A comparative assessment of data-driven flood susceptibility mapping in Nigeria

Abstract Flood risk in West Africa, particularly Nigeria, has significantly increased over the past five decades due to changing hydrological conditions, insufficient mitigation measures, and limited adaptation efforts. This study addresses the need for accurate and high-resolution data to support effective disaster risk management. Leveraging open-access remote sensing and geospatial data, we trained machine learning models to produce 30-meter resolution flood susceptibility maps for Nigeria. We compared four Digital Elevation Models (DEMs) and four hydrological methods (D8, D-inf, FD8, and Rho8) to model water flow direction and accumulation. Additional flood-influencing factors, such as land cover, soil characteristics, and proximity to water bodies, were also incorporated. Three models were developed and evaluated: random forest (RF), binary logistic regression (LG), and linear discriminant analysis (LDA). Across all models, the highest accuracy was achieved using the Copernicus DEM in combination with the D8 and FD8 methods. Model performance was validated against a major flood event in 2022, demonstrating a strong predictive capability. To reconcile differences among model outputs, we created an ensemble map that consolidates their strengths while accounting for uncertainty. We also estimated the population exposed to flood risk and found that approximately 11 million people in Nigeria currently live in flood-prone areas. This approach offers valuable insights for stakeholders seeking to strengthen localized disaster risk management. We discuss study limitations and outline directions for future research. To promote transparency and reproducibility, we provide the scripts used to generate the flood susceptibility maps, along with our final output maps for Nigeria: figshare.com.