Machine Learning-Based Flood Susceptibility Mapping Using Geoenvironmental Factors in Central Morocco

Flood susceptibility mapping using geoInformation and machine learning-based models is of vital importance to predict future flood occurrences and make informed decisions on mitigation strategies. This study aims to assess the applicability of three widely used machine learning models, Classification and Regression Trees (CART), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost), and to evaluate their performance in mapping flood susceptibility in the Tensift Watershed, located in the central-western part of Morocco within the Marrakech province. Sixteen conditioning factors spanning topographic, geologic, climatic, and land cover domains were used as model inputs. A total of 228 flood inventory points, consisting of 114 flood and 114 non-flood locations, were used to train and test the models. The area under the receiver operating characteristic curve (AUC) was used to assess the performance of models. The results indicate that the CART model achieved the highest predictive accuracy with an AUC of 0.882, followed by SVM (AUC = 0.860) and XGBoost (AUC = 0.833). These findings suggest that CART is the most suitable model for flood susceptibility assessment in the study area. The outcomes of this research are also expected to support flood risk management activities in Marrakech province by identifying vulnerable areas, guiding infrastructure planning, and enhancing community preparedness to minimize adverse impacts of future flood events on both people and the environment.