Résumé
This study introduces a novel pixel-based satellite mapping approach for classifying coral island seabed. The model combines a pixel-based approach and a segmentation technique, to smooth the predictions into coherent objects. The model is applied to Maupiti Island (French Polynesia) and compared with an expert-based mapping mostly based on the Reef Cover classification. Results demonstrate high accuracy, ranging between 87% and 90% for various spatial resolutions. The developed tool is open-source and flexible, allowing users to retrain it for different classification schemes and environments. The study highlights the potential of automated satellite mapping for monitoring coral reef ecosystems and supporting conservation efforts.