A novel approach to produce NDVI image series with enhanced spatial properties

Abstract

A novel multi-step method is presented to improve the spatial properties of MODIS NDVI data series based on one or few single-date higher spatial resolution (HR) images. This method does not rely on the classification of the HR imagery, which is often inadequate in characterizing all main vegetation types that are present in the observed area. An unmixing strategy is instead applied to identify these vegetation types from the low spatial resolution (LR) MODIS imagery, which offers a more effective description of seasonal NDVI evolutions. In particular, an annual multitemporal MODIS NDVI data series is preliminarily decomposed by an automatic technique, which produces abundance images representative of the main vegetation types. These images are then used to extract spatially variable NDVI endmembers. Next, a statistical method is applied to improve the spatial features of the abundance images based on these endmembers and the available HR NDVI imagery. The final recombination of the spatially enhanced abundance images and NDVI endmbemers allows the production of synthetic imagery, which maintains the temporal information of the MODIS NDVI data and most spatial properties of the HR images. The new method is preliminarily tested using an annual MODIS NDVI data series and five Landsat 8 OLI images taken in a study area of Tuscany (Central Italy). The results obtained support the potential of the method and indicate some possibilities for future methodological advancement.