Climate change puts the prefecture of Tangier-Assilah under increased risk of flooding, manifested by an upsurge and intensification of rainfall episodes (40 to 130 mm), thus overloading drainage systems and heightening the area’s vulnerability to extreme weather. This study seeks to answer the key research question: how can flood vulnerability be accurately assessed using machine learning and GIS to support spatial risk management in this fast-growing flood-prone area? Six critical variables were selected for their direct impact on runoff generation, water retention capacity, flood exposure and its dynamics: Elevation, Slope, Distance to permanent waters and wetlands, Land Use/Land Cover (LULC), Soil Moisture and Precipitation. The Flood Vulnerability Index (FVI) was developed using the Principal Component Analysis (PCA) to assign weights to each normalized factor involved and reduce the dimensionality while preserving essential variability. Sample points were then extracted and classified by vulnerability level, providing labeled data to train a Support Vector Machine (SVM) model that produced a detailed flood vulnerability map for precise large-scale spatial risk assessment. The model validation showed strong performance across multiple metrics: Coefficient of Determination (R² = 0.921) and Root Mean Square Error (RMSE = 0.230)—chosen for ordinal output suitability—alongside classification metrics including accuracy (99%), Kappa (0.98), F1-score (0.99), and confusion matrix analysis, demonstrating comprehensive fit between predictions and observed data. The results highlighted a significant disparity in vulnerability, with the highest levels found in dense urban areas, low-lying terrains, the surroundings of the Mharhar, El Hachef and Tahaddart Oueds, and coastal areas, emphasizing urgent targeted mitigation and effective risk handling.
Yacoubi et al. (Fri,) studied this question.