Downscaling algorithm for AMSR2 land surface temperature over the Tibetan Plateau using random forest

Land surface temperature (LST) is an important parameter in land surface process research, significantly influencing the energy radiation balance and climate system changes. The Tibetan Plateau, a sensitive and ecologically vulnerable region, is particularly susceptible to global change, making high-precision surface temperature data essential for clarifying regional climate change and its impacts. This study utilizes surface temperature observations from meteorological stations, Advanced Microwave Scanning Radiometer (AMSR) passive microwave brightness temperature data, and topographic factors to construct an LST downscaling model based on Random Forest. The model generates daily LST with a resolution of 1 km for the year 2013. Results indicate that the downscaled LST is highly accuracy, with the downscaling inversion accuracy for nighttime LST being higher than that for daytime. The root mean square errors (RMSE) for daytime and nighttime LST are 4.31 K and 1.33 K, respectively. Compared to the MYD11A1 LST product, the RMSE for downscaled daytime and nighttime LST is reduced by 0.92 K and 0.31 K, respectively, with a smaller bias of -1.96 K at night. Furthermore, the model exhibits good stability under different land cover types, latitudes, and slope conditions. This research provides valuable insights for the development and improvement of LST downscaling models and products.