790 
has been constructed for the Barrax test site (10 km x 10 km), in order to validate the method for mapping 
emissivity described in section 2.2.1. To do this, we have applied the method to a Landsat-TM image and 
compared the results with the digitized map. The comparison shows a standard deviation in e better than 0.005, 
which is sufficient for most LST applications. It indicates a good performance of the method and encourages for 
applying it at the NOAA-AVHRR spatial resolution. This has been done for a 100 km x 100 km area using a 
NOAA-11 image on June 23, 1991. The results indicate a large homogeneity in at the NOAA pixel scale. The 
method for determining the spectral variation Ae has been also applied to the same AVHRR data. Coincident 
radiosounding registered by the Spanish Meteorological Service (INM), and the University of Karlsruhe in 
different sites inside the study area have been processed in the LOWTRAN 7 computer code (Kneizys et al„ 
1988) for radiative transfer calculations. From the different radiosonde data, the vertical water vapor content 
ranges between 1.3 and 1.5 g/cm 2 , showing certain atmospheric homogeneity. We have calculated the brightness 
temperatures T;* from equation (1), and the coefficients b; according to Coll et al. (1993b). Then the spectral 
difference emissivity, Ae, have been determined from equation (9), which also requires the retrieved value of e. 
The Ae image is shown in Figure (4a). As a final result we have applied the split-window algorithm (equation 5) 
to obtain a LST image of the study area, which is shown in Figure (4b). A value (3=120 K has been used 
according to the total water vapor content estimated from the radiosoundings. 
3.2. HAPEX-Sahel Project 
We have applied the methodology developed to the whole HAPEX-Sahel area (1° x 1°) using AVHRR images 
acquired in 1991. We have used field emissivity measurements performed during the 1992 Intensive Observation 
Period (IOP) by the University of Strasbourg, in combination with NDVI images to produce e maps of the 
whole study area. An example of this is shown in Figure (5a). Besides, laboratory spectral measurements have 
been made by the University of Strasbourg for different soil types present in the area, showing a small 
emissivity difference (with an order of magnitude of 10 4 ), which is similar to the magnitude of the spectral 
difference for vegetation. This question will be investigated by applying the Ae method (equation 9) to the 1992 
AVHRR data and coincident radiosoundings adquired during the IOP. 
In tropical atmospheres, the atmospheric attenuation is large and consequently the efficiency of 
the split-window technique is reduced. However, the larger reflection of the downwelling atmospheric radiance 
contributes to the decreasing of the emissivity effect. Bcrguc and Bcsscmoulin (1993) report total column water 
vapor content for the CNRM radiosoundings during the IOP, showing values of W about 4 g/cm 2 . For W=4 we 
have obtained a=45 K and (3=30 K. This fact, combined with the expected small importance of the emissivity 
spectral variation suggests that the term (3Ae=0. Thus, wc have not included this term in the LST algorithm. 
According to equation (5) we have constructed a LST image, which is shown in Figure (5b). 
4 - CONCLUSION 
We have proposed an operational method for producing emissivity and land surface temperature images, by using 
a modified split-window equation wich takes into account the atmospheric variability at global scale, and the 
surface emissivity effects through the mean emissivity, e, and the channel emissivity difference, Ae. The surface 
emissivity must be known at the NOAA-AVHRR spatial and spectral resolution with sufficient accuracy, for 
which we have developed methods for retrieveing and mapping both e and Ae using AVHRR data. We have used 
the operational model in two different areas in Spain and Niger, for showing that the method can be easily 
applicable. An important task is the validation of the LST retrievals in various environments (different 
atmospheric and surface types), which is being currently carried out by the authors. For doing that, extensive 
ground truth and satellite data are necessary that can be obtained from past and future research projects. 
ACKNOWLEDGEMENTS 
The authors wish to express their gratitude to the Environment Programme of the Comission of the European 
Communities (Contracts No. CT90-0030, CT91-(X)33, and CT91-(X)35) for financial support. We also thank Dr. 
Y. Kerr and the HSIS Team for providing the AVHRR images of 1991, and the Air Force Geophysics 
Laboratory (Massachusetts, USA) for supplying the LOWTRAN 7 computer code.