The Utility of Simpler Spatial Disaggregation Models for Retrieving Land Surface Temperature at High Spatiotemporal Resolutions

Abstract

The thermal infrared (TIR) sensors aboard geostationary satellites provide multi-temporal land surface temperature (LST) observations for characterizing the diurnal temperature cycle (DTC) at a coarse spatial resolution. This study aims to test if simpler spatial disaggregation models developed to improve the spatial resolution of LST data can be applied across the temporal domain to obtain LST at relatively higher spatial resolution and to capture the DTC. LST from geostationary satellites Kalpana-1 and INSAT-3D were downscaled from 8- and 5-km spatial resolutions respectively to 1 km resolution using Moderate Resolution Imaging Spectroradiometer (MODIS) data acquired during its daytime overpass. Two disaggregation models, DisTrad and NL-DisTrad and a data fusion model spatiotemporal integrated fusion model (STITFM) were selected for the purpose. The downscaled LSTs were compared against the MODIS LST observations from Terra nighttime and Aqua daytime overpasses in addition to in situ data at four sites in India. The results indicated that the DisTrad model (RMSE: 3.61 K, R²: 0.95) was able to capture the spatiotemporal patterns with an accuracy comparable to that of the STITFM model (RMSE: 3.22 K, R²: 0.96). However, the NL-DisTrad performed inconsistently, leading to higher errors.