JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 107, NO. B10, 2256, doi:10.1029/2001JB000804, 2002
Characteristics of Lg attenuation in the Tibetan Plateau
Guangwei Fan and Thorne Lay1
Center for the Study of Imaging and Dynamics of the Earth, Institute of Geophysics and Planetary Physics, University of
California, Santa Cruz, California, USA
Received 23 July 2001; revised 5 March 2002; accepted 2 May 2002; published 25 October 2002.
[1] Lg, a regional seismic wave comprised of multiple shear wave reverberations trapped
in the crustal waveguide, is important for magnitude estimation and source
discrimination for monitoring nuclear testing treaties. In stable continental regions, Lg
propagates with a group velocity of about 3.5 km/s and can often be observed at distances
up to 4000 km. To better understand the absence of high-frequency Lg arrivals for paths
traversing the northern boundary of the Tibetan Plateau, we investigate spatial variations of
broadband (0.15–5.0 Hz) energy in the Lg group velocity window (3.6–3.0 km/s) using
regional waveforms recorded at the Chinese Digital Seismic Network station WMQ.
Vertical component seismograms are analyzed for 90 events with magnitudes of 4.4 mb
6.4 that occurred between 1987 and 1999 in the Tibetan Plateau and around its margins.
The Lg amplitude spectra for events located near the northern margin of the plateau have
apparent corner frequencies of 1–2 Hz, nearly identical to those for comparable size events
at similar distances outside the plateau. High-frequency (>1 Hz) Lg energy recorded at
WMQ decreases rapidly as a function of source distance into the plateau. A path length of
300–400 km within the northern Plateau suffices to eliminate 2–5 Hz Lg energy. For
events in southern Tibet with paths crossing the central portion of the Tibetan Plateau,
almost total Lg extinction occurs, even for energy in the low-frequency band of 0.2–1 Hz.
Corresponding apparent corner frequencies of the ‘‘Lg’’ amplitude spectra range between
0.2 and 0.4 Hz. The corner frequency shift is found to vary systematically with path length
across the plateau. Linear regressions demonstrate that the shift in apparent corner
frequency of Lg amplitude spectra is negatively correlated with features of the Tibetan
Plateau, such as mean elevation along the paths or travel distance within the plateau
above specified elevation thresholds. The systematic variations in the amplitude and
frequency content of energy in the Lg window as a function of path length within the
plateau indicate that strong crustal attenuation plays an important role in Lg extinction
for paths traversing central and northern Tibet, superimposed on any structural blockage
effects associated with abrupt thinning of the crust near the northern boundary of the
plateau. Spectral ratios of many event pairs along great circle paths give estimates of
frequency-dependent Lg attenuation for paths crossing western, central, and eastern
sectors of Tibet. The region of strong Sn attenuation in northern central Tibet also has
strong Lg attenuation with QLg for 1 Hz on the order of Qo = 80–90, while in southern
central Tibet, Qo increases to about 316, and in eastern and western Tibet, Qo is on the
order of 120–200 for paths traversing the entire plateau. The strong Lg attenuation in
northern central Tibet is responsible for the so-called Lg blockage and may be associated
with partial melting in the crustal low-velocity layer in northern Tibet. Our QLg values for
Tibet are significantly lower than most earlier estimates, primarily as a result of not
excluding blocked observations along with allowing for lateral variations within
INDEX TERMS: 7203 Seismology: Body wave propagation; 7255 Seismology: Surface waves and
Tibet.
free oscillations; 7205 Seismology: Continental crust (1242); KEYWORDS: Lg, Tibetan Plateau, nuclear event
discrimination, crustal attenuation, Lg blockage, crustal partial melting
Citation: Fan, G., and T. Lay, Characteristics of Lg attenuation in the Tibetan Plateau, J. Geophys. Res., 107(B10), 2256,
doi:10.1029/2001JB000804, 2002.
1
Also at Earth Sciences Department, Institute of Tectonics, University
of California, Santa Cruz, California, USA.
Copyright 2002 by the American Geophysical Union.
0148-0227/02/2001JB000804$09.00
ESE
1. Introduction
[2] Lg is a regional seismic phase commonly observed in
continental areas. It is variously described as a superposition
of many higher mode surface waves or trapped postcritical S
14 - 1
ESE
14 - 2
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
waves, with most of its energy being confined to the crustal
waveguide [e.g., Xie and Lay, 1994]. When observed on
vertical component seismograms, Lg can be primarily
associated with Rayleigh wave overtone modes, although
scattering may mix Love and Rayleigh energy. Lg is rather
insensitive to earthquake radiation patterns, thus it has
particular value for seismic magnitude estimation. Because
Lg is dominated by shear wave energy, it tends to be more
strongly excited by earthquakes than by explosions. Thus,
various P/Lg amplitude measures have been developed for
source identification applications, with good success for
frequencies above 2 – 3 Hz [e.g., Walter et al., 1995; Taylor,
1996; Hartse et al., 1997]. Lg propagation is affected by
laterally heterogeneous crustal thickness and velocity structure [e.g., Kennett, 1989; Bostock and Kennett, 1990; Zhang
et al., 1994] and Lg does not propagate efficiently in thin
oceanic crust [e.g., Press and Ewing, 1952; Zhang and Lay,
1995], but it is otherwise extensively observed. In Eurasia,
Lg is the most stable phase observed at regional distances
[Sereno et al., 1990; Baumgardt, 1990; Rapine et al., 1997],
with efficient broadband Lg transmission over large distances in much of the continental crust.
[3 ] Strong attenuation or blockage of Lg has been
observed in some continental areas where significant
changes in crustal thickness are located, such as the mountain belts in central Asia [Ruzaikin et al., 1977], the Tibetan
Plateau [Båth, 1954; Ni and Barazangi, 1983; McNamara et
al., 1996; Rapine et al., 1997], in the Bolivian Altiplano
[Baumont et al., 1999] and near the Alpine mountains
[Campillo et al., 1993]. It is believed that scattering of Lg
energy from fractures within the shallow crust may be a
major cause of strong Lg attenuation in tectonically active
regions, while in regions such as Tibet partial melt in the
crust may also play an important role. It is also possible that
the lack of a continuous waveguide in areas with significant
variations of crustal thickness and sedimentary basin thickness may result in disruption of Lg [Baumgardt, 1990].
Understanding the causes of inefficient Lg propagation in
various settings is very important for event discrimination,
particularly in the context of nuclear test monitoring efforts;
anomalously low Lg amplitudes may cause earthquake
signals to resemble those for explosions.
[4] Of particular interest for this study are observations
that Lg is extremely weak or totally extinguished for paths
traversing the northern margins of the Tibetan Plateau.
Ruzaikin et al. [1977] found that a path length of 200 km
within the plateau is sufficient to attenuate 1 Hz Lg signals
down to noise levels for observing stations located north of
the plateau. Those authors suggested that either strong
attenuation or scattering by a thinning crustal waveguide
was responsible, but their narrow-band data did not resolve a
progressive shift of Lg spectral content to lower frequencies
with increasing path length in Tibet, as would be expected if
attenuation is primarily responsible. They did note that very
low QLg values of 20– 40 would be required to attenuate 1
Hz Lg energy by a factor of 100 over 100 –200 km path
lengths, as their qualitative observations suggested. Qualitative analysis of Lg propagation efficiency relative to P
wave coda for stations west and south of Tibet indicated that
Lg signals are strongly attenuated within central and northern Tibet while the phase does propagate along the Himalayan mountain belt [Ni and Barazangi, 1983]. This
inefficient Lg propagation was again attributed to either
strong crustal attenuation within Tibet or to waveguide
disruption on the southern margin. Rapine et al. [1997]
found inefficient Lg propagation across most of Tibet, with
no Lg from western Himalayan events reaching either station
WMQ to the north of the Tarim Basin or station LSA within
Tibet (LSA did record Lg on some paths from events in the
eastern Himalayas and eastern Tibet, suggesting internal
variations in propagation efficiency within the plateau).
[5] McNamara et al. [1996] analyzed data recorded at 11
broadband stations in eastern central Tibet, finding that highfrequency Lg is generated by events within the plateau, and
can propagate to distances of at least 600 km; however, the
phase was not observed for paths traversing western or
central Tibet or for events in the Himalayas. For eastern
Tibet, where sufficient Lg signal-to-noise ratios were
observed, a frequency-dependent QLg function given by
QLg( f ) = (366 ± 37) f (0.45±0.06) was estimated for the passband (0.5 –16 Hz) [McNamara et al., 1996]. This is comparable to the estimate of QLg( f ) = (448 ± 82) f (0.426±0.157)
estimated for 1 – 6 s Lg waves traversing Tibet by Shih et al.
[1994]. This latter value is likely also dominated by eastern
Tibet paths, as ‘‘blocked’’ observations were omitted from
the calculation. QLg value at 1 Hz was also estimated to be
about 340 for events north of LSA using spectral methods
by assuming a constant QLg model [Reese et al., 1999].
These attenuation values are typical of tectonically active
areas, and are not low enough to cause rapid extinction of 1
Hz Lg over just a few hundred kilometers, as observed for
paths traversing the northern margin of Tibet. One could
perhaps infer that waveguide disruption must cause extinction of Lg in the latter regions; however, this is not a secure
interpretation because QLg may vary laterally within Tibet.
Indeed one might expect this to be the case, for it is true of Sn
attenuation, which is strongest in northern central Tibet. The
possibility of stronger Lg attenuation in northern and central
Tibet is suggested by the lack of high-frequency Lg detections for events to the west of the broadband stations in the
data set of McNamara et al. [1996].
[6] Although the failure of Lg to propagate on oceanic
paths has been relatively well explained by numerical modeling [Kennett, 1986; Cao and Muirhead, 1993; Zhang and
Lay, 1995], the physical nature of Lg blockage for continental
paths is still debated. For paths traversing the northern margin
of Tibet the question is whether a step in crustal thickness on
the plateau margins plays a significant role or whether other
properties of the crust are important. It is probable that crustal
attenuation plays some role as moderate necking of the crust
alone is not likely to impede high-frequency Lg propagation
to the extent observed [Regan and Harkrider, 1989; Gibson
and Campillo, 1994]. We analyze broadband, digital seismic
data to characterize general features of Lg wave propagation
across the northern Tibetan Plateau margin and to explore the
relative roles of Lg attenuation and crustal disruption.
Assuming a frequency-dependent QLg model, we estimate
the values of the Lg attenuation for different parts of Tibet to
characterize attenuation heterogeneity in Tibet.
2. Data Analysis
[7] Our main data are regional waveforms recorded at
broadband station WMQ, the westernmost station of the
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
ESE
14 - 3
Figure 1. Topography in western China and its vicinity. The location of station WMQ is shown. The
squares, circles, triangles, and inverted triangles indicate the epicenters of the events used in this study,
keyed to other figures. Squares with origin dates indicate events for which spectra are shown. The events
within Tibet are grouped into eastern (inverted triangles), central (circles) and western (triangles) groups.
Chinese Digital Seismic Network. This station lies just
north of the Tarim Basin, about 650 km from the northern
boundary of the Tibetan Plateau (Figure 1). While threecomponent data were obtained, we restrict our analysis to
the vertical component seismograms due to the attenuated
nature of the Lg phases traversing Tibet, which leads to
confusion with fundamental mode Love wave energy on
the transverse components. We examine energy in the
standard Lg group velocity window of 3.6 to 3.0 km/s,
which precedes short and intermediate period fundamental
mode Rayleigh wave energy. We analyze recordings for 90
earthquakes with magnitudes of 4.4 mb 6.4 that
occurred between 1987 and 1999 in the Tibetan Plateau
and around its margins. A few events with paths well
removed from Tibet (Figure 1) were also analyzed.
[8] The recorded seismograms show great variability in
broadband waveforms, and our analysis is constrained to
relative measures because we use only one station and
cannot fully account for source effects. For events prior to
1998 the source parameters are taken from the International Seismological Center (ISC) bulletin. For more
recent events the source parameters are from the USGS
Preliminary Determination of Epicenters (PDE) catalog.
Only events with catalog focal depths less than 50 km are
included. While catalog source depths in this region are
subject to tens of kilometers of uncertainty, this criterion
probably ensures that all events analyzed are crustal
events, given that the crustal thickness in the central
Tibetan Plateau is about 65 to 75 km [e.g., Hirn et al.,
1984; Molnar, 1988; McNamara et al., 1995; Rodgers
and Schwartz, 1998; Zhao et al., 2001]. All selected
events have signal-to-noise ratios greater than 2 for broadband measurements of the Pn signal and pre-Pn noise. A
list of the events used is available from the authors upon
request.
[9] Figure 2 demonstrates the predominant feature of
high-frequency Lg propagation across the northern Tibetan
Plateau boundary first documented by Ruzaikin et al.
[1977]. The records shown are for a roughly northwest
to southeast profile of events, with the signals filtered in
the bandpass 1.0 –5.0 Hz. The first three records are for
events located near the 4000 m topographic contour along
the northern boundary of the plateau. These show typical
earthquake-like high-frequency signals, with predominant
Lg energy that is much stronger than the Pn arrivals. The
largest arrivals in the Lg window (brackets) tend to have
group velocities near 3.5 km/s. The lower two records are
from events 250– 400 km further to the southeast, and the
maximum relative amplitude ratio of Lg to P energy is
about a factor of ten lower than for the northwestern
events, giving these signals explosion-like character.
While P wave amplitudes tend to vary more than Lg
due to focal mechanism effects, this relative behavior is
observed for all waveforms. The distribution of events in
Tibet (Figure 1) presents a challenge in that there is not a
continuous distribution of events from the plateau margin
to the interior; this makes it difficult to separate possible
waveguide disruption effects from strong crustal attenuation effects as causes for the high-frequency Lg energy
blockage. However; we can consider behavior of Lg at
lower frequency on longer profiles to assess the role of
attenuation in the plateau, as this should be manifested in
a progressive shift of the frequency content of Lg with
propagation length across the plateau as suggested by
Molnar (in a personal opinion expressed by Ruzaikin et
al. [1977]).
14 - 4
ESE
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Dist=632 km
BAZ=158.9
5000
specific energy, but a group velocity of 3.5 km/s is a
reasonable value to use for the larger arrivals early in the
Lg signal.
[11] We lack detailed knowledge of the source process
and radiation effects for our events, and the station coverage
is not adequate to enable empirical separation of source and
propagation effects. However, our broadband digital data
indicate that propagation effects dominate in shaping the Lg
window. We examine windowed and tapered velocity
amplitude spectra with no source corrections, measuring
the apparent corner frequency of the velocity amplitude
4.49
0
-5000
940907
-10000
30000
Dist=701 km
BAZ=154.3
2.75
0
-30000
870225
80000
Dist=759 km
BAZ=150.2
3.68
Dist=632 km
(940907)
0
-80000
1000
900114
(1/3 x Amp.)
Dist=1006 km
BAZ=154.4
500
0.42
Dist=701 km
(870225)
0
890513b
-500
300
Dist=983 km
(980113)
Dist=1116 km
BAZ=159.5
0.59
Dist=1116 km
(881125)
0
881125
-300
0
100
200
TIME (s)
300
Dist=1339 km
(900602)
400
Dist=1415 km
(940723)
Figure 2. Demonstration of the rapid decrease in Lg
energy near the northern Tibetan Plateau boundary. Events
are identified with numbers indicating year/month/day.
WMQ records are shown for a short NW – SE profile of
events, with bandpass filters isolating the 1.0– 5.0 Hz
passband, for events near the northern boundary (top 3
records) and a few hundred kilometers into the plateau
(lower 2 records). The Lg group velocity window (3.6 – 3.0
km/s) is indicated by the bracket below each seismogram.
Small arrows represent the Pn arrivals after the onset of
visible signal in each trace. There is about a factor of 10
decrease in the Lg/Pn amplitude ratio (shown by the
numbers on upper right of each trace) common to all events
several hundred kilometers south of the plateau boundary.
Dist=1526 km
(980825)
Dist=1644 km
(880125)
Dist=1781 km
(970131)
Dist=1839 km
(980926)
0
[10] Figure 3 shows a longer profile of records, retaining
the full broadband energy with amplitudes normalized
relative to the 20 s Rayleigh wave energy in each signal.
The dramatic shift of frequency content in the Lg group
velocity window over the first few hundred kilometers is
again apparent, but now it is clear that there is also a
progressive shift of frequency content and a decrease of
relative amplitudes in the Lg window for lower frequencies
as path length across the plateau increases. This suggests
that while high-frequency Lg energy is eliminated over
short path segments in the plateau, lower frequency energy
survives much longer and can be used to constrain any
progressive path effect. The rapid amplitude decrease with
distance and sparse sampling make it difficult to track
200
400
TIME (s)
600
800
Figure 3. Broadband recordings at WMQ for events along
a profile traversing the entire central portion of the Tibetan
Plateau. The records have been amplitude-normalized
relative to the 20 s Rayleigh wave energy in each signal
(note the amplitude reduction for the first record) to help
display the rapid decrease of high frequency energy in the
Lg group velocity window (brackets). Small arrows
represent the first P arrivals after the onset of visible signal
in each trace. There is a steady shift of the spectral content
and relative amplitude of energy in the Lg window as path
length within the plateau increases. The two most distant
recordings show increased P amplitudes due to upper
mantle triplications.
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
900114 Event
107
106
Amplitude
105
104
103
102
101
Lg Wave
Pre-Pn Noise
100
10-1
10-2
a)
10-1
100
Frequency (Hz)
101
920730 Event
106
105
Amplitude
104
103
102
101
Lg Wave
100
Pre-Pn Noise
10-1
b)
10-2
10-1
100
Frequency (Hz)
101
Figure 4. Amplitude spectra for Lg signal and pre-Pn
noise recorded at seismic station WMQ for: (a) an event
located near the northern margin of the plateau, and (b) for
an event located in southern Tibet. The event locations are
indicated in Figure 1. The event magnitudes are comparable, so most of the tremendous variation in Lg spectral
content is due to propagation effects.
spectra for each event. The observed apparent corner
frequency is defined as the frequency where the amplitude
spectra start to fall-off. We call this corner frequency, which
is not to be confused with the source radiation effect, and
use its values below to infer relative attenuation of the Lg
energy.
[12] To explore whether the Lg group velocity significantly affects the spectral corner frequency estimate, we
considered several group velocity windows (3.6 to 2.8, 3.8
to 2.8, 3.6 to 3.0 and 3.8 to 3.0 km/s). The corner frequency
measurements are relatively insensitive to the value of the
larger group velocity, but the choice of lower group velocity
affects the low-frequency part of the amplitude spectra. This
is because fundamental Rayleigh wave energy enters the
window as the lower Lg group velocity decreases from 3.0
km/s to 2.8 km/s. For events in southern Tibet, the Lg wave
defined by a group velocity window of 3.6 to 3.0 km/s is
well isolated, albeit almost totally attenuated (Figure 3), so
we hold this group velocity window fixed.
[13] We compare the Lg amplitude spectra with pre-Pn
noise as a guide to signal bandwidth. One can also consider
the pre-Lg signal strength, but our purpose is not to define a
threshold for defining the existence of an Lg phase, only to
characterize the overall spectral evolution of the Lg window
ESE
14 - 5
with distance. The noise spectra are computed from 10 –15 s
time windows preceding the Pn arrival. Figure 4 shows some
comparisons of the Lg and background noise spectra. Our
data generally have good low-frequency signal-to-noise
ratios, but the spectral character is strongly dependent on
source location. Figure 4a shows the velocity spectra for a
representative event located near the northern margin of the
plateau. The signal-to-noise ratio is high across the entire
frequency band, and the Lg amplitudes are much higher than
noise amplitudes at frequencies up to 6 Hz, where the
instrument limitations flatten the spectra. The corner frequency is not very well defined for this spectrum, but fitting
of asymptotes to the 0.1 – 0.7 Hz band and the 1 – 4 Hz band
gives a corner frequency near 0.8 to 1.0 Hz. Around 4 – 5 Hz
the spectral roll-off steepens further. For events near the
southern margin of the plateau, the WMQ spectra for Lg are
above background noise for frequencies below 1 Hz, but
there is no high-frequency Lg signal even relative to the
background noise level (see Figure 4b). A comparison with
spectra for the pre-Lg window (Sn coda window) indicates
that all energy above 0.5– 1.0 Hz in this window may be
scattered Sn. This limits the range used for QLg estimation. A
corner frequency of around 0.18– 0.22 Hz can be defined.
There is clearly a huge difference in spectral content that
greatly exceeds any plausible effect of the source radiation
for these events. Lacking detailed source corrections, we will
seek systematic behavior in our large data population to
effectively average out minor source effects.
3. Distance Dependence of Lg Spectra in Tibet
[14] For most events located near the northern margin of
the Tibetan Plateau, the WMQ Lg velocity spectra have a
corner frequency of 1 – 2 Hz. The windowed and tapered Lg
energy decays fairly slowly at frequencies from 2 – 5 Hz and
more rapidly after 5 Hz, as shown in Figure 4a. Many
associated paths traverse either the central or eastern regions
of the Tarim Basin, and thus these Lg signals may still be
anomalous relative to Lg on other paths in Eurasia. To
assess this, we compare the amplitude spectra for an event
on the Tibetan Plateau margin with an event located directly
west from WMQ. The two events have the same magnitude,
similar seismic moments, similar Harvard Centroid Moment
Tensor depths and focal mechanisms, and similar propagation distances (Figure 5a). The waveforms and spectra are
quite similar, with corner frequencies in the vicinity of 1 Hz
and steepening of the high-frequency spectra near 5 Hz due
to instrument and windowing effects. Comparisons with
other distance- and magnitude-matched pairs indicate that
there is nothing anomalous about the Lg signals from events
on the northern Tibetan Plateau margin relative to events
elsewhere in central Eurasia. This indicates that Lg propagation across the eastern Tarim Basin is similar to that
elsewhere in the China platform, consistent with results
from Lg attenuation tomography by Phillips et al. [2000]. It
also appears that there is no abrupt effect on Lg associated
with the rapid increase in topography and presumably rapid
crustal thickening along the northern margin of the plateau,
as similar waveforms and spectra are observed for events
located beneath topography ranging from 2000 to 4000 m.
This is consistent with prior observations. Numerical simulations of Lg wave propagation indicate that a stepwise
14 - 6
ESE
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
a.
b.
931230
50000
10000
0
0
-50000
-5000
0
100
200
300
400
Time (sec)
mb = 5.7
Mo = 1.87E+17 N m
Dist. = 715 km
-10000
0
200
400
600
Time (sec)
106
mb = 5.5
Mo = 8.27E+16 N m
Dist. = 1685 km
106
105
105
104
Amplitude
Amplitude
880103
5000
103
102
101
931230
870225
100
10-1
10-2
104
103
102
101
880103
880125
100
10-1
100
101
10-1
10-2
Frequency (Hz)
5000
870225
100000
10-1
100
101
Frequency (Hz)
880125
0
0
-5000
-100000
0
100
200
Time (sec)
300
400
mb = 5.7
Mo = 5.75E+17 N m
Dist. = 701 km
0
200
400
Time (sec)
600
mb = 5.4
Mo = 8.86E+16 N m
Dist. = 1645 km
Figure 5. Comparisons of Lg amplitude spectra for events located on the margin or within the Tibetan
Plateau with events outside the plateau that have similar path lengths, magnitudes and focal mechanisms:
(a) The top seismogram is for event 931230 (year/month/day), an event located west of WMQ, and the
lower seismogram is for event 870225 near the northern Plateau margin (see Figure 1). Note the
similarity of the Lg spectra and waveforms. (b) The top seismogram is for event 880103 in central China
while the lower seismogram is for event 880125 in southern Tibet (see Figure 1). Note the relative
depletion of high frequency Lg energy for the path traversing the plateau. Lower hemisphere best double
couples are shown from the Harvard Centroid Moment Tensor solutions for all events, along with
corresponding seismic moments.
decrease in crustal thickness does not cause Lg blockage
[e.g., Gibson and Campillo, 1994], and this would be
particularly true for events located very close to the step
change. However, because there is a two hundred kilometer
distance gap to the next concentration of seismic activity in
the plateau, and those events lack high-frequency Lg signals, a possible role of blockage by crustal necking for the
events further to the south remains open.
[15] For events near the Himalayan mountains, Lg signals
at WMQ traverse the entire Tibetan Plateau. Without
exception, Lg signals at WMQ for long travel paths across
the plateau are weak or entirely extinguished. For the
Himalayan and southern Plateau events, the corresponding
corner frequencies of velocity spectra range between 0.2
and 0.4 Hz, and the spectra are similar to Figure 4b. The
question then arises as to how anomalous such spectra are,
so we again compare with signals for comparable propagation distances outside the plateau. Figure 5b shows one such
comparison, for events of similar magnitude, seismic
moment, focal mechanism and propagation distance. The
event in Central China is quite typical of Eurasian observations at this distance, with strong, high-frequency Lg
energy arriving ahead of the fundamental mode Rayleigh
wave, and a spectral corner frequency around 0.7 Hz. The
seismogram for the event in southern Tibet is visibly
depleted in high-frequency Lg, and the corner frequency
is shifted to about 0.2 Hz, despite the presence of comparable high-frequency energy in direct P and the P coda and
comparable fundamental mode amplitudes. Such comparisons confirm that the Tibetan paths are truly anomalous
relative to standard Eurasian paths, and that the spectral
character is robust enough that we can proceed to characterize the evolution of this spectral behavior across Tibet even
though detailed source models are not available.
[16] We observe that the Lg spectra evolve systematically
between the extremes in Figures 5a and 5b as a function of
path length across Tibet. The pattern is most evident for the
longest paths through central Tibet, and somewhat more
subdued for paths across western and eastern Tibet. We thus
divide the events into three subgroups based on the back
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Corner Frequency (Hz)
101
CC = -0.820
SD = 0.156
K = -0.0605
Lg
100
10-1
500
1000
a)
1500
2000
Distance (km)
Corner Frequency (Hz)
101
CC = -0.938
SD = 0.0950
K = -34.383
Lg
100
10-1
1
b)
2
3
4
Mean Elevation (km)
Figure 6. Apparent corner frequency of the Lg amplitude
spectra for events in the Tibetan Plateau as a function of (a)
path length from WMQ and (b) mean path elevation along
the path to WMQ. Symbols in (a) are the same as those in
Figure 1, and these indicate the shift to lower corner
frequencies for paths in central Tibet; solid circles in (b)
represent all events used. CC, linear correlation coefficients;
SD, standard deviation of linear regression; K, slope of the
regression line. The dotted lines indicate ±1 standard
deviation about the regression lines.
azimuth of each event (see Figure 1). While the corner
frequency of the velocity spectra is a rather uncertain,
subjective parameter, we find it useful for conveying the
overall evolution of the spectra manifested in the raw data.
Figure 6a shows the corner frequency estimates as a
function of total path length to WMQ, with symbols differentiating the three spatial subgroups. While one would
expect some trend due to attenuation, a much subtler trend
would be found for events outside Tibet, effectively a shift
from 1 – 2 Hz near 700 km to 0.7 –1 Hz near 2000 km, as
demonstrated in Figure 5. While source size should affect
corner frequency, this is also expected to be a small effect
because most of our data span only one unit of magnitude.
We actually find no correlation between the observed corner
frequencies and event magnitude, suggesting that the source
effect is swamped by propagation factors. We also could not
detect any systematic variation with source focal mechanism for the events with Harvard CMT solutions, although
such dependence may be rather complex. Lg radiation
patterns are often treated as isotropic for gross measure-
ESE
14 - 7
ments such as ours [e.g., Shih et al., 1994], and we will
make this assumption below.
[17] Overall, the data in Figure 6a display a systematic
decrease in corner frequency with path length, with a
baseline shift between central Tibet and eastern or western
regions of the plateau. In detail, the data exhibit some
clustering that reflects distributions in the seismicity, but
if one focuses on the circles, which are for paths through
central Tibet, it is clear that the primary factor appears to be
propagation distance from the Tibetan Plateau margin. It has
been observed that a sharp change in crustal thickness
across the northern margin of the Tibetan Plateau results
in anomalous double-pulse teleseismic P waves at one
station near the margin [Zhu and Helmberger, 1998].
Because Lg propagation at regional distances is more
complicated than for teleseismic P waves, it is unclear
whether there is any abrupt affect on Lg. Nevertheless, it
is reasonable to infer that if crustal necking at the northern
margin is responsible for the observed Lg blockage, one
would expect far less of a uniform trend between central and
southern Tibet events than observed. The logarithmic corner
frequency values decay rather linearly with distance. Effectively, these spectra fulfill Molnar’s prediction (as individually summarized by Ruzaikin et al. [1977]) that strong
frequency shifts with distance would indicate a progressive
attenuation effect. This is, however, a much stronger effect
than would be expected for the moderate attenuation values
previously found for eastern Tibet [McNamara et al., 1996].
[18] The distribution of seismicity and the geographic
pattern of the Tibetan Plateau relative to WMQ (Figure 1)
are such that propagation distance alone does not isolate the
contributions from the plateau portions of the paths very
well. We thus consider the behavior of the Lg amplitude
spectra relative to various measures that reflect the plateau
path contribution more directly using a parametric analysis
method [e.g., Fan and Lay, 1988]. We mainly considered
topographic features [Gesch et al., 1999] and parameters
from the crustal waveguide structure [Fielding et al., 1992].
Regression analysis shows that the Lg corner frequency
measurements have strong, negative correlations with several physical parameters of the Tibetan Plateau. The highest
correlation coefficient (0.94) we found is for mean elevation along each path, which reflects the relative portion of
the paths within Tibet (see Figure 6b). This correlation
appears to suppress any east to west variations within the
plateau. If travel distance along the plateau above a certain
elevation threshold, such as 4000 m, is used rather than the
entire path length to WMQ, the value of correlation coefficient varies from 0.82 (Figure 6a) to 0.92, again
suppressing variations from east to west. Strong correlations
are also found using mean crustal thickness and mean
topographic roughness (variance) of each path (0.90 and
0.85, respectively); these parameters covary strongly with
mean elevation. The mean thickness of the sediment layer
on each path displays only a weak, positive correlation
(0.35) with Lg corner frequency. The available sedimentary
thickness information for Tibet is rather low resolution, so it
is premature to discount some control on Lg energy by
sedimentary basin scattering. The slopes of the higher
frequency portion of the Lg amplitude spectra for each
event were also computed, but these measures do not show
any simple relationship with other physical parameters. This
ESE
14 - 8
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Figure 7. A map showing interpolated estimates of apparent corner frequency for the Lg amplitude
spectra for events in the Tibetan Plateau as observed at WMQ. Interpolation of the observations is done
using a standard kriging algorithm. Small circles represent the event locations used in this study; thick
solid lines are 3000-meter topographic contours. See color version of this figure at back of this issue.
may reflect an interaction of frequency-dependent attenuation and source radiation controlling the slope.
[19] The two-dimensional spatial pattern of the corner
frequency measures made at station WMQ was further
examined using an ordinary kriging algorithm to assess
whether the single parameter regressions were missing any
behavior. The interpolated pattern basically follows the
integrated path sampling of plateau topography, surrounded
by a base level in other regions of China, and no subtle spatial
pattern emerges (see Figure 7). We found similar results
when kriging RMS measures of Pg/Lg in various passbands.
Thus, physical properties of the crustal waveguide in the
Tibetan Plateau are responsible for the systematic pattern in
the low-frequency Lg observations. Attenuation strong
enough to account for the variation in spectra over long
paths transiting central Tibet certainly requires more pronounced QLg values than previously reported. The next step
in our analysis is to demonstrate whether this attenuation is
uniform or needs to have particularly low values in the
northernmost region in order to account for the rapid loss
of high-frequency energy.
4. Determination of Lg Attenuation
[20] Our observations of Lg spectral content shifting to
lower frequency with increased path length in Tibet demonstrate a strong, progressive Lg attenuation in the crust
beneath the plateau. Using a simple two-event method, we
estimate Lg attenuation coefficient g( f ) for the paths crossing several sectors of Tibet. Some constraints are placed on
regional heterogeneity of Lg attenuation.
[21] The displacement spectral amplitude of the Lg wave
arriving from a source at a distance d can be expressed in
the form
Að f ; d Þ ¼ Sð f ÞRð f ; qÞGðd Þ ð f ; qÞI ð f ÞSS ð f ; Þ;
ð1Þ
where f denotes the frequency, q the station azimuth, S( f ) is
the source excitation function, R( f, q) is the source radiation
pattern, G(d ) is the Lg geometrical spreading function, I( f )
is the instrument response, SS( f, ) is the station site
response which can vary with the backazimuth, , ( f, q) is
anelastic attenuation function which can be written by
ð f ; qÞ ¼ egd ;
ð2Þ
where g is the attenuation coefficient , which is related to
the quality factor Q and the group velocity U as
g ¼ ð pf Þ=ðQU Þ:
ð3Þ
This parameter will account for both intrinsic attenuation
and scattering attenuation losses. We assume a constant
group velocity of 3.5 km/s, as this is generally consistent
with the larger arrivals in the early part of the Lg wave
train as seen in Figures 2 and 3, along with corresponding
to Rayleigh overtone energy apparent in time-variable
filtered plots of some of the signals. Certainly some error
is incurred by assuming a single, somewhat uncertain
group velocity for all frequencies, but the path geometry
precludes a detailed analysis of individual modes as
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
necessary to do better. Given the interference effects of
the multimode contributions to Lg, it is unlikely that a
more robust group velocity behavior can be determined
for our strongly attenuated data, given the present data
distribution.
[22] Restricting our attention to two events that lie within
a few degrees of the great circle path connecting the receiver
and the two epicenters, we obtain the ratio of Lg wave
spectral amplitude for event 2 (the distant event) to that for
event 1 (the nearer event) as
Að f ; d2 Þ=Að f ; d1 Þ ¼ ½S2 ð f Þ=S1 ð f Þ½Gðd2 Þ=Gðd1 Þexp½g;
ð4Þ
Where the indices 1 and 2 represent event 1 and event 2,
respectively, and is the distance between the two
events. Note that the source radiation pattern R( f, q) does
not appear in (4) because we adopt the common
assumption that Lg source radiation is azimuthally
independent to first order [Sereno et al., 1990; Shih et
al., 1994]. This is the most difficult assumption to
validate, and we rely on stacking of multiple ratios to
overcome any bias due to neglect of azimuthal radiation
effects. The site amplification term cancels 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, (versus G(d) = d0.83 in the time domain) [Ewing
et al., 1957; Campillo et al., 1984; Hasegawa, 1985;
Chun et al., 1987; Shin and Herrmann, 1987]. The precise
exponent to use for a laterally heterogenous structure is
not clear, however, the study by Xie [2002] and our own
calculations indicate that the choice of different reasonable
geometrical spreading parameters does not lead to
significant changes in QLg estimates.
[23] To solve for the attenuation coefficient, we need to
estimate the effects of the source excitation term S( f ), which
can be represented by the w2 model as
S ð f Þ ¼ S0 =ð1 þ f 2 =f c2 Þ;
ð5Þ
where S0 is a constant for a given event, and fc is the
corner frequency. The radiation corner frequency fc is
related to the stress drop, s, the shear wave velocity at
the source, ub, and the seismic moment of the event, M0
[Brune, 1970]. If the difference in fc for the two events is
small, then, as a first order approximation, (5) can be
simply reduced to S( f ) ffi S0. One can assume constant
stress drop and scale corner frequency with moment, but
this is rather arbitrary given observed scatter in measured
stress drops. Since our spectral ratios are extremely
bandwidth limited by strong attenuation, we ignore small,
high-frequency corrections and restrict our analysis to
frequencies below 0.5– 1.0 Hz. Thus, we can rewrite (4)
as
Að f ; d2 Þ=Að f ; d1 Þ ¼ ½S02 =S01 ½d2 =d1 0:5 exp½g;
ð6Þ
and solve for the attenuation coefficient g(f ). Our
simplified source correction is approximately equivalent
to shifting the spectral ratio to a zero baseline at its low –
ESE
14 - 9
frequency limit. We found that use of the seismic
moments from the Harvard CMT solutions provided
spectral ratios with small scatter for different event pairs
with about the same magnitudes. Because not all of the
events studied have Harvard CMT solutions available, we
first constructed a reference spectral ratio curve by
averaging all spectral ratios after source corrections.
Other spectral ratios lacking moment information were
then shifted relative to the reference spectral ratio by
correlating over the frequency band of 0.2– 0.5 Hz.
Finally, a stacked average is calculated from all spectral
ratios for all event pairs to provide a representative Lg
attenuation, g(f ), along a profile.
[24] We constructed four profiles approximately in the
north-south direction. Figure 8 shows the locations of
selected events used for estimating Lg attenuation. The
first three profile are positioned in the western, central,
and eastern parts of the Tibetan Plateau, respectively, with
events located near the southern and northern margins
and the associated paths crossing the entire plateau (see
Figure 8). The fourth profile covers a relatively small
area in the central part of northern Tibet, where strong Lg
attenuation was observed. In order to approximate great
circle paths, only events with backazimuths confined to a
range of 30 degrees are used in calculating g(f ). The
number of events used for each profile is 4 3, 6 6, 4
7, and 5 6, respectively, where the first number
represents the number of events in the north and the
second is the number of events in the south. The events
that have Harvard CMT solutions available were used in
calculating the reference Lg attenuation curves, which are
the average of many spectral ratios after source corrections for selected event pairs. The construction of a
moment-corrected reference Lg attenuation curve is demonstrated in Figure 9. The source corrections do not
eliminate scatter, but do appear to give a well-defined
mean curve. Any additional ratios for which absolute
amplitude information is lacking are correlated with the
mean, shifted to that level and averaged in, mainly to
smooth and stabilize the curve. In constructing reference
attenuation curves, 15 event pairs were used for profile
III, and 8 event pairs for profiles II and IV. For profile I,
the reference curve was calculated using only three event
pairs, and is thus less reliable. The spectral ratios are
calculated out only to 1 Hz, given the low signal-to-noise
ratio at higher frequencies.
[25] The Lg attenuation coefficients g(f) for each profile
show clear frequency dependence across the band of 0.1–
1.0 Hz (see Figure 10). Scatter in the 0.2– 0.5 Hz band
was minimized by our processing. The largest scatter is
observed for profile IV. In general, the Lg attenuation g(f)
increases with frequency up to about 1 Hz for profiles I,
II, and III and up to about 2 Hz for profile IV, then it
gradually drops off. This drop off may be caused in part
by lack of high-frequency source corrections, but mainly is
due to high-frequency Sn coda appearing in the Lg
window at large distances. The rate of increase in Lg
attenuation g(f ) is not a constant function of frequency
and differs for each profile. The highest attenuation and
the best linearity for LOG[g(f )] of Lg are observed in
profile IV, where rapid attenuation is observed across a
short distance range.
ESE
14 - 10
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Figure 8. A map showing locations of four profiles used for the Lg attenuation study. The event
epicenters in each profile are shown by different symbols with triangles for profile I, circles for profile II
and crosses for profile III. An expanded map on the top shows event locations (squares) for profile IV.
Those squares with origin dates indicate events for which waveforms are shown in Figure 2.
[26] Assuming a power law frequency-dependent model,
Lg attenuation can be written in terms of quality factor QLg as
QLg ð f Þ ¼ Q0 f h ;
ð7Þ
where Q0 is the value of QLg at 1 Hz, and h is the power law
frequency dependence. Based on equations (3) and (7), we
fit a QLg(f ) model to observed g(f ) to estimate the value of
Lg attenuation in various areas of the Tibetan Plateau. Our
data lack sufficient bandwidth and spatial sampling to
pursue a separation of intrinsic and scattering attenuation
effects.
[27] The logarithmic Lg attenuation coefficients g(f ) are
fit to a QLg(f ) model by least squaress linear regression in
two frequency bands, 0.2– 0.5 Hz and 0.2– 1.0 Hz, for
each profile. Figures 11a –11d show our results. The
numerical values of the best fit model parameters Q0 and
h are listed in Table 1. Previous research on Lg attenuation
in Asia indicates that h is very sensitive to lateral hetero-
geneity in the crust, thus reliable h estimates are difficult to
obtain [Shih et al., 1994], particularly with band-limited
data.
[28] We found that Q0 values differ for each profile. There
is little difference in Q0 for profile IV between the two best
fit models because the logarithmic spectra are very linear.
However, there are variations in the Q0 estimates for the
other profiles between the 0.2 – 0.5 Hz and 0.2 – 1.0 Hz
bands. Profiles II and III show a reduced slope in logarithmic
g(f ) commencing at about 0.5 Hz. Thus, broader band
estimates give higher Q0 values. We believe that this is
likely due to contamination at higher frequencies (above 0.5
Hz) by increased contribution of high-frequency Sn coda
rather than true Lg signals. The Q0 value of Lg for the
Tibetan Plateau as a whole is in a range of about 120– 150
for the 0.2 – 0.5 Hz band, and in a range of 160 –230 for the
0.2– 1.0 Hz band based on the results for profiles I, II and III.
Our observations indicate that a region in northern central
Tibet close to the plateau margin is characterized by very
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Lg Attenuation Coef.
0.015
0.010
0.005
0.000
-0.005
No Source Correction
-0.010
10-1
a)
100
Frequency (Hz)
Lg Attenuation Coef.
0.015
0.010
0.005
0.000
-0.005
After Source Correction
-0.010
10-1
b)
100
Frequency (Hz)
Figure 9. The construction of a reference Lg attenuation
curve by averaging spectral ratios of 15 event pairs after
simplified source corrections. Dashed lines are (a) observed
spectral ratios of 15 event pairs, and (b) the same spectral
ratios after simplified source corrections. Thick solid line in
(b) represents the reference Lg attenuation curve, which is the
average of all event pairs after simplified source corrections.
strong Lg attenuation, with a Q0 value of 85– 90. This is
consistent with observed high-frequency amplitude decay by
a factor of ten or more in that region (see Figure 2).
5. Discussion
[29] The Lg attenuation estimates obtained in section 4
are much lower than those discussed in the introduction,
most of which are for eastern Tibet. One concern is that
there is still some bias due to traversing the northern
margin of Tibet despite our observation of progressive
evolution of the Lg spectra. We thus consider Lg observations at station LSA, which is located within Tibet, for 31
events throughout the plateau, none of which have paths
traversing the northern margin of the plateau. Figure 12 is
a plot of observed apparent corner frequency versus path
length from LSA, showing great similarity to the WMQ
observations in eastern and western Tibet, with only the
central part having a stronger pattern. We infer that paths
to LSA sample Q0 attenuation values of 122 – 150 as in
profiles I, II and III. Recently, Xie [2002] has reanalyzed
the Lg spectra data used by McNamara et al. [1996],
ESE
14 - 11
obtaining a QLg model with Q0 = (126 ± 9) and h = (0.37
± 0.02), which is very consistent with our estimates in
profiles II and III, where the data overlap (however, note
that our estimates are for paths across the entire plateau).
Xie’s estimate is for the 0.2 – 3.6 Hz band, with the
extension to higher frequency being viable due to the
short path lengths involved. This extrapolates smoothly
up from our 0.2– 0.5 Hz band prior to the decrease in slope
of LOG[g(f)]. Phillips et al. [2000] obtain tomographic
models of Q0 with values of 200 in Tibet south of the
Qaidam and Tarim basins. Their results give Q0 = 500 for
the Tarim Basin, compatible with our observation of
efficient propagation there. Their coverage of western
Tibet is very limited due to constraining the data to
0.75– 1.5 Hz, which is strongly attenuated over longer
paths in Tibet.
[30] We resolve a region in northernmost Tibet with
much lower QLg than previously reported by any study of
Tibet. Figure 13 summarizes our Q0 estimates, illustrating
the localized region with Q0 values of 85– 90. Profile II
gives overall Q0 values of 141 for the 0.2 –0.5 Hz band,
which requires that Q0 increase in southern central Tibet.
Phillips et al. [2000] indicate such an increase in Q0 from
north to south, with higher Q0 values of 450 – 500 south
and east of Lhasa. We estimated QLg for a fifth profile
using four southern events in profile IV as reference events
relative to three events near the southern margin of the
plateau. Fitting data in the 0.2– 0.5 Hz band gives an
estimated Q0 value of 316 for the region from 35N to the
southern margin of Tibet. This is very compatible with
recent Lg analysis sampling to the west of LSA in the
0.5– 1.0 Hz band, which gives values of Q0 = 300– 350,
somewhat reduced from the earlier results of Phillips et al.
[2000] after elimination of events with uncertain mb values
(S. Phillips, personal communication, 2002). Because the
reference events in profile V are already attenuated significantly, we cannot extend our frequency band to higher
frequency, but the consistency is encouraging. Our value
for southern central Tibet is also consistent with the results
of McNamara et al. [1996] and Reese et al. [1999], but
our values are not as low as in the analysis by Xie [2002].
If the region of very low Q0 in northern Tibet extends
further west and east than can be resolved by our data,
there may also be north-to-south gradients in Q0 in western and eastern Tibet. Phillips et al. [2000] do not detect
strong trends from north to south in these regions, but
detailed work is needed.
[31] Our low Q0 values of 85– 146 are comparable to
Lg attenuation values found in other tectonically active
areas, such as California [Herrmann, 1980; Nuttli, 1986]
and Iran [Nuttli, 1980]. In the central Andes, where the
crustal thickening is comparable to that in Tibet, an
average Qs of 100 has been reported for the Bolivian
Altiplano [Baumont et al., 1999]. The latter study interpreted the frequency-dependent attenuation of Lg as
mainly being caused by strong crustal scattering, although
a significant contribution from small amounts of partial
melting in the crust was suggested. 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. The strong attenuation of Lg waves in Tibet and the
ESE
14 - 12
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Profile II
Lg Attenuation Coef.
Lg Attenuation Coef.
Profile I
0.010
0.005
0.000
-0.005
-0.010
10-1
0.010
0.005
0.000
-0.005
-0.010
100
10-1
Frequency (Hz)
a
Profile IV
Lg Attenuation Coef.
Lg Attenuation Coef.
Profile III
0.010
0.005
0.000
-0.005
-0.010
10-1
100
Frequency (Hz)
b
0.015
0.010
0.005
0.000
-0.005
100
10-1
Frequency (Hz)
c
100
Frequency (Hz)
d
Figure 10. Lg attenuation coefficient g(f ) as a function of frequency for each profile. The dashed lines
are calculated from the spectral ratio of each event-pair after simplified source correction, the solid line
represents a stacked average for all Lg attenuation coefficient g(f ) observations in each profile.
10-2
Profile I
scale crustal waveguide variations do not dominate the
attenuation effect, it is certainly possible that scattering
from small scale heterogeneity within the crust plays a
role, similar to the Alpine environment [e.g., Campillo et
Lg Attenuation Coef.
Lg Attenuation Coef.
very sparse station coverage preclude an attempt to
apportion Lg attenuation estimates into intrinsic anelasticity versus small-scale scattering attenuation, as attempted
in the Altiplano. While we have demonstrated that large
10-3
10-2
Profile II
10-3
100
100
Frequency (Hz)
a
b
Profile III
Profile IV
Lg Attenuation Coef.
Lg Attenuation Coef.
Frequency (Hz)
10-3
10-2
100
Frequency (Hz)
c
100
Frequency (Hz)
d
Figure 11. Lg attenuation coefficient g(f ) and the best fit QLg models (straight lines) for each profile.
Crosses are g(f ) values used for the data fitting. Solid and dashed lines represent the best fit QLg models
in frequency bands of 0.2 – 0.5 Hz, and 0.2– 1.0 Hz, respectively.
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
ESE
14 - 13
Table 1. Values of Lg Attenuation in Different Parts of Tibet
Profile I
Profile II
Profile III
Qo
h
146 ± 20
0.67 ± 0.12
141 ± 17
0.15 ± 0.11
0.2 – 0.5 Hz
122 ± 20
0.19 ± 0.15
Qo
h
160 ± 5
0.75 ± 0.04
224 ± 10
0.56 ± 0.06
0.2 – 1.0 Hz
195 ± 14
0.24 ± 0.08
al., 1993]. As improved resolution of Tibetan crustal
complexity is achieved, it may become viable to assess
the scattering contribution, but for this study it is lumped
into the overall effective Lg attenuation.
[32] The region of northern central Tibet where we find
very strong effective attenuation is the most volcanically
active area of Tibet [e.g., Molnar, 1988; Turner et al.,
1993, 1996; Arnaud et al., 1992]. In general, mechanisms
Profile IV
Profile V
90 ± 20
0.15 ± 0.10
316 ± 31
0.40 ± 0.09
85 ± 2
0.10 ± 0.04
–
–
of intrinsic shear wave attenuation are very sensitive to
temperature conditions, and the very low Qo of 85– 90
may be associated with partial melting of the crust in
northern Tibet, for which there is some independent
evidence. Owens and Zandt [1997] presented evidence
for a lower crustal low-velocity zone likely to involve
partial melt in northern Tibet. This region has inefficient
Sn propagation [Ni and Barazangi, 1983; McNamara et
Figure 12. Top figure is a topography map in western China and its vicinity. The location of station
LSA is shown. The squares with crosses indicate the epicenters of the 31 events used in this study.
Bottom figure shows apparent corner frequency of observed amplitude spectra versus distance from
station LSA. The solid and dotted lines represent the regression line observed from stations LSA and
WMQ, respectively. CC, linear correlation coefficients.
ESE
14 - 14
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Figure 13. A summary map of Qo values estimated for
different parts of Tibet. Thick solid lines are 3000-meter
topographic contours, thin solid lines represent three
subregions, where QLg values are estimated. The central
corridor has very low Qo in the north and higher Qo in the
south, with overall Qo values of 141– 224. The thick dashed
line represents the region with inefficient Sn propagation
based on the work of McNamara et al. [1995]; thin dashed
lines are political borders.
al., 1995], low Pn velocity [McNamara et al., 1997], 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. Our profile IV,
which provides the lowest Qo values, samples the northern half of the Qiangtang Terrane. Figure 14 shows a
north-south profile through Tibet adapted from Owens and
Zandt [1997], indicating the coincidence of the proposed
partial melt zone in the deep crust of northern Tibet and
the region of low QLg. Further analysis is required to
constrain the spatial extent of the very low QLg region,
particularly its southward extent, but it is clear that Tibet
cannot be viewed as a region with uniform Lg attenuation,
nor as a region with Lg attenuation values only in the
n
ya ya
th la
Te ima
H
as
h ay
ig l
H ima
H er ys
ss la
Le ima
H
S
[34] The Tibetan Plateau appears to have very strong
attenuation of Lg phases for paths within and traversing
the northern boundary of the plateau. This is demonstrated
by the broadband spectral behavior of Lg, which undergoes a systematic shift of apparent corner frequency from
2 to 0.2 Hz as a function of path length. The apparent
corner frequency shift is consistent with progressive
energy losses proportional to path length within the
plateau, and is particularly well correlated with physical
properties of the plateau. Spectral ratios for various path
geometries across Tibet indicate that Lg attenuation values
at 1 Hz are on the order of Qo = 80– 90 in the northern
central region of Tibet, near 316 in southern central Tibet
and on the order of 122 – 195 for whole plateau averages
in eastern and western Tibet. This implies that so-called
blockage of Lg energy for paths in Tibet is mainly caused
by very strong attenuation in the crust. The very low QLg
values in northern Tibet may be a result of partial melt in
the low-velocity layer in the lower crust.
Qiangtang
terrane
Moho
n
ya ic re
th
Te ean phe
Oc thos
Li
200
240 km
200
Qaidam
basin
N
40
Indian Continental
Lithosphere
100
Songpan-Ganz
terrane
Very low Lg Q
Indian Crust
120
160
0
6. Conclusions
MBT
40
80
Lhasa terrane
Tsangpo
Suture
range of Qo = 340 – 450 as previously reported. There is
some evidence for partial melt and crustal low-velocity
zones existing north of the Tsangpo suture in southern
Tibet [Nelson et al., 1996; Kind et al., 1996; Cotte et al.,
1999]. This low-velocity zone may also relate to a
localized low QLg region, however, it is not resolved by
our data set due to lack of spatial resolution in southern
Tibet.
[33] Further work is also needed to constrain the depth
distribution of attenuation in the crust, as QLg involves
complex sampling of the crustal waveguide. With relatively low overall crustal velocities likely to exist in
northern Tibet [e.g., Rodgers and Schwartz, 1998; Owens
and Zandt, 1997], Lg waves may provide overall sampling
of the thick crust, averaging the entire attenuation structure. Shallow crustal attenuation needs to be constrained
by fundamental mode attenuation in order to establish the
Qs structure of the upper and lower crust [e.g., Mitchell
and Xie, 1994].
300
400
500
Partially Molten Lower Crust
80
Inefficient Sn
Asian
Low Pn Velocity Continental
Lithosphere
200
240 km
600 km
Figure 14. A schematic north – south profile through Tibet adapted from Owens and Zandt [1997],
indicating the coincidence of the partial melt zone in the deep crust of northern Tibet and the region of
low QLg.
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
[35] Acknowledgments. We thank D. Baumgardt for sharing his
unpublished work on Lg blockage in Tibet. Comments from S. Phillips
and an associate editor helped us to improve the paper. R.-S. Wu and X.-B.
Xie provided helpful discussion and suggestions. Broadband waveform
data used in this study were collected at the IRIS data center. This research
was partially supported by the Defense Threat Reduction Agency through
contracts DSWA01-98-C-0161 and DTRA01-00-C-0211. CSIDE contribution 439, IGPP, University of California, Santa Cruz.
References
Arnaud, N. O., P. Vidal, P. Tapponnier, P. Matte, and M. Deng, The high
K2O volcanism of northwestern Tibet: Geochemistry and tectonic implications, Earth Planet. Sci. Lett., 111, 355 – 367, 1992.
Båth, M., The elastic waves Lg and Rg along Euroasiatic paths, Ark. Geofys., 2, 295 – 342, 1954.
Baumgardt, D. R., Investigation of teleseismic Lg blockage and scattering
using regional arrays, Bull. Seismol. Soc. Am., 80, 2261 – 2281, 1990.
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.
Bostock, M. G., and B. L. N. Kennett, The effects of 3-D structure on Lg
propagation pattern, Geophys. J. Int., 101, 355 – 365, 1990.
Brune, J. N., Tectonic stress and the spectra of seismic shear waves from
earthquake sources, J. Geophys. Res., 84, 2262 – 2272, 1970.
Campillo, M., M. Bouchon, and B. Massinon, Theoretical study of the
excitation, spectral characteristics, and geometrical attenuation of regional seismic phases, Bull. Seismol. Soc. Am., 74, 79 – 90, 1984.
Campillo, M., B. Feignier, M. Bouchon, and N. Béthoux, Attenuation of
crustal waves across the Alpine Range, J. Geophys. Res., 98, 1987 – 1996,
1993.
Cao, S., and K. J. Muirhead, Finite difference modeling of Lg blockage,
Geophys. J. Int., 115, 85 – 96, 1993.
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. Seismol. Soc. Am., 77, 398 – 419, 1987.
Cotte, N., H. Pedersen, M. Campillo, J. Mars, J. F. Ni, R. Kind, and E. Zhao,
Determination of the crustal structure in southern Tibet by dispersion and
amplitude analysis of Rayleigh waves, Geophys. J. Int., 138, 809 – 819,
1999.
Ewing, W. M., W. S. Jardetzky, and F. Press, Elastic Waves in Layered
Media, p. 358, McGraw-Hill, New York, 1957.
Fan, G.-W., and T. Lay, Statistical analysis of irregular waveguide influences on regional seismic discriminants in China, Bull. Seismol. Soc. Am.,
88, 74 – 88, 1988.
Fielding, E. J., B. L. Isacks, and M. Barazangi, A geological and geophysical information system for Eurasia, Tech. Rep. 2, F19601-91-K-DB08,
Philips Lab., Hanscom Air Force Base, Mass., 1992.
Gesch, D. B., K. L. Verdin, and S. K. Greenlee, New land surface digital
elevation model covers the Earth, EOS Trans. AGU, 80, 69 – 70, 1999.
Gibson, R. L., and M. Campillo, Numerical simulation of high- and lowfrequency Lg-wave propagation, Geophys. J. Int., 118, 47 – 56, 1994.
Hartse, H. E., S. R. Taylor, W. S. Phillips, and G. E. Randall, A preliminary
study of regional seismic discrimination in central Asia with emphasis on
western China, Bull. Seismol. Soc. Am., 87, 551 – 568, 1997.
Hasegawa, H. S., Attenuation of Lg waves in the Canadian shield, Bull.
Seismol. Soc. Am., 75, 1569 – 1582, 1985.
Herrmann, R. B., Q estimates using the coda of local earthquakes, Bull.
Seismol. Soc. Am., 70, 447 – 468, 1980.
Hirn, A., A. Nercessian, M. Sapin, G. Jobert, Z. Xu, E. Gao, D. Lu, and
J. Teng, Lhasa block and bordering suture—A continuation of a 500-km
Moho traverse through Tibet, Nature, 307, 25 – 27, 1984.
Kennett, B. L. N., Lg waves and structural boundaries, Bull. Seismol. Soc.
Am., 76, 1133 – 1141, 1986.
Kennett, B. L. N., On the nature of regional seismic phases, 1, Phase
representations for Pn, Pg, Sn, Lg, Geophys. J., 98, 447 – 456, 1989.
Kind, R., et al., Evidence from earthquake data for a partially molten crustal
layer in southern Tibet, Science, 274, 1652 – 1654, 1996.
McNamara, D. E., T. J. Owens, and W. R. Walter, Observations of regional
phase propagation across the Tibetan Plateau, J. Geophys. Res., 100,
22,215 – 22,229, 1995.
McNamara, D. E., T. J. Owens, and W. R. Walter, Propagation characteristics of Lg across the Tibetan plateau, Bull. Seismol. Soc. Am., 86, 457 –
469, 1996.
McNamara, D. E., W. R. Walter, T. J. Owens, and C. J. Ammon, Upper
mantle velocity structure beneath the Tibetan Plateau from Pn travel time
tomography, J. Geophys. Res., 102, 493 – 505, 1997.
Mitchell, B. J., and J.-K. Xie, Attenuation of multiphase surface waves in
the Basin and Range Province, 3, Inversion for crustal anelasticity, Geophys. J. Int., 116, 468 – 484, 1994.
ESE
14 - 15
Molnar, P., A review of geophysical constraints on the deep structure of the
Tibetan Plateau, the Himalaya and the Karakoram, and their tectonic
implications, Philos. Trans. R. Soc. London, Ser. A, 327, 33 – 88, 1988.
Nelson, K. D., et al., Partially molten middle crust beneath southern Tibet:
Synthesis of project INDEPTH results, Science, 274, 1684 – 1688, 1996.
Ni, J., and M. Barazangi, High-frequency seismic wave propagation beneath the Indian Shield, Himalayan Arc, Tibetan Plateau and surrounding
regions: High uppermost mantle velocities and efficient Sn propagation
beneath Tibet, Geophys. J. R. Astron. Soc., 72, 665 – 689, 1983.
Nuttli, O. W., The excitation and attenuation of crustal seismic phases in
Iran, Bull. Seismol. Soc. Am., 70, 469 – 485, 1980.
Nuttli, O. W., Yield estimates of Nevada test site explosions obtained from
seismic Lg waves, J. Geophys. Res., 91, 2137 – 2152, 1986.
Owens, T. J., and G. Zandt, Implications of crustal property variations for
models of Tibetan plateau evolution, Nature, 387, 37 – 43, 1997.
Phillips, W. S., H. E. Hartse, S. R. Taylor, and G. E. Randall, 1 Hz Lg Q
tomography in central Asia, Geophys. Res. Lett., 27, 3425 – 3428, 2000.
Press, F., and M. Ewing, Two slow surface waves across North America,
Bull. Seismol. Soc. Am., 42, 219 – 228, 1952.
Rapine, R. R., J. F. Ni, and T. M. Hearn, Regional wave propagation in
China and its surrounding regions, Bull. Seismol. Soc. Am., 87, 1622 –
1636, 1997.
Reese, C. C., R. R. Rapine, and J. F. Ni, Lateral variation of Pn and Lg
attenuation at the CDSN station LSA, Bull. Seismol. Soc. Am., 89, 325 –
330, 1999.
Regan, J., and D. G. Harkrider, Seismic representation theorem coupling:
Synthetic SH mode sum seismograms for non-homogeneous paths, Geophys J. R. Astron. Soc., 98, 429 – 446, 1989.
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.
Ruzaikin, A. I., L. Nersesov, V. I. Khalturin, and P. Molnar, Propagation of
Lg and lateral variations in crustal structure in Asia, J. Geophys. Res., 82,
307 – 316, 1977.
Sereno, T. J., S. Bratt, and T. Bache, Frequency-dependent attenuation in
eastern Kazakhstan and implications for seismic detection thresholds in
the Soviet Union, Bull. Seismol. Soc. Am., 80, 2089 – 2105, 1990.
Shih, X. R., K.-Y. Chun, and T. Zhu, Attenuation of 1 – 6 s Lg waves in
Eurasia, J. Geophys. Res., 99, 23,859 – 23,874, 1994.
Shin, T.-C., and R. B. Herrmann, Lg attenuation and source studies using
1982 Miramichi data, Bull. Seismol. Soc. Am., 77, 384 – 397, 1987.
Taylor, S. R., A procedure for estimation of source and propagation amplitude correlations for regional seismic discriminants, Bull. Seismol. Soc.
Am., 86, 1042 – 1053, 1996.
Turner, S., C. J. Hawkesworth, J. Liu, N. Rogers, S. Kelley, and P. van
Calsteren, Timing of Tibetan uplift constrained by analysis of volcanic
rocks, Nature, 364, 5053, 1993.
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, J. Petrol.,
37, 45 – 71, 1996.
Walter, W. R., K. M. Mayeda, and H. Patton, Phase and spectral ratio
discrimination between NTS earthquakes and explosions, 1, Empirical
observations, Bull. Seismol. Soc. Am., 85, 1050 – 1067, 1995.
Xie, J., Lg Q in the eastern Tibetan Plateau, Bull. Seismol. Soc. Am., 92,
871-876, 2002.
Xie, X.-B., and T. Lay, The excitation of Lg waves by explosions: A finitedifference investigation, Bull. Seismol. Soc. Am., 84, 324 – 342, 1994.
Zhang, T.-R., and T. Lay, Why the Lg phase does not traverse oceanic crust,
Bull. Seismol. Soc. Am., 85, 1665 – 1678, 1995.
Zhang, T.-R., S. Schwartz, and T. Lay, Multivariate analysis of waveguide
effects on short-period regional wave propagation in Eurasia and its
application in seismic discrimination, J. Geophys. Res., 99, 21,929 –
21,945, 1994.
Zhao, W., et al., Crustal structure of central Tibet as derived from project
INDEPTH wide-angle seismic data, Geophys. J. Int., 145, 486 – 498,
2001.
Zhu, L., and D. Helmberger, Moho offset across the northern margin of the
Tibetan Plateau, Science, 281, 1170 – 1172, 1998.
G. Fan, Center for the Study of Imaging and Dynamics of the Earth,
Institute of Geophysics and Planetary Physics, University of California,
Santa Cruz, CA 95094, USA.
T. Lay, Earth Sciences Department, Institute of Tectonics, University of
California, Earth and Marine Science Building, Santa Cruz, CA 95094,
USA. (thorne@es.ucsc.edu)
FAN AND LAY: CHARACTERISTICS OF Lg ATTENUATION IN THE TIBETAN PLATEAU
Figure 7. A map showing interpolated estimates of apparent corner frequency for the Lg amplitude
spectra for events in the Tibetan Plateau as observed at WMQ. Interpolation of the observations is done
using a standard kriging algorithm. Small circles represent the event locations used in this study; thick
solid lines are 3000-meter topographic contours.
ESE
14 - 8