A Two-Step Framework for Generating 0.01°, Hourly, and Gapless Land Surface Temperature

By depicting high-frequency surface thermal dynamics, hourly gapless land surface temperature (LST) data at a moderate spatial scale are crucial for various thermal environment investigations. However, cloud cover and the sensor hardware limitations have constrained the access to such LST data. In this work, a two-step framework is proposed to generate hourly gapless LST, which is made up of two steps: first, a machine learning model is used to reconstruct moderate-resolution imaging spectroradiometer (MODIS) daily (four times from Terra and Aqua) LST; and second, the generated daily gapless LST is then fused with hourly community land model (CLM) simulated LST based on the spatial and temporal nonlocal fusion model. A 0.01°, hourly, gapless LST dataset was generated over the middle and upper sections of the Heihe River Basin in China from 2008 to 2011. Validation was conducted using clear-sky MODIS LST and four sets of all-sky ground-based LST measurements. The results reveal that the daily LST reconstruction model performs well, with a Pearson correlation coefficient (R) of 0.97–0.98 and a root-mean-square error (RMSE) of 3.01–3.6 K in cloudy conditions. Validation using hourly in situ measurements also indicated a high accuracy, with the RMSE between 2.56 and 3.76 K under all-sky conditions. The daily mean LST obtained by averaging the hourly gapless LST resulted in an RMSE of 1.6–1.92 K. Spatial and temporal analysis further demonstrated that the proposed method can accurately characterize the spatial details and temporal dynamics of LST at both daily and hourly scales. Compared with other hourly and daily mean LST products, the generated LST data have better validation accuracy. The proposed method offers a practical and robust approach to produce hourly gapless LST at a moderate spatial scale, which is essential in a variety of regional-scale thermal applications.