Machine learning-based flood susceptibility mapping in the Ourika Watershed, Morocco

Flooding is a prevalent natural hazard that poses significant challenges for accurate prediction and risk assessment. Effective spatial prediction is essential for developing strategies to mitigate the impacts of floods. This study aimed to map flood susceptibility in the Ourika watershed using three machine learning algorithms: eXtreme Gradient Boosting (XGBoost), Random Forest (RF), and K-Nearest Neighbors (KNN). The analysis used 557 binary data points (1 = flood zone, 0 = non-flood zone) to generate a flood inventory map. Fifteen potential flood-influencing factors were considered, and the most relevant and independent variables were selected through a multi-collinearity analysis. The final flood susceptibility maps revealed that 90% of the area was classified as non-flood susceptible, while the remaining 10% was identified as highly prone to flooding. The performance of the machine learning models was evaluated using training and validation tests, with the models demonstrating satisfactory results. The average area under the curve (AUC) for the receiver operating characteristic (ROC) curve was 0.89, indicating good predictive accuracy across all three algorithms. This study provides an effective approach for flood susceptibility mapping in the Ourika watershed and highlights the potential of machine learning techniques for improving flood risk management. The results can support more informed decision-making for flood mitigation strategies, helping to reduce the impact of future flood events. The developed flood susceptibility maps offer valuable insights for local authorities and stakeholders involved in flood risk management and disaster preparedness.