Content uploaded by Kou-Cheng Chen
Author content
All content in this area was uploaded by Kou-Cheng Chen on Jul 27, 2017
Content may be subject to copyright.
1190
Bulletin of the Seismological Society of America, 91, 5, pp. 1190–1198, October 2001
Estimates of Source Parameters for the 1999 Chi-Chi, Taiwan,
Earthquake Based on Brune’s Source Model
by Win-Gee Huang, Jeen-Hwa Wang, Bor-Shouh Huang, Kou-Cheng Chen, Tao-Ming Chang,
Ruey-Der Hwang, Hung-Chie Chiu, and Chu-Chuan Peter Tsai
Abstract The general features of the rupture of the 1999 Chi-Chi, Taiwan, earth-
quake (M
s
7.6) can be explained by the displacement waveforms derived from the
accelerograms recorded at short distances from the fault traces. Applying Brune’s
model, we have determined important source parameters, such as rise time, stress
drop, offset, and particle velocity. Generally, the earthquake is characterized as hav-
ing had two distinct fault segments. The southern segment, dominated by thrust
motion, started from the focus on a fault plane raking at 78⬚and extended about 30
km to the north. The northern segment, dominated by thrust with significant strike-
slip motion, began next to the end of the southern segment on a fault plane raking
at 53⬚and extended northward for 25 km. Slips in the southern segment were fol-
lowed by a small dislocation (⬃1 m), while those in the northern segment were
followed by a much larger dislocation (⬃9 m). The average slip velocity was dis-
tributed at 34–49 cm/sec, along the southern segment, and an unusual slip velocity
exceeding 2 m/sec was observed along the northern segment. Furthermore, the south-
ern segment experienced a rise time of 1.8 sec and a stress drop of 65 bars, in contrast
to a rise time longer than 4 sec and a stress drop larger than 300 bars registered to
the north. Our results also indicate that, along the southern segment, the rupture
propagated northward at an average velocity of 2.84 km/sec, but along the northern
segment, the rate declined to less than 2 km/sec. The difference in the source param-
eters between these two segments suggests that the rupturing associated with the Chi-
Chi earthquake may have encountered a resistive patch and changed course in the
middle part of the fault. After crushing that resistance, the long rise time and high
stress drop probably caused substantially slower motion and larger slip along the
northern segment.
Introduction
Dislocation rise time (s) is defined as the time required
for the final slip at a point on a fault plane to occur during
an earthquake rupture process. Several studies for the rise
time have been presented in theories on the source model
(Brune, 1970; Savage, 1972; Sato, 1989; Heaton, 1990).
Savage (1972) assumed that the rise time on a rectangular
rupture surface is related to the fault width (W) and rupture
velocity (v
r
); hence, s⳱W/4.6v
r
. Based on a compilation
of various source models, Sato (1989) set forth a scaling
relationship between the rise time and the magnitude (M)of
an earthquake, by which s⳱10
0.5
M
ⳮ1.4
/80. Heaton (1990)
investigated the dislocation time histories of models from
waveforms of earthquakes and concluded the rise time is of
the order of 10% of the overall duration (i.e., total rupture
time) of the earthquake, albeit with considerable variation
between different models. Despite various source models on
rise time, sis usually to scale as a source dimension. From
source dimension and seismic moment, the average (global)
stress drop over the entire fault plane can then be deter-
mined.
The average stress drop deduced from the rise time de-
scribed is always model dependent, and none is an actual
stress changes during the faulting. In fact, to date there has
been no theory that adequately explains the rise time of an
earthquake. Nevertheless, some relationships have been de-
veloped to best fit the observations. In actual fault zones, the
stress drop varies in general complexity from place to place.
Locally, the stress drop can be much higher than the average
stress drop of an earthquake. The deduction of large ground
motions is often based on a localized region of high stress
and hence a greater degree of slip during the faulting. To
obtain the localized stress drop, the most straightforward
way has been through observations made during ruptures in
the vicinity of the fault. In this study, we utilize the near-
Estimates of Source Parameters for the 1999 Chi-Chi, Taiwan, Earthquake Based on Brune’s Source Model 1191
fault observations to determine local stress drop following
Brune (1970).
Recently, on 21 September 1999, the Chi-Chi earth-
quake (M
s
7.6) ruptured the ground surface along the Che-
lungpu fault in central Taiwan. The earthquake triggered al-
most all strong-motion stations operated by the Central
Weather Bureau (CWB) around the epicentral area. This data
set is important for at least two reasons. First, the near-fault
ground displacement, as obtained from accelerograms by
double integration, clearly shows a ramp-type waveform (cf.
Chung and Shin, 1999), thus indicating a permanent dis-
placement of the ground. Such fault offset is very similar to
the idealized slip-time function in the vicinity of the fault
(e.g., Brune 1970). Second, these recordings provide a rare
opportunity to investigate the fault movements of an earth-
quake using near-fault measurements.
Results from ground-motion observations at close-in
sites (Chung and Shin, 1999) as well as from far-field seis-
mograms (Kikuchi et al., 2000) led one to believe that the
fault plane of the Chi-Chi earthquake can essentially be di-
vided into two segments that break the surface. Based on the
near-fault recordings, Chung and Shin (1999) showed that
the character of ground motions exhibits a major change
between the southern and northern segments of the fault. At
the southern segment, the ground motions were dominated
in short-period motions with large accelerations. Con-
versely, long-period motions with large velocity and dis-
placement were predominant in the northern segment. Kik-
uchi et al. (2000) constructed a rupture model for the
earthquake by inversion of teleseismic body waves. Their
results present estimates of the fault width (⬃40 km), fault
length (⬃100 km), and average rupture velocity (⬃2.5 km/
sec). Also, they pointed out that there were two significant
moment releases during the faulting. The first, a smaller one,
was close to the epicenter. The second, which was much
larger, occurred 35 km north of the first. Both studies expand
our insight into the macroscopic features of the fault rupture.
However, in order to get a more detailed understanding of
the rupture properties, a study using local strong-motion data
is required.
The purpose of this study, therefore, is to analyze the
displacement waveforms at the stations nears the Chelungpu
fault for inference of slip velocities and permanent offsets.
Here, we employed Brune’s (1970) model to quantify some
source parameters, such as the rise time and stress drop.
Based on the results, we are able to determine the physical
properties of the fault plane of the Chi-Chi earthquake.
Observed Data
Earthquake Location and Fault
The epicenter (Fig. 1) of the Chi-Chi earthquake as de-
termined by Chang (2000) using both strong motion and
real-time monitoring data was 23⬚49.2⬘N and 120⬚51⬘E. Its
focal-depth was determined to be 8 km (Chang, 2000) and
resulted in a ⬃80 km surface break. The fault trace is shown
in the curve in Figure 1. In the southern section, the trace is
nearly due north. In the middle section, the trace turns
slightly to the northeast, but further north, the trace returns
to a more northerly direction once again. At the northern
end, the trace rotates N60⬚E. Observations of surface slip
revealed a peak value of about 1–2 m in the southern section
and about 6–8 m in the northern section. The fault-plane
solution (Chang, 2000) shows that the earthquake occurred
as a thrust on a plane dipping approximately 34⬚E with an
average strike of N5⬚E and rake of 65⬚.
Instrumentation
Eight stations nearest (all within 2 km, 6 within 1 km)
to the Chelungpu fault trace were chosen for this study
(Fig. 1). Each station is equipped with a triaxial force-
balance accelerometer (Teledyne Geotech A900) that has a
flat instrument response from 0 to 50 Hz. The accelerometer
is capable of recording a full scale of Ⳳ2g. The outputs
were digitized and recorded with a 16-bit resolution at a rate
of 200 samples/sec.
Ground Motion Near the Fault
Figure 2 depicts the accelerograms, within the most sig-
nificant 40-sec time windows, at these eight stations. The
direct P-wave-arrival alignment exists between these traces.
As shown in Figure 2, the degree of complexity in the wave-
forms decreases northward from the epicenter and some
properties can be figured out. The records at the southern
three stations (TCU129, TCU076, TCU075) are relatively
more complex, compared to the northern three stations
(TCU052, 102, 068). Moderate complexity appears at the
remaining two central stations (TCU065 and TCU067). The
moderate complexity accelerograms in these two stations
may be responsible for the release of energy caused by the
southern and northern segments of the fault, since both sta-
tions lie near to the intersection of the two fault segments
(Fig. 1). The southernmost station (TCU129) recorded the
largest peak east–west horizontal acceleration of 982 gal
(cm/sec
2
), whereas the northernmost station (TCU068) reg-
istered the largest peak vertical acceleration of 519 gal.
The velocity and displacement waveforms, integrated
once and twice from the accelerograms, are shown in Figures
3 and 4, respectively. Like the accelerograms, the velocity
waveforms degenerate in complexity from the south to the
north. At the northern three stations, each trace is essentially
characterized by one single pulse with a duration of about
6–8 sec.
Noteworthy are displacement waveforms (Fig. 4).
Firstly, all east–west components are characterized by ramp-
like features, indicative of permanent ground displacement.
Secondly, the horizontal displacement at two northern sta-
tions (TCU052 and TCU068) points westerly and northerly,
while at the other six stations, the ground displacements are
in the opposite directions, east and south. Thirdly, the two
largest peak horizontal displacements, which exceed 5 m
1192 W. G. Huang, J. H. Wang, B. S. Huang, K. C. Chen, T. M. Chang, R. D. Hwang, H. C. Chiu, and C. C. Tsai
Figure 1. The regional map around the
source area in west-central Taiwan. Solid star
represents the epicenter of the Chi-Chi earth-
quake. The solid and shaded bold curves de-
note the southern and northern segments of the
Chelungpu fault, respectively. Solid triangu-
lars denote the major cities near the fault trace.
Numbered solid squares show are the seismo-
graphic stations
each, occur at these same two northern stations, in contrast
to the peak values of 0.4 to 2.6 m at other stations. Fourthly,
the waveforms for the vertical displacements at stations
TCU052 and TCU068 are quite simple compared with other
stations where some shorter-period ripples are abundant.
Upward-rising motion at stations TCU052 and TCU068 is
consistent with the fact that both stations are on the hanging
wall of the thrust. For other stations, we cannot recognize
their locations relative to the fault trace from the complexity
of waveforms. Fifthly, the peak vertical displacements at
stations TCU052 and TCU068 are greater than 4 m, while
at the other stations, the recorded peak values range only
from 0.3 to 0.7 m.
Basic Theory
The expression for Brune’s ground displacement u(t)
caused by stress drop Dr near the earthquake source is given
by the following:
ⳮt/s
u(t)⳱(Drb/l)s(1 ⳮe), (1)
where lis the shear modulus and bis the shear-wave speed
in the source volume. The time constant, s, can be approx-
imated by the process time a/b, i.e., s⬃a/b, with abeing
the equivalent radius of the fault surface area. Equation
(1) can be approximated by the following expressions: at
t⳱s,
u(s)⳱0.63 (Drb/l)s,(2)
and at t⳱⬁
u(⬁)⳱(Drb/l)s⳱D.(3)
The Din equation (3) denotes the static offset. If the particle
displacement is averaged over the process time a/b, we have
an average particle speed ( ). The stress drop can then be
˙
¯
U
expressed in terms of ˙
¯
U
˙
¯
Dr⳱Ul/0.63b(4)
If is measurable with sufficient precision, we can estimate
˙
¯
U
the stress drop Dr through equation (4). From equations (3)
and (4), scan be determined by the expression
˙
¯
s⳱0.63D/U.(5)
At a site close to the fault, the ramp-function-like displace-
ment records should reflected the movement either at the
hanging wall or at the footwall rather than their relative mo-
tion. Before the source parameters were estimated, we ro-
tated the displacement waveforms from recorded orientation
to the fault plane to obtain fault dip-slip, strike-slip, and
normal-slip motion components. According to the rotated
seismograms, we calculated the static offset Dand estimated
the average particle velocity by finding an asymptotic line
˙
¯
U
to fit the rising part of the displacement waveforms in the
least-squared sense. The rise time and stress drop were then
calculated from Dand for each station. Based on the ve-
˙
¯
U
locity model of Chen (1998) and the focal depth (about 8
km) for the Chi-Chi earthquake, we used b⳱3 km/sec and
l⳱3⳯10
11
dyne/cm
2
for the shallow crust in our cal-
culation.
Analysis and Results
Rotated Displacements
Figure 5a shows the original displacement waveforms
at station TCU052. The rotated one, based on the focal
Estimates of Source Parameters for the 1999 Chi-Chi, Taiwan, Earthquake Based on Brune’s Source Model 1193
Figure 2. Accelerograms (vertical, east, and north components) at eight stations
arranged from the north to the south (Fig. 1). Station identifications precede seismo-
grams. Each accelerogram is normalized to its peak value (cm/sec
2
) as enumerated
above each trace
Figure 3. Ground velocities integrated once from accelerograms shown in Figure 2.
See Figure 2 caption. The peak ground velocity (cm/sec) is shown at the beginning of
each time series
1194 W. G. Huang, J. H. Wang, B. S. Huang, K. C. Chen, T. M. Chang, R. D. Hwang, H. C. Chiu, and C. C. Tsai
Figure 4. Ground displacements integrated twice from accelerograms shown in Fig-
ure 2. See Figure 2 caption. The peak ground displacement (cm) is shown at the be-
ginning of each time series
Figure 5. Comparison of the (a) original and (b) rotated ground displacements at
station TCU052. All traces are normalized to the same scale. Arrows demarcate a time
window for the rising part of the dip- or strike-slip components. The bold line segments
represent the best fit of the envelope of the rotated displacements, which gives the
average particle velocity
Estimates of Source Parameters for the 1999 Chi-Chi, Taiwan, Earthquake Based on Brune’s Source Model 1195
Figure 6. Comparison of the displacement waveforms of (a) vertical and (b) dip-
slip components at all eight stations. See Figure 2 caption.
mechanism determined by Chang (2000), is depicted in Fig-
ure 5b. The result reveals the ramplike displacement com-
ponent along the fault dip or strike exceeds 6.8 m, while a
pulselike normal component has a peak displacement of 2.2
m. The partition of displacement components suggests that
the rotated displacement waveforms can reasonably explain
an idealized slip motion along the fault length and width.
Figure 6a depicts the original vertical displacement
waveforms for all stations. Shown in Figure 6b is the dip-
slip displacement of the rotated waveforms. Comparing
these two sets of displacement waveforms, we found that all
the rotated waveforms are indicative of a permanent ground
displacement. Furthermore, the relative motion between the
two sides of the fault can be recognized in Figure 6b. The
waveforms at stations TCU052 and TCU068 show positive
displacement, but those at the other stations always exhibit
negative displacement. This indicates that stations TCU052
and TCU068 are located on the up-thrown hanging-wall
side, and the others stations are located on the down-thrown
footwall side. At these two stations, the permanent ground
displacement exceeds 8 m. In contrast, other stations expe-
rienced considerably less permanent displacements (0.87 to
1.5 m). In addition, Figure 6b exhibits two distinctive groups
of displacement waveforms. At the five southern stations
(i.e., TCU129, TCU076, TCU075, TCU065, and TCU067),
the ground displacements rise immediately and propagate
substantially to the north. The duration (T) for the ground
displacement arrests within about 4.0 sec at these five sta-
tions. The fault slip seems to delay about 3 sec between
stations TCU067 and TCU052 before continuing farther
north. At the northern three stations (i.e., TCU052,
TCU0102, and TCU068), the ground displacements rise
relatively slow and arrest after enduring for 10 sec.
Determination of Static Offset, Slip Velocity,
and Rise Time
Based on the data presented in Figure 5b, an example
is given to illustrate how the value of was estimated. A
˙
¯
U
data segment was demarcated (between arrows) from the
onset of the rising part at a time of about 12 sec to a time
of about 16 sec when further displacement began to fade
away. Using a fit of least squares, a solid line was drawn to
approximate a ramp envelope. Its slope gives the slip veloc-
ity: ⳱127 cm/sec for dip slip and ⳱110 cm/sec for
˙˙
¯¯
UU
ds
strike slip, both with a linear correlation coefficient of 0.97.
At this step, both and may be considered as average
˙˙
¯¯
UU
ds
values over the demarcated time segment because there are
some variations in the details. In reality, the slip history can
be fairly complex from place to place on the fault plane.
The rise time is estimated as the ratio of to D. The
˙
¯
U
last step is to determine the final displacement when the
earthquake rupture almost arrests. Here, the final displace-
ment is taken from the mean value of the late arrivals (time
⬎22 sec), which stay at a relatively static level. The esti-
mates of static offset for the two aforementioned compo-
nents are D
d
⳱798 cm for dip slip and D
s
⳱636 cm for
strike slip. The corresponding rise times are s
d
⳱4.0 sec
and s
s
⳱3.6 sec, respectively. The negligible discrepancy
(0.4 sec) between the rise times for the two components
supports our continuing usage of Brune’s slip model and our
processes.
Following the procedures outlined previously, the dip-
1196 W. G. Huang, J. H. Wang, B. S. Huang, K. C. Chen, T. M. Chang, R. D. Hwang, H. C. Chiu, and C. C. Tsai
Table 1
Source Parameters Determined by Offset and Particle Velocity from the Observations Using the Brune’s Model
Station D
d
(cm) D
s
(cm) (cm/sec)
˙
¢
U
d
s
d
(sec) T(sec) T
r
(sec) Dr(bars) c
TCU068 887 559 247 (157) 2.3 (2.2) 9.0 6.7 392 0.98
TCU102 87 72 12 4.6 10.5 5.9 19 0.97
TCU052 798 636 127 (110) 3.9 (3.6) 9.5 5.6 202 0.97
TCU067 134 13 44 1.8 4.0 2.2 70 0.96
TCU065 150 54 49 1.9 4.0 2.1 78 0.97
TCU075 97 4 44 1.4 3.0 1.6 70 0.98
TCU076 89 22 34 1.6 3.3 1.7 54 0.91
TCU129 102 30 34 1.9 3.6 1.7 54 0.95
D
d
, static offset of dip-slip component; D
s
, static offset of strike-slip component; average particle velocity of dip-slip component; s
d
, rise time
˙
¢
U,
d
determined from dip-slip component; T, duration for the ground displacement to reach its final offset; T
r
, rupture time that separated by s
d
and T;Dr, stress
drop based on Brune’s model; c, correlation coefficient. Number in parentheses denotes the average slip velocity and rise time estimated for the strike-slip
component, respectively.
and strike-slip displacement waveforms were used to esti-
mate Dand . The for each station was estimated with a
˙˙
¯¯
UU
correlation coefficient better than 0.91. This would increase
the accuracy of the calculation for Dr and s, since they are
proportional to . Table 1 lists the estimated values of D
d
,
˙
¯
U
D
s
, , and s
d
at each station. Also, the and s
s
at stations
˙˙
¯¯
UU
ds
TCU052 and TCU068 on the hanging-wall side are included.
Because of the complexity in the rising part of the displace-
ment waveforms at other stations, we did not estimate ˙
¯
U
s
and s
s
for all stations. At present, we only use D
d
and to
˙
¯
U
d
calculate the stress drop at each station.
Source Parameters of the Chi-Chi Earthquake
On the footwall side, the static offset D
d
ranges from 87
to 150 cm, whereas a D
d
with amplitude greater than 8 m
was found on the hanging-wall side. Similar disparity ap-
pears also for the strike-slip component: D
s
ranges from 4
to 72 cm on the footwall side but 559 to 636 cm on the
hanging-wall side. Also, the D
d
/D
s
ratios give a steeper slip
angle of 70⬚–88⬚at the five southern stations than that of
50⬚–58⬚at the three northern stations. This distinction in slip
angle implies that the southern segment mainly acts as a
thrust fault mechanism, while a thrust fault with strike-slip
mechanism is for the northern segment. The changing slip
is an important factor for any fault modeling.
The average particle velocity determined from the
˙
¯
U
d
footwall side varies from 12 to 49 cm/sec, which is much
less than the 127 cm/sec and 247 cm/sec determined at sta-
tions TCU052 and TCU068 on the hanging-wall side, re-
spectively. Since the stress drop Dr is directly proportional
to , large stress drop occurs on the hanging wall (202 bars
˙
¯
U
d
at station TCU052 and 392 bars at station TCU068); and
relatively small stress drops (19 to 78 bars) occur on the
footwall side.
The near-fault observations indicated that the rise times
vary by a factor of 3.3 (1.4–4.6 sec) in this study. In general,
the rise times are nearly identical for the five southern sta-
tions (average, 1.8 sec; range, 1.4–1.9 sec), but are greater
and more varied for the northern three stations (average, 3.6
sec; range, 2.3–4.6 sec). The increased rise time is particu-
larly evident for stations TCU052 (⬃4.0 sec) and TCU102
(⬃4.6 sec). Based on Brune’s model, the rise times yield
source radii of about 5.4 km and 10.8 km in the southern
and northern sections of the fault, respectively.
Anderson and Richard (1975) estimated the ground mo-
tion from different dislocation models and demonstrated that
the rise time and rupture velocity (i.e., rupture time) can be
traded off to produce very similar waveforms. This similarity
in waveforms means it will be difficult to separate the effect
of rise time and rupture velocity unless their relationship is
assumed. Because the rise times differ between two seg-
ments of the fault, the difference in duration (T) that the
dislocation takes to reach its final state may mostly be caused
by the rupture time difference. From the rise time, the rup-
ture time (T
r
) can be separated by examining T. The esti-
mated T
r
and Tfor each station is given in Table 1. On
average, T
r
is 1.9 sec in the southern segment and 6.0 sec in
the northern segment. This observed difference suggests that
rupture along the northern segment required a rupture ve-
locity lower than that along the southern segment. The es-
timated rupture times and source radii indicate that the fault
ruptured from the focus (near station TCU129) toward the
north at an average velocity of 2.84 km/sec until it reach
station TCU067, and slowed down to about 1.8 km/sec along
the northern section.
Discussion
In the present analysis, the observations were made
along the Chelungpu fault, which follows approximately
along the front of the Western Foothills that define the east-
ern border of the coastal plain area (Fig. 1). The area around
the fault zone is covered by the Pleistocene–Recent allu-
vium. Since our stations are close to the fault traces (within
2 km), the transmission properties are simple for shorter and
more homogeneous paths. Hence, the near-surface alluvium
layers would probably have affected the latter part of the
seismograms but not the rising part or the amplitude of the
Estimates of Source Parameters for the 1999 Chi-Chi, Taiwan, Earthquake Based on Brune’s Source Model 1197
early arrivals. Therefore, the sedimentary overburden should
not have impacted the major conclusions reached in this
study.
Static Offset and Average Slip Velocity
The static offset D
d
was large (8–9 m) on the hanging-
wall side, but much less (0.9–1.5 m) on the footwall side.
We believe that this offset distribution is reasonable not only
because it is compatible with the changes in ground defor-
mations, but also because it is in agreement with the results
obtained by Ma and Lee (2000), whose finite-fault inversion
model shows a maximum surface slip of about 10 m at the
northern section of the fault with relatively small slips in the
southern section. Besides, we found that the average slip
velocity, , greater than 1.3 m/sec in the hanging-wall side
˙
¯
U
d
is at least 3.4 times greater than that on the footwall side.
The smaller static offset and the slower average slip velocity
on the footwall side imply that the footwall remained almost
stationary during the faulting.
In addition, the variation in D
d
/D
s
ratios signals a chang-
ing slip during rupture propagation. The average slip angle
declines from 78⬚in the southern part to 53⬚in the north.
The change in slip angle implies that the faulting changed
from a thrust fault to a strike-slip fault as the rupture prop-
agated northward. The two-segmented fault geometry was
supported by evidence from teleseismic waveform inversion
(Lee and Ma, 2000) and from near-field records (Wu and
Ma, 2000). Both studies concluded that one single focal
mechanism could not adequately represent the observed and
synthetic seismograms at all stations. In order to achieve a
good agreement with the waveforms, they allowed the slip
angle to change at 40 km northward from the epicenter (i.e.,
close to station TCU052). Chen et al. (2000) found a vari-
able slip through fault-trace mapping. They reported that the
fault motion acted primarily as a thrust in the focal region
but was accompanied by strike-slip motion away from the
hypocenter.
Rise Time and Rupture Velocity
Apart from the focal mechanism, the rise times (Table
1) also characterize and distinguish two segments of the
fault. In the southern segment, the static deformation began
with a sharp rise time and grew relatively slowly in the
northern segment. The long rise times of 4.0 and 4.6 sec at
stations TCU052 and TCU102 agree with the rise time of
greater than 4 sec used by Huang et al. (2000) in their syn-
thetic seismograms. In addition to the long rise time, our
results also indicate a decreasing rupture velocity of 1.8 km/
sec in the northern segment. This value of 1.8 km/sec seems
consistent with the rupture velocity of 2 km/sec in the north-
ern section of the fault proposed by Huang et al. (2000). As
mentioned earlier, the southern segment was bombarded
with more short-period radiation as compared with the north-
ern segment (Figs. 3, 4). One possible reason for such dis-
tinction is that each segment has its own rise time and rup-
ture velocity. For a shorter rise time and a higher rupture
velocity, the ground accelerations and velocities are marked
with high-frequency components, while a longer rise time
and a slower rupture velocity yield low-frequency wave-
forms.
Characters of the Two-Segmented Fault
In terms of source parameters, both fault segments show
similarities and differences. The southern segment runs from
station TCU129 (closest to the epicenter) to station TCU067
(in the middle part of the fault) and is about 30 km long.
The northern segment, about 25 km long, starts somewhat
at station TCU067 (about 40 km from the epicenter) and
ends at station TCU068 (at the northern end of the fault). As
the rupture initiated and propagated to the north, the fault
experienced predominantly a thrust motion, raking at 78⬚.
Slip in this segment can be described as a uniform disloca-
tion with an amplitude of 1 m and an average rise time of
1.8 sec. The local low slip velocity (several decimeters per
sec) demonstrates that the region around the southern seg-
ment is within the low-stress area (several tens of bars).
Along the northern segment, the strike-slip component
catches up with the thrust component. The rise time averages
3.6 sec, which is twice that along the southern segment, and
implies that the rupture size was growing significantly in the
northern segment. The high slip velocities (several meters
per sec) at stations TCU052 and TCU068 indicate that the
northern segment was highly stressed (in the several hun-
dreds of bars range). In their teleseismic study of the Chi-
Chi earthquake, Kikuchi et al. (2000) pointed out that the
rupture pattern might have resulted from the rupturing of
two separate large-slipped areas (i.e., asperity) at shallow
depth. The smaller one is close to the epicenter, and the
larger one is located 35 km north. Our results are consistent
with theirs.
The highly stressed area around the northern segment
can be viewed to have resulted from relatively higher rock
strength. The abrupt change in stress drop between the two
segments suggests that during the course of rupturing the
rupture propagation may have encountered a patch that was
strong enough to temporarily bear the increasing strain with-
out breakage. This suggestion of a stronger patch is sup-
ported by the fact that the fault trace changes its course from
due north at the southern end to northeast in the middle part
of the Chelungpu fault. The course change served as a geo-
metrical barrier that prevented the fault from rupturing fur-
ther. This temporary suspension of rupture is supported by
the evidence of a relatively large time delay in the rising
motion between stations TCU052 and TCU067 (Fig. 6b).
Because the northward rupturing stopped momentarily at the
intermediate branch of the fault (around station TCU067),
the stress would have been enhanced near this region until
the concentrated stress exceeded the strength of the localized
patch. Such high-stress resistance may slow or suspend the
motion of the rupture front and delay the static deformation
reaching its final offset. A longer rise time and a slower
rupture velocity result in the broadening of the pulse width
1198 W. G. Huang, J. H. Wang, B. S. Huang, K. C. Chen, T. M. Chang, R. D. Hwang, H. C. Chiu, and C. C. Tsai
(i.e., decreasing the high-frequency content), which is evi-
dent in the velocity and acceleration waveforms recorded at
stations TCU052, TCU102, and TCU068 (Figs. 3, 4). As the
patch obstacle was breached, the heightened stress was re-
leased promptly. This relaxation of stress at the strong patch
behaved like crushing an asperity to induce an extensive
sliding on the fault plane. High slip velocity and large slip
along the northern segment are consequences of such patch
or asperity removal. This may be a reason why severe dam-
ages aligned with the northern fault segment.
Conclusions
The source parameters of the Chi-Chi earthquake of
1999 were determined using near-fault recordings. In order
to assess the behavior of relative motion between the fault
blocks during the rupture, the observed displacement on the
ground surface was transformed onto the fault plane. Using
this approach, we have characterized the source parameters
in terms of offset and particle velocity of Brune’s model. It
is obvious that the rupture did not progress with the same
rise time, nor did it result in a uniform rate of stress release
along the fault. The gross features of the Chi-Chi earthquake
are summarized for the two identified fault segments: (1)
The southern segment: thrust fault type, east dipping, slip
with a raking of 78⬚; fault length, 30 km; average disloca-
tion, 1 m; average rise time, 1.8 sec; average slip velocity,
34–50 cm/sec; average stress drop, 65 bars; average rupture
velocity, 2.84 km/sec. (2) The northern segment: fault type,
east dipping, strike slip with a raking of 53⬚; fault length, 25
km; average dislocation, ⬎8 m; average rise time, 3.6 sec;
average slip velocity on the fault plane, ⬎2 m/sec; stress
drop, ⬎300 bars; average rupture velocity, 1.8 km/sec.
Acknowledgments
The authors would like to express their sincere gratitude to the strong-
motion data processing group at the Central Weather Bureau for providing
the accelerograms. We gratefully acknowledge comments by an anony-
mous reviewer. This study was supported by Academia Sinica and the
National Science Council, R.O.C., under Grant NSC89-2116-M-001-038-
EAF.
References
Anderson, J. G., and P. G. Richards (1975). Comparison of strong ground
motion from several dislocation models, Geophys. J. R. Astr. Soc. 42,
347–373.
Brune, J. N. (1970). Tectonic stress and the spectra of seismic shear waves
from earthquakes, J. Geophys. Res. 75, 4997–5009.
Chang, C. H. (2000). Relocated Chi-Chi, Taiwan earthquake and its char-
acteristics, EOS Trans. Am. Geophys. Union 81, no. 22, 104.
Chen, W. S., Y. G. Chen, T. K. Liu, N. W. Huang, C. C. Lin, S. H. Sung,
and K. C. Lee (2000). Faulting characteristics of 921 Chi-Chi earth-
quake, in 2000 Annual Meeting of the Geol. Soc. of China, Taipei,
Taiwan, 30–31 April, 5–8.
Chen, Y. L. (1998). Three-dimensional velocity structure and kinematic
analysis in the Taiwan area, Master’s Thesis, National Central Univ.,
Taiwan, 172 pp. (in Chinese).
Chung, J. K., and T. C. Shin (1999). Implications of rupture processes from
the displacement distribution of strong motions recorded during the
21, September 1999 Chi-Chi, Taiwan earthquake, TAO 10, 777–786.
Heaton, T. H. (1990). Evidence for and implications of self-healing pulses
of slip in earthquake rupture, Phys. Earth Planet. Interiors 64, 1–20.
Huang, B. S., K. C. Chen, W. G. Huang, J. H. Wang. T. M. Chang, R. D.
Huang, H. C. Chiu, and C. C. Tsai (2000). Characteristics of ground
motion across a thrust tip from the September 21, 1999, Chi-Chi,
Taiwan earthquake, Geophys. Res. Lett. 27, 2729–2932.
Kikuchi, M., Y. Yagi, and Y. Yamanaka (2000). Source process of the Chi-
Chi, Taiwan earthquake of September 21, 1999 inferred from tele-
seismic body waves, Bull. Earthquake Res. Inst. 75, 1–13.
Lee, K. S., and K. F. Ma (2000). Rupture process of the 1999 Chi-Chi,
Taiwan earthquake from the inversion of teleseismic data, TAO 11,
591–608.
Ma, K. F., and S. J. Lee (2000). Damage patterns associated with the dy-
namic rupture behavior of the 1999 Chi-Chi, Taiwan earthquake, in
Proceeding of Taiwan-France Symposium, on Natural hazard miti-
gation, Taipei, Taiwan, 22–25 May, 89–92.
Sato, R. (1989). Handbook of Fault Parameters for Japanese Earthquakes,
Kajima Institute Publishing Co. Ltd, Tokyo (in Japanese).
Savage, J. C. (1972). Relation of corner frequency to fault dimensions, J.
Geophys. Res. 77, 3788–3795.
Wu, H. I., and K. F. Ma (2000). Rupture process of the Chi-Chi earthquake
from the inversion of near source strong motions, in 2000 Annual
Meeting of the Geol. Soc. of China, Taipei, Taiwan, 30–31 April,
138–140.
Institute of Earth Sciences, Academia Sinica
P.O. Box 1-55, Nankang
Taipei, 115 Taiwan, R.O.C.
Manuscript received 10 November 2000.