Generating a 30 m Hourly Land Surface Temperatures Based on Spatial Fusion Model and Machine Learning Algorithm

Abstract

Land surface temperature (LST) is a critical parameter for understanding climate change and maintaining hydrological balance across local and global scales. However, existing satellite LST products face trade-offs between spatial and temporal resolutions, making it challenging to provide all-weather LST with high spatiotemporal resolution. In this study, focusing on Chengdu city, a framework combining a spatiotemporal fusion model and machine learning algorithm was proposed and applied to retrieve hourly high spatial resolution LST data from Chinese geostationary weather satellite data and multi-scale polar-orbiting satellite observations. The predicted 30 m hourly LST values were evaluated against in situ LST measurements and Sentinel-3 SLSTR data on 11 August 2019 and 21 April 2022, respectively. The results demonstrate that validation based on the in situ LST, the root mean squared error (RMSE) of the predicted LST using the proposed framework are around 0.89 °C to 1.23 °C. The predicted LST is highly consistent with the Sentinel-3 SLSTR data, and the RMSE varies from 0.95 °C to 1.25 °C. In addition, the proposed framework was applied to Xi’an City, and the final validation results indicate that the method is accurate to within about 1.33 °C. The generated 30 m hourly LST can provide important data with fine spatial resolution for urban thermal environment monitoring.