Optimizing satellite-derived bathymetry in coastal turbid waters through integration with optically active substances: Insights from the SATCORE project

Abstract

Satellite-Derived Bathymetry (SDB) has long been a focus of research, yet its practical implementation faces persistent challenges, primarily due to suspended materials, especially in nearshore regions. This research aims to enhance SDB accuracy using Landsat-7 and Landsat-8 spectral bands by systematically analyzing the spatial distribution of Optically Active Substances (OAS) in coastal waters. This study explores the spatial interdependencies among bio-geo-chemical water quality variables collected by Indian National Center for Ocean Information System (INCOIS) initiated SATellite Coastal and Oceanographic Research (SATCORE) project and bathymetry using Lasso Regression, Elastic Net, and Random Forest (RF) models. The study evaluates the predictive power and importance of various features in deriving SDB. Feature importance analysis showed that the spectral band 0.561 μm played a dominant role across all datasets, while other environmental variables had varying levels of influence. The results highlight the effectiveness of advanced machine learning models, particularly RF, in optimizing SDB mapping, enhancing the accuracy of predictions in complex coastal regions with high levels of suspended materials.