Generating Spatiotemporal Seamless Data of Clear-Sky Land Surface Temperature Using Synthetic Aperture Radar, Digital Elevation Mode, and Machine Learning over Vegetation Areas

Abstract

The continuous retrieval of clear-sky land surface temperature (LST) holds paramount importance in monitoring vegetation temperature and assessing water stress conditions. Nonetheless, the extensive cloud cover results in a widespread lack of LST data, posing challenges in accurately forecasting LST in regions characterized by diverse vegetation types and complex terrains. Therefore, this paper proposes a synthetic aperture radar (SAR)- and digital elevation model (DEM)-integrated LST reconstruction model (SDX-LST) to generate realistic and high-spatial-resolution (30 m) clear-sky LST data. To assess the practicality and robustness of the SDX-LST model, this paper selects the study areas of Loess Plateau (LP), Qinghai-Tibet Plateau, Northeast China Plain, Nanling Mountains, and North China Plain in China, Desert Rock, Nevada in America, spanning a wide range of longitude and latitude and having obvious differences in topography, landforms, and vegetation. The analysis of the reconstruction results in different spatial location distributions, vegetation cover types, and multidate and time distributions throughout the year indicate that the SDX-LST model achieves excellent performance and high stability (with a mean absolute error lower than 2K). The SDX-LST predictions demonstrate a commendable level of consistency with the ERA5-hourly product and in situ data. We conclude that the integration of SAR and DEM within the SDX-LST model enables precise predictions of LST for various vegetation types and intricate terrains. The study quantitatively analyzes the effects of SAR and DEM on LST and introduces novel insights for exploring SAR-based reconstruction of LST.