Abstract
In the Wadden Sea, intertidal mussel beds of the blue mussel ( Mytilus edulis ) and the Pacific oyster ( Magallana gigas ) form distinct epibenthic communities and local hotspots of high biomass and biodiversity. To detect and evaluate natural and anthropogenic processes, a ground-based monitoring program started over 25 years ago in the German Wadden Sea. In this study, we describe the potential of drones and machine learning approaches for a remote sensing-based integration into an existing monitoring program of intertidal mussel beds. A fixed wing drone was used to cover an area of up to 39ha in a single flight, with an overall time saving potential of 50%. Applying machine learning approaches, a detailed extraction of the intertidal blue mussel bed coverage with an overall accuracy (OA) up to 95.6% was reached, applying a Support Vector Machine (SVM). The application of a multispectral sensor improved the classification performance. Compared to ground-based monitoring, the drone-based method provided significantly more information on the area extension, coverage, and associated algae of the mussel beds. The results show that drones can provide a non-invasive way to survey large and difficult to access areas providing detailed maps of mussel beds and their internal structures.