Proceedings of the 18 Annual Seismic Research Symposium on Monitoring a comprehensive Test Ban Treaty

Teleseismic P- and S-wave Attenuation under the Baikal Rift Zone and Adjacent Areas: Method, Measurements, and Interpretation

Paul M. Davis and Shangxing Gao
Department of Earth and Space Sciences, University of California, Los Angeles, CA90095
Contract No. F49620-94-1-0161
Sponsored by AFOSR


Teleseismic P- and S-wave attenuation have been measured using the data set that we gathered along a 1280 km profile traversing the Siberian platform, the Baikal rift zone (BRZ), and the Mongolian fold belt. We compared the commonly used spectral ratio method and a common spectrum method for the calculation of relative tex2html_wrap_inline89 using synthetic data. The comparison suggests that the common spectrum method is more stable than the spectral ratio method. We used the common spectrum method to calculate tex2html_wrap_inline91 and tex2html_wrap_inline93 from 37 teleseismic events. It was found that the Baikal rift zone is associated with low tex2html_wrap_inline95 and tex2html_wrap_inline97 , with corresponding tex2html_wrap_inline91 of 0.2 s and tex2html_wrap_inline93 of 0.8 s relative to the Siberia platform and the Mongolian fold belt. Under the assumption that the mean tex2html_wrap_inline97 is 400 and tex2html_wrap_inline95 is 200 beneath the Siberian platform, we obtain a tex2html_wrap_inline97 of 130 beneath the BRZ, and a tex2html_wrap_inline95 of 60 beneath the BRZ. The best fitting relation between tex2html_wrap_inline91 and tex2html_wrap_inline93 was found to be tex2html_wrap_inline115 . The difference in tex2html_wrap_inline91 and tex2html_wrap_inline93 implies that the observed low Q in the BRZ is probably caused by intrinsic attenuation in the upper mantle rather than by scattering.

Keywords: Attenuation, Anisotropy, Baikal, Siberia, Mongolia, Continental Rift Zone, Bayesian Inversion


The objective is to study the propagation of seismic waves in a region of large lateral heterogeneity in seismic velocity, anelasticity, and anisotropy. We used digital data that along with colleagues from the University of Wisconsin (UW), and the Institute of Earth's Crust of Russian Academy of Sciences at Irkutsk (IEC) we collected in field experiments the summers of 91 and 92 in south central Siberia and Mongolia, as well as data digitized from the regional Russian seismic network that operates in that area. Continental rifts lie in regions of some of the largest lateral heterogeneity in velocity and attenuation in the continental crust and mantle. It is therefore important for nuclear monitoring purposes to quantify how seismic waves are affected by such lateral heterogeneities so that estimates of yield and location are accurate, and source type discrimination is reliable.


1. Introduction

In previous reports we have presented results from our ongoing study of the Baikal rift zone dealing with P wave attenuation, anisotropy, and P wave tomography (Davis and Gao, 1995). We have found the region is underlain by a low velocity zone extending from the Moho beneath the rift to depths of about 200 km and two to three hundred km wide. At almost every station SKS splitting of about 1 second has been observed with a pattern of fast directions that is variable, but shows some spatial correlation, suggesting a relation to the tectonics (Gao et al., 1994, Davis and Gao, 1995). We have reported P wave attenuation measurements along a 570 km east-west profile of seismic stations crossing lake Baikal which exhibit a relative tex2html_wrap_inline91 of 0.1 seconds. In this report we present P wave attenuation estimates along a longer 1280 km NW-SE profile and obtain a larger contrast of 0.2 seconds. We have also calculated S wave attenuation along this profile. Characterizing the seismic properties of such a tectonic zone has relevance to CTBT in that nuclear tests in continental zones of high absorption can disguise yield. Also, explosions detonated in highly anisotropic regions may generate S waves making discrimination from earthquakes more difficult if not recognized.

2. Method: A Bayesian Approach for the Calculation of Seismic Body-wave Attenuation Factors

Most of the seismic wave attenuation studies used spectral ratio method (e.g. Teng, 1968; Solomon and Toksoz, 1970; Der and McElfresh, 1976; Taylor et al., 1986). In this method the ratio of spectra between two stations, station i and the reference station j is used to determine tex2html_wrap_inline129 , the relative attenuation factor.^M The spectral ratio method uses a single reference record to determine tex2html_wrap_inline131 . We have performed synthetic tests that indicate that in the presence of noise the results from this method can be unstable. They are strongly affected by the spectrum of the reference record which, because it appears in the denominator, can cause infinities in the division if there are nulls in the spectrum. Thus the largest values in the fit have the greatest uncertainties and weighted methods should be applied. The method requires careful choice of parameters such as spectral smoothing window, or where to truncate the spectrum when the signal to noise ratio is low. To overcome such problems, Halderman and Davis (1991) proposed a non-linear inversion procedure, called the "Common Spectrum method" (CS), using the routines in Bevington (1969), which makes use of a gradient search method. The CS method uses all the spectra from an event to simultaneously invert for t*, the receiver amplification, and a common spectrum for the event. After the seismograms have been corrected to a standard response, the spectrum recorded by the ith station from the kth event, tex2html_wrap_inline143 can be expressed as


where f is frequency, tex2html_wrap_inline147 is the common spectrum for the event, and tex2html_wrap_inline149 are the near-receiver effects which are assumed to be frequency and source-location independent.

We performed the non-linear inversions using a Bayesian approach (Jackson and Matsu'ura, 1985) to search for tex2html_wrap_inline151 , and tex2html_wrap_inline147 . The Bayesian approach uses some prior information such as the starting parameters and weights the data and prior data according to their tex2html_wrap_inline155 uncertainties. Data processing parameters include:

tex2html_wrap_inline157 - starting guesses for tex2html_wrap_inline159 , and tex2html_wrap_inline161 . We use the mean spectrum over all the records from a given event as the starting guesses for tex2html_wrap_inline147 ; tex2html_wrap_inline165 are initially set to 1; and the starting guesses for the tex2html_wrap_inline161 values are taken as the values found by the spectral ratio method where the reference spectrum is the mean spectrum.
tex2html_wrap_inline169 - standard deviations of tex2html_wrap_inline157 . Together with tex2html_wrap_inline157 , the selection of tex2html_wrap_inline169 is the most important step in Bayesian inversion. It reflects the prior uncertainty of the tex2html_wrap_inline157 . A parameter with a larger tex2html_wrap_inline169 is more likely to be modified during the inversion than the ones with smaller tex2html_wrap_inline169 .
tex2html_wrap_inline183 - number of iterations for the inversion. We use tex2html_wrap_inline185 .

Note that no smoothing is performed; and no reference record is needed. We have conducted extensive tests on synthetic data of different signal to noise ratios and find that the Common Spectrum approach is significantly more accurate than the spectral ratio method and gives robust estimates for signal to noise ratios as low as 1.25 where the spectral ratio method breaks down.

3. Application to Baikal Data

We applied the common spectrum method to P wave and S wave data from Baikal.^M The stations analyzed were located along a 1280 km long profile traversing the Siberian platform, the BRZ, and the Mongolian fold belt; and in a 300 by 300 km cluster around the southern part of Lake Baikal (Figure 1). All the seismographs were digital and timed by the global Omega clock system. Several types of seismometers were used, most of them were three-component, short period (with free-period 0.5-2 s) sensors. The seismograms were standardized to a uniform response function with free-period 0.5 s, and damping factor 0.707.

Strong teleseismic events with direct P-wave as the first arrival were used. According to the ISC catalog, during the time of the experiment (day 180 and 265, 1992), 48 events with tex2html_wrap_inline187 occurred in the distance range of tex2html_wrap_inline189 from station 24, which was located on the southern shore of Lake Baikal. Among the 48 events, 37 are found to have sufficient P-wave signal to noise ratio on at least 10 of the stations to be used for P-wave attenuation studies, and 5 of them are found to be suitable for S-wave studies. The locations of the 37 events are shown in Figure 2. The events can roughly be divided into two groups. Events in the southern group are from a backazimuth range of 130-210 tex2html_wrap_inline191 , and those in the northern group are in the range of -30-60 tex2html_wrap_inline191 . The number of events in the southern and northern groups is 29 ^M and 8 respectively.^M Note that all the 5 events with strong S-waves are in the southern group.

The starting values for tex2html_wrap_inline147 are taken as the mean spectrum for all the stations from event k. The tex2html_wrap_inline155 standard deviations were set at about 30% of the peak value. The starting parameters for tex2html_wrap_inline161 and their standard deviations are taken as those found by the spectral ratio method. All the starting parameters for tex2html_wrap_inline165 are taken as 1.0 with tex2html_wrap_inline205 . The frequency band of 0.1 - 3.0 Hz is used. Most of the energy recorded by the seismometers, which are short period sensors, is in the frequency band of 0.3 to 2.0 Hz for both P- and S-waves. In this frequency band the Q values vary slowly with period (Anderson and Given, 1982). Therefore the assumption of frequency-independent Q is close to reality.

For the calculation of both tex2html_wrap_inline91 and tex2html_wrap_inline93 , a time window of 12.8 s was taken starting approximately 6 s before the onset of the signal, and was tapered and Fourier transformed. A noise window of the same length was also selected before the signal window and Fourier transformed. The noise spectrum was subtracted from the signal spectrum. We found that the removal is necessary, especially for stations in the Siberian platform where a 2.09 Hz continuous seismic signal has been detected (Gao et al., 1996 in preparation).

To minimize the effects of stations contaminated by noise, a two-step procedure was used. The first step applies the method to all the stations recording a given event. The second step uses stations with misfit (reduced chi squared) smaller than a cutoff value. The misfits were normalized by the amplitude of the common spectrum and therefore a single cutoff value of the misfit is used for all the events, which is 0.05 for P-waves and 0.2 for S-waves. The reason for a larger cutoff value for S-waves is that S-wave data is nosier than P-wave data and if the same cutoff value were used, there would be only a few qualified S-wave measurements. The number of stations rejected after the first step is about 4, out of typically more than 20 stations recording the event.

4. Results and Discussion

The number of measurements for tex2html_wrap_inline91 and tex2html_wrap_inline93 is 596 and 82, respectively. The mean values at each station and their standard deviations are shown in Figure 3. The solid curves in Figure 3 are obtained by averaging the measurements in a window of 200 km in width and moving at 10 km each step. The peak to peak anomaly of 0.2 s for tex2html_wrap_inline91 , and 0.8 s for tex2html_wrap_inline93 is observed in the smoothed mean curves. The mean standard deviation for tex2html_wrap_inline91 is 0.06 tex2html_wrap_inline225 0.03 s, and for tex2html_wrap_inline93 is 0.14 tex2html_wrap_inline225 0.08 s. The tex2html_wrap_inline91 and tex2html_wrap_inline93 anomalies associated with the BRZ are about 6 times the standard deviation. For tex2html_wrap_inline91 , the peak values are located in the area of tex2html_wrap_inline237 km about Lake Baikal, for tex2html_wrap_inline93 , the peak is located at -160 km (i.e., to the Northwest). Because all the events for tex2html_wrap_inline93 were from the south and with an S-wave angle of incidence of about 35 tex2html_wrap_inline191 , the northward shift may indicate that the depth of the center of the low-Q body is located at about 240 km.

Given that most of the lateral variation of shear wave velocities are found in the top 300 km of the Earth (e.g. Woodhouse and Dziewonski, 1984), we assume that the upper-most 300 km of the Earth beneath the array is responsible for the observed tex2html_wrap_inline91 and tex2html_wrap_inline93 . Under the assumption that the mean tex2html_wrap_inline97 is 400 beneath the Siberian platform, the 0.2 s in tex2html_wrap_inline91 results in a tex2html_wrap_inline97 of 130 beneath the BRZ, and the 0.8 s in tex2html_wrap_inline93 yields a tex2html_wrap_inline95 of 60 beneath the BRZ if a tex2html_wrap_inline95 of 200 is assumed beneath the Siberian platform.

Figure 4 shows the relation between the smoothed tex2html_wrap_inline91 and tex2html_wrap_inline93 . The best fitting relation is tex2html_wrap_inline115 . The cross correlation coefficient between the two is 0.83. In comparison, similar measurements between the eastern and western United States have tex2html_wrap_inline93 / tex2html_wrap_inline91 ranges from 2.7 to 3.4 (Der et al., 1980). In addition, using short period core phases, Kanamori (1967) found tex2html_wrap_inline271 . This ratio indicates that the observed tex2html_wrap_inline131 is the result of difference in purely intrinsic attenuation rather than difference in scattering (Walsh, 1969; Solomon and Toksoz, 1970; Richards and Menke, 1983).


Using synthetic data, we tested the common spectrum method using a Bayesian non-linear ^M inversion procedure to calculate relative attenuation factors for P- and S-waves. We found that it is more stable then the spectral ratio method. The common spectrum method was applied to the recordings from 37 teleseismic events recorded by about 60 stations in the Siberia, Lake Baikal, and Mongolia areas in the summer of 1992. The Baikal rift zone was found to be associated with a tex2html_wrap_inline89 of 0.2 s for P-waves and 0.8 s for S-waves relative to the Siberian platform. The best fitting relation between the smoothed tex2html_wrap_inline91 and tex2html_wrap_inline93 is tex2html_wrap_inline115 . The cross correlation coefficient between the two is 0.83. This difference indicates that the observed tex2html_wrap_inline131 is the result of difference in purely intrinsic attenuation rather than difference in scattering. The seismologic contrast between the Baikal rift zone and the Siberian craton is in many regards similar to that between the Basin and Range rift province and east North America.


Anderson, D. L., and J. W. Given, Absorption band Q model for the Earth, J. Geophys. Res., 87, 3893-3904, 1982.

Bevington, P. R., Data Reduction and Error Analysis for the Physical Sciences, pp 237-239, McGraw-Hill, New York, 1969.

Davis, P.M., and S. Gao, Seismic Propagation in the Baikal Rift Zone: A Transition from a Craton to an Orogenic Zone, in Proceedings of the 17 Annual Seismic Research Symposium on Monitoring a comprehensive Test Ban Treaty, 12-15 September, 1995, PL-TR-95-2108, editors James F. Lewkowicz, Jeanne M. McPhetres, and Delaine T. Reiter, pp. 99-110, 1995.

Der, Z. A., and T. W. McElfresh, The effect of attenuation on the spectra of P waves from nuclear explosions in North America, SDAC-TR-76-7, Teledyne Geotech, Alexandria, Virginia, 1976.

Der, Z. A., E. Smart, and A. Chaplin, Short-period S-wave attenuation under the United States, Bull. Seismo. Soc. Ame., 70, 101-125, 1980.

Gao, S., P. M. Davis, H. Liu, P. D. Slack, Yu. A. Zorin, V. V. Mordvinova, V. M. Kozhevnikov, and R. P. Meyer, Seismic Anisotropy and Mantle Flow beneath the Baikal Rift Zone, Nature, 371, 149-151, 1994.

Gao, S., P. M. Davis, and Yu. A. Zorin, A continuous 2.09 Hz seismic signal detected by a portable seismic network in Siberia and Mongolia, In preparation, 1996.

Halderman, T. P., and P. M. Davis, tex2html_wrap_inline97 beneath the Rio Grande and east African rift zones, J. Geophys. Res., 96, 10,113-10,128, 1991.

Jackson, D. D., and M. Matsu'ura, A Bayesian approach to nonlinear inversion, J. Geophys. Res., 90, 581-591, 1985.

Kanamori, H., Spectrum of short-period phases in relation to attenuation in the mantle, J. Geophys. Res., 72, 2181-2186, 1967.

Richards, P. G., and W. Menke, The apparent attenuation of a scattering medium, Bull. Seismol. Soc. Am., 73, 1005-1021, 1983.

Solomon, S.C., and M.N. Toksoz, Lateral variation of attenuation of P and S waves beneath the United States, Bull. Seismol. Soc. Am., 60, 819-838, 1970.

Teng, T.L., Attenuation of body waves and the Q Structure of the mantle, J. Geophys. Res., 73, 2195-2208, 1968.

Walsh, J. B., New analysis of attenuation in partially melted rock, J. Geophys. Res., 74, 4333-4337, 1969

Woodhouse, J. H., and A. M. Dziewonski, Mapping the upper mantle: Three-dimensional modeling of earth structure by inversion of seismic waveforms, J. Geophys. Res., 89, 5953-5986, 1984.


Figure 1: A Mercator projection map showing locations of stations. The lower-left inset shows coast lines and national borders of Asia and the location of the study area.

Figure 2: An azimuthal equidistant projection map showing epicenters of events, and their back-azimuths and epicentral distances to the array. The center of the map is station 24, which was on the southern shore of the Lake. The four circles represent tex2html_wrap_inline289 and tex2html_wrap_inline291 epicentral distances.

Figure 3: Results of tex2html_wrap_inline131 measurements. The dots are averaged values over all the events for tex2html_wrap_inline91 (Figure A) and tex2html_wrap_inline93 (Figure B), and the solid lines are the results of spatial averaging in windows of 200 km wide, moving at steps of 10 km. The thickness of the shaded area is 2 times the standard error of the mean.

Figure 4: Relations between tex2html_wrap_inline91 and tex2html_wrap_inline93 . Note that the values used are from spatial averaging of the mean values (the solid curves in Figure 3). The best fitting line and the cross correlation coefficient are shown in the figure.

S. Gao

Gao's Home Page