Prediction of Debris Flow Initiation Points using High Resolution DEM and GIS Analysis in South Korea

This study presents a GIS analysis model that utilizes high-resolution Digital Elevation Map data to predict the initiation points of debris �ow through the overlap of slope stability analysis techniques and rainfall �ow analysis. The debris �ow that occurred on Mount Woomyun in Seoul, South Korea in 2011 was selected as the case study for comparing and analyzing the prediction model against the actual case. In South Korea, occurrences of debris �ow mostly happen during the summer monsoon season. When a substantial amount of rainfall accumulates and heavy rainfall events occur, the pore water pressure within slopes reaches saturation. Consequently, �ow is initiated along "temporary streams." In this study, areas with a combination of low slope stability and abrupt changes in velocity vectors were assumed to have the highest probability of debris �ow occurrence due to the overlapping of these factors. The research model was conducted under this assumption. To evaluate the performance of the model, 3D coordinates of the model's result point, and actual occurrence points were obtained, and a Pearson correlation analysis was conducted to compare them. With these results, R 2 values of X = 0.9147, Y = 0.8625, and H = 0.8942 were obtained. These high R 2 indicate a strong correlation between the model's predictions and actual occurrences.


Introduction
Debris ow in South Korea predominantly occur during the concentrated heavy rainfall of the summer monsoon season.With over 70% of the country's territory consisting of mountainous terrain, the risk of debris ows and landslides is exceptionally high compared to other disasters.South Korea is thus considered one of the countries with a very high vulnerability to these hazards.The model proposed in this study utilizes high-resolution aerial LiDAR DEM-based terrain data (with a resolution of 1m x 1m) to analyze the topographical stability of slopes and simulate rainfall ow.
We assume that areas with extreme changes in rainfall vector velocity on slopes with low stability generate abrupt impact forces, leading to a higher likelihood of debris ow occurrence.We identify points with these characteristics using GIS overlay analysis.The reliability of this technique will be validated by comparing it with representative real-world cases of debris ow occurrence.Takahashi(2007) classi ed the initiation of debris ows in steep mountainous areas during concentrated rainfall into three cases: (1) natural slope failure, (2) gully erosion and incision on valley oors and slopes, and (3) collapse of natural earthen dams.Similarly to Japan, the initiation of natural slope failure in domestic areas is primarily due to an increase in pore pressure resulting from rainfall in ltration.Considering this, the following assumptions are made in this study to estimate the starting point of debris ows: (1) During rainfall events, water generally collects on slopes through various temporary drainage networks and converges in catchment areas, forming temporary channels and moving down the slopes.
(2) Debris ows are more likely to occur in areas with low stability index grades.(3) The ow generated by rainfall temporarily alters velocity and impact force along the channels, in uenced by the topography.
The probability of debris ow occurrence is higher at points where there is a rapid change in velocity vectors and impact force.(4) After a su cient cumulative rainfall(200ml), we assume the pore water pressure of soil on the slope is close to the saturated state.
The soil composed of ne particles maintains the slope in a state of cohesion.When uid ow generated by rainfall induces shear stress in the slope, exceeding the yield stress limit due to cohesion, the soil particles mix with water and begin to move.The movement of debris ow can be considered as the movement of a viscous uid, and rheological factors such as yield stress and viscosity play a crucial role in determining the mobility of the failure surface (Jeong, 2010).
To re ect the characteristics of debris ow behavior in South Korea, this study simulated the ow and extent of debris ow based on speci c similarity concentrations using the FLO-2D model.
Figure 1 shows the conceptual process of estimating the occurrence points of debris ows proposed in this study.
The locations where the ow velocity undergoes a rapid change, and the impact force abruptly increases along the path are determined through overlapping analysis.To consider the ow motion of the discharge during concentrated rainfall events, a numerical analysis program based on rheological models is used to simulate the hydraulic ow and calculate the similar velocity vectors, thus identifying the points of abrupt change in the watercourse.
The reliability of the analysis results is evaluated by comparing them with the actual locations of debris ow occurrence obtained through eld surveys.Figure 2. shows an overview of the owchart for this study.

Study Area
In this study, we selected the representative case of the debris ow that occurred on Mount Woomyun in Seoul, South Korea in 2011, the most prominent example of domestic debris ows, as the research site.
We analyzed the terrain data from 2009, created using aerial LiDAR techniques, prior to the occurrence of the landslide.By examining the vectors associated with the movement of temporary streams and in owing discharge during concentrated rainfall events, we calculate the location where debris ow initiation occurs, predicting the starting point of debris ow.

Create Terrain Data
Base terrain data for this study utilized a Digital Elevation Model (DEM) with 1m × 1m resolution, which is obtained through aerial LiDAR surveying techniques.Aerial LiDAR systems involve laser scanners mounted on aircraft that emit laser pulses, measure the time it takes for the pulses to reach the surface, and calculate the three-dimensional spatial coordinates of the re ected points, thus extracting terrain information about the surface.Recently, integration with GPS/INS equipment has been used to achieve an average accuracy of 0.089m ± 0.062m, surpassing the accuracy of 1/1,000-scale topographic maps (Wie et al., 2007).The coordinate data of the point clouds extracted from aerial LiDAR survey results, along with GPS/INS data, can be preprocessed and irregular point data can be post-processed to generate high resolution DEM.Fig. 3

Slope stability analysis
In this study, the Stability Index MAPping (SINMAP) method, which is based on the in nite slope stability model, was used to classify the slope stability of the study area.Instead of using default parameter values typically employed in general analyses, the parameter settings were determined based on the results of in-situ soil laboratory experiments conducted at the time of debris ow occurrence.The parameter settings are speci ed in Table 1.

Catchment Area Analysis
To dividing Watershed and Catchment Area, ArcGIS was used to derive watersheds and catchment areas in the study area from the 1m × 1m resolution DEM based on the temporary water ow generated by rainfall.The catchment area delineation was performed using the D8 algorithm, which is commonly used in ArcGIS to derive drainage networks.
The D8 algorithm is based on the strict assumption that water always ows in the steepest direction within the 9-grid cell neighborhood.
Through this process, a channel network is derived, which is temporarily generated during the occurrence of concentrated rainfall.In the ow analysis, movement follows this channel network.Figure 5. Shows the result of channel network analysis.

Stability Index Analysis
The accuracy of output is heavily reliant on the accuracy of the digital elevation model (DEM) data input (2005, SINMAP user's manual).The analysis results are presented in terms of Stability Index ratings, which indicate the slope stability regarding debris ow occurrence at each location.The slope stability index in this analysis represents relative risk rather than numerically precise values.the Upper Threshold area (0.0 < SI < 0.5), which is highlighted in red in Fig. 6. is de ned as the Low Stability Zone.

Flow Vector Analysis
For Rainfall ow modeling, we used the FLO-2D model.The model is based on the physical laws of the debris ow.Also, model uses a numeric grid made from quadratic cells, the size of which can be altered, to describe the geometry of the given area.The basic Model equations in all directions are the continuity equation.The governing equations of the FLO-2D model consist of the continuity equation (Eq. 1) and the momentum equations (Eqs. 2 and 3).
Eqs. ( 2) is shown only equations for the x direction.
Where the friction slope is composed of four sources of ow resistance.The rst term is a yield stress where is a cohesive and Mohr-Coulomb yield stress and is the speci c weight of the debris ow.The second term describes a viscous stress where K is a resistance parameter that grows based on roughness and is the Bingham dynamic viscosity.The third and fourth terms are turbulent and dispersive stress, respectively, merged into one variable that is described as a turbulent dispersive roughness connected with uid concentration and Manning's (FLO-2D, 2018).Table 2 shows input data of this model.It was investigated that intense rainfall occurred in the area where the debris ow took place.In this study, the observed rainfall was considered to lead to a condition similar to that of a temporary channel forming due to saturated pore pressure in the slope of the study area, leading to a situation akin to ooding.As a result, this rainfall data was adopted as the input value for the ow rate.

FLO-2D Model Analysis
The rainfall data input values were used to simulate the ow over the entire slopes of the study area, and similar velocity vectors for the owing slope were computed.
The simulation was based on 1m × 1m high-resolution terrain data.Figure 8. shows a) -> b) -> c) -> d) simulated the ow of rainfall over time, and the velocity vectors of the soil mixture were calculated.This enabled the calculation of the ow's acceleration, and the points of rapid change in the initial velocity vector could also be derived.Figure 8. also shows the movement of rainfall over a duration of 6 minutes, which is equivalent to 0.1 hour.

Velocity Vector Calculation
Through FLO-2D analysis, Fig. 9 shows calculated results of velocity vector.The maximum velocity vector calculated was 21.95 m/s, and the acceleration at the initial vector abrupt change point was calculated to be 28.88 m/s².Figure 9 (a) represents the analysis results for the entire study area, while Fig. 9 (b) is a zoomed-in image showing the location where the maximum velocity vector occurred.

Compare Model Results with Actual Occurrence Locations
To compare the model's predicted results with the actual locations of debris ow, coordinates of the model result points marked with orange dots and the observed occurrence points from eld surveys were obtained.Pearson correlation analysis was attempted between the coordinates to analyze the correlation between the model's results and actual occurrences.The coordinates for each location were obtained for the purpose of comparing the model's results with the actual occurrence locations.The reference for the coordinates is as follows: for horizontal positions, the ellipsoid used is GRS80, the projection is Transverse Mercator, False Easting is 200000 meters, and False Northing is 600000 meters.the observed actual occurrence locations are presented in Fig. 10.The coe cient of determination, R 2 , signi es the statistical correlation between the model's analytical results and the actual cases.the calculated R 2 values are X = 0.9147, Y = 0.8625, and H = 0.8942, indicating a high correlation between the predicted values of the model and the actual occurrence cases.The results of this model represent a very strong positive correlation with observed actual occurrence debris ow initiation points.

discussion
The model in this study aims to predict the initiation points of debris ow, allowing for effective estimation of the volume and extent of soil loss.Such estimations are crucial for analyzing the total debris ow volume and are essential for developing scenarios for risk analysis.additionally, this model could be highly useful for studying disaster scenarios by calculating the impact of debris ow propagation and estimating the potential damage areas.Moreover, analyzing points of maximum velocity vectors could assist in identifying areas prone to debris ow, aiding in the construction of reinforcing dams.The results of the analysis presented in this model, represented by the "orange dots," generally align with the starting points of "temporary channels" that form during concentrated rainfall events.However, cases where such temporary channel starting points are not identi ed as the initiation points of debris ow occur when slope stability analysis indicates higher stability or when non-erosionrelated slope failures or terrain deformations take place.In such situations, alternative analysis methods divergent from those proposed in this model are necessary.
When using high-resolution terrain data based on 1m x 1m LiDAR resolution, advantages are gained in analyzing rainfall dynamics, yet there remains a limitation in accounting for erosion initiation points caused by terrain deformation.

Conclusions
In this study, a combined model of slope stability analysis and rainfall ow analysis was proposed using high-resolution LiDAR-based terrain data.This model was utilized to predict erosion-prone areas during concentrated rainfall events with a cumulative rainfall exceeding 270mm and an hourly rainfall intensity of 50mm/hr.To evaluate the model's performance, Pearson correlation analysis was conducted between actual occurrence cases and the model's results, yielding R 2 of X = 0.9147, Y = 0.8625, and H = 0.8942.
These high R 2 indicate a strong correlation between the model's predictions and actual occurrences.
The accuracy of model proposed in this study is directly in uenced by the resolution of the spatial terrain data.In areas where high-resolution terrain data is not available, it is di cult to expect highly reliable results when using this model.furthermore, while advantageous for rainfall ow analysis, the model is limited in its ability to consider erosion initiation points caused by terrain deformation.although, this model could be highly useful for studying disaster scenarios by calculating the impact of debris ow propagation and estimating the potential damage areas.Moreover, analyzing points of maximum velocity vectors could assist in identifying areas prone to debris ow, aiding in the construction of reinforcing dams.

Declarations Data availability
The datasets generated during the current study are available from the corresponding author on reasonable request.

Corresponding author
Correspondence to Tae-Yun Kim.E-mail address: taeyoon4488@gmail.com

Ethics declarations
The authors declare that they have no known competing nancial interests or personal relationships that could have appeared to in uence the work reported in this paper.
(a) represents location of this study area and Fig. 3(b) represents the DEM of Mount Woomyun area in 2009, prior to the occurrence of debris ow.

Figure 7 .
Figure 7.  shows the observed hourly rainfall in the study area.The rainfall intensity recorded by the meteorological agency at 7 AM, when the debris ow occurred, was 52.4 mm/hr, and the accumulated rainfall was measured at 220 mm.

Figure 2 Flow
Figure 2

Figure 7 Observed
Figure 7

Table 1
Input Data of SINMAP analysis

Table 3
shows the coordinate of actual occurrence events.

Table 4 .
shows the coordinate of model results.To assess the reliability of the model's results, a statistical analysis was conducted by calculating the Pearson Correlation Coe cient.The equation for calculating the Pearson Correlation Coe cient, denoted as "r," is as follows:

Table 5 .
Shows R 2 value of model's results.Figure11.shows overlaying image of observed actual occurrence debris ow initiation points and model results.