A spatiotemporal satellite image fusion model with autoregressive error correction (AREC)

Abstract

To overcome the trade-off between spatial and temporal resolutions of satellite sensors, various spatiotemporal image fusion methods have been developed to generate synthetic imagery with both fine spatial and temporal resolutions. While such methods have achieved different levels of success, they have not considered spatiotemporal autocorrelation in the fusion process. Herein, we propose a novel spatiotemporal model incorporating autoregressive error correction (AREC) for fusing Moderate Resolution Imaging Spectroradiometer (MODIS) and Landsat Enhanced Thematic Mapper Plus (ETM+) images. This AREC model minimizes autoregressive errors when fitting a relationship between coarse- and fine-resolution pixels in an existing MODIS–Landsat image pair (on a base date (Formula presented.)). The derived relationship is then applied to the corresponding pixel in the prediction pair (on a prediction date (Formula presented.)). This method enables the relationship to be optimized and the phenological and land-cover changes to be treated in a unified manner. The AREC model was tested using a simulated data set and two actual data sets from Shenzhen, China, and Saskatchewan, Canada. The model was compared with the spatial and temporal adaptive reflectance fusion model (STARFM) and the flexible spatiotemporal data fusion (FSDAF) method in terms of changed and unchanged regions, and whole data sets, using visual analysis and quantitative indices. The results show that the AREC model can effectively predict phenological and land-cover changes with higher accuracies than other algorithms as a result of AREC.