Content uploaded by Hossein Tahghighi
Author content
All content in this area was uploaded by Hossein Tahghighi on Feb 25, 2019
Content may be subject to copyright.
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
Finite Source Simulation of Near-Fault Strong Motion Records
from the 1999 Chi-Chi, Taiwan Earthquake
Hossein Tahghighi1, Kazuo Konagai2
1University of Kashan, Ravand Street, Kashan, P.O.BOX: 87317-51167, Iran
2Institute of Industrial Science, University of Tokyo, Tokyo 153-8505, Japan
tahghighi@kashanu.ac.ir
Abstract
The stochastic method for finite fault is applied to simulate near-source ground motions from the 21
September 1999, Mw 7.6 Chi-Chi, Taiwan dip-slip earthquake. We simulated accelerograms, peak ground
motions, and response spectra from 35 strong motion stations within a distance of about 10 km from the
ruptured fault plane. The simulations are in good agreement with the examined records. However, the
stochastic predictions are better constrained at distances >1.5 km. This primarily is attributed to the
significant near-fault effects at the close vicinity of the ruptured plane related to this earthquake.
Keywords: Stochastic, Finite source, Near-fault, Strong motion.
1. Introduction
We have been expanding cities in the past several decades without paying enough attention to seismic faults.
With a rapid population growth in the 20th century, many people are now living in disaster prone areas where
nobody used to live. The 1999 Kocaeli earthquake in Turkey and 1999 Chi-Chi earthquake in Taiwan
showed the devastating effects of fault rupture on dwellings and civil infrastructures. This is a serious threat
to mega cities spreading over active fault traces, and is posing us difficult problems about minimizing the
fault-rupturing-related damage. Near-fault motions are noticeably influenced by the forward directivity when
the fault rupture propagates toward a site and by permanent ground displacement, so called ‘fling-step’,
resulting from tectonic movement. These pulse-type motions have been identified as critical in the elastic and
inelastic design of engineering structures subjected to near-fault records [1- 4].
Stochastic modeling of earthquake radiation has been widely applied to predict strong ground motions
treating causative fault as a point source. The point-source approach resulted in successful predictions at high
frequency (f ≥2 Hz) and as far as the fault is located at distance large compared to its dimension [5, 6].
However, at near-source regions of large earthquakes and at low-to-intermediate frequencies (0.1< f <2 Hz),
finite-source modeling technique has been an important part of ground motion prediction [7-10]. Beresnev
and Atkinson [11] have developed a procedure for the stochastic simulation of strong motion from finite fault
ruptures. The fault plane is discretized into elements, each element is treated as a small source; similar to
Boore [6]. This method has been applied to consider rapture along finite fault plane in various tectonic
environments.
In this paper, we simulate the strong ground motion of the Mw 7.6, 1999 Chi-Chi, Taiwan earthquake
from a vast near-source database by using the stochastic method for finite faults [11, 12]. We modeled
accelerograms, peak ground motions, and response spectra from all available strong motion stations. The
database includes all records within a distance of about 10 km from the ruptured fault [13]; mainly located at
sites with very stiff soils. We simply used generic site amplification based on estimated average shear-wave
velocity over the upper 30 m, Vs30 [14]. Our primary targets are to investigate how near-fault effects related
to this earthquake affected the distribution of strong ground motions and whether the employed stochastic
method is capable of predicting broadband ground motion time histories. From an engineering standpoint,
these results are important for structures with intermediate-to-long natural periods (T ≥0.5 Sec). Such
structures, if located near a large active fault, could be subjected to significantly higher long-period
amplitudes in the direction normal to the fault.
2. Near-source strong ground motions database
The database compiled in this study consists of a comprehensive strong ground motions that were observed at
stations located within about 10 km from the activated Chelungpu fault during the Mw7.6 1999 Chi-Chi,
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
2
Taiwan earthquake (see Fig. 1). Detailed information including station names, site classifications,
hypocentral distances, closest distances to the fault plane, and geographical coordinates of the stations are
provided in Table 1. The hypocentral distances, R, range from 9.4 to 63.8 km.
Stations TCU065, TCU084, TCU129, and WNT were recorded the top fourPGA . However, these
four stations have questionable records from the 1999 Chi-Chi earthquake and their records should be used
with caution [16]. These stations are denoted by an asterisk at the right side of their name in Table 1.
Excluding these problematic stations, the mean and median of PGA at the examined near-fault stations are
281.9 and 231.9 Gal respectively which is fairly low for an event of magnitude Mw7.6. The recorded low
peak ground accelerations during the mainshock of the Chi-Chi event has been related to the effect of
faulting mechanism and geometry [17, 18].
(b) (a)
Figure 1. (a) Distribution of 441 Accelerograph stations that recorded the 1999 Mw 7.6 Chi-Chi, Taiwan
earthquake (after [15]). (b) 35 near-fault stations with closest distance to rupture plane less than about 10 km;
used in this study.
3. Simulation method
This section provides a summary and application to the stochastic finite-fault radiation modeling proposed by
Beresnev and Atkinson [11] to synthesize near-source ground motions from the 1999, Mw7.6 Chi-Chi,
Taiwan earthquake. This method involves discretization of a rectangular fault plane into smaller elements
(subfaults), each of which is assigned an 2
ω
Brune point source spectrum. The contributions from all
subfaults are empirically attenuated to the observation site and summed to produce the synthetic acceleration
time history. A simple kinematic model of the Hartzell [7] type is used to simulate the rupture propagation,
which is assumed to start at the hypocenter and radially propagate from it. There is no intention to
reintroduce the stochastic modeling of the near-source Chi-Chi earthquake (see [16] for more details).
Following [16], a rectangular fault with 89 km long and 30 km wide, a strike of 5 deg, and an easterly dip of
30 deg was divided into 10×3 subfaults (see Fig. 2). The seismic moment of th
ij subfault, 0ij
m, is controlled
by its relative slip weight,Sij , and seismic moment for the entire fault plane, 0
M
, as follows [11].
00/( )
11
nl nw
mMS S
pq
ij ij pq
=∑∑
==
(1)
The total number of subfaults is equal to
N
nl nw
=
×. The seismic moment of the subfault in Eq. 1 is related
to its length
L
∆and stress drop of σ [11] by
03
.mLσ= ∆ (2)
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
3
Table 1. Near-fault strong ground motions database considered in this study.
R d Lat. Lon.
PGAEW PGANS PGA
No. Station Site Side (km) (km) (oN) (oE) (cm/s/s) (cm/s/s) (cm/s/s)
1 CHY024 S Fw 25.4 7.7 23.76 120.61 276.4 162.1 219.3
2 CHY028 sr Fw 33.6 2.3 23.63 120.61 624.4 749.9 687.2
3 CHY101 S Fw 33.0 7.7 23.69 120.56 333.0 390.5 361.8
4 NSY+ sr Fw 63.8 9.7 24.42 120.76 119.4 115.1 117.3
5 TCU+ S Fw 37.1 5.0 24.15 120.68 200.5 187.1 193.8
6 TCU049 s Fw 39.7 2.7 24.18 120.69 273.4 240.9 257.2
7 TCU051 s Fw 39.4 6.9 24.16 120.65 156.8 231.0 193.9
8 TCU052 sr Hw 40.4 0.8 24.20 120.74 349.8 438.1 394.0
9 TCU053 s Fw 42.0 4.6 24.19 120.67 225.0 132.1 178.6
10 TCU054 s Fw 38.5 4.7 24.16 120.68 143.3 190.4 166.9
11 TCU055 s Fw 36.7 6.5 24.14 120.67 257.1 208.5 232.8
12 TCU060 s Fw 46.1 8.5 24.22 120.64 196.9 101.1 149.0
13 TCU065
*
s Fw 27.9 0.1 24.06 120.69 773.3 563.3 668.3
14 TCU067 s Fw 29.8 0.2 24.09 120.72 488.6 312.9 400.8
15 TCU068 s Hw 48.5 0.2 24.28 120.77 500.6 364.1 432.4
16 TCU071 r Hw 17.4 4.1 23.99 120.79 517.8 637.5 577.7
17 TCU072 s Hw 22.9 6.8 24.04 120.85 467.2 370.5 418.9
18 TCU074 sr Hw 20.7 11.4 23.96 120.96 585.0 368.4 476.7
19 TCU075 sr Fw 22.2 0.6 23.98 120.68 325.4 257.3 291.4
20 TCU076 r Fw 17.9 2.3 23.91 120.68 340.5 419.7 380.1
21 TCU078 s Hw 9.4 5.4 23.81 120.85 438.2 302.4 370.3
22 TCU082 s Fw 37.1 5.0 24.15 120.68 222.3 182.5 202.4
23 TCU084
*
r Hw 12.0 10.4 23.88 120.90 989.4 422.8 706.1
24 TCU087 sr Fw 56.2 5.8 24.35 120.77 119.3 111.7 115.5
25 TCU089 sr Hw 10.7 6.2 23.90 120.86 346.1 223.9 285.0
26 TCU101 sr Fw 45.8 1.5 24.24 120.71 207.7 253.6 230.7
27 TCU102 r Fw 46.3 0.6 24.25 120.72 298.3 170.0 234.2
28 TCU103 s Fw 53.0 4.4 24.31 120.71 126.6 149.4 138.0
29 TCU110
*
s Fw 29.5 11.6 23.96 120.57 177.7 187.8 182.8
30 TCU116 s Fw 25.7 11.5 23.86 120.58 185.5 132.9 159.2
31 TCU120 sr Fw 26.8 6.1 23.98 120.61 223.2 193.6 208.4
32 TCU122 s Fw 23.2 8.5 23.81 120.61 206.2 255.7 231.0
33 TCU128 sr Fw 63.8 9.7 24.42 120.76 141.1 162.9 152.0
34 TCU129
*
sr Fw 16.3 1.5 23.88 120.68 981.8 609.8 795.8
35 WNT+
*
sr Fw 16.3 1.5 23.88 120.68 920.5 602.0 761.3
+: Stations WNT, NSY and TCU have same location with TCU129, TCU128 and TCU082 respectively.
*: TCU065, TCU084, TCU110, TCU129, and WNT have questionable records.
s: Soil, r: Rock, sr: Soft rock.
Hw: Hanging wall, Fw: Footwall
R: Hypocentral distance.
d: Closest distance to the rupture plane.
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
4
Lat., Lon.: Geographical coordinates, latitude and longitude.
PGAEW, PGANS: Peak Ground Acceleration in East-West and North-South directions, respectively.
PGA: Geometric average of the recorded Peak Ground Acceleration in the two horizontal directions.
Figure 2. 3D perspective of assumed fault rupture plane geometry (dimensions, dip, strike, and hypocenter) in the
1999 M7.6 Chi-Chi earthquake simulation.
Table 2. Modeling parameter
Parameter Value
Fault orientation Strike5o/Dip30o
Depth to the upper edge of fault (km) 0
Fault dimensions along strike and dip (km) 89 by 30
Subfault dimensions (km) 8.9
×
10
Crustal shear-wave velocity (km/sec) 3.2
Rupture velocity (km/sec) 0.8
×
(shear-wave velocity)
Crustal density (g/cm3) 2.7
Geometric spreading 1/R (R<50 km)
1/R0 ( 50 km
≤
R<170 km)
1/R0.5 ( 170 km ≤R)
Stress parameter (bar) 50.
Kappa, high-cut filter parameter (sec) 0.07
Anelastic attenuation; Q(f) 117f0.77
Site amplification Generic rock/soil
Windowing function Cosine-tapered boxcar
Radiation strength factor, controlling max. slip rate
Slip distribution 1.0
Homogeneous
Stress drop was kept constant at the value of 50 bars [12]. The attenuation effects of the propagation path
were taken into account through the empirical anelastic and geometric attenuation operators [18]. The crustal
amplification effect is then modeled by multiplying the spectrum by the frequency-dependent factors
proposed for generic rock sites in Western North America, WNA, described by shear-wave velocity Vs30
[14]. Table 2 summarizes all parameters, i.e. simulation domain, source parameters, wave propagation
effects, used in the employed stochastic method for synthesizing of ground motions (see [16] for details).
4. Results and discussions
After screening out the problematic records, ground accelerations, response spectra, and peak ground motions
are simulated from the stochastic scheme for finite fault. In Fig. 3, the observed and simulated acceleration
time histories, synthesized by stochastic finite fault model, at four representative near-fault stations have been
compared. As shown in this figure, the PGA, S-wave part of both the east-west and north–south observed
acceleration components (top two traces) and ground motion durations are well matched with the random
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
5
synthesized horizontal accelerograms (bottom traces). The simulated motions have the same sampling
interval (0.005 sec) as the recorded traces to which they are compared. Figure 4 compare the observed and
simulated peak ground accelerations, PGAs, synthesized by stochastic finite fault model at the near-source
stations shown in Table 1. The observed PGA at each site was calculated as the geometric average of the
PGAs of the two horizontal components; i.e. PGA in Table 1. Bias of simulated ground motions as a
function of hypocentral distance and closest distance to the rupture plane were shown in Figs. 4a and 4b
respectively. Bias is defined as the logarithm (base 10) ratio of the observed to the predicted PGAs. Peak
ground accelerations are generally well reproduced, except from very few stations where an error of more
than a factor of 2 is observed. Taking into account the complexity of the examined event, the simplicity of
the method and the fact that local site effect were basically calculated based on amplification factors
suggested for generic sites, the fit can be considered very satisfactory. The obtained results in [16] showed a
satisfactory match between synthetics and observed response spectra as well.
Figures 5 shows comparison between the stochastic simulations and observed response spectra at
frequencies of 0.2 Hz and 5.0 Hz. Comparisons are plotted as function of distance to fault according to GSA
and NSR models, respectively. The results are in a general agreement with each other in the distance range
considered in this study. However, the observed spectra exceed the predictions by a significant amount at low
frequencies (f = 0.20 Hz) for distances less than 1.5 km from the fault. At larger distances, the simulations
agree well with the recorded spectra.
Figure 6 plots the mean residuals for the near-fault stations in the two distance ranges of 0 – 1.5 km
and 1.5 – 10 km, respectively. At distances less than 1.5 km from the fault and at lower frequencies (f <#1
Hz), the stochastic finite-fault method tends to under-predict the ground motions by a significant factor of 2.5
(0.4 log units). It is possible that this underestimation is due to the effects of near-source forward directivity
and fling-related long-period behaviors.
20 30 40 50 60
-150
0
150
CHY024
Obs SN, PGA = 0.28 g
20 30 40 50 60
-150
0
150
Acceleration (cm/sec/sec)
Obs SP, PGA = 0.17 g
010 20 30 40
-150
0
150
Time (sec)
Sim, PGA = 0.26 g
20 30 40 50 60
-150
0
150
TCU075
Obs SN, PGA = 0.33 g
20 30 40 50 60
-150
0
150
Acceleration (cm/sec/sec)
Obs SP, PGA = 0.26 g
010 20 30 40
-150
0
150
Time (sec)
Sim, PGA = 0.32 g
20 30 40 50 60
-150
0
150
TCU120
Obs SN, PGA = 0.23 g
20 30 40 50 60
-150
0
150
Acceleration (cm/sec/sec)
Obs SP, PGA = 0.20 g
010 20 30 40
-150
0
150
Time (sec)
Sim, PGA = 0.19 g
20 30 40 50 60
-150
0
150
TCU122
Obs SN, PGA = 0.21 g
20 30 40 50 60
-150
0
150
Acceleration (cm/sec/sec)
Obs SP, PGA = 0.26 g
010 20 30 40
-150
0
150
Time (sec)
Sim, PGA = 0.24 g
Figure 3. Comparison between observed and synthetic acceleration time histories at four representative ground
motion stations. The upper two wave forms are the recorded horizontal components (strike–normal and strike–
parallel), whereas the bottom traces are the predicted stochastic horizontal component using Generic Site
Amplification (GSA) scheme of Boore and Joyner [14]. The PGA is also shown for each accelerogram.
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
6
010 20 30 40 50 60 70
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
BJ, Obs(avg of EW and NS) & Sim(Sto)
Hypocentral distance, dH (km)
PGA; Log(Obs/Sim)
(a)
0 2 4 6 8 10 12
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
BJ, Obs(avg of EW and NS) & Sim(Sto)
Distance-to-fault, dR (km)
PGA; Log(Obs/Sim)
(b)
Figure 4. Bias of simulated strong ground motions, showing the logarithmic ratio of average observed horizontal
PGA to synthetic PGA versus (a) earthquake hypocentral distance, dH, at the near-fault stations depicted in
figure 1-b. (b) closest distance to rupture plane, dR, at the same stations. Simulated spectra have been amplified
according to GSA model.
5. Conclusions
We simulated a comprehensive near-source ground motions that were recorded during the 1999 Chi-Chi
earthquake using the stochastic finite-fault method. The near-source zone was assumed to be restricted to
within a distance of about 10 km from the ruptured fault. The results show a satisfactory match between
synthetics and observed accelerograms and also response spectra, although significant discrepancies were
observed at individual sites. To generate more accurate ground motion time histories at specific locations,
more reliable source and attenuation parameters as well as site-specific response functions would be useful.
We concluded that the mean ratio of simulated to observed spectra is very close to unity for frequencies
larger than about 1 Hz, whereas at lower frequencies and very close-to-fault distances (d <1.5 km) there is a
systematic under-prediction. Therefore, the stochastic method for finite-fault provides a sound basis for
estimation of ground motions at near-fault regions on average. However, the method may lack in adequate
prediction of exceptional waveforms with strong long-period velocity pulses and large permanent ground
displacements at near-fault sites sufficiently close to the rupture plane. These special aspects of near-fault
motions will be addressed in another paper.
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
7
0 2 4 6 8 10 12
10
1
10
2
10
3
BJ, f = 0.2 Hz, Obs & Sim(Sto)
Distance-to-fault, dR (km)
Accel. response spectrum (cm/sec/sec)
Obs.(EW)
Sim.(Sto)
(a)
0 2 4 6 8 10 12
10
1
10
2
10
3
BJ, f = 5.0 Hz, Obs & Sim(Sto)
Distance-to-fault, dR (km)
Accel. response spectrum (cm/sec/sec)
Obs.(EW)
Sim.(Sto)
(b)
Figure 5. Distribution of simulated and observed 5% damped acceleration response spectra with distance to
rupture plane. Comparisons are plotted at frequencies of (a) 0.2 Hz and (b) 5.0 Hz. Simulated spectra have been
amplified according to GSA model.
10
-1
10
0
10
1
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
BJ, Obs(EW) & Sim(Sto)
Fre
q
uenc
y
(
Hz
)
Accel. response spectrum residual; Log(Obs/Sim)
0-1.5 KM
1.5-10 KM
Figure 6. Mean residual showing the ratio of simulated to recorded spectrum, averaged over the near-source
stations depicted in Table 1, for closest distance to fault ranges, d, of 0-1.5 km and 1.5-10 km.
8th International Congress on Civil Engineering,
May 11-13, 2009, Shiraz University, Shiraz, Iran
8
6. Acknowledgment
Partial financial support for this study was provided by the Japanese government. The first author would like
to acknowledge this research sponsorship under the postdoctoral fellowship at the University of Tokyo.
7. References
1. Hall JF, Heaton TH, Halling MW, Wald DJ (1995), “Near source ground motion and its effects on flexible buildings,”
Earthquake Spectra, 11, pp 569–606.
2. Sasani M, Bertero VV (2000), “Importance of severe pulse-type ground motions in performance-based engineering:
Historical and critical review,” Proceedings of the 12th World Conference on Earthquake Engineering (12WCEE),
Auckland, New Zealand (CD-ROM).
3. Alavi B, Krawinkler H (2000), “Consideration of near-fault ground motion effects in seismic design,” Proceedings
12th World Conference on Earthquake Engineering (12WCEE), Auckland, New Zealand (CD-ROM).
4. Alavi B, Krawinkler H (2004), “Behavior of moment-resisting frame structures subjected to near-fault ground
motions,’” Earthquake Engng Struct. Dyn. 33, pp 687–706.
5. Hanks TC, and McGuire RK (1981), “The character of high frequency strong ground motion,” Bull. Seism. Soc. Am.
71, pp 2071-2095.
6. Boore DM (1983), “Stochastic simulation of high-frequency ground motions based on seismological models of the
radiated spectra,” Bull. Seism. Soc. Am. 73, pp 1865–1894.
7. Hartzell S (1978), “Earthquake aftershocks as Green’s functions,” Geophys. Res. Lett. 5, pp 1–14.
8. Irikura K (1983), “Semi-empirical estimation of strong ground motions during large earthquakes,” Bull. Disaster
Prevention Res. Inst., Kyoto Univ. 33, pp 63–104.
9. Somerville PG, Sen M, and Cohee B (1991), “Simulations of strong ground motions recorded during the 1985
Michoacan, Mexico and Valparaiso, Chile, earthquakes,” Bull. Seism. Soc. Am. 81, pp 1-27.
10. Atkinson GM, and Silva W (1997), “An empirical study of earthquake source spectra for California
earthquakes,”Bull Seism. Soc. Am. 87, pp 97-113.
11. Beresnev IA, and Atkinson GM (1997), “Modeling finite-fault radiation from the n
ω
spectrum,” Bull. Seism. Soc.
Am. 87, pp 67–84.
12. Beresnev IA, and Atkinson GM (1998), “FINSIM: a FORTRAN program for simulating stochastic acceleration time
histories from finite faults,” Seism. Res. Lett. 69, pp 27–32.
13. Lee WHK, Shin TC, Kuo KW, Chen KC, and Wu CF (2001a), “CWB free-field strong-motion data from the 9-21-
1999 Chi-Chi earthquake,” Seismological Observation Center, Central Weather Bureau, Taipei, Taiwan (CD-ROM).
14. Boore DM, and Joyner WB (1997), “Site amplifications for generic rock sites,” Bull. Seism. Soc. Am. 87, pp 327–
341.
15. Lee WHK, Shin TC, Kuo KW, Chen KC, and Wu CF (2001b), “CWB free-field strong-motion data from the 21
September Chi-Chi, Taiwan, earthquake,” Bull. Seism. Soc. Am., 91, pp 1370–1376.
16. Tahghighi, H. and Konagai, K., Johansson, J. (2008), “Hybrid Stochastic Simulation of Near-Fault Strong Motion
Records from the 1999 Chi-Chi, Taiwan Earthquake,” Report on JSPS Research Project for Rational Design of
Lifelines Near Seismic Faults, Grant-in-Aid for Scientific Research, The University of Tokyo.
17. Aagaard BT, Hall JF, and Heaton TH (2004), “Effects of fault dip and slip rake angles on near-source ground
motions: Why rupture directivity was minimal in the 1999 Chi-Chi, Taiwan, Earthquake,” Bull. Seism. Soc. Am. 94,
pp 155–170.
18. Roumelioti Z, and Beresnev IA (2003), “Stochastic finite-fault modeling of ground motions from the 1999 Chi-Chi,
Taiwan, earthquake: Application to rock and soil sites with implications for nonlinear site response’” Bull. Seism. Soc.
Am. 93, pp 1691–1702.
