Inferring Bathymetry from Sentinel-2 Satellite Images Using Machine Learning Algorithms Based on Chlorophyll Concentration Data in the Absence of Ground Measurement

Abstract Bathymetric mapping is vital for navigation, coastal management, and marine habitat assessment. Traditional methods use satellite reflectance data and machine learning (ML), supported by echo-sounding field data. This study explores the underutilized potential of chlorophyll concentration for water depth inference, introducing it as a novel alternative for bathymetry and highlighting advanced ML techniques for improved mapping precision. The novel approach, applied at two sites in Egypt, Jemsha region (Gulf of Suez coast) and New Heaven Resort (south of Marsa Alam on the Red Sea coast), uses two main strategies. First, water depth was estimated from MODIS satellite chlorophyll data and validated with echo-sounding field data, yielding an RMSE of 1.5 m, R 2 of 0.55, and precision of 0.836 for Jemsha, and an RMSE of 2.5 m, R 2 of 0.1, and precision of 0.979 for New Heaven. Second, water depth was derived from Sentinel-2 satellite reflectance data using a new ensemble ML (EM) technique, refined from three well-known bathymetry models, and validated similarly. Results showed an RMSE of 1.3 m, R 2 of 0.5, and precision of 0.836 for Jemsha, and an RMSE of 2 m, R 2 of 0.3, and precision of 0.979 for New Heaven. These findings are globally significant, addressing bathymetric data scarcity in areas with limited field data or logistical constraints, while advancing methods for sustainable coastal management and marine conservation.