Applying fuzzy approaches to map Natura 2000 and EUNIS habitats on coastal dunes from WorldView-3 imagery

Habitat mapping on coastal dunes, which are highly endangered and crucial ecosystems, demands objectivity and repeatability, elements still lacking in the implementation of the Habitats Directive. While remote sensing holds promise, the effectiveness of distinguishing Natura 2000 habitats on coastal dunes from satellite imagery remains untested. Fuzzy approaches to image classification could improve habitat mapping but have not yet been applied to coastal dunes for this purpose. The aim of this work is to compare crisp and fuzzy classification methods using WorldView-3 satellite imagery to map coastal dune habitats in two Parks of Tuscany (Italy). Vegetation data were collected from 244 plots and classified into Natura 2000 habitats using both expert assessment and noise clustering, and into EUNIS habitats through the EUNIS Expert System. Using field data as a reference, we classified WorldView-3 images with a crisp Random Forest method and three fuzzy methods, comparing the results through overall accuracy (OA) and Mantel tests. The highest accuracy (OA = 0.90) was achieved for EUNIS habitats, due to their simpler classification scheme. We observed a great disparity among habitats, with coastal dune scrubs and white dunes generally exhibiting the highest accuracy (maximum = 1.00). Although fuzzy classifications produced lower accuracy than crisp ones (mean OA = 0.40 vs 0.58), they provided a more realistic representation of vegetation patterns (mean R = 0.36 vs 0.32), indicating the fuzzy nature of vegetation in coastal dunes. Despite challenges concerning habitat heterogeneity and spatial resolution of images, the integration of field surveys and satellite imagery allowed to produce a detailed cartography, suitable for monitoring the area, structure and functions of dune habitats, as required by Article 17 of the Habitats Directive.