A comparative study of algorithms for estimating land surface temperature from AVHRR Data

Abstract

A knowledge of land surface temperature (LST) is strongly required for many applications, notably agrometeorology, climate, and environmental studies. Satellite remote sensing in the infrared provides an interesting alternative for the global and continuous measurements of this parameter. Two main problems arise in the use of remote sensing data for evaluation of LST. These are the atmospheric effect and the land emissivity effect. Different algorithms have been developed to correct these two effects. In this work we present the results of a comparative test of LST algorithms. We have used AVHRR images corresponding to a flat homogeneous region characterized by the presence of natural grassland with patches of bare soil. The images cover different seasons, thus allowing for changes in surface emissivity due to changes in vegetation cover. The results show that the Ulivieri et al. algorithm and the Price algorithm are statistically indistinguishable, according to the Kolmogorov-Smirnov test. These two algorithms provide LST estimates close to surface temperature used as ground data. We have tested also the algorithm proposed by Kerr et al., which uses a “surrogate” for the emissivity, the NDVI based on the visible and near-infrared radiances measured by the AVHRR firsts two channels. The use of this approach provides results close to the Ulivieri et al. algorithm. Thus suggesting the convenience of this method that does not require previous emissivity estimates. Alternative formulations for the vegetation index has been also tested with the Kerr et al. algorithn, but the best results correspond to the use of NDVI. A study of the error propagation in the different algorithms due to errors in land surface emissivity has been also performed. The results of this study indicated that the minimum error propagation corresponds to the Ulivieri et al. algorithm while the Price algorithm presents the maximum one. The error propagation shows a thermal dependence for all the tested algorithms except for the Ulivieri et al. algorithm. This thermal dependence is very important for Prata and Platt algorithm.