A retrieval method for passive microwave LST integrating radiative transfer-informed land-air environmental conditions

Passive microwave (PMW) remote sensing is a significant data source for generating spatiotemporally continuous land surface temperature (LST). Although empirical methods can retrieve PMW LST that approximates the physical meaning of thermal infrared (TIR)-derived LST, they generally fail to fully consider land-air variables implied by PMW radiative transfer principles. Particularly, the impacts of atmospheric situation have often been overlooked. Here, we propose a machine learning-based retrieval method that integrates a broader range of land-air environmental variables in accordance with PMW radiative transfer principles (PLST-ILA). Validation against pre-extracted MODIS LST shows that the proposed method achieves annual daytime and nighttime root mean square errors (RMSEs) of 3.17 K and 2.21 K, respectively, showing reductions of 9.9 %∼26.1 % relative to the five reference methods. Validation at six in-situ sites indicates that PLST-ILA yields an annual daytime RMSE of 4.21 K and a nighttime RMSE of 4.02 K, reducing the daytime RMSE by 0.33 ∼ 1.68 K and the nighttime RMSE by 0.27 ∼ 1.32 K benchmarked by four advanced methods. The monthly-scale model generally achieves superior performance compared with the seasonal-scale and annual-scale models. Moreover, accounting for topographic and geomorphological characteristics, anthropogenic emissions, atmospheric humidity, and aerosol optical depth may further improve retrieval accuracy. Finally, the proposed method demonstrates the capability to reconstruct PMW observations within orbital gaps. Validation against gap-retained MODIS LST yields annual RMSEs of 3.68 K during the daytime and 2.69 K at night. The results of this study provide a technical reference for PMW LST retrieval and may contribute to optimizing the precision of thermal-related research.