Machine learning approaches for satellite-derived bathymetry in tropical coastal waters: A comparative study from Nha Trang marine protected area, Vietnam

Bathymetry mapping plays a critical role in coastal zone management, marine conservation, and navigation safety. With the increasing availability of high-resolution satellite imagery, such as PlanetScope (3−5 m), remote sensing-based bathymetry retrieval offers a cost-effective and scalable alternative to traditional in-situ surveys. This study explores the capability of PlanetScope imagery to retrieve a wide range of bathymetry (-0.5 − ~ -40 m) in the southern area of the Nha Trang Marine Protected Area (MPA), Vietnam - an ecologically significant and dynamic coastal region. We conduct a comprehensive comparison between traditional approaches, including the Stumpf ratio model and Multiple Linear Regression (MLR), and a suite of advanced machine learning (ML) algorithms, including Random Forest (RF), Support Vector Machine (SVM), Light Gradient Boosting Machine (LGBM), Extreme Gradient Boosting (XGB), CatBoost (CB), and Gradient Boosting (GB). Among these, RF achieved the highest performance with an R2 of 0.85, RMSE of 2.66 m, and MAE of 1.85 m, significantly outperforming the Stumpf model (R2 = 0.29) and MLR (R2 = 0.57). This study represents one of the most extensive model comparisons to date for satellite-derived bathymetry using PlanetScope data, offering a benchmark for future applications in tropical coastal environments. Results underscore the potential of machine learning to advance spatially detailed and accurate bathymetric mapping from space.