Comparative study of different methods for bathymetric mapping based on multi-source remote sensing data

Satellite-derived bathymetry (SDB) technology for mapping shallow water around islands and reefs plays a crucial role in the management of coastal zone resources. The accuracy of SDB depends strongly on the selection of satellite remote sensing data and bathymetric models. This study evaluated the performance of five SDB inversion methods, including the log-transformed linear model, log-transformed band-ratio model, k-nearest neighbor regression, random forest, and support vector regression, using Landsat-8, Sentinel-2, and WorldView-2 imagery together with measured bathymetric data from Lingyang Reef. We compared and analyzed the bathymetric results of remote sensing data with different spatial resolutions in empirical models and machine learning models. The results show that the bathymetric results derived from different data across all models were able to reflect the spatial variation trends of water depth. The machine learning models achieved higher overall accuracy than the empirical models, with the random forest performing best across different data conditions. Sentinel-2 data exhibited greater stability in empirical model fitting, while WorldView-2 imagery achieved higher bathymetric accuracy when used with machine learning models. This study provides an empirical evidence for the appropriate selection of multispectral imagery with varying spatial resolutions and corresponding shallow water bathymetry models, contributing to enhanced accuracy and broader applicability of SDB.