Normalizing land surface temperature data for elevation and illumination effects in mountainous areas: A case study using ASTER data over a steep-sided valley in Morocco

Abstract

The remotely sensed land surface temperature (LST) is a key parameter to monitor surface energy and water fluxes but the strong impact of topography on LST has limited its use to mostly flat areas. To fill the gap, this study proposes a physically-based method to normalize LST data for topographic - namely illumination and elevation - effects over mountainous areas. Both topographic effects are first quantified by inverting a dual-source soil/vegetation energy balance (EB) model forced by 1) the instantaneous solar radiation simulated by a 3D radiative transfer model named DART (Discrete Anisotropic Radiative Transfer) that uses a digital elevation model (DEM), 2) a satellite-derived vegetation index, and 3) local meteorological (air temperature, air relative humidity and wind speed) data available at a given location. The satellite LST is then normalized for topography by simulating the LST using both pixel- and image-scale DART solar radiation and elevation data. The approach is tested on three ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) overpass dates over a steep-sided 6km by 6km area in the Atlas Mountain in Morocco. The mean correlation coefficient and root mean square difference (RMSD) between EB-simulated and ASTER LST is 0.80 and 3°C, respectively. Moreover, the EB-based method is found to be more accurate than a more classical approach based on a multi-linear regression with DART solar radiation and elevation data. The EB-simulated LST is also evaluated against an extensive ground dataset of 135 autonomous 1-cm depth temperature sensors deployed over the study area. While the mean RMSD between 90m resolution ASTER LST and localized ibutton measurements is 6.1°C, the RMSD between EB-simulated LST and ibutton soil temperature is 5.4 and 5.3°C for a DEM at 90m and 8m resolution, respectively. The proposed topographic normalization is self-calibrated from (LST, DEM, vegetation index and in situ meteorological data) data available over large extents. As a significant perspective this approach opens the path to using normalized LST as input to evapotranspiration retrieval methods based on LST.