133 
34. Natural surfaces and derived products 
The operational utility of the method is evaluated using a model of natural surface reflectance. Soil bi 
directional reflectance is approximated by Hapke’s functions. The bi-directional reflectance of a vegetated 
surface is modelled by the SAIL radiative transfer approximation (Verhoef 1984), extended to include the hot 
spot effect of finite leaf size (Jupp and Strahler 1991). Soil spectra are taken from the Purdue Soils Database 
(Stoner et al. 1980) and leaf optical properties generated using the PROSPECT model (Jacquemoud and Baret 
1990). 
Table 3 shows results of operation on three surfaces under clear and turbid atmospheric conditions. The 
sensitivity of the algorithm to violation of the assumption of Lambertian reflectance is apparent in the retrieved 
albedos for vegetated surfaces. However it is apparent that the errors are consistent between clear and hazy 
atmospheres. This suggests that these recovered albedos may be used in vegetation analysis. Examination of 
NDVI index values derived from these reflectance shows the desired sensitivity to leaf area index but 
independence from atmospheric effects. Figures 6 a and 6 b show a comparison between NDVI values derived 
from top-of-atmosphere radiances and NDVI based on values corrected using this method. Further work is 
currently being undertaken to evaluate the sensitivity of the derived index on canopy structure. 
Surface 
Channel 
Surface reflectance 
Retrieved parameters at 
lOKm visibility. 
Retrieved parameters at 
40Km visibility. 
Nadir 
50° 
albedo 
visibility 
albedo 
visibility 
Bare soil 
Cl 
2.53 
2.56 
233 
9.96 
2.53 
36.97 
C2 
3.11 
3.14 
3.13 
10.07 
3.02 
38.93 
C3 
6.96 
7.05 
6.99 
10.00 
7.00 
40.07 
C4 
8.41 
8.48 
8.47 
10.09 
8.49 
41.12 
Homogenous 
Cl 
9.62 
10.77 
7.13 
7.09 
7.96 
26.18 
canopy 
C2 
3.26 
3.59 
2.49 
9.05 
2.77 
34.75 
LAI=2.0 
C3 
37.81 
41.00 
44.73 
136 
42.86 
2.14 
C4 
16.43 
18.15 
14.91 
4.15 
14.75 
8.02 
Homogenous 
Cl 
10.25 
1132. 
7.78 
7.03 
8.77 
26.91 
canopy 
C2 
337 
3.61 
2.50 
9.04 
396 
36.32 
LAI=4.0 
C3 
4921 
52.67 
6837 
2.18 
69.01 
2.17 
C4 
17.65 
19.30 
1635 
4.06 
16.06 
7.86 
Table 3. Results of operation on natural surface reflectance models at two visibilities 
4 - DISCUSSION 
A method has been presented for atmospheric correction of radiances measured by ATSR-2 optical channels. 
The method uses the dual-look capability of the sensor to simultaneously retrieve atmospheric aerosol loading 
and surface reflectance. Correction is performed for each channel separately. The method suggested allows the 
use of accurate forward models but is operationally fast due to exploitation of pre-computation. A sensitivity 
analysis has been performed using an end-to-end simulation of radiative transfer. Results indicate highly 
accurate retrieval of surface albedo for a near Lambertian surface, and robustness to tropospheric aerosol 
variation. The results show considerable sensitivity to deviation of the surface reflectance from Lambertian, 
and that this sensitivity is implicit in the information content of two views of a surface at a single wavelength. 
Initial results using a model of vegetated surface reflectance suggest indices derived from the algorithm output 
are robust to variation in atmospheric turbidity. 
Further work is being undertaken to explore use of the information content of multiple wavebands to allow 
inversion of more sophisticated surface and atmospheric models. This necessitates assumptions about the likely 
spectral variation of surface bi-directional reflectance. Realistic models of natural surface bi-directional 
reflectance and exact models of atmospheric scattering are being developed for algorithm evaluation.