Using machine learning and satellite data to improve flood forecasting: the case of the Ouémé basin at the Bétérou outlet

Abstract

Flooding is a major concern for the scientific community, and it has been exacerbated by climate change. Accurate prediction of these extreme events is crucial for adequate preparedness. This study investigates the potential of advanced artificial intelligence (AI) techniques to enhance the accuracy of flood prediction and provide actionable insights for flood management. This study focuses on the African context, where data are scarce and the weak capacity of governments to react to floods makes populations vulnerable. It adopted advanced recurrent neural network architectures such as the long short-term memory (LSTM) and the convolutional long short-term memory (ConvLSTM) models, focusing on hydrological modeling innovation. The results indicated a high performance of these models in simulated runoff. The coefficient of determination (R2) and Nash–Sutcliffe efficiency between observed and simulated runoff are approximately 0.96 and 0.95, respectively, for the ConvLSTM model and 0.95 and 0.95 for the LSTM model. This study generated the flooding risk area maps. These maps represent a significant decision-making tool for flood management. This research confirms the effectiveness of deep learning in hydrology and proposes an innovative methodological framework for sustainable water resource management in the African context.