Flood susceptibility mapping for sustainable water management in the Upper Awash River Basin, Ethiopia: integrating machine learning and geospatial techniques

Flooding is among the most severe natural hazards globally, with particularly devastating impacts in developing countries such as Ethiopia. The Upper Awash River Basin (UARB), a socio-economically vital region, faces increasing flood frequency and intensity driven by urbanization, land use change, and climate variability. This study develops a high-resolution flood susceptibility map for the UARB using geospatial datasets integrated with machine learning algorithms. A flood inventory from historical records, field surveys, and satellite imagery was split into 70% training and 30% validation subsets. Six algorithms: Logistic Regression, Random Forest, Decision Tree, XGBoost, Support Vector Machine, and K-Nearest Neighbors were evaluated using ROC-AUC, Precision, Recall, Accuracy, F1-score, and spatial consistency. Ensemble models (XGBoost and RF) performed best, achieving AUCs of 0.92 and 0.90. About 15.89% of the basin was classified as very highly susceptible and 22.31% as highly susceptible. The findings offer a practical framework for flood risk management in data-scarce regions.