Coastal depth monitoring plays a crucial role in navigation and environmental management but is often constrained by the high operational costs and accessibility issues of conventional surveys. To address these challenges, this study implements "BathySAR-Net," an Artificial Neural Network (ANN) model designed to predict sea depth using Sentinel-1 Synthetic Aperture Radar (SAR) imagery combined with BATNAS bathymetric data. The model utilizes a Multilayer Perceptron (MLP) architecture to capture the complex, non-linear relationships between radar backscatter coefficients and water depth. Evaluation using robust K-Fold Cross Validation yielded a Root Mean Square Error (RMSE) of 8.7060 meters and a coefficient of determination (R²) of 0.0838. While the model demonstrated improved stability in deeper zones (10–15 meters), predictive performance in shallow intertidal areas (<5 meters) remained limited due to significant radar signal noise and data scarcity. These findings suggest that while SAR-based ANN offers a promising, accessible alternative for bathymetry, further integration of hydrodynamic variables is essential to resolve nearshore dynamics effectively.
Putri et al. (Thu,) studied this question.