1

Earthquake-induced deep-seated landslide and landscape evolution

process at Hungtsaiping, Nantou County, Taiwan

Chia-Ming Lo1* Wei-Kai Huang2 Ming-Lang Lin3

4*1 Department of Civil and Disaster Prevention Engineering, National United University, Miaoli, Taiwan.
52 Disaster Prevention Technology Research Center, Sinotech Engineering Consultants, Inc., Taipei 105, Taiwan, R.O.C.
63Department of Civil Engineering, National Taiwan University, Taipei, Taiwan, R.O.C.
7*Corresponding Author: E-mail: ppb428@gmail.com or ppb428@yahoo.com.tw

8Abstract
9

The Chi-chi earthquake (magnitude 7.6) on September 21, 1999 triggered numerous landslides

10in central Taiwan. A particularly large, deep-seated landslide occurred at Hungtsaiping, 15

11kilometers north of the epicenter. Field investigation and landscape evolution analysis revealed an
12irregular pattern of displacement vectors, indicating that this failure mechanism is highly complex.
13This paper presents the results of site investigations in conjunction with simulations of landscape
14evolution associated with this landslide. Analysis of landscape evolution was based on topographic
15maps produced at seven different times as well as aerial photographs taken between 1966 and 2006.
16Landslide site investigation included field reconnaissance, geomorphologic analysis, geophysical
17exploration, borehole logs, and laboratory experiments. Comprehensive geologic investigation led
18us to conclude that the materials involved in the landslide are colluvial deposits. Landscape
19evolution involved at least three large landslide events, two previous rockslides (the first triggered
20during the 1916 Nantou earthquake and another event occurring in 1934) as well as the 1999
21colluvium slide (triggered during the Chi-chi earthquake). By taking into account the source of the
22collapsed mass, we were able to reproduce the three landslide events using a 3D discrete element
23model. Calibration of the parameters used in the numerical model was based on the strength
24characteristics of the rock mass as well as the morphology of the landslide deposits.
25
26Key Words: the Chi-chi earthquake, deep-seated landslide, landscape evolution, 3D discrete
27element model.

281. Introduction
29

The Chi-chi earthquake that struck Taiwan on September 21, 1999 (ML 7.3) triggered over

309,000 landslides in the mountains in the central region of the country, the most severe of which
31occurred in Chiufenerhshan, Tsaoling, Chiuchifeng, the Central Cross-Island Highway, and the
32Dajia River (Liao, 2000). The landslides at Hungtsaiping in the Zhongliao Township of Nantou
33County and on Chiufenerhshan to its east were located on opposite sides of an anticline. However,
34the complex mechanism of the small displacement landslide at Hungtsaiping (28 m at the most),
35differed significantly from that of the landslide on Chiufenerhshan, which was a swift translation
36landslide with substantial sliding distance (exceeding 1 km) (Lee et al. , 2004; Chang et al., 2005;
37Lo et al., 2006; Wei and Lee, 2006; Dong et al., 2009). Surface displacements in the deep-seated
38landslide at Hungtsaiping following the earthquake were not as apparent as those at
39Chiufenerhshan. This fact and the unique surface deformations make it difficult to grasp the entirety
40of the deformation through field investigation. This prompted Lee et al. to compare aerial photos
41and employ numerical methods in their investigation (Lee et al., 2004). Tseng et al. and Lo et al.
42utilized particle image velocimetry (PIV) to analyze the displacement, identify the boundaries, and
43infer the sliding mechanisms of the Hungtsaiping landslide (Tseng et al., 2004; Lo et al., 2006).
44Unfortunately, the complexity of the geological conditions in that area (Dong et al., 2009) prevented
45them from deducing the overall landslide mechanism, sliding depth, or deformation characteristics
46using only analysis of surface displacement.
47

Field investigations, core drilling, geophysical surveys, and long-term monitoring enabled

48Chang et al., Lo et al., and Dong et al. to identify the range of three primary sliding masses (Sliding
49Mass A, Sliding Mass B, and Sliding Mass C) at Hungtsaiping during the Chi-chi earthquake (Fig.
501). Among these three sliding masses, Sliding Mass A and Sliding Mass B slid along the interface
51between an old colluvium and the bedrock. The researchers speculate that the sliding material
52originated from the deposits of earlier collapses on the upper slopes, most of which exceeded 80 m
53in thickness (Chang et al., 2008; Lo et al., 2008; Dong et al., 2009). Sliding Mass C was situated
2

54near the source of the sliding zone and the axis of the anticline, which presented numerous talus
55deposits, continuous recession along the cliff lines, and dip slope landslides. Later monitoring and
56surveys revealed severe landslide disasters throughout the range of the three sliding masses
57following torrential rain events, thereby indicating that the source area and the colluvium remain
58prone to landslides. In previous topographic maps, Lo et al. discovered large-scale changes in the
59topography at Hungtsaiping between 1904 and 1934, which led to the recent covering, collapse, and
60erosion of the old colluvium (Lo et al., 2009). Referral to earthquakes in Taiwan during the last
61century showed that topographic changes in that area were likely associated with a series of major
62earthquakes in Nantou during 1916 and 1917 (Huang et al., 2012). However, little research has been
63conducted on these historical landslide events or the evolution of the terrain.
64

Many studies have examined the mechanisms underlying the large-scale, deep-seated landslide

65at Hungtsaiping. Nevertheless, limitations in survey data and monitoring conditions still leave a
66number of matters to be investigated, and many other large-scale landslide cases have indicated that
67the kinematics and velocity characteristics are important analysis indices (Chang et al., 2005; Dong
68et al., 2009; Lo et al., 2011). During large-scale rockslide, entire sliding masses generally break up
69into numerous pieces, thereby transforming into debris avalanches (sturzstroms), are remarkable
70geological phenomenon (Steven and Simon, 2006). Kinematics during a landslide and changes in
71velocity are determinant of the landslides damage potential, which can occur within just seconds
72(Hungr, 2007). Dong et al. (2009) also mentioned that the direction of the earthquake shockwaves,
73the direction in which the sliding mass is moving, and landslide mechanisms are crucial to research
74on large-scale landslides. However, the complex topographical evolution resulting from the large75scale landslide at Hungtsaiping required numerical simulations to aid in explanation. Describing the
76various changes that occur during landslide events could help to gain understanding into the
77complex mechanisms underlying landslides, assist in planning treatment projects, and facilitate
78predictions related to the future evolution of the site.
79
3

Recently, discrete element method has been widely applied in the simulation of landslides. The
3

80advantage of this approach lies in its ability to simulate the bonds and separation of strata material
81as well as the behavior of landslides with large displacements after collision interactions. This
82approach is based fundamentally on the results of mechanical testing and field investigations.
83Velocities and displacements recorded during simulations are important parameters capable of
84gradually unveiling the process of large-scale landslides (Chang et al., 2005; Tang et al., 2009; Lo et
85al., 2011; Lo et al., 2014). Chang (2011) used PFC2D to reproduce the Hungtsaiping landslide
86caused by the Chi-chi earthquake and further integrated data from core drilling and field
87investigations to enhance descriptions of changes and deposition characteristics associated with
88recent landslides in the area. However, complex geological structures, such as the distribution of
89anticlines and synclines, regulate the direction of landslide movement, such that using a 3D
90approach to reproduce this landslide is a more reasonable approach.
91

Accordingly, we examined the history and topographical evolution at Hungtsaiping using

92topographic maps from various time periods to deal with the unique geological structures and
93complex kinematic mechanisms of this large-scale, deep-seated landslide. The use of PFC3D made
94it possible to recreate the landslides that occurred at Hungtsaiping in 1916 and 1999 in order to
95examine the landslide processes and mechanisms in detail. Therefore, the study is unique in its use
96of three-dimensional simulations to represent the sliding behavior of old colluvium under complex
97geological conditions. We also took into account the characteristics of previous earthquakes to
98present a more comprehensive history of landslides over the last century. The substantial
99availability of data and survey results provided an invaluable reference for use in clarifying how
100earthquakes induce the development of landslides. The primary objectives of this study are as
101follows:
102
103
104
105

(a) Investigating topographical changes and landslide mechanisms at Hungtsaiping over the
last century;
(b) The application of 3-D discrete element method for the re-creation of earthquake-induced
landslides at Hungtsaiping in 1916 and 1999.

106
107Fig. 1 Location of Hungtsaiping landslide area
4

1082. History of earthquake-induced landslides at Hungtsaiping

109

Hungtsaiping is located in the catchment of the Zhangping Stream, which is upstream of the

110Wu River originating at an elevation of 1079 m on Kandoushan, in the Zhongliao Township of

111Nantou County, covering approximately 2,110 ha. The Zhangping Stream is formed by the
112confluence of the Yonglu Creek and the Longfeng Creek, and the Hungtsaiping landslide area
113resides on the left bank of the Yonglu Creek (Fig. 1). Surveys following the Chi-chi earthquake
114revealed several apparent landslide areas in the Hungtsaiping area, affecting an area of 389 ha with
115substantial deposits of debris below the ridges and steep slopes. The most significant slide, which
116was approximately 10 m in length, occurred under a main escarpment. It was accompanied by
117bulging or shifted roadbeds, cracked pavement, fallen retaining walls, and rockfall. Landslide
118deposits were also observed on both banks of the Yonglu Creek. Wei and Lee (2006) employed
119aerial triangulation and 3D mapping to draw the largest landslide area in Hungtsaiping after the
120Chi-chi earthquake, which covered more than 100 ha. This large-scale, deep-seated landslide
121occurred primarily on the southeast bank of the Yonglu Creek (Fig. 1), which displayed a slope
122inclination towards the northeast. Lee et al. (2004) uncovered evidence of multiple sliding events in
123the Hungtsaiping landslide area, including tension cracks, gaps, and extrusions in the ground
124surface, shifted cliffs, compressed or collapsed slope toes, and canted trees.
125

The use of contour maps to interpret landslide terrain is a basic technique used in the

126investigation of landslides. Large-scale landslide movements generally take place within the sliding
127zones of earlier landslides, such that interpreting the topographical characteristics of old landslides
128can assist in the identification of potential landslide sites and reduce the occurrence of disasters in
129the future. Thus, to understand the geological changes and spatial distributions in the history of
130Hungtsaiping, we analyzed the variations in contour maps and water systems from various years.
131The changes in hummocky surfaces on contour maps present geomorphological characteristics
132associated with landslides in the past. We deduced the geological evolution of Hungtsaiping using
133the geomorphological changes found in contour maps of the area produced in 1904, 1934, 1977,
5

1341984, 1991, 1998, 1999, and 2003 as well as data related to earthquakes and heavy rain disasters
135over the last century.
136

Based on our analysis of the maps between 1904 and 2003 (Fig. 2) and the times of various

137landslides in the past, we divided the entire landslide area into six zones, numbered from I to VI.
138Our analysis results are as follows:
139

(a)

A comparison of the terrain in 1904 and 1934 revealed that large-scale landslides may

140

have occurred in Zones I and II between 1904 and 1934. The landslide in Zone I

141

(approximately 162 ha in area, sliding direction towards N22W) cut off the Longfeng

142

Creek, forcing it to veer north, resulting in a ninety-degree turn near Siasuiku Bridge No.

143

2. The colluvial layer that formed after the recent landslide became Sliding Masses A, B,

144

and C at Hungtsaiping. The landslide in Zone II was larger (approximately 250 ha in area,

145

sliding direction towards N83W). The 1904 map showed that upper slopes in Zone II

146

were gentler, and lower slopes steeper, whereas the 1934 map indicated the opposite. The

147

resulting deposits formed the sliding mass on the northwest side of Hungtsaiping today.

148

(b)

To understand the cause of this landslide, we referred to the research conducted by Cheng

149

and Yeh (2004) on earthquakes of the last century and found that a series of earthquakes

150

occurred in Nantou in 1916 and 1917. Since the establishment of instrumental observation

151

in Nantou in 1898, few major earthquakes have occurred. However, four severe

152

earthquakes took place starting in August of 1916, resulting in substantial alterations in the

153

terrain. Our preliminary speculation is that the earlier landslides at Hungtsaiping were

154

triggered by these earthquakes, and the significant topographical changes that ensued

155

between 1904 and 1934 help to explain the existence of thick colluvium over the bedrock

156

within the ranges of Sliding Masses A, B, and C, and also explains why the masses did not

157

result in a swift landslide following the Chi-chi earthquake.

158
159
6

(c)

Analysis of topography for the period between 1977 and 2003 revealed that the sliding
mass in Zone II was already dividing into two masses. In terms of the geological
6

160

characteristics, this zone was stabilizing. With regard to the focus of this study (the sliding

161

mass in Zone I), a new sliding mass was forming at the top of the slope behind this mass

162

in Zone III. We speculate that at this time, the sliding range of Sliding Mass C was

163

covered by colluvium comprising the sliding mass of Zone III. The upper portions of the

164

colluvium in the ranges of Sliding Masses A and B may also have contained colluvial

165

deposits involved in the landslide event following the Chi-chi earthquake. The newer

166

landslides that occurred in Zones IV, V, and VI may have been caused by earthquakes or

167

heavy rains between 1934 and 2003. Being newer, they tend to be less stable.

168

169Fig. 2 Geographical analysis of the history of Hungtsaiping

1703. Surface geological surveys and core drilling analysis

171

According to the geological map compiled by Dong et al. (2009) (Fig. 3), the exposed strata

172around Hungtsaiping includes shale, sandstone, sandstone interlaced with a thin layer of shale, and
173shale with siltstone. The exposed strata in the landslide zones mainly comprise shale and sandstone
174covered by colluvium between 30 m and 85 m thick, and inclining roughly to the northwest.
175However, the unique geological structures at Hungtsaiping result in atypical dip slope conditions, in
176which the distribution of the syncline axes controls the direction in which the strata are inclined as
177well as where old sliding masses (colluvium) accumulate (Dong et al., 2009). We therefore included
178these unique geological structures within our numerical model.
179

Based on the results of core drilling records and resistivity image profiling (RIP) (Fig.3b), the

180contact surface between the shale and colluvium (the red dashed line) is primarily a layer. Thus, our
181preliminary speculation regarding topographical evolution is that large-scale sliding on the rock
182slopes caused the accumulation of massive piles of rock debris, resulting in the formation of
183colluvium that was later shaken loose and deformed by the Chi-chi earthquake. The source of the
184sliding masses may be the moderately thick stratum of sandstone interlaced with a thin layer of
185shale, which is exposed at the clifftop. Thus, the crown is likely the same sandstone scarp currently
186exposed on the southeast side. Clearly, a natural weak plane (the thin shale layer) exists beneath the
187ten-meter-thick sandstone outcrop that was exposed in earlier times in the source area at
188Hungtsaiping. As geological forces gradually eroded the slope toe, it was easy for torrential rain
189events or strong earthquakes to trigger large-scale sliding, thereby producing the massive colluvium
190deposits observed today (Chang et al., 2011).
191

The geological cross-sections in Fig. 3c display the distribution of the syncline axis

192corresponding to the original riverbed of the Yonglu Creek in the topographic map from 1904. We
193can therefore infer that earlier landslides caused all of the syncline axes as well as the coverage of
194the original riverbed with a large volume of massive colluvium, with the result being that the river
195was pushed to the north. The syncline axes and the terrain of the original riverbed also shortened the
8

196sliding distance for the earlier landslide, such that most of the landslide material came to a rest on
197the slope face. Although the thickness of the resulting colluvium reached 30 m to 85 m, the material
198was weak in terms of strength and stiffness. On account of this, the Chi-chi earthquake merely
199caused the sliding masses to move along the rock and soil interface rather than producing a major
200landslide disaster. The above, is a description of landslides and geological evolution in the
201Hungtsaiping areas over the last century. We combined existing data to construct numerical models
202of two major landslides at Hungtsaiping (1916 & 1999) in order to provide an accurate explanation
203of the landslide history and geological evolution in the area.
204

205Fig.3 Investigation of geology and borehole logs obtained in the surveying and mapping of strata
206underlying the Hungtsaiping landslide. We also list the locations of boreholes and resistivity
207exploration (Dong et al., 2009).
208a. Geological map of the landslide area.
209b. Profile AA-AA and BB-BB across the slide directions with resistivity survey lines (Chang et
210

al., 2011).

211c. Cross-sections A-A, B-B, C-C, and D-D show the topography, geological structures, and
212

underlying strata (Dong et al., 2009).

2134. Connection between the 1916 Nantou earthquake and the 1999 Chi-chi
214earthquake
215

Few major earthquakes had occurred since the establishment of instrumental observation in

216Nantou (1898), until a series of severe earthquakes rocked the area in 1916 and 1917. The four
217primary earthquakes included the Nantou earthquake on August 28, 1916 (measuring 6.8 on the
218Richter scale at a focal depth of 45 km), the Taichung-Nantou earthquake on November 15, 1916,
219(measuring 6.2 on the Richter scale at a focal depth of 3 km), and the Puli earthquakes on January 5
220and 7 in 1917, (measuring 5.5 on the Richter scale at a focal depth of 1 km) (Fig. 4). These shocks
221resulted in 70 deaths, 98 severely injured, 208 slightly injured, 1,212 private homes completely
222demolished, 1,821 private homes partially collapsed, 6,588 private homes damaged, and 14 private
223homes buried. Large-scale landslides in numerous locations (including Hungtsaiping) brought
224unprecedented disasters to the Nantou area (Cheng and Yeh, 2004). The Chi-chi earthquake struck
225Central Taiwan in the middle of the night on September 21, 1999. It occurred 12.5 km to the west of
226Sun Moon Lake at 8 km on the highway and measured 7.3 on the Richter scale. The main quake
227and its aftershocks resulted in a total of 2,489 deaths, 50 missing, 11,465 injured, and over 106,287
228buildings completely or partially collapsed. Over 100,000 people were left homeless, and the losses
229in property damage reached NTD 14 billion, making the Chi-chi earthquake the most severe natural
230disaster in the recorded history of Taiwan (Cheng and Yeh, 2004).
231

Although earthquake records in Taiwan begin in the 20th century, complete information on

232seismic waves and velocities were not included until after 1950. The records before this year
233contain only the scale of the earthquake and the location of the epicenters, which makes later
234analysis more difficult. Wang et al. (2000) divided earthquake activity in Western Taiwan in the last
235century into two primary regions. In the northern region, seismicity forms a linear pattern trending
236in the NW-SE direction, called the Tunghsiao-Puli Linear seismic zone (TPL). The seismic zone
237south of the Choshuihsi stream near Chia-Yi is called the Chia-Yi Group seismic zone (CYG). The
238TPL seismic zone is linked to the CYG seismic zone, forming a semi-circular belt surrounding the
10

10

239PHK (Fig.4 b).

240

Seismic activity is distributed in this semi-circular belt rather than being associated with the

241Chelungpu fault, which was activated by the main shock. Even more interesting, the earthquake
242focal mechanisms for some of these significant aftershocks were nearly the same as those associated
243with the main shock (Wang et al., 2000). Thus, this study used the same location, depth, and
244magnitude of aftershocks associated with the Jiji earthquake for the simulation of PGA waveforms
245for the Nantou earthquake. Actual measurements of strong motion associated with the Chi-chi
246earthquake and aftershocks were applied to the numerical model used to simulate Hungtsaiping
247landslide events.
248

249Fig.4 Tectonic boundaries and the locations of seismic activity in the western area of Taiwan.
250a. Tectonic boundaries of Peikang Basement High (modified by Huang et al., 2001).
251b. Relationship between semi-circular belt and source locations of 1916 and 1999 earthquakes
252

11

(modified by Wang et al., 2000).

11

2535. Numerical Simulation

2545.1 Numerical Concept
255

Particle Flow Code (PFC) is based on the theory of Newtons 2nd law and discrete element

256method (Cundall and Strack, 1979). It is used to model the movement of spherical particles as well
257as their interactions and current position. Calculations related to contact between particles or
258particles and walls are detected automatically. Particles can be bonded to form various shapes
259comprising various quantities of elements. Bonding strength between two particles is in the form of
260a contact bond or parallel bond. Contact bonds transfer only force, whereas parallel bonds comprise
261a set of virtual springs with effects in the vertical as well as tangential directions, resulting in the
262transfer of stress as well as moment. These bonds tend to fracture gradually when the mass is under
263stress or moving, resulting in a separation into mutually independent masses. Displacement distance
264is free from any limits within the allowable range of the program (Tang et al, 2009).
265

The above-mentioned dynamic behavior is represented numerically using a time stepping

266algorithm in which the velocities and accelerations are assumed to be constant within each time
267step. The use of an explicit (as opposed to an implicit) numerical scheme makes it possible to
268simulate nonlinear interactions among a large number of particles without the need for a large
269amount of memory or an iterative computation method (Itasca, 2002). Thus, the time step plays a
270critical role throughout the simulation, with a considerable effect on calculations and the resulting
271system performance. This study compared landslide simulations with different time steps as they
272pertain to the length and width of the landslide zone. These results led to the selection of a time step
273parameter of 0.005 (sec/step).
274

2755.2 numerical model of Hungtsaiping landslide

276

Discrete element method does not impose any limits on the scale of separation or the

277displacement behavior of elements. Thus, the movement of masses from fracture to separation can
278be entirely simulated, making this approach highly applicable to the simulation of landslides (Poisel
12

12

279and Roth, 2004; Poisel et al., 2005). Thus, we employed PFC3D for the simulation and
280interpretation of the kinematic processes in three events associated with the Hungtsaiping landslide.
281The elements of the PFC model include particles and walls. The sliding surface of the collapse area
282in the Hungtsaiping landslide model was constructed using 87,000 wall elements based on a
2831515m DEM from the mapping of underlying strata and geological structures. Figure 5a presents a
284reconstruction of the geological structure associated with the sliding surface prior to the series of
285earthquake events in Nantou in 1904. The total length from East to West was 3,765m and the total
286width from North to South was 2,625m. The sliding mass was constructed using 20,158 spherical
287elements with a radius of 7.5 meters as well as 18,379 spherical elements divided into three blocks
288(Fig.5b, 6), resulting in a total volume of approximately 33 million cubic meters.
289

The behavior of granular media at the macroscopic scale depends on the contact mechanical

290properties; however, there is no straightforward solution to the selection of these parameters.

291Ensuring reliability in the use of PFC models as simulation tools requires the establishment of
292reasonable relationships between numerical parameters and the mechanical characteristics found in
293actual situations (Potyondy and Cundall, 2004). We adopted experimental methods for the
294optimization of PFC models with regard to uniaxial compression tests in order to calculate
295appropriate macro-parameters for model generation with the aim of obtaining an accurate
296reproduction of the macro-parameters of the rock material. These parameters included uniaxial
297compression strength (UCS), Young's modulus, and Poisson's ratio (Yoon, 2007). For bond
298modelling, macro-parameters for the interaction of two circular particles include normal and shear
299stiffness, normal and shear bonds, and the Coulomb friction coefficient, which must be obtained
300during the calibration step. Fakhimi (2004) proposed a slightly overlapping interaction of circular
301particles to resolve the failure envelope and the ratio of unconfined compressive strength to tensile
302strength, which tends to be lower than that of rock. We applied dimensional analysis in the
303calibration of DEM for Sandstone with the aim of mimicking the deformation and failure
304characteristics in stress paths (Fakhimi and Villegas, 2007). Although there is no straightforward
13

13

305solution to the problem of progressing from micro-properties to macro-parameters, there is a

306relationship between the two properties that can be employed in initial calibration. The Young's
307modulus of grains and cement are expressed as follows:
Ec

308
309

kn
4R

[1]

where kn and R are the normal stiffness and radius of particles, respectively (Potyondy and

310Cundall, 2004). Thus, Poisson's ratio depends on the ratio of shear contact stiffness to normal
311contact stiffness as well as packing geometry. The peak strength of the material depends on the
312friction coefficient as well as bond strength (Itasca, 2002).
313

Thus, we performed a series of numerical tests on the compression of granular samples in

314order to derive the mechanical properties of the rock material at the macro scale. We also obtained
315samples and geological strength index (GSI, Fig.7) in order to modify the strength properties found
316in the results of sandstone borehole logs (Chang et al., 2011), in accordance with the HoekBrown
317failure criterion. The 3D granular samples consisted of 8,574 spherical elements. The numerical
318parameters obtained for the compression test included a Young's modulus (E) as 407.21MPa and
319compression strength (UCS) as 5.249MPa with an internal friction angle of 17.63 (Tab.1). The
320macroscopic properties of the numerical sample are similar to those of the sandstone samples, as
321determined in laboratory testing (Tab.1 and Fig. 8). Table 2 lists the numerical parameters used for
322PFC modelling.
323

The numerical model of the Hungtsaiping landslide was reconstructed in the form of three

324landslide events, as follows:

325

(a)

1916 Nantou earthquake: There are no records on the seismic accelerations of this

326

earthquake event; therefore, we referred to the aftershocks of the Chi-chi earthquake that

327

were greater than 6.0 on the Richter scale and less than 10 km deep to aid in the simulation

328

of landslides at Hungtsaiping. When the seismic forces during the earthquake began

329

displacing Masses A, B, and C (Fig. 9(a)-Step 2), the large sandstone stratum began to

330

slide, thereby exposing the entire sandstone cliff near the crown of the source area. Due to

14

14

331

the distributions of the syncline axis and strata, Masses A and B moved toward the

332

syncline axis, coming to a rest when they hit Mass C. A gradual loss of bonding strength in

333

Masses A and B during the sliding and impact resulted in their breaking up into pieces and

334

led to the accumulation of a substantial amount of sandstone debris over Mass C (Fig.

335

9(a)-Step 2). The impact of Masses A and B accelerated the movement of Mass C toward

336

the northwest, which compressed and then cut off the Yonglu Creek, forcing it to veer

337

toward the northwest (Fig. 9(a)-Step 3). A significant quantity of landslide material also

338

accumulated in the new riverbed, at which point approximately between 37% and 42% of

339

the bonds in the three sliding masses had already broken. This caused the entire rock mass

340

to gradually fragment into pieces.

341

(b)

1916-1917 Nantou earthquake events: The seismic forces during the earthquakes initiated

342

the displacement of Masses A, B, and C (Fig. 9(b)-Step 4) and increased the proportion of

343

broken bonds in the three masses to between 71% and 87%. These three masses gradually

344

formed a massive colluvial deposit, mainly comprising soil centered over the syncline

345

axis. With the sliding of Masses A and B, collisions and fragmentation shifted their

346

movement in the same direction to the west (Fig. 9(b)-Step 5). Mass C continued

347

compressing the riverbed of the Yonglu Creek, which eventually deviated northwest,

348

ultimately arriving at its current location (Fig. 9(b)-Step 6). Furthermore, as Mass C

349

moved northwest, it undermined the support of Masses A and B, thereby causing them to

350

slide slightly to the northwest. Approximately 17 % of the material in the front portion

351

actually passed the erosion gullies to the west, such that they accumulated in the riverbed

352

(Fig. 9(b)-Step 6). Following the 1916-1917 Nantou earthquake events, the sliding masses

353

temporarily stabilized; however, a significant quantity of unstable colluvial material still

354

resided on the strata and syncline axis. This material eventually played a critical role in the

355

Chi-chi earthquake.

356
15

(c)

1999 Chi-chi earthquake: Following the series of earthquakes in Nantou, the primary
15

357

sliding masses at Hungtsaiping remained fairly stable for 82 years until the Chi-chi

358

earthquake caused the old sliding masses to move once again. The main landslide involved

359

the large colluvial deposit that had accumulated over the syncline axis. Small-scale

360

movements in the large colluvial deposit occurred along the rock and soil interface,

361

causing the ground surface to shift and deform (Dong et al., 2009). Seismic acceleration

362

data (Fig.10) was collected from TCU072, the seismic station nearest Hungtsaiping, for

363

conversion into the seismic parameters (velocity or displacement) required for PFC3D

364

programming. This made it possible to monitor the displacement of elements in various

365

locations (Fig. 11) and thereby strengthen our explanation of how the Chi-chi earthquake

366

moved the old landslide material. Our simulation results show that the Chi-chi earthquake

367

propelled the old sliding masses to move in roughly three directions (Fig. 9(c)-Step 7);

368

Masses A and B mainly moved west (becoming the current Sliding Mass A), and the 23%

369

of the material in the front spread over Mass C. The syncline axis divided Mass C into two

370

portions, sliding northwest and north, respectively to become the current Sliding Masses B

371

and C (Fig. 9(c)-Step 8). Despite persistent seismic forces, the sliding masses did not

372

display large-scale deformation or slide for very long distances. However, observation of

373

the elements in various locations revealed that in addition to differences in sliding

374

direction, the surface and bottom elements of the sliding masses also differed in the degree

375

of displacement, which resulted in varying degrees of geomorphological change and/or

376

destruction. During the Chi-chi earthquake, Sliding Mass A moved primarily toward the

377

west, resulting in displacement of 10.52 m at the top and 11.31 m at the bottom being,

378

which is slightly less than the displacements associated with Sliding Masses B and C. For

379

this reason, the upper structure of Sliding Mass A presented damage. In contrast, Sliding

380

Mass B moved northwest and underwent the largest displacements at the: 5.61 m at the top

381

and 19.96 m at the bottom. This led to the formation of tension near the top, which created

382

minor scarps and small ponds. Slightly bulging from compression, the slope of Sliding

16

16

383

Mass B displayed relatively little displacements of 17.54 m at the top and 19.93 m at the

384

bottom, which resulted in a small, shallow landslide near the left bank of the Yonglu

385

Creek. Finally, Sliding Mass C moved mainly toward the north with displacements of

386

19.11 m at the top and 19.83 m at the bottom.

387

Discrete element numerical modeling of the three earthquake-caused landslides provided

388important physical constraints, which were compared with field observations. This led to
389adjustments in the reported characteristics of the events. Research into the series of earthquakes in
390Nantou, as well as studies on the Chi-chi earthquake (Lo et al., 2008; Chang et al., 2011; Lee et al.,
3911999) and earlier studies associated with those earthquakes provided abundant background
392information with which to investigate possible factors governing the kinematics of these large
393earthquake-induced landslides. Based on previous studies using numerical simulation, we can
394discuss in greater detail the possible reasons for the movement of slide masses following these
395events. A comparison of the simulation results related to the Chi-chi earthquake obtained using
396PFC3D with particle image velocimetry technology (PIV) (Lo et al., 2008) revealed that the
397boundaries and movement direction of colluvium were nearly a perfect match (Fig.12). We found
398evidence of a Syncline, i.e., M-symmetry minor folds of thin-bedded siltstone and sandstone layers
399within a massive shale (Tanliaoti Shale) outcrop on the river bank of Yonglu stream, which appears
400to have played a crucial role in these events. Between 1904 and 1934, it played an important role in
401determining the direction of sliding (Fig. 9(a) mass B and mass C) and deflected the direction of the
402Yonglu stream toward the northwest. This exposed the toe of Sliding Mass A, which resulted in
403instability and eventual failure. The Chi-chi earthquake caused slip displacement in Sliding Masses
404A, B, and C of approximately 25.61 meters between 1998 and 2003. The conditions associated with
405phenomenon are worth discussing. In the simulation, the areas surrounding ball IDs 2020 and 2037
406presented large displacement of approximately 19.1 m. In comparison, the area surrounding ball IDs
4077289 and 7756 on the lower slope had displacement of only 1.48 m. The conditions that led to such
408large differences in displacement may have caused deformations in the mass of the colluvium,
17

17

409resulting in the formation of tension cracks on the sliding surface.

410
411Table 1 Comparison results of uniaxial compression test in accordance with HoekBrown failure
412criterion and PFC model
413Table 2 Numerical parameters used in PFC modeling
414Fig.5 3D illustration of geological layers based on the mapping of underlying strata
415a. Mid-thick sandstone showing geological syncline and anticline structures
416b. Shale: Reconstruction of sliding surfaces in Fig 3 with modifications based on topographic
417

maps from 1904. Also shown is the scope of possible landslide sources: masses A, B, and C

418Fig.6 Numerical modeling of the Hungtsaiping landslide

419Fig. 7 Geological strength index and distribution of samples obtained during field investigation
420Fig. 8 The results of simulation uniaxial compression test
421Fig. 9(a) Simulation results of landslide event associated with the series of earthquakes in Nantou
422in 1916
423Fig. 9(b) Simulation results of landslide event associated with the series of earthquakes in Nantou
424in 1916-1917
425Fig. 9(c) Simulation results of old landslide caused by Chi-chi earthquake
426Fig.10 Seismic records at TCU072 in (a) the vertical, (b) North-South, and (c) East-West directions
427Fig.11 (a) the monitoring location of numerical model; (b) the displacement of ball elements
428Fig.12 Comparison results of simulation of Chi-chi earthquake events
429a., b. Morphologic changes were identified by studying surface displacement using particle image
430

velocimetry technology (PIV) to obtain detailed landslide zonation from topographic maps

431

(Lo et al., 2008).

432c., d. Surface displacement identified using PFC 3D for the simulation of kinematics at the site of
433the Hungtsaiping landslide

18

18

4346. Conclusions
435

This study examined topographical changes and landslide mechanisms of a large-scale

436landslide area at Hungtsaiping, using topographic maps from the previous century as well as onsite
437surveys, drilling logs, geophysical data, and the findings of long-term monitoring. We simulated
438earthquake-induced landslides in 1916, 1916-1917, and 1999 using PFC3D to clarify the sliding and
439deposition behavior of the large-scale landslides at Hungtsaiping under complex geological
440conditions. The results of terrain analysis show that the series of earthquakes in Nantou in 1916 and
4411917 caused massive landslides in the Zone I between 1904 and 1934, which cut off the Younglu
442Stream and forced its channel to move north (presenting a right angle bend). This produced the
443current Sliding Masses A, B, and C and covered the strata in the study area in colluvial material to
444depths of 30 m to 85 m. The sliding that took placed in Zone II during this time created the colluvial
445layer that exists to the northwest of Hungtsaiping today. Geological surveys revealed that due to the
446atypical cataclinal slope conditions formed by the unique geological structure at Hungtsaiping, the
447syncline axis determined the direction of strata inclination and the distribution of the colluvial layer.
448Sliding Masses A, B, and C were situated on the syncline axis, which prevented longer runout
449distances. As a result, the Chi-chi earthquake caused the three masses to slide only 20 m to 35 m
450along the stratum-soil interface and did not lead to a large-scale landslide disaster.
451

PFC3D simulations that the earthquake events from 1916 to 1917 produced large-scale

452rockslide on the cataclinal slopes, primarily involving sandstone, which exposed the entire cliff in
453the source area. During the sliding process, the terrain caused the strata to fragment and break up on
454a large-scale, thereby forming a colluvium layer that was deposited on the syncline axis and
455changing the original path of the Younglu Stream to the valley terrain observed today. In 1999, the
456Chi-chi earthquake caused sliding between the colluvial layer and the bedrock on the syncline axis,
457in which Sliding Masses A and B moved westward, and Sliding Mass C moved northwest and north
458as divided by the syncline axis. Fortunately, the Chi-chi earthquake caused deformation only in the
459colluvial layer and small-scale sliding of 10 m to 25 m, which did not result in the formation of
19

19

460landslide dams or large-scale disasters. In the future, the Hungtsaiping sliding area should continue
461to be monitored and analyzed.

20

20

462References
463[1] Chang K.J., Taboada A., Chan Y.C. (2005) Geological and morphological study of the
464

Jiufengershan landslide triggered by the Chi-Chi Taiwan earthquake. Geomorphology 71(3-4):

465

293-309.

466[2] Chang, K.J., Taboada, A., Lin, M.L., and Chen, R.F. Analysis of landsliding by earthquake
467

shaking using a block-on-slope thermo-mechanical model: example of Jiufengershan landslide,

468

Central Taiwan, Engineering Geology 80: 151163 (2005).

469[3] Chang K.T., Lai C.C., Lin D.G. (2008) Analyses of the Huntsaiping Landslide Induced by the
470

Chichi Earthquake, Journal of Chinese Soil and Water Conservation, 39(3): 329-344 (in

471

Chinese).

472[4] Chang, K.T., Lin, M.L., Dong, J.J., Chien, C.H. (2011) The Hungtsaiping landslides: from
473

ancient to recent, Landslides (3 September 2011): 1-10.

474[5] Cheng, S.N., and Yeh, Y.T. (2004) Taiwan centuries earthquake, Journal of Scientific
475

Development 373: 68-75.

476[6] Cundall P.A. (1971) A computer model for simulating progressive large scale movement in
477

blocky rock systems. Proceedings of the Symposium of the International society of rock

478

mechanics, 1. Nancy, France: II-8.

479[7] Cundall P.A., Strack O.D. (1979) A discrete numerical model for granular assemblies.
480

Geotechnique 29: 47-65.

481[8] Dong J.J., Lee W.R., Lin M.L., Huang A.B., Lee Y.L. (2009) Effects of seismic anisotropy and
482

geological characteristics on the kinematics of the neighboring Jiufengershan and

483

Hungtsaiping landslides during Chi-Chi earthquake. Tectonophysics 466 (34): 438457.

21

21

484[9] Fakhimi A., Villegas T. (2007) Application of dmensional anlysis in calibration of a discrete
485

element model for rock deformation and fracture. Rock Mechanics and Rock Engineering 40:

486

193-211.

487[10] Hoek, E., Carranza-Torres, C. and Corkum, B. (2002), Hoek-Brown Failure Criterion-2002
488

Edition, 5th North American Rock Mechanics Symposium and 17th Tunneling Association of

489

Canada Conference: NARMS-TAC: 267-271.

490[11] Huang C.Y., Xia K.Y., Yuan P.B., Chen P.G. (2001) Structural evolution from Paleogene
491

extension to Latest Miocene-Recent arc-continent collision offshore Taiwan, J. Asian Earth Sci

492

19: 619-639.

493[12] Huang, W.K., Lo, C.M., Lin, M.L., Dong, J.J., Chang, K.T., Chang, Y.L. (2012) Effects of
494

geological characteristics on the kinematics of historical Hungtsaiping landslide, 2012 Taiwan

495

Rock Engineering Symposium, Miaoli, Taiwan: 679-688.

496[13] Hungr, O. (2007) Dynamics of rapid landslides, Progress in Landslide Science (pp. 47-56),
497

Chapter 4.

498[14] Itasca, Consulting Group Inc. (2002) PFC3D Particle Flow Code in 3 Dimensions. User's
499

Guide. Minneapolis.

500[15] Lee, J.F., Wei, C.Y., and Huang, C.C. (2004) The study of Hungtsaiping landslide using digital
501

aerial photogrametric technique, Proceedings International Symposium on Landslide and

502

Debris Flow Hazard Assessment: 5-1 - 5-9.

503[16] Lee W.H.K., Shin T.C., Kuo K.W., Chen K.C., Wu C.F. (1999) CWB free field strong-motion
504

data from the 921 Chi-Chi earthquake: volume 1. Digital Acceleration Files on CD-ROM.

505[17] Liao H.W. (2000) Landslides triggered by Chi-Chi earthquake. Master's thesis, Institute of
506

22

Geophysics, National Central University (in Chinese).

22

507[18] Lo, C.M., Lee, K.C., Lee, W.C., Lin, M.L., and Jeng, F.S. (2006) Landslide zonation of
508

Hungtsaiping area based on aerial photograph and PIV technology, Proceedings of 2nd Japan-

509

Taiwan Joint Workshop on Geotechnical Hazards from Large Earthquakes and Heavy Rainfall,

510

Nagaoka, Niigata, Japan: 3036.

511[19] Lo C.M., Lin M.L., Lee W.C., Lee K.C., Chien S.Y., Dong J.J., Chang K.T., Huang A.B.
512

(2008) Landslide characterization and zonation of Hungtsaiping area based on topography,

513

aerial photograph and PIV technology. The 3rd International Conference on Site

514

Characterization, Taipei, Taiwan: 467-472.

515[20] Lo, C.M., Lin, M.L., Dong, J.J., Chang, K.T., Chien, S.Y., Huang, A.B. (2009) Landslide
516

Characterization and Zonation of Hungtsaiping Area based on Topography, Iimage of Rrmote

517

Sensing, and PIV Technology, Journal of the Chinese Institute of Civil and Hydraulic

518

Engineering 21(2): 395-410 (in Chinese).

519[21] Lo C.M., Lin, M.L., Tang C.L., Hu, J.C. (2011) A kinematic model of the Hsiaolin landslide
520

calibrated to the morphology of the landslide deposit, Engineering Geology 123: 22-39.

521[22] Lo C.M., Lee C.F., Chou, H.T., Lin, M.L. (2014) Landslide at Su-Hua Highway 115.9k
522

triggered by Typhoon Megi in Taiwan, Landslides 11(2):. 293-304.

523[23] Poisel R., Roth W. (2004) Runout models of rock slope failures. Felsbau 22: 46-50. Poisel R,
524

Bednarik M, Holzer R, LIK P (2005) Geomrchanics of Hazardous landslides. Journal of

525

Mountain Science 2(3): 211-217.

526[24] Potyondy D., Cundall P. (2004) A bonded-particle model for rock. International Journal of
527

Rock Mechanics & Mining Sciences 41: 13291364.

528[25] Steven N.W., Simon D. (2006) Particulate kinematic simulations of debris avalanches:
529

interpretation of deposits and landslide seismic signals of Mount Saint Helens, 1980 May 18.

530

International Journal of geophysics 167: 991-1004.

23

23

531[26] Tseng, C.H., Hu, J.C., Chan, Y.C., Lee, J.F., and Wei, J.Y. (2004) A slow slide in central
532

Taiwan induced by chi-chi earthquake revealed by PIV analysis, American Geophysical Union

533

journal-T11D-1307: 1-4.

534[27] Tang C.L., Hu J.C., Lin M.L., Angelier J., Lu C.Y., Chan Y.C., Chu H.T. (2009) The Tsaoling
535

landslide triggered by the Chi-Chi earthquake, Taiwan: insights from a discrete element

536

simulation. Eng Geol 106: 1-19.

537[28] Wang C.Y., Chang C.H., Yen H.Y. (2000) An interpretation of the 1999 Chi-Chi earthquake in
538

Taiwan based on the thin-skinned thrust model, TAO 11: 609-630.

539[29] Wei C.Y., Lee J.F. (2006) The application of digital aerial photography in the study of
540

Hungtsaiping landslide, Chungliao, Nantou County. Bulletin of the Central Geological Survey

541

19: 39-59 (in Chinese).

542[30] Yoon J (2007) Application of experimental design and optimization to PFC model calibration
543

in uniaxial compression simulation. International Journal of Rock Mechanics & Mining

544

Sciences 44: 871-889.

24

24