Short Notes Lg Q in the Eastern Tibetan Plateau

Transcription

1 Buetin of the Seismoogica Society of America, Vo. 92, No. 2, pp , March 2002 Short Notes Q in the Eastern Tibetan Pateau by Jiakang Xie Abstract spectra are coected from the Tibetan Pateau Passive Experiment to measure Q vaues. Using a standard two-station method that virtuay eiminates source and site effects, I obtain a mode of Q 0 (26 9) and g ( ) in a frequency range between 0.2 and 3.6 Hz, where Q 0 and g are Q at Hz and its power-aw frequency dependence, respectivey. The estimated Q 0 vaue is among the owest ever reported for continenta areas; it quaitativey supports the observation by McNamara et a. (996) that can not be observed inside the pateau beyond about 700 km, a imiting distance that is much shorter than those in the other ow Q 0 ( 200) regions, such as Iran and the western United States. The ow Q 0 vaue may be the cause of bockage across the northern boundary of the pateau and may indicate abnormay high temperature and fuid content in the Tibetan crust. Quantitativey, the estimated Q 0 vaue is ower by a factor of 3 than the vaue of 366 estimated by McNamara et a. (996), who used data from the same experiment. Since there are severa differences in the data processing and inversion procedures used in this and the previous studies, I investigated the effects of these differences on the Q estimates. I concude that the most probabe cause of the discrepancy is in the different inverse methods used. The previous inversion soved for a arge number of free parameters that incude the source and site terms. In this study ony two free parameters (Q 0 and g) are soved for, thus avoiding the instabiity caused by parameter trade-offs. Introduction The seismic wave can be treated as mutipe supercritica S-wave refections or many overtone surface waves traveing in the continenta crust. The attenuation rate, or Q, of generay correates with the tectonic environment. The Q has often been observed to fit a power-aw frequency dependence: η Q( f) = Q 0 f, () where Q 0 and g are Q at Hz and its power-aw frequency dependence, respectivey. Q 0 vaues in tectonicay active regions are typicay ower than about 300. For exampe, Q 0 vaues were estimated to be about 50 in Caifornia (Herrmann, 980; Nutti, 986), 200 in Iran (Nutti, 980), and 200 to 267 for the various regions in the western United States (Xie and Mitche, 990; Xie, 998). These vaues are generay about a factor of 2 to 5 ower than the vaues found in the stabe centra and eastern United States. The highest attenuation rates ever documented are those observed in the Tibetan Pateau. Ruzaikin et a. (977) and Ni and Barazangi (983) reported that, on short-period seismograms observed over paths crossing the northern and southern boundaries of the pateau, the wave is either abnormay weak, or absent. These phenomenon are quaitativey known as partia or compete bockage of by regions aong, or behind, these boundaries. McNamara et a. (996) used seismic data from a portabe network depoyed inside the pateau to study attenuation. They found that for paths that ie competey inside the pateau, the phase coud be observed out to a imiting distance of about 600 to 700 km, beyond which was absent owing to a high attenuation, or ow Q. They aso quantitativey fitted the ampitudes observed over paths in eastern Tibet, with a Q (f ) mode of Q and g A puzzing aspect of that mode is that the estimated Q 0 vaue is consideraby higher than the vaue of about 200 estimated for Iran and the vaues beow 267 for the western United States, whereas the -Hz signa can be observed at distances out to 400 km in Iran (Nutti, 980) and out to more than 2000 km in the western United States (Xie and Mitche, 990; Xie, 998). In this artice, I report a new anaysis of the data used by McNamara et a. and a resuting Q (f ) 87

2 872 Short Notes mode that is characterized by a much ower Q 0 vaue for eastern Tibet. Data Processing The raw waveform data used in this study are the same as those used by McNamara et a. (996) and are coected from broadband PASSCAL seismic stations depoyed during the , passive Tibetan Pateau experiment. Detaied specification of the stations and seismic events that were recorded during this experiment are given by McNamara et a. (996). Station ocations are aso shown in Figure of this artice. The data seection and processing procedures in this study somewhat differ from those used by McNamara et a. (996). The data processing procedure used by Xie and Mitche (990) and Xie (998) is used in this study to cacuate the ampitude spectra from the vertica component waveforms. Specificay, a 20% cosine taper window is used in the Fast Fourier Transform (FFT) to obtain spectra. The two corners of the window are ocated around group veocities of 3.0 and 3.5 km/sec, respectivey, and are aowed to be sighty adjustabe to best isoate the waveform. A noise reduction procedure, in which the power spectra is subtracted by a moving window average of the pre-p noise spectra, is appied. Ony those spectra estimates with signa-to-noise (S/N) ratio greater than 2.0 are kept for further anaysis. The origina purpose of the spectra anaysis in this study was to find source spectra parameters of the events recorded. During an initia anaysis of the spectra from ony a few events, it was found that the interstation spectra decay rate of coud not be fit by a Q 0 near 366. Rather, a Q 0 of no higher than 50 was required by the subset of spectra. Subsequenty, a systematic effort was undertaken to obtain a ampitude spectra from the stations and events that were shown in figure 2 of McNamara et a. (996). It is found that there is a fairy arge Figure. Locations of the PASSCAL stations depoyed during the passive Tibetan Pateau experiment (soid trianges), earthquakes (open circes), and exposion (star) used in this study. Soid paths are those satisfying a (dh) max (the maximum aowabe difference between the event-to-station azimuths of two stations; see text) of 2 when two-station pairs are seected. Dashed paths are those satisfying a (dh) max of 30. More information of the stations and events can be found in McNamara et a. (996). Note the simiarity of the 74 paths potted here and the 06 paths potted in figure 7 of McNamara et a. (996).

3 Short Notes 873 number (37) of two-station pairs that recorded the same events aong about the same great-circe paths. This permitted an area averaged Q (f ) mode to be deveoped using the two-station spectra ratio method, which is described in the next section. Inverse Method The two-station spectra ratio method is commony used for measuring interstation phase deay or ampitude decay of seismic waves (e.g., Aki and Richards, 980). A version of this method to measure interstation Q (f ) was given by Xie and Mitche (990). In this method the source effects in the recorded spectra are canceed by taking ratio of the spectra from a pair of stations that are aigned aong the same great circe path from the source. Spectra ratios from many combinations of the two-station pairs are then averaged to suppress the station site responses. The averaged (stacked) spectra ratio is then used to measure path Q, with a minima error caused by the source and site effects. In the foowing paragraphs, we present the method and its potentia errors more rigorousy foowing Xie and Mitche (990), with minor adaptations. The Q (f ) in the study area is parameterized as being ateray homogeneous. When there is an ith event recorded by two stations, j and, aong the same great circe path, we denote the ampitude spectra recorded at the th discrete frequency and station j as A ( j = ), and define a ij, scaed ogarithmic spectra ratio D V ij, ij, ( R R ) n = 2 π ij,, R R ij, ij, A A ij, 2 ij, 2 (2) where Vg is the group veocity and R i,j and R i, are the epicentra distances. For each of the station pairs ( ), we denote the number of events that are recorded aong the same ij,, j great circes as I( ). We average a avaiabe D 2 to obtain the stacked spectra ratio (SSR): where D = N I( ) ij, j D, 2 d i= Nd = I( ), (3) (4) ij,, j is the tota number of avaiabe D 2 at the th frequency. Xie and Mitche (990) show that the mathematica expectation of D is given by f V ED ( ) = + Q ( f ) N d π I( ) i= j C n C, (5) where the term C j ( j = ) describes the station site response. Ignoring the second term in the above equation (e.g., j assuming C are a unity), D can be used to fit a Q (f ) mode: n( D) = n Q0 + ( η ) n( f), (6) where the power-aw frequency dependence (equation ) has been assumed. In genera, ignoring the second term of equation (5) subjects the estimated Q (f) to a systematic bias δ Q 2 I( j, j ) j V C n j ( f ) =, π fn d i= C. (7) Q (f ) (or Q 0 and g) estimated using equation (6) is aso subject to a random error that decays in proportion to/. N d Resuts Two-station pairs aigned to the same event-to-station azimuth can sedom be obtained. In practice the approximations of the SSRs (equation 2) are obtained by requiring that the differences between the azimuths to the two stations are smaer than a preset maximum aowabe vaue, (dh) max. The choice of (dh) max is ess restrictive for the than for many other phases, since the contains a minima source radiation pattern in reaistic, 3D structures (e.g., Xie, 998). In this study two different vaues of (dh) max of 30 and 2 are used. These vaues resut in 37 and 22 two-station pairs, respectivey. SSRs are formed and used to fit Q (f ) modes using equation (6). Figure shows the path coverage by the two sets of SSRs, and Figure 2 shows the SSRs and the fit of the Q (f ) modes. SSRs obtained with (dh) max 30 are fitted by ( 037. ± 002. ) Q( f) = ( 26 ± 9) f 0. 2 Hz f 3. 6 Hz. (8) SSRs obtained with a (dh) max of 2 are very simiar to those with a (dh) max of 30 (Fig. 2) and are fitted by a Q (f) with Q 0 of (34 0) and g of ( ), respectivey. These vaues are virtuay the same as those in equation (8) but are theoreticay more subject to random errors and bias owing to a smaer N d Q (f ) in equation (8) is therefore my preferred mode. The upper bounds of the bias in Q estimates, caused by nonunity vaues of C j in equation (7), can be estimated foowing Xie and Mitche (990). For the preferred Q (f) mode, we have N d 37, V 3.5 km/sec, and D between 200 and 500 km (Fig. ). Aso, some combinations j of n( C / C ) cance. Contributions to nonunity vaues of C j by site response and focusing/defocusing are extensivey discussed by Xie and Mitche (990) and Xie (998). A fairy conservative estimate of the upper bound of C j is 2.0.

4 874 Short Notes Figure 2. Stacked spectra ratios (SSRs) from many two-station pairs potted in Figure, and the fit of best Q modes (straight ines). Back and gray symbos are SSRs obtained using a (dh) max of 30 and 2, respectivey. The Q modes from fitting both sets of SSRs are simiar. The Q mode written on the top of the pane is from fitting the back symbos. f Hz; Q ; g Substituting the terms on the right-hand side of equation (7) by these estimates, at f.0 Hz, we have δ < (9) Q 0 Such a bias is sma. As an independent and empirica test of effects of the nonunity C j, I form new SSRs by seecting eight pairs of spectra ratios under a criterion that, when the summations in equations (3), (5), or (7) are taken, the effect j of n( C / C ) competey cance (for exampe, station pair AMDO-BUDO is samped by reversed geometry, resuting in such a canceation). The new SSRs are fitted by a Q 0 of (22 8), which is virtuay the same as that in equation (8) except the uncertainty sighty increased owing to a smaer N d. These demonstrate the robustness of the Q (f) estimated using the two-station method. A drawback of this method is that it requires spectra to be recorded at two stations that are approximatey ocated on the same great circes, eading to ower cutoff frequencies at the more distant stations. As a resut, the highest frequency used in this study is beow 4 Hz, much ower than that of about 2 Hz used by McNamara et a. (996). Discussion and Concusion McNamara et a. (996), using data from the same seismic experiment as that used in this study, observed a imiting distance of about 700 km, beyond which disappears owing to a high attenuation. That imiting distance is at east a factor 2 smaer than the maximum distances of observation in other ow Q 0 regions, where the estimated Q 0 vaues are beow 267 (see Introduction). The Q 0 vaue estimated in this study is much ower than 267 and is therefore quaitativey in agreement with the observed short imiting distance in Tibet. Quantitativey, the Q 0 estimate of about 26 by this study is ower by a factor of 3 than the vaue of 366 estimated by McNamara et a. (996). To expore the cause of this arge discrepancy, I note that the signa processing and inversion procedure of this study differs from that of McNamara et a. in severa aspects. These differences incude () different upper frequency imits as mentioned in the ast section; (2) different components of the seismograms used: vertica and horizonta components are used in this and the previous studies, respectivey; (3) different methods used to measure ampitude: straightforward Fourier ampitude spectra are used in this study, whereas maximum ampitudes of bandpass fitered, time domain signas were used in the previous study; (4) different data censoring criteria used: the spectra used in this study are seected based on the S/N ratios cacuated using pre-p noise, whereas the previous study used S/N ratios cacuated using the mean eve of 50 sec of coda; (5) nonidentica samping areas by data used, and (6) different agorithms used for the inversion of Q (f). Difference on the upper frequency imits used shoud not affect the Q 0 estimate since it is measured around Hz, a frequency that is much ower than the high-frequency imits in either study. The effect of the different components used in estimating Q (f ) has been extensivey investigated by severa authors, incuding McNamara et a. (996), and has been found to be very sma. Different methods used for ampitude measurements shoud make itte difference in Q measurement as ong as the windowing and bandpass fitering do not cause significant processing artifacts and appropriate geometrica spreading terms (GSTs) are used. A GST with distance decay of D 0.83 was used by McNamara et a., treating the time domain signa as an Airy phase. In this study, a GST with D 0.5 is used in the frequency domain. To ensure that the choice of GST, in practice, does not significanty affect the Q (f) estimates, I reinverted the SSRs obtained using the GST with D 0.83 to see its effect. The estimated Q 0 and g vaues are (50 4) and ( ), respectivey. Obviousy, such changes are too sma to account for a Q 0 discrepancy of a factor 3. To test the possibiity that different data censoring criteria has somehow biased the Q (f ) mode obtained in this study, I reran the inversion by seecting the 22 pairs of SSRs from seismograms on which the has distincty arger ampitudes than its proceeding phases and coda. This seection procedure excudes any emergent signas. The resuting Q 0 and g estimates are (25 5) and ( ). I do not favor this mode since the emergent appearance of is expected at the more distant stations when Q is ow; there

5 Short Notes 875 is no reason to excude the respective spectra in Q estimates, whose robustness and reiabiity are dependent on the number of avaiabe two station pairs, Nd. In any case, the Q 0 estimate of about 25 obtained here is unchanged from that in equation (8). Different nonidentica samping areas coud change Q 0 estimates if they caused much difference in the geographica areas samped and if Q 0 vaues vary drasticay in these areas. The 37 pairs of SSRs (from 74 ampitude spectra) used in this study roughy sampe the area covered by the network used (Fig. ), whereas the 06 records used by McNamara et a. (in the 2 4 Hz band; their figure 7) mosty samped the same area, with a sma fraction (about 20% or ess) of paths aso samping the areas that are adjacent to the network but sti ying inside the eastern Tibetan pateau. It is very unikey that the reativey minor difference in samping areas coud expain a factor of 3 difference in the Q 0 estimates. Contrary to the first five differences, the difference in the inverse agorithms used can potentiay resut in drasticay different Q 0 estimates. In the method used by McNamara et a. (996), waveforms were fitered into five discrete frequency bands. Maximum ampitudes were then read from the fitered time series and used as data in the inverse scheme. In each frequency band, there were no more than 06 ampitude data and many unknown free parameters to be soved for. These unknowns incuded the Q (f ) vaue, pus twenty parameters describing the sources and eight parameters describing the site responses. In a there were more than 00 free parameters to be soved for, making the inverse probem ess stabe. The Q 0 vaue shoud have a significant trade-off with the source and site parameters because the observed ampitudes coud be fit by simutaneousy increasing the estimated Q 0 vaue and decreasing the source and/or site terms. By contrast, in this study effects of the source radiation and site responses are virtuay canceed by taking the SSRs, eaving ony two free parameters, Q 0 and g, that are soved for. The Q 0 vaue of about 26, estimated for the eastern Tibetan pateau in this study, is the owest ever documented for any continenta areas. It is comparabe or sighty ower than the recenty reported vaues near the study area by Phiips et a. (2000) and Fan and Lay (200). An impication of this ow vaue is that the observed bockage of for paths crossing the boundaries of the pateau (see Introduction) may be argey, or even entirey, attributed to the ow Q 0 vaues in the pateau (see Fig. 3 for a detaied discussion). Another impication of the ow Q 0 is that the crust in Tibet may be characterized by higher-than-norma temperature and fuid content, which are responsibe for the ow Q 0 vaues. Future research shoud be directed toward anayzing more seismic data from recent seismic experiments in the pateau to resove detais of the atera variations of Q (f) in the pateau. Acknowedgments I thank Dan McNamara for his kind discussions on the possibe causes of the difference between the Q modes obtained in this study and in Figure 3. An exampe showing how ow Q vaues aone can cause an bockage across the northern boundary of Tibetan Pateau. Potted are topographic profies from the Lop Nor exposion (asterisk; shown as star in Fig. ) to station ERDO (curve in gray) and AMDO (curve in back). The observed ampitudes show a partia bockage at ERDO (McNamara et a., 996, their figure 0) and compete bockage at AMDO. Numbers in parentheses are Q 0 vaues: the eft segments of the profies are in the Tarim Basin where Q 0 is 450 or higher from previous works (e.g., Xie et a., 996). Between ERDO and AMDO, Q 0 shoud be cose to 26 (this study). Using () these Q 0 vaues, (2) the spectra ampitude in the expected window at AMDO, and (3) the known source eve of the exposion (Xie et a., 996), I estimate that Q 0 between the topographic boundary and station ERDO is 208 or ower. This means that a ow average Q 0 vaue in the pateau of between 26 and 208 is capabe to bock at AMDO, and a strong scattering at the topographic boundary is not required. This is consistent with the simuation of Kennett (986), who shows that the scattering can not fuy account for the bockage.

6 876 Short Notes McNamara et a. (996). This research was supported by the Defense Threat Reduction Agency Grants DSWA and DSWA This is Lamont-Doherty Earth Observatory Contribution No References Aki, K., and P. G. Richards (980). Quantitative Seismoogy: Theory and Methods, W. H. Freeman, New York. Fan, G., and T. Lay (2002). Characteristics of attenuation in the Tibetan pateau, J. Geophys. Res. (in revision). Herrmann, R. B. (980). Q estimates using the coda of oca earthquakes, Bu. Seism. Soc. Am. 70, Kennett, B. L. N. (986). waves and structura boundaries, Bu. Seism. Soc. Am. 76, McNamara, D. T., T. J. Owens, and W. R. Water (996). Propagation characteristics of across the Tibetan Pateau, Bu. Seism. Soc. Am. 86, Ni, J., and M. Barazangi (983). High-frequency seismic wave propagation beneath the Indian shied, Himaayan arc, Tibetan pateau and surrounding regions: high uppermost mante veocities and efficient Sn propagation beneath Tibet, Geophys. J. R. Astr. Soc. 72, Nutti, O. W. (980). The excitation and attenuation of crusta seismic phases in Iran, Bu. Seism. Soc. Am. 70, Nutti, O. W. (986). Yied estimates of Nevada test site exposions obtained from seismic waves, J. Geophys. Res. 9, Phiips, W. S., H. E. Hartse, S. R. Tayor, and G. E. Randa (2000). Hz Q tomography in Centra Asia, Geophys. Res. Lett. 27, Ruzaikin, A. I., I. L. Nersesov, V. I. Khaturin, and P. Monar (977). Propagation of and atera variation in crusta structure in Asia, J. Geophys. Res. 82, Xie, J. (998). Spectra inversion using from earthquakes: improvement of the method with appications to the 995, western Texas earthquake sequence, Bu. Seism. Soc. Am. 88, Xie, J., and B. J. Mitche (990). Attenuation of mutiphase surface waves in the Basin and Range Province. I. and coda, Geophys. J. Int. 02, Xie, J., L. Cong, and B. J. Mitche (996). Spectra characteristics of the excitation and propagation of from underground nucear exposions in Centra Asia, J. Geophys. Res. 0, Lamont-Doherty Earth Observatory of Coumbia University Route 9W Paisades, New York 0964 Manuscript received 5 Apri 200