干旱区生态环境知识资源中心

This study compares six satellite-retrieved land surface emissivity (LSE) products over gravel plains and sand dunes of the hyper-arid Namib desert in Namibia and validates them with in-situ measurements performed with the ’emissivity box method’. The following products are compared: LSE derived by the Land Surface Analysis - Satellite Application Facility (LSA-SAF) for the Spinning Enhanced Visible and Infrared Imager (SEVIRI) onboard Meteosat Second Generation (MSG), LSE products MOD11A2.C5, MOD11B1.C4.1, and MOD11B1.C5 derived for the Moderate Resolution Imaging Spectroradiometer (MODIS) onboard EOS-Terra, LSE derived with the Temperature Emissivity Separation (TES) algorithm for the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) onboard EOS-Terra, and LSE derived with the TES algorithm for EOS-Terra/MODIS data. The LSA-SAF, MOD11A2.C5, and MOD11B1.C5 algorithms directly or indirectly utilize land cover classification and vegetation cover fraction data with the result that for arid regions their LSE are practically identical to the bare ground emissivities assigned to those classes. Over the gravel plains, mean LSA-SAF, ASTER-TES, and MODTES LSE are about 0.950 in the 11 mu m range, whereas mean MOD11A2.C5 and MOD11B1.C5 are about 1.5% (similar to 1 K) higher. The LSA-SAF algorithm misclassifies the sand dunes as ’open & closed shrubland’, which results in an overestimated mean LSE (0.969). Since MOD11A2.C5 and MOD11B1.C5 utilize a similar classification and similar emissivity library data, their LSE estimates for the sand dunes are also too high (mean of 0.972 and 0.980, respectively). In contrast, the physics-based ASTER-TES and MODTES algorithms estimate mean sand dune LSE as 0.952 and 0.948, respectively. The physics-based MOD11B1.C4.1 algorithm produced noisy LSE estimates with frequent outliers at 5 km resolution: spatial averaging yielded mean LSE of 0.950 and 0.954 for the gravel plains and the sand dunes, respectively. Based on a combined analysis of in-situ LSE and TES retrieved LSE, and also accounting for uncertainty in the fraction of dry grass (only gravel plains), for future work it is recommended to use SEVIRI ch10.8 emissivities of 0.941 +/- 0.004 for the sand dunes and 0.944 +/- 0.015 for the gravel plains, respectively. The results suggest that split window algorithms would benefit significantly from using physically based MODTES LSE. (c) 2012 Elsevier Inc. All rights reserved.