A wavelet–LSTM framework for mapping large inland water bodies using CYGNSS observations

Study region This study investigates six representative large lakes distributed across different climatic and hydrological environments: Lake Victoria, Lake Titicaca, Lake Malawi, Lake Tanganyika, Lake Nicaragua, and Lake Maracaibo. These lakes provide diverse conditions for evaluating large-scale inland water mapping using spaceborne Global Navigation Satellite System Reflectometry (GNSS-R). Study focus A wavelet–Long Short-Term Memory (LSTM) framework is developed using Cyclone Global Navigation Satellite System (CYGNSS) observations. Wavelet transform is applied to Delay-Doppler Map (DDM) data to extract multi-scale scattering features, while the LSTM network captures temporal dependencies within the feature sequences. The model is trained using 2021 CYGNSS observations over Lake Victoria and Lake Titicaca, with Global Surface Water data as reference labels. New hydrological insights for the region The proposed framework achieved overall accuracies of 98.53% and 99.10% for Lake Victoria and Lake Titicaca, respectively. Validation over Lake Malawi, Lake Tanganyika, Lake Nicaragua, and Lake Maracaibo yielded overall accuracies of 95.48%, 96.34%, 89.76%, and 93.74%, respectively. The results show that multi-scale DDM features and temporal modeling enable robust large-lake mapping under diverse and complex scattering conditions. Application to Lake Poyang further captured quarterly water variations in 2021 with an average accuracy of 85.24%, demonstrating the potential of GNSS-R for monitoring seasonal lake dynamics.