Long-term accuracy assessment of land surface temperatures derived from the Advanced Along-Track Scanning Radiometer

Abstract

The accuracy of land surface temperatures (LSTs) derived from the Advanced Along-Track Scanning Radiometer (AATSR) was assessed in a test site in Valencia, Spain from 2002 to 2008. AATSR LSTs were directly compared with concurrent ground measurements over homogeneous, full-vegetated rice fields in the conventional temperature-based (T-based) method. We also applied the new radiance-based (R-based) method over bare soil and water surfaces, where ground LST measurements were not available. In the R-based method, ground LSTs are simulated from AATSR brightness temperatures in the 11μm band and radiative transfer simulations using surface emissivity data and atmospheric water vapor and temperature profiles. The accuracy of the R-based ground LSTs depends on how well the profiles used in simulations represent the actual atmosphere at the time of AATSR observations. This can be checked with the difference δ(T11-T12) between the actual AATSR and the profile-based simulated difference in the 11 and 12μm brightness temperatures (T11 and T12, respectively). We found that for −0.6K<δ(T11-T12)<0.6K, the R-based LSTs were accurate within ±1.0K and can be used for LST validation. For the data analyzed here, the AATSR operational algorithm overestimated the ground LST by 2 to 5K, showing that the auxiliary data utilized within the retrieval scheme (biome classification and fractional vegetation cover maps at 0.5°×0.5° resolution) should be improved and provided at the same spatial resolution as the AATSR data (1km2). When the AATSR algorithm was optimized with biome and fractional vegetation cover selected according to the nature of each surface, LST errors showed negligible average biases and rmse=±0.5K for full vegetation and water, and ±1.1K for bare soil. Furthermore, we checked an alternative algorithm explicitly dependent on emissivity, which provided accurate LSTs for all the surfaces studied, with small biases, rmse from ±0.4 to ±0.6K and most LST errors within ±1.0K. The algorithm requires monthly emissivity maps at 1km2, which can be derived from classification and fractional vegetation cover estimated from optical AATSR data. The results of this paper show the high LST accuracy achievable with AATSR data in ideal conditions. While it is necessary to establish and maintain highly homogeneous T-based validation sites, the R-based method provides an alternative for the semi-operational, long-term evaluation of LST products at global scale, since it is applicable over surfaces with varied LST and atmospheric regimes where ground LST measurements are not feasible.