LAUSR

Spatiotemporal fusion of land surface temperature (LST) has a vital significance in studying the temporal and spatial variation of urban heat islands. But most existing LST fusion methods do not consider the highly heterogeneous urban surface and complexity of the spatial layout. In this study, a Comprehensive Flexible Spatiotemporal DAta Fusion (CFSDAF) method was proposed to generate a high spatiotemporal resolution urban LST image, which was an improvement of the Flexible Spatiotemporal DAta Fusion (FSDAF). The CFSDAF first adjusted the differences between coarse-resolution LST and fine-resolution LST. Then, the visible and near-infrared image of a fine resolution was introduced to execute spectral unmixing and to conduct soft classification, which considered the mixed pixel of fine-resolution LST. The inverse distance weighting (IDW) interpolation was used in improving the computational efficiency, and the constrained least square was selected to better distribute the residual. The performance of CFSDAF was compared with the temporal adaptive reflectance fusion model (STARFM) and FSDAF. The results indicate that the predicted images by CFSDAF are better than STARFM and FSDAF from both visual comparison and quantitative assessment in two experiments, and CFSDAF can reserve more spatial details and accurately restore the spatial continuity of urban LST than others. Moreover, CFSDAF has high computational efficiency and can monitor land cover changes the same as FSDAF. Due to the above advantages, the CFSDAF has great potential for studying spatiotemporal changes of LST and UHI in an urban area.