Supervised segmentation algorithm for coastal dunes delimitation and classification

Coastal dunes systems play an important role as a defensive barrier against erosive and flood processes caused by storms and sea level rise. Delimiting coastal dunes is important for management and analysing local dynamics. Although, it is a complex task due to their inherent variability. This process can be optimized using remote sensing methods, showing to be essential, especially when dealing with extensive and complex systems or when analysing spatio-temporal changes.This contribution shows a preliminary approach to the automatic segmentation of coastal dunes using the Multiresolution Segmentation (MRS) algorithm. The case of study is located in the Trafalgar Cape (Cádiz), in the southwest of the Iberian Peninsula. This area, that is part of the La Breña and Marismas de Barbate Natural Park, is composed of a combination of cliffs, beaches, dunes, and marshes, creating a diverse ecosystem. Dune vegetation, dominated by species such as stone pine, mastic, coastal juniper, and palmetto, is adapted to the climatic conditions, playing an important role in stabilizing the dunes and soils.MRS was applied to different data sets i) RGB and NIR reflectances from Sentinel-2 L2 2017 composites at 10 m spatial resolution, ii) NDVI for the same year, and iii) a Digital Terrain Model (MDT). Different combinations of the algorithm hyper-parameters: “image layer weights”, “Scale Parameter”, shape/colour (0-0.9), and compactness/smoothness (0-0.9) were evaluated using the algorithm ESP2.The proportion of different land cover categories in this area (built-up, cropland, barren/sparse vegetation, trees, grassland, open water, herbaceous wetland and shrubland), was compute for every segment, considering the ESA WorldCover map at 10 m spatial resolution. The segments with the highest dune cover were compared to those obtained from a manual interpretation of very high resolution digital orthoimage.