Deep Learning-Based Spatiotemporal Fusion Approach for Producing High-Resolution NDVI Time-Series Datasets

摘要

The availability of concurrently high spatiotemporal resolution remote sensing data is highlydesirable as they represent a key element for effective monitoring in various environmentalapplications. However, due to the tradeoff between the spatial resolution and acquisitionfrequency of current satellites, such data are still lacking. Many studies have been undertakentrying to overcome these problems; however, a couple of long-standing limitationsremain, including accommodating abrupt temporal changes, dealing with complex and heterogeneouslandscapes, and integrating other satellite datasets as well. Accordingly, thispaper proposes a deep learning spatiotemporal data fusion approach based on Very DeepSuper-Resolution (VDSR) to fuse the NDVI retrievals from Sentinel-2 and Landsat 8 images.The performances of VDSR are analyzed in comparison with those of two other classicalmethods, the enhanced spatial and temporal adaptive reflectance fusion model (ESTARFM)and the flexible spatiotemporal data fusion (FSDAF) method. The results obtained indicatethat VDSR outperforms other data fusion algorithms as it generated the least blurred imagesand the most accurate predictions of synthetic NDVI values, particularly in areas with heterogeneouslandscapes and abrupt land-cover changes. The proposed algorithm has broadprospects to improve near-real-time agricultural monitoring purposes and derivation of cropstatus conditions in the field-scale.