Reconstructing daily clear-sky land surface temperature for cloudy regions from MODIS data

Land surface temperature (LST) is a critical parameter in environmental studies and resource management. The MODIS LST data product has been widely used in various studies, such as drought monitoring, evapotranspiration mapping, soil moisture estimation and forest fire detection. However, cloud contamination affects thermal band observations and will lead to inconsistent LST results. In this study, we present a new Remotely Sensed DAily land Surface Temperature reconstruction (RSDAST) model that recovers clear sky LST for pixels covered by cloud using only clear-sky neighboring pixels from nearby dates. The reconstructed LST was validated using the original LST pixels. Model shows high accuracy for reconstructing one masked pixel with R2 of 0.995, bias of −0.02 K and RMSE of 0.51 K. Extended spatial reconstruction results show a better accuracy for flat areas with R2 of 0.72‒0.89, bias of −0.02-0.21 K, and RMSE of 0.92-1.16 K, and for mountain areas with R2 of 0.81-0.89, bias of −0.35-−1.52 K, and RMSE of 1.42‒2.24 K. The reconstructed areas show spatial and temporal patterns that are consistent with the clear neighbor areas. In the reconstructed LST and NDVI triangle feature space which is controlled by soil moisture, LST values distributed reasonably and correspond well to the real soil moisture conditions. Our approach shows great potential for reconstructing clear sky LST under cloudy conditions and provides consistent daily LST which are critical for daily drought monitoring.