The management of small Pacific coastal territories has become a crucial issue; these are insular units that often display a
high level of biodiversity in a context of changing climate and
sea-level rise. In order to preserve and protect
populations, infrastructure and living resources, there is a need to understand inland processes that may influence the
behaviour of coastal systems and, more particularly, active erosion zones. Cartography of stripped surfaces by remote
sensing has become routine and we propose here the first step of a method that aims to monitor erosion features using an
automated process. Managing catastrophic erosion and/or landslides needs high frequency image acquisition so as to
optimize hazard prevention.
On the basis of a single remote sensing map, we propose a generic method for automatically assigning expert-designed
labels to erosion areas. Our automated process follows three steps: first, we use classical algorithms to detect stripped
zones; second, we assign a label to each extracted zone using domain knowledge. Finally, as a post-processing phase,
detected and labelled erosion areas are checked by experts. This method has been validated in an erosion-sensitive area
of south-eastern New Caledonia.