Anisotropic Tomography and Dynamics of the Big Mantle Wedge

We determine high‐resolution 3‐D tomographic images of P‐wave isotropic velocity, radial anisotropy and azimuthal anisotropy beneath NE Asia down to 800 km depth. Our results show negative radial anisotropy (i.e., Vhorizontal < Vvertical) in the asthenosphere of the big mantle wedge (BMW), which may reflect mineral alignment caused by vertical flow in the asthenosphere. Across the Tanlu fault zone (TLF), the western and eastern parts of the BMW exhibit high and low P‐wave velocities, respectively. Combining our tomographic results with surface geological features, we speculate that convection in the BMW includes upwelling asthenosphere beneath the Japan Sea and the Korean Peninsula and downwelling asthenosphere beneath the Songliao and North China basins. The downwelling asthenosphere beneath the two basins is associated with diminishing volcanism and anomalous tectonic subsidence since ∼110 Ma. The great TLF is an important boundary for the BMW structure and dynamics.

mantle is mainly caused by mantle flow in the asthenosphere (Holtzman et al., 2003), tectonic stress in the lithosphere (S. Zhang & Karato, 1995), and fossil preferred orientation in the past frozen-in deformation (Fouch & Rondenay, 2006). The upper mantle anisotropy above 400 km depth mainly depends on the lattice preferred orientation of olivine, the most abundant mineral in the mantle. In deformation induced by mantle flow, five types of olivine slip system corresponding to A-E fabrics are formed in different conditions of temperature, stress and water content (Karato et al., 2008). A fast seismic wave parallel to the mantle flow direction can be seen in A-, C-, D-, and E-type olivine fabrics corresponding to [100](010), [001](100), [100]{0kl}, and [100](001) slip systems (Mainprice et al., 2005). Only one exception (B-type olivine fabric corresponding to [001](010) slip system) may appear in low-temperature and hydrous-rich environment, especially the subducting slab bending zone near the forearc mantle (Kneller et al., 2005). Previous seismic studies focused on the azimuthal anisotropy (AAN) structure of variable scales in East Asia (H. X. Fan et al., 2020;Z. Huang et al., 2014;Lü et al., 2019;J. Ma et al., 2019;Wei et al., 2015;Jia et al., 2022) and radial anisotropy structure beneath the North China craton Cheng et al., 2013;Fu et al., 2015;J. Wang et al., 2014). However, there has been no study of 3-D P-wave radial anisotropy of the BMW in East Asia, especially in its asthenosphere.
To better understand the BMW structure and dynamics, we conduct anisotropic tomography using 673,208 P-wave travel-time data of local earthquakes and teleseismic events recorded at 2,388 seismic stations in NE Asia ( Figure 1). Our high-resolution tomographic models of isotropic P-wave velocity (Vp), radial anisotropy and AAN shed new light on the mantle structure and dynamics beneath NE Asia. Figure 1. Distributions of (a) seismic stations, (b) local earthquakes and (c) teleseismic events used in this study. In (a and b), the three green lines denote the North-South Gravity Lineament (NSGL), the Tanlu fault zone (TLF) and the Japan trench, whereas the two yellow patches denote the Songliao basin (SLB) and the North China basin (NCB). In (a), the black lines denote depth contours of the subducting Pacific plate, and the red dots denote Cenozoic intraplate basalts from the GEOROC database. In (b), the dot colors denote focal depths whose scale is shown at the bottom, and the beach balls denote focal mechanism solutions of three deep earthquakes. The red rectangle denotes an area in NE China (NEC) and the NW Japan Sea (NWJS) where deep earthquakes (red dots) occurred actively. In (c), the black dashed lines denote plate boundaries at the surface, and the red trapezoid denotes the present study region.

Data and Method
In this work, we use two sets of P-wave arrival-time data (Figure 1). The first data set is selected from that of C. , which includes 4,311 local/regional earthquakes recorded at 2,388 seismic stations in Japan, South Korea and East China. The second data set is provided by the China Earthquake Networks Center, National Earthquake Data Center (CENC; http://data.earthquake.cn), which includes 5,969 local events recorded at 202 seismic stations in NE China. Details of the two data sets are shown in Table S1 in the Supporting Information S1. We carefully selected and processed these data according to the following criteria: (a) All overlapping or inconsistent earthquakes in the different data sets are removed; (b) The average epicentral distance of each local earthquake is larger than 100 km ( Figure S1 in the Supporting Information S1), because this work focuses on the regional mantle structure rather than local crustal structure; (c) After relocating all the local events using a 1-D velocity model including depth variations of the sedimentary layer, the Conrad and the Moho discontinuities derived from the CRUST1.0 model ( Figure S2 in the Supporting Information S1; Laske et al., 2013), the uncertainty of the epicentral location is smaller than 0.1° (∼10 km), and the focal depth error is smaller than 15 km ( Figure S3 in the Supporting Information S1); and (d) the cut-off travel-time residual is taken to be 2.5 s in the tomographic inversion. As a result, our local data set contains 399,697 P-wave arrival times of 10,280 local/regional earthquakes (Figures 1 and S4 in the Supporting Information S1). The root-mean-square (RMS) travel-time residual for all the local/regional event data is reduced from 1.726 to 1.503 s after the earthquake relocation ( Figure S1 in the Supporting Information S1).
The teleseismic events are selected from the data set of C.  with the following criteria: (a) Each event was recorded at more than 15 seismic stations in the study region; (b) the epicentral distance of each event is limited to 30° to 90°; and (c) the cut-off travel-time residual is taken to be 2.5 s in the tomographic inversion ( Figure S1 in the Supporting Information S1). As a result, our data set contains 273,511 relative travel-time residuals of 8,988 teleseismic events (Figures 1c and S4 in the Supporting Information S1).
We apply tomographic methods (Zhao et al., 2009;J. Wang & Zhao, 2008 to our data set to determine 3-D images of isotropic Vp, radial anisotropy (RAN) and AAN beneath NE Asia (Figures 2, 3 and S5-S8 in the Supporting Information S1). The RMS travel-time residual is reduced to 0.691, 0.664 and 0.672 s after the isotropic, AAN and RAN tomographic inversions, respectively ( Figure S5 in the Supporting Information S1). We performed extensive resolution tests to examine the robustness of the obtained tomographic models (Table S2 and Figures S9-S43 in the Supporting Information S1). The test results show that the P-wave rays of our data set crisscross very well in the study volume ( Figure S4 in the Supporting Information S1). The lateral resolution is ∼1.0° for the isotropic Vp model and ∼1.5° for the RAN and AAN models. The depth resolution is ∼80 km for the isotropic Vp model and ∼150 km for the RAN and AAN models (Figures S10-S19 in the Supporting Information S1). A potential problem of anisotropic tomography is that there may be trade-off between isotropic Vp structure and anisotropy (e.g., Z. Huang et al., 2015;Z. W. Wang et al., 2022). We performed extensive synthetic tests to examine the trade-off issue, and the test results show that our isotropic and anisotropic images are robust and the trade-off effect is very small (Figures S30-S41 in the Supporting Information S1). Details of the resolution and synthetic tests can be found in the Supporting Information S1.

Results
Our tomographic model shows an expected geometry of the subducting Pacific slab beneath the Japan Islands and the Japan Sea, as well as the stagnant slab in the MTZ beneath the Korean Peninsula and NE China (e.g., J. Huang & Zhao, 2006;Wei et al., 2015;C. Chen et al., 2017;J. Ma et al., 2018J. Ma et al., , 2019Zhao, 2021). Positive RAN, that is, V horizontal > V vertical , occurs in the stagnant slab, reflecting horizontal movement and deformation of the slab in the MTZ. The boundary between the W-BMW and E-BMW roughly runs along the Tanlu fault zone (TLF), which is a great trans-lithospheric fault with strike-slip motions in East China (e.g., Lei et al., 2020). G. Zhu et al. (2018) suggested that the TLF has been a tectonics-sensitive weak zone in the lithosphere since the Mesozoic. Our tomographic images of isotropic Vp and anisotropy clearly show different patterns across the TLF ( Figure 2). Hence, we deem that the great TLF has acted as an important boundary for the structure and dynamics of the BMW.
Previous regional tomographic models show a low-Vp anomaly in the asthenosphere beneath the Songliao basin and the North China basin (e.g., J. Huang & Zhao, 2006;Koulakov, 2011;J. Ma et al., 2019;Ward et al., 2021), but local tomographic models (C. X. Fan et al., 2021;Guo et al., 2016;J. Ma et al., 2018;Wei et al., 2019;Jia et al., 2022) and our present model show a high-Vp anomaly there. The discrepancy is attributed to the resolution differences between the local and regional models. The larger-scale regional models are generally determined with larger grid intervals and stronger damping and smoothing regularizations, so small-scale velocity anomalies do not show up. However, even in large-scale tomographic models (e.g., J. Huang & Zhao, 2006;Wei et al., 2015;J. Ma et al., 2019), the mantle beneath the two basins exhibits relatively higher velocities than that beneath the SW Japan Sea and South Korea, reflecting lateral structural variations in the BMW.
Our anisotropic Vp models also reveal significant differences between the W-BMW and E-BMW (Figures 2  and 3). The two domains exhibit negative RAN (V horizontal < V vertical ) but are separated by a narrow zone with Map views of (a) isotropic Vp perturbation (dVp), (b) radial anisotropy (RAN), and (c) azimuthal anisotropy (AAN), whose scales are shown at the bottom. The layer depth is shown at the upper-left corner of each map. The North-South Gravity Lineament (NSGL), the Tanlu fault zone (TLF) and the slab upper boundary are shown as white, pink and blue lines in (a, b, and c), respectively. The blue and red dashed lines outline the isotropic high-Vp and low-Vp anomalies, respectively. The horizontal and vertical bars in (b) denote positive and negative RANs, respectively, whose length denotes the RAN amplitude. The orientation and length of the black bars in (c) denote the AAN fast velocity direction (FVD) and amplitude, respectively. The blue and red colors in (a) denote isotropic high-Vp and low-Vp anomalies, respectively. The blue and green colors in (b) denote positive and negative RANs, respectively. The orange and cyan colors in (c) denote nearly eastwest and north-south FVDs, respectively. W-BMW, the western big mantle wedge (BMW); E-BMW, the eastern BMW.  (Figures 2b and 3b). Previous RAN tomographic studies of East Asia focused on the Japan Islands and the North China craton. Beneath SW Japan, body-wave and surface-wave tomographic results show trench-normal fast-velocity directions (FVDs) and negative RANs, which are consistent with our results and interpreted to reflect the C-type fabric of olivine (X. Liu & Zhao, 2016. Beneath the eastern North China craton, negative RANs are revealed in the asthenosphere (J. Wang et al., 2014), which are consistent with our model beneath the North China basin, although some models show positive RANs in the crust Cheng et al., 2013;Fu et al., 2015).
Our  (Bi et al., 2020). Beneath the North China basin, the FVDs change slightly at depths of 100-300 km (Figures 2c and S8 in the Supporting Information S1), which may reflect the contribution of fossil anisotropy in the crust and/or lithosphere. In other areas of the W-BMW, our AAN model is generally consistent with the SWS results ( Figure S8 in the Supporting Information S1).

The Eastern Big Mantle Wedge (E-BMW)
In the E-BMW, rotational motions of SW Japan yielded a board pull-apart zone during the Miocene, where a newly created oceanic crust outcrops in some extensional areas (Otofuji et al., 1991;Jolivet et al., 1994). Cenozoic intraplate volcanoes (e.g., Ulleung, Changbai, Longgang, etc.) are located above the E-BMW. Hence, wet and hot upwelling flow in the BMW (Zhao et al., , 2009Lei & Zhao, 2005) may explain the low-Vp anomaly and negative RAN in the E-BMW. Tang et al. (2014) attributed the Changbai basaltic magmatism to lower-mantle material upwelling through a slab hole in the MTZ, but the slab hole is not observed in our images (Figure 3) and other recent tomographic models (see a detailed review by Zhao, 2021). Most tomographic studies made so far indicate that the intraplate volcanoes in NE Asia (including Changbai) are more likely to originate from hot and wet upwelling flows in the BMW (Zhao et al., , 2009C. Chen et al., 2017;Zhao, 2021), which are also supported by most petrological and geochemical studies (e.g., Su et al., 2017;Xu et al., 2018;Kuritani et al., 2019;H. Choi et al., 2020;Q. Ma & Xu, 2021). Assuming that the AAN FVD is parallel to the regional mantle flow (Holtzman et al., 2003), melt-rich upwelling may have a 10.1029/2021GL097550 6 of 10 component orientated nearly in the N-S direction, which is related to the oceanic trench shape and the subducting slab geometry (Funiciello et al., 2006). For example, laboratory modeling shows that the width of a subducting plate strongly affects the vigor of both poloidal and toroidal advection since initial subduction (Funiciello et al., 2006).
Large deep earthquakes (M ≥ 7.0; focal depths >500 km) take place actively in the subducting Pacific slab in the MTZ beneath the NW Japan Sea and NE China (Figure 1b) where the slab bends abruptly (Figure 3a). These large deep earthquakes may invoke release of fluids from the slab in the MTZ to the overlying BMW, which may facilitate higher degree of partial melt feeding the Changbai and Ulleung intraplate volcanoes . Hence, the bottom of the E-BMW may be a relatively melt-rich layer (J. Yang & Faccenda, 2020), which forms a source pool of the intraplate magmatism and provides a crucial driving force for the upwelling flow in the E-BMW and mantle convection in the entire BMW.

The Western Big Mantle Wedge (W-BMW)
In the W-BMW, the high-Vp anomaly beneath the Songliao and North China basins is distinct and robust according to our extensive resolution tests (Figures 2a, 3a and Table S2 in the Supporting Information S1). A common explanation of this high-velocity anomaly is a delaminated lithospheric fragment (Cheng et al., 2013;Wei et al., 2019;Y. Zhang et al., 2011), but Guo et al. (2016) attributed it to downwelling asthenosphere. Our model cannot exclude the contribution from the delaminated lithosphere, but we prefer the latter explanation because the bottom of the high-Vp anomaly in the W-BMW reaches ∼300 km depth and a lithospheric fragment is insufficient to form such a thick anomaly (see Figures S31 and S32 in the Supporting Information S1). Numerical modeling shows that the surface topography associated with delamination is rapid uplift (Gogus & Pysklywec, 2008). Since such uplift is not observed in the Songliao and North China basins, thermal-mechanical erosion associated with ongoing mantle flow in the asthenosphere is more likely to occur in the W-BMW (Xu, 2007) instead of lithospheric delamination.
Anomalous tectonic subsidence is observed in intraplate rift basins (e.g., the Songliao and North China basins) above the W-BMW, which is thought to be a dynamic topography caused by the negative buoyancy due to mantle cooling (Ren et al., 2002;C. Li & Liu, 2015;Yao et al., 2017). Therefore, the high-Vp anomaly in the W-BMW beneath the two basins should represent a colder and denser mantle, and its negative RANs (i.e., V vertical > V horizontal ) most likely reflect mineral alignment caused by vertical shear due to asthenospheric downwelling (Karato et al., 2008). In addition, the downwelling asthenosphere in the W-BMW is consistent with the absence of volcanism within the two basins since ∼100 Ma, because upwelling mantle flows would produce intraplate volcanism (Zhao et al., , 2007X. Wang et al., 2017;Wei et al., 2019;J. Yang & Faccenda, 2020). In fact, most Cenozoic volcanism occurs at the basin margins in NE China (Figure 1a; Q. Fan & Hooper, 1991;Jia et al., 2022). This feature of the Cenozoic basalt distribution suggests the absence of hot and wet upwelling material beneath the two basins. Hence, the asthenospheric downwelling very possibly extends southward to the North China basin.

Convection in the Big Mantle Wedge
Previous geophysical studies have found some fingerprints of mantle convection in the BMW. Xu (2007) integrated geophysical data, isotopic composition, and paleogeographic configuration and considered the NSGL to be the western boundary of vigorous convection induced by the stagnant slab in the MTZ and the old major lithospheric weak zones. R. Zhang et al. (2014) and A. Zhang et al. (2021) investigated the seismic lithosphere-asthenosphere boundary (LAB) and the thermal LAB, respectively, in the W-BMW beneath the Songliao basin. A. Zhang et al. (2021) suggested that interactions between asthenospheric circulation and lithosphere control the intraplate volcanism. Surface wave tomography shows the existence of local sub-lithosphere downwelling in the W-BMW beneath the Songliao basin (Guo et al., 2016). X. Fan et al. (2020) suggested that the regional flow field in the E-BMW is oriented from the SW Japan Sea to East and NE China, which is consistent with a SWS result (S. . Our 3-D AAN model (Figures 2c and 3c) reveals NW-SE orientated mantle flow between the E-BMW and W-BMW at 100-300 km depths, which may represent volume compensation direction due to vertical flow in the asthenosphere. Since the subducting Pacific and Philippine Sea plates block recharge from the east and the southeast, respectively, the recharge of asthenospheric upwelling in the E-BMW is likely from the W-BMW. Thus, asthenospheric horizontal flow from the W-BMW to E-BMW may cause the NW-SE FVDs beneath the TLF and the northern Korean Peninsula at ∼300 km depth (Figure 2c), which coincides well with a previous result about mantle flow beneath the eastern North China craton .
Mantle flow above 200 km depth could be variable because asthenospheric upwelling is blocked by the lithospheric lid. The NW-SE FVDs beneath the TLF may be explained by asthenospheric flow from the E-BMW to W-BMW due to the volume compensation of vertical mantle flow in the BMW, which is consistent with the flow pattern beneath NE China derived from surface wave tomography (Guo et al., 2016). On the basis of the anisotropic tomographic models and geological evidence, we deem that 3-D mantle convection occurs in the whole BMW (Figures 3a and 4), which may include asthenospheric upwelling beneath the SW Japan Sea and South Korea, asthenospheric downwelling beneath the Songliao and North China basins, and horizontal flow beneath the TLF.
All these results suggest that mantle convection associated with the Pacific plate deep subduction may have controlled the lithospheric deformation and occurrence of intraplate volcanism. On the one hand, volcanism disappeared in both two basins since the middle-late Cretaceous (Figure 1a), in line with the asthenosphere downwelling detected in the W-BMW. A similar tectonic setting is passive continental margin, where an absence of volcanism is thought to result from mantle downwelling of edge-driven convection (King & Anderson, 1998).
On the other hand, the anomalous tectonic subsidence of the Songliao basin also occurred during the late Cretaceous, which is attributable to deficit in the negative buoyancy induced by mantle convection (C. Li & Liu, 2015). Hence, the BMW convection could be traced back at least to the late Cretaceous, which is consistent with a recent study based on the tempo-spatial pattern of late Mesozoic magmatism in northern North China Craton (Q. Ma & Xu, 2021). Moreover, mantle xenoliths brought to the surface by the Cenozoic basalts in the W-BMW show an oceanic mantle affinity in terms of bulk rock Mg#, olivine Fo, spinel Cr#, and Sr-Nd-Os isotopes (Gao et al., 2002;Xu & Bodinier, 2004). Because they are much more fertile than typical cratonic peridotites, the replacement of the ancient cratonic root by newly accreted lithosphere must have taken place (Xu & Bodinier, 2004

Conclusions
We obtain the first 3-D model of P-wave radial anisotropy beneath NE Asia in addition to new 3-D models of isotropic P-wave velocity and AAN of the region. Our high-resolution tomographic results reveal east-west variations in the structure and dynamics of the BMW under NE Asia. On the basis of the tectonic history and surface geology since the late Cretaceous, we speculate that 3-D convection occurs in the whole BMW beneath NE Asia, which contains upwelling asthenosphere beneath the Japan Sea and the Korean Peninsula and downwelling asthenosphere beneath the Songliao and North China basins. The great TLF has acted as an important boundary for the structure and dynamics of the BMW.

Data Availability Statement
In this work, we used two sets of P-wave arrival-time data (Figure 1). The first data set was selected from the data used by C. , which contains 4,311 local/regional earthquakes and 8,988 teleseismic events recorded at 2,388 seismic stations. The second data set was downloaded from the database of the China Earthquake Networks Center, National Earthquake Data Center (http://data.earthquake.cn), which contains 5,969 local events recorded at 202 seismic stations in NE China. Locations of the Cenozoic intraplate basalts are downloaded from the GEOROC database (http://georoc.mpch-mainz.gwdg.de/georoc/Start.asp). We used the seismic data provided by the China Earthquake Networks Center, National Earthquake Data Center (http://data.earthquake. cn), as well as data centers of the Japanese and South Korea seismic networks. of Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou) (GML2019ZD0202), and Director's Fund of Guangzhou Institute of Geochemistry (CAS). Locations of the Cenozoic intraplate basalts in Figure 1a are compiled from the GEOROC database (Sarbas & Nohl, 2008). Most of the figures are plotted using the GMT software (Wessel & Smith, 1998). We are very grateful to Prof. Lucy Flesch (the Editor) and two anonymous reviewers for their thoughtful review comments and suggestions, which have improved this paper. This is contribution NO. IS-3144 from GIGCAS.