Landslide susceptibility, ensemble machine learning, and accuracy methods in the southern Sinai Peninsula, Egypt: Assessment and Mapping

Abstract

Abstract Each year, thousands of tourists visit Egypt's Wadi Feiran region, one of the most popular tourist sites in the Sinai Peninsula. The region’s topography is distinctive and diverse, making it particularly susceptible to “natural disasters” (such as floods and landslides). The current study deals with landslide hazards as a critical hazard type, where, after rainfall, hundreds of landslides occur annually, and landslide disaster assessments are becoming more necessary to reduce mountain hazards. The current research mapped "landslide susceptibility" in the Wadi Feiran basin using three different modeling strategies: “Logistic Regression” -LR, “Artificial Neural Network”-ANN, and an "ensemble" of LR and ANN. A “landslides” map was first created as a preliminary stage, using 800 landslide locations acquired from multiple data sources (30% validation datasets, 70% training datasets), including historical records, field surveys, and high-resolution satellite imagery. In addition, fourteen landslide causative parameters (LCPs), elevation (El), “distance to wadis” (DtW), “distance to fault” (DtF), “distance to road” (DtR), lithology (Lith), aspect (As), “profile-curvature” (PrC), “plan-curvature” (PlC), “slope length” (LS), slope-angle (Sa), “topographic wetness index” (TWI), “relative slope position” (RSP), rainfall (Ra), and “topographic roughness index” (TRI) were employed. These models' accuracy was evaluated using “receiver operating characteristics and area under the curve (ROC - AUC),” “root mean square error”-RMSE, and “kappa index”-K. According to the findings, the AUC for LR, ANN, and ensemble of LR &ANN were 82%, 89%, and 91%, respectively. The results showed that the ensemble model outperformed ANN and LR by 2.3% and 10.9%, respectively, whereas ANN model outperformed LR by 8.5%. Other statistical indices also revealed that the RMSE and kappa index values obtained by LR were the highest and the lowest, respectively, whereas the RMSE and kappa index values generated by the LR&ANN ensemble were the lowest and the highest, respectively. These results indicate that landslides are influenced by a wide variety of natural and anthropogenic factors. To better manage and avoid landslides, it is important to create maps that show the area’s most susceptible to landslides.