Combination of GEOBIA and data-driven approach for grassland habitat mapping in the Alta Murgia National Park

This study aims to discriminate semi-natural dry grassland habitats (namely: 6210, 6220, 62A0, according to the Annex I of the European Habitat Directive) in the Alta Murgia National Park, in southern Italy. These Mediterranean habitats are often characterized by small and fragmented patches, therefore, multi-season very high spatial resolution satellite images and data-driven Geographic Object-Based Image Analysis (GEOBIA) approach were considered to obtain grassland habitats mapping by an automatic classification process. Different classifiers such as Support Vector Machine (SVM) and Random Forest (RF) were evaluated, and their performance was compared by varying different input feature configurations such as the number of seasonal images used. Pléiades and Worldview-2 satellite images were considered. A dual nomenclature was adopted to consider the set of vegetation mosaics and transitional stages occurring in the field. RF performed better than SVM. Although the F1-scores of the different habitat classes were not greater than 0.75 and further improvements are needed, the results can be considered a satisfying preliminary attempt to automatically reproduce, at fine scale, the fragmentation of grassland habitats on the large area of Alta Murgia National Park. The mapping can be a useful tool for local authorities involved in the periodic monitoring of habitats in protected areas according to the European Habitat Directive and the fine scale can support focused local decision-making process for the conservation of natural ecosystems. • GEOBIA and machine learning applied to multi-season VHR satellite images. • VHR images for fragmented pattern of grasslands habitats in Mediterranean landscape. • Random Forest classifier performs better than Support Vector Machine. • Dual nomenclature for grassland habitats to reproduce mosaic pattern. • VHR grasslands habitats maps as a useful support for local decision-making process.