Large-scale flood susceptibility mapping along major tributaries of the Nile River system in Sudan using convolutional neural network

Abstract Every fall season, Sudan experiences devastating floods that result in significant fatalities, displacement of populations, and severe disruptions to agricultural and economic activities. The 2020 floods were particularly catastrophic, reaching the highest Nile River levels in over a century and affecting more than 800,000 people. To the best of the authors’ knowledge, this study represents one of the first large-scale, data-driven flood susceptibility assessments for Sudan using deep Convolutional Neural Network (CNN) model. Twelve conditioning factors spanning topographical, hydrological, geological, and anthropogenic parameters were systematically analyzed. The flood inventory dataset was obtained from the Humanitarian Data Exchange (HDX), comprising geospatially verified flood polygons compiled from remote sensing and field observations. A five-fold cross-validation strategy was implemented to ensure model robustness, yielding a classification accuracy of 97% and demonstrating reliable generalizability across spatial partitions. The analysis generated probability maps with susceptibility values ranging from zero to 1, delineating three discrete risk zones: High, Medium, and Low susceptibility. Results reveal that very high susceptibility zones are concentrated along immediate river corridors, the Khartoum metropolitan confluence area, and low-lying Quaternary alluvial plains in Jazirah and Sennar states, where Sudan’s most critical agricultural production occurs. The susceptibility maps provide essential decision-support tools for evidence-based urban planning, agricultural management, and water resources development, enabling targeted risk reduction strategies that can protect lives, livelihoods, and development investments.