Development of hazard assessment framework for Debris-flow induced by shallow landslide under extreme rainfall

Abstract

This research aims to develop a debris-flow hazard assessment framework to provide quantitative hazard information by estimating three input parameters (initial volume, entrainment growth rate, and bulk basal friction angle) of a dynamic DAN3D model for debris-flow predictive modeling. The proposed framework comprises of the five following main processes: (1) artificial neural network (ANN)-based shallow landslide susceptibility analysis to determine potential hazardous areas at a large scale

(2) debris-flow volume prediction analysis to provide a potential value of debris-flow volume, which can be utilized to obtain the entrainment growth rate value

(3) predictive debris-flow hazard analysis: (3-1) physically-based landslide analysis to estimate landslide sources among the highly susceptible areas

(3-2) application of a local geomorphological initiation criterion to extract initial debris-flow volume among the landslide sources

(3-3) debris-flow run-out analysis based on a parameterized bulk friction angle database established through the back-analysis of past debris-flow events within similar geo-environmental conditions by applying the Monte Carlo method. Through these processes, debris-flow starting points, initial volume, final volume, behavior change along the channel path, and probabilistic information on debris-flow velocity and thickness can be obtained. First, in the ANN-based landslide susceptibility analysis stage, 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. Overall, the model with the best performance was the ANN 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. This model was used to reduce the area to be applied when performing physically-based landslide analysis. Second, in the debris-flow volume prediction analysis stage, an ANN model was developed to predict debris-flow volume based on 63 historical events. A total of 15 predisposing candidates of morphological, rainfall, and geological types were constructed from data obtained from the central region of South Korea. Among these data, four predisposing factors (watershed area, channel length, watershed relief, and continuous rainfall) were selected based on Pearson’s correlation analysis to check for significant correlations with debris-flow volume. In addition, in a comparative study with other existing regression models, the ANN model showed better results in terms of adjusted $R^2$ value (0.911) using all datasets. Third, in the predictive debris-flow run-out analysis stage, the factor of safety was calculated according to rainfall duration by combining 1-dimensional rainfall infiltration analysis and infinite slope stability analysis in a Raemian watershed located in the northern slope of Mt Umyeon. The results showed that the factor of safety decreased to 1.3 or less two hours before the actual landslide occurred, and the unstable area increased as rainfall continued. As a result of applying the debris-flow initiation criterion in unstable areas, it was confirmed that the initial volume was generated at the time of the actual debris-flow occurrence. Lastly, Monte Carlo simulation was applied to analyze the velocity and thickness of the debris-flow in the Raemian region using the basal friction angle (rheological parameter) database constructed by performing back-analysis of 37 past debris-flow events. As a result of the Monte Carlo simulation, the measured debris-flow velocity and thickness corresponded to the 99% quantile of the cumulative distribution function in the Raemian region. This was attributed to the fact that a large amount of rainfall run-off occurred simultaneously at the actual debris-flow initiation time. In order to validate the applicability of the suggested framework, a case study was applied and tested in the Raemian watershed, where catastrophic damage was caused by debris-flows that were recorded in detail. According to a comparison between the landslide inventory map and recorded debris-flow evidence, the application of the framework produced reasonable estimates of the debris-flow initiation time and location, thickness, velocity, volume, and flow-path. Based on the hazard information, efficient locations for check dams were suggested adjacent to the locations of potential initiation areas and perpendicular to the flow-path with consideration of vulnerability analysis results. As a result of numerical analysis, the vulnerability of structures was analyzed as acceptable when two check dams are installed at the start and end paths of potential initiation points. Thus, it was confirmed that it is possible to design check dams on the basis of the hazard information of debris-flow derived from the suggested framework in this study. However, since this study did not consider the size of the check dams and its effects, a follow-up study should focus on developing high-efficiency and eco-friendly check dam designs. Nevertheless, the framework developed in this study is vital as it is necessary to consider predicted debris-flow hazards when planning countermeasures and developing cities around mountainous regions. As a final conclusion, the suggested framework can become a powerful tool for decision-makers for disaster preparation if it is verified and corrected with improvements according to local conditions.