Land surface temperature (LST) plays a crucial role in the energy and water cycles of the Earth's climate system. The uncertainty of LST retrieval from satellites is a fundamental and long-standing issue, especially in plateau areas (such as the Tibetan Plateau (TP)), due to its high altitude, unique hydrometeorological conditions, and complex underlying surfaces. To improve the accuracy of LST retrieval over the TP, different methods, including the single channel (SC) algorithm, the split-window (SW) algorithm, and four machine learning (ML) models, were used to retrieve the LST based on SLSTR data in this study. The validation results indicated that the RMSEs of the LSTs retrieved by the SC and SW algorithms were 3.48 and 2.64 K, respectively, which shows better performance than the official SLSTR LST products (5.23 K). In addition, the random forest (RF) model has the highest accuracy among the four ML models, with an RMSE of 3.26 K. By comparing the performance of various methods, the SW algorithm is more stable and reliable for LST retrieval over the TP. In addition, the accurate spatiotemporal distribution of the LST based on the SW algorithm was also analyzed, which would benefit the understanding of the physical processes of energy and water cycles over the TP.
IndexTerms-Land surface temperature (LST), moderate resolution imaging spectroradiometer (MODIS)/Terra, sea and land surface temperature radiometer (SLSTR) data, Tibetan Plateau (TP). I. INTRODUCTION and surface temperature (LST) is a crucial parameter for water and energy flux exchange between the land surface and atmosphere [1-3]. It has been widely used to estimate evapotranspiration (ET) [4] and the urban heat island effect [5] and to monitor climate change [6]. The Tibetan