Landslides and flood multi-hazard assessment using machine learning techniques: Fid-Bau Portal

Landslides and flood multi-hazard assessment using machine learning techniques

Youssef, Ahmed M. / Mahdi, Ali M. / Pourghasemi, Hamid Reza

Abstract Saudi Arabia is affected by various types of natural hazards that affect people's lives and property. In this paper, the effects of landslides and floods in Wadi Dawqah in Bahah region, Saudi Arabia, are studied. Mountainous regions are exposed to natural hazards that require modeling to produce multi-hazard maps. Three models were used in this study: logistic regression (LR), random forest (RF), and support vector machine (SVM). Seventeen influencing factors are used to map multi-hazards using inventory data. The effectiveness and significance of the factors were determined for both landslides and floods. Model results were compared using area under the curve (AUC), overall hazard accuracy, mean absolute error, root mean square error, and kappa index. The results show that RF has the best performance in predicting landslides with a high AUC of 94.9%, while SVM, LR, and RF have high prediction for floods (AUC of 92.7%, 98%, and 98.7%, respectively). Since RF has the highest values for both landslides and floods, other statistical comparisons show that the overall accuracy is 94% and 99%, the kappa index is 0.870 and 0.978, the MAE values are 0.064 and 0.010, and finally the RMSE values are 0.253 and 0.10 for landslides and floods, respectively. Accordingly, the models RF for landslides and LR, SVM, and RF for floods were used to generate three multi-hazard maps for Wadi Dawqah basin. As a preliminary step, the map produced could provide a cornerstone for planners and decision making for future development.