Susceptibility Analysis of the Mt. Umyeon Landslide Area Using a Physical Slope Model and Probabilistic Method

Every year, many countries carry out landslide susceptibility analyses to establish and manage countermeasures and reduce the damage caused by landslides. Because increases in the areas of landslides lead to new landslides, there is a growing need for landslide prediction to reduce such damage. Among the various methods for landslide susceptibility analysis, statistical methods require information about the landslide occurrence point. Meanwhile, analysis based on physical slope models can estimate stability by considering the slope characteristics, which can be applied based on information about the locations of landslides. Therefore, in this study, a probabilistic method based on a physical slope model was developed to analyze landslide susceptibility. To this end, an infinite slope model was used as the physical slope model, and Monte Carlo simulation was applied based on landslide inventory including landslide locations, elevation, slope gradient, specific catchment area (SCA), soil thickness, unit weight, cohesion, friction angle, hydraulic conductivity, and rainfall intensity; deterministic analysis was also performed for the comparison. The Mt. Umyeon area, a representative case for urban landslides in South Korea where large scale human damage occurred in 2011, was selected for a case study. The landslide prediction rate and receiver operating characteristic (ROC) curve were used to estimate the prediction accuracy so that we could compare our approach to the deterministic analysis. The landslide prediction rate of the deterministic analysis was 81.55%; in the case of the Monte Carlo simulation, when the failure probabilities were set to 1%, 5%, and 10%, the landslide prediction rates were 95.15%, 91.26%, and 90.29%, respectively, which were higher than the rate of the deterministic analysis. Finally, according to the area under the curve of the ROC curve, the prediction accuracy of the probabilistic model was 73.32%, likely due to the variability and uncertainty in the input variables.