A New Method to Enhance the Spatial Features of Multitemporal NDVI Image Series

A novel spatiotemporal fusion (STF) method is presented to enhance the spatial features of low-spatialresolution (LR) normalized difference vegetation index (NDVI) image series based on single-date high-spatial-resolution (HR) imagery. The method is particularly suitable for areas where the main vegetation types show asynchronous NDVI evolutions whose spatial distribution cannot be properly characterized by a single-date HR image. In contrast with previous STF methods, the new algorithm identifies these vegetation types by automatically decomposing the LR multitemporal data series, which offers a complete description of major seasonal NDVI evolutions. The new method, named Spatial Enhancer of Vegetation Index image Series (SEVIS), is tested in two Italian study areas using annual MODIS NDVI data sets and some Landsat 8 OLI images taken in different seasons. The performances of SEVIS are analyzed in comparison with those of two other STF methods, the classical Spatial and Temporal Adaptive Fusion Model (STARFM) and the more recent Flexible Spatiotemporal DAta Fusion (FSDAF) algorithm. The results obtained indicate that the three methods perform differently depending mainly on the synchronicity of the NDVI evolutions from the base to the prediction dates. Specifically, SEVIS outperforms the other two methods when the NDVI values evolve differently during the prediction period, i.e., when the base and prediction images are poorly correlated.