A transferable satellite-derived bathymetry model without site-specific calibration for Case-1 optically shallow waters using Sentinel-2/MSI imagery

Accurate bathymetry in optically shallow waters (OSWs), particularly across coral reef environments, is essential for coastal ecosystem management and benthic habitat monitoring. Satellite-derived bathymetry (SDB) offers a scalable alternative, but reliance on in situ depth measurements constrains its application in data-sparse OSW regions, while the coupling between water-column and benthic signals limits retrieval accuracy. Here, we developed OSW-SDB, a simulation-based SDB model for Case-1 OSWs that is not tailored to a specific region and accounts for variations in benthic reflectance. Using a newly developed benthic reflectance library comprising 2,100 spectra, we generated 200,000 pairs of water depth and remote sensing reflectance (Rrs) by radiative transfer simulations spanning diverse water-column optical properties and benthic reflectance conditions. An XGBoost model was trained based on the simulated dataset using SHAP-selected multispectral features, with 6,711 matched ICESat-2/ATL24 observations additionally used for hyperparameter optimization. OSW-SDB achieved coefficient of determination (R2) values above 0.96 and root mean square error (RMSE) values below 1 m on the simulated training and testing datasets. Two independent simulation-based validations using benthic spectra excluded from model development yielded R2 values of 0.84 and 0.94, and RMSEs of 2.26 and 1.35 m, respectively. Median absolute percentage errors (MdAPEs) were below 15% across all simulated datasets. This model was applied to Sentinel-2/MSI imagery after OSW extraction, producing spatially coherent 10-m bathymetry maps across seven OSW regions. Independent validation using 13,296 ATL24 depths covering 20 reef sites yielded an R2 of 0.80, an RMSE of 0.75 m, and an MdAPE of 26.13%. Additional evaluations using airborne LiDAR and in situ SeaBASS observations, together with comparisons against two existing SDB models, demonstrated stable performance across geographically diverse OSWs. The model is designed for Case-1 OSWs shallower than 20 m, with better performance below 10 m, consistent with the depth range of benthic habitat mapping. OSW-SDB enables practical and transferable bathymetric mapping at 10-m resolution across diverse Case-1 OSWs using only Sentinel-2/MSI imagery, without requiring site-specific bathymetric or benthic priors during application.