using GCPs from handheld GPS (with an
estimated positional error of c. 4-5 m) provides 5-8 m RMSE. This accuracy allows desk
based mapping and subsequent field navigation to be undertaken with improved confidence.
It cannot be overstated how much time and money this simple technique will save in
comparison to traditional Total Station survey (Newson 2002). This re-geo-corrected Ikonos
imagery became the basemap upon which all subsequent co-registration of Corona and
Landsat was undertaken.
Figure 85 The location of two GCPs, raw Ikonos and corrected
Ikonos.
5.4.1 Co-registration
Most modern satellite imagery is delivered to the user as geo-referenced data. All such data
has an error associated with its spatial location. This means that it is extremely rare that two
images collected by the same satellite at different times will have the same location in the
same space. Figure 86 shows this for two different panchromatic Ikonos images. There is an
163
offset of 7.3m East and 9.9m North between the images. For most applications this is an
insignificant difference. However, for image fusion techniques, such as pan sharpening (see
section 5.6.1) or time change analysis (see section 9.4), it is important that the pixels in the
layer stacks all represent the same object. This process is referred to as co-registration.
Figure 86 Geo-referencing errors. The Ikonos image on the right
is offset from the left image by 7.3m East and 9.9m North.
As the Corona imagery was scanned from photographic negatives they have no spatial
component. These images were co-registered to the re-geo-corrected Ikonos imagery using
concurrent GCPs. It should be noted that the Corona imagery used non-metric cameras.
This means that there can be a significant non-linear displacement of features in the negative.
Therefore higher order polynomial functions were required to warp the imagery in order to
ensure low error margins at the GCP points (see Table 16).
Image
Corona 1108 North
Corona 1108 mid
Corona 1108 South
Corona 1110 North
Corona 1110 mid
Corona 1110 South
Corona 1111 mid
Corona 1111 South
No. of GCPs
34
25
20
21
12
26
27
8
Correction Method
3rd order polynomial
3rd order polynomial
3rd order polynomial
3rd order polynomial
2nd order polynomial
3rd order polynomial
3rd order polynomial
2nd order polynomial
Table 16 Corona co-registration RMSE accuracy.
164
RMSE (1m Pixels)
1.90
2.63
1.79
1.93
2.15
2.63
2.66
13.40
The anomalous 13.40 RMSE for the Corona 1111 South image is due to the lack of
concurrent control between the Corona and Ikonos. It should also be noted that the area of
intersection between the two images is very small in relation to the full extent of the Corona
image.
Figure 87 Comparison of rescaling techniques. Although the 4,3,2
FCC look similar the histogram of the image using SD rescaling is
significantly altered from the original.
The Landsat imagery was co-registered to the re-referenced Ikonos imagery by selecting
concurrent tie points. However, the difference in scale between 1m and 30m spatial
resolutions made the choice of points very difficult. This situation was somewhat ameliorated
with Landsat 7 imagery. The panchromatic band has a ground resolution of 15m. This band
allows the collection of tie points with greater confidence. After the Landsat 7 images were
co-registered the panchromatic bands were used as basemaps for the Landsat 5 imagery.
5.5 Image rescaling
In order to evaluate whether the 11 bit radiometric depth was of significant value, the Ikonos
imagery was re-scaled to 8 bit image depth. This procedure was undertaken in Erdas Imagine
8.4 using the re-scale command and the parameters outlined in Figure 88. Each new Ikonos
file was provided with the suffix ‘_8bit’.
165
A minimum-maximum rescale was used as this simulates the 11 bit sensor recording in 8 bits
(the nature of the histogram is preserved). However, from an image compression perspective
a standard deviation rescale was also conducted (see Figure 87). The standard deviation
rescale uses more of the DN bins that the minimum-maximum rescale and therefore
maintains more of the original data. If processing time or disk space is an issue this technique
could be used in order to reduce file size with minimal loss of content. It is also interesting to
note that the standard deviation rescale makes some archaeological residues more obvious
(see site 508 in Figure 87).
Figure 88 The rescale parameters used in Erdas Imagine to
convert 11 bit Ikonos to 8 bit.
5.6 Image fusion
Image fusion is the process of analysing more than one image source. Image fusion
techniques are valuable because of their ability to integrate the various characteristics of
different imagery into a single image stack. Imagery is fused in order to exploit the spectral
and spatial characteristics of different sensors (pan-sharpening see section 5.6.1) or to
166
evaluate changes in image structure over time (time change analysis see section 9.4). As such,
fusion is not normally an end in itself but an interim process for different analytical goals
(Pohl and Van Genderen 1998). Image fusion can only occur if the images to be analysed are
co-registered.
5.6.1 Pan-sharpening
Remote sensing systems usually exhibit high spatial resolution and low spectral resolution or
low spatial resolution and high spectral resolution. High spatial resolution is necessary for the
definition of shape and structure, whereas improved identification (as opposed to detection)
comes from high spectral resolution (Ranchin et al. 2003). Panchromatic sharpening is a
generic term for the process of merging any high spatial resolution image with a low spatial
but higher spectral resolution image. The resultant imagery combines both the high spatial
and spectral resolution of the input products (see Figure 89). These images can improve
visual interpretation by integrating the texture of the higher resolution imagery with the
colour composition of the multispectral bands (Dare and Fraser 2001).
Successful pan-sharpening requires that the images are co-registered (see section 5.4.1) and
are collected on or around the same date. The specifics of the pan-sharpening algorithms are
beyond the scope of this research but are discussed in detail by Chavez et al. (1991), Wald et
al. (1997), Pohl and Van Genderen (1998) and Ranchin et al. (2003). Campbell (2002 pp. 313315) summarises the most common pan-sharpening techniques. Given the spatial, spectral
and temporal variations in the imagery only co-collected imagery was pan-sharpened: the 4m
Ikonos MS bands with the 1m Ikonos pan band and the 30m Landsat ETM+ MS bands with
the 15m Landsat ETM+ pan band. This will allow further evaluation of a 1m resolution MS
Ikonos image (potentially useful in the basalt) and a 15m resolution MS Landsat ETM+
image (potentially useful in the marl).
Most image fusion techniques employ information from different sensors (for example Sunar
and Musaoglu 1998). Increasingly, sensors are developed that have the capability to co-collect
co-registered multispectral and higher resolution panchromatic imagery. Both the Ikonos and
Landsat ETM+ imagery sets have these characteristics. These data sets have limited or no
error in their co-registration and because they are co-collected they have the same solar
illumination (Liu 2000a).
167
Figure 89 Ikonos pan and MS imagery with a PCT and SFIM
resolution merge derivative.
168
Liu (2000a) compared the Smoothing Filter-based Intensity Modulation (SFIM: Liu 2000b),
Intensity, Hue and Saturation (IHS) and Brovey image sharpening techniques on an image
stack of Landsat 7 ETM+ data. The panchromatic band was used as the high resolution
image. SFIM was determined as the most robust technique primarily due to the sensitivity of
the technique to co-registration accuracy. Furthermore, the quality of the Landsat merge for
all techniques was better with the 15m panchromatic than with a higher resolution SPOT
10m panchromatic image. Again this is due to the imperfect co-registration of the SPOT to
the Landsat. Liu (2000b) also states that SFIM retains the majority of the spectral component
of the imagery. However, this assertion is refuted by Wald and Ranchin (2002). Frank (2001)
compared automated classification techniques on raw TM imagery and SFIM and IHS
sharpened Landsat TM using 5m IRS panchromatic imagery. Considering the 6 fold
improvement in spatial resolution the classification accuracies of 96.87% SFIM and 93.45%
IHS are very good against the 98.51% of Landsat TM.
Given the relatively robust nature of the SFIM technique with co-located imagery it was the
method chosen for pan-sharpening in this research. A Principal Components
Transformation (PCT) was also used. The PCT method calculates principal components
from the multispectral image and then substitutes the high resolution image into the first
principal component (brightness). An inverse principal component transformation is then
applied resulting in an image with the spectral resolution of the multispectral image and the
spatial resolution of the panchromatic image. This method assumes that PC-1 corresponds to
image brightness (the same assumption as the Intensity (brightness), Hue and Saturation
method).
The Principal Component method is best used in applications that require the original scene
radiometry (colour balance) of the input multispectral image to be maintained. As this
method rescales the high spatial resolution data set to the same data range as Principal
Component 1, before the Inverse Principal Component calculation is applied, the band
histograms of the output file closely resemble those of the input multispectral image.
The PCT transformation is a built-in function within ERDAS imagine and was run directly
for each data set. The SFIM transformation was built by the author in the Model Maker
component of Erdas Imagine. Figure 89 compares the results of these transformations on
the Ikonos imagery.
169
5.7 Discussion
The image preparation stage is an essential component of any modelling exercise. It
necessitates a clear understanding of the nature of the problem in hand. Sensors are selected
on their ability to provide appropriate imagery to answer the problem. Local factors, such as
geology, agricultural regime and rainfall must also be assessed in order to collect imagery
from an appropriate time frame. The acquisition of imagery for this research has highlighted
that this can be a time consuming and difficult process. It is essential that enough time is
programmed into image acquisition as unforeseen difficulties inevitably arise.
The image pre-processing stage is where the different data sets are co-ordinated to facilitate
interpretation and analysis. An appropriate projection scheme is chosen and where necessary
imagery is co-registered and fused. In particular fused imagery (with different spatial, spectral
and temporal characteristics) can potentially produce a significant improvement in
interpretative techniques.
170
CHAPTER 6 SATELLITE IMAGERY FOR THEMATIC INFORMATION
EXTRACTION
6.1 Landscape themes
Landscape archaeology is a geographical approach whereby a region is investigated in an
integrated manner, studying sites and artefacts not in isolation, but as aspects of living
societies that once occupied the landscape. To do this it is necessary to collect and analyse
archaeological and environmental data over large areas. Satellite imagery can be utilised to
derive information about the contemporary landscape, and under certain circumstances it is
possible to make inferences regarding former changes in the environment.
(Clark et al. 1998 p. 1461)
Thematic variables such as soil type, hydrology, topography and elevation may be used to
either contextualise archaeological data during Exploratory Data Analysis, as data layers in
predictive modelling exercises or as a backdrop for CRM applications (see section 3.1). From
a landscape survey perspective these themes can be extremely useful, providing information
on such diverse themes as land management (will archaeological residues be masked by
vegetation?) and geomorphology (was this terrace formed before or after a certain date?). The
thematic approach analyses different landscape components in an integrated manner. These
landscape components commonly include the following themes:
•
Land use and cover (topography).
o Communication networks.
o Hydrology networks.
o Settlements (discussed in Chapter 7).
o Field Systems (discussed in Chapter 7).
o Soil/geology maps.
•
Elevation (Digital Terrain Modelling).
171
The resultant thematic maps can be analysed independently or in conjunction with other data
sets. Many different techniques are available for the production and analysis of these themes.
The components can be analysed with standard GIS techniques such as network analysis,
shape analysis and image overlay. Synthesis is achieved by composite mapping or multivariate
statistical techniques between layers or individual layer analysis. Overlay techniques are
frequently used to look for spatial associations and relationships between the different
themes (e.g. between settlement distribution and soil types).
For many areas of the world thematic data is already available from the national mapping
agency (such as the Ordnance Survey). Although this information is available in Syria, it is at
too coarse a scale, poor or difficult to access. Thematic information is only available at small
scales (greater than 1:500,000, such as the Russian geology maps (Ponikarov et al. 1967)).
Furthermore, cartography generalises the real world (Morehouse 1995; Müller et al. 1995; Lee
1996): the generalisation process is normally undertaken for economic or asset management
purposes and therefore can have limited archaeological content. Satellite and aerial imagery
can be used to improve information extraction by allowing user-defined generalisation and
interpretation.
The use of remote sensing in this context is dependent upon the type of ‘theme’ desired and
the scale of interpretation. For example, in the absence of geological maps multispectral
imagery can be used to identify different surficial soils and geologies. Landsat TM imagery is
regularly used for this type of identification (Ebert 1989). Geological themes can be extracted
from Landsat data which has a large ground footprint, a relatively high spatial resolution for
the application (a 30m cell size is much smaller than any geological unit) and an appropriate
spectral resolution. Ikonos imagery can supply information related to modern topography
and can be used akin to aerial photography to update digital mapping. Corona imagery can
supply information on broadly the same scale as Ikonos but it relates to a relict landscape and
hence information can be elucidated regarding landscape change. Furthermore, the Ikonos
and Corona data sets are available as stereo pairs, which allow the production of digital
terrain models using photogrammetric techniques.
6.2 Land cover mapping
The analysis of land use patterning is used to infer the relationship between people and their
environment. As a result the delineation of modern land use patterns is fundamental to the
172
work of geographers and economists amongst others. Landsacpe archaeologists are
interested in land use patterns for curatorial reasons and as a means to reconstruct past
systems of inhabitation. Reconstructing these ‘past’ variables can be difficult For example,
surface vegetation changes on a yearly basis and has a positive correlation to ambient
environmental factors. On the other hand soils change over a much longer time-frame and
modern soil characteristics may not be positively related to many past soil conditions. Finally,
the parent regolith changes slowly over time and for archaeological purposes can normally be
assumed as constant.
The delineation of land use information can involve a complex integration of data from a
variety of sources. The choice of these sources is in part determined by the scale of land use
and the analytical requirements of the end product. For example, the requirements of a
regional planner, where regional generalisation may be important, are very different from
those of a social geographer, for whom high resolution geodemographic data may be
important. Remote sensing sources, particularly aerial photography, are commonly integrated
with map bases for mapping purposes. The benefit of imagery in this process is the lack of
cartographic generalisation. High spatial resolution imagery can be used to identify parcels at
large scales. At correspondingly smaller scales, which normally increase the footprint of the
area to be studied, lower spatial and higher spectral resolutions become more important.
Whatever the scale of analysis collateral information from ground observation is always
required.
6.2.1 Land cover classification systems
Creating land use maps from imagery is essentially a process of segmenting the imagery into
contiguous parcels with different characteristics. If specific archaeological themes have not
been identified (see Chapter 3) it is advised that a standard classification system is employed
so that other land use identification programmes can be easily integrated (Campbell 2002 p.
559). The classification schema should also be tallied with the requirements of the analytical
problem. For example although ‘suburban area’ may seem like a natural choice for analytical
purposes it is inappropriate if the end user needs to identify sub categories such as
‘residential’, ‘commercial’ and ‘industrial’.
Many classification systems, including the USGS system (Anderson et al. 1976 cited in;
Campbell 2002), use a nested hierarchy of levels which allow the identification and
173
subsequent generalisation of data between different levels. Levels I and II in the USGS
system are defined in the framework with Level III (high resolution identification) defined by
the user (see Table 17). The USGS system has many benefits; not the least being that it was
designed with remotely sensed imagery in mind. This allows the allocation of different land
use categories based upon the scale of information available. For example Level I is designed
for broad scale imagery such as Landsat. Levels II and III are more detailed classifications
that can be attributed to finer resolution (both spatial and spectral) imagery. The CORINE
programme has employed a similar approach in creating an integrated European Union-wide
land cover data set using satellite imagery (Gerard 2002).
ClassID
55
65
8
54
56
57
58
59
60
61
62
64
53
USGSLevel_I
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
Agricultural land
USGSLevel_IIcode
21
21
21
21
21
21
21
21
21
21
21
21
21
9
Agricultural land
22
38
63
17
1
2
3
45
4
48
46
49
43
44
50
5
6
7
18
39
40
41
42
47
19
51
20
52
21
Agricultural land
22
Barren land
Forest land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Urban or built up land
Water
Water
Water
Water
Water
Water
Water
Water
Water
Water
Water
74
43
11
12
13
14
14
14
14
14
14
14
15
15
16
17
51
51
51
51
51
51
52
52
53
53
54
USGSLevel_II
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Croplands and pasture
Orchards, groves, vineyards, nurseries and
ornamental horticultural areas
Orchards, groves, vineyards, nurseries and
ornamental horticultural areas
Bare exposed rock
Mixed forest land
Residential
Commercial and services
Industrial
Transportation, communications and utilities
Transportation, communications and utilities
Transportation, communications and utilities
Transportation, communications and utilities
Transportation, communications and utilities
Transportation, communications and utilities
Transportation, communications and utilities
Industrial and commercial complexes
Industrial and commercial complexes
Mixed urban or built up land
Other urban or built up land
Streams and canals
Streams and canals
Streams and canals
Streams and canals
Streams and canals
Streams and canals
Lakes
Lakes
Reservoirs
Reservoirs
Bays and estuaries
SHRLevel_III
Cropland: well drained basalt
Cropland: wadi silts/Marl
Croplands and pasture
Cropland: alluvial
Cropland: poorly drained basalt
Cropland: thick southern marl
Cropland: lacustrine deposits
Cropland: alluvial fan
Cropland: wadi silts
Cropland: southern marl
Cropland: thin southern marl
Cropland: irrigated southern marl
Cropland: irrigated northern marl
SHR_Geol
Basalt: Well drained
Marl/Wadi silts
N/A
Alluvium
Basalt: Poorly drained
Marl: Thick southern
Lacustrine
Alluvial fan
Wadi silts
Marl: Southern
Marl: Thin southern
Marl: Irrigated southern
Marl: Irrigated northern
Orchards/Groves
N/A
Orchards, groves, vineyards,
nurseries and ornamental
N/A
horticultural areas
Bedrock slope
Mixed forest land
Residential
Commercial and services
Industrial
Road (tarmac)
Transportation, communications
Water pipe
Main Road
Airport
Railway (disused)
Railway
Military zone
Industrial and commercial
Mixed urban or built up land
Other urban or built up land
Streams and canals
Channel (wadi)
Channel (palaeo)
River
Channel (concrete canal)
Channel (anthropogenic)
Lakes
Seasonal lakes
Reservoirs
Birka
Bays and estuaries
Bedrock slope
N/A
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Urban
Water
Water
Water
Water
Water
Water
Water
Water
Water
Water
Water
Table 17 Combined USGS and SHR land use codes.
The USGS system maintains flexibility for specific projects and allows wider scale
generalisation and incorporation of the data within other programmes of collection and
analysis. However, the classification system must also correspond to the data used to compile
174
the land use map. It is obviously inappropriate to expect a high level of identification when
one is employing imagery collected at a coarse scale.
The classification system employed in this research will be based upon the USGS system (see
Table 17). Other SHR project specific levels have been built into the hierarchy to enable the
allocation of land use classes specific to the application area and the project goals. Over time
this classification schema can be extended to not only produce traditional classifications but
also archaeologically specific classifications (i.e. land-segmentation that might have had
currency in the past). However, each of these classes can be generalised through the USGS
system. The flexibility of this system allowed the production of a joint land use and land
cover classifications system to identify economic, landform and surface sedimentary zones.
6.2.2 Land cover mapping methodology
As discussed in section 2.2 imagery can be classified by either qualitative or quantitative
methods. For the purposes of this research the majority of thematic classification has
occurred through qualitative rather than quantitative classification as this is the easiest system
to implement. This is primarily due to the fact that thematic data, although useful for analysis
and synthesis, are not archaeological in nature. Furthermore, quantitative image classification
techniques are highly skilled: it is difficult for an interpreter who is not trained in the specifics
of multispectral analysis of different land cover types to incorporate these skills. Some
specific image classification did occur. Where quantitative techniques have been used to
elucidate land use information they are explained in the text. Great use, however, was made
of image enhancement techniques during the visual interpretation component (particularly
histogram stretches, false colour composites and band ratios).
The best practice methodological guidelines outlined by Campbell (2002 pp 559-576) were
followed (as summarised in Figure 90). All the metadata information concerning the image
quality (sensor type, cloud cover index and collection date) were recorded within the
metadata for each data set. The classifications schema is defined in Table 17 where the field
‘SHRlevel_III’ delineates the bespoke classification values for the application area. The 2000
Landsat imagery and 2002 Ikonos imagery were the primary image sources for the
classification.
175
Figure 90 Land use digitising schema (after Campbell 2002 p. 559).
The basic approach to digitising the land cover themes was to digitise linear and land parcel
elements directly from the satellite imagery within ArcGIS. Ikonos imagery would provide
the primary resource for landcover mapping in the northern and southern application areas
and Landsat imagery would provide the primary resource for the soil, geology and urban
mapping. The Corona imagery, Syrian map and Russian geology map were also used as
ancillary data sets. Ground observation during previous fieldwork seasons had provided a
general understanding of most of the region. This was augmented by the geomorphological
work undertaken by Drs Bridgland and Westaway (Bridgland et al. 2003).
176
Finally, decisions need to be made about the form of the thematic output. Manual
interpretation techniques traditionally delineate parcels or linear features using vector
techniques. On the other hand, image classification techniques produce a raster output. As
both techniques are used in the production of thematic information it was decided to
produce all themes, with the exception of the DTM, as vector layers. Vector delineation of
this information also removes misclassification artefacts commonly associated with pixel
mixing and natural heterogeneity in class composition (Tso and Mather 2001; Campbell
2002; Harrower et al. 2002).
6.2.3 Vector digitising methodology
Whereas Campbell (2002 p. 560) advocates working on translucent overlays the digitising
process was conducted directly using ArcGIS. The imagery, or its derivatives, provided the
digitising backdrop. Land use and cover was separated into polygon features and polyline
networks. Polyline networks were employed so that direction of flow and other attributes
could be appended for the appropriate feature types. Polyline networks include
transportation systems and hydrology. Polygon layers were created for land cover and
surficial geology/soil types. Each polyline network and land cover polygon was attributed an
identifier from the classification schema based upon the most refined ‘SHRlevel_III’ field
and linked through the primary key field ‘CodeID’ (see Table 17). This allowed the
generalisation of the themes to the USGS levels I and II. To reflect this, three new vector
layers were created within the Geodatabase (see Appendix I) within the theme
‘satellite_themes’: a polyline network layer called ‘Line_landcover’ and two polygon coverages
called ‘Polygon_landcover’ and ‘Soils_geology_urban’.
6.2.3.1 Polyline networks
Linear elements digitised for land cover are best represented as polylines. These are vector
lines that represent linear units such as road and river networks (based on centre lines). GIS
systems can enhance the utility of connected polylines by analysing them as polyline
networks. These networks can be embedded with information determining how the line
segment can be used. For example, in a road network a line segment could be a one way
street with a specific speed limit. These attributes can be embedded into the polyline. When a
network is analysed for the shortest or quickest route from place A to place B these attributes
can form part of the analysis. Figure 91 defines the creation of a polyline network for
digitising the hydrology in the application area. In this model water is only allowed to flow
177
downhill (i.e. from node X to node W). When digitising in ArcGIS polyline networks are
created dynamically, however, attribute data (ClassID and flow direction) needs to be added
for each individual polyline. The attribute ‘ClassID’ is the linking field between the drawing
and the USGS code database described in Table 17.
Figure 91 Digitising topologically intact polygons and networks for
GIS analysis.
6.2.3.2 Polygon layers
Enclosed areas digitised for land cover are best represented as polygons. These are closed
vector lines that represent land parcels such as a geological unit or an urban area. It is
178
possible to just describe a polygon by digitising its outline (for example the technique which
is applied for digitising site extents). However, in land cover applications many of the
polygons are contiguous (i.e. they share adjacent borders) or are contained within other
polygons. Digitising topology (the spatial relationships which make up a polygon) can
substantially reduce the work involved in the digitising process and ensure that polygons
actually share common boundaries. Figure 91 defines the creation of a polygon topology
using archaeological excavation data. Burroughs (1986) discusses extensively the formal
process of polygon topology creation.
When digitising in ArcGIS polygon topology is created dynamically, but attribute data
(ClassID) needs to be added for each individual polygon. The attribute ‘ClassID’ is the
linking field between the drawing and the USGS code database described in Table 17.
6.3 Hydrology networks
The hydrology network consists of rivers, seasonal wadis, anthropogenic canals and channels
(see Figure 92 and Figure 93). These were digitised directly from the pan-sharpened Ikonos
imagery. However, use was also made of the panchromatic Ikonos imagery overlaid by the
multispectral imagery with a 60% transparency setting. This provided a combination of high
resolution imagery and colour contextual information without the need for pan-sharpening.
Any co-registered resource can be used in this fashion (for example panchromatic aerial
imagery and SPOT multispectral imagery). A few specific enhancements were used to aid in
the identification of water bodies. Particularly important were differences between the visual
bands and near-infrared. The value of the near-infrared lies in the fact that the contrast
between water, vegetation and other surface phenomena, which are not obvious in the visible
spectrum, are more enhanced. All wet areas tend to absorb infrared radiation, and this leads
to a lower reflectance value (Ebert and Lyons 1983 p. 1256; Campbell 2002 p. 525).
NDI =
Infra Re d − Blue
Infra Re d + Blue
Equation 1 NDI equation
179
Figure 92 Hydrology network image interpretation key.
180
To further illuminate these differences between the visible and NIR a ratio of the pansharpened Ikonos bands 1 and 4 was also calculated. This was augmented with a 4, 4, 1 false
colour composite. Finally a Normalised Difference Index (NDI) between the NIR band and
the blue band (see Equation 1) was also conducted. In this technique water bodies are dark
areas.
The Corona imagery was also used to provide extra contextual information, particularly for
the location of wadi channels. These are seasonal channels that for the majority of the year
exist as soil marks with a higher reflectance than the surrounding soil matrix (presumably
from wadi silts and gravels). Modern agricultural practices (including deep ploughing,
bulldozing, and increased vegetation cover due to irrigation) have masked or blurred many of
these features in more recent imagery. These agricultural practices had limited impact on the
older Corona imagery. The same situation applies for the features identified as ‘channels
(anthropogenic)’ (see Figure 92). These features have the same reflectance characteristics as
the wadi channels however they differ in one significant aspect: instead of flowing directly
through the contours on a path of least resistance they hug contours observed on the
digitised Syrian 1:25,000 maps. This led to the interpretation that these channels were
anthropogenic in origin. One of these wadi channels appears to run into the ditch
surrounding Tell as-Safinat Nebi Noah (site 14), giving the impression that this fortified site
may have been moated. For the same reasons Corona imagery has been exploited by Stone
(2003) for her research into the course of the ancient Tigris.
In the northern marl zone extensive concrete irrigation channels fed from Lake Qatina were
constructed under the French mandate during the 1930s. Since the late 1980s the combined
effects of capping the spring at ‘Ain at-Tannur (to divert the water to the expanding city of
Homs) and a reduction in rainfall has led to shrinkage of Lake Qatina (see Figure 64 and
Figure 169). This has effectively meant that during periods of low rainfall (such as the winter
of 2001/2002) these irrigation channels are redundant. Consequently these hydrological
features can not always be identified on the basis of the reflection characteristics of water.
However, these linear features have a distinct morphological signature. They run in straight
lines with distinct periodic curves and roads have regularly been built on their margins. They
can be distinguished from normal roads due to their curvature and where necessary can be
interpreted by proxy.
181
Figure 93 The digitised hydrological and communication networks.
182
Figure 94 Communication network image interpretation key.
183
The digitised components were compared with the 1:25,000 Syrian maps and driving surveys
and showed a high degree of correlation. For many of the features there was, in fact, an
improved level of detection.
6.4 Communication networks
A similar methodology was used for digitising the communication networks. However, in
this instance only the panchromatic and multispectral Ikonos imagery were employed. Main
roads, asphalt roads, water pipes, railways (both used and disused) and an airport runway
were all digitised (see Figure 93 and Figure 94). Due to time constraints tracks were not
digitised, however, they were easily identifiable on the imagery. There was very little
interpretative difference between the pan and MS Ikonos for this theme. No specific
enhancements were required to detect these features. The result were correlated against the
Syrian 1:25,000 maps and compared during driving surveys and showed a much improved
product in comparison to the mapping.
6.5 Land cover parcels
The land cover parcels contained a greater range of variation than either the communication
or hydrology networks (see Figure 95 and Figure 96). This was predicated on the basis that
discrimination of the different land cover elements should be as refined as possible. All the
land cover parcels for this theme were identified using the Ikonos imagery so that this theme
could provide a benchmark for the state of the landscape in 2002. It is a future intention to
digitise a set of Corona imagery in the same way. Comparison of these data sets would be
particularly useful for time change analysis based solely on interpretative data (see section
9.4). The benefit of this technique is that instead of quantifying the change between pixel DN
values, one can evaluate the changes in specific land use categories (i.e. from cropland to
orchard) and hence extrapolate changes in agricultural, economic or political factors.
Water features (lakes, seasonal lakes, reservoirs and birka) were identified using the same
techniques and enhancements as described in section 6.2.3.2. Where seasonal lakes were
identified the surrounding corona which delineated the full extent of the lake when full was
also included in the parcel. Reservoirs were defined as any dam like structure. The only
exception to this rule is lake Qatina, which although dammed is referred to as a lake. Birka
are small reservoirs associated with settlements in the basalt. Sometimes these features are
delineated by surrounding basalt walls.
184
Olive groves and orchards were identified by textural variation in the imagery. Areas with
mature plants were easier to identify on the Ikonos multispectral imagery (particularly when
using the near infrared (band 4)) whereas areas with younger plants were easier to identify on
the Ikonos panchromatic imagery. Hence, the most effective way to locate groves and
orchards was to overlay a 4, 3, 2 false colour composite of the Ikonos multispectral with a
60% transparency setting over the Ikonos panchromatic. Forested areas were detected in the
same way. Differentiation between grove and forest was based upon differences in location
(forests occur in the basalt zone only), texture and shadow length (see Figure 96).
The high reflectance observed at urban and other concreted areas was very easy to identify.
The separation of each of these into the different classes of military, residential, industrial,
commercial and industrial complexes was based upon morphological differences and by
proxy along major communication networks. The default designation was residential; these
areas are identified by their distinctive tone and texture. Military zones were extremely easy to
identify due to their distinctive morphology. Virtually everything else would be classed as
industrial. Most small scale industry is confined to long buildings (chicken farms). Industrial
complexes were attributed on the basis of scale and include the fertiliser factory and oil
refinery. Where there was a combination of urban usage the parcel was allocated to the
‘mixed urban or built up land’ zone. No specific enhancement techniques were required to
identify these features.
Cropland accounted for the vast majority of the land surface. The bespoke level III schema
for cropland took into account the formation processes and parent regolith unique to each
environmental zone. Hence, the marl was split into Northern marl, thick marl, irrigated marl,
thin marl and wadi silts/marl (see Figure 95) to aid the delineation of different marl types.
The basalt was split into poorly drained and well drained units (the well drained deposits are
probably areas of stable floodplain). Alluvial cropland is related to areas of overbank flooding
or modern river terraces. In the area SW of lake Qatina the alluvial area blends into an area
where lacustrine deposits have been revealed (and exploited as agricultural land) by the
shrinking of the lake. Wadi silts are areas of high reflectance associated with migrating and
alluviating seasonal wadi channels. At the south end of lake Qatina west of the current course
of the Orontes there is an alluvial fan extending from the Anti-Lebanon range. Finally there
is an area of exposed bedrock slope in the south of the area.
185
Figure 95 The digitised land cover parcels from the Ikonos 2002
imagery.
186
Figure 96 Land cover parcels image interpretation key.
187
6.6 Geology/soil mapping
Remote sensing imagery can be used to classify geology and soil attributes (Siegal and
Gillespie 1980; Williams 1983; Irons et al. 1989; Drury 1993; Mattikalli 1997; Way and Everett
1997). The scale of information captured means that high spatial resolution data is not
necessarily required. The use of high spatial resolution imagery can increase heterogeneity
and hence makes the classification/identification process more difficult by producing a
classification schema with many different variables rather than landscape groups. Rather,
improved spectral resolution imagery can be more important for defining geological
attributes as different units are easier to discriminate when spectral resolution is increased.
The Landsat imagery was used for the majority of the soil/geology mapping. One of the
major benefits of Landsat in this instance is the size of footprint. This imagery extended well
beyond the application area. This substantially increased the overall context of the imagery,
improving interpretation by having access to broader scale environmental systems. Arid and
semi-arid environments, such as the landscape around Homs, offer better potential for the
characterisation and identification of surficial soil and rocks due to the limited vegetation
cover. It is advisable to select an image that has minimal vegetation cover to conduct this
process or to conduct comparable analyses on multiple scenes and generalise the results
accordingly.
Computerised classification and image enhancement techniques were used to help determine
soil and geology units. Classification involves the categorisation of the multispectral image
using statistical procedures. As discussed in section 2.2.7 supervised classification procedures
involve user intervention to define training areas which are then statistically extrapolated to
the rest of the imagery. Alternatively unsupervised classifications define their own clusters by
employing a statistical algorithm on the data. These techniques produce spectral signatures
that respond to each cluster or training area. These spectral signatures vary due to a number
of factors including sensor attitude, wavelength, ground cover, vegetation cover, moisture
content, atmospheric conditions, slope and aspect (Campbell 2002). Multi-band visualisation,
density slicing and band ratioing (particularly bands 7 and 5) can help distinguish specific
geological units. False colour composites using band combinations 7, 5, 1 and 3, 5, 1 were
used to help discriminate different geology types. In areas of high reflectance, such as the
wadi silts, it is appropriate to enhance the histogram (by using a minimum-maximum stretch
or density slicing) to improve contrast and hence interpretation. Tasselled cap
188
transformations were also employed to assess whether feature space transformations
improved residue detection.
Figure 97 The unsupervised
geology/soil classification
classification
algorithm
for
Using the methodology outlined by Harrower et al. (2002) an unsupervised classification was
conducted on all the co-registered and atmospherically corrected Landsat imagery (see
Chapter 5 for details on this pre-processing). 70 classifications were produced, for each
Landsat scene, using an Isodata unsupervised algorithm (see Figure 97). This algorithm
produced an output image with colour determined by the red, green and blue bands and a
spectral signature file for each cluster.
189
One of these classified images was chosen as the primary scene. It was initially assumed that
the October 1987 imagery would be the most appropriate imagery to conduct lithological
analysis based upon the assumption that in the summer of 1987 less surface vegetation would
be visible as the irrigation techniques were not as extensive in 2000. While classifying the
imagery and viewing the tasselled cap transformation it became apparent that the northern
marls were heavily vegetated. This is due to the high rainfall in 1987 and successful irrigation
through the network of concrete canals (discussed in section 6.2.3.2). Therefore the October
2000 scene was employed as the primary scene as it contained the least amount of vegetation
as determined from the greenness index of the tasselled cap images.
Using the ancillary data, including the land cover map techniques, each of the 70
classifications were attributed to a different landuse type in the USGS schema (see Table 17).
Wherever possible geology and soil zones were subdivided based on the classification results
and ground observations (i.e. the number of different marl zones in Figure 98). The digitising
of these data sets into vector files meant that rogue, misclassified, pixels (see section 6.2.2)
did not occur.
The Ikonos-based land cover mapping (see section 6.5) was used in the geology map. The
SHR_LevelIII agricultural classification contained the geological zone of the cropland. An
extra field was added into the USGS schema to generalise the classification to the parent
materials. This allowed the original schema to be expanded so that it incorporated both land
cover components and surficial deposits. This simple application reduces data complexity as
one single data set can be used to answer a range of different questions from land cover to
surficial geology. Furthermore, the higher spatial resolution of the Ikonos imagery allowed
more accurate spatial referencing of features.
In summary, the following data and visualisation methods were used for the creation of the
geology/soil map (see Figure 98):
•
False colour composite of Landsat imagery.
•
Unsupervised classifications.
•
The land cover mapping.
190
Figure 98 The soil/geology/urban classification
191
Figure 99 The effects of increased rainfall. Note the left hand
images are all summer images where there has been normal rainfall
(hence the irrigation channels are empty).
6.7 Digital terrain modelling
Digital Terrain Models (DTMs) are an essential component of many archaeological
modelling exercises. Elevation data can aid in solving a wide range of spatial problems
(Toutin 2001). Terrain models are important within archaeology for studying individual
monuments as well as their larger scale topographic context (Redfern et al. 1999).
192
Furthermore, terrain models offer a range of thematic information from a primary elevation
model to the secondary slope and aspect models. For a discussion of DTMs and their raster
and vector derivatives (digital surface models, digital elevation models, triangular irregular
networks, slope derivatives and aspect derivatives) consult Burroughs (1986), Fujimura and
Kuo (1999), Hageman and Bennett (2000) and Wheatley and Gillings (2002).
Many archaeological predictive modelling and visualisation exercises place significant
emphasis on DTMs and their derivative layers (Gaffney and Stancic 1991; Hageman and
Bennett 2000; Wescott and Brandon 2000). Wheatley (1995), in particular, has argued for the
use of cumulative viewshed analysis to elucidate aspects of site and network locations in
respect to their visibility from prominent points in the landscape. Furthermore, DTMs are
used within remote sensing to model and correct for reflectance from different illumination
angles (Campbell 2002).
Most DTMs used in archaeology are created by digitising the contours of an available base
map. However, there has been some criticism of DTMs derived in this manner (Kvamme
1995; Redfern et al. 1999). Hageman and Bennet (2000) produced a best-practice guideline
for contour digitising. One of the major criticisms of DTMs created from digitised contours
is that they require interpolation of non-primary data. Contours themselves are produced by
interpolating primary elevation data (collected either from photogrammetry or land survey),
hence contours are secondary data. The process of digitisation introduces further errors into
the data set.
6.7.1 DTM from contour data
Although the inaccuracy of creating Digital Terrain Models (DTM) from contour data is
recognised (Redfern et al. 1999 p. 212), if the mapping data is available at an appropriate scale
it is the easiest source for creating a DTM.
The digital contours created from the 1:25,000 basemap were used to create a Triangular
Irregular Network (TIN). Best practice guidelines for the contour digitising were followed to
improve the DTM accuracy (Burroughs 1986; Hageman and Bennett 2000). Contours were
derived from the TIN and compared to the original basemap contours to evaluate the
accuracy of the TIN. The TIN was converted into a raster Digital Elevation Model (DEM)
with a 20m cell resolution. Slope and aspect models were derived from the DEM (see Figure
193
100). It is interesting to note that artefacts from the original contour data itself are obvious in
the slope model.
These models were used throughout the research project for a variety of archaeological and
visualisation procedures already discussed. In this context they also provide an important
information resource from which to evaluate the DTM generation by photogrammetric
techniques from Corona and Ikonos imagery.
Figure 100 DEM, aspect and slope derived from contour data.
6.7.2 DTM generation from remotely sensed data
Remote sensing provides two primary solutions to the creation of DTMs. One solution
encompasses the variable wavelength detection and ranging techniques such as RADAR and
LiDAR (respectively, RAdio and LIght Detection And Ranging). These active sensors record
the duration of travel. As the wavelength is known distance can therefore be calculated
(Holden et al. 2002).
The other solution, which is more familiar to archaeologists, is that of Photogrammetry:
More specifically the exploitation of the differences in stereoscopic parallax between two
images taken from slightly different perspectives. Stereo photography was exploited by
archaeologists as early as 1975 (Astorqui 1999). Photogrammetry is defined as:
194
The art or science of obtaining reliable measurements by means of photography.
(Colwell 1997 p. 3)
If two images are taken of an object from slightly different perspectives (a stereo pair) then a
displacement in the object can be observed in the images. This phenomenon is referred to as
stereoscopic parallax. If these images are placed side by side and viewed through a stereoscope a
3-d effect is observed. These phenomena can be exploited for measuring distances or
heights. This technique is called photogrammetry, or more specifically for those who work
on digital imagery, soft-copy photogrammetry. In order to extract a DTM from a stereo pair
a sophisticated work-station is required that runs specific photogrammetric software (such as
SocetSET or Erdas Imagine Stereo Analyst) and a camera model is required. The camera
model refers to the interior geometry of the camera and its orientation (Teng 1997 pp. 82102; Campbell 2002 pp. 77-84).
Any overlapping satellite imagery taken from different perspectives can be used to create a
stereo pair (for example Toutin 2001; Li et al. 2002; Toutin 2002; Zomer et al. 2002).
Furthermore, these images can come from different platforms and sensor devices. However,
better results are obtained with imagery from the same sensor taken with a limited time
differential. High resolution satellite sensors such as Ikonos and Quickbird recognise the fact
that users would want to extract DTMs using stereo images and provide Rational Polynomial
Coefficients (RPCs) with their metadata to facilitate DTM extraction. RPCs contain all the
necessary metadata about the camera model and orientation required for the interpolation
algorithms (ERDAS 2001).
Galiatsatos (in prep) extensively discusses the creation of DTMs by soft-copy
photogrammetry using SocetSET and Erdas Imagine Stereo Analyst for the application area.
Galiatsatos employs Corona-Corona (mission 1110 stereo pair), Ikonos-Ikonos and CoronaIkonos stereo pairs. It is important to note that the Ikonos-Ikonos imagery was not
purchased as a stereo pair and hence the RPC files were not supplied. Therefore, Galiatsatos
extracted the DTM using traditional photogrammetric techniques. Furthermore, the camera
model and ephemeris data for the Corona imagery are difficult to obtain (it requires a visit to
NARA). This necessitated the use of an empirical non-metric camera model to determine the
195
terrain model. Altmaier and Kany (2002) also had difficulties when attempting to create a
DEM from Corona stereo pairs.
Galiatsatos (in prep) proposes that the Corona – Ikonos stereo model can be used for time
change analysis. He postulates that if one were to analyse the error surface associated with
the DTM then locations with large errors will be due to changes (such as house
construction).
Figure 101 Graph highlighting the relationship between
photogrammetric derived DTM heights from Corona and DGPS
check heights (after Galiatsatos in prep).
During the 2002 season a DGPS was permitted for the collection of geomorphological
features (mainly terraces) in the application area. Unfortunately it was not permitted to record
areas of ‘hard’ detail which could have been used as GCPs for any of the rectification
procedures. However, these points did provide an effective control from which to
independently assess the accuracy of the DTMs. The mean height difference was 0.46m with
196
a maximum difference of 1.35m and a minimum of 0.01m. Figure 101 shows the correlation
between these check heights and the photogrammetrically derived DTM. The correlation
coefficient (r) is 0.999.
6.7.3 DTM evaluation
Photogrammetric DTM creation when using a metric measuring device with an appropriate
camera model produces accurate results while at the same time being highly efficient in terms
of labour (Redfern et al. 1999). This contrasts sharply with traditional archaeological DTM
generation techniques such as contour digitising (which is an inefficient use of labour and
relatively inaccurate) and ground survey using Total Stations or DGPS (which is a highly
inefficient use of labour and very accurate). However, LiDAR technology does provide
highly accurate and labour effective terrain models relatively cheaply, although its application
is currently limited due to sensor costs. Photogrammetric software is rarely used by
archaeologists, predominantly due to the expense of the software, the need for accurate
camera information, a lack of skilled users and lack of awareness (Redfern et al. 1999).
Gillings and Goodrick (1998) have used the low cost PhotoModeller software for 3-d
reconstruction of standing buildings in archaeology but it has yet to be ascertained how this
software would work with aerial and satellite imagery.
High resolution elevation data is an important interpretative data source within archaeological
analyses. Many archaeological features produce a topographic effect, this effect is exploited in
oblique aerial imagery, or occur within specific areas identifiable through elevation analyses,
such as river terraces (Holden et al. 2002).
6.8 Discussion
This chapter has discussed the creation of thematic layers from satellite imagery. These layers
are relevant for a number of archaeological management and analysis needs and can help to
frame further research. A combination of medium and high spatial resolution imagery has
been integrated to produce a number of thematic layers which were not previously available
at this scale. The ability to produce bespoke thematic generalisations from satellite imagery is
one of its major benefits. This is even more significant in an area which has limited
alternative mapping and little published data on soils, hydrology, etc.
197
In general terms the spatial resolution of the imagery dictates the type of classification
technique. The medium-resolution Landsat imagery was successfully employed for
quantitative classification. However, classification becomes increasingly more complex when
spatial resolution is increased. Even classification techniques used by experienced interpreters
do not produce the same kind of results commonly seen in the classification of medium
spatial resolution data (Palumbo and Powlesland 1996 p. 126). This is due to increased
heterogeneity in the imagery and the inclusion of artefacts not necessarily relevant to the scale
of analysis (e.g. groves in the Ikonos when one is classifying for surficial geology). However,
the differences in spatial resolution are complementary in that high spatial resolution imagery
can be used to ‘truth’ classified results from lower resolution sensors.
High spatial resolution imagery, from any platform, can be used in conjunction with any
contemporaneous coarser spatial resolution imagery (such as Landsat) to aid any
interpretation or classification of the imagery. For example, it is possible to identify crop
types or genus directly from the high resolution imagery and use this to directly interpret the
Landsat imagery.
With the techniques employed and the lack of alternative resources at appropriate scales it is
difficult to evaluate the accuracy of the classifications. This is particularly important for the
more qualitative themes (such as geology and soil, see section 8.2) which tend to change
gradually over the landscape and do not occur in discrete parcels. However, this classification
is not seen as the finished product. It is envisaged that further fieldwork, ground observation
and consultation with the project geomorphologists will continue to refine its accuracy.
As a mechanism to improve engagement with the landscape it is recommended that a land
cover and soil classification is undertaken at a very early stage in the interpretation process.
This engages the investigator with the landscape and provides important information on its
structure and morphology. The outlined techniques provide a broader and more
representative understanding of the landscape than, for example, a driving survey.
198
CHAPTER 7 SATELLITE IMAGERY AS A PROSPECTION TOOL
7.1 Archaeological prospection
This chapter addresses the issues of extrapolating elements of the relevant archaeological data
structure from satellite imagery. The complex theoretical issues in the delineation and
interpretation of archaeological phenomena (as discussed in Chapter 3) are somewhat
simplified when one is employing satellite imagery for archaeological prospection: the level of
interpretation will normally only occur at the detection or recognition levels rather than the
much more complex identification level. Hence, there are two primary issues when using
satellite imagery for prospection:
•
Can aspects of the relevant data structure actually be identified?
•
If so, how much of the data structure can be identified?
The answer to both of these questions requires ground observation. The first question can be
addressed by visiting the potential areas identified from the imagery as part of a more
rigorous recording exercise (recognition or identification of the residues). The second
question can be addressed by visiting ‘blank’ areas of the landscape in search of
archaeological residues. The second question is much more difficult as the extent of the total
relevant data structure is unknown and locating other ‘undetected’ archaeological residues is a
difficult task. Effective quantification requires an appropriate sampling methodology (see
section 0).
In a macroscopic landscape context, the archaeological record can be thought of as a more or
less continuous spatial distribution of artefacts, structures, organic remains, chemical residues
and other less obvious modifications. Microscopically, the distribution is far from even, with
large areas where archaeological remains are widely and infrequently dispersed. There are
other areas, however, where materials and other remains are abundant and clustered. It is
these peaks of abundance that are commonly referred to as sites. In areas where there is
limited understanding of the archaeological resource, particularly with reference to the
distribution of ‘sites’, then some form of archaeological residue discovery is required. As
discussed in section 3.5.2 the term site can be a misleading. In this context site refers to any
archaeological residue (including field systems, hinterlands and settlements).
199
Discovery requires the detection of one or more site constituents. These are sufficient to
suggest that the site might be present (McManamon 1984). Many archaeological projects
detect archaeological residues through an intensive process of ground survey (Shennan 1985;
Shennan 1997; Gillings et al. 1999; Francovich et al. 2000; Banning 2002). However, such
approaches are costly in terms of time and human resource allocation. It is proposed that
satellite remote sensing techniques can be incorporated into archaeological landscape
evaluations, at an early stage, to help determine the strategic deployment of the ground
survey teams.
Analysis of remotely sensed data involves identifying features and correlating ground-based
measurements with recorded reflectance or emmittance values. The co-registered bands are
multiple layers of numeric information that have spatial and spectral structure. Some of this
structure relates to archaeological phenomena. Analysis of this data is a creative synthetic
process that transforms data into information. The creative act of ‘interpretation’ itself
requires that the interpreter has an understanding of the data and its structure. This means
that the interpreter is aware of underlying processes during the act of archaeological
‘discovery’ (Aldenderfer 1987 p. 92). It is hoped that this chapter will apply these processes
and outline an appropriate methodology for this and other environmentally similar
application areas.
The important points for archaeological residue detection (after McManamon 1984) are that:
1. Archaeological sites are physical and chemical phenomena.
2. There are different kinds of site constituents.
3. The abundance and spatial distribution of different constituents vary both
between sites and within individual sites.
4. These attributes may be masked or accentuated by a variety of other phenomena.
Archaeological prospection applies a series of principals aimed at the detection, localisation
and identification of previously unknown archaeological sites or features. For purposes of
definition this section is focussed on site discovery using satellite imagery. Thus, the primary
focus of archaeological prospection is to detect locations that have a high potential of
containing archaeological residues. The recognition or even identification (see section 1.3.5) of
archaeological residues is desirable at this stage but not essential. It is envisaged that residue
200
detection using satellite imagery is quicker, more cost effective and may provide a more
representative sample than surface survey. However, for successful archaeological
‘interpretation’ to occur, increased resolution, or more likely ground observation data may
also be required to provide adequate interpretative definition. Site discovery, on the other
hand, focuses on a study area with the goal of locating as high a proportion as possible of
sites and potential factors which may account for bias in their discovery. In this context,
satellite imagery can be used as part of a Desk Based Assessment to provide a rapid overview
of the archaeological potential within different environmental areas. Thus, satellite imagery
provides a platform from which to perform landscape identification at a scale hitherto
unprecedented.
However, this physical approach to site definition is not well documented. Most
archaeological site analysis aims mainly to interpret sites and structure them culturally or
behaviourally, rather than physically or chemically. There is, therefore, a limited consolidated
body of reference data. Yet this sort of information would be extremely useful for
determining the effectiveness of different identification techniques (however, notable
exception include: McManamon 1984; Scollar 1990; Bintliff et al. 1992; Spoerry 1992; Pollard
and Heron 1996; Taylor 2000; Heron 2001). These author employ different techniques
including, satellite and aerial imagery, soil geochemistry, artefact variations and geophysics.
Sites can be identified by the variation in frequency and distributions of various physical,
chemical and biological constituents. Chapter 8 addresses some of these issues analysing the
physical and chemical nature of samples collected in the marl zone of the application area.
7.2 Impact of the environmental zones on prospection
Ground observation has shown that the three different environmental zones produce
contrasting and complementary types of archaeological residue. In each zone the
archaeological record has also been subjected to different formation and de-formation
events. The basaltic zone contains a well preserved combination of sites and hinterlands as a
palimpsest of stone walls and concentrations of rubble. By contrast, the marl zone, where a
mud-brick architectural tradition dominated, contains a few mounded tell sites and many
ploughed out artefact scatters. The marl zone only contains site loci as the hinterland has
been masked, or eradicated, by subsequent anthropogenic activity. The alluvial zone contains
a proportion of each type, depending upon the location of the alluvium, with a
preponderance of artefact scatters. Each of these zones employs distinctive urban and rural
201
management strategies that impact upon the visibility of the archaeological residues. Some
strategies can effectively destroy, or severely mask, archaeological residues (such as land
clearance through bulldozing or urban expansion). This chapter will focus on each of the
environmental zones separately as each zone requires different methodological techniques.
Figure 102 Changes in crop cover at site 339 at different times of
the year.
The visibility of most of the archaeological residues is primarily influenced by the type of
surface cover. Surface cover is positively related to the environmental conditions and the
agricultural regime (see Figure 102). For this research ground observation at different times
of the year indicated that crop cover is the most debilitating factor in the detection of
archaeological residues from satellite imagery (see section 5.2.4):
•
Unless crop marks are produced the crop canopy effectively masks any
underlying archaeological residues (see Figure 102).
•
It has yet to be determined if negative archaeological features produce the same
kind of crop mark evidence in the region as observed in Europe (i.e. the
construction techniques, materials and deformation are different).
•
Even if crop marks do occur, demonstrating that the spatial or temporal
resolution of the satellite images are adequate to detect crop vigour or stress is
inappropriate with the imagery available.
7.3 Methodological background
When an image is available in digital form, the interpreter can use several approaches to
extract information. Campbell (2002) distinguished two complementary approaches (see
202
Figure 4): photointerpretation (or qualitative image interpretation), where an analyst extracts
information according to their experience and understanding of the phenomena under study;
and quantitative analysis, where classification schemes are used to examine the image and
assign classifications according to their pixel attributes. Each approach has its own strengths
and weaknesses and the final combination depends upon the scope of the application.
Identifying the physical or chemical contrast between archaeological residues and the natural
landscape can be undertaken using either quantitative or qualitative techniques. Qualitative
prospection is the norm for archaeological image interpreters (see Donoghue 2001 for a
general discussion of archaeological remote sensing interpretation techniques). Qualitative
interpretation requires an experienced interpreter and imagery of the appropriate resolution
for the interpretative task. This process can be improved by having access to contextual
information to enable rapid verification. Quantitative interpretations in their purest form rely
on digital manipulation of the imagery to extract the relevant data from feature space.
Quantitative approaches are rarely used for archaeological prospection using satellite or aerial
imagery. This is in all likelihood a legacy trait due to two important issues: the fact that most
remote sensing techniques are ‘site’ rather than landscape focused (such as geophysical
surveys or site focused aerial photography) and the perception that spatial variations in the
physical and chemical residues at different sites would make quantitative techniques difficult,
if not impossible to apply. Furthermore, the human visual system and brain is much more
adept at interpreting and classifying complex information than a computer system (Ebert
1984 p. 348). This is particularly pertinent for archaeological residues which tend to exhibit
complex tonal, textural and spatial patterning which are difficult to classify by automated
techniques.
7.3.1 Qualitative methodologies
Archaeological residues can be distinguished on the basis of a number of image qualities
including tone, texture, colour, pattern, shape and size (Ebert 1984 pp. 313-315). Most
interpreters apply these techniques manually, relying solely on visual inspection of images
collected, predominantly from the visual wavelengths. False colour composites produced
from bands outside the visual wavelengths create representations which are unfamiliar to the
inexperienced analyst. When applying qualitative techniques to digital satellite data an
interpreter is provided not only with the potential to analyse information from other
wavelengths but also with a number of sophisticated tools to enhance these data (Scollar
203
1990 pp. 126-204; Philipson 1997; Campana and Pranzini 1999; Campbell 2002). Digital
image interpretation can improve on analogue interpretation by providing the interpreter
with geometric and radiometric enhancement techniques. Radiometric enhancement
techniques include contrast enhancement, histogram equalisation and histogram matching.
Geometric enhancement techniques include filtering, edge detection, spatial derivatives and
shape detection.
Multispectral images can also be transformed to generate new sets of image components or
bands, which represent an alternative, feature space, description of the data. Image
transformations include principal components analysis, image arithmetic, vegetation indices,
and the tasselled cap transformation.
Qualitative interpretation can be difficult to reproduce, though, many interpreters produce
image interpretation keys. These keys highlight the salient aspects used during image
interpretation and can be considered as a form of interpretative metadata (see appendix I.4).
7.3.2 Quantitative methodologies
When we think of the structure of archaeological data, most of us consider the sets of cases
and observations on the cases we may have collected from a site or series of sites. Data in
this sense are “things” or objects and their attendant descriptors. We can easily think of
other structural attributes of data, such as the three “dimensions” of archaeological data
postulated by Spaulding – time, space and culture – or of somewhat less general categories of
data such as “settlement” data or “spatial” data. Those of us with some mathematical or
statistical training may think of data in reference to some scale or measurement, such as the
now familiar quartet of nominal, ordinal, interval and ratio data. Some others may think
of data as swarms of points in Euclidean hyperspace and be concerned with the degree to
which these point swarms resemble standard models of statistical distributions.
(Aldenderfer 1987 p. 89)
204
The quantitative approach uses classification techniques for image analysis. It is hoped that
the numerical structure of the relevant archaeological residue data embedded in satellite
imagery can be extrapolated using these techniques. These include supervised classifications
like maximum likelihood, Mahalanobis, minimum distance, parallelepiped and unsupervised
classification techniques based on various algorithms.
Objects distributed in space have a relationship with one another. This relationship can be
expressed in a variety of ways but does tend to conform to the assumption that objects that
are close to one another are more related than distant objects. This phenomena is referred to
as spatial autocorrelation (Wheatley and Gillings 2002 pp. 131-134, 183). Hence, spatial data
does not conform to the traditional assumptions of statistics as levels of spatial dependence
and heterogeneity deviate from the norm. This body of statistics is referred to as spatial
statistics or geo-statistics.
Only a few researchers have applied quantitative remote sensing techniques for
archaeological prospection using imagery with a large footprint (for example see Gaffney et
al. 1996; Clark et al. 1998; Sever 1998; for example see Campana and Pranzini 1999; Harrower
et al. 2002). Other quantitative prospection applications, such as predictive modelling, tend to
occur within CRM bodies (Wheatley 1996; Wescott and Brandon 2000). One of the
difficulties of quantitative analyses is that it can be difficult to ascertain if clustering within the
data is relevant or irrelevant. Further, one needs to question whether the observed
relationship is a reflection of clustering in the data or is a residue of the quantitative process
(Aldenderfer 1987 p. 105). One of the major criticisms of predictive modelling is the amount
of irrelevant data that is produced. The decision of relevance is based upon the knowledge
and experience of the interpreter.
7.3.3 Digitising methodology
From a GIS perspective the archaeological residues identified during the prospection phase
can be broken down into point, linear and polygonal residues. There is a broad division
between the basalt zone and all other zones in respect of these feature types. The vast
majority of the residues identified in the basalt are cairns (currently identified as point
features) and wall segments (belonging to either field systems or structures and digitised as
polyline features). Where there is a complex of structures then these are agglomerated,
digitised as a polygon and issued a site number (see Figure 103). In all the other areas the
205
archaeological residues predominantly consist of polygonal features. However, some possible
linear features have also been identified. As discussed in section 6.2.3 all the digitising
occurred within ArcGIS.
One of the most important aspects of digital image interpretation is that the images are
normally already geo-referenced. Therefore, if an appropriate digitising methodology is
employed, the vector or raster derivatives defining residue location will also be georeferenced.
Figure 103 ‘Site' 358 in the basalt landscape and its surface
collection subsidiaries.
7.3.3.1 Data sources
Digitised data is primarily derived from two sources: digitised from the satellite imagery (see
Figure 104) or collected by GPS. The satellite imagery provides snapshots of the landscape
under different environmental conditions. The measured extent of residues from different
imagery and GPS may have subtle spatial differences. These differences will in part be caused
by spatial errors, but more significantly, they will also be caused by different environmental
conditions and post-depositional processes (such as bulldozing). Although these different
206
extents are potentially difficult to model, having a record of them will be useful for
determining the impact of cultural and natural modifications over time (see section 9.4). Due
to problems of co-registering satellite imagery, GPS data is considered as the primary
resource and, where appropriate, supplants all other data. However, GPS data can only be
collected for areas which have been ground observed, and in some cases use of GPS would
be inappropriate. For example, recording basalt walls by GPS would be extremely time
consuming and reflects the contemporary ‘modified’ landscape of the 21st century and not
necessarily the more intact archaeological landscape as described by, for example, Corona in
the late 1960s.
Figure 104 Digitised cairns and walls in the basalt area (scale
1:11,000).
207
7.3.3.2 Digitising ‘point’ features
Data represented as points are derived from two sources: points collected by GPS (the
location of sample points (e.g. the soil samples discussed in Chapter 8), archaeological
‘installations’ (such as oil-presses, discussed in section I.1.2)), and points digitised for the
location of cairns (see Figure 104). The research into the basalt area is still in a preliminary
phase and specific attributes for the cairns have not yet been finalised. Those attributes which
have been defined are currently held within the structure of the geodatabase.
7.3.3.3 Digitising ‘linear’ features
The same basic technique is used for digitising polyline networks as described in section
6.2.3.1. Polyline network digitising was employed as this reduces each line segment into a
feature type which can have attributes associated with it (e.g. from a site visit: see Figure 104).
Attributes for the wall segments have also not been finalised. Those attributes which have
been defined are currently held within the structure of the geodatabase. The same is true for
all other linear features identified in other areas.
7.3.3.4 Digitising ‘site’ polygons
The same basic technique is used for digitising ‘site’ polygons as described in section 6.2.3.2.
However, the unique referencing systems employed are different. The archaeology feature
data set contains a simple three tier hierarchy for the linking of all spatial and attribute data
(see Figure 105).
Figure 105 Schematic example of the a-spatial and spatial linkages
for the archaeological data.
208
The most generalised element is referred to as the Site Level. This contains a polygon outline
for the extent of the site, non-site or background landscape. This is the level where most of
the satellite imagery is digitised.
The next element is referred to as the Site Sub Unit Level. This contains a polygon outline for
any sub-units within the site. This allows large and complex sites to be broken down into
sub-units (see Figure 106). Although this layer may appear to be unnecessary it is employed
so that if any future excavations occur this can be used to reflect any feature or group
numbers assigned during excavation. For example, when conducting surface survey all
artefacts are collected from the ‘surface plough soil’ layer. For simplicity the ploughsoil layer
across the whole landscape is given a default value of 1. However, each of these layers is
spatially unique as they belong to a ‘Site’ with known spatial extent. For example any non
gridded surface collections at Tell Nebi Noah would belong to UnitID = 14 and SubUnID =
1 as this is the most refined level of spatial control (see Figure 105).
Figure 106 Example of the hierarchical structuring system at site
173.
209
The most refined element in the hierarchy is the Sample Square Level (note that these are not
always squares and could be excavated contexts). This contains the identifier for each sample
unit. Hence, artefacts collected from a 2m square on the surface of Tell Nebi Noah would
belong to UnitID = 14, SubUnID = 1 and SqrID = 100 (where 100 is the unique square
identifier for that site).
This simple hierarchical system has proven to be relatively robust. For example site 173 (Tell
Qatina) has been heavily eroded on its western side by Lake Qatina and a portion of its
surface levelled by bulldozing. The hierarchical structuring employed (see Figure 106) allows
each of these post-depositional events to be individually modelled. This structuring will not
only facilitate landscape analyses at virtually all scales but will also allow hinterlands and other
off-site related artefacts to be articulated within a single model. This is exemplified by site
sub-unit 173-3 which falls outside the modern extent of site 173. From a CRM perspective
this information is essential as it delineates areas of destruction outside of what could be
argued as the ‘scheduled’ area.
7.4 Prospection in the basalt zone
The basalt zone is distinct from all the other zones in that the majority of the archaeological
residues exist as extant physical features in the form of cairns, field walls and other structural
features. This in its own right has ramifications for the nature of the formation and
deformation processes that have occurred in this environmental zone (i.e. a stable matrix
with minimum deflation or aggradation in the landscape). This has resulted in a complex
multi-period palimpsest of field boundaries, structures and tracks.
Unfortunately this is now a landscape which is under significant threat. In the past 30 years,
enhancements to the road and rail networks (and the concomitant increase in associated
settlement activity (cf Sever 1998)) have removed some archaeological residues. Even more
significance is the clearance of fields, walls and cairns by bulldozing. Hence, in this
environmental zone, historical imagery should provide a more representative view of the
archaeological landscape than present day imagery (see Figure 107). However, this will only
be the case if the historical Corona imagery can be used to detect the same range of residues
as the present day Ikonos imagery.
210
Figure 107 The basalt landscape. Modern destruction is contrasted
between the Ikonos and Corona imagery (in italics). Points A, B, C
and D in the Ikonos and Corona images refer to the locations of
the photographs (building foundations, olive-press, wall, road and
birka respectively).
211
Figure 108 Comparison of the different spatial resolutions of the
imagery and their effects on identification in the basalt (scale
1:5,000).
212
Figure 109 Comparison of the different spatial resolutions of the
imagery and their effects on identification in the basalt (scale
1:1,000).
213
7.4.1 Image selection in the basalt zone
There is a positive relationship between the size of the objects under study and the resolution
of the imagery required to identify them. In the case of the basalt zone, where the smallest
wall width is c. 0.5-1m, (see Figure 108 and Figure 109) Jensen’s (2000) approximation that
the spatial resolution of the sensor should be one half of the feature’s smallest dimension
appears to hold.
At a scale of 1:5,000 (see Figure 108) all the images, with the exception of image B (due to
atmospheric haze and vegetation), clearly delineate the field-systems and clearance cairns.
However, in order to digitally map these features, increased magnification is required. At a
scale of 1:1,000 (see Figure 109) it becomes obvious that the lower (4 m) resolution of the
Ikonos MS imagery (images C’ and D’) is too coarse to provide an accurate backdrop.
It is interesting to note that both the cairns and field-systems are identifiable in these images,
but producing an accurate digital plan is out of the question. Image A’ at 2m resolution still
provides enough detail for mapping, although this is in part due to the shadow cast by the
walls. The panchromatic and pan-sharpened (resolution merged) Ikonos imagery provide the
best backdrop for digitising. Pan-sharpened false colour composites provide a better
interpretative medium than the panchromatic imagery due to the increased spectral
resolution (contrast E’ and F’ in Figure 109). Given the specific reliance on spatial resolution
in this environmental zone, the Landsat imagery is obviously of nominal value for
prospection purposes.
7.4.2 Qualitative techniques in the basalt zone
Qualitative techniques are the main mechanism for archaeological residue detection in the
basalt zone. The procedure effectively involves the visual interpretation of walls, structures
and cairns and digitising them directly within ArcGIS.
Images were enhanced through the normal histogram manipulations of standard deviation
stretch and histogram equalisation. Minimum-maximum and linear stretches tended to
reduce image contrast and hence were not used.
Kernel filters were also applied to the images to help enhance the visualisation of
archaeological residues. As the majority of residues are linear, edge detection algorithms were
214
employed. Both the Sobel and Prewitt non-directional kernel filters were applied within
Figure 110 Kernel filters applied in the basalt zone (scale 1:5,000).
It should be noted that as demonstrated in Figure 108 and Figure
109 that interpretative clarity is related to image scale.
215
Erdas Imagine (see Figure 110). However, these did not add any value to the other analytical
techniques. Of more interest were the ‘crisp’ filters. This filter did allow the detection of
features that were more ephemeral. Unfortunately, due to temporal fieldwork constraints, it
was impossible to verify if these were actual artefacts or anomalies inherent in the data
structure. However, from the general alignment of features one is inclined to place them with
the former.
7.4.3 Quantitative techniques in the basalt zone
Quantitative techniques were applied within the basalt zone with limited success.
Unsupervised classifications were conducted on both the Corona and Ikonos imagery. The
results of these analyses were, unsurprisingly, disappointing. Image segmentation did produce
slightly more worthwhile results. However, segmentation values needed recalculating for each
zone. Furthermore, no new information was extracted using these techniques and the vector
information still required digitising. Quantitative techniques were not pursued further.
7.4.4 Case study in the basalt zone
Initially it was decided to select a 4 km2 area around the village of Borg el-Qaí. The primary
reason for this choice was that it has just been placed under the protection of the Directorate
General of Antiquities and Museums (Damascus). This represents one of the last remaining
intact palimpsests of archaeological residues in any basalt area in Syria (Abdulkareem, pers.
comm.). Unfortunately this area is at the extreme north of the application area and is only
covered by the poor quality Corona mission 1110 imagery.
As an alternative a 4 km2 area around the village of Krad ad-Dâsiniya (Akrad) was selected.
Akrad lies on the edge of a plateau overlooking a gentle slope down towards Râm Sheikh
Hanifa in the poorly drained basalt zone (see Figure 98). The original village nucleus (site
866) is a low tell-like structure, comprising small rectangular houses and narrow
interconnecting alleyways, all constructed in roughly-dressed basalt. Much of the construction
material is re-used. The Corona imagery indicates that by the late 1960s the village had
expanded to the south and west (following the main road routes) from its original nucleated
core.
The hinterland for the village appears to be split into a number of zones (see Figure 111).
The immediate environs are used for market gardening (vines, vegetables and orchards) in a
216
dense pattern of small walled plots. Beyond these plots are larger fields which would
traditionally have been used for dry-farming cereal cultivation. Sheep and possibly cattle are
grazed in the more marginal ground.
Figure 111 Hinterland zones at the SE of Krad ad-Dâsiniya
(digitised walls are derived from Ikonos imagery). The parallel
walls represent access tracks.
Within the hinterland zone it is possible to extract at least four different systems of land
management from the geometry and stratigraphy of the walls and the form of the enclosed
areas (see Figure 178):
1. Large fields cleared with a bulldozer (modern): these are demarcated by large
piles of boulders up to 5m wide. These bulldozed areas do not tend to appear on
the Corona imagery.
217
2. Long straight dry stone walls (modern: < 30 years old): Built of regular basalt
blocks these walls line access routes and surround groups of smaller fields. These
walls do not tend to appear on the Corona imagery.
3. Traces of multiple parallel, low linear walls probably related to the past Mouchaa
farming system (pre 20th century land reform): Large fields (probably from
Centuriation) evenly divided into smaller units.
4. Centuriation: extensive field system with large fields instituted (mainly Roman).
Not easy to detect across the whole area.
Figure 112 Comparison of the observable detail between Corona
and Ikonos imagery and their resultant digitised interpretations
(scale 1:2,000). The lack of parallel walls on the Corona could be
due to its lower resolution. Alternatively the tracks could have
been created to allow tractor access.
Across the study area these systems maintain the same broad field orientation of 352º
(Newson, pers. comm.). Vestiges of the centuriation, as the earliest identified form of land
division, are still incorporated into many of the later systems.
218
Figure 111 and Figure 112 display digitising from the south eastern quadrant centred on
Akrad. The higher spatial resolution in the Ikonos imagery is translated into a much more
detailed vector output. Even though there are differences in the digitised outputs the core
structure of the systems are very similar (see Figure 114).
A broader scale evaluation of the settlement and hinterland systems in the basalt zone is
currently being undertaken by Dr. Paul Newson (Department of Archaeology, University of
Durham) as part of an AHRB post-doctoral research project.
Figure 113 Comparison of the Corona and Ikonos imagery in a
bulldozed area of the basalt (scale 1:3000). Even though bulldozed
the Ikonos resolution merge image exhibits some of the residues
seen on the Corona image.
219
7.4.5 Evaluation of sensors in the basalt zone
In this application area the Corona imagery predates the significant use of bulldozing and has
recorded an effectively intact landscape with minimal destruction, disturbance or masking of
archaeological residues by present day agricultural or settlement expansion. This is a
potentially fortuitous set of circumstances: the application area was, and still is, considered
militarily sensitive therefore a relatively large number of Corona missions intersected the
application area.
Of the relatively cloud free missions only 1108 and 1111 provided the appropriate spatial
resolution and contrast for digital mapping in the basalt (compare image A against image B in
Figure 108). Unfortunately the 1108 imagery does not cover the whole basalt zone and
mission 1111 does not intersect it at all (see Figure 115). The mission 1111 collection time (c.
18:30 hours) would have recorded shadows which could have improved interpretation.
Although the spatial resolution of the Corona imagery is high enough for mapping purposes
it is not as refined as the Ikonos imagery. Hence a more generalised mapping product is
generated (see Figure 113 and Figure 114). Furthermore, Corona imagery does not display
the full range of residues as seen on the Ikonos imagery and noted during ground
observation (see Figure 116). The cause of this is unknown although it may have something
to do with the time the imagery was collected, the radiometric resolution of the film or the
scanning process. Most importantly, the Corona imagery requires extensive geo-referencing
(see section 5.4). As experienced in the first field season, without the Ikonos imagery as a
geo-referencing base (as at the time differential GPS was impossible and handheld GPS
correction was too time consuming) the Corona imagery would have highlighted many
residues but locating them in this complex landscape would have been difficult if not
impossible.
The Ikonos imagery records the landscape under the environmental conditions determined
during the collection window (see section 5.2). Therefore, in comparison to Corona, there is
much more flexibility in obtaining an image where archaeological residues are easier to
distinguish. Much like any source of evidence the Ikonos imagery may not fully represent the
archaeological resource due to recent landscape modification (examine the wooded area,
centre right, in Figure 117). In comparison to the Corona the spatial resolution of the
panchromatic Ikonos imagery allows detection with a higher degree of confidence.
Consequently, the accuracy of the digitising is improved.
220
Figure 114 Comparison of digitising using the Corona and Ikonos
imagery as backdrops (scale 1:3,000).
221
The multispectral Ikonos imagery should not be used for digitising purposes as the spatial
resolution is too coarse. However, the increased spectral resolution of the pan-sharpened MS
Ikonos imagery improves the detection of residues that are difficult to distinguish in the
panchromatic (see Figure 112 and Figure 114). Alternatively placing the Ikonos multispectral
over the panchromatic with a 40-60% transparency produced a similar result. This technique
has the advantage of reducing storage requirements. For example, where an Ikonos pan and
MS image have file sizes of 235MB and 60MB respectively the resolution merged image has a
file size of 940MB. Furthermore, the pan-sharpened Ikonos imagery has allowed the
identification and mapping of wall elements where the above ground component has been
removed by bulldozing (see the Ikonos 3,2,1 resolution merge in Figure 113).
In summary, both Corona and Ikonos imagery are appropriate sensors for the prospection
and mapping of archaeological residues in the basalt zone. The Corona imagery provides a
synoptic view of the landscape prior to recent destructive modifications. Thus the Corona
imagery is potentially a more representative snapshot of the archaeological resource than the
Ikonos. However, the Ikonos imagery produces a less generalised view of the archaeological
residues. When the Ikonos and Corona imagery are used in conjunction with one another
further benefits are realised. From a CRM perspective the analysis of both data sources
provides an overview of the archaeological residues and the range and number of destructive
modifications over the past thirty years. The Ikonos imagery provides a present day snapshot
of a landscape under destruction and is therefore useful as a CRM tool to determine the level
of threat to the landscape. This form of information is essential when determining the
management strategy for this environment (this is discussed in more detail in Chapter 9).
From a methodological perspective the most productive approach would be to ensure that
the imagery is geo-referenced and co-registered to a high degree of ground precision. The
archaeological residues should then be digitised from the Ikonos imagery (either 1m pan or
1m pan-sharpened (see Figure 112)). This would produce a present day ‘benchmark’ of the
potential archaeological residues essential for constructing a CRM strategy. The Corona
imagery can then be digitised or alternatively evaluated against the Ikonos digitising. Residues
that have been masked or eradicated in the intervening period can be mapped, integrated into
the data model and evaluated. As regards user interpretation of the imagery, qualitative
techniques for residue detection are more appropriate than quantitative techniques. However,
some image enhancement algorithms, particularly sharpening and edge enhancement kernels,
222
can help improve detection. Specific quantitative techniques and classification algorithms do
not appear to produce adequate results. Although it could be argued that automated
archaeological residue detection techniques are desirable, from an archaeological
interpretative viewpoint, at least in this zone, this means that the interpreter is less likely to
immerse themselves in the data and hence subtle relevant nuances may be overlooked
Figure 115 The extent of the Russian aerial photographs and
Corona imagery in the northern application area.
Using such a technique it is possible to record the extent and impact of landscape
modifications that have occurred in the intervening years. The resultant vector map when
used in conjunction with ArcPAD and a GPS facilitates the navigation to specific residue
elements with confidence (see section 9.3.1). The accurate co-registration between desk based
digitising and field reconnaissance is a significant methodological advance that has resulted in
substantial financial savings. For example, digitising these field systems using traditional total
station technology, apart from being extremely difficult and time consuming, would have
been very expensive (cf. Newson 2002 pp. 166-172).
223
7.4.6 Russian aerial photographs in the basalt zone
In addition to the Corona and Ikonos imagery, Russian vertical aerial photography was made
available by Dr. Abdulkareem in 2003 (see section 5.1.4.2). The five overlapping aerial
photographs covered a small portion of the Northern study area. Unfortunately these
photographs did not correspond with the extent of the Corona 1108 imagery (see Figure
115). Therefore, only the Ikonos imagery was directly comparable with the aerial
photography. Comparisons between the Corona 1108 imagery and the aerial photography are
extrapolated (see Figure 117).
Figure 116 Comparison of the Russian aerial photography, Corona
and Ikonos imagery.
Using the analogue to digital conversion technique described in section 5.1.4.2 the Russian
aerial photographs have similar spatial resolution to the Ikonos panchromatic and spectral
characteristics of the Corona imagery. Furthermore, it was assumed that this aerial imagery
was collected for mapping purposes. If correct, this imagery would have been taken with a
metric camera at a relatively stable (low) platform with a relatively low cloud cover index.
224
Figure 117 Comparison of the Russian aerial photography, Ikonos
and Corona imagery over an unknown site and a comparative
evaluation over a different area using the mission 1108 Corona.
225
This combination of factors has produced crisp imagery with good geometric properties.
Although it is not possible to directly contrast the Corona 1108 and Russian aerial
photography, extrapolation from the 1970 Corona 1110 gives the impression that the
landscape was not significantly modified between 1958 and 1970. This adds further credence
to the assumption that the Corona imagery is a representative snapshot of a preserved
archaeological palimpsest. However, like the Corona imagery the contrast of the photography
does not highlight all the archaeological residues (see Figure 116). This could be a result of
poor storage or the scanning process. It is hoped that a more appropriate scanner with better
spatial, radiometric and geometric fidelity would produce even more detail and compensate
for any potential inadequacies in the original scanning process.
On a point of methodology it is important to stress that without the Ikonos imagery accurate
rectification of these photographs would have been extremely time consuming (see section
5.1.4.2). Furthermore, it is interesting to note that although Dr. Abdulkareem stated that
these photographs are Russian in origin they are labelled in Roman rather than Cyrillic script.
7.5 Prospection in the marl zone
The marl zone contains archaeological residues that contrast strongly with those found in the
basalt zone. The majority of these residues do not exist as hard wearing structural material
(such as basalt blocks); rather it appears they are remnants of the ongoing formation and
deformation processes associated with settlement and the creation and decay of mud-brick
architecture. The residues can be broadly considered as anthropogenically induced physical
and chemical variations of the localised soil matrix. Settlement evidence consists of mounded
tells and other ‘flatter’ sites producing a distribution of discrete ‘site loci’ across the zone.
These residues are at least an order of magnitude larger than those in the basalt. The smallest
radius attributed to a site is 23m (site 478).
At a very early stage in the research it was observed that these sites showed a distinct soil
colouration in comparison with the background soil matrix (see Figure 118). This
phenomenon, which is investigated in detail in Chapter 8, is the basis for residue prospection
in this zone. Hinterland evidence in the form of field systems associated with these
settlements has not been recognised either on the satellite imagery or during ground survey.
226
Figure 118 The marl landscape and ‘sites’. Modern development is
contrasted between the Ikonos and Corona imagery (in italics).
227
Figure 119 Comparison of the different spatial resolutions of the
imagery and their effects on identification in the marl (scale
1:12,500). Sites are highlighted in image A.
228
Figure 120 Comparison of the different spatial resolutions of the
imagery and their effects on identification in the marl (scale
1:5,000).
229
Figure 121 Different visualisations of raw and processed Landsat
imagery overlaid by ‘sites’. Yellow sites are flat, black sites are tells.
230
The archaeological residues in this area fulfil many of the expected norms for the Middle
East. This is reflected in the tell focused archaeological studies that have been conducted in
Syria for many years and which have, in part, been criticised by Wilkinson (2000).
In common with the basalt zone the marl zone is also experiencing residue modification
through landscape clearance, urban expansion and changes in agricultural technique
(particularly deep ploughing, see for example Lambrick (1977)). The contrast in construction
technique between the zones is most striking. The basalt environment exhibits extensive
structural evidence which are not evident in the marl. At present only areas of settlement
have been located in the marl and associated hinterland systems appear to have been
eradicated. Continuous ploughing has not yet eradicated the settlement residues in the marl.
However, current survey evidence indicates that over time the marl landscape was more
intensively settled and exploited than the basalt landscape. While recognising that this is a
simplified comparison this does imply that there is an increased likelihood of cultural
transforms in the marl landscape. This landscape will have seen a number of profound
variations in landscape structure over time. Both cadastration and musha’a farming would
have impacted on previous landscape structures. For example, Van Liere (1959) identified
centuriation in the marl landscape East of Homs and one would expect this system to extend
into all the Homs hinterlands. However, no trace of this land management structure has been
identified in our marl areas.
7.5.1 Image selection in the marl zone
The larger size of residues in the marl zone increases the range of imagery which is effective
for prospection. Obviously the Ikonos (4m multispectral and 1m panchromatic) and Corona
imagery are appropriate. Theoretically the 15m panchromatic band on the Landsat ETM
sensor should detect all residues and the other 30m resolution bands on the Landsat platform
should detect most residues. The increased size meant that pan-sharpened Ikonos imagery
was not employed. However, the transparent overlay technique of MS over pan Ikonos
imagery was used.
Figure 119 compares the resolution of the primary sensors within the marl zone. Without any
major image manipulation most of the residues are relatively easy to distinguish in the
Corona and Ikonos panchromatic imagery (compare images A or B with C). However,
231
present day modifications (i.e. infrastructure building, settlement expansion and deeper
ploughing) have increased complexity in the Ikonos imagery. It is interesting to note that the
reduced image fidelity of the mission 1110 Corona image, which made it an inappropriate
resource for prospection in the basalt, is not so significant for the marl zone. The increased
spectral resolution of the Ikonos multispectral allows the creation of colour composites
improving visual detection (compare image C with image D). The pan-sharpened Landsat
imagery still allows the identification of the sites. However, the decreasing spatial resolution
means that more ‘potential’, but negative, sites are identified. This would necessitate
increased time for ground observation. This is further exacerbated in the 30m Landsat
imagery. However, the increased spectral resolution of the Landsat imagery allows greater
flexibility in image processing and visualisation. Different bands can be combined to produce
false colour composites and alternative visual representations such as tasselled cap and
principal components (see Figure 121). These same techniques can be applied to the Ikonos
multispectral imagery. Of particular relevance are Landsat bands 5 and 7 in the infrared
which provide useful information about soil and surficial geology.
Figure 122 Spectral profile of on and off-site point locations
derived from Ikonos MS imagery.
As the scale is decreased (see Figure 120) other interpretative artefacts become clearer. The
Corona imagery appears to display the greatest contrast between archaeological residues and
the background soil and vegetation. The Ikonos imagery provides the best backdrop for
digitising features (such as the military entrenchments on site 256 (compare C’ against all
232
other images). It also interesting to note that the 4m multispectral Ikonos provides better
clarity for mapping purposes than the Corona (compare image D’ against A’ and B’). At these
scales the spatial resolution of the Landsat imagery makes the scenes relatively unintelligible
(compare E’ and F’ with E and F in Figure 119).
7.5.2 Spectral nature of the residues in the marl zone
As previously discussed, the main mechanism for detection in the marl is through differences
in soil colour associated with archaeological sites. This necessitates a change in reflectance
within, at least, the visual portion of the EM spectrum. This phenomenon will be discussed
in more detail in this section. For simplicity, throughout this section the Ikonos MS imagery
is used for illustrative purposes, as it has a relatively high spatial resolution (useful for
examining transects and surfaces) and a medium spectral resolution (useful to examine
changes across the EM spectrum).
Figure 122 displays the spectral profile from a single on and off-site point for a number of
different site types in the marl. In all instances, with the exception on site 279 (discussed at
length in Chapter 8), there is a significant increase in DN values for ‘site’ pixels of between
20 and 30%. The graph also reveals that there is not a single spectral signature for
archaeological residues. However, most sites do show a general increase for bands 2 (c. 0.55
microns) and 3 (c. 0.65 microns) against bands 1 (c. 0.5 microns), which may be useful for
band ratioing.
A closer examination of the spectral content of the archaeological residues was deemed
necessary. Using ERDAS Imagine, transects and surface spectral profiles were taken from
the Ikonos MS imagery for a number of sites within the marl zone (see Figure 123). These
sites were of a range of different types (tells to flat sites), chronological periods and in
different soil zones (as defined in section 6.6).
Most sites displayed an increase in each band in comparison to the off-site readings. Sites 218
and 279 were the exceptions. Site 218 exhibits very strong reflectance in the NIR and
consistent readings in all other bands. It is likely that this site was largely masked by crop at
the time of image collection. For detection purposes the NIR is the least reliable band of all.
Vegetation makes it difficult to detect sites in the NIR band in a number of profiles (218,
233
221, 259, 339, 454, 478 and 602). Although, this band is not valuable for detecting residues it
does allow one to simply plot the location of vegetation.
Figure 123 The distribution of archaeological residues where
transect spectral profiles and surface profiles were conducted.
234
Figure 124 Transect spectral profile and surface profile of sites 97256. The line and rectangle in the location represent the transect
and surface profile respectively.
235
Figure 125 Transect spectral profile and surface profile of sites
259-339. The line and rectangle in the location represent the
transect and surface profile respectively.
236
Figure 126 Transect spectral profile and surface profile of sites
454-602. The line and rectangle in the location represent the
transect and surface profile respectively.
Where there is limited vegetation cover the blue, green and red bands appear to provide the
most accurate reflection of the site extents. There is generally a 20-40% increase in DN value
between off and on-site values. Most off-site reflectance values were in the range of 200-250.
These values were seen to increase the further south one moved through the marl zone
(possibly due to a decrease in soil organic matter content, a decrease in moisture content or
237
thinner soils). Site 238, which lay in an area of silts deposited by an active seasonal wadi,
exhibited the lowest percentage change (c. 10%), also had the highest general reflectance
values (an average of c. 350). Sites 238 and 279 do not exhibit the same consistent change in
reflectance as seen on the other sites. Rather, these sites exhibit high frequency sinusoidal
variations in reflectance which are possibly related to changes in texture.
The red and green bands in particular show the greatest percentage increase between off-site
and site soils. Hence, the red band was used to display the surface profile. Although the
surface profiles are difficult to interpret they work in a similar way to a digital elevation
model, but the DN value is plotted in the z axis. In general these plots display the gradually
increasing reflectance between off-site and site soils. Roads (i.e. the peaks in 271, 454 and 508
and the trough in 339) and buildings (site 97, 271 and 508) are in a number of the plots.
There is generally quite a close correlation in the shape of the red, green and blue bands
along transects. The red and green bands tend to have approximately the same DN values
while those of the blue band tend to be lower. These curves also indicate that a ratio between
the red or green bands and the blue band may be useful (off-site it would approach unity).
Once again the plots reinforce the fact that there is no single spectral signature for
archaeological residues. However, each site does express quantifiable changes that can be
identified from the background soil and vegetation responses.
7.5.3 Qualitative techniques in the marl zone
Qualitative techniques are the main mechanism for archaeological residue detection in the
marl zone. The procedure effectively involves the visual interpretation of sites and delineating
their extent by digitising the area of distinctive reflectance in ArcGIS. As has already been
discussed many of the archaeological residues are relatively easy to detect as they display a
marked increase in reflectance. However, initial supervised and unsupervised clustering
algorithms covering the whole marl area did not produce specific spectral signatures that
responded to archaeological residues (discussed in section 7.5.4). This led to the hypothesis
that archaeological sites in the marl area change the physical and chemical properties of the
soil to such an extent that the ‘site’ exhibits a localised quantifiable increase in reflectance.
This hypothesis was the basis for the majority of image manipulations to improve residue
detection.
238
Although no specific spectral signature can be defined that describes archaeological residues
across all areas one can apply a variety of different techniques to improve detectability. Five
major avenues were chosen (note that the word zone in this context refers to any defined
geographical extent with some level of similarity):
•
Traditional histogram manipulation and false colour composites.
•
Histogram enhancements within spatial zones.
•
Archaeological residues as background soil variations.
•
Band indices in the Ikonos MS.
•
Image visualisation using alternative feature space transformation.
7.5.3.1 Traditional histogram manipulation and false colour composites
Both the Corona and Ikonos responded well to the normal histogram enhancements
(standard deviation stretch, minimum-maximum stretch and histogram equalisation). The
application of stretches depended primarily on where the imagery was located. For example
the minimum-maximum stretch did not work well with the Ikonos imagery as it tended to
reduce contrast. However, in certain Corona scenes (such as Cor_1108_mid) the minimummaximum stretch was the most effective. For the Ikonos MS images false colour composites
provided better residue visualisation than individual band stretches. All combinations worked
well although a 3, 2, 1 band combination provided the best contrast particularly when applied
in conjunction with a histogram equalisation stretch (it also produces a familiar colour
scheme). As discussed in section 7.5.2 the red, green and blue bands provide the best
response for archaeological residues in this environment.
In the southern part of the Southern marl zone the Ikonos imagery is relatively bright with a
low contrast making residue detection quite difficult. As was discussed in section 7.5.2, this is
because the soils in this area have a higher reflectance than those in other areas. However,
this phenomenon was not evident in the Corona imagery that intersected the same region.
This could, in part, be due to the increased radiometric resolution of the Ikonos imagery or
recent landscape modifications. It is, however, more likely that it is due to changes in the
nature of the surface soil as a consequence of present day land management practices (a
decrease in soil organic matter content, a decrease in moisture content or thinner soils).
239
Figure 127 Comparison of zonal and normal Ikonos MS imagery
(3,2,1 FCC), with a close up of site 238.
7.5.3.2 Histogram enhancements within spatial zones
Histogram enhancements are based upon manipulating the statistical distribution of pixels
(O'Brien et al. 1982). If an image contains many contrasting zones then the enhancement of
variations within each zone is likely to be reduced. Detecting residues in zones where the
240
contrast has been reduced is much more difficult. In order to maximise contrast in each zone
it was decided to subdivide the Corona and Ikonos imagery based upon zonal data derived in
section 6.6 (see Figure 123). These new images correspond to areas with similar soil
characteristics. These should show improved contrast when histogram enhancements are
applied, hopefully improving detection. The soil/geology/urban image was chosen because it
corresponds with changes in reflectance in the soil zones which appear to impact on the
visual detection of archaeological residues. Urban areas have been removed, as they tend to
have very high reflectance and therefore skew the histogram's distribution.
Figure 127 compares the original Ikonos MS imagery with a set of zonal Ikonos images with
a standard deviation histogram stretch. As expected, the zonal imagery exhibits much
improved contrast across the whole marl zone. Therefore archaeological residues were much
easier to detect visually. This is particularly well demonstrated in the inset image of site 238,
from the wadi silts, where the original Ikonos image had very low contrast. All the standard
histogram stretches proved useful for identifying residues in the zonal data sets.
The subdivision of a large image into areas of similarity (or ecologically these could be viewed
as large scale patches) is an interesting technique that could produce further dividends for
qualitative interpretation. The subdivision in the imagery improves contrast manipulation in
each zone allowing a greater range of values to be expressed, which improves visual
detection. Trimming each image into the different zones does not affect the numerical
structure of the data, therefore it can still be considered as raw data. Hence, a range of
quantitative manipulations can still be conducted on the data, such as band ratios and
classifications.
7.5.3.3 Archaeological residues as background soil variations
Before the 20th century the major building material in the marl zone was mud-brick. The
destructional debris of these structures exacerbated by incorporation with other
anthropogenic refuse (particularly ash) has resulted in ‘tells’ which dominate the landscape.
Even though mud-brick is normally of local origin, it is postulated that the incorporation of
these archaeological residues into the soil matrix would reduce the average but increase the
range of particle sizes and increase the organic content (discussed in detail in Chapter 8).
Archaeological sites reflect the complex interaction between the sub-surface characteristics
and surface characteristics of the soil and the vegetation growing on it. This interaction
241
disguises the nature of the archaeological residues by distorting the underlying geometry of
the feature or masking it completely. These variations in the nature of the soil matrix are
exploited in many archaeological prospection techniques including geophysical, geochemical
and aerial survey techniques (Kouchoukos 2001).
As has been demonstrated in section 7.5.2, the archaeological residues tend to exhibit a
distinct increase in reflectance in comparison to the background soil matrix. This change in
reflectance is not consistent and it is, therefore, difficult to define a distinct archaeology
spectral curve that will detect residues across the marl zones. However, subtracting an
averaged background soil DN value from an on-site pixel DN value will generally produce a
positive value. It was decided to apply a moving average kernel (see section 2.2.6.4) to the
Corona and Ikonos imagery in order to evaluate whether residues were easier to locate in the
resultant statistical surface. In theory, after processing, areas of unmodified soil should have
an average approaching zero. Features that significantly deviate from these background
values, such as archaeological residues, roads, buildings, crops and water, should exhibit
positive or negative values. It is likely that this kernel would create a statistical surface which
approximates to a normal distribution with a mean value of zero.
Figure 128 Statistical distribution of the Red band after 200m
averaging kernel.
242
Figure 129 Comparison of 200m kernel average and normal
Ikonos MS imagery (3,2,1 FCC), with a close up of cluster of sites.
After empirical trial and error approaches to define an appropriate radius for the averaging
kernel, a compromise of 200m was reached. This provides a large enough area that extends
beyond the outlines of most sites and can be processed in a reasonable time frame. However,
even this relatively conservative figure demands a large amount of processing power. For
243
example the Ikonos panchromatic imagery, at 1m resolution, requires a 400 by 400 kernel
(note that most kernels employ 3 by 3, 5 by 5 or 7 by 7 matrix). Thus each new pixel value is
the average of the 160,000 surrounding pixels, however, only 125,664 pixels are used (as this
is a radius kernel), all external pixels have a weight of 0. Therefore, for the eastern marl
panchromatic Ikonos image a total of 2*1013 calculations is required. To overcome these
significant processing problems the 1m resolution data was degraded to 4m and a 200m
radius was calculated from this derivative. This significantly reduced the number of
calculations to 3*1011. Although this reduces the accuracy of the kernel algorithm, spatial
autocorrelation means that this should not significantly skew the results. For comparative
purposes the same technique was used to degrade the Ikonos MS imagery to 8m for the
calculation of a 400m radius kernel algorithm. The results of the processing are real numbers
which has a concomitant increase in storage size for each image.
Figure 130 Transect spectral profiles for 200m mean Corona 1111
and Ikonos MS imagery at sites 256, 279 and 339. Note the use of
negatives for Corona resulting in a lower DN value for sites.
The results of the averaging kernel on all the imagery were excellent. Figure 129 compares a
3,2,1 false colour composite of the 200m kernel average and unprocessed Ikonos MS
imagery. Although from the small scale image it looks like this technique provides no benefit,
the close up of the cluster of sites displays how the processed imagery aids visual detection:
244
site 478 (the prehistoric site (see the profile in Figure 126)) is significantly enhanced as is site
458 (an Islamic settlement in the lee of tell site 256). The small scale imagery appears
effectively in grey scale as most off-site values are close to zero in all bands. When these are
combined in a colour composite shades of grey are produced (see the off-site vales in Figure
130). As expected, this algorithm did produce a near normal distribution with a mean
approaching zero (see Figure 128).
Figure 130 shows spectral profiles of the averaging kernel for sites 256 (tell), 279 (flat site)
and 339 (flat site). For reasons of clarity the NIR band was not displayed in this figure (see
Figure 124 and Figure 125 for comparative profiles). As expected site 256 and 339 have a
higher value than the local mean (this is inverted for the Corona negative). It is also possible
to identify site 458 to the SW of 256. The site 279 Ikonos profile does produce the same
decrease as displayed in Figure 125. Of particular interest is the blue band which exhibits
minimum variation from the mean. On the other hand the Corona profile for site 279 neither
exhibits a consistent increase or decrease. The mean across the whole length of the transect
appears to approximate to zero. However, the frequency of peaks decrease and their
amplitude increase when over the site. This probably reflects the variations in texture at this
site (as discussed in section III.6).
For comparative purposes, a 400m radius averaging kernel was applied to the Ikonos MS
image. Figure 131 shows an example area of the 200 and 400m kernels. The residues exhibit
an increased difference from the mean in the 400m radius kernel. This is to be expected as a
larger radius should move the kernel average closer to the scene mean. The 400m radius
kernel has the distinct advantage of improving the contrast for larger sites (such as 256). On
larger sites a 200m radius kernel may contain a majority of site pixels. Hence, the contrast
against background soil values is reduced. Theoretically there should be a distance limit
beyond which the benefits of an averaging kernel would be reduced. It is recommended that
further research is conducted into defining these limits.
This technique provided a number of significant benefits: not only does the averaging kernel
improve contrast across the whole image without the need to subdivide it (see section 0) it
also corresponds to theoretical expectations. In order to detect archaeological residues then
there needs to be a local contrast between a variable on the site and off the site. This
technique examines those local variations. Furthermore, there could be a whole range of
245
quite simple statistical kernels that improve on this relatively simple tool. Unfortunately this
technique significantly disrupts the structure of the original data. The results of this analysis
should, therefore only be incorporated into further quantitative calculations with care.
7.5.3.4 Band indices in the Ikonos MS.
As previously discussed (see section 7.5.2), ratioing bands 2 or 3 over band 1 may highlight
areas that contain archaeological residues. The same techniques used to derive vegetation
indices were employed for two reasons:
1. The difference between Band 1 and Band 2 or Band 1 and Band 3 for contrasting
on and off-site reflectance is similar to the red-NIR shift exploited in vegetation
indices (see Figure 122).
2. A large amount of research has been conducted on differentiating vegetative
from non-vegetative material.
The following analogous indices were produced:
•
NIR/R (near infra-red/red).
•
SQRT (NIR/R).
•
Vegetation Index = NIR-R.
•
Normalized Difference Vegetation Index (NDVI).
When producing these indices the NIR band was replaced by band 2 or 3 and the red band
was replaced by Band 1. All of these indices did highlight archaeological residues to some
degree. However, none of them provided any more information than was already available
from a 3,2,1 false colour composite. Substituting Band 2 for NIR proved to be slightly better
for visual detection than substituting Band 3.
7.5.3.5 Image visualisation using alternative feature space transformation
Tasselled cap and Principal Component Analyses (PCA, see section 2.2.6.3.2) were
conducted on the Ikonos MS imagery. The tasselled cap transformation has been developed
specifically by Space Imaging (SI) for its Ikonos satellite (Horne 2003). This transformation
has been devised to maximise the separation between different surface types. The PCA
algorithm was instructed to produce four principal components.
246
Figure 131 Comparison of the 200 and 400m radius averaging
kernel on an Ikonos MS image.
247
Ikonos MS SW
Component 1 Component 2 Component 3 Component 4
0.419
0.278
-0.496
-0.708
Band 1
0.491
0.452
-0.303
0.680
Band 2
0.433
0.344
0.814
-0.179
Band 3
0.629
-0.775
0.007
0.064
Band 4
83.916
14.383
1.391
0.311
Percentage variation
Ikonos MS SE
Component 1 Component 2 Component 3 Component 4
0.376
0.142
0.750
0.525
Band 1
0.476
0.270
0.265
-0.794
Band 2
0.497
0.578
-0.571
0.304
Band 3
0.620
-0.757
-0.201
0.047
Band 4
91.481
6.950
1.485
0.083
Percentage variation
SI TasCap
Component 1 Component 2 Component 3 Component 4
0.326
-0.311
-0.612
-0.650
Band 1
0.509
-0.356
-0.312
0.719
Band 2
0.560
-0.325
0.722
-0.243
Band 3
0.567
0.819
-0.081
-0.031
Band 4
73.240
25.060
1.530
0.160
Percentage variation
Table 18 PCA (Southern Marl only) and Tasselled Cap component
matrix and percentage variation for Ikonos MS (the tasseled cap
statistics are after Horne 2003).
Table 18 displays the component matrix and the percentage variation contained within each
component. The first two components in all the transformations account for c. 98% of the
total scene variance. Component 1 in all the transformation is approximately the average of
all the bands and corresponds very well with the Ikonos pan image (albeit at 4m resolution:
see Figure 132). Component 2 approximates to NIR minus the visible bands for the tasselled
cap transformation and visible minus the NIR for the PCA transformations. This component
will primarily indicate vegetation and water. Component 3 approximates to Red minus Blue
for the tasselled cap and the south western (SW) PCA transformation and Blue minus Red
for the south eastern (SE) PCA transformation. This should also indicate some
archaeological residues due to the predominance of bands which have been identified as
useful for ratioing. Component 4 approximates to Green minus Blue for the tasselled cap
and the SW PCA transformation and Blue minus Green for the SE PCA transformation.
This should indicate some archaeological residues due to the predominance of bands which
have been identified as useful for ratioing. However, this component has low variance and is
too noisy for detection purposes. The correlation between the components in the SI tasselled
cap transformation and the SW Ikonos PCA transformation is probably due to the presence
248
of Lake Qatina in the western image and the lack of a significant water body in the eastern
image.
Figure 132 Components of Tasselled Cap and PCA analysis in the
marl.
249
Figure 132 shows examples of the transformations in the marl zone. There is no significant
improvement for residue detection in these transformations although component 3 in the
PCA did provide some improvements in the southern marl. However, these transforms
should prove quite beneficial for any future re-interpretation of the land cover classifications.
7.5.4 Quantitative techniques in the marl zone
In general the application of quantitative techniques in the marl zone was of limited success.
Image classification and segmentation techniques were evaluated.
Figure 133 Classification of the marl and thick marl zones
(residues in Blue and Green and sites outlined in white).
Supervised and unsupervised classifications were applied to the Corona and Ikonos imagery.
The results of these classifications were generally disappointing. The unsupervised
classifications produced a better correlation than the supervised classification. Some
250
archaeological residues were identified with each technique although there was no
consistency. The problems of pixel mixing, the coarse sampling of the electromagnetic
wavelengths and the non-standardised residue response meant that spectral classifications
grouped different features as the same entity.
Figure 134 Ikonos MS 200m mean image segmentation (100 block
size, 60 spectral threshold and 250 region size)
251
It was felt that subdividing the image into spectrally similar areas might improve the accuracy
of the image classification. Twenty class unsupervised classifications were undertaken in each
of the individual soil zones for all the sensors. Again, in general these did not produce
favourable results. However, the marl and thick marl zones (see Figure 123 and Figure 133)
did produce some worthwhile results where the classification groups correlated relatively well
with the known archaeological residues. Even with this correlation the classifications do not
indicate any previously unknown residues.
Image segmentation examines the spatial variation in changes in reflectance. It was hoped
that this technique would produce better results as the archaeological residues actually exhibit
these spatial variations (see the surface plots from Figure 124 to Figure 126). Separate block
sizes, spectral thresholds and region sizes were produced by trial and error for the Ikonos
MS, pan and Corona. Different parameters were required as each image has a different spatial
resolution (impacting block size) and radiometric resolution (impacting spectral threshold).
Region size was based upon the size of archaeological residues (a minimum area of 200m2
was applied). These parameters represent a trade-off between complex and generalised
segmented images. The spectral threshold value is particularly important for delineating the
break points from which a new polygonal area is produced: a lower threshold will produce
many more polygons. The region size determines the minimum size of the resultant
polygons. Prior to conducting the segmentation the NIR band was removed from the Ikonos
MS imagery. This band was removed as it responds to vegetation rather than archaeological
residues.
The results of image segmentation were somewhat more encouraging than spectral
(supervised or unsupervised) classification. Although the results are quite complex to
interpret there was a relatively good correlation with archaeological residues. Unfortunately, it
appears that different segmentation parameters are required for each sub-zone in the marl.
Rather than establishing these parameters for each individual zonal image for each satellite (a
very time consuming process) it was decided to apply image segmentation to the 200m radius
mean image (see section 7.5.3.3). It was hoped that the effective ‘normalisation’ of this image
would negate the variable effects introduced by the different zones. A 100 block size, 60
spectral threshold and 250 region size were used for image segmentation on the Ikonos MS
200m mean image.
252
Figure 134 displays the results of this segmentation process. The correlation between residues
and segmented areas is very good. However, image segmentation gives no indication of
which discontiguous polygons may share levels of similarity. Therefore, each polygon needs
verifying individually (it is easiest to do this by selecting a ‘no colour’ fill and overlying the
polygons over the appropriate satellite image). Some areas can be removed automatically (for
example very large polygons that relate to the general background soil).
7.5.5 Evaluation of sensors in the marl zone
Both the Corona and Ikonos imagery are very effective mediums for residue detection in the
marl zone. Once again qualitative interpretation proved the best mechanism for detection. A
number of statistical and zonal approaches were tested which aided visual detection.
In general the Corona images provide a very good backdrop. The imagery is distinctly less
complex than the Ikonos. The lack of modifying processes in the landscape aids the
detection of residues (compare images A and B against C and D in Figure 119) as the
associated changes in reflectance are much easier to locate. All of the Corona missions were
useful for detection, though Corona mission 1110 produced the fewest potential residues. As
has already been discussed in section 0 the 1110 mission has poor image quality. In this case
it is the timing of the imagery which is more important. The Corona 1110 was collected on
28th May 1970 when cereal crop would have been at its most vigorous. Therefore, a
proportion of the landscape is masked by vegetation. Without a NIR band it is impossible to
quantify how much of the landscape this affects. Missions 1108 and 1111 were equally good
at detecting residues. At the time of collection (17th December 1969 and 31st June 1970
respectively) the marl area would have been effectively soil only. It is interesting to note that
the strip fields coming off wadi al-Rabaya have a higher reflectance in the December image
(probably due to ploughing or to the c. 18:30 collection time of the June image: see Figure
135). In the southern part of the southern marl zone archaeological residues were much
easier to detect using Corona than the Ikonos due to the clear textural component obvious in
the Corona imagery. This may have been lost to deeper ploughing in the intervening period
or alternatively irrigation may have reduced the contrast by equalising the soil moisture
content (see section 9.4.2). It must be remembered that the high quality Corona imagery
available for this area may not be available elsewhere.
253
Figure 135 Strip fields off wadi al-Rabaya are easier to identify in
the winter (mission 1108) as opposed to the spring (mission 1111)
image. Also note the different illumination of the foothills in the
SE.
The Ikonos pan and MS imagery are also good media for residue detection. However, the
MS imagery is a better resource for detection as the increased spectral resolution allowed
more confident interpretation. It was hoped that the high resolution Ikonos pan imagery
would aid in enhanced residue recognition or interpretation (such as the mapping of internal
structures) but this was not the case. Although it did allow the improved mapping of postdepositional modifications (such as bulldozing events or military entrenchments on the tells).
The red, green and blue bands tended to show an increase in DN values over archaeological
residues in the Ikonos MS. It was expected that the NIR band would be a significant
advantage for detection purposes. However, due to the delay in image collection, which took
place further into the growing season than planned, there was a significant amount of
vegetation across the landscape: this made this band very difficult to interpret. The NIR band
may still prove to be beneficial if the imagery is collected when there is minimum vegetation
cover. However, the NIR is important for detecting areas of vegetation masking. The MS
imagery is much better for identifying cloud cover than the panchromatic (see Figure 136).
Over the past 30 years landscape modifiers have affected micro and macro-scopic reflectance
characteristics making the Ikonos imagery more complex to interpret than Corona. This has
resulted in a greater number of potential sites identified from Ikonos imagery. Some of these
are negative sites that are artefacts of landscape processes (see Figure 177).
254
Figure 136 Problems of cloud and crop cover.
On both the Ikonos and Corona images most residues could be detected by employing
standard histogram techniques. Subdividing the imagery into different zones substantially
improved visualisation. Of more significance was the moving average kernel. This effectively
treated the image as a statistical surface and improved the contrast of the residues across the
whole image. Although this did not add many new residues it did improve the interpreter’s
confidence during detection. Quantitative classification techniques did not improve on visual
detection. However, image segmentation on the moving average kernel did correlate quite
well with the residues.
From a methodological viewpoint each image should be processed with the moving 200m
radius average kernel. The resultant images should then be visually interpreted. It is more
appropriate to examine the Corona imagery first as the residues are more prominent. The
255
Ikonos MS imagery can then be compared and analysed. Where ‘sites’ correlate there is a
strong likelihood that they are archaeological. Where they do not correlate the Ikonos
response may be due to some recent landscape modification. This will produce a number of
‘high potential’ and ‘potential’ archaeological residues.
7.6 Prospection in the alluvial zone (and floodplain)
The alluvial zone contains a range of archaeological residues and has been a continual focus
for anthropogenic activity. Of the 35 tells (including low tells) identified in the application
area 20 are located adjacent to the current course of the Orontes or Lake Qatina. Not only
does this indicate the importance of water for settlement location it also infers that the
current course of the Orontes has not significantly migrated in the past 8-10 millennia.
Consequently, this area has a complex natural and cultural formation and deformation
sequence.
The obvious proximity of the alluvial zone to the Orontes creates a number of significant
problems in the applications of satellite imagery for residue detection. With the exception of
tell sites, where most can be readily located due to their topographic component, the
identification of discrete areas of activity is complicated by near consistent crop cover. In
Figure 137 it is much easier to discriminate between the different zones in Corona (the new
road and the extension of irrigation have blurred the zonal boundaries in the Ikonos image).
The Corona image also indicates the preferential use of the floodplain for agriculture use
prior to extensive irrigation. In the Ikonos imagery the crop cover (denoted as a redder
colouration) and to a lesser extent the road, make it impossible to locate site 483 (Arjoune),
whereas it is an obvious feature in the Corona imagery. Furthermore, alluviation (overbank
silting) is likely to mask any deposits on the lower terraces.
With the exception of the larger sites (such as Arjoune), no potential sites were detected from
the satellite imagery even though a range of quantitative and qualitative approaches were
applied. At best only broad areas of potential activity could be inferred. In this situation the
Corona imagery proved to be a more useful resource as there was less landscape modification
than in the Ikonos imagery. Small scale fieldwalking and driving survey identified a range of
residues. However, given the importance of this zone for agricultural purposes and the
complexity of the landscape modifiers it was difficult to define many of these residues as
sites. Therefore, it is recommended that due to the somewhat arbitrary nature of the
256
detection process (the problems associated with masking from alluviation, irrigation and
crop) and that as the alluvial zone only accounts for c. 20 km2 that a programme of intensive
fieldwalking is conducted in this area. These results can then be reviewed in conjunction with
the satellite imagery to ascertain, retrospectively, if there is a correlation between sites and
reflectance.
Figure 137 Comparison of Corona and Ikonos in the alluvial zone.
257
Figure 138 Evidence from site 494 including a tenuous linear
feature in the foundations of a reservoir.
258
The area west of the Orontes and south west of Lake Qatina was originally interpreted as a
floodplain created during river migration. Again a range of quantitative and qualitative
detection techniques were applied in this area with limited success. Overall it demonstrated a
negative response for archaeological residues as only 8 sites have been identified (only 3 of
these with any high degree of certainty and each was impacted by bulldozing). Fieldwork in
this area did produce material. However, due to the paucity of residues and surface material,
any part of the landscape displaying only a small number (<10) of clustered artefacts was
identified as a potential site. Coins and a gaming piece were handed over by a farmer who
discovered them while excavating the foundations for a reservoir (site 494). These objects
were provisionally dated as Hellenistic. Further fieldwalking produced a handful of sherds.
Hence, it was assumed that, unless these are stray artefacts, any associated settlement is
masked beneath an unknown depth of alluvium. The possible identification of a ditch (see
Figure 138) supports this assumption. From this evidence it was extrapolated that the
floodplain was prone to extensive alluviation over the past millennia masking the majority of
archaeological residues in the area.
However, if this was the case then all the archaeological residues in the alluvial zone and
those west of Orontes would be masked by the same sediments. This is not the case, as sites
displaying prehistoric activity (such as Arjoune) are still surficial. After extensive consultation
with the geomorphologists Drs David Bridgland and Rob Westaway (University of Durham
and the Open University respectively) it is felt more likely that the sediments in this zone are
related to sheet wash or an alluvial fan extending from the foothills of the Anti-Lebanon
mountain range. The exact geomorphological origin of this zone is still under investigation.
Although this zone has a diverse range of local resources and would therefore be attractive
for settlement, it is likely that residues are masked by a significant depth of sediments.
Successful detection with satellite imagery is therefore highly unlikely. However, if there are
appropriate negative archaeological features it may be possible to detect them as crop marks.
7.7 Discussion
Multi-resolution satellite imagery has been employed as an archaeological residue detection
tool in three contrasting environments. With the exception of the alluvial zone the
application of satellite imagery was an unqualified success. However, different sensor
259
resolving characteristics and interpretative techniques are required for the each
environmental zone. These are discussed in this chapter and summarised in Chapter 10.
To date with the exception of extrapolated morphological evidence from extensive
monumental sites the imagery gives little indication of the form (the current workable
classification covers large site (possibly tell) to small site (probably flat)), function and dates of an
archaeological site. It is also unlikely that the site constituents identified from satellite imagery
will allow a high degree of temporal identification. Therefore, satellite imagery should only be
used in conjunction with other Cultural Resource Management (CRM) techniques.
The use of other techniques (such as surface survey or even basic ground observation) are
essential to enhance the resource. Thus, the data sets can be used to analyse and hypothesise
about the archaeological resource and to produce a feedback loop allowing increased
accuracy of the identification procedure. This will allow improved interpretation, particularly
for sites not initially identified from the imagery, and an understanding of the scale threshold
with which the imagery can be confidently interpreted.
260
CHAPTER 8 ANALYSIS OF MARL SOILS
8.1 Introduction
Low mounds with phosphate-rich, artefact-rich or ashy sediments will often appear quite
different in colour than their surroundings.
(Banning 2002 p. 42)
Chapter 7 demonstrated that archaeological residues can be identified from satellite imagery.
It was proposed that the soil in the marl area be examined in an initial attempt to determine
the physical cause for the localised increases in reflectance associated with ‘sites’. This should
provide a platform for future research and analyses in similar environments. The sites that
have been recognised fall into two main categories: flat sites and tells. Tells are artificial
mounds that have developed though long-term re-building employing degradable building
material (Rosen 1986; Stevanovic 1997). The distinct topography of tells makes them easy to
identify both from the ground and from a vertical perspective (see Figure 139).
Figure 139 Ikonos image (vertical view) and photograph
(horizontal view) of a tell (site 191). Note the increased and
decreased reflectance of the SE and NW part of the ramparts due
to differential illumination from the Sun and site topography.
As has been demonstrated in Chapter 7, flat sites do not exhibit such extreme topographic
variations and are located by the detection of anthropogenically formed sediments and other
physical residues (Schuldenrein 1995). Here, it is assumed that anthropogenic action has
261
caused the colour change and not that lighter coloured soils were preferentially chosen for
settlement locations. The question is what are the anthropogenically introduced physical or
chemical constituents that have produced the observable variations in the colour of the soil
matrix?
Soil colour is one of the most obvious and easily determined attributes of soil, and is a
primary element widely used by soil scientists in the identification, field description,
characterisation and classification of soils (White 1997). Soil colour is almost entirely an
indirect measure of other more important characteristics or qualities that are not so easily
observed. Surface colour that differs from that of the parent material is usually an indication
of the processes involved in soil formation and may also be indicative of anthropogenic
actions which have disturbed a localised soil matrix. Myers (1983) and Horvath et al. (1984)
state that the most important factors influencing soil colour are mineralogy and chemical
constituents, soil moisture, soil structure, particle size and organic matter content.
In the field, a pedologist determines soil colour by a direct comparison between a soil sample
and Munsell colour chips (as discussed in section 5.2). The Munsell colour system utilises the
descriptive system of hue, value and chroma (Munsell 1981). Hue is the dominant spectral
colour and corresponds to the wavelength regions used in optical remote sensing. Value
refers to the relative brightness of the colour and is a function of the total amount of light
reflected. Chroma is the relative purity of strength of the spectral colour. The solar radiation
reflected from the surface microns of soils can also be observed by remote sensing
instruments that offer measurement capabilities beyond those of the naked eye. Obukhov
and Orlov (1964, cited in Ben-Dor et al. 1999a) concluded that the red region and the near
infrared region are the most favourable for a qualitative and quantitative description of soils.
Fernandez et al. (1988) calculated soil colour from DN values and reported increased
accuracy and precision from satellite sensors in comparison to visual interpretation, thus
making it possible to quantify small differences in soil colour that would otherwise be
difficult to assess. An understanding of how reflectance changes as soil properties alter is
required to infer soil properties and processes from the observed reflectance (Irons et al.
1989; Mattikalli 1997).
In order to reduce complexity during interpretation it was assumed that the soils were stable
during periods of occupation and that subsequent post depositional soil formation or
262
deformation has not significantly impacted on the original sediments (Wilkinson, pers.
comm.). Further, it was hypothesised on the basis of known patterns of human settlement
loci that the increase in soil reflectance observed at flat sites was due to either one or a
combination of the following reasons:
•
Variation in soil moisture between on and off-site soils.
•
Changes in soil particle size distribution between on and off site soils (see section
8.2.3).
•
Different levels of organic matter content.
•
Different levels of trace elements or chemical compounds on sites.
•
Increased ash content on sites.
An understanding of soil formation processes and how the above changes may affect the
reflectance of electromagnetic energy was required.
8.2 Pedology
Pedology is a study of soils as three dimensional natural bodies resulting from weathering
processes. The processes are conditioned by climate, biota and topography, acting on parent
material over time. Pedology includes the description, quantitative characterisation,
classification and mapping of soils (White 1997). Since the characteristics of radiation
reflected from a material are a function of material properties, observations of soil reflectance
can provide information on the properties and state of soils. These properties can be
quantified using a variety of laboratory techniques (Olson 1984; Goldberg 1992). The most
familiar application of this concept is the observation of soil colour to describe and help
classify soils (Irons et al. 1989; Curran 1990).
Soil maps typically divide the landscape into discrete units with distinct boundaries. Although
the delineation of mapping units is useful for descriptive purposes, abrupt soil boundaries are
rare, due to the nature of the soil-forming processes (as observed when producing the
soil/geology map as described in section 6.6). More often, a gradual transition in soil
properties and profiles occurs over the landscape. This variation is best observed from
remote platforms, which can afford a synoptic perspective of a geographic area. Remotely
263
acquired images show variations in surface reflectance, which can indicate corresponding
variations in underlying soil profiles.
Figure 140 Soil profile (after Irons et al. 1989 p. 77).
Soil profiles continue to develop and change over the years, but the changes are generally
gradual. The sequence and character of horizons in a profile (see Figure 140) often persist for
decades or even centuries. This variation in the nature of soils is a common criticism of
predictive modelling techniques that assume that present day soil variables adequately
represent past values. Soil surface roughness, structure, moisture content and other
properties are readily altered by weather, cultivation, erosion, plant growth and other surface
phenomena. These temporal variations affect the surface reflectance properties. Changes in
reflectance properties can complicate the use of remote sensing for soil mapping and
characterisation. Alternatively, observations of reflectance variation over time can also
indicate characteristics of the underlying soil properties. Thus, for certain applications an
appreciation of spatial and temporal soil variability is needed to understand soil reflectance
and the application of remote sensing to pedology (Irons et al. 1989).
264
8.2.1 Soil biochemistry
Soil may be thought of as a chemical mixture formed en route from rock to infinitely diluted
ions. Inevitably its composition is variable, representing various stages of weathering in
different environments. Furthermore, microbiological action takes place within the chemical
system altering its chemistry by modifying the inorganic balance and generating complex
organic components. More advanced life forms, by their growth and decay and the
circulation of air and water, further modify the soil structure. It should be noted that at the
surface layer the amount of carbon and hydrogen increase significantly due to the presence of
organic matter. Soil water within this medium is the solute for a variety of ions, some of
which are essential for plant growth.
8.2.2 Organic matter
From the moment death occurs organic matter begins to decompose. Almost instantaneously
microbial organisms attack tissues. The majority of the initial decomposition of large objects
is carried out by mammals, termites and earthworms (Stein 1992). Many plants and
organisms obtain their energy from this partially decomposed matter and in turn reduce the
organic compounds into residues and humus. As organic matter becomes more finely divided
the size of the decomposing organisms decrease. All decomposition is affected by
environmental factors, especially variations in temperature, moisture and available oxygen.
Organic soil matter consists of decomposing plant material, animal material and substances
derived from the products of decomposition by micro-organisms and small animals in the
soil. The elements of organic matter are joined in various compounds such as proteins,
glucose, carbohydrates, fats, tannins and lignin. These are readily decomposed by soil microorganisms. The products of decomposition form complex mixtures of brown or dark brown
amorphous and colloidal substances called humus (Brady 1984). Humus, composed of
humin, humic acid and fulvic acid, constitutes approximately 65 to 75 per cent of the organic
matter in mineral soils and can occur as a discreet substance in soil or as coatings on mineral
particles. It can also act as a binder between particles in aggregates. Those elements of
organic matter not degraded to humus are referred to as non-humic. Some non-humic
substances are still recognisable as physical or chemical components of plant or animal tissue
and include proteins, peptides, amino acids, fats, waxes and organic acids (Schnitzer 1982).
265
At every stage of this reduction, carbon dioxide gas is released and escapes from the soil.
Moulds and spore-forming bacteria are especially active in consuming the proteins, starches
and cellulose of fresh organic matter and they release carbon dioxide, water, ammonia,
hydrogen sulphide, sulphur dioxide and organic acids as by-products. Further reduction by
other micro-organisms results in the creation of humus and the further release of carbon
dioxide. Carbon, released as carbon dioxide, is transferred from the soil to the atmosphere or
dissolved in soil water to produce carbonic acid which lowers ambient pH. In temperate
regions half of the organic carbon produced during soil genesis is lost to the atmosphere in
the first 3-4 months and in the tropics half is lost in only 3-4 weeks. Thus, the organic carbon
content of the soil organic matter, or organic sediment, is lowered through decomposition.
In the context of sediments, deposition of organic matter is a single event. The organic
matter will support a micro-organism population that is related to the amount of organic
matter originally deposited. Lacking a continuing source of organic matter, a steady state is
not reached. The micro-organism population will be supported until the original organic
sediments decompose (i.e. until all the carbon dioxide is lost to the atmosphere and the other
by-products reduced to humus). As decomposition occurs, the micro-organism population
will change. If the process occurs for millennia the organic matter will be reduced to only the
most resistant humus. On the other hand, if the organic sediment is augmented by repeated
deposition, then the rate of decay and the size of the micro-organism population will depend
on the conditions within the buried deposits. Decomposition is slowed when oxygen or
water is not available (or available in excess), or in places experiencing extreme acidity or
cold. In these circumstances animal and plant material can survive for millennia and in certain
cases (i.e. desert, bog and frozen sites) can be almost perfectly preserved (Stein 1992).
8.2.3 Soil particle size
Soil is an extremely complex physical and biological medium. Soils consist of material in the
three common physical phases: solid, liquid (water) and gas (mostly air). The solid particles
are of different sizes. The grain size distribution is the principal factor which governs the
majority of the physical and mechanical properties with the exception of magnetic anomalies
(Scollar 1990 p. 9).
266
Figure 141 Soil description matrix (after Asrar 1989 p. 75) and
archaeological field soil description chart (after Middleton 2000).
267
The solid phase consists of irregular particles. Particle size is determined by the minimum
size of sieve through which the particle will pass or by the diffraction of light through a
suspension of material when using laser granulometry (i.e. Coulter). The combinations of
these fractions give the general soil classification. Archaeologists coarsely gauge soil fractions
during fieldwork by rubbing damp soil between a thumb and forefinger. The rubbing breaks
down aggregates into primary particles and the feel, or texture of the broken-down soil is
determined to a large degree by the particle size distribution of the soil minerals (see Figure
141).
The continuum of particle size in a soil sample can range over three to four orders of
magnitude. For ease of interpretation this large range is divided into different particle size
fractions. Different soil classification schemas use different ranges for grouping particle size
fractions (such as the British Soil Survey and Wentworth/Udden classifications). Texture
descriptions are further facilitated by the definition of texture classes on the basis of relative
portions of sand, silt and clay sized particles (Olson 1984; Asrar 1989; White 1997).
However, this classification technique does not take into account any organic material that
may exist in the soil.
8.3 Soil spectral reflectance
Analysis of remotely sensed data involves identifying features and correlating ground-based
measurements with recorded reflectance or emmittance values. In the case of soils, these
ground-based measurements include several properties such as texture, composition, grain
size, soil moisture and colour. Earlier studies have identified significant relationships between
some of these properties and spectral reflectance of soil in the visible and infra-red portions
of the electromagnetic spectrum (Irons et al. 1989; Ben-Dor et al. 1999a).
The observation and mapping of soil conditions through optical remote sensing is restricted
to reflectance of the surface microns (Leone and Escadafal 2001). Because the characteristics
of radiation reflected from a material are a function of its physical and chemical properties,
the observation of surface reflectance principally carries information on the properties and
the state of the topsoil. This means that only effects which cause significant changes to
surface characteristics can be observed and mapped. The spectral reflectance of soils is a
cumulative property which derives from the inherent spectral behaviour of heterogeneous
268
combinations of minerals (and their textural components), organic matter and soil water
(Baumgardner et al. 1985; Irons et al. 1989; Sommer et al. 1998 p. 198).
A substantial number of researchers have measured spectral reflectance factors from soil
samples in the laboratory and from soil surfaces in the field. Perhaps the most
comprehensive study of soil reflectance in different regimes was conducted by Stoner and
Baumgardner (1981) in the United States. Using laboratory techniques they measured spectral
bidirectional reflectance factors, over the 0.52 - 2.32 µm wavelength region, from samples of
over 240 soil series. The samples were selected by a stratified sampling strategy of soil series
within 17 temperature and moisture regimes across the United States. Several samples of
tropical soils from Brazil were included.
Figure 142 Bidirectional reflectance of soils (after Asrar 1989 p.
89).
Stoner and Baumgardner (1981) defined five spectral curves that were characteristic of the
observed soil reflectance spectra. In other words, they believed each of the observed spectra
resembled one of the five curves shown in Figure 142, the resemblance being determined by
soil properties, especially organic matter and iron-oxide content. The discriminating features
of the five curves are shape and the presence or absence of spectral absorption bands.
269
8.3.1 Organic matter response
Despite the generally low content of organic matter in most soils it exerts a profound
influence on soil properties such as structure, fertility, water holding capacity and, in
particular, reflectance. The influence is highly dependent on climate and environment (Irons
et al. 1989). Under different climatic conditions the organic matter constituents will vary
significantly. Soils developed in semi-arid grassland environments contain abundant humus,
which imparts a very dark pigmentation to the soils. That pigmentation is less intense in soils
of humid temperate regions and is least apparent in the soils of the tropics and semi-tropics.
UnitID
Unit Type
6 Structures
14 Tell
83 Tell
90 Tell
97 Tell
184 Scatter
191 Tell
208 Tell
210 Tell
222 Scatter
245 Tell
252 Scatter
253 Scatter
268 Tell
279 Scatter
280 Scatter
283 Scatter
318 Scatter
348 Indeterminate
351 Indeterminate
496 Scatter
498 Scatter
508 Scatter
521 Scatter
Solid Geology
Basalt
Marl
Limestone
Limestone
Limestone
Marl
Marl
Marl
Marl
Marl
Marl
Marl
Marl
Limestone
Marl
Marl
Marl
Marl
Marl
Marl
Marl
Marl
Limestone
Marl
Soil Colour Wet Soil Colour Dry Off-Soil Colour Wet Off-Soil Colour Dry
L2
N13
K2
N14
I19
G12
K18
L14
N14
B11
L19
L19
N13
C11
M19
N14
N13
C11
N13
N13
M14
K13
M15
M14
N13
C11
M14
L14
K2
C11
L19
G14
M13
C11
A10
B10
N13
B11
M14
G13
A11
C11
K19
J18
M13
L13
K18
J15
M14
L13
L14
K13
L19
B11
N13
M14
L14
G13
M14
L14
M14
L13
K14
L13
M14
G13
M14
G13
L13
G12
M14
L14
M14
L13
M15
M14
M14
K13
L17
K15
N13
M13
K19
A11
M13
G12
A10
B10
M14
L13
M14
L13
M13
C11
J19
L14
Table 19 Comparison of on and off site soil colours when wet and
dry. Note this requires the colour chips described by Middleton
(2000).
Kristof and Zachary (1971, cited in Myers 1983) and Baumgardner (1970, cited in Myers
1983) classified multispectral data to delineate organic matter content in soil classes (see
Figure 142). They generated five different ranges of organic matter content for mineral soils
containing from 1.5 to 7 per cent organic matter. Da Costa (1979, cited in Ben-Dor et al.
1999a), showed that colour values can be used to estimate organic matter content from soil,
but that this correlation is more pronounced in some soils than in others (Myers 1983). The
strongest correlations between organic matter and reflectance are usually observed in the
visible spectral region (see spectra E in Figure 142), whereas the relationships between
reflectance and the clay fraction is better observed in the short wave infrared ((SWIR) Ben270
Dor and Banin 1995; Ben-Dor et al. 1997; Galvao et al. 2001). Spectral reflectance generally
decreases over the entire SWIR region as organic matter content increases. At organic matter
contents greater than 2%, the reflectance decrease may mask other absorption features in soil
spectra. The spectra of soils with organic matter content greater than 5% often have a
concave shape between 0. 5 and 1 .3 µm (see Figure 142, curve E) as compared to the convex
shape of spectra for soils with lower organic matter content (Stoner and Baumgardner 1981).
Figure 143 Soil reflectance for a silt loam soil with varying
moisture content (after Asrar 1989 p. 90)
8.3.2 Moisture response
Increasing moisture content generally decreases soil reflectance across the entire visible and
SWIR spectrum (Asrar 1989). Wet soils usually appear darker to the eye than dry soils for this
reason. The decrease in reflectance in the visible spectrum has been attributed to internal
reflections within the film of water covering soil surfaces and particles (see Figure 143). For
this reason colours were taken from both ambient and wet site and off-site soils to help
determine image acquisition times (as discussed in section 5.2.4: see Table 19).
271
8.3.3 Iron and Iron-Oxide response
Iron commonly occurs as a principal constituent of some soil minerals and as ions in soil
water. Many of the absorption features in soil reflectance spectra are due to the presence of
iron in some form. The absorption features are caused by either crystal field effects or charge
transfers involving Fe2+ or Fe3+ ions. For example, the steep decrease in reflectance toward
the blue and ultraviolet wavelengths is a characteristic of almost all soil reflectance spectra
(Figure 142). Increasing the proportion of iron content results in increased levels of energy
absorption in the visible wavelengths and decreases in the reflectance curve (compare curves
B and C in Figure 142). Curve A represents the spectra of soils with a high iron-oxide
content and display generally low reflectance (Stoner and Baumgardner 1981).
8.3.4 Particle size response
Soil reflectance is not only influenced by the chemical composition of the soil constituents,
but also by the size and arrangement of the soil particles. Soil texture refers to the size
distribution of soil mineral particles. The physical arrangement and aggregation of these
particles provide a soil with structure. Texture and structure determine the amount of pore
space available in a soil for occupation by water and air. Soils of similar mineral composition
and texture can have distinctly different structures and hence porosity.
Bidirectional reflectance generally increases and absorption features decrease as particle size
decreases. This behaviour is characteristic of transparent materials, and most silicate minerals
behave transparently in the shortwave region. In contrast, the bidirectional reflectance of
opaque materials decrease as particle size decreases (Ben-Dor et al. 1999a).
Soil texture is fundamental to understanding the characteristics of a soil. The mineralogy of
the inorganic solids, for example, is related to particle size. Sandy soils contain a high
proportion of quartz and other primary minerals, whereas clay rich soils contain a higher
proportion of secondary minerals. In particular, sandy soils tend to be brighter than clay rich
soils. The difference may be explained in part by the different mineralogies, and hence
reflectance properties, of clay and sand particles, but it may also be due to the tendency of
clay particles to aggregate. This aggregation into colloids larger than sand grains can
contribute to the darker appearance of clay soils. Additionally the presence of rock fragments
(i.e. particles greater than 2 mm in size) modifies texture and hence affects the physical
properties and reflectance of soils. Since the size distribution of the mineral particles is not
272
generally subject to rapid alterations in topsoils, texture is considered a relatively constant soil
property and its determination is a basic requirement for any soil description (Irons et al.
1989).
Decrease in particle size has been seen to increase soil reflectance among sandy textured
soils, possibly by forming a smoother surface with fewer voids to trap incoming light.
However, the inverse appears to be true with medium to fine textured soils. This is possibly
because the increased moisture and organic matter content associated with clay rich soils
leads to lower reflectance.
8.3.5 Theoretical reasons for changes in reflectance
The relationships between soil constituents and in-situ or laboratory spectral reflectance has
been discussed by several researchers who have studied the spectral effects of soil parameters
such as moisture, organic matter, iron oxide and particle size.
Moisture reduces overall soil reflectance and produces strong and broad absorption bands at
1.4 and 1.9 µm that affect the shape of a soil spectrum (see Figure 143). Soil colour can
provide extremely valuable insights into the hydrological regime or drainage status of the soil.
Bright (high chroma) colours throughout the profile are typical of well-drained soils through
which water easily passes, and in which air is generally plentiful. The presence of grey, low
chroma colours, either alone or mixed in a mottled pattern with brighter colours, is indicative
of waterlogged conditions during at least a major part of the plant growing season. Colour
can also provide qualitative information about the current moisture status of the soil, dry
soils generally having lighter (higher value) colours than moist soils.
Organic matter has a profound influence on soil colour. In general, decreases in organic
matter produce an increase in reflectance. Increases in iron oxides tend to decrease soil
reflectance. In the optical region a reduction in particle size tends to increase reflectance.
The archaeological residues produce a distinctly higher reflectance than the background soil
colour. Therefore, theoretically this increased reflectance is related to one or a combination
of the following phenomena:
•
Low organic matter content.
273
•
Low iron oxide content.
•
Relatively finer textured particles or differences in soil structure.
•
Improved drainage reducing soil moisture.
8.4 Soil texture analysis
In the 2001 fieldwork season, Dr. Keith Wilkinson (King Alfred’s College, Winchester)
collected 68 soil samples from various locations in the marl zone in order to examine
variations in present day soil properties from on and off-site locations (Wilkinson, pers.
comm.). These samples were taken from consolidated material below the ploughsoil (c. 1015cm below topsoil) and the location of each sample was recorded using handheld GPS.
These samples were analysed for trace metals (copper, iron, manganese, potassium and
sodium), magnetic susceptibility and moisture using standard methodologies within King
Alfred’s College (Wilkinson, pers. comm.). Digital number (DN) values were taken from the
pixels surrounding each sample location on the Ikonos multispectral and Corona mission
1111 imagery. As there may be errors associated with the spatial location of the imagery the
values for the Ikonos imagery were an average of the four closest pixels and for the Corona
imagery the nine closest pixels were used. It should be noted that there may be difficulty in
the quantitative analysis of the Corona imagery due to variations in view angle and that
different filters were employed on different missions.
On completion of the analysis outlined above the soil samples were returned to Durham
where particle size analysis and reflectance measurements were undertaken within the
Department of Geography. Particle size was determined using a Coulter LS230 by
Galiatsatos. The raw particle size readings could be generalised into the British Soil Survey
classification through the use of a look-up table (see Table 20). It should be recognised that
the determination of particle size using laboratory techniques requires the disaggregation of
the particles (i.e. the reduction of aggregated lumps to their constituent particles).
Furthermore, Coulter analysis is only conducted on particles with sizes of less than 2mm.
Therefore this does not measure in-situ particle size which is the component that directly
affects reflectance. However, it is assumed that there is a positive relationship between the
laboratory measurements and field sample behaviour. Laboratory based reflectance
measurements were undertaken by Galiatsatos using a GER 1500 handheld spectroradiometer. This instrument measures reflectance between 0.3 and 1.1 µm which means that
274
the moisture absorption bands at 1.4 and 1.9 µm could not be examined. Four measurements
were taken as follows:
•
A sample in a Petrie dish.
•
A sample in an aluminium tin.
•
A sample dried overnight at 100º in a Petrie dish.
•
A sample dried overnight at 100º in an aluminium tin.
These measurements correspond to ambient and dry conditions. Different containers were
used to see if they impacted on the reflectance measurements.
Organic matter and moisture readings were not taken as the timeframe between collecting
the samples and their arrival in the UK meant that this component would be severely
compromised. It is recommended that field based recording of organic matter, particle size
and moisture values are conducted in the future. However, for future research it may be
possible to evaluate organic matter and soil moisture by reference to the satellite imagery and
the spectro-radiometry profiles.
Fraction Size
<2µm
>2 and <6µm
>6 and <20µm
>20 and <60µm
>60 and <200µm
>200 and <600µm
>600 and <2000µm (2mm)
>2000 and <6000µm
BSS Particle Classification
Clay
Fine Silt
Medium Silt
Coarse Silt
Fine Sand
Medium Sand
Coarse Sand
Fine Gravel
Generalised Classification
Clay
Silt
Silt
Silt
Sand
Sand
Sand
Stone
Table 20 British Soil Survey particle size classification
Hence a range of different analytical results were collected for each soil sample. The fact that
these samples have their transect location plotted by GPS means that the results can be
analysed in conjunction with the satellite imagery. This allows a number of comparisons to
occur:
•
The spectro-radiometry results can be compared directly with sensor reflectance.
•
Sensor reflectance can be simulated directly from the spectro-radiometry.
275
•
Trace metals can be analysed to see if chemical variations impact upon spectroradiometry and reflectance readings.
•
Grain size distributions can be analysed to see if they impact upon spectroradiometry and reflectance readings.
There are further scale implications in this analysis: the samples represent results from a 3
dimensional point, whereas the satellite pixel represents a generalised average of a 2
dimensional surface.
The results of these analyses (discussed in section 8.4.1) were very encouraging. During the
2003 season a further 122 soil samples were taken across representative features in the marl
zone (and one further in the alluvial zone) for particle size analysis.
8.4.1 The 2001 soil sampling programme
Of the samples collected by Wilkinson, four groups formed transects across sites (259, 279,
339 and 602: see Figure 144). Each of these transects was analysed individually to determine
if any of the variables showed a correlation with reflectance variations from the background
soil to the site. Each of the samples was given a locational value depending on where they
were in relation to the site (pre-site (off-site at the start of the transect), pre-transition (the
boundary between pre-site and site), site, post-transition and post-site). Even though these
samples were collected from below the topsoil layer it was felt that the results would correlate
well with topsoil samples due to the shallow topsoil. If the results showed promise then
further samples could be collected from the topsoil.
The data for each site is presented in three figures denoting:
•
Site location, transect position and particle size response.
•
Grouped (site, transition and off-site ) spectro-radiometry curves and simulated
Ikonos MS readings across the transect.
•
A summary of sensor reflectance and the laboratory analyses conducted by
Wilkinson.
276
Figure 144 Location of sites where soil samples were taken by
Wilkinson.
277
Figure 145 Locations of soil samples collected by Wilkinson over
site 259 and results of particle size analysis (generalised to the
British Soil Survey classification).
278
Figure 146 Averaged spectral curve and simulated Ikonos MS
readings from the spectro-radiometer readings of the soil samples
from site 259.
279
8.4.1.1 Analysis at site 259
Site 259 is a flat site west of a wadi in the marl zone (as defined in section 6.6) with a local
place name of 'Um Al-Sakhr. During the last visit the site was under olive plantation and
various cereals. Surface material included pottery and architectural fragments. A significant
portion of the eastern edge of the site has been bulldozed revealing interleaved conglomerate
and marl sediments. The sample transect was located across the site in a WSW-ENE
direction and sampling took place at approximately 20 metre intervals (see Figure 145).
Unfortunately the transect does not exit the site on the eastern edge due to the bulldozing,
thus the transect is considered completed at sample point 26. Sample points 19 to 20
highlight the transition between off-site and site respectively. The site boundary has been
recorded from satellite imagery and not from field observation.
As shown in Figure 145, Figure 146 and Figure 147, there is an increase in DN values and
reflectance readings from the MS Ikonos sensor between off-site and site soils and a
corresponding decrease on the Corona imagery (note that the Corona is digitised from a
negative so the decrease is actually an increase. Hence the Corona and Ikonos sensors
represent the same pattern). This curve is also seen in the simulated Ikonos MS using the labbased radiometry data. The general trends between the simulated and real Ikonos readings
compare favourably, including the dip between samples 21 and 23.
Both the low and high frequency magnetic susceptibility readings show a reduction in value
at the site boundary. These data should indicate that on-site soils show less anthropogenic
activity than off-site soils. This is an unexpected result requiring more research. Moisture
contents shows a broad correlation with the site. Trace copper and sodium do not correlate.
There is a possible correlation with trace manganese, potassium and iron and a large increase
in the percentage of clay on the site and a large decrease in coarse and medium silt. There is
also a slight, but observable, increase in fine silt.
The spectral data show an increase in reflectance from pre, through transition to site soils.
After comparison with Figure 142 it can be extrapolated that the soil has low organic matter
content (< 2%; see curves B, C and D). In conclusion the increase in the spectral response is
most likely to have occurred through changes in particle size, although moisture may play
some part.
280
8.4.1.2 Analysis at site 279
Site 279 is a flat site east of a wadi in the thin marl zone (as defined in section 6.6) with a local
place name of Khirba (ruin) Al-Qatisiyya. When last visited the site was under a combination
of wheat and potato cultivation. Surface material included pottery, basalt and numerous
architectural fragments (which provide a large textural change across the site). A small
amount of bulldozing has occurred on this site. The transect was located across the site in an
ENE-WSW direction. Samples were taken at c. 20 metre intervals (see Figure 149). Sample
points 3 to 5 and 11 to 13 highlight the transition between off-site and site. This is one of the
few sites in the whole application area where the satellite imagery actually shows a decrease in
DN values (however, in dry conditions on the ground the site soil appears lighter). The site
boundary has been recorded from the satellite imagery.
Figure 147 Analysis of the soil samples from site 259 and the DN
value from the Corona and Ikonos MS satellite sensors for the
corresponding GPS point.
As shown in Figure 148, Figure 149 and Figure 150 there is a decrease in DN values and
reflectance readings from the Ikonos sensor on site (which are more pronounced in the blue
and near infrared bands) and a corresponding increase on the Corona imagery. Although
slight, this is reflected in the simulated Ikonos MS in the red and near infrared bands. The
281
simulated Ikonos displays an increase in the blue and green bands (which tallies with the
ground observation).
Both the low and high frequency magnetic susceptibility data produce an increase in value
over the site, indicating anthropogenic activity. Moisture values decrease strongly within the
site. This is a particularly interesting result considering that there is no concomitant increase
in reflectance. None of the trace metals indicate any correlation with the site extent. There is
a large decrease in the percentage of clay and fine silt and a large increase in coarse sand.
The spectral curves show a complex pattern across the different parts of the transect. The
site shows the highest reflectance between 400 and 600 nm (blue and green). The site shows
the lowest reflectance value between 600 and 800 nm (red and near infrared). The transition
soils are particularly difficult to classify and indicates soil complexity at the site boundary.
After comparison with Figure 142 it can be extrapolated that the soil has low organic matter
content (< 2%; see curves B, C and D). In conclusion, the increase in reflectivity is most
likely to have occurred through changes in particle size. The heterogeneous frequency of
massive clasts (architectural fragments) must play some part in the variation in texture across
the site, further emphasising the lack of clarity at the site boundary.
Figure 148 Analysis of the soil samples from site 279 and the DN
value from the Corona and Ikonos MS satellite sensors for the
corresponding GPS point.
282
Figure 149 Locations of soil samples collected by Wilkinson over
site 279 and results of particle size analysis (generalised to the
British Soil Survey classification) on these samples.
283
Figure 150 Averaged spectral curve and simulated Ikonos MS
readings from the spectro-radiometer readings of the soil samples
from site 259.
284
8.4.1.3 Analysis at site 339
Site 339 is a flat site in the marl zone (as defined in section 6.6) with a local place name of
Khirba (ruin) Al-Tahisah. This is one of a number of sites which is associated with a large
topographic depression. In this instance the depression is located in the centre of the site.
When last visited the site was under cereal cultivation. Surface material included pottery and
architectural fragments (including tile). The transect was located across the site in a SW-NE
direction with samples taken at approximately 20 metre intervals (see Figure 151). Sample
points 57 to 59 and 63 to 64 highlight the transition between off-site and site. The site
boundary has been recorded from satellite imagery.
As shown in Figure 151, Figure 152 and Figure 153 there is an increase in DN values and
reflectance readings from the Ikonos sensor on site (note the peak in all the bands between
samples 61 and 62). There is also a corresponding decrease on the Corona imagery (the peak
at sample 64 on the Corona is possibly due to interference from the road). These results are
also seen in the simulated Ikonos MS.
Both low and high frequency magnetic susceptibility data produce an observable decrease in
value over the site. Moisture values decrease strongly within the site. In this instance a
reflectance increase is associated with the moisture decrease. None of the trace metals
indicate any correlation with the site extent. There is a large increase in the percentage of clay
(with the exception of the initial anomaly) and sand and a decrease in silt.
As seen at site 259, the spectral data for this site show an increase in reflectance from off-site,
through transition to site soils. After comparison with Figure 142 it can be extrapolated that
the soil has low organic matter content (< 2%; see curves B, C and D). In conclusion the
increase in spectral response is most likely to have occurred through changes in particle size
and decrease in moisture content.
285
Figure 151 Locations of soil samples collected by Wilkinson over
site 339 and results of particle size analysis (generalised to the
British Soil Survey classification) on these samples.
286
Figure 152 Averaged spectral curve and simulated Ikonos MS
readings from the spectro-radiometer readings of the soil samples
from site 339.
287
Figure 153 Analysis of the soil samples from site 339 and the DN
value from the Corona and Ikonos MS satellite sensors for the
corresponding GPS point.
8.4.1.4 Analysis at site 602
Site 602 is a flat site in the lee of tell 254 in the thin marl zone (as defined in section 6.6) to
the west of a wadi. When last visited the site was under an olive plantation and various
cereals. Surface material included pottery (mostly Islamic), basalt and architectural fragments
(including tile). A trench (area 601 see Figure 155) has been cut into the western part of the
site. The transect was located across the site in a SW-NE direction with samples taken at
approximately 30 metre intervals (see Figure 155). Due to its proximity to tell 254 it is
difficult to define the exact limits of this site, the current boundary is extremely arbitrary.
Hence the transect was extended to ensure off-site coverage. From the satellite imagery it has
been extrapolated that the transition between off-site and site starts between samples 39 and
42 and finishes between samples 46 and 49. It should also be noted that sample 50 is taken
from the wadi margins.
As shown in Figure 154, Figure 155 and Figure 156 there is an increase in DN values and
reflectance readings from the Ikonos sensor on site and a corresponding decrease on the
Corona imagery. This, however, is unsurprising in this context given that the imagery was
used to define the site limits. The fact that this trend is reflected in the simulated Ikonos MS
288
is encouraging and it is recommended that intensive surface survey is conducted at this site to
determine the site extent by artefact fall-off.
Both low and high frequency magnetic susceptibility data produce an observable decrease in
value over the site. Moisture values do not seem to correlate with the site although there is an
increase in moisture content as the samples approach the wadi. None of the trace metals
indicate any correlation with the site extent. The SW edge of the site can be seen through the
increase in the clay component between samples 39 and 40. It is difficult to isolate trends in
the particle sizes and to locate the NE edge of the site, although there could be interference
at the NE edge due to the proximity of the wadi and its associated deposits (see Figure 155).
The spectral curves for this site show an increase in reflectance from off-site, through
transition to site soils. The on and off-site soils do appear to correlate well, although defining
a site edge in the transitional samples is difficult. This should be re-examined after the site
extent is re-surveyed. After comparison with Figure 142 it can be extrapolated that the soil
has low organic matter content (< 2%; see curves B, C and D). In conclusion, the increase in
spectral response is most likely to have occurred through changes in particle size.
Figure 154 Analysis of the soil samples from site 602 and the DN
value from the Corona and Ikonos MS satellite sensors for the
corresponding GPS point.
289
Figure 155 Locations of soil samples collected by Wilkinson over
site 602 and results of particle size analysis (generalised to the
British Soil Survey classification) on these samples.
290
Figure 156 Averaged spectral curve and simulated Ikonos MS
readings from the spectro-radiometer readings of the soil samples
from site 602.
291
8.4.1.5 Summary of the analysis of Wilkinson’s 2001 soil samples
After evaluation of the soil samples collected by Wilkinson the following results were
interpreted. Anthropogenic action may have altered the quantities of trace metals on sites
(see in particular site 259). However, in respect to their impact on reflectance the results are
inconclusive. However, lead, zinc and Al203 which produced positive results for Rimmington
(2000) in Greece were not analysed by Wilkinson. On sites 259, 279 and 339 moisture
content does appear to decrease. However, these results may be skewed by changes in
moisture content between sampling in the field, transit from Syria and laboratory analysis.
This is supported by the similarity of the ambient and dry spectral curves at each site. Both
low and high frequency magnetic susceptibility data appear to be related to archaeological
residues. There is a very strong correlation between these readings and sites 259 and 279 and
weak correlations on sites 339 and 602. It is interesting to note that only site 279 exhibits an
increase in magnetic susceptibility. These changes in magnetic susceptibility should be related
to ash content (as should the potassium measurements). However, magnetic susceptibility is
not related to changes in reflectance.
Figure 157 Combined spectral curve from sites 259 and 339 by
sample location.
292
Figure 157 shows the combined spectral curves by sample location from the most
representative sites (259 and 339: site 602 is ephemeral on the ground and 279 exhibits a DN
value decrease). All the curves have a similar shape (dominated by Fe3+ absorption), but the
site curve has a 15-20% increase in reflectance in comparison to the off-site (pre and post
site) curves. A similar increase in DN value is also observed in the satellite imagery itself.
Analysis of these curves in ENVI against NASA standards (Grove et al. 1992) revealed that
the marl soil appears to be a closely related to hematite (Fe2O3) and goethite (Fe3+O(OH)).
The marl soil displays the characteristic iron absorption curves of these minerals. As
expected, the transitional samples describe curves somewhere in between the site and off-site
curves. In comparison to Figure 142, the off-site curve appears to correlate quite closely with
curve C (< 2% organic matter and 1-4% iron oxide (FeO)). The site soils appear to correlate
quite closely with curve B (< 2% organic matter and <1% FeO). However, analysis of the
trace iron content in the samples does not correspond with this identification. Alternatively,
curves B and D are chemically and biologically the same soils with a different arrangement of
particle sizes. Hence, this change in texture and/or structure produces a 15% increase in
reflectance from curve D to curve B over the 0.3 to 1.1 µm range (the same range as the
GER 1500 spectro-radiometer). Therefore, if one assumes that there is a close physical and
chemical relationship between on and off-site soils then a change in particle size distribution
and a reduction in moisture content is the most likely cause for the increase in DN value.
This hypothesis is supported by the correlation across all the sites between changes in DN
value and corresponding changes in particle size distribution.
8.4.2 The 2003 soil sampling programme
After the encouraging results from Wilkinson’s samples a further 122 samples were collected
for particle size analysis from 10 sites (and located with GPS) during the 2003 study season.
As optical reflectance is based on the surface microns of soil these samples were collected
from the topsoil. Unfortunately 8 of these samples were misplaced during collection and
transit resulting in 114 samples being available for analysis. Sites were carefully chosen so that
a range of site types of different dates and in different soil environments were sampled (see
Figure 159 and Table 22). For control purposes sites 279 and 339, where Wilkinson placed
transects in the 2001 season, were re-sampled. Transects were located so that they started and
finished off-site and traversed the site itself. Sample intervals were variable, although
293
nominally between 20 and 30 metres. However, emphasis was placed on collecting more
samples at site boundaries.
Figure 158 Comparison of marl soil curve (from spectroradiometry) against standard samples of hematite and goethite.
These samples were processed and analysed for particle size using the methodology
described in Appendix II. The results of these analyses are contained in Appendix III.
However, there were differences between the particle size analyses from the 2001 and 2003
collection on the control samples at sites 279 and 339. In contrast to the 2001 samples the
2003 samples did not contain particles greater than 1mm. Only two reasons were
immediately obvious for these results: either there was human error during the preparation of
the 2001 or 2003 samples or the Coulter LS230 was out of calibration during one of the
processing runs.
294
Figure 159 Location of sites sampled in the 2003 season.
295
It was determined that there were subtle differences in the processing techniques employed
by Galiatsatos and myself. Galiatsatos did not riffle the samples and therefore increased the
likelihood that the Coulter sample was unrepresentative. However, this is highly unlikely to
produce the observed variations. The Coulter LS230 was checked for calibration by running
a range of standard samples and was determined to be in specification. It was subsequently
ascertained that Wilkinson’s samples were collected from consolidated material below the
ploughsoil (c. 10-15cm below topsoil: Wilkinson, pers. comm.). Hence, variations in particle
size were to be expected, but once again such extreme results are unlikely.
To ensure reliability 6 comparable samples collected from the control sites (279 and 339)
during 2001 and 2003 were re-analysed (resulting in a total of thirteen samples). Minor
variations were observed between the two runs of the 2003 samples. Significant variation was
observed between the first and second runs of the 2001 samples (samples 4, 8, 13, 58, 61 and
64: see Figure 160). Like the 2003 samples the second run of the 2001 samples contained no
particles greater than 1 mm. The second run 2001 samples produced comparable particle size
distributions with the 2003 samples.
The causation of these variations in particle size results is still unclear. It is recommended that
the methods employed in Coulter sample preparation and analysis are reviewed so that multitemporal sample results can be integrated with confidence.
8.4.3 Interpretation of the analysis
As discussed in section 8.4.1.5, the 2001 samples and their analysis provided a promising
basis for explaining the increase in satellite DN value over archaeological residues in the marl
zone. However, the results of the particle size analysis on the 2003 samples are generally
disappointing. With the reliability of the grain size analyses still in question any interpretations
are, at the best, tentative. In 2001 the surface scatter sites 259, 279 and 339 produced very
clear differences in particle size at the predicted site boundaries. This clarity is only exhibited
in the tell and low tell sites (97 and 218 respectively) for the 2003 samples. However, less
clear, but identifiable correlations occur at the scatter sites 221, 271, 279, 339 and 508.
296
Figure 160 Comparison of the different Coulter 'runs' on selected
2001 and 2003 samples at sites 279 and 339.
This difference in results could be due to a number of issues:
•
Particle size modification in the intervening two years.
•
The reliance on the satellite imagery to define the sample location (i.e. pre, post,
transitional or site samples).
•
Unresolved problems in the Coulter analysis.
•
Laboratory analysis of particle size does not adequately reflect in-situ particle size.
•
Particle size variation has no relationship with archaeological residues.
Whatever the cause, variation in particle size still has a positive relationship with DN value
across many of the sites studied. However, these variations are not consistent across the
range of sites. Table 21 and Table 22 outline a summary of the results of the particle size
analysis.
297
Sand %
Silt %
Clay %
46.00
9.00
34.00
61.90
16.60
7.60
7.00
5.20
13.70
31.60
43.00
27.00
47.00
12.30
68.20
53.20
59.10
20.00
58.50
49.10
9.00
14.00
15.00
25.80
15.20
39.20
33.90
39.80
27.80
19.30
Sample
No
1
2
3
4
5
6
7
8
9
10
Upland
Mesopotamian soil
2.40
71.10
26.50
11
Lowland
Mesopotamian soil
4.10
45.75
50.15
12
4.10
13.00
1.90
6.40
7.80
8.30
22.00
2.38
6.24
6.59
5.74
4.57
9.74
6.71
6.04
7.47
5.47
11.88
18.24
16.99
26.11
61.70
65.30
61.20
73.60
65.70
65.80
57.50
70.27
59.53
62.24
73.18
60.85
63.59
62.57
61.07
61.93
66.50
55.26
62.41
58.94
54.50
34.20
21.70
36.90
20.00
26.50
25.90
20.50
27.35
34.23
31.17
21.09
34.59
26.67
30.73
32.89
30.60
28.03
32.86
19.35
24.07
19.39
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
Source
Place of extraction or number of
Gravel %
the sample
Building stucco
2.00
Edge E4
15.00
Brick no. 3B
Brick no. 2
Brick no. 2
Brick no. 5
Brick no. 8
Site
Tureng Tépé
Nush-i-Jan
Samarra
Ur
Choche
Sauvage
Aqar Quf
Tell Umar
Tell Nebi Mend
(SHR site 315)
SHR wadi
Irrigated Northern
Marl
SHR
Irrigated Southern
Marl
Southern Marl
Thin Southern Marl
Wadi silts/Thin
Marl
Sample 69: Rampart red brick
Sample 70: Rampart grey brick
Sample 71: Rampart white brick
Sample 72: TrVIII yellow brick
Sample 73: Modern mudbrick
Sample 74: Modern surface
Average across the marl zone
Off site average
SHR tell 97
SHR scatter 271
SHR scatter 508
Off site average
SHR low tell 218
Off site average
SHR scatter 221
SHR scatter 339
SHR scatter 478
Off site average
SHR scatter 279
Off site average
SHR scatter 238
Table 21 Particle size analysis of mud brick and soil. Non SHR
samples are taken from Sauvage (1998 p. 19, Tables 2 and 3).
If the hypothesis that the change in particle size and hence reflectance is primarily caused by
the incorporation of degraded mud-brick into the soil matrix, one would expect to see an
increase in clay particulates. This is based on the assumption that mud-brick is created with
local material and that manufacture increases the clay component. However, this may be
complicated by the amount of re-used material used in the construction of mud-bricks. For
example, it is fair to assume that, mud-bricks on flat sites do not contain much re-used
material. Conversely, mud-bricks on tell sites are likely to be made with material quarried
from the tell itself (i.e. the residues of collapsed mud-brick structures are used as a source for
creating new mud-bricks). In such a situation re-used tell mud-bricks are likely to contain a
range of anthropogenic and natural material that will alter the particle size distribution (see
Figure 161).
298
Figure 161 Mud-brick structures in varying degrees of collapse.
In addition, Table 21 and Figure 162 (displaying a range of particle sizes for mud-brick
samples and on and off-site soils) shows that particle sizes can vary irrespective of local or
regional soil conditions. The range of components in mud-brick is extremely variable (sand:
1.9%-61.9%, silt: 12.3%-73.6% and clay 9%-39.8%). Therefore, the impact of degraded mudbrick on soil particle size is much more complex than originally postulated.
This observation is confirmed in Table 22 which summaries the percentage change of each
soil fraction between off and on site soils. The sites fall into four categories, those that
exhibit an:
•
Increase in clay and silt and a decrease in the sand fraction (site 602).
•
Increase in clay and sand and a decrease in the silt fraction (site 97, 259 and 339
(2001 and 2003 samples)).
•
Increase in sand and a decrease in the clay and silt fraction (site 238, 271, 279
(2001 sample), and 508).
299
•
Increase in sand and silt and a decrease in the clay fraction (site 218, 279 (2003
sample) and 478).
100%
Percentage
80%
60%
Gravel
Clay
Silt
Sand
40%
20%
0%
1
3
5
7
9
11
13
15
17
19
21
23
25
27
29
31
33
Sample Number
Figure 162 Graph of particle size distributions from mud-brick
and soil. The sample numbers relate to Table 21.
The only possible permutations not encountered are:
•
Increase in the clay and a decrease in the sand and silt fraction.
•
Increase in the silt and a decrease in the sand and clay fraction.
Table 22 also subdivides the results on a number of other criteria (the period of the site, the
zone the site is located in, if a depression occurs on the site, the year of collection and the
type of site) in an attempt to allow the visual extrapolation of relationships between these
criteria and soil changes. There seems to be no correlation between any of the four categories
of sample and any of the other site criteria.
The raw Coulter results expressed in Table 22 have been further generalised to display soil
texture in Table 23 (see Figure 141 for cross referencing purposes). Unsurprisingly no real
correlation is expressed. However, the anomalous Coulter readings, between different
seasons, at sites 279 and 339 are displayed in sharp contrast.
300
Table 22 Pivot table displaying a summary of the average
percentage change of soil fractions between off-site and site.
301
UnitID
97
Year
2003
218
2003
221
2003
238
2003
259
2001
271
2003
279
2001
2003
339
2001
2003
478
2003
496
2003
508
2003
602
2001
Location
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
Off site
Site
loam
sandy loam silt loam
1
silty clay loam
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Table 23 Variations in soil texture calculated from the average of
the Coulter results for each location.
Scatter sites 238, 478 and 496 did not display a clear correlation between the site boundaries
and changes in particle size. Site 496 is a site located in the alluvial floodplain/fan zone where
post-depositional mixing of alluvium or alluvial fan deposits may have caused variations in
the particle size distribution. This is reinforced by the results where the average percentage
change between off and on-site soils is less than 1% for all components. Site 478 is a small
prehistoric site where the assumptions about mud-brick construction may not hold.
Furthermore this site was extremely difficult to locate using satellite imagery. It is also
difficult to accurately define the site extent for site 238, therefore the samples may have been
given wrong locations in relation to the site (i.e. off, transition and site). Site 339, gave
positive results in the 2001 sample programme but less clear results in the 2003 programme.
As yet no convincing reason has been established as to why this site did not produce
comparable results during the 2003 sampling programme. However, this is one of the four
302
sites which contain a depression. In this instance the depression lies effectively in the centre
of the site (see Figure 102). Therefore these depressions could very easily skew transect
interpretations of particle size distributions by becoming repositories for different particle
sizes (an aeolian or water trap or an area of increased colluviation). These are issues which
should be addressed in a continuing programme of more refined analyses.
Figure 163 Comparison of curve profiles for a number of
attributes for site 339.
Irrespective of the particle size analysis this research has produced a number of empirical
observations that require explanation. Archaeological site soils do demonstrate an increase in
reflectance in comparison to off site soils. This increase in reflectance is observed in both the
satellite sensors and the spectro-radiometry readings. The similarity of the spectral curves (see
Figure 157) demonstrate that no new absorption features were introduced, indicating that the
soils are chemically similar. Hence, the structure of the soil has changed resulting in an
increase in reflectance. This research has demonstrated that whatever changes have occurred
will be complex, possibly involving combined changes in particle size, moisture content and
organic matter (see Figure 163).
303
Subsequently, Galiatsatos (in prep) conducted loss on ignition on the 2001 samples to
determine the percentage organic content of the soils. Galiatsatos found a slight negative
correlation between variations in organic matter and variations in laboratory reflectance (see
Figure 164). This decrease in organic matter content associated with sites would theoretically
increase reflectance
Figure 164 Variations in organic matter content across sites 279,
259, 602 and 339.
8.5 Discussion
The laboratory analysis of the soil samples produced a number of interesting results. Changes
in particle size do seem to correlate with areas of increased DN value. However, these
changes in particle size are not consistent between different size fractions and do not
highlight any correlation with other site attributes except moisture content. Furthermore,
there is some ambiguity regarding the accuracy of the Coulter particle size readings. The
changes in particle size will be discussed, including their implications for changes to soil
structure and drainage.
On a point of methodology, samples should be collected perpendicular to the normal
ploughing direction. For example at site 221 (see Figure 200) the imagery displays a diffuse
304
spread in the NE-SW direction which corresponds with the direction of ploughing. This
spread is less intense in the NW-SE direction. Samples taken in a transect perpendicular to
the ploughing direction should show a more distinct boundary change.
8.5.1 Non-organic imports
For the purposes of this discussion non-organic imports to an archaeological site fall into two
categories on the basis of their size: micro imports (those with a particle size of less than
2mm) and macro imports (those with a particle size greater than 2mm)
Micro imports have been the major focus of this chapter. It has been assumed that degraded
building material, particularly mud-brick, and pottery is the major cause for the change in
particle sizes between on and off-site soils. However, mud-brick itself has been recognised as
a complex medium, and although degraded mud-brick will change the particle size
distribution of a soil it does not do so in a predictable way. For example, one cannot say that
there is always an increase in the clay component. Therefore the increase in DN value is a
product of the different mineralogies, the different particle size and their structure. The
ability of clay to aggregate should not be completely discounted in this process. Therefore
other factors, related to particle size, which produce a consistent increase in soil reflectance
must be sought. It is likely that, associated with the change in the particle size distribution,
there is also a change in structure and porosity which has a profound impact on reflectance.
This is seen when examining curves B and D in Figure 142; these are chemically and
biologically similar soils with a different arrangement of particle sizes. The laboratory analysis
undertaken does not give an indication of these important components. The changes in
structure and porosity could also indirectly impact on reflectance by changing the nature of
drainage. If the soil has a loose structure (as at site 221) then there is improved drainage and
an increase in reflectance. However, the spectro-radiometry readings of the ambient and
dried samples did not display any moisture variations. In-situ spectro-radiometry readings are
under analysis by Galiatsatos (in prep).
Imported material such as stone, brick and tile (even after robbing) may be incorporated into
the site matrix. These have a massive grain size in comparison to the local matrix. Material
may also be masked by the dumping of other material over them. Field-walking is used to
identify clusters of these macroscopic artefacts. However, the sizes of these surface artefacts
are massive in comparison to the soil matrix and will affect its texture and porosity.
305
Subsurface layers and masonry have a large impact on the surface qualities of the soil and soil
moisture characteristics. Whatever formation and de-formation processes are associated with
the arrival of massive material they do have an impact on site reflectance. This is highlighted
on site 279 which has the largest number of artefacts with massive grain size and exhibits a
decrease in reflectance. In this context the particle size analysis has ignored massive particles
as they were sieved prior to Coulter analysis.
8.5.2 Organic matter imports
Organic matter has frequently been used to define some aspect of an archaeological site,
especially the boundaries of occupation (horizontally and vertically), the presence of features,
the source of the deposits, or the presence of post-depositional alterations. However,
archaeological residue prospection employing surficial organic matter should be approached
with care. A salutary lesson on organic matter analysis is discussed by Stein (1992), where
Heidenreich and Narratil expected organic matter to rise within site confines. In fact organic
matter did not change significantly. Their expectations were based on the assumption that
greater amounts of organic matter were deposited within the village boundaries and that,
consequently, the level of organic matter within village samples would be higher than those in
the surrounding soils. They failed to recognise that decomposition of the organic material by
micro-organism activity would re-establish the equilibrium. However, another study
conducted by Goffer et al. (1983) at the tell site of Beer-Sheba, Israel showed a different
response in a more regionally representative environment. Two large pits were discovered
that were filled with very dark sediment. The large size and organic-rich content of the pits
led the authors to suggest that the pits were used for making compost. The percentages of
organic carbon, nitrogen, and phosphorus were determined for samples from the pit fill, the
nearby tell, an off-site location and from compost found in an adjacent modern town. The
largest percentages of carbon were found in the sample from the modern compost, with
decreasing values found in the pit fill, the tell sediment, and the lowest percentages in the offsite soil. Decomposition had resulted in the loss of organic matter in the tell sediments. The
percentages of organic matter in the pit fill were elevated above those expected by
occupation or pedogenesis alone. However, these pit deposits were not surficial and hence
underwent different post-depositional processes.
Aeolian erosion in the application area has led to the removal of the majority of the ‘O’
horizon. This means that the background soils may have their humic content exposed. This
306
degraded organic matter, which has a red colour, might provide a good contrast against sites
which do not express such colouration. The laboratory analysis of the 2001 samples also
indicated that there is a slight negative correlation between site organic content and
reflectance. It is recommended that further organic matter analyses are undertaken in the
future to confirm these results over a more representative sample.
8.5.3 Reflectance implications of the soil analysis: summary and conclusions
In section 8.3.5 the following reasons were postulated for the increased reflectance observed
at sites in the marl zone:
•
Low organic matter content.
•
Low iron oxide content.
•
Relatively finer textured particles or differences in soil structure.
•
Improved drainage reducing soil moisture.
The loss on ignition analysis on the 2001 soil samples (Galiatsatos in prep) did show that
there was a slight decrease in organic matter content associated with sites. In order to more
rigorously verify this further samples should be analysed.
With the possible exception of site 259, trace element analysis of the 2001 samples effectively
discounted iron as an important factor. However, this analysis was conducted on iron and
not its oxides (as discussed in section 8.4.1.5). In fact the lack of any significant correlation
with trace element analysis implies that the site soils still share many chemical characteristics
with the background soil and parent material.
Particle size does display a correlation with the boundaries of sites as defined by the satellite
imagery. However, it is unclear whether this is a direct result of changes in reflectance due to
changes in particle size or a change in reflectance due to some other factor such as structure
or porosity (and hence drainage). Further analyses, such as bulk density (Stanjek and
Fabbinder 1995; Scudder et al. 1996) and in-situ moisture and organic matter readings, would
be beneficial to define these causative factors rigorously. Further benefits may be derived if
the results are analysed in conjunction with micro and macro scale formation models (see for
example Ward and Larcombe 2003).
307
Even though no specific reason for the increase in reflectance associated with archaeological
sites has been explicitly defined, the analysis has empirically identified a number of promising
areas for further enquiry:
•
Reflectance differences are observed at sites in the visible and NIR wavelengths
by all sensors.
•
Ground and laboratory radiometry correlates with these sensor readings and with
soil moisture differences at key sites.
•
Ground and laboratory radiometry relationships with grain size are unclear.
However, there is evidence to show positive relationships.
•
Reflectance data from soil samples matches with theoretical expectations (i.e. no
major absorption features other than Fe3+ and Fe2+ transitions (see section 8.3.3
and Figure 146)).
This analysis has provided a platform from which to conduct further research to explain the
increase in soil reflectance associated with archaeological sites. The relationships between
grain size, organic matter, moisture and soil structure on archaeological sites requires further
work as there is surprisingly little literature available on this topic. Further, radiometry is an
objective way of observing reflectivity in the field and has thus allowed spatial and spectral
scale issues to be addressed between a point and a pixel. It is hoped this phase of the research
will have significant impact in this and other environmental similar areas.
8.5.4 Recommendations for field collection and analysis
In addition to the collection and analysis of in-situ bulk density, moisture and organic matter
readings (discussed above), it is recommended that further grain size analysis is undertaken.
Samples can be coarsely analysed in the field by sieving the major fractions (clay, silt, sand
and gravel) on site. Furthermore, a handheld radiometer should be used to collect reflectance
values of these samples.
308
CHAPTER 9 SATELLITE IMAGERY AS A CRM TOOL
9.1 Introduction
The previous chapters have been concerned with the prospection and documentation of the
archaeological record within the project area. As Syria does not currently have an accessible
national archaeological inventory, the SHR data set can be used by the DGAM for the
purposes of management and monitoring to ensure long term stability. This is critical as the
archaeological resource is under considerable stress as a result of population growth and
developments in both industry and agriculture. However, CRM applications do not solely
concern themselves with management and preservation; they also cover a range of associated
activities including research and aspects of public participation.
All of the data has been integrated into a GIS framework which allows the data sets to be
analysed in many complementary ways. Satellite imagery can provide an alternative
mechanism for visualising the landscape in a digital environment. It can be used as an
alternative basemap or as a different layer for computerised landscape visualisation. Not only
are these useful for research purposes, particularly for phenomenological approaches to
interpretation; they are also a powerful tool for public engagement in the archaeological
resource. For example, 3-dimensional fly-overs are employed in a number of successful
museum applications (Forte and Williams 2003). The application of multi-temporal imagery
provides valuable insights into how the landscape has changed over time and how these
changes have impacted on archaeological residues. When coupled with digital in-field
recording devices, satellite imagery provides another layer of information that can improve
in-situ interpretations and aid in field navigation.
9.2 Satellite imagery as a contextual backdrop
Air photographs are not maps, but mirrors: they are not selective: they simply show what
happens to be visible at the time; they are undiscriminating and should be valued with this
in mind.
(Johnson, 1982, 6 in Maxwell)
309
The context of archaeological residues is an important aspect for their interpretation.
However, ‘context’ can mean a number of different things to a number of different people
depending on their analytical goals. For example an environmentally determinist
archaeologist may be interested in environmental variables that determine site placement (as
discussed in Chapter 6); a CRM archaeologist will also be interested in localised
developments that may impact on the archaeological residues, and archaeologists researching
settlement dynamics will be interested in the spatial relationships of contemporaneous sites
and infrastructure networks. As has been discussed in earlier chapters, satellite images are
flexible resources that can be applied to a wide range of spatial problems providing different
mechanisms of contextualisation. With this in mind two areas of visualisation are discussed:
the comparison of satellite against digital base maps and the use of satellite imagery as a
visualisation tool.
9.2.1 The comparison of satellite imagery against digital map bases
Remotely sensed imagery and maps are both representations of the Earth. However, there
are profound differences between the two data sources. Maps are generalisations of reality
and reflect the mapmaker’s selection of what is important to represent (Müller et al. 1995).
Hence, information of low import is discarded to improve interpretative clarity.
Unfortunately this normally includes the majority of non-monumental archaeological
residues. Remotely sensed imagery, on the other hand, are an ‘objective’ data source. The
interpretable content of the imagery is subject to different biases relating to the resolution of
the sensors and the ambient environmental conditions during collection (Liverman et al.
1998). As the spatial resolution of satellite sensors continues to improve they are more
regularly being used as alternative sources for updating map bases (Armenakis et al. 2003)
As has already been discussed in Chapter 5, the availability of base mapping in the application
area was generally poor, although a 1:25,000 topographic map was acquired. Comparisons
will be drawn between this mapping and the satellite imagery (primarily the Ikonos). From a
European perspective, this may be viewed as a biased comparison in favour of satellite
imagery. However, in the Middle East and other areas of the world it is difficult to access
high quality topographic and thematic mapping at appropriate scales.
From the 1:25,000 topographic mapping 16 sites were identified from antiquity symbols, 187
potential sites from contours, 33 pen sites (see Hinterland System 1 in Figure 165) and 106
310
sites from possible place name evidence. An image interpretation key was produced for this
process of identification (see Figure 165). As has already been demonstrated in Chapter 7 the
satellite imagery allowed the detection of many more residues. From a prospection basis
satellite imagery is a much more useful product, providing more accurate interpretation and
spatial referencing (particularly in the basalt zone).
Figure 165 Image interpretation key for the Syrian topographic
map.
This is further reinforced by the improved contextualisation of archaeological residues from
satellite imagery. As is demonstrated in Figure 165 the satellite image provides improved
insights into local conditions. For example, the imagery provides information about residue
311
type, residue extent and vegetation cover. However, experience is required to interpret these
indicators.
Satellite imagery was a much improved resource for compiling thematic data. Thematic
mapping for the area was only available at inappropriate scales. Therefore, landscape themes
were produced directly from the satellite imagery (see Chapter 6). Thematic variables such as
soil, vegetation, hydrology, topography and elevation are used to either contextualise the
archaeological data or in predictive modelling exercises. Many projects have employed
Landsat imagery to classify soil and vegetation themes (Gaffney and Stancic 1991; Gaffney et
al. 1996; Rothaus and De Morett 1999; Harrower et al. 2002). Ikonos and Corona imagery can
immediately supply information related to present day and relict topography and hydrology
and can be used akin to vertical aerial photography to update digital mapping. Furthermore,
the Ikonos and Corona data sets are available as stereo pairs which can be used to create
digital elevation models. However, environmental thematic variable production is not just an
issue of reproducing maps:
The real issue is that the variables are not usually chosen because of some theoretical model
or even based on a hypothesis about the past. Rather, they are most often selected because
they are the only data available. This is not good science.
(Maschner 1996).
This quote highlights one of the most fundamental problems facing archaeological thematic
data analysis through its use of readily available yet potentially inappropriate data sets.
Satellite imagery has uses beyond the accurate collection of information reflected in present
day mapping it also has the capacity to model more archaeologically pertinent information.
However, very limited research has been conducted into the production of specific landscape
archaeological themes. This lack of research is in stark contrast when one compares the
research into this area by landscape ecologists (Griffiths et al. 2000; Griffiths and Mather
2000; Luque 2000; Wu and Qi 2000). Much of the archaeological research has been focussed
on terrain based viewshed analysis (Wheatley 1995; Lake et al. 1998). However, Duncan and
Beckman’s (2000 pp. 41-42) site location model created a theme to represent solar insolation
gain on the shortest day of the year using DEM derived variables. This theme was created on
312
the assumption that, in their application area at least, warmer areas are preferable to colder
areas during the colder parts of the year. Gaffney et al. (1996) employed Landsat TM to
define soil types. However, their criteria only classified agricultural value from poor to good.
Ebert (1988) gives an overview of holistic or ecosystemic approaches employing satellite
imagery.
Mapping does have some benefits over satellite imagery. Map generalisation allows the
presentation of complex spatial entities and relationships in a digestible format. However, as
has been demonstrated by many of the illustrations in this thesis these elements can be
replicated by satellite imagery derivatives to produce equivalent resources. More importantly,
mapping can have more stable spatial accuracy than satellite imagery. Due to the problem of
projection encountered with the Syrian mapping in this research this was not the case.
However, neighbouring countries (such as Jordan and Lebanon (Seif, pers. comm.)) do have
accessible, regularly updated mapping. These resources are particularly useful as a reference
source when co-registering satellite imagery. This impact has, however, been reduced by the
accuracy of handheld and differential GPS.
Remotely sensed data provide an alternative representation of geographical context to that
given by maps. The ‘objectivity’ of a satellite image in most instances provides an improved
archaeological backdrop. Furthermore, satellite imagery may allow the creation of more
archaeologically specific thematic data. In the study area the high spatial resolution effectively
made the available topographic and thematic mapping redundant. However, given that other
projects may not have comparable access to the quality and quantity of high resolution
satellite imagery, mapping sources should still be included in any project.
Finally, it is important to identify the fact that neither remotely sensed imagery nor map data
provide insights into how landscapes were perceived by the people who inhabited them.
9.2.2 Satellite imagery as a presentation and visualisation tool
The last decade has seen the archaeological community embrace computerised virtual
visualisation and reconstruction tools (Gillings et al. 1999; Goodrick and Gillings 2000; Forte
and Williams 2003). Forte in particular has championed the integration of multiple data
sources, including satellite imagery, for landscape analysis, reconstruction and visualisation
purposes (Forte 1995; 1999; 2000; 2003b; 2003a; Forte and Kay 2003). Forte (2003a)
313
employs the term ‘mindscape’ for an immersive computer generated ecosystem that facilitates
multi-scalar archaeological cognition, contextualisation and re-interpretation. He also
provides a series of protocols for the integration of such data. Although most of this work is
beyond the scope of this research it provides a useful template for the presentation and
visualisation requirements of satellite imagery within landscape archaeological applications.
Figure 166 Isometric visualisation (with 15 time exaggeration) of
different imagery draped over a terrain model.
At a basic level satellite imagery provides a geo-referenced backdrop for the articulation and
visualisation of landscape data. As discussed in section 9.2.1 satellite imagery provides a more
‘objective’ representation of space which is comparatively realistic to the end user.
Visualisation is significantly enhanced by ‘draping’ satellite imagery over a terrain model to
produce a 2.5 dimensional representation (see Figure 166). The inclusion of higher spatial
resolution imagery, extracted derivatives and archaeological data means that a user can
visualise a range of information in pseudo 3-d. This facilitates the articulation of a range of
landscape data for the researcher and can provide an engaging presentation medium for
museums. These static presentations can be further enhanced by producing ‘fly’ or ‘walkthrough’ movies or alternatively a fully interactive virtual environment.
314
From a presentation perspective, the integration of satellite imagery within a GIS facilitates
the rapid production of publication quality mapping (see, for example, Figure 54 and Figure
65). The benefit of GIS applications in this instance is that a variety of bespoke maps can be
produced whereby the user controls the content and degree of generalisation.
Once again, however, the vast majority of these presentations are prepared in Cartesian space
and do not represent how the mental or cognitive maps of the landscape are perceived by
actors (Kvamme 1999 p. 182). These actors may have used different, and at times, non-linear
representations of space.
Figure 167 Problems of site location (Site 308).
9.3 Satellite imagery and mobile applications
The benefits of field-based GIS for spatial sciences are outlined by Pundt and BrinkkotterRunde (2000). The future for field archaeology will inevitably see increased usage of
computerised techniques. Global Positioning Systems (GPS) and total stations are
supplanting traditional approaches in many applications (particularly landscape survey) and
cheap handheld Personal Digital Assistants (PDAs) are already used to record ‘digital’ context
sheets in a relational database (Ryan et al. 1998; Beck and Beck 2001; Beck 2002).
315
There are many layers of data which an archaeologist can access while using mobile computer
applications. Amongst other things satellite imagery will be of value for site navigation and
field interpretation applications.
Figure 168 Comparison of corrected and uncorrected Ikonos
imagery with GPS overlay
9.3.1 Site navigation
The 1999 to 2000 fieldwork seasons highlighted how difficult it was to locate sites even after
they had been identified on the satellite imagery. Figure 167 shows the physical colour
difference that led to the identification of site 308. However, the inset photograph indicates
the difficulty in actually locating this scatter on the ground. After correction of the satellite
imagery to the UTM 37N projection system (see discussion in section 5.4) it was possible to
use GPS to locate the sites. This simple procedure saved a considerable amount of time as
locating sites from out of date maps or imagery in a non-standard projection was very time
consuming. This underlines the importance of selecting an appropriate projection.
Furthermore, incorporating the Ikonos and Corona satellite images into ArcPAD (in MrSID
file format) allowed the use of this significant data resource in the field to aid interpretation.
This is particularly important for the field systems in the basalt landscape (see Figure 168).
316
9.3.2 Spatial data collection
From analysis of the satellite imagery it was possible to map the extent of sites within the
project GIS (see Chapter 7). However, in order to ensure that what was seen in the imagery
was comparable to what was observed on the ground the survey teams also mapped the
extent of surface scatters/soil discolourations in the field. Each ‘site’ was mapped as a simple
polygon within ArcPAD by walking around the extent of the archaeological residues and
logging GPS readings directly. Upon completion of a circuit the polygon was closed and its
unique GIS identifier was added.
The techniques used for site navigation and mapping proved essential for the recording of
the basalt field systems. Using the derived vector interpretations of the archaeological
residues (as discussed in Chapter 7) in ArcPAD the survey team can accurately navigate to
each individual wall segment. However, the survey team was faced with a recording paradox.
Figure 168 shows an uncorrected and corrected Ikonos satellite image overlaid by GPS
points. These GPS points were taken by handheld GPS without reference to an ArcPAD
image backdrop. The corrected image (on the right) demonstrates that basalt walls have a
high correlation with many of the GPS tracks collected. However, the imagery contains more
information than was recorded by GPS alone. On the ground some features may appear
ephemeral or insignificant but when viewed on the satellite imagery a whole wealth of
information can be revealed. Hence, a more rounded interpretation can be produced in the
field by exploiting the synergy of field observations with supporting vertical imagery and
other GIS data.
9.4 Time change analysis
Change detection is a technique used to extract and analyse differences in imagery acquired at
different times. These differences can be a result of anthropogenic modifications (for
example, new construction and de-forestation) or natural/environmental changes (for
example variation in crop vigour). Analysis of image change involves a relatively simple group
of procedures. The challenge is to develop techniques that highlight change in the
phenomena of interest. For example an urban planner, who is interested in monitoring urban
sprawl, would find changes related to crop health a distraction.
Change detection techniques can be broken down into the following areas:
317
•
Visual comparison: when change is obvious, visual comparison can be the
simplest technique. No processing is required beyond image rectification.
•
Layer stacking: changes in co-registered imagery can be highlighted by assigning
the different imagery to different colour guns (red, green and blue). Changes in
reflectance values are observed as colour variations.
•
Image algebra: Changes can be quantified by the mathematical manipulation of
two images using, for example, subtraction, regression and ratioing.
•
Post-classification comparison: Independently classified images from different
dates are compared. This is the only method in which ‘from’ and ‘to’ classes can
be evaluated (forest zone to urban zone).
9.4.1 Data preparation
Image selection and preparation are important for successful change detection analysis. If
comparing imagery from different sensors it is important that the spatial, radiometric and
spectral resolutions are actually comparable (Song et al. 2000; Furby and Campbell 2001).
9.4.1.1 Atmospheric effects
Consideration should also be given to ambient atmospheric effects which can produce
erroneous artefacts for time change analysis. Haze, smoke and clouds can produce undesired
areas of change. Clouds should be masked out of any analysis. Haze and smoke can be more
difficult to mitigate. Both atmospheric correction and histogram matching techniques can be
used to improve analysis. However, Song et al. (2000) discuss a variety of temporal analysis
techniques which do not require atmospheric calibration.
9.4.1.2 Seasonal effects
Vegetation differences are the over-riding cause for most change in imagery over time.
Unless one wants to detect seasonal changes then it is advisable to minimise their impact by
selecting imagery that has been collected at approximately the same month and day. Hence,
seasonal effects and Sun angle differences can be assumed to be constant.
9.4.1.3 Soil moisture
Moisture content affects the reflectance values of both soil and vegetation. Hence, change
detection techniques can highlight reflectance changes due solely to differences in moisture
content. Although this may have some useful applications for agronomists it can mask other
318
more subtle effects. It is therefore not advisable to employ imagery that has been taken after
heavy rainfall. Historical climatological data, if available, can be used to determine the effects
of rainfall.
9.4.1.4 Image preparation
The amount of image preparation to ameliorate for atmospheric, Sun-angle, vegetation and
moisture effects are dependent on the type of analysis to be performed. Visual comparison
was deemed the most appropriate analytical method for the high spatial resolution Ikonos
and Corona imagery for the following reasons:
•
Sub-pixel co-registration could not be guaranteed.
•
The imagery is taken at different times of the year, predominantly highlighting
vegetation changes.
•
Concurrent atmospheric information is not available for the Corona imagery.
•
The radiometric depth between the Corona and Ikonos imagery was different.
As the Landsat imagery was only used for providing information on the effects of changes in
land management over time this was also analysed visually. This meant that no specific
preparation techniques were used prior to conducting the analysis.
9.4.2 Impact of changes over time
The temporal component of the imagery was of profound significance in relation to residue
detection. There is approximately 30 years difference between the Corona and Ikonos data
sets. In the intervening periods the landscape has been significantly modified, directly and
indirectly impacting on the archaeological residues.
Bulldozing is the most significant modification in the basalt zone, so much so that the
DGAM is preserving a 2x2 km area. Bulldozing has also impacted on the marl landscape,
although to a lesser extent. The residues in this zone are also harder to eradicate completely
as they are fine grained deposits. Site 97 demonstrates that bulldozing (or some other heavy
earth moving equipment) was in use prior to 1970 (see Figure 176). Consequently, bulldozing
cannot be discounted as a residue modifier in the Corona imagery. Other infrastructure
modifications have also occurred. The road network has been extensively upgraded (a northsouth dual carriageway, ring road around Homs and surface metalling for b-roads). An
319
underground water pipe has been laid between the spring at ‘Ain at-Tannur and Homs
cutting across the southern marl. The rail network has also been extended.
These changes to infrastructure have also had an impact on settlement locations and land
management practices. Settlement expansion in particular is closely linked to high quality
infrastructure networks such as roads (cf Sever 1998). However, urban expansion itself has
not heavily impacted on residues. Although, a number of sites are enclosed within military
areas and the cutting of trenches into tells has compromised their archaeological integrity.
In this time frame agricultural practices have changed. Deeper ploughing techniques have
been introduced to the area. In the marl zone the Ikonos imagery contains many more high
reflectance areas (potential archaeological residues) than the Corona. It is likely that this is
either underlying marl being bought to the surface by the plough or possibly increased
erosion due to the aridity of the past decade. A number of these areas were examined in the
2003 season: no obvious archaeological residues were encountered. Higher quality
agricultural soil has been moved and re-deposited in the landscape. This has been reported
from both tell sites and lacustrine deposits (Bshesh, pers. comm.).
Irrigation has extended across all the zones allowing an increased number of crop rotations in
a year. The increased amount of surface vegetation could potentially mask many
archaeological sites if the time of image collection is not chosen carefully. A combination of
these factors probably accounts for the eradication of subtle features such as some wadi
channels (see section 6.3).
The Landsat imagery (see Figure 169) highlights the effects of irrigation over a thirteen year
period. As late as the 1980s the majority of the application area was reliant on rain-fed
agriculture with one cropping season (except in the fluvial margins). However, irrigation and
other pumping has meant that more crop rotations can occur in a single season. This has had
a significant impact on the local water table. 25 years ago the water table was c. 2 – 5 metres
below the surface, but now some pumps have depths of between 20 and 40 metres (Bshesh,
pers. comm.).
The ramifications of this pumping are seen in the extent of Lake Qatina which has shrunk
dramatically (see Figure 64 and Figure 169). A study of ground water use (Rodriguez et al.
1999) has been undertaken by the International Centre for Agricultural Research in the Dry
320
Areas at Atareb, northwest Syria. A 15 year study of the water-table at Tel Hadya determined
that the water table was shrinking by, on average, 1.44 m per year (see Figure 170).
Figure 169 Comparison of October Landsat scenes from 1987 and
2000. Note the increase in vegetation in the 2000 scene (denoted
by the red colouration) and shrinkage of Lake Qatina due to
irrigation.
321
This could explain why reflectance has increased between 1987 and 2000 in the Landsat
imagery in the southern marl zone. As the water table has dropped there is less soil moisture
and hence reflectance is increased. Furthermore, the culmination of the effects of the
different landscape modifiers has effectively eradicated some features from the imagery. As
discussed in Chapters 6 and 7 some natural and cultural features, particularly wadis, appeared
on the Corona imagery but not on the Ikonos.
Figure 170 Depths of the water table in Tel Hadya (1983 – 1997
(Rodriguez et al. 1999 p. 9))
Although some landscape features have disappeared on the Ikonos imagery underlying
artefacts can sometimes be inferred. For example, Figure 171 displays a Corona image with
an indistinct geological feature which, from its morphology, looks hydrological. The
comparable Ikonos image shows that modern house construction respects the geological
feature (which is now not visible) rather than the NE-SW road which appears in both images.
In summary changes in landscape management techniques have had a profound impact on
this landscape in the past 30 years. Improvements in infrastructure and urban expansion have
impacted on a small number of archaeological residues. Of greater concern, however, are the
changes in agricultural practices and bulldozing. Extensive irrigation has helped to lower the
water table which has significantly altered reflectance characteristics in the marl zone. Deeper
322
ploughing has eradicated or partially destroyed some features and produced many more
negative but ‘potential’ features (see Figure 177). Finally bulldozing in the basalt zone has
reached such a level that the DGAM has had to protect a portion of the landscape.
Figure 171 Urban expansion in the marl zone.
At a more refined level, images can be studied on a seasonal basis to evaluate changes in
landscape pattern. From a detection perspective, this allows the evaluation of archaeological
residue visibility under different environmental regimes. As part of the purchasing
arrangement for the Ikonos imagery it was attempted to get repeat coverage of a small area of
known archaeological residues at monthly interval. Amongst other things this study would
allow one to explicitly define the environmental conditions for optimal residue detection
using empirical, rather than theoretical, reasoning. However, the cost implications for such a
study were prohibitive. It is hoped that closer academic relationships will be forged with the
current and future providers of high resolution commercial imagery to facilitate such studies.
The three different Corona missions were originally collected over a six month period
(December 1969 – June 1970) and intersected pre-harvest, harvest and post-harvest (see
Figure 172 and Figure 173). Mission 1108 was taken prior to the heavy rainfall (Lake Qatina
is not full). Although some rain has occurred there is still a good contrast between sites and
soil across the landscape. Mission 1110 was taken just prior to harvest. The vegetation means
that with the exception of tells many sites are masked (particularly those near the Orontes
and Lake Qatina). This mission had the lowest image quality, probably due to atmospheric
323
dust (see Figure 173). Mission 1111 was taken after the harvest. Although some patches of
vegetation can still be identified, there is a good contrast between sites and soil across the
landscape. It should be noted that the atmospheric problem that probably affected the
Mission 1110 imagery is not exhibited in the 1111 imagery. Mission 1111 was collected at c.
18:30: at this time the wind driven through the Homs-Tripoli gap has generally subsided.
Figure 172 Seasonal effects on the satellite imagery.
Two Landsat images were collected on 14th January 2000 and 28th October 2000 (the 1987
images had approximately the same time difference but can not be pan-sharpened; see Figure
172). The January image was taken during heavy rainfall and with immature vegetation. Sites
are still quite obvious in this landscape, although the full range can not be detected due to the
spatial resolution of the sensor. Comparatively, the October scene exhibits a much brighter
residue response resulting in improved detection. This demonstrates that increased aridity
increases the visibility of archaeological residues.
324
Two Ikonos images were collected on 5th September 1999 and 3rd February 2002. Much like
the October Landsat image, the residues in the September Ikonos appear much brighter.
However, the spatial resolution has been somewhat degraded due to atmospheric dust. In the
wet February image, all of the residues are still visible although contrast between site and soil
has been reduced. This is paralleled by the results of soil colours from on and off site soil
samples when wet and dry. However, rain has removed atmospheric dust and the image is
comparatively sharp.
Figure 173 Comparison of Corona missions of the same areas in
the basalt and marl zones. Note the much better relative image
quality of the winter scene.
Hence, seasonal environmental conditions have a significant effect on the ability to detect
archaeological residues. Periods of intense vegetation should be avoided as sites are masked,
whereas increased aridity increases the relative reflectance. Even during the periods of heavy
rain residues can still be detected, although their contrast with the background soil is likely to
be reduced. The results of this study compare quite closely with the theoretically defined
acquisition time variables defined in section 5.2.4.
325
9.4.3 Site monitoring
In order to evaluate if seasonality or changes in landscape management impacted on
archaeological residues, site extents from different satellite imagery were mapped and
compared. An area in the marl zone was chosen as this zone is subject to a wider range of
landscape modifications. Figure 175 displays these different extents in the Corona missions
and the Ikonos MS. The results of this comparison are not conclusive. Some of the sites
show quite subtle changes (sites 251, 339, 475) which could be due to geo-referencing errors.
Sites 319 and 477 show quite different extents and it is difficult to explain the cause for this
change. Other sites, such as 221, do exhibit much larger extents in the Ikonos imagery
probably due to the spread of surface material during ploughing or bulldozing.
Figure 174 Monitoring of tell (site 173) on the lake edge.
However, it was possible to record other more specific impacts on archaeological residues. In
addition to the destruction of the basalt landscape by bulldozing (as discussed in section 7.4)
sites at the lake margins are subject to the longer term destructive effects of water. Figure 174
displays the extent of tell erosion between 1970 and 2002 taken directly from the Corona and
Ikonos satellite imagery. In the intervening timeframe 3,800m2 (approximately 10%) of the
stable tell platform has been eroded.
326
Figure 175 Comparison of site extents over time for sites 251, 319,
339, 471 and 477.
327
Hence, multi-temporal imagery can be used to monitor archaeological sites. However,
ground observation data is required to provide a context for any interpretation. From a
monitoring perspective it may be financially untenable to buy imagery regularly. However, an
increasing amount of archive imagery is made available at discount or at nominal cost.
Many archaeological sites have been subject to unforeseen catastrophic disasters. In recent
years floods, earthquakes and military action have destroyed a number of world class sites.
Although the application of satellite imagery cannot stop such disasters from occurring it can
help to rapidly evaluate their impact. High resolution commercial sensors can easily collect
imagery in these potentially dangerous areas. This would be a significant resource in any
mitigation scheme.
Satellite imagery may also aid in the curation of world heritage sites. At present the spatial
component of the world heritage site database is not complete and many developing
countries find it difficult to provide appropriate spatial data (Stott, pers. comm.). This could
be ameliorated by including medium to high spatial resolution data (such as Landsat ETM+)
with every regional application. It is, however, understood that satellite imagery will not
resolve such issues in its own right and some degree of supportive capacity building is also
required.
Although not observed in this environment a particularly novel time change application was
employed by Palumbo and Powlesland (1996 pp. 126-127) at Pueblo Buenito and Chentro
Ketl. By integrating aerial and satellite imagery from over a thirty year period they were able
demonstrate how the management infrastructure (footpaths, roads, car-parks etc.) had
changed and developed and how this knowledge would be useful in the future management
of the archaeological parks. However, it does strike a cautionary note: in some instances these
management strategies had had a significant negative impact on the archaeological resource
(Powlesland, pers. comm.).
9.5 Discussion
Satellite imagery can play an important role within CRM activities. This role is increased in
areas where mapping is poor and the curatorial authorities do not have a full inventory of
their archaeological resource. From an inventory basis satellite imagery has already
demonstrated its utility and, in this application area, is a more useful medium than
328
topographic mapping. Furthermore, satellite imagery can be a very flexible resource in the
presentation of data to both academic peers and the public. Although it can be argued that
aerial photographs provide imagery with a higher level of interpretable content, satellite
imagery provides a much larger synoptic footprint (Schmidt 2004). Satellite imagery are,
therefore, much more appropriate for landscape survey and archaeological inventory
applications. Furthermore, from a cost-benefit perspective it is cheaper to incorporate
satellite imagery at an early stage rather than relying on long term coverage from aerial
photography. This is even more important where GIS management techniques are used, as
rectification of small footprint vertical and oblique photographs can be time consuming and
expensive.
From a practical viewpoint the incorporation of satellite imagery within digital mobile
applications provides a number of benefits. Although quite simplistic, the ability to navigate
directly to potential residues saves significant amounts of time. Of more import is the use of
the imagery in the field as an interpretative tool. The different perspective offered by vertical
imagery can improve the detection and interpretation of features that look ephemeral on the
ground.
Time change analysis provided some valuable insights into how the landscape has changed
since 1969. This analysis highlights the important utility of historical satellite imagery in any
study such as this. The Corona imagery provides a snapshot of a landscape prior to
significant urban and rural modification. Furthermore, many of the most archaeologically
destructive heavy machinery (i.e. bulldozers and tractors with deep ploughs) were not in
common use. The imagery was unable to provide any significant monitoring information for
sites in an agricultural setting (see Figure 175). However, it was possible to monitor the
destruction of sites on the lake edge and those subject to large scale destructive modification
through bulldozing or infrastructure expansion.
329
SECTION 3 SUMMARY, RECOMMENDATIONS AND CONCLUSIONS
330
CHAPTER 10 DISCUSSION OF THE RESULTS
Chapters 6-9 have described a number of different techniques for preparing satellite imagery
and extracting archaeologically pertinent information. This information has been particularly
beneficial for the goals of the SHR project. Prior to the incorporation of the satellite imagery
the Desk Based Assessment revealed that, with the exception of tell sites, little was known
about the range and location of archaeological residues in the environmental zones and, for
that matter, about variations within the environmental zones themselves. What follows in this
chapter is a summary of the salient points of the interpretation process across all
environmental zones. The main focus is on archaeological prospection.
10.1 Archaeological prospection summary
In comparison to traditional ground survey aerial photography can be a very economical way
of getting an initial overall perspective of objects, phenomenon and their spatial relationships;
furthermore, as a reconnaissance tool, it can prove to be highly cost-effective efficient way of
exploring a large area for discovery purposes.
(Ebert and Lyons 1983 p. 1246)
The use of satellite imagery for archaeological prospection has been the main thrust of this
research. In general, the application of Corona and Ikonos satellite imagery to archaeological
prospection in the application area has been an unqualified success. The techniques
employed have a high correlation with archaeological residues observed during surface survey
and have the benefit of being rapid, spatially accurate and cost-effective. Furthermore, they
allow the prospection phase to occur as part of the Desk Based Assessment, which not only
reduces costs but enables the contextualisation of the resources with other geographical data
sets, enhancing the research framework at an early stage. Within this context the rerectification of the satellite imagery using GPS derived GCPs is of particular significance.
This allows each of the spatial datasets to be overlaid with a higher degree of spatial
precision.
331
High resolution satellite imagery has enhanced the understanding of the archaeological
resource within the application area. Prior to the analysis of the imagery only 63 sites were
known (i.e. place name (tell, khirbet) or antiquity symbol). After the analysis a total of 189
archaeological sites have been positively identified on the ground (tells, scatters and
structures) and 271 potential sites have been located but not visited (see Table 24). It should
also be noted that this does not include the 133 km2 of field systems have been mapped in
the basalt zone. This significant amount of work has only been one facet of the SHR project
over the previous 5 years.
Unit Type
Field System
Indeterminate
Installation
Non-site
Scatter
Structures
Tell
TOTAL
Count
3
271
4
60
91
66
32
527
Table 24 Summary of sites in the whole application area.
UnitType
Count
UnitID
130
132
133
135
170
171
198
199
200
201
202
203
205
219
225
244
258
260
261
263
277
278
281
322
326
327
329
350
355
364
408
483
950
951
952
954
955
Grand Total
Indeterminate
Evidence source Evidence
Corona 1108 Corona 1110 Corona 1111 Ikonos MS
Soil Colour
Soil Colour
Soil Colour Soil Colour
1
1
1
1
1
Ikonos Pan Syrian 1:25,000
Soil Colour
Antiquity
Contour
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Grand Total
Khirbet
Place Name
Quarry
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
8
1
1
1
1
1
8
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
11
1
1
1
1
1
12
1
1
19
6
Table 25 Summary of evidence for indeterminate (potential) sites
in the southern marl.
332
1
1
3
1
1
1
2
1
1
1
1
1
1
1
1
3
2
3
1
1
1
1
3
3
1
1
2
2
3
1
3
5
2
3
2
2
1
3
3
68
UnitType
Non-site
Count
UnitID
174
185
208
215
226
227
228
230
231
232
233
234
235
236
239
241
242
245
247
248
316
330
331
335
341
342
343
346
348
349
352
353
354
Grand Total
Evidence source Evidence
Corona 1108 Corona 1110 Corona 1111 Ikonos MS
Soil Colour
Soil Colour Soil Colour Soil Colour
1
1
Ikonos Pan Syrian 1:25,000
Soil Colour
Contour
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Grand Total
Khirbet
1
2
2
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
3
1
1
2
1
1
1
2
1
1
2
1
40
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
9
4
1
4
2
1
17
3
Table 26 Summary of evidence for non-sites in the southern marl.
UnitType
Count
UnitID
14
173
179
191
206
207
210
212
216
218
229
251
254
255
256
264
265
315
484
Grand Total
Tell
Evidence source Evidence
Corona 1108 Corona 1110 Corona 1111 Ikonos MS
Soil Colour
Soil Colour Soil Colour Soil Colour
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
16
1
1
14
Ikonos Pan Syrian 1:25,000
Soil Colour
Antiquity
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
15
1
1
14
1
1
13
Grand Total
Contour
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
15
Table 27 Summary of evidence for tell sites in the southern marl.
333
Khirbet
Tell
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
16
7
7
4
7
6
6
7
7
5
4
2
7
7
7
7
1
2
8
4
105
UnitType
Count
UnitID
127
172
177
178
181
184
193
194
196
197
204
211
213
217
221
222
224
238
249
252
257
259
266
267
275
279
280
308
317
318
319
320
328
332
334
336
339
344
345
351
454
459
461
475
477
478
480
486
487
496
498
521
523
524
734
737
Grand Total
Scatter
Evidence source Evidence
Corona 1108 Corona 1110Corona 1111 Ikonos MS Ikonos Pan Syrian 1:25,000
Soil Colour
Soil Colour Soil Colour Soil Colour Soil Colour
Antiquity
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
33
Khirbet
Place Name
Tell
Um
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
Grand Total
Contour
1
1
1
1
1
28
37
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
48
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
20
8
8
21
5
1
2
4
5
6
3
5
7
7
5
6
7
3
4
1
2
6
1
5
3
3
3
1
6
5
1
2
6
4
6
4
7
5
5
2
3
4
4
6
4
5
2
5
1
4
4
3
2
2
2
2
3
1
1
3
4
4
2
211
Table 28 Summary of evidence for scatter sites in the southern
marl.
Tables 25 to 28 describe the Desk Based Assessment (DBA) evidence that was used to
identify tells, scatters, non-sites and indeterminate (potential sites) in the southern marl only.
Given the continuous nature of the residues in the basalt zone the same quantitative
approach cannot be undertaken. In each of the tables sites are cross-tabulated against
evidence. This allows the comparative evaluation of different evidence types at each site.
Table 29 summarises these tables into evidence types. From the sites visited to date the
satellite imagery produces by far the most accurate results.
334
Table 27 describes the evidence for tell sites. There is a good correlation between all the
satellite imagery and tells and as would be expected there is a good correlation between place
name and contour evidence. Table 28 describes the evidence for scatter sites, which are less
obvious on the ground than tell sites. In this instance there is a reasonable relationship
between mapping evidence and sites visited (62.5%). However, there is a very high
correlation between satellite evidence and sites visited (92.9%).
It should also be noted that most of the non-sites from the Syrian mapping are derived from
contour features (see Table 26). Of the 56 scatter sites positively identified contours have
only been evidence for 8 sites (see Table 28). If one assumes that all tell sites have been
identified during the course of the survey (they are, after all, the easiest residue to detect) then
the remaining contours can only refer to scatters. Therefore, extrapolating from Table 26 and
Table 28 then statistically only 25% of these contours will be scatters. From the
indeterminate (potential sites) evidence (see Table 25) 19 of these are based on contour
evidence. This means that only 4 of these are likely to actually be sites. To make matters
worse in Table 28, with the exception of 1 site, every site with contour evidence has
supporting evidence from satellite imagery. Of the remaining 19 only 4 are supported by
satellite evidence. As would be expected the place name evidence and antiquity symbols were
very good indicators of residues. However, it should be noted that for place name evidence
in particular, the spatial accuracy on the maps is very poor. For example place name evidence
can up to 5km away from the actual location of a site. 8 indeterminate (potential sites) have
associated place name or antiquity symbol evidence. Unfortunately only 2 of these are
associated with contours. Therefore, when the southern marl has been fully surveyed one can
expect the accuracy of the mapping resource in Table 29 to decrease dramatically.
Evidence
Satellite
imagery
1:25,000
mapping
Not Visited (potential)
Not Sites
Scatters
Tells
Total Sites
Total Sites Visited
% Accuracy from sites visited
20
14
52
17
103
83
83.13%
25
20
35
16
96
71
71.83%
Table 29 Summary of results in the southern marl by method.
335
Combined
Satellite and
Mapping
37
33
56
19
145
108
69.44%
From the satellite imagery the Corona mission 1110 has produced the most non-sites. In
many respects this is to be expected as this is the poorest quality imagery. Of all the sensors
the Ikonos multispectral correctly highlights the largest number of residues. In general,
however, the scatters and tells are well represented by all sensors.
However, this should not be taken solely at face value: it is also important to establish if the
results are representative of the full range of expected archaeological residue types. Given
that the total data structure is unknown this is a difficult task. However, it can be appraised
theoretically and empirically. From an empirical standpoint the landscape has had a
programme of ‘off-site’ sample transects (see sections 3.5.1.4 and 3.5.1.5) located across all
zones. Amongst other things, it is hypothesised that some of these transects should locate
archaeological residues that have not been identified from the satellite imagery. From these
transects, not including the alluvial zone, only three sites were identified that were not
observed on the satellite imagery. Of these two were subsequently identified on the imagery
after ground location. Hence, there is an excellent correlation between the number of extant
physical surficial residues and the residues identified by satellite imagery.
Theoretically, however, the techniques do exhibit some serious bias. With a few exceptions,
the vast majority of sites have both sedentary architectural and pottery traditions. Hence, the
vast majority of the residues are post-Neolithic. Therefore, the majority of pre-sedentary
human inhabitation has not been identified, although this does assume that inhabitation
actually occurred at this time. Furthermore, no evidence has been identified for transitory
sites (such as non-modern Bedouin camps). As discussed by Wilkinson (2001 p. 531)
archaeological residues derived from sedentary occupation are easier to detect than those of
nomadic pastoralists. The impact of nomadic communities on the landscape is slight and any
residues are easily eradicated or masked by post-depositional processes (especially ploughing
and deflation). Hence, it is not surprising that neither of these residue types have been
identified. Furthermore, the fact that they have not been identified during surface survey
makes it very unlikely that they would be identified from satellite imagery.
Given these limitations the evidence in the application area falls into three categories:
positive, negative and masked.
336
10.1.1 Positive evidence
Positive evidence is the identification of an actual archaeological residue, or the
interpretation, by proxy, of objects that would lead one to assume that archaeological
residues exist. Figure 176 and Figure 178 display the range of positive evidence within the
marl and basalt zones respectively.
In the basalt zone the archaeological residues take the form of site and hinterland structural
evidence (field walls, enclosures and buildings). These are relatively easy to identify on both
the Corona and Ikonos panchromatic imagery. The Ikonos pan and particularly the Ikonos
pan-sharpened imagery provide the best level of interpretation in this environment due to the
higher spatial resolution of the sensor (see Figure 112). However, the Ikonos imagery,
although representative of the present day conditions, does not encompass the same range of
residues as observed in the Corona imagery. This is due to destructive modifications that
have occurred over the past 30 years. Hence, the Ikonos and Corona imagery are
complementary resources that provide a greater understanding of the archaeological resource
and its destructive modifications.
Archaeological residues in the marl zone take the form of discrete settlement sites that are
easy to identify as colour or textural variations in soil. These residues are an order of
magnitude larger than those found in the basalt zone. Hence, there is not such a reliance on
high spatial resolution data. The Ikonos MS and the Corona imagery provide the best
interpretative sources for this zone. The Ikonos MS is a particularly useful resource for
displaying changes in soil colour. Furthermore, there is more scope for interpretation by
proxy in this zone. Sites 279 and 308 in Figure 176 show kinks in the road network where the
road respects the archaeological site. These are useful indicators when interpreting the
satellite imagery. Once again this zone has been subject to a range of landscape
modifications, but due to the nature of the residues few sites have been eradicated. Rather,
deeper ploughing has removed some of the surface textural components in the Ikonos and
brought sub-surface marl deposits to the surface creating a number of potential but negative
features.
10.1.2 Negative evidence
Negative evidence is the identification of features that appear to be archaeological but are in
fact natural features or residues of other processes (see Figure 177 and Figure 178).
337
In the basalt zone no negative features have been identified per se. However, modern field
boundaries have been created in the past 30 years (see Wall type 1 and Wall type 2 in Figure
178) and could, therefore, be classed as negative features.
In the marl zone a range of negative evidence occurs (see Figure 177). The vast majority of
these appear to be military in origin. This can normally be attested to by the regularity of the
features. Recent Bedouin encampments are also observed in the imagery as rectangular areas
(such as around sites 191 and 256 and see Figure 177). This is due to localised compaction of
the soil in the areas where the tents were pitched changing reflectance characteristics. Finally,
it is important to be aware of irrigation marks. These are by-products of the irrigation
systems used in the zone and either form darker areas (through water leakage) or brighter
areas in the NIR (due to localised increased crop vigour). These should not be confused with
archaeological crop marks (cf. GORS 2002 p. 28). In general the vast majority of the negative
evidence is exhibited in the Ikonos imagery. Therefore, once again, the Corona imagery is a
complementary resource for evaluating and understanding these effects.
10.1.3 Masked evidence
While it is useful to consider whether the perceived archaeological anomalies are positive or
negative evidence it is also necessary to delineate areas that are masked and may not respond
to satellite prospection. Masking can take two primary forms:
•
Temporary masking, such as vegetation cover or contrast equalisation after
rainfall
•
Permanent masking, such as burial by sediments or eradication by bulldozing
Temporary masking effects can be resolved by acquiring satellite imagery at a more
appropriate timeframe (i.e. when crop has been harvested or during periods of peak aridity).
However, it is essential that the environmental regime is thoroughly understood prior to
purchasing any satellite imagery where temporary masking may be a factor.
338
Figure 176 Residue image interpretation key: positive features in
the marl.
339
Figure 177 Residue image interpretation key: negative and masking
features in the marl.
340
Figure 178 Residue image interpretation key: basalt zone.
341
Temporary masking is more prevalent in the Ikonos imagery in the marl zone. This is
predominantly due to the increase in irrigation between 1970 and the present day. Since the
Corona imagery was collected extensive irrigation has had two effects: crop cover has been
increased across the zone and increased water content after irrigation has reduced the
contrast between background soil and archaeological residues. This potentially masks more
residues over a longer time frame than observed in the Corona imagery
Permanent masking is a much more serious barrier for interpretation. Where residues have
been eradicated (e.g. through bulldozing in the basalt zone) it may be possible to infer some
elements of the archaeological residues (see Figure 178). However, where they have been
buried or subject to extensive post-depositional processes (such as in the alluvium and
alluvial fan zones) satellite imagery may be of little value for prospection purposes. Optical
sensors only reflect the surface microns of an object, therefore unless a buried residue
expresses some surface anomaly (such as a crop mark) it is unlikely to be detected. Even
RADAR imagery, which can penetrate arid soils (Holcomb 2001), would require high spatial
resolution to detect most archaeological residues.
10.1.4 Image interpretation key
Given the qualitative nature of archaeological residue detection it is unlikely that two
different interpreters will classify a landscape in the same way. Image interpretation keys have
been developed in order to maintain a semblance of consistency (see Figure 176 to Figure
178). They are valuable aids for organising and presenting the knowledge of expert
interpreters. As such, they can be used for training novice personnel or as a general reference
aid (Colwell 1997 pp. 19-27; Campbell 2002 p. 133). The image interpretation keys cover all
aspects of evidence (positive, negative and masked). It is hoped that further collaboration
with the DGAM and other CRM bodies in the region will extend this image interpretation
key to cover different areas throughout Syria. This will be a significant resource for any future
research employing remotely sensed imagery.
10.1.5 Quantitative summary
Although the vast majority of the residues were interpreted qualitatively, quantitative analysis
was undertaken on soil samples in the marl zone. This had the aim of establishing the
physical properties that accounted for increased reflectance in this zone. Although these
analyses are in their early phase, initial results indicate that increased soil reflectance
342
associated with archaeological residues is related to variations in particle size distribution and
soil structure (see Chapter 8).
Changes in particle size impact on a variety of soil properties such as texture, structure and
drainage. However, due to differences in the materials of mud-brick construction and postdepositional deformation processes no single variant has been identified for increased
reflectance. Further research into these variations is recommended, including further analysis
of a number of other soil properties (e.g. in-situ water content and bulk density).
In general, however, no specific spectral signature that responds to archaeological residues
has been recognised during the course of this research. Rather, archaeological residues
produce a localised positive or negative variation in the DN values when compared to the
background reading. This has been demonstrated during the recognition of soil marks in the
marl zone and should be equally true for the application of aerial imagery for crop mark
identification (Powlesland et al. 1997). In essence when prospecting for residues in crop or
soil they can only be identified if they exhibit sufficient contrast against the ‘natural’
background level.
10.1.6 Effects of resolution on archaeological detection
Remote sensing instrumentation provides archaeologists with a range of imagery with
different spatial, spectral and temporal resolutions consisting of numeric data (with different
radiometric resolution). Changes in any of the axes of resolution can lead to different
classifications and interpretations.
The benefit of increasing any axis of resolution is that more objects can be confidently
resolved and classified (Tso and Mather 2001 p. 9). However, increasing any axis of
resolution produces a concomitant increase in complexity, storage size and analysis costs.
This is recognised in hyperspectral imaging systems which produce an extremely large
amount of data within which, for any specific problem, there is a large amount of data
redundancy.
10.1.6.1 Spatial resolution and archaeological detection
Spatial resolution is particularly important for the visual detection of objects. In order to
detect a feature, the spatial resolution of the sensor should be roughly one half of the
feature’s smallest dimension (Jensen 2000). There is, therefore, a positive relationship
343
between the size of the objects under study and the resolution of the imagery required to
identify them. In the case of the basalt zone, where the smallest wall width is c. 1m (see
Figure 108 and Figure 109), a pixel size of 0.5m should be required. However, features were
accurately detected by Corona with a 2m spatial resolution, although the higher resolution
Ikonos imagery gave a much more informative digital product. In part this can be explained
by shadows cast by the wall increasing its perceived ground footprint. Further, the
combination of shadow and wall may produce a signicant increase or decrease in local
contrast which predominates in the reflectance recorded by the sensor. Even so, only the
highest spatial resolution sensors are appropriate (i.e. with a ground pixel size ≤2m) within
this zone. It will be very interesting to evaluate panchromatic Quickbird imagery in this
environment as a 0.6m pixel size is very close to the postulated 0.5m required.
In comparison to the basalt zone the residues in the marl are much larger. Thus lower
resolution sensors are suitable for detection. In this context pan-sharpened Landsat imagery
(15m) did appear to highlight archaeological residues more accurately than 30m Landsat (see
Figure 119). Therefore, sensors with a spatial resolution of 3-15m would seem to be most
appropriate for prospection in the marl zone. Obviously, higher resolution sensors would
make visual interpretation much easier. However, at certain scales higher resolution sensors
appear to increase heterogeneity which makes quantitative detection more difficult.
Furthermore, unlike some European examples, increased spatial resolution has not led to a
higher level of understanding (i.e. internal artefacts have not been identified leading to feature
recognition or interpretation). The decreased spatial resolution required for this zone means
that a number of cheaper satellite sensors are appropriate for detection (e.g. SPOT).
10.1.6.2 Spectral resolution and archaeological detection
The benefit of increasing spectral resolution is that from the comparison of two or more
images, from different parts of the spectrum, we may learn more than studying tonal
variations of just one image (Parker and Wolff 1973 p. 31). Images from different bands can
also be combined to produce false colour composites. These make identification easier by
focussing on aspects of the EM spectrum that exhibit greater contrast for the objects under
study.
The Ikonos MS imagery is the only data set with a multispectral component that has been
fully evaluated. Although this imagery only provided four bands (blue, green, red and NIR) it
344
was a significant resource in all zones. In the basalt zone it was most effective as pansharpened high spatial resolution imagery and as a transparent overlay with the panchromatic
imagery. True and false colour composites improved detection. However, given the
requirement for high spatial resolution in this zone, it is unlikely that new high spatial
resolution sensors will be developed that extend much beyond the near infra-red.
In the marl zone, colour composite combinations were also exploited to great effect. In
particular the red, green and blue bands showed a high correlation with archaeological
residues. The DN variations within these bands at archaeological sites permits a range of
statistical manipulations to improve detection. In both zones increasing spectral resolution
will improve detection. Although not yet available, the use of bands in the short wave
infrared chosen for distinguishing geology or soil (such as Landsat bands 5 and 7) would be
of particular benefit as the ability to discriminate different soil types and conditions could
significantly improve archaeological interpretation.
The NIR band was disappointing for this study. However, this is probably due to the time of
image collection. The Ikonos imagery was collected after the winter rains and while the crops
were growing. The NIR band is sensitive to soil moisture and vegetation vigour. Given the
window of collection the soil moisture contrast, which would be correlated to a difference in
soil structure (discussed in Chapter 8) and hence archaeological residues in the marl, is
reduced. Hence the NIR band has not been fully evaluated in this region. Data collected
under different environmental conditions would be needed to fully evaluate this band. This
was attempted in this programme of work but was found to be too expensive.
Another benefit of the MS imagery is that it is much easier to identify clouds. It is very easy
to misinterpret small areas of cloud cover as potential sites in the Ikonos panchromatic (see
Figure 136). Furthermore, the near infra-red band is very useful for determining areas of
masking vegetation (which is difficult to identify in panchromatic imagery). For example site
218 is identifiable in some of the Corona imagery. However, it is difficult to detect it in the
Ikonos imagery. As Figure 136 illustrates, site 218 is masked by surface vegetation (the NIR
band is displayed in red).
345
10.1.6.3 Radiometric resolution and archaeological detection
Increasing radiometric resolution improves the recording sensitivity of the sensor. The
benefit of increasing radiometric resolution is that subtle diagnostic variations within an
image, which would otherwise have been grouped with other values, can be differentiated
(see Figure 25 and section 5.5).
In this region no significant benefit has been observed by the increased radiometric
resolution of the Ikonos imagery. A summary qualitative interpretation of the Ikonos imagery
degraded to 8 bit (with a minimum-maximum rescale) displayed the same residues as the 11
bit imagery. In the marl zone there is no improvement in detection (see Figure 87). In the
basalt zone the improved radiometric quality of the Ikonos may aid in the detection of wall
components. However, it is likely that the improved spatial resolution is more important. In
this context it would be interesting to compare Quickbird and Ikonos imagery in the basalt
zone. Quickbird imagery has higher spatial resolution (0.6m pan and 2.4m MS) but a lower
radiometric resolution (8 bit). Evaluation of Quickbird imagery in a different environmental
context (Beck et al. in prep) has shown that it may be susceptible to saturation.
Given that the 11 bit radiometric depth of the Ikonos imagery is of limited value for
archaeological interpretation in the application area the raw imagery can be rescaled to reduce
file size with little loss of interpretative detail. This would reduce the storage requirements for
the Ikonos imagery by a factor of 2. As discussed in section 5.5, a standard deviation rescale
would maintain most of the dynamic range present in the original 11 bit imagery. Space
Imaging offer 14 different combinations of Ikonos imagery (panchromatic, multispectral and
merged) with different radiometric image depths, although the documentation is unclear
about the rescaling technique used. The lower radiometric imagery does not, at present, come
with a cost reduction. In the future it would be hoped that this degraded imagery would be
retailed at a premium discount.
It is difficult to discuss radiometric resolution in the context of Corona but, as outlined in
section 7.4, the Corona imagery does not highlight all the residues in the basalt zone that
were observed in the Ikonos imagery. This may be due to variations in the radiometric
resolution between the two sensors or unidentified modifications in the intervening years.
However, given that the film may have deteriorated over thirty years, that it is scanned and
that the images were collected at different times of day and year this is impossible to verify.
346
10.2 Thematic extraction summary
The analysis of co-registered Landsat and Ikonos imagery allowed a number of thematic
layers to be extrapolated at different scales. The Ikonos imagery was essential for creating a
number of high resolution land cover themes. In this context the MS imagery was important
for the recognition of different vegetation and hydrological components.
At a broader scale the Landsat imagery provided a range of information pertaining to larger
scale soil, geological and other environmental components. Of particular import was the
United States Geological Society (USGS)-based interpretation schema which provided a
common framework for both analytical systems. Furthermore this allows the incorporation
of contiguous data collected by other researchers. It is strongly recommended that any future
researchers use this, or another, co-ordinated approach for environmental identification.
Both the land cover and soil mapping provided essential contextual data which was
unavailable at this scale from any other source.
10.3 CRM application summary
In addition to the utility of satellite imagery for residue prospection it provides a number of
benefits for other CRM applications. The imagery itself is a complementary resource to
traditional mapping. It can provide important contextual information which, through the
process of generalisation, has been removed from cartography. Satellite imagery also lends
itself to a number of practical field applications. The convergence of technological tools
(such as handheld computers and GPS) facilitates the use of satellite imagery for navigational,
mapping and interpretative purposes in the field. Satellite imagery is also a useful presentation
tool. Not only does it provide a different representation to traditional cartography; it can also
be incorporated easily into more immersive digital presentation environments. Finally multitemporal imagery can have a significant monitoring component.
10.3.1 Impact of landscape change on archaeological residues and site
monitoring
Changes in agricultural techniques, infrastructure expansion and urban sprawl have all had
substantial impacts on the landscape over the past 30 years. More profound landscape
changes have probably occurred during this period than at any other time. Analysis of the
Landsat, Corona and Ikonos imagery has produced a much greater understanding of the
347
types and extent of these modifications. These types of change are rarely recorded and it is
difficult to ascertain if this important information would have been available from any other
source.
High resolution Corona and Ikonos imagery allows archaeological residues themselves to be
monitored. Natural processes have impacted on a number of sites (such as erosion and burial
of sites by Lake Qatina) and the imagery has allowed a coarse quantification of this rate of
change. This information may help in protecting sites which are at significant risk. Cultural
processes have had an extensive impact in the landscape, particularly bulldozing in the basalt
zone. The satellite imagery has helped identify the extent of this destruction and was pivotal
in defining a zone which was placed under protection by the DGAM.
10.4 Limitations of archaeological interpretation from satellite imagery
Although satellite imagery has provided significant benefits for the evaluation of the study
area there are limitations to its application. From an archaeological perspective in this
environment satellite imagery only allows the ability to confidently detect residues. Recognition
and interpretation of residues is complicated (with the exception of tell sites). Hence,
archaeological residues in the marl zone have been interpreted as tell or flat sites and residues
in the basalt zone as structures, enclosures and walls. However, in the basalt zone it is
possible to examine the stratigraphic relationships of the field systems to establish a relative
chronology (i.e. the use of the t-junction rule). Therefore, although areas of high
archaeological potential are located, without ground observation the analytical and research
value is limited.
10.4.1 Crop mark identification
Considering the reliance on crop mark identification in European contexts, there has been
very little comment on the use of crop marks for the detection of archaeological residues in
this application area. As discussed in section 5.2.4, it is unclear whether crop marks actually
occur. Furthermore, the likelihood of being able to programme a collection sequence which
intersects with crop mark appearance is low. This is further exacerbated when one considers
the large footprint of satellite imagery and variations in crop ripening within that footprint.
There is a much greater window of opportunity to collect imagery which will show soil
marks, and soil marks appear to be representative of the bulk of archaeological residues.
348
Therefore, unless specific research questions are to be addressed then satellite imagery should
not be purchased with the sole aim of crop mark detection in similar environments.
Resolution
Sensor
Corona
Ikonos Pan
Ikonos MS
Ikonos Pansharpened
Landsat TM
Landsat ETM+
Zone
Spatial (m)
Spectral (bands)
Basalt
Marl
Alluvium
2
1
4
1
1
4
Medium-Good
Good
Medium-Poor
Good-Excellent
Good
Good-Excellent
Medium-Poor
Poor-Medium
Poor-Medium
1
30
15
4
7
9
Excellent
N/A
N/A
Good-Excellent
Poor
Poor-Medium
Poor-Medium
Poor
Poor
Table 30 Sensor summary by zone.
10.5 Recommendations
This research programme has successfully applied a number of satellite sensors for landscape
archaeological prospection and analysis. Table 30 summaries the utility of each sensor within
each environmental zone.
What follows are recommendations for researchers wishing to apply satellite imagery to their
own projects in similar environments.
Prior to purchasing any imagery it is essential that the nature of the environmental zones and
the archaeological residues are understood and that a Desk Based Assessment is undertaken.
This information can be contextualised by one of the number of free satellite images available
over the internet (for example Landsat imagery at the Global Land Cover Facility in
Maryland: http://glcf.umiacs.umd.edu/index.shtml). A preliminary field visit is required to
record the following environmental, archaeological and background information:
•
The type and extent of different environmental zones.
•
Surface cover in each zone.
•
Agricultural seasons in each zone.
•
Extent of irrigation.
•
Rainfall average per month.
•
Atmospheric variations over the year.
•
The nature and extent of archaeological residues in each zone.
349
•
Establish a range of rectification points by GPS on features that are likely to be
identifiable on present day and historic imagery.
Regarding archaeological residues, it is particularly important to understand how their
contrast changes against any ‘background’ readings during different environmental
conditions. For example, in the marl zone residues exhibit greater contrast during periods of
peak aridity (i.e. the soil colour difference between sites and soils is at its maximum). After
rainfall and when under crop this contrast is reduced or eradicated.
This information can then be integrated within a GIS with the DBA data (including geology,
present day, historic and archaeological CRM mapping and aerial photographic archives).
This allows the project team to evaluate if satellite imagery is required and if so of what type
and within which time frame.
When deciding upon the spatial, spectral and temporal resolution of the sensor for
archaeological residue detection and landscape evaluation one should understand the nature
of the residues to be encountered and the level of identification that one is hoping to achieve.
Inevitably this will mean that a range of different sensors are appropriate for a survey: for
example the following sensors could be employed in this study area:
1. Low spatial (>100m) and high spectral (>10 bands) resolution imagery for coarse
landscape identification (particularly soils and geology).
2. Medium spatial (10-60m) and medium spectral (> 6 bands) resolution imagery
for refined landscape identification e.g. Landsat ETM+ or equivalent.
3. High to medium spatial (2-15m) and low to medium spectral (>3 bands)
resolution to detect larger features (ploughed out sites, tells etc.) e.g. Quickbird
MS, Ikonos MS or SPOT 5.
4. High spatial (<1m) and low spectral (pan) resolution imagery to detect very small
features (walls, linear soil marks, pits, postholes etc.) e.g. Quickbird pan or
Ikonos pan.
Wherever it exists historic satellite imagery should be purchased, even if it turns out to be of
nominal value. The purchase and evaluation costs are minimal and it has the benefit of being
collected prior to many countries adopting deep (tractor) ploughing, extensive irrigation and
other heavy earth moving equipment, each of which has profoundly altered the present day
350
landscape. Archive high resolution commercial image sets should also be consulted.
However, if the commercial archives fall outside the required time frame for image collection
they should be discounted as this imagery is only available at a small discount.
If bespoke or archive high resolution imagery is to be purchased then an evaluation sub-set
of imagery could be procured prior to collecting the whole application area. This will allow
the full utility of the imagery to be evaluated without committing to costly imagery that may
not provide the desired results. However, this could have the disadvantage that it may take
over one year to acquire the full data set. For this research, the purchase of £21,770.26
(excluding VAT) worth of inappropriate Ikonos imagery would have been disastrous. Finally,
other avenues of acquisition should also be pursued. For example purchase costs can be
shared between a number of projects or institutions working in the same geographical locale.
Furthermore, none of these data sets need to cover the whole application area. If the
environmental conditions and the reflectance characteristics of the objects of interest are well
understood then one can discriminate where imagery should be purchased. Significant
economies can then ensue (see section 10.5.2).
From a detection viewpoint this research has defined a number of different quantitative and
qualitative techniques. However, the techniques used in this region may be inappropriate in
other areas. Fortunately, the majority of the techniques used rely on the detection of residues
in the visual wavelengths. Hence, it should be relatively easy to create bespoke analytical
systems in similar environments. To ensure long-term utility of the imagery, it is
recommended that image interpretation keys are produced for each environmental zone and
disseminated to the regional and national CRM bodies.
10.5.1 Issues of implementation
As with any development the cost benefit of the application needs to be justified. The
organisational issues of implementation should also be addressed as there can be significant
barriers to success if key stakeholders are excluded during any phase of implementation. This
phase should also cover appropriate software purchase and training. Any benefits should be
articulated through the project information systems. The techniques employed in this
application area may require re-appraisal for individuals wishing to test the methods in other
environments. Finally, copyright requires addressing: archaeologists have a long history of
351
sharing information. However, the stringent copyright attached to satellite imagery may
reduce its full research potential.
10.5.1.1 Cost
As with any project the cost of purchase is of over-riding import. From a cost basis the
Corona photography is the cheapest satellite source to purchase. However, it does have
processing costs. Landsat is reasonably cheap for archive imagery and is becoming
increasingly available at no cost through certain websites, although the environmental
suitability of this imagery is generally reduced. High resolution imagery is the most significant
financial investment. The launch of the Quickbird satellite increased competition within the
market which has generally reduced image acquisition costs.
In general, existing imagery is of course the least expensive to use; the costs of acquisition
have been met by others and copies of the results can be obtained at a discount. However,
re-using imagery collected by a partner may have copyright limitations. The most costeffective mechanism for imagery collection is through a strategic partnership purchase with
other bodies.
Costs can only be fully evaluated against any benefits that ensue. The costs of purchase for
the satellite imagery in this research at c. £30,000 (including VAT) equate to approximately
three seasons of fieldwork funding. Most projects would consider this an extreme expense
and it could be argued that only the most visionary funding bodies would support such a
proposal. However, without the purchase of this imagery at an early stage, it would take many
seasons of fieldwork to produce a commensurate archive. Therefore on a pound for pound
research value the satellite imagery has proven to be extremely cost effective. If the
recommendations outlined in section 10.5.2 are followed then the purchase cost is only c.
£15,000 (including VAT) providing more significant returns on investment.
10.5.1.2 Copyright
The last ten years has seen a revolution in remote sensing technology and management.
Advances in sensor resolving characteristics have occurred, the industry is deregulated
(particularly from the dominance of NASA) and the majority of hard and software analysis
systems have matured (Slonecker et al. 1998). Furthermore, the internet has facilitated the
rapid dissemination of data (raw and classified), method and synthesis. The shift from
352
research led to commercially driven Earth observation has had a significant impact on the
industry: particularly in the area of copyright. Commercial data is generally subject to more
stringent copyright than research data. The nature of this copyright varies between the
different commercial organisations, the sensors and the age of the imagery.
Onerous copyright may become the major inhibiting factor in the application of satellite
imagery for archaeology. Archaeologists are great borrowers and the discipline has benefited
from techniques and data imported from other researchers and related fields of study. It is
essential that mechanisms are sought whereby archaeologists, and other researchers, can
share remote sensing data without breaching copyright.
One such method is to purchase data for an organisation rather than a sub-division (in this
instance I am considering an academic framework, whereby the data is licensed to a
University rather than to a Department). However, this does not make the data available to
peers in other establishments. It is recommended that umbrella organisations (such as
NERC) could be a central conduit for data purchase, making the data available to national or
even international researchers. Whatever mechanism is chosen it is essential that
archaeological researchers have affordable access to appropriate remote sensing resources.
10.5.1.3 Rectification and Co-registration
Image rectification and co-registration are vital stages in image preparation. For this research
Ground Control Points recorded by handheld GPS were used to re-rectify the Ikonos
imagery which in turn was used as a basemap for the co-registration of all the other satellite
sensors. It is likely that differential GPS will be employed for the collection of GCPs in most
other areas due to its improved positional accuracy. Whichever GPS technique is used points
should be collected so as to satisfy statistical rigour and future satellite systems or
declassifications. From a statistical viewpoint many individual readings and their metadata
should be recorded at each control point (as opposed to the averaging technique discussed in
section 5.4). This will allow more rigorous statistical procedures to deduce the most accurate
position using, for example, least squares techniques. Further, these points should be as
accurate as possible: there is very little point in conducting a GCP survey with only one
image product in mind as the GCPs will have utility for the correction of future images with
different spatial and spectral characteristics.
353
10.5.1.4 Use in other geographical locations
The re-application of the techniques described in this research in other geographical locales is
particularly important. Changes in environmental conditions will change how residues are
observed by sensors but will also structure the natural and cultural formation and
deformation of the residues themselves. In this instance it is advised that the
recommendations outlined in sections 10.5.1 and 10.5.2 are followed.
10.5.1.5 Archaeological information systems and data management
In the wider context multispectral data can be integrated with any other form of spatially
referenced information, particularly within a Geographic Information System, and many of
the image enhancements methods applied to multispectral data can be used with suitably
transformed, i.e., digitised, conventional aerial photography. Therefore multispectral remote
sensing has the potential to become a fully integrated and routine technique in archaeological
research. At present it is neither.
(Shennan and Donoghue 1992 p. 224)
As discussed by Parrington (1983 p. 107), Shennan and Donoghue (1992 p. 231) and
Palumbo and Powlesland (1996 p. 127) satellite imagery and other remote sensing resources
will play an increasingly significant role in the management of cultural resources. However, in
isolation these resources will have limited capacity. It is much more beneficial for
archaeologists to integrate remote sensing data with other resources, such as ground survey
and other ‘monuments’ records, through an information system. GIS and image processing
technology will be essential as will the collection and, more importantly, the database
structuring of attribute data so that it can be used to address research problems. The web will
have an important part to play for the future of archaeological data analysis and management.
It is proposed that a full web-GIS data entry, visualisation and analysis system is created for
this archive. Researchers could articulate the full archive within a web browser from
anywhere in the world. Furthermore, it would mean that all users are querying the most upto-date data at any one time. Developments such as this should significantly improve the
mechanisms archaeologists employ to analyse the landscape and should thus impact upon
theory and practice.
354
10.5.1.5.1 Software and training
However, these benefits come at a cost: they can only be realised if the appropriate software
packages have been purchased and effectively integrated and if the project personnel have
been taught how to use them. In a research project such as this it is difficult to advise on
specific software packages as the situation changes very rapidly. Furthermore, some packages
are better than others at certain tasks. However, one thing can be guaranteed and that is the
continued functionality convergence between GIS, image processing software and database
systems. In this environment it is more important that the data formats employed are
interoperable (i.e. they can be used in any software package and on any platform without loss
of data content). Appendix I defines the structures used in this project and although the data
structures are not fully interoperable they are one processing step away from such a position.
Furthermore, the system has been designed so that it can be easily archived within the
Archaeology Data Service.
Training is a separate issue and once again it is difficult to determine where the responsibility
for training should lie (the individual, the project, a teaching institution or the professional
body (through continuing professional development)). The Institute for Field Archaeologists
has highlighted that computer skills, geophysics and remote sensing will play an increasingly
important role in the future of archaeological practice and are currently under-resourced
(Aitchison and Edwards 2003). It is hoped that the professional archaeological body can
increase pressure to improve training in this sphere.
10.5.2 Impact of the recommendations on the application area
The following section re-evaluates theoretically the decision-making process of the SHR
project in light of the above recommendations. After the preliminary field study it would
have been established that:
•
The application area consists of three environmental zones (basalt, marl and
alluvium).
•
Aerial photographic, geological and land-use mapping resources were minimal or
at inappropriate scales.
•
The 1:25,000 mapping, although appropriate, was in an undefined projection.
355
•
The CRM dataset only contained monumental archaeological residues with a bias
towards the marl zone.
•
The residues in the basalt zone are a palimpsest of complex structural
components and would require high resolution (< 1m) imagery.
•
The residues in the marl zone are discrete tells or ploughed out sites. Ploughed
out sites exhibit a distinct soil colour change in comparison to the background
soil. This soil colour varies with moisture content.
•
The alluvial zone contains a range of archaeological residues. It is difficult to
establish the extent of residues in this area due to a range of post-depositional
processes.
•
The main agricultural season is between December and May, although a second
crop may be realised in irrigated areas (mainly in the marl).
•
The majority of rainfall occurs between December and April.
This background information provides the framework from which the application of the
remote sensing programme must be structured (see Figure 179).
From a thematic perspective multiple sensor systems can be used to establish a range of
layers that define the different environmental zones, geology, soil types and land use. Soil
variations in particular appeared to be important for the detection of archaeological residues.
In this context a sensor with high spectral resolution and coarse spatial resolution to define
broad geological, soil, vegetation and urban zones. Two images should be acquired
encompassing low crop and peak aridity (October-December) and high crop (April-May)
This low spatial resolution resource can be augmented by the incorporation of higher spatial
resolution Landsat imagery. Further time change analysis should be conducted employing
imagery from different seasons and years. This will elucidate the impact of cropping, rainfall
and other landscape modifications over time. Higher spatial resolution imagery can be
integrated into this framework to provide even greater recognition or identification. This is
the exact form of framework employed when integrating the Ikonos imagery with the
Landsat imagery for the land cover classification (see section 6.2). It should also be advised
that one of the Landsat images should be an October Landsat 7 scene so that the 15m band
can be evaluated for residue prospection. The cost implication of this phase (2 low cost high
spectral and low spatial resolution scenes (for example ASTER) and 4 Landsat (2 at January356
May and 2 at July-November with a 10+ year gap) would be c. £100 for ASTER and £1350
for Landsat.
Figure 179 Revised sequence for the incorporation of satellite
imagery into an archaeological landscape project.
From a residue prospection viewpoint, different sensors would be required for the different
environmental zones. The marl zone requires high but not the highest spatial resolution
sensors. A multispectral image focused on the visual wavelengths is much more appropriate
for detection in this zone. Appropriate sensors include SPOT 5 (5m panchromatic and 10m
MS), Ikonos MS and Quickbird MS. This imagery should be collected between October and
December. The cost implication for this imagery, excluding VAT, would be £5700, £6000
357
and £8333 (for SPOT (Orthorectified, 3600 km2), Ikonos (Geo, 500 km2) and Quickbird
(Standard, 500 km2) respectively). In this instance SPOT 5 imagery would appear to be most
appropriate. Not only is it the cheapest, it also has the largest ground footprint, greatest georeferencing accuracy and would improve land-cover identification.
The basalt zone would require the highest spatial resolution imagery. As multispectral
imagery would also be useful in this zone then either co-collected Ikonos or Quickbird pan
and MS imagery should be purchased for the whole 133 km2 basalt zone. This imagery
should be taken in December or January to coincide with minimal crop cover and maximum
visibility of the basalt structures after the first rains. For Ikonos or Quickbird imagery this
would have a cost implication of £2660 (excluding VAT).
In addition, archive satellite photography would be incorporated. All scenes with a high
spatial resolution sensor and low cloud cover index should be purchased. Since the inception
of this research programme the KH-7 missions have been declassified (with ground
resolution of between 0.6 and 1.2m).
Hence a range of complementary satellite imagery has been purchased and processed that
fulfils the diverse goals of the landscape survey (see Figure 179). The full cost of the imagery
would be c. £15,000 (including VAT). This compares very favourably against the c. £30,000
spent during the current research programme.
10.6 Future systems
Inevitably sensor systems will improve in the future. In the short to medium term there will
be improvements on spatial resolution and possibly on spectral resolution. It can be hoped
that multispectral satellite sensors will be produced that have more bands located in the infrared so that differences in crop vigour and soil can be analysed in more detail leading to
improved archaeological detection and interpretation.
Hyperspectral systems and LiDAR have been referred to in passing throughout this research.
These sensors are discussed here in more detail as they have immense potential for the future
of archaeological remote sensing.
As has been discussed, electromagnetic remote sensing techniques in archaeology take
advantage of differential responses of vegetation or soil which are affected by chemical,
358
physical and topographic traces of archaeological activity. These traces produce different
responses in different portions of the electromagnetic spectrum. This research has exploited
the different characteristics of satellite imagery by analysing the components of each spectral
band through radiometric, spatial and spectral manipulation. The benefit of hyperspectral
systems is that more finer spectral resolution bands are recorded (see Figure 22). Under
certain environmental conditions cropmarks (stressed or vigorous vegetation), thermal
anomalies (expressing different emissivity characteristics) and soil variations may be analysed
in much more detail than with multispectral and panchromatic systems. The benefit of
hyperspectral systems is that the small window of opportunity which is used to detect these
residues in the visual part of the spectrum can theoretically be substantially enhanced in other
parts of the spectrum. Computer enhancement of this data places less reliance on specific
environmental conditions to reveal archaeological anomalies (Donoghue 1999; 2001). Hence
environmentally extreme conditions are not required to detect archaeological anomalies.
Amongst other things, aerial LiDAR sensors are used to create high resolution digital terrain
models. Holden et al. (2002) have demonstrated the utility of LiDAR DEMs for
archaeological detection and recognition. One of the major benefits of LiDAR is that the
high resolution DEM can be manipulated in a number of ways within image processing and
GIS software. For example, many aerial photographic techniques rely on discriminating
shadows cast by subtle archaeological topographic disturbances. The light source for these
observations is fixed by the position of the Sun. This means that only objects perpendicular
to the Sun’s direction will be in full shadow. Virtual reconstructions allow the Sun to be
placed anywhere in the sky allowing non-natural illumination conditions to occur. Holden et
al. (2002) exploited this virtual technique to improve archaeological detection and
interpretation.
Specialised sensors and techniques could be focussed on specific topographic, soil and
vegetation responses during the appropriate environmental conditions. This may start to
structure future archaeological research agendas by discriminating between topographic, soil
and vegetation response variations. As a trivial example hyperspectral bands that correlate
with crop stress or crop vigour could identify a range of ‘new’ crop marks in environments
which are currently marginal for aerial photographic prospection.
359
CHAPTER 11 CONCLUSIONS
Remote sensing has broad multidisciplinary applications to geology, geomorphology, biology,
pedology, hydrology and climatology, as well as to anthropology…The remote sensing
perspective provides not only the synoptic overview otherwise unobtainable, but more
importantly, a synergistic grasp of obscured physical and cultural phenomena.
(Lyons and Avery 1977 p. 53)
11.1 General discussion
Satellite imagery offers immense possibilities in terms of its deployment in the context of
archaeological survey and Cultural Resource Management (CRM) applications. This is
particularly the case in areas where the traditional desk based resources of archaeological
catalogues, cartographic and air-photographic data are sparse or difficult to access. The
experience, of working in western Syria, suggests that the critical element is 'fitness for
purpose', that is, the employment of satellite imagery at appropriate scales and in appropriate
environments.
Imagery has the potential to provide critical information on both present and past
environments, the ability to assist in the location of concentrations of human activity and to
provide bespoke thematic mapping. When integrated within a Geographical Information
System, which contains other appropriate data, satellite imagery provides a multifunctional
analytical tool.
In environments such as Syria it is inevitable that aerial photography will be available for
most if not all areas. This photography is normally collected for topographic or military
reconnaissance purposes. However, it is rare for this resource to become available as it is
normally classed as militarily sensitive. It was only after 6 seasons of fieldwork that samples
of the 1958 Russian photographs were made available. If these had been made available at
the initiation of the research then a significant amount of fieldwork would have been
required in order to accurately geo-correct these images. This is due to the small footprint of
360
the photography and the lack of any appropriate geo-referencing base. Hence, the utility of
these images is increased by the ability to rapidly co-register them against a satellite backdrop.
11.2 Satellite imagery as a complement to landscape archaeological survey
High resolution satellite imagery has immense potential as a complementary tool to frame
landscape survey. Traditional applications tend to employ a battery of ground focussed
survey techniques such as fieldwalking and geophysical survey (Banning 2002). Larger scale
interpretations only accrue after many years of intensive fieldwork. Many of these are aligned
to the theoretical premise that the landscape is a blanket of archaeological residues of varying
intensity (see section 3.5.1.2). Alternatively, some landscape surveys tailor their fieldwalking
based upon assumptions about human settlement. Many of these surveys use proximity to
water as their main criterion for site location (for example Adams 1981 p. 28). Satellite
imagery provides an effective tool with which to evaluate rapidly these assumptions prior to
entering into the field.
Ground survey has demonstrated that the residues in the marl zone are islands of discrete
settlement loci with very little archaeological material in between. This equates to the
settlement class ‘nucleated 1a’ as discussed by Wilkinson et al. (2004 p. 191). In this
environment the satellite imagery has framed the survey programme by locating these ‘peaks’
of archaeological activity. Hence, resources can be efficiently deployed during field seasons
resulting in improved modes of data collection and analysis.
By contrast the basalt zone does exhibit off-site scatters. However, the complex relationships
between the structural palimpsest and the artefact scatters currently makes interpreptation
difficult. The satellite imagery has also helped frame the survey of the basalt zone by allowing
the rapid recording of the nature and extent of the structural evidence. The stable internal
geometry of the Ikonos imagery allowed its re-rectification providing high ground precision.
Due to this accuracy individual structural components can be digitised with confidence and
subsequently visited, using GPS, when further attributes could be recorded. The significance
of this simple rectification procedure to the accuracy of the subsequent mapping should not
be underestimated. If differential GPS were permitted for the collection of ground control
points the positional accuracy could be improved to the sub meter or decimetre level.
Traditional ground survey using a total station or differential GPS would have produced a
more spatially precise representation of this landscape. However, the costs involved in
361
conducting such a survey would have been prohibitive. Equally, aerial reconnaissance would
have been costly in terms of the flight and the extensive rectification required.
The contrast in the environmental zones, the nature of inhabitation in these zones and the
consequence for the structure of the archaeological residues provides a good framework for
evaluating the satellite imagery. The satellite imagery has framed interpretative approaches
and survey methodology across the study area. This is in part due to selecting imagery from
seasons which make the identification of archaeological residues easier. If residue location
occurred only through fieldwalking then only a small portion of the 600 km2 area would have
been covered in any detail and the project would still be in its preliminary stages. Many
surveys still use this approach based upon assumptions about archaeological residue
dispersal. One should not assume that continuous patterns of artefact scatters are always
present, as is sometimes assumed in Mediterranean survey (Cherry et al. 1991; Alcock et al.
1994; Knapp 1997; Bintliff 2000; Francovich et al. 2000). However, it must be stressed that
this can only occur in environments where the archaeological resource is distributed in this
manner: the satellite imagery provide a resource where the project team could question this
assumption at an early stage.
The most fundamental impact is that satellite imagery is an important tool for visualising and
evaluating landscapes. For example, the change in perspective, large footprint and pseudoobjective nature of the imagery can make satellite imagery a more powerful contextual and
interpretative medium than a map. The ability to articulate all the data directly within the GIS
(see Appendix I) has enabled the project team to rapidly redefine survey objects at a micro
and macro-scopic level in light of new data.
Furthermore, the vast majority of landscape survey models discussed are primarily focussed
on detecting and interpreting archaeological residues on the basis of statistical variations in
artefact concentrations over large areas. The scale and scope of such operations correspond
closely with many applications of satellite imagery. Archaeological aerial photography, on the
other hand, is normally an unsystematic tool that locates archaeological residues over small
areas.
However, in order to define what form of sampling is appropriate one needs to understand
the general spatial and temporal attributes of the population. This application area was
362
unsurveyed, hence, little was known about the nature of the archaeological population.
Satellite imagery has the inherent property of being able to provide synoptic observations
with high sampling density of relatively large areas. This in itself is an improvement on
traditional aerial photographic techniques. This also means that satellite imagery can be
classed as total survey with high intensity. In this respect the scales of analysis between
satellite imagery and landscape archaeology are complementary. Thus, satellite imagery can be
considered as a landscape sample. This assumes the remotely sensed measurements may be
less accurate at a given point, but, because of the great number of observations, may produce
greater accuracy over large areas (Salomonson 1983 p. 1497). This allows the imagery to be
analysed statistically in order to extract information of interest. Hence, the detection and
interpretation systems employed by landscape survey archaeologists and aerial archaeologists
have relevance when interpreting satellite imagery. Natural and cultural formation and
deformation processes are contained in the structure of the satellite imagery (i.e. the total data
structure see section 1.4). The challenge for the image interpreter is to extract the relevant
components. Logically this will employ variations on the methodologies employed in site and
non-site modelling, as satellite imagery does not immediately define the temporal dimension
during interpretation.
In conclusion, in this environment satellite imagery provides a reasonably high level of
residue detection and thematic content which, due to the relatively large footprint of the
imagery, can be used to frame subsequent research enquiries and field techniques. It must be
stressed that the full benefits of satellite imagery can only be realised if it is an integral
component of a field observation programme. This may appear self evident, however even
after 100 years of use aerial photographic techniques are still not routinely employed in
conjunction with excavation (Aerial Archaeology Committee 1983; Palmer 2000).
As a prospection tool satellite imagery has proven to be accurate and cost effective.
However, as was discussed in Chapter 3, there are a number of other techniques that could
have produced the same form of data.
11.2.1 Satellite imagery comparison against aerial photography
Satellite imagery and aerial photography share a common heritage. It is likely that aerial
photography will remain the preferred resource for archaeological applications due its
flexibility of collection (oblique, vertical and at different times of day), large archive, generally
363
higher spatial resolution and enhanced archive duration. However, satellite imagery does
provide some significant benefits over aerial photography. Most satellite imagery is already
geo-referenced, meaning that it can be immediately integrated within a GIS project.
However, it should not be forgotten that improving the rectification accuracy can
significantly enhance this resource. Once orbit is achieved satellite imagery is cheaper than
bespoke photography on an area by area basis (contra Schmidt 2004). Satellite sensors also
exploit more of the electromagnetic spectrum which may yield other archaeologically
pertinent information (aerial hyperspectral systems have yet to be fully evaluated for
archaeological purposes). Satellite imagery can also be acquired for areas where aerial surveys
and archives are restricted. Finally the large footprint of the imagery means that it is a more
representative snapshot of a survey area (Schmidt 2004). Most archaeological aerial
photography is non-systematic: the results from aerial photographic surveys are biased
(Cowley 2002). This is due to selective flight lines, a focus on sites and not on the
surrounding landscape and a reliance on visual applications. Consequently, areas that look
blank from AP interpretation are not necessarily archaeologically sterile. Although satellite
imagery is subject to the same criticisms, the larger sampling footprint can improve
quantitative and qualitative analysis. Importantly, the nature of the environmental and land
use regime dictates if a response will be observed from aerial or satellite sensors.
In reality, however, it is not an either / or situation. Satellite imagery and aerial photography
are complementary resources. Both can provide valuable archaeological insights. The
discriminatory application of both sources allows archaeological problems to be framed in a
number of different ways.
11.2.2 Satellite imagery comparison against other landscape survey
techniques
Ground reconnaissance and particularly survey (such as driving survey, village to village
survey and fieldwalking) are the traditional techniques for locating archaeological residues in
Mediterranean environments. These approaches have many elements to recommend them,
not the least being that archaeological interpretations are often significantly enhanced,
artefactual data can be collected or analysed in-situ (providing information relating to site
form, function, date and duration) and formation and deformation processes can be
understood in greater detail. However, these are expensive strategies in terms of resource
deployment and unless total survey is being conducted (the most expensive option) then
364
inevitably much of the landscape will not be surveyed. Furthermore, it can take many seasons
of survey before it is possible to contextualise the whole landscape over time with these
techniques.
However, remote sensing approaches have very little meaning unless they are augmented
with ground reconnaissance data. The techniques employed by the SHR project means that
most of the field teams are directed to areas that are likely to be archaeologically productive.
This means that the teams themselves spend proportionally more time evaluating
archaeological residues than ‘blank’ areas. The other field teams have been conducting
control survey to assess if the satellite imagery has located the full range of archaeological
residues. Hence, these techniques are complementary and in the semi-arid landscape around
Homs, where the archaeological resource is poorly understood, they are an effective tool.
11.3 Some limitations of archaeological remote sensing
The limitations of satellite based archaeological remote sensing techniques, as discussed by
several researchers (for example Ebert 1988; Gaffney and Stancic 1991), require reappraisal
in light of technological developments. The recent generation of high resolution sensors
allow a number of new archaeological applications to be realised as discussed above.
However, even though technological developments have increased the utility of satellite
imagery there are still limitations.
Some limitations are inherent in the sensor itself: the spatial, spectral, radiometric and
temporal resolutions may impose limits on what can be analysed. Of particular import are
spatial and spectral resolutions. While the spatial resolution of Ikonos and Quickbird imagery
is approaching that of aerial photography they also offer greater flexibility in spectral
resolution. However, as yet these sensors are limited to the visible and near infra-red
component of the spectrum. Airborne hyperspectral systems will allow different wavebands
across the electromagnetic spectrum to be compared for archaeological value.
Human limitations (such as the ability to perceive colours, experience in remote sensing and
archaeological knowledge) can affect any remote sensing applications. Training is essential to
address some of these limitations. Furthermore, the research questions must be
unambiguously stated and the relationship of remote sensing data to this problem must be
defined.
365
Environmental factors impose further limitations on the interpretative capacity of imagery.
Clouds, mist and haze can obscure or degrade the phenomena of interest in any scene.
However, these atmospheric effects affect different sensors in different ways depending
upon the wavelength of collection (for example RADAR is not affected by cloud or
vegetation cover). The ability to detect different archaeological residues is based upon the
ambient environmental conditions: some sites are easier to distinguish as crop marks, others
as soil marks. Identifying the optimal environmental conditions for the research problem may
take many years of experimentation.
The most important point is that any measuring device is dependent upon the conditions
under which it is employed. The failure of a sensor to reach its full interpretative potential in
one situation does not mean that they will always be less than useful. In essence, the
environmental conditions are of paramount importance in order to appropriately interpret
any image set. The optimum environmental conditions for archaeological interpretation will
vary in each geographical locale.
11.4 Classification, classification, classification
Why such prominence on the term ‘classification’. If nothing else, this programme of
research has demonstrated the importance of classification techniques within archaeology
and remote sensing. The development of classification techniques for archaeology and
remote sensing are very different and respond to the needs, aims and goals of each discipline.
Generally, remote sensing applications can successfully classify their material due to a-priori
knowledge of the materials under study. There is a large body of reference material that aids
the remote interpreter to classify materials. The ability of an interpreter to adequately identify
materials is based upon the amount of information required and the resolution of the data
available to them. Conversely archaeologists do not have this a-priori luxury. Objects and
sites are classified in artificial ways that do not necessarily represent how they were physically
or chemically constructed or how they were observed, understood and incorporated into past
societies and life-styles. Theoretically, archaeological entities can be validly classified into a
variety of different groups for the purpose of attempting to understand them from a different
contextual viewpoint (Hodder 1991; Hodder 1999; Lucas 2001).
From a technical perspective there are a wide range of techniques that can be used to model
and classify data. The ‘great borrowing’ of statistical and mathematical techniques from
366
associated hard and soft sciences (Aldenderfer 1987 p. 90) has been given greater validity by
the use of GIS and associated computer technology. However, it is rare for these techniques
to have a methodological or theoretical consistency (Doran 1987 p. 74). However, many
archaeologists see the need for methodological transparency in their modelling exercises (for
example the papers in Wescott and Brandon 2000). It is hoped that this research programme
adds to this body of methodology which will enable flexible and theoretically sound
application of statistical and interpretative technique.
11.5 Summary
Even in the early 21st century many non-archaeologists believe that archaeological enquiry is
predicated by excavation. On the other hand archaeologists are increasingly contextualising
their sites within landscapes. Landscape analysis at or beyond the regional scale provide an
appropriate backdrop to study many archaeological problems and cultural processes
(Wilkinson 2001).
Survey techniques, as discussed by Banning (2002) and Wilkinson (2001), sample the
landscape to evaluate surface and sub-surface archaeological residues employing a range of
techniques. The use of high resolution satellite imagery in this semi-arid environment has
enabled the detection of productive areas for archaeological survey. Furthermore the imagery
itself provides an interpretative backdrop that is easy to integrate and analyse. Top-down
intensive survey has also been demonstrated to be very expensive as each part of an
application area is analysed with the same techniques. This technique is very useful when one
has no prior understanding of the landscape, the distribution of residues and any important
landscape modifications. However, satellite imagery provide a large scale synoptic coverage
of an application area allowing one to focus survey resources effectively saving both time and
money. This is particularly appropriate when one considers that Alcock et al. (1994 p. 138)
recommend that field survey should be guided by the known distribution of archaeological
residues. In environments where the distribution of archaeological residues is unknown,
poorly understood or subject to significant biases then high resolution satellite imagery can
be considered an essential resource for evaluating the landscape prior to conducting surface
survey.
It is expected that the cost of satellite imagery will continue to decrease as the market
expands and as the general utility of archive imagery decreases. Initiatives such as the Global
367
Land Cover Facility website at Maryland where one has access to a vast range of free satellite
imagery (this site includes Landsat imagery for virtually the entire land surface) will promote
the uptake of satellite imagery within a variety of industries. It is not difficult to imagine that
other sites like this will emerge containing more archaeologically pertinent data.
Theoretical agendas now require archaeologists to evaluate the archaeological resource at a
number of different scales and interpret it from different perspectives. Within this framework
high resolution satellite imagery is a significant tool, particularly in environments which are
poorly understood.
The archaeological potential of remotely sensed satellite data is about to grow rapidly.
…..….With these (higher) resolutions it will be possible to not only conduct within-site
investigations from space that will be able to identify individual surface or near-surface
features but also undertake such things as the periodic monitoring of sites for the effects of
erosion or vandalism.
(Kvamme 1999 p. 184)
This prescient quote from Kvamme neatly encapsulates many of the conclusions from this
research. Present day and historic high resolution satellite imagery are now part of the
archaeologists arsenal that offer new insights into our cultural heritage. The next generation
of satellite sensors will, inevitably, offer the archaeologist even greater benefits.
368
APPENDIX I : THE ARCHAEOLOGICAL DATA MODEL AND
ENVIRONMENT
I.1 Introduction
Data modelling is the process of defining the structure, vocabulary, content and environment
to represent information in a digital system. The structure defines the relationships between
individual elements, or objects, in the system; the vocabulary defines the terminology to be
used to describe individual elements; the content defines what is and what is not included in
the system; and the environment defines the specific hardware and software required to store
and manipulate the content.
Figure 180 Schematic of an Archaeological Data Model (after
Martin 1991 p. 55) . Note all arrows perform some transformation.
Figure 180 outlines a schematic for the processes of archaeological enquiry employing a
digital analytical system. The real world is observed and transformed by the process of
recording (or data collection). Where a non-digital collection system is employed this raw
data is then further transformed during the process of data input into the data model. In an
archaeological scenario the raw data is augmented by our current knowledge (this could
369
include theories, syntheses and other data). These data are manipulated (and further
transformed) within the model to create new information. This information is added to our
current knowledge and used to reappraise any theoretical models. This is used to create a
hermeneutic loop between theory, data and synthesis (see Figure 181).
Figure 181 Hermeneutic loop of archaeological enquiry.
The model concept is to provide a single set of co-ordinated digital spatial and a-spatial data
for use in all aspects of the project process: from concept design through collection,
interpretation and eventually on to synthetic analysis, dissemination, archivation and CRM.
Therefore, the data model will outline a framework which will ensure successful data
collection, analysis, management, curation and dissemination. The environment should allow
the following to occur:
•
Data re-use throughout the duration of the project.
•
Access to an up to date model at all times.
•
Creation of a collaborative environment to allow the sharing of data.
370
•
Robust data management and back-up regime.
In comparison to only 10 or 15 years ago, computers are now an accepted part of
archaeological life. The use of integrated ‘office’ products is now ubiquitous in the discipline.
Archaeological data management, analysis and visualisation applications using software such
as Geographical Information Systems (GIS), Relational DataBase Management Systems
(RDBMS), Virtual Reality Mark-up Language (VRML) viewers and statistical analysis
packages have proliferated over the past decade. The use of technology has led to the
phenomena of information explosion. Hence, defining the modelling environment becomes
even more important in order to manage this important and potentially volatile resource.
The use of GIS and database applications, in particular, has allowed archaeologists to not
only rapidly analyse their data sets but to employ a variety of different analytical mechanisms
(Kvamme 1999). GIS and other computerised analysis software should take us beyond the
simplistic, but useful, production of distribution maps into the realms of multi-criteria and
other statistical analysis. This can only occur if the data model has been defined with analysis
in mind and the data that is recorded is of consistent quality. The accumulated benefits of
running sophisticated statistical analyses at a variety of generalisations should transform
current analytical frameworks. The key to understanding these information resources is to
find appropriate mechanisms to collect, integrate, analyse, generalise and synthesise the
archaeological record.
I.1.1 Project requirements of the data model
The thrust of this research is to evaluate the effective integration of three important sources
of satellite imagery as aids to archaeological analyses;
•
Recently declassified military photography (Corona).
•
Multispectral imagery (Landsat TM).
•
New sub-metre imagery (Ikonos).
To facilitate analysis and interpretation the processed and co-registered satellite imagery had
to be integrated with ground-observation information captured as part of the SHR project.
These data, predominantly morphological, artefactual and geo-archaeological, will provide the
basis for understanding the history of human inhabitation of the application area. The
integrated study will consist of data sets from the following sources:
371
1. Present day and historic mapping.
2. Multi-resolution satellite imagery.
3. Ground survey (e.g. total station and GPS survey, field walking and geophysical
survey).
4. Site, artefact and environmental attribute recording.
The comparative study of these data sets will allow the following to occur:
1. Identification of features/anomalies from past archaeological landscapes.
2. Recognition of additional features based upon spectral and spatial characteristics
determined from (1).
3. Assessment of the different environmental zones and their impact on
archaeological visibility and preservation during different seasonal and
agricultural regimes.
4. The creation of thematic data sets as an analytical and predictive framework (e.g.
geomorphology, hydrology and relict-hydrology).
5. The delineation of relationships between archaeological landscapes and the
environment.
6. Long term evaluation and management of the regional cultural resources.
In essence the research entails the elucidation of as many of the archaeological components
of the landscape as possible from the satellite resource. However, to realise the full CRM
potential of the application area and the long-term potential of the SHR project the
information system must be able to respond to archaeological investigation at any scale
(including excavation) and have the ability to synthesise this information for different scales
of analyses.
A major requirement of the SHR project is to investigate the feasibility of applying wholly
digital recording and analysis techniques throughout the project from inception to
deposition. Integrated digital applications are rarely applied during the process of data
collection itself. The vast majority of digital techniques are used during the creation of
synthetic information (report preparation) or, potentially more dangerously, to generalise the
fieldwork record to perform landscape analyses (for example the English Sites and
Monuments Record). Generalisation in this way can severely inhibit analytical enquiries as the
372
raw data is separated from the analytical data i.e. scalar generalisation occurs through a
synthetic intermediary rather than directly on the raw data. As the research employs data sets
with different spatial resolutions, one goal is to develop analytical and management models
that enable the raw data sets to be analysed at a variety of scales (from landscape down to
intrasite analysis).
I.1.1.1 The archaeological evidence
The archaeological evidence within the study area falls into the following categories and subcategories:
1. Sites and their sample units.
a. Tells (sites of sustained, intensive duration, usually identified through
contour interpretation).
b. Other sites (artefact scatters, soil features and structures).
2. Non-sites (or background).
3. Hinterlands.
4. Field systems.
a. Those identified through ‘hard’ physical features (walls).
b. Those identified through ‘soft’ physical features (vegetation and ditches).
c. Those identified as ‘relict’ or ‘out of use’ systems (soil marks etc.).
5. Communication networks.
6. Seasonal systems.
7. Contextual environments.
I.1.1.2 The project environment
This data set will primarily be used by researchers within the departments of Archaeology
and Geography at the University of Durham. However, many different specialists
(particularly environmental and artefact specialists) will need to contribute to and have access
to the data set. Many of these specialists (mainly university affiliates) will be in different
locations and countries. For periods of the year, the fieldwork programme requires a
significant proportion of the data set to be ‘off-line’. Furthermore, the project has Syrian
373
partners within the Directorate General of Antiquities and Museums (DGAM), who in the
longer term, will need easy access to this information resource.
The most obvious practical problem is the distributed nature of users and work
environments. The traditional solution to this problem provides each individual with their
own copy of the archive. Each researcher inputs their data, these multiple data sets are then
collated and then re-integration to a single model is attempted. This approach creates a
significant data management problem which is further exacerbated in an archive which is
predominantly digital in nature.
I.1.2 Model brief
The model data set will consist of 2-d and 3-d graphical data and attribute data. These data
will be in raster, vector and RDBMS format. The model will be used to provide spatial and
attribute information to the user in the required format by querying the data model. This set
of data will be geographically and spatially correct (projected in Universal Transverse
Mercator (UTM) 37N).
Attribute data will be linked through to vector data via the use of unique identifiers and other
compound keys. These links will allow spatial and attribute queries to occur at any level of
abstraction. For example, the whole landscape can be spatially queried for the distribution of
artefact attributes based upon their most refined spatial location, alternatively, the same query
could be performed at the level of ‘site’ where the artefact attributes would be grouped by a
‘containing’ site which generalises their spatial location.
Some data sets will be ‘stand alone’ that is. they will have limited or no direct links to other
data sets and will be used for visualisation and resource purposes only (e.g. raster digital
mapping). Although satellite imagery and TIN models will be used to derive other thematic
products (such as land use maps, archaeological sensitivity maps and slope and aspect
models), where such a situation occurs all processing metadata will be maintained to allow
subsequent recreation of these data sets.
374
Figure 182 Conceptual modelling schema.
The data model will be structured predominantly to produce results from satellite processing
techniques and archaeological surface survey. However, the data model must allow future
invasive excavation episodes to be adequately incorporated and queried. Furthermore,
isolated architectural evidence (for example olive presses or milestones), referred to as
installations, must also be able to be recorded and queried without the attribution of any ‘site’
details.
Where appropriate the model will conform to the following requirements:
1. The adherence to cultural collection standards to ensure data conformity and
longevity.
2. Where possible all data will be captured by digital means.
3. All database tables will be normalised (Date 2000).
4. Referential integrity will be maintained in the database.
5. Files will be maintained in interoperable data formats (but not at the expense of
data management and integrity).
6. Metadata will be created and maintained in appropriate formats.
7. Where possible an audit trail for all information conversion and analytical
reasoning will be maintained.
375
8. There will be a strict regime for backup.
To summarise, the data model aims to produce a flexible environment where high quality
data can be collected, integrated, generalised and disseminated for analysis at any scale. The
environment outlined will provide a research framework that will allow the integration and
generalisation of ALL archaeological data whatever the scale and mechanism of collection.
The ability to access multiple high quality data sets should unify some of the disparate
agendas of archaeological theory and practice by re-establishing the causal links between
practice, data, analysis, synthesis and theory. Current interest in the complex relationships
between theory versus practice, particularly on the point of reflexivity, have re-opened many
discussions on the application of technology (Beck and Beck 2001; Lucas 2001; Chadwick in
press). Reflexive approaches to field recording, by their very nature, require rapid feedback of
complex information in a digestible format (synthesis on the fly), a process which is best
conducted through a robust data model. Technology itself is a facilitator; however, the
implementation and use of this technology is subject to organisational pressures (for example
see Campbell and Masser 1993).
Finally, a data model is not rationalised by what data it can store (that is the domain of a
catalogue), rather upon what information can be extracted from it. Hence, information
extraction is the primary driving force in the model design.
I.2 Model environment
I.2.1 Software environment
The backbone of the software environment will be provided by the functionality of
Environmental Systems Research Institute’s (ESRI) ArcGIS suite of GIS software, Research
System’s ENVI and ERDAS’s Imagine Image Processing (IP) software. MicroSoft (MS)
Access will be the Relation DataBase Management System (RDBMS).
ArcGIS is an industry standard Geographical Information Software (GIS) suite with
advanced ‘object orientated’ raster and vector modelling capabilities. It allows users to access
separate project files as if they were a single model and at the same time allows multiple users
to work on individual files simultaneously. All vector files are stored within the co-ordinated
ESRI personal Geodatabase format (an MS Access file). All raster files will employ the
interoperable formats determined by the Image Processing (IP) systems. ArcPAD is used in
conjunction with ArcGIS. ArcPAD is a mobile GIS data collection application designed to
376
run on the Pocket PC O/S. ArcPAD is designed as a seamless interface for multiple users
wishing to collect and update spatial and attribute data within a collaborative work
environment.
ENVI and Imagine are both dedicated image processing packages with advanced raster
modelling capabilities (including dedicated hyperspectoral analysis programs) and GIS
functionality. All raster files are stored as GeoTiff or multi-band Imagine files. All vector files
will be shared in interoperable formats determined by the GIS.
Microsoft Access is a popular and inexpensive RDBMS which is widely supported within
many GIS packages. Files are stored in native MS Access format. ThinkDB will be used in
conjunction with MS Access. ThinkDB is a mobile relational database application designed to
run on the Palm OS. ThinkDB is a multi-user system which allows users to bi-directionally
synchronise records with the central database (see Figure 184). This allows users to have the
most up-to-date records with them at all times.
Other software systems will also be included in the data model. However, these will be
predominantly intermediary software for the conversion of data from a collection device into
the required formats for the data model (i.e. digital camera, total station or GPS downloading
and conversion software) and are hence subject to rapid change.
I.2.1.1 File Formats
The software environment employs a suite of different applications, each of which has
specific strengths, weaknesses and their own preferred, and in some instances proprietary, file
format. These applications will access the same data in the data model. In order for this to
occur seamlessly, wherever possible, the data should be interoperable. Interoperability is the
ability of a system or a product to work with other systems or products without special effort
on the part of the user (whatis.com). Although interoperability is normally used in a hardware
and software context, it also pertains to data itself. With the increase in volume of data being
produced and stored digitally, data interoperability is a topic of concern in all collaborative
projects and programmes. Table 31 and Table 32 outline the major file format groupings the
choice of file formats for active usage and archive formats.
377
Type
Comments
Structured Texts
Documents produced using word-processing, desk top publishing
and text mark-up applications.
Raster Graphics
Vector Graphics
GIS
Image Processing
Database
Format
MS Word (.doc), MS Excel (.xls),
Adobe Portable Document Format
(.pdf) and Adobe PageMaker (.pmd)
TIF, JPEG, BMP and Adobe Photoshop
(.psd)
WMF, AutoCAD (.dwg), ESRI Shape
(.shp) and Adobe Illustrator (.ai)
Bitonal, greyscale and colour raster images. Produced for
recording and illustration purposes.
Vector images, including 3d models. Produced for recording and
illustration purposes.
Although this comprises a combination of other datasets it is
ESRI Geodatabase (.mdb), Erdas
grouped separately due to its integrated nature. The arbitrary focus
Imagine (.img) and GeoTIFF (.tif)
for GIS datasets are vector files.
Although this comprises a combination of other datasets it is
ESRI Shapefile (.shp), Erdas Imagine
grouped separately due to its integrated nature. The arbitary focus
(.img), MrSID (.sid) and GeoTIFF (.tif)
for IP datasets are multiband raster files.
Integrated raw data and analysis systems. No spreadsheets will be
created for raw data; rather they will be included as database
MS Access (.mdb)
tables.
Table 31 Types of Digital Resource
Over the medium to long time frame software and file convergence between GIS, IP and
RDBMS will continue, which will ultimately provide the full functionality of these diverse
software packages within one framework. From a file format perspective this has already
been achieved within ESRI’s Geodatabase, a single RDBMS containing raster, vector and
attribute data. However, this file format has yet to demonstrate its interoperability.
Data Type
Structured Text:
W ord Processed
Structured Test: DTP
Raster Graphics:
Publication
Vector Graphics: Geo
Use Format
Microsoft W ord
Abobe Portable Document Format
Adobe Pagemaker
TIFF
JPEG
Adobe Photoshop
GeoTIFF
Erdas Imagine
Erdas Imagine
MrSID
Adobe Illustrator
W indows Metafile
AutoCAD Drawing W eb Format
ESRI GeoDatabase
Data Sets: Database
Microsoft Access
Data Sets: Spreadsheets
Microsoft Excell
Raster Graphic:
Geo Single Band
Raster Graphics:
Multi Band
Vector Graphics:
Publication
Archive Format
DOC
PDF
PDF
TIF
JPEG
TIF
GeoTIFF
GeoTIFF
GeoTIFF (one per band)
No archive
AI
W MF
DW F
ESRI shape
ASCII text (comma
seperated with text in
quotes) or MDB
XLS (these files should be
integrated into the database)
Table 32 Active and archive file formats
I.2.1.2 Naming conventions and file structures
For deposition and re-use purposes an 8 character-naming convention is preferred (based on
DOS). All directories are currently in this format with the exception of directories
378
automatically generated through theme creation in ArcGIS. However, in order to make some
filenames intelligible (particularly when working with processed raster imagery) long
filenames are used. No names shall include the following characters <spaces> / \ < > * % $
£ “ ! + -.
Figure 183 outlines the project directory structure. Each directory and sub-directory contains
an ascii text file called ‘Readme.txt’ which provides more information about the directory and
any subdirectories. The directories called unknown contain information created during the
project which have been misfiled and consequently need integrating or deleting.
Figure 183 Project Directory Structure.
I.2.1.2.1 Archive directory
The archive directory contains all archived and legacy data sets. These will normally reside
solely on the Geography server at the University of Durham.
I.2.1.2.2 Database directory
The database directory contains all RDBMS files and their derivatives. The subdirectories
Panorama and photos contain the project panoramic (360 degree) and digital photographs. The
mangment subdirectory contains any database management files.
379
I.2.1.2.3 Document directory
The document directory contains all validated project (and related) documentation. Each
document in this directory has an associated metadata database record. This also means that
all documentation can be accessed directly through the information system. For publication
purposes some documentation files contain separate textual and graphical components. For
simplicity of data management each graphic is prefixed by the filename of the linking
document and suffixed by fig and the appropriate figure number. For example the document
sienna2001.doc has three images stored as tiff files called sienna2001fig1.tif, sienna2001fig2.tif and
sienna2001fig3.tif. Note that ALL unvalidated documentation is stored in the unknown
directory.
I.2.1.2.4 Download directory
The download directory contains all the raw downloaded data sets. The subdirectories refer to
the nature and type of device. These data sets are maintained to ensure that if a critical system
failure or malicious deletion occurs then the archive can be recreated from proprietary data.
I.2.1.2.5 Raster directory
The raster directory contains all the geo-referenced raster data sets. This does not include
digital photography unless it has been geo-registered. Each sub-directory contains data set
groupings which are mainly self explanatory.
I.2.1.2.6 Speclst directory
The speclst directory contains all interim information supplied by the project specialists and is
therefore a holding directory until all the primary data can be integrated into the model.
I.2.1.2.7 Unknown directory
The unknown directory contains all unvalidated information of unknown purpose that has
arrived into the project data structure but may be important. These files will be integrated
into the main project structure as appropriate.
I.2.1.2.8 Vector directory
The vector directory contains all the geo-referenced data sets held in the vector format. As the
project has moved to a geodatabase model (held in MS Access database format in the database
directory) for the primary vector data this directory contains commonly used vector data sets
exported from the geodatabase. Hence all files in the directory are secondary files.
380
I.2.2 Hardware environment
The modelling environment has been established on the principal of a co-ordinated yet
distributed situation. This design is to reflect the stable data management and back-up
facilities available at the University of Durham and the less stable fieldworking environment.
A worst-case data scenario would involve multiple field collection teams in Syria and multiple
users accessing and altering the on-line information content in Durham.
Figure 184 Hardware schema for the project data model.
Figure 184 outlines the basic hardware schema for the project. This currently entails making
the server effectively ‘offline’ during fieldwork to ensure that there is no duplication of data
or data loss at either end. However, it is the long-term goal to define a global data model
where the fieldwork data is bi-directionally synchronised with the stable data set stored on
the Geography Server. This would ensure that local and remote users can work
simultaneously on the same up-to-date data set.
In addition to the basic hardware schema defined by Figure 184 the project also employs
digital cameras, handheld GPS devices and PDAs for digital mobile data collection.
381
Figure 185 Schematic data flowline
I.3 SHR project data model
I.3.1 Data flowline
Figure 185 defines the project data flowline. Formal delineation of this flowline is essential
for the data model as it provides information on how data enters into the model and how it is
transformed by the modelling process. This information is then used to help determine the
nature of the data structure. For example, artefactual evidence, with the exception of
installations, all come from somewhere within a geo-referenced site or off-site collection unit.
It is essential that when an artefact’s attributes are entered into the database that at least the
382
basic information for this unit exists. Otherwise, particularly in an environment where
referential integrity is applied, the user will be unable to add any data about the artefact.
Figure 186 Database schema.
I.3.2 A-spatial model
There are many GIS data sets within the project yet only a few of these will have the need for
a complicated attribute data set. It should be possible to store most information within the
spatial data set itself. For example, the hydrology network can contain the river network
dendritic hierarchy weighting within the drawing. Where external information is stored
separately from the drawing it will initially be maintained within the Access2000
environment, however, migration to AccessXP may occur if this database format is utilised
within the university.
Complex a-spatial information will be maintained for the ground based reconnaissance and
the interpretative elements of the landscape. This data set will include information about
finds, site sub-divisions, sites, morphology and interpretation.
383
The use of the geodatabase will allow the seamless integration of all spatial and attribute data
within one data environment. However, until this storage model has been rigorously tested in
the distributed work environment it will not be fully implemented.
I.3.2.1 Database architecture
For data management purposes the Access data sets have been separated into five Access
databases, including the geodatabase SHR_GeoBase.mdb (see Figure 186). All referential
integrity has been turned off for the database PDADB.mdb as this is essential for the
synchronisation process between the PDAs and Access. Referential integrity ensures that a
‘parent’ table must have a record with the appropriate primary key before any ‘child’ table can
refer to this record. For example, with referential integrity turned on you can not add any
sub-table attributes about site 454 if site 454 does not already exist in the parent table.
Referential integrity is now enforced through the Graphical User Interface (GUI) rather than
at the table level. When the software and hardware architecture change it is hoped that
referential integrity will be enforced at the table level.
I.3.2.2 PDADB.mdb
PDADB.mdb holds the raw data and data management queries for the tables which are
synchronised with the PDAs (see Figure 187).
I.3.2.3 DskTopDB.mdb
DskTopDB.mdb (Desk Top Database) holds the raw data and data management queries for
the tables which are not synchronised with the PDAs (see Figure 188).
I.3.2.4 Analysis and visualisation databases
PDADB.mdb and DskTopDB.mdb contain the raw database tables and management queries
for the a-spatial data model. These databases have been physically separated for ease of data
management and maintenance when, for example, implementing the mobile collection
database on the PDAs. However, when users need to access or analyse the full data set then
both PDADB.mdb and DskTopDB.mdb need incorporating into one conceptual model.
This has occurred by linking all the tables from the two databases into any number of new
databases. This provides a flexible mechanism to create as many user-definable databases as
required which all have direct access to the raw data contained in PDADB.mdb and
DskTopDB.mdb.
384
Figure 187 PDADB.mdb entity relationship model. . Note the
fields in bold text are the primary (or compound primary) keys.
385
Figure 188 DskTopDB.mdb entity relationship model. Note the
fields in bold text are the primary (or compound primary) keys.
Figure 189 The main switchboard for Forms.mdb.
386
Figure 190 Fully integrated database entity relationship model.
Figure 190 describes the entity relationship model after integrating the tables from both
PDADB.mdb and DskTopDB.mdb. For the sake of clarity lookup tables have been excluded
from the model as they provide little new information concerning the overall model.
For ease of use, design and management two databases have been designed following this
structure: Form.mdb and Queries.mdb. Multiple copies of these databases can be distributed
to different users without affecting data integrity.
I.3.2.4.1 Form.mdb
Form.mdb contains the Graphical User Interface (GUI), form related queries and linked
tables from DskTopDB.mdb and PDADB.mdb (see Figure 186, Figure 189, Figure 191 and
Figure 192). The form database has been networked (on all the machines) through the R
drive. The R drive on each computer should be ‘mapped’ to the SHR_proj directory. This
387
means that all changes in design can be easily distributed over any number of computers on
the network.
Figure 191 The metadata switchboard for Forms.mdb.
I.3.2.4.2 Queries.mdb
Query.mdb contains user defined queries, reports and linked tables from DskTopDB.mdb and
PDADB.mdb (see Figure 186). This database should be used as a link for any pivot-table
querying in MS Excel and as a basis for any GIS attribute queries. If the project upgrades to
Office XP pivot table analysis will occur directly in MS Access.
Figure 192 An example of a site form (after clicking Add/View
data on the main switchboard). Note the tabs to access other data
about the site.
388
I.3.3 Spatial model
Spatial data refers to any data set that has an implicit spatial component. Archaeological
examples include maps, aerial photographs, satellite imagery, site plans, geophysical surveys
and terrain models. Unfortunately, spatial data standards are not as robust as RDBMS
standards although organisations such as the Open GIS consortium (www.opengis.org) are
facilitating interoperable data formats. This project is employing both raster and vector
formats to store its data.
Figure 193 Comparison of vector versus high and medium
resolution raster representations (Courtesy NERC, GetMapping
and English Nature).
I.3.3.1 Raster model
The raster model employs a simple array of cells located in space. Each cell contains a value
that represents a real world object. The cells can be of any shape which can form a
tessellation although rectangles are the most common examples. Due to the regular nature of
the data structure raster models are easily manipulated within computers . Raster models are
spatially less precise than vector models as the representational scale is governed by the size
389
of the cell (see Figure 193). However, many spatial data sets (satellite imagery, aerial imagery
and geophysical surveys) are collected in raster format.
The raster model is maintained through the interoperable file formats described earlier. These
formats are used so that seamless analysis can occur in all the software environments.
I.3.3.2 Vector model
Vector representations use points, lines and polygons to represent reality. Vector GIS have
more accurate spatial referencing and are thus commonly used in environments where spatial
precision is essential (such as cartography). Hence, there has been a long standing
relationship between Computer Aided Design (CAD) and vector GIS (Burroughs 1986). The
vector model is maintained within a single ESRI Geodatabase called SHR_GeoBase.mdb.
The geodatabase is hierarchically structured into many feature data sets which contain many
related feature classes (see Table 33).
Feature Dataset
3d_Mapping
3d_Mapping
3d_Mapping
3d_Mapping
3d_Mapping
Archaeology
Archaeology
Archaeology
Archaeology
Archaeology
Boundary
Boundary
Boundary
Communication
Communication
Communication
Communication
Communication
HistoricMapping
HistoricMapping
Hydrology
Hydrology
Hydrology
Hydrology
Hydrology
Hydrology
Hydrology
Soil
Feature Class
Contour
Contour_point
GPS_Breaklines
Spotheight
Trigheight
Depressions
Non_Arch_Soil_marks
Poss_Linears
Site_Sub_units
Sites
Grid
Grid_done
Study_Area
A_Road
B_road
Bridge
Track
Train
Fields_Syrian1to50
Structure_Syrian1to50
Canal
Irrigationchannel
Lake
Marsh
River_cent
River_edge
Wadi
Soil_Mapping
Topology
Network
Point
Network
Point
Point
Polygon
Polygon
Network
Polygon
Polygon
Network/Polygon
Point
Network/Polygon
Network
Network
Network
Network
Network
Network/Polygon
Polygon
Network
Network
Network
Network
Network
Network
Network
Polygon
Description
3D Contour lines
3D points 'weeded' from the contour lines
3D and 2D breakline data collected from GPS
3D Spot height
3D Triangulated trig. points
Depressions associated with sites
Soil marks of non-archaeological origin
Possible linear features
Outlines of site sub units
Outlines of sites
2km sampling grid
Sampled 2km sampling grid
Extent of the study area
Network of major roads
Network of minor roads
Bridges
Network of tracks
Train network
Field boundaries identified from the mapping
Amorphous structures identified from the mapping
Concrete irrigation canals
Irrigation canals
Lake edge
Marsh edge
River centerline
River network
Wadis and other seasonal water courses
Soil polygons
Table 33 Definition of the feature data sets and their classes within
SHR_GeoBase.mdb.
The specific methodologies for converting any raster-derived themes or classifications into
vector themes are discussed in the appropriate sections of the methodology. The majority of
attribute data is contained directly within the geodatabase itself except for the archaeology
feature data sets.
390
I.3.3.3 Legacy information
The spatial modelling environment has already gone through many incarnations as the core
spatial software environment has changed. Initially ESRI’s ArcView GIS was the main vector
modelling environment. Unfortunately the vector data management capabilities of ArcView
were quite poor. To address this issue AutoDesk’s AutoCAD MAP software was employed
to manage the vector data as AutoCAD is one of the most widely supported packages for
surveying instrumentation; furthermore, the functionality of MAP enabled high-end GIS data
cleaning, topology management and export to occur.
LatinGenus
Common Name
Genus
Family
Suborder Order
Vertebrate
Code
Accipiter nisus
Alauda arvensis
Anas crecca
Anas platyrhynchos
Anguis fragilis
Anura sp.
Apodemus cf. Sylvaticus
Apodemus flavicollis
Apodemus sp.
Apodemus sylvaticus
Apodemus/ Mus sp.
Ardea cinerea
Arvicola terrestris
Asio flammeus
Asio otus
Athene noctua
Bird sp.
Bufo bufo
Bufo calamita
Bufo sp.
Buteo buteo
Buteo cf. Buteo
Sparrowhawk
Accipiter sp. Accipitridae
Skylark
Alauda sp.
Alaudidae
Teal
Anas sp.
Anatidae
Mallard/ Domestic Duck
Anas sp.
Anatidae
Slow Worm
Anguis sp.
Anguidae
Frog/ Toad sp.
Anura sp.
Anura
Wood Mouse
Apodemus sp. Muridae
Yellow Necked Field Mouse Apodemus sp. Muridae
N/A
Apodemus sp. Muridae
N/A
Apodemus sp. Muridae
Wood
Mouse/
House
Mouse
Apodemus sp. Muridae
Grey Heron
Ardea sp.
Ardeidae
Water Vole
Arvicola sp. Crecitidae
Short-Eared Owl
Asio sp.
Strigidae
Long-Eared Owl
Asio sp.
Strigidae
Little Owl
Athene sp.
Strigidae
Bird sp.
Bird sp.
Other
Common Toad
Bufo sp.
Bufonidae
Natterjack Toad
Bufo sp.
Bufonidae
Toad sp.
Bufo sp.
Bufonidae
Common Buzzard
Buteo sp.
Accipitridae
Buzzard sp.
Buteo sp.
Accipitridae
Accipitridae
Alaudidae
Anatidae
Anatidae
Anguidae
Anura
Myomorpha
Myomorpha
Myomorpha
Myomorpha
Accipitriformes
Passeriformes
Anseriformes
Anseriformes
Sauria
Anura
Rodentia
Rodentia
Rodentia
Rodentia
Avian
Avian
Avian
Avian
Reptilian
Amphibian
Mammalian
Mammalian
Mammalian
Mammalian
Myomorpha
Ardeidae
Myomorpha
Strigidae
Strigidae
Strigidae
Other
Bufonidae
Bufonidae
Bufonidae
Accipitridae
Accipitridae
Rodentia
Ciconiformes
Rodentia
Strigiformes
Strigiformes
Strigiformes
Other
Anura
Anura
Anura
Accipitriformes
Accipitriformes
Mammalian
Avian
Mammalian
Avian
Avian
Avian
Avian
Amphibian
Amphibian
Amphibian
Avian
Avian
Table 34 Genus lookup table with ‘Value Added’ information
(LGenus is the primary key): Courtesy of Dr. Philip Piper.
I.3.4 Data validation
Many analyses are limited by attempting to query information in a model which is incomplete
or inconsistent. Data validation is the process of ensuring that all data is complete and
consistent and where there are problems these are flagged. Many of the validation procedures
are transparent to the user by the use of digital recording. The databases on the PDAs require
that all their fields contain a value, even if this value is ‘unknown’ thus enforcing validity.
Furthermore, many management queries have been developed in the databases to test for
391
logical consistency within the data itself. However, the most prevalent form of data validation
is in the use of lookup tables.
Lookup tables (attribute table or data dictionaries) are commonly employed to standardise
data entry. They contain a list of pre-defined terms that can be entered into a field without
the user introducing typographical errors. However, lookup tables can also be used to extend
meaning from the ‘value laden’ terms used in recording by incorporating a powerful
generalisation and analytical functionality. Table 34 is a genus table for an animal bone
specialist. The field ‘LatinGenus’ is the primary key (a field that contains only unique values)
and is all that is required for a look-up table. However, the other fields ‘add value’ to the table
by including information implicitly contained within the ‘Latin Genus’. As a trivial example
‘Common Name’ can be used to replace the Latin name for popular publications, or more
importantly ‘Order’ can be used to generalise data into taxonomic order groupings. This
technique could also be extended to include other ‘indicator’ attributes such as habitat. This
will allow more sophisticated and generalised querying to occur while maintaining the
primacy of the raw data (i.e. show the distribution of all bones grouped in their taxonomic
‘Order’ that occur in enclosure ditches and are of animals that do not prefer moist habitats).
I.3.4.1 Confidence
It is important to be aware of the level of error associated with data sets before they are
modelled. Most of the raster data sets will have their locational and classification errors
embedded directly into their meta-data. It is more difficult to describe confidence for
attribute data: error is introduced from a variety of sources and is closely correlated to the
subjective experience and goals of the individual recorder (Richards 1998 pp. 65, 224-225;
Banning 2002 p. 40). Attempting to create wholly objective field records is viewed by many
archaeologists as a fruitless goal (Shennan 1985; Banning 2002). Hence, a subjective system
has been adopted whereby a recorder can evaluate their own confidence in their
interpretation by use of a modifier. For example the ‘Unit Interpretation’ field UnitInt has an
associated modifier field IntMod (see Figure 187) whereby the recorder can add one of the
following values to represent their own confidence in the interpretation: low, medium or
high. These fields are used extensively throughout the database. Although not ideal, the
system does provide greater awareness to an end-user of the vagaries of field recording.
Future developments may include a landscape version of the innovative excavation recording
systems used by the Archaeology Services at the University of Durham (Adams 2001).
392
I.3.5 Audit trail
Currently only a few stages of auditing have been implemented into the data model. The
maintenance of raw data and the processing methodology allows the re-creation of data sets
from first principals, if, for example, new information is incorporated into the model which
requires global data model changes (i.e. a change in the geoid model). From an intellectual
perspective an extensive audit trail is maintained in the comments table LandScom (see Figure
187). This table has been designed to provide a date-stamped list of comments relating to all
forms of data collection, processing and analysis for each ‘site’. This means that any future
researcher can examine the formal and intellectual processes that occurred in the
interpretation of a site. This design also has significant CRM benefits as it delineates every
visit to a site and its purpose. Upon achieving long-term stability for the data model a final
audit system will be invoked which will involve a separate database within which any physical
changes to a field value will be maintained. This system, which will also provide a robust
backup resource, will also highlight any changes in the intellectual process not documented in
the current approach.
I.3.6 Data generalisation
Data generalisation is an essential component of archaeological analysis. Archaeological
processes and mechanisms of collection are scale-dependent and require changing degrees of
detail when analysis occurs. Hence, there is a need for variable levels of abstraction for
different modelling and analytical purposes (Adams 2001 p. 3). Excavation data are normally
synthesised for inter-site and landscape analysis. However, the general rule has been to
synthesise these data sets with limited long-term reference to their raw information sets.
Hence, over time, these data sets can lose their precise analytical value. Data are collected at
different scales which, in turn, are integrated in a variety of ways for analytical purposes.
Ideally these data should be integrated into a single spatial and related attribute data set that
can be automatically generalised rather than multiple data sets that need maintenance and
refreshing.
Database generalisation has been a long overlooked area within all the spatial disciplines. The
cartographers’ challenge is to create meaningful maps at different scales (the ratio between
the size of an object on a map and its real size) and resolutions (the smallest object that can
be represented). This same analogy is being applied within the GIS and Computer Aided
Mapping (CAM) industries to manage automated and semi-automated generalisation of
393
spatial data (Müller et al. 1995). Until recently the software to enable cost-effective spatial data
generalisation has been either non-existent or prohibitively expensive. However, modern
software can, at least theoretically, automatically generalise spatial and a-spatial data sets.
Encoding a generalisation index or algorithm which is applied to the spatial component and
its a-spatial attribute information is, potentially, one technique of successful generalisation
(Morehouse 1995). The use of object-orientated data models such as ESRI’s geodatabase
could be pivotal (Müller et al. 1995 pp. 3-5 gives an overview of proposed generalisation
functions for ESRI; Lee 1996). Ideally this information should be encoded within metadata
(MacDonald 2001). These metadata indices should be created for all potential scales of
analysis (both spatially and a-spatially). For example, for inter-site analysis the metadata index
would automatically generalise raw data (spatial and a-spatial) to its context groupings (e.g.
enclosures and structures).
In order to integrate information from multiple scales into the data model, metadata
recording is essential. Figure 194 details a possible data schema for this integration process.
This data model conforms to this general schema, however, it must be emphasised that
incorporating other project data structures into this approach would be extremely difficult for
exactly the same reasons defined by CIDOC (see section I.5 (Wise and Miller 1997)).
An approach of this type could still retain any level of complexity and could be employed at a
variety of information levels automatically reducing the data to fit into any scale of analysis
without influencing the primacy of the raw data.
Figure 194 Data schema for multi scalar data integration
394
I.4 Documentation
The range of documentation includes reporting (in the form of academic and popular
syntheses in a variety of media formats) down to basic file and structure documentation
created as part of data creation itself. Metadata will be created for every piece of
documentation. It is expected that this high level of metadata will be useful for reconstruction (and hence re-use) of the raw archaeological data and as an audit trail for both
the processing of data and the creation of any project syntheses or narrative. Therefore, all
documentation is accessible through the project information system.
Figure 195 Example of Dublin Core resource discovery metadata
maintained in table MetaT.
395
I.4.1 Metadata
Metadata is data about data (Wise and Miller 1997; Crofts et al. 2003). It is a specific subset of
documentation that systematically catalogues information about data that is not normally
available or obvious. For example, information pertaining to the creation, content, accuracy,
use, ownership and geographic coverage are all included in metadata.
Figure 196 Example of table level metadata stored in table
MetaTDB.
The project has extensive metadata information stored throughout the data model. The vast
majority are stored in tables: MetaTDB, MetaTDD and DocuT within DskTopdb.mdb and MetaT
within PDADB.mdb. MetaT (see Figure 195) contains an ADS compatible Dublin Core based
metadata architecture (ESRI 1995; Gillings and Wise 1998; Bewley et al. 1999). This table is
used to describe all the data sets and/or data set groups for subsequent resource discovery in,
for example, the ADS ArcHSearch catalogue. MetaTDB (see Figure 196) contains metadata for
the Access databases at the table level with descriptions about each table and how they relate
to one another (note that field metadata is maintained within the raw table design). MetaTDD
contains the project Data Dictionary (glossary or thesaurus). DocuT (see Figure 197) contains
bibliographic information about all the documentation held in the document directory.
396
I.5 Archiving and re-use
Archiving is becoming an increasingly important issue within archaeology and other cultural
heritage sectors. This is reflected in the emergence or participation of many bodies over the
past decade dedicated to providing best practice guidelines for preparing, preserving and
warehousing digital resources (for example; the Arts and Humanities Data Service, The
Centre for the Study of Archaeology (CSA), Electronic Resource Preservation and Access
NETwork (ERPANET) and English Heritage).
Figure 197 Example of bibliographic metadata held in table
DocuT.
Many researchers are being encouraged to subscribe to these facilitators either through
internal best practice mechanisms or as stipulations from a funding body (i.e. NERC, AHRB
and EH). This NERC funded research is no exception. However, there are intrinsic storage
and copyright difficulties with the deposition of this archive. The archive is currently 35GB,
of which the vast majority is satellite imagery and derived thematic information. Hence, the
curation of an archive of this magnitude would be very expensive. Many of the data sets have
been purchased through funding bodies (e.g. NERC for the Ikonos satellite imagery). These
bodies and the original data providers have their own licensing systems which may conflict
with the open copyright policies of warehousing organisations, such as the ADS, making
deposition of this data difficult or impossible.
397
Furthermore, the breadth of the data model and the requirement of other users to access and
develop this resource make it possible that this data set will not have a clear termination date.
Hence, it is difficult to assign a clear time-frame to archive a ‘live’ data resource. However, as
the data model, and particularly cutting edge software formats change, high priority will be
given to ensure backward compatibility with standard deposition formats or the prioritised
use of standard deposition formats. Fortunately the use of, for example, the ESRI
geodatabase storage architecture allows the integration of diverse data sets within one file,
providing significant data management advantages (Richards and Robinson 2000; ESRI
2002). Furthermore, the project already employs many standard formats which by their very
nature are interoperable between many software systems with limited or no data loss
(MacDonald 2001).
Comité International pour la DOcumentation du Conseil international des musées (CIDOC)
(Brown 2000) have acknowledged that many cultural resources suffer these problems of
archivation. Their original concept recommended the integration of data sets through a single
data model. Although it is possible to produce a framework which will directly integrate with
other CRM systems this approach has many drawbacks; not the least being that the specific
challenges faced in the SHR project would be under-represented using a standardised CRM
data model. The difficulties in extending the original CIDOC data model to cater for all
structuring eventualities led to its abandonment. CIDOC now advocate the use of mediation
systems capable of managing data from heterogeneous sources. Mediation systems allow
access to multiple information sources facilitating distributed queries without the need to
aggregate the data within a single data model. Integration of this data set within such a model
will require re-modelling to ensure concept conformity, but the benefits of the proposed
system will allow access to multiple data source at multiple levels of generalisation.
Once an integrated data set has been produced a coherent dissemination structure is
required. Other geographical science industries have consolidated their provision of spatial
information through a few web-based geographic data portals (e.g. ESRI's geography
network www.geographynetwork.com). These portals tend to be collaborative, multiparticipant systems allowing discriminatory access to geographical data. The majority of
leading GIS and remote sensing software vendors include functionality to access data sets
over such distributed networks. Higher speed communication systems (such as broadband)
and improved compression algorithms mean that larger file sizes can be easily accessed. For
398
example, the use of Enhanced Compressed Wavelet (ECW) technology by ER Mapper Inc.
has facilitated the deployment of terabytes of data over the web (e.g. a single 25cm resolution
mosaiced aerial photograph of the whole of Denmark is available at www.kortal.dk). It is
acknowledged that this technology is predominantly available in Europe and America,
although it is hoped that improvements in communication technology will alleviate this
problem in the near future. However, given the size and nature of modern digital data sets
the ability to adequately archive and provide access to information outweighs short-term
problems in user access.
I.6 Future requirements
A requirement of the model design has been to provide scope for future developments of the
data model. Hence this data model does not stand in isolation. Rather it outlines current best
practice guidelines and is expected to evolve to meet the requirements of changing research
and management needs and the general computing environment.
Considering the scope of the project, the range of specialists involved and the distributed
nature of their working environments it would seem appropriate to enable online updating
and querying of the data set. This will ensure that all participants conform to the data model
and are always working on up-to-date information. It was impossible to incorporate this
approach into the initial data models as until quite recently the use of the internet was illegal
in Syria. Although, in such a scenario, the deployment of many of the large satellite imagery
files might be restricted, access to the other geographical and attribute data resources should
be easily facilitated through the use of dedicated internet mapping software (such as ESRI’s
ArcIMS). This will also have an impact upon the attribute recording system which currently
employs two different mobile platforms running two different operating systems
(Handspring PDAs running Palm O/S and Compaq iPAQ running Pocket PC O/S). These
handhelds are used off-line and synchronised with the local data management system at the
end of each day. However, the development of multi-platform mobile application
synchronisation systems (such as the Freedom! Platform developed by ThinkingBytes
Technology) will enable the bi-directional synchronisation of these data sources between the
field, the fieldwork laptop and the server in Durham.
399
I.7 Using the data model and recording system
Once the data sets are downloaded only minimal post-processing is needed. The a-spatial
data is already in MS Access format and available through the desktop database. The digital
photographs need their numbers converting followed by a cursory check. The only difficult
issue is the integration of the GPS data. This is primarily due to the limited functionality of
handheld GPS. Other mechanisms to capture spatial data directly within a GIS are being
researched. However, the approach used provides the project team with a fully accessible
GIS ready archive at the end of each day. Consequently, hypotheses can be quickly reevaluated and amended. Furthermore, summary statistical information can be quickly and
easily generated to support any reports. The ability to dynamically engage with a project
archive in this way has far reaching consequences for the feedback between hypothesis
development and hypothesis testing. Furthermore, it provides valuable insights into the
quality of the record, the validity of the recording system and how these impact on analysis.
400
APPENDIX II : COULTER SAMPLE PROCESSING METHODOLOGY
The LS230 measures particle size distribution using the principal of laser diffraction. A
sample placed in the fluid module is circulated through a sample cell at a constant speed. A
beam of laser light shone through the cell is diffracted by particles within the sample, and the
forward scattered (or diffracted) light is collected by a series of detectors. The distribution of
light falling on the sensors enables the size distribution of the sample to be calculated. This
method enables the measurement of particles from 0.4µm to 2000µm (0.0004mm to 2mm).
Preparation of a solution for Coulter analysis requires a representative sample of 0.5 ±0.2 g
(all less than 2mm) in solution with the organic component removed. In order to achieve the
appropriate sample weight each sample was sieved through a 2mm sieve into a blank. The
residue was subdivided in a riffle so that the resultant sample was representative of the range
of particle sizes. 87.5% (i.e. three 50% boxes) of the sample was retained and re bagged. The
remaining sample was subdivided in the riffle until a weight of 0.5 ±0.2 g was achieved. The
sample was placed into a test tube marked with the sample number.
The organic material was removed by the addition of 20 ml of 10% Hydrogen Peroxide
(H2O2) to each sample. The sample was left in a water bath for 3 hours to allow the complete
oxidation of the organic material. Distilled water was added to each sample and any other
coarse organic material was removed. The sample was placed in a centrifuge for 6 minutes at
4000rpm. The excess liquid was decanted and 20 ml of distilled water was added. 2ml of
Sodium Hexameta Phosphate was added and mixed vigorously with the sample. This agent
reduces the likelihood of particles re-agglomerating. The sample is now ready for analysis in
the Coulter LS230.
401
APPENDIX III PARTICLE SIZE ANALYSIS RESULTS FROM SITES 97, 218,
221, 238, 271, 279, 339, 478, 496 AND 508.
III.1 Particle size analysis at site 97
Site 97 is a tell site in the northern irrigated marl zone (as defined in section 6.6) with a local
place name of Khirba Al-Ramadi. When last visited the site was under olive plantation and
various cereals. Surface material included pottery and basalt. A deep N-S bulldozed cut
(down to soil level) has removed all the site soil in the centre of the site. Sample 270 was
taken from the base of this bulldozed area and should therefore correspond to the underlying
soil or the earliest deposits. It appears that some of this has been re-deposited to the south of
the site (see Figure 198). The transect was located across the site in a S-N direction. Sample
points 276 to 277 and 282 to 283 highlight the transition between off-site and site. The site
boundary has been recorded from satellite imagery.
The particle size analysis displays a 10% increase in clay, a 5% decrease in fine silt, a 10%
decrease in medium silt, a 5% decrease in coarse silt and a c. 5% increase in fine sand
between off and on-site soils. Medium and coarse sand do not occur on the site. The
transitional boundaries defined by the satellite imagery correlate with the boundary changes
from the particle size analysis. Sample 270 has approximately the same particle size response
as the off-site soils.
402
Figure 198 Locations of soil samples collected over site 97 and
results of particle size analysis on these samples.
403
Figure 199 Locations of soil samples collected over site 218 and
results of particle size analysis on these samples.
404
III.2 Particle size analysis at site 218
Site 218 is a low tell site in the southern irrigated marl zone (as defined in section 6.6) close to
the Orontes with a local place name of Khirba Kafr Musa. The site was fallow when last
visited. Surface material included pottery, glass, basalt, tile and architectural remains. A deep
SW-NE bulldozed cut is on the NW edge of the site and a shallower NW-SE cut is on the
SW edge of the site. The transect was located across the site in a NE-SW direction (see
Figure 199). Sample points 114 to 116 and 121 to 122 highlight the transition between offsite and site. Sample 110 was misplaced from this site. The site boundary was recorded from
field collection.
The particle size analysis displays a 10% decrease in clay, a 5% decrease in fine silt, a <5%
increase in medium silt, a 5% increase in coarse silt and a c. 5% increase in fine sand between
off and on-site soils. Medium and coarse sand do not occur on the site. The transitional
boundaries defined by the satellite imagery do not correlate with the boundary changes from
the particle size analysis. The site extent should probably be moved to near sample 115. If
possible, this should be verified during the next fieldwork season.
III.3 Particle size analysis at site 221
Site 221 is a scatter in the southern marl zone (as defined in section 6.6). Like site 339 this
site is associated with topographic depressions (three in total). The site was fallow when last
visited. Surface material included pottery, basalt, tile and architectural remains. The transect
was located across the site in a NW-SE direction (see Figure 200). Sample points 131 to 133
and 137 to 139 highlight the transition between off-site and site. Samples 132, 133 and 138
were misplaced. The site boundary has been recorded from satellite imagery.
The particle size analysis displays a small 5% decrease in clay, no discernable difference in
fine silt, a 5% increase in medium silt and a tenuous <5% increase in coarse silt between off
and on-site soils. Fine sand produces a 10% increase on the post-transition and post-site
soils. Medium and coarse sand do not occur on the site. The transitional boundaries defined
by the satellite imagery correlate with the boundary changes from the particle size analysis for
the post site change. However, this correlation does not occur for pre-site. This indicates that
the transect for pre-site should have been extended further to the NW. During sample
collection it was noticed that the site soil had a noticeably looser structure (so much so that
walking the site was difficult).
405
Figure 200 Locations of soil samples collected over site 221 and
results of particle size analysis on these samples.
406
Figure 201 Locations of soil samples collected over site 238 and
results of particle size analysis on these samples.
407
III.4 Particle size analysis at site 238
Site 238 is a scatter on the margins of thin marl and wadi silts (as defined in section 6.6), just
north of a wadi with a local place name of Khirba Ramzoun. This site is also associated with
a topographic depression. When last visited the site was under fruit and olive plantation.
Surface material included pottery, basalt, tile, glass and a coin (undated). The transect was
located across the site in a NW-SE direction (see Figure 201). Sample points 222 to 223 and
228 to 230 highlight the transition between off-site and site. Please also note that samples 229
and possibly 230 are in a wadi. The site boundary has been recorded from satellite imagery.
The results of the particle size analysis are unclear and interpretations are difficult. There is a
decrease in clay, fine silt and medium silt and an increase in coarse silt and fine sand between
off and on-site soils. Medium and coarse sand do not occur on the site.
III.5 Particle size analysis at site 271
Site 271 is a scatter in the northern irrigated marl zone (as defined in section 6.6) with a local
place name of Khirba Khair Jamali. This site is also associated with a topographic depression.
The site was fallow when last visited. Surface material included pottery (possibly Islamic),
basalt, tile and architectural remains. The transect was located in a S-N direction (see Figure
202). Sample points 252 to 254 and 258 to 259 highlight the transition between off-site and
site. The site boundary has been recorded from satellite imagery.
The particle size analysis displays no discernable change in clay, a 5% decrease in fine silt, a
10% decrease in medium silt, a 5% increase in coarse silt and a 5% increase in fine sand
between off and on-site soils. Medium and coarse sand do not occur on the site. The
transitional boundaries defined by the satellite imagery correlate with the boundary changes
from the particle size analysis with the exception of medium silt.
408
Figure 202 Locations of soil samples collected over site 271 and
results of particle size analysis on these samples.
409
Figure 203 Locations of soil samples collected over site 279 and
results of particle size analysis on these samples.
410
III.6 Particle size analysis at site 279
Site 279 is discussed in section 8.4.1.2. The second transect was located across the site in a
SW-NE direction (see Figure 203). Unfortunately the transect was not long enough and
sample 208 remains in the transition zone. Sample points 201 to 203 and 207 to 208 highlight
the transition between off-site and site. The site boundary has been recorded from satellite
imagery.
The particle size analysis displays a 10-15% decrease in clay, a 5% decrease in fine silt, a 5%
increase in medium silt, a 10% increase in coarse silt and a c. 10% increase in fine sand
between off and on-site soils. Medium and coarse sand do not occur on the site. The
transitional boundaries defined by the satellite imagery correlate with the boundary changes
from the particle size analysis. This is an extremely interesting result as the correlation
between the two particle size analyses is poor (see Figure 149). Particle size analysis on the
2001 samples gave very clear results from this site (see section 8.4.1.2).
III.7 Particle size analysis at site 339
Site 339 is discussed in section 8.4.1.3. The second transect was located across the site in a SN direction (see Figure 204). Sample points 151 to 153 and 157 to 158 highlight the
transition between off-site and site. Sample 152 and 159 were misplaced from this site. The
site boundary has been recorded from satellite imagery.
The particle size analysis displays an uncorrelated change in clay, a 5% decrease in fine silt, a
5% decrease in medium silt, a 5% increase in coarse silt and a c. 7% increase in fine sand
between off and on-site soils. Medium and coarse sand do not occur on the site. The
transitional boundaries defined by the satellite imagery correlate with the boundary changes
from the particle size analysis. In comparison to the original transect (see Figure 151), the
results from the second transect are less clear. However, this change in clarity could be a
more representative reflection of how particle size varies through a site with a central
depression.
411
Figure 204 Locations of soil samples collected over site 339 and
results of particle size analysis on these samples.
412
Figure 205 Locations of soil samples collected over site 478 and
results of particle size analysis on these samples.
413
III.8 Particle size analysis at site 478
Site 478 is a scatter in the southern marl zone (as defined in section 6.6) and is one of the few
distinct single period prehistoric sites in the application area. The site was fallow when last
visited. Surface material included pottery (coarse tempered) and flint. The transect was
located across the site in a NE-SW direction (see Figure 202). Sample points 182 to 184 and
187 to 188 highlight the transition between off-site and site. An extra sample (190) was taken
next to an animal burrow where the majority of pottery was located. The site boundary has
been recorded from satellite imagery.
The particle size analysis displays no discernable change in any of the particle sizes with the
exception of a small (2-3%) increase in fine sand. This reflects the general difficulty in
detecting the site from the imagery and locating the site on the ground. In all probability this
difficulty is due to a different construction tradition in this period.
III.9 Particle size analysis at site 496
Site 496 is a scatter in the alluvial fan to the west of the Orontes (as defined in section 6.6).
When last visited the site was under scrub, beans and wheat. Surface material included
pottery, basalt, tile and architectural fragments. This site has been bulldozed creating a
distinct line of cleared basalt blocks (see Figure 206). Sample points 102 to 103 and 105 to
106 highlight the transition between off-site and site. Samples 100 and 102 were misplaced.
The transect was located across the site in an E-W direction. The site boundary has been
recorded from satellite imagery.
The particle size analysis displays no discernable change in any of the particle sizes. However,
there is a 10% increase in clay and a 10% decrease in medium sand across the transect.
Although this site is not in the marl zone it was included for comparative purposes in the
belief that it employed a similar construction tradition. The post-depositional processes in
this zone mean that many of the archaeological residues are likely to be buried. Hence, any
surviving surficial deposits will be impacted by mixing with alluvial wash deposits making
identification using this technique rather difficult.
414
Figure 206 Locations of soil samples collected over site 496 and
results of particle size analysis on these samples.
415
Figure 207 Locations of soil samples collected over site 508 and
results of particle size analysis on these samples.
416
III.10 Particle size analysis at site 508
Site 508 is a scatter in the northern irrigated marl zone (as defined in section 6.6). The site
was fallow when last visited. Surface material included pottery, basalt, glass, tile and
architectural remains. A shallow bulldozed cut ran E-W across the centre of the site. The
transect was located across the site in an E-W direction (see Figure 207). Sample points 291
to 293 and 299 to 301 highlight the transition between off-site and site. The site boundary
has been recorded from satellite imagery.
The particle size analysis displays an uncorrelated change in clay, a tenuous 5% increase in
fine silt, a tenuous 5% increase in medium silt, a 10% decrease in coarse silt across the
transect and a <5% increase in fine sand between off and on-site soils. Medium and coarse
sand do not occur on the site.
417
REFERENCES CITED
.
Adams, M. (2001). The Optician's Trick: an approach to recording excavation using an Iconic
Formation Process Recognition System. Interpreting Stratigraphy Site evaluation, recording
procedures and stratigraphic analysis. S. Roskams, (Ed.), Tempus Reparatum. 910: 91-102.
Adams, R. M. (1981). Heartland of cities: Survey of Ancient Settlements and Land Use of the
Central Floodplains of the Euphrates. Chicago, University of Chicago press.
Aerial Archaeology Committee (1983). Research objectives in aerial archaeology. Research
objectives in British archaeology. C. Thomas, (Ed.). London, CBA: 1-6.
Aitchison, K. and R. Edwards (2003). Archaeology Labour Market intelligence: Profiling the
Profession 2002/03. Bradford, Institute of Field Archaeologists: 129.
Alcock, S. E. (1994). Breaking up the hellenistic world: survey and society. Classical Greece.
Ancient histories and modern archaeologies. I. Morris, (Ed.). Cambridge, Cambridge
University Press: 171-190.
Alcock, S. E., J. F. Cherry and J. L. Davis (1994). Intensive survey, agricultural practice and
the classical landscape of Greece. Classical Greece. Ancient histories and modern
archaeologies. I. Morris, (Ed.). Cambridge, Cambridge University Press: 137-170.
Aldenderfer, M. S. (1987). On the Structure of Archaeological Data. Quantitative Research in
Archaeology: Progress and Prospects. M. S. Aldenderfer, (Ed.). London, London: 89-113.
Aldenderfer, M. S. (1998). "Quantitative Methods in Archaeology: A review of recent trends
and developments." Journal of Archaeological Research 6(2): 1998.
Allan, J. A. and T. S. Richards (1983). "Use of Satellite Imagery in Archaeological Surveys."
Libyan Studies 14: 4-8.
Allen, K. (2000). Consideration of Scale in Modelling Settlement Patterns Using GIS.
Practical Applications of GIS for Archaeologists: A predictive modelling toolkit. K. L.
Wescott and R. J. Brandon, (Eds.). London, Taylor and Francis.
Altmaier, A. and C. Kany (2002). "Digital surface model generation from CORONA satellite
imagery." ISPRS Journal of Photogrammetry and Remote Sensing 56: 221-235.
Ammerman, A. (1981). "Surveys and Archaeological Research." Annual Review of
Anthropology 10: 68-88.
Anderson, J. R., E. E. Hardy, J. T. Roac and R. E. Witmer (1976). A land use and land cover
classification for use with remote sensing data. Washington, US Government printing office:
18.
418
Anschuetz, K. F., R. H. Wilshusen and C. L. Scheick (2001). "An Archaeology of
Landscapes: Perspectives and Directions." Journal of Archaeological Research 9(2): 157-207.
Armenakis, C., F. Ledue, I. Cyr, F. Savopol and F. Cavayas (2003). "A comparative analysis
of scanned maps and imagery for mapping applications." ISPRS Journal of Photogrammetry
and Remote Sensing 57: 304-314.
Asrar, G. (1989). Theory and applications of optical remote sensing. New York, Wiley.
Aston, M. and T. Rowley, (Eds.) (1974). Landscape archaeology: an introduction to fieldwork
techniques on post-Roman landscapes. Newton Abbot, David & Charles.
Astorqui, A. (1999). Studying the archaeological record from photogrammetry. New
techniques for old times: CAA 98, computer applications and quantitative methods in
archaeology. J. A. Barceló, I. Briz and A. Vila, (Eds.). Oxford, Archaeopress.
Banning, E. (2002). Archaeological Survey. New York, Kluwer Academic.
Barnsley, M. J., D. Allison and P. Lewis (1997). "On the information content of multiple
view angle (MVA) images." International Journal of Remote Sensing 18(9): 1937 - 1960.
Barrett, J. C. (1994). Fragments from antiquity: An archaeology of social life in Britain, 29001200BC. Oxford, Blackwell Publishers.
Barrett, J. C. (1995). Some Challenges in Contemporary Archaeology. IFA conference,
Oxbow.
Baumgardner, M. F., L. F. Silva, L. L. Biehl and E. R. Stoner (1985). "Reflectance properties
of soils." Advances in agronomy 38: 1-44.
Beauchemin, M. and K. P. B. Thomson (1997). "The evaluation of segmentation results and
the overlapping area matrix." International Journal of Remote Sensing 18(18): 3895-3899.
Beck, A., G. Philip and D. Donoghue (in prep). "The preliminary evaluation of high
resolution satellite imagery for landscape applications in Syria and India."
Beck, A. R. (2002). "Getting your Hand dirty: In the field with the Handspring PDA and
ThinkDB." Archaeological Computing Newsletter(59): 11-15.
Beck, A. R. (2003). Integrated mobile applications: A case study from the SHR project, Syria.
WAC, Washington.
Beck, A. R. and M. Beck (2001). Computing, Theory and Practice: establishing the agenda in
contract archaeology. Interpreting Stratigraphy Site evaluation, recording procedures and
stratigraphic analysis. S. Roskams, (Ed.), Tempus Reparatum. 910: 173-182.
Beck, A. R., G. Philip, D. N. M. Donoghue and N. Galliatsatos (in press). Geo-locating
declassified CORONA satellite photography for archaeological surveys and cultural resource
419
management applications: an interim paper. Workshop on Aerial Archaeology and Remote
Sensing, Siena.
Ben-Dor, E. and A. Banin (1995). "Near Infrared Analysis as a rapid method to
simultaneously evaluate several soil properties." Soil Science Society of America 59: 364-372.
Ben-Dor, E., Y. Inbar and Y. Chen (1997). "The Reflectance Spectra of Organic Matter in
the Visible Near-Infrared and Short Wave Infrared Region (400-2500 nm) during a
Controlled Decomposition Process." Remote Sensing of Environment 61(1): 1-15.
Ben-Dor, E., J. R. Irons and G. F. Epema (1999a). Soil Reflectance. Remote Sensing for the
Earth Sciences. A. N. Rencz, (Ed.). London, John Wiley and Sons. 3.
Ben-Dor, E., J. Portugali, Kochavi, M. Shimoni and V. Lipaz (1999b). "Airborne thermal
video radiometry and excavation planning at Tel leviah, Golan Heights, Israel." Journal of
Field Archaeology 26.
Bewley, R. (1993). Aerial photography for archaeology. Archaeological Resource
Management in the UK: An Introduction. J. Hunter and I. Ralston, (Eds.). Stroud, Sutton
Publishing Limited.
Bewley, R. (2000). Aerial Survey: Learning from a Hundred Years of Experience? Aerial
archaeology - developing future practice, Leszno, Poland, IOS Press.
Bewley, R., D. Donoghue, V. Gaffney, M. Van Leusen and A. Wise (1999). Archiving Aerial
Photography and Remote Sensing Data: A Guide to Good Practice. Oxford, Oxbow.
Bewley, R. and W. Raczkowski (2000). Past Achievements and Prospects for the Future
Development of Aerial Archaeology: An Introduction. Aerial archaeology - developing future
practice, Leszno, Poland, IOS Press.
Binford, L. R. (1982). "The archaeology of place." Journal of anthropological archaeology(1):
5-31.
Binford, L. R. (1992). Seeing the Present and Interpreting the Past - and keeping things
straight. Space, time, and archaeological landscapes. J. Rossignol and L. Wandsnider, (Eds.).
New York, Plenum Press.
Binford, L. R. (1996). Willow Smoke and Dogs' Tails: Hunter-Gatherer Settlement Systems
and Archaeological Site Formation. Contemporary Archaeology in Theory: a reader. R. W.
Preucel and I. Hodder, (Eds.). Oxford, Blackwell: 39-60.
Bintliff, J. (2000). The concepts of 'site' and 'off-site' archaeology in surface artefact survey.
Non-destructive techniques applied to landscape archaeology. M. Pasquinucci and F.
Trement, (Eds.). Oxford, Oxbow.
Bintliff, J., B. Davies, C. Gaffney, A. Snodgrass and A. Waters (1992). Trace metal
accumulations in soils on and around ancient settlements in Greece. Geoprospection in the
Archaeological Landscape. P. Spoerry, (Ed.). Oxford, Oxbow.
420
Brady, N. C. (1984). The nature and properties of soils. New York, Macmillan.
Bridgland, D. R., G. Philip, R. Westaway and M. White (2003). "A long Quaternary terrace
sequence in the Orontes River valley, Syria: A record of uplift and occupation." Current
Science 84(5): 1080-1089.
Brivio, P. A., M. Pepe and R. Tomasoni (2000). "Multispectral and multiscale remote sensing
data for archaeological prospecting in an alpine alluvial plain." Journal of Cultural Heritage
1(2): 155 - 164.
Brown, A. (2000). "Centre for Archaeology Digital Archiving Strategy." 1.0. Retrieved 3rd
March, 2003, from http://www.english-heritage.org.uk.
Brugioni, D. A. (1996). "The Art and Science of Photoreconnaissance." Scientific
American(March): 78-85.
Buck, C. E., W. G. Cavanagh and C. D. Litton (1996). Bayesian Approach to Interpreting
Archaeological Data. New York, John Wiley and Sons.
Buck, P. E., D. E. Sabol and A. R. Gillespie (2003). "Sub-pixel artifact detection using remote
sensing." Journal of Archaeological Science 30: 973-989.
Burroughs, P. A. (1986). Principles of GIS for Land Based Assessment. Oxford, Clarendon
Press.
Butzer, K. W. (1982). Archaeology as human ecology: method and theory for a contextual
approach. Cambridge, Cambridge University Press.
Campana, S. and R. Francovich (2003). Landscape Archaeology in Tuscany: Cultural resource
management, remotely sensed techniques, GIS based data integration and interpretation. The
reconstruction of archaeological landscapes through digital technologies: proceedings of the
1st Italy-United States workshop, Boston, Massachusetts, USA, November 1-3, 2001. M.
Forte and P. R. Williams, (Eds.). Oxford, BAR: 15 - 29.
Campana, S. and E. Pranzini (1999). Il telerilevamento in Archeologia. Ciclo di Lezioni sulla
ricerca applicata in archeologia; Remote sensing in archaeology, Siena, Italy, Edizioni All'
Insegna del Giglio.
Campbell, H. and I. Masser (1993). Implementing GIS: the organisational dimension.
Association of Geographic Information, AGI.
Campbell, J. B. (2002). Introduction to remote sensing. New York, Guilford Press.
Carr, C. (1985). Perspective and basic definitions. For concordance in archaeological analysis.
C. Carr, (Ed.). Kansas City, Westport: 1-17.
Chadwick, A. (in press). What have the post-processualists ever done for us? Towards an
integration of theory and practice, and a radical field archaeology. Proceedings of the 2001
421
Interpretting Stratigraphy Conference. S. Roskams and C. M. Beck, (Eds.). Oxford, Tempus
Reparatum.
Chatwin, B. (1987). The Songlines. London, Cape 1987.
Chavez, P. S. (1996). "Image based atmospheric corrections - revisited and revised."
Photogrammetric Engineering and Remote Sensing 62(9): 1025-1036.
Chavez, P. S., S. C. Sides and J. A. Anderson (1991). "Comparison of three different methods
to merge multiresolution and multispectral data: Landsat TM and Spot panchromatic."
Photogrammetric Engineering and Remote Sensing 57: 265-303.
Cherry, J. F., J. L. Davis and E. Mantzourani (1991). Landscape archaeology as long-term
history: northern Keos in the Cycladic Islands from earliest settlement until modern times.
Los Angeles, Institute of Archaeology University of California Los Angeles.
Clark, C. D., S. M. Garrod and M. M. Parker Person (1998). "Landscape archaeology and
remote sensing in southern Madagascar." International Journal of Remote Sensing 19(8):
1461-1477.
Clark, G. A. (1987). Paradigms and Paradoxes in Contemporary Archaeology. Quantitative
Research in Archaeology: Progress and Prospects. M. S. Aldenderfer, (Ed.). London,
London: 30-60.
Clarke, D. L. (1978). Analytical Archaeology. London, Methuen.
Coburn, C., A. Roberts and K. Bach (2001). "Spectral and spatial artifacts from the use of
desktop scanners for remote sensing." International Journal of Remote Sensing 22(18): 38633870.
Colwell, R. N. (1997). History and place of photographic interpretation. Manual of
Photographic Interpretation. W. R. Philipson, (Ed.). Bethesda, American Society of
Photogrammetry and Remote Sensing: 3-48.
Colwell, R. N., J. E. Estes and G. A. Thorley (1983). Manual of remote sensing. Falls Church,
Va., American Society of Photogrammetry.
Comfort, A. (1997). Remote sensing and archaeological survey on the Euphrates: 1997
Report: 19.
Comfort, A. (1999). Remote sensing and archaeological survey on the Euphrates: 1998
Report: 19.
Comfort, A. (2000). Satellite remote sensing and archaeological survey on the Euphrates:
1999 Report: 25.
Condron, F., J. Richards, D. Robinson and A. Wise (1999). Strategies for digital data: findings
and recommendations from Digital Data in Archaeology: a survey of user needs. York,
Archaeology Data Service University of York 1999.
422
Cowley, D. C. (2002). "A case study in the analysis of patterns of aerial reconnaissance in a
lowland area of Southwest Scotland." Archaeological Prospection 9: 255-265.
Cox, C. (1992). "Satellite Imagery, aerial photography and wetland archaeology." World
Archaeology 24(2): 249-267.
Cracknell, A. P. and L. Hayes (1991). Introduction to remote sensing. New York, Taylor &
Francis.
Crawford, O. G. S. (1953). Archaeology in the field.
Crofts, N., M. Doerr and T. Gill. (2003, February 2003). "The CIDOC Conceptual Reference
Model: A standard for Communicating Cultural Content." Retrieved 2nd March, 2003, from
http://cidoc.ics.forth.gr/docs/martin_a_2003_comm_cul_cont.htm.
Crumley, C. L. (1994). Historical Ecology: A Multidimensional Ecological Orientation.
Historical Ecology. C. L. Crumley, (Ed.). Sante Fe, School of American Research Press.
Crumley, C. L. and W. H. Marquardt, (Eds.) (1987). Regional dynamics: Burgundian
landscapes in historical perspective. London, Academic Press.
Curran, P. J. (1990). Remote sensing of soils and vegetation in the USSR. London, Taylor &
Francis.
Daels, L. and K. M. Al Saadi (1990). The detection of old irrigation patterns in Mesopotamia
based upon the interpretation of aerial photographs and satellite images. Aerial photography
and geophysical prospection in archaeology. C. Léva, (Ed.). Brussels, CIRA - ICL: 83-86.
Dalla Bona, L. (1994). "Predictive Modelling Methodology." Retrieved 3 January, 2003,
from http://modelling.pictographics.com/method.htm.
Dare, P. M. and C. S. Fraser (2001). "Mapping informal settlements using high resolution
satellite imagery." International Journal of Remote Sensing 22(8): 1399-1401.
Date, C. J. (2000). An introduction to database systems. Reading, Mass., Addison-Wesley.
Day, D., A., J. Logsdon, M. and B. Latell (1998). Eye in the sky: the story of the CORONA
spy satellites. Washington, Smithsonian Institution Press.
Dial, G. and J. Grodecki (2003). Applications of Ikonos imagery. ASPRS. Anchorage, Alaska.
Donoghue, D. N. M. (1999). Multispectral Remote Sensing for archaeology. Ciclo di Lezioni
sulla ricerca applicata in archeologia; Remote sensing in archaeology, Siena, Italy, Edizioni
All' Insegna del Giglio.
Donoghue, D. N. M. (2001). Remote Sensing. Handbook of archaeological sciences. D. R.
Brothwell and A. M. Pollard, (Eds.). Chichester, Wiley.
423
Donoghue, D. N. M., N. Galiatsatos, G. Philip and A. R. Beck (2000). Satellite Imagery for
Archaeological Applications: A Case Study from the Orontes Valley, Syria. Aerial archaeology
- developing future practice. R. H. Bewley and W. Raczkowski, (Eds.). Leszno, Poland, IOS
Press: 211-223.
Doran, J. (1987). Modelling and Expert Systems. Quantitative Research in Archaeology:
Progress and Prospects. M. S. Aldenderfer, (Ed.). London, Sage.
Drury, S. A. (1993). Image interpretation in geology. London; New York, Chapman & Hall.
Duncan, R. B. and K. A. Beckman (2000). Site Location in Pennsylvania and West Virgina.
Practical Applications of GIS for Archaeologists: A predictive modelling toolkit. K. L.
Wescott and R. J. Brandon, (Eds.). London, Taylor and Francis.
Dunnel, R. C. (1992). The Notion of Site. Space, time, and archaeological landscapes. J.
Rossignol and L. Wandsnider, (Eds.). New York, Plenum Press.
Ebert, I. (1988). Remote sensing in archaeological projection and prediction. Quantifying the
present and predicting the past: theory, method, and application of archaeological predictive
modelling. W. J. Judge and L. Sebastian, (Eds.). Denver, U.S. Department of the Interior
Bureau of Land Management: 429-492.
Ebert, I. (1991). Distributional Archaeology. Albuquerque, University of New Mexico Press.
Ebert, J. I. (1984). "Remote Sensing Applications in Archaeology." Advances in
Archaeological Method and Theory 7.
Ebert, J. I. (1989). Techniques, methods and theoretical goals in American archaeological
remote sensing: 'Predictive Modelling' as an example. Into the sun: essays in air photography
in archaeology in honour of Derrick Riley. D. L. Kennedy, (Ed.). Sheffield [England], Dept.
of Archaeology and Prehistory University of Sheffield.
Ebert, J. I. and T. R. Lyons (1983). Archaeology, anthropology, and cultural resources
management. Manual of Remote Sensing. R. N. Colwell, J. E. Estes and G. A. Thorley,
(Eds.). Falls Church, Va., American Society of Photogrammetry.
El-Baz, F. (1997). "Space Age Archaeology." Scientific American(August): 60-65.
Elachi, C. (1987). Introduction to the physics and techniques of remote sensing. New York;
Chichester, John Wiley.
ERDAS (1999). ERDAS Field Guide. Atlanta, ERDAS.
ERDAS (2001). IKONOS sensor model support. Atlanta, ERDAS: 62.
ESRI (1995). Metadata Management in GIS. Redlands, ESRI.
ESRI (2002). Working with the Geodatabase. Redlands, ESRI.
424
Fernandez, R. N., D. G. Schulze, D. L. Coffin and G. E. Van Scoyoc (1988). "Color, organic
matter and pesticide adsorption relationships in a soil landscape." Soil Science Society of
America 52: 1023-1026.
Fisher, C. T. and T. L. Thurston (1999). "Forward to Dynamic Landscapes and Sociopolitical
Process: The Topography of Anthropogenic Environments in Global Perspective." Antiquity
73(281): 630-632.
Foley, R. A. (1981). Off-site archaeology: An alternative approach for the short-sited.
Patterns of the Past: Studies in honour of David Clarke. I. Hodder, G. Isaac and N.
Hammond, (Eds.). Cambridge, Cambridge University Press: 157-183.
Forte, M. (1995). Scientific visualization and archaeological landscape: the case study of a
terramarra, Italy. Archaeology and Geographical Information Systems: a European
perspective. G. Lock and Z. Stancic, (Eds.). London, Taylor & Francis.
Forte, M. (1999). Telerilevamento e paesaggi archeologici tridimensionali. Ciclo di Lezioni
sulla ricerca applicata in archeologia; Remote sensing in archaeology, Siena, Italy, Edizioni
All' Insegna del Giglio.
Forte, M. (2000). GIS, Aerial Photographs and Microtopography in Archaeology: Methods
and Applications. Aerial archaeology - developing future practice, Leszno, Poland, IOS Press.
Forte, M. (2003a). Mindscape: ecological thinking, cyber-anthropology and virtual
archaeological landscapes. The reconstruction of archaeological landscapes through digital
technologies: proceedings of the 1st Italy-United States workshop, Boston, Massachusetts,
USA, November 1-3, 2001. M. Forte and P. R. Williams, (Eds.). Oxford, BAR: 81-93.
Forte, M. (2003b). The remote sensing project for the archaeological landscape of Askum
(Ethiopia). The reconstruction of archaeological landscapes through digital technologies:
proceedings of the 1st Italy-United States workshop, Boston, Massachusetts, USA,
November 1-3, 2001. M. Forte and P. R. Williams, (Eds.). Oxford, BAR: 81-93.
Forte, M. and S. Kay (2003). Remote sensing, GIS and virtual reconstruction of
archaeological landscapes. The reconstruction of archaeological landscapes through digital
technologies: proceedings of the 1st Italy-United States workshop, Boston, Massachusetts,
USA, November 1-3, 2001. M. Forte and P. R. Williams, (Eds.). Oxford: 109-115.
Forte, M. and P. R. Williams (2003). The reconstruction of archaeological landscapes through
digital technologies: proceedings of the 1st Italy-United States workshop, Boston,
Massachusetts, USA, November 1-3, 2001. Oxford, Archaeopress: for British Archaeological
Reports.
Fowler, M. J. F. (1993). "Satellite archaeology. Stonehenge from space." Spaceflight 35: 130132.
Fowler, M. J. F. (1994). "Satellite image processing for the archaeologist." Archaeological
Computing Newsletter 39: 2-7.
425
Fowler, M. J. F. (1996). "High-resolution satellite imagery in archaeological application: a
Russian satellite photograph of the Stonehenge region." Antiquity 269: 667-671.
Fowler, M. J. F. (2001). "A high resolution satellite image of archaeological features to the
south of Stonehenge." International Journal of Remote Sensing 22(7): 1197-1171.
Fowler, M. J. F. (2002). "Satellite remote sensing and archeology: a comparative study of
satellite imagery of the environs of Figsbury Ring, Wiltshire." Archaeological Prospection 9:
55-69.
Francovich, R., H. Patterson and G. Barker, (Eds.) (2000). Extracting meaning from
ploughsoil assemblages. The POPULOUS project. Oxford, Oxbow.
Frank, S. (2001). "Comparison of classification of intensity-Hue-Saturation and Smoothing
Filter based Modulation (SFIM) image fusion techniques." Retrieved 11th June, 2003, from
http://www.geo.wvu.edu/geog655/spring2001/03/project.htm.
Franklin, S. E. and P. T. Giles (1995). "Radiometric processing of aerial and satellite remote
sensing imagery." Computers and Geosciences 21(3): 413-423.
Fraser, C. S., E. Baltsavias and A. Gruen (2002). "Processing of Ikonos imagery for submetre
3D positioning and building extraction." ISPRS Journal of Photogrammetry and Remote
Sensing 56: 177-194.
Fujimura, K. and E. Kuo (1999). "Shape reconstruction from contours using isotopic
deformation." Graphical Models and Image Processing 67: 127-147.
Furby, S. L. and N. A. Campbell (2001). "Calibrating images from different dates to 'likevalue' digital counts." Remote Sensing of Environment 77(2001): 186-196.
Gaffney, V., K. Ostir, T. Podobnikar and Z. Stancic (1996). Satellite imagery and GIS
applications in mediteranean landscapes. Interfacing the past: CAA95. H. Kamermans and K.
Fennema, (Eds.). Leiden, Institute of Prehistory University of Leiden: 337-342.
Gaffney, V. and Z. Stancic (1991). GIS approaches to regional analysis: a case study of the
island of Hvar. Ljubljana, Research Institute Faculty of Arts and Science University of
Ljubljana.
Galiatsatos, N. (in prep). Assessment of satellite imagery in Landscape Archaeology
applications: Case study from Orontes valley, Syria. Department of Geography. Durham,
UK, University of Durham.
Galvao, L. S., A. P. Marco and N. E. José Carlos (2001). "Variations in Reflectance of
Tropical Soils: Spectral-chemical composition relationships from AVIRIS data." Remote
Sensing of Environment 75: 245-255.
Gerard, F. (2002). "A CORINE map of Great Britain by semi-automated means." Retrieved
8th December, 2002, from http://www.ceh.ac.uk/subsites/CORINE/index.html.
426
Gerlach, F. (2000). "Characteristics of space imaging’s one-meter resolution satellite imagery
products." International archives of Photogrammetry and Remote Sensing 33 (B1): 128-135.
Gillings, M. and G. Goodrick (1998). "Sensuous and Reflexive GIS: Exploring visualisation
and VRML." Internet Archaeology.
Gillings, M., D. J. Mattingly and J. v. Dalen, (Eds.) (1999). Geographical information systems
and landscape archaeology. The archaeology of Mediterranean landscapes. Oxford, Oxbow.
Gillings, M. and A. Wise (1998). GIS Guide to Good Practice. Oxford, Oxbow.
Goffer, Z., M. Molcho and I. Geit-Arich (1983). "The disposal of wastes in ancient BeerSheba." Journal of Field Archaeology 10: 231-235.
Goldberg, P. (1992). Micromophology, Soils and archaeological sites. Soils in archaeology:
landscape evolution and human occupation. V. T. Holliday, (Ed.). Washington, D.C,
Smithsonian Institution Press.
Golledge, R. G. and R. J. Stimson (1987). Analytical behavioural geography. London; New
York, Croom Helm.
Goodchild, M. F. and D. A. Quattrochi (1997). Scale, Multiscaling, Remote Sensing, and
GIS. Scale in Remote Sensing and GIS. D. A. Quattrochi and M. F. Goodchild, (Eds.).
London, Lewis Publishers: 1-11.
Goodrick, G. and M. Gillings (2000). Constructs, simulations and hyperreal worlds: the role
of Virtual Reality (VR) in archaeological research. On the Theory and Practice of
Archaeological Computing. G. Lock and K. Brown, (Eds.). Oxford, Oxbow.
GORS (2002). Syria: Archaeology from Space. Damascus, General Organisation of Remote
Sensing.
Greene, K. (1990). Archaeology: an introduction. London, Batsford.
Griffiths, G. H., J. Lee and B. C. Eversham (2000). "Landscape pattern and species richness;
regional scale analysis from remote sensing." International Journal of Remote Sensing
21(13&14): 2685-2704.
Griffiths, G. H. and P. M. Mather (2000). "Remote sensing and landscape ecology: landscape
patterns and landscape change." International Journal of Remote Sensing 21(13&14): 25372539.
Grove, C. I., S. J. Hook and E. D. Paylor II (1992). Laboratory Reflectance Spectra of 160
minerals, 0.4 to 2.5 Micorometers. Jet Propulsion Laboratory, Pasadena, California.
Hageman, J. B. and D. A. Bennett (2000). Construction of Digital Elevation Models for
Archaeological Applications. Practical Applications of GIS for Archaeologists: A predictive
modelling toolkit. K. L. Wescott and R. J. Brandon, (Eds.). London, Taylor and Francis: 113
- 127.
427
Hall, R. C. (1997). Post War Strategic Reconnaisance and the Genesis of the project Corona.
Corona between the Sun and the Earth the first NRO reconnaisance eye in space. R. A.
McDonald, (Ed.). Bethesda, American Society for Photogrammetry and Remote Sensing.
Harris, J. R., D. W. Viljoen and A. N. Rencz (1999). Integration and Visualization of
Geoscience Data. Remote Sensing for the Earth Sciences. A. N. Rencz, (Ed.). London, John
Wiley and Sons. 3.
Harrower, M., J. McCorriston and E. A. Ocher (2002). "Mapping the results of agriculture in
Southern Arabia: the application of satellite remote sensing, global positioning system and
geographic information system technologies." Archaeological Prospection 9(1): 35-42.
Heron, C. (2001). Geochemical Prospecting. Handbook of Archaeological Sciences. D. R.
Brothwell and A. M. Pollard, (Eds.). Chichester, Wiley.
Hirata, M., N. Koga, H. Shinjo, H. Fujita, G. Gintzburger and A. Miyazaki (2001).
"Vegetation classification by satellite image processing in a dry area of North-Eastern Syria."
International Journal of Remote Sensing 22(4): 507-516.
Hodder, I. (1991). Reading the past: current approaches to interpretation in archaeology.
Cambridge, Cambridge University Press.
Hodder, I. (1993). Changing Configurations: the relationships between theory and practice.
Archaeological Resource Management in the UK: An Introduction. J. Hunter and I. Ralston,
(Eds.). Stroud, Sutton Publishing Limited.
Hodder, I. (1999). The Archaeological Process: an introduction. Oxford, Blackwell.
Holben, B. and C. Justice (1981). "An examination of spectral band ratioing to reduce the
topographic effect of remotely sensed data." International Journal of Remote Sensing 2: 115133.
Holcomb, D. W. (2001). "Imaging Radar and Archaeological Survey: An Example from the
Gobi Desert of Southern Mongolia." Journal of Field Archaeology 28: 131 - 141.
Holden, N., P. Horne and R. Bewley (2002). High-Resolution Digital Airborne Mapping and
Archaeology. Aerial Archaeology: Developing Future Practice. R. H. Bewley and W.
Raczkowski, (Eds.). Oxford, IOS press: 173-180.
Horne, J. H. (2003). A Tasseled Cap transformation for Ikonos images. ASPRS, Anchorage,
Alaska.
Horvath, E. H., D. F. Post and J. B. Kelsey (1984). "Relationship among Landsat digital data
and the properties of Arizona rangelands." Soil Science Society of America 56: 865-872.
IFA. (2001, September 2001). "Standard and guidance for archaeological desk-based
assessment." 2nd. Retrieved 6 January, 2003, from
http://www.archaeologists.net/docs/codes/dba2.pdf.
428
Irons, J. R., R. Weismiller and G. W. Petersen (1989). Soil Reflectance. Theory and
applications of optical remote sensing. G. Asrar, (Ed.). New York (N.Y.); Chichester, Wiley.
Jensen, J. R. (1996). Introductory digital image processing: a remote sensing perspective.
Upper Saddle River, N.J., Prentice-Hall.
Jensen, J. R. (2000). Remote sensing of the environment: an earth resource perspective.
Upper Saddle River, N.J., Prentice Hall.
Jones, B. (2000). Aerial archaeology around the Mediterranean. Non-destructive techniques
applied to landscape archaeology. M. Pasquinucci and F. Trement, (Eds.). Oxford, Oxbow:
49-60.
Judge, W. J., J. I. Ebert and R. K. Hitchcock (1975). Sampling in Regional Archaeological
Survey. Sampling in Archaeology. J. W. Mueller, (Ed.). Tucson, The University of Arizona
Press.
Judge, W. J. and L. Sebastian, (Eds.) (1988). Quantifying the present and predicting the past:
theory, method, and application of archaeological predictive modelling. Denver, U.S.
Department of the Interior Bureau of Land Management.
Karpouzli, E. and T. Malthus (2003). "The empiracle line method for the atmospheric
correction of IKONOS imagery." International Journal of Remote Sensing 24(5): 1143-1150.
Kartikeyan, B., A. Sarkar and K. L. Majumder (1998). "A segmentation approach to
classification of remotely sensed imagery." International Journal of Remote Sensing 19(9):
1695-1709.
Kennedy, D. L. (1998). "Declassified satellite photographs and archaeology in the Middle
East: case studies from Turkey." Antiquity 72: 553-561.
Key, J. (2001). "FW: Syrian - Ikonos." Retrieved 12th November, 2001.
Knapp, A. B. (1997). The archaeology of late Bronze Age Cypriot society: the study of
settlement, survey and landscape. Glasgow, Department of Archaeology.
Knapp, A. B. and W. Ashmore (1999). Archaeological landscapes: Constructed,
conceptualised, ideational. Archaeologies of landscape: contemporary perspectives. W.
Ashmore and A. B. Knapp, (Eds.). Oxford, Blackwell Publishers: 1-30.
Kodak. (2003, 6th February 2003). "KODAK PLUS-X AERECON II Film 3404."
Retrieved 2nd June, 2003, from
http://www.kodak.com/US/en/government/aerial/technicalPubs/tiDocs/ti1669/ti1669.sh
tml#specsens.
Kouchoukos, N. (2001). "Satellite Images and Near Eastern Landscapes." Near Eastern
Archaeology 64(1-2): pp. 80 - 91.
429
Kruckman, L. (1987). "The role of remote sensing in Ethnohistorical research." Journal of
Field Archaeology 14.
Kvamme, K. L. (1988). Development and testing of quantitative models. Quantifying the
present and predicting the past: theory, method, and application of archaeological predictive
modelling. W. J. Judge and L. Sebastian, (Eds.). Denver, U.S. Department of the Interior
Bureau of Land Management.
Kvamme, K. L. (1995). A view from across the water: the North American experience in
archaeological GIS. Archaeology and Geographical Information Systems: a European
perspective. G. Lock and Z. Stancic, (Eds.). London, Taylor and Francis: pp. 1-14.
Kvamme, K. L. (1999). "Recent Directions and Developments in Geographical Information
Systems." Journal of Archaeological Research 7(2): 153-201.
Lake, M. W., P. E. Woodman and S. J. Mithen (1998). "Tailoring GIS software for
archaeological applications: An example concerning multiple viewshed analysis." Journal of
Archaeological Science 25: 27-38.
Lambrick, G. (1977). Archaeology and agriculture: a survey of modern cultivation methods
and the problems of assessing plough damage to archaeological sites. Oxford, Council for
British Archaeology and the Oxfordshire Archaeological Unit.
Lang, S. A. (1992). An investigation of image processing techniques at Pincevent habitation
No. 1, a late Magdalenian site in Northern France, Arizona State University.
Langran, G. (1992). Time in geographic information systems. London; Washington, Taylor
and Francis 1992.
Layton, P. and P. J. Ucko (1999). Introduction: gazing on the landscape and encountering the
environment. The archaeology and anthropology of landscape: shaping your landscape. P. J.
Ucko and R. Layton, (Eds.). London, Routledge.
Lee, D. (1996). "Automation of Map Generalization." Retrieved 6th April, 2002, from
http://arconline.esri.com/arconline/whitepapers/ao_/mapgen.pdf.
Leone, A. P. and R. Escadafal (2001). "Statistical analysis of soil colour and
spectroradiometric data for hyperspectral remote sensing of soil properties (example in a
southern Italy Mediterranean ecosystem)." International Journal of Remote Sensing 22(12):
2311-2328.
Li, R., G. Zhou, N. J. Scmidt, C. Fowler and G. Tuell (2002). "Photogrammetric processing
of high-resolution airborne and satellite linear array stereo images for mapping applications."
International Journal of Remote Sensing 23(20): 4451-4473.
Liang, S., H. Fang, J. Morisette, M. Chen, C. Shuey, C. Walthall and C. Caughtry (2002).
"Atmospheric Correction of Landsat ETM+ Land Surface Imagery: II. Validation and
Applications." IEEE Transactions on Geoscience and Remote Sensing: 1-10.
430
Lillesand, T. M. and R. W. Kiefer (1999). Remote sensing and image interpretation. New
York, John Wiley & Sons.
Liu, J. G. (2000a). "Evaluation of Landsat-7 ETM+ Panchromatic Band for image fusion
with Multispectral Bands." Natural Resources Research 9(4): 269-276.
Liu, J. G. (2000b). "Smoothing Filter-based intensity modulation: a spectral preserve image
fusion technique for improving spatial details." International Journal of Remote Sensing
21(18): 3461-3472.
Liverman, D., E. F. Moran, R. R. Rinfus and P. C. Stern (1998). People and pixels: Linking
Remote Sensing and Social Science. Washington, National Academy Press.
Lucas, G. (2001). Critical approaches to fieldwork: contemporary and historical
archaeological practice. London, Routledge.
Luque, S. S. (2000). "Evaluating temporal changes using Multi-Spectral Scanner and
Thematic Mapper data on the landscape of a natural reserve: the New Jersey Pine Barrens, a
case study." International Journal of Remote Sensing 21(13&14): 2589-2611.
Lyons, T. R. and T. E. Avery (1977). Remote sensing: a handbook for archaeologists and
cultural resource managers. Washington, D.C., Cultural Resources Management Division
National Park Service U.S. Dept. of the Interior.
Lyons, T. R. and F. J. Mathien (1980). Cultural resources remote sensing. Washington, D.C.,
Cultural Resources Management Division National Park Servce.
MacDonald, A. (2001). Building a geodatabase: GIS by ESRI. Redlands, CA, Environmental
Systems Research Institute c1999-2001.
Macklin, M. G. and S. Needham, (Eds.) (1992). Alluvial archaeology in Britain. Oxbow
monograph; 27. Oxford, Oxbow Books.
Macklin, M. G., D. Passmore and B. Rumsby (1992). Climatic and cultural signals in
Holocene alluvial sequences: the Tyne basin, northern England. Alluvial archaeology in
Britain. G. Macklin Mark and S. Needham, (Eds.). Oxford, Oxbow Books.
Mandelbrot, B. B. (1967). "How long is the coastline of Britain? Statistical self-similarity and
fractal dimension." Science 155(636).
Marceau, D. J. (1999). "The scale issue in social and natural sciences." Canadian Journal of
Remote Sensing 25(4): 347-356.
Martin, D. (1991). Geographic Information Systems and their Socioeconomic Implications.
London, Routledge.
Maschner, H. D. G. (1996). Theory, Technology, and the Future of Geographic Information
Systems in Archaeology. New methods, old problems: geographic information systems in
431
modern archaeological research. H. D. G. Maschner, (Ed.). Carbondale, Southern Illinois
University: 301-308.
Mather, P. M. (1999). Computer processing of remotely sensed images: an introduction. New
York, John Wiley.
Mattikalli, N. M. (1997). "Soil colour modelling for the visible and near infrared bands of
Landsat sensors using laboratory spectral measurements." Remote Sensing of Environment
59: 14-28.
McDonald, R. A. (1997). Corona between the Sun and the Earth the first NRO
reconnaisance eye in space. Bethesda, American Society for Photogrammetry and Remote
Sensing.
McManamon, F. P. (1984). "Discovering Sites Unseen." Advances in Archaeological Method
and Theory 7.
Middleton, M. (2000). The Archaeological Soils Recording Chart. Fife, Artacorn.
Morain, S. A. (1998). A Brief History of Remote Sensing Applications with Emphasis on
Landsat. People and pixels: Linking Remote Sensing and Social Science. D. Liverman, E. F.
Moran, R. R. Rinfus and P. C. Stern, (Eds.). Washington, National Academy Press: 28-50.
Morehouse, S. (1995). GIS-based map compilation and generalization. GIS and
generalization: methodology and practice. J. C. Muller, J. P. Lagrange and R. Weibel, (Eds.).
Compiegne, France, Taylor & Francis: 21-30.
Müller, J. C., J. P. Lagrange and R. Weibel, (Eds.) (1995). GIS and generalization:
methodology and practice. Gisdata. 1. London, Taylor & Francis.
Mumford, G. and S. Parcak (2002). "Satellite image analysis and archaeological fieldwork in
El-Markha Plain (South Sinai)." Antiquity 76: 953-954.
Munsell, A. H. (1981). A Color Notation. Baltimore, Macbeth.
Munsell Colour Company (1975). Munsell Soil Colour Charts. Baltimore, Macbeth Division
of Kollmorgen.
Myers, V. (1983). Remote sensing applications in agriculture. Manual of remote sensing. R.
N. Colwell, J. E. Estes and G. A. Thorley, (Eds.). Falls Church, Va., American Society of
Photogrammetry.
Newson, P. G. (2002). Settlement, land use and water management systems in Roman
Arabia: an integrated archaeological approach. School of archaeology and ancient history.
Leicester, University of Leicester: 266.
Nunnally, N. R. (1973). Introduction to Remote Sensing: The physics of electromagnetic
radiation. The Surveillance Science: Remote Sensing of the Environment. R. K. Holz, (Ed.).
Boston, Houghton Mifflin Company: 18-26.
432
O'Brien, M. J., J. L. Beets, R. E. Warren, T. Hotrabhavananda, T. W. Barney and E. E.
Voight (1982). "Digital enhancement and grey-level slicing of aerial photographs: techniques
for archaeological analysis of intrasite variability." World Archaeology 14(2): 173-190.
Olson, G. W. (1984). Field guide to soils and the environment: applications of soil surveys.
New York; London, Chapman and Hall.
Ostir, K., Z. Stancic and M. Trusnovec (1999). Multispectral classification of satellite images.
Geographical information systems and landscape archaeology. M. Gillings, D. J. Mattingly
and J. v. Dalen, (Eds.). Oxford, Oxbow.
Palmer, R. (2000). Air Photo Interpretation and Mapping to Guide Fieldwork in Commercial
Archaeology in England. Aerial archaeology - developing future practice, Leszno, Poland,
IOS Press.
Palumbo, G. (1992). JADIS (Jordan Antiquities Database and Information System): An
example of national archaeological inventory and GIS applications. Computing the past:
computer applications and quantitative methods in archaeology. J. Andresen, T. Madsen and
I. Scollar, (Eds.). Aarhus; Denmark, Aarhus Denmark: 183-188.
Palumbo, G. and D. Powlesland (1996). Remote sensing and geographic database
management systems applications for the protection and consrevation of cultural heritage.
Remotoe Sensing for Geography, Geology, Land Planning and Cultural Heritage. D. ArroyoBishop, R. Carla, J. B. Lurie, C. M. Marino, A. Panunzi, J. J. Pearson and E. Zilioli, (Eds.).
Washington, SPIE. 2960: 124-129.
Parker, D. C. and M. F. Wolff (1973). Remote Sensing. The Surveillance Science: Remote
Sensing of the Environment. R. K. Holz, (Ed.). Boston, Houghton Mifflin Company: 29-37.
Parrington, M. (1983). "Remote Sensing." Annual Review of Anthropology(12): 105-124.
Pecknold, S., S. Lovejoy, D. Schertzer and C. Hooge (1997). Multifractals and Resolution
Dependence of Remotely Sensed Data: GSI to GIS. Scale in Remote Sensing and GIS. D. A.
Quattrochi and M. F. Goodchild, (Eds.). London, Lewis Publishers: 361-394.
Philip, G., A. R. Beck and D. N. M. Donoghue (in press). The contribution of satellite
imagery to archaeological survey: an example from Western Syria. Workshop on Aerial
Archaeology and Remote Sensing, Siena.
Philip, G., D. N. M. Donoghue, A. R. Beck and N. Galiatsatos (2002a). "CORONA satellite
photography: an archaeological application from the Middle East." Antiquity 76(291): 109118.
Philip, G., F. Jabour, A. R. Beck, M. Bshesh, J. Grove, A. Kirk and A. R. Millard (2002b).
"Settlement and Landscape Development in the Homs Region, Syria: Research Questions,
Preliminary Results 1999-2000 and Future Potential." Levant 34.
433
Philipson, W. R. (1997). Manual of Photographic Interpretation. Bethesda, Md., American
Society of Photogrammetry and Remote Sensing.
Plog, S., F. Plog and W. Wait (1978). Decision making in modern surveys. Advances in
Archaeological Method and Theory. M. Schiffer, B., (Ed.). London, Academic Press. 1: 383421.
Pohl, C. and J. L. Van Genderen (1998). "Multisensor image fusion in remote sensing:
concepts methods and applications." International Journal of Remote Sensing 19(5): 823-854.
Pollard, A. M. and C. Heron (1996). Archaeological Chemistry. Cambridge, Royal Society of
Chemistry.
Ponikarov, V. P., V. G. Kazmin, I. A. Mikhailov, A. V. Razvaliayev, V. A. Krasheninnikov,
V. V. Kozlov, E. D. Soulidi-Kondratiyev, K. Y. Mikhailov, V. V. Kolakov, V. A. Faradzhev
and K. M. Mirzayev (1967). The Geology of Syria: Explanatory notes on the 1:500,000
geological map of Syria: Part I. Stratigraphy, igneous rocks and tectonics. Moscow.
Powlesland, D., J. Lyall and D. Donoghue (1997). "Enhancing the record through remote
sensing: the application and integration of multi-sensor, non-invasive remote sensing
techniques for the enhancement of the Sites and Monuments Record." Internet
Archaeology(2).
Pundt, H. and K. Brinkkotter-Runde (2000). "Visualization of spatial data for field based
GIS." Computers and Geosciences 26: 51-56.
Raczkowski, W. (2000). Beyond the Technology: Do we need `Meta-Aerial Archaeology'?
Aerial archaeology - developing future practice, Leszno, Poland, IOS Press.
Ramenofsky, A. F. and A. Steffen, (Eds.) (1998). Unit issues in archaeology: measuring time,
space, and material. Foundations of archaeological inquiry. Salt Lake City, University of Utah
Press.
Ranchin, T., B. Aiazzi, L. Alparone, S. Baronti and L. Wald (2003). "Image-fusion- the
ARSIS concept and some successful implementation schemes." ISPRS Journal of
Photogrammetry and Remote Sensing 58: 4-18.
Redfern, S., G. Lyons and R. M. Redfern (1999). "Digital Elevation Modelling of individual
monuments from Aerial Photographs." 6: 211-224.
Rees, W. G. (2001). "Improving the accuracy of low-cost GPS measurements for remote
sensing applications." International Journal of Remote Sensing 22(5): 871-881.
Renfrew, C. and P. G. Bahn (1991). Archaeology: theories, methods, and practice. New
York., Thames and Hudson.
Richards, J. (1998). "Recent trends in Computer Applications in Archaeology." Journal of
Archaeological Research 6(4): 331-382.
434
Richards, J. D. and D. Robinson (2000). Digital archives from excavation and fieldwork: a
guide to good practice. Oxford, Oxbow.
Rimmington, J. N. (2000). Soil geochemistry and artefact scatters in Boeotia, Greece. Nondestructive techniques applied to landscape archaeology. M. Pasquinucci and F. Trement,
(Eds.). Oxford, Oxbow.
Robinson, J. M. and E. Zubrow (1999). Between Spaces: Interpolation in Archaeology.
Geographical information systems and landscape archaeology. M. Gillings, D. J. Mattingly
and J. v. Dalen, (Eds.). Oxford, Oxbow.
Rodriguez, A., H. Salahieh, R. Badwan and H. Khawam (1999). Groundwater Use and
Supplemental Irrigation in Atareb, northwest Syria. Aleppo, Syria, ICARDA.
Romano, D. G. and O. Tolba (1996). Remote sensing and GIS in the study of Roman
Centuriation in the Corinthia, Greece. Interfacing the past: CAA95. H. Kamermans and K.
Fennema, (Eds.). Leiden, Institute of Prehistory University of Leiden: 457-463.
Rosen, A. M. (1986). Cities of clay: the geoarcheology of tells. Chicago, University of Chicago
Press.
Roskams, S. (2001). Excavation. Cambridge, Cambridge University Press.
Rossignol, J. (1992). Concepts, Methods and Theory Building: A Landscape Approach.
Space, time, and archaeological landscapes. J. Rossignol and L. Wandsnider, (Eds.). New
York, Plenum Press.
Rossignol, J. and L. Wandsnider, (Eds.) (1992). Space, time, and archaeological landscapes.
Interdisciplinary contributions to archaeology. New York, Plenum Press.
Rothaus, R. M. and A. A. De Morett (1999). Landsat TM imagery in landscape archaeology:
Detection and modelling. Ciclo di Lezioni sulla ricerca applicata in archeologia; Remote
sensing in archaeology, Siena, Italy, Edizioni All' Insegna del Giglio.
Ryan, N. (1991). Beyond the relational database: Managing the variety and complexity in
archaeological data. Computer applications and quantitative methods in archaeology, 1991.
G. Lock, J. Moffett and J. Castleford, (Eds.). Oxford, BAR.
Ryan, N. S., D. R. More and J. Pascoe (1998). Enhanced Reality Fieldwork: the Context
Aware Archaeolgical Assistant. Computer Applications in Archaeology, 1997. V. Gaffney, S.
Exon and M. van Leusen, (Eds.). Oxford, BAR.
Sabins, F. F. (1997). Remote sensing: principles and interpretation. New York, W.H.
Freeman & Co.
Salomonson, V. V. (1983). Water resources assessment. Manual of remote sensing. R. N.
Colwell, J. E. Estes and G. A. Thorley, (Eds.). Falls Church, Va., American Society of
Photogrammetry.
435
Sauer, C. O. (1938). The morphology of landscape. Berkeley, Calif., University of California
Press.
Sauvage, M. (1998). La brique et sa mise en œuvre en Mésopotamie: des origins à l'époque
achéménide. Paris, Centre de recherche d'Archaéologie Orientale, Université de Paris.
Schiffer, M., B. (1987). Formation processes of the archaeological record. Albuquerque,
University of New Mexico Press.
Schiffer, M. B. (1996). Formation processes of the archaeological record. Salt Lake City,
University of New Mexico Press.
Schiffer, M. B., A. P. Sullivan and T. C. Kilinger (1978). "The design of archaeological
surveys." World Archaeology 10(1): 1-28.
Schmidt, A. (2004). "Remote Sensing and Geophysical Prospection." Internet Archaeology
25.
Schnitzer, M. (1982). Organic matter characterisation. Methods of soil analysis. A. L. Page, R.
H. Miller and R. Keeney Dennis, (Eds.). Madison, Wis., American Society of Agronomy: Soil
Science Society of America.
Schuldenrein, J. (1995). Geochemistry, Phosphate Fractionation, and the Detection of
Activity Areas at Prehistoric North American Sites. Pedological Perspectives in
Archaeological Research. M. E. Collins, B. J. Carter, B. G. Gladfelter and R. J. Southard,
(Eds.). Madison, Soil Science Society of America.
Scollar, I. (1990). Archaeological prospecting and remote sensing. Cambridge, Cambridge
University Press.
Scudder, S. J., J. E. Foss and M. E. Collins (1996). "Soil Science and Archaeology." Advances
in agronomy 57: 1-76.
Sever, T. L. (1988). Remote Sensing. Benchmarks in time and culture: an introduction to the
history and methodology of Syro-Palestinian archaeology. J. F. Drinkard, G. L. Mattingly and
J. M. Miller, (Eds.). Atlanta, Scholars Press.
Sever, T. L. (1998). Validating prehistoric and current social phenomena upon the landscape
of eten, Guatemala. People and pixels: Linking Remote Sensing and Social Science. D.
Liverman, E. F. Moran, R. R. Rinfus and P. C. Stern, (Eds.). Washington, National Academy
Press: 145-163.
Sever, T. L. and D. W. Wagner (1991). Analysis of prehistoric roadways in Chaco Canyon
using remotely sensed digital data. Ancient road networks and settlement hierarchies in the
New World. C. D. Trombold, (Ed.). Cambridge, University of Cambridge: 42-52.
Shell, C. (2000). Airborne High-Resolution Digital, Visible, Infra-Red and Thermal Sensing
for Archaeology. Aerial archaeology - developing future practice, Leszno, Poland, IOS Press.
436
Shennan, I. (1997). Quantifying Archaeology. Edinburgh, Edinburgh University Press.
Shennan, I. and D. N. M. Donoghue (1992). "Remote sensing in archaeological research."
Proceedings of the British Academy 77: 223-232.
Shennan, S. (1985). Experiments in the Collection and Analysis of Archaeological Survey
Data: The East Hampshire survey. Sheffield, Department of Archaeology and Prehistory.
Siegal, B. S. and A. R. Gillespie (1980). Remote sensing in geology. Chichester, Wiley.
Skirvin, S. (2003). "Atmospheric and radiometric correction of Landsat Thematic Mapper
data using the COST model of Chavez, 1996." Retrieved 10th June, 2003, from
http://www2.erdas.com/supportsite/downloads/models/user_models/user_model_2.html.
Slonecker, E. T., D. M. Shaw and T. M. Lillesand (1998). "Emerging legal and ethical issues
in advanced remote sensing technology." Photogrammetric Engineering and Remote Sensing
64: 589-595.
Sommer, S., J. Hill and J. Mégier (1998). "The potential of remote sensing for monitoring
rural land use changes and their effects on soil conditions." Agriculture Ecosystems and
Environment 67: 197-209.
Song, C., C. E. Woodcock, K. C. Seto, M. P. Lenney and S. A. Macomber (2000).
"Classification and change detection using Landsat TM data: When and how to correct
atmospheric effects?" Remote Sensing of Environment 75: 230-244.
Space Imaging. (2003). "Ikonos Product Guide." Retrieved 30th May, 2003, from
www.spaceimaging.com/whitepapers_pdfs/IKONOS_Product_Guide.pdf.
Spoerry, P., (Ed.) (1992). Geoprospection in the Archaeological Landscape. Oxford, Oxbow.
Stanjek, H. and J. W. E. Fabbinder (1995). "Soil aspects affecting archaeological details in
aerial photographs." Archaeological Prospection 2: 91-101.
Stein, J. K. (1992). Organic Matter in Archaeological Contexts. Soils in archaeology:
landscape evolution and human occupation. V. T. Holliday, (Ed.). Washington, D.C,
Smithsonian Institution Press.
Stevanovic, M. (1997). "The Age of Clay: The Social Dynamics of House Destruction."
Journal of anthropological archaeology(16): 334-395.
Stone, E. (2003). Remote sensing and the location of the ancient Tigris. The reconstruction
of archaeological landscapes through digital technologies: proceedings of the 1st Italy-United
States workshop, Boston, Massachusetts, USA, November 1-3, 2001. M. Forte and P. R.
Williams, (Eds.). Oxford, 157-162: 81-93.
Stone, E. and P. Zimansky (1995). "The Tapestry of Power in a Mesopotamian City."
Scientific American(April): 118-123.
437
Stoner, E. R. and M. F. Baumgardner (1981). "Characteristic variation in the reflectance of
surface soils." Soil Science Society of America 45: 1161-1165.
Sunar, F. and N. Musaoglu (1998). "Merging multiresolution SPOT P and Landsat TM data:
the effects and advantages." International Journal of Remote Sensing 19(2): 219-224.
Taylor, C. C. (1972). The study of settlement patterns in pre-Saxon Britain. Man, Settlement
and Urbanism. P. J. Ucko, R. Tringham and G. W. Dimbleby, (Eds.). London, Duckworth.
Taylor, J. (2000). Soil phosphate survey. Non-destructive techniques applied to landscape
archaeology. M. Pasquinucci and F. Trement, (Eds.). Oxford, Oxbow.
Teng, W. L. (1997). Fundamentals of Photographic Interpretation. Manual of Photographic
Interpretation. W. R. Philipson, (Ed.). Bethesda, American Society of Photogrammetry and
Remote Sensing: 49-116.
Tilley, C. (1994). A Phenomenology of Landscape. London, Routledge.
Toutin, T. (2001). "Elevation modelling from satellite visible and infrared (VIR) data."
International Journal of Remote Sensing 22(6): 1097-1125.
Toutin, T. (2002). "DEM from stereo Landsat 7 ETM+ data over high relief areas."
International Journal of Remote Sensing 23(10): 2133-2139.
Townshend, J. R. G., J. R. Hardy, J. Cushnie and A. Wilson (1988). Thematic mapper data:
characteristics and use, NERC/SERC reprographics.
Tso, B. and P. M. Mather (2001). Classification methods for remotely sensed data. London,
Taylor & Francis.
United Nations. (1985). "Principles Relating to Remote Sensing of the Earth from Outer
Space." Retrieved 24 January, 2003, from
http://www.oosa.unvienna.org/SpaceLaw/rstxt.htm.
Ur, J. (2002). "Settlement and landscape in Northern Mesopotamia: The Tell Hamoukar
Survey 2000-2001." Akkadica 123: 57-88.
Ur, J. (2003). "CORONA satellite photography and ancient road networks: A Northern
Mesopotamian case study." Antiquity 77(295): 102-115.
USGS. (2003a). "Declassified Satellite Imagery - 1." Retrieved 2nd June, 2003, from
http://edc.usgs.gov/products/satellite/declass1.html.
USGS. (2003b). "Declassified Satellite Imagery - 1 (1996)." Retrieved 2nd June, 2003, from
http://edc.usgs.gov/guides/disp1.html.
USGS. (2003c, Friday, May 09, 2003). "EROS Data Center, Sioux Falls, SD." Retrieved
28th May, 2003, from http://edc.usgs.gov/products/satellite/tm.html.
438
USGS. (2003d, October 9, 2002). "LANDSAT-7 LEVEL-0 AND LEVEL-1 DATA SET
DOCUMENT." Retrieved 28th May, 2003, from
http://eosims.cr.usgs.gov:5725/DATASET_DOCS/landsat7_dataset.html.
USGS. (2003e, Monday, April 14, 2003). "USGS Landsat Project - Landsat 7 Home page."
Retrieved 28th May, 2003, from http://landsat7.usgs.gov/index.php.
Van Liere, W. J. (1959). "Ager centuriatus of the roman colonia of Emesa." Les annales
archaeologiques de Syries 8-9: 55-58.
Voûte, C. (1955). "Climate or tectonics? Some remarks on the evolution of the valley of the
Orontes (Nahr el-Aassi) between Homs and the marshy plains of the Ghab (Syria)."
Geologie en Mijnbouw 17: 197-206.
Wald, L. and T. Ranchin (2002). "Comment: Liu "Smoothing Filter-based intensity
modulation: a spectral preserve image fusion technique for improving spatial details"."
International Journal of Remote Sensing 23(3): 593-597.
Wald, L., T. Ranchin and M. Mangolini (1997). "Fusion of satellite images of different spatial
resolutions: Assessing the quality of resulting images." Photogrammetric Engineering and
Remote Sensing 63: 691-699.
Wandsnider, L. (1992a). Archaeological Landscape Studies. Space, time, and archaeological
landscapes. J. Rossignol and L. Wandsnider, (Eds.). New York, Plenum Press.
Wandsnider, L. (1992b). The Spatial Dimension of Time. Space, time, and archaeological
landscapes. J. Rossignol and L. Wandsnider, (Eds.). New York, Plenum Press.
Ward, I. and P. Larcombe (2003). "A process-orientated approach to archaeological site
formation: application to semi-arid Northern Australia." Journal of Archaeological
Science(30): 1223-1236.
Warren, R. E. and D. L. Asch (2000). A Predictive Model of Archaeological Site Location in
the Eastern Prairie Peninsula. Practical Applications of GIS for Archaeologists: A predictive
modelling toolkit. K. L. Wescott and R. J. Brandon, (Eds.). London, Taylor and Francis.
Waters, M. R. (1992). Principles of geoarchaeology: a North American perspective. Tucson,
University of Arizona Press.
Way, D. S. and J. R. Everett (1997). Landforms and geology. Manual of Photographic
Interpretation. W. R. Philipson, (Ed.). Bethesda, American Society of Photogrammetry and
Remote Sensing: 117-166.
Wescott, K. L. and R. J. Brandon (2000). Practical Applications of GIS for Archaeologists: A
predictive modelling toolkit. London, Taylor and Francis.
Wheatley, D. (1995). Cumulative viewshed analysis: a GIS-based method for investigating
intervisibility, and its archaeological application. Archaeology and Geographical Information
439
Systems: a European perspective. G. Lock and Z. Stancic, (Eds.). London, Taylor and
Francis: 171 - 186.
Wheatley, D. (1996). Between the lines: the role of GIS-based predictive modelling in the
interpretation of extensive survey data. Interfacing the past: CAA95. H. Kamermans and K.
Fennema, (Eds.). Leiden, Institute of Prehistory University of Leiden: 275-291.
Wheatley, D. and M. Gillings (2002). Spatial technology and archaeology: the archeaological
applications of GIS. New York, Taylor & Francis.
White, R. E. (1997). Principles and practice of soil science: the soil as a natural resource.
Oxford, Blackwell Science.
Wilkinson, T. J. (1998). "Water and human settlement in the Balikh Valley, Syria:
investigations from 1992 - 1995." Journal of Field Archaeology 25: 63-87.
Wilkinson, T. J. (2000). "Regional Approaches to Mesopotamian Archaeology:
The Contribution of Archaeological Surveys." Journal of Archaeological Research 8(3): 219 267.
Wilkinson, T. J. (2001). Surface collection techniques in field archaeology: Theory and
Practice. Handbook of archaeological sciences. D. R. Brothwell and A. M. Pollard, (Eds.).
London, John Wiley and Sons Ltd.
Wilkinson, T. J., C. A. I. French, W. Matthews and J. Oates (2001). Geoarchaeology,
Landscape and the Region. Excavations at Tell Brak Vol. 2: Nagar in the third millennium
BC. D. Oates, J. Oates and H. McDonald, (Eds.). Cambridge, McDonald Institute for
Archaeological Research. 2: 1-14.
Wilkinson, T. J., J. Ur and J. Casana (2004). From Nucleation to Dispersal: Trends in
Settlement Pattern in the Northern Fertile Crescent. Side by Side Survey: Comparative
regional studies in the Mediterranean World. S. E. Alcock and J. F. Cherry, (Eds.). Oxford,
Oxbow: 251.
Williams, R. S. (1983). Geological Applications. Manual of remote sensing. R. N. Colwell, J.
E. Estes and G. A. Thorley, (Eds.). Falls Church, Va., American Society of Photogrammetry.
Wilson, D. R. (2000). Air photo interpretation for archaeologists. Stroud, Tempus.
Winterhalder, B. (1994). Concepts in Historical Ecology: The view from evolutionary
ecology. Historical Ecology. C. L. Crumley, (Ed.). Sante Fe, School of American Research
Press.
Wirth, E. (1971). Syrien eine geographische Landeskunde. Darmstadt, Wissenschaftliche
Buchgesellschaft.
Wise, A. and P. Miller (1997). "Why metadata matters." Internet Archaeology(1).
440
Wolfart, R. (1967). Syria und Lebanon: Beiträge zur regionalen geologie der erde. Berlin,
Gebrüder Borntraeger.
Wu, J. and Y. Qi (2000). "Dealing with scale in landscape analysis: An overview." Geographic
Information Sciences 6(1): 1-5.
Zomer, R., S. Ustin and J. Ives (2002). "Using satellite remote sensing for DEM extraction in
complex mountainous terrain: landscape analysis of the Makalu Barun National Park of
eastern Nepal." International Journal of Remote Sensing 23(1): 125-143.
441