[en] Mapping surface water dynamics is critical for hydrological research and environmental monitoring, yet traditional remote sensing methods face limitations under vegetation or cloud cover and offer low temporal resolution. In this study, we propose an enhanced surface water mapping technique over the Amazon Basin using Global Navigation Satellite System-Reflectometry (GNSS-R) data from the CYGNSS mission, combined with a Deep Learning (DL) approach. By leveraging delay-Doppler maps (DDMs) and ancillary data, we trained convolutional and artificial neural networks to predict surface water presence at 1-km and 3-km resolutions. Validation against the Global Surface Water dataset and external CYGNSS-based products demonstrated high accuracy (0.90) and robust performance across varying hydrological conditions. Feature importance analysis identified peak reflectivity and topographic factors as critical predictors. Despite challenges in regions consistently classified as land or water and reliance on optical references beneath vegetation, the method highlights the potential of GNSS-R for future water monitoring applications, including upcoming missions like HydroGNSS.
