AN ABSTRACT OF TIlE THESIS OF
Xiao-Oing Li for the degree of Master of Science in Geophysics presented on August
6. 1991.
Title: The 1988 Lancang-Gengma. China. Earthquake Sequence: Body Wave.
Abstract
Redacted for Privacy
Dr. John L. Nabelek
On November 6, 1988, two strong earthquakes (Mw: 7.0 and 6.8) separated
by about 13 minutes occurred in Yunnan Province, China. The aftershocks located
by Kunming Telemetered Seismic Network form a lineament approximately 120 km
long and 20 km wide with the long dimension oriented approximately N30°W. The
epicenter of the first event lies about 30 km from the southern terminus of the
aftershock zone while the epicenter of the second event is 60 km further to the
northwest. Field investigations indicate that the surface fault ruptures associated with
the first and second shock and a variety of ground deformations.
We analyze teleseismic data recorded by the GDSN network to determine the
rupture process of these two mainshocks (referred to as Ml and M2) and the two
largest aftershocks (referred to as Al and A2).
Inversion of long-period body waves gives the following centroid source
parameters for Ml: strike 154°±4°, dip 86°±1°, slip 181°±1°, centroid depth shallower
than 15 km (least-misfit centroid depth 12 1cm), and seismic moment 4.5-4.9 X 1026
dyn cm (least-misfit seismic moment 4.6 X 1026 dyn cm). The source time function,
further constrained by broadband seismograms, indicates that the source duration for
this event is 12 seconds.
Due to signal interference with Ml, body wave inversion techniques cannot
be applied to M2. The Rayleigh waves provide a better look at this event. In order to
identify the energy contributions from the two events, group velocity analysis was
performed on the surface wave trains. The energy from the individual events was
then isolated based on their dispersion patterns. The amplitude spectra in the period
range of 100 to 66 s were inverted for the source parameters. The inversion
constrains the strike of M2 precisely (155°±3°), however, dip and slip angles were
not well resolved by the inversion. Similar Rayleigh wave amplitude spectra and
radiation patterns of Ml and M2, however, suggest that they had very similar
mechanisms and centroid depths. On the average, the amplitude spectra of M2 are
smaller than those of Ml by a factor of 2.2, indicating the seismic moment of M2 is
2.1 X 1026 dyn cm.
The two largest aftershocks, Al (Mw 6.1) and A2 (Mw 5.3), which occurred
at the southern terminus of the aftershock zone, were analyzed by modeling
teleseismic and strong ground motion data. Teleseismic body wave inversion gives
source orientation of Al: strike 165°±2.5°, dip 900±1.50, slip 178°±0.5°, centroid
depth shallower than 12 km (least-misfit centroid depth 7 km from broadband
waveform inversion), and seismic moment 1.5-1.6 X 1025 dyn cm. The inversion of
A2 gives the source orientation and centroid depth very similar to those of Al. The
seismic moment for this event is 1.3-1.6 X 1024 dyn cm. Modeling of strong ground
motion seismograms adds more constraints on centroid depths and source time
functions of Al and A2. To minimize the effect of scattering caused by upper crustal
heterogeneity, we confined our analysis to frequencies lower than 1 Hz. A crustal
model, with a low velocity sedimentary layer, was found that predicts common
features of observed strong ground motion seismograms for both events. Derived
source orientation is consistent with that found from teleseismic body wave
inversion. The centroid depths of Al and A2 were constrained to be between 4 and
12km. A source duration of 7 s and 2 s was obtained for Al and A2, respectively.
Derived rupture parameters of Ml and M2, aftershock distribution, field
investigations, geological information and concepts of geometrical barriers and fault
asperities, indicate that the preexisting fault intersections played the key role in
rupture terminations and initiations. The 12 s source duration of Ml and about 60 km
long zone of ground deformation along the strike suggest that Ml rupture was
bilateral. The rupture initiated near a fault intersection and propagated to NNW and
SSE along the strike. The SSE propagating rupture was terminated by a preexisting
fault which intersects the ruptured fault 30 km to the south. The aftershock Al and
A2 as well as a dense group of small aftershocks were associated with the termination
of the SSE segment. The NNW propagating rupture was also terminated by a NE
striking preexisting fault on which several of the largest aftershocks appear to have
occurred. This NE slrildng fault right-laterally offsets the fault on which Ml and M2
occurred forming a geometrical barner for the rupture. M2 presumably nucleated near
this barrier and unilaterally ruptured about 25 km toward NNW where it was
terminated by a well documented preexisting fault.
The 1988 Lancang-Gengma, China, Earthquake Sequence: Teleseismic
Body Wave, Surface Wave and Strong Ground Motion Studies
by
Xiao-Qing Li
A THESIS
submitted to
Oregon State University
in partial fulfillment of
the requirements for the
degree of
Master of Science
Completed August 6, 1991
Commencement June 1992
r
.x'i '
Redacted for Privacy
jØ'r Professor: Associate Professor of Geophysics
Redacted for Privacy
College of
Redacted for Privacy
Dean of
Date thesis is presented August 6. 1991
Typed by the author Xiao-Oing Li
ACKNOWLEDGEMENTS
The most thanks go to my advisor Dr. John Nabelek for his continuous
support, encouragement and insightful guidance. I have benefited very much from
his broad knowledge and acute insight in "finding the truth from the facts". I am
very proud of being a student of him.
Thanks also go to Drs. Gary Egbert, John Chen, Roy Rathja, and Douglas
Keszler who were my committee members and kindly provided helpful comments.
I also like to thank Drs. Anne Trehu, Shaul Levi, Robert Ducan and Robert
Yeats who, in addition to the faculty members mentioned above, made contributions
to my graduate education at OSU.
It is doubtless that, without the help from Jefferson Gonor, Donna Obert and
Marcia Turnbull, I would not have to deal properly with the "requirements" set by
administration. Thank you for helping me to find the thread to do these "complicated"
things in right way.
It was enjoyable to be with my fellow geophysics student at OSU. We came
from different countries with different cultural background, and fortunately, we all
speak English. I enjoyed the discussion and conversations with Akbar, Ann, Daniel,
Ganyuan, Guibiao, Haraldur, Jochen, John Shaw, Juan, Luiz, Maureen, Pordur,
Rainer and Soofi. I appreciate whatever the help you provided. Thank you, Steven, I
will always remember our wonderful cooperations that we did in North Carolina,
Washington and Oregon.
Chris Goldfinger read part of the thesis and improved the manuscript. Dr.
Francis Wu provided us the strong ground motion data and Janyou Li sent us the
copy of their publication.
Finally, I like to express my thanks to my wife and my parents. At the last
stage of this project, my parents visited us and provided tremendous help. They felt
this earthquake and had real experience of" 12 s source duration". Of course, without
the love and support from my wife and our little baby, Amy, it would be impossible
to finish this work.
This work was funded by the National Science Foundation grant EAR
8896187/8940229.
TABLE OF CONTENTS
1
INTRODUCTION ......................................................................
1
GEOLOGICAL SETTING OF THE REGION .....................................
3
Fault Distribution and Seismicity in the Region..............................
3
1.2 Observed Ground Deformation ................................................
5
The Aftershock Distribution....................................................
6
1.1
1.3
2 TELESEISMIC BODY WAVE ANALYSIS .......................................
15
Introduction ......................................................................
15
2.1
2.2 Teleseismic Body Wave Inversion Method .................................. 16
2.3 Data ...............................................................................
18
2.3.1
Long-period data .......................................................
18
2.3.2
Broad-band data........................................................
19
2.4 Modeling of Mainshock Ml ...................................................
20
2.4.1
Long-period body wave inversion................................... 20
2.4.2
Broad-band body wave inversion.................................... 22
2.5 Modeling of Aftershock Al .................................................... 24
2.5.1
Long-period body wave inversion................................... 24
2.5.2
Broad-band body wave inversion .................................... 25
2.6 Modeling of Aftershock A2 ....................................................
26
2.7 Discussion .......................................................................
27
3 SURFACE WAVE ANALYSIS ..................................................... 44
3.1 Introduction ...................................................................... 44
3.2 Moving Window Analysis ...................................................... 46
3.2.1
Principle ................................................................. 46
3.2.2
Application to synthetic data .......................................... 47
3.3 Time-variable Filter ............................................................. 48
3.3.1
Principle
3.3.2
Application to synthetic data.......................................... 49
3.4 Rayleigh Wave Inversion Technique.........................................
.
48
51
3.5 Data and Data Analysis ......................................................... 53
3.5.1
Group velocity determinations of Ri and R2 waves ............... 54
3.5.2
Retrieving Ri and R2 waves by time-variable filtering............ 55
3.6 Rayleigh Wave Inversion of Mainshock Ml ................................. 56
3.6.1
Data ...................................................................... 56
3.6.2
Attenuation coefficient ................................................. 56
3.7 Rayleigh Wave Inversion of Mainshock M2 ................................. 59
3.7.1
Data ...................................................................... 59
3.7.2
Results and discussion ................................................
59
4 STRONG GROUND MOTION ANALYSIS ...................................... 87
4.1 Introduction ...................................................................... 87
4.2 Data and Data Analysis .........................................................
88
4.2.1
High frequency features ..............................................
4.2.2
Effect of medium heterogeneity ...................................... 90
88
4.3 Forward Modelling for a Point Source......................................
91
4.3.1
Crustal model and centroid depth ...................................
92
4.3.2
Effect of source orientation ..........................................
93
4.3.3
The best fit source time function....................................
94
4.4 Discussion ......................................................................
95
5 RUPTURE PROCESS OF THE LANCANG-GENGMA ...................... 137
EARTHQUAKE SEQUENCE
6 CONCLUSIONS .....................................................................
142
BIBLIOGRAPHY .................................................................... 144
LIST OF FIGURES
Page
Figure
1.1
Fault distribution and seismicity in Lancang-Gengma area.
11
1.2
Ground surface deformation reported by Zhou et al. (1989).
12
1.3
Ground surface deformation reported by
Zhao et al. (1989, unpublished data).
13
1.4
Distribution of the aftershocks (2M6.9) determined by KTSN
in two months following the mainshock.
Observed (solid) and synthetic (dashed) long-period P-, SV- and
SH-wave seismograms for Ml.
14
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
Normalized variance as a function of the deviations of the strike,
dip and slip from the best fit point source model of Ml.
Observed (solid) and synthetic (dashed) broadband P- and SHwave seismograms for Ml.
Normalized variance (top) and seismic moment (bottom) as
a function of centroid depth for Ml.
Observed (solid) and synthetic (dashed) long-period P-, SV- and
and SH-wave seismograms for Al.
Normalized variance as a function of the deviations of the strike,
dip and slip from the best fit point source model of Al.
Observed (solid) and synthetic (dashed) broadband P wave
seismograms for Al.
Normalized variance (top) and seismic moment (bottom) as
a function of centroid depth for Al.
Observed (solid) and synthetic (dashed) long-period P- and SHwave seismograms for A2.
Normalized variance as a function of the deviations of the strike,
dip and slip from the best fit point source model of A2.
Normalized variance (top) and seismic moment (bottom) as
32
33
34
35
36
37
38
39
40
41
42
a function of centroid depth for A2.
2.12
A comparison of P wave seismograms of Ml, Al and A2 recorded
at station HIA and GRFO.
43
3.1
3.2
3.3
3.4
Group velocity measurements of synthetic seismograms by moving
window analysis.
Measured (dots) and theoretical (straight line) group velocity.
66
Time-variable filtering of seismograms.
68
Time-variable filter retrieving linearly dispersed wave from complex
69
67
seismogram.
3.5
3.6
Station distribution and Ri paths for the Lancang-Gengma earthquake.
70
Two examples of group velocity analysis of Ri waves for stations
71
TOL (a) and HRV (b).
3.7
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
Group velocities of Ri waves.
Two examples of group velocity analysis of R2 waves for stations
ANMO (a) and AR (b).
Group velocities of R2 waves.
Two examples of time-variable filtering of Ri waves for stations
TOL (top) and HRV (bottom).
Two examples of time-variable filtering of R2 waves for Stations
AR (top) and ANMO (bottom).
Amplitude spectra period range 60-100 s of Ri (a) and R2 (b)
waves for Mi (solid) and M2 (dashed).
Attenuation models used in Rayleigh wave inversions.
Normalized variance of Ml as a function of centroid depth for
different attenuation models.
Observed (solid) and synthetic (dashed) amplitude spectra of Ml
72
73
74
75
76
77
80
81
82
at all stations used in the inversion.
3.16
3.17
Normalized variance as a function of centroid depth for Mi.
Normalized variance of M2 as a function of centroid depth for
3.18
4.1
different attenuation models.
Rayleigh wave amplitude spectra of Mi (solid) and M2 (dashed).
Observed three-component (EW, NS and Z from top to bottom)
84
85
86
99
ground acceleration (top), velocity (middle) and displacement
4.2
(bottom) for Al.
Observed three-component (EW, NS and Z from top to bottom)
100
ground acceleration (top), velocity (middle) and displacement
(bottom) for A2.
4.3
Low-pass filtering of ground acceleration record of Al.
101
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12
4.13
High-pass filtering of ground acceleration record of Al.
Acceleration spectra of Al.
Acceleration spectra of A2.
High frequency (1-10 Hz) acceleration particle motions of Al.
Low frequency (0.0667-1 Hz) acceleration particle motions of Al.
Synthetic (dashed) three-component (EW, NS and Z from top to
bottom) ground acceleration, velocity and displacement computed
for a half space crustal model and a centroid depth of 11 km for Al.
Synthetic (dashed) three component (EW, NS and Z from top to
bottom) ground acceleration, velocity and displacement computed
for a crustal model, which includes the Moho discontinuity, and a
centroid depth of 11 km for Al.
Synthetic (dashed) EW acceleration seismograms computed for six
crustal models and a source centroid depth of 4 km for Al.
Synthetic (dashed) EW acceleration seismograms computed for six
crustal models and a source centroid depth of 4 km for A2.
Synthetic (dashed) EW acceleration seismograms computed for six
crustal models and a source centroid depth of 7 km for Al.
103
105
106
107
111
115
116
117
118
119
4.14
Synthetic (dashed) EW acceleration seismograms computed for six
120
4.15
crustal models and a source centroid depth of 7 km for A2.
Synthetic (dashed) EW acceleration seismograms computed for six
121
4.16
4.17
4.18
4.19
crustal models and a source centroid depth of 10 km for Al.
Synthetic (dashed) EW acceleration seismograms computed for six
crustal models and a source centroid depth of 10 km for A2.
Synthetic (dashed) three-component (EW, NS and Z from top to
bottom) ground acceleration, velocity and displacement computed for
the best crustal model CM 1 and source centroid depth 7 km for Al.
Synthetic (dashed) three-component (EW, NS and Z from top to
bottom) ground acceleration, velocity and displacement computed for
the best crustal model CM1 and source centroid depth 10 km for A2.
Synthetic (dashed) three-component (EW, NS and Z from top to
122
123
124
125
bottom) ground displacement computed for five different strikes for Al.
4.20
Synthetic (dashed) three-component (EW, NS and Z from top to
127
bottom) ground displacement computed for five different dips for Al.
4.21
Synthetic (dashed) three-component (EW, NS and Z from top to
129
4.22
bottom) ground displacement computed for five different slips for Al.
The best fit synthetic (dashed) three-component (EW, NS and Z from
top to bottom) ground acceleration, velocity and displacement for Al.
131
4.23
The best fit synthetic (dashed) three-component (EW, NS and Z from
top to bottom) ground acceleration, velocity and displacement for A2.
132
4.24
4.25
Observed (solid) and synthetic (dashed) broadband P waves of Al.
133
The far-field source time functions of Al determined from teleseismic
134
4.26
broadband data (dashed) and strong ground motion data (solid).
Acceleration amplitude spectra ratio of P (taking from vertical
135
component) and S (taking from E-W component) waves of Al
4.27
5.1
5.2
for total 7.2 s source duration.
Acceleration amplitude spectra ratio of P (taking from vertical
component) and S (taking from E-W component) waves of A2
136
for total 2.0 s source duration.
Rupture model for Ml and M2.
Aftershocks in 3 hours follow the occurrence of Ml.
140
141
LIST OF TABLES
Table
Page
1.
Epicenter parameters of the events investigated in this thesis.
7
2.
3.
Epicentral parameters of the historical events in the region'.
8
The epicentral parameters of the aftershocks determined both by
9
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
PDE and KTSN.
Station parameters for body wave inversion of Ml.
Station parameters for body wave inversion of Al.
Station parameters for body wave inversion of A2.
Error lists forrecords of Al andA2.
Ri and R2 waves used in Rayleigh wave inversion of Ml and M2.
Ml Rayleigh wave inversion results for different Q models
(all parameters free).
Ml Rayleigh wave inversion results for different Q models
(some parameters fixed).
M2 Rayleigh wave inversion results for different Q models
different Q models.
M2 Rayleigh wave inversion results for different Q models
(some parameters fixed).
Parameters of strong motion station.
Layer parameters of the crustal models used for modeling of strong
ground motion data.
28
29
30
31
61
62
63
64
65
97
98
THE 1988 LANCANG-GENGMA, CHINA, EARTHQUAKE
SEQUENCE: TELESEISMIC BODY WAVE, SURFACE
WAVE AND STRONG GROUND MOTION STUDIES
INTRODUCTION
On November 6, 1988, two strong earthquakes (Mw: 7.0 and Mw: 6.8)
separated by about thirteen minutes occurred in Lancang-Gengma area, Yunnan
Province, China. The mainshock doublet caused widespread damage in the area. Two
thirds of the buildings in Lancang city were destroyed and 730 people were killed
(USGS). The highest intensity reported by Yunnan Province Seismological Bureau is
IX and field investigations revealed surface fault ruptures and a variety of ground
deformations in the meizoseismal area (Zhou et al., 1989; Zhao et al., 1989). In the
two months following the mainshocks, more than 4500 aftershocks (M
1.5)
occurred, according to Kunming Telemetered Seismic Network of Yunnan Province
Seismological Bureau, China. The largest aftershock, Mw: 6.1, occurred on
November 30, 1988. The two mainshocks and the largest aftershock were well
recorded by GDSN (Global Digital Seismographic Network). After the mainshocks,
strong ground motion instruments were set up in the meizoseismal area. Two largest
aftershocks in the aftershock series, which occurred on November 27 (Mw 5.3) and
November 30 (Mw 6.1), were well recorded by the strong ground motion recorders.
In addition to being well recorded at teleseismic and short distances, the
Lancang-Gengma earthquake sequence was also well monitored by Kunming
Telemetered Seismic Network which is a short-period instrument network, mainly
designed to locate the earthquakes in Yunnan Province. By January 5, 1989, two
months after mainshocks, 1029 aftershocks with magnitude greater than 2.0 have
been located by the network.
In this thesis, body waves and Rayleigh waves recorded at teleseismic
distances and strong ground motion records were used to study the source
mechanisms of the earthquake sequence (two mainshocks and two largest
aftershocks). The events studied in this thesis are listed in Table 1. The epicenters
marked KTSN were determined by Kunming Telemetered Seismic Network (KTSN)
and the epicenters marked PDE were from the USGS Preliminary Determination of
Epicenters listing. Ml and M2 refer to the two mainshocks, and Al and A2 refer to
the two largest aftershocks that we studied.
2
Teleseismic body wave analysis was applied to Ml, Al and A2. Since the
two mainshocks were separated by only 12.5 minutes, the body waves of M2 were
buried in the signal from Ml and could not be analyzed. Moving window filtering
was used to retrieve the Rayleigh waves of M2 and the amplitude spectra were
inverted to get the source parameters of that event. In addition to body waves
analysis, forward modeling of strong ground motion data was used to study the detail
source time histories of Al and A2 as well as the upper crustal structure in the source
region.
From the estimated source parameters, as well as the aftershock distribution,
geological information and observed surface fault ruptures, the rupture model for this
earthquake sequence was established.
c]
1. GEOLOGIC SETTING OF THE REGION
1.1 Fault Distribution and Seismicity in the Region
Towns of Lancang and Gengma are located in southwest Yunnan Province,
China, near the border with Burma. The area is close to the India-Eurasia collision
zone and is the most seismically active region in southwest China. Fig. 1.1 shows
fault distribution and six large histrorical earthquakes since 1900 in the area. The
faults mainly strike NW-SE and NE-SW. Four major faults are dominant in the area:
Lancang fault, Mujia fault, Nantinghe fault (E. Nantinghe fault and W. Nantinghe
fault) and Menglia fault (Jiang et al., 1989).
About 150 km long, a straight line Lancang fault strikes N30°W from its
intersection with Daluo fault in the south toward its junction with Mujia fault in the
north. 1.5-2 km right-lateral horizontal movement on the fault is evidenced by offset
Mesozoic folds (Jiang et al., 1989). Immediately south of its intersection with the
Mujia fault, the Lancang fault and several NE striking fault segments form a
complicated conjugate faulting system. The mainshock, Ml, was located near the
intersection of the Lancang fault and Mujia fault, while M2 was in a area north of
Mujia fault where surface trace of Lancang fault is not clear. The NE striking
Nantinghe fault formed in late Paleozoic era and is divided into two segments in the
area. Both west-segment and east-segment have dominant left-lateral strike-slip
movement with some thrust component (Jiang et al., 1989). The NE striking leftlateral strike-slip Menglian fault is about 140 km long and is conjugate to the Lancang
fault. The Mujia fault is curved and striking NW. It is divided into two parallel
segments north of the Lancang city. The movement on the Mujia fault had a change
according to Jiang et al. (1989): before Himalayan epoch, the movement on the fault
was left-lateral, and since the Himalayan epoch, the movement has been right-lateral.
The faulting system in the area is charactered by a conjugate faulting complex
in the area south of the Mujia fault, a faulting gap in the area immediate north of the
Mujia fault and a complicated NE striking faulting system in the north of the gap. On
the major faults, strike-slip movement is dominant.
Six large historical earthquakes (H1-H6) since 1900 with magnitude greater
than and equal to 6.0 are shown in Fig. 1.1. They were compiled by Lee et al. (1978)
and were instrumentally determined. Some of the magnitudes were recomputed by
Abe (1981). The epicentral parameters of these six events are listed in Table 2. Four
4
of them occurred in the area south of the Mujia fault and two occurred in the area
north of it. Only two events (Hi and H6), which have the smallest magnitudes in the
list, occurred in the area north of the Menlian fault and south of E. Nantinghe fault
suggesting that the area was devoid of large recent events before Mi and M2
occurred. The locations of H3 and H6 are near the Lancang fault. Even though the
epicenter of H4 determined by Lee et al. (1978) is near Daluo fault, Jiang et al.
(1989) reported that both H3 and H4 had ruptured 20-30 km on Lancang fault,
suggesting that H4 occurred on Lancang fault too. Most large historical earthquakes
were associated with the Lancang fault, indicating that the Lancang fault is the most
active fault in the area. Along this fault, the oldest event (H3) occurred on the middle
of the fault, then the south (H4) and finally the north (H6 and Mi). Also note that the
H5 occurred only about 3.5 hours after H4. The magnitudes and relative location of
this pair of events are quite similar to those of Ml and M2.
1.2 Observed Ground Deformation
The surface fault ruptures and a variety of ground deformations were reported
by Thou et al. (Fig. 1.2) and Zhao et al. (Fig. 1.3). The surface fault ruptures (heavy
lines) and ground deformation reported by Zhou et al. (1989) are distributed in two
main areas (Fig. 1.2). The northern part of the distribution is presumably associated
with M2 and southern part is associated with Ml. The southern part of ground
deformation zone is characterized mainly by linear ground fissure segments
distributed along the Lancang fault. Unlike the M2, the surface fault expression of
Ml is less obvious (Thao et al., 1989, unpublished data). It is characterized primarily
by the ground fissures located on the old fault, i.e. Lancang fault. The surface
rupture of Ml reported by Thou et al. (1989) is only in south of the intersection of
the Mujia and Lancang fault (see Fig. 1.2). However, the rupture probably extended
to the north of the intersection, because the longest ground fissure (> 1 1cm) striking
paralleled to the strike of the Lancang fault was found in the area north of the
intersection (Zhao et al., 1989). Probably, because the surface trace of the Lancang
fault is less obvious in the area north of the intersection, Thou et al. did not consider
these large ground fissures as faulting. The surface ruptures of Ml trend
approximately N30°W. Thou et al. (1989) reported right-lateral motion on the surface
ruptures. However, Zhao et al. (1989) reported left-lateral movement on a ground
fissure in the area north of the intersection of Lancang and Mujia fault. The fissures
and other deformation associated with Ml are in a narrow zone trending about
N30°W, with a long dimension of 60 km and short dimension of 10 km (Fig. 1.2,
1.3).
The surface fault rupture of M2, reported by Thou et al. (1989), is roughly
parallel to that of Ml with an perpendicular offset about 10 km and is about 14 km
long (Fig. 1.2). The right-lateral motion is quite clear. The maximum horizontal
displacement observed on the surface fault is 0.9 m and maximum vertical
displacement is 1.8 m (Zhao et al., 1989, unpublished data). The length of total
deformation zone associated with M2, reported by Thao et al., is about 25 km (Fig.
1.3).
1.3 The Aftershock Distribution
The 1029 aftershocks and major faults are plotted in Fig. 1.4. The
aftershocks were located by the Kunming Telemetered Seismic Network (KTSN) in
the two months following the mainshock. The magnitudes of these aftershocks range
from 2.0 to 6.9. The type of the magnitude scale used by KTSN is not known.
Usually, it is much higher than Ms magnitude for large magnitude events. For ease
comparison, the aftershocks which were also determined by PDE are listed in Table
3. However, unless otherwise noted, all the magnitudes mentioned in following
section are those of KTSN.
The aftershock zone is bounded by the W. Nantinghe fault in the north and
Menglian fault in the south and forms a lineament approximately 120 km long and 20
km wide, with the long dimension oriented approximately N30°W, which is
approximately parallel to the strike of the Lancang fault. Six aftershocks have
magnitudes between 6.0 and 6.9. Four of them occurred near the middle and the
other two occurred near the southern terminus of the aftershock zone. Most of small
aftershocks, magnitude ranging from 2.0 to 4.0, occurred at the two ends of the
aftershock lineament. The epicenters of Ml, Al, and A2, lie near the southern
terminus of the aftershock zone. The epicenter of M2 is about 60 km northwest of
Ml. The PDE locations of the two mainshocks and the two largest aftershocks are
shown in Fig. 1.2. The epicenter locations determined by PDE and KTSN are
similar, differing by about 10 km for Ml, Al and A2, while for M2, PDE's location
is about 25 km southwest of the KTSN location. The mainshock locations by KTSN
fit the surface fault ruptures (heavy lines in the figure) better than the PDE locations.
The PDE's locations are determined teleseismically and are probably less accurate.
Table 1. Epicentral parameters of the events investigated in
this.
PDE
Event
Date
No.
Oiigin Time
Lat.
h:m:s
(0)
KTSN
Lon.
Depth
(°)
(1cm)
mb
Ms
Lat.
Lon.
Depth
(1cm)
Ml
11/06/88
13:03:19.3
22.789
99.611
18
6.1
M2
11/06/88
13:15:43.3
23.181
99.439
10
6.4
Al
11/30/88
08:13:29.7
22.773
99.844
15
5.6
A2
11/27/88
04:17:56.2
22.749
99.852
16
5.0
7.3
22.833
99.717
13
23.383
99.600
8
6.0
22.717
99.833
11
5.1
22.683
99.800
12
Table 2. Epicentral parameters of the historical events in the region1
No. Year
Mo. Day
On. Time
h:m:s
Lat.
Lon.
Mag.2
Ms3
Hi
1938
5
14
12:03:03
23.0
99.8
6.0
H2
1941
5
16
07:14:32
23.6
99.4
6.9
H3
1941
12
26
14:48:09
22.2
100.1
7.0
7.0
H4
1950
2
2
23:33:39
21.7
100.1
7.0
7.0
H5
1950
2
3
02:51:52
22.1
99.9
6.75
H6
1952
6
19
12:12:59
22.7
99.8
6.5
1. The data is from Lee et al. (1978).
2. The magnitudes are from Lee et al. (1978). The type of the magnitude is
unknown.
3. Surface wave magnitude computed by Abe (1981).
Table 3. The epicentral parameters of the aftershocks determined both by PDE and
KTSN.
KTSN
PDE
Date
y. m. d.
On. Time
h:m:s
13:03:19.3
13:07:54*
1988
11
6
1988
11
6
1988
11
6
1988
11
6
1988
11
6
13:13:57*
1988
11
6
13:15:43.3
1988
11
6
1988
11
6
1988
11
6
1988 11 6
Lat.
Lon.
Mag.
Lat.
Lon.
(°)
(mb)
(°)
(0)
(c')
22.789 99.611 6.1
13:12:29*
23.181 99.439 6.4
13:21:04.6 22.82
99.56 5.8
13:36:33*
13:39:48.1
13:43:09*
22.980 99.710 5.0
1988
11
6
1988
11
6
16:00:32.0 23.424 99.500
1988
11
6
16:19:15.1
1988
11
6
1988
11
6
18:09:03.8
20:01:16.8
1988
11
6
1988
11
6
1988
11
6
1988
11 6
1988
11 7
1988
11 7
1988
11 7
20:24:24.6
20:35:28.2
21:19:04.3
22:40:16.4
02:39:56.2
20:50:28.1
22:00:24.6
1988
11
8
07:01:08.1
1988 11 6
1988 11 6
99.717 7.6#
5.0
34
13:12:21*
13:44:58.9 23.967 99.173 4.9
14:13:24.2 23.239 99.440 5.1
14:29:46.2 23.301 99.454 4.9
15:49:16.1 22.514 99.816 4.6
1988 11 6
22.833
Mag.
23.74
22.799
23.906
23.029
23.389
23.196
22.010
23.408
23.593
23.263
23.109
98.79
98.993
99.441
99.789
99.306
99.541
99.575
99.495
99.513
99.510
99.523
5.1
4.4
4.3
3.8
5.4
4.9
4.2
4.6
4.9
4.4
4.2
4.8
23.383
23.167
23.017
22.850
23.467
23.467
23.183
23.417
22.633
23.433
23.067
23.400
23.167
23.217
23.500
23.150
23.533
23.500
23.633
4.8
4.1
99.600 7.2#
99.367 6.0
99.533 5.2
99.633 4.7
99.550 4.7
99.583 5.0
99.533 6.0
99.483 4.9
100.000 4.4
99.467 5.2
99.683 4.1
99.533 4.2
99.567 4.2
99.767 6.1
99.467 5.0
99.550 4.9
99.433 4.3
99.517 5.9
99.517 4.3
4.4
4.2
10
Table 3. (Continued)
1988
11
8
1988 11 9
1988 11 9
1988 11 10
1988 11 10
1988 11 11
1988 11 11
1988 11 12
1988 11 15
1988 11 17
1988 11 18
1988
11 18
1988
11 18
1988
11 19
1988 11 20
1988 11 27
1988 11 28
1988 11 30
1988 12 08
1988 12 09
1988 12 19
1988 12 19
1988 12 22
1988 12 25
1988 12 26
*
09:53:07.9
01:14:06.8
12:32:18.7
09:43:34.6
22:49:38.8
07:41:16.0
10:18:12.7
09:20:13.8
10:28:14.3
13:14:47.8
14:29:19.5
18:21:44.8
22:18:44.2
01:37:14.8
01:35:33.6
04:17:56.2
03:38:11.2
08:13:29.7
15:32:47.1
22:42:10.8
11:06:57.5
14:01:47.2
03:49:46.8
03:41:48.2
02:48:46.5
Determined by KTSN.
Events studied in this thesis.
22.609
22.88
23.250
23.436
22.781
22.936
23.363
23.445
23.145
23.360
23.414
23.210
22.722
23.058
23.456
22.749
23.044
22.773
23.413
22.19
23.360
23.459
23.255
22.857
23.409
99.806
100.07
98.919
99.345
99.965
99.504
99.373
99.560
99.685
99.289
99.140
99.685
99.629
99.771
99.385
99.852
99.717
99.844
99.470
4.6
4.7
4.5
5.2
4.9
4.9
5.0
4.4
5.0
4.7
5.0
4.9
5.6
4.9
100.51
99.514
99.462
99.370
99.958
99.486
5.0
4.6
5.0
4.4
22.633
23.100
23.500
23.467
22.683
22.983
23.333
23.500
23.217
23.467
23.617
23.283
22.917
23.150
23.533
22.683
23.050
22.717
23.433
23.083
23.333
23.283
23.267
23.250
23.500
99.833
99.750
99.483
99.400
99.767
99.650
99.517
99.483
4.2
3.8
4.4
5.0
4.5
4.9
4.4
4.2
99.583
99.467
99.467
99.550
99.750
100.083
99.500
99.800
99.617
99.833
99.467
99.600
99.467
99.533
99.600
99.500
99.450
6.1
5.0
4.1
5.3
4.8
5.1
4.7
6.3#
4.8
6.7#
5.3
4.3
5.5
4.3
4.2
4.0
4.7
11
24,
23.
23.
23.
22.
22.
21.
21.50
98.70
WI.).,-
WI.I
Fig. 1.1. Fault distribution and seismicity in Lancang-Gengma area. The faults are
modified from Jiang and Thang (1989). Circles (H1-H6) are historical earthquakes
(M 6.0) since 1900. They were instrumentally determined by Lee et al. (1978). The
epicentral parameters for these events are listed in Table 2. Stars indicate the location
of Ml and M2 determined by KTSN. Inset indicates the location of the area.
12
:-
M2(KTSN)
+ 23.5°
//\
M2(PDE)
+ 23.00
M1(KTSN)
_StrOfl
Suace fault
[Eii
Fissure
i Collapse
[j
I
Motion Sta.
A l(PDE)
FEi] Landslide
0]
T]
F
Rock-mud stxeam
Sand blow
Subsidence
I
[jj
I*1
Conozoic basin
River
Epicenter
M1(PDE)
/
L\1
A1(KTSN)
A2(KTSN)
±
9950
+
100.00
Fig. 1.2. Ground surface deformations reported by Zhou et al. (1989). The surface
fault ruptures (heavy lines) caused by Ml and M2 are roughly parallel with a
perpendicular offset of -10 km. The stars show the epicentral locations determined
by PDE and KTSN for the events studied in this thesis. The epicenters of Ml and
M2, determined by KTSN, fit the surface fault ruptures better than those of PDE.
PDE's locations are probably less accurate.
13
Gengma
23.
'
'
:
4 %
.,
.\
:
_' : 7
"
:
I.
*L
'I
Fissure
Water emission
Sand blow
-
Collapse
Landslide
o
Subsidence
S
S.
4
+
Lancang
22.5° L
99.00
99.5°
100.00
Fig. 1.3. Ground surface deformations reported by Zhao et al. (1989, unpublished
data)
14
25
23.E
23.4
23.2
23.0
22.8
22.6
22.4
99.2
99.4
99.6
99.8
100.0
100.2
Fig. 1.4. Distribution of the aftershocks (2M6.9) determined by KTSN in two
months following the mainshock. Aftershock magnitudes used here were determined
by KTSN. Some of the aftershocks are listed in Table 3 for comparison with PDE
determinations. Ml and M2 are indicated by stars. Major faults (solid lines) and
surface fault ruptures (dash lines) are also shown.
15
2. TELESEISMIC BODY WAVE ANALYSIS
2.1 Introduction
The use of teleseismic body waves to determine earthquake focal mechanisms
is well established (Langston and Helmberger, 1975; Kanamori and Stewart, 1976;
Bouchon, 1976; Langston, 1981; Nabelek, 1984). According to seismic ray theory,
seismic body waves travel along the paths specified by Fermat's principle. In the
epicentral distance range from 300 to 900, the rays travel steeply through the crust and
upper mantle beneath the source and receiver regions. Most of the path is in the
relatively homogeneous lower mantle. The seismic phases are mainly affected by the
complexities of crustal stratification, which can be incorporated by taking into
account conversions and reflections at boundaries in both source and receiver
regions. The wave packets in this epicenter range represent P. S, and other phases
well separated in time. This allow us to model P and S phases separately. With the
current availability of high quality digital data and good station coverage, the
earthquake source parameters can be well inverted for by the quantitative comparison
of observed body wave seismograms with synthetic seismograms calculated for a
postulated source model.
The inversion method used in this study was developed by Nabelek (1984).
The source parameters of inversion are double-couple point source orientation (strike,
dip, slip), seismic moment, source time function and centroid depth. If necessary,
more source complexities can be introduced, such as line sources (unilateral and
bilateral propagation), and multiple events with different orientation and separated in
space and time. The inversion scheme is an interactive iterating process minimizing
the misfit between synthetic and observed seismograms in a least-squares sense.
In this chapter, we first present theory behind teleseismic body wave
modeling. Then, teleseismic long-period P, SH and SV are used to determine average
source parameters of Ml, Al and A2. The source time functions of Ml and Al were
further constrained by the inversions of broad-band body waves. Because the body
waves of M2 were masked by the signal of Ml, we will discuss that event later in
chapter 3.
16
2.2 Teleseismic Body Wave Inversion Method
In a Cartesian coordinate system, the displacement fields of body waves in a
linear isotopic homogeneous infinite elastic medium due to an arbitrary moment
tensor rate, Mjk(t). acting at the origin are represented as
u(r,t) = gk G1x * MJk(t)
where
(2.1)
is the direction cosine and Cx is wave velocity. u(r,t) is the ith component
of displacement. G is the elastodynamic Green's function, (x) denotes the type of
wave, either P or S and "" denotes the time convolution.
Assuming all com1onents of the moment tensor rates have the same time
dependence, S(t), namely Mjk(t)=MjkS(t), the equation (2.1) becomes
x
u1
(r,t) =
MG
* S(t)
(2.2)
where S(t) is called the far-field source time function. From definition, it represents
normalized seismic moment release rate.
For a dislocation earthquake, its equivalent is a double couple forces with
zero net force and moment. In this case, each component of moment tensor has a
simple form and can be represented in terms of strike, dip, slip of the dislocation,
representing the orientation of the dislocation, and scalar seismic moment, Mo,
representing its strength.
In practice, the displacement is computed in a cylindrical coordinate system
centered on the source. The displacement at a point (,cD) and time t due to an
arbitrary source orientation then can be split into P-SV and SH components and each
component shows its own radiation pattern around the azimuth CD.
There are many ways to compute the medium response, i.e., Green's
function. Nabelek (1984) split the computation of the Green's function of the earth
into three parts: the contributions from the crustal and free surface effects in the
source and receiver regions and the contribution from the mantle:
G(t) = GX(t) * M(t) * cR(t)
(2.3)
17
Because of the homogeneity of the mantle, M(t) involves only geometrical
spreading, inelastic attenuation, t', and travel time.
With the help of propagator matrixes, the conversions and reflections of a
plane wave at the free surface and layered crust in the receiver region can be easily
computed. For a general case, an average crustal model is assumed for all stations.
When a large number of stations is used in the analysis, the differences between
receiver areas shotld average out (Nabelek, 1984).
At the source region, for an earthquake at depth, d, the Green's function is
calculated only for four rays, namely, up- and down-going P and S waves. In the
case of a homogeneous half space earth crustal model, the computed body wave
packet consists of the phases P, pP, sP for P waves; S, pS, sS for SV waves and S,
sS for SH waves.
The synthetic seismograms are first computed for an initial guess of the
source model, then the source parameters are inverted using an iterative technique
which minimizes the misfit between synthetic and observed seismograms in a leastsquare sense.
2.3 Data
2.3.1 Long-period data
Teleseismic data used in this study are long-period and broad-band records
from Global Digital Seismographic Network (GDSN). Tables 4, 5, 6 give lists of
stations and wave types used in body wave analysis for each event.
Before modeling, all long-period records were corrected for possible manmade errors, such as mis-labeled channels and polarity. Two main criteria were used
for error checking: (1) The polarities of the first motions of N-S. E-W and vertical
components should be consistent. For example, if one station has down polarity of
vertical component first motion, and its back-azimuth is in the first quadrant, then, its
first motion polarities for the other two components should be north and east. (2)
After rotating the seismograms from the station coordinates to source coordinates,
there should be no P waves in the transverse component. In addition, the wave-form
similarity was also checked for those stations with similar azimuth and epicentral
distance. Based on these criteria, no errors were found in the data set for Ml.
However, several errors were found in the data set for Al and A2. Table 7 gives a
list of these errors. Error corrections for instrument response were also made based
on those published by the USGS.
Because both long-period and broad-band seismorneters are used by GDSN,
in order to normalize different types of seismometers, an instrument response
normalization was performed on all records. In this processing, original instrument
response of each record was removed and station GRFO instrument response, which
is more sensitive to longer period, was inserted.
Only those stations with epicentral distance of 300l010 for P and SH and 700
800 for SV waves were used. The crustal model of source and receiver regions is a
homogeneous half space with a compressional wave velocity of 6 km/s, Poisson's
ratio of 0.25, and density of 2800 kg/rn3. Thus, the wave packet in our inversion
includes direct P, pP, sP for P waves; direct 5, p5, sS for SY waves; direct S, sS for
SH waves. The mantle inelastic attenuation coefficients (t*) of 1 and 4 s were
assumed for P and S waves, respectively. All seismograms are normalized to a
magnification of 5000 and epicentral distance of 40°. In the inversion, SH and SV
were weighted so that the total rms amplitude of the S waves was the same as the
total rms amplitude of P waves. In this way, both P and S waves are assigned equal
19
importance (Nabelek, 1984). The length of inversion window was chosen such that
all energy contributed from P. pP, sP for P waves and S. pS, sS for S waves were
included in the window, and other phases like PP, SS and SKS were eliminated.
2.3.2 Broadband data
Broadband seismograms were used to refine the details of the rupture
parameters determined from long-period data. The broad-band records are either from
original broadband seismograms or simultaneous deconvolution of the long- and
short- period records (Harvey and Choy, 1982). Instrument response was removed
from all broad-band records. Then, data were filtered using a bandpass filter of 0.5 to
100 s and resampled to 5 samples per second. At this sampling rate, the Nyquist
frequency is 2.5 Hz and the highest frequency we analyzed is 2 Hz, thus, there is no
frequency aliasing problem in our analysis.
20
2.4 Modeling of Mainshock Ml
2.4.1 Long-period body wave inversion
P, SH waves from 10 stations and SV waves from 2 stations were used in
long-period body wave inversion. Because most short-period first arrivals for this
event are not impulsive, the theoretical first arrival times computed for PDE's location
were used for alignment of observed and theoretical seismograms. These first arrival
times are only as initial guess, the synthetic seismograms were later realigned with
the observed seismograms to maximize correlation. Usually, 3 and 5 s uncertainties
were assigned to the first arrival times of P and S for realigning the seismograms in
the inversion. Due to short epicentral distance of station HIA, the PP phase arrived
50 s after direct P arrival at this station. To eliminate the PP phase, which we did not
model, an inversion time window of 50 s, starting at theoretical first arrival, was
used for this station. For other stations, inversion windows of 60 and 70 s long for
P and S waves were used, respectively. After the best fit point source orientation
(strike, clip, slip), centroid depth, source time function, and seismic moment are
obtained, the uncertainties in the source orientations were estimated in this way: each
parameter of source orientation was deviated by up to ±10° from the best fit model,
while other parameters were kept unchanged. The normalized variance, which is
variance divided by power of total modeled signal, was computed and plotted as a
function of the deviation of the parameter. The uncertainty in the varied parameter
then was interpreted as the range with 90% confidence using a t-test (Huang et al.,
1986). To estimate the uncertainty of the depth, inversions were performed at a series
of trial depths (other source parameters were left free). In those inversions, the
inversion windows and theoretical arrival times from the best fit point source solution
were used. But the theoretical first arrival times were not fixed, a common 1 s
uncertainty is allowed for P and S waves in order to realign the synthetic and
observed seismograms. The normalized variances and seismic moment were plotted
as a function of depth. The acceptable depth is interpreted as the range in which the
normalized variance is within 5% the smallest value, which is a conservative
estimate.
21
Long-period P, SV and SH wave inversion for this event indicates that Ml is
a purely right-lateral strike-slip event which occurred on a nearly vertical fault plane.
The inversion gives the best fit point source orientation: strike 154°; dip 86°; slip 181°;
centroid depth 12 km and seismic moment 4.6 x 1026 dyn cm (Mw: 7.0). Fig. 2.1
shows the long-period observed and the best fit point source synthetic seismograms.
The source time function in this figure indicates that the source had a total duration of
12 s, and only a little energy (18%) was released in the first 6 s of rupture. Most
energy (82%) was released in the following 6 s. The synthetic seismograms for the
best fit point source model fit the observed data well. However, relatively large misfit
occurred for the eastern stations (P waves of stations LON and AR; SH waves of
stations HIA, LON, HON and AFT; see Fig. 2.1). Most eastern stations have large
epicentral distances (LON: A=101.18°; HON: =92.54°; AFT: z=94.06°). The
structural complexities near the core-mantle boundary may contribute to the large
misfits at those stations. Although station HIA has a short epicentral distance
(it=3O.85°), large misfit of SH wave at this station was apparently caused by longperiod noise (see Fig. 2.1).
In Fig. 2.2, the normalized variances are plotted as a function of the deviation
of the strike, dip and slip from the best fit point source model. Although the
normalized variance in this figure is only computed for the case of deviating one
parameter while other parameters were kept unchanged, the plot does give the relative
sense of how the available station coverage is sensitive to the source orientation. The
rapid convergences of the normalized variances of strike, dip and slip shown in the
figure indicate that all parameters of source orientation were well resolved. The good
resolution of the source orientation was due to presence of both nodal stations (SLR,
HON for P waves; COL for SH waves) and anti-nodal stations (COL, HIA for P
waves; SLR for SH waves). With 90% confidence, the t-test gives the uncertainties
in each parameter as follows: strike 150°-157.8°; dip 85.5°-87.5°; slip 179.9°-181.3°.
Thus, at same confidence level, the largest uncertainty is in strike. The uncertainties
in depth and seismic moment will be discussed in the broadband body wave
inversion section.
22
2.4.2 Broad-band body wave inversion
Broadband seismograms are usually used to get more detailed information on
source time histoiy because of their broad range of frequency contents. With a good
coverage of broadband stations, space distribution of the source rupture can be also
studied. When the Lancang-Gengma earthquakes occurred, only limited broad-band
stations were available. Within the epicentral distance 300 to 900, four broadband
stations (HIA, TOL, COL, KEY) recorded Ml. Three other broadband P-wave
records (GRFO, SLR, GDH) were formed by simultaneous deconvolution of the
long- and short-period records (Harvey and Choy, 1982). Among these stations, six
are located north of the epicenter. One station (SLR), which was near the P-wave
nodal plane, is located west-southwest of the epicenter. Assuming that the source
propagated along strike (154°), all broadband stations were on the same side of
propagation direction. This makes it difficult to invert propagating source parameters.
For this reason, broadband P and SH waves are used only to determine a point
source time function in detail. Station KEY appeared to have digitizing problems and
was not used. The problems were not serious in the NS component, and the backazimuth for this station was 98°, so the S wave of the NS component was used as the
SH wave for this station. As mentioned in previous section, SH wave of station HIA
was contaminated by long-period noise and was also excluded from the inversion.
Thus, in the broadband body wave inversion, six P waves of the epicentral distance
range 30°-86°, three SH waves of the epicentral distance range 62°-85° were used.
The source and receiver region crustal models were half spaces, same as those of the
long-period body wave inversions. The source orientation (strike, dip, slip) was
taken from the long-period body wave inversion and was fixed. A small uncertainty
(1 s) was assigned to the first arrival times of those stations whose P wave first
motions can be determined from short-period or broadband records, if first motions
could not be determined, the theoretical first arrival times were used, and a large
uncertainty (usually 5 s) was assigned to the first arrival times of those stations. The
inversion parameters were source time function and centroid depth. After the best fit
depth and source time function were obtained, the inversion time window and first
arrival times of best fit solution were then used in inversion for the source time
23
function at each fixed depth to assess the uncertainties on depth and seismic moment.
In these inversions, an 0.5 s uncertainty was assigned to P and SH first arrival times.
Fig. 2.3 shows the broadband observed and best fit synthetic seismograms.
A seismic moment of 4.2 x 1026 dyn cm and centroid depth of 9 km were obtained
from broadband body inversion. The source time function is also shown at the lower
portion of the figure. The total source duration of 12 seconds was obtained. The
source time function obtained from the broadband inversion further confirms that
little seismic moment was released during the first 6 s of fault rupture, and most of
the energy was released in the following 6 seconds. In Fig. 2.4, normalized variance
and seismic moment are plotted as a function of depth. The results from the longperiod body wave inversions are also plotted in this figure. That both variance curves
have a jump near the depth of 15 km indicates that centroid depth for this event was
shallower than 15 km. However, at depth shallower than 15 km. the centroid depth
resolution is reduced. This is common for shallow focus strike-slip events as
indicated by Nabelek (1984). It is difficult to get the precise centroid depth for
shallow strike-slip earthquakes unless a clear reversed pP phase can be identified.
The best depth resolution occurs when the direct and reflected phases have opposite
polarities because we require the source time function to be positive (no back slip on
the fault). For a strike-slip event, the only reversed phase is the pP phase and its
amplitude is small. At depths shallower than 15 km, the 12 s source duration of Ml
exceeds the travel time difference between P and pP. Thus, the misfit between
synthetic and observed seismograms due to centroid depth deviation from the true
centroid depth can be improved by modifying the source time function. However, the
best fit centroid depth from the long-period body wave inversion is 12 km and and
the broadband body wave inversion is 9 km. All aftershock hypocenters determined
by KTSN have depths shallower than 25 km. If we take 25 km as the lower
boundary of the ruptured fault for this event, the centroid depth should be 12.5 km in
good agreement with our results. Between 7.5 km and 15 km. the long-period body
wave inversions give a stable solution for the source orientation, and seismic moment
varies from 4.5 X 1026 dyn cm at depth 7.5 km to 4.9 X 1026 dyn cm at depth 15
km.
2.5 Modeling of the Aftershock Al
2.5.1 Long-period body wave inversion
P, SH waves from 10 stations and SV waves from 4 stations (Table 5) were
used to invert point source parameters for this event. Due to difficulty in reading first
motions from most short-period records, the theoretical first arrivals computed for the
PDE location were used in inversion, but were not fixed. 3 and 5 s uncertainties on
first arrival times were assigned to P and S waves in the inversion, respectively.
When the best fit point source solution was obtained, the normalized variances were
computed and plotted for the deviation of the source orientation (strike, dip, slip).
The normalized variance and seismic moment were also plotted as a function of depth
as in the analysis of Ml.
Results
Long-period P, SH and SY inversion gives the best fit point source
parameters: strike 165°; dip 900; slip 178°; centroid depth 11 km and seismic moment
1.5 x 1025 dyn cm (Mw 6.1) for this event. It is a nearly pure right-lateral strike-slip
event which occurred on a vertical fault plane. This event is thus very similar to Ml
except the strikes differ by 110. Fig. 2.5 shows the long-period observed and best fit
point source synthetic seismograms. The far-field source time function is also shown
in the figure. The total duration of the estimated source time function is about 2
seconds. However, because long-period data are not sensitive to short source
duration, this source lime function might not be well constrained.
Fig. 2.6 shows the normalized variance vs. deviations of strike, dip and slip
from the best fit model. The plot indicates that the strike, dip and slip are well
constrained from the long-period body wave inversion, because used stations have a
good azimuth distribution. With 90% confidence, the uncertainties in the source
orientation are: strike 162.5°-167.5°; dip 88.5°-90.5°; slip 177.5°-178.5°. Again, at
same confidence level, the largest uncertainty is in the strike.
25
2.5.2 Broadband body wave inversion
Due to the small size of the event, only broadband stations with short
epicentral distance have good signal to noise ratio records. Among them, two stations
(lilA, BJI) were within an epicentral distance of 2O03O0. Due to the short epicentral
distance, only the waveforms (not absolute amplitudes) of P waves were inverted in
broadband body wave analysis. The source orientation (strike, dip and slip) is that
derived from the long-period body wave analysis and fixed. Because waveform
inversion employs an arbitrary seismic moment to match the amplitudes of the
seismograms, the inversion parameters were source time function and centroid depth.
The inversion started with P wave first motions, which were carefully read from
short-period records, and lasted 30 s. An uncertainty of 0.5 s on P wave first motion
was assigned to both stations. A half space earth model, same as the previous one,
was used in both source and receiver regions. The source time functions were
inverted for different depths, and normalized variances were piotted as a function of
the depths. The best fit source time function and centroid depth were obtained when
minimum variance occurred.
The inversion results are shown in Fig. 2.7. The total source duration is
about 7 s and the best centroid depth is 7 km. In Fig. 2.8, normalized variance and
seismic moment were plotted as a function of depth. The broad-band body wave
inversion indicates that the centroid depth for this event is shallower than 12 km.
Again, the precise determination of the centroid depth is difficult. However, the least
misfit centroid depth from the broad-band P waveform inversion is 7 km and from
the long-period body wave inversion is 11 km. Between 6 and 12 km, the longperiod body wave inversions show the seismic moment from 1.5 X 1025 dyn cm to
1.6 X 1025 dyn cm and a stable source orientation.
26
2.6 Modeling of Aftershock A2
Due to the small magnitude of this aftershock, only a few long-period stations
are available for body-wave inversion. The inversion gives the source orientation:
strike 166°, dip 92°, slip 177°, centroid depth 5 km. seismic moment 1.3 x 1024
dyn.cm (Mw 5.3) and source duration of 1.6 s. Fig. 2.10 shows the observed and
best fit synthetic long-period P and SH waves for this event.
In Fig. 2.11, the normalized variances are plotted as a function of the source
orientation deviation from the best fit model. The plot indicates that the available
stations for this event can only constrain the strike precisely. The convergence of the
variance for the deviation of the dip and slip are very slow indicating that the
resolution for dip and slip is poor. With 90% confidence, the uncertainty of the strike
is 163°-170°. In Fig. 2.12, the normalized variance and seismic moment are plotted as
a function of the depth. Because the variance increases significantly for centroid
depth deeper than 12 km, the centroid depth for this event is shallower than 12 km.
Between depth 2.5 km and 12.5 km. the seismic moment varied from 1.3 X 1024
dyn cm to 1.6 X 1024 dyn cm.
27
2.7 Discussion
The long-period P waves from Ml recorded at some stations show complex
wave shapes. A comparison of long-period P waves from Ml, Al and A2 suggests
that such complexities were due to path effects rather than source complexities. Fig.
2.12 shows such a comparison. P waves of Ml, Al and A2 recorded at station HIA
are shown in this figure. Because A2 was not recorded at station GRFO, only P
waves from Ml and Al are shown. The amplitudes in Fig. 2.12 have been
normalized to unity and have been time shifted for easy comparison. Fig. 2.12
indicates that the wave shapes of Ml, Al and A2 at station HIA, which have short
epiceniral distances (HIA:
30°), are almost identical; while wave shapes from Ml
and Al recorded at station GRFO (i 72°) are very similar. Both records at this
station have an energy arrival 70 s after the direct P and 70 s before PP arnval. The
similarity in observed wave shapes suggests that this arrival was caused by
propagation path effects. Wang et al. (1989) split Ml into three subevents. The third
subevent, in their study, has a 70 s time delay with respect to origin time. It appears
that the propagating path complexities have been mistakenly incorporated into their
source modelling.
Table 4. Station parameters for body wave inversion of Ml.
Station
Azimuth
(°)
Instrument type
Phase
(°)
COL
23.6
78.53
GDSN
P SH
HIA
26.1
30.85
GDSN
P SH
LON
27.6
101.18
GDSN
P SH
HON
65.6
92.54
GDSN
P SH
AFI
103.2
94.06
GDSN
P SH
SLR
239.3
84.32
GDSN
P SH
TOL
311.5
85.42
GDSN
P SH
GRFO
317.0
71.89
GDSN
P SV SH
KEY
338.1
62.75
GDSN
P SY SH
GDH
350.7
86.07
GDSN
P SH
29
Table 5. Station parameters for body wave inversion of Al.
Station
Azimuth
Instrument type
Phase
P SV SH
(°
()
COL
23.7
78.46
GDSN
REA
25.8
30.77
GDSN
LON
27.7
101.10
GDSN
KIP
65.6
92.30
GDSN
HON
65.7
92.35
GDSN
P --
--
GIJMO
94.0
43.59
GDSN
P --
--
TAU
146.4
78.75
GDSN
- SV
SH
SLR
239.4
84.5
GDSN
P --
SH
TOL
311.6
85.60
GDSN
P --
SH
GRFO
317.1
72.04
GDSN
P SY SH
KONO
327.5
70.68
GDSN
P SV SH
KEY
338.1
62.84
GDSN
P --
SH
--
SH
P --
SH
-
P --
SH
30
Table 6. Station parameters for body wave inversion of A2.
Station
Azimuth
(0)
Instrument type
Phase
P -- SH
HIA
25.8
30.77
GDSN
GUMO
94.0
43.59
GDSN
SLR
239.4
84.5
GDSN
GRFO
317.1
72.04
GDSN
KONO
327.5
70.68
GDSN
P
-
-- --
-- SH
-- SH
-
-- SH
31
Table 7. Error lists for records of Al and A2.
Al
Station
Type of error
HIA
Timing problem: three components of long-period records have
time shift about 340 seconds respect to short-period records.
GUMO
Original labeled LPE is LPN and original LPN is LPE.
GRFO
Original labeled LPE is LPZ, LPN is LPE and LPZ is LPN.
KIP
Polarity of LPE component was reversed.
A2
Station
Type of error
GUMO
Original labeled LPE is LPN and original LPN is LPE.
GRFO
Original labeled LPE is LPZ, LPN is LPE and LPZ is LPN.
32
Ml (11/06188)
Long-period P waves
key
1\Jf
lol
pon
sv
SH
Wa
ton
Souco Tune Function
o
to is
P=6.O cm
SH=32.5 cm
-'l
limo(s)
limo (s)
Fig. 2.1 Observed (solid) and synthetic (dashed) long-period P-, SV- and SH-wave
seismograms for Ml. The far-field source time function is shown in the lower left
corner.
P
Variance vs. source orientation (Ml)
0.4
" '
/
U
I!
Ii
'I
0.3
I!
'I
I!
I-.
Ii
Strike
N
Slip
E
I.
0.2
0.1
-20
-10
0
Deviation from the best fit
10
20
(°)
Fig. 2.2 Normalized variance as a function of the deviations of the strike, dip and slip from the best fit point source model
of Ml. All curves are rapidly convergent to the best fit point source model, indicating that the source orientation of Ml is
well resolved by long-period body wave inversion.
(J
34
Ml (11/06/88)
Broad-band P waves
fcoI
f'1hia
grfo
(01
sir -
SH
key
(ci
Source Time Function
P=lO7cm
024 681024
time (s)
SH=45.Ocm
2mp
0
10 20 30
time (s)
Fig. 2.3 Observed (solid) and synthetic (dashed) broadband P-and SH-wave
seismograms for Ml. The far-field source time function indicates that the source
duration of Ml is 12s.
35
Variance vs. depth (Ml)
1.0
U
0.8
/
0.6
--
.s
S..
.5
Long-period
Broad-band
0.4
E
1
/
/
/
I-
N
/
.5.
0.2
0.0
0
1
/
20
40
Depth
60
80
(km)
Seismic moment vs. depth
1 .40e+27
1.20e+27
1.00e+27
Long-period
8.00e+26
0
E
Broad-band
6.00e+26
4.00e+26
2.00e+26
0
20
40
60
80
Depth (km)
Fig. 2.4 Normalized variance (top) and seismic moment (bottom) as a function of
ceniroid depth for Ml.
36
Al (11/30/88)
Long-penod P waves
__CQt
key
kono
grlo .;_-\/\.
tot
-{z°9- 1v"
/
tc.p
L' '
sv
gilo
SH
cot
koLj\f
tot _/\
hia.
Ion
N
tau
Source Tune Function
P=O.3 cm
tune (s)
SV=l.Ocm
SH=1.4cm
____
0 '000
time (s)
Fig. 2.5 Observed (solid) and synthetic (dashed) long-period P-, SV- and SH-wave
seismograms for Al.
Variance vs. source orientation (Al)
0.4
V
I..
Strike
Dip
C)
N
E
0
-
0.2
0.1+-20
-10
Deviation
0
from
10
Slip
20
the best fit (°)
Fig. 2.6 Normalized variance as a function of the deviations of the strike, dip and slip from the best fit point source model
of Al. All curves are rapidly convergent to the best fit point source model, indicating that the source orientation of Al was
well resolved by long-period body wave inversion.
Al (11/30/88)
Broad-band P waves
hia
rAl
bji
Source Time Function
amp
pO.147cm
0 2 4 6 8 101214
time (s)
J
0
10
20
30
time (s)
Fig. 2.7 Observed (solid) and synthetic (dashed) broadband P wave seismograms
for Al. The source time function, shown at the lower left corner, indicates that the
source duration of Al is 7 s.
39
Variance vs. depth (Al)
1.0
0
a
ii
I..
0
>
Long-period
Broad-band
0.4
0
E
z
0.2
0.0-f
0
10
20
Depth (km)
30
Seismic moment vs. depth
1 .90e+25
E
U
=
1 .80e+25
-a
=
1.70e+25
Long-period
1.60e+25
U
ci,
1.50e+25
1.40e+25
0
10
Depth
(km)
20
30
Fig. 2.8 Normalized variance (top) and seismic moment (bottom) as a function of
centroid depth for Al. The broadband waveform inversion constrained the centroid
depth of Al to be shallower than 12 km.
3']
A2
(11/27/88)
Long-penad P waves
_______
-0.034 as
0
10 20 30
time (s)
SH waves
hia
Source Time Function
0.206 all
5
________
0
time (s)
10 20 30
time (s)
Fig. 2.9 Observed (solid) and synthetic (dashed) long-period P- and SH-wave
seismograms for A2.
Variance vs. source orientation (A2)
U
0.4
I.
Strike
0.3
Dip
Slip
z0
0.2
0.1 +-20
-10
0
Deviation from the best fit
10
20
(°)
Fig. 2.10 Normalized variance as a function of the deviations of the strike, dip and slip from the best fit point source model
of A2. Convergence rates indicate that only strike was well constrained by long-period body wave inversion for this event.
Variance vs. Depth (A2)
V
U
0.
Variance
1
0
z
Depth (km)
Seismic moment vs. depth
1 .70e+24
E
1.60e+24
1.50e+24
Mo
1.40e+24
1.30e+24
1.20e+24
1.1 Oe+24
0
10
30
20
Depth
40
(km)
Fig. 2.11 Normalized variance (top) and seismic moment (bottom) as a function of
centroid depth for A2. The centroid depth to be shallower than 12 km was well
constrained by long-period body wave inversion.
Ml
:
-1.I2
Al
I..
I
./'\
\,_/'
/_'
If_'
\.44
'.
-
.,I
'' 211
A2
.1.I
Ml
0
i.825
Al
It
II
388
I.885
Fig. 2.12 A comparison of P wave seismograms of Ml, Al and A2 recorded at stations HIA and GRFO.
44
3. SURFACE WAVE ANALYSIS
3.1 Introduction
Due to interference with the signal of Ml, teleseismic body wave analysis can
not be applied to M2. However, because of the large amplitude and dispersion
properties of surface waves, it is possible to retrieve the Rayleigh waves of Ml and
M2 separately from the interfering seismograms. Surface wave inversion technique
can then be used to study the source mechanisms of the events. In this way, we can
determine the mechanism of M2, for which no source parameters have previously
been determined.
Dispersion, i.e., group velocity variation as a function of period, is a well
known property of surface waves. From it, much information about the earth's
structure over the wave propagation path can be obtained. Over the past few decades,
an effort has been made to improve the measurement of surface wave dispersion
patterns in terms of group velocity. Moving window analysis, developed by
Landisman et al. (1969), improves the group velocity measurement when recorded
signal is noisy or not well dispersed. In this analysis, a variable length window is
incorporated in order to give the same importance to each frequency. Precise
measurement of group velocity then makes it possible to design ifiters based on the
dispersion pattern to isolate selected mode of surface waves from a multi-mode
record, or from a record contaminated by arrivals from other events (Dziewonski et
al., 1969, 1972). Isolated surface wave from complex recording improves the signal
quality of the desired wave and consequently improves the resolution of earthquake
source mechanism determination from the surface wave inversion.
Several researchers have used amplitude spectra of Rayleigh waves to study
earthquake source mechanisms (e.g., Tsai and Aki, 1969, 1970; Romanowicz and
Suarez, 1983). The results have shown that source parameter determination from
amplitude spectra of Rayleigh waves can be effective. If source earth model, path
effect, in terms of attenuation coefficient, as well as source fmiteness can be properly
marie, some critical parameters of the source such as centroid depth can also be
determined precisely (e.g., Tsai and Aki, 1970; Zhang and Kanamori, 1988).
In this chapter, synthetic data with known dispersion pattern were used to test
the group velocity determination of the moving window analysis. Complex synthetic
seismograms were filtered with a time-varible filter, which was based on dispersion
45
pattern inferred from the moving window analysis (Landisman et al., 1969), to
isolate desired waves. After showing that the filter works correctly in the frequency
band of our interest, the group velocities of Ri and R2 waves (Ri represents
Rayleigh wave which propagates in the direction of the minor arc and R2 in the
direction of major arc) from Ml were determined using the moving window analysis,
and the Ri and R2 waves of Ml and M2 were extracted using time-varible filtering.
The isolated amplitude spectra of Ri and R2 waves were inverted to get source
parameters of Ml and M2. The results of Ml were compared with those of body
wave analysis which were well constrained. The comparison allows us to evaluate
the effectiveness of source parameter determination by Rayleigh wave inversion. This
help us in estimating which source parameter of M2 can be determined precisely.
Because, for this event, only Rayleigh waves are avaiable.
3.2
Moving Window Analysis
3.2.1
Principle
Moving window analysis, which displays seismic energy as a function of
period and group velocity, is a fundamental tool in surface wave dispersion
measurements. The analysis is applied to a portion of seismogram extracted from the
observed time series. The length of the portion (window) is proportional to the
product of a fixed window factor W (usually between 4 and 6, in our case 4 was
used) and period, Tm, which is currently analyzed. Therefore for each period, Tm,
there are always four complete cycles in the window. This gives the same relative
frequency resolution over all periods analyzed.
For a selected period, Tm, and a group velocity, u, the analysis window is
centered on group arrival time t0=
with a length of 4 X Tm, where i is the
epicentral distance for the path and u is the maximum group velocity to be analyzed.
To reduce the side lobes introduced by windowing and give more weight to the
portion of seismogram near the center of the window, the windowed data are
modulated by a symmetrical window function q(t)=cos2(
). In order to remove
long-period waves and constant offset originating in the seismogram and introduced
by truncation, the modulation of window function is split into two stages. In each
stage, the selected portion of seismogram is modulated by a square root window
function, q1(t)os(,
). Before and after first modulation, the mean and trend of
the seismogram are removed.
Modulated data are then Fourier transformed to get the amplitude of the period
Tm. This amplitude is plotted corresponding to the period Tm and group velocity u
and represents the energy arrival of period Tm with a group velocity u. The time
window is then moved to center a new group arrival with a group velocity of (u-O.2)
km/s, and amplitude of this portion of seismogram at period Tm is obtained and
plotted representing the energy arrival of period Tm with group velocity (u-O.2).
When the minimum group velocity has been analyzed, the analysis loops back to the
maximum group velocity u with a new time window length. Hence, amplitudes at a
new period are computed and plotted for all group velocities. This procedure repeats
47
until all periods have been analyzed. All seismic energy is then plotted as a function
of period and group velocity.
3.2.2 Application to synthetic data
Fig. 3.1 shows an example of group velocity measurement using the moving
window analysis. The synthetic data used in (a) is a linearly dispersed wave with the
group velocity of 4.3 km/s for the period 200 s and 2.6 km/s for the period 10 s. The
synthetic seismogram was computed for a sampling interval of 3 s and an epicentral
distance of 90°. In (b) a complex dispersed wave is used, which consisted of two
linearly dispersed waves as used in (a) but one was delayed by 13 minutes. The time
delay is similar to that of Ml and M2 and thus similar to the interfering Rayleigh
wave of Ml and M2, except that the Rayleigh waves are not linearly dispersed. The
outputs of the moving window analysis are shown below each wave. In these plots,
the vertical axis is period and the horizontal axis is velocity. The size of the amplitude
is represented by an alpha-numeric symbol 0-9-a-z, from small to large. The group
velocity is interpreted as the velocity of the envelope peak for each period (lines in the
plots). The values at the right side of the plots give the maximum amplitude for each
period. In plot (b), the energy contributions from each linearly dispersed wave are
well resolved for all considered periods, indicating that the two linearly dispersed
waves can be well separated by their dispersion pattern.
Fig. 3.2 is the plot of theoretical group velocity (straight line), which is 4.3
km/s at 200 s and 2.6 km/s at 10 s, and measured group velocity (dots) of figure
3.la. The maximum error of measurement, 0.054 km/s (1.9% of measured group
velocity), occurred at a period of
s. Between period 70 and 100 s, the error of
measurement is smaller than 0.043 km/s. which is 1.3% of the measured group
velocity. From this plot, it is apparent that the group velocity measurement of the
moving window analysis couls be very accurate, measurement error is smaller than
1.9% of measured group velocity using these synthetic data.
Time-variable Filter
3.3
3.3.1
Principle
The design of a time-variable filter is based on the dispersion pattern inferred
from the moving window analysis. It can be used to filter complex seismograms so
that the output of the ifitering can closely approximate the group arrival time pattern
for the selected wave (Landisman et al., 1969).
The time-variable filter used in this study was developed by Landisman et al.
(1969). Filtering is accomplished in the frequency domain. The original time series
(seismograms) are Fourier transformed by FFT. Within the pass-band Tmjn and
Tm, the amplitude and phase of the period Tm forms the monochromatic wave with
period Tm in time domain. This monochromatic wave is multiplied by a window
function, w(t), which depends on the group velocity ,u, of the period Tm. The output
of the multiplication then can closely approximate the group arrival of period Tm With
the group velocity of u. However, the total number of the group velocities obtained
from the moving window analysis is 26 with the period ranged 10-200 s, while in
time-variable filtering, the frequency sampling rate is dependent on the time sampling
rate of the seismograms. Therefore, for those intermediate periods, the group velocity
is obtained by interpolating the 26 group velocities over the period 10-200 s.
In practice, the following window function is used:
=0
w(t) = cos(
=0
lr(t-Nu(Tm))
tat1,
t<ta
)
ta
t
tb
t>tb
with
ta
A
du(T)
A
duct')
Tm( a +131 cIT 1T=Tm)
U(Tm)
/9, \ + Tm( CX +131 ,IT IT=Tm)
where A is epicentral distance, U(Tm) is the group velocity, T and Tm are the periods,
a and 13 are constants.
From equation above, it is easy to identify that tO_ut)
is the group arrival
time, and the window function, w(t)=1 at t=to; w(t)=O at both t=ta and t=tb. Thus, the
window function, basically, is an envelope function of the monochromatic wave.
This envelope is centered at the arrival time of grounp velocity u(Tm), which is
tO_U(Tm)
and the length of the envelope is 2T( a +
'IT=Tm). So the velocity
of the envelope peak is u(Tm), which is group velocity of the period Tm.
The final output of the time-variable filter is the linear superposition of all
windowed monochromatic waves over all analyzed periods.
However, the constants a and are dependent on the type of wave analyzed
(Landisman et al., 1969), In our case, a=3.5 and
(i is in 1cm) were used.
3.3.2 Application to synthetic data
The synthetic data used in this time-variable filtering test are those of Fig. 3.1
(a) and (b). One is a simple linearly dispersed wave and another is a complex
dispersed wave which consisted of two linearly dispersed waves.
Fig. 3.3 shows the result of the time-variable filter and a butterworth filter
with 3 poles. The amplitude spectra are shown at the bottom of the figure. Unfiltered
data, in this figure, is a simple linearly dispersed wave (trace (a) in the Fig. 3.3).
Trace (b) is the filtered wave trace based on the group velocities of Fig. 3.1 (a) with
band-pass of 118-17 s. Trace (c) is that of a Butterworth filter with the same
passband. In this example, the amplitude spectra of the time-variable filter are same
as those of the Butterworth filter, indicating that the desired band has been well
passed by time-variable filter. In top of the Fig. 3.3, all three seismograms are
exactly in phase, indicating that no phase shift occurred in time-variable filtering. In
addition, the amplitude spectra of the time-variable filter decays beyond bandpass
limits faster than the conventional Butterworth filter, indicating that the time-variable
filter has better filtering resolution.
In Fig. 3.4, the complex dispersed wave, as used in Fig. 3.1(b), is filtered
by the time-variable filter to retrieve the two linearly dispersed waves which were
used to form the complex wave. The complex wave (a) was first filtered using the
group velocities of Fig. 3.1(a) to retrieve the first linearly dispersed wave. The
output is trace (b). Then trace (a) was delayed by 13 minutes and the same group
50
velocities were used to isolate the second linearly dispersed wave. Trace (c) in Fig.
3.4 is the isolated second linearly dispersed wave. In both cases, a filtering
bandwidth of 118-17 s was used. Again, the unifitered and filtered wave traces are all
in phase, indicating no phase shift occurred during filtering. The amplitude spectra of
these three wave traces are shown at the bottom of the figure. Compared with the
original amplitude spectrum of the linearly dispersed wave in Fig. 3.3 (a), the
amplitude spectra of the two isolated waves (b) and (c) in Fig. 3.4 are the same as
that in Fig. 3.3 (a) within the passband (118-17 s). This indicates that two linearly
dispersed waves have been well retrieved by the time-variable filter within the desired
passband.
The amplitude spectrum of the unfiltered complex wave (a) in Fig. 3.4 shows
that it is very complicated, and thus conventional frequency domain filtering can not
isolate the linearly dispersed waves from the complex wave. On the other hand, by
carefully selecting the filtering window, a conventional frequency domain filter, e.g.,
a Butterworth filter, can resolve the linearly dispersed waves from the complex wave
trace for a certain range of frequencies. For example, in the complex wave trace, the
second linearly dispersed wave arrived 13 minutes later than first one. The second
arrival was contaminated only by the waves of the first arrival whose periods are
shorter than about 80 s. Therefore, the period between 200-80 s of the two linearly
dispersed waves can be resolved by conventional frequency domain filtering for these
synthetic data. However, for a realistic, non-linearly dispersed seismic wave, it is
hard to select such a specific filtering time window. More importantly, when
epicentral distance increases, the seismic wave is more dispersed, thus the resolved
pass-band becomes narrow in conventional frequency domain filtering. For instance,
in the above example, if the epicentral distance is 115°, the pass-band which can be
resolved becomes 200-100 s in conventional frequency domain filtering.
In summary, the dispersed property of the surface wave provides a basis for
designing filter and this kind of filter has shown high resolution in filtering
contaminated seismogram.
51
3.4 Rayleigh Wave Inversion Technique
The inversion method used here was developed by Suarez (1983) and
modified by Nabelek (personal communication), which uses the absolute amplitude
spectra of vertical component Rayleigh waves for the period range 10-108 s to
determine point source orientations, seismic moments and centroid depths of
earthquakes.
For a point source in a flat-layered medium, the observed spectrum of vertical
component Rayleigh waves can be expressed as
u(co,O,r) = S.I(o)exp(
-iwr
c(o),e)
)exp(-(o,9)r)A(c,O)exp(14(a),O))
where A(0,O)exp(i4)((o,O)) is the source spectrum. I(co) is the instrument response.
c(o),9) is the average phase velocity. i(co,O) is the attenuation coefficient along the
propagation path and S is the geometrical spreading factor (Romanowicz et al.,
1983).
S='4I
\4
rp
Re sinA
where A is the epicentral distance in radians, r is normalized epicentral distance,
usually 4000 km is used, and Re is the earth radius.
After making the correction for propagation effects and instrument response
of the spectrum, the source spectrum can be expressed as a linear combination of the
moment tensor components (Mendiguren, 1977). Furthermore, for a double couple
source, moment tensor elements can be expressed in terms of strike, dip, slip and
seismic moment, and thus, the source spectrum can be expressed as the following
form:
A(co,9)e14(°'°) = a + 3i
(3.8)
absolute amplitude spectrum
A(cok,9j)=\Ia2+32
k=1, ..... , n; j=1, ..... , m
(3.9)
52
where j is the index for the number of stations used, k is the index for the number of
frequencies modeled, and
a = Ak + Bkcos2Oj + Cksin2OJ
=
+ EksineJ
(3.10)
with
Ak = Mo(sin26sinA.)(G2(0k,h)-G1 (cok,h))
Bk = Mo(2sjn&ossjjij4+sjn23sjcos24)G 1 (0)k,h)
Ck =
(3.11)
Dk = Mo(-cos&osos-cos2ösjnXsjn)G3(CUk,h)
Ek Mo(cos&osXsjn4+cos2sjnXcos4)G3(0)k,h)
where Gi, G2 and G3 are the excitation functions for a known earth model, 9 is
station azimuth, 4) is strike, & is dip, . is slip, M0 is scalar seismic moment and
A(o,9) is the observed Rayleigh wave amplitude spectrum.
Through equations (3.9), (3.10) and (3.11), we have total m x n nonlinear
equations, where m is the number of stations used and n is the number of frequencies
modeled. These nonlinear m x n equations are solved to invert source orientation and
scalar seismic moment in the least-square sense for each trial depth. The residuals of
the inversions are plotted as a function of depth. The depth corresponding to the
minimum residual is the best solution.
53
3.5 Data and Data Analysis
Long period vertical component Rayleigh waves along the path Ri and R2
were used in this analysis. In this analysis, we only used vertical component
Rayleigh waves because there are no Love waves in this component.
The raw data were carefully checked. The Ri waves at the stations LON,
HON. GDH, SCP, GAC, BJI and SLR were clipped. In order to identify clipping
and other non-linear instrument behavior, which particularly show up at long
periods, for all Ri and R2 waves, instrument responses were removed and traces
were filtered by a Butterworth filter with pass band of 100-200 s. At this period
range, Ri and R2 waves should show reversed dispersion, namely, shorter periods
should arrive before longer periods. The clipped waves generated anomalous longperiod transients not consistent with the expected dispersion. In this way, we
identified that the Ri waves of Ml at stations BJI, GDH, LON, HON. SLR and the
Ri waves of M2 at stations HON. SLR, GAC, SCP were contaminated by clipping.
With this procedure, we also identified non-linear instrument behavior for Ri wave
of M2 at station GRFO and large background noise for Ri wave at station ANMO
and R2 waves at station LON, GDH and HON. All these waves were not used in the
analysis.
The stations used in Rayleigh wave inversions are listed in Table 8. Fig. 3.5
shows the station distribution and great circle paths of Ri from Ml. Most stations
were located north of the epicenter, so the use of R2 waves improves the station
coverage.
The group velocities of Ri and R2 waves were determined using the moving
window analysis for each station. Since the two events are very close (separated by
about 60 km), at teleseismic distances, the path differences due to the such a small
difference in epicentral location can be neglected. Therefore, the group velocities
determined from Mi are also used to retrieve Ri and R2 waves of M2 in timevariable filtering, but the epicentral distance for each station was computed for M2
location.
54
3.5.1
Group velocity determination of Ri and R2 waves
The original Rayleigh waves were extracted from observed seismograms
using a group velocity window of 4.3 km/s to 2.3 km/s. A group velocity range from
2.4 km/s to 4.2 km/s, and a period range from 10 s to 200 s was analyzed by the
moving window analysis. The mean of all records were removed before analysis.
The extracted seismograms were resampled to a sampling interval of 3 s. The energy
arrivals are plotted as a function of velocity and logarithmic period. Fig. 3.6 gives
these plots for stations TOL (a) and HRV (b). In these plots, the x-axis is group
velocity from 2.4 km/s to 4.2 km/s with linear scale, the y-axis is 26 periods from 10
s to 200 s with logarithmic scale and, for the better visibility, each line for a given
period has been plotted twice. The plots show that the energy contributions from
individual events (Ml and M2) can be well distinguished for periods 40-200 s,
indicating that in this period range an effective time-variable filter can be designed
based on this dispersion pattern to separate Ri waves from the contaminated records.
The group velocities were picked for the periods 16.1 to 139.6 s and were
used later in time-variable filtering. Fig. 3.7 shows the group velocities for the Ri
waves of Ml. Even though the group velocities in Fig. 3.7 are plotted up to the
period 18.2 s, for most stations with longer epicentral distances, only group
velocities for the period 42.1 to 139.6 s are reliable. However, the less reliable group
velocity for periods shorter than 40 s does not affect source parameter estimation,
because the waves in this period range are not used in the inversion, even though
they
are also isolated
by time-variable filtering.
A velocity window of 3.0 to 4.0 km/s was used in the moving window
analysis for R2 waves, and all extracted seismograms were resampled to a sampling
interval of 3 s. For the stations which were close to the epicenter (KMI, LZH,
CHTO), R2 and R3 waves are too close to be analyzed separately, while for station
TOL, the original seismogram was too short to contain the R2 wave, so this station's
group velocity of R2 wave was also not analyzed.
55
In Fig. 3.8, the dispersion patterns of R2 waves for the station AFI (a) and
ANMO (b) are shown. The energy contributions from the Ml and M2 can be
identified at the period range of 60-200 s, indicating that, within this range of period,
R2 waves of Ml and M2 can be separated effectively by time-variable filtering.
In Fig. 3.9, the group velocities of R2 waves for all the stations are shown.
Again, the group velocities are plotted up to the period of 18.2 s, but only the
velocities with periods greater than about 60 s are reliable.
3.5.2 Retrieving Ri and R2 waves by time-variable filtering
In order to separate Ri and R2 waves contributed from Ml and M2, timevariable filtering (period range 18-118 s) is performed. This filtering is based on the
dispersion pattern for each path obtained from the moving window analysis.
The Ri and R2 waves of Ml were first isolated. The M2 were isolated using
the group velocity determined from Ml. For station BJI, the Ri wave of Ml was
clipped, the group velocity determined from M2 was used in isolating Ri wave of
M2. Because station WMQ is nodal and it is difficult to pick the group velocity of R2
wave, the group velocity of R2 wave for station HIA was used for this station.
Fig. 3.10 shows two examples, stations TOL (top) and HRV (bottom), of the
time-variable filtering for Ri waves. Fig. 3.11 are for R2 waves of stations AR (top)
and ANMO (bottom). In these figures, trace (a) is the original vertical-component Ri
(R2) wave; trace (b) is the result of bandpass (50-250 s) filtering of a Butterworth
filter from trace (a); trace (c) and (d) are retrieved Ri (R2) waves for Ml and M2,
respectively. In Fig. 3.10 and 3.11, long-period energy contributions from Ml and
M2 are also well separated by conventional frequency domain filtering for stations
TOL and ANMO. While for stations HRV and API, the separation by Butterworth
filter becomes obscure because, at these two stations, the Rayleigh waves are more
dispersed.
Fig. 3.12 shows the amplitude spectra for the period range of 60-100 s of all
selected Ri (a) and R2 (b) waves for Ml (solid) and M2 (dashed). The spectra of Mi
and M2 at each station are very similar, indicating M2 had a similar source
mechanism to Ml.
56
3.6
Rayleigh Wave Inversion of Mainshock Ml
3.6.1
Data
The Ri waves from 10 stations and R2 waves from 14 stations for the period
range of 66-100 s were used in the inversion. Within this range of periods, both Ri
and R2 waves of Ml and M2 were well resolved by time-variable filtering. Because
Ml and M2 were continental earthquakes, the Gutenberg continental earth model was
used to compute Rayleigh wave excitation functions. The propagation effects due to
lateral inhomogeneities were not considered here because these effects can not be
modeled by our method. However, the longest period of Rayleigh waves reported
which can be laterally refracted and reflected by continental margins and mid-ocean
ridges is about 50 s (Capon, 1970, 1971), modeling 66-100 s Rayleigh waves may
reduce some of the error due to lateral refractions and reflections. On the other hand,
the mutipathing problems may affect this range of periods, but the effect is reduced
by the time-variable filtering. The propagation effect is mainly attenuation due to the
inelasticity of the earth.
The global average attenuation model for Rayleigh wave period range 60 to
100 s is not well constrained, so, in this modeling, several attenuation models were
tested to see which model gives the source parameters of Ml which are in the best
agreement with those of body wave inversion. This attenuation model is then used in
source parameter inversions of M2. Because the source duration of Ml was well
constrained by broadband body wave inversion, the source duration effect in
Rayleigh wave inversion was also tested.
3.6.2
Attenuation coefficient
Three attenuation models were used in the modeling (Fig. 3.13). The model
MH78 is that of Mills and Hales (1978b). PREM model was computed using PREM
Q and group velocity models (Dziewonski and Anderson, 1981) and the model ABM
was computed using ABM Q model (Anderson and Given, 1982) and PREM group
velocity model. For all considered periods, model MH78 has smallest attenuation
coefficients, while model ABM has the largest.
Each attenuation model was used to correct for propagation effect of Rayleigh
wave amplitude spectra and spectra were inverted at different trial depths, ranging
57
from 5 km to 70 km, for source orientation, centroid depth and seismic moment of
Ml. The normalized variance for these three attenuation models are plotted as a
function of depth in Fig. 3.14. The normalized variance vs. depth shown in Fig.
3.14 indicates that the Rayleigh wave amplitude spectrum inversion for this event has
a poor resolution in centroid depth. All attenuation models give almost same variance
at shallow depths. Among these models, model ABM gives the best centroid depth
resolution, which constrains centroid depth to be shallower than 40 km. At 40 1cm,
the normalized variance increased by 7% from the minimum. The inversion results at
depth 10, 20 and 30 km for these three attenuation models are listed in Table 9.
Except strike, the inverted source parameters varied in a large range for different
depths and attenuation models. At depth 10 1cm, source orientations inverted for three
attenuation models are very similar. Comparing these parameters with those of body
wave inversion (strike: 154°, dip: 86°, slip: 181°, Mo: 4.6 X 1026 dyn cm and
centroid depth: 12 km), strikes and slips are in good agreement but not dips,
indicating that the Rayleigh wave inversion can not constrain all the parameters of
source orientation for this event. In next inversion, we fixed the dip to that derived
from body wave inversion and inverted other source parameters for depths 10, 20
and 30 km. The inverted results are listed in Table 10 (upper part). The results shown
in this table indicate that, for three depths and attenuation models, the inversions gave
the almost same strike of 153°, in good agreement with 154° from body waves.
However, slips for all the attenuation models at all the depths show a large difference
with 181° of body wave inversion. Thus, for source orientation of Ml, because
neither slip nor dip were resolved, we can not estimate seismic moment and centroid
depth from these inversions. In order to estimate the seismic moment and centroid
depth from the Rayleigh waves, we fixed source orientation to that found from body
waves. The inversion results are listed in Table 10 (lower part). Among three Q
models, model ABM shows the best centroid depth resolution (see variations of the
variance) and gives the least-misfit centroid depth of 10 km. At this depth, a seismic
moment of 4.8 X 1026 dyn cm was obtained which is in an excellent agreement with
4.6 X 1026 dyn cm determined from body waves.
In Fig. 3.15, observed Rayleigh wave amplitude spectra corrected with
different attenuation models are shown (solid). Dashed lines are the best fit synthetic
spectra inverted for seismic moment at centroid depth 10 km using the source
orientation from body wave inversion. The fits are rather poor for all the attenuation
models. Large misfits occur particularly for R2 waves which are strongly affected by
assumed attenuation model. Large difference in spectra, corrected for attenuation
model ABM and PREM, indicates that attenuation models have a big effect on the
inversions and that without a precise knowedge of the attenuation the results are
poorly constrained.
In previous inversions, the source was a point source with a step source time
function. However, 12 s point source duration of Ml was well constrained by
broadband body wave inversion. Fig. 3.16 shows that correcting for the source
duration does not affect the results.
59
3.7
Rayleigh Wave inversion of Mainshock M2
3.7.1
Data
Ri waves from 8 stations and R2 waves from 10 stations were used (see
Table 8). The modeled periods are the same as those of Ml and again the Gutenberg
continental earth model was used to compute excitation functions. Even though we
estimated source parameters for three attenuation models, the emphasis was put on
the results determined for model ABM, because this model gave the best centroid
depth resolution and seismic moment for Ml. The source duration for this event is
not known, however, as discussed in previous section, this effect is negligible for an
event of this magnitude, and a point source with a step source time function was used
in the inversion.
3.7.2 Results and discussion
In Fig 3.17, the normalized variance is plotted as a function of the depth for
three different attenuation models. At the same resolution level for ceniroid depth as
that of Ml (normalized variance increase by 7% from the minimum), attenuation
model ABM constrained the centroid depth to be shallower than 40 km (least-misfit
centroid depth is 15 km). Inversion results for the three attenuation models at depth
10, 20 and 30 km are listed in Table 11. Again, except strike, inverted dip and slip
varied in a large range. For the ABM model, at depth 10 km. inverted source
orientation is similar to that in the inversion of Ml. As suggested by the inversion of
Ml, available Rayleigh waves can only constrain strike of the source orientation. In
the following inversions, we used some source parameters of Mi as a priori values
because the amplitude spectra of both events shown in Fig. 3.12 are very similar,
indicating they had similar source mechanisms. In order to put more restriction on the
strike of M2, the dip of 86°, which is same as that of Ml, was fixed in the inversion
and other source parameters were inverted using attenuation coefficient model ABM.
The inversion results are listed in Table 12 (upper part). At different depths, the
inverted strike varied in a small range of 153°-158°. The least-misfit strike of 1550
occurs at depth of 20 km, which is very similar to the strike of Ml. To obtain an
better resolution on centroid depth and seismic moment, source orientation of Ml
was assumed. The inverted source parameters were only seismic moment at different
depths. The inversion results are listed in Table 12 (lower part). The least-misfit
occurred at depth 10 km (see variance variation in Table 12), and at this depth, a
seismic moment of 2.2 X 1026 dyn cm was obtained. Thus, for this event, only
strike of 155° and seismic moment of 2.2 X 1026 dyn cm can be constrained. The
least-misfit depth is 10 km. which is similar to that of Ml. However, any differences
of dip and slip between two events can not be distinguished from the inversions.
On the other hand, a comparison of Rayleigh wave radiation patterns,
especially for Ri's of Ml and M2 shown in Fig. 3.18 indicates that the mechanisms
are very similar. The amplitude spectra shown in this figure have been corrected for
attenuation using ABM attenuation model; due to small differences in epicentral
locations of Ml and M2, the corrections should be same for same wave type at a
given station. Taking the average amplitude spectrum value, Ri waves of Ml are
greater than those of M2 by a factor 1.8 to 2.9 (stations WMQ: 1.8; HRV: 1.9; COL:
2.0; HIA: 2.0; PAS: 2.0; AR: 2.9; TOL: 2.8). The average factor is 2.2. This factor
and 4.6 X 1026 dyn cm seismic moment of Ml, which was well constrained in body
wave inversion, gives a 2.1 X 1026 dyn cm (Mw: 6.8) seismic moment for M2. This
seismic moment is consistent with that estimated from the Rayleigh wave inversion at
centroid depth of 10 km (2.2 X 1026 dyn cm).
61
Table 8. Ri and R2 waves used in Rayleigh wave inversion of Ml and M2.
Mi
Sta.
Azim.
i
M2
Wave types
Azim.
(c,
ANMO
23.6
23.9
HIA
26.1
PAS
33.4
35.4
103.2
239.3
311.5
317.0
337.7
352.9
356.3
357.9
COL
BJI
AFI
SLR
TOL
GRFO
WMQ
HRV
GAC
SCP
78.53
117.33
30.85
112.81
22.21
94.06
84.32
85.42
71.89
23.19
114.36
Ri,R2
--,R2
R1,R2
R1,R2
--,R2
R1,R2
--,R2
Ri,-R1,R2
R1,R2
R1,R2
111.53
Ri,R2
116.54
R1,R2
23.6
23.7
26.5
33.2
36.2
103.1
239.2
311.4
316.9
337.6
352.7
356.2
357.7
i
Wave types
(cr)
78.24
117.04
30.57
R1,R2
--,R2
112.57
R1,R2
21.99
94.31
84.39
85.05
71.49
22.77
113.95
111.13
116.14
Ri,-Ri,R2
Ri,R2
--,R2
Ri,---,R2
Ri,-R1,R2
--,R2
--,R2
62
Table 9. Ml Rayleigh wave inversion results for different Q models (all parameters
free).
Depth
(km)
Strike
Dip
Slip
Mo
()
(')
()
(1026 dyn cm)
Variance
PREM
186
156
40
62
153
69
170
10
157
20
30
199
4.9
4.8
5.2
0.47
0.47
0.48
4.4
4.4
4.9
0.45
0.45
0.46
7.5
7.4
8.4
0.63
0.63
0.64
MH78
10
154
43
190
20
156
62
199
30
154
66
181
ABM
10
158
38
183
20
30
156
69
160
68
204
206
*1
Table 10. Ml Rayleigh wave inversion results for different Q model (some
parameters fixed).
Depth
(km)
Strike
Dip
(e)
Slip
Mo
()
(1026 dyncm)
Variance
PREM
10
153
86*
228
20
153
86*
205
30
154
86*
190
4.9
4.4
5.3
0.48
0.50
0.50
4.3
3.9
4.8
0.47
0.48
0.48
0.64
0.66
0.68
MH78
10
153
86*
227
20
30
153
86*
203
154
86*
188
ABM
10
153
86*
235
8.5
20
30
153
86*
218
154
86*
203
7.4
8.2
3.1
10
154*
86*
PREM
181*
20
30
154*
86*
181*
154*
86*
181*
10
154*
86*
MH78
181*
20
30
154*
86*
181*
154*
86*
181*
4.0
5.3
0.50
0.50
0.50
2.8
3.6
4.8
0.49
0.48
0.48
0.68
0.68
0.70
10
154*
86*
ABM
181*
20
154*
86*
181*
4.8
6.2
30
154*
86*
181*
8.1
* parameter is from body wave inversion and fixed
Table 11. M2 Rayleigh wave inversion results for different Q model (all parameters
free).
Depth
Strike
(km)
()
Dip
(
Slip
Mo
()
(1026 dyn cm)
Variance
PREM
10
162
39
180
2.1
20
30
160
83
208
162
73
199
2.0
2.4
0.17
0.17
0.17
MH78
10
162
-39
180
1.9
20
162
60
180
1.8
30
163
72
194
2.1
0.16
0.16
0.16
3.4
3.5
3.9
0.30
0.29
0.29
ABM
10
162
38
179
20
30
156
85
222
162
72
211
65
Table 12. M2 Rayleigh wave inversion results for Q model ABM (some parameters
fixed).
Depth
(km)
Strike
Dip
Slip
(0)
(0)
(C)
10
158
86*
233
20
155
86*
228
30
40
156
86*
36
153
86*
323
10
154*
86*
181*
20
154*
86*
181*
30
154*
86*
181*
40
154*
86*
181*
Mo
Variance
(1026 dyn cm)
3.7
3.7
3.7
4.9
0.30
0.29
0.30
2.2
2.8
3.7
5.2
0.31
0.32
0.33
0.35
* parameter is from body wave inversion of Ml and fixed.
0.31
p.40.4
14.0
11.3
12.7
14.3
14.1
14.2
20.5
23.1
24.1
21.4
33.1
31.4
q..
0.8331.401
0.1015.402
0. 1i3.402
0.1315.401
0.1410*442
0.1444.402
0.1154.442
0.2103.402
0.2448.402
0.284,.+02
0.3143.402
0.3440.402
0.4040.442
0.4503.402
0.3182.402
0.5144.442
0.4532.402
0.1348*402
0.0283.402
0.1321.402
0.1043*403
0.1171*403
0.1311.403
0.1442*403
0.1404.403
42.1
47.5
53.3
40.3
44.0
74.,
$4.4
07.3
100.1
123.1
131.4
157.4
117.4
200.0
p.4i40
10.0
11.3
12.7
14.3
10.1
14.2
20.5
23.1
24.1
20.4
33.1
31.4
42.1
41.5
33.5
40.3
'I.'
74.7
04.4
$1.5
101.0
123.0
130.4
157.4
171.4
200.0
0. 1217.403
q*
..locity
2.40 to 4.20
/..o
0.0340.401
0.1102*402
0.1102.402
0.1401*402
0.1404*402
0.142.402
0.1042.402
0.2183.402
0.2448.402
0.2141.+02
0.3125.402
0.3484*402
0.4004.402
0.4400.402
0.3114.+02
0.5147.402
4.0545.402
0.7483.402
0.8314*402
0. 0340.402
0.1043*403
0. 1111*443
0. 1311*403
0.1402.403
0.12484003
Fig. 3.1 Group velocity measurements of synthetic seismograms by moving
window analysis. (a) Measurement of a linearly dispersed synthetic seismogram. (b)
Measurement of a complex seismogram which consisted of two linearly dispersed
waves as used in (a) but one was shifted by 13 minutes.
E
4-.
0
Group velocity (km/s)
Theoretical group velocity
2
100
PerIod (s)
200
300
Fig. 3.2 Measured (dots) and theoretical (straight line) group velocity. The maximum error of measurement is 0.054 km/s
at period of 37.4 s.
*AAA
,
II1')FThflbI'i
Scteendrp an progress
Fig. 3.3 Time-variable filtering of seismograms. Top are seismograms and bottom are the corresponding amplitude
spectra. (a) is the linearly dispersed wave. (b) is the filtered wave of (a) by time-variable filter, and (c) is the output of a
bandpass Butterworth filter. In both filterings, a passband of 118-17 s was used.
00
a
C
TIE. r.t.r*,: 0
$30SO
Il4fliIiFFM.IiISi±1t4t
002h27,300
Fig. 3.4 Time-variable filter retrieving linearly dispersed wave from complex seismogram. (a) is the complex seismogram.
(b) and (c) are the two linearly dispersed waves retrieved from (a). Bottom are corresponding amplitude spectra.
'0
70
POLE
22.79
99.61
Fig. 3.5 Station distribution and Ri paths for the Lancang-Gengma earthquake.
Azimuthal projection centered on the epicenter of Mi was used. The Ri and R2 paths
in this map are straight lines.
71
10.0
'0
0
C,
0
42
I
2.4
I
Goup velocity (km/s)
M2
Ml
(a)
10.0
(I,
0
0
0
1
200.0
2.4
I
M2
I
Group velocity (kmls)
Ml
4.2
(b)
Fig. 3.6 Two examples of group velocity analysis of Ri waves for stations TOL (a)
and HRV (b). In the period range of 200 to 40 s, the energy contributions of two
events can be well resolved. The period axis is logarithmic.
COL RI
NRV Ri
HIA Ri
PAS Ri
U
60
120
180
60
0
SCP Ri
120
180
0
60
CAC Ri
120
180
I
60
120
th0
0
I
60
1111]
120
[111111
0
60
TOL Ri
180
Il
0
60
120
120
180
GRFO Ri
2.8
0
AFI Ri
2.8
2.8
0
BJI Ri
0
4.0
4.0
3.6
3.6
3.2
3.2
2.8
2.8
60
120
180
0
ITII
60
120
I
I
180
WHO Ri
I
180
0
60
120
180
0
60
120
180
Fig. 3.7 Group velocities of Ri waves.
t)
HIA R2
COL R2
3.2
2.8
60
0
120
3.6
0
180
BJI R2
4.0
3.2
2.8
60
liii
120
60
120
180
3.2
3.2
3.2
2.8
2.8
2.8
0
60
120
0
180
SCP R2
60
120
180
GAC R2
0
3.2
3.2
2.8
2.8
60
120
180
0
60
120
180
0
60
120
Fig. 3.9 Group velocities of R2 waves.
180
120
180
GRFO R2
4.0
3.6
60
0
60
120
PAS R2
4.0
3.6
3.6
0
ANMO R2
4.0
3.6
180
II III III
0
SLR R2
3.6
AFI R2
4.0
J
111111
HRV R2
4.0
180
0
60
120
180
75
--:
-
z4,.
-
'1
d
-
--
-
---
d
Fig. 3.10 Two examples of time-variable filtering of Ri waves for stations TOL
(top) and HRV (bottom). (a) are original seismograms. (b) are the outputs of
bandpass Butterworth filtering from 50 to 250 s. (c) and (d) are Ri waves retrieved
by time-variable filtering for Ml and M2, respectively.
76
[-212.1
25*
:
-212.1
25*
-
2323
25*5
-232.5
2*5
ill
\J\t
-w
---
-III
-212*
7152*
,.*-
-7*23*
[
(?
15*5
25*5
2325
2*55
2323
2555
2323
-7*235
7*535
-7153*
7*5,,
I_fl
S
lIst
-7',,.
Fig. 3.11 Two examples of time-variable filtering of R2 waves for stations AR (top)
and ANMO (bottom).
77
Fig. 3.12 Amplitude spectra period range 60-100 s of Ri (a) and R2 (b) waves for
Ml (solid) and M2 (dashed).
;;q
/
II
0I
'.Oi
.01
(j
,
0I
I-UI
t.0I
0I
/;:os
.01
1.01!
t0I
.0l
)
'UI
tØ
,
sO.(.
.Oi
I
(-).
Os
/
.0l
-
(l
sir
wmq
grfo
hrv
scp
10
toe
toe
to-I
too
I0
to-I
to.,
10.1
to.,
p.do
.0'
(,)
P.,I.d (3)
goc
(3)
on m o
col
__.
r,O-'
10-I
7
S
0
S
to
10.1
t0'
to-I
to
ipos
to'
to'
to-I
-.
,
p.'ed ()
,
p.7Ie
p0Ied (.)
p..bd (.)
._Io.3
S
tI
toe
toe
10-I
30
'./
,.,
S
7
S
(5)
0
tO -I
'
.
,.I.d (I)
hon
bji
tOo
100
to'
Fig. 3.12 (b)
.....
to-'
I
-S
p.Ied C.)
5
to-'
I
1Ø
p.4od (,)
-3
to-'
to'
P3I4 (5)
10
'
too
ponod (5)
Attenuation
coefficient
models
1.4
D
.
MH78
PREM
ABM
I
0.4
40
60
100
80
Period
120
(S)
Fig. 3.13 Attenuation models used in Rayleigh wave inversions.
1.1
-0- MH78
-.- PREM
-a- ABM
0
1.0
>
N
0.8
0.7
0
20
40
60
80
depth (Ian)
Fig. 3.14 Normalized variance of Ml as a function of centroid depth for different
attenuation models. Among the three attenuation models, model ABM, which
constrained the centroid depth to be shallower than 40 km. gave the best centroid
depth resolution.
Fig. 3.15 Observed (solid) and synthetic (dashed) amplitude spectra of Ml at all
stations used in the inversion. Period ranges from 66 to 100 s from left to right.
Results for different attenuation models are shown. The synthetic amplitude spectra
were computed at centroid depth 10 km and the source orientation determined from
body waves. Seismic moments estimated for each attenuation model are listed in
Table 10.
WMQRI
CRFORI
ANMOR2
WMQ R2
GRIVK2
0
C
Amplitudc spectrum (cm s)
0
-
0
Amplitude spcctrum (cm s)
I
0
0
-
Amplitude spectrum (cm s)
1.1
F
::
0.8
0.7
0
20
40
depth
60
80
(km)
Fig. 3.16 Normalized variance as a function of centroid depth for Ml. Blank squares
for a point source with a step source time function, filled squares for a point source
with 12 s duration. In both inversions, ABM attenuation model was used.
Variance vs. depth (M2)
1:
PREM
MH78
ABM
I
I)
'U)
Depth (km)
Fig. 3.17 Normalized variance of M2 as a function of ceniroid depth for different
attenuation models. For this event, all the attenuation models give the same ceniroid
depth resolution, which consirain the centroid depth to be shallower than about 40
km.
-o
- <
-
-
c1
rID
(4
C
c
C
c
-
-
E
C-)
H
E
L
I.
C-)
a)
0
(I)
a)
-o
I.
1
.4
Ini
Fig. 3.18 Rayleigh wave amplitude spectra of Ml (solid) and M2 (dashed). For each wave, plotted amplitude spectrum is
from 66 to 100 s (from left to right). The spectra are very similar, indicating that Ml and M2 had similar source
mechanisms.
.
4. STRONG GROUND MOTION ANALYSIS
4.1
Introduction
Aftershocks Al and A2 were well recorded by strong ground motion
recorders. These data provide us with a chance to look at the near-field features of the
earthquakes and to verify the previous determinations of the centroid depths and
source time functions of the events.
Strong ground motion data have been used to study earthquake source
process by many researchers (e.g., Hartzell and Heimberger, 1982; Liu and
Heimberger, 1985; Olson and Aspel, 1982; Bouchon, 1982; Gariel et al., 1990).
Strong ground motion data are more sensitive to the time and spatial distribution of an
earthquake fault rupture than teleseismic data, especially for an intermediate size
event. In this case, teleseismic body waves can not constrain source time history
precisely because long-period data are not sensitive to the short source duration.
In this chapter, forward modelling discrete wavenumber technique developed
by Bouchon (1981) was used to study aftershocks Al and A2. The characteristics of
the acceleration in different frequency bands were tested and the particle motions
were plotted to see which frequency band provides stable data because at high
frequencies, crustal heterogeneities can severely influence the data and such effects
can not be modeled by our method. We found only frequencies lower than 1 Hz
behaved in a predictable fashion and this band of data were modeled for different
crustal structures and source centroid depths. Due to small sizes of the events (from
teleseismic body wave inversion, a seismic moment of 1.54 x 1025 dyn cm and 1.5 x
1024 dyn cm for Al and A2, respectively) and low pass filtering, only point sources
were considered. The source orientations were tested to see if they agree with those
obtained from teleseismic body wave inversion. Finally, the source time functions,
centroid depths, and seismic moments of Al and A2 were obtained from the
modeling. The results were compared in detail with those from teleseismic longperiod and broad-band body wave inversions.
4.2 Data and Data Analysis
The data used in this study are the three component acceleration records
recorded at station FUB. The records were provided by the Institute of Geophysics,
State Seismological Bureau, China, and Department of Geological Sciences,
University of New York, Binghamton, who deployed this station after the
mainshock. The instrument is a three component DCS-302 digital accelerograph with
sampling rate of 100 sps (Chen and Wu, 1989). The station was located nearly due
north of Al and A2 (solid square in Fig. 1.2). The parameters of the station are given
in Table 14. The epicentral parameters of Al and A2 used in this table are from the
PDE; the KTSN locations are very similar, but the origin times are earlier by about 3
seconds than those of PDE for both events. For a half space earth model, with
compressional velocity of 6.0 km/s, Poisson's ratio of 0.25, and source depth of 7
km. the theoretical first arrivals at station FUB using KTSN's locations are early by
1.42 s and 0.33 s compared to the observed first arrivals of Al and A2, respectively.
On the other hand, using PDE's locations, the first motions are delayed by 0.58 s
and 1.45 s. It is hard to discriminate which location is better only from one station's
strong ground motion record, however, available teleseismic short-period records of
both events indicate that the theoretical first arrivals computed for PDE's parameters
match the observed first arrivals significantly better than for KTSN's. Therefore, the
PDE's epicenter locations of Al and A2 were used in this part of the study. Three
component ground accelerations, velocities and displacements of Al and A2 are
shown in Fig. 4.1 (Al) and 4.2 (A2). The originally recorded lengths of Al and A2
were 17.52 and 21.05 s, respectively. In the figures, the record of A2 was truncated
to the length of Al to facilitate easier comparison. Later in forward modeling, the
synthetic seismograms were also computed for this time length. The ground velocity
records were obtained by integrating the accelerations once. Then the trend and mean
were removed and velocities were integrated again to obtain the displacements. To
avoid truncation effects, the time series were tapered with a 2 second (Al) and 1
second (A2) cosine taper on sides of the velocity records.
4.2.1
High frequency features
An examine of the observed ground acceleration data of both events indicates
that the frequency content of vertical component is much different from that of
horizontal components; the vertical component is much richer in high frequencies (see
Fig. 4.1 and 4.2).
To look in more detail at this phenomenon, the high-pass and low-pass
filtering were performed on the acceleration records of Al with filter cutoffs changing
in 5 Hz increments. Fig. 4.3, 4.4 show the low-pass and high-pass filtering results,
respectively.
Low-pass filtering (Fig. 4.3) shows that the three components seem to have
the same frequency content below 10 Hz; for cutoffs higher than 10 Hz, richer high
frequency content of vertical component than that of horizontal component is
obvious.
Moreover, the high-pass filtering (Fig. 4.4) indicates that, compared to the
vertical component, there is little energy on the two horizontal components for the
frequencies higher than 10 Hz. Clearly, the high frequency content of the vertical
component, higher than 10 Hz, was amplified for some reason.
The acceleration amplitude spectra of Al and A2 are shown in Fig. 4.5 and
4.6, respectively.
The idealized ground displacement spectrum model is the 'co-square' model
(e.g. Aki, 1967), according to this model, displacement spectra are flat for the
frequencies lower than fo and decay as co2 for the frequencies higher than f0. The f0
is called corner frequency and is dependent on the dimensions of an earthquake
rupture. From this idealization displacement spectrum model, ground acceleration
spectra (displacement spectra multiplied by &) should be flat at frequencies higher
than f0. However, in reality, as noted by many researchers (e.g., Hanks and
McGuire, 1981; Hanks, 1982), acceleration spectra are band limited between corner
frequency fQ and cut-off frequency fmax. Beyond fm, spectrum amplitude decay
abruptly.
The acceleration spectra shown in Fig. 4.5 and 4.6 are band limited. For both
events, estimated fm of two horizontal components are very similar, which is about
10 Hz. However, the amplitude spectra of two vertical components were amplified
between 10 and 20 Hz. The amplification made the
of vertical component be
higher than those of horizontal components. We will present further discussion of
this problem in the discussion section.
4.2.2
Effect of medium heterogeneity
To examine the effect of medium heterogeneity, which can not be modeled by
our method, the acceleration particle motions of Al were plotted. The particle
motions were plotted for high frequencies between 1 and 10 Hz and low frequencies
between 1 and 0.0667 Hz. The 10 Hz upper limit was used in order to avoid vertical
component amplification discussed above; while the 0.0667 Hz lower limit was used
in order to remove long-period trends. The high frequency particle motions are
shown in Fig. 4.7, and the low frequency particle motions are shown in Fig. 4.8. In
both cases, the plots start at the first motion and span a 9 second time window in 0.5
s interval (1-18). In each plot of Fig. 4.7 and 4.8, from left to right, are NS vs. EW;
Z vs. EW and Z vs. NS. The positive axes are up, north, and east.
With the aid from the results of teleseismic body wave analysis, some
features of particle motions at station FUB can be predicted. Because Al is a nearly
pure strike-slip event with source duration of 7 s, as indicated by teleseismic longperiod and broadband body wave inversions, and station FUB is located along the
strike north of the epicenter, S waves at FUB should have horizontal polarization in
the EW direction (SH waves) and P wave should be nearly nodal and have principal
motion in the vertical and NS direction. Therefore, in spite of 7 s source duration,
the P waves should have little effect on the EW component. The source-station
configuration indicates that the first S arrival is about 2 s after the first motion and
therefore we mainly examine the particle motion of in the first 2 s after the first
motion (plots 1-4 in Fig. 4.7 and 4.8) and in a 7 s time window (source duration)
which started 2 s after the first motion (plots 5-18 in Fig. 4.7 and 4.8). Within this 7
s time window, principal arrivals were S waves.
The first 2 s of high frequency particle motions (plots 1-4, Fig. 4.7) show
that the P wave motions are nearly vertical, indicating a presence of a low velocity
sedimentary layer.
From plots 5-18 (Fig. 4.7), we observe that at high frequencies, S wave
particle motions show great scattering in horizontal plane, only a few plots (9, 11,
18) have dominant EW motions. On the other hand, at low frequencies (Fig. 4.8),
the particle motions clearly show the S-type EW motions (plots 7,9, 10, 11, 12, 13,
14, 15), indicating that this frequency band data behave in a predictable fashion.
Thus, in following modeling, we confined our analysis to frequencies lower than 1
Hz.
91
4.3 Forward Modeling for a Point Source
Three component (N-S, E-W, Z) theoretical displacement seismograms of Al
and A2 were computed using discrete wavenumber technique for a point source
model. Synthetic velocity and acceleration seismograms were obtained by taking first
and second derivative of the displacements in frequency domain. Modeled parameters
were the seismic moment, centroid depth, source time function and a flat layered
crustal model, which was specified by compressional velocity, shear velocity,
density and thickness of the layers. Synthetic seismograms were first computed for a
step source time function, then convolved with a triangle to produce a smooth rise
time. All synthetic seismograms had 256 samples in a time window of 17.52 s. The
maximum frequency computed is that of Nyquist frequency (7.3 Hz). A bandpass
filter (1-15 s) was used to filter the synthetic and observed seismograms so that they
can have the same frequency content.
The modeling was performed by trial and error. Synthetic seismograms were
computed for different crustal models, orientations, centroid depths and source time
functions and compared with observed seismograms by eye. The emphasis of
comparison was put on the E-W component because the main energy arrivals of Al
and A2 on this component were anti-nodal SH waves. When computing the synthetic
seismograms, the origin times of Al and A2 were adjusted so that both synthetic and
observed seismograms can have same first arrival times.
Using the source orientation determined in teleseismic body wave analysis
and a half space crustal model, computed synthetic seismograms indicated that the
polarities of the horizontal components were reversed relative to the observed
seismograms. Due to small amplitude, it is hard to identify whether or not the
polarities of the vertical components were also opposite. Because the main phases in
EW components were the anti-nodal SH waves, it is difficult to accept that the
reversals of polarities were caused by wrong estimates of the source orientations or
source mislocations. It appears that the observed data were incorrectly labelled, and
therefore, polarities of all three components of observed data for Al and A2 were
reversed, hereafter.
92
4.3.1 Crustal model and centroid depth
The details of the crustal structure are unknown at this time. So, we have to
find the crustal model which can correctly account for the main path effects.
At first, the importance of reflection from the Moho was tested. The synthetic
acceleration, velocity and displacement of Al were computed for a half space and one
layer over a half space crustal models. The specifications of the half space crustal
model are: compressional velocity of 6.0 km/s, shear velocity of 3.5 km/s and
density of 2800 kg/cm3. The synthetic seismograms computed for this model are
shown as dashed lines in Fig. 4.9; solid lines are observed seismograms. The farfield source time function, which is the first derivative of the source time function, is
shown at the bottom of the figure. The 1.6 s source duration was used. The synthetic
seismograms (dashed lines) computed for a one layer over a half space crustal model
and observed seismograms (solid lines) are shown in Fig. 4.10. The layer parameters
are the same as those for the half space in the previous model, the thickness is 33 km.
and the half space has compressional velocity of 8.0 km/s, shear velocity of 4.6 km/s
and density of 3300 kg/cm3. In both models, a centroid depth of 11 km and seismic
moment of 1.0 X 1025 dyn cm were used. Fig. 4.9 and Fig. 4.10 indicate that, at this
epicentral distance range, Moho reflection is insignificant.
The particle motions of observed data suggest that there is a low velocity
sediment layer in the receiver region. To investigate the effect of sediments on the
seismograms, several crustal models with a sediment layer over a half space were
tested for three ceniroid depths (4,7 and 10 kin). Six such crustal models are listed in
Table 14. Among these, only one parameter was varied with respect to starting
model, either compressional velocity (model CM1, CM2 and CM3) or thickness of
the sedimentary layer (model CM4, CM5 and CM6). A Poisson's ratio of 0.25 is
assumed to both sedimentary layer and half space. Throughout this modeling, source
orientations are those derived from long-period teleseismic body wave inversion.
Seismic moments were varied so that the synthetic seismograms can match the
observed seismograms. A simple source time function with a duration of 1.6 and 1.4
s was used for Al and A2, respectively, because we want to match the common
features in the seismograms, which are likely to be caused by the crustal structure,
because of the similarity of the mechanisms and epicenters of these two events.
E-W component synthetic accelerations (dashed lines) of Al computed for six
crustal models and different centroid depths are shown in Fig. 4.11 (depth: 4 km),
93
4.13 (depth: 7 1cm) and 4.15 (depth: 10 1cm). Fig. 4.12, 4.14 and 4.16 are for A2.
Solid lines in all these figures are observed E-W component accelerations. Fig. 4.11
and 4.12 show that the source centroid depth of 4 km is too shallow for all crustal
models, except CM3. At this depth, contrary to observations, large amplitude trapped
phases were generated in the sedimentary layer. Even though no large amplitude
reverberations were generated in the model CM3, the fit between synthetic and
observed accelerations is poor, especially for A2 (see Fig. 4.12). Fig. 4.13, 4.14,
4.15 and 4.16 show that the fit is improved for source centroid depths of 7 km and
10 km. For these two depths, the synthetic seismograms of all six models fit the
observed data reasonably well. However, the fit for the second cycle of the observed
seismograms, models CM1 and CM5 are better than the other models. Model CM1
and CM5 are very similar (see Table 14); CM1 was selected because this model
predicts a more vertical incident angle than CM5 for P waves which was suggested
by particle motions. The data were insensitive to centroid depth between 7 and 10
km; the centroid depths of 7 km for Al (consistent with teleseismic broadband body
wave inversion) and 10 km for A2 were used in following modelling. The computed
seismograms (dashed) of ground acceleration, velocity and displacement for crustal
model CM1, centroid depth 7 km for Al and 10 km for A2 are shown in Fig. 4.17
(Al) and 4.18 (A2). Again, the solid lines in these figures are observed
seismograms.
4.3.2
Effect of Source Orientation
The source mechanisms used in previous modeling were those of teleseismic body
wave inversions. In order to find the best source orientations from strong ground
motion data, different strikes, dips, slips were modeled for Al. In this modeling,
each parameter of strike, dip and slip was deviated by up to ±10° from the source
orientation derived from teleseismic body wave inversion. The crustal model of
CM1, centroid depth of 7 km. seismic moment of 1.0 x i0 dyn cm and source
time function with a duration of 1.6 s were used. Fig. 4.19 gives three component
synthetic displacements (dashed) computed for five different strikes. The results
indicate that all three component displacements are sensitive to the deviation of strike
and the best fit strike is 165°, which is that of teleseismic body wave inversion. In
Fig. 4.20 synthetic displacements computed for five different dips are shown. The
SV waves (N-S component) seem to be more sensitive to deviations of dip than SH
94
waves (E-W component). The best fit dip is 900, which is that of teleseismic body
wave inversion. The deviations of the slip, however, do not have much effect on
either SH or SV (Fig. 4.21). Thus, the source orientation from teleseismic body
wave analysis is consistent with those of strong ground motion modeling.
4.3.3
The best fit source time function
Using the best crustal model, centroid depth and source orientation, computed
seismograms for a simple source time function can not entirely fit the major features
of observed data, indicating that the source had more complicated time history (see
Fig. 4.17 and 4.18). Large misfit between synthetic and observed seismograms of
Al occurs at about 2-5 s and 7-10 s into the E-W displacement record (see Fig.
4.17). A2 also shows some misfits, but less significant (see Fig. 4.18).
We assume that the remaining misfits were caused by complexities in the
source time functions. The best fit source time functions for Al and A2 were found
by trial and error. In this modeling, the synthetic seismograms were first computed
for a step source time function, then the results were convolved with complex farfield source time functions parameterized in the same fashion as in the teleseismic
body wave analysis. The best fit synthetic (dashed) and observed (solid)
seismograms are shown in Fig. 4.22 for Al and 4.23 for A2. The source time
function and its first derivative (far-field source time function) are shown at the
bottom. A source duration of 7 s and 2 s, a seismic moment of 2.2 X 1025 dyn cm
and 2.5 X l0 dyn cm are obtained for Al and A2, respectively. A comparison of
source time functions between Al and A2 indicates that Al is a slowly starting event.
95
4.4
Discussion
Using CM1 as source crustal model, a new estimate of the source time
function and centroid depth of Al was obtained by inverting teleseismic broad-band
P waves. The source mechanism from teleseismic long-period body wave inversion
was used. Only the wave-forms of the stations BJI and HIA were used in the
inversion. The observed and synthetic seismograms are shown in Fig. 4.24. The
least-misfit centroid depth of 7 km was obtained and the source time function
indicates that the main energy release was in three pulses of total duration of 6 s. A
comparison of far-field source time functions from strong motion modelling and
teleseismic broadband waveform inversion is shown in Fig. 4.25 (solid line: strong
motion data; dashed line: teleseismic data). Two source time functions are similar.
The differences in the beginning parts of the rupture could be due to the emergent
first motions at teleseismic stations because Al is a slow starting event. The
difference in the third pulse could be due to unaccounted structural effects, e.g.,
complexities of the upper mantle structure, because both broadband stations BJI and
HIA have a short epicentral distance.
The modeled crustal structure from strong ground motion data includes a low
velocity sedimentary layer. This low velocity sedimentary layer then can separate
principal P and S waves to vertical component and horizontal components,
respectively. Thus, the different high frequency content between P and S waves leads
to the different fmax, as discussed earlier, in vertical component and horizontal
components. Cranswick and Mueller (1985) suggested that high frequency P waves
are relatively amplified due to SV-to-P conversions at the free surface. They
suggested that the amplification is associated with a thin UnSaturated Surface Layer
(USSL), which consists of dry, unconsolidated sediments. The lower boundary of
USSL is defined by water table which forms a velocity discontinuty for P waves but
not for S waves. S waves then have a non-vertical incident angle and generate SV-toP conversions at free surface. The SV-to-P conversions reberate in USSL and are the
function of frequency. According to this hypothesis, the S V-to-P conversions occur
at free surface, predicting that high frequency P wave amplification does not occur
before the first SV-to-P conversion arrival. To check this idea, the acceleration
amplitude spectrum ratios of P and S waves from consecutive time windows were
plotted in Fig. 4.26 (Al) and 4.27 (A2) to see when the amplification begins to
occur. In Fig. 4.26, the P and S wave spectrum ratio of Al for total 7.2 s source
duration is plotted. The P waves are taken from vertical component starting at P wave
first motion for three consecutive time windows and each window spans 2.4 s. The S
waves are taken from EW component starting from the S wave first motion for the
three same time windows as used for P waves: top is the ratio for the first 2.4 s
source duration (first time frame), middle is for the following 2.4 s (second time
frame) and bottom is for the last 2.4 s (third time frame). Fig. 4.27 is the P and S
wave spectrum ratio of A2 for the total 2 s source duration. The S-P travel times, 2.4
s for Al and 2.8 s for A2, were computed using crustal model CM1 and a source
depth of 7 km (Al) and 10 km (A2). All these ratios show an abrupt increase above
10 Hz suggesting that P waves are richer in high frequencies than S waves for whole
source duration (-7 s for Al and 2 s for A2). Furthermore, because the first SV-to-P
conversion at free surface should arrive about 2.4 s and 2.8 s after the first P arrivals
for Al and A2, respectively, the observed amplification in the first time frame
excludes SV-to-P conversions (Cranswick and Mueller, 1985) as the mechanism for
the amplification.
Other possible mechanisms for the distinctive different high frequency content
between P and S accelerations are either from medium anelastic attenuation or
inhomogeneity of the fault: Liu and Helmberger (1985) show that P and S
accelerations can have compatible frequency content if proper Qa and Q values are
used to correct for attenuations. From the data of a 1500 m deep, three-level
downhole seismometer array, Hauksson et al. (1987) shows that the medium has a
high attenuation for high frequency (higher than 10 Hz) S waves. Rather than from
the medium attenuation effect, Papageorgiou and Aki (1983) suggest that fm, which
is inversely proportional to the size of the fault nonelastic zone (cohensive zone size),
is primarily a source effect. If so, the different max for P and S waves, its equivalent
is the different high frequency content between P and S accelerations, is a pure
source characteristic. However, from our data, we can not distinguish whether the
different high frequency content in P and S waves is a medium attenuation effect or a
source effect.
97
Table 13. Parameters of the Strong Motion Station.
Event
A (km)
Azi. (°)
Back-Azi.(°)
Al
15.17
342.6
162.6
A2
17.95
342.6
162.6
Table 14. Layer parameters of the crustal models use for modeling of strong motion
data.
CM1
CM2
CM3
CM4
CM5
CM6
Thick. (km)
1.5
1.5
1.5
1.5
a (km/s)
2.5
1.4
2.0
3.5
2.0
2.2
4.5
2.6
2.4
3.0
2.0
3.0
2.5
3.0
1.7
1.7
1.7
2.1
2.1
2.1
(km/s)
p(g/cm3)
HS
6.0
3.5
2.8
Here a is P-wave velocity, 13 is S-wave velocity, p is density and thick. is the
thickness of the sedimentary layer. HS are parameters of the half-space.
F
:n.
.-
1
Acceleration
u.
-
r
..
.
-r-
-
.-
VeloCitY
tV
I7.52
4.
C
-2
3.3*4
a
*
-a
-a.3z.
-
2
32
-
-,
Displacement
Fig. 4.1 Observed three-component (EW, NS and Z from top to bottom) ground
acceleration (top), velocity (middle) and displacement (bottom) for Al. Arrow
indicates the possible first motion. Acceleration is in cm/s2, velocity cm/s and
displacement cm.
100
file
C
-2e
2
-2e
C. 23
2e
Acc&eration
F-2.241
V
[
2.3d1
I
Ve1oC ty
::
C_____
Dispf ace ment
Fig. 4.2 Observed three-component (EW, NS and Z from top to bottom) ground
acceleration (top), velocity (middle) and displacement (bottom) for A2. Arrow
indicates the possible first motion. The units are the same as those in Fig. 4.1.
101
-
-
a
.
's__/
-
3t._
'-4.
<1Hz
--
--
--
<5 Hz
1:
-.
'
<10 Hz
Ic....
1'NW
r.'Vfr
<15 Hz
Fig. 4.3 Low-pass filtering of ground acceleration record of Al.
-
102
1
<20 Hz
<25 Hz
<30 Hz
<35 Hz
Fig. 4.3 (continued)
103
>1 Hz
I
.f
-.
- r--
>5Hz
.,
.
.
>iO Hz
t
-,--
J
-t t
-
't
>15 Hz
Fig. 4.4 High-pass filtering of ground acceleration record of Al.
-'-
104
.
.
..
fwfl. rrU.t
..
s
t
t
.
'-
4swj
..Lij
-
4
'
>20 Hz
i t-.*4I. tUl
7.
I-s
'I I
* j.
il 41' pl1 tI.t ii
t
f I
*frI
'
4,--.
:T
>25 Hz
-.
I
7.:
I
i4t.--i4.Ib-i.
ll
P1 yt4P4jJØII34I4ff111r
i-t* -I f,-.
I -
-,
4
$-i'u1 i.ti.i*
>30 Hz
>35 Hz
Fig. 4.4 (continued)
tir.** 4tic.
105
Fig. 4.5 Acceleration spectra of Al. From top to bottom is EW, NS and Z
component fmax is estimated by arrows.
106
Fig. 4.6 Acceleration spectra of A2. From top to bottom are EW, NS and Z
component fm
is estimated by arrows.
107
Fig. 4.7 High frequency (1-10 Hz) acceleration particle motions of Al. The plot start
at the first motion and span a 9 s time window in 0.5 s interval (plots 1-18). From
left to right are NS vs. EW; Z vs. EW and Z vs. NS. The positive axes are up, north
and east.
C)
2
NI
2
i
CM
2
Hr
Fl
2
HI
1
HI
2
I
A°
1.
2
2
[J
2
O[)
2
___
2Or
-.
2
2
NI
8
1
8828
-f')
228202
2$82°jj
2
2
1
Ia
882820
82882
2
8
2a
0
920
8288
2
8
a
U
2
2
C'
\T
2
8
C
o
0
!11
2°
qk
a
2°
A
A
9
2°
A
HI
0
A
I
___________________
a
a
A
A
0)
-
A
NI
A
A
Ui
1
NI
A
A
2°
441
1
A
A
____________
NI
111
Fig. 4.8 Low frequency (0.0667-1 Hz) acceleration particle motions of Al. The plot
start at the first motion and span a 9 s time window in 0.5 s interval (1-18). From left
to right are NS vs. EW; Z vs. EW and Z vs. NS. The positive axes are up, north and
east.
00
f
(\J
00000ff 111
,eo0o0ffi?
0
A
A
A
A
0000ff
I
70
'1
I
'1
I
ii
U
0
i
q
0 OUff
70
0
I
2
U.)
P)
q
9
a
0
a
C I C I C C C C C7
70
i
(0
70
C
fT -
113
rH
-
0
-2
)
o
7
-2
0
-2
2
-
4
2
0
*6
-2
2
o
8
-2
-2
-.
0
Lw
2
0
-2
-
*
2
4
*15
9
-
-2
0
-
4
2
-2
0
Lw
0
4
2
*15
20
20
*5
*3
10
10
7
o
0
-3
-s
-*0
-*0
-IS
-13
-20
-20
0
-*0
20
*0
-20 _
-70
-*0
10
-*0
-*5
-20
0
-*0
('0
('0
0
*6
*0
2
20
*3
*0
0
o
c
11
.10
.10
-*5
.15
20
-*0
..lO
0.
*0
2
-20
*0
i0
l6
('0
20
2*'
/
12
I
-*0
-20 ____
-*5
-20
-*0
-*0
-*0
0
*0
20
-20
-20
.20 _
-I5
-is
I
-*0
0
Lw
*0
70
Fig. 4.8 (continuec
-20
-*0
0
-S
*0
70
c)
to
N
(0
Q
a
9
a
0
Mi
0*
U
j
0
T
0
a
a
-Th
**flav eq
III
C
OM
j
115
-1
2
-1
2
16
2
.O61ty
16
1.
-2
1.
-2.
1_
dl.pI.c..nt
16
-2.
Fig. 4.9 Synthetic (dashed) three-component (EW, NS and Z from top to bottom)
ground acceleration, velocity and displacement computed for a half space crustal
model and a centroid depth of 11 km for Al. Solid lines are the observed
seismograms.
116
I
16
'.o.ty
2
I
I
16
I
1TiTTT.
-2.
2
I
I
I
16
-2.
TTJ\
.
.
...
Fig. 4.10 Synthetic (dashed) three component (EW, NS and Z from top to bottom)
ground acceleration, velocity and displacement computed for a crustal model, which
includes the Moho discontinuity, and a centroid depth of 11 km for Al. Solid lines
are the observed seismograms.
117
CM1
CM2
CM3
\j
i
CM4
z::----
I
I
I
I
CM5
CM6
1.
2
ourc.-tm.-fncto.,
16
Fig. 4.11 Synthetic (dashed) EW acceleration seismograms computed for six crustal
models and a source centroid depth of 4 km for Al. Solid lines are the observed
seismograms.
118
CM1
acceleration
3
16
CM2
2
accel oration
16
- CM3
CM4
CM5
CM6
1.
2
*ourc.-tln..-functlo.,
16
Fig. 4.12 Synthetic (dashed) EW acceleration seismograms computed for six crustal
models and a source centroid depth of 4 km for A2. Solid lines are the observed
seismograms.
119
CM1
17
I
2
r,
acceleration
16
:----- CM2
16
CM3
CM4
2
_
CM5
fli.1cc
1
2
.ourc.-t1..-funct1o,
Fig. 4.13 Synthetic (dashed) EW acceleration seismograms computed for six crustal
models and a source centroid depth of 7 km for Al. Solid lines are the observed
seismograms.
120
CM1
CM2
CM3
CM4
CMS
CM6
2
.ourco-t1n.-funct1on
Fig. 4.14 Synthetic (dashed) EW acceleration seismograms computed for six crustal
models and a source centroid depth of 7 km for A2. Solid lines are the observed
seismograms.
121
CM1
28
2
acceleration
16
CM2
-1.
213
2
accol9ratiOn
16
-1
CM3
-1.
CM4
CM5
CM6
1.
2
.on-c.-t1m.-funct1ofl
i.e
Fig. 4.15 Synthetic (dashed) EW acceleration seismograms computed for six crustal
models and a source centroid depth of 10 km for Al. Solid lines are the observed
seismograms.
.22
.çClS alid a
seisli01
ç ctUSt
ati0 se'$'°
ifr 4.16
dde%'tl0
lcOl /
Solid. lilies
axe
123
2
-t
2
.cc.,_e.t;..,
16
-6
2
16
1.
-2.
I-
-2.
16
-2.
Fig. 4.17 Synthetic (dashed) three-component (EW, NS and Z from top to bottom)
ground acceleration, velocity and displacement computed for the best crustal model
CM1 and source centroid depth 7 km for Al. Solid lines are the observed
seismograms.
124
16
2
2
41I l6t
16
Fig. 4.18 Synthetic (dashed) three-component (EW, NS and Z from top to bottom)
ground acceleration, velocity and displacement computed for the best crustal model
CM1 and source centroid depth of 10 km for A2. Solid lines are the observed
seismograms.
125
Fig. 4.19 Synthetic (dashed) three-component (EW, NS and Z from top to bottom)
ground displacement computed for five different strikes for Al. The dip and slip
derived from teleseismic body wave analysis, the crustal model CM1 and a source
centroid depth of 7 km were used. Solid lines are the observed seismograms.
126
E-W
N-S
1.
Strike: 145°
-2.
16
2
z
-2.
Strike: 155°
1.
-z
2
dt.pI.c
16
--------------.=--
-2.
Strike: 165°
2
d12.c'
26
Strike: 175°
16
2
Strike: 185°
-2.
2
-2.
41I.C*
16
Fig. 4.19
127
Fig. 4.20 Synthetic (dashed) three-component (EW, NS and Z from top to bottom)
ground displacement computed for five different dips for Al. The strike and slip
derived from teleseismic body wave analysis, the crustal model CM1 and a source
centroid depth of 7 km were used. Solid lines are the observed seismograms.
128
Dip: 700
-2.
-2.
2
2
14
16
Dip:
900
Dip: 100°
Dip: 1100
Fig. 4.20
129
Fig. 4.21 Synthetic (dashed) three-component (EW, NS and Z from top to bottom)
ground displacement computed for five different slips for Al. The strike and dip
derived from teleseismic body wave analysis, the crustal model CM1 and a source
centroid depth of 7 km were used. Solid lines are the observed seismograms.
130
Slip: 158°
-2-
2
16
._,.c...,t
-2
Slip: 168°
-2
1C
Slip: 178°
1.
-2
1.
2
44a.c.t
2
dl.e1.Cfl
I.
-2.
S.
-2
a.
Slip: 188°
-2.
1-
2
dII.CS.St
16
-2.
S.
-2.
1.
Slip: 198°
-2
1.
2
d.pI.o.....S
16
Fig. 4.21
-2.
131
iiiiiI\"TIii;
-1
-i
16
2
16
2
16
2
2
-2.
1.
d.p1o..t
16
d1p
16
Fig. 4.22 The best fit synthetic (dasjied) three-component (EW, NS and Z from top
to bottom) ground acceleration, velocity and displacement for Al. Source time
function and its first derivative are shown at the bottom. Solid lines are the observed
seismograms.
132
2
2
2
....toC*ty
16
16
Fig. 4.23
Fig. 4.23 The best fit synthetic (dashed) three-component (EW, NS and Z from top
to bottom) ground acceleration, velocity and displacement for A2. Source time
function and its first derivative are shown at the bottom. Solid lines are the observed
seismograms.
133
Al
o_._A___A C
hia i
me Function
8roaand (1) - 0.147cm
024 6$1a24
time (s)
w.PIi_
o
time (s)
;o
0
Fig. 4.24 Observed (solid) and synthetic (dashed) broadband P waves of Al. The
source crustal model CM1 was uaed in the inversion.
I
I
'I
I'
I'
I
ji
1
I
I
A
I
1
I
I
I
I
I
I
I,
I
F
I
1
I
I
/
1
F
V
I
I
/
I
/
F
I
I
I
I
.
'S
I
)
I
I
I
1
0'
F
2
8
Fig. 4.25 The far-field source time functions of Al determined from teleseismic broadband data (dashed) and strong
ground motion data (solid).
135
P
102
101
10°
10-I
10-2
10_I
100
101
102
10'
I 0°
10'
10-2
10-1
100
10'
102
10'
100
10-1
102
I'''_
100
11
Fig. 4.26 Acceleration amplitude spectra ratio of P (taking from vertical
component)
and S (taking from E-W component) waves of Al for total 7.2 s source duration.
Top is the ratio for the first 2.4 s source duration, middle is for following
2.4 s and
bottom is for last 2.4 s.
136
P
1 02
lot
100
10_i
10-2
10_i
100
lot
Fig. 4.27 Acceleration amplitude spectra ratio of P (taking from vertical component)
and S (taking from E-W component) waves of A2 for total 2.0 s source duration.
137
5. RUPTURE PROCESS OF THE LANCANG-GENGMA
EARTHQUAKE SEQUENCE
Even though we could not determine the finite source parameters of Ml
directly due to insufficient azimuthal coverage offered by the broad-band data, some
estimates of the fault dimensions can be inferred from the characteristics of the point
source model.
For a kinematic rupture model, assuming a rupture velocity of 2.5 km/s. a 12
s point source duration of Ml suggests a 30 km and 60 km long rupture for unilateral
and bilateral ruptures, respectively. Estimated seismic moment of 4.6 X 1026 dyn cm
and 24 km fault width (twice of the centroid depth), combined with the medium
rigidity of 3.35 X 1010 N nr2 assumed in our half space earth model, suggest an
average slip of 1.90 and 0.95 m on the fault for unilateral and bilateral rupture
models. The length of 60 km of observed ground fissures presumably caused by
Ml and not well developed surface fault slip suggest that the rupture of Ml was
bilateral.
Based on the observed ground deformation, aftershock distribution, the
bilateral rupture model of Ml, geological information, and using the concepts of
geometrical barriers and fault asperities (King and Nabelek 1985), we propose the
following scenario for Lancang-Gengma earthquake sequence.
In Figure 5.1, the aftershock epicenters and geological faults are plotted. The
heavy lines with arrows, where the surface fault ruptures and large ground fissures
were found, indicate the possible rupture areas and rupture propagation directions of
Ml and M2. Assuming the rupture of Ml initiated at the middle of the rupture area
(blank star), which is near the intersection of Lancang and Mujia fault and about 10
km northeast of the KTSN epicenter (solid star), and propagated about 30 km to
NNW and 30 km to SSE along the Lancang fault.
SSE propagating rupture presumably stopped in the area near 22.6° latitude,
where a preexisting northeast striking fault segment, referred to as Fl, intersects and
probably offsets the Lancang fault. The rupture was probably terminated by the
intersection of Lancang fault and Fl. In the vicinity of Fl, we observe a confluence
of faults which formed the geometrical barrier for the rupture. The aftershock Al and
A2 as well as the dense group of small aftershocks at the south terminus of the
aftershock zone may be associated with the termination of the SSE propagating
rupture and increased stress load on this complex region.
138
NNW propagating rupture presumably reached the area near 23.2° latitude.
To examine what caused the rupture to be terminated, the aftershock epicenters which
occurred in the first 3 hours following the occurrence of the mainshocks are plotted in
Figure 5.2. It is clear that the rupture zone of the M2 was bounded by two groups of
aftershocks: one group was located near 23.4° latitude north of the epicenter of M2
and another was south of it near 23.2° latitude. The northern group of the aftershocks
seem to be associated with the northeast striking E. Nantinghe fault; the southern
group of the aftershocks, which consisted of four aftershocks including the two
largest aftershocks in the period, suggests that there is a northeast striking fault
segment at 23.2° latitude (dash line in Fig. 5.1 and 5.2) associated with these
aftershocks. Jiang et al. (1989) have reported a geological fault complex running
through the area and we refer to this fault segment as F2. F2 probably right-laterally
offsets the Lancang fault on which Ml ruptured and formed the geometric barrier for
the rupture. Thus, the NNW propagating rupture of Ml appears to be terminated by
this geometric complexity.
The role of preexisting faults in the earthquake rupture initiation and
termination has been extensively discussed by King and Nabelek (1985). They
suggested that the initiation and termination of earthquake ruptures occurs near fault
bends. The termination of the rupture has to cause other deformation near the
termination zone. This deformation may break the asperities on the other main fault
and initiate another major earthquake. The loading associated with the termination of
NNW propagating rupture of Ml broke the asperities on the offset fault segment
northeast of the intersection of Lancang fault and F2 and thus initiated the M2.
The rupture of M2 probably started near F2 at 23.2° latitude and unilaterally
propagated to the NNW. The rupture was fmally stopped by E. Nantinghe fault near
23.4° latitude and most aftershocks at the north terminus of the aftershock zone are
probably associated with the rupture termination of M2 (Fig. 5.1).
In the southernmost and northernmost end of the aftershock zone, the
preexisting faults also had a influence on aftershocks. At southern end, the
aftershocks were bounded by northeast striking Menglian fault; at northern end, the
aftershocks were bounded by W. Nantinghe fault.
Two additional points should be pointed out:
1. The rupture length of the Ml from Fl to F2 is about 60 km. The surface
fault rupture reported by Zhou et al. (Fig. 1.2), however, was in south segment of
the rupture zone. As mentioned earlier, the surface fault rupture caused by Ml is less
139
obvious and the main feature of ground deformation caused by Ml is a linear
distribution of ground fissures. However, the ground fissures which are presumably
caused by Ml, reported by Zhao et al. (1989), were distributed up to 23.2° latitude
(see Fig. 1.3).
2. The rupture zone of M2 between two groups of aftershocks in Fig. 5.2 is
about 25 km along the strike of M2. This length is longer than 15 km of surface
fault rupture reported by Zhou et al. (Fig. 1.2) but in good agreement with 30 km
ground fissures distributed observed in the area (see Fig. 1.3). The estimated seismic
moment of 1.9 X 1026 dyn cm and the fault width of 20 km (twice of the centroid
depth estimated in surface wave analysis) suggest an average slip of 1.13 m which is
consistent with 0.9 m maximum horizontal displacement on the surface fault rupture
reported by Zhao et al. (1989, unpublished data).
141
23.8
23.6
oQ
23.4
)X
'
F2
/
1
23.2
U/Ia p
23.0
0
0
0
Ml
22.8
0
Fl
22.6
22.4
99.2
99.4
99.6
99.8
100.0
Fig. 5.2 Aftershocks in 3 hours follow the occurrence of Ml.
100.2
142
6. CONCLUSIONS
Lancang-Gengma earthquake sequence has been studied using teleseismic
body waves, surface waves and strong ground motion data. The following
conclusions have been obtained:
1. The point source orientation, which was well constrained by teleseismic
long-period body wave inversion, indicates mainshock Ml is a pure right-lateral
strike-slip event on a nearly vertical striking 154° fault plane. The strike of the focal
plane is coincident with preexisting Lancang fault and the field investigations indicate
Ml occurred on Lancang fault. The ceniroid depth for this event is constrained to be
shallower than 15 km. The estimated best centroid depth is about 12 km and seismic
moment is 4.6 X 1026 dyn cm. The point source duration from broad-band inversion
is 12 s.
2. In order to study the source mechanism of M2, the moving window
analysis has been used to retrieve the Ri and R2 waves of Ml and M2. The Ri and
R2 waves can be well resolved from the complex seismograms within the period 60100 s. Because of uncertainties in attenuation and other path complexities, the
amplitude spectra of the Ri and R2 waves can only resolve the strike (155°) for M2.
This strike is consistent with observed surface fault rupture. On the other hand,
similar Rayleigh wave radiation patterns of Ml and M2 indicate that they had very
similar source mechanisms and centroid depths; the minor differences are statistically
indistinguishable. A seismic moment of 2.1 X 1026 dyn cm was estimated for this
event.
3. The source mechanism of the largest aftershock Al was also well
determined by teleseismic body wave inversion. It is also a strike slip event which
occurred on a vertical fault plane at shallow depth (between 4-12 km). However, it is
difficult to identify which nodal plane is the fault plane because this event occurred
near a geological fault confluence. Even though source orientation of A2, determined
from the teleseismic long-period body wave inversion, is almost the same as that of
Al, however, only strike of A2 was well resolved.
4. The strong ground acceleration particle motions indicate that scattering
caused by upper crustal heterogeneity has a strong effect on high frequencies, and
only low frequencies behave in a predictable fashion. Modeling of low frequency (<
1 Hz) acceleration, velocity and displacement seismograms required a low velocity
sedimentary layer. The modeling of Al and A2 indicates ceniroid depths of Al and
143
A2 greater than 4 km and less than 12 km. A point source duration of 7 s and 2 s,
seismic moment 2.2 X 1025 and 2.5 X 1024 dyn cm were obtained for Al and A2,
respectively.
5. The distinctive different high frequency content in P and S accelerations
has been observed. Comparing the acceleration spectra between P and S accelerations
for different time windows indicates that this phenomenon is not due to SV-to-P
conversions at free surface. However, further studies are needed to verify if this
phenomemon is due to medium attenuation or source effects.
6. The rupture processes of Ml and M2, established from the aftershock
distribution, field investigations and geological information suggest that the ruptures
of Ml and M2 were terminated by preexisting faults or fault complexities. The
initiation of M2 was the result of the asperities broken by the deformations which
were caused by the termination of the NNW propagating rupture of Ml and the stress
loading. Thus, the preexisting faults and their intersections play a key role in
Lancang-Gengma earthquake rupture process.
144
BIBLIOGRAPHY
Abe, K. (1981). Magnitudes of large shallow earthquakes from 1904 to 1980, Phys.
Earth Planet. Inter., 27,72-79.
Aki, K. (1967). Scaling law of seismic spectrum, J. Geophys. Res., 72, 1217-123 1.
Anderson, D. L. and J. W. Given (1982). Absorption band Q model for the earth, J.
Geophys. Res., 87, 3893-3904.
Bouchon, M. (1976). Teleseismic body wave radiation from a seismic source in a
layered medium, Geophys. J. R. Astr. Soc., 47, 5 15-530.
Bouchon, M. (1981). A simple method to calculate Green's functions for elastic
layered media, Bull. Seis. Soc. Am., 71, 959-971.
Bouchon, M. (1982). The rupture mechanism of the Coyote lake earthquake of 6
August 1979 inferred from near-field data, Bull. Seis. Soc. Am., 72,745-
757.
Capon, J. (1970). Analysis of Rayleigh-wave multipath propagation at LASA, Bull.
Seis. Soc. Am., 60, 1701-1731.
Capon, J. (1971). Comparison of Love- and Rayleigh-wave multipath propagation at
LASA, Bull. Seis. Soc. Am., 61, 1327-1344.
Chen, Y. and F. T. Wu (1989). Lancang-Gengma earthquake, a preliminary report
on the November 6, 1988, event and its aftershock, Trans. Am. Geophys.
Union, 49, 1527 & 1540.
Cranswick, E. and C. S. Mueller (1985). High-frequency vertical particle motion
in the Imperial Valley and S V-to-P coupling: The site-response of differential
array and the "greenhouse effect", Trans. Am. Geophys. Union, 66,
968.
145
Dziewonski, A., S. Bloch and M. Landisman (1969). A technique for the analysis of
transient seismic signals, Bull. Seis. Soc. Am., 59, 427-444.
Dziewonski, A., J. Mills and S. Bloch (1972). Residual dispersion measurement - A
new method of surface wave analysis, Bull. Seis. Soc. Am., 62, 129-139.
Dziewonski, A. and D. L. Anderson (1980). Preliminary reference Earth model,
Phys. Earth Planet. Inter., 25, 297-356.
Gariel, J.-C., R. J. Archuleta and M. Bouchon (1990). Rupture process of an
earthquake with kilometeric size fault inferred from the modeling of near-field
records, Bull. Seis. Soc. Am., 80, 870-888.
Hanks, T. C. and R. K. McGuire (1981). The character of high-frequency strong
ground motion, Bull. Seis. Soc. Am., 71, 2071-2095.
Hanks, T. C. (1982).
fmax,
Bull. Seis. Soc. Am., 72, 1867-1879.
Hartzell, S. and P. V. Heimberger (1982). Strong motion modeling of the Imperial
Valley earthquake of 1979, Bull. Seis. Soc. Am., 72(2), 571-596.
Harvey, D. and G. L. Choy (1982). Broad-band deconvolution of GDSN data,
Geophys. J. R. Astr. Soc., 69, 659-668.
Hauksson, E, T. Teng and T. L. Henyey (1987). Results from a 1500 m deep, three
level downhole seismometer array: site response, low Q values, and
Bull. Seis. Soc. Am., 77, 1883-1904.
f,
Jiang, K. and J. Zhang. (1989). Seismological background and tectonic in the
seismic region, in Corrected accelero grams and response spectra of
Lancang-Gengma earthquake, 13-16, edited by Y. Wang and J. Li,
Seismological Press, Beijing.
Kanamori, H. and G. S. Stewart (1976). Mode of the strain release along the Gibbs
fracture zone, mid-Atlantic ridge, Phys. Earth Planet. Inter., 11, 3 12-332.
146
King, G. and J. Nabelek (1985). Role of fault bends in the initiationand termination
of earthquake rupture, Science, 228, 984-987.
Landisman, M., A. Dziewonski and Y. Sato (1969). Recent improvements in the
analysis of surface wave observations, Geophys. J. R. asrr. Soc., 17, 369-
403.
Langston, C. A. (1981). Source inversion of seismic waveforms: The Koyna, India,
earthquake of 13 September 1967, Bull. Seis. Soc. Am., 71, 1-24.
Langston, C. A. and D. V. Helmberger (1975). A procedure for modeling shallow
dislocation sources, Geophys. J. R. astr. Soc., 42, 117-130.
Lee, W. H. K., F. T. Wu and S. C. Wang (1978). A catalog of instrumentally
determined earthquakes in China (magnitude 6) compiled from various
sources, Bull. Seis. Soc. Am., 68, 383-398.
Liu, H.-L. and D. V. Helmberger (1985). The 23:19 aftershock of the 15 October
1979 Imperial valley earthquake: More evidence for an asperity, Bull. Seis.
Soc. Am., 75, 689-708.
Mendiguren, J. A. (1977). Inversion of surface wave data in source mechanism
studies, J. Geophys. Res., 82, 889-894.
Mills, Jr. J. M. and A. L. Hales (1978). Great-circle Rayleigh wave attenuation and
group velocity, Part II: observations for period between 50 and 200 seconds
for nine great-circle paths and global average for periods of 50 to 600
seconds, Phys. Earth Planet. Inter., 17, 209-23 1.
Nabelek, J. L. (1984). Determination of earthquake source parameters from
inversion of body waves, Ph.D. thesis, Mass. Inst. of Technol., Cambridge.
147
Olson, A. H. and R. J. Aspel (1982). Finite faults and inverse theory with
applications to the 1979 Imperial Valley earthquake, Bull. Sets. Soc. Am.,
72(6), 1969-2001.
Romanowicz, B. A. and G. Suarez (1983) On an improved method to obtain the
moment tensor and depth of earthquakes from the amplitude spectrum of
Rayleigh waves, Bull. Sets. Soc. Am., 73, 1513-1526.
Papageorgiou, A. S. and K. Aki (1983). A specific barrier model for the quantitative
description of inhomogeneous faulting and the prediction of strong ground
motion. I. Description of the model, Bull. Seis. Soc. Am., 73, 693-722.
Tsai, Y.-B. and K. AId (1969). Simultaneous determination of the seismic moment
and attenuation of seismic surface waves, Bull. Sets. Soc. Am., 59, 275-
287.
Tsai, Y.-B. and K. Aki (1970). Precise focal depth determination from amplitude
spectra of surface waves, J. Geophys. Res., 75, 5729-5743.
Wang, K., Z. Yao, L. Gao and T. C. Wallace (1989). Source mechanism of the
1988 Lancang-Gengma, China, earthquake. Trans. Am. Geophys. Union,
70, p. 1218.
Zhao, X. and Z. Xang (1989). The preliminary report of the comprehensive survey
on Gengma-Lancang earthquake, Yunnan Province, China (in Chinese),
Seismology and Geology, 11, 135-136.
Zhang, J. and H. Kanamori (1988). Depth of large earthquakes determined from
long-period Rayleigh waves, .1. Geophys. Res., 93, 4850-4868.
Zhou, B., P. Huang and J. Cheng (1989). Intensity distribution and ground failure,
in Corrected accelerograms and response spectra of Lancang-Gengma
earthquake, 17-18. edited by Y. Wang and J. Li, Seismological Press,
Beijing.