Artificial Intelligence classification of multispectral and multitemporal aerial images for the detection of polychaeta reefs (Sabellaria alveolata)

Dorothée James¹,Antoine Collin¹, Laurent Barille², Stanislas Dubois³, Pierre-Olivier Liabot³, Simon Oiry², Agathe Bouet^1,4, Eric Feunteun^5,1

¹Ecole Pratique des Hautes Etudes, PSL, France - ²Institut Des Substances et Organismes de la Mer ISOMer, France - ³Ifremer, France - ⁴Association Planete-Mer, France - ⁵UMR BOREA, France

Temperate polychaeta reefs, formed by the worm Sabellaria alveolata, are sandy biogenic habitats essential for marine biodiversity. They play an important role in stabilising sediments, reducing coastal erosion, filtering water and providing a refuge for many species. However, these fragile reefs are under threat from pressures such as trampling, fishing, and climate change. Protecting these ecosystems requires measures such as raising awareness, regulating access, establishing marine protected areas, and ecological restoration.
Satellite monitoring of polychaeta reefs is an innovative approach to large-scale monitoring. Remote sensing technologies can be used to map the extent, dynamics and health of these habitats. However, their location in intertidal zones necessitates periods of low tide for optimal detection. In France, Ortholittorale® is a national coastal mapping program coordinated by IGN and Cerema. This program provides very high resolution (0.50 m) multispectral ortho imageries taken at low tide and freely accessible. Two campaigns have already been carried out in 2000, 2011/2014, and 2020, depending on the site, while a third is currently being acquired. For the reefs in the Bay of Mont St Michel, the aerial images come from an aerial campaign conducted in 2020 by the Coastal Monitoring Network (ROLNHDF).
This study focuses on the spatial evolution over two decades of three temperate polychaeta reefs: La Barbâtre (Noirmoutier island), Champeaux (Bay of Mont-Saint-Michel) and Sainte-Anne, the largest reef in Europe, also located in the Bay of Mont-Saint-Michel. A classification algorithm based on artificial intelligence was used to map these reefs. Six land use classes were identified: Sabellaria alveolata, sediment, brown algae, green algae, water and oyster. Spectral predictors and their derivatives, such as NDVI, were tested to optimize the classification. Performance was assessed using confusion matrices to analyze the contribution of each predictor.
The results provided valuable information for the long-term management and conservation of temperate polychaeta reefs, allowing their spatial evolution to be mapped.