Novel Physical and Statistical Models for Landslide Spatial Probability Prediction and Landslide Risk Assessment

Abstract

Landslides are catastrophic natural events. The frequency of landslide occurrence and magnitude of the landslide which causes damages to humans and infrastructures are increasing worldwide due to global climate change. Rainfall and earthquake are key triggering factors for landslides. The simultaneous occurrence of rainfall and seismic activity increases the likelihood of landslides. Korea is one of the countries that have been affected by extreme rainfall, especially in the summer season. Besides, Korea is located at a moderate seismic-hazard zone, which corresponds to a maximum acceleration on a rock site of 0.2 – 0.8 m/s2 with a 10% probability of exceedance in 50 years. Therefore, landslides become the most dangerous geo-hazard in the Korean mountains. To prevent landslides more efficiently, more studies related to predicting landslide initiation during rainfall and earthquake as well as evaluating regional-scale landslide risk are necessary. This thesis represents an attempt for predicting landslide spatial probability during rainfall and earthquake as well as assessing regional-scale landslide risk by proposing novel frameworks to improve the reliability and accuracy of the obtained results through using Python and GIS environments. Saturated depth in the soil layer due to rainfall is one of the most important factors controlling landslide. This thesis developed a hydrological model using Python and GIS environments to determine the saturated depth in the soil layer due to rainfall. The hydrological model considered both processes including rainfall infiltration and sub-surface flow in the soil layer, therefore the developed model can assess the saturated depth more reasonable and reliable than previous models. The saturated depth maps obtained from the developed model in Mt. Umyeon indicated that the saturated depth from the developed model is reasonable and reliable for using in landslide assessment. The effect of seismic activity on the shear strength of saturated soil is often neglected in earthquake-induced landslide assessment. This can lead to underestimation in evaluating landslide. This thesis proposed the processes to estimate changes of shear strength of saturated soil during earthquake through calculating the excess pore-water pressure ratio and strength softening ratio. The results obtained at Mt. Umyeon showed that the proposed frameworks were successful in assessing the effect of seismic activity on the shear strength of saturated soil. Two novel physical models were proposed to predict rainfall-earthquake-induced landslide spatial probability. The proposed physical models are an integration of the hydrological model and the effect of an earthquake on the shear strength of saturated soil in the infinite slope model. The MPS model combined the hydrological model and excess pore-water pressure ratio in the infinite slope model for calculating the factor of safety. While the RS model incorporated the hydrological model and strength softening ratio in the infinite slope model. Landslide spatial probability was estimated by two proposed models based on Monte Carlo simulation (MCs). The results obtained at validation areas (Atsuma, Padang) and application area (Mt. Umyeon) indicated that the two proposed models were successful in rainfall-earthquake-induced landslide spatial probability mapping and showed better performance than the previous model. To improve the accuracy of the rainfall-earthquake-induced landslide spatial probability maps from the physical models, an ensemble of deep learning (DL) techniques and physical model was proposed. Two DL models were used that is the multilayer perceptron (MLP) neural network and the convolutional neural network (CNN). The results obtained at Atsuma and Padang presented that both DL architectures were successful in the improved performance of the physical models, in which CNN showed more efficient than MLP. The researches related to landslide risk assessment which is a difficult and complicated task have not much done for the Korean mountains. This thesis proposed a model for assessing regional landslide risk based on the risk index. The proposed risk index equation considered both the effect of the magnitude of the landslide on structures and the number of the damaged structures from the landslide. The result from Mt. Umyeon and Mt. Hwangnyeon expressed that the proposed model was successful in regional landslide risk assessment with high reliability.