Earth and Planetary Science Letters

Transcription

1 Earth and Planetary Science Letters 306 (2011) Contents lists available at ScienceDirect Earth and Planetary Science Letters journal homepage: Seismic anisotropy and mantle dynamics beneath China Zhouchuan Huang a,b,, Liangshu Wang a,, Dapeng Zhao b,, Ning Mi a, Mingjie Xu a a School of Earth Sciences and Engineering, Nanjing University, Nanjing , China b Department of Geophysics, Tohoku University, Sendai , Japan article info abstract Article history: Received 24 December 2010 Received in revised form 29 March 2011 Accepted 31 March 2011 Available online 29 April 2011 Editor: P. Shearer Keywords: Shear-wave splitting Anisotropy Continent Subduction of the Pacific plate India Asia collision Absolute plate motion We analyzed the shear-wave splitting at 138 permanent seismograph stations to study seismic anisotropy and mantle dynamics under Mainland China. To obtain reliable results we used three different methods to measure the shear-wave splitting parameters using core phases (SKS, SKKS, SKiKS and PKS) as well as the direct S waves from regional and distant earthquakes. Our results show that the fast orientations of the anisotropy (WNW ESE) in eastern China are generally consistent with the absolute plate motion (APM) direction of the Eurasian plate, suggesting that the anisotropy is mainly located in the asthenosphere resulting from the lattice-preferred orientation of olivine due to the shear deformation there. The fast axes in western China generally agree with the strikes of the orogens and active faults, while they are perpendicular to the direction of the maximum horizontal stress, suggesting that the anisotropy in the lithosphere contributes significantly to the observed shear-wave splitting. The fast axes in western China are also consistent with the APM direction, suggesting that the APM-driven anisotropy in the asthenosphere is another source of the shear-wave splitting there. These results suggest that APM-driven anisotropy commonly exists under continents, similar to that under oceanic regions, even though the continental lithosphere has suffered extensive deformation Elsevier B.V. All rights reserved. 1. Introduction Seismic anisotropy describes the directional dependence of seismic velocity within the Earth and it is a characteristic feature of the Earth's interior structure. It may exist at different depth ranges in the crust, mantle and inner core (see Mainprice, 2007 and references therein). Different factors dominate in producing the anisotropy at various depths, such as aligned cracks in the upper crust (e.g., Crampin, 1984) and the lattice-preferred orientation (LPO) of minerals in the lower crust and upper mantle (see Karato et al., 2008; Mainprice, 2007 for comprehensive reviews). The LPO (or the anisotropic fabric) in the upper mantle is generally considered to be the result of dislocation creep of the minerals (mainly olivine) above a depth of ~300 km (e.g., Gung et al., 2003; Panning and Romanowicz, 2006). The anisotropic fabric depends on both the type and extent of strain (Savage, 1999). Many experimental studies and theoretical models have focused on the development of different fabrics for simple shear, pure shear, axial and uni-axial compression. In general, the fast orientation of the anisotropy (e.g., a-axis of olivine) is subparallel to the extension or shear direction in the upper mantle (e.g., Karato et al., 2008; Nicolas, 1993; Savage, 1999; Silver and Chan, 1991; Zhang and Karato, 1995). For simple horizontal mantle flow, the fast direction is usually parallel to the flow direction (Karato et al., 2008). Corresponding author at: Z. Huang, School of Earth Sciences and Engineering, Nanjing University, Nanjing , China. Corresponding authors. addresses: zhouchuan.huang@gmail.com (Z. Huang), lswang@nju.edu.cn (L. Wang), zhao@aob.gp.tohoku.ac.jp (D. Zhao). Because of the close relationship between the anisotropy and strain in the upper mantle, observations of seismic anisotropy can in principle be used to constrain the lithospheric and sublithospheric mantle deformation that produces this anisotropy (Conrad et al., 2007). Shear-wave splitting is a popular tool for characterizing anisotropy in the Earth (e.g., Long and Silver, 2009; Silver and Chan, 1988, 1991; Vinnik et al., 1992). Shear-wave splitting, also called seismic birefringence, is a phenomenon in which a shear wave splits into two polarized shear waves with different velocities when traveling though an anisotropic medium. Two splitting parameters (φ, δt) can be measured from seismograms, which correspond to the polarization direction of the fast quasi-s phase (φ)and the delay time (δt) between the fast and slow components, respectively (Long and Silver, 2009). By using the long-period core phases, such as SKS, SKKS, SKiKS and PKS (hereafter, we call them XKS phases), the observed shear-wave splitting is usually considered to reflect the anisotropy in the crust and upper mantle under the seismograph stations. Hence the splitting parameters can be used to study the anisotropy and deformationintheuppermantle. Many researchers have used splitting parameters to constrain the global mantle flow and the absolute plate motion (APM) (e.g., Becker et al., 2006, 2008; Conrad et al., 2007; Kreemer, 2009; Kustowski et al., 2008). Comparisons of the XKS observations with the LPO derived from the numerical modeling show that the predictions of upper mantle anisotropy made by the global mantle circulation models match the observations well beneath the oceans but poorly under the continents (Conrad et al., 2007; Long and Becker, 2010). The fit under the continents, however, can be improved when considering lateral variations in the lithospheric X/$ see front matter 2011 Elsevier B.V. All rights reserved. doi: /j.epsl

2 106 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) thickness (Conrad et al., 2007). These results suggest that the anisotropy in the continental lithosphere, which may suffer extensive deformation, contributes significantly to the observed shear-wave splitting (e.g., Fouch and Rondenay, 2006; Savage, 1999; Silver, 1996; Vinnik et al., 1992). It is far from clear, however, which of the sources the anisotropic structure in the lithosphere or the contemporary flow in the asthenosphere dominates in the observed anisotropic signal in the continental regions (Long and Becker, 2010). China provides an ideal site to estimate the contribution from the continental lithosphere. The Chinese continent is composed of various lithospheric blocks formed during its long geological history (e.g., Ma, 1987, 1988; Ren et al., 1999). While retaining several stable Archean blocks, China has been suffering extensive deformation in the Cenozoic (see Yin, 2010 for a comprehensive review). The eastern and western parts of China (Fig. 1), however, have experienced completely different tectonic evolutions. The Cenozoic tectonics of western China (or even the entire Asia region) is most dramatically expressed by the development of the Tibetan Plateau resulting from the India Asia collision (Tapponnier et al., 1982, 2001). In the Middle to Late Miocene(i.e., 18 8 Ma), the N S contraction of the early stage was replaced by the coeval development of conjugate strike-split faults and E W extension which continues until today (Tapponnier and Molner, 1977; Yin, 2010 and references therein). The far-field effect of the India Asia collision is considered to have reached the Tienshan orogen and the Baikal rift zone thousands of kilometers northward (Molnar and Tapponnier, 1975; Tapponnier and Molner, 1979). Eastern China, in contrast, is characterized by the development of the back-arc extensional system as a result of the subduction of the Pacific and Philippine Sea plates (e.g., Ren et al., 2002; Tian et al., 1992; Zhao et al., 2011a). Debates, however, are continuing on whether the India Asia collision has influenced the tectonic evolution of eastern China (e.g., Liu et al., 2004). The existence of various tectonic elements (blocks, orogens, faults, etc.) in China allows us to better understand the origin of seismic anisotropy in the lithosphere and asthenosphere under a continental region. In this study, we analyzed the shear-wave splitting at 138 permanent digital seismograph stations in Mainland China. The measured splitting parameters are then used to estimate the first-order anisotropic patterns in the upper mantle, including the lithosphere and asthenosphere. Although many previous studies have investigated the anisotropic structure in various parts of China, the present work has the following advantages over the previous studies. (1) We analyzed the shear-wave splitting at 138 permanent stations across China. Many of the stations have been deployed for more than ten years (Fig. 1), and so much more high-quality data are available for us to better understand seismic anisotropy and mantle dynamics under the entire Chinese continent. (2) Three different methods are used to make the shear-wave splitting measurements, and the results by the different methods are carefully compared and analyzed to avoid any potential bias. While our results are generally consistent with many of the previous results, the present work has provided important new insights into the seismic anisotropy and deformation under the Chinese continental region. 2. Data and method The 138 seismograph stations used in this study (Fig. 1) are broadband, permanent stations operated by the China Seismic Network Data Center. Most of the stations are equipped with CTS-1, KS-2000 and JCZ-1 seismometers, while several stations have GS-13, STS-2 and CMG-3 type Fig. 1. Distribution of the 138 seismograph stations used in this study. The surface topography is shown in color with its scale shown at the bottom. The white triangles denote the stations of group A with data available during 2000 to 2009, while the black triangles show the stations of group B with data only in The dashed lines indicate the boundaries between different tectonic blocks (Ren et al., 1999). The bold arrows denote the motion directions of the Indian, Pacific and the Philippine Sea plates relative to the Eurasian plate. The vertical dashed line shows the rough boundary between western and eastern China investigated in this study.

3 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) seismometers (Table S2). These stations can be divided into two groups: Group A consists of 46 stations with data available from 2000 to 2009 (white triangles in Fig. 1), while group B consists of 92 stations with available seismic records of one-year in 2009 (solid triangles in Fig. 1). Earthquakes with magnitude 6.0 at epicentral distances from 85 to 140 were selected for analysis and 377 events were finally included in our data set (cross symbols in Fig. 2; 53 events in 2009). Most of the earthquakes with clear core phases (such as SKS and SKKS) occurred in the Tonga subduction zone with back-azimuths of We also checked many earthquakes with epicentral distances of and selected some of the events if their PKS phases are clearly visible, which improves the azimuthal coverage of the events used. We used the SplitLab software by Wüstefeld et al. (2008) to measure the splitting parameters (φ, δt). The rotation cross-correlation method (RC) (e.g., Levin et al., 1999) and the transverse-component minimization method (SC) (Silver and Chan, 1991) were used simultaneously, which increases the reliability when both methods give consistent results (e.g., Long and Silver, 2009; Vecsey et al., 2008). Both of the methods utilize a grid-search approach to identify the best-fitting splitting parameters by rotating and time-shifting the Q T components in a ray-coordinate based L Q T coordinate system(vecsey et al., 2008). The RC method tries to identify a pair of splitting parameters (φ, δt) that maximizes the cross-correlation between the corrected Q and T components, while the SC method seeks to minimize the amount of energy on the transverse component when the effect of splitting is accounted for (Long and Silver, 2009). The Q and T components are rotated between 90 and 90 with a step of 1 and the time is shifted from 0 s to 4 s with a step of 0.02 s. Figs. S1 and S3 show examples of shear-wave splitting measurements on the SKS and PKS phases carried on SplitLab, respectively, while Fig. S2 shows an example of a null measurement. The multi-channel method (MC) introduced by Chevrot (2000) is an alternative to the single-record methods (RC and SC). The MC method takes advantage of the predicted variation in the amount of energy on the uncorrected transverse component with incoming polarization angle for a single, horizontal layer of anisotropy (Long and Silver, 2009). For the vertically propagating, long-period shear wave (δt T) that has traversed a single layer of anisotropy with a horizontal axis of transversely isotropic symmetry, the predicted transverse component T(t) can be written as a function of the time derivative of the radial component R(t) (Chevrot, 2000; Vinnik et al., 1989): Tt ðþ 1 2 ðδtsin2βþdr ðþ t dt where β is the angle between φ and the incoming polarization direction (equivalent to the back-azimuth for XKS phases). After calculating the splitting intensity/vector (S=δt sin 2β) for the seismograms of various back-azimuths, we can retrieve the best-fitting splitting parameters (φ, δt) by fitting a sin 2β curve to the splitting vector (Long and Silver, 2009). The values of φ and δt can be inferred from the phase and amplitude of the sinusoid, respectively. The splitting intensity measurements can be either stacked in azimuthal bins to improve the signal-to-noise ratio (Chevrot, 2000) or used individually (Long and van der Hilst, 2005; Monteiller and Chevrot, 2010). Nevertheless, the MC method requires good coverage of events in the incoming polarizations (Chevrot, 2000). As shown above, the backazimuthal coverage of events for most of the stations used in this study is poor if only the SKS/SKKS phases are used, but it can be improved when the PKS phases are added. In addition, direct teleseismic S phases from deep earthquakes can potentially improve the back-azimuthal coverage after their initial polarizations are measured directly from the seismogram (Long and van der Hilst, 2005; Vidale, 1986). The teleseismic S phases from 164 deep events (N200 km) with epicentral distances of (see Long and Silver, 2009 for details) were also included in our data set (red circles in Fig. 2; 11 events in 2009). These earthquakes mainly occurred in the Indonesian subduction zone and they dramatically improve the back-azimuthal coverage (Fig. 2). 3. Results 3.1. Measurements with the RC and SC methods Fig. 2. Epicenter distribution of the earthquakes used in the study. The blue crosses denote the 377 events with XKS (i.e., SKS, SKKS, SKiKS and PKS) phases, while the red circles represent the 164 deep earthquakes with direct S phases. The three concentric circles indicate the epicentral distances of 50, 100 and 150 from the center of the study area. The open triangles denote the seismograph stations used. The purple dashed lines show the plate boundaries (Bird, 2003). We define the quality of measurements with the RC and MC methods by using the differences between the results by the two methods (Wüstefeld and Bokelmann, 2007): the angular difference of φ (Ψ= φ RC φ SC ) and the ratio of δt (ρ=δt RC /δt SC ) (Fig. S4). The good splitting measurements are identified as 0.8 ρ 1.1 and Ψ 10, while the fair measurements have 0.7 ρ 1.2 and Ψ 15. The quality of nulls is also defined: the good nulls with 35 Ψ 55 and ρ 0.2, while the fair nulls with 30 Ψ 60 and ρ 0.3. The remaining ones are considered to be poor measurements. Following these criteria, we obtained 266 good and 152 fair measurements, in addition to 1457 good and 341 fair nulls. The above criterion for the non-null measurements is very rigorous (Fig. S4). In fact, the SC method can always yield a correct estimate of φ even if it is parallel or perpendicular to the back-azimuth (Wüstefeld and Bokelmann, 2007). We also tried to identify the good and fair non-null measurements based on the angular difference between φ (with the SC method) and the back-azimuth ( φ-baz, projected to 0 45 ) and the signal-to-noise ratio (SNR) of the original transverse component (Fig. S4) (e.g., Liu et al., 2008). The good measurements are identified as φ-baz 30 and SNR 10, while the fair measurements have φ-baz 20 and SNR 5. The uncertainties of φ and δt for these measurements are constrained to be smaller than 45 and 1.0 s, respectively. Following the new criterion, we obtained 263 good and 459 fair non-null measurements, which contain almost all those defined with the method of Wüstefeld and Bokelmann (2007) (Fig. S4b). Note that the good and fair measurements only make up approximately 17% of the total measurements we made, while the nulls make up more than 40%.

4 108 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) Due to the reasons mentioned above, we focus on the results with the SC method. The δt values range from 0.4 to 1.4 s with a peak at s (Fig. 3a). As a whole, the fast directions are generally oriented E W in eastern China, whereas they are parallel to the strikes of the orogens and active faults in western China (Fig. 4a). The clockwise rotation of φ around the eastern Himalayan syntax and its abrupt change from N S to E W in the southeastern margin of the Tibetan Plateau are also revealed, similar to previous results (Flesch et al., 2005; Fu et al., 2008, 2011; Huang et al., 2000, 2007; Lev et al., 2006; McNamara et al., 1994; Sol et al., 2007; Wang et al., 2008). Azimuth-dependent splitting is found at many stations (Fig. S5) and the corresponding splitting parameters (both φ and δt) are scattered. For example, at HIA, the measured φ ranges between 60 and 30 and δt between 0.5 s and 2 s (Fig. S6; Table S1), which cannot be attributed to measurement errors alone. We applied the two-layer fitting algorithm (Silver and Savage, 1994) and found that the measurements at HIA can be well fitted with two-layer anisotropy (Fig. S6). Two-layer anisotropy has also been identified at many other IRIS stations (BJT, TYA, HIA, KMI, LSA, WUS, XAN, and ULN) by this and previous studies (Table S3) (Bai et al., 2010; Barruol et al., 2008; Gao and Liu 2009; Li et al., 2010, 2011), which suggests complex anisotropic structures beneath China and adjacent regions. The measured nulls are also summarized in Fig. 4b. However, the null pattern is somewhat biased from the fast axis φ with the SC method, especially under eastern China. Note that many of the earthquakes used in this work occurred in the Tonga subduction zone. At periods N8 10 s, the splitting detection limit for broadband data with typical noise levels using the RC and SC methods is ~0.5 s (Long and Silver, 2009). It is possible that, with weaker anisotropy, the pattern of the nulls is biased to reflect the earthquake distribution. At several stations of group A (with data of 10 years), the null measurements cover a large or even the entire back-azimuthal range (HIA, MDJ, SSE, LSA, and WMQ) (Fig. 4b). The results may indicate isotropy beneath the stations because the splitting is smaller than the lower detection limit of the RC and MC methods, or they are the result of extreme lateral heterogeneity that is not coherent over the length scales associated with the seismic wavelengths under study (Long, 2010) Measurements with the MC method As mentioned above, the MC analysis needs good azimuthal coverage of earthquakes at each station. For the stations of group A, the highquality records of 10 years have led to many reliable splitting measurements (Fig. S7a), and so the splitting parameters (φ, δt) were determined for most of the stations (Table S2). There are several stations where the splitting vector does not follow a sinusoid (Fig. S7d), which may indicate the existence of complex structures under these stations, such as azimuth-dependent or multi-layer anisotropy (Long and Silver, 2009; Long and van der Hilst, 2005). For some of the stations in group B, robust splitting parameters are also obtained (Fig. S7b), although seismograms are available for only one year. For many stations in both groups, however, the XKS phases themselves are not enough to determine reliable splitting parameters. For those stations, the direct S phases from the teleseismic deep earthquakes could be used to better constrain the measurements (Fig. S7c; Table S2). Using the S phases, however, has a risk that source-side anisotropy may be mistaken for receiver-side anisotropy, although we have used only deep events (N200 km) to minimize this effect. Comparing the measurements from the S phases at neighboring stations can help to clarify this issue (Long and Silver, 2009; Long and van der Hilst, 2005). If the splitting vectors at neighboring stations from the events in the same regions show notably different patterns, the observations are most possibly the result of anisotropy under the stations. Otherwise, if the splitting parameters determined by using only the XKS phases and using both the S and XKS phases (S+XKS) for the same station are similar, the source-side anisotropy of the teleseismic S phases can also be excluded (Long and van der Hilst, 2005). We compared the results determined by using the XKS and the S+XKS phases to confirm if both approaches yield reliable solutions. The comparison shows that the differences between them are generally 20 for φ and 0.3 s for δt. These results suggest that using the S phases from the teleseismic deep earthquakes is a good choice for the MC analysis if there are insufficient XKS phases available. But there are also some stations where source-side anisotropy cannot be ruled out from the observations. Therefore in this study, we mainly rely on measurements made from only the XKS phases. Only the measurements at six stations(gul,ksh,lyn,wzh,cn2andhuc)aremadefromboththe XKSandSphasesbecausetheXKSphasesalonedidnotyieldrobust results, and we have carefully checked these results to avoid possible source-side anisotropy. The δt measurements with the MC method mainly range between 0.4 s and 1.4 s (Fig. 3b), being consistent with that with the SC method. The dominant value of δt( s), however, is slightly smaller than that from the SC method( s) (Fig. 3a) because individual measurements with the SC method can often overestimate the delay time when there is noise in the components (Monteiller and Chevrot, 2010). The φ measurements with the MC method show a similar pattern to that with the SC method, such as fault- and orogen-parallel φ in western China (Fig. 5). However, there seems a systematic discrepancy in the fast axes in eastern China: the dominant WNW ESE φ by the MC method as compared with the nearly E W φ bythescmethod(fig. 5). Note that, at stations with two-layer anisotropy, the MC method can also yield robust measurements that are the combined result of upper- and lower-layer anisotropy (Silver and Long, 2011) Comparison with previous results Fig. 3. Histograms of δt values measured by using the SC (a) and MC (b) methods (see text for details). Many researchers have studied seismic anisotropy under China using shear-wave splitting. The previous studies can be divided into two types. One approach is to analyze the shear-wave splitting on individual seismogram using the RC or SC method (e.g., Huang et al., 2000;

5 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) Fig. 4. (a) Distribution of seismic stations with good (red) and fair (blue) measurements of shear-wave splitting with the SC method. The orientation and length of the short bars denote the fast polarization direction (φ) and the delay time (δt), respectively. The scale for δt is shown in the inset. The curves show the epicentral distances (in degrees) from a point (180, 20 S) in the Tonga subduction zone. (b) Distribution of the good and fair null measurements in our data set. Nulls are plotted as crosses at each station with the bars oriented in the direction of the XKS back-azimuth and its orthogonal direction. The other labeling is the same as Fig. 1. McNamara et al., 1994; Zhao et al., 2007). The other is to use all the data available at one station to find the optimal solution with the stacking or the MC method (e.g., Chen et al., 2010; Fu et al., 2011; Lev et al., 2006; Li and Niu, 2010). In general, the results obtained by using the stacking or the MC methods are consistent with each other, while they may show significant discrepancies from the results by the RC and SC methods.

6 110 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) Fig. 5. Distribution of the measured shear-wave splitting parameters with the MC method from the XKS phases (bold red bars) and the S+XKS phases (open red bars). The measurements with the SC method are also shown in blue bars, while the results of the previous studies are shown in gray bars. The splitting parameters (φ, δt) are indicated by the orientation and length of the bars with the scale shown in the upper-left inset. The orange lines denote the active faults. The other labeling is the same asfig. 1. Synthetic studies show that the splitting parameters determined by the RC and SC methods on individual seismograms may deviate considerably from the true values, especially when the initial polarizations (backazimuths for the XKS phases) are parallel or perpendicular to the fast axis of the anisotropy in the media under the stations (e.g., Monteiller and Chevrot, 2010; Vecsey et al., 2008; Wüstefeld and Bokelmann, 2007). The analysis also demonstrates that the stacking method (Wolfe and Silver, 1998) as well as the MC method can determine accurate splitting parameters (Monteiller and Chevrot, 2010). Many of the earthquakes used in this study occur in the Tonga subduction zone with a back-azimuth of (Fig. 2). Because the back-azimuths of the events (the initial polarization of the XKS phases) are generally parallel with the fast axis of the anisotropy (φ by the MC method) in eastern China, analysis based on individual seismograms is highly risky. For example, Zhao et al. (2007) showed that the dominant φ in South China is ENE WSW to NE SW, which is similar to the results by the SC method in this study (Fig. 4a). But the results by the MC method, as well as the previous studies using the stacking method (e.g., Chang et al., 2009; Fu et al., 2011), show dominant ESE WNW φ in South China (Fig. 5). Hence the results by the RC and SC methods on the individual seismogram are probably biased. Because the incoming polarizations are subparallel to the fast orientations, the energy of the transverse component is very weak and so the noise may dominate in the measurement (Monteiller and Chevrot, 2010; Wüstefeld and Bokelmann, 2007). This possibility is confirmed by the fact that null measurements make up as much as 40% of the total measurements in our analysis. Fortunately, the situation is better for western China. The previous studies using the RC and SC methods on individual seismograms focused on the anisotropic structure in southern Tibet and the Tienshan orogen. The φ in those regions (NE SW)isneitherparallelnorperpendicularto the incoming polarizations, so the results obtained from the individual seismograms in those regions are reliable (e.g., Huang et al., 2000; McNamara et al., 1994). 4. Discussion The shear-wave splitting signals extracted from the XKS phases and the S phases from the teleseismic deep earthquakes are generally considered to reflect the anisotropy in the upper mantle, either in the lithosphere or asthenosphere (e.g., Long and Silver, 2009; Savage, 1999; Silver, 1996). In regions where the lithosphere lacks significant deformation and the lithospheric anisotropy is weak, the observations mainly reflect the anisotropy in the asthenosphere (e.g., Conrad et al., 2007). As a result of simple shear induced by the APM, the LPO in the asthenosphere is rotated toward the infinite strain axis. Thus the fast axes of anisotropy (φ) are parallel with either the mantle flow or the shear direction (i.e., the APM directions) (Karato et al., 2008; Zhang and Karato, 1995). In contrast, in regions with extensive lithospheric deformation, anisotropy in the lithosphere may be another source for shear-wave splitting observations. Three major categories of lithospheric deformation that may be encountered are transcurrent, uniaxis compression (actually transpression in most cases), and extensional regimes (Silver, 1996). For either transcurrent or transpression deformation, the fast orientations of anisotropy (φ) are generally parallel with the major surface features, such as the strikes of orogens and faults (e.g., Nicolas, 1993; Savage, 1999; Silver, 1996). In the case of an extensional regime, φ would be parallel with the extension direction (Savage, 1999; Silver, 1996). We compared the φ by the MC method with the APM directions derived from the HS3-NUVEL1A model (Gripp and Gordon, 2002) atthe 90 stations where the MC method yielded reliable measurements (Fig. 6a; Table S2). As a whole, φ is consistent with the APM direction with an angular difference smaller than 25 (Fig. 6b). The results suggest

7 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) that the observed shear-wave splitting is closely related to APM-driven anisotropy in the asthenosphere due to the mineral LPO there. The correlation between φ and the APM direction is relatively low for the stations in western China; some of the angular differences reach 35 (Fig. 6b). In western China φ appears to be better related to the surface features and lithospheric structures, e.g., φ is generally parallel with the strikes of orogens and boundary-faults (Fig. 10) while perpendicular to the maximum horizontal stress σ H (Heidbach et al., 2010)(Fig. 6e and f). Therefore, anisotropy in the lithosphere may significantly contribute to the observed shear-wave splitting in western China. We also compare our measurements with the GPS results relative to the ITRF97 (International Terrestrial Reference Frame, epoch ) (Wang et al., 2001)(Fig. 6c). In eastern China, the GPS results are consistent with both φ and the APM direction, suggesting that the crust and upper mantle are highly coupled and APM-driven anisotropy exists in the asthenosphere. In western China, φ seems to be consistent with the GPS results (especially in south Tibet), but not correlated with the APM direction (Figs. 6c and 10). This result also argues for lithospheric anisotropy in western China rather than asthenospheric anisotropy. In summary, to first order, the observed shear-wave splitting is closely related to asthenospheric anisotropy resulting from mineral LPO caused by the APM. In western China, lithospheric anisotropy as a result of lithospheric deformation may significantly contribute to the observations. In the following we discuss the mechanisms that may affect the observations in different parts of China Northeast China Northeast (NE) China is characterized by widespread intraplate volcanism in the Great Xing'an Ranges and Changbai Mountain and the accompanying deformation caused by multi-episode extension since the Late Mesozoic (Ren et al., 2002; Tian et al., 1992; Wang et al., 2006). Global and regional seismic tomography shows that the subducting Pacific slab from the Japan Trench becomes stagnant in the mantle transition zone under eastern China, and a big mantle wedge has formed in the upper mantle above the Pacific slab and the stagnant slab under East Asia (Huang and Zhao, 2006; Zhao, 2004). Mantle convection in the big mantle wedge and deep slab dehydration may induce hot and wet upwelling flows, leading to the formation of active intraplate volcanism in NE China, such as the Changbai and Wudalianchi volcanoes (Duan et al., 2009; Lei and Zhao, 2005; Zhao et al., 2004, 2009). Iidaka and Niu (2001) studied anisotropy in the crust using Mohoconverted phases and suggested that the crust in eastern China is almost isotropic. Li and Niu (2010) proposedthatfossilanisotropyfrozeninthe lithosphere due to extension is the source for their observed shear-wave splitting with NW SE φ in NE China. In contrast, Liu et al. (2008) suggested that the measurements could be explained by the LPO of metastable olivine within the Pacific slab(e.g., Jiang et al., 2008) and by back-arc asthenospheric flow in the mantle wedge above the slab. These mechanisms, however, cannot explain the observations that uniform NW SE oriented anisotropy exists in not only NE China but also broad areas in East Asia (Fig. 5). Two-layer anisotropy was revealed at HIA and ULN, suggesting that the NW SE oriented anisotropy (being parallel with APM and the subduction direction) spreads across the Great Xing'an Range and continues to Central Mongolia thousands of kilometers westward (Barruol et al., 2008; Gao et al., 1994) (Fig. 5). Thus it is improper to attribute the large-scale splitting observations to any localscale structures. Note that the frontier of the subducted Pacific slab has reached to the eastern margin of the Great Xing'an Range, not further westward (e.g., Huang and Zhao, 2006). The anisotropy in the lithosphere and that related to the subducted Pacific slab may contribute to or affect our observations, but can hardly be the dominant factor in such broad regions. For example, the observed δt is smaller in the southeastern part (i.e., the Changbai Mountain) than that in northwestern part of NE China (Fig. 7). As mentioned above, ongoing upwelling flow is expected in the upper mantle under the Changbai volcano (Zhao et al., 2009), and it must have canceled some of the APM-driven mineral LPO. The widespread nulls at station MDJ (Fig. 4b) also suggest weak anisotropy in this area. The NNE SSW φ (at GNH, MOH and also in the upper layer under HIA) measured in the Great Xing'an Range is subparallel with the strike of the range and faults (Fig. 7). It may reflect anisotropy in the lithosphere that wrapped around the Siberian craton (Barruol et al., 2008; Li and Niu, 2010; Liu et al., 2008). A similar measurement is obtained at station CN2 near the Tanlu fault (Fig. 7); the fault-parallel φ couldbecausedby transcurrent deformation in the lithosphere (Ren et al., 2002; Silver, 1996; Zinke and Zoback, 2000) North China Fig. 6. Comparisons of the fast polarization direction φ (red bars) (a) with the direction of the absolute plate motion (APM) (Gripp and Gordon, 2002), (c) with the GPS results (purple arrows) (Wang et al., 2001), and (e) with the orientations of maximum horizontal stress σ H (black bars) (Heidbach et al., 2010). The GPS results in (c) and σ H in (e) are the average values within 2 around each seismic station (Audoine et al., 2004). The right panels (b, d and f) show the histograms of the angular differences between φ and APM (b), between φ and GPS results (d), and between φ and σ H (f). The orange columns show the results for all the stations in China, while the blue columns show the results for the stations in western China. North China, usually named the North China Craton, is in a quite similar situation to that of NE China. Most of the above arguments about the anisotropic structure under NE China remain suitable for North China. However, North China may have suffered more specific tectonic evolution. It is considered to be an Archean craton which consists of three sub-blocks: the Western Block, Eastern Block and the Trans-North China Orogeninthemiddle(Fig. 8)(Zhao et al., 2001).Theorogeninthemiddle is a suture zone formed during the Precambrian when the Western and Eastern blocks amalgamated (Zhao et al., 2001). The Western Block, or the Ordos Block, is a stable block characterized by a thick mantle root and

8 112 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) Fig. 7. Distribution of the shear-wave splitting measurements (red bars) in Northeast China. The splitting parameters (φ, δt) are indicated by the orientation and length of the bars. At station HIA where multi-layer anisotropy has been revealed, the fast axes in the upper and lower layers are shown as open and solid black bars, respectively. The bold blue arrows and thin purple arrows indicate the APM directions and the GPS results with the scales shown in the upper-right inset. The dark red bars denote the orientations of σh. The orange lines denote the active faults, and the yellow rose diagram shows the statistics of the fault strikes. The two solid triangles denote the Wudalianchi and Changbai volcanoes. The other labeling is the same as Fig. 1. Fig. 8. Distribution of the shear-wave splitting measurements in North China. The labeling is the same as Fig. 7.

9 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) the lack of internal deformation since the Precambrian (Huang and Zhao, 2006; Tian et al., 2009; Zhao et al., 2001). The Eastern Block, in contrast, has been strongly reactivated since the Late Mesozoic and Cenozoic; the lithosphere has been thinned from ~200 km in the Archean to the present ~80 km (e.g., An and Shi, 2006; Chen et al., 2008; Griffin etal., 1998; Menzies et al., 2007). The anisotropic features in the three sub-blocks are different. In the Eastern Block, φ aligns WNW ESE, which is consistent with the directions of APM and the Pacific plate subduction (Fig. 8) (Bai et al., 2010; Chang et al., 2009; Liu et al., 2008; Zhao and Xue, 2010). The observations thus reflect asthenospheric anisotropy due to the mineral LPO in the mantle flow. However, the role of Pacific platesubductionis not clear. Seismic tomography has revealed that the stagnant Pacific slab has reached to the middle orogen and the upper mantle under the Eastern Block is characterized by extensive low-velocity anomalies (Huang and Zhao, 2006; Tian et al., 2009; Xu and Zhao, 2009). Both the high heat flow and thin lithosphere suggest a hot mantle beneath the Eastern Block (An and Shi, 2006; Chen et al., 2008; Hu et al., 2000). Note that high temperatures can reduce viscosity, and so the development of LPO becomes much easier under the shear induced by the APM (Karato et al., 2008). Therefore the Pacific plate subduction has played a very important role in the formation of anisotropy under the Eastern Block. In the middle orogen, orogen- and fault-parallel φ (NNE SSW) is measured (Fig. 8, SHZ and upper layer at BJT and TAY) (Bai et al., 2010; Zhao et al., 2008; Zhao and Xue, 2010). The orogen is the transition zone from thick lithosphere in the Western Block (N120 km) to thin lithosphere (~80 km) in the Eastern Block (An and Shi, 2006). Zhao and Xue (2010) suggested that there is a sublithopheric corridor under the orogen; the mantle material in the asthenosphere may flow along the corridor and generate the LPO with the NNE SSW orientation. This model is supported by the tomographic images, which revealed a low-velocity zone in the uppermost mantle right beneath the orogen (Tian et al., 2009; Xu and Zhao, 2009; Zhao and Xue, 2010). Bai et al. (2010), however, carefully analyzed the shear-wave splitting at the permanent stations and argued for a two-layer model under the orogen. The φ in the lower layer aligns WNW ESE, which is consistent with the APM direction, while φ in the upper layer is subparallel with the orogen and fault strikes, reflecting anisotropy in the lithosphere (Bai et al., 2010; Huang et al., 2008). There are few measurements in the Western Block (Ordos) because of the lack of seismograph stations there. Station YCH is located just in the western margin of the Ordos Block. The events recorded by YCH mainly occurred in the Tonga subduction zone and most of the rays pass through the upper mantle under the Ordos Block. Thus the measurements at YCH can be used to constrain the anisotropy within the stable block. The results show a NW SE φ under the Ordos Block, which is consistent with the results at the portable stations in the northern Ordos (Fig. 8)(Zhao et al., 2008). The NW SE, even NNW SSE (HHC in Fig. 8) φ is different from the APM direction and so cannot be explained by the APM-driven LPO in the asthenosphere. The Ordos Block may be very cold and deep-rooted and so the development of LPO is not easy (Hu et al., 2000; Huang and Zhao, 2006; Tian et al., 2009; Zang et al., 2005). This cannot be the result of the present lithospheric deformation either, because there is no significant internal deformation within the Ordos Block since the Precambrian (Zhao et al., 2001). Thus, the most plausible explanation for the anisotropy in the Ordos Block is that it is fossil anisotropy frozen in the lithosphere when the block formed (Zhao et al., 2008) South China South China was formed by the collision of the Yangtze and Southeast (SE) China blocks in the Precambrian (e.g., Charvet et al., 1996), and later superimposed by broad thin-skinned fold-and-thrust belts (trending NE SW) formed by the intracontinental shortening in the Mesozoic (Fig. 9) (Charvet et al., 1996; Li and Li, 2007). While the dominant WNW ESE φ (APM-parallel, Fig. 9) suggests that the anisotropy exists mainly in the asthenosphere, frozen anisotropy in the lithosphere may also contribute to the shear-wave splitting signals in areas with strong lithospheric deformation. At station ENH in the Sichuan Block, the observations reveal a complex anisotropic structure (Figs. S5 and S7d), which may indicate multilayer anisotropy. The measurements may also reflect lateral variations of the anisotropy under ENH (Fig. 9) because the multi-layer anisotropy itself can be revealed as an accumulative result by themcmethod(silver and Long, 2011). SE China is located near the boundary between the Eurasian and Philippine Sea plates. Seismic tomography has revealed significant lowvelocity anomalies, which may reflect strong mantle upwelling occurring in this region in the Late Mesozoic and Cenozoic (Huang and Zhao, 2006; Huang et al., 2010; Zhou et al., 2006). Thus the APM-driven LPO cannot be well developed in the upper mantle under SE China. As the Philippine Sea plate is moving toward the northwest at a rate of ~82 mm/yr (Fig. 9) (Seno et al., 1993), SE China is under extensive NW SE contraction (Sibuet and Hsu, 2004). The NE SW φ in this area is parallel with the fault strikes (Fig. 9), and so it should reflect the anisotropy in the lithosphere under NW SE compression (Savage, 1999; Silver, 1996; Zinke and Zoback, 2000) Western China The western China region is characterized by extensive lithospheric deformation resulting from the India Asia collision (e.g., Tapponnier et al., 2001; Yin, 2010 and references therein). The Himalayan orogen, Tibetan Plateau, Tienshan orogen, and even the Baikal rift zone are all considered to be in the tectonic domain of the India Asia collision (Molnar and Tapponnier, 1975; Tapponnier et al., 1982, 2001; Tapponnier and Molner, 1977; Yin, 2010). The φ orientation in western China is generally parallel with the strikes of the orogens and active faults while perpendicular to the orientations of the maximum horizontal stress σ H (Figs. 6e, f and 10) (Heidbach et al., 2010), as revealed by many previous studies (Chen et al., 2006, 2010; Dricker et al., 2002; Flesch et al., 2005; Herquel and Tapponnier, 2005; Huang et al., 2000; Li and Chen, 2006; Li et al., 2010; McNamara et al., 1994; Silver, 1996; Sol et al., 2007; Vinnik et al., 2007; Wang et al., 2008). Thus the observed shear-wave splitting is most probably caused by anisotropy in the lithosphere and its spatial variations mainly reflect the large-scale pattern of lithospheric deformation (e.g., Flesch et al., 2005; Wang et al., 2008). Many studies have suggested that the crust and the underlying mantle are completely decoupled beneath Yunnan in the southeastern margin of the plateau where φ suddenly rotates from N S to nearly E W(Fig. 10)(e.g., Flesch et al., 2005; Huang et al., 2007; Lev et al., 2006; Royden et al., 1997, 2008; Sol et al., 2007; Tapponnier et al., 2001). But Wang et al. (2008) argued for crust mantle mechanical coupling during deformation; the observed dramatic spatial rotation of φ reflects the deformation style from simple shear in the Tibetan Plateau transitioning to pure shear (E W extension) in Yunnan. In general, φ in Yunnan agrees well with the APM direction and the GPS results (relative to the ITRF97) (Fig. 10) (Bai et al., 2009; Huang et al., 2007; Lev et al., 2006), supporting the APM-driven anisotropy contributing to the observations. Our measurements at KMI can be fitted well with two-layer anisotropy and φ in the lower layer is parallel with the APM direction (Fig. 10), which confirms the existence of APM-driven anisotropy under Yunnan. Whether APM-driven anisotropy in the asthenosphere exists widely under western China is actually a critical issue, because it is concerned with the effect of the India Asia collision on the evolution in the sublithospheric mantle (e.g., Liu et al., 2004). As mentioned above, the small angular differences between φ and the APM direction (b35 ; Fig. 6) indicate that the observed shear-wave splitting is closely related to the APM. As increasing data have become available, the complex anisotropic structure, e.g., the multi-layer anisotropy, has been clarified. Two-layer anisotropy has been revealed at the stations in SE Tibet (KMI) and NE Tibet (Li et al., 2011), near the Tienshan orogen that is over 1000 km away from the frontier of the India Asia collision (WUS, WMQ; Li et al., 2010), and even in Mongolia (ULN, Ulaanbaatar; Barruol et al., 2008) and

10 114 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) Fig. 9. Distribution of the shear-wave splitting measurements in South China. The measurements with the SC method at ENH are shown as black bars, and the blue circle denotes the incidence angle of 10. The other labeling is the same as Fig. 7. the Baikal rift zone. The fast axes of the lower-layer anisotropy at all these stations show good consistency with the APM direction. Many previous studies showed that the fast axes of one-layer anisotropy are also parallel with the APM direction (e.g., Barruol et al., 2008; Dricker et al., 2002; Gao et al., 1994; Huang et al., 2007; Lev et al., 2006; Li et al., 2011). All these results suggest that APM-driven anisotropy exists widely in the asthenosphere under western China, i.e., the Tibetan Plateau, Tienshan orogen and even Mongolia and the Baikal rift zone (Figs. 5 and 10). Although the lithosphere in this region may have suffered stronger deformation than any other region in the world as a result of the India Asia collision (Tapponnier et al., 2001; Yin, 2010), from the viewpoint of seismic anisotropy, the asthenosphere may have evolved in its own way and been less affected by the mountain buildings in the overlying lithosphere. Special attention, however, should be paid on the interpretation of splitting in south Tibet. As only four permanent stations (GZE, LIZ, LSA and SQH) are located in south Tibet, our results could not provide a comprehensive view on anisotropy there. Fortunately, the results accumulated in the past decades provide important information on this issue (e.g., Chen et al., 2010; Gao and Liu, 2009; Huang et al., 2000). Because south Tibet is located in the frontier of the India Asia collision where the Indian plate has subducted into the upper mantle and even reached the mantle transition zone (e.g., Li et al., 2008; Zhao et al., 2011b), consistent mineral LPO in the asthenosphere is unlikely. Thus the APM-driven anisotropy may not exist in this region. The measurements in south Tibet show strong lateral variations and tend to be well correlated with the strikes of active faults and GPS results (Figs. 5 and 10)(e.g.,Chen et al., 2010; Huang et al., 2000), which also argue against APM-driven anisotropy while favoring anisotropy in the lithosphere. The δt splitting timesmayreachmorethan2s(gze;fig. 4;TableS1)(Chen et al., 2010; Huang et al., 2000), suggesting that the thickness of the anisotropic layer should be at least 230 km (for anisotropy of 4%; Silver and Chan, 1988). Because the depth of the lithosphere asthenosphere boundary is only ~200 km (Zhao et al., 2011b) and the crustal thickness is ~70 80 km (Yuan et al., 1997), the thickness of the lithospheric mantle beneath Tibet is no more than 130 km. Hence the crustal anisotropy must contribute to the observed splitting. Two origins of crustal anisotropy should be considered. One is the LPO of middle-lower crustal minerals such as amphibole developed in deep crustal flow, which may cause shear-wave splitting with fast axes (φ) consistent with the GPS results (Gao and Liu, 2009; Royden et al., 2008). The other is fault-parallel φ near active faults, which may result from the crustal fault-fabric accompanying the fault activities (Wang et al., 2008; Zinke and Zoback, 2000). Gao and Liu (2009) analyzed the shear-wave splitting measurements at LSA and confirmed the multi-layer anisotropy (Fig. 10), i.e., upper-layer anisotropy in the lower-middle crust and lower-layer anisotropy in the lithospheric mantle. 5. Conclusions We have made a detailed study of the shear-wave splitting at 138 permanent seismograph stations in China. Our results have provided new insights into the anisotropic structure and mantle dynamics under the Chinese continent. Three different methods (the RC, SC and MC) were used simultaneously to better estimate the splitting parameters and avoid potential bias. Most of the events used in our analysis occurred in the Tonga subduction zone with back-azimuths of , which are subparallel to the fast orientations of the anisotropy (i.e., φ for the MC method) in eastern China. Thus the splitting parameters in eastern China determined by a routine (RC or SC) method on the individual seismograms may be seriously biased. The φ fast directions in eastern China, to first-order, are consistent with the APM directions and GPS results, suggesting that the shear-wave splitting reflects asthenospheric anisotropy due to mineral LPO driven by APM. In NE China and North China, the subduction of the Pacific plate plays an important role in the LPO development. Under the orogens in

11 Z. Huang et al. / Earth and Planetary Science Letters 306 (2011) Fig. 10. Distribution of the shear-wave splitting measurements in western China. The labeling is the same as Fig. 7. the western part of eastern China, anisotropy in the lithosphere contributes significantly to the observations, while the observations in the Ordos Block reflect fossil anisotropy frozen in the stable Archean block. In SE China, the NE SW φ is caused by anisotropy in the lithosphere due to the strong NW SE contraction between the Eurasian and Philippine Sea plates. The most significant feature in western China is that φ is perpendicular to the orientations of the maximum horizontal stress σ H while parallel with the strikes of the orogens and faults. The observed spatial variations in anisotropy reflect the large-scale pattern of lithospheric deformation, accompanying a transition from simple shear in the Tibetan Plateau to pure shear in the surrounding regions. In south Tibet, crustal anisotropy, either due to deep crustal flow or the fault-fabric, contributes to the observed splitting. In addition, φ also exhibits good correlation with the APM directions and GPS results, and many studies have revealed that APM-driven anisotropy exists widely in the asthenosphere under western China. These results suggest that the mountain building has caused significant deformation in the lithosphere, but has less affected the underlying asthenosphere, from the viewpoint of seismic anisotropy. Acknowledgments Data used in the study were archived and managed by the China Seismic Network Data Center and IRIS DMC. We thank Prof. S. Chevrot and Dr. V. Monteiller for providing us the code for multichannel splitting analysis. This work was supported partially by the National Natural Science Foundation of China (Grant No ), a grant (Kiban-A ) to D. Zhao from the Japan Society for the Promotion of Science and the Scientific Research Foundation of Graduate School of Nanjing University. D. Zhao and Z. Huang were also supported by the Global-COE program of Tohoku University. Prof. P. Shearer (editor), M. Savage and an anonymous reviewer provided constructive comments which improved the manuscript. Most of figures were made by using GMT(Wessel and Smith, 1998). Appendix A. Supplementary data Supplementary data to this article can be found online at doi: / j.epsl References An, M., Shi, Y., Lithospheric thickness of the Chinese continent. Phys. Earth Planet. Inter. 159, Audoine, E., Savage, K., Gledhill, K., Anisotropic structure under a back arc spreading region, the Taupo Volcanic Zone, New Zealand. J. Geophys. Res. 109, L Bai, L., Iidaka, T., Kawakatsu, H., Morita, Y., Dzung, N., Upper mantle anisotropy beneath Indochina block and adjacent regions from shear-wave splitting analysis of Vietnam broadband seismograph array data. Phys. Earth Planet. Inter. 176, Bai, L., Kawakatsu, H., Morita, Y., Two anisotropic layers in central orogenic belt of North China Craton. Tectonophysics 494, Barruol, G., Deschamps, A., Deverchere, J., Mordvinova, V.V., Ulziibat, M., Perrot, J., Artemiev, A.A., Dugarmma, T., Bokelmann, G.H.R., Upper mantle flow beneath and around the Hangay dome, Central Mongolia. Earth Planet. Sci. Lett. 274, Becker, T.W., Chevrot, S., Schulte-Pelkum, V., Blackman, D.K., Statistical properties of seismic anisotropy predicted by upper mantle geodynamic models. J. Geophys. Res. 111, B Becker, T.W., Kustowski, B., Ekstrom, G., Radial seismic anisotropy as a constraint for upper mantle rheology. Earth Planet. Sci. Lett. 267, Bird, P., An updated digital model of plate boundaries. Geochem. Geophys. Geosyst. 4, Chang, L., Wang, C., Ding, Z., Seismicanisotropy ofupper mantle ineasternchina. Sci. China Earth Sci. 39, (in Chinese).