See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/222667289

Impacts of the Chi-Chi earthquake on

subsequent rainfall-induced landslides in
central Taiwan

Article in Engineering Geology August 2006

DOI: 10.1016/j.enggeo.2006.02.010

CITATIONS READS

116 252

4 authors, including:

Ching-Weei Lin Cheng-Chien Liu

National Cheng Kung University National Cheng Kung University
57 PUBLICATIONS 1,409 CITATIONS 54 PUBLICATIONS 670 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Ching-Weei Lin on 27 April 2015.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
 Engineering Geology 86 (2006) 87 101
www.elsevier.com/locate/enggeo

Impacts of the Chi-Chi earthquake on subsequent rainfall-induced

landslides in central Taiwan
Ching-Weei Lin a,b,, Shou-Heng Liu b , Sen-Yuan Lee b , Cheng-Chien Liu a,b
a
Department of Earth Sciences, National Cheng Kung University, Taiwan, ROC
b
Disaster Prevention Research Center, National Cheng Kung University, Taiwan, ROC
Received 13 December 2004; accepted 20 February 2006
Available online 27 April 2006

Abstract

The influence of the Chi-Chi earthquake on subsequent rainfall-induced landslides was evaluated by comparing the occurrence
of landslides in the Choushui River watershed through eight SPOT images that covered the period from 1996 to 2001. The Chi-Chi
earthquake not only triggered serious coseismic landslides itself but also extensively disturbed surface strata around the epicentral
area. After surface strata have been highly disturbed heavy rainfalls in 2000 and 2001 have triggered more landslides than the
earthquake itself. Compared with the data obtained from images before the earthquake, it is obvious that the density of rainfall-
induced landslides increased significantly after the earthquake, and the places where landslides occurred changed, as well.
Although typhoon Herb in 1996 brought more precipitation than typhoon Toraji in 2001, only 9.77 km2 landslides were induced by
Herb, while 48.8 km2 landslidesalmost five times larger than those caused by Herbwere triggered by Toraji. Besides, landslides
induced by Herb were mainly distributed in mid-slope areas, with slopes of 2030. After the Chi-Chi earthquake, the rainfall-
induced landslides occurred mainly in places with slopes between 40 and 50.
2006 Published by Elsevier B.V.

Keywords: Chi-Chi earthquake; Landslide; SPOT image; Choushui River; Taiwan

1. Introduction induced landslides, many studies have been concerned

In mountainous areas, landslides and debris flows
are major natural hazards and threats to both human
lives and environmental ecology. Previous studies
have shown that rainfall and earthquakes are two main
mechanisms that trigger landslides (Keefer, 1984;
Schuster et al., 1996; Crosta, 2004). For earthquake-
Corresponding author. Department of Earth Sciences, National
Cheng Kung University, 1 Ta-Hsueh Road, Tainan, Taiwan 701,
Republic of China. Tel.: +886 6 2757575x65427; fax: +886 6 274
0285.
E-mail address: chingwee@mail.ncku.edu.tw (C.-W. Lin).
0013-7952/$ - see front matter 2006 Published by Elsevier B.V.
doi:10.1016/j.enggeo.2006.02.010
with the identification and description of coseismic
landslides, particularly those caused by catastrophic
earthquakes (see, for example, Keefer, 1984; Harp et
al., 1991; Jibson et al., 1994; Harp and Jibson, 1996;
Khazai and Sitar, 2004). However, a catastrophic
earthquake always intensively disturbs ground strata
and affects the stability of slopes for a long period of
time. Therefore, to predict the landslide behavior for a
region that has suffered a catastrophic earthquake it is
necessary to keep tracking the occurrences and density
of rainfall-induced landslides for a longer period.
Unfortunately, far less research work has been done on
this topic.
88 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
Located on an active mountain-building region and
having experienced the catastrophic Chi-Chi earthquake
in 1999, Taiwan provides an excellent opportunity to
investigate how a catastrophic earthquake affects
subsequent rainfall-induced landslides. In this study,
the Choushui River watershed, located near the
epicenter of the Chi-Chi earthquake, is selected as the
study area. Landslides identified from eight SPOT
images that cover the period 19962001 are used to
differentiate the variations of landslide occurrence prior
to and after the earthquake.
To understand the impact of the Chi-Chi earthquake
on the occurrence of subsequent rainfall-induced land-
slides, this study mainly focuses on the variations in the
characteristics of rainfall-induced landslides prior to and
after the earthquake. We are especially concerned with
(1) the variations in the density of landslides, (2) the
variations in the places where landslides occurred, (3)
the influence of lithology on the density and the places
of landslide occurrence, and (4) the impact of ground
motion on the occurrence of landslides.
2. Study area
The Choushui River watershed in central Taiwan,
which is the largest drainage basin on the island, was
selected as the study area (Fig. 1). Previous studies (Lin
and Jeng, 2000, Lin et al., 2004) have shown that this
watershed is one of the most susceptible areas for
landslides and debris flow hazards on the island. In the
Choushui watershed, the horizontal peak ground
acceleration (PGA) induced by the Chi-Chi earthquake
was in the range of 100 to 1200 gal, and over 70% of the
area suffered horizontal PGA over 600 gal (Fig. 1).
Thus, the Choushui River watershed suffered violent
shaking, and the ground was highly disturbed during the
earthquake. Besides coseismic landslides, serious land-
slides and debris flows were also triggered in the
watershed during the heavy rainfall brought by typhoon
Toraji after the earthquake. Therefore, the Choushui
River watershed probably is one of the best places to
study the impact of an earthquake on the occurrence of
subsequent rainfall-induced landslides.
Before the Chi-Chi earthquake, severe landslides had
been triggered by several major catastrophic earth-
quakes in the study area (Cheng et al., 1999). The 1906
Meishan earthquake (ML = 7.1; at 23.55N, 120.45E),
with a focal depth of 6.0 km, generated surface ruptures
of over 13 km and caused severe landslides in the
southern part of the Choushui River watershed. The
1941 Chungpu earthquake (ML = 7.1; at 23.40N,
120.47E) induced intensive landslides in the water-
shed. The Tsaolien landslide, similar to the landslide that
occurred at the same place in the Chi-Chi earthquake, is
the most famous example. Over 1.5 108 m3 of debris
slid down to the Chinshui River and formed a debris
dam during the earthquake. The 1998 Reili earthquake
(ML = 6.2; at 23.5N, 120.7E) caused some landslides
around the southern boundary of the study area that
were also reported. However, no quantitative data were
recorded for the 1906 and 1941 earthquakes, and
landslides triggered by the 1998 earthquake only
occurred in a small epicentral area near the southern
boundary of the Choushui River watershed.
The physiography of the Choushui River watershed
is primarily composed of a series of approximately NS
trending mountain ranges, including the Yushan and
Central Range, the Chunda Range, and the Alishan
Range, and some NS trending tributaries, including the
Chunda River, the Chenyulan River, and the Chinshui
River (Fig. 2). The Central Range, with a maximum
elevation of over 3900 m, forms the backbone ridge of
the island. A series of EW trending rivers initiate from
the backbone range and join the NS trending Chunta
and Choushui Rivers when they meet the southern
extension of the Hsuehshan Range and the Chunda
Range. The Chenyulan River, which closely follows the
Chenyulanchi fault line, is the major river flowing
through the area between the Chunda Range and the
Alishan Range (Fig. 2). The Chinshui River mainly
flows through the area between the Alishan Range and
the low hills of the Western Foothills.
Almost all important geological units in the Western
Foothills and the Western Central Range are present in
the study area. Four geologically distinct areasthe
Western Central Range, the inner belt of the Western
Foothills, the outer belt of the Western Foothills, and the
Coastal Plaincan be differentiated in the Choushui
River watershed (Fig. 3). For convenience of discussion,
the exposed rocks in the study area are grouped (roughly
according to their age and mechanical behavior) into six
stratigraphic units, including, in ascending order,
OligoceneEocene metamorphic sandstone, Miocene
Eocene slate, Miocene sedimentary rocks, Pliocene
sedimentary rocks, Pleistocene sedimentary rocks, and
HolocenePleistocene terrace and alluvium deposits
(Table 1). The Western Central Range primarily consists
of Miocene and Pre-Miocene dark gray, well-cleaved
slates inter-bedded with lenticular, thick-bedded to
massive metamorphic sandstone. The inner belt of the
Western Foothills is the region bounded by the
Chelungpu fault in the west and the metamorphic slate
belt in the east, and it is mainly composed of well-
cemented, Miocene sandstone and shale. The outer belt
 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
Fig. 1. (a) Landslides identified from the SPOT image taken after the Chi-Chi earthquake in the Choushui River watershed. (b) Landslides identified from the SPOT image taken after typhoon Toraji in
the Choushui River watershed. Solid line represents surface rupture of the Chi-Chi earthquake. Isolines represent PGA in gal. Black dots are landslides. Solid circles indicate the locations of strong
ground motion stations. The epicenter is indicated by a star.

89
90 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
Fig. 2. 3-D topographic map of the study area.
of the Western Foothills is primarily composed of
weakly cemented, Pliocene to Pleistocene sandstone,
shale, and conglomerate. The Coastal Plain is mainly
covered by PleistoceneHolocene sediments.
3. Study methods
In this study, a geographic information system was
used to conduct a spatial characterization of landslides.
Landslides in the study area were mainly identified from
eight SPOT4 multi-spectral images with 12.5 m pixel
resolution (Table 2). These images were taken after a
typhoon or heavy rains with cumulative precipitations of
over 100 mm (Table 3), except for the image G09619,
which was taken before the Typhoon Herb. On a SPOT
image, shallow debris slides are the easiest and most
reliable type of sliding phenomena to be detected
because they strip off the vegetation cover and are thus
readily discernable. Therefore, in this study, the shallow
debris slide is the main sliding phenomena recognized in
SPOT images. To prevent misinterpretation of landslides
in images, only landslides with areas over 4 pixels
(representing areas larger than 625 m2) were recognized.
However, if vegetation of deep-seated slides, such as the
Tsao-Liang landslide, and lateral erosion along the gully
bed caused by debris flows were stripped off too, they
were also included. In addition, in order to make
comparisons between different images, landslides rec-
ognized in previous images are always treated as
landslides in later images, even if the vegetation had
recovered. Besides the area of landslides, to make
comparisons with the results of similar studies conducted
elsewhere in the world (e.g., Keefer, 2000), variations in
landslides were also presented in terms of landslide
concentration (LC), expressed as the number of landslide
sources per square kilometer of surface area of a given
interval of slope, stratigraphic unit, or level of ground
shaking.
After the SPOT images were interpreted, all data
were superimposed on a geological map, the Digital
Terrain Model (DTM), and the ground motion of the
Chi-Chi earthquake in the Arc View 8.2 geographic
information system to conduct a spatial characterization
of the landslides. Because the 1:50,000 quadrangle
maps from Central Geological Survey only covered one-
third of the study area, the geological map used was
modified from the 1:100,000 Chiayi and Taichung maps
published by the Chinese Petroleum Cooperation. A
40 m 40 m grid DTM created by the Center for Space
and Remote Sensing Research at National Central
University in 1989 was used to calculate the slope
angles of landslides by averaging the slopes of all pixels
that fell within a given landslide. The ground motion of
Chi-Chi earthquake was calculated by using cubic spline
interpolation from data recorded at 390 stations of the
strong ground motion network established by the
 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101 91

Fig. 3. Simplified geological map of the Choushui River watershed. Star is epicenter of the Chi-Chi main shock. At upper right corner, the focal
mechanism (from Central Weather Bureau) of the Chi-Chi earthquake main shock shows thrust motion is dominant in the quake.

Central Weather Bureau, and the topographic amplifi-
cation effect was not incorporated in the calculation.
4. Rainfall condition in the Choushui River
watershed
According to the recoded data between 1965 and
2001 (Water Resources Agency, 2001), the annual
precipitation in the Choushu River watershed varied
from a low of 1411 mm in 1980 to a high of
2915 mm in 1998. The annual mean for the 6 year
period from 1995 to 2001 was 2200 mm (Table 4).
The rain season of the year is from May to September,
and the dry season runs from October to March of the
next year. Heavy rainfall brought by typhoons
commonly occurs in July, August, and September,
and particularly in August. Mean monthly precipita-
tion during the rain season is in the range of 284 mm
to 387 mm, and in the dry season it is in the range of
24 mm to 105 mm (see Table 4).
Due to the discrepancy in topography, rainfall
distributed in the Choushui River watershed is highly
varied, which can be seen from the data on precipitation
brought by typhoons during the period 19962001 (see
Table 3). Within the Choushui river watershed, mean
annual precipitation is about 1800 mm in the west part
and over 2200 mm in the east part of the watershed.
Along the eastern and southeastern boundaries of the
watershed, average annual rainfall even reaches
4000 mm.
During the duration 1 month before and 4 months
after the Chi-Chi earthquake, there was no typhoon or
storm that brought heavy, widespread rainfall to the
study area. Before the Chi-Chi earthquake, only two
local rainfall events in the watershed, one on August 23
24 and the other on September 1315, brought
cumulative precipitation of over 100 mm in a small
region. The recorded maximum cumulative precipitation
is 193 mm in 18 h on August 2324, and 159 mm in 27 h
on September 1315. After the Chi-Chi earthquake, no
92 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101

Table 1 5. Major disaster events occurred in the Choushui

Stratigraphic units in the Choushui River watershed River watershed from 1996 to 2001
Simplified Stratigraphic Lithology
stratigraphic unit unit The Choushui River watershed has been recognized
HolocenePleistocene Loose sands, silts, and as the most susceptible region for landslide and debris
sediments and gravels. flow hazards on the island. During the period from 1996
terrace deposits
to 2001, seven typhoons brought mean cumulative
Pleistocene Toukoshan Massive to thick conglomerate
sedimentary rock Formation with 50% quartzite and rainfalls of over 100 mm in the Choushui river
sandstone fragments in the watershed (see Table 3). In the same period, three
upper part. In the lower part, major eventsthe typhoon Herb, the Chi-Chi earth-
massive, fine-grained, light quake, and the typhoon Torajihave caused serious
gray, thick-bedded to massive,
landslides, debris flow hazards, and heavy causalities in
loosely cemented sandstone
with intercalations of shale. the study area.
Pliocene sedimentary Cholan Light gray, fine-grained, On July 29, 1996, typhoon Herb, with wind speeds
rock Formation thick-bedded to massive, up to 53 m/s and a radius of 350 km, brought record-
loosely cemented sandstone breaking precipitation to central Taiwan. The Choushui
with thin intercalations of
River watershed experienced a cumulative precipitation
shale.
Chinshui Dark gray, thick-bedded to in the range of 101 mm to 1994 mm during typhoon
Shale massive shale with lenticular Herb. On one day (July 31) and two consecutive days
siltstone and mudstone. (July 31 to August 1), at the Alishan weather monitoring
Miocene sedimentary Yutengping Light gray, fine-grained, station, near the southern boundary of the study area,
rock Sandstone thick-bedded to massive
rainfall intensities of 1094.5 mm and 1986.5 mm were
sandstone inter-layered with
thin- to thick-bedded shale. recorded, respectively (Lin and Jeng, 2000). The
Shihliufeng Greenish gray to dark gray precipitation brought by Herb caused 51 deaths, with
Shale massive sandy shale. 22 others missing and presumed dead. The watershed
Kuantaoshan Gray, fine-grained, was reported as one of the areas most severely damaged
Sandstone thick-bedded to massive,
by typhoon Herb (Lin and Jeng, 2000). More than 32
clayey sandstone.
Nanchuang Light gray, fine-grained, debris flows were reported to have occurred along the
Formation thin-bedded to thick-bedded main tributary, in the Chenyulan River watershed (Lin et
alternations of sandstone al., 2004).
and dark gray shale. The 1999 Chi-Chi earthquake (ML = 7.3, Mw = 7.6, at
Nankang Inter-bedded, thin to thick
Formation layers of light, fine-grained
23.85N, 120.81E), with a focal depth of 8.0 km, was
sandstone and dark gray shale. triggered by reactivation of the Chelungpu fault in
MioceneEocene Lushan Black to dark gray argillite, central Taiwan on September 21, 1999. The earthquake
slate Formation slate, and phyllite with caused 2400 deaths, 8373 casualties, and over US$10
occasional inter-beds of billion in damages. It not only produced surface ruptures
sandstone and disseminated
marly nodules.
Hsinkao Black argillite, slate, and Table 2
Formation phyllite with occasional Variations in area of different types of landslides in each SPOT image
inter-beds of sandstone and
conglomerate band. SPOT Date All Incremental Enlarged New
OligoceneEocene Paileng Massive to thick beds, white, landslides landslides landslides landslides
metamorphic Formation fine- to coarse-grained (km2) (km2) (km2) (km2)
sandstone quartzitic sandstone with G09619 1996/06/05 9.56
inter-beds of dark argillite 1996/07/29 Typhoon Herb
or slate. G14166 1996/11/08 19.33 9.77 3.39 6.38
G14475 1997/12/15 27.46 8.13 5.08 3.05
G13804 1998/10/12 32.12 4.64 2.74 1.90
G12561 1999/01/05 35.52 3.40 2.73 0.67
significant rainfall event was recorded. Cumulative 1999/09/21 Chi-Chi earthquake
rainfalls in the range of 20 mm to 50 mm in 820 h G14162 2000/01/08 57.37 21.85 14.38 7.47
were recorded at a few stations on October 78. Mean G14525 2001/03/05 100.39 43.02 30.40 12.62
monthly precipitation from October of 1999 to January 2001/07/28 Typhoon Toraji
of 2000 is in the range of 13 mm to 64 mm. G14806 2001/08/12 148.81 48.42 34.30 14.12
 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101 93

Table 3
Rainfall data for typhoons with mean cumulative precipitation over 100 mm in the study area during the period 19962001
Typhoon Date Rainfall duration (h) Cumulative rainfall (mm) Rainfall intensity (mm/h)
(mm)
Mean Range Mean Range Mean Range
Herb 1996/07/29 44.8 1079 486.0 101.01994 41.6 12.072.0
Winnie 1997/08/16 25.5 333 131.6 50.0366.0 15.1 6.054.5
Amber 1997/08/27 27.0 350 149.2 54.0600.0 24.8 8.569.5
Otto 1998/08/03 24.8 1636 176.8 53.5432.0 41.5 10.0143.0
Zeb 1998/10/10 38.4 2755 277.9 83.5781.0 29.1 8.092.5
Bilis 2000/08/21 41.57 970 203.4 38.0913.0 21.3 5.573.0
Toraji 2001/07/28 22.1 1531 363.8 88.0634.0 69.1 24.5140.5
Rainfall data were collected and calculated from 46 rainfall recording stations within and around the Choushui River watershed.

over 100 km in length but also caused serious coseismic
landslides in central Taiwan (e.g., Kaima et al., 2000;
Lin et al., 2004). Over 20,000 landslides, with surface
areas totaling approximately 113 km2, occurred in an
area of 2400 km2 in central Taiwan. More than 90% of
the landslides were smaller than 0.01 km2 in scale, and
most were shallow debris slides, although a few were
large and deep-seated.
On July 28, 2001, typhoon Toraji, with wind speeds
up to 38 m/s and a radius of 250 km, struck Taiwan.
Total precipitation in the Choushui River watershed was
in the range of 88 mm to 634 mm, and a maximum
hourly rainfall intensity of 140 mm/h was recorded. The
typhoon caused 103 deaths, with 111 missing and
presumed dead and 189 injured. Serious landslide and
debris flow hazards were reported to have occurred in
central and eastern Taiwan (Lin et al., 2004). In the
Choushui River watershed alone, 290 debris flows were
reported.
6. Coseismic landslides associated with the Chi-Chi
earthquake
A large number of landslides were triggered by the
Chi-Chi earthquake in the study area. Most of them were
located on the hanging wall of the Chelungpu fault (Lin
et al., 2000; Khazai and Sitar, 2004). Several studies
have focused on these landslides to discuss the
characteristics of earthquake-induced landslides on a
regional scale (e.g., Lin et al., 2000; Khazai and Sitar,
2004; Lin et al., 2004), and some studies (Shou and
Wang, 2003; Wang et al., 2003) have focused on
specific, large, deep-seated landslides such as the Tsao-
Liang landslides and the Chiu-Fen-Erh-Shan landslide.
Here, we briefly review the results of the studies on
landslides on a regional scale.
On the basis of the Chi-Chi earthquake Database
Analysis and Management System, which includes
9272 landslides identified from SPOT images and aerial
photos, Lin et al. (2000) and Khazai and Sitar (2004)
reported that most landslides triggered by the Chi-Chi
earthquake were shallow, disrupted, small-to-medium
slides and rock falls. Ground motion was found to be the
most significant factor in triggering the shallow land-
slides. Seventy-four percent of slope failures occurred in
the area with vertical ground motion greater than
200 gal, and 81% of landslides occurred in the region
with mean horizontal peak ground acceleration in the
range from 150 to 450 gal (Lin et al., 2000; Khazai and
Sitar, 2004). In addition, according to 436 field-
investigated landslides and 2070 SPOT-identified land-
slides, 90% of the landslides occurred on slopes with
angles greater than 45. Khazai and Sitar (2004) also
noted that the shallow landslides were not associated

Table 4
Precipitation in the study area during the period 19962001
January February March April May June July August September October November December Total
1996 16.5 41.0 34.2 224.0 455.0 129.5 543.8 548.1 131.6 57.9 37.2 11.8 2230.5
1997 43.1 126.4 177.3 75.5 216.1 482.1 290.9 435.1 166.5 17.9 0.00 12.2 2043.2
1998 119.0 346.7 200.7 249.7 278.7 432.2 96.8 424.8 195.1 501.6 3.4 66.5 2915.5
1999 26.0 3.58 80.2 65.2 313.4 274.9 380.6 459.9 121.5 46.8 12.3 63.2 1847.7
2000 31.0 155.8 54.7 256.4 141.7 323.0 298.3 465.0 64.2 157.3 77.5 51.3 2076.2
2001 108.4 5.8 83.2 179.0 359.0 361.5 687.3 148.0 605.5 25.8 37.0 12.6 2613.2
Mean 53.1 113.0 105.9 164.8 284.4 341.3 375.5 387.2 199.4 118.0 24.8 32.5 2200.2
Mean monthly rainfall data were collected and calculated from 46 rainfall recording stations within and around the Choushui River watershed.
94 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101

Fig. 4. A cartoon shows different terms in landslides in a comparison of two consecutive images. In the first image, only landslide A was recognized.
In the second image, landslides A, B, and C were recognized. Therefore, All landslides in the second image were A + B + C. Incremental landslides
in the second image were B + C. Landslide C in the second image was a new landslide, and landslide B was an enlarged landslide.

with particular geological units. They found that 45% of
the landslides occurred in Neogene sedimentary rocks,
22% in Quaternary alluvium, 17% in Pleistocene
sandstone and conglomerate, and 16% in slate and
metamorphic sandstone.
7. Variations in area and density of landslides
In this paper, to better determine the characteristics of
landslides in different SPOT images, landslides recog-
nized in each image are further described by three terms
incremental landslides, enlarged landslides, and new
landslides. The incremental landslides represent the
difference in landslides between two consecutive
images. Therefore, the incremental landslides can better
represent the landslides induced by events that occurred
between two images. In order to know how many
landslides are enlarged from pre-existing landslides,
incremental landslides are further classified into new
landslides and enlarged landslides (see Fig. 4). New
landslides are landslides that can only be recognized in
the new image, and enlarged landslides are landslides
enlarged from pre-existing landslides in a previous
image.
Recognition and calculation of landslide areas from
the SPOT images indicates that the Chi-Chi earthquake
had greatly enhanced the development of landslides (see
Fig. 5 and Table 2). Typhoon Herb brought in heavy rain
in July 1996, and there were 9.77 km2 of incremental
landslides between images in June 1996 (image
G09619) and November 1996 (image G14166). During
the period after typhoon Herb and before the Chi-Chi
earthquake, the area of incremental landslides gradually
decreased from 8.13 km2 to 4.64 km2, and then to
3.40 km2. However, in January 2000 (after the Chi-Chi
earthquake but before the heavy rain of April 2000) the
incremental landslide area was nearly six times larger, at
21.85 km2. Because there was no abnormally high
precipitation after the Chi-Chi earthquake, the abrupt
increase in landslide area was considered most likely
attributable to the Chi-Chi earthquake. Image measure-
ment for later periods showed that the incremental
landslide area had further expanded to 43.02 km2 in
March 2001 and to 48.42 km2 in August 2001. This

Fig. 5. Plot of different types of landslides in each SPOT image.

 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
shows that a strong shock such as the Chi-Chi
earthquake did more than cause severe coseismic
landslidesit also substantially induced landslides
during subsequent heavy rains. We believe this is due
to numerous extension cracks generated on hill slopes
during the earthquake, which accelerate landslides
during downpours.
The variations in the enlarged landslides and the
new landslides exhibit trends similar to what we see in
the incremental landslides (see Table 2 and Fig. 5).
Among 9.77 km2 of incremental landslides induced by
typhoon Herb, 6.38 km2 were classified as new
landslides and 3.39 km2 were recognized as enlarged
landslides. The amount of both new landslides and
enlarged landslides then gradually decreased to
0.7 km2 and 2.7 km2, respectively, in the image
taken before the Chi-Chi earthquake (see Table 2 and
Fig. 5). After the Chi-Chi earthquake, the new
landslides expanded to 7.5 km 2 and then to
12.6 km2 in images of March 2001, and finally to
14.1 km2 in images of August 2001. Similarly, the
enlarged landslides dramatically increased from
2.7 km2 to 14.4 km2 after the Chi-Chi earthquake,
then to 30.4 km2 in images of March 2001, and to
34.3 km2 in images of August 2001. Except for the
image taken after typhoon Herb, landslides in all
images show that the area of enlarged landslides is
larger than that of new landslides. This indicates that
most of the incremental landslides in the study area
mainly come from the enlargement of pre-existing
landslides except in extreme rainfall conditions such as
typhoon Herb, when more new landslides develop.
8. Variations in places where landslides occurred
When we compared the occurrence of landslides in
terms of area in each image, there were also evident
shifts in the locations where landslides took place.
Before the earthquake, all types of landslides recog-
nized in each image were relatively uniformly
distributed in places with slopes between 20 and
50 (Fig. 6ad), although two small peaks at 2030
and 4050 can be differentiated. However, after the
earthquake, a significant peak located in a place with a
slope of 4050 was recognized, which represents the
characteristics of typical earthquake-induced land-
slides. In addition, in the following 2 years, most of
the rainfall-induced landslides also occurred in areas
with slopes between 40 and 50. This implies that the
impact of the Chi-Chi earthquake still highly influ-
enced the locations in which rainfall-induced land-
slides occurred. During or after the Chi-Chi
95
earthquake, the shift of landslide locations to places
with a steep slope can probably be attributed to
topographic amplification of ground motion and a
lower external stress required to trigger surface
material to slide down on a steep slope.
In terms of landslide concentration, all types of
landslides were concentrated in places with slopes of
20 in the images taken before the Chi-Chi earthquake
(Fig. 6eh). The landslide concentrations in places with
20 slopes are typically the largest of any slope interval.
However, a large value was also recognized in the 70
slope, which is intuitive because sliding is also easy on
such very steep slopes. In the images taken after the Chi-
Chi earthquake, landslide concentration has similar
values in slope intervals of 2030 and 6070 (Fig.
6e). For incremental landslides in the image taken in
January 2000 (G14162) and the image taken in March
2001 (G14525), more landslides were counted in places
with slopes of over 50 than that in places with 20
slopes (Fig. 6f). However, in the image taken after Toraji
(G14806), the numbers of landslides are very similar in
slope intervals from 30 to 70 (Fig. 6f). This trend is
more easily observed in the plot of new landslides (Fig.
6g). In contrast, the numbers of enlarged landslides are
greatest in places with 6070 slopes in the image taken
after Toraji (Fig. 6h).
9. The influence of lithology on the characteristics of
landslides
In order to determine the influence of lithology on
landslide concentration and the places where land-
slides occur, incremental landslides recognized in each
image are plotted for the different stratigraphic units
(Fig. 7). Because the distributions of incremental
landslides in different stratigraphic units are similar in
images taken before the Chi-Chi earthquake, only the
results from the image G14166 taken after typhoon
Herb is shown here.
In the image taken after Herb (G14166), more
incremental landslides were observed in Holocene
Pleistocene terrace and alluvium deposits than in the rest
of the units, and a significant peak was seen in Fig. 7a.
The extremely large number of landslides in places with
70 slopes is mainly attributed to landslides that
occurred at the edges of river terraces. In addition,
incremental landslides in all stratigraphic units were
mainly distributed in places with slopes of 20 to 30,
which is the same as for the typical rainfall-induced
landslides that have been documented before. In the
images taken in January 2000 (G14162) and March
2001 (G14525), the numbers of incremental landslides
96 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101

Fig. 6. (a) Plot of the area of all landslides at different hill slopes. (b) Plot of the area of incremental landslides at different hill slopes. (c) Plot of the
area of new landslides at different hill slopes. (d) Plot of the area of enlarged landslides at different hill slopes. (e) Plot of the concentration of all
landslides at different hill slopes. (f) Plot of the concentration of incremental landslides at different hill slopes. (g) Plot of the concentration of new
landslides at different hill slopes. (h) Plot of the concentration of enlarged landslides at different hill slopes.

dramatically increased in places with slopes over 40
without regard to lithology (Fig. 7b and c). Also, in all
stratigraphic units, more incremental landslides were
counted in the image G14525 than in the image G14162.
This indicates that only some of the disturbed surface
material slid down in the Chi-Chi earthquake and that
more disturbed strata poured down during the subse-
quent rainfall. However, landslide concentration slightly
decreased and was more uniformly distributed in places
with slopes of 2060 in the image taken after Toraji
 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
Fig. 7. (a) Plot of the incremental landslides by different stratigraphic
units in the image G14166 (1996/11/08). (b) Plot of the incremental
landslides by different stratigraphic units in the image G14162 (2000/
01/08). (c) Plot of the incremental landslides by different stratigraphic
units in the image G14525 (2001/03/05). (D) Plot of the incremental
landslides by different stratigraphic units in the image G14806 (2001/
08/12).
97
(G14806) when compared with image G14525. This
may imply that the impact of the Chi-Chi earthquake on
the rainfall-induced landslides gradually decreased.
However, the total area of incremental landslides in
the image G14806 is still larger than that in image
G14525 (Table 2), which indicates that the mean area of
each incremental landslide in G14806 is larger than that
in G14525.
The distribution of new landslides is similar to that
of incremental landslides (Fig. 8). Before the Chi-Chi
earthquake, new landslides mainly occurred in places
with a slope of 2030, except for those that occurred
in HolocenePleistocene terrace and alluvium deposits
and in Pleistocene sedimentary rocks (Fig. 8a). The
distribution of new landslides induced by the Chi-Chi
earthquake (Fig 8b) shows a peak at 50 with a value
near 30 (landslides/km2) in Pleistocene sedimentary
rocks, at 70 with a value of about 50 in Pliocene
sedimentary rocks, and at 60 with a value below 20
in the rest of the stratigraphic units. This indicates that
more coseismic landslides were triggered in Pliocene
and younger sedimentary rocks in terms of landslide
concentration. In addition to the peaks mentioned in
image G14162, all new landslides in places with
slopes of 2030 significantly increased, without
regard to the lithology, in the image taken in March
2001(G14525) (Fig. 8c). However, in the image taken
after Toraji (G14806), new landslides were more
uniformly distributed in places with different slopes
(Fig. 8d).
Enlarged landslides in the image taken after Herb
(G14166) were mainly located in areas with a slope of
20 in all stratigraphic units (Fig. 9a). However, more
enlarged landslides were also recognized in places with
70 slopes in Miocene sedimentary rocks. In the image
taken after the Chi-Chi earthquake (G14162), more
enlarged landslides occurred in places with slopes of
2030 than in locations with different slopes in all
stratigraphic units. This is reasonable because most of
the pre-existing landslides that had occurred before the
Chi-Chi earthquake were located in places with slopes
of 2030. The results from the image taken after the
Chi-Chi earthquake (G14162) also shows that more
enlarged landslides occurred in HolocenePleistocene
terrace and alluvium deposits than in the rest of the
units, without regard to the slope of the landslide
location (Fig. 9b). This is also reasonable because
more landslides occurred in HolocenePleistocene
terrace and alluvium deposits before the earthquake
(Fig. 7a). Although in the images G14525 and G14806
landslides enlarged from pre-existing landslides con-
centrated at three places, with slopes of 20, 50, and
98 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
Fig. 8. (a) Plot of the new landslides by different stratigraphic units in
the image G14166 (1996/11/08). (b) Plot of the new landslides by
different stratigraphic units in the image G14162 (2000/01/08). (c) Plot
of the new landslides by different stratigraphic units in the image
G14525 (2001/03/05). (d) Plot of the new landslides by different
stratigraphic units in the image G14806 (2001/08/12).
70 in all stratigraphic units (Fig. 8c and d), more
enlarged landslides still occurred in places with steeper
slope.
10. The impact of ground motion on the occurrence
of landslides
To evaluate the impact of ground motion on the
occurrences of rainfall-induced landslides, the concen-
tration of incremental landslides and new landslides
Fig. 9. (a) Plot of the enlarged landslides by different stratigraphic units
in the image G14166 (1996/11/08). (b) Plot of the enlarged landslides
by different stratigraphic units in the image G14162 (2000/01/08). (c)
Plot of the enlarged landslides by different stratigraphic units in the
image G14525 (2001/03/05). (d) Plot of the enlarged landslides by
different stratigraphic units in the image G14806 (2001/08/12).
 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
Fig. 10. (a) Plot of incremental landslide concentration recognized in
the image G14166 (1996/11/08) versus ground motion of the Chi-Chi
earthquake. (b) Plot of incremental landslide concentration recognized
in the image G14162 (2000/01/08) versus ground motion of the Chi-
Chi earthquake. (c) Plot of incremental landslide concentration
recognized in the image G14525 (2001/03/05) versus ground motion
of the Chi-Chi earthquake. (d) Plot of incremental landslide
concentration recognized in the image G14806 (2001/08/12) versus
ground motion of the Chi-Chi earthquake.
versus ground motion were plotted. For convenience of
discussion, landslides in the image G14166 that was
taken before the Chi-Chi earthquake were also plotted
against the ground motion of the earthquake, although
99
this shaking in fact had not yet occurred. In the
Choushui River watershed, 6070% of the catchment
suffered 600700 gal of ground motion during the Chi-
Fig. 11. (a) Plot of new landslide concentration recognized in the
image G14166 (1996/11/08) versus ground motion of the Chi-Chi
earthquake. (b) Plot of new landslide concentration recognized in the
image G14162 (2000/01/08) versus ground motion of the Chi-Chi
earthquake. (c) Plot of new landslide concentration recognized in the
image G14525 (2001/03/05) versus ground motion of the Chi-Chi
earthquake. (d) Plot of new landslide concentration recognized in the
image G14806 (2001/08/12) versus ground motion of the Chi-Chi
earthquake.
100 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
Chi earthquake (Fig. 1), which meant that the area of
incremental landslides in the image G14166 were
mainly distributed in an area subject to this magnitude
of ground motion. However, the distribution of landslide
concentration is quite uniform, and it can be treated as
the background value of landslide concentration even
though there is a higher value in the area with 801
900 gal.
In the images G14162, G14525, and G14806 that
were taken after the Chi-Chi earthquake, the concentra-
tion of incremental landslides and new landslides shows
a similar trend in that the number of slope failures
increased with increasing magnitude of ground motion
in an area that had experienced ground motion over
700 gal (Figs. 10 and 11). For earthquake-induced
landslides, increasing numbers of landslides with
increasing magnitudes of ground motion is what we
expected. Since this trend is absent in pre-earthquake
landslides (Figs. 10a and 11a) the influence of seismic
shaking effects on landslide activation appears convinc-
ing. However, a more irregular pattern appears in areas
with a PGA below 700 gal. This may imply that only
when the shaking exceed 700 gal, then the seismic
shaking effects on landslide activation become prevail-
ing. Otherwise, the impact of seismic shaking may be
masked by other controlling factors such as slope angle,
lithology, and rainfall spatial distribution.
11. Discussion and conclusions
In comparison with field surveys, certain errors
might be created by using SPOT images to count and to
locate landslides. The resolution of SPOT images does
not allow a distinction to be made between the fresh
erosional surfaces generated by flowing water and
surfaces with vegetation stripped off by shallow
landsliding. Therefore, the landslide area covered by
the SPOT images may over-estimate when fresh
erosional surfaces were also included. However, the
vastness of the study area and poor accessibility to the
disaster region made SPOT images and DTM the most
efficacious tools for conducting the current research.
Therefore, although the slope distributions cannot be
calculated precisely by using the 40 m 40 m DTM, this
study has shown that the variations in slope distributions
of landslides prior to and after the Chi-Chi earthquake
can be obtained in spite of the fact that the slope angles
may be underestimated.
Comparing the rainfall conditions of typhoons Herb
and Toraji, the maximum hourly rainfall intensity
brought by Toraji was larger than that brought by
Herb, but the cumulative precipitation brought by Herb
was significantly larger than that brought by Toraji. Yet,
only 9.77 km2 of landslides were induced by Herb,
whereas 48.8 km2 of landslides were triggered by Toraji.
This shows the great impact of the Chi-Chi earthquake
on the density of landslides in the Choushui River
watershed. In addition, the Chi-Chi earthquake induced
coseismic landslides of 21.85 km2, which is also much
less than those triggered by subsequent heavy rainfalls
and by Toraji. Thus, it seems evident that after surface
strata have been highly disturbed a heavy rainfall may
trigger more landslides than the earthquake itself.
In typhoon Herb, landslides in the Choushui River
watershed were mainly triggered by heavy rains.
Therefore, they were mainly distributed in mid-slope
areas with slope angles of 2030, as is typical for
rainfall-induced landslides. In addition to the shifting of
the locations of coseismic landslides to places with
steeper slopes during the Chi-Chi earthquake, the
subsequent rainfall-induced landslides also shifted
from mid-slope (before the earthquake) to ridge tops
(after the quake). This evidence indicates that most of
the rainfall-induced landslides after the Chi-Chi earth-
quake were still highly influenced by the earthquake
itself.
The variations in the amount of landslide area derived
from eight SPOT images consistently shows increased
areas of different types of landslides from the period after
the Chi-Chi earthquake to the end of typhoon Toraji.
However, if different types of landslides were plotted in
terms of landslide concentration, it shows the landslide
concentration slightly decreased between the image
taken in March 2001 (G14525) and the image taken in
August 2001 (G14806). This may imply that the
influence of the Chi-Chi earthquake on subsequent
rainfall-induced landslide gradually diminished.
After the Chi-Chi earthquake, the study results show
the density of rainfall-induced landslides in Pleistocene
Holocene terrace and alluvium deposits are slightly
larger than that occurred in the rest of stratigraphic units.
However, there is no specific trend that can be
recognized in the density of landslide in terms of
stratigraphic units. This indicates that the impact of
ground motion is similar in all stratigraphic units, and
rainfall-induced landslides were not associated with any
specific stratigraphic unit. This result is consistent with
the result from Khazai and Sitar (2004) that noted that the
coseismic shallow landslides were not associated with
particular geological units.
The catastrophic Chi-Chi earthquake not only
triggered serious coseismic landslides itself but also
extensively disturbed surface strata around the epicen-
tral area. During the earthquake, although surface strata
 C.-W. Lin et al. / Engineering Geology 86 (2006) 87101
were disrupted and brought downhill, there was still
abundant debris left behind on the slopes, which would
later serve as source material for rainfall-induced
landslides. Also, some substrata became preconditioned
to fail through loss of cohesion and frictional strength of
rock mass and because of numerous extension cracks
that were induced on hill slopes by the strong ground
motion. These factors undoubtedly accelerated the
landslides during the subsequent heavy rains brought
by typhoon Toraji (Dadson et al., 2004). Furthermore,
due to the increase in the density of landslides during
and after the Chi-Chi earthquake, suspended sediments
in the Choushui River are as much as four times higher
than background value before the Chi-Chi earthquake
(Dadson et al., 2004). The total sediments removed from
the Choushui River watershed during the period
between the Chi-Chi earthquake and the end of 2001
was 326 Mt, 175 Mt of which was removed by Toraji
over a period of 2 days. This also induced serious
problems in the balance of sediment transportation and
deposition along the river channel and caused the river
channel topography to change. Therefore, understand-
ing the duration and the impact of the Chi-Chi
earthquake is an important problem for landslide and
debris flow hazard mitigation in the study area. We
believe that more studies to continue tracking rainfall-
induced landslides and debris flows are necessary.
Acknowledgements
This work was supported by Grant Nos. NSC 92-
2216-M-006-004 and NSC92-2625-Z-006-016 from the
National Science Council, Taipei, Taiwan. The authors
also appreciate Dr. David Keefer, Dr. Vincenzo Del
Gaudio, and Dr. Janusz Wasowski for their critical
reviews and constructive comments.
References
Cheng, S.N., Yeh, Y.T., Hsu, M.T., Shin, T.C., 1999. Photo Album of
Ten Disastrous Earthquakes in Taiwan. Central Weather Bureau,
Taiwan. 290 pp.
Crosta, G.B., 2004. Introduction to the special issue on rainfall-
triggered landslides and debris flows. Engineering Geology 73,
191192.
View publication stats
101
Dadson, S., Hovius, N., Chen, H., Dade, W.B., Lin, J.C., Hsu, M.L.,
Lin, C.W., Horng, M.J., Chen, T.C., Miliman, J., Stark, C.P., 2004.
Earthquake-triggered increase in sediment delivery from an active
mountain belt. Geology 32-8, 373376.
Harp, E.L., Jibson, R.W., 1996. Landslides triggered by the 1994
Northridge, California earthquake. Bulletin of the Seismological
Society of America 86, 319332.
Harp, E.L., Schmidt, K., Wilson, R., Keefer, D.K., Jipson, R.W., 1991.
Effects of landslides coseismic fractures triggered by the 17
October 1989 Loma Prieta, California Earthquake. Landslide
News 5, 1822.
Jibson, R.W., Prentice, C.S., Borissoff, B.A., Rogozhin, E.A., Langer,
C.L., 1994. Some observations of landslides triggered by the 29
April 1991 Racha earthquake, Republic of Georgia. Bulletin of the
Seismological Society of America 84, 964973.
Kaima, T.W., Wang, W.N., Sguzui, H., 2000. The landslide disaster
induced by the Taiwan Chi-Chi earthquake of 21 September 1999.
Landslide News 13, 812.
Keefer, D.K., 1984. Landslides caused by earthquakes. Bulletin of
Geological Society of America 95, 406421.
Keefer, D.K., 2000. Statistical analysis of an earthquake-induced
landslide distributionthe 1989 Loma Prieta, California event.
Engineering Geology 58, 231249.
Khazai, B., Sitar, N., 2004. Evaluation of factors controlling
earthquake-induced landslides caused by Chi-Chi earthquake and
comparison with the Northridge and Loma Prieta events.
Engineering Geology 71, 7995.
Lin, M.L., Jeng, F.S., 2000. Characteristics of hazards induced by
extremely heavy rainfall in Central Taiwantyphoon Herb.
Engineering Geology 58, 191207.
Lin, M.L., Wang, K.L., Chen, T.C., 2000. Characteristics of the slope
failure caused by the Chi-Chi earthquake. International Workshop
on Annual Commemoration of Chi-Chi Earthquake, September
1820, 2000, Taipei, vol. 3, pp. 199209.
Lin, C.-W., Shieh, C.-L., Yuan, B.-D., Shieh, Y.-C., Liu, S.-H., Lee, S.-
Y., 2004. Impact of Chi-Chi earthquake on the occurrence of
landslides and debris flows: example from the Chenyulan River
watershed, Nantou, Taiwan. Engineering Geology 71, 4961.
Schuster, R.L., Nieto, A.S., O'ouke, T.D., Crespo, E., Plaza-Nieto, G.,
1996. Mass wasting triggered by the 5 March 1987 Ecuador
earthquakes. Engineering Geology 42, 123.
Shou, K.J., Wang, C.F., 2003. Analysis of the Chiufengershan
landslide triggered by the 1999 Chi-Chi earthquake in Taiwan.
Engineering Geology 68, 237250.
Water Resources Agency, 2001. Hydrological Year Book of Taiwan,
Republic of China, ISSN 02575442.
Wang, W.N., Chigira, M., Furuya, T., 2003. Geological and
geomorphological precursors of the Chiu-fen-erh-shan landslide
triggered by the Chi-Chi earthquake in Central Taiwan. Engineer-
ing Geology 69, 113.