Spatial-Location Gradient Boosting Machine for High-Resolution Bathymetric Inversion in Optically Complex and Turbid Waters

Accurate bathymetric mapping is vital for maritime safety and ecosystem management. Conventional satellite-derived bathymetry (SDB) models often struggle in turbid waters due to the inherent spatial non-stationarity of underwater topography. We propose the Spatial-Location Gradient Boosting Machine (SL-GBM), a novel inversion framework that integrates spectral signatures with explicit geospatial coordinates. By incorporating longitude, latitude, and their interaction terms as exogenous variables, SL-GBM effectively captures localized geomorphological variations and mitigates spectral confusion common in heterogeneous aquatic environments. The model was rigorously evaluated at Yantian Port (coastal) and Sancha Lake (reservoir). Results demonstrate that SL-GBM significantly outperforms traditional Stumpf, Log-linear, and Random Forest models. At Yantian Port, it achieved an R2 of 0.95 and RMSE of 0.90 m; at Sancha Lake, it maintained an R2 of 0.87 and RMSE of 2.05 m despite a 24-meter depth gradient. Feature importance analysis revealed that the synergy between spatial priors and spectral derivatives accounts for over 60% of the model gain. These findings suggest that SL-GBM is a robust, spatially-aware framework for high-resolution bathymetric mapping within the evaluated coastal and inland environments. Future cross-regional validation is recommended to firmly establish its broader operational applicability across globally diverse aquatic systems.