Wüest, Alfred JohnyAmrari, Selim2019-07-092019-07-092018https://infoscience.epfl.ch/handle/20.500.14299/158964Bathymetry is a crucial variable for many researches, such as monitoring of the coral ecosystems. Traditional bathymetric measurements are made by echo-sounder, a type of sonar, measuring water depth from sea surface. Nevertheless, this acquisition process remains expensive and cannot be applied everywhere. Despite less accurate results, passive satellite derived bathymetry can provide a large-scale resolution at a lower cost. This study explores the potential of empirical models applied on images provided by Sentinel-2 sensor for bathymetry retrieval. Technical issues and suitable conditions are described here. In addition, a different approach is proposed. It is composed of an initial step designed to classify areas relying on their bottom reflectance and their water optical properties. Tests are achieved in New Caledonia, in the South West part of the lagoon. The method showed its ability to produce bathymetric data at a large scale, including shallow coral areas, which are difficult to measure by boat. The proposed approach allowed to improve the results of traditional empirical models. For this reason, satellite derived bathymetry can provide a valuable alternative for researchers needing to study coral ecosystems.remote-sensingmachine-learningclassificationstudent work::master thesis