Landslide susceptibility mapping is well recognized as an essential element in supporting decision-making activities for preventing and mitigating landslide hazards as it provides information regarding locations where landslides are most likely to occur. The main purpose of this study is to produce a landslide susceptibility map of Mt. Umyeon in Korea using an artificial neural network (ANN) involving the factor selection method and various non-linear activation functions. A total of 151 historical landslide events and 20 predisposing factors consisting of Geographic Information System (GIS)-based morphological, hydrological, geological, and land cover datasets were constructed with a resolution of 5 x 5 m. The collected datasets were applied to information gain ratio analysis to confirm the predictive power and multicollinearity diagnosis to ensure the correlation of independence among the landslide predisposing factors. The best 11 predisposing factors that were selected in this study were randomly divided into a 70:30 ratio for training and validation datasets, which were used to produce ANN-based landslide susceptibility models. The ANN model used in this study had a multi-layer perceptron (MLP) structure consisting of an input layer, one hidden layer, and an output layer. In the output layer, the logistic sigmoid function was used to represent the result value within the range of 0 to 1, and six non-linear activation functions were used for the hidden layer. The performance of the landslide susceptibility models was evaluated using the receiver operating characteristic curve, Kappa index, and five statistical indices (sensitivity, specificity, accuracy, positive predictive value (PPV), negative predictive value (NPV)) with the training dataset. In addition, the landslide susceptibility models were validated using the aforementioned measures with the validation dataset and were compared using the Friedman test to check the significant differences among the six developed models. The optimal number of neurons was determined based on the aforementioned performance evaluation and validation results. Overall, the model with the best performance was the MLP model with the logistic sigmoid activation function in the output layer and the hyperbolic tangent sigmoid activation function with five neurons in the hidden layer. The validation results of the best model showed a sensitivity of 82.61%, specificity of 78.26%, accuracy of 80.43%, PPV of 79.17%, NPV of 81.82%, a Kappa index of 0.609, and AUC of 0.879. The results of this study highlight the effectiveness of selecting an optimal MLP model structure for shallow landslide susceptibility mapping using an appropriate predisposing factor section method.

Debris flows are one of the perilous landslide-related hazards due to their fast flow velocity, large impact force, and long runout, in association with poor predictability. Debris-flow barriers that can minimize the energy of debris flows have been widely constructed to mitigate potential damages. However, the interactions between debris flows and barriers remain poorly understood, which hampers the optimal barrier installation against debris flows. Therefore, this study examined the effect of barrier locations, in particular source-to-barrier distance, on velocity and volume of debris flows via the numerical approach based on smoothed particle hydrodynamics (SPH). A debris-flow event was simulated on a 3D terrain, in which a closed-type barrier was numerically created at predetermined locations along a debris-flow channel, varying the source-to-barrier distance from the initiation point. In all cases, the closed-type barrier significantly reduced the velocity and volume of the debris flows, compared to the cases without a barrier. When the initial volume of source debris was small, or when the flow path was short, the barriers effectively blocked the debris flow regardless of the source-to-barrier distance. However, with a long flow path, installation of the barrier closer to the initiation location appeared more effective by preventing the debris volume from growing by entrainment. Our results contribute to a better understanding of how source-to-barrier distance influences debris-flow behavior, and show that the methodology presented herein can be further used to determine optimum and efficient designs for debris-flow barriers.

Soil water content is one of the most important physical indicators of landslide hazards. Therefore, quickly and non-destructively classifying soils and determining or predicting water content are essential tasks for the detection of landslide hazards. We investigated hyperspectral information in the visible and near-infrared regions (400–1000 nm) of 162 granite soil samples collected from Seoul (Republic of Korea). First, effective wavelengths were extracted from pre-processed spectral data using the successive projection algorithm to develop a classification model. A gray-level co-occurrence matrix was employed to extract textural variables, and a support vector machine was used to establish calibration models and the prediction model. The results show that an optimal correct classification rate of 89.8% could be achieved by combining data sets of effective wavelengths and texture features for modeling. Using the developed classification model, an artificial neural network (ANN) model for the prediction of soil water content was constructed. The input parameter was composed of Munsell soil color, area of reflectance (near-infrared), and dry unit weight. The accuracy in water content prediction of the developed ANN model was verified by a coefficient of determination and mean absolute percentage error of 0.91 and 10.1%, respectively.

In this study, the back analysis was conducted using the DAN3D model for debris flows simulation in the Raemian and Sindonga watersheds to quantitatively analyze the influence of entrainment growth rate and rheological properties on the debris flow behavior. The numerical simulation was verified by using the field observation data obtained on July 27, 2011, at Mt. Umyeon, Seoul. The DAN3D model reasonably simulated the increase in the volume of the debris flow, due to entrainment phenomenon. Additionally, the model also provided a good estimation of the debris flow velocity. Furthermore, parametric studies confirmed that interactions between the entrainment growth rate and rheological parameters affect the debris flow velocity, thickness, volume increase, and run-out distance. The results of this study demonstrate that the DAN3D model is highly applicable for numerical simulation of debris flows with extremely rapid velocities. However, additional research should be conducted on methods to estimate the initial volume, entrainment growth rate, and rheological parameters as input data in the DAN3D model for predicting debris flow hazards.

A regional-scale landslide early warning system was developed in collaboration with a city government. The structure and distinctive features of the system are described in detail. This system employs the principles of the sequential evaluation method that consecutively applies three different evaluation stages: statistical, physically based, and geomorphological evaluations. Based on this method, the system determines five phases of warning levels with improved levels of certainty and credibility. In particular, the warning levels are systematically derived to enable the discrimination of slope failures and debris flows. To provide intuitive and pragmatic information regarding the warning capabilities of the system, a comprehensive performance analysis was conducted. Early warning level maps were generated and a historical landslide database was established for the study period from 2009 to 2016. As a result, 81% of historical slope failures and 86% of historical debris flows were correctly predicted by high-class warning levels. Miscellaneous details associated to the timing efficiency of warnings were also investigated. Most notably, five high-class warning level events and four landslide events were recorded for a study region during the eight-year period. The four landslide events were all successfully captured by four out of the five warning events.