Reconstructing all-weather daytime land surface temperature based on energy balance considering the cloud radiative effect

Land surface temperature (LST) has been used in many applications as its strong relationships with land surface processes. However, the greatest limitation of the use of LST is the missing data caused by cloud contamination and weather conditions. In this study, we first used the XGBoost method to describe complex relationships of LST with surface characteristics from clear-sky pixels, and applied the model to retrieve hypothetical clear-sky LST under cloudy sky. Secondly, cloud radiative effect (CRE) on the LST was calculated based on energy balance using the reanalysis data. The models were applied to reconstruct the all-weather LST over the Tibetan Plateau (TP). The spatial patterns of reconstructed LST indicated that our model could produce completely spatial-seamless LST and depict the detailed information. The accuracy of the XGBoost LST was validated against the clear-sky MODIS LST (average R2 = 0.92, MAPE = 0.52, RMSE = 2.32 K). The CRE-EB LST was evaluated using data from six in-situ sites from the TP. The validation results were separated into three conditions: clear sky (RMSE = 3.01 K–3.52 K, R2 = 0.88–0.93, bias = −1.08 K-1.88 K), cloudy sky (RMSE = 3.31 K–4.06 K, R2 = 0.87–0.92, bias = −0.21 K-1.11 K), and overall (RMSE = 3.31 K–3.82 K, R2 = 0.88–0.93, bias = −0.42 K-1.24 K). Compared to existing all-weather LST datasets, the temporal variability of our LST data shows similar seasonal and daily changes, and the CRE-EB LST has advantages in terms of image quality and accuracies under cloudy condition. This study demonstrated the utility of proposed models to reconstruct all-weather LST.