A Hybrid Method for Temporal Normalization of Land Surface Temperature Under All-Sky Conditions

Land surface temperature (LST) is a vital parameter that reflects the land–atmosphere interaction in the land surface energy process. However, due to the temporal effect induced by the wide swath of satellites, differences in local solar time for each pixel observation can result in large differences in LST. Although many studies have been conducted to address this problem, these methods still suffer from spatial discontinuities due to cloud contamination. In this context, this article aims to develop a hybrid method that is combined with surface solar irradiance parameterization and random forest regression. The developed method is suitable for the temporal normalization of LST under all-sky conditions. This method was tested with terra moderate resolution imaging spectroradiometer (MODIS) data and skin temperature in the land component of the fifth generation of European ReAnalysis under clear and cloudy conditions and was validated by three in-situ measurements. An enhancement can be observed with the root mean square error (RMSE) reduction from 2.73 K (3.39 K) to 2.45 K (2.66 K) under clear-sky conditions compared with the original MODIS LST. The RMSE of the normalized LSTs was reduced from 3.60 K (4.14 K) to 2.96 K (3.74 K) under cloudy-sky conditions. From a comparison between the hybrid method and current temporal normalization methods, it is found the former outperforms the latter in terms of spatial completeness and accuracy. These results demonstrate this method has good potential for the temporal normalization of LST under all-sky conditions.