GEOPHYSICAL RESEARCH LETTERS, VOL. 30, NO. 10, 1530, doi:10.1029/2002GL016211, 2003
Strong Lg wave attenuation in the Northern and Eastern Tibetan
Plateau measured by a two-station/two-event stacking method
Guang-Wei Fan and Thorne Lay
Center for the Study of Imaging and Dynamics of the Earth, Institute of Geophysics and Planetary Physics,
University of California Santa Cruz, USA
Received 2 September 2002; revised 19 December 2002; accepted 30 January 2003; published 27 May 2003.
[1] The regional seismic phase, Lg, which involves shear
wave reverberations in the crust, is strongly attenuated
along propagation paths in northern and eastern Tibet.
Robust estimates of the attenuation quality factor, QLg, in
these regions are obtained using a two-station, two-event
stacking method for broadband, vertical component Lg data
in the passband 0.2– 1.0 Hz. We find average 1 Hz QLg
values, Qo, of 79 ± 4 in northern central Tibet and 119 ± 17
in eastern Tibet for the 0.2 –0.5 Hz passband. Localized
regions of eastern Tibet have Qo estimates ranging from 66
to 121. These results confirm and augment the spatial
coverage of recent estimates of low Qo values in north
central and south central Tibet, and support the notion of
widespread partial melting of the crust throughout the
INDEX TERMS: 7205 Seismology: Continental crust
Plateau.
(1242); 7203 Seismology: Body wave propagation; 7218
Seismology: Lithosphere and upper mantle; 7219 Seismology:
Nuclear explosion seismology; 8102 Tectonophysics: Continental
contractional orogenic belts. Citation: Fan, G.-W., and T. Lay,
Strong Lg wave attenuation in the Northern and Eastern Tibetan
Plateau measured by a two-station/two-event stacking method,
Geophys. Res. Lett., 30(10), 1530, doi:10.1029/2002GL016211,
2003.
1.0 Hz, with a region in northern central Tibet having Qo =
85– 90. The attenuation is attributed to either strong smallscale scattering or partial melting in the tectonically
deformed crust. Confirming and mapping the regions of
strong Lg attenuation in Tibet is necessary for assessing its
tectonic significance.
2. Data
[4] We analyze broadband, vertical component recordings of Lg for moderate size earthquakes within Tibet
obtained from stations WMQ, LSA and KMI (Figure 1). In
Paper 1 a single-station/multiple event analysis was used
for WMQ observations, with attenuation estimates being
made for several corridors traversing Tibet. This paper
applies a more robust method using pairs of stations and
events along a corridor, yielding results for Profiles III and
IV defined in Paper 1 (Figure 1). The new results have
superior suppression of source and receiver effects, along
with better spatial resolution of Lg attenuation in eastern
Tibet.
1. Introduction
[2] The Tibetan Plateau is the largest and highest plateau
on Earth, and understanding its evolution plays a key role in
continental tectonics [e.g., Tapponnier et al., 2001]. With
extensive regions of the Plateau at elevations above 4000 m
(Figure 1), and crustal thickness of 65– 75 km, it is no
surprise that regional seismic phases have unusual propagation characteristics in the Plateau, with poor transmission of
S wave energy in the crust and upper mantle [e.g.,
McNamara et al., 1996]. One of the important regional
phases is Lg, which involves trapped post-critical S waves
propagating in the crustal waveguide. Throughout Eurasia,
Lg is the most stable phase observed at regional distances
[Rapine et al., 1997], with efficient transmission of
broadband Lg over large distances in much of the
continental crust.
[3] Early estimates of the attenuation quality factor for
Lg, QLg, in Tibet gave 1 Hz, Qo, values of 300 – 448 [e.g.,
McNamara et al., 1996]; higher than found in other
tectonically active regions. Recent work [Xie, 2002; Fan
and Lay, 2002] indicates much lower values. For example,
Fan and Lay [2002; hereinafter called Paper 1], estimate
Plateau-wide average Qo = 120 – 200 for the passband 0.2–
Copyright 2003 by the American Geophysical Union.
0094-8276/03/2002GL016211$05.00
37
Figure 1. Regional map of the Tibetan Plateau, the
locations of the three stations used in the attenuation
analysis (triangles), and the locations of event epicenters
used for analysis of WMQ and LSA data (plusses) and
WMQ and KMI data (circles). Profile IV and Profile III
labels correspond to similarly named regions in Fan and
Lay [2002].
- 1
37 - 2
FAN AND LAY: STRONG WAVE ATTENUATION IN TIBETAN PLATEAU MEASURED
larger distance in each pair. The distribution of earthquakes
in Tibet allows us to spatially isolate regions in northern and
eastern Tibet for this robust approach (Figure 1).
3. Determination of Lg Attenuation
[6] Our formal procedure is a two-station/two-event
method, as described by Chun et al. [1987], applied to Lg
signals for two events that lie within a few tens of degrees of
azimuth from the great-circle path connecting the two
receivers and the two epicenters. The procedure is
augmented by stacking of the spectral estimates for multiple
pairs of events. Following Chun et al. [1987], the ratio of Lg
wave spectral amplitude for event 2 (the distant event) to
that for event 1 (the nearer event) at station 1 (the first index
of d) is:
Að f ; d12 Þ=Að f ; d11 Þ ¼ ½S2 ð f Þ=S1 ð f Þ½R2 ð f ; q1 Þ=R1 ð f ; q1 Þ
½Gðd12 Þ=Gðd11 Þexp½g;
ð1Þ
while the ratio using the same event labeling at station 2 is:
Að f ; d21 Þ=Að f ; d22 Þ ¼ ½S1 ð f Þ=S2 ð f Þ½R1 ð f ; q2 Þ=R2 ð f ; q2 Þ
½Gðd21 Þ=Gðd22 Þexp½g:
Figure 2. Bandpass (1 –5 Hz) filtered seismograms for
events recorded by both WMQ and LSA. The upper panels
are for an event at the northern edge of Tibet, closer to
WMQ. Underlying brackets indicate the Lg group velocity
window. The amplitude ratio of energy in the Lg window
relative to the Pn window for each trace is shown on the
right (the onset of Pn is marked by the arrowheads). The
lower panel is for an event several hundred kilometers to
the south, in northern central Tibet.
[5] Figure 2 provides an example of observations used in
this study. High-pass filtered (>1 Hz) seismic waveforms are
shown for two events recorded at both WMQ and LSA. The
events are located at the northern and southern ends of Profile
IV in Figure 1. For the event near the northern end, WMQ
records clear high frequency Lg energy, while LSA does not.
The path to WMQ crosses the northern margin of Tibet and
the Tarim Basin, which has very high QLg typical of most
paths in China. Reciprocal behavior is seen for the event
near the southern end of Profile IV, with LSA recording
higher amplitude Lg than WMQ. High frequency Lg signal
is eliminated over a path length of 350 km or so between the
event pairs. This behavior is systematic, and allows us to
reliably estimate the Lg attenuation in the region between
the events, which are close to being on a great-circle path
(Figure 1). An approximate time domain attenuation value
at 1 Hz based on the factor of 25 variation in Lg/P ratio for
the upper event in Figure 2 suggests a Qo value near 100.
Using spectra for all four records simultaneously we can
explicitly cancel out both source and site effects, isolating
the inter-event attenuation effect on Lg, although we must
use frequencies lower than 1 Hz because the higher frequency energy is down to the noise level for the signal at the
ð2Þ
Here dij are the distances to the ith station from the jth event,
Sj( f ) are the source spectra, Rj( f, qi) are the source radiation
patterns at azimuths qi, G(dij) are the geometric spreading
1
U 1,
functions, g is the attenuation coefficient [g = p f QLg
with U being the group velocity at frequency f ]. is the
Figure 3. Lg attenuation coefficient estimates, with
standard deviations, from stacking of two-station, two-event
combinations for WMQ-LSA and WMQ-KMI recordings.
The text identifies the regions and the regression results are
summarized in Tables 1 and 2.
FAN AND LAY: STRONG WAVE ATTENUATION IN TIBETAN PLATEAU MEASURED
Table 1. Estimates of Lg Attenuation in Northern Central Tibet
0.2 – 0.5 Hz
0.2 – 1.0 Hz
Qo
h
Qo
h
Profile IV
(Paper 1)
Profile IV
WMQ-LSA
Velocity
Profile IV
WMQ-LSA
Displacement
90 ± 20
0.15 ± 0.10
85 ± 2
0.10 ± 0.04
79 ± 4
0.13 ± 0.04
94 ± 3
0.29 ± 0.04
74 ± 4
0.08 ± 0.05
92 ± 3
0.28 ± 0.05
distance between the two events, = d12 d11 or d21 d22.
The instrument response cancels out in each ratio, and any
station site response term is assumed to cancel out for
small differences in event backazimuth. For the geometrical spreading function G(d), we adopted G(d) = d0.5, as
appropriate for Lg waves analyzed in the frequency
domain.
[7] To solve for the attenuation coefficient, we eliminate
the source excitation and radiation pattern terms (assuming
quadrapolar or isotropic radiation) by multiplying the two
spectral ratios (1) and (2):
½ Að f ; d12 Þ=Að f ; d11 Þ½ Að f ; d21 Þ=Að f ; d22 Þ½d12 d21 =d11 d22 0:5
¼ exp½2g;
ð3Þ
thus, we can solve (3) for the attenuation coefficient g(f ). In
practice, we allow backazimuth variations to deviate by up
to 30, as Lg is found to have nearly isotropic radiation in
many studies. A stacked average spectral ratio is calculated
for all data combinations in a corridor to estimate g( f ), and
assuming a group velocity (3.5 km/s), we estimate QLg( f ).
[8] The long pathlengths and strong attenuation in the
region constrain our spectra to frequencies of 0.2 –1.0 Hz,
with the most reliable band being 0.2 – 0.5 Hz. Stacking
reduces error incurred by imprecise cancellation of source
and receiver terms due to non great-circle or non-point
source effects. The method is more robust than the twostation method used by Xie [2002] or the two-event method
used in Paper 1, and it can thus be applied to smaller
populations of events.
[9] Assuming a power-law frequency dependent model,
Lg attenuation can be written in terms of quality factor QLg
as QLg ( f z) = Q0 f h, where Q0 is the value of QLg at 1 Hz,
and h is the power-law frequency dependence. Using the
stacked estimates from (3), we fit QLg( f ) models by leastsquares linear regression in several frequency bands, 0.2–
0.5 Hz (or 0.35– 0.6 Hz), 0.2– 1.0 Hz and 0.35– 1.0 Hz, to
estimate the value of Lg attenuation in Profiles III and IV of
Figure 1.
[10] Figure 3 shows our stacked spectra, g( f ), for various
subregions. The spectral ratios are only shown out to
frequencies of 1 Hz, given the low signal-to-noise ratio at
higher frequencies. The numerical values of the best-fit
model parameters Q0 and h are listed in Tables 1 and 2,
37 - 3
along with values found in Paper 1. Previous studies have
shown that h is sensitive to lateral heterogeneity in the crust,
thus reliable h estimates are difficult to obtain, particularly
when only relatively narrow bandwidth is available, so we
focus on the quality factor Q0.
[11] The most stable result is for Profile IV, based on data
from WMQ and LSA. The corresponding spectra in Figure 3
are averaged over 12 estimates (4 events in the north, 3 in
the south of the profile), with small variance and a smooth
linear variation. Processing of either velocity or displacement spectra give very comparable results (Table 1).
[12] We constructed five attenuation estimates for regions
of eastern Tibet using data from WMQ and KMI, ranging
from whole Plateau averages to more localized subregions.
The number of events used for each profile is 5 4 for the
entire region, and 3 2, 2 2, 3 2, and 2 2 for subregions in western (W), eastern (E), northwestern (NW) and
southeastern (SE) portions of eastern Tibet, respectively,
where the first number represents the number of events at
the northern end of each profile and the second is the
number of events at the southern end. Given that two
stations and two events are used for each estimate, the total
number of distinct spectral ratios is 24 in north central Tibet
and 60 in eastern Tibet. Subdivision of eastern Tibet into
smaller areas is viable due to the robust nature of the twostation/two-event method and the stability of the stacked
spectra for periods of 0.2– 0.5 Hz (Figure 3). At least 8
spectral ratios are used in the subregions.
[13] The Lg attenuation coefficients for each WMQ-KMI
profile show flattening at frequencies above about 0.7 Hz.
This results in significant dependence of estimated QLg
parameters on the frequency band used for the regression
analysis: as higher frequencies are included we obtain
higher QLg estimates. Regressions using the 0.2 – 0.5 Hz (or
0.35– 0.6 Hz) band yield Qo values near 110, while for the
0.2– 1.0 Hz band Qo values are close to 180 (Table 2).
These are very compatible with the single station estimates
from WMQ. We believe the high frequency flattening of the
spectra is caused by contamination of the Lg window by
scattered high frequency P energy, and thus prefer the
results constrained to frequencies lower than 0.6 Hz. For the
relatively small data sets in the (W) and (SE) subregions
there is a steep drop-off of low frequency values, which
appears to also be a signal-to-noise issue; these regions give
estimates with large uncertainties, but the spectra in the
0.2– 0.5 Hz band are consistent with the regional average.
4. Discussion and Conclusions
[14] The Lg attenuation estimates in Tables 1 and 2
indicate very strong attenuation in the crust of northern and
eastern Tibet. Our results for northern central Tibet confirm
the very low Qo reported in Paper 1, and are summarized in
Figure 4. Recently, Xie [2002] obtained a QLg model with
Table 2. Estimates of Lg Attenuation in Eastern Tibet
0.2 – 0.5 Hz
*(0.35 – 0.6 Hz)
0.2 – 1.0 Hz
Qo
h
Qo
h
Profile III
(WMQ: Paper 1)
Profile III
(WMQ-KMI)
Profile (W)
(WMQ-KMI)
Profile(NW)
(WMQ-KMI)*
Profile(SE)
(WMQ-KMI)*
Profile (E)
(WMQ-KMI)
122 ± 20
0.19 ± 0.15
195 ± 14
0.24 ± 0.08
119 ± 17
0.20 ± 0.13
183 ± 13
0.22 ± 0.09
100 ± 20
0.60 ± 0.17
180 ± 15
0.03 ± 0.11
110 ± 10
0.14 ± 0.11
-
66 ± 47
1.95 ± 0.69
-
121 ± 18
0.03 ± 0.13
178 ± 12
0.41 ± 0.08
37 - 4
FAN AND LAY: STRONG WAVE ATTENUATION IN TIBETAN PLATEAU MEASURED
has inefficient Sn propagation (Figure 4), low Pn velocity,
and high Poisson’s ratios of 0.34– 0.35 over a 30 km
thickness [Owens and Zandt, 1997]. Rodgers and Schwartz
[1998] find very low Qs values of 44– 89 in the Qiangtang
Terrane, along with high Poisson’s ratio, which they
attribute to partial melting of the crust. There is evidence
for partial melt and crustal low-velocity zones existing north
of the Tsangpo suture in southern Tibet [Nelson et al.,
1996]. Overall it seems likely that partial melt in the thick
Tibetan crust plays a significant role in regional Lg
attenuation.
Figure 4. Map of Tibet highlighting regions where 1 Hz
Lg attenuation coefficient has been estimated by twosource/two-station spectral stacking analysis. There is a
broad region of very low Qo in northern Tibet with the
lowest values in the central region. The average values of
Qo for eastern Tibet is about 110. The shaded area with
dashed line corresponds to the region of inefficient Sn
propagation in northern Tibet [McNamara et al., 1996].
Q0 = 126 ± 9 for the 0.2– 3.6 Hz band, in the region from
LSA up into the center of our Profile III. His result is very
consistent with our estimates for the 0.2– 0.5 Hz band in
eastern Tibet, where the paths overlap. Xie [personnal
communication, 2002] analyzed INDEPTH II, III data,
finding that areas in southern Tibet west of LSA have Qo
values of 70– 90, as low as we find in northern Central
Tibet.
[15] Our low Qo values of 79– 121 are comparable to Lg
attenuation values found in other tectonically active areas,
such as the Bolivian Altiplano [Baumont et al., 1999]. The
similarity of Lg attenuation values in the Altiplano and in
Tibet may suggest a common effect of thickened, deformed
crust within major continental plateaus behind active
mountain belts. Our very sparse station coverage precludes
an attempt to apportion Lg attenuation estimates into
intrinsic anelasticity versus small-scale scattering attenuation, as was attempted for the Altiplano.
[16] In general, mechanisms of intrinsic shear wave
attenuation are sensitive to temperature conditions, and
the very low Qo of 79– 94 may be associated with partial
melting of the crust in northern Tibet. The region of
northern central Tibet is the most volcanically active area of
Tibet [e.g., Turner et al., 1996]. Owens and Zandt [1997]
presented evidence for a lower crust low-velocity zone
likely to involve partial melt in northern Tibet. This region
[17] Acknowledgments. J. Xie and S. Phillips provided helpful
discussion, and J. Xie and an anonymous reviewer made constructive
comments on the manuscript. This research was supported by the Defense
Threat Reduction Agency through contract DTRA01-00-C-0211. CSIDE
contribution #458, IGPP, University of California, Santa Cruz.
References
Baumont, D., A. Paul, S. Beck, and G. Zandt, Strong crustal heterogeneity
in the Bolivian Altiplano as suggested by attenuation of Lg waves,
J. Geophys. Res., 104, 20,287 – 20,305, 1999.
Chun, K.-Y., G. F. West, R. J. Kokoski, and C. Samson, A novel technique
for measuring Lg attenuation – Results from Eastern Canada between 1 to
10 Hz, Bull. Seism. Soc. Am., 77, 398 – 419, 1987.
Fan, G.-W., and T. Lay, Characteristics of Lg Attenuation in the Tibetan
Plateau, J. Geophys. Res., 107(B10), 2256, doi:10.1029/2001JB000804,
2002.
McNamara, D. E., T. J. Owens, and W. R. Walter, Propagation characteristics of Lg across the Tibetan plateau, Bull. Seism. Soc. Am., 86, 457 –
469, 1996.
Nelson, K. D., et al., Partially molten middle crust beneath southern Tibet:
Synthesis of project INDEPTH results, Science, 274, 1684 – 1688, 1996.
Owens, T. J., and G. Zandt, Implications of crustal property variations for
models of Tibetan plateau evolution, Nature, 387, 37 – 43, 1997.
Rapine, R. R., J. F. Ni, and T. M. Hearn, Regional wave propagation in
China and its surrounding regions, Bull. Seism. Soc. Am., 87, 1622 –
1636, 1997.
Rodgers, A. J., and S. Y. Schwartz, Lithospheric structure of the Qiangtang
Terrane, northern Tibetan Plateau, from complete regional waveform
modeling: Evidence for partial melt, J. Geophys. Res., 103, 7137 –
7152, 1998.
Tapponnier, P., X. Zhiqin, F. Roger, B. Meyer, N. Arnaud, G. Wittliner, and
Y. Jingsui, Oblique stepwise rise and growth of the Tibet Plateau,
Science, 294, 1671 – 1677, 2001.
Turner, S., N. Arnaud, J. Liu, N. Rogers, C. Hawkesworth, N. Harris,
S. Kelley, P. van Calsteren, and W. Deng, Post-collision, shoshonitic
volcanism on the Tibetan Plateau: Implications for convective thinning
of the lithosphere and the source of ocean island basalts, Petrology, J., 37,
45 – 71, 1996.
Xie, J., Lg Q in the Eastern Tibetan Plateau, Bull. Seism. Soc. Am., 92,
871 – 876, 2002.
G.-W. Fan and T. Lay, Center for the Study of Imaging and Dynamics of
the Earth, Institute of Geophysics and Planetary Physics, University of
California Santa Cruz, USA. ([email protected])