~1 1 1 r 
8 9 10 11 
jre (Degree C) 
the brightness 
field and those 
;ode for ATSR, 
images were acquired in the morning (about 10h30) and AVHRR in the afternoon (about 14h30). The 
atmospheric condition was relatively stable in the morning but changed very rapidly in the afternoon in the test 
site area during the experimental campaign. 
Figure 4 shows also that the two ground level brightness temperatures estimated respectively from the 11 pm and 
12 pm channels for the same target are different. This spectral difference is highly correlated to the residual error 
of the average of the two estimated temperatures compared with the field measurement: the more important is the 
spectral difference, the larger is the error. The positive difference between the ground level brightness 
temperature estimated by using 11 pm measurement and that estimated by using the 12 pm measurement 
indicates an under-estimation of the real ground level brightness temperature. The negative difference indicates 
an over-estimation of the real ground level brightness temperature. This analysis can be used to check the 
representativeness of used atmospheric profiles. But we must note out that the comparison of Split Window 
results with Lowtran corrections shows that a well checked Split-Window method can provide a result as good as 
that obtained by using a special coincident radiosounding. 
For TM images, the two days results obtained with special coincident radiosoundings and standard 
radiosoundings show that the estimated brightness temperatures are cooler than the field measurements. The 
problem of the quality of radiosounding can not explain this phenomena: the coincident radiosounding for ATSR 
on the 27 July has shown its accuracy and the similar results obtained with standard radiosoundings exclude the 
rapid changes of atmospheric conditions. The two possible explanations are due to errors in the absolute 
calibration coefficients used for the calculation of the brightness temperature at the satellite level and the 
uncertainty on the input spectral response of TM 6 sensor. 
4.3 Evaluation of ground level brightness temperatures estimated by new method 
The comparison of different Split-Window methods has shown that two of them can be considered as the best 
compared with the field measurements: RAL93 and Li93. But from the zoomed graph of Figure 5, we can see 
that both of the two methods produce up to about 1.5 °C errors. This is because that Split-Window methods are 
accurate only when the atmospheric conditions corresponding to the image acquired are not very far from the 
average atmospheric conditions. The results obtained by the proposed new method (Figure 5) show that 
significant improvement are achieved compared to the best Split-Window methods: the largest difference 
between the estimated temperatures and the field measurements is reduced from 1.5°C to 0.5°C. We must note 
that the targets considered in our studies have negligible spectral difference on emissivity. In other cases, spectral 
variation of emissivity in the two considered channels will have to be taken into account. 
Figure 5 Differences between the brightness temperatures measured in the field and those estimated by the new 
method compared with the two best Split-Window methods 
799