63 
3.3.2. Application for different situations over the Coral Sea. The set of coincident ATSR temperatures 
and in-situ measurements over the Coral Sea has been used again. The differences have been plotted on 
Fig. 9. The mean difference is equal to 0.06 K and the rms is 0.44 K. These results are worth noticing 
since no ancillary information has been used. The variability of the situations must be noted too : the 
data, ranging from September 1991 to December 1991, correspond to different atmospheric situations 
with different water vapor amounts and different emissivities. 
1 
0.5 
K 
0 
-0.5 
- 1 
A = 0.06 K 
o = 0.44 K 
• Tsec-Tsea (using the simplified method) 
Figure 9. Application of the double-viewing method over the Coral Sea 
(September to December 1991) 
3.4. The case of LST retrieval with the double-viewing method 
As shown previously, two sources of error appear when using Eq. 2 (secante method equation) : the 
solution would be right if there were no reflected radiance from the ground, and if both nadir and forward 
emissivities were equal. This last condition may be complete over a lambertian surface. And the problem 
of the surface reflected radiance may be resolved if only the surface emitted radiance B(T s b) is required, 
instead of B(T S ): 
B(T „ ) = e,B(T,) + ( 1 - e, )Rad i (5) 
T sb is the Surface Brightness Temperature which would be measured with a surface radiometer. 
The simplified double-viewing equation reduces then to (4). In the case where Nadir and 
Forward emissivities are not equal, the remaining error is a function of 8 (difference between E n and ef). 
A simulation with mean atmospheric conditions gives the following relation between 8 and this error (in 
K): 
AT = 100 8 (6) 
For instance, with 8=10*3 ( a realistic directional emissivity difference) the error on the retrieved surface 
brightness temperature is about 0.1 K. Therefore the double-viewing method may be used on land pixels 
to retrieve the surface radiance (or surface brightness temperature). 
4 - CONCLUSION 
Concerning the Split-Window method, it clearly appears that the water vapor amount has to be taken into 
account when using the Split-Window technique. A simple measurement of the total water vapor content 
really improves the surface temperature retrieval: in that case the mean error is between -0.2 K and 0.2 K 
for both applications. These good results have to be confirmed on other datasets, and especially on land