796 
(2) Identification of the best atmospheric model introducing the smallest difference between 
Ts(4) and Ts(5) 
(3) Calculation of the optimal ground -level brightness temperature Ts by 
Ts=Ts(4)+(Ts(4)-Ts(5) 
where Ts(4) and Ts(5) are calculated by using the best atmospheric model. 
4 RESULTS 
Table 2 displays the brightness temperatures measured by ESR1-ATSR or NOAA11-AVHRR sensors and those 
measured at ground level at satellite overpass time. It shows that the atmospheric effects play an very important 
role. The difference between the satellite-level brightness temperatures and the ground-level brightness 
temperatures in the best atmospheric window (centred at 11 pm) can reach up to -12.6°. This difference is well 
correlated with the spectral difference of brightness temperature. 
Table 2: Satellite measured brightness temperatures and filed measured brightness temperatures at ESR1-ATSR 
or NOAA11-AVHRR overpass times. Tsat (11 pm) and Tsat (12 pm) indicate the brightness temperatures 
measured in the AVHRR4 and AVHRR5 channels or equivalent channels. The temperatures are expressed in °C. 
Number 
1 
2 
3 
4 
5 
6 
7 
8 
Sensors 
ATSR 
ATSR 
AVHRR 
AVHRR 
ATSR 
AVHRR 
TM 
TM 
Date 
7/27 
7/27 
7/27 
7/27 
7/24 
7/24 
7/20 
8/5 
Target 
Crau 
Lake 
Crau 
Lake 
Crau 
Crau 
Crau 
Crau 
Tsat (11 pm) 
39.2 
25.3 
36.2 
25.4 
34.4 
30.2 
27.7 
31.8 
Tsat (12 pm) 
37.2 
24.5 
32.7 
23.8 
31.3 
25.9 
Tsat(l 1 pm)-Tsat (12 pm) 
2.0 
0.8 
3.5 
1.6 
3.1 
4.3 
Tsoi 
44.1 
27.8 
45.3 
28.6 
41.8 
42.8 
36.9 
39.5 
Tsat(ll pm)-Tsol 
-4.9 
-2.5 
-9.1 
-3.2 
-7.4 
- 12.6 
-9.2 
-7.7 
4.1 Comparison of Split Window methods 
Figure 3 represents the differences between the brightness temperatures measured at ground level and those 
estimated by ten published Split-Window methods. The abscissa from 1 to 6 corresponds to the six field 
brightness temperature measurements indicated in table 2. Additional analyses have shown that the effect of the 
spectral difference between homologous bands is very small on the Split-Window methods and can be neglected 
in our case. 
In Figure 3, we can see that the temperature estimated by RAL93, Li93 give very good correspondence with 
field-measured brightness temperatures. Ottl692, Ulivieri85 and Nesdis92 Split-Window coefficients give also 
acceptable results. The temperatures estimated by Becker90, Price84 are systematically warmer (about 2°C) than 
the field measurements, but those estimated by Kerr92 and Deschamps80 as well as Ulivieri92 are systematically 
cooler (about 2°C). This confirms also the two recent results obtained by Li et al. (1993) and Schmugge et al. 
(1993), who have pointed out the 1.5 °C over-estimation of Becker90 method (Li et al., 1993) and 1.5 °C under 
estimation of Kerr92 method (Schmugge et al., 1993). 
Figure 3 shows also that there are not significant effects when we mix the Split-Window coefficients established 
for NOAA7,9, 11 AVHRR and ERS1 ATSR sensors. 
4.2 Evaluation of ground temperature derived from satellite data by the use of LOW TRAN model 
The atmospheric correction resulting of the use of LOWTRAN7 radiative transfer code is presented in Figure 4. 
We can see that for ATSR on the 27 July, the difference is quite small between the field-measured brightness 
temperature and that estimated by Lowtran7 with a special coincident radiosounding. If a standard radiosounding 
performed at about 12 h is used as input data, the residual errors on atmospheric correction, for two days, are 
between 1 and 2°C for ATSR, but up to 11°C for AVHRR. This effect can be explained by the fact that ATSR