Multi-temporal MODIS and Landsat reflectance fusion method based on super-resolution reconstruction

Abstract

A new Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat reflectance fusion method is proposed based on the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM) and super-resolution reconstruction, which fuse observed MODIS and Landsat images to produce a Landsat synthetic reflectance image at the prediction date. Super-resolution reconstruction via sparse representation is first applied to enhance the resolution of a MODIS image. The results show that this operation can enhance the spatial details of the original MODIS image and can improve the prediction accuracy of the STARFM algorithm. On the other hand, considering the problem of "temporal smoothing" attributed to large differences between two input pairs of MODIS and Landsat images, this method adds a patch-based selection strategy to the original STARFM algorithm. This strategy constrains each prediction of STARFM to use only one pair of MODIS and Landsat images at a base date. The optimal prediction of each patch is then selected from two images, which are predicted by two input pairs of MODIS and Landsat images. The results show that the proposed method outperforms the original STARFM algorithm in terms of prediction accuracy.