INTEGRATION OF STUMPF'S RATIO MODEL AND RANDOM FOREST FOR SATELLITE-DERIVED BATHYMETRY ESTIMATION

Abstract

The development of remote sensing for coastal and marine environment mapping has significantly enhanced our understanding of these ecosystems, enabling improved mitigation strategies against the impacts of human activities. However, remote sensing must consider the complex interplay of the atmosphere and water column. Ongoing research focuses on refining water column correction techniques, including Depth Invariant Indices (DII), Radiative Transfer models, and bathymetry models. This study specifically aims to enhance the Stumpf's Ratio model (SRM) for bathymetry by employing the Random Forest (RF) machine learning regression algorithm. The resulting bathymetry model, which incorporates visible bands from a Sentinel-2 MSI image, and their Stumpf's ratios, outperforms other methods, yielding the highest accuracy with RMSE and R2 values of 1.25 m and 0.854, respectively. This was followed by the multivariate SRM with RMSE and R2 values of 2.196 m and 0.554 respectively These findings demonstrate the promising potential of using RF machine learning regression with SRM for bathymetry modelling.