Seismic constraints and geodynamic implications of differential lithosphere-asthenosphere flow revealed in East Asia

Significance

This study tackles the longstanding question of how the expanding Tibetan Plateau has influenced the continental deformation in east China using an approach that maps seismic azimuthal anisotropy, the directional dependence of seismic wave speeds in the lithosphere and asthenosphere, which are the rigid outer shell and mechanically weak part of the upper mantle, respectively. The results depict the eastward extrusion of asthenosphere beneath northeast Tibet whose motion is being blocked and deflected by Ordos and Sichuan cratonic keels. The consequence is the asthenosphere migrates around the Ordos and then turns into an east-west flow within the narrow channel of thinner lithosphere between these cratons. A broad region of surface continental deformation in eastern China is influenced by this asthenospheric flow.

Abstract

During the last 50 Ma, the East Asian continent has been a zone of massive continental collision and lithospheric deformation. While the consequences of this for Asian surface and lithospheric deformation have been intensively studied over the past 4 decades, the relationships between lithospheric deformation and underlying asthenospheric flow have been more difficult to constrain. Here we present a high resolution 3-D azimuthal anisotropy model for the northeastern Tibetan Plateau and its eastward continuation based on surface-wave tomography and shear-wave splitting measurements. This model shows that eastward lateral flow of asthenosphere beneath the northeastern Tibetan Plateau is being blocked by thick Ordos and Sichuan cratonic keels. The damming effect of these keels induces flow to first rotate around the Ordos keel and then transition into strong east-west flow beneath the thinner lithosphere that forms the lithospheric suture between the two cratonic keels. We further find that asthenosphere flow directions can differ from those of overlying lithosphere, with the asthenosphere neither being passively dragged by overlying lithosphere, nor being able to drag the overlying plate to mimic its subsurface flow. Finally, the region of eastward-channeled asthenospheric flow from Tibet underlies a belt of stronger intracontinental deformation in eastern China.

The eastern Eurasian continent is where the most significant intracontinental deformation on Earth has taken place during the Cenozoic. Here the collision between the Indian and Eurasian continents since ∼50 Ma has resulted in the development of high-elevation and ultra-thick crust in the Tibetan Plateau (1–5). Most geodynamic scenarios for asthenospheric flow have either envisioned the asthenosphere to be passively dragged by the overlying lithosphere (6) or alternatively that the shear stresses from asthenosphere flow will strongly shape surficial plate motions (7). East Asia’s rich set of geophysical and geological observations, and complex present-day surface deformation field constrained by relatively dense GPS surveys (Fig. 1A) (8, 9), make this region an ideal site to test proposed links between surficial motions and deeper asthenospheric flow. To date, seismic observations of possible lateral asthenospheric flow have been sparse, and its potential role in East Asian intracontinental deformation has remained unclear (5, 10–13).

Fig. 1.

(A) GPS map of surface displacements in East Asia. GPS-inferred surface motions are shown with respect to the “absolute” hotspot reference frame defined by the azimuths of global hotspot tracks within the past 10 Myr (8). The red dot at 37°N,103°E marks the location of the azimuthal phase velocity variation observations shown in SI Appendix, Fig. S2. (B) Map of the study area and the 790 seismic stations used for anisotropic surface tomography analysis. Blue line segments represent shear-wave splitting measurements (17–22). The length and orientation of each blue segment denotes the fast direction and delay time, with reference 1 s and 2 s lengths shown in the top right corner. Black triangles mark SOSArray and Himalaya II seismic stations; red triangles mark permanent CEA stations.

Here we show a promising approach to simultaneously infer the directions of crustal and lithospheric mantle deformation and deeper asthenospheric flow. Shear-wave splitting provides a direct constraint on the pattern of azimuthal anisotropy in the crust and upper mantle. This technique has been widely applied to infer crustal and upper mantle deformation (14–16). Recent studies reveal that a strong and coherent shear-wave splitting pattern surrounds the Ordos block and Sichuan basin (cf. Fig. 1B) (17, 18), implying the presence of substantial mantle deformation at depth (17–22). However, shear-wave splitting measurements are known to have poor vertical depth resolution; hence with these alone it is impossible to constrain whether/where the source of the splitting anisotropy lies within the lithosphere or asthenosphere.

Here we combine surface-wave anisotropic tomography results with shear-wave splitting measurements to make an improved image of depth variations in seismic anisotropy (23, 24) that lets us better infer the overall depth variation of directions of crustal, lithospheric, and asthenospheric flow. We determine a 3-D crustal and upper mantle shear-wave azimuthal anisotropy model that combines surface-wave phase dispersion measurements derived from ambient noise and teleseismic earthquakes with shear wave splitting measurements. The results reveal the presence of a strong channel of eastward asthenospheric flow beneath the Qinling belt (CQ), with further implications for a potential role of asthenospheric flow in shaping where lithospheric deformation concentrates within East Asia.

Results

Surface Crust-Lithosphere Motions (constrained by GPS).

Fig. 1A shows results from a recent GPS synthesis of surface displacement in East Asia for surface motions with respect to the absolute motion frame given by the trends of recent hotspot volcanism (8, 9). The modern surface displacement field shows the broad effects of northward collision of the Indo-Australian plate into Asia, and the resulting Eastward/Southeastward extrusion of East Asia. The shear strain rate field estimated from GPS (Fig. 1A) indicates significant deformation along major faults in NE Tibet (the Qilian Orogen [QL], West Qinling [WQ], and Songpan-Ganzi Block [SPGZ]). Our aim will be to compare these surface motions to deeper crustal, lithospheric, and asthenospheric regional flow.

Fig. 2.

Shear-wave velocity and azimuthal anisotropy maps at different depths in the crust and uppermost mantle. (A–D) Shear-wave velocity at depths of 10, 30, 50, and 90 km. Green bars show the azimuthal anisotropy as measured by the third inversion approach discussed here, which includes both crustal and upper mantle anisotropy.

Crustal and Upper-Mantle Azimuthal Anisotropy.

Fig. 2 presents our new 3-D shear-wave velocity and azimuthal anisotropy model at different depths within the final inversion. In general, the crustal shear-wave velocity structure in the present study is consistent with a recent joint inversion model of surface waves and receiver functions by Guo and Chen (25). In the upper crust (10 km, Fig. 2A), the shear-wave velocity structure is highly correlated with known regional geology: low shear velocities are observed beneath NE Tibet and basins with thick deposits, such as the western Ordos block and the northern Sichuan basin, while relatively high velocities are found beneath the CQ and the Trans-North China Orogen (TNCO). In the lower crust (30 km, Fig. 2B), NE Tibet, the TNCO, and the eastern CQ are dominated by low shear velocities. Thick crust with a very low shear velocity is observed at 50 km depths beneath NE Tibet and the western Daba belt (Fig. 2C) (25).

Fig. 3.

Vertical cross-sections of shear-wave velocities. Vertical cross-sections of S-wave velocities beneath profiles A–A’ (B), B–B’ (C), and C–C’ (D). The location of each profile is shown in panel A. LV1 and LV2 show the two low-wavespeed anomalies discussed in the text. Note that there could exist strong trade-offs between Moho depths and the near-Moho velocity structure, as further discussed in the text.

In the upper mantle, the western Ordos block and Sichuan basin are characterized by fast velocities (Fig. 3). Since only surface-wave dispersion is used to invert for shear velocity, the velocity jump across the Moho is not well constrained, and there may exist trade-offs between the near-Moho velocity structure and Moho depth. For example, the thin high velocity layer beneath the Moho in Fig. 3 is not well resolved and could be an artifact linked to a discontinuity in the starting model instead of a real contrast in layering or chemical composition at this depth. In contrast, lithospheric fast velocities only extend to depths of ∼100 km beneath the CQ. These high velocity lids represent the seismological lithosphere beneath the study region. Previous S-wave receiver function studies have shown that cratonic keels beneath the Ordos block and Sichuan basin extend to depths of over 200 km and that the lithosphere thickness beneath the CQ, NE Tibet, the TNCO, and the Eastern North China Craton (ENCC) is less than ∼120 km (26–28).

Fig. 4.

Upper crustal (A), mid to lower crustal (B), and uppermost mantle (C) anisotropy compared with shear-wave splitting measurements. Crustal (A and B) and upper mantle (C) seismic azimuthal anisotropy is compared to shear-wave splitting measurements. The pink color represents the anisotropy from surface-wave inversion, while the blue color shows the shear-wave splitting results.

The most prominent features in the uppermost mantle (Fig. 2D) are two distinct low shear velocity anomalies, here called LV1 and LV2, beneath NE Tibet and the Trans-North China Orogen, respectively. The spatial distribution and depth extent of the LV1 and LV2 anomalies are easier to visualize in vertical cross-sections (Fig. 3). In the north (A–A’ profile, Fig. 3B), LV1 and LV2 are separated by cratonic lithosphere beneath the Ordos block. LV1 and LV2 are connected by a low shear velocity channel (LVC in the B–B’ and C–C’ profiles in Fig. 3 C and D) below ∼100 km. The north-south trending C–C’ profile shows that the LVC is ∼200 km wide beneath the CQ.

Fig. 4 presents the crustal and uppermost mantle azimuthal anisotropy. The anisotropy model from surface wave has limited lateral resolution, whereas shear-wave splitting data are sensitive to local structure beneath station, which sometimes shows abrupt fast directions changes. To better compare to azimuthal anisotropy from surface-wave inversion with shear-wave splitting measurements, we collect splitting measurements from previous studies (Fig. 1B) but omit those data with significantly different fast orientations from their nearby stations. These splitting measurements are then spatially averaged in the present study to a 0.5° × 0.5° grid (Fig. 4). Seismic anisotropy from surface-wave tomography is generally weak in the upper crust beneath the NE Tibet. Fast directions generally rotate from W-E in the west that is following strikes of surface structure to SW-NE in the east, which is parallel to the Longmenshan fault (LMSF) (Fig. 4). In the mid to lower crust, we observe strong azimuthal anisotropy (>0.85%) with a predominant NW-SE fast direction beneath NE Tibet, consistent with shear-wave splitting measurements. The fast polarization slightly rotates toward the W-E direction beneath the CQ and Weihe rift. Beneath the Ordos block, azimuthal anisotropy is relatively weak (∼0.3–0.5%), with diffusely distributed fast polarizations. In the uppermost mantle, the fast polarization of azimuthal anisotropy gradually changes from a predominant NW-SE direction beneath the Alxa to an NNW-SSE direction beneath NE Tibet. In the central section of CQ (Fig. 4) is also dominated by anisotropy with an E-W fast direction. Beneath the southwestern Ordos, the fast direction strikes NW-SE with an anisotropic strength reaching almost −1.3%, broadly consistent with observed splitting patterns.

Fig. 5.

Differential lithosphere-asthenosphere flow in NE Asia. (A) GPS map of NE Asia. 20 mm/yr GPS rate symbol size shown in the right. (B) Distribution of principal strain rates estimated from GPS velocity field. For reference, the 20 × 10−9 strain-rate symbol size shown to the right of this panel. (C) Accumulated splitting time in crust; the background color denotes S velocity at 20 km. (D) Accumulated splitting time in upper 120 km; the background color denotes S velocity at 90 km. (E) Anisotropy below 120 km depths inferred from the joint analysis of surface-wave anisotropy and shear-wave splitting measurements. The background indicates the average S velocity between 120 and 300 km as determined by Yu et al. (34). The gray-shadow area highlights the region of shallow asthenospheric flow beneath this region. Note the pronounced E-W low velocity channel that underlies the CQ. This appears to be the primary path for lateral asthenosphere flow from Northeast Tibet.

Discussion

Origin of Seismic Anisotropy.

Seismic anisotropy in the crust and uppermost mantle is believed to primarily originate from the lattice preferred orientation (LPO) of anisotropic minerals such as mica and/or amphibole in the mid to lower crust and olivine in the upper mantle that change their LPO in response to finite strain (6, 29, 30).

To quantitatively evaluate the relative contributions of crustal and lithosphere mantle seismic anisotropy to shear-wave splitting, the theoretical splitting is estimated by assuming that an incident shear wave propagates vertically through the resulting anisotropic model at each geological grid point from 120 km depth to the surface (31).

The predicted shear-wave splitting pattern is shown in SI Appendix, Fig. S13A. The inferred fast directions have an average deviation of ∼35° relative to the observations (SI Appendix, Fig. S13D). Beneath the Ordos Block and Sichuan Basin, up to 50% of the observed delay time can be attributed to anisotropy within the lithosphere above 120 km depths (SI Appendix, Fig. S13). Moreover, if we assume vertically coherent anisotropy throughout the lithospheric mantle, then the ∼180–200 km thick lithosphere beneath the Ordos block and Sichuan basin (cf. 26) could account for over 80–90% of the observed splitting delay time, implying that the observed splitting could primarily originates within the thick lithospheric mantle in these regions.

However, seismic anisotropy in the upper 120 km only contributes a small portion of the observed splitting time (∼30%) beneath NE Tibet, northern Qinling, and the Trans-North China Orogen, all places where the lithosphere is relatively thin (e.g., less than 120 km, SI Appendix, Fig. S13). This implies that in places where the lithosphere is relatively thin, the asthenosphere is the main contributor to shear-wave splitting.

We further use shear-wave splitting measurements to constrain anisotropy in the asthenosphere at depths ≥120 km (more details can be found in SI Appendix).

Fig. 5 shows the resulting splitting within the crust (Fig. 5C), above 120 km (Fig. 5D), and from depths greater than 120 km depth (Fig. 5E). Overall, the layer below 120 km (asthenosphere) accommodates ∼0.75–1.4 s delay time beneath NE Tibet (the gray shadow area in Fig. 5E), northern Qinling, and the TNCO regions, which are characterized by relatively thin lithosphere (≤120 km). The eastern CQ contains the largest asthenospheric delay time, where delays locally reach up to 1.0–1.4 s. SI Appendix, Fig. S15 presents the shear-wave splitting measurements (32) collected from 4 nearby stations in NE Tibet (within a 50 km aperture, SI Appendix, Fig. S15). Here, we use the method of Silver and Savage (33) to compute the theoretical curve of apparent splitting parameters as a function of back-azimuth in a two-layer anisotropy model (one lithospheric layer and one asthenospheric layer). The shear-wave splitting measurements exhibit clear azimuthal variations, in good agreement with estimates based on the surface-wave tomographic model. In particular, both observed and predicted fast directions and anisotropy strengths show local variations in back-azimuth of ∼120° and support the presence of multilayered anisotropy beneath NE Tibet.

Geodynamic Implication: Eastward Asthenospheric Flow Is Strongly Shaped by the Deeper Lithospheric Keels of the Ordos and Sichuan Basin Blocks.

The two low wavespeed anomalies (LV1 and LV2) observed in the upper mantle beneath NE Tibet and the TNCO (see Figs. 2 and 3) are intriguing features in the 3-D shear velocity model. In the north, these anomalies are separated by the cratonic keel beneath the Ordos block. However, it is likely that they are connected by a low velocity anomaly at depths greater than 120 km beneath the CQ (Figs. 2, 3, 5). This low velocity anomaly is ∼200 km wide. It is bounded by thicker lithospheric roots beneath the Ordos basin to the north and the Sichuan basin to the south. The spatial pattern of the low velocity anomaly appears to closely mimic the geometry of the region with strong asthenospheric anisotropy (0.8–0.9 s of delay time) around the Ordos block (gray shadowed region shown in Fig. 5). We propose that this low velocity region indicates the presence of significant eastward asthenospheric flow that links asthenosphere associated with the Indian collision+subduction and Pacific subduction systems.

A recent teleseismic body-wave tomographic model shows that the low velocity anomaly beneath the CQ may extend to as much as ∼300 km depths (34). If so, the ∼0.8–1.4 s of splitting time accumulated within an ∼180 km thick asthenospheric layer would imply about 1.5% azimuthal anisotropy within this layer (Fig. 5).

The geometry of the low velocity anomaly may indicate that the thicker mantle lithospheric keels of the Ordos block and Sichuan basin play a crucial role in channeling the lateral spread of asthenospheric flow from beneath Tibet. If we make the common approximation that the fast crystalline fabric-induced anisotropy direction aligns with the direction of lateral flow as generally holds for large finite strains of olivine-rich aggregates (35–37), then asthenosphere flow appears to originate beneath NE Tibet with a NW-SE orientation. It then streams into a more focused channel beneath the CQ, to finally spread out below the now thin-lithosphere region of the ENCC into a more diffuse pattern.

Lateral Flow-Induced Melting.

When strong lateral asthenosphere flow is present, then melting has the potential to occur whenever asthenosphere rises and decompresses beneath an upwardly sloping region of the lithosphere. This phenomenon was originally proposed to explain the southward wave of volcanism along the East Coast of Australia that mimics, in time, when the Tasman Plume was migrating southward beneath more interior regions of the continent (38). Beneath East Asia, this melting mode could be triggered in regions where lateral asthenospheric flow encounters rapid horizontal variations in thickness of the lithosphere, like beneath the TNCO where the lithospheric thickness changes by over 50 km within a lateral distance of 100 km, and lateral and vertical variations in temperature appear to be large (26). Furthermore, it is also possible that local shear-driven convection (39) or edge-driven convection (40) could be triggered in regions with strong lateral variations in lithospheric thickness. Either the upwelling associated with coastward lateral flow itself or alternatively the upwelling associated with possible local edge-driven convection have been suggested to provide a potential explanation for the patterns of volcanism and regional stresses around the Ordos block (41, 42), such as the Hebi and Datong volcanoes.

Coupled and Decoupled Lithospheric Deformation.

The central section of CQ between Sichuan basin and Ordos block (Fig. 1B) displays coherent E-W orientated fast directions throughout its lithosphere, subparallel to the structural axis of the orogeny (Fig. 4), that generally resembles the shear-wave splitting pattern (17–22). We suggest that the regional-scale anisotropy in the lithospheric portion of the CQ primarily reflects the lithospheric fabric created during the N-S collision of the NCC and Yangtze Cratons during the Triassic that led to the formation of the CQ.

In contrast, NE Tibet, a region where mountain building has been ongoing since the Miocene, exhibits differing anisotropy patterns within its lower crust and uppermost mantle (Fig. 4). In the lower crust, the predominantly NW-SE oriented fast direction generally parallels the direction of large-scale strike-slip faults (Fig. 4B), i.e., the Kunlun fault and Haiyuan fault.

However, the direction of fast polarization of anisotropy systematically changes to NNW-SSE in the uppermost mantle beneath the Western Qinling and Songpan-Ganzi blocks of NE Tibet, which is an average 30° deviation from shear wave splitting measurements (NW-SE) (Fig. 4C). We propose that this indicates that crust and mantle deformation beneath NE Tibet is currently stratified/decoupled. At crustal levels, anisotropy in NE Tibet is generally associated with the lateral extrusion of crustal materials along the major strike-slip faults in response to continuous indention of the Indian continent, as has been proposed by numerous recent works (43–45). In contrast, the relatively warm and weak uppermost mantle of NE Tibet appears to be ductilely expanding eastwards in response to the indentation of the Indian subcontinent. However, the lateral expansion of deforming NE Tibetan mantle is being partially blocked by the relatively stronger Ordos lithosphere to its east. This may cause the NNW-SSE alignment lithospheric mantle texture, and thus the observed anisotropy. A weak and ductile mid to lower crust would allow decoupled deformation between the Tibetan crust and uppermost (ductile) mantle.

The lithosphere of western Ordos also exhibits layered deformation. Weak crustal anisotropy with incoherent fast polarization directions (Fig. 4) implies that the crust of the Ordos block has remained relatively stable with little to no deformation, consistent with GPS measurements and surface geology (9, 46, 47). The lithospheric mantle layer exhibits strong anisotropy with a predominant NW-SE fast-axis direction that continuously crosses the geological boundary between Ordos and Qilian block, the present-day frontier of NE Tibet. The measured anisotropy agrees well with the substantial shear wave splitting with delay times locally up to ∼1.5 s and uniformly aligned NW-SE fast polarization (Fig. 4). More generally, the preexisting morphology of lithospheric keels is suggested to significantly influence the spread of asthenospheric flow beneath continental regions, with asthenospheric flow being deflected around or beneath a deep lithospheric keel.

Further Implications of Decoupled Lithospheric and Asthenospheric Flow.

We see that this region of East Asia has clearly differing seismic anisotropy in the crust and mantle beneath NE Tibet and its eastern continuation (Fig. 5). One plausible explanation is that there is differential flow of extruding Tibetan crust, mantle lithosphere, and asthenosphere beneath this region of intensive transpressive shortening and deformation (4). Further to the East, depth-varying seismic anisotropy suggests the possibility of active flow decoupling between the lowermost lithosphere and uppermost asthenosphere, with net asthenospheric flow from Tibet toward the Pacific Rim of the Asian continent appearing to be shaped by a complex interaction between a “source” of asthenosphere arising from ongoing Tibetan compression, a net asthenosphere sink associated with down-dragging at W. Pacific subduction zones (5, 12, 13), and barriers that deflect lateral asthenosphere flow related to regions with thicker lithospheric keels like the Ordos and Sichuan Blocks. The interactions between these regional asthenosphere sources, sinks, and the spatial pattern of deep cratonic keels is what appears to govern the lateral path of East Asian asthenosphere as it migrates from Tibet toward the Pacific Rim (Fig. 5).

These findings provide useful observational constraints to test future geodynamic models of deformation within East Asia, both in their predictions for the pattern of lateral flow and for the time-history of the melting that would arise from upwards sloping lateral flow of asthenosphere during its lateral migration beneath East Asia. Finally, this work demonstrates that a relatively dense network of seismic stations, in combination with proper measurements of regional surface movements, will let us not only determine surficial “plate tectonics” motion, but also the horizontal flow directions of underlying asthenosphere. When researchers can extend this approach to cover more of Earth’s surface—in particular, beneath oceanic and continental regions—then we will have fully realized this powerful observational tool to better understand the origins of intraplate melting and the potential coupling between plate tectonics, underlying mantle flow, and melting.

Materials and Methods

The seismic stations used in this study consist of several temporary and permanent seismic arrays (Fig. 1B), including 171 seismic stations from the SOSArray (South Ordos Seismic Array) deployed between July 2011 and May 2014, 432 stations belonging to the Himalaya II array that operated from November 2013 to March 2016, and 187 permanent CEA (China Earthquake Administration) stations that operated between September 2011 and September 2014 (Fig. 1B). In total, we used records from 790 seismic stations, with most stations having at least 2 y of recordings.

Ambient Noise Eikonal Tomography (ANET).

At short periods (8–30 s), we use seismic ambient noise to measure the phase velocities of Rayleigh waves. The data processing procedures for ambient noise are similar to those described by numerous previous studies (e.g., 48, 49). First, the vertical components of raw seismic recordings are cut into a series of 1-day segments and resampled to 1 sample per s. Then, seismic data are band-pass filtered between 5 and 150 s, with both time and spectral domain prewhitening applied. Finally, all daily cross-correlations are stacked to obtain the final stacked cross-correlations between all possible station pairs. From these, fundamental Rayleigh wave phase velocities are measured using frequency-time analysis (FTAN) (50). We select phase velocity measurements with SNR > 20 (signal-to-noise ratio) for further analysis.

We employ Eikonal tomography (51) to obtain Rayleigh wave phase velocity and azimuthal anisotropy maps at periods between 8 and 30 s. According to the Eikonal equation, the local phase slowness is directly related to the gradient of the phase travel-time. To determine this, we first construct a phase travel-time surface at a selected frequency across the array, centered at each virtual source station, by interpolating travel-time measurements onto a 0.2° × 0.2° grid. Following Lin et al. (51), two methods are used to interpolate the travel-time surface, and the regions in which differences between the two methods exceed 1 s are discarded. Furthermore, we remove regions separated by distances less than 2 wavelengths. Then, the gradient of each travel-time surface is calculated to derive the local phase slowness, with the direction of the gradient regarded as the direction of wave propagation. Finally, the isotropic phase velocity and azimuthal anisotropy at each node is obtained by fitting and averaging all phase velocity measurements from different directions at nearby 3 × 3 grids using the 2ψ term of the cosine function (51):

$$c\left(w,\theta \right) =c_0\left(w\right)\left[1+c_2\left(w\right)\cos 2\left(\text{θ}-\mathrm{\text{ψ}}\right)\right];$$

where $c\left(w,\theta \right)$ is the azimuthal variability of Rayleigh wave phase velocity; ω is the angular frequency; $\theta$ is the azimuthal propagation of the Rayleigh wave measured with respect to north; $c_0\left(w\right)$ is the isotropic phase velocity and the $c_2\left(w\right)$ term is the zero-to-peak amplitude of the 2ψ anisotropy; ψ being the direction of the fast-axis.

Teleseismic Wave Helmholtz Tomography (TWHT).

Helmholtz tomography using teleseismic earthquakes (52) is applied to construct Rayleigh wave phase velocities and azimuthal anisotropy maps between 30 and 55 s. We use a total of 356 teleseismic earthquakes with magnitudes greater than 5.8 Mw and epicentral distances between 20° and 160° (SI Appendix, Fig. S1). First, Rayleigh wave phase delays between all nearby stations within 200 km are measured by waveform cross-correlation. Then, an apparent phase velocity map at each period for each earthquake is constructed based on the Eikonal Equation (Eq. 51). Finally, the Laplacian of the amplitude term in the Helmholtz equation is used to account for the finite-frequency effects and to correct the apparent phase velocity and to construct the structural phase velocity (52).

In contrast to ambient noise measurements, Rayleigh wave phase velocity measurements from teleseismic earthquakes exhibit strong 180° and 360° periodic azimuthal variabilities called the 2ψ and 1ψ anisotropies, respectively, instead of the predominant 2ψ anisotropy that would be expected for the propagation of Rayleigh waves in weakly anisotropic media (24, 53). The 1ψ component of azimuthal anisotropy represents a nonphysical bias that we seek to reduce by fitting the phase velocity measurements with both the 1ψ and 2ψ terms of cosine functions.

Phase Velocity Azimuthal Anisotropy.

SI Appendix, Fig. S2 A–D shows examples of azimuthal-dependent phase velocity measurements at different periods from both ambient noise and teleseismic earthquake data, for the grid located in NE Tibet (red point in Fig. 1A). Strong azimuthal anisotropy is observed at 8–12 s, with peak-to-peak anisotropy of almost 4.46%. Phase velocities using the TWHT method are relatively scattered compared with those obtained from the ANET technique, because of the superposition of a 1ψ variation onto the 2ψ anisotropy for the TWHT measurements. The fast orientation of anisotropy ranges between 91.3° and 101.3° at periods between 8 and 20 s.

The isotropic phase velocity anomalies agree well with results from previous ambient noise tomography studies (25) and exhibit gradual changes between maps at different periods. SI Appendix, Fig. S3 shows the isotropic phase velocity and azimuthal anisotropy maps for Rayleigh waves at different periods, as determined from the ANET (12 and 20 s) and TWHT (40 and 55 s) observations. At an intermediate period (30 s), both the ANET and TWHT techniques yield highly consistent phase velocity and azimuthal anisotropy maps (SI Appendix, Fig. S4). Large differences occur at the margins of the study region, e.g., the eastern Qinling and central Trans-North China Orogen, where station coverage is relatively poor for the ANET technique, since permanent stations were not available for measuring ambient noise. The average isotropic phase velocity difference between the two methods is −8.53 m/s, with a SD estimated to be ∼29.94 m/s. For most regions, the fast directions determined using the ANET method match values from the TWHT method to within 20°, with an average anisotropy strength difference of −0.14% (see SI Appendix, Fig. S4). Phase velocity and anisotropic measurements from the ANET and TWHT techniques at their overlapping period range are averaged with weights defined by their uncertainties. With this approach we obtain maps of Rayleigh wave phase velocity and anisotropy between 8 and 55 s.

Phase velocity and azimuthal anisotropy uncertainties are derived from the statistical analysis of multiple events (or stations, for ANET) (51, 52). Estimated uncertainties for isotropic phase velocity are relatively small (<15 m/s) beneath the western study region and increase toward the east (15–50 m/s) (SI Appendix, Fig. S5). In most study areas, uncertainties in the anisotropy strength lie within 0.5%, however, large uncertainties are observed in the southeastern regions (1.2–1.5%) (SI Appendix, Fig. S6). The average uncertainty in the fast direction is less than 20°, but much larger uncertainties (>60°) are associated with long periods (40–55s) (SI Appendix, Fig. S7), where observed anisotropy strengths are generally weak (SI Appendix, Fig. S3).

It is difficult to estimate spatial resolution using traditional checkerboard resolution tests since the Eikonal and Helmholtz tomography adopted here do not involve forward and inverse operators (51). Following Lin et al. (51), the lateral resolution of phase velocities from ANET is approximated by the coherence length of phase measurements, which provides the information of scales of features that can be resolved by the array. The statistical correlations, which vary between 0 and 1, of a particular point to neighboring points are calculated and then summarized to generate the correlation surface. The coherence length is then estimated by fitting the correlation surface with a cone, and the base radius of the cone is taken as the coherence length (51). In terms of TWHT, a similar method is employed to estimate spatial resolution but ignore the amplitude term. Given that the amplitudes correction term has more effect on absolute velocity variations and the length scale of Laplacian operator is almost 2 times the station spacing, ignoring it would not significantly underestimate spatial resolution for TWHT. The resulting spatial resolutions of two methods are summarized in SI Appendix, Fig. S8. For most of the study region, the coherence length is close to station spacing. For example, the spatial resolution of ANET is 30–40 km in the west where station coverage is dense and over 70–80 km in the east of the region, since only temporary arrays are used for ANET, while all seismic stations are used for TWHT and the resulting coherence length is less than 40 km in most of study region but increases to over 80 km in the Sichuan basin.

Shear-Wave Velocity and Azimuthal Anisotropy Inversion.

We adapt a two-step inversion scheme to invert for 1-D depth-dependent shear-wave velocity and azimuthal anisotropy (anisotropy strength and fast polarization) using the anisotropic dispersion curve extracted for each geological grid. The resulting 3-D anisotropic shear-wave velocity model is constructed by assembling all 1-D profiles measured at each 0.5° × 0.5° grid point.

In the first step, we invert for the reference isotropic shear-wave velocity using the Markov Chain Monte Carlo technique and the Delayed Rejection and Adaptive Metropolis algorithm (49). The 1-D V_sv model is parameterized as a crystalline crustal layer and a mantle layer from the surface down to 200-km depth. We use four B-spline coefficients to describe velocity variations in the crystalline crustal layer, and five B-spline coefficients to describe velocity variations in the mantle layer. The starting model is chosen to be the 3-D shear-wave velocity model of Guo et al. (25) in the upper 60 km and the 1-D PREM model (54) at greater depths. The searching range of V_s is ±20% relative to the starting model. Though crustal thickness is treated as an unknown parameter in the Markov Chain Monte Carlo inversion (55, 56), it is primarily constrained by a prior information from receiver function study (57). The search range for crustal thickness is ±5 km relative to the starting model of He et al. (57). The bulk crustal V_p/V_s ratio of He et al. (57) and average V_p/V_s of 1.79 for the upper mantle is used to scale for V_p from the V_s measurement in the inversion. We then relate V_p to density using Birch’s law (58). A physical dispersion correction is applied to our S wave model using the Q values from PREM (54). We estimate the statistics of the posterior probability density function from the last 3,000 accepted samples of the Markov chain. Shear velocity sensitive kernels (SI Appendix, Fig. S9) show that the maximum depth sensitivity of 55s surface-wave measurements is at 80–90 km, however, these still provide important information on structure down to 120–130 km.

We then invert for the crustal and upper mantle azimuthal anisotropy using the optimal shear wavespeed maps obtained during the first step. We apply the above-mentioned Markov Chain Monte Carlo Delayed Rejection and Adaptive Metropolis inversion technique to search the model space of anisotropic parameters of G_c and G_s, as these are most sensitive to the 2ψ azimuthal variations of the V_sv vertically polarized shear wave. The smaller 4ψ term related moduli (C_c,s) (59) is ignored in the inversion. We also neglect the moduli H_c,s in the inversion, which are associated with modulus F. Following Lin et al. (24) and Yao (60), the compressional moduli B_c,s are related to G_c,s through the simple relation B_c,s/A = G_c,s/L, where A = ρ$V_{ph}^2$, and L = ρ$V_{sv}^2$. Here, ρ is density and horizontally propagating (V_pv) and vertically-propagating (V_ph) P wavespeeds are assumed to be equal in the inversion. The partial derivatives of Rayleigh wave dispersion measurements respective to A and L for the azimuthal terms ($\partial \text{c}/\partial \text{A}$ and $\partial \text{c}/\partial \text{L}$) are calculated using the CPS code (61).

We have performed several inversion tests to investigate the possible origin of azimuthal anisotropy in the study area. Each test assumes a constant anisotropic layer in the crust and/or upper mantle. The resulting χ² misfits at each node are shown in SI Appendix, Fig. S10 A–L. In the first inversion, we allowed anisotropy in the crust but kept the upper mantle isotropic. A large misfit is shown beneath all study regions (SI Appendix, Fig. S10 A–D). In the second test, we allowed anisotropy in the upper mantle, but kept the crust as an isotropic layer, which slightly improves the data fit in the southwestern Ordos block (SI Appendix, Fig. S10 E–H). However, the large misfits in the NE Tibet show little improvement. In the third inversion, we simultaneously invert for azimuthal anisotropy in both crustal and upper mantle layers. The resulting χ² misfits significantly decrease in the whole study region as shown in SI Appendix, Fig. S10I, implying the existence of anisotropy in both crust and upper mantle. SI Appendix, Figs. S11 and S12 summarize model uncertainties in shear wavespeeds and anisotropy structure.

Finally, we also used two crustal layers (upper crust and mid to lower crust) and one uppermost mantle layer to represent the shear-wave azimuthal anisotropy in the NE Tibet plateau region (QL, WQ, and SPGZ), since NE Tibet has thick crust and low wavespeeds in its mid to lower crust. The thickness of lower crustal layer in the plateau region varies between 30 km and 40 km based on the thickness of low velocity zone found in Fig. 3. In the rest of the region, we used one crustal layer for the anisotropy inversion. Here, we focus on the mid to lower crustal structure since we use phase velocities ≥8s, which is most sensitive to structure deeper than 8–10 km.

Spatially Averaged Shear-Wave Splitting.

To directly compare surface-wave anisotropy and shear-wave splitting measurements, we needed to obtain an evenly spaced splitting map. In multilayered anisotropic structure, shear-wave splitting measurements will depend on the back-azimuth. We thus removed extreme values and outliers of splitting data at stations where multiple measurements are available, from which we obtain the station-averaged splitting parameters (62). Finally, we constructed spatially averaged split maps following the approach in (63). We obtained $2\mathrm{\Psi }$ terms of splitting measurements from fast directions and delay times:

$$\begin{array}{l}S=\delta t\sin 2\mathrm{\Psi }\\ C=\delta t\cos 2\mathrm{\Psi }\end{array}$$

Finally, the S and C terms at each station are interpolated onto an evenly spaced 0.5° × 0.5° grid with a spatial Gaussian smoothing at the length of ∼70 km (about twice the station spacing).

Shear-Wave Splitting Forward Modeling and Grid Search for Splitting in Asthenosphere.

For most stations, we do not have back-azimuthal information of splitting measurements. Also most of splitting data were measured using long period teleseismic shear waves (>5s), which are significantly longer than the splitting delay times. We thus use the method of Montagner et al. (31) to estimate the predicted shear-wave splitting parameters using our tomographic model. The predicted splitting time ($\delta t$) and fast direction ($\mathrm{\Psi }$) is computed under the assumption of subvertical propagation of long period shear wave through weak anisotropic layers with horizontal symmetry axes (59, 64), which can be expressed as:

$$\begin{array}{l}\delta t=\sqrt{{\left({\displaystyle {\int }_0^h\sqrt{\frac{\rho }{L}}\frac{G_c(z)}{L(z)}dz}\right)}^2+{\left({\displaystyle {\int }_0^h\sqrt{\frac{\rho }{L}}\frac{G_s(z)}{L(z)}dz}\right)}^2}\\ \tan 2\mathrm{\Psi }={\displaystyle {\int }_0^h\sqrt{\frac{\rho }{L}}}\frac{G_s(z)}{L(z)}dz/{\displaystyle {\int }_0^h\sqrt{\frac{\rho }{L}}}\frac{G_c(z)}{L(z)}dz\end{array}\text{,}$$

where $h$ is the maximum depth extent of anisotropic layer from tomography.

We fix the anisotropic parameters (crustal and lithospheric mantle fast direction and anisotropy strength) above 120 km determined from surface-wave tomography and carry out a grid search to find the best fit splitting time and associated fast direction below 120 km depths by minimizing the misfit between predicted and observed shear-wave splitting parameters as follows:

$$\mathit{misfit}={\left({(\delta t\sin 2\mathrm{\Psi })}_{\mathit{observed}}-{(\delta t\sin 2\mathrm{\Psi })}_{\mathit{predicted}}\right)}^2+{\left({(\delta t\cos 2\mathrm{\Psi })}_{\mathit{observed}}-{(\delta t\cos 2\mathrm{\Psi })}_{\mathit{predicted}}\right)}^2\text{.}$$

Here, the observed shear-wave splitting is the spatially-averaged splitting data (delay time $\delta t$ and fast direction $\mathrm{\Psi }$).

Note that the azimuthal dependence of splitting measurements shown in SI Appendix, Fig. S15 may affect the results based on removing the calculated effect of shallow structure from a single averaged splitting result. However, it is difficult to quantitatively estimate the uncertainties caused by this effect in the whole study region since splitting measurements in most stations are obtained from a narrow back-azimuth range. We try to reduce such effect by removing extreme values and outliers of splitting data (which could locate at the peaks or troughs of a theoretical curve similar to SI Appendix, Fig. S15) at stations where multiple measurements are available.

Strain Rate Estimated from GPS.

We used the open-source software STRAINTOOL (https://github.com/DSOlab/StrainTool; 65) to estimate strain rates from the GPS data compilation. This algorithm uses the Velocity Interpolation for Strain Rate algorithm described in Shen et al. (66).

Data, Materials, and Software Availability

The seismic data used in this study, including ambient noise cross-correlations and teleseismic waveforms, and azimuthal phase velocities determined in this study are available from https://doi.org/10.5281/zenodo.5234893 (67). The TWHT code is available at https://github.com/jinwar/matgsdf (68). The other codes mentioned in this paper are available from the corresponding author.