SEBAL

Surface Energy Balance Algorithms for Land

Idaho Implementation

Advanced Training and Users Manual

August, 2002
Version 1.0
Waters Consulting

University of Idaho

WaterWatch, Inc.

Nelson, British Columbia

Kimberly, Idaho

Wageningen, the Netherlands

Ralf Waters

Richard Allen
Masahiro Tasumi
Ricardo Trezza

Wim Bastiaanssen

Funded by
a NASA EOSDIS/Synergy grant from the Raytheon Company
through The Idaho Department of Water Resources

TABLE OF CONTENTS
Page
4

List of Symbols
List of Figures

Introduction

1.
2.
3.
4.

Background
Concept
Objectives
Who can use this manual?

The Theoretical Basis of SEBAL

1. Overview How SEBAL computes evapotranspiration (ET)
2. Data requirements and data preparation
A. Cloudy conditions
B. Header file information
C. Land-use map
D. Weather data

9
11
11
12
12
12

The Operation of the SEBAL Model

1. Setting up the image
2. Acquiring header file information, weather data, and reference ET values
3. Surface radiation balance equation
A. Surface albedo
B. Incoming shortwave radiation
C. Outgoing longwave radiation
D. Choosing the cold and hot anchor pixels
E. Incoming longwave radiation
F. Solving the surface energy balance equation
4. Surface energy budget equation
A. Soil heat flux
B. Sensible heat flux
C. Latent Heat Flux, ET, Instantaneous ET, and ETr Fraction

14
14
14
16
19
19
23
23
24

5. 24-hour evapotranspiration
6. Seasonal evapotranspiration

35
38

Appendices
1.
2.
3.
4.
5.

24
25
34

40

Acquiring header file information

Land-use map and momentum roughness length
Reference evapotranspiration calculation
Preparing the original satellite image for use in SEBAL
Time issues around weather data and reference evapotranspiration
2

6. Tables for surface albedo computations

7. Methods for selecting cold and hot pixels
8. Iteration spreadsheet used for the atmospheric stability correction
9. SAVI and LAI for southern Idaho
10. G/Rn computations for water surfaces and snow
11. The stability of the atmosphere
12. The SEBAL Mountain Model

LIST OF SYMBOLS

cp
d
de-s
dr

Ta
Ts
u
u*
z
zom

air specific heat

zero plane displacement
relative earth-sun distance
inverse squared relative earth-sun
distance
actual evapotranspiration rate
reference evapotranspiration
soil heat flux
sensible heat flux
von Karmans constant
Monin-Obukhuv length
spectral radiance for band
aerodynamic resistance to heat
transport
net radiation flux
incoming shortwave radiation
incoming longwave radiation
outgoing longwave radiation
corrected thermal radiance from the
surface
path radiance in the 10.4 12.5 m
band
narrow band downward thermal
radiation for a clear sky
near surface air temperature
surface temperature
wind velocity
friction velocity
height
momentum roughness length

K
K
m/s
m/s
m
m

path-

surface albedo
albedo path radiance

albedo at top of atmosphere

solar elevation angle
aspect angle of the pixel

degrees
radians

ET
ETr
G
H
k
L
L
rah
Rn
RS
RL
RL
Rc
Rp
Rsky

radiance

toa

J/kg/K
m
mm/hr
mm/hr
W/m2
W/m2
0.41
m
W/m2/sr/m
s/m
W/m2
W/m2
W/m2
W/m2
W/m2/sr/m
W/m2/sr/m
W/m2/sr/m

ET

sw

h
m

declination of the earth

broad band surface emissivity
narrow band surface emissivity
atmospheric emissivity
solar incidence angle
latent heat of vaporization
latent heat flux
air density
reflectivity for band
Stefan-Boltzmann constant
shortwave transmissivity of air
narrow band transmissivity of air
latitude of the pixel
stability correction for heat transport
stability correction for momentum
transport
mountain wind speed weighting
coefficient
weighting coefficient for band

radians
degrees
J/kg
W/m2
kg/m3
5.67 10-8 W/m2/K4
radians
-

LIST OF FIGURES
Page
Figure 1. Surface Energy Balance

Figure 2. Surface Radiation Balance

10

Figure 3. Flow Chart of the Net Surface Radiation Calculation

15

Figure 4. Flow Chart of the Iterative Process for the Calculation of

Sensible Heat (H)

29

Figure 5. Plot of ET and ETrF vs. time

36

Figure 6. A Map of 24-hour Evapotranspiration

37

Figure 7. A Map of Seasonal Evapotranspiration

39

INTRODUCTION
1. BACKGROUND
Land managers today are faced with multiple challenges. They must, as a result of
expanding economic pressures, become advocates for the sustainable use of our natural
resources while ensuring that the quality of the resource is maintained. The public and the
courts are increasingly demanding objective and effective strategies for managing these
exhaustible resources.
Fresh water has become our most precious natural resource and the wise management of
this resource is one of our greatest challenges. An understanding of the natural systems
and the physical laws that govern each component of the hydrologic cycle is very important
for the water resource manager. In particular, the evaporation processes from the various
surfaces of the earth need to be understood in order to achieve a sustainable development
of our water resources. Cropland irrigation is a major consumer of water in semiarid and
arid regions and an efficient and reliable method for determining the consumptive use of
water by crops is crucial for adequate water management. As fresh water becomes
scarcer, competition for fresh water intensifies and better irrigation management will be
required to achieve greater efficiency in the use of this valuable resource.
The combination of the two separate processes whereby water is lost to the atmosphere
due to evaporation from the soil and transpiration from vegetation is referred to as
evapotranspiration (ET). The capability to predict levels of ET would be a valuable asset for
water resource managers. ET is a good indicator of irrigation effectiveness and total water
consumption from vegetation. Evapotranspiration information is useful for irrigation supply
planning, water rights regulation, and river basin hydrologic studies.

2. CONCEPT
Experimentally, the calculation of evapotranspiration can be made quite accurately using
weighing lysimeters, eddy correlation techniques, and the Bowen ratio technique. These
methods are limited, however, because they provide point values of ET for a specific
location and fail to provide the ET on a regional scale. This limitation has motivated the
development of using remotely sensed data from satellites to evaluate ET over vast areas.
The major advantage of remote sensing is that ET can be computed without quantifying
other complex hydrological processes.
ET is highly variable in both space and time. It is variable in space due to the wide spatial
variability of precipitation, hydraulic characteristics of soils, and vegetation types and
densities. It is variable in time due to variability of climate. Satellite images provide an
excellent means for determining and mapping the spatial and temporal structure of ET.

Remote sensing has great potential for improving irrigation management, along with
other types of water management by providing ET estimations for large land surface
areas using a minimal amount of ground data.

3. OBJECTIVES
This manual explains a remote image-processing model for predicting ET termed SEBAL
(Surface Energy Balance Algorithm for Land). SEBAL calculates ET through a series of
computations that generate: net surface radiation, soil heat flux, and sensible heat flux to
the air. By subtracting the soil heat flux and sensible heat flux from the net radiation at the
surface we are left with a residual energy flux that is used for evapotranspiration (i.e.
energy that is used to convert the liquid water into water vapor). This manual describes the
theoretical basis of SEBAL using images from Landsat 5 and 7 satellites. However, the
theory is independent of the satellite type and this manual could be applied to other satellite
images if used with appropriate coefficients.

4. WHO CAN USE THIS MANUAL?

The user of this manual should have background knowledge in hydrologic science or
engineering and environmental physics (theories of mass and momentum transfer), as well
as solar radiation physics. SEBAL was developed for use with ERDAS IMAGINEs Model
Maker tool and experience with this software is a prerequisite. As with any computer model,
only knowledgeable users who understand the theory and who can consistently recognize
the reliability of the computations should use SEBAL. As will be shown, it is also important
that the user be familiar with the area being studied and has some on the ground
knowledge of land use and topography for the area of interest. Following is a list of useful
references, which provide a good background of the scientific methods used in SEBAL:
1.
2.
3.
4.

Monteith, J. L. and Unsworth, M. H. (1990). Principles of Environmental Physics,

Second Edition, Butterworth Heinemann. ISBN 0-7131-2931- X
Campbell, G. S. and Norman, J. M. (1998). An Introduction To Environmental
Biophysics, Second Edition, Springer. ISBN 0-387-94937-2
Allen, R. G., Pereira, L., Raes, D., and Smith, M. (1998). Crop
Evapotranspiration, Food and Agriculture Organization of the United Nations,
Rome, Italy. ISBN 92-5-104219-5. 290 p.
Allen, R.G. (2000). REF-ET: Reference Evapotranspiration Calculation Software
for FAO and ASCE Standardized Equations, University of Idaho.
www.kimberly.uidaho.edu/ref-et/

THE THEORETICAL BASIS OF SEBAL

1. OVERVIEW HOW SEBAL COMPUTES EVAPOTRANSPIRATION (ET)
In the SEBAL model, ET is computed from satellite images and weather data using the
surface energy balance as illustrated in Figure 1. Since the satellite image provides
information for the overpass time only, SEBAL computes an instantaneous ET flux for the
image time. The ET flux is calculated for each pixel of the image as a residual of the
surface energy budget equation:
ET = Rn G H

(1)

where; ET is the latent heat flux (W/m2), Rn is the net radiation flux at the surface (W/m2),
G is the soil heat flux (W/m2), and H is the sensible heat flux to the air (W/m2). The surface
energy budget equation is further explained in part 4 of this section.

Figure 1. Surface Energy Balance

The net radiation flux at the surface (Rn) represents the actual radiant energy available at
the surface. It is computed by subtracting all outgoing radiant fluxes from all incoming
radiant fluxes (Figure 2). This is given in the surface radiation balance equation:
Rn = RS - RS + RL - RL - (1-o)RL

(2)

where; RS is the incoming shortwave radiation (W/m2), is the surface albedo

(dimensionless), RL is the incoming longwave radiation (W/m2), RL is the outgoing
longwave radiation (W/m2), and o is the surface thermal emissivity (dimensionless).

shortwave radiation

longwave radiation

RL
RS

(incident
shortwave)

(incident
longwave)

RS

(1-o)RL

RL

(emitted
longwave)

(reflected
longwave)

(reflected
shortwave)

vegetation surface

Net surface radiation = gains losses

Rn = (1 - ) RS + RL - RL - (1-o)RL
Figure 2. Surface Radiation Balance
In Equation (2), the amount of shortwave radiation (RS) that remains available at the
surface is a function of the surface albedo (). Surface albedo is a reflection coefficient
defined as the ratio of the reflected radiant flux to the incident radiant flux over the solar
spectrum. It is calculated using satellite image information on spectral radiance for each
satellite band. The incoming shortwave radiation (RS) is computed using the solar
constant, the solar incidence angle, a relative earth-sun distance, and a computed
atmospheric transmissivity. The incoming longwave radiation (RL) is computed using a
modified Stefan-Boltzmann equation with atmospheric transmissivity and a selected
surface reference temperature. Outgoing longwave radiation (RL) is computed using the
Stefan-Boltzmann equation with a calculated surface emissivity and surface temperature.
Surface temperatures are computed from satellite image information on thermal radiance.

10

The surface emissivity is the ratio of the actual radiation emitted by a surface to that
emitted by a black body at the same surface temperature. In SEBAL, emissivity is
computed as a function of a vegetation index. The final term in Equation (2), (1-o)RL,
represents the fraction of incoming longwave radiation that is lost from the surface due to
reflection. The radiation balance equation is explained in detail in part 3 of this section.
In Equation (1), the soil heat flux (G) and sensible heat flux (H) are subtracted from the net
radiation flux at the surface (Rn) to compute the residual energy available for
evapotranspiration (ET). Soil heat flux is empirically calculated using vegetation indices,
surface temperature, and surface albedo. Sensible heat flux is computed using wind speed
observations, estimated surface roughness, and surface to air temperature differences.
SEBAL uses an iterative process to correct for atmospheric instability due to the buoyancy
effects of surface heating.
Once the latent heat flux (ET) is computed for each pixel, an equivalent amount of
instantaneous ET (mm/hr) is readily calculated by dividing by the latent heat of vaporization
(). These values are then extrapolated using a ratio of ET to reference crop ET to obtain
daily or seasonal levels of ET. Reference crop ET, termed ETr, is the ET rate expected
from a well-defined surface of full-cover alfalfa or clipped grass and is computed in the
SEBAL process using ground weather data (see Appendix 3).
SEBAL can compute ET for flat, agricultural areas with the most accuracy and confidence.
Appendix 12 describes additional SEBAL applications for mountain areas where elevation,
slope, and aspect need to be considered.

2. DATA REQUIREMENTS AND DATA PREPARATION

SEBAL requires a satellite image and some weather data. A land-use map for the area of
interest is also helpful. This manual applies specifically to images produced from the
Landsat satellites. Landsat Thematic Mapper (TM) bands 1-5 and 7 provide data for the
visible and near infrared bands. The pixel size for these bands is 30m by 30m. TM band 6
provides data for longwave (thermal) radiation. The pixel size for this band is 60m by 60m
for Landsat 7 and 120m by 120m for Landsat 5. We recommend using images from
Landsat 7 since there is more error possible with recent Landsat 5 images due to sensor
degradation. Some Landsat 5 satellite constants that are used in the computations are
updated to 1986 and may be in error for 2002 without adjustment.
A. Cloudy Conditions
It is important that the image used is for a totally clear sky. ET cannot be computed for
cloud covered land surfaces, because even a thin layer of cloud can considerably drop the
thermal band readings and cause large errors in calculation of sensible heat fluxes.
Therefore, all satellite images should be thoroughly checked for the occurrence of cloud
cover and if found, these areas should be masked out and dealt with individually. In order
to detect clouds in an image, view the thermal band using a range of colors to differentiate
temperature values. Clouds will then show up as uniquely colored areas in the image and
11

can be flagged. Masking can be done following the processing of ET, but is an important
final step.
B. Header File Information
TM images are generally created with an associated header file. The header file for the
satellite image is a relatively small file that contains important information for the SEBAL
process. The following information must be obtained from the header file for entry into
SEBAL:
The satellite overpass date and time
The latitude and longitude of the center of the image
The sun elevation angle () at the overpass time
Gain and Bias levels for each band
The satellite overpass time is expressed as Greenwich Mean Time (GMT) and must be
converted to local time (see Appendix 5).
For Landsat 7, the header file includes the gains and biases for bands 1-5 and 7. These
parameters are used to convert digital numbers (DNs) in the original files into energy units.
For band 6, the thermal band, both high and low gain images are supplied and the user
must decide which one to use for SEBAL. We recommend using the low gain image, which
yields slightly lower resolution, but is less likely to suffer from saturation . When the low
gain image is selected, use the gain and bias information for low gain. If the header file is
missing, the user must use the internet to find gain and bias data for the image. Appendix 1
gives information about acquiring header file information.
C. Land-Use Map
A land-use map is not a requirement for SEBAL but is highly recommended since it is
useful in estimating the surface roughness parameter (zom). The land-use map divides the
area of interest into various general classes of land-use such as agriculture, city, water,
desert, forest, grassland, etc. Appendix 2 lists the general land-use types used in the Idaho
application of SEBAL.
D. Weather Data
The following weather data are recommended for processing SEBAL:
Wind speed (required, hourly data is preferred)
Precipitation (daily data are recommended for a several week period prior to the
image)

In addition to wind speed data, the following weather data are recommended for
calculating hourly and daily reference ET. If the reference ET data are available for
the image area, then the following data are not needed.
Humidity (hourly data such as vapor pressure or dew point temperature)
Solar radiation (hourly is preferred)
Air Temperature (hourly is preferred)

12

Reference evapotranspiration, ETr (hourly is preferred, see Appendix 3 for

calculation). The alfalfa ETr is preferred over the clipped grass ETr since it more
nearly represents the upper limit of ET from well-watered vegetation.

The wind speed (u) at the time of the satellite overpass is required for the computation of
sensible heat flux (H) and for the ETr calculation. Precipitation data is used to evaluate the
general wetness of areas that have received rain within four or five days before the image
date. Humidity data are necessary for the ETr calculation. Solar radiation data are useful
for the estimation of the cloudiness of the image and for adjusting the atmospheric
transmissivity (sw). Reference evapotranspiration (ETr) is the estimated ET for a wellwatered reference crop, usually alfalfa. It is used to compute H at the cold anchor pixel
and to compute the ETr fraction (ETrF) that is used to predict 24-hour and seasonal ET.
Weather conditions might not be uniform within a satellite image. If the area of interest has
a variety of terrains and land-uses such as mountains, valleys, agriculture, desert, and
grassland, the weather conditions may vary significantly. If the primary interest is ET from
agricultural fields, one should use a weather station located in an agricultural area. In
cases of widely varying terrain or land-use, one should consider using data from two or
more weather stations. It is important that the weather data come from a location that is
near (within 50 km) the locations of the hot and cold anchor pixels (described later).

13

THE OPERATION OF THE SEBAL MODEL

1. SETTING UP THE IMAGE and ACQUIRING HEADER FILE
INFORMATION
First, a satellite image for the area of interest is required. This manual uses Landsat 5 TM
images or Landsat 7 ETM+ images. Next, the original satellite image must be georectified
so that the final ET map is accurately referenced to earth coordinates. The
georectification can be done by the user or the image can be sent to a third party such as
Earth Satellite Corporation. It is recommended that the georectified image be saved in the
GeoTIF format and that nearest neighbor resampling be used during georectification to
preserve spectral information.
The geo-rectified disk is used with the ERDAS IMAGINE software to produce a layered
spectral band image necessary for the SEBAL procedure. Once a layered image is
produced, a subset image can be made if a smaller area of interest is desired (Appendix 4
contains the ERDAS steps required for this image set up).
Then, header file information is obtained as explained in Appendix 1. Original images for
which the header file information is most accessible are Landsat 7 ETM+ in the NLAPS or
FAST format.

2. WEATHER DATA and REFERENCE ET VALUES

Gather weather data from a relevant weather station (see section 2D above) and organize
it in a table. Now, compute the reference ET as described in Appendix 3 or using other
means. The results are hourly estimates of ETr for the day of the image.
Now, time issues regarding how the hourly weather data were gathered and what time
period the data represent need to be examined (see Appendix 5).

3. SURFACE RADIATION BALANCE EQUATION

The first step in the SEBAL procedure is to compute the net surface radiation flux (Rn)
using the surface radiation balance equation:
Rn = (1-)RS + RL - RL - (1-o)RL

(2)

This is accomplished in a series of steps using the ERDAS Model Maker tool to compute
the terms in the above equation. A flow chart of the process is shown in Figure 3:

14

Figure 3. Flow Chart of the Net Surface Radiation Computation

Rn = (1-)RS + RL - RL - (1-0)RL
model F09

Surface
albedo

Incoming
shortwave

Outgoing
longwave

Incoming
longwave

RS

RL

RL

model F04

spreadsheet

model F08

Albedo-top of
atmosphere

spreadsheet

Surface
temperature

toa

Surface
emissivities

model F03

&

TS
model F07

model F06

Reflectivity

model F02

NDVI
SAVI
LAI

Spectral
radiance

model F05

L
model F01

The computer model number used for each computation is given along with the variable
name. The computation steps begin at the bottom of the figure with model F01 and
continue upward to model F09 for the computation of Rn. The two terms RS and RL are
computed with a spreadsheet or a calculator rather than the Model Maker tool.
15

The terms of Equation (2) and how they are computed are now explained.
A. Surface Albedo ()
Surface albedo is defined as the ratio of the reflected radiation to the incident shortwave
radiation. It is computed in SEBAL through the following steps:
1. The spectral radiance for each band (L) is computed (model F01). This is the
outgoing radiation energy of the band observed at the top of the atmosphere by the
satellite. It is calculated using the following equation given for Landsat 5 and 7:

LMAX LMIN
( DN QCALMIN ) + LMIN
L =
QCALMAX QCALMIN

(3)

where; DN is the digital number of each pixel, LMAX and LMIN are calibration
constants, QCALMAX and QCALMIN are the highest and lowest range of values for
rescaled radiance in DN. The units for L are W/m2/sr/m.
For Landsat 5, QCALMAX = 255 and QCALMIN = 0 so that Equation (3) becomes:

LMAX LMIN
L =
DN + LMIN
255

(4)

Table 6.1 in Appendix 6 gives the values for LMAX and LMIN.
For Landsat 7 with header file data on gains and biases, a simpler equation for L is
given:

L = (Gain DN) + Bias

(5)

where; Gain and Bias refer to the values given in the header file. The low gain data for
band 6 is recommended.
If header file data is not available for Landsat 7, Equation (3) must be used, where
QCALMAX is 255 and QCALMIN is 0 or 1. For the NLAPS generating system,
QCALMIN is 0. For the LPGS generating system, QCALMIN is 1.
Table 6.2 in Appendix 6 gives LMIN and LMAX values for low and high gain. For
Landsat 7, the gain for a band can shift from low to high or vice-versa for any given row.
Therefore, it is important to confirm, for each scene (path/row), the specific gain of each
band. The header file on the original image CD or obtained from the internet provides
information on whether to use low or high gain for a given band (see Appendix 1).
To complete this first step, model F01_radiance, is loaded and run. Values for LMIN and
LMAX, or Gains and Biases are entered into the tables of the model along with the
subset image. The model is saved along with the output file for radiance (L).
16

2. The reflectivity for each band () is computed (model F02). The reflectivity of a
surface is defined as the ratio of the reflected radiation flux to the incident radiation flux.
It is computed using the following equation given for Landsat images:

ESUN cos d r

(6)

where; L is the spectral radiance for each band computed in model F01, ESUN is the
mean solar exo-atmospheric irradiance for each band (W/m2/m), cos is the cosine of
the solar incidence angle (from nadir), and dr is the inverse squared relative earth-sun
distance. Values for ESUN are given in Table 6.3, Appendix 6. Cosine is computed
using the header file data on sun elevation angle () where = (900 - ). The term dr is
defined as 1/de-s2 where de-s is the relative distance between the earth and the sun in
astronomical units. dr is computed using the following equation by Duffie and Beckman
(1980)1, also given in FAO 56 paper: Crop Evapotranspiration (Allen et al., 1998):

d r = 1 + 0.033 cos DOY

365

(7)

where; DOY is the sequential day of the year (given in Table 6.5, Appendix 6), and the
angle (DOY 2/365) is in radians. Values for dr range from 0.97 to 1.03 and are
dimensionless.
To complete step 2, model F02_reflectivity, is loaded and run. Values for ESUN are
entered into the table of the model. (Since band 6 is not shortwave, it is not used in this
computation and is given a dummy value of 1 for ESUN). The constants dr and cos
are also entered along with the radiance file computed in model F01. Again, the output
file for reflectivity () and the model are saved.
3. The albedo at the top of the atmosphere (toa) is computed (model F03). This is the
albedo unadjusted for atmospheric transmissivity and is computed as follows:

toa = ( )

(8)

where; is the reflectivity computed in step 2 and is a weighting coefficient for each
band computed as follows:

ESUN
ESUN

(9)

Duffie, J.A. and W.A. Beckman. 1980. Solar Engineering of Thermal Processes. John Wiley and
Sons, New York, p 1-109.

17

Values for the weighting coefficient, , are given in Table 6.4, Appendix 6, for Landsat
images.
To complete step 3, model F03_albedo_toa, is loaded and run. Values for are
entered into the table of the model with a dummy value of zero for band 6. The file for
reflectivity () computed in model F02 is also input. The model and output file are
saved.
4. The final step is to compute the surface albedo (model F04). Surface albedo is
computed by correcting the toa for atmospheric transmissivity:

toa path _ radiance

sw 2

(10)

where; path_radiance is the average portion of the incoming solar radiation across all
bands that is back-scattered to the satellite before it reaches the earths surface, and sw
is the atmospheric transmissivity.
Values for path_radiance range between 0.025 and 0.04 and for SEBAL we recommend a
value of 0.03 based on Bastiaanssen (2000). Atmospheric transmissivity is defined as
the fraction of incident radiation that is transmitted by the atmosphere and it represents
the effects of absorption and reflection occurring within the atmosphere. This effect
occurs to incoming radiation and to outgoing radiation and is thus squared in Equation
(10). sw includes transmissivity of both direct solar beam radiation and diffuse
(scattered) radiation to the surface. We calculate sw assuming clear sky and relatively
dry conditions using an elevation-based relationship from FAO-56:

sw = 0.75 + 2 10-5 z

(11)

where; z is the elevation above sea level (m). This elevation should best represent the
area of interest, such as the elevation of the relevant weather station.
To complete step 4, model F04_surface_albedo, is loaded and run. The file for toa from
model F03 is entered along with the constants for path_radiance and sw. The model and
output file are saved.
The surface albedo for all pixels has now been computed and these values can be
checked to see if they are realistic. Use the ERDAS Viewer tool to view the surface
albedo image and values. Compare the values for various known surfaces with typical
albedo values given in Table 6.6, Appendix 6.
For conditions having high levels of atmospheric aerosols or dust, the user may wish to
predict separately for each individual band using an atmospheric radiation transfer
model such as Modtran along with radiosonde data for the location and image date. In
most situations, however, because of the design of SEBAL, where the energy balance

18

and ET are specified for the two extreme conditions (maximum ET and minimum ET),
some error in estimated abledo caused by uncertainty in values for transmissivity and
path_radiance, cause little error in the final ET computation. Therefore, for general
application, Equation (10) is adequate.
B. Incoming Shortwave Radiation (RS)
Incoming shortwave radiation is the direct and diffuse solar radiation flux that actually
reaches the earths surface (W/m2). It is calculated, assuming clear sky conditions, as a
constant for the image time using:
Rs = Gsc cos dr sw

(12)

where; Gsc is the solar constant (1367 W/m2), cos is the cosine of the solar incidence
angle as in step 2 above, dr is the inverse squared relative earth-sun distance, and sw is
the atmospheric transmissivity. This calculation is done with a spreadsheet or calculator.
Values for Rs can range from 200 1000 W/m2 depending on the time and location of the
image.
C. Outgoing Longwave Radiation (RL)
The outgoing longwave radiation is the thermal radiation flux emitted from the earths
surface to the atmosphere (W/m2). It is computed in SEBAL through the following steps
(see the flow chart, Figure 3):
1. Three commonly used vegetation indices are computed (model F05). Normalized
Difference Vegetation Index (NDVI), Soil Adjusted Vegetation Index (SAVI), and Leaf
Area Index (LAI) are computed using the reflectivity values found in model F02. Any
one of these indices can be used to predict various characteristics of vegetation,
according to preferences of the user.
The NDVI is the ratio of the differences in reflectivities for the near-infrared band (4)
and the red band (3) to their sum:
NDVI = (4 3) / (4 + 3)

(13)

where; 4 and 3 are reflectivities for bands 4 and 3 and are found in the output file of
model F02. The NDVI is a sensitive indicator of the amount and condition of green
vegetation. Values for NDVI range between -1 and +1. Green surfaces have a NDVI
between 0 and 1 and water and cloud are usually less than zero.
The SAVI is an index that attempts to subtract the effects of background soil from
NDVI so that impacts of soil wetness are reduced in the index. It is computed as:
SAVI = (1 + L) (4 3) / (L + 4 + 3)

(14)

where; L is a constant for SAVI. If L is zero, SAVI becomes equal to NDVI. A value of
0.5 frequently appears in the literature for L. However, a value of 0.1 is used to better

19

represent soils of southern Idaho. The value for L can be derived from analysis of
multiple images where vegetation does not change, but surface soil moisture does
(Appendix 9).
The LAI is the ratio of the total area of all leaves on a plant to the ground area
represented by the plant. It is an indicator of biomass and canopy resistance. LAI is
computed for southern Idaho using the following empirical equation:

0.69 SAVI ID
ln

0.59

LAI =
0.91

(15)

where; SAVIID is the SAVI for southern Idaho calculated from Equation (14) using a
value of 0.1 for L. The maximum value for LAI is 6.0, which corresponds to a maximum
SAVI of 0.687. Beyond SAVI = 0.687, the value for SAVI saturates with increasing LAI
and does not change significantly. Users must realize that the relationship between
SAVI and LAI can vary with location and crop type. Equation (15) is for southern Idaho.
See Appendix 9 for more details of how these equations worked in southern Idaho.
To complete this step, model F05_vegetation_indicies, is loaded and run. The file for
reflectivity () is entered and the three output files saved along with the model.
2. Surface emissivity () is the ratio of the thermal energy radiated by the surface to the
thermal energy radiated by a blackbody at the same temperature. Two surface
emissivities are used in SEBAL. The first is an emissivity representing surface behavior
for thermal emission in the relatively narrow band 6 of Landsat (10.4 to 12.5 m),
expressed as . The second is an emissivity representing surface behavior for
thermal emission in the broad thermal spectrum (6 to 14 m), expressed as . is
used in calculation of surface temperature (Ts) and is used later on to calculate total
long wave radiation emission from the surface.
The surface emissivities are computed in model F06 using the following empirical
equations, where NDVI > 0:

NB = 0.97 + 0.0033 LAI;

for LAI < 3

(16a)

0 = 0.95 + 0.01 LAI;

for LAI < 3

(16b)

and = 0.98 and 0 = 0.98 when LAI 3.

For water and snow we use filters in the model to set the value of and o:
For water; NDVI < 0 and < 0.47,
= 0.99 and o = 0.985
For snow; NDVI < 0 and 0.47,
= 0.99 and o = 0.985

20

To complete step 2, model F06_surface_emissivity, is loaded and run. The files for
NDVI, LAI and surface albedo () are input and the output file and model are saved.
3. The corrected thermal radiance from the surface (Rc) is calculated following Wukelic
et al. (1989)2 as:
L6 R p
Rc =
(1 NB ) R sky
(17)

NB

where; L6 is the spectral radiance of band 6 from model F01 (W/m2/sr/m), Rp is the
path radiance in the 10.4 12.5 m band (W/m2/sr/m), Rsky is the narrow band
downward thermal radiation for a clear sky (W/m2/sr/m), and NB is the narrow band
transmissivity of air (10.4 12.5 m). Units for Rc are W/m2/sr/m.
The corrected thermal radiance (Rc) is the actual radiance emitted from the surface
whereas L6 is the radiance that the satellite sees. In between the surface and the
satellite two things occur. First, some of the emitted radiation is intercepted by the
atmosphere (transmissivity). Second, some thermal radiation is emitted by the
atmosphere in the direction of the satellite (path radiance) and the satellite thinks that
this is from the surface is confused as being emitted from the surface.
Values for Rp and NB require the use of an atmospheric radiation transfer simulation
model such as Modtran and radiosonde profiles representing the image and date. In the
absence of values for these terms, they can be ignored by setting Rp = 0 and NB = 1
and the Rsky term can also be ignored by setting Rsky = 0. This converts Rc into an
uncorrected radiance (L6). Fortunately, the effects of the three parameters on Rc are
largely self-canceling. However, the result of no correction to L6 will be a general
underestimation of surface temperature (Ts) by up to about 5o C for warmer portions of
an image. However, because SEBAL tailors the dT function, described later, around
the Ts data calculated for an image, the impact on final ET values is generally small,
especially for areas having low and high values of ET. Errors may be larger for
midrange ET values, but generally less than a few percent. The Rsky term is small
enough that it can probably be ignored all of the time. If calculated, it can be based on
an Idso-Jackson style of empirical formula as applied by Wukelic et al. (1989):

R sky = (1.807 x 10 10 ) Ta4 1 0.26 exp 7.77 x 10 4 [273.15 Ta ]2

)]

(18)

where; Ta is the near surface air temperature at the time of the image (K).

Wukelic, G.E., D.E. Gibbons, et al., (1989). Radiometric calibration of Landsat thematic mapper
thermal band. Remote Sensing of Environment 28:339-347.

21

The corrected thermal radiance (Rc) is computed in model F07 along with the surface
temperature as shown below.
4. The surface temperature (Ts) is computed in model F07 using the following modified
Plank equation:

Ts =

K2

(19)

K
ln NB 1 + 1

Rc

where; Rc is the corrected thermal radiance from the surface using L6 from model F01.
K1 and K2 are constants for Landsat images (Table 1). Units for Rc must be the same as
those for K1.
Table 1. Constants for Equation 19 for Landsat 5 TM in mW/cm2/sr/m (Markham and
Barker, 1986), and for Landsat 7 ETM+ in W/m2/sr/m (Landsat 7 Science User Data
Handbook Chap.11, 2002)

Landsat5 TM Band6
Landsat7 ETM+ Band6

K1

K2

607.76
666.09

1260.56
1282.71

To complete step 4, model F07_surface_temp, is loaded and run. Values for K1 and K2
and Rp, NB, and Rsky are input along with the files for spectral radiance (L) and
narrow band emissivity (). The output file and model are saved. (0, 1, and 0 are
default values for Rp, NB, and Rsky).
5. The outgoing longwave radiation (RL) is computed (model F08). This is computed
using the Stefan-Boltzmann equation:
RL = o Ts4

(20)

where; is the broad-band surface emissivity (dimensionless), is the StefanBoltzmann constant (5.67 10-8 W/m2/K4), and Ts is the surface temperature (K).
To complete this step, model F08_longwave_out, is loaded and run. Files for the
surface emissivity (o) and surface temperature (TS) are entered and the output file and
model saved. Values for RL can range from 200 700 W/m2 depending on the location
and time of the image.

22

D. Choosing the Hot and Cold Pixels

The SEBAL process utilizes two anchor pixels to fix boundary conditions for the energy
balance. These are the hot and cold pixels and are located in the area of interest. The
cold pixel is selected as a wet, well-irrigated crop surface having full ground cover by
vegetation. The surface temperature and near-surface air temperature are assumed to be
similar at this pixel. The hot pixel is selected as a dry, bare agricultural field where ET is
assumed to be zero. Both of these anchor pixels should be located in large and
homogeneous areas that contain more than one band 6 pixel (i.e. 60m60m for Landsat 7
and 120m120m for Landsat 4&5).
The selection of these anchor pixels requires skill and practice. The quality of the ET
computations in SEBAL depends on a careful selection of these two pixels. Appendix 7
contains a guide for choosing the hot and cold pixels. Once selected, the surface
temperature and x and y coordinates for these two pixels are recorded for future
computations.
E. Incoming Longwave Radiation (RL)
The incoming longwave radiation is the downward thermal radiation flux from the
atmosphere (W/m2). It is computed using the Stefan-Boltzmann equation:
RL = a Ta4

(21)

where; a is the atmospheric emissivity (dimensionless), is the Stefan-Boltzmann constant

(5.67 10-8 W/m2/K4), and Ta is the near surface air temperature (K). The following
empirical equation for a by Bastiaanssen (1995)3 is applied using data from alfalfa fields in
Idaho:

a = 0.85 (-ln sw).09

(22)

where; sw is the atmospheric transmissivity calculated from Equation (11). The original
coefficients by Bastiaanssen (1995), derived for western Egypt, were a = 1.08 (-ln
sw).265.
Substituting Equation (21) into Equation (20) and using Tcold from our cold pixel for Ta
yields the following equation:
RL = 0.85 (-ln sw).09 Tcold4

(23)

This computation is done using a spreadsheet or calculator. Values for RL can range from
200 500 W/m2, depending on the location and time of image. Some automatic weather
stations include a net radiometer that measures all incoming and outgoing short and
longwave radiation. Measured RL data can be used to check computed values.
3

Bastiaanssen, W.G.M. 1995. Regionalization of surface flux densities and moisture indicators in
composite terrain: A remote sensing approach under clear skies in Mediterranean climates. Ph.D.
dissertation, CIP Data Koninklijke Bibliotheek, Den Haag, The Netherlands. 273 p.

23

F. Solving the Surface Radiation Balance Equation for Rn

The net surface radiation flux (Rn) is now computed using Equation (2). Load and run
model F09_net_radiation. Files for surface albedo (), outgoing longwave radiation (RL),
and surface emissivity (o) are input along with the incoming shortwave radiation (RS) and
the incoming longwave radiation (RL). The output file for Rn and the model are saved.
Values for Rn can range from 100 700 W/m2, depending on the surface. This completes
the first step of the SEBAL procedure.

4. SURFACE ENERGY BUDGET EQUATION

The second step of the SEBAL procedure is to compute the terms G and H of the surface
energy budget equation. The net radiation flux (Rn) is the net amount of radiant energy that
is available at the surface for warming the soil, warming the air, or evaporating soil
moisture. This is written as the surface energy budget equation:
Rn = G + H + ET

(23)

where; Rn is the net radiation at the surface (W/m2), G is the soil heat flux (W/m2), H is the
sensible heat flux to the air (W/m2), and ET is the latent heat flux (W/m2).
SEBAL computes ET as a residual of net radiant energy after soil heat flux and sensible
heat flux are subtracted:

ET = Rn G H

(1)

This equation is solved through the following steps using the ERDAS Model Maker tool.
A. Soil Heat Flux (G)
Soil heat flux is the rate of heat storage into the soil and vegetation due to conduction.
SEBAL first computes the ratio G/Rn using the following empirical equation developed by
Bastiaanssen(2000) representing values near midday:
G/Rn = Ts/ (0.0038 + 0.00742)(1 - .98NDVI4)

(24)

where; Ts is the surface temperature (o C), is the surface albedo, and NDVI is the
Normalized Difference Vegetation Index. G is then readily calculated by multiplying G/Rn by
the value for Rn computed in model F09.
Soil heat flux is a difficult term to evaluate and care should be used in its computation. One
must understand the area of interest in order to evaluate the accuracy of Equation (24).
Values of G should be checked against actual measurements on the ground. Land
classification and soil type will affect the value of G and a land-use map is valuable for
identifying the various surface types (Appendix 2). Equation (24) predicts G measured for

24

irrigated crops near Kimberly, Idaho quite accurately (M.Tasumi and R.Allen, 2002, pers.
commun.).
In the Idaho applications of SEBAL, values of G/Rn for water and snow are assigned as
follows (Appendix 10), representing values near midday:
If NDVI < 0; assume surface is water;
G/Rn = 0.5
If Ts < 4 oC and > 0.45; assume surface is snow; G/Rn = 0.5
The estimation of G/Rn for deep, clear lakes is complex. It may be large in early summer
when the lake is colder than the air and smaller in the fall when the lake is warmer. Any
significant water bodies in the area of interest should be flagged and G/Rn carefully
analyzed (Appendix 10). G/Rn for turbid or shallow water bodies will be less than 0.5 due to
absorption of short-wave radiation more near the water surface.
To complete this step, model F10_soil_heat_flux is loaded and run. Files for NDVI, surface
temperature, surface albedo, and Rn are entered and the output files and model are saved.
View the image for G/Rn and check the values for various land types (Table 2).
Table 2. Estimates of G/Rn for various surfaces
Surface type
Deep, clear water
Snow
Desert
Agriculture
Bare soil
Full cover alfalfa
Rock

G/Rn
0.5
0.5
0.2 0.4
0.05 0.15
0.2 0.4
0.04
0.2 0.6

B. Sensible Heat Flux (H)

Sensible heat flux is the rate of heat loss to the air by convection and conduction, due to a
temperature difference. It is computed using the following equation for heat transport:
H = ( cp dT) / rah
(25)
where; is air density (kg/m3), cp is air specific heat (1004 J/kg/K), dT (K) is the
temperature difference (T1 T2) between two heights (z1 and z2), and rah is the
aerodynamic resistance to heat transport (s/m).
The sensible heat flux (H) is a function of the temperature gradient, surface roughness, and
wind speed. Equation (25) is difficult to solve because there are two unknowns, rah and dT.
To facilitate this computation, we utilize the two anchor pixels (where reliable values for H
can be predicted and a dT estimated) and the wind speed at a given height.
The aerodynamic resistance to heat transport (rah) is computed for neutral stability as:

25

z
ln 2
z
rah = 1
u* k

(26)

where; z1 and z2 are heights in meters above the zero plane displacement (d) of the
vegetation, u* is the friction velocity (m/s) which quantifies the turbulent velocity fluctuations
in the air, and k is von Karmans constant (0.41). The friction velocity (u*) is computed using
the logarithmic wind law for neutral atmospheric conditions:

u* =

ku x
z
ln x
z om

(27)

where; k is von Karmans constant, ux is the wind speed (m/s) at height zx, and zom is the
momentum roughness length (m). zom is a measure of the form drag and skin friction for the
layer of air that interacts with the surface.
Figure 4 shows how the following steps are used to compute sensible heat flux (H):
1. The friction velocity (u*) at the weather station is calculated for neutral atmospheric
conditions using Equation (27). (An explanation of neutral conditions is given in
Appendix 11) The assumption of neutral conditions for the weather station is reasonable
if the weather station is surrounded by well-watered vegetation. The calculation of u*
requires a wind speed measurement (ux) at a known height (zx) at the time of the
satellite image. This measurement is taken from the weather data that was set up
earlier in the SEBAL process. The momentum roughness length (zom) is empirically
estimated from the average vegetation height around the weather station using the
following equation (Brutsaert, 1982):
zom = 0.12h

(28)

where; h is the vegetation height (m). (At the Kimberly weather station in southern
Idaho, h=0.3m was chosen to represent the weather station surrounded by a small 50 m
expanse of 0.15 m grass with other taller agricultural crops upwind of that.) This
calculation is done on a calculator or spreadsheet.
2. The wind speed at a height above the weather station where one can assume no
effect from the surface roughness is calculated. This height is referred to as the
blending height and 200 meters is used. u200 is calculated using a rearranged
Equation (27):

u 200

200

ln
z om

= u*
k

26

(29)

where; u* is the friction velocity at the weather station. This calculation is also done on a
calculator or spreadsheet.
3. The friction velocity (u*) for each pixel is computed using the ERDAS Model Maker
tool. u200 is assumed to be constant for all pixels of the image since it is defined as
occurring at a blending height unaffected by surface features. From Equation (27):

u* =

ku 200
200

ln
z om

(30)

where; zom is the particular momentum roughness length for each pixel.
Momentum Roughness. The momentum roughness length (zom) for each pixel can be
computed by two methods:
1) Using a land-use map:
When a land-use map is available, values of zom for non-agricultural surface features
(such as water, forest, desert grass, snow, manmade structures, etc.) can be
assigned using the values given in Appendix 2.
For agricultural areas, zom is calculated as a function of Leaf Area Index (LAI):
zom = 0.018 LAI

(31a)

To compute zom, model F11_surface_roughness, is loaded and run. The file names
for LAI and the land-use map are specified. Only the section of the land-use map
that corresponds to the area of interest is used, so one should define the proper
extent of the land-use map (the map has been previously created outside of
SEBAL). The extent is specified by transferring the four coordinates defining the
extent of the LAI image into the land-use map using the following steps: View the
LAI image and record the four coordinates from the image info icon. In the model,
click on the land-use map and enter it from a CD or disk. Enter the four coordinates
of the LAI image. Save the output file and the model.
2) Using NDVI and surface albedo data:
When a land-use map is not available, the second method for computing the
momentum roughness length using NDVI and surface albedo data by Bastiaanssen
(2000) and modified by Allen (2001)4 can be used. In this method, zom is computed
from the following equation:
4

Bastiaanssen, W.G.M. 2000. SEBAL-based sensible and latent heat fluxes in the
irrigated Gediz Basin, Turkey. J. Hydrology 229:87-100.
Allen, R.G. 2001. Prediction of regional ET for the Tampa Bay, Florida watershed
using SEBAL. personal communication.
27

zom = exp[(a NDVI/) + b]

(31b)

where; a and b are correlation constants derived from a plot of ln(zom) vs NDVI/ for
two or more sample pixels representing specific vegetation types. Equation (31b) is
an empirical expression that must be fitted to the local vegetation and conditions.
The use of albedo () to modify NDVI helps to distinguish between some tall and
short vegetation types that may have similar NDVI. For example, trees generally
have smaller albedo than agricultural crops. Other variations on Equation (31b) can
be explored by the user to improve the predictive accuracy.
To determine a and b in Equation (31b), a series of sample pixels representing
vegetation types and conditions of interest are selected from the respective image
and the associated values for NDVI and surface albedo are obtained. For each type
or condition, a characteristic vegetation height (h) is assigned. A value for zom is
computed using Equation (28), however the constant (0.12) may vary depending on
the type of vegetation represented by the pixel (for agricultural crops, 0.12 is used;
for an open forest, 0.2 may be more appropriate; for a tight canopy forest, a value
less than 0.2 may be more appropriate). The ln(zom) is plotted against NDVI/ in a
spreadsheet and the correlation coefficients, a and b, are derived by regression or
visual fitting.
Zom for each pixel is now computed using model
F11_surface_roughness_using_NDVI that utilizes a fitted Equation (31b). The files
for surface albedo and NDVI are input along with values for the correlation
coefficients a and b. Save the output file and the model.

28

Weather station
H for each pixel

u, zx, zom , u*

wind speed at
200 meters

friction velocity
at each pixel

u 200

z
ln 200
z
= u * 0m
k

u* =

rah for each pixel

dT
rah
3

L=

ku200
z
ln 200
z0 m

Cp u * Ts
kgH

m (blending)
h (z1)

z
ln 2
z
rah = 1
u* k

u* =

Cold Pixel

Hot Pixel

Hcold=Rn-G-ETcold

Hhot=Rn-G

dTcold=Hcold*rah/(Cp)

dThot = Hhot*rah /(Cp)

dT for each pixel

H = Cp

h (z2)

k u 200
z
ln 200 m(200)
z 0m

z
ln 2 h (z 2 ) + h (z1 )
z
rah = 1
u* k

dT=aTs+b

Figure 4. Flow Chart of the Iterative Process for the Calculation of Sensible Heat (H)

29

4. The aerodynamic resistance to heat transport (rah) is computed in model F12 as is u* in

step 3. A series of iterations is required to determine the value for rah for each period that
considers the impacts of instability (i.e., buoyancy) on rah and H. Assuming neutral
atmospheric conditions, an initial rah is computed using Equation (26). z1 is the height just
above the zero plane displacement (d 0.67 height of vegetation) for the surface or
crop canopy and z2 is some distance above the zero plane displacement, but below the
height of the surface boundary layer. Based on experienced analysis, values of 0.1 meter
for z1 and 2.0 meters for z2 are assigned. Steps 3 and 4 are completed by running model
F12 to compute initial values of u* and rah. The file for zom from model F11 is input along
with u200, which was calculated above with equation (29). Save the output files and the
model5.
5. To compute the sensible heat flux (H) from Equation (25), the near surface
temperature difference (dT) for each pixel needs to be defined. This is given as
dT = Tz1 Tz2. The air temperature at each pixel is unknown, along with explicit values
for Tz1 and Tz2. Therefore, only the difference dT is utilized. SEBAL computes dT for
each pixel by assuming a linear relationship between dT and Ts:
dT = b + aTs

(32)

where; b and a are the correlation coefficients. To define these coefficients, SEBAL
uses the two anchor pixels where a value for H can be reliably estimated. Values for H
and dT at these anchor pixels are computed in a spreadsheet format as explained in
Appendix 8 and in the following steps. The linearity of the dT vs. Ts function is a major
presumption of SEBAL. However, research by Bastiaanssen and others and by the
University of Idaho at Kimberly, indicate that this assumption appears to fit a large range
of conditions.
a) At the cold pixel, we define the sensible heat flux as Hcold = Rn G - ETcold
from Equation (23). Experience in Idaho shows that the most cold (wet)
agricultural fields have an ET rate about 5% larger than the reference ET (ETr).
Hence, ETcold is assumed to be 1.05 ETr (which was calculated earlier in the
SEBAL process, Appendix 3). Hcold is now calculated in the spreadsheet as:
Hcold = Rn G 1.05ETr. dTcold is computed in the spreadsheet using the
inverse of Equation (25):
dTcold = Hcold rah_cold / (cold cp)

(33)

Note that because dT is defined as T between z1 and z2, the estimate of Tair as Tair = Ts + dT is not a
real, representation of air temperature. This is because the T at z1 is generally somewhat less than Ts. In
addition, Ts is a radiometrically derived surface temperature, whereas T at z1 is an aerodynamically
representative temperature.

30

Note: An alternative method used by Bastiaanssen is to choose the cold

pixel at a body of water where one assumes Hcold = 0; and therefore, ETcold =
Rn G and dTcold = 0. The use of Equation (33) and ETr helps to consider the
impact of regional advection and wind speed on ET. It also helps to account
for any error in image estimates for albedo and Rn by SEBAL.
b) At the hot pixel, Hhot = Rn G - EThot; where EThot is assumed to be zero for a
hot (dry) agricultural field having no green vegetation and with dry soil surface
layer. The weather data should be checked to see if this assumption is correct. If
there was some precipitation 1-4 days before the image date, then EThot should
be estimated using a water balance model and tracking the soil moisture of the
hot pixel. Hhot is now calculated in the spreadsheet and dThot computed from
Equation (33).

Plotting dTcold vs. Ts_cold and dThot vs. Ts_hot gives the following graph:

The correlation coefficients, b and a, are now computed for the linear relationship: dT =
b + aTs. The temperature differential (dT) for each pixel can now be computed using the
coefficients b and a and the surface temperature (Ts). The determination of dT at the
anchor pixels and b and a are done in the spreadsheet. Figure 4 shows a flow chart of
the process.
6. An approximation for air temperature (Ta) for each pixel is computed as Ta = Ts dT
and an approximation for air density () is calculated in the spreadsheet for the anchor
pixels.
7. The sensible heat flux (H) for each pixel is computed in SEBAL using Equation (25).
This is the first estimation of H assuming neutral atmospheric conditions. Model F13 is

31

used for this computation. The file for the initial value of rah from model F12, the surface
temperature file, and the correlation coefficients b and a are input. The output is saved
as H1 and the model is saved.
8. In order to account for the buoyancy effects generated by surface heating, SEBAL
applies the Monin-Obukhov theory in an iterative process (Figure 4). Atmospheric
conditions of stability have a large effect on the aerodynamic resistance (rah) and must
be considered in the computation of sensible heat flux (H), especially for dry conditions.
Appendix 11 gives details of different stability conditions that can occur in the
atmosphere. SEBAL repeats the computation of H through a number of iterations, each
one correcting for buoyancy effects, until the value for rah stabilizes.
The Monin-Obukhov length (L) is used to define the stability conditions of the
atmosphere in the iterative process (note, this is not the same L as used in the SAVI
computation). It is a function of the heat and momentum fluxes and is computed as
follows:

L=

c p u* 3Ts
kgH

(34)

where; is the density of air (kg/m2), cp is the air specific heat (1004 J/kg/K), u* is the
friction velocity (m/s), Ts is the surface temperature (K), g is the gravitational constant
(9.81 m/s2), and H is the sensible heat flux (W/m2). This computation is carried out in
model F14. Values of L define the stability conditions of the atmosphere. If L<0, the
atmosphere is considered unstable; if L>0, the atmosphere is considered stable; if L=0,
the atmosphere is considered neutral (Appendix 11).
Depending on the atmospheric conditions, the values of the stability corrections for
momentum and heat transport (m and h) are computed by the model F14 as follows,
using the formulations by Paulson (1970) and Webb (1970)6:
If L<0; unstable conditions:

1 + x( 200 m ) 2
1 + x( 200 m )
2 ARCTAN (x
+ ln
m ( 200 m ) = 2 ln
( 200 m ) ) + 0.5

2
2

2
1 + x
( 2m )

h ( 2m ) = 2 ln

(35)

(36a)

Paulson, C.A. 1970. The mathematical representation of wind speed and temperature profiles in the
unstable atmospheric surface layer. . Appl. Meteorol. 9:857-861.

Webb, E.K. 1970. Profile relationships: the log-linear range, and extension to strong
stability. Quart. J. Roy. Meteorol. Soc. 96:67-90.
32

2
1 + x
( 0.1m )

h ( 0.1m ) = 2 ln

(36b)

where;

200

x (200m ) = 1 16

x (2m ) = 1 16
L

0.1

x (0.1m ) = 1 16

If L>0; stable conditions:

0.25

(37a)

0.25

(37b)

0.25

(37c)

2
L

m ( 200 m ) = 5
2
L
0.1
h (0.1m ) = 5
L

h (2m ) = 5

(38)

(39a)

(39b)

If L=0; neutral conditions: m and h = 0

9. A corrected value for the friction velocity (u*) is now computed for each successive
iteration as:

u* =

u 200 k
200
m ( 200 m )
ln
z0m

(40)

where; u200 is the wind speed at 200 meters (m/s), k is von Karmans constant (0.41),
zom is the roughness length for each pixel (m), and m(200m) is the stability correction for
momentum transport at 200 meters (equation 35 or 38). This computation is carried out
in model F14.
10. A corrected value for the aerodynamic resistance to heat transport (rah) is now
computed during each iteration as:

33

z
ln 2 h ( z ) + h ( z )
2
1
z
rah = 1
u* k

(41)

where; z2 = 2.0 meters, z1 =0.1 meters, and h(z2) and h(z1) are the stability corrections
for heat transport at 2 meters and 1 meters (equations 36 or 39). This computation is
carried out in model F14.
11. Return to step 5 and the spreadsheet to compute new dT values for the cold and
hot pixel using the corrected rah. New values for the correlation coefficients b and a are
also computed and then dT for each pixel is revised as dT = b + aTs.
12. Step 6 is repeated for a revised value for air temperature (Ta) and air density ().
13. Step 7 is repeated to compute a corrected value for H.
14. Step 8 is repeated to compute a new stability correction.
15. This iterative process is repeated until the successive values for dThot and rah at the
hot pixel have stabilized.

Refer to Figure 4 for a flow chart of the iteration process and to Appendix 8 for
instruction on the use of the spreadsheet.
16. Next, compute the final corrected value for the sensible heat flux (H) at each pixel,
which will be used in the computation of the instantaneous ET at each pixel.
C. Latent Heat Flux (ET), Instantaneous ET (ETinst), and Reference ET
Fraction (ETrF)
Latent heat flux is the rate of latent heat loss from the surface due to evapotranspiration. It
can be computed for each pixel using Equation (1):

ET = Rn G H

(1)

where; ET is an instantaneous value for the time of the satellite overpass (W/m2).
1. An instantaneous value of ET in equivalent evaporation depth is computed as:

ETinst = 3600

ET

34

(42)

where; ETinst is the instantaneous ET (mm/hr), 3600 is the time conversion from
seconds to hours, and is the latent heat of vaporization or the heat absorbed when a
kilogram of water evaporates (J/kg).
2. The Reference ET Fraction (ETrF) is defined as the ratio of the computed
instantaneous ET (ETinst) for each pixel to the reference ET (ETr) computed from
weather data:
ETrF =

ETinst
ETr

(43)

where; ETinst is from Equation 41 (mm/hr) and ETr is the reference ET at the time of the
image from the REF-ET software (mm/hr). ETrF is similar to the well-known crop
coefficient, Kc. ETrF is used to extrapolate ET from the image time to 24-hour or longer
periods.
One should generally expect ETrF values to range from 0 to 1. At a totally dry pixel,
ET = 0 and ETrF = 0. A pixel in a well established field of alfalfa or corn can occasionally
have an ET slightly greater than ETr and therefore ETrF > 1, perhaps up to 1.1.
However, ETr generally represents an upper bound on ET for large expanses of wellwatered vegetation. Negative values for ETrF can occur in SEBAL due to systematic
errors caused by various assumptions made earlier in the energy balance process.
ETinst and ETrF are computed in model F25 along with 24-hour ET as shown in Section 5
below.

5. 24-Hour Evapotranspiration (ET24)

Daily values of ET (ET24) are often more useful than instantaneous ET. SEBAL computes
the ET24 by assuming that the instantaneous ETrF computed in model F25 is the same as
the 24-hour average. Figure 5 shows a plot of ET and ETrF vs. time. This graph shows how
ET varies throughout the day while ETrF is relatively constant. Finally, the ET24 (mm/day)
can be computed as:

ET24 = ETrF ETr _ 24

(44)

where; ETr-24 is the cumulative 24-hour ETr for the day of the image. This is calculated by
adding the hourly ETr values over the day of the image.
ET24 is computed in model F25. Enter the files for H, G, Rn, Ts, and the values for ETr and
ETr-24. Save the output files and the model.

35

4/18/1989, Sugar Beet

Data by Dr. Jim Wright, Kimberly ID

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

ET SB(mm)
ETrF

19:00

17:00

15:00

13:00

11:00

9:00

0
7:00

ETrF

ET (mm/hr)

ETr (mm)
ETrF24

ETo (mm)

5/20/1989, Sugar Beet

Data by Dr. Jim Wright, Kimberly ID

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

ET SB(mm)
ETrF

ETr (mm)
ETrF24

19:00

17:00

15:00

13:00

11:00

9:00

0
7:00

ETrF

ET (mm/hr)

ETo (mm)

Figure 5. Plot of ET and ETrF vs. time (from observations of Dr. Jim Wright, Kimberly,
Idaho

36

Figure 6 shows an example of a daily evapotranspiration map. The range of values for ET24
can be colored using the ERDAS Imagine software so that ET intensity can be readily
observed on the image.

Path 39: Am. Falls - 24-hour ET

8/07/00

Figure 6. A Map of 24-hour Evapotranspiration

37

6. Seasonal Evapotranspiration (ETseasonal)

A seasonal evapotranspiration map that covers an entire growing season is often valuable.
This can be derived from the 24-hour evapotranspiration data by extrapolating the ET24
proportionally to the reference evapotranspiration (ETr). We assume that the ET for the
entire area of interest changes in proportion to the change in the ETr at the weather station.
ETr is computed for a specific location and therefore does not represent the actual
condition at each pixel. This does not matter, however, since ETr is used only as an index
of the relative change in weather, and therefore ET, for the image area. We also assume
that the ETrF computed for the time of the image is constant for the entire period
represented by the image. The following steps show the process for computing seasonal
ET:
1. The first step is to decide the length of the season for which ET is desired (i.e.,
March 1 to October 31).
2. The second step is to determine the period represented by each satellite image
within the chosen season (i.e., if the first two images for the above season are on
March 15 and April 8, then the period represented by the March 15th image would
be March 1 to March 27).
3. The third step is to compute the cumulative ETr for the period represented by the
image. This is simply the sum of daily ETr values over the period. These 24-hour
values can be computed using the University of Idaho REF-ET software
described in Appendix 3. The same ETr method must be used through the
SEBAL process. ETr should represent the alfalfa reference ET, which is a larger
value than for the clipped grass reference (ETo).
4. The fourth step is to compute the cumulative ET for each period as follows:
n

ET period = ETr F period ETr 24

(45)

where; ETrFperiod is the representative ETrF for the period, ETr-24 is the daily ETr,
and n is the number of days in the period. Units for ETperiod will be in mm when
ETr-24 is in mm/day.
5. The fifth step is to compute the seasonal ET by summing all of the ETperiod values
for the length of the season.
The following difficulties encountered in the computation of seasonal ET must be
understood:

38

1. If there is some cloud cover in one of the images used for the seasonal ET
computation, then there can be no ET or ETrF values represented for this area
during the period represented by the image. One can assign ETrF values to the
cloud-covered areas by interpolating between the ETrF values for the images on
either side of the cloud-covered image.
2. When the ETr at the weather station is not representative of the entire area (i.e.,
where the image spans many mountain valleys or includes both coastal and
inland areas having different effects from clouds or wind that are not well
correlated across the image), then the image should be broken down into subimages, each with its own value of ETr. In this situation, simple SEBAL runs
should be made for each sub-image.
3. If an agricultural field is dry on the day of the image and is then irrigated on a
following day, the increase in ET for this field during the representative period of
the image will not be considered in the computation of seasonal ET. This
problem can be minimized if many satellite images are used for the computation
of seasonal ET with each image representing a shorter time period. One can also
apply a water balance model to each pixel that includes irrigation and
precipitation as inputs.
Figure 7 shows an example of a seasonal evapotranspiration map that can be created for
the area of interest:

Seasonal ET
for Bear
River Basin

Figure 7. A Map of Seasonal Evapotranspiration

39

Appendix 1
Acquiring Header File Information

The header file of the satellite image contains important information for SEBAL processing.
The following information must be obtained from the header file:
The satellite overpass date and time
The latitude and longitude of the image
The sun elevation angle, , at the overpass time (used in the Flat Model, only)
Gain and Bias levels for each band (applicable for Landsat 7 ETM+, only)
There are three methods for obtaining the header file information depending on the satellite
type (Landsat 5TM or Landsat 7ETM+) and on the format used to store the image on the
distribution disk.
Method A: Method A is the most straightforward method for acquiring header file
information. It is applicable for the following satellite types and data formats:
Landsat 5 TM, where the original image format is NLAPS
Landsat 7 ETM+, where the original image format is NLAPS or FAST

For the NLAPS format, explore the original image disk. Find and open the small file titled
NLAPS Correction Processing Report. This file contains all the information needed
including the gains and biases for Landsat 7. For band 6 of Landsat 7, the top row of the
gain and bias section is for the low gain image and the second row is for the high gain
image. We recommend using the low gain image as explained in section 2 B of the main
text.
For Landsat 7 FAST format, explore the original image disk. Open the bottom two small
FST files. These contain the gains and biases for bands 1-7 with biases in the first column
and gains in the second column. The list with six entries is for bands 1-5 and 7. The list with
two entries is for band 6 (low gain on top, high gain below). The satellite overpass time is
not included in the FAST header files and Method B must be used to acquire exact image
time from the internet.
Print the header file information for future reference.
Method B: Method B is used to obtain header file information via the internet. It is
applicable for the following data formats and satellite types:
Original image format is NLAPS and satellite type is Landsat 5TM (if the header file
is missing)
Original image format is FAST and satellite type is Landsat 5TM (regardless the
header file availability)
Original image format is FAST and satellite type is Landsat 7ETM+ (only if header
file is available)

40

For all other formats (including GeoTIFF) where the satellite type is Landsat 5 TM
(regardless the header file availability)

(1) First, go to the Earth Explorer web page: http://edcsns17.cr.usgs.gov/EarthExplorer/

(2) Click Enter as a guest.
(3) In the Data Set Selection box, select TM (Landsat 5) or ETM+ (Landsat 7) depending
on the image, and click Continue.

(4) Define Acquisition Date and Path Row Search, and click the search button.

(5) When the search is finished, select the appropriate image and click Result.

41

(6) Click Show in the Show All Fields category, for the appropriate image.

(7) All necessary header file information including the satellite overpass time (use either
start time or stop time) is presented.
It is recommended that the header file information be printed for future reference.
Method C: Method C is applicable for the following data formats and satellite types:
Original image format is NLAPS and satellite type is Landsat 7ETM+ (if the header
file is missing)
Original image format is FAST and satellite type is Landsat 7ETM+ (if the header
file is missing)
For all other formats (including GeoTIFF) when satellite type is Landsat 7ETM+
(regardless the header file availability)

The following is the procedure for obtaining header file information (including the satellite
overpass time) from the internet for the specified data types. This procedure was obtained
from Dr. Satya Kalluri of Raytheon Company:
A. The core metadata that contains much needed info like solar zenith
and azimuth angles, gain settings etc. will not be in the header of the
GeoTIFF files! You should get this as a separate ascii file when you
get your L7 data. If you do not have this, you can get it online for
each scene using the ECS data gateway. This is how you go about doing it:
Go to http://edcimswww.cr.usgs.gov/pub/imswelcome/
Login as a guest or register
Under Choose Search Key Words - Method 2 - click on DATA SET and click
on SELECT
On the next screen, select LANDSAT-7 LEVEL-0R WRS-SCENE V002 under Data
Set list 2, and click on OK. This will take you back to the main page.

42

On the main page, select Type in Path/Row Range under Choose Search Area
Enter your path and row
Enter your start date and end date
Click on Start Search!
The search should return your granule (image) as a result.
Click on "Granule attributes" button, and this will give you all the
metadata you need including the gain settings and solar angles for that
particular scene.
When you follow these instructions, you obtain the equivalent of the header file information
as shown below.
Landsat7 ETM+ P25/R39

9/20/1999

All of the data in the following table are available in the "Granule attributes" (Header of the image has equiverent information too)
Date and Time, in GMT
Gain combination
Center of the image
Sun Elevation
Band12345(66)7
Lat
Lon
Degree
Start Date 20 Sep 1999, 16:43:32
HHHHH(HL)H
30.31 Lat
-94.79 Lon
53.9359
Stop Date 20 Sep 1999, 16:43:32
Band 6 has both High and Low gains
I recommend to use Low gain (but High gain is okay too)
Comment
(Low gain has bit less resolution but safer)
We do not use band 8 then I do not includes band 8
(band 8 is always Low gain)

Note that Band 6 has both High and Low gain images (files). For SEBAL, we
recommend using the Low gain image, even though it has slightly less resolution,
since it is less likely to suffer from saturation. SEBAL does not use Band 8 (the
Panchromatic Band), so header information for this band is not described here.

43

Appendix 2
Land-Use Map and Momentum Roughness Length
A. Land-Use Map
A land-use map is used in SEBAL for determining the momentum roughness length (zom)
and for establishing G/Rn values for various land-use types. If a land-use map is not
available for the area of interest, the user can develop one using satellite data. The landuse map helps the user develop a good working knowledge of the area being studied so he
can better recognize the reliability of the SEBAL computations. The land-use map should
be in a raster format with the same pixel size as the satellite image (30m 30m).

The land-use map for the Snake River Plain of southern Idaho contains the following landuse types along with a given pixel value:
Pixel value
0
1
2
3
4
5
6
7
8
9
10
11
12
13
21
22

Land-use type
background
water
city and manmade structures
vegetated field
forest (relatively flat)
grassland
sage brush
bare soil
burned area
salty soil
basalt (dark gray)
basalt (black)
basalt (gray)
basalt (light gray)
forest (mountain)
bare soil (mountain), dead grass and other small vegetation

B. Momentum Roughness Length

The land-use classification is used to assign values of zom to specific land types. The
following are examples of zom estimations given in a land-use map:

a)
b)
c)
d)
e)
f)

Water;
Cities;
Forests;
Grassland;
Desert with vegetation;
Snow;

zom = 0.0005 m
zom = 0.2 m
zom = 0.5 m
zom = 0.02 m
zom = 0.1 m
zom = 0.005 m

44

These values for zom can vary by region and their use should be carefully considered.
For agricultural areas, zom is computed as a function of crop height (h) which in turn is
defined as a function of Leaf Area Index (LAI). For crops such as alfalfa, potatoes, beans,
beets, and peas, h = 0.15LAI. Other empirical relationships between h and LAI for specific
crops are given below (based on the work of M. Tasumi, Univ. Idaho, using data from Dr.
J.L. Wright of the USDA-ARS, Kimberly, Idaho):

Alfalfa: H = 0.009LAI2 + 0.076LAI,

Corn: H = 0.03LAI3 - 0.2194LAI2 + 0.7243LAI,
Potato and Tall beans: H = 0.0043LAI3 - 0.0694LAI2 + 0.3783LAI,
Beans, Beet and Peas: H = 0.0025LAI3 - 0.0417LAI2 + 0.2754LAI,
Spring Wheat: H = 0.0414LAI3 - 0.1848LAI2 + 0.3214LAI,
Winter Wheat: H = 0.0039LAI3 - 0.0411LAI2 + 0.1747LAI,

R2 = 0.9881
R2 = 0.9938
R2 = 0.9837
R2 = 0.9756
R2 = 0.9896
R2 = 0.9914

The data from the figures above are compared against the h = 0.15 LAI function in the
following figure (from Univ. Idaho, 2002):

45

LAI vs Crop Height, Kimberly ID

(Data by Dr. J.Wright USDA/ARS)

2
1.8

1971 Alfalfa
(3seasons)

1.6

1972 Potato

Height (m)

1.4

1973 Beans

1.2

1974 Beans

1975 Beets

0.8

1976 Corn

0.6
0.4

1977 Pias

0.2

1978 Wheat

1979 Wheat

1:0.15 Line

LAI
Note that in an area dominated by corn production, a different height-LAI function should be
developed.

46

Appendix 3
Reference Evapotranspiration, ETr, Calculation

The reference evaotranspiration (ETr) is the ET rate expected from a well-defined surface
of full-cover alfalfa or clipped grass. ETr is used in SEBAL to estimate the ET at the cold
pixel and to calculate the reference ET fraction (ETrF).
ETr can be computed for a given weather station using the REF-ET software made
available by the University of Idaho:
Allen, R.G. (2000). REF-ET: Reference Evapotranspiration Calculation Software for
FAO and ASCE Standardized Equations, University of Idaho
www.kimberly.uidaho.edu/ref-et/
This software is included with this SEBAL manual and contains the ETr program and
instructions.
A. Requirements for running REF-ET:
Weather data for hourly or shorter time periods surrounding the date and time of
satellite image is required. The weather data should include air temperature, solar
radiation, wind speed, and humidity (in the form of dewpoint temperature, relative
humidity, or vapor pressure). Data for other days can be processed at the same time
and will be useful for predicting ET for adjacent time periods in SEBAL.

The data file should be in text format (i.e., ASCII), and data columns can be separated
by tabs, spaces, or commas. The data columns can have headings and can be in any
order and can use a wide range of data units. The order of the weather parameters and
their units are defined within REF-ET. The weather data file should be created before
running REF-ET.
The following figure is an example of hourly weather data with headings for use in REFET. Weather station information is defined in section B-5 of this appendix.

47

B. Steps for running REF-ET

1. Open REF-ET and click on Proceed

2. Use the file browser as shown in the above box, identify the name of the data file,

48

and click on Open

3. Provide the name for a new REF-ET definition file (i.e., ref_et.def) to be created by
REF-ET. The definition file defines the order of weather data in the data file and the
associated measurement units. If a definition file has been previously created for the
weather data set, then this definition file can be reused by REF-ET as shown in the
following dialogue box.

49

4. Indicate the data file parameter order and units by double clicking on the list of data
parameters and measurement units in the box in the upper right hand of the
following REF-ET screen. Selected parameters will appear in the list in the upper left
hand side of the following screen (green area).

Click on Continue

50

5. Describe weather station and weather data file

Enter 1 (or other value) for Initial Lines to be skipped if the first line(s) of the weather
data is/are headings.
Click on Continue

51

6. Enter output modes and reference ET equations. The following selections are
recommended:

Select SI units
Select file and screen output to create a disk file of results
Select the ETr - ASCE Penman-Monteith Standardized Form equation. This method is
recommended as it represents a tall alfalfa surface that best represents the ET from
highly vegetated pixels.
The intermediate files contain an echo of the data file and various parameters involved
in calculating reference ET.
Click on Continue
7. Save the definition file for future use

52

8. Save the output file using a name selected by the user. Results from REF-ET that
are echoed on the screen should look similar to the following screen:

Click on End

53

Appendix 4
Preparing the Original Satellite Image for Use in SEBAL

The original Landsat image needs to be geo-rectified for use in SEBAL. This can be done
locally or by sending it to a commercial entity such as the Earth Satellite Corporation. It is
recommended that the rectified image by saved in the GeoTIF format and that nearest
neighbor resampling be used during rectification to preserve spectral information.
SEBAL uses the following seven bands of the spectrum:
Band 1
Band 2
Band 3
Band 4
Band 5
Band 6
Band 7

visible (blue)
visible (green)
visible (red)
near infrared
near infrared
thermal
near infrared

These bands must be layered inside ERDAS Imagine, in order, from 1 to 7 to create an
image file for use in the SEBAL process. After that, a smaller subset image can be created
for the area of interest. The steps for the layering a Landsat 7ETM+ image are as follows
(Figure 4.1):
1. Insert the geo-rectified disk (GeoTIF format)
2. View the details
The six large files with a t in the name are the bands 1 - 5 and 7, in order.
The two large files with a k in the name are for band 6, low gain and high gain
in that order (the low gain is recommended).
The large file with a p in the name is for band 8 (PanChromatic) and is not used
in SEBAL.
3. In ERDAS, go to the Interpreter tool Utilities - Layer stack and follow these steps:
a. The input file will come from the geo-rectified CD in a geo-TIFF format.
b. Create an output file having a unique name. It is recommended that the name
include the date, path, and row of the image.
c. Add the seven layers in order beginning with band 1 (the first large t file). After
adding band 5, add band 6 (the first large k file for low gain). Then add band 7
(the last large t file).
3. A layered image is now created.
For Landsat 5TM images, layering is similar to above except that spectral bands 1 7 are
listed in order in the header file (Figure 4.2).
To create a smaller subset image follow these steps:
1. In ERDAS, go to the Data Prep tool - subset image.
2. Input the image file created above.
3. Create an output file name.

54

4. Go to the Viewer tool and view the layered image. Go to Utility - Inquire Box and fit
the box to the desired area. Click on Apply.
5. Return to Data Prep and click on From Inquire Box.
6. The subset image file is now created.
The area of interest can now be viewed with the ERDAS Viewer tool. The image is shown
in false color. ERDAS has many tools for working with the image which the user must be
familiar with. Explaining all of these tools is outside the scope of this manual.

Figure 4.1 Layering the image for Landsat 7ETM+

Band 6 (low & high)

Bands 1-5,7

55

Figure 4.2 Layering the image for Landsat 5TM

Bands 1-7
in order

56

Appendix 5
Time Issues Around Weather Data and Reference Evapotranspiration
A. Converting Greenwich Mean Time (GMT) to Local Time

Satellite image times are given in GMT and must be converted to the local time of the area
of interest. For SEBAL processing, use winter standard time (local time without applying
any daylight savings time shift):
timage (Local Time) = GMT + Correction

where timage (Local Time) is the image time using local time, and Correction is the time
correction that depends on the time zone.
This correction term is found on the internet. Table 5.1 gives samples of the correction term
for some locations:
Table 5.1. List of the correction term (From Date/Time Property of Microsoft
Windows)

For example, if the weather station is in the Mountain Time zone, the correction term is 7
hours. Therefore, if the satellite image time is 17:25 GMT, the local time of the image is
17:25 7 = 10:25 AM.
The correction can also be determined by dividing the longitude of the center of the time
zone by 15o. For example, the Mountain Time zone center is -105o. Therefore, the
Correction = -105/15 = -7 hours. The centers of Eastern, Central, Mountain and Pacific
Time zones are 75O, -90O, -105O, and 120o respectfully.

57

B. Determining the Time for Weather Data and Reference Evapotranspiration (ETr)

This section describes how the instantaneous wind speed and the reference ET at the
satellite overpass time are determined from hourly (or shorter) weather data. This
procedure assumes that one has hourly (or shorter) meteorological data from a weather
station. Refer to part 2D in The Theoretical Basis of SEBAL for the selection of weather
stations and required weather data.
B-1. Data Investigation
Any data from a weather station must be studied to determine what time interval it
represents. The table below shows weather data obtained from the following weather
station: (6/20/2000, Aberdeen Idaho weather station in the Agrimet Public Weather System,
http://mac1.pn.usbr.gov/agrimet/). ETr was computed using the REF-ET software
(Appendix 3).

DoY HrMn Year

Tmax Tmin

172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172
172

C
10.9
10.2
10.3
9.3
7.3
5.8
3.6
3.7
5.7
10.7
12.9
14.7
16.2
17.6
18.8
20.3
21.2
21.7
22.2
22.0
20.9
19.1
17.1
14.6

0
100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2100
2200
2300

0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

C
10.9
10.2
10.3
9.3
7.3
5.8
3.6
3.7
5.7
10.7
12.9
14.7
16.2
17.6
18.8
20.3
21.2
21.7
22.2
22.0
20.9
19.1
17.1
14.6

ASCE
Wind DewP stPM
ETr
W/m2 m/s
C mm/h
0 1.7 8.8 0.00
0 2.3 7.3 .01
0 4.1 3.6 .04
0 1.3 2.6 .01
0
.3 3.6 -.02
0 1.0 3.1 .02
1 1.2 2.0 .01
64 1.7 1.6 .05
156 1.9 2.6 .10
415
.4 2.8 .25
591
.7 1.2 .38
747 1.8
.3 .54
868 3.4
.9 .68
941 4.5
.5 .79
964 6.2 1.5 .87
939 4.5
.3 .87
869 4.8 -.3 .87
619 5.4
.2 .76
469 5.6 1.6 .69
402 5.7
.7 .66
226 3.9 1.4 .41
69 3.7 2.5 .12
3 3.8
.8 .10
1 3.3 2.0 .06

Rs

When using these data, one must ask two questions:

1. Was daylight savings time, DST, applied to the time of record?

58

2. What time interval does the data represent? (i.e., does the posted time represent the
end of the measurement period, the center of the measurement period, or the
beginning of the measurement period?)
In the example above, DST was applied and the data represent averaged values over the
hour ending at the reported time. For example, 11:00 AM data represents the average of
values from 10:00 AM 11:00 AM.
B-2. Instantaneous Wind Speed and ETr Calculation
This section describes how to determine the instantaneous wind speed at the image time
using the weather data. Reference ET (ETr) can be determined by the same method.

(1) Calculate the satellite overpass time in local standard time (i.e., without applying the
daylight saving time):
timage (Local Time) = GMT + Correction

(2) Determine the weather time periods to use for interpolation. Quite likely, the weather
parameter for the instantaneous image time will need to be interpolated from averages
representing two time periods. A general formula to determine which time periods to use
in the interpolation is the following:

t image ( local time) 1

+ Flag period t + Flag DST

t1 = int
t
2

t 2 = t1 + t
where;
t1
= time identifier for the first weather period in hours
t2
= time identifier for the second weather period in hours
t
= length of weather period in units consistent with t1 and t2 (i.e., hours)
timage (local time)
= the satellite image time (local time, with no adjustment for
daylight savings)
Flagperiod = a flag for the way that the weather data period is represented by its time
label. Flagperiod = 0 if the time label is the endpoint of the weather data
period; = 1 if the time label is the beginning of the weather data period; =
0.5 if the time label represents the midpoint of the weather data period.
FlagDST = a flag for use of DST in the weather data set. FlagDST = 0 if no DST is
used and FlagDST = 1 hour if DST is used in the weather data set.
(3) Estimate the instantaneous wind speed applying a simple linear interpolation
assumption.

59

For interpolation of data we assume that the averages for the periods represent
measurements occurring at the midpoints of the periods. A general interpolation formula
based on the t1 and t2 period identifiers from step (2) is:
Data image = Data t 1 +

(Data t

Data t 1
t

t image ( Local time ) t1 Flag DST + t Flag period

t
2

where;
Datat1

= the average value for the specific data parameter (i.e., wind or ETr) for
the t1 time period

Datat2

= the average value for the specific data parameter (i.e., wind or ETr) for
the t2 time period

Example:
Using the weather data from section B-1.
The Image Overpass Date&Time: 6/20/2000, 17:49 GMT
Location of weather station: In Mountain Time zone (time correction is -7)
Weather Data: Given in section B-1 (Aberdeen, Idaho; based on DST)
timage (Local Time) = 17:49 7:00 = 10:49 am

For this weather data DST was applied and the time identifier (label) for the weather data
represented the endpoint of the data period (FlagDST = 1, Flagperiod = 0). t = 1 hour.
Therefore, t1 and t2 are calculated as:

(10 + 49 / 60) 1

t1 = int
+ 0 (1) + 1 = int[11.3](1) + 1 = 12 hours
1
2

t 2 = 12 + 1 = 13 hours

Therefore, the two weather data periods used to determine the wind speed and ETr at the
image time are 1200 and 1300 hours, Mountain Standard Time.
Calculate the instantaneous wind speed at 10:49 AM (Mountain Standard Time) by
applying a linear interpolation using the reported time periods of 1200 and 1300 hours
(Mountain Daylight Savings Time, MDT).

60

The wind speed is 3.4 m/s for the 1200 period and 4.5 m/s for the 1300 period (see
weather data).
The wind speed at 10:49 AM (i.e., image time) is calculated as:

Data image = 3.4 +

(4.5 3.4) (10 + 49 / 60) 12 1 + 1(0) 1 = 3.75 m / s

1

61

Appendix 6
Tables for Surface Albedo Computations
Table 6.1. LMIN and LMAX values for Landsat 5 TM (Left: for 1984 from Markham and
Barker, 1986, Right: for 2000 calibrated by Univ. Idaho 2002)
After 15, Jan, 1984
Band W*m-2*ster-1*m-1
Number
LMIN
LMAX
1
-1.500
152.100
2
-2.800
296.800
3
-1.200
204.300
4
-1.500
206.200
5
-0.370
27.190
6
1.238
15.600
7
-0.150
14.380

For 2000
Band W*m *ster-1*m-1
Number
LMIN
LMAX
1
-1.765
178.941
2
-3.576
379.055
3
-1.502
255.695
4
-1.763
242.303
5
-0.411
30.178
6
1.238
15.600
7
-0.137
13.156
-2

Table 6.2. LMAX and LMIN values for Landsat 7 ETM+ (Landsat 7 Science User Data
Handbook Chap.11, 2002)

W*m-2*ster-1*m-1
Band
Number
1
2
3
4
5
6
7
8

Before July 1, 2000

Low Gain
High Gain
LMIN
LMAX
LMIN
LMAX
-6.20
297.50
-6.20
194.30
-6.00
303.40
-6.00
202.40
-4.50
235.50
-4.50
158.60
-4.50
235.00
-4.50
157.50
-1.00
47.70
-1.00
31.76
0.00
17.04
3.20
12.65
-0.35
16.60
-0.35
10.932
-5.00
244.00
-5.00
158.40

After July 1, 2000

Low Gain
High Gain
LMIN
LMAX
LMIN
LMAX
-6.20
293.70
-6.20
191.60
-6.40
300.90
-6.40
196.50
-5.00
234.40
-5.00
152.90
-5.10
241.10
-5.10
157.40
-1.00
47.57
-1.00
31.06
0.00
17.04
3.20
12.65
-0.35
16.54
-0.35
10.80
-4.70
243.10
-4.70
158.30

Band 8 (panchromatic band) is not used in SEBAL. The gain combination for the
particular image must be clear when selecting the LMIN and LMAX values in Table 2.
These calibration constants are best used for 1999 to 2001 images. For the future use, it is
recommended to access the Landsat 7 Science User Data Handbook to see if the latest
calibration constants are available.

62

Table 6.3. ESUN for Landsat 5 TM (Markham and Barker, 1986), and for Landsat 7 ETM+
(Landsat 7 Science User Data Handbook Chap.11, 2002), both are in W/m2/m
Landsat 5
Landsat 7

Band 1
1957
1969

Band 2
1829
1840

Band 3
1557
1551

Band 4
1047
1044

Band 5
219.3
225.7

Band 6
-

Band 7
74.52
82.07

Band 6
-

Band 7
0.011
0.012

Note: a dummy value of 1 is entered for band 6.

Table 6.4. Weighting coefficients,

Landsat 5
Landsat 7

Band 1
0.293
0.293

Band 2
0.274
0.274

Band 3
0.233
0.231

Band 4
0.157
0.156

Band 5
0.033
0.034

Note: a dummy value of 0 is entered for band 6.

Table 6.5. Number of the day of the year.
Number of the day in the year (J)
Day

January

February

March*

April*

May*

June*

1
2
3
4
5
6
7
8
9
10

1
2
3
4
5
6
7
8
9
10

32
33
34
35
36
37
38
39
40
41

60
61
62
63
64
65
66
67
68
69

91
92
93
94
95
96
97
98
99
100

121
122
123
124
125
126
127
128
129
130

152
153
154
155
156
157
158
159
160
161

11
12
13
14
15
16
17
18
19
20

11
12
13
14
15
16
17
18
19
20

42
43
44
45
46
47
48
49
50
51

70
71
72
73
74
75
76
77
78
79

101
102
103
104
105
106
107
108
109
110

131
132
133
134
135
136
137
138
139
140

162
163
164
165
166
167
168
169
170
171

52
53
54
55
56
57
58
59
(60)
-

80
81
82
83
84
85
86
87
88
89
90

111
112
113
114
115
116
117
118
119
120
-

141
142
143
144
145
146
147
148
149
150
151

172
173
174
175
176
177
178
179
180
181
-

21
21
22
22
23
23
24
24
25
25
26
26
27
27
28
28
29
29
30
30
31
31
*add 1 if leap year

63

TABLE 6.5 (continued)

Number of the day in the year (J)
Day

July*

August*

September*

October*

November*

December*

1
2
3
4
5
6
7
8
9
10

182
183
184
185
186
187
188
189
190
191

213
214
215
216
217
218
219
220
221
222

244
245
246
247
248
249
250
251
252
253

274
275
276
277
278
279
280
281
282
283

305
306
307
308
309
310
311
312
313
314

335
336
337
338
339
340
341
342
343
344

11
12
13
14
15
16
17
18
19
20

192
193
194
195
196
197
198
199
200
201

223
224
225
226
227
228
229
230
231
232

254
255
256
257
258
259
260
261
262
263

284
285
286
287
288
289
290
291
292
293

315
316
317
318
319
320
321
322
323
324

345
346
347
348
349
350
351
352
353
354

233
234
235
236
237
238
239
240
241
242
243

264
265
266
267
268
269
270
271
272
273
-

294
295
296
297
298
299
300
301
302
303
304

325
326
327
328
329
330
331
332
333
334
-

355
356
357
358
359
360
361
362
363
364
365

21
202
22
203
23
204
24
205
25
206
26
207
27
208
28
209
29
210
30
211
31
212
*add 1 if leap year

The number of the day in the year, J, between 1 (January 1) and 365 or 366 (December 31) can be calculated
as follows:

3
M Mod (Y ,4 )
J = DM 32 + INT 275 + 2 INT

+ 0.975
+ INT
9
4

M + 1
100
where; DM = the day of the month (1-31)
M = the number of the month (1-12)
Y = the number of the year (1996 or 96)
The INT function finds the integer number of the argument in parenthesis by rounding downward.
The Mod(Y,4) function finds the modulus (remainder) of the quotient Y/4.

64

Table 6.6. Typical albedo values.

Fresh snow
0.80 0.85
Old snow and ice
0.30 0.70
Black soil
0.08 0.14
Clay
0.16 0.23
White-yellow sand
0.34 0.40
Gray-white sand
0.18 0.23
Grass or pasture
0.15 0.25
Corn field
0.14 0.22
Rice field
0.17 0.22
Coniferous forest
0.10 0.15
Deciduous forest
0.15 0.20
Water
0.025 0.348 (depending on solar elevation angle)
(Horiguchi, Ikuo (Ed.) 1992. Agricultural Meteorology, Buneidou, Tokyo, Japan)

65

Appendix 7
Methods for Selecting Cold and Hot Pixels

The selection of the two anchor pixels requires knowledge and skill. Below is a description
of the steps for selecting these pixels, followed by an illustrative example.
A. Cold Pixel Selection
Philosophy of the Cold Pixel. The cold pixel is used in SEBAL to define the amount of
ET (or more directly, the amount of sensible heat, H) occurring from the most well-watered
and fully vegetated areas of the image. It is presumed that these areas represent instances
where the maximum amount of available energy is being consumed by evaporation. The
impact of the near maximum evaporation rates is to cool the surface below that of areas
having less ET.

In the traditional application of SEBAL by Bastiaanssen, the cold pixel is generally selected
from a water body and it is assumed that ET = Rn G (i.e., that all available energy is
used to evaporate water, so that H = 0 for the cold pixel. In the Idaho extension to SEBAL,
we have elected to presume that ET at the cold pixel is closely predicted by the ET rate
from a large expanse of alfalfa vegetation. We therefore assume that ET = 1.05 ETr at the
cold pixel, where ETr is the rate of ET from the alfalfa reference, defined using the ASCE
Standardized Penman-Monteith equation for alfalfa. The ETr is calculated using weather
data measured at a weather station within or very close to the image. The use of 1.05 ETr
to represent the ET from the cold pixel helps to incorporate effects of advection of heat
energy from surrounding desert areas and the impact of wind speed on ET. The 1.05 factor
is applied because it is likely, in a large image, that some fields may have a wet soil surface
beneath the vegetation canopy that might increase the total ET rate about 5% above that of
the ETr standard. H for the cold pixel is then calculated as H = Rn G 1.05 ETr.
The cold pixel should therefore be selected to represent an agricultural area that has
vegetation at full cover and that is well irrigated. Full cover means that the LAI is greater
than about 3.
Instructions
The following steps necessary for the selection of the cold pixel:
1. Open the image or subset image and view it in true color (layer combination of 3,2,1)
and in false color (layer combination of 4,3,2).
2. Open the Ts image in a new viewer with the Pseudo Color option. Link the two
images.
3. On the Ts image use Raster Attributes to observe the range of Ts values.

66

4. Make a first guess for Tcold by looking at some fully covered agricultural fields (dark
green). Insert a color gradation around the selected temperature.
5. View the colored Ts image and select a point that represents a cold (wet)
agricultural field. Do not select an extreme cold point but one that is representative
of a well-watered full cover crop. For the best results, the cold pixel should have a
surface albedo in the range of 0.22 to 0.24 (corresponding to a full covered
reference alfalfa field). It should have a Leaf Area Index in the range of 4 to 6
(corresponding to full covered agricultural field). Finally, the cold pixel should be
located nearby the weather station so that the weather conditions are similar.
6. Observe other areas of the image for similar temperatures.
7. Select Tcold from a pixel in the center of the chosen field that represents a very cold,
but not extremely cold, point in the image.
8. Record the coordinates and temperature of the cold pixel.

Illustrative Example (Landsat 7 ETM+, 4/8/2000)

The following example illustrates some of the thought processes employed during selection
of a cold pixel.
In the following example, many well-watered agricultural fields, during an initial viewing, are
found to have surface temperatures of around 290 K. This can serve as an initial guess for
the cold pixel temperature.
Next, we create a graduated color map for Ts over a range of 284-295 K (Figure 1) that
includes the 290 K temperature. In this picture, the blue colored pixels are near 284 K and
the red pixels are near 295 K. The image in Fig. 1 is an early April image, and not many
fields are well-watered for this period in this region. One can see that most of the
agricultural fields are still bare fields and cultivation has not been started. In the colored Ts
image, bluish fields are more likely to be well-watered fields having some vegetation cover.

67

Fig.1. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image(left, 284K:Blue to

295K:Red) and the true color image (right).The coordinates of the inquire cursor are x/y =
436892/194372)

First, lets focus on the northwest region. There are many cold pixel candidate fields in this
region. Some of the candidate fields are circled in Figure 2.

Fig.2. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (left, 284K:Blue to

295K:Red) and the false color image (right).The coordinates are x/y = 435805/188719)

One of the candidate fields is displayed (marked by the cross-hairs) at the center of the
next image (Figure 3). The temperature is quite homogeneous in the field, around 289.5 K.
This is a good candidate field because the false color image indicates a high intensity of
vegetation. In addition, there are some fields in other areas having similar temperatures.
Based on the local area (south-central Idaho) and time of this image, this field is probably
full cover, well-watered, winter wheat.

68

Fig.3. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (left, 284K:Blue to

295K:Red) and the false color image (right). The coordinates of the inquire cursor are x/y =
439452/194132)

There is another center pivot irrigated field in Fig.3 to the left of the candidate field. The
false color image indicates that the intensity of the vegetation is not very strong. It is
interesting that the Ts image shows some variation of temperature in the field, and it looks
as if the image has captured the impact of wet vegetation near the center pivot lateral.
It is expected that the temperature be relatively homogeneous within a dense wheat field
irrigated by center-pivot, and that the impact of the wet soil/vegetation be more apparent in
a field having less than full-cover. The surface temperature of the second field near the
center pivot lateral is 290 K, which is slightly higher than for the dense wheat field. This is
understandable, because the dense, well-watered wheat field should have a higher ET than
a sparse, well-watered field due to higher transpiration rates. The higher ET has decreased
the surface temperature by 0.5 K.
Next, we will look at a different region within the area of interest. In the center region, it
appears that there are not as many well-watered agricultural fields (Figure 4).

Fig.4. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (left, 284K:Blue to

295K:Red) and the false color image (right). The coordinates are x/y = 488062/151266 )

69

The circled field in Fig.4 (486446/147416) appears to be the only good candidate field in
this area. The temperature is 289.5 K.
We are evaluating an April image. If this were a July image, we would not select this field
as a candidate because there would be many well-watered fields at this time and this field,
with a lower temperature than surrounding fields, would probably represent an extreme
condition.
Since this is an early April image, we do not expect many of the fields to be in full cover and
this field may be a good candidate for the cold pixel.
The temperatures of the two cold pixel candidates from two different parts of the image
are the same. This is good confirmation that we have located an expected temperature
condition representing typical full cover vegetation that is well-watered. It is common for
similar temperatures to occur in well-watered, well-developed agricultural fields. However, it
is not necessary that the cold pixel candidates in different areas have similar
temperatures. The temperatures of cold pixel candidates can vary by region due to
regional climate differences within an image, vegetation development, and soil type.
However, if the regional weather is known to vary significantly within an image, the SEBAL
user should carefully investigate how to best select the cold pixel, because one basic
assumption in SEBAL is that climate conditions are uniform within the area of interest.
Where weather varies, one may consider splitting the image into two or more parts and
apply SEBAL to each part separately, with unique cold and hot pixels for each part.
In the northeast region of our example image, there are some cold bare fields (the blue
circled fields in Figure 5). The false color image shows that many center-pivot irrigated bare
fields in this area were just starting to be irrigated. One could select such a wet bare field;
however, it is best if the cold pixel is within a well-watered field having a full cover crop.
This is because we relate ET for the cold pixel to ETr representing ET from a full cover
alfalfa crop. There is only one field in this area that meets this criterion in Figure 5 (circled
green).

70

Fig.5. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (left, 284K:Blue to

295K:Red) and the false color image (right). The coordinates of the inquire cursor are x/y =
538010/186916)

The surface temperature in this field is 291.1 K, which is 1.5o higher than the two other
candidate fields selected earlier. If this image was for July or early August rather than early
April, more fields would be well-watered. In this case, we would not select this field
because this may represent an extremely cold condition (the coldest in the region). In our
April situation, however, the false color image tells us that this field is one of the few fully
vegetated fields in the area and that it may be a good candidate for a cold pixel.
Finally, we chose the candidate field from Figure 4 with Tcold = 289.5 K for our cold pixel.
This field is close to the weather station (20km) and should experience similar weather
conditions. If the cold pixel is far from the weather station, it is possible that the field is
cold because of varying weather conditions.
After selecting the cold pixel, we record Tcold and its coordinates for future reference.
B. Hot Pixel Selection
Philosophy of the Hot Pixel. The magic of SEBAL is in its prediction of sensible heat
(H) for each pixel of an image. H is needed to solve for ET at each pixel using the energy
balance equation ET = Rn G H. Values for H are distributed across an image in
SEBAL according to the surface temperature (Ts). This is done using a dT vs. Ts function
where dT is the difference between the air temperature very near the surface (at 0.1 m
above the zero plane displacement height, d) and the air temperature at 2 m above the
zero plane displacement height. SEBAL presumes a linear change in dT with Ts. The linear
equation for dT vs. Ts is developed by using the dT values for the cold and hot pixels. The
dT at the hot pixel is determined from H, as described in the main text, by assuming that
H = Rn G. Therefore, the hot pixel should be one where there is no ET.

71

Instructions

The following steps are necessary for the selection of the hot pixel:
1. The selection of the hot pixel follows the same procedure as for the cold pixel. It
is more difficult, however, because there is a broader range of temperatures for hot
pixel candidates to select from.
2. The hot pixel should be located in a dry and bare agricultural field where one can
assume there is no evapotranspiration taking place. It is recommended that one not
use a hot desert area, an asphalt parking lot, a roof, a south facing mountain slope,
or other such extremely hot areas, since the relationship between dT and Ts may not
follow the same linear relation as for agricultural and bare agricultural soils. In
addition, the prediction of soil heat flux, G, is less certain for man-made structures
and desert. Desert is difficult because the thin canopy creates a two-source system
for heat transfer (canopy and underlying soil).
3. The hot pixel should have a surface albedo similar to other dry and bare fields in
the area of interest. It should have a LAI in the range of 0 to 0.4 (corresponding to
little or no vegetation). In the Mountain Model, the hot pixel should have a slope of
less than one percent. As with the cold pixel, it should be located near the weather
station.
4. If there has been precipitation within the past three or four days prior to the image,
then it is possible that the hot pixel may exhibit some ET. In this case, it cannot be
assumed that the H = Rn G at the hot pixel. Instead, H must be predicted as Rn
G ETbare soil, where ETbare soil is predicted by operating a daily soil water balance
model of the surface soil using ground-based weather measurements. An example
model is included in Annex 8 of FAO-56 available at:
http://www.fao.org/docrep/X0490E/X0490E00.htm
A downloadable spreadsheet from Annex 8 is available from:
http://www.kimberly.uidaho.edu/water/fao56/index.html
Generally, the ET from a bare soil can be roughly estimated to be equal to 0.8 ETr,
0.5 ETr, 0.3 ETr, 0.2 ETr, and 0.1 ETr, 1, 2, 3, 4, and 5 days following a substantial
rain event (15 mm or more).
Illustrative Example (Landsat 7 ETM+, 4/8/2000)

In the following image, a simple daily water balance model predicted that bare soil in this
area is totally dry so that we can assume zero ET for the hot pixel.

72

We open the image or subset image and select true and false color. We open the Ts image
using Raster Options-Pseudo Color and link the two viewers. We go to Raster Attributes
in the Ts viewer to view the attribute table shown below:

The hottest pixel in the image is around 316 K. However, the purpose is not to select an
extreme hot pixel but to select a dry pixel where ET is zero.
The problem is that the surface temperatures of totally dry surfaces are not necessarily
uniform. For example, Ts of a dry black asphalt road might be extremely high while Ts of a
dry desert might be lower but still quite high. Ts of a very dry agricultural field might be
moderately high. ET from these three different surfaces could all be zero or very close to
zero.
The selection of an extremely high temperature pixel for the hot pixel could cause SEBAL
to predict positive ET for other dry pixels. For example, if the cold pixel is in a field
having Ts = 18o C and our hot pixel is selected in the middle of a desert with Ts = 43o C,
SEBAL would regard any totally dry and fallow agricultural field having Ts = 30o C as a field
with some moisture. This is because the dT for this dry agricultural field would be predicted
to be less than that for the hot pixel based on our dT vs. Ts function, and therefore H would
be under predicted. Therefore, it is best to limit the land surface condition for candidate hot
pixels to agricultural fields.
Our first step in selecting a hot pixel is to find a representative, totally dry agricultural field
in the area of interest. It is important to check precipitation records in order to get an idea of
how wet an area is. One can apply a water balance model to predict the residual moisture
in non-irrigated bare soil. For our example, many fallow fields are evident in the image and

73

weather records indicate that there was no rainfall prior to the image time. We can then
assume that there should be many dry and hot agricultural fields in the area.
We do not want to look at the extremely hot pixels so we ignore anything higher than 311
K (from the attribute table). These values represent mostly desert areas. Next, we will
colorize the Ts image using a blue-red gradation (blue for 302 K and red for 310 K):

Fig.6. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (302K:Blue to 310K:Red).

It appears that our first guess for a temperature range that includes the hot pixel (302-310
K) is too high, because the red color appears only in the desert areas. There is some bluish
color in the agricultural areas, so we will change our range to 300-306 K. However, we are
not yet certain that the hot pixel is below 306 K so we will assign a light blue color to 307 K,
yellow to 308.5 K and pink to 310 K in addition to the blue-red gradation for 300-306 K
(Figure 7).

Fig.7. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (300K:Blue to 306K:Red,

307K:light blue, 308.5K:yellow, and 310K: pink).

74

Now, we look at the northwest area (Figure 8).

Fig.8. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (300K:Blue to 306K:Red,

307K:light blue, 308.5K:yellow, and 310K: pink). The coordinates of the cursor are x/y =
440393/175789.

In the area shown in Fig.8, Ts of the yellow-circled field is around 302.7 K. Actually, the
values for Ts in this specific field vary from 301.7-303.2 K, but we select 302.7 K as the
representative value.
`
We assume that some fields in this area are very dry. The green-circled field in Figure 8
appears to be a fallow field with a lower temperature of 301.5 K. We must decide if this field
is very dry or if it has some moisture and ET.
The temperature of the pink-circled field in Fig.8 is 301.6 K, which is very similar to the
green-circled field (301.5 K). However, this field contains a little vegetation since it is
slightly reddish in the false color image so we assume that the pink colored field is relatively
dry but probably has some moisture and ET.
There are two possible explanations for why the pink-circled field has a lower temperature
than the yellow-circled field:
(1) The green-circled field may have some surface moisture since it has a lower
temperature than the yellow-circled field (which may be very dry).
(2) The green-circled field may be totally dry and bare, but has a lower surface temperature
than the yellow-circled field because it is surrounded by more humid fields than the other
field.

75

The pink-circled field is darker than other fields in the area (see Figure 9). It is possible that
this field is not a bare field but is partially covered with dark desert-type vegetation.

Fig.9. P40/R30, Landsat 7 ETM+ 4/8/2000, true color image, x/y = 442873/179798.

We decide that the best candidate for the hot pixel in the northwest region is the yellowcircled field (302.7 K).
Next, we will look at the southeast region (Figure 10).

Fig.10. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (300K:Blue to 306K:Red,

307K:light blue, 308.5K:yellow, and 310K: pink). The coordinates of the cursor are x/y =
518565/159301. (Note that the cross-hairs point to the eastern part of the city of Burley.
Cities should be avoided during anchor pixel selection).

It appears that this region is hotter than the northwest region. The light blue, yellow and
pink areas are where the temperature is 307-310 K. which is much hotter than our previous
hot pixel candidate (302.7 K) selected from the other area. This is explained as follows:
1) The regional weather or soil conditions are different between the northwest and southeast regions of this image. If so, we might think about separating the image into two areas,
since SEBAL assumes that the regional weather is relatively uniform within the area of
interest.
76

(2) The regional weather or soil conditions are the same but there is another reason why
the surface temperature conditions are so different between two areas of this image.
We must now decide how to proceed.
First, we doubt that the northwest region is colder because of rainfall but we are not
completely sure of this since the precipitation data is for the center of the area where the
weather station is located. We therefore investigate Ts in the center of the image where the
weather station is located and where we know that there has been no observed
precipitation (Figure 11).

Fig.11. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (300K:Blue to 306K:Red,

307K:light blue, 308.5K:yellow, and 310K: pink). Center of the plain (the coordinates are x/y
= 466958/160477)

The temperature range for the center of the image is very similar to northwest region which
indicates that our hot pixel candidate from the northwest region (302.7 K) might be a good
selection and representative of bare soils having ET = 0.
Now, we go back to the southeast region (Figure 12). If we look at the region from the
center to the east, it is clear that surface temperatures increase toward the east. Since the
east tends to contain more sandy soil, this temperature increase might be due to soil type.

77

Fig.12. P40/R30, Landsat 7 ETM+ 4/8/2000, Colored Ts image (300K:Blue to 306K:Red,

307K:light blue, 308.5K:yellow, and 310K: pink). Center to the east of the plain.

When we scan the entire area of interest, we find that the surface temperature in the
southeast region is higher. A type of heat island can be recognized in the following Ts
image (Figure 13).

Fig.13. P40/R30, Landsat 7 ETM+ 4/8/2000, Ts image (Black and white, entire snake river
plain)

78

We conclude that this heat island is a local phenomenon, occurring during the early spring
image, and that our hot pixel of 302.7 K from a more typical part of the image is valid. This
value for Thot is probably not a good value to apply for the southeast region. It would be
best to separate this region into a separate SEBAL application. In our application, we will
include this region, but we will ignore the ET estimates for this region since it is outside our
area of interest.
After selecting a hot pixel, Thot should be recorded along with its coordinates.

79

Appendix 8
Iteration Spreadsheet Used for the Atmospheric Stability Correction

The SEBAL process described in this manual includes a spreadsheet named

Calculations_for_05_H_Automatic_GMD_model.xls that is used to facilitate the iteration
process used in SEBAL for the correction of the aerodynamic resistance to heat transport
(rah) due to buoyancy effects in the atmosphere. In SEBAL, first, initial values for the friction
velocity (u*) and rah are computed assuming neutral atmospheric conditions. The sensible
heat flux (H) is computed next. Then, using the Monin-Obukhov theory, u* and rah are
computed again using the stability correction. The corrected rah is then used to compute a
new dT function and a new value for H. This iteration is repeated until dT and rah at the
hot pixel stabilize.
In the SEBAL process and ERDAS Imagine Model Maker, the iterative process can be
done manually, i.e., an updated Imagine model (for example, F13_h_1.gmd) is operated
during each iteration. Alternatively, the iterative process for H can be done automatically
for all iterations using the F13toF24_h_automatic.gmd model. In both cases, the
spreadsheet is useful for calculating the correlation coefficients b and a of the dT function
that is used for the calculation of H inside SEBAL.
The spreadsheet calculates values for Hcold, dTcold, Hhot and dThot. The correlation
coefficients b and a for the dT function are computed and a corrected value for Ta and are
then calculated. As the iteration continues with stability corrections, new values for dTcold
and dThot are calculated along with new coefficients, b and a.
On the following page is an example of the iterative spreadsheet followed by instructions
for the manual operation and the automatic operation of the process.

80

Part 1: using the manual iteration process

1. Go to the spreadsheet.
2. Enter values in the colored, marked boxes for the following parameters:
a. The blending height for high level wind speed (generally 200 m)
b. z1 in m (generally 0.1 m)
c. z2 in m (generally 2 m)
d. ETr at the time of the image in mm/hour
3. For the cold and hot pixels, enter the following information in the colored, marked
boxes:
a. The x and y coordinates of both pixels
b. Elevation in meters of both pixels (elevation of the weather station for the Flat
Model)
c. The expected ETrF (i.e,. fraction of ETr ) at both pixels (this is generally 1.05 for
the cold pixel and 0 for the hot pixel)
d. Surface temperatures for both pixels (Ts) in Kelvin
e. Elevation corrected surface temperatures (Ts_dem) for both pixels. (For the Flat
81

Model, Tsis used)

f. Net Radiation (Rn) for both pixels in W/m2 at the time of the image (computed by
SEBAL)
g. Soil Heat Flux (G) for both pixels in W/m2 at the time of the image (computed by
SEBAL)
h. Momentum roughness length (zom) for both pixels
i. Wind speed at the blending layer height (u200m) in m/s. This value is generally the
same for both pixels.
4. Following the entry of the data in step 3, the spreadsheet will automatically calculate
the b and a coefficients that are used in the dT = a Ts + b equation. These
coefficients are entered into an updated F13_h_1.gmd model, one set at a time
(i.e., per iteration).
a) Load model F13_h_1.gmd.
b) Specify the file names for Ts and initial rah, files along with the elevation and enter the
first set of coefficients b and a.
c) Save the output files for h_1, airdensity_1, and the model.
5. Corrected values of u* and rah are computed using the Model Maker:
a) Load model F14__u_star_2_rah_2.gmd.
b) Specify file names for h_1, airdensity_1, initial u*, Ts, zom, and the value for u200.
c) Save output files for u_star_2 and rah_2.
6. Modify model F13_h_1.gmd to use new file names and enter the second (or third,
or fourth, so on) set of values for b and a.:
a) Enter rah_2, and new coefficients b and a.
b) Save output as h_2 and airdensity_2
c) Save model as; model F15_h_2.gmd
7. Corrected values of u* and rah are computed (similar to step 5).
a) Use model F14 from step 5
b) Enter file names for h_2, airdensity_2, and u_star_2
c) Save output as u_star_3 and rah_3
d) Save model as; model F16__u_star_3_rah_3.gmd.
8. Repeat steps 6 and 7, etc. using corrected values.
9. Iteration continues until dThot and rah for the hot pixel stabilize (usually 3 to 7
iterations).
10. Finish with a final run of the model for H. Save the output files for h and airdensity
along with the model.
.

82

Part 2: Using the automatic iteration process

The model F13 to 24_h_automatic.gmd is used with the support spreadsheet,

Calculations_for_05_H_Automatic_GMD_model.xls. This model should be
operated after F12_initial_u_star_rah.gmd. Since this one large model gives the
final H value in W/m2, one can skip all the iteration models. Therefore, after
running this model, only the final model F25_et24.gmd need be run.
The following are the steps necessary to setup and run this model:
1. Select the cold and hot pixels
2. Open the support spreadsheet and input the necessary values
Enter values in the colored boxes for the following parameters:
a) The blending height for high level wind speed (generally 200 m)
b) z1 in m (generally 0.1 m)
c) z2 in m (generally 2 m)
d) ETr at the time of the image in mm/hour
For the cold and hot pixels, enter the following information in the colored boxes:
a) The x and y coordinates of both pixels
b) Elevation in meters of both pixels (elevation of the weather station for the Flat
Model)
c) The expected ETrF (i.e,. fraction of ETr ) at both pixels (this is generally 1.05 for
the cold pixel and 0 for the hot pixel)
d) Surface temperatures for both pixels in Kelvin (read from the Ts image)
e) Elevation corrected surface temperatures for both pixels. (read from Ts_dem
image; for the Flat Model, use Ts values)
f) Net Radiation (Rn) for both pixels in W/m2 at the time of the image (read from Rn
image)
g. Soil Heat Flux (G) for both pixels in W/m2 at the time of the image (read from Rn
image )
h. Momentum roughness length (zom) for both pixels (read from zom image)
i. Wind speed at the blending layer height (u200m) in m/s. (calculated in the support
spreadsheet SEBAL Model constants for flat model.xls)
Once the above values are input, the correlation coefficients b and a for 7 iteration steps
are computed. One must make sure that the rah and dT values of the cold and hot
pixels are numerically stable (i.e., they do not diverge with the iteration time). This can
be checked by looking at the red-circled area in the following snapshot:

83

The blue-circled cell tells you how many iterations are recommended.
The F13to24_h_automatic.gmd model automatically runs 7 iterations with the 7 sets of
coefficients b and a.
As long as the rah and dT values for cold and hot pixels converge, the number of
iterations is not important. If these parameters diverge, we recommend increasing the
wind speed at 200m to 4 m/s, regardless what was computed using the actual weather
data
3. Setup and Run the F13to24_h_automatic.gmd model
Complete the following steps to run the model:
a) Load model F13to24_h_automatic.gmd.
b) Define the following input values and files:
Surface temperature image
Representative elevation value
Representative Elevation value
rah_1 image
zom image
calculated wind speed (m/s) at 200m height
c) Enter all seven sets of coefficients b and a into the appropriate tables. (Note that the
first entry in the tables is numbered as 0).
d) Save the output files for h and airdensity along with the model.

84

Appendix 9
SAVI and LAI for Southern Idaho
Note: Equations (14) and (15) for SAVI and the LAI are not well calibrated for universal
use with all vegetation types, and some uncertainties remain on their use. They are,
however, utilized in SEBAL for two reasons: (1) LAI might be a better index than NDVI,
since it has a physical meaning, and (2) LAI might be a better index for estimating the soil
heat flux (G) and the momentum roughness length (zom).

Some of the development or validation of these equations in Idaho is shown below:

Change of SAVI number with various L values
for 5 soil / dead grass conditions in Southern Idaho, 1989

0.15
0.14

0.009

0.13
0.008

SAVI

0.12
0.11

0.007

0.1
0.006

0.09
0.08

0.005

0.07

Standard Deviation

0.01

0.004

0.06

#1
#2
#3
#4
#5

0.05

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.003

STDEV

SAVI vs. L, where L is the coefficient in the SAVI equation for five soil or dead grass conditions in
southern Idaho; #1: Dry bare field with bright color (4/18/89), #2: Wet bare field with dark color
(5/4/89), #3 to #5: Totally dead grassland from desert area with different soil color. The standard
deviation (STDEV) between the five soil locations or conditions becomes minimum when L = 0.1.
This indicates that L = 0.1 tends to reduce the impact of soil wetness on the vegetation index. The
value of L for SAVIID was chosen as 0.1 due to this analysis.

85

A value of L for a particular area can be derived from an analysis of multiple images for the area.
Using three or more images, look at some pixels of vegetation that is not likely to change much (for
example; sage brush during mid to late summer, conifer trees, etc.). Compute SAVI using a range of
values for L and plot SAVI vs. image date. Select the L that provides for a consistent value of SAVI
over time.
Compareison of calculated and measured LAIs
1989 Sugar Beets
6
5

LAI

4
3
2
Estimated by SEBAL

Measured LAI (Beet)

12/8/89

10/19/89

8/30/89

7/11/89

5/22/89

4/2/89

Appendix 10

G/Rn Computations for Water Surfaces and Snow

SEBAL uses an empirical function (equation 24) to compute G/Rn based on

surface temperature, surface albedo, and the NDVI. For surfaces such as water,
snow, and wetlands other methods for defining G/Rn are needed.
A. G/Rn for Water Surfaces
The heat balance at water surfaces works very differently than at land surfaces. Generally,
albedo values for clear water are very small (less than .05) when the sun angle is more
than 10o above the horizon. Shortwave radiation penetrates into a water body according to
the transparency of the water and is absorbed at a range of depths below the surface
where it is converted into heat (G). For oceans and rivers, a large amount of energy is
transferred horizontally with the currents, and the vertical heat balance equation is no
longer consistent. Hence, the equation for G/Rn for a water surface is difficult to define. The
function for G/Rn that is most suited for a specific water surface must be evaluated.

For the inland states such as Idaho, the primary water surfaces are lakes where the depth
of solar radiation penetration will vary with the sun angle and the turbidity of the water body.
In the 2000 Idaho project, a large, deep lake (Bear Lake) was the primary water surface in
the study area. Therefore, a deep lake condition was focused on during development of the

86

equation for G. The following are examples that may provide ideas on how to compute G
for lakes:
1. G equation for deep and clear lakes
Yamamoto and Kondo reported the seasonal trends in the heat balance for a deep lake
from a set of actual measurements (Yamamoto and Kondo, 1968). Figure 10.1 shows a
graph of the monthly averages for Rn and G values that they measured in a deep lake
(mean depth was 21m) in Japan. This graph indicates how the ratio G/Rn varies with the
time of year. In the spring and early summer it is large since the water is cold and the air is
warm. In the fall it is small since the water is warm and the air is cool. One can note also
that for a deep, clear body, the net daily value for G can be strongly positive or negative
due to change in heat storage, whereas, it is generally assumed for vegetation that the net
G for a 24-hour period is nearly zero (i.e., no net change in storage for the day).

Figure 10.1 The plot of the monthly averages of Rn and G values measured in a deep lake (mean
depth of 21m) in Japan (Yamamoto and Kondo, 1968).

In our 2000 project (Tasumi et al., 2000), we separated the year into two periods; JanuaryJune and July-December. We derived a Rn-G function for each period using Yamamotos
measurements. Figure 10.2 is for January-June. One extreme data point in January was
rejected. Figure 10.3 is for July-December. One extreme data point in July was rejected.

87

y = 0.8694x - 35.286
R2 = 0.8174

G (W/m2)

150
100
50
0
0
-50

100

150

200

Rn (W/m2)

y = 1.0656x - 114.28

50

G (W/m2)

50

Figure 10.2. A Rn-G curve and

measurements for estimating G for water
during the January to June period (based on
Yamamoto and Kondo, 1968).

R2 = 0.9472

0
0

50

100

-50
-100
-150

150

Figure 10.3. A Rn-G curve and

measurements for estimating G for water
during the July to December period (based on
Yamamoto and Kondo, 1968)

Rn (W/m2)

This data represents the monthly averages and G may be larger if the day time is focused
on.
In addition to Yamamotos work, Amayreh (1995) measured G for Bear Lake, Idaho/Utah
using eddy covariance and heat storage change and reported that daily values of G during
the summer and fall could be predicted using: G = 0.984Rn 62 . This equation does not
take into account the seasonal difference in G since Amayrehs work was limited to
summer and fall. However, Amayrehs measurements are similar to those of Kondos in
both magnitude and trends and this simple prediction approach based on Rn is therefore
assumed to be valid for deep clear water bodies in general. The following equations were
developed for estimating G for water surfaces based on Yamamotos and Amayrehs work:
From July to December, the instantaneous G during midday (and at the satellite image
time):
Gwater = Rn 90

(10.1)

Gwater = Rn 100

(10.2)

and for 24-hours:

From January to June, the instantaneous G during midday (and at the satellite image time):
Gwater = 0.9 Rn 40

88

(10.3)

and for 24-hours:

Gwater = 0.9 Rn 50

(10.4)

The penetration of solar radiation into water decreases as the depth of the water body
decreases and/or as the turbidity of the water increases. Highly turbid bodies of water and
very shallow bodies of water will have a value of G that is close to the value of Rn and the
offsets of the above equations will be reduced or may completely disappear.
2. G equation for shallow (less than 2m) or turbid lakes
The value of G for shallow lakes is estimated to be between the value for deep lakes and
for wetlands. This results in the following equation developed from Table 10.1:
for instantaneous at midday:
Gwater = 0.50Rn
and for 24-hours:
Gwater = 0

89

(10.5)

Table 10.1. Calculated G/Rn ratio for deep lake and wetland for various Rn ranges. For
deep lakes, G/Rn equation for Bear Lake was used. For Wetlands, Burba et al (1999:
G = -51 + 0.41Rn) was used.
Rn
200
250
300
350
400
450
500
550
600

G/Rn
Deep Lake Wetland
0.63
0.16
0.69
0.21
0.73
0.24
0.76
0.26
0.79
0.28
0.81
0.30
0.82
0.31
0.83
0.32
0.84
0.33

B. G/Rn for Snow Surfaces

In the application of SEBAL to Idaho, the value of G for snow surface was predicted for
daytime periods as:

Gsnow = 0.5 Rn

(10.6)

This is a very crude estimate for G and assumes that one half of the net radiation
incident on the snow surface penetrates in the form of light and is absorbed into the snow
mass as G. For 24 hours, the average G was assumed to be zero. Additional research and
literature review is needed in this area.

Estimated LAI values from eight TM images using equation (15) compared with actual LAI values
measured by Dr. J.L.Wright, USDA/ARS, in a sugar beet field at Kimberly, Idaho 1989.
Equation (15), when used with SAVIL=0.1, closely estimates the measured LAI of sugar
beets at Kimberly, especially during the crop-growing season. The Idaho application has
shown that the LAI equation (15) is quite sensitive to SAVI after LAI > 3. This confirms the
known phenomenon that the NDVI (which is similar to SAVI) saturates, i.e., it does not
increase, once the LAI reaches a certain amount (Kondoh, 1999). The sensitivity problem
of LAI values greater than 3 does not affect the G or H estimation. The accuracy of
equation (15) has been found to be extremely sensitive to the value for L used to compute
SAVI. Field verification is needed for new areas or for new vegetation.

90

Appendix 11
The Stability of the Atmosphere
Air has three stability conditions; unstable, neutral, and stable. Stability
conditions must be considered during the computation of sensible heat flux
(H) because they affect the aerodynamic resistance to heat transport (rah). In
SEBAL, a neutral atmospheric condition is initially assumed and a stability
correction is later applied using the Monin-Obukhov length (L) as the indicator.
The three stability conditions for air are illustrated in Figure 11.1. If a dry air
mass at 10o C is suddenly moved upward 100 meters, its temperature will
decrease about 1o C due to the lapse effect. If the surrounding air temperature
is lower than this air mass, an upward force on the air mass will occur. This
condition is called unstable. If the surrounding air temperature is the same
as the air mass, no force will occur and the condition is called neutral. If the
surrounding air temperature is higher than the air mass, a downward force on
the air mass will occur. This condition is called stable.
Generally, air temperature decreases by about 0.65oC when elevation
increases by 100 m under neutral stability conditions. In places where positive
sensible heat flux (H) occurs, the change of temperature with elevation
becomes smaller because the air mass is heated by the positive H. In this
case, vertical air movement is easier and therefore aerodynamic resistance
becomes smaller as H becomes higher. In general, air is in a neutral condition
at a well-watered agricultural field at noon and an unstable condition at a dry
bare field at noon. The stable condition will most likely occur at night and
sometimes in the afternoon in irrigated areas surrounded by desert.

91

Tendency of vertical movement of air at near land surface,

o

assuming 6.5 C/km air lapse rate

o
19.35 C 19.7 C

o
19.35 C 19.35 C

19.35 C

19.1 C

20 C

100m
o

20 C

20 C

20 C

20 C

20 C

Stable

N eutral

U nstable

: Direction of Force
: Direction of Sensible Heat Flux

Relative shapes of eddy at near land surface

U nstable

N eutral

Figure 11.1. Three Conditions of Atmospheric Stability

92

Stable

Appendix 12
The SEBAL Mountain Model
SEBAL provides accurate estimations of ET for relatively flat agricultural areas. For use in
mountainous areas where there is significant relief and a wide range of slopes and aspects,
the SEBAL Mountain Model was developed. The SEBAL Mountain Model contains
modifications that correct for slope, aspect, and elevation in the computations. The 24-hour
solar radiation is typically much higher on a south slope than on a north slope in the
Northern Hemisphere. The result of not making the corrections for slope, aspect, and
elevation is that a higher than actual ET will be computed for north slopes and high
elevations because SEBAL incorrectly interprets the cooler surface temperatures as
indicating wet (therefore cool) areas. The cooler surface temperatures on north facing
slopes and high elevations is not due to wetness however, but is from the temperature
lapse caused by higher elevation and from less incident solar radiation caused by the
aspect.
Predictions of air flow and wind speed over mountainous areas are much more difficult. The
effects from the orthographic drainage of air caused by cooling, the acceleration of air
streams passing over mountains due to the venturi effect, and the impacts of drag due to
undulating topography all combine to make wind speed prediction difficult and complicated.
The following describes the modifications that are made to the SEBAL Flat Model in order
for it to be used in mountainous areas (SEBAL Mountain Model):
A. Data Requirements
A Digital Elevation Model (DEM) is required to provide elevation data. The units of the DEM
should be in meters with a pixel size the same as that of the satellite data (30m 30m). As
in the flat model, a layered image must first be created followed by a smaller subset image
if desired (Appendix 4). A DEM subset image must then be created which is the same size
as the subset image. To do this, one must go to the ERDAS Data Prep-subset image tool
and input the DEM from a disk and enter the coordinates of the subset image. The output
will be the DEM subset image.
B. Slope and Aspect Images
The Mountain Model requires slope and aspect information for its computations. The slope
and aspect for each pixel are derived using the DEM subset image and the ERDAS
Interpretertopographic analysisslope/aspect tool. The slope and aspect images are then
saved as the output.
C. Calculation of the Cosine of the Solar Incidence Angle (cos )
The solar incidence angle is the angle between the solar beam and a vertical line
perpendicular to the land surface. In the flat model, we assume that the land surface is
horizontal and the calculation of cos is very simple and is a constant over the area of
interest. In the Mountain Model, cos is different for each pixel depending on the slope and

93

aspect of the land surface. SEBAL uses the following equation for computing cos (Duffie
and Beckman, 1991):
cos = sin( ) sin( ) cos( s ) sin( ) cos( ) sin( s ) cos( )
+ cos( ) cos( ) cos( s ) cos( )

(12-1)

+ cos( ) sin( ) sin( s ) cos( ) cos( )

+ cos( ) sin( ) sin( s ) sin( )
where;
=
=
s=

declination of the earth (radians; positive in summer in northern hemisphere)

latitude of the pixel (radians; positive for northern hemisphere)
slope (radians) where; s=0 is horizontal and s=/2 is vertical downward (s is
always positive and represents a downward slope in any direction)
= surface aspect angle (radians) where; = 0 for due south, = -/2 for east, =
+/2 for west and = for north (Figure 12.1).
= hour angle. = 0 at solar noon, is negative in morning and is positive in
afternoon

Figure 12.1. The definitions of slope and aspect in SEBAL

In order to solve equation (12-1), the following steps are taken:

1. The sine and cosine of the slope (s) and aspect () are calculated. This is done with
model_01_sin_cos_slope_aspect. The slope and aspect image files are input and
the sine and cosine of the slope/aspect are output and saved along with the model.
2. The declination (), latitude (), and hour angle () are computed. The declination (in
radians) is computed as:
= .409 sin{(2/365 DOY) 1.39}

(12-2)

The latitude (in radians) is computed as:

= latitude in degrees /180

(12-3

The hour angle (in radians) is computed as:

= /12{(t + 0.06667(Lz - Lm) + Sc) - 12}

94

(12-4)

where; t is the standard clock time for the satellite overpass time (hours; i.e. for
14:30, t = 14.5), Lz is the longitude of the center of the local time zone (degrees west
of Greenwich; i.e., Lz = 75, 90, 105, and 1200 for the Eastern, Central, Rocky
Mountain, and Pacific time zones respectively), Lm is the longitude of the center of
the satellite image (degrees west of Greenwich), and Sc is a seasonal correction for
solar time (hours).
The seasonal correction for solar time is computed as:
Sc = 0.1645 sin(2b) 0.1255 cos(b) 0.025 sin(b)
b = 2(DOY 81)/364

(12-5)

(12-6)

3. Import and run model_02_cos_theta. Inputs are the sine and cosine of the
slope/aspect, the declination, the latitude, and the hour angle. The output file for the
cosine of the incidence angle is saved along with the model. In model_02, an initial
value for cos is computed and is then divided by the cosine of the slope to correct
it to a horizontal equivalent. This assures that all of the energy fluxes computed by
SEBAL will be analyzed on a horizontal equivalent basis.
D. Reflectivity
In the model for reflectivity, the input of cos is a file rather than a constant.
E. Transmissivity
The transmissivity is no longer a constant but varies for each pixel as a function of the
elevation. It is computed with the DEM subset image as the input.
F. Surface Albedo
The surface albedo is computed using the transmissivity file as input.
G. Incoming Shortwave Radiation
The incoming shortwave radiation (RS) is computed for each pixel since cos is now
variable for each pixel.
H. Surface Temperature
Generally, air temperature decreases 6.5oC for each 1 km elevation increase under neutral
stability conditions. Since surface temperatures are in strong equilibrium with air
temperature, one can usually observe similar decreases in surface temperature.

During the prediction of the near surface temperature difference (dT), SEBAL assumes a
linear relationship between dT and the surface temperature (Ts). The surface temperature,
however, needs to be adjusted to a common reference elevation for accurate prediction of
dT. Otherwise, high elevations that appear to be cool may be misinterpreted as having
high ET. Therefore, in the Mountain Model, a lapsed (artificial) surface temperature map
is created for purposes of computing dT by assuming that the rate of decrease in surface
temperature due to elevation increase is the same as that for a typical air profile. The

95

elevation data is provided by the DEM. This artificial lapse-adjusted surface temperature
is referred to as a DEM corrected surface temperature (Ts-dem).
The DEM corrected surface temperature is calculated by the following equation:

Ts _ dem = Ts + 0.0065z

(12-7)

where; z is the elevation of each pixel minus the elevation of a datum (meters). The term
z is positive if the elevation of a pixel is higher than the datum. The representative
elevation for the image, usually at the weather station, is used for the datum elevation.
I. Outgoing Longwave Radiation
The outgoing longwave radiation (RL) is computed as in the Flat Model using the
uncorrected surface temperature (Ts).
J. Cold and Hot Pixel Selection
The corrected surface temperature (Ts-dem) is used for the selection of the two anchor
pixels. The Ts image shows high and low pixels rather than wet or dry pixels because of the
elevation effects. For higher than datum pixels, the lapse correction is added and Ts-dem is
higher than Ts. Using the Ts-dem data allows us to find truly wet and dry pixels. The methods
described in Appendix 7 are followed except that the Ts-dem image is used rather than the Ts
image.
K. Incoming Longwave Radiation
The incoming longwave radiation (RL) is computed since the transmissivity of air is no
longer a constant for all pixels. The reference temperature (Tcold) is adjusted for the lapse
effect in a similar way as in equation (12.7):

Tcold(each pixel) = Tcold + 0.0065z

(12-8)

where; z is the elevation of the cold pixel minus the elevation of each pixel. The
elevation of the cold pixel is taken from the DEM subset image by entering its
coordinates.
L. Net Surface Radiation
The net surface radiation (Rn) is computed in model_14 similar to the Flat Model except
that incoming longwave radiation (RL) and incoming shortwave radiation (RS) are now
files rather than constants.
M. Soil Heat Flux
The soil heat flux (G) and G/Rn are computed as in the Flat Model using the uncorrected
surface temperature (Ts).

96

N. Momentum Roughness Length

The momentum roughness length (zom) is initially computed as in the Flat Model using a
land use map and LAI as input. It is then adjusted for the slope using the following
equation:

zom(mnt) = zom [1 + (slope - 5)/20]

(12-9)

O. Initial Friction Velocity and Aerodynamic Resistance to Heat Transport

The wind velocity at 200 meters (u200) is adjusted for elevation effects before computing the
friction velocity (u*) and the resistance to heat transport (rah). The elevation effects on wind
speed are complicated and difficult to quantify. A mountain wind speed weighting
coefficient () is used to adjust u200 as follows:

= 1 + 0.1[ ({Elevation Elevationstation}/1000)]

(12-10)

where; Elevation is the elevation of each pixel and Elevationstation is the elevation where
the wind speed is measured. u200 for each pixel is then adjusted by multiplying it by .
P. Sensible Heat Flux
The sensible heat flux (H) is computed as in the Flat Model. The uncorrected surface
temperature (Ts) is used to compute a new air density. The corrected surface temperature
(Ts-dem) is used for the computation of dT.
Q. Atmospheric Stability Correction
The iterative process for the atmospheric stability correction is similar to the Flat Model and
uses the same spreadsheet.
R. Applications and Precautions For The SEBAL Mountain Model

The Mountain Model is intended to improve the computation of components of the energy
balance for sloping areas. However, some uncertainties remain, so that while the Mountain
Model does provide improved estimates of ET in mountainous areas, the ET estimates in
mountains have higher uncertainty than those for flatter areas. When integrated over a
sub-basin scale, the ET from the SEBAL mountain model is probably reasonably accurate.
This is because some of the uncertainties (due to unknown wind speed and direction for
example) tend to cancel.
At the present time, the SEBAL mountain model is offered by the University of Idaho for a
purchase fee. The fee is intended to recoup development costs for the model and to
support user inquiries and needs, which may be extensive, due to the higher level of
sophistication in the mountain model.

97

98