Comparing the applicability of five typical spatio-temporal information fusion algorithms based on remote sensing data in vegetation index reconstruction of wetland areas

This study aims to explore the applicability of various spatio-temporal information fusion algorithms based on remote sensing data to wetland areas characterized by frequent land-water conversion and diverse surface features. With the Poyang Lake sample area as the study area, this study examined five typical spatio-temporal information fusion algorithms (STARFM, ESTARFM, FSDAF, Fit-FC, and STNLFFM). Considering the differences in surface features among different periods, Landsat and MODIS remote sensing data were selected to conduct image fusion experiments for normalized difference vegetation indices (NDVIs) during low- and normal-water periods. Moreover, the accuracy of these algorithms was evaluated in spatial and spectral dimensions. The results of this study are as follows: ① In the case of only one pair of coarse- and fine-resolution images as input, the FSDAF exhibited the optimal fusion prediction effect for the low-water period, with an overall error of 0.433 5, whereas the STNLFFM manifested the optimal fusion prediction effect for the normal-water period, with an overall error of 0.514 7; ② In the case of two pairs of coarse- and fine-resolution images of low- and normal-water periods as input, the ESTARFM demonstrated the optimal fusion prediction effect, with an overall error of 0.467 0; ③ The applicability of different algorithms to a wetland area is associated with the proportion of water bodies in the study area. The STNLFFM displayed the optimal fusion prediction effect for water bodies.