You are on page 1of 17

AN OPERATIONAL METHOD FOR DERIVING STANDARDISED

SURFACE REFLECTANCE FROM LANDSAT TM AND ETM+


IMAGERY FOR QUEENSLAND AND NEW SOUTH WALES

Tony Gill1,2, Neil Flood1,2, Sam Gillingham1,3, Tim Danaher1,4, James


Shepherd5, John Dymond5

1. Joint Remote Sensing Research Program.

2. University of Queensland. School of Geography, Planning and


Environmental Management, St Lucia, QLD, 4072.

3. Queensland Department of Environment and Resource Management. 80


Meiers Road, Indooroopilly, QLD, 4068

4. New South Wales Department of Environment, Climate Change and


Water. PO Box 856, Alstonville, New South Wales, 2477.

5. Land Care Research. Private Bag 11052, Palmerston North, 4442,New


Zealand.

Abstract
Operational monitoring of vegetation cover and vegetation cover change over
large areas requires the use of satellite sensors that measure radiance reflected
from the Earth’s surface. Monitoring programs use multiple images for complete
spatial coverage over time. Accurate retrievals of cover and change estimates
can be made difficult by variation, in both space and time, in the measured
radiance caused by atmospheric conditions, topography, sensor location, and
sun elevation. In order to obtain estimates of cover that are comparable
between images, and to retrieve accurate estimates of change, these sources
of variation must be removed. In this paper we present a preprocessing scheme
for minimising atmospheric, topographic and bi-directional reflectance effects on
Landsat 5 TM and Landsat 7 ETM+ imagery. The approach involves the
following sequence of steps: conversion of image digital numbers to calibrated
at-sensor radiances; atmospheric correction to compute surface-leaving
radiance; bi-directional reflectance modelling to remove the effects of
topography and bi-directional reflectance. The result is surface reflectance
standardised to a fixed viewing and illumination geometry. Validation shows that
there is a reduction in variation in reflectance for overlapping pixels on adjacent
Landsat scenes. Further validation with independently derived estimates of
surface reflectance is required. The method was used to process over 15000
Landsat 5 TM and Landsat 7 ETM+ scenes and is therefore operationally
viable.

1
Introduction
Large area monitoring of vegetation cover depends on relationships between
field observations of cover and the spectral radiance measured by satellite
image sensors. These relationships are collectively called vegetation indices,
and may be as simple as using the normalised difference vegetation index, or
be more complicated multiple regression relationships. The field observations
used in the development of the indices for an operational program are collected
at a range of locations and at different times of the year in order to capture the
variation of cover values across the landscape. Therefore, the location and
timing of the acquisition of satellite imagery will vary to match, as closely as
possible, the location and timing the field measurements. With the index
developed, it is applied to imagery collected at a range of times and locations,
for making predictions of vegetation cover. The method is currently used by the
Queensland Department of Environment and Resource Management (DERM)
and NSW Department of Environment and Climate Change (DECC) for woody
vegetation cover and grass cover monitoring.
Variation in measured radiance, caused by factors other than variation in
vegetation cover, reduces the accuracy of the derived vegetation cover
estimates if not accounted for. The factors include differences in the sensor's
sensitivity to incoming radiation, atmospheric conditions, and sun and sensor
positions. A change in atmospheric conditions and sun and sensor elevations
will alter the amount of light scattered and absorbed by the atmosphere. These
changes affect both the amount of light illuminating the surface and the amount
of light entering the sensor. The topography and land cover vary how light
illuminates and is reflected from the surface, and this varies with the relative
locations of the sun and the sensor.
Assuming the sensor radiometrics are well characterised, as is the case for
Landsat-5 TM and Landsat-7 ETM+, then obtaining accurate vegetation cover
retrievals from satellite imagery requires reducing the variation caused by
atmospheric conditions, topographic and bi-directional reflectance effects. The
aim of the work outlined is this paper was to develop a preprocessing scheme
for deriving standardised surface reflectance from Landsat imagery.
Standardised means reflectance free from topographic and bi-directional
effects. The scheme must be simple enough to implement within an operational
environment where thousands of images are to be processed. A preliminary
analysis of the validation of the scheme is provided.
Methods

Standardised reflectance
The equation used to model the brightness of a sloping surface was
      


(1)

where L is the surface-leaving radiance in the direction of the sensor,  and
 are the surface reflectances in the direction of the sensor for direct and

2
diffuse irradiance respectively, and   and   are the direct and diffuse
irradiances respectively. All quantities have implied band dependencies, and
  and   are also dependent on altitude. Solving equation 1 requires an
estimate of the diffuse reflectance. The direct reflectance is the major
contributor to the total signal received by the sensor in most cases, therefore
diffuse reflectance was modelled as a multiple, β, of the direct reflectance
(Shepherd and Dymond 2003), so that equation 1 becomes
      


(2)

Computing standardised reflectance requires that the sloping surface


reflectance be transformed to a value that would be obtained at a standard set
of sun and sensor positions relative to the ground topography. We define a
correction factor,  , as the value that transforms a measured direct
reflectance, 

, to a predicted (or standardised) direct reflectance, 
   

(3)

By substituting equation 3 into 2 and rearranging, the predicted direct


reflectance was computed from measurements of surface radiance,  , direct
irradiance,  , and diffuse irradiance,  , as
 

 

 
   
  (4)

where  is the diffuse reflectance multiple for the measured reflectance. The
quantities  ,  , and  were obtained from radiative transfer modelling of
 

the atmosphere. The correction factors  and  were obtained from


bidirectional reflectance modelling. The methods that were used to derive these
values are described in the next two sections.

Atmospheric radiative transfer modelling


The atmospheric transfer modelling software, 6S (Vermote et al. 1997),
provided estimates of direct and diffuse irradiance onto a horizontal surface,
 and  respectively. 6S also provided estimates of apparent surface


reflectance, , from the observed top of atmosphere radiance. The top of


atmosphere radiance was obtained from the satellite imagery by applying the
supplied gains and offsets to each pixel digital number. The apparent surface-
leaving radiance was computed from 6S as
   

 

(5)

The total surface radiance onto a sloping surface can be written as

3
cos 
         

cos 
(6)

where  =   is the total irradiance onto a horizontal surface;  is the




incidence angle between the surface normal and the sun vector;  is the
incidence angle for a horizontal surface (solar zenith angle); Θ is a binary
coefficient that is set to zero when the incidence angle is greater than 90
degrees, and set to one otherwise;  is a sky-view factor;  is a terrain view
factor; and  is the average reflectance of surrounding pixels (Shepherd and
Dymond 2003). The first term can be substituted for  
in equation 4, and the
sum of the second and third terms can be substituted for  in equation 4.


The sky view factor,  , is the proportion of sky visible to a given pixel, relative
to a hemisphere of sky which would be visible if no surrounding terrain
obstructed the view. The hemisphere can be thought of as having its base
aligned with the slope of the pixel. The sky view factor was calculated using the
horizon search methods detailed by (Dozier and Frew 1990). For each pixel, the
angle to the horizon is found in 16 search directions, and a sky view factor
along that azimuth is computed. The sky view factor for the whole hemisphere
is approximated as the average of these 16 azimuthal view factors. The terrain
view factor,  , is assumed to be whatever remains of the hemisphere over the
slope at that point, thus   1  . The sky view factor,  , was pre-
computed for every pixel, using the Shuttle Radar Topography Mission (SRTM)
digital elevation model (DEM).
6S was run for each Landsat scene to be corrected. The model was run at a
range of angle combinations, with the resolution of angle change (in radians)
being 0.01 for sun zenith, 0.02 for satellite zenith and azimuth, and 0.04 for sun
azimuth. 6S was also run for a range of altitudes, for each angle combination,
with an altitude step size of 200 m starting at sea level up to the maximum
height as indicated by the SRTM. Each pixel is then corrected using the
resulting 6S parameters for the nearest corresponding elevation and angles.
Table 1 lists the parameters required by 6S and the source of those parameters
used for this study.

Topographic and bi-directional reflectance modelling


The reflectance from a target on the ground varies with both incidence angle
and exitance angle (the angle between the surface normal and the view vector).
The variation is caused by both topography and the anisotropic nature of how
light reflects from the Earth’s surface due to different land cover types. The light
reflected from the surface is therefore highly dependent on the incidence and
exitance angles, and the phase angle between the incidence and exitance
vectors. One combination of incidence, exitance and phase angles results in
one possible, bi-directional, reflectance value. An infinite number of
combinations results in the bi-directional reflectance distribution function
(BRDF) for a surface.

4
Table 1: The source of the parameters used in atmospheric transfer modelling with 6S.
Parameter Source
Continental aerosol model 6S (Vermote et al. 1997)
Ozone TOMS climatology 1996-2003
(Ziemke et al. 2006)
Aerosol Optical Depth (AOD) at 550 Fixed at 0.05
nm
Precipitable water Daily interpolated vapour pressure
point observations (Jeffrey et al.
2001)
Solar zenith and azimuth angles Pre-calculated
View zenith and azimuth angles Pre-calculated

We chose to consider the topographic and bi-directional effects together with


one model. This implicitly assumes that canopy roughness is the same for both
horizontal and inclined surfaces (Dymond and Shepherd 1999). Topographic
correction is then correction for the BRDF that slopes, by their nature, introduce.
Therefore we seek to obtain the direct and diffuse reflectance correction factors,
γ and β, in equation 4, from the BRDF.
One of the challenges of modelling the BRDF of the land surface is that it must
be done from a finite number of bi-directional reflectance observations. Ideally,
a large number of observations would be obtained for a target over a range of
incidence and exitance angles within a very short period of time to reduce the
effect of on-ground change. For example, a field spectrometer mounted on a
goniometer permits the collection of a large number of bi-directional
observations over the course of a day.
Scale differences between field and satellite-measured reflectance results in the
need to measure the bi-directional reflectance at the scale that the BRDF model
will be used at. For example, the MODIS BRDF product is derived for the
Earth’s surface every 500 m from imagery with a nominal spatial resolution of
500 m (Schaaf et al. 2002). However, 16 days of bi-daily observations are
required to obtain enough points to reliably parameterise the MODIS BRDF.
This is not possible with Landsat data where one observation for a point on the
Earth is, nominally, only available every 16 days, and is always from the same
view angle. Thus the range of angles is restricted to that obtained by the
seasonal variation in sun position, and even this is only available over a long
time period, which is then susceptible to on-ground change in the pixel being
observed. Therefore a different scheme for obtaining a BRDF of the land
surface from Landsat data is required.
The scheme used to derive a BRDF model of the land’s surface requires
collecting a set of pairs of bi-directional reflectance observations. Each pair
represents two observations of the same location obtained close in time, but
under different illumination and viewing conditions. Ground conditions are

5
assumed to be the same; therefore any difference in reflectance is due solely to
the difference in phase angle (the angle between illumination and viewing
vectors). The scheme has been used previously to obtain BRDF models from
AVHRR and Landsat imagery (Shepherd and Dymond 2000, Danaher 2002).
Pairs of observations were obtained from the atmospherically corrected Landsat
imagery. Two important effects resulting from the nature of the Landsat orbit
need to be considered when selecting pairs of observations (pixels). Firstly,
images between adjacent paths overlap. Secondly, a location is observed at
approximately the same local time each cycle. View angle variation was
obtained by selecting pairs within the overlap regions. Sun angle variation was
obtained by selecting pairs on either side of the equinox. Using pairs eight
weeks apart, on either side of the equinox provides a maximal change in sun
angle variation for a minimal change in ground conditions.
Using pairs of observations constrains how the BRDF model can be solved and
implemented. One limitation is that a per-pixel BRDF model cannot be obtained.
A land cover dependent BRDF model can be obtained by stratify the pairs of
observations by land cover. Such a scheme requires prior knowledge of the
land cover, which will change both spatially and temporally. If the standardised
imagery is to be used to derive land cover estimates, then those estimates can
be used to solve the land cover dependent BRDF model. However, the process
would be recursive, needing an initial estimate of the BRDF to begin with. A
simpler solution is to derive a single BRDF model for the entire landscape.
Therefore a BRDF model that is flexible enough to handle different land cover
conditions is required. The pairs of observations used to parameterise the
model must also be selected from a range of land cover and topographic
conditions. A single BRDF model, once derived, has the added advantage that
it is straight-forward to implement operationally.
The kernel-driven Ross-Thick Li-Sparse (RTLS) reciprocal BRDF model was
chosen as it has been found to perform well for a range of land cover types
(Privette et al. 1997). Variation in land cover and topography were included in
the set of observation pairs by sampling across the landscape. Pairs were
obtained in both the spring and autumn equinoxes to reduce bias towards one
particular type of change in land cover associated with a change in sun angle
(Table 2). From each overlap pair, a set of pixels was selected to satisfy certain
criteria. A number of sets were selected from horizontal ground (slope < 3%)
and steep ground (slope > 50%), for a range of land cover conditions as
determined using foliage projective cover (FPC) classes (0-5%, 5-10%, 10-15%,
15-20%, 20-30%, 30-40%, 40-50%, 50-60%, 60-70%, 70-80%, 80-90%), and
these sets were combined to make a single large set of pixels.

6
Table 2: The dates, locations and solar angles for the overlapping pairs of Landsat
imagery from which pairs of pixels were obtained for parameterising the BRDF model.
High-Sun Path/Row Sun Low-Sun Path/Row Sun Time Gap
Date Zenith Date Zenith (days)
19/10/2003 095/074 32o 9/9/2003 096/074 43o 40
1/3/2009 096/074 38o 27/4/2009 095/074 47o 57
12/10/2003 094/078 37o 18/8/2003 093/078 53o 55
19/2/2005 094/078 39o 17/4/2005 093/078 49o 57
3/10/2003 095/079 40o 24/8/2003 096/079 53o 40
o o
21/2/2004 090/079 41 4/5/2004 089/079 56 76
o o
2/10/2004 090/079 39 24/8/2004 089/079 52 39
o o
24/2/2004 095/079 42 20/4/2004 096/079 52 56

The RTLS model is comprised of a linear combination of three different types of


surface scattering. The first is isotropic and represented by a single parameter
"#$ . The second is volume scattering and modelled using the kernel,
%&$' , (, ), of Ross (1981), where ) is the relative azimuth angle. The third is
geometric scattering modelled using the geometric shadow casting kernel,
%*+$ , (, ), of Li & Strahler (1992). The kernels are both dependent on
incidence, , exitance, (, and relative azimuth, ), angles, which are all functions
of their respective zenith, ,, and azimuth angles, -. Direct reflectance from a
surface can be modelled as
 , (, )  "#$  "&$' %&$' , (, )  "*+$ %*+$ , (, ) (7)

The task is to determine a single set of parameters, "#$ , "&$' , "*+$ , from the set
of observation pairs. Recall that  is defined as the value that transforms a
measured direct reflectance,  
, to a predicted (or standardised) direct
reflectance,  , as described by equation 3. Therefore  can be obtained for


a pair of pixels from equation 7 by substituting in the appropriate angles. Recall


that the diffuse reflectance,  was modelled as a multiple, , of the direct
reflectance,  . Therefore, determining requires an estimate of diffuse
reflectance. Diffuse reflectance for a fixed exitance angle can be determined
numerically from the BRDF model as
2⁄1 12
1
 
(  0 0  , (, )
./
(8)
45 345

where . and / are the number of incidence and relative azimuth angles
respectively for which the integral is to be evaluated. Appendix A shows the
derivation of equation 8. Hence is given by

7
, (, )   (/  , (, ) (9)

Because we only require a reflectance ratio,  , rather than a reflectance, we


must normalize the reflectance expression to avoid non-unique solutions. Thus
"#$  "&$' %&$' 8 , ( , ) 9  "*+$ %*+$  , ( , ) 
 
"#$  "&$' %&$'  , ( , )   "*+$ %*+$  , ( , ) 

" "*+$
(10)
1  &$' %&$' 8 , ( , ) 9  %  , ( , ) 
"#$ "#$ *+$   

" "*+$
1  &$' %&$'  , ( , )   %  , ( , ) 
"#$ "#$ *+$   

where the subscripts : and ;, refer to the measured and standardised angles
respectively, and we are able to solve for the two normalized parameters
"&$' ⁄"#$ and "*+$ ⁄"#$ . To determine these parameters we minimised

1= 
1

0 <= <
1  1 1

(11)

Equation 11 is the same formulation as equation 4. The process was to predict


the measured reflectance of observation 1 from the measured reflectance of
observation 2 for each observed pair, to compute the absolute difference
between the observed and predicted values, and to minimise the sum of these
differences.

Operational implementation
There are currently over 15000 Landsat 5 TM and Landsat 7 ETM+ images in
the Queensland DERM archive as downloaded from the United States
Geological Survey (USGS) archive. All of these images were converted to
standardised surface reflectance using the following sequence of steps.
• The image digital numbers were converted to top of atmosphere
radiance using the provided gains and offsets.
• The apparent surface radiance, and horizontal-surface direct and diffuse
irradiances were computed using 6S.
• The horizontal-surface irradiances were converted into sloped surface
equivalents.
• The standardised surface reflectance was computed using the BRDF
model parameters with equation 4. The images were standardised to an
incidence angle of 45 degrees and exitance angle of 0 degrees.

8
Validation
An image-based validation approach was used to validate the capacity of the
and correction factors in equation 4 to adjust a reflectance value from one set
of angles to another. Four independent sets of observation pairs were obtained
to test the model under a range of conditions (Table 3). In each case the
surface reflectance of the first pixel (with the higher sun elevation angle) in each
pair was left unchanged. The surface reflectance of the second pixel (with the
lower sun elevation angle) was used to predict the reflectance at the same sun
and viewing geometry as the first pixel using equation 4. A comparison of the
difference between the unchanged observations, and the difference between
the observations with one pixel modified was undertaken for four different
cases.

Table 3: The dates, locations and solar angles for the overlapping pairs of Landsat
imagery from which pairs of pixels were obtained for validating the standardised
imagery.
High-Sun Path/Row Sun Low-Sun Path/Row Sun Time Gap
Date Zenith Date Zenith (days)
21/10/2003 093/077 34o 25/8/2003 094/077 50o 57 days
21/10/2009 093/078 32o 27/8/2009 092/078 49o 55 days
o o
10/3/2007 089/080 42 4/5/2007 090/080 55 55 days
o o
1/3/2009 096/074 38 27/4/2009 095/074 47 57 days

In all cases, pixels were screened for null values, cloud contamination and only
pixels with incidence angle < 80 degrees were used, to ensure that at least
some direct sunlight was illuminating the pixel.
Case 1 involved flat terrain (slope less than 2 degrees). This case was analysed
as the majority of pixels to be corrected in Australian conditions are on flat
ground. The pixels were chosen to have slope < 3%.
Case 2 involved observations made on steep terrain, in order to cover a wide
range of incidence and exitance angles. Pixels were restricted to those having
slope > 50%.
Case 3 was a special case of case 2, that of pixels lying on the principle plane.
When the land surface is horizontal, it is almost impossible in Australia to have
a Landsat configuration along the principle plane. However, when the land
surface is sloping in just the right way, it becomes possible, although even then
the reflection hotspot can never be observed. Pairs of pixels in this set were
selected as steep, as in Case 2, but further restricted to have the relative
azimuth in the plane of the surface to be in the ranges 0 to 5 or 175 to 180
degrees, for one of the pair of dates. This implies that the sun-surface-satellite
configuration lies along the principle plane. This will only be true for one of the
dates in each pair, and in the other date the configuration will be off the principle
plane. This test was included because this seemed likely to be the part of the

9
th the most structure, and therefore more challenging for a
BRDF shape with
BRDF adjustment.
For Case 4 pixels were selected from steep terrain, as in Case 2, but
exclusively on hill slopes facing away from the sun. The angle between the
aspect of the hillside and the solar azimuth was restricted to the range 120 to
240 degrees, for the low
low-sun
sun image of each image pair. These slopes would be
the darker side of the hill, and would therefore be most sensitive to the
illumination by diffuse irradiance in shorter wavelength
wavelengths.
s. Running the same test
as above would test the part of the algorithm in which reflectance is calculated
from radiance, using the combination of direct and diffuse light, as deduced
from the outputs of 6S.

Results
Figures 1 to 4 show the scatter plots for pairs of pixels under test cases 1 to 4.
On all plots, the X axis is the high sun reflectance value, while the Y axis is
reflectance value from the low sun image, both with or without adjustment to the
angles for the high sun image. The solid line is the 11-to-1
1 line, and the dashed
line is the regression
ssion line (orthogonal distance regression), constrained to pass
through the origin. The slope of the regression line gives a measure of
systematic bias in the prediction.

Figure 1: Validation
alidation plots of the BRDF correction on flat terrain.. The left plot shows
unadjusted data, and the right plot shows the low sun pixels adjusted to match the
angles of the high sun pixels.

10
Figure 2: Validation
alidation plots of the BRDF correction on steep terrain.. The left plot shows
unadjusted data, and the right plot shows the low sun pixels adjusted to match the
angles of the high sun pixels.

Figure 3: Validation
alidation plots of the BRDF correction along the principle plain. The left plot
shows unadjusted data, and the right plot shows the low sun pixels adjusted to match
the angles of the high sun pixels.

11
Figure 4: Validation
alidation plots of the BRDF correction on the dark side of the hill. The left
plot shows unadjusted data
data, and the right plot shows the low sun pixels adjusted to
match the angles of the high sun pixels.

Figure 5 shows a region in South East Queensland. The image on the left is the
standardised without allowing for topography. T The
he image on the right is the
standardised image using the angles ap appropriate
propriate for the terrain at each pixel.
pixel
The removal of the topographic effects is apparent in the image on the right.
right
Figures 6 and 7 show Landsat 5 TM images mosaics for the states of
Queensland and New South Wales respectively. In both Figures the image to
the left is the at-sensor
sensor radian
radiancece with no correction applied, and the image to
the right is the standardised surface reflectance standardised to nadir view and
an exitance angle of 45 degrees. In both cases the spatial variation in
brightness, due to varying solar zenith angles between image acquisition dates
and view angles across the Landsat scenes, is greatly reduced. Note that
variation in brightness due to different vegetation cover conditions is retained,
and is particularly evident in the New South Wales image.

12
Figure 5: Removal of topographic effects. Both Landsat 5 TM images are of the same
location in South east Qld, in the Gold Coast hinterland. The image on the left is at-
sensor radiance with no correction applied. The image on the right is the standardised
surface reflectance and shows that most topographic effects have been removed. Note
that areas in full shadow cannot be reliably estimated, and are masked out in the image
on the right (white areas). Images are false colour composites of bands 5, 4, 2 shown
as red, green, and blue respectively.

Figure 6: At-sensor radiance (left) and standardised surface reflectance (right) Landsat
5 TM mosaics for the state of Queensland. Images are false colour composites of
bands 5, 4, 2 shown as red, green, and blue respectively.

13
Figure 7: At-sensor radiance (left) and standardised surface reflectance (right) Landsat
5 TM mosaics for the state of New South Wales. Images are false colour composites of
bands 5, 4, 2 shown as red, green, and blue respectively.
Discussion
The results show that the BRDF modelling captured the variation caused by
differences in topography and viewing and illumination geometry. The scatter
plots show that in the majority of cases for the majority of bands, the variation in
reflectance is reduced (correlation coefficients and slope of line of best fit get
closer to 1). The validation plots are supported by the imagery that shows that
the shading caused by topography is reduced along with a reduction in the
variation in reflectance between overlapping pairs of images.
The nature of the Landsat data means that the BRDF modelling is not
performed for each pixel. The model, therefore, does not capture fine-scale
variation in bi-directional reflectance. Instead, the model accounts for the large-
scale variation in bi-directional reflectance caused by changes in topography,
sun and sensor positions.
Future work will focus on validating the estimates of surface reflectance from
alternate data sources. One approach is to compare the reflectance values with
scaled-up field observations acquired using a spectrometer. Another approach
is to compare the Landsat estimates to those obtained from airborne
hyperspectral imagery.

Conclusions
A method for producing standardised surface reflectance estimates was
developed. The method utilised atmospheric transfer modelling code and
subsequent modifications to obtain estimates of apparent surface radiance and
direct and diffuse irradiance onto a sloping surface. The radiative transfer
modelling outputs were used along with a parameterised BRDF model, to derive
standardised surface reflectance estimates. Validation of the BRDF model
showed that it was capable of reducing variation caused by topography and
viewing and illumination conditions. The method was shown to be operationally
viable as it was used to standardise over 15000 Landsat TM and ETM+ images
of Queensland.

14
Appendix A. Derivation of the diffuse reflectance factor
In equation 8 the diffuse reflectance is modelled as the average sum of all direct
reflectance factors for a fixed view zenith and azimuth (exitance) angle. This
result can be derived as follows. The diffuse reflectance factor is
 (

 
( 
 
(12)

where  is the radiance from a surface due to the diffuse irradiance and
reflectance only. Diffuse irradiance can be thought of as an infinite number of
point light sources covering the hemisphere above the target. Therefore  is
the sum of the all radiance reflected from the surface towards the sensor
emanating from the infinite number of point light sources. This can be
expressed as

 >  ? @, >  @A@ (13)


B

where C is the hemisphere, and @ and > are the sun and view vectors
respectively. Assuming the diffuse light is isotropic, then for . incidence and k
relative azimuth angles equation 13 can be evaluated as
2⁄1 12
1  
 (  0 0  , (, )

./
(14)
45 345

Finally, equation 14 can be substituted into equation 12 to give equation 8.

15
Appendix B. Table of symbols
Symbol Meaning
, (, ) Incidence, exitance and relative azimuth angles
 , ( Incidence and exitance angles onto a horizontal surface
(equivalent to view zenith and solar zenith angles).
 ,  Surface reflectance: direct, diffuse
 , 
 Surface reflectance: the subset p is for a predicted set of
, (, ), and the subset m is for an observed set of , (, ).
,  ,  ,  , Surface-leaving radiance: total, direct and diffuse. The

 , 
 subscript m is for a specific set of observed , (, ).

  ,   ,  , Surface irradiance: direct and diffuse. No subscript denotes


 ,  , 
   values for a sloping surface. The subscript h denotes the
values for a horizontal surface. The subscript m is for a
specific set of observed , (, ).
 Correction factor that transforms a measured direct
reflectance, 

, to a predicted direct reflectance,  .
 diffuse reflectance multiple
 Binary coefficient. Zero for incidence angle > 90°.
 ,  Diffuse irradiance sky and terrain view factors.
 Average reflectance of pixels surrounding a given pixel.
"#$ , "&$' , "*+$ Ross Thick Li Sparse reciprocal BRDF model coefficients
%&$' , %*+$ Ross Thick Li Sparse BRDF kernels
., / The number of incidence and exitance angles used to
evaluate the diffuse reflectance factor.
@, > Sun and view vectors.

References

DANAHER, T. J., 2002, An empirical BRDF correction for Landsat TM and ETM+
imagery. In 11th Australasian Remote Sensing and Photogrammetry
Conference (Brisbane, Australia.
DOZIER, J., and FREW, J., 1990, Rapid calculation of terrain parameters for
radiation modeling from digital elevation data. IEEE Transactions on
Geoscience and Remote Sensing, 28, pp. 963-969.
DYMOND, J. R., and SHEPHERD, J. D., 1999, Correction of the topographic effect
in remote sensing. IEEE Transactions on Geoscience and Remote
Sensing, 37, pp. 2618-2620.

16
JEFFREY, S. J., CARTER, J. O., MOODIE, K. B., and BESWICK, A. R., 2001, Using
spatial interpolation to construct a comprehensive archive of Australian
climate data. Environmental Modelling & Software, 16, pp. 309-330.
LI, X., and STRAHLER, A. H., 1992, Geometric – optical bidirectional reflectance
modeling of the discrete crown vegetation canopy: effect of crown shape
and mutual shadowing. IEEE Transactions on Geoscience and Remote
Sensing, 30, pp. 276-292.
PRIVETTE, J. L., ECK, T. F., and DEERING, D. W., 1997, Estimating spectral albedo
and nadir reflectance through inversion of simple BRDF models with
AVHRR/MODIS-like data. Journal of Geophysical Research, D-102, pp.
29529– 29542.
ROSS, J. K., 1981, The radiation regime and architecture of plant stands
(Norwell, MA: Dr. W. Junk), pp. 392.
SCHAAF, C. B., GAO, F., STRAHLER, A. H., LUCHT, W., LI, X., TSANG, T.,
STRUGNELL, N. C., ZHANG, X., JIN, Y., MULLER, J.-P., LEWIS, P., BARNSLEY,
M., HOBSON, P., DISNEY, M., ROBERTS, G., DUNDERDALE, M., DOLL, C.,
D'ENTREMONT, R. P., HU, B., LIANG, S., PRIVETTE, J. L., and ROY, D., 2002,
First operational BRDF, albedo nadir reflectance products from MODIS.
Remote Sensing of Environment, 83, pp. 135-148.
SHEPHERD, J. D., and DYMOND, J. R., 2000, BRDF correction of vegetation in
AVHRR imagery. Remote Sensing of Environment, 74, pp. 397-408.
SHEPHERD, J. D., and DYMOND, J. R., 2003, Correcting satellite imagery for the
variance of reflectance and illumination with topography. International
Journal of Remote Sensing, 24, pp. 3503-3514.
VERMOTE, E. F., TANRE, D., DEUZE, J., HERMAN, M., and MORCRETTE, J., 1997,
Second simulation of the satellite signal in the solar spectrum, 6S: an
overview. IEEE Transactions on Geoscience and Remote Sensing, 35.
ZIEMKE, J. R., CHANDRA, S., DUNCAN, B. N., FROIDEVAUX, L., BHARTIA, P. K.,
LEVELT, P. F., and WATERS, J. W., 2006, Tropospheric ozone determined
from Aura OMI and MLS: Evaluation of measurements and comparison
with the Global Modeling Initiative's Chemical Transport Model. Journal
of Geophysical Research, 111.

17

You might also like