Does Hyperspectral Imagery Improve Satellite-Derived Bathymetry? A Case Study from a Posidonia oceanica-Dominated Mediterranean Region

Coastal bathymetric mapping is essential for marine conservation, navigation, and environmental management. Satellite-derived bathymetry (SDB) is a cost-effective solution to mapping bathymetry over large shallow areas. However, traditional multispectral instruments can produce poor depth estimates for several reasons, including image noise, atmospheric interference, waves and white caps, and where the seafloor-reflected signal is weak, e.g., in areas with deep water or a low-albedo seafloor. This study investigates the potential of PRISMA hyperspectral imagery to improve SDB performance. Through an iterative process, hyperspectral bands were added to a base Random Forest model, and model performance was assessed across different water pixel classes, including bright shallow substrates, seagrass, and deep water. The model’s performance was then compared to that of multispectral Landsat 8 imagery. The results demonstrated that adding hyperspectral bands to the base model improved bathymetric accuracy, particularly in deeper waters (25 m–30 m), where Mean Absolute Error decreased by 2.51 m from a 3-band to a 24-band model. However, the best-performing model was achieved using Landsat 8, resulting in a lower Mean Absolute Error (1.88 m) than the optimized 24-band PRISMA model (2.01 m). Our findings suggest that although additional hyperspectral bands can improve bathymetry estimation, multispectral imagery may still be more effective for general coastal bathymetry mapping despite its lower spectral resolution.