Integrating Morphometric Controls for Runoff Dynamics in Bayelsa State, Nigeria: Enhancing Flood Susceptibility Mapping with Machine Learning

Abstract Flooding remains one of the most frequent and destructive hazards in Nigeria’s Niger Delta, where flat terrain and intense rainfall combine with rapid land-use change to heighten risk. This study integrates morphometric basin analysis with supervised machine learning to enhance flood-susceptibility mapping across four representative catchments in Bayelsa State Forcados, Nun, Ekole, and Seibri. Using a 30 m SRTM digital elevation model, fourteen morphometric parameters were derived to quantify drainage efficiency, infiltration potential, and relief characteristics that influence runoff generation. These parameters served as predictors for Random Forest, Support Vector Machine, and XGBoost classifiers trained with flood and non-flood samples extracted from Sentinel-1 SAR data (2018–2024). Among the tested models, XGBoost achieved the highest performance (accuracy = 93.1%, AUC = 0.95), reliably delineating high-risk sub-catchments. The resulting probability maps revealed micro-zones localized clusters of elevated flood potential within generally moderate basins such as Ekole, emphasizing the spatial heterogeneity of flood processes. Drainage density (Dd), relief ratio (Rh), and infiltration number (If) were the most influential variables controlling flood susceptibility. Forcados exhibited the greatest hazard (Dd = 3.57 km km⁻²; If = 41.80), whereas Seibri and Ekole showed lower overall susceptibility but contained critical micro-zones requiring attention. The approach demonstrates that combining morphometric controls with data-driven classification provides a transferable, high-resolution tool for flood-risk assessment in other tropical deltaic environments lacking dense hydrological networks. These findings support more precise local planning, early warning, and adaptation strategies in vulnerable low-lying regions.