SHALLOW WATER BATHYMETRY RETRIEVAL USING MACHINE LEARNING ALGORITHMS FOR HYPERSPECTRAL PRISMA IMAGERY

Accurate shallow-water bathymetric maps are essential for coastal resource management, transportation, military navigation, and scientific research. Satellite-based bathymetry provides an alternative to traditional ship-based methods. In this study, Random Forest (RF) and Multilayer Perceptron (MLP) algorithms were applied to PRISMA hyperspectral satellite imagery to map bathymetry in a protected area on the northeast coast of Brazil. Depth data were collected using a single-beam echosounder. The combination of PRISMA hyperspectral imagery and these ML algorithms proved to be effective for bathymetry retrieval, achieving high accuracy. The results showed R² values of 0.96 for RF and 0.98 for MLP, with RMSE values of 1.16 and 0.82, respectively.