This paper is to evaluate and compare the most used atmospheric correction methods for satellite thermal data 
(Split Window methods and LOWTRAN radiative transfer code) and to propose a new method to improve the 
atmospheric correction quality for 2 spectral channels' data. The studies are based on the high precision ground 
brightness temperature measurements performed with a well-calibrated thermal camera and coincident 
radiosounding data. Three satellite data were considered: ERS1-ATSR, NOAA11-AVHRR and LANDSAT5- 
TM. The results show that two of the ten tested Split Window methods provide satisfactory correspondence to 
ground level measurements. The use of LOWTRAN 7 code has not given expected result with standart 
radiosounding data, the difference between the ground measurements and the satellite data derived ground level 
temperature remain about 2°C. The proposed new method is based on the LOWTRAN 7 transfer code and uses 
the two channels’ measurements as input data. The results shown that the new method can improve significantly 
the atmospheric correction precision for 2 channels' thermal measurements. The largest residual error is reduced 
from 1.5°C (Split-Window methods) to 0.5°C. 
1. INTRODUCTION 
Satellite thermal measurements are very important in applications such as energy balance and crop water stress 
studies. These measurements are strongly affected by atmospheric effects, particularly by water vapour 
absorption, and a correction is absolutely necessary for deriving surface thermal characteristics. Many 
atmospheric correction methods are available (radiative transfer codes, simple formulations, Split Window 
methods, etc.) But they are generally developed by theoretical or simulation analysis, and few comparisons and 
validations have been achieved with accurate field ground truth measurements. This has complicated the task of 
the users for making the suitable choice. 
Otherwise, Split Window coefficients, generated by using a large number of atmospheric profiles, can give 
satisfactory correction only when atmospheric conditions are not far from the average condition. In the other 
cases, some large correction error can exist. However, the difference between the measurements in two spectral 
channels used for the Split Window correction can provide some useful information about the real atmospheric 
condition, it will be interesting to use this information to improve the atmospheric correction quality. 
In this paper, we have used accurate field measurements performed on the "Crau" test site with the following 
objectives: (1) test the most used Split-Window methods; (2) evaluate the correction accuracy performed by 
Lowtran7 code with special and standard radiosounding data; and (3) try to develop a more accurate correction 
method. 
2 DATA ACQUISITION 
The experiment has been conducted in the summer 1992 on the Crau test site . The data of three satellites were 
used: ERS1-ATSR, NOAA11-AVHRR and LANDSAT5-TM. We have focused our field work on accurate 
brightness temperature measurements with a calibrated thermal camera in the spectral band of the sensors and on 
atmospheric radiosoundings at the time of satellite overpass.