Shallow water bathymetry based entirely on Satellite data: a case study in the South China Sea

Abstract Shallow water depth information plays a critical role in coastal and marine applications, and satellite-derived bathymetry (SDB) has emerged as a valuable due to its extensive coverage and capability to generate comprehensive bathymetric maps. This study integrates ICESat-2 photon-counting LiDAR data with WorldView-2 multispectral imagery, employing two empirical models (Band Ratio (BR) model and Linear Band (LB) model) and two machine learning models (Support Vector Regression (SVR) model and Random Forest (RF) model) to assess the accuracy and reliability of water depth estimation in three regions of the South China Sea: Wuzhizhou Island, Qilian Island, and Gongshi Reef. The results show that all four models can effectively achieve high-precision shallow water bathymetry, with the Random Forest (RF) model exhibiting the highest overall accuracy (RMSE = 1.13 m, R² = 0.90). Additionally, although the two machine learning models outperform the empirical models in terms of accuracy and robustness, they exhibit higher sensitivity to environmental variations, particularly in shallow-water areas with reef platforms and whitecaps.