Landslide and debris flow susceptibility zonation using TRIGRS for the 2011 Seoul landslide event

This paper presents the results from the application of a regional, physically based stability model: Transient Rainfall Infiltration and Grid-based Regional Slope-stability analysis (TRIGRS) for a region on Woomyeon Mountain, Seoul, South Korea. This model couples an infinite-slope stability analysis with a one-dimensional analytical solution to predict the transient pore pressure response to the infiltration of rainfall. TRIGRS also adopts the geographic information system (GIS) framework for determining the whole behaviour of a slope. In this paper, we suggest an index for evaluating the results produced by the model. Particular attention is devoted to the prediction of routes of debris flow, using a runoff module. In this context, the paper compares observed landslide and debris flow events with those predicted by the TRIGRS model. The TRIGRS model, originally developed to predict shallow landslides, has been extended in this study for application to debris flows. The results predicted by the TRIGRS model are presented as safety factor (FS) maps corresponding to transient rainfall events, and in terms of debris flow paths using methods proposed by several researchers in hydrology. In order to quantify the effectiveness of the model, we proposed an index called LR class (landslide ratio for each predicted FS class). The LR class index is mainly applied in regions where the landslide scar area is not well defined (or is unknown), in order to avoid overestimation of the model results. The use of the TRIGRS routing module was proposed to predict the paths of debris flow, especially in areas where the rheological properties and erosion rates of the materials are difficult to obtain. Although an improvement in accuracy is needed, this module is very useful for preliminary spatio-temporal assessment over wide areas. In summary, the TRIGRS model is a powerful tool of use to decision makers for susceptibility mapping, particularly when linked with various advanced applications using GIS spatial functions.

This discussion paper is/has been under review for the journal Natural Hazards and Earth System Sciences (NHESS). Please refer to the corresponding final paper in NHESS if available.

Landslide and debris flow susceptibility zonation using TRIGRS for the 2011 Seoul landslide event 1 Introduction
Shallow landslides involving colluvium are generally the most common in Korea and often mobilize into destructive debris flows. Shallow landslides are typically 1-3 m deep and often occur at boundaries between the colluvium and the underlying more solid parent rock . In most parts of Korea, including Seoul, the thick-5 ness of the colluvium is generally less than 2 m because of the relatively shallow depth of the bedrock, and hence shallow landslides are frequent. Furthermore, the climate of Korea is typical of the Indian Ocean Monsoon, with pronounced seasonal precipitation (Kim et al., 2010). Thus, rainfall-triggered landslides are a recurring problem in Korea. Due to the mountainous terrain with a shallow layer of colluvium, and associ-10 ated weather conditions, landslides have proven a hazard across most of the country. The socio-economic impact, moreover, has become much higher than before because of the current population levels in the hazardous zones. During 26-27 July 2011, in particular, a heavy rainfall (470 mm in two days) occurred in Seoul, an amount approximately equal to 20 % of the total annual rainfall for that 15 region. During this precipitation event, 147 catastrophic landslides occurred on Mt. Woomyeon. Most of the landslides were accompanied by debris flows, and these mixtures of debris flowed down roads into the surrounding communities. Sixteen people were killed and ten buildings damaged by these debris flows. During the storm, shallow landslides on steep mountainous terrain were mostly triggered by heavy rainfall that in- 20 creased the pore pressure of soil in the near-subsurface, with an attendant decrease in its shear strength. Under these conditions, precipitation-induced landslides caused translational mass movements that occurred suddenly.
In order to understand when and where rainfall-induced landslides have occurred in mountainous regions, and how topographic, geotechnical and hydraulic parameters Introduction  (Shaw and Johnson, 1995), which stands for Slope MORPHology Model, is an empirical model adapted to include the contributing area with creeping process. One advantage of this model is that it only uses parameters derived from the Digital Elevation Model (DEM) to calculate susceptibility, and does not require field mapping.
In contrast, physically based or deterministic models are more frequently used for 5 specific catchments, because there are physical descriptions that can be used to inform mathematical equations about slope failure processes. Five such models are introduced below, and compared with TRIGRS, the model upon which this work is focused. LISA (Hammond et al., 1992) stands for Level I Stability Analysis. It identifies the effects of the tree root strength and tree surcharge on slope stability as an important 10 parameter of forested, hill-slope areas. LISA enables the user to compute the probability of slope failure using up to 1000 iterations of a Monte Carlo simulation, by varying input values involved in the infinite slope equation. The Monte Carlo simulation estimates the probability of failure rather than a single factor of safety value. SHALSTAB (Dietrich et al., 1993Montgomery and Dietrich, 1994;Mont-15 gomery et al., 1998) stands for Shallow Land-sliding Stability Model. It is a coupled, steady-state runoff and infinite-slope stability model which can be used to map the relative potential for shallow sliding. This model correctly predicts the observed tendency for soils to be thick in the un-channelled valleys and thin on ridges. The dSLAM (Wu and Sidle 1995;Dhakal and Sidle, 2003), distributed Shallow Land-20 slide Analysis Model, is a distributed, physically based model that combines an infinite slope model, a kinematic wave groundwater model, and a model simulating continuous changes in vegetation root strength, to analyze shallow, rapid landslides. This also includes results on the spatial distribution of safety factors in steep, forested terrain. This model is characterized by its focus on the stochastic influence of rainfall on pore water 25 pressure. SHETRAN (Ewen et al., 2000;Birkinshaw et al., 2010), which stands for System Hydrology European TRANsport, provides a hydrological and sediment transport framework for simulating landslides triggered by rain and snowmelt, along with sediment yield. In this model, the occurrence of shallow landslides is predicted as a function of the time-and space-varying soil saturation conditions, using an infinite slope model for safety analysis. SINMAP (Pack et al., 1998(Pack et al., , 2001, for Stability INdex MAPping, was developed in British Columbia with the support of the Canadian government. Compared to other 5 models, the slope stability model SINMAP has merit in that it calculates the potential slide risk for shallow translational slides via the specific-slope water-balance (Chinnayakanahalli, 2004).
TRIGRS (Baum et al., 2002 stands for Transient Rainfall Infiltration and Gridbased Regional Slope-stability. It is written in FORTRAN code and based on Iverson's (2000) linearized solution of the Richards equation, and the extension of that solution. The TRIGRS model, used for either saturated or unsaturated soils, is able to improve the effectiveness of susceptibility analysis by accounting for the transient effects of varying rainfall on conditions affecting slope stability. It has been used successfully around the world for quantitatively evaluating rainfall-triggered landslides, and 15 a number of those applications follow.
1. TRIGRS was used in a case study to account for the transient effects of rainfall on shallow landslide initiation, and verified with pilot catchments. Some examples being the Seattle area, Washington, USA ; Mt. Tenliao, Taipei, Taiwan (Chen et al., 2005); and Mt. Gaemyung, Yangjoo, Korea (Kim et al., 2010). 20 2. TRIGRS was used to evaluate and compare other physically based models including SLIP (Montrasio, 2011), SHALSTAB (Sorbino, 2010) 3. Some research has focused on parametric analyses to estimate material properties (Salciarini et al., 2006;Vieira, 2010). They proved that reasonable approximations of soil parameters, based on a limited number of measurements in the study area, were able to produce satisfactory results.
4. The TRIGRS model was augmented with a statistical technique. In the proba-5 bilistic approach with TRIGRS, the simulated landslide potential map created was generally comparable to field observations when using the Monte Carlo simulation (Liu and Wu, 2008) and the General Extreme Value probability distribution .
5. TRIGRS code has been revised and converted for specific purposes. There is . TRIGRS-P adopts a stochastic approach to compute and input parameters to be sampled randomly from a given probability distribution. MaTRIGRS offers unique computational efficiency in multi-dimensional matrix data and in real-time visualization of the simulation during modelling. 15 The main objective of this study was to predict shallow, rainfall-triggered landslides using TRIGRS, in the region of Woomyeon Mountain. The landslide ratio of each predicted class of safety factors was employed in evaluating the performance of the landslide model. Finally, this paper discusses the applicability of the flow routing model; then concludes with a discussion of the effectiveness of this approach and the poten-20 tial for further research.

Study area
The study area was Woomyeon Mountain, which is located in the Seocho district of Seoul City, South Korea ( Fig. 1 above sea level. This area is completely encircled by buildings and roads amounts to 5 104 162 m 2 and is predominantly covered by forest, mostly oak trees. The Mt. Woomyeon range is basically composed of Pre-Cambrian banded biotite gneiss and granitic gneiss as depicted in Fig. 2. The banded biotite gneiss was moderately weathered and has stripes called gneissic banding, which develops under con-5 ditions of high temperature and pressure. Because of the gneissic banding, it is clear that the study area has been exposed to extreme shearing. The soil profile can be divided into three main layers (Korean Geotechnical Society, 2011): 1. A colluvium layer extends to a maximum depth of 3.0 m from the ground level 10 and the upper part of this layer was formed from previously transported soil. This layer is generally loose material composed of gravel and silty-sand, according to the Unified Soil Classification System (USCS), a heterogeneous, incoherent and permeable soil.

2.
A transition zone is composed of mainly a clay layer (thickness is 0.2 m to 0.5 m 15 below colluvium layer) characterized by the colors taupe and dark brown. It was anticipated that landslides would be generated by conditions in this layer between the colluvium and bedrock.
3. A subsoil of stiff weathered bedrock is followed by a clay layer. This subsoil layer can be considered impervious according to the low hydraulic conductivity indi-20 cated by the modelling that follows.

Landslide event
In the area of Mt. Woomyeon, about 147 catastrophic, shallow, landslide events were triggered by a localized torrential rainfall from 26 July to 27 July in 2011. Most of the landslides were accompanied by debris flows, and mixtures of debris flowed down the Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | roadways into local communities. Sixteen people were killed and ten buildings were damaged by the debris, leading to economic losses of about US$15 million. Figure 3 depicts the locations of the damaged districts (deaths, buildings, inundated areas, landslide scarps and landslide areas) after the disaster. Figure 4 contains all the landslides and debris flows documented for this event on 5 27 July 2011. These have been registered in an official archive of disaster survey reports and publications for the government of Seoul by the Korean Society of Civil Engineers. To recognize shallow landslides, satellite images and aerial photographs taken after landslide events, as well as during field surveys, were used to spatially describe the geomorphic features of the landslide area. Figure

Description of the TRIGRS model
TRIGRS ) models rainfall infiltration, resulting from storms that have durations ranging from hours to a few days. To do so, it uses analytical solutions of partial differential equations that represent one-dimensional, vertical flow in isotropic, homogeneous materials for either saturated or unsaturated soil conditions (Fig. 5). 20 This combines the theoretical bases of the models for infiltration and subsurface flow of storm water, routing of runoff, and slope stability, to calculate the effects of rainfall on the analysis of stability over large areas. Following is a brief description of the models and formulas used by TRIGRS to represent these processes.

Infiltration model
The infiltration models in TRIGRS for initial wet conditions are based on Iverson's linearized solution of the Richards equation and extensions by Baum et al. (2002Baum et al. ( , 2008 to that solution. TRIGRS also uses a series of Heaviside step functions to implement Iverson's suggested summation of his original solution for rainfall of constant intensity, 5 to represent a general time-varying sequence of surface fluxes of variable intensities and durations. As an alternative to the solution with an infinitely deep basal boundary, Baum et al. (2002Baum et al. ( , 2008 added to TRIGRS a solution for pore pressure in the case of an impermeable basal boundary at a finite depth d LZ . The pore pressure for an impermeable basal boundary at a finite depth is given by: (2) 15 where ψ is the ground-water pressure head; Z = z/ cos δ, where Z is the vertical coordinate direction (positive downward) and depth below the ground surface, z is the slope-normal coordinate direction (also positive downward), and δ is the slope angle; d is the steady-state depth of the water table measured in the vertical direction; d LZ is the depth of the impermeable basal boundary measured in the vertical direction; where K S is the saturated hydraulic conductivity and S S is the specific storage); N is the total number of time intervals; H(t − t n ) is the Heaviside step function and t n is the time at the nth time interval in the rainfall infiltration sequence; erfc(η)is the complementary error function: series displaying odd term in complementary error function.

Slope stability model
The model of slope stability, using an infinite-slope stability analysis, is characterized by the ratio of resisting friction to gravitationally induced downslope driving stress. F S < 1 denotes unstable conditions and the depth Z where FS first drops below "1" will be 10 the depth of landslide initiation. The equation to calculate the safety factor of the slope according to the infiltration of rainfall for an infinite slope model is given by: where c is soil cohesion for effective stress; φ is the soil friction angle for effective stress; γ W is unit weight of groundwater; γ S is unit weight of soil.

Runoff model
TRIGRS computes the infiltration of each cell. The amount of infiltration, I, is the sum of the precipitation, P , and runoff from adjacent cells, R u , if the infiltration cannot exceed the saturated hydraulic conductivity, K S , as: 2013 Landslide and debris flow susceptibility zonation The runoff, R d , is calculated by the following equation: Further theoretical details of the model have been fully described in TRIGRS open file reports (Baum et al., 2002.

Description of the flow routing model 5
The model TRIGRS, uses a method for routing runoff flow cell-by-cell in the mass balance calculations. Several methods for the representation of flow directions, using rectangular grid digital elevation models, are presented later, along with flow routing features. Figure 6 shows the designation of the eight flow directions used by following flow routing model, and the numbering scheme according to ESRI direction codes in 10 ArcGIS. This approach is commonly referred to as an eight-direction (D8) flow model because of the eight valid output directions relating to the eight adjacent cells into which flow could travel. Based on the Grid DEM, there are many models for predicting the flow. 15 The earliest and simplest method for estimating flow directions, is to distribute flow from each cell to one of its eight neighbors, on the steepest downslope path. That is, flow is diverted only to the one neighboring cell that is on the steepest direction. This model, named the D8 method, was suggested by O'Callaghan and Mark (O'Callaghan and Mark, 1984), and has been widely used. In the D8 method approach, however, the 20 resulting flow distribution is irregular and somewhat unrealistic, because flow can occur in only steepest direction, either adjacent or diagonally, of eight possible directions.

The Multiple flow direction method
Multiple flow direction method (Quinn, 1991) offers advancement over the D8 method (restricts flow to one among eight possible directions). The fraction of the flow through each grid cell to each downslope direction is proportional to the gradient of each downhill flow path, so that steeper gradients will naturally attract more of the stream area.

5
Every cell surrounding the central point can be flow directions if they have lower elevations than the initial one. The pattern, of course, becomes more strongly concentrated toward the steepest downslope path. Equation (5) expresses the relative amounts for the downhill directions. 10 where n is the total number of downhill directions; ∆A i is the amount passed onto the i th downhill cell; A is the total upslope area accumulated in the current cell; tan β i is the gradient (difference in elevation/distance between the elevation values) in the i th downhill directions; L i is the contour length of the i th direction either cardinal or diagonal;

The D-infinity method
Tarboton's D-infinity method (Tarboton, 1997) assumes that water flows down one or two cells by partitioning the flow between the two cells nearest to the steepest slope direction. Figure 7 illustrates the calculation of flow directions. The single flow direction is determined by the steepest downwards slope among the eight triangular facets. This 20 direction is calculated by apportioning flow between two downslope pixels according to how close the flow direction is to the direct angle to the downslope pixel. In other words, the procedure is based on representing the flow direction as a single angle taken as the steepest downwards slope among the eight triangular facets. Like for the NHESSD Table 1 presents a summary of the flow routing methods used in simulations. The TRIGRS runoff module is used to compare the flow direction routing models above, with the observed debris flow routes in the study area.
5 Application of the model 5.1 Rainfall characteristics 10 There are two meteorological monitoring stations (Namhyun and Seocho) near the Mt. Woomyeon region. All weather stations are operated by the Korea Meteorological Administration.
The climate in the study area is mainly characterized by an average annual rainfall of 1400-1500 mm; highest in July and lowest in January. The climatic conditions occurring 15 in July 2011, however, were significantly different from the average. During July alone, Mt. Woomyeon received about 55 % of its total annual precipitation of 2039 mm. Hourly maximum rainfalls were 114 mm h −1 (07:44 ∼ 08:44 UTC on 27 July) and 87 mm h −1 (07:41 ∼ 08:41 UTC on 27 July). The first record was at Namhyun Station, the second one at Seocho Station. From the Intensity-Duration-Frequency (IDF) curves for Seoul 20 City, the rainfall recurrence intervals were 120 and 20 yr, respectively. Figure 8 shows the hourly rainfall history from 25 July to 27 July in 2011. Shallow landslides were triggered by the localized torrential rainfall during this period, characterized by a cumulative rainfall of about 350 mm, of which 42 % poured down during the last two hours (06:00 ∼ 09:00 UTC) before the failures. The landslides started at 9 a.m.

Input parameters
Many important parameters are involved in the TRIGRS model, for example, topographic factors, soil thickness, as well as strength properties and hydraulic parameters of the soil. Accuracy and reliability of the results depend mainly on a detailed knowledge of the study site, and on the quality of the input parameters.

5
For Mt. Woomyeon, topographic analyses for elevation, slope angle and aspect were calculated from 1 : 5000 maps developed by the National Geographic Information Institute. The ArcGIS was used to create grids with 10 m cells and to quantify the aforementioned information above for each cell of the DEM.
All available data were obtained from the engineering geological investigation for 10 landslides hazards restoration work conducted by the National Forestry Cooperative Federation, Korean Society of Civil Engineers and Korean Geotechnical Society. After the landslides occurred on 27 July, a total of 58 geotechnical investigation boreholes were drilled for collecting soil, hydrologic and geological information. Among these, available data from 13 boreholes and 19 soil samples were used in this analysis. The 15 locations of the investigation boreholes and profiles are depicted in Fig. 9. Determination of the soil water characteristic curve was accomplished by using pressure plate extractor and filter paper method as shown in Fig. 10. Using van Genuchten formula was best fits among several fitting equation. From the soil-water retention curves, saturated and residual volumetric water content were 50 % and 18 %, respectively.

20
Hydraulic parameters including hydraulic saturated conductivity (K S ), diffusivity (D 0 ) and steady infiltration rate (I Z ) were obtained from laboratory tests, and derived according to soil classes and empirical references. The values of D 0 and I Z were not well defined, since they had wide ranges according to the complex properties of soil (e.g., void, fine content, and soil density). These parameters are quite 25 different for various samples, even though they were collected from the same site. In the literature review about colluvium soil, D 0 was discovered to have a value about 10-500 times the value of the hydraulic conductivity. For this reason, the D 0 value was NHESSD 1,2013 Landslide and debris flow susceptibility zonation  (Liu and Wu, 2008). The information about the I Z rate, however, is rare in the literature. The I Z value is affected by soil characteristics including porosity, storage capacity, and transmission rate through the soil. The soil texture and structure, vegetation types and cover, water content of the soil and soil temperature also play a role in controlling the infiltration rate. If the soil is saturated, I Z can 5 be the same as hydraulic conductivity, while it can be zero for dry soil. In this research, the reasonable value 0.01 of the K S Liu and Wu, 2008;Kim et al., 2010) was selected for I Z , because of the hot, dry conditions during the summer of this event. In the simulations carried out in this study, uniformly homogeneous soil conditions were considered present due both to similar soil properties in reports, and 10 the presence of almost the same geological conditions. The input values, and units of the parameters for analysis, are listed in Table 2.
6 Results and discussion 6.1 Elevation, slope, aspect and curvature The simulations described below, were carried out at two levels: (i) considering land- 15 slides of the study area where 147 shallow landslides occurred during an intense rainfall event in July 2011, and (ii) taking into account the debris flow routes, but not debris mass, velocity and deposits.
In the study area, elevation, slope, aspect and curvature, all of which are relevant to landslides, were calculated from topographic information. These results for both the 20 study area, and the landslide occurrence points, are shown in Fig. 11. Bar graphs represent the percentage of the area of each category, in relation to the total study area. Polygonal lines are the ratio of the number of landslides in each category, in relation to the total number of landslides.
Seventy percent of the study area was between 50 m and 150 m elevation, and 67 % NHESSD 1,2013 Landslide and debris flow susceptibility zonation triggered on terrain with mid-to-high altitudes ranging from 100-250 m (average 119 m), and on steeper slopes (> 25 • , average: 19 • ). The aspect ratios in each category were similar to each other while the largest landslide orientation was east with 18 %. Curvature graph showed nearly the same number of concave and convex profiles in the study area, but, concave was predominant in failure spots. 5

Prediction of landslides
One main objective of this research was to evaluate the spatiotemporal predictability of landslide events in Mt. Woomyeon, using the TRIGRS model for regional landslide hazard assessment. The factor of safety (hereafter FS) over the entire study area was calculated for each cell, and plotted over time during this severe storm. Figure 12a, b 10 and c show the spatial distributions of FS in different periods of time. In other words, these depict the temporal and spatial dynamics of FS values induced by heavy rainfall during the 48 h duration (from 9 a.m., 25 July to 9 a.m., 27 July). The three FS maps are for the 0, 46, and 48 h, and 46 h corresponded to the start of extraordinarily heavy rainfall for two hours. The TRIGRS model correctly simulated the time that the landslides 15 were triggered. The areas characterized as having a safety factor close to FS = 1.0 progressively expanded when the rainfall became more intense. This implies that large numbers of the landslides were triggered by the intense rainfall. The Korean peninsula has a lot of curvy and steep nature of mountainous terrain. This is why FS maps are very complex and tortuous. Nevertheless, the performance of the TRIGRS model for 20 prediction, which has been evaluated by field investigation, can be considered reasonably applicable as shown in Fig. 12c. The success of landslide prediction models has been typically evaluated by comparing locations of measured landslides with the predicted results. Thus, a proper index or an estimator for measuring performance is essential. Most previous studies (Crosta 25 et al., 2003;Salciarini et al., 2006;Kim et al., 2010;Vieira et al., 2010) used agreement of parts (cells) between the predicted and the actual landslides to evaluate the performance of their models. However, it can be seen that the model output with more unstable areas are better than the underestimated results, since it covers more landslides. An ideal landslide assessment model simultaneously maximizes the agreement between known and predicted locations of landslides, and minimizes predicted unstable area to give useful information for prediction. In order to overcome the disadvantages and limitations of such comparisons, various indices have been proposed: SR and MSR stand for Success Rate and Modified Success Rate (Huang and Kao, 2006), ROC stands for Receiver Operating Characteristic using confusion matrix Montrasio, 2011;Raia et al., 2013), SI and EI stand for Success Index and Error Index (Sorbino et al., 2010), SC and LP stand for Scar Concentration and Landslide Potential (Vieira et al., 2010), POD, FAR and CSI stand for Probability Of Detection, False Alarm Ratio and Critical Success Index (Liao et al., 2011) and the D index (Liu and Wu, 2008).
Although the performance factors above are useful for quantifying the accuracy of a model, the precise area of known landslides is necessary for applying them. In the Mt. Woomyeon event, it is difficult to know the size of the landslides that occurred, since a 15 number of the debris flows occurred after the landslides. Figure 4a depicts this problem associated with unclear boundaries of landslides. Most landslide areas are connected with debris flow channels. This is the reason why counting landslide sites, instead of calculating landslide area, was used in the following.
In this paper, the landslide ratio of each predicted FS class (hereafter LR class ) was and debris flows. The performance value derived by LR class enables consideration of predicted stable areas as well as predicted unstable areas, and thus substantially reduces the over-prediction of landslide potential. Unlike the numerator, the number of predicted and total cells is used in denominator. The numerator, also, is the same as the SR (Success Ratio) index. Table 3 and Fig. 13 show that 2.99 % of the area was classified as unstable (FS ≤ 1.0), and that 33.33 % of the actual landslides were correctly localized within this predicted unstable area. LR FS<1 was about 11 with 33.33 % over 2.99 %. By calculating the % of LR class , we can get quantitative result. The % of LR FS<1 is 70.30 %, in other words, if a landslide happens, then predicted unstable area (FS < 1) has 70.30 % 10 chance of including the landslide. In other words, lower safety factor classes showed higher values of LR class percentages. The results show significant agreement between the simulated and known landslide map from a quantitative point of view, despite missing information of landslide area. 15 During severe storms, the failed soil mass rapidly propagates downslope and increases its initial volume through erosion of in-place soils producing a dangerous mobilized volume called a debris flow. A large number of landslides evolved into debris flows during the torrential rainfall from 26 to In this study, we suggested a debris flow routing method for the TRIGRS runoff module. This concept aims at giving a prompt assessment of debris flow path analysis at a 25 regional scale with minimum data requirement. Most debris flow susceptibility models obviously depend on a lot of information about the area of interest. Due to the complexity of debris flows relative to the modelling parameters, a simplified model was required 2564 Introduction for predicting flow paths on a regional scale. Thus, by using this module, it was possible to provide a quick, simple preliminary debris flow assessment. In order to obtain a qualitative comparison of the model presented, Flow-R (Horton et al., 2013) has been chosen as the reference method. Flow-R is a distributed, semiempirical model for susceptibility assessments of debris flow developed at University 5 of Lausanne. It assesses propagations using three critical factors of sediment availability, water input and gradient (Takahashi, 1981). The following calibrated parameters were used in the model: (1) flow accumulation and slope relationship -extreme event threshold, (2) Holmgren's exponent -4, (3) slope angle algorithm of the energy loss function -11 • , and (4) velocity threshold: 15 m s −1 (Horton et al., 2011).  First, comparing the results with the debris flow inventory map, good agreement can be found between the predicted debris flow paths and the inventory, despite the paucity of parameters for rheological properties and erosion rate. It was shown that the topography of the DEM is an important factor to determine debris flow propagation (Horton et al., 2011). In the framework of debris flow mapping, the predicted results 20 have more routes than the observed debris flow routes shown in Fig. 14a. This is because the predicted results should be representative of the worst cases; i.e., flow in every potential route even where a landslide did not occur. To solve this problem, a coupled analysis with landslide and debris flow is needed. On the other hand, this module has the advantage by predicting susceptible zone for future extreme storm.

25
The second interesting issue arises from the comparison of two runoff schemes in the module as shown in Fig. 14c and d. The runoff scheme in TRIGRS can calculate the amount of flow through each cell, and use various hydrological routing models. The results from the two model applications shown in Fig. 14  for the same topographic input, without any data related to sediment availability using geological, lithological information or land use. In Fig. 14, the width of the flow paths calculated by the Multiple flow direction method is wider than that produced by the Dinfinity method, due to the routing of flow to all the adjacent pixels of lower elevation. The module in TRIGRS allows the user to control the width using a weighting factor 5 proportional to the slope raised to an exponent. However, in the scheme to predict debris flow susceptibility, it is not important to consider the flow width.

Conclusions
This paper presented an approach to assess rainfall induced shallow landslides and debris flows in a mountainous region in Seoul, Korea. Model simulations resulted in 10 reasonable estimates of the mountain hazards based on a deterministic approach. One purpose of this paper was to suggest and verify a factor LR class (landslide ratio of each predicted FS class) for a pilot study. The biggest strength of this index is that it can estimate model performance quantitatively by minimizing the overestimated area, even in landslide-debris flow regions where the area of landslide scars is unknown or 15 ambiguous. According to the results, the percentage of LR class of unstable area is 70 %, and well reflects the effect of transient rainfall.
Another purpose was to introduce a GIS based TRIGRS runoff module for predicting debris flow routes. By applying hydrological routing models, the results provide approximate information about debris flow routes. This means that the model, originally 20 developed for landslide assessment, has also been proved applicable for susceptibility analysis of debris flow in catchments with limited data availability. It is noteworthy that the proposed approach is useful when there are only DEM and its derivative.
In summary, though TRIGRS output is dependent on the resolution of DEM and precision of the geotechnical parameters used, this paper provides a practical approach 25 for mapping the susceptibility to landslide and debris flow of each pixel in an area. This kind of approach has advantages in that: (1) it considers "dynamic" (i.e., time varying)  (Quinn, 1991) w i j = s w i j / j =n j =1 s w i j D-infinity method (Tarboton, 1997) w i 1 = ( π 4 − δ d )/( π 4 ) ,w i 2 = δ d /( π 4 ) ; if two cells w i1 = 1 ; if only one cell NHESSD 1,2013 Landslide and debris flow susceptibility zonation