LST Reconstruction on the Remote-Sensed LST Products Under Clouds

A surface energy balance-based neighboring pixel algorithm was implemented to enhance the MODIS Land Surface Temperature (LST) product. This algorithm focused on the reconstruction of cloud-contaminated pixels in the MODIS LST product over the entire year of 2012 in the Heihe River Basin. Improved MODIS LST products were generated based on the reconstruction results and the original MODIS LST data with clear-sky observations. During the algorithm implementation, regionalization of parameters was conducted using long-term ground observation data, land cover data, soil type data, and other relevant data for the Heihe River Basin. The reconstructed under-cloud pixel data were compared and validated against on-site measured LST data and the original MODIS LST data with good quality. The results indicated that the reconstructed under-cloud LST data exhibited a similar level of bias to the good-quality data in the original MODIS LST, albeit with a higher root mean square error (RMSE). However, the spatial and temporal continuity of the improved MODIS LST product was significantly enhanced after the reconstruction of under-cloud pixels. Therefore, the improved MODIS LST product can be directly applied to studies with less stringent accuracy requirements for LST, greatly increasing the usability of the original MODIS LST data.