Pan-European Grassland Mapping Using Seasonal Statistics From Multisensor Image Time Series

Grasslands cover approximately 40% of the Earth's surface. Low-cost tools for inventory, management, and monitoring are needed because of their great expanse, diversity, and the importance for environmental processes. Remote sensing is a useful technique for providing accurate and reliable information for land use planning and large-scale grassland management. In the context of “GIO land” (Copernicus Initial Operations land program), which is currently contracted by the European Environment Agency, a high-resolution grassland layer of 39 European countries is being created with an overall classification accuracy of better than 80%. Since grassland canopy density, chlorophyll status, and ground cover (GC) are highly dynamic throughout the growing season, no unique spectral signature can be used to map grasslands. Therefore, it is necessary to use image time series to characterize the phenological dynamics of grasslands throughout the year in order to discriminate between grasslands and other vegetation with similar spectral responses. This paper describes an operational approach based on a multisensor concept that employs optical multitemporal and multiscale satellite imagery to generate the high-resolution pan-European grassland layer. The approach is based on the supervised decision tree classifier C5.0 in combination with previous image segmentation and seasonal statistics for various vegetation indices (VIs). Results from the grassland classification for Hungary are presented. The accuracy assessment for this classification was carried out using 328 independent sample points derived from a ground-based European field survey program (LUCAS) and current CORINE Land Cover data. The grassland classification approach is explained in detail on the example of Hungary where an overall accuracy of 92.2% has been reached.