Shallow Water Bathymetry Mapping Using Sentinel-2 and Machine Learning in Raja Ampat Coastal Waters

Abstract

This study aims to evaluate the effectiveness of Satellite-Derived Bathymetry (SDB) using Sentinel-2 imagery integrated with machine learning approaches for shallow water depth mapping in Raja Ampat, Southwest Papua a region characterized by complex seafloor topography and exceptionally clear waters. The methodology combines empirical regression models (green band power regression, blue band power regression, and Stumpf logarithmic ratio method) with Random Forest machine learning to predict bathymetry from spectral reflectance data. Field bathymetric data from fishfinder measurements were used for calibration and validation through Google Earth Engine (GEE) platform. Empirical regression results demonstrate that the green band model (B3) achieved the highest accuracy (R² = 0.7097, RMSE = 1.80–14.12 m across depth classes), followed by the blue band (R² = 0.6194) and Stumpf method (R² = 0.5693). The Random Forest model exhibited superior performance in capturing non-linear depth-reflectance relationships, particularly in complex substrate conditions. The green band model performed optimally for shallow to medium depths (0–20 m), while the Stumpf method showed greater stability at depths >20 m. These findings provide a cost-effective and scalable approach for bathymetric mapping in remote archipelagic regions, supporting marine conservation and coastal resource management in Raja Ampat and similar tropical marine ecosystems.