Upper-mantle tomography and dynamics beneath the North China Craton

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi:10.1029/2012jb009212, 2012 Upper-mantle tomography and dynamics beneath the North China Craton Jianshe Lei 1 Received 8 February 2012; revised 30 April 2012; accepted 10 May 2012; published 23 June 2012. [1] A high-resolution tomographic model of the upper mantle beneath the North China Craton (NCC) is determined using a large number of precisely hand-picked teleseismic P wave arrival times. The results are generally consistent with previous results but high-quality arrivals provide new insights into the dynamics beneath the NCC. Obviously north-south trending low-velocity (low-v) zones are revealed down to 300 400 km depth under the Shanxi rift and Tanlu fault zone, while a north-south trending high-velocity (high-v) zone representing the remainder of detached lithosphere is visible down to 200 km depth under the western portion of eastern NCC. High-V anomalies representing the detached lithosphere are detected at 200 400 km depth under central and eastern NCC. Under the Ordos block high-v anomalies are visible above 400 km depth, indicating intact lithosphere. Broad high-v anomalies representing the stagnant Pacific slab are imaged with a low-v anomaly from Datong volcano to the edge of Bohai Sea in the mantle transition zone beneath eastern and central NCC, suggesting that the Pacific slab has subducted to central NCC but with a gap. A continuously Y-shaped low-v structure is clearly imaged under Datong volcano and Bohai Sea from the lower mantle through this gap in the mantle transition zone to the upper mantle, indicating the existence of a lower mantle plume. These results suggest that in addition to the subduction of the Pacific plate, the plume has also played an important role in lithospheric destruction by thermal erosion of the asthenosphere and detachment of the lithosphere beneath the NCC. Citation: Lei, J. (2012), Upper-mantle tomography and dynamics beneath the North China Craton, J. Geophys. Res., 117,, doi:10.1029/2012jb009212. 1. Introduction [2] The present study region, the North China Craton (NCC), is situated in Northeast Asia and is characterized by a complex tectonic environment that is dominated by the subduction of the Pacific plate in the east and the collision of the Indian plate with the Eurasian plate in the southwest [e.g., Yin, 2000; Zhao, 2004; Liu et al., 2004; Lei and Zhao, 2005, 2006]. The region contains the Ordos block, Shanxi rift, Taihangshan and Yanshan mountains, North China basin, and some large seismic fault zones, such as the Zhang-Bo (from Zhangjiakou to Bohai Sea) seismic zone and Tanlu fault zone (Figure 1). Tectonically speaking, the region can be divided into three parts: the eastern, central and western NCC. The western NCC mainly includes the Ordos block. The central NCC contains the Shanxi rift, Taihangshan mountain, and western Yanshan mountain, while the eastern NCC mainly covers the North China basin, Songliao basin and eastern Yanshan mountain. Furthermore, the boundary 1 Key Laboratory of Crustal Dynamics, Institute of Crustal Dynamics, China Earthquake Administration, Beijing, China. Corresponding author: J. Lei, Key Laboratory of Crustal Dynamics, Institute of Crustal Dynamics, China Earthquake Administration, Beijing 100085, China. (jshlei_cj@hotmail.com; leijs@eq-icd.cn) 2012. American Geophysical Union. All Rights Reserved. 0148-0227/12/2012JB009212 between central and eastern NCC is in general agreement with the North-South Gravity Lineament (NSGL) (Figure 1). [3] The cratons that formed from Archaean to early Proterozoic are relatively stable tectonic units on Earth, such as Kaapvaal in South Africa, Siberia in Russia and Slave in Canada. Under these cratons the continental keels are about 250 300 km thick or thicker [e.g., Polet and Anderson, 1995; Artemieva and Mooney, 2001; James et al., 2001]. The cratons can be destroyed and the lithospheric keels become thinner, due to the influence of certain tectonic environments. Recently, many petrological, geochemical and geophysical investigations have been conducted in the NCC [e.g., Fan and Menzies, 1992; Menzies et al., 1993; Gao et al., 2002; Xu, 2001, 2007; Xu et al., 2004; Wu et al., 2008; Chen, 2010], but its specific mechanism by which the deep lithospheric root was destroyed is still under debate. Some researchers suggested that the NCC destruction was caused by thermomechanical erosion due to the asthenospheric upwelling [e.g., Griffin et al., 1998; Xu, 2001], while others indicated that it can be attributable to the detachment of lithospheric material [e.g., Gao et al., 2002; Wu et al., 2008]. However, few researchers mentioned the significant effect of hot material upwelling related to the Datong volcanism on the destruction of the NCC. [4] In addition to many smaller earthquakes, over 26 earthquakes of M > 6.0 occurred since AD 294 in the region, 1of29

Figure 1. Sketch map of the study area. Thick lines correspond to the boundaries between eastern, central and western NCC (ENCC, CNCC, and WNCC), while thin lines indicate major active faults [Deng et al., 2002]. The blue line marks the North-South Gravity Lineament (NSGL). HTR, Hetao rift; DT (yellow triangle), Datong volcano; WA (white star), the 4 July 2006 Wen-An county, Hebei, China, earthquake (M 5.1); YSM, Yanshan mountain; ZBZ, Zhang-Bo seismic zone; BHS, Bohai Sea; SXR, Shanxi rift; THS, Taihangshan mountain; HHS, Huanghai Sea; TLF, Tanlu fault zone; QNL, Qingling; SCB, Sichuan basin. White circles denote small earthquakes (4 < M < 7) that occurred from 1964 to 2005 [Engdahl et al., 1998], while color circles show large historic earthquakes (M 7.0) that occurred since BC 780 in the region [Song et al., 2011]. The scale for earthquake magnitude is shown in the top left corner. among of which 7 earthquakes had M 7.0 8.0, such as the 1679 Sanhe-Pinggu earthquake (M 8.0), the 1969 Bohai earthquake (M 7.4), and the 1976 Tangshan earthquake (M 7.8) (Figure 1) [Song et al., 2011]. However, since 2000 there have been no large earthquakes (M > 6.0) in the region, but it is worth noting that on 4 July 2006 an earthquake of M 5.1 occurred in Wen-An county, Hebei province (Figure 1), which is a unique earthquake larger than M 5.0 in North China after a quiescence of over six and a half years. Although this was a moderate-size event, its seismic shaking was felt quite widely over a large area of North China, including Beijing, Tianjin and Hebei province. No significant damage was caused by the Wen-An earthquake, but most people in the region paid much attention to it [Diao et al., 2006; Lei et al., 2008, 2011]. [5] Seismic tomography is a powerful technique to investigate the deep structure of the craton and understand the dynamics of destruction of the NCC, origin of Datong volcano, and seismotectonics in the study region, because it can provide significant information on the structural heterogeneity under the tectonic belts and the source areas. Therefore, many seismic tomographic investigations have recently been conducted in the region with different approaches and on various scales. Although some global or large regional tomographic studies have been made, they are too rough to image the detailed structure beneath the NCC [e.g., Zhao, 2004; Huang and Zhao, 2006; Lei and Zhao, 2006; Li et al., 2006, 2008; Li and van der Hilst, 2010]. Therefore, in addition to the application of the finite-frequency tomographic method of Hung et al. [2004] by Zhao et al. [2009] to only teleseismic data, several researchers applied the ray tomography method of Zhao et al. [1994] to local and teleseismic arrival-time data to infer the detailed deep structure of the NCC [e.g., Huang and Zhao, 2009; Tian et al., 2009; Xu and Zhao, 2009]. Although these studies generally showed similar patterns of velocity anomalies, they demonstrated some significant differences in details, which may be attributable to the use of different data sets. In the work of Xu and Zhao [2009] and Tian et al. [2009], some of the arrival-time data chosen from the observation bulletins of China Earthquake Administration were picked from analog rather than digital 2of29

waveforms, while Huang and Zhao [2009] used the arrivaltime data manually picked from digital seismograms but in a region of interest limited to a relatively small area, the Chinese capital region. Zhao et al. [2009] only used the teleseismic data from denser but linear portable seismic networks and only few permanent stations in the region, so their raypath coverage is very inhomogeneous in the shallow mantle. [6] The recent rapid growth of broadband seismic observation in China [Zheng et al., 2009, 2010] has opened a new era for imaging the mantle structure beneath China. The waveform data were collected from the Data Management Center of the China National Seismic Network at the Institute of Geophysics, China Earthquake Administration [Zheng et al., 2009, 2010]. In the present work, a large number of high-quality teleseismic arrival times were hand-picked from high-quality digital seismograms recorded by as many as possible Chinese permanent seismic stations in the region to invert for the 3-D upper mantle structure so as to reveal the detailed deep structure and dynamics beneath the NCC. The stations used generally have a uniform distribution in the region, in addition to some more densely spaced stations in Beijing and Tianjin. The present results provide some new insights into the mechanism for the destruction of the NCC, origin of Datong volcano, and seismotectonics in the study region. 2. Geotectonic Setting [7] The NCC, which contains crustal rocks older than 3.6 Ga [e.g., Liu et al., 1992], is one of the major Archean cratons in eastern Eurasia. It is bounded by the Yinshan- Yanshan mountains to the north and by the Qinling-Dabie- Sulu orogenic belt to the south (Figure 1). The Craton underwent a series of tectonic events in late Archean and Paleoproterozoic [e.g., Zhai et al., 2001], but it became stable in late Paleoproterozoic after the collision of eastern and western NCC [e.g., Zhao et al., 2000] and was magmatically and tectonically quiescent until middle Ordovician [e.g., Lu et al., 2000]. However, the craton began to be eroded from late Ordovician to middle Carboniferous, and became tectonically reactivated since late Mesozoic. [8] The NCC evolutional process could be related to the Tanlu fault zone, Shanxi rift, Datong volcanism, and Zhang-Bo seismic zone (Figure 1). The Tanlu fault zone, an approximately northeastward trending fault zone, extends over 2400 km in the Chinese Territory from Zhaoxing near the border between Russia and China through Bohai Sea to Guangji near the northern bank of the Changjiang River [e.g., Xu et al., 1987; Xu, 1993; Okay and Sengor, 1992]. It was called the Tanlu fault zone by Xu et al. [1993] because it was originally thought that it extends only from Tancheng to Lujiang. The Tanlu fault zone has experienced a complicated evolutional process. In late Triassic to middle Cretaceous time it was a left-laterally striking fault zone with displacement of 740 km [e.g., Xu, 1984], but in late Cretaceous period it became a well-developed rift and accumulated clastic and pyroclastic sediments with a thickness of up to 10 km [e.g., Xu, 1984]. However, in Paleogene time the rift graben was closed in response to the variations of the regional tectonic stress from WNW extension to ENE contraction [e.g., Gao et al., 1980], and gradually evolved into a right-laterally striking fault zone in late Eocene time [e.g., Lu et al., 1983] resulting from the Indo-Asian collision and the Pacific subduction. [9] The Shanxi rift, situated in central NCC, is bounded to the north by the Yinshan-Yanshan mountains and to the south by the Qinling mountains. Its western and eastern flanks correspond to the Lüliang and Taihang mountains (Figure 1). It is composed of a string of asymmeteric basins and extends about 1200 km in an S-like shape from Yanqing- Huailai, Hebei Province in the north through Datong and Taiyuan to Yuncheng, Shanxi Province in the south. The width of Shanxi rift varies from 20 to 80 km. The Shanxi rift began its extension during the Miocene [e.g., Zhang et al., 1998; He et al., 2004], and became active since the Pliocene [e.g., Xu et al., 1993], which resulted in devastating seismicity along the rift [e.g., Tapponnier and Molnar, 1977]. The best examples are the 1303 Hongtong (M 8.0) and 1695 Linfei (M 7.75), Shanxi, earthquakes [Song et al., 2011]. He et al. [2004] demonstrated the active extension of Shanxi rift using numerical simulations. [10] The Quaternary Datong volcano is situated at the northern end of Shanxi rift. Morphologically speaking, the Datong volcanic field can be divided into two parts. One is the region to north of Datong county, where there are at least 13 volcanic cones, while the other is the region to southeast of Datong county, where there exist subordinate volcanic cones with a relatively smaller height along the Sanggan river [e.g., Fan et al., 1992; Xu et al., 2005]. Furthermore, Potassium-Argon (K-Ar) dating suggested that the volcanism to north of Datong county took place in late Pleistocene (0.4 Ma) [Chen et al., 1992], later than that to southeast of Datong county, where it took place since early Pleistocene (0.74 Ma) [Chen et al., 1992]. Such a spatial distribution is consistent with the development of Shanxi rift from the south in late Eocene or early Oligocene to the north in Pleistocene [Ye et al., 1987; Ren et al., 2002]. [11] The Zhang-Bo seismic zone is also called Zhangjiakou- Penglai fault zone, which was originally defined by Zheng et al. [1981]. The zone was formed in Cenozoic time [Xu et al., 1998] and extends roughly in the NW-SE direction over a distance of about 700 km (Figure 1). Along this zone numerous earthquakes occurred since AD 780 [Song et al., 2011]. Some well-known examples of strong crustal earthquakes are the 1976 Tangshan earthquake (M 7.8) and the 1679 Sanhe-Pinggu earthquake (M 8.0). Therefore, the Zhang-Bo zone is one of the most significant and important tectonic features in the region. 3. Data and Method [12] In the present study, for the first time, 126,534 direct P wave arrival times were hand-picked, rather than picked by the multichannel cross-correlation technique of VanDecar and Crosson [1990] or the adaptive staking method of Rawlinson and Kennett [2004], from high-quality original seismograms recorded at 397 seismic stations (Figure 2) from 784 teleseismic (30 90 ) events (Figure 3) during August 2007 to March 2009. Figure 4 shows examples for original seismograms and hand-picked arrival times. The magnitude of these events is 5.3 or greater. The events are mainly located in the northwestern and southwestern Pacific subducting zones, but some events on the Indian and Atlantic 3of29

Figure 2. Distribution of 397 seismic stations (white triangles) used in the present study. Thick lines denote the boundaries of eastern, western and central NCC. Thinner lines denote major active faults. DT (yellow triangle), Datong volcano. ridges, in the Red and Mediterranean seas and on the Hawaiian islands are also included (Figure 3). Thus, although the events have an uneven distribution, they have a good azimuthal coverage around the study center. Some stations are very densely spaced in the Beijing and Tianjin areas, but other stations basically have an even distribution in the region (Figure 2). These stations have a large spatial coverage in eastern China and belong to 14 provincial seismic networks: Beijing, Tianjin, Hebei, Liaoning, Jilin, Neimeng, Shaanxi, Shanxi, Shandong, Henan, Anhui, Jiangsu, Hubei, and BU. Table 1 shows the number of seismic stations that are contained in each network. Most seismic stations recorded over 300 to around 600 teleseismic events on the global scale, except for those in Beijing and Hebei Province (Figure 5). Each event was recorded by at least 10 to over 350 seismic stations (Figure 6). The number of teleseismic arrivals recorded by each network is shown in Table 1. [13] The observed travel times versus epicentral distance are shown in Figure 7a. It is found that the hand-picked data form clear curves but with some misidentified data. To avoid the effect of misidentified data and to obtain a reliable tomographic image, it is necessary to winnow the data using absolute travel-time residuals that were deduced by subtracting theoretical from observed travel times. I calculated theoretical travel times (Figure 7b) by using the 3-D ray tracer of Zhao [2001] in a 1-D velocity model (Figure 8a). This ray tracer can compute not only the travel time and raypath accurately and efficiently, but also can deal with a model containing complex velocity discontinuities. For details of the technique, see Zhao et al. [1992] and Zhao and Lei [2004]. The finite-frequency kernels were neglected here, mainly because tomographic results obtained with ray theory and finite-frequency approaches show a high level of similarities [e.g., Hung et al., 2004; Tong et al., 2011]. The starting 1-D velocity model in the crust was inferred from the CRUST2.0 model [Bassin et al., 2000] by averaging the velocities at the same depth in the region, and the IASP91 velocity model [Kennett and Engdahl, 1991] was directly adopted for the upper mantle (Figure 8a). Depth variations of the Moho discontinuity (Figure 8b) adopted from the CRUST2.0 model [Bassin et al., 2000] were taken into account in ray tracing and travel time calculation in the 4of29

Figure 3. Distribution of 784 teleseismic events (circles) used in the present study. The numbers denote the distance in degrees from the study center (triangle). The scales for earthquake magnitude and focal depth are shown at the bottom. study. Travel times were further corrected for the Earth s ellipticity [Dziewonski and Gilbert, 1976] and station elevations. The number of positive absolute residuals is relatively larger than that of negative residuals (Figures 7c and 7d), indicating more rays passing through low-velocity (low-v) anomalies in the mantle, but most of the absolute residuals are within 5.0 s, so I winnowed the data using this threshold. [14] To minimize the effect of the uncertainties of the hypocentral parameters and the lateral heterogeneity outside the model, relative travel-time residuals were used in the teleseismic tomography to determine a 3-D upper mantle structure. The relative residuals were obtained by subtracting the absolute residuals from the average residual for each event. For more details of the calculation of the relative residuals, see Zhao et al. [1994] and Lei and Zhao [2005]. Finally, I obtained 121,606 relative travel-time residuals (Figures 7e and 7f). Figure 7g shows the curves of traveltime data contained in the final data set. From Figures 7e and 7f it can be seen that some relative residuals are relatively large, mainly due to these rays passing through a very inhomogeneous structure in the upper mantle, but about 95.0% relative residuals are between 2.5 s and +2.5 s. Figure 9a demonstrates the distribution of the average relative residuals at each station from all events. The maximal negative and positive residuals can amount to 1.8 s, possibly suggesting that there are strong lateral heterogeneities in the upper mantle. There is a clear pattern of negative and positive residuals in the region. Obviously delayed arrivals appear at the stations around Shanxi rift and Taihangshan mountain, and some relatively weakly delayed arrivals are shown around the Tanlu fault zone and southern portion of the NCC, perhaps suggesting the existence of low-v anomalies in the upper mantle under these areas. Prominently 5of29

Figure 4. (a c) Examples showing original seismograms recorded by NM (Neimeng), HA (Henan), and AH (Anhui) provincial seismic networks for an earthquakes (M 7.6) that occurred near New Guinea on 3 January 2009. The vertical bars show the first P wave arrivals hand-picked. The number on the left of each trace denotes the epicentral distance in degrees. The letters on the left above each trace denote the code of seismic stations in provincial networks. (d) Distribution of seismic stations (solid triangles) in NM, HA, and AH provincial networks and hypocenter (star) of the 3 January 2009 New Guinea earthquake (M 7.6). negative residuals exist in the North China basin, perhaps indicating some high-velocity (high-v) anomalies in the upper mantle and mantle transition zone. Note that the amplitude of the relative residuals in the southwest corner of the study region is very small (Figure 9a), mainly due to a large variation of the residuals there from negative values at NW and NE source quadrants (Figures 9b and 9c) to positive values at SW and SE quadrants (Figures 9d and 9e). 6of29

Table 1. Information About Provincial Seismic Networks, Teleseismic Events, and Arrival Times Used Number of Events Recorded Number of Arrivals Chosen Network Code Full Name Number of Stations AH Anhui 24 532 8,065 BJ Beijing 18 385 4,309 BU Institute of 28 452 5,724 Geophysics, CEA HA Henan 20 586 6,866 HB Hubei 28 620 11,093 HE Hebei 55 655 13,942 JL Jilin 1 287 287 JS Jiangsu 37 449 8,537 LN Liaoning 29 591 10,948 NM Neimeng 25 579 8,035 SD Shandong 37 575 12,468 SN Shaanxi 33 573 11,970 SX Shanxi 31 588 12,350 TJ Tianjin 31 467 7,012 TOTAL 397 784 121,606 This indicates that high-v anomalies may exist in the upper mantle to the north, while low-v anomalies may be present to the south. Nevertheless, the pattern of relative residuals at most stations in the region from all quadrants (Figure 9a) is similar to that from each quadrant (Figures 9b 9e). [15] A 3-D grid was set up in the model to express the 3-D velocity structure. The model was parameterized by an optimal grid spacing of 0.7 0.7 (70 km 78 km) in map view and with grid nodes at depths of 50, 120, 190, 260, 330, 400, 500, 600, and 700 km, as shown in Figure 10, after making many resolution tests with different grid intervals. Such a parameterization of the model is chosen mainly because the pattern of velocity anomalies assumed is generally recovered in most portions of the region (Figures 11 and 12) and the model can provide as much as possible tectonic information. Velocity perturbations at the grid nodes were taken as unknown parameters. I resolved 2758 velocity perturbations at the grid nodes with hit counts (the number of rays passing near each grid node) larger than 10. Note that although seismic anisotropy was documented previously in the region [e.g., Zhao et al., 2007], it was not accounted for here because not considering seismic anisotropy will not significantly affect major structural features inferred [e. g, Wang and Zhao, 2008]. [16] Figure 10 delineates the distribution of hit counts in map view. At 50 190 km depth hit counts have an uneven distribution (Figures 10a 10c) and they are closely related to the distribution of seismic stations (Figure 2), but at 260 400 km depth hit counts become more uniform in the entire study region, though they are still higher in the Chinese capital region (Figures 10d 10f). In the mantle transition zone the region of high hit counts gradually moves eastward to under North China basin and Bohai sea (Figures 10g and 10h), while in the lower mantle hit counts decrease significantly (Figure 10i). Such a pattern in the variation of hit counts with depth is closely related to the fact that most teleseismic events occurred in the northwestern and southwestern Pacific subduction zones (Figure 3). [17] In the present study a conjugate-gradient algorithm LSQR [Paige and Saunders, 1982], with damping and firstorder smoothing regularizations [Zhao, 2001, 2004; Lei and Zhao, 2006], was used to resolve the large and sparse system of observational equations. By considering the balance between the reduction of travel-time residuals and the smoothness of the 3-D velocity model obtained [e.g., Eberhart-Phillips, 1986], the optimal value of the damping parameter is found to be 30.0 in the present study after performing many inversions with different values of the damping parameter. The optimum horizontal and vertical smoothing parameters are determined to be 0.025 and 0.025, respectively, by the cross validation contour, where yields minimal prediction errors. The cross validation contour is constructed by a series of trial solutions with many different pairs of horizontal and vertical smoothing parameters. For details, see Inoue et al. [1990]. [18] Teleseismic rays are nearly vertical in the shallow crust, so teleseismic tomography cannot resolve the crustal structure well, which is usual for teleseismic tomography. To better image the mantle structure, some researchers applied a joint inversion of complementary body and surface wave data that can close the gap between well resolved lithospheric and mantle structures [e.g., West et al., 2004; Obrebski et al., 2012] or the integration of local and teleseismic data in tomographic inversion that can yield a good crisscross ray coverage in the shallow crust [e.g., Zhao et al., 1994; Lei et al., 2009], while others corrected the theoretical teleseismic travel times by using a 3-D velocity crustal model with an undulating Moho discontinuity [e.g., Hung et al., 2004; Lei and Zhao, 2005; Zhao et al., 2006]. Here, the latter was adopted by correcting the travel times for the crust heterogeneities by using the CRUST2.0 model [Bassin et al., 2000] (http://igpweb.ucsd.edu/gabi/rem.html) which is an updated version of CRUST5.1 [Mooney et al., 1998] and is specified on a 2 2 grid in the horizontal directions. 4. Resolution Analyses and Results 4.1. Resolution Analyses [19] The checkerboard resolution test was used to examine the resolution scale of seismic data and evaluate the ray coverage in the study region. Many checkerboard resolution tests were conducted by changing the grid spacing of the model. In this paper I only show the results of the checkerboard resolution test with grid intervals of 0.7 0.7 in the horizontal directions (Figures 11 and 12). Alternative negative and positive velocity anomalies of up to 2% were assigned to the 3-D grid nodes that were arranged in the modeling space. Random noise with zero mean and a standard deviation of 0.2 s was added to the synthetic travel times to account for data errors usually present in a real data set. Output models were obtained using the same algorithm and the same numbers of seismic stations, events, and raypaths as in the real inversion. By examining the inverted image of the checkerboard, one can easily understand where the resolution is good and where it is poor. [20] Figure 11 illustrates the results of one of the checkerboard resolution tests. At 50 km depth the areas with good resolution (Figure 11a) are closely related to the distribution of seismic stations (Figure 2), which is consistent with the distribution of hit counts (Figure 10a). At 120 and 190 km depth the good resolution area gradually increases, including Bohai Sea and northern Ordos block (Figures 11b and 11c), 7of29

Figure 5. (a) Spatial distribution of the numbers of teleseismic events recorded at each seismic station. The number scale is shown in the top left corner. The maximum number is over 540. The triangle denotes Datong volcano (DT). Thick lines denote the boundaries of eastern, western, and central NCC, while thinner lines denote major active faults [Deng et al., 2002]. BHS, Bohai Sea; HHS, Huanghai Sea. (b) Histogram of the number of seismic stations versus the number of teleseismic events recorded. though the amplitude of anomalies at some grid nodes is not retrieved completely yet. At 260 km depth the pattern of velocity anomalies is generally reconstructed, but the amplitude of either high-v or low-v anomalies are recovered better (Figure 11d), perhaps indicating a non-uniform crisscrossing distribution of seismic rays there. At 330 600 km depth the resolution is improved greatly (Figures 11e 11h), maybe attributable to a much better crisscross ray coverage there, but at 700 km depth the good resolution area is much reduced (Figure 11i). [21] In order to display the spatial resolution along vertical cross-sections for which tomographic images are shown, here the star-cross way is used to express the resolution [Lei and Zhou, 2002], because these cross sections do not pass right through the grid nodes in the modeling space (Figure 12). Stars and crosses denote the grid nodes where the pattern of 8of29

Figure 6. (a) Spatial distribution of the numbers of seismic stations that recorded the corresponding teleseismic event. The scale for the numbers of seismic stations is shown on the bottom left corner. The numbers in the map denote the distance in degrees from the study center (triangle). (b) Histogram of the numbers of teleseismic events recorded by seismic stations. input velocity anomalies is recovered correctly and wrongly after the inversion, respectively. The size of star and cross symbols denotes the ratio of the recovered amplitude of velocity anomaly to that in the initial model. The stars with values of 100% show the grid nodes where the checkerboard model is recovered perfectly. For details of this expression method, see Lei and Zhou [2002] and Lei and Zhao [2005]. It is found that the resolution is generally good along all vertical cross sections, except for some places around Ordos block and in the shallower mantle, where the amplitude of velocity anomalies is not retrieved well but the pattern has been retrieved (Figure 12), suggesting that these vertical cross section areas generally have a good resolution. 4.2. Tomographic Results [22] Figure 13 illustrates the resulting tomographic images in map view. There exist obvious differences in pattern of velocity anomalies in the upper mantle between eastern, 9of29

Figure 7. (a) Travel times of P wave data hand-picked versus epicentral distance in degrees. (b) The same as Figure 7a but for calculated travel times. (c) Absolute travel-time residuals computed by subtracting observed travel times in Figure 7a from calculated ones in Figure 7b versus epicentral distance in degrees. Black dots show the residuals smaller than 5.0 s, while gray dots denote the residuals larger than 5.0 s. (d) Histogram of absolute travel-time residuals. (e) Relative travel-time residuals versus epicentral distance in degrees. (f) Histogram of relative travel-time residuals. (g) The same as Figure 7a but excluding misidentified data. western, and central NCC. In the eastern part of eastern NCC, prominent low-v anomalies are visible under the Tanlu fault zone at depths of 50 260 km (Figures 13a 13d), while in the western part of eastern NCC, an obvious NE-SW trending high-v anomaly ( Re ) is found at 50 190 km depths (Figures 13a 13c), maybe indicating the remainder of detached lithosphere. In central NCC prominent and continuous low-v anomalies are detected at depths of 50 260 km beneath the Shanxi rift, Datong volcano, and Taihangshan mountain (Figures 13a 13d). Moreover, at greater depth the low-v anomaly under Datong volcano is separated into two branches. One shifts westward with depth and extends down to 400 km depth under Hetao rift (Figures 13d 13f), while the other moves eastward with depth and extends down to 700 km depth under eastern NCC (Figures 13d 13i). Under the western NCC, some high-v 10 of 29

Figure 8. (a) The 1-D velocity model. The velocity used in the crust (dotted line) is deduced from the CRUST2.0 model [Bassin et al., 2000], while the IASP91 velocity model [Kennett and Engdahl, 1991] was directly adopted for the mantle. Dashed lines correspond to the Moho, 410-km and 660-km discontinuities, respectively. (b) The Moho discontinuity in map view. The depth scale is shown at the bottom. Black lines indicate the boundaries of eastern, western and central NCC, while the gray line denotes the North-South Gravity Lineament (NSGL). Dashed lines mark the outlines of the Hetao rift (HTR) and Shanxi rift (SXR) and the traces of the Zhang-Bo seismic zone (ZBZ) and Tanlu fault zone (TLF). anomalies ( In ) are visible down to 400 km depth under the Ordos block (Figures 13a 13f), suggesting intact lithosphere there. Some high-v anomalies ( De ) are also found under eastern and central NCC at depths of 260 400 km (Figures 13d 13f), perhaps indicating detached lithospheric material there. At 600 km depth a broad high-v anomaly representing the stagnant Pacific slab is imaged under eastern and central NCC but with a low-v gap from under Datong volcano to the edge of Bohai Sea (Figure 13h), suggesting that the Pacific slab has subducted under central NCC but left room for hot material upwelling from the lower mantle. At 700 km depth a similar pattern of velocity anomalies is imaged (Figure 13i), suggesting that the collapsed slab material has fallen down to the lower mantle. At depths of 500 700 km there is one more obvious low-v anomaly imaged under the Qinling orogenic belt (Figures 13h 13i). In addition, it is found that most strong earthquakes are underlain by obvious low-v anomalies in the mantle (Figures 13a 13f). [23] Figure 14 shows tomographic images along seven vertical cross sections. Along cross section AA it is clearly seen that there exist large differences in pattern of velocity anomalies under eastern and western Yanshan (Figure 14a). Under the eastern Yanshan, high-v anomalies are imaged in the upper mantle, indicating thick lithosphere, while under western Yanshan low-v anomalies are visible, likely indicating the northward extension of Shanxi rift. Along cross section BB an obvious Y-shaped low-v anomaly is detected under Datong volcano and Bohai Sea from the upper to lower mantle (Figure 14b), perhaps suggesting that it is a lower mantle plume. Along cross section CC the Ordos block appears as high-v anomalies ( In ) above 400 km depth, while central and eastern NCC shows prominent low-v anomalies (Figure 14c). Along cross section DD (Figure 14d) the image is somewhat similar to the one shown along CC in the upper mantle but there are also some differences in pattern of anomalies between the two images. Along DD high-v anomalies ( In ) under the Ordos block are more pronounced than those along CC, possibly due to the cold lithospheric material affected by hot asthenospheric material under the Hetao rift (Figure 14b), as suggested by Tian et al. [2011], but low-v anomalies under eastern NCC along CC are generally replaced by high-v anomalies ( Dr ) along DD. Cross section EE mainly passes through Shaanxi graben to eastern NCC (Figure 14e). Low-V anomalies are imaged under Shaanxi graben above 300 km depth, suggesting the existence of hot material upwelling there, while high-v anomalies ( De ) are found at 300 400 km depth under central NCC and at 200 300 km depth under eastern NCC (Figure 14e), perhaps representing the detached lithosphere there. Cross section FF generally runs along Shanxi rift (Figure 14f). Above 300 km depth obvious low-v anomalies likely indicating upwelling hot material are observed, while at 300 400 km depth high-v anomalies ( De ) representing the detached lithosphere are 11 of 29

Figure 9. (a) Distribution of relative travel-time residuals at each station from all events. Blue colors denote negative residuals, while red colors denote positive residuals. The scale for the residuals is shown on the right. Thick lines indicate the boundaries of eastern, western, and central NCC, while thin lines denote major active faults [Deng et al., 2002]. The yellow triangle corresponds to Datong volcano. (b e) Distributions of relative residuals at each station from NW, NE, SW, and SE source quadrants. imaged (Figure 14f). Cross section GG mainly runs along the eastern portion of eastern NCC (Figure 14g). High-V anomalies ( Re ) representing the remainder of detached lithosphere are imaged above 200 km (Figure 14g), while obvious low-v anomalies are detected at 200 400 km depth. Similar to the images as shown in map view (Figure 13), all vertical cross sections show obvious high-v anomalies in the mantle transition zone and lower mantle under eastern and central NCC (Figure 14), suggesting that the Pacific slab has subducted westward to under central 12 of 29

Figure 10. Distribution of hit counts (number of rays passing around each grid node) in map view. Small triangles denote the seismic stations used in the study, while the big triangle denotes Datong volcano. Thick lines denote the boundaries of eastern, western and central NCC. Depth values are shown on the top of each slice. The scale for hit counts is shown at the bottom. Here the square root (SQRT) of hit counts is shown. NCC and collapsed down to the lower mantle. Moreover, a strong low-v anomaly is detected under the Qinling orogenic belt, suggesting the existence of hot material upwelling there (Figures 14e and 14f). 4.3. Synthetic Tests [24] To confirm some specific features of the images obtained in the present study, a number of synthetic tests have been performed by changing the diameter and morphology of velocity anomalies assumed in the input model. Among them, four synthetic tests are illustrated in Figure 15. All the procedures are the same as in the checkerboard resolution test, and only differences between them are in input models. In the synthetic test, only specific features similar to those in the obtained images are assumed in the input model. In the first and second tests, a continuously and discontinuously Y-shaped low-v anomaly of up to 2% under Datong volcano and Bohai Sea extending down to the lower mantle and a high-v anomaly of up to 2% above 200 km depth between these two low-v anomalies are put in the input model, respectively (Figures 15a and 15c). In the third test, 13 of 29

Figure 11. Results of one of the checkerboard resolution tests. Grid spacing is 0.7 0.7 in the horizontal directions. Solid and open circles denote high-v and low-v anomalies, respectively. The scale for velocity perturbations (in %) is shown at the bottom. The open triangle denotes Datong volcano. Depth values are shown at the top of each slice. Thick lines denote the boundaries of eastern, western and central NCC. a broad low-v anomaly of up to 2% is put in the mantle transition zone under the Qinling orogenic belt in the input model (Figure 15e). In the fourth test, a 100-km-thick high-v anomaly of up to 2% around 600 km depth, having a gap with no velocity anomalies, is put in the mantle transition zone in the input model (Figure 15g). The output models (Figures 15b, 15d, 15f, and 15h) show a similar pattern as the input models, though there are some differences between them in amplitude and there is some smearing in the output models. In particular, in the second test, some separated low-v blocks under Datong volcano to Bohai Sea are not connected to a continuous low-v anomaly after the inversion (Figures 15c and 15d), strongly implying that the continuously Y-shaped low-v anomaly is a reliable feature. Other features in the resulting model, such as a strong low-v anomaly under the Qinling orogenic belt in the mantle transition zone, and broad high-v anomalies representing the Pacific slab with a narrow low-v gap in the mantle transition zone, have also been demonstrated to be generally robust. 4.4. A Restoring Test [25] To evaluate all the main features that are present in the obtained results, a restoring test was performed (Figure 16) by taking the current images (Figure 14) obtained from the 14 of 29

Figure 12. The same as Figure 11 but for vertical cross sections. The star-cross way is used to express the resolution because these cross sections are not passing right through the grid nodes of the model. Stars denote the grid nodes where the pattern of the input velocity anomalies is retrieved correctly after the inversion, while crosses denote the grid nodes where the pattern of input velocity anomalies is wrongly recovered after the inversion. The scale for the degree of recovery (in %) is shown below Figure 12f. For details, see Lei and Zhou [2002] and Lei and Zhao [2005]. YSM, Yanshan mountain; BHS, Bohai Sea; CNCC and ENCC, central and eastern NCC; OB, Ordos block; DBM, Dabie mountain; DT, Datong volcano. Locations of cross sections are shown on the insert map. 15 of 29

Figure 13. The resulting tomographic images in map view. Red and blue colors denote low-v and high- V anomalies, respectively. The scale for velocity perturbation (in %) is shown at the bottom. Thick lines denote the boundaries of eastern, western and central NCC. Dashed lines correspond to the Hetao rift, Shanxi rift, Zhang-Bo seismic zone, and Tanlu fault zone. The open triangle denotes Datong volcano, while color circles mark the earthquakes with magnitude larger than and equal to 7.0 since BC 780 [Song et al., 2011]. All the earthquakes (M 7.0) are plotted on the slices (Figures 13a 13f) to see how large earthquakes are related to mantle images. The scale for earthquake magnitude is shown at the bottom. Depth values are shown above each map. In, the intact lithosphere; De, the detached lithosphere; Re, the remainder of detached lithosphere. real data set as an input model. The procedure is the same as that used in the checkerboard resolution test. The results of the restoring test show that the pattern of velocity anomalies is well retrieved, though there are some minor differences between input and output models in amplitude and morphology (Figure 16). This restoring test along with the checkerboard resolution test (Figures 11 and 12) and synthetic tests (Figure 15) demonstrates that all the structural features that appear in the tomographic images (Figure 14) are generally robust. 5. Discussion 5.1. Comparison With Previous Tomographic Models [26] The present tomographic model not only displays the general features contained in the previous models, but also reveals some new features. For example, above 200 km 16 of 29

Figure 14. The same as Figure 13 but for seven vertical cross sections. Red and blue colors denote low-v and high-v anomalies, respectively. The scale for velocity perturbation (in %) is shown below Figure 14f. Crosses denote small earthquakes (M < 7.0) between 1964 and 2005, which were reprocessed by Engdahl et al. [1998], while color circles denote large earthquakes (M 7.0) since BC 780 [Song et al., 2011] within 35 km off the profile. The scale for earthquake magnitude is shown at the bottom of Figure 14g. Dashed lines denote the Moho, 410-km and 660-km discontinuities, respectively. The Moho discontinuity is extracted from the CRUST2.0 model [Bassin et al., 2000]. The polygon above each map denotes the topography along the cross-section. The triangle denotes Datong volcano (DT). YSM, Yanshan mountain; BHS, Bohai Sea; CNCC and ENCC, central and eastern NCC; OB, Ordos block; DBM, Dabie mountain. Locations of cross sections are plotted on the insert map. In, the intact lithosphere; De, the detached lithosphere; Re, the remainder of detached lithosphere; Dr, the dripping lithosphere. 17 of 29

depth an arc-shaped strong low-v anomaly under the CNCC is imaged from east to south of the Ordos block in the present model (Figures 13a 13c), while the models of Tian et al. [2009] and Zhao et al. [2009] show a similar feature east of the Ordos block but some high-v anomalies south of the Ordos block. In the western portion of the ENCC, the present model shows a prominent, continuous high-v anomaly at 50 and 120 km depth (Figures 13a and 13b), Figure 15 18 of 29

while in the models of Tian et al. [2009] and Zhao et al. [2009] there are only some intermittent, weaker high-v anomalies. More importantly, the present model reveals a continuously Y-shaped low-v anomaly under Datong volcano and Bohai Sea ascending from the lower mantle through a gap of the stagnant Pacific slab in the mantle transition zone (Figures 13h, 13i, 14b, 17, and 18). These significant improvements can be attributable to the highquality arrival-time data used. In this sub-section, the results are only compared with the models of Tian et al. [2009] and Zhao et al. [2009] mainly because these two models have a similar resolution scale and size of the study region as in the present study. For details of further comparison, see the following sub-sections. 5.2. Deep Structure Under the NCC [27] The present tomographic model illustrates obvious lateral structural contrasts not only between eastern, central and western NCC, but also between the eastern and western portions of eastern NCC (Figures 13 and 14). In the eastern part of eastern NCC, obvious low-v anomalies were detected (Figures 13a 13d, 14b, and 14c), which is consistent with previous studies [e.g., Huang and Zhao, 2009; Tian et al., 2009; Zhao et al., 2009], but there are some differences between them in depth extent of low-v anomalies. Zhao et al. [2009] showed the low-v anomalies extending down to the lower mantle, and Tian et al. [2009] illustrated the low-v anomalies above 400 km depth, while the present results document the low-v anomalies only above 200 300 km depth (Figures 14b and 14c), which is generally consistent with the results of Huang and Zhao [2009]. Either way, these results all suggest the existence of the upwelling of hot material in the asthenosphere under the eastern portion of eastern NCC. However, in the western portion of eastern NCC, the present tomographic model shows an obvious arcshaped high-v anomaly ( Re, Figures 13a 13c, 14b 14e, and 14g). This feature confirms what was already observed by previous studies [e.g., Tian et al., 2009; Zhao et al., 2009; Li et al., 2011], although there exist some differences in details between the present work and previous results. Tian et al. [2009] showed an approximate north-south trending high-v anomaly extending down to only 300 km depth, while Zhao et al. [2009] showed a weaker high-v anomaly extending down to the mantle transition zone. Li et al. [2011] also showed a prominent high-v anomaly in a relatively small area in the uppermost mantle, due to using only Pn arrival-time data. In contrast, the present results show a strong high-v anomaly only above 200 km depth ( Re, Figures 13a 13c, 14b 14e, and 14g), which has been demonstrated to be robust by extensive resolution tests (Figures 11, 12, 15, and 16). If this depth denotes the bottom of the lithospheric mantle, then it is well consistent with depth estimates from receiver function analyses [Chen, 2010] and surface wave tomography [Huang et al., 2009]. All these results suggest that the lithospheric remainder of the Archean Craton still exist in the western part of eastern NCC. [28] Under central NCC some strong low-v anomalies are visible down to 300 400 km depth or deeper (Figures 13 and 14), which confirms what was already observed by Tian et al. [2009] and Zhao et al. [2009]. Under the Ordos block some prominent high-v anomalies are observed down to about 400 km depth ( In, Figures 13a 13f, 14c, and 14d), which has further been demonstrated by previous tomographic studies [e.g., Tian et al., 2009; Zhao et al., 2009], suggesting that western NCC has not been destructed yet. In addition, some obvious high-v anomalies ( De ) exist at 200 400 km depth under central and eastern NCC, but they are not connected to those above 200 km depth (Figures 13d 13f and 14b 14g), suggesting the existence of detached lithosphere in the upper mantle. These results are well consistent with those as shown in previous tomographic results [e.g., Huang and Zhao, 2009; Tian et al., 2009; Xu and Zhao, 2009; Zhao et al., 2009], although there exist some significant differences between them in details. For example, the models of Huang and Zhao [2009] and Zhao et al. [2009] show some intermittent and nearly vertical high-v anomalies at 200 400 km depth, and the model of Tian et al. [2009] exhibits small localized high-v anomalies at these depths. In addition to some localized ( De, Figures 14b 14d and 14g) and nearly vertical ( Dr, Figures 14d, 14e, and 14g) high-v anomalies, the present model also displays some horizontal high-v anomalies ( De, Figures 14e and 14f), while Xu and Zhao [2009] only exhibit horizontal high-v anomalies above 400 km depth. Petrological and geochemical studies also suggested that most of the ancient lithospheric mantle was lost due to the detachment, and the present lithospheric mantle was formed after late Mesozoic [e.g., Wu et al., 2000; Xu, 2001; Gao et al., 2002]. [29] In contrast, some high-v anomalies at 200 400 km depth under the southern portions of the NCC are connected to those above 200 km depth with weaker anomalies, maybe indicating the NCC lithosphere is dripping ( Dr, Figures 14d, 14e, and 14g). Other examples have been documented in the western United States such as the Colorado Plateau [e.g., Roy et al., 2009; Obrebski et al., 2011], and the southern and central Sierra Nevada [e.g., Zandt et al., 2004; West et al., 2009]. The formation of a lithospheric drip may be a common type of a viscous Rayleigh-Taylor gravitational instability where the lithosphere deforms. The primary difference from the detachment is that the mantle lithosphere is Figure 15. Four synthetic tests. (a, c, e, g) Input and (b, d, f, h) output models, respectively. Red and blue colors denote low-v and high-v anomalies. Color circles denote large earthquakes (M 7.0) since BC 780 within 35 km off the profile. The scales for velocity perturbation and earthquake magnitude are all shown below Figure 15e. Locations of cross sections in Figures 15a 15f are shown on the insert map below Figure 15f. Depth values of Figures 15g and 15h are shown in the upperleft corner. Other symbols in Figures 15a 15f and Figures 15g and 15h are the same as those shown in Figures 13 and 14, respectively. The purpose of the tests (Figures 15a 15d) is to confirm if the low-v anomalies under Datong volcano and Bohai Sea continuously ascend from the lower mantle. The objective of the tests (Figures 15e and 15f) is to affirm whether the broad low-v anomaly representing hot material upwelling under Qinling orogenic belt is a reliable feature. The aim of the test (Figures 15g and 15h) is to clarify if the broad high-v anomaly representing the stagnant Pacific slab with a gap under eastern and central NCC in the mantle transition zone is robust. 19 of 29

Figure 16. Results of one of the restoring tests. Input models are tomographic images, as shown in Figure 14, obtained from the real data set. Symbols are the same as Figure 14. Random noise with zero mean and a standard deviation of 0.2 s was added to the synthetic travel times. not removed as a coherent lithospheric slice, but deforms in a dripping manner [Zandt et al., 2004; Gogus and Pysklywec, 2008]. These results suggest that some portions of the NCC lithosphere could be suffering from a dripping process due to a gravitational instability. [30] Differences found between eastern and western NCC in the upper mantle generally confirm what has already been inferred from other geophysical observations and geochemical and petrological studies. For example, the NCC is divided into two parts by the NSGL that runs over 3500 km from south China to northeast China and is approximately parallel to Tanlu fault zone. East of the NSGL, the craton is characterized by a thin crust and lithosphere, high heat flow, and weakly negative to positive regional Bouguer anomalies, 20 of 29

Figure 17. (a, b) The same as Figures 13h 13i but with interpretative graphics in thick dashed lines, providing a basis for understanding the stagnant Pacific slab in the mantle transition zone and the collapsed slab material in the lower mantle under the NCC. (c) The configuration of the stagnant Pacific slab deduced from the tomographic images in the mantle transition zone and lower mantle, as shown in Figures 17a and 17b. CNCC and ENCC, central and eastern NCC. The depth is not to scale, so the vertical axis is shown with a zigzag. 21 of 29

Figure 18. A cubed view of the P wave velocity model with the interpretation of observed structure. Red and blue colors denote low-v and high-v anomalies, respectively. The color scale for velocity perturbation is shown on the right. The topography on the top is not to scale. White arrows indicate the directions of the hot material upwelling. Re, the remainder of detached lithosphere; De, the detached lithosphere; HTR, Hetao rift; SXR, Shanxi rift; NCB, North China basin; BHS, Bohai Sea; DT (triangle), Datong volcano. while west of the NSGL the craton is characterized by thick crust and lithosphere, low heat flow, and strongly negative Bouguer gravity anomalies [e.g., Ma, 1987; Griffin et al., 1998; Hu et al., 2000; Huang et al., 2009; Chen, 2010]. Furthermore, mantle xenoliths from Paleozoic diamondbearing kimberlites [Griffin et al., 1998; Zheng and Lu, 1999] and Cenozoic basalts [Xu et al., 1996; Fan et al., 2000; Gao et al., 2002] demonstrate a significant variation in lithosphere thickness from 200 km in the Paleozoic [e.g., Griffin et al., 1998] to 80 110 km in the Cenozoic [e.g., Xu et al., 1995; Chen et al., 2001]. Compositional variations suggest a diachronous lithospheric thinning process between eastern and western NCC [e.g., Xu et al., 2004; Xu, 2007]. The lithospheric thinning of eastern NCC occurred during late Mesozoic, while that of western NCC occurred in Cenozoic time [e.g., Xu, 2007]. The lithospheric thinning beneath eastern NCC was accompanied by heterogeneous replacement by fertile peridotites in Mesozoic-Cenozoic time [e.g., Zheng et al., 1998, 2005]. [31] Various causes for the thinning of the lithosphere of the NCC have been proposed, such as the Indo-Asian collision [e.g., Menzies et al., 1993], the piecing together of South and North China blocks [e.g., Li et al., 1993; Yin and Nie, 1993; Xu, 2001; Gao et al., 2002], a mantle plume [e.g., Zheng et al., 1998; Xu, 2002], and the westward subduction of the Pacific plate [e.g., Ye et al., 1987; Griffin et al., 1998; Wu and Sun, 1999; Wu et al., 2000]. The present tomographic model favors the roles of both the subduction of the Pacific plate under eastern China and a mantle plume from the lower mantle in the reactivation and destruction of the NCC lithosphere. This conclusion is based on the present results which clearly show some broad high-v anomalies in the mantle transition zone and lower mantle under eastern and central NCC (Figures 13h, 13i, 14, and 17) and a 22 of 29

continuously Y-shaped low-v anomaly under Datong volcano and Bohai Sea extending from the upper mantle to lower mantle (Figure 14b). The present tomographic model shows the high-v anomalies in the mantle transition zone (Figures 13h, 13i, 14, and 17), which confirms what was previously observed by tomographic studies [e.g., Huang and Zhao, 2009; Tian et al., 2009; Xu and Zhao, 2009; Obrebski et al., 2012] and receiver function analyses [e.g., Chen and Ai, 2009; Wang and Niu, 2011; Xu et al., 2011]. Furthermore, the Y-shaped low-v anomaly clearly imaged in the present model (Figure 14b) is consistent with a thinned mantle transition zone there from receiver function analyses [Chen and Ai, 2009; Xu et al., 2011] and has been demonstrated by extensive resolution tests (Figures 11, 12b, 15a 15d, and 16b). These results suggest that this low-v structure may denote a mantle plume. [32] Therefore, I infer that the reactivation and destruction of the NCC were primarily caused by the detachment of cold lithospheric material and thermomechanical erosion due to the hot material upwelling. The hot material upwelling could be not only due to the dehydration of the stagnant Pacific slab in the mantle transition zone under eastern and central NCC, but also could be directly related to a mantle plume under Datong volcano and Bohai Sea that originates from the lower mantle. 5.3. Deep Origin of the Datong Volcano [33] Global and large-scale regional tomographic models revealed a prominent low-v anomaly under Datong volcano down to 400 km depth, and this anomaly is connected with low-v zones under North China basin, Bohai Sea, and Japan Sea. These low-v anomalies are underlain by the subducted and stagnant Pacific slab. These results suggest that the Datong volcano could be related to mantle upwelling induced by the stagnancy and dehydration of the Pacific slab in the mantle transition zone [e.g., Zhao, 2004; Huang and Zhao, 2006; Lei and Zhao, 2006; Li et al., 2008; Li and van der Hilst, 2010]. However, these results also demonstrated that the western edge of high-v anomalies representing the stagnant Pacific slab in the mantle transition zone is about 119 120 E, which is far away from Datong volcano (about 113.3 E). In addition, recent small-scale seismic tomographic studies showed that low-v anomalies in the upper mantle under Datong volcano and the Japan Sea are separated by prominent high-v anomalies representing the detached lithosphere and its remainder [e.g., Tian et al., 2009; Zhao et al., 2009]. The present tomographic model illustrates that two prominent low-v anomalies are visible under Datong volcano and Bohai Sea above 200 km depth, and are separated by a strong high-v anomaly ( Re ) representing the remainder of detached lithosphere (Figure 14b). Below 200 km depth, these two low-v anomalies are together connected to a broad low-v anomaly that ascends from the lower mantle. Furthermore, some high-v anomalies ( De ) representing the detached lithosphere above 400 km depth are clearly seen to the east of this broad low-v anomaly (Figure 14b). These low-v anomalies formed a Y-shaped structure, suggesting that the low-v anomalies under Datong volcano and Bohai Sea are the mantle plume, having the same origin, from the lower mantle. [34] The volcanic activity is considered to be one of the most important factors in the formation and evolution of the ancient Bohai Sea. In Mesozic time from Jurassic to Cretaceous the volcanism was very active, and formed extensive basalts and pyroclastic rocks there. Some geophysical features around Baohai Sea were summarized by Teng et al. [1997] as follows: (1) a thinner crust of 28 29 km thickness, (2) a high heat flow value of 56 77 mw/m 2, (3) the existence of low-v anomalies in the crust and upper mantle down to about 120 km depth, (4) a shallower highconductivity layer of 50 60 km depth, (5) the mantle material scattered from a mantle convective center around Bohai Sea when the upwelling material reached the bottom of the crust, (6) the presence of hot material upwelling from the deep mantle to the crust. Based on these geophysical observations above, the Bohai Sea plume was hypothesized by Teng et al. [1997]. However, due to the data limited in previous studies, it has been difficult to understand the origin of the Bohai Sea plume. Some recent tomographic models showed a low-v anomaly under Bohai Sea at depths of about 60 300 km [Tian et al., 2009], 150 300 km [Zhao et al., 2009], and 50 150 km [Huang and Zhao, 2009]. These differences in depth range may be due to different data sets used. The present results from high-quality arrival-time data show that a strong low-v anomaly under Bohai Sea extends down to only 200 km but is connected to the low-v anomaly from the lower mantle under North China basin (Figure 14b). This structure strongly supports the existence of the Bohai Sea plume and its origin from the lower mantle. [35] In the mantle transition zone, in addition to a Y-shaped low-v anomaly under Datong volcano to the edge of Bohai Sea, there also exist broad high-v anomalies representing the stagnant Pacific slab (Figures 13h and 17). This not only confirms what was already observed by tomographic results of Tian et al. [2009], but also has been demonstrated to be a reliable feature by extensive resolution tests (Figures 11, 12, 15, and 16). The stagnant Pacific slab in the mantle transition zone could dehydrate and cause the hot material upwelling of asthenosphere in the upper mantle [e.g., Zhao, 2004; Lei and Zhao, 2005; Li and van der Hilst, 2010]. At 700 km depth there is also a similar pattern of velocity anomalies (Figures 13i and 17b) as shown in the mantle transition zone, suggesting that the Pacific slab has been stagnant in the mantle transition zone for a long time (ca. 100 140 Ma) [e.g., Fukao et al., 1992; Zhao, 2004; Li and van der Hilst, 2010] and then finally collapsed down to the lower mantle. Receiver function analyses showed a thickened mantle transition zone under eastern and central NCC but a thinned mantle transition zone under Datong volcano [e.g., Chen and Ai, 2009; Wang and Niu, 2011; Xu et al., 2011]. Either way, all these results suggest that the origin of the Datong volcano and Bohai Sea upwelling may be related to the dehydration of the stagnant Pacific slab in the mantle transition zone, but the present study emphasizes that the low-v anomalies under Datong volcano and Bohai Sea could represent a mantle plume from the lower mantle. The upwelling from the lower mantle could have many causes, such as the tearing of the slab [e.g., Ai et al., 2008; Zhu et al., 2011; Liu and Stegman, 2012] or the existence of a thermal boundary layer [e.g., Sleep et al., 1988; Steinberger and Torsvik, 2012]. In the future, higher-resolution seismic imaging along with geological, geochemical, and other geophysical investigations would be still required to provide 23 of 29

a better understanding of the origin of the Datong volcano and Bohai Sea upwelling. 5.4. Deep Structure of the Zhang-Bo Seismic Zone, Tanlu Fault Zone, and Shanxi Rift 5.4.1. The Zhang-Bo Seismic Zone [36] Several investigations have been made around the Zhang-Bo seismic zone using various geophysical approaches. Corresponding to the intersection of NW with NE oriented faults, the Zhang-Bo zone is divided into five segments: Zhangbei-Huailai, Nankou-Sanhe, Tianjin-Tanggu, center of Bohai, and Penglai-Yantai segments [Xu et al., 1998]. A recent high-resolution tomographic model of the crust [Lei et al., 2008] inferred from direct P and PmP (Moho reflected wave) data supports the existence of some significant tectonic differences along this zone. Geometries of low-v anomalies within the Zhang-Bo zone are complex and low-v anomalies become narrower from the southeast to the northwest along the zone. These contrasts may reflect different dynamic evolutional processes of the Taihangshan and Yanshan uplifts and North China depression basin. Furthermore, low-v anomalies along the zone may deepen to the uppermost mantle in the northeast and southwest portions [Lei et al., 2008]. The present mantle tomographic model illustrates obvious low-v anomalies in the Zhangbei- Huailai, Bohai Sea, and Penglai-Yantai segments at 50 km depth, and they shift northward to the north of the Nankou- Sanhe and Tianjin-Tanggu segments (Figure 13a). Such a structural feature may extend down to 120 km depth (Figure 13b), but at 190 km depth these low-v anomalies are right located under the Zhang-Bo seismic zone (Figure 13c). Combining previous studies [e.g., Lei et al., 2008], the present results suggest that the Zhang-Bo seismic zone may provide a channel of hot material upwelling from the upper mantle, which confirms what was previously observed by tomographic studies [e.g., Huang and Zhao, 2009]. 5.4.2. The Tanlu Fault Zone [37] Petrological and geochemical studies demonstrated that the Cenozoic peridotites with fertile fine-grained and foliated textures are mainly composed of lithospheric mantle beneath the eastern part of the NCC [e.g., Griffin et al., 1998; Chen et al., 2001; Gao et al., 2002], especially along the Tanlu fault zone [e.g., Xu et al., 1996]. A receiverfunction analysis by Chen et al. [2006] showed a 60 80 km thick present-day lithosphere around the Tanlu fault zone, which is much thinner than the Paleozoic lithosphere with 180 km thickness. Furthermore, both the Moho discontinuity and lithosphere-asthenosphere boundary are uplifted right under the Tanlu fault zone [Chen et al., 2006]. The present tomographic model illustrates obvious low-v anomalies under the Tanlu fault zone at 50 260 km depth (Figures 13a 13d and 14), which is generally consistent with previous studies [e.g., Tian et al., 2009]. All these results may indicate that the Tanlu fault zone may have acted as a major channel of upwelling of hot asthenospheric material during the Mesozoic-Cenozoic continental extension and lithospheric thinning in eastern China. 5.4.3. The Shanxi Rift [38] The crustal velocity model showed a clear low-v anomaly zone under the Shanxi rift [e.g., Chang et al., 2007]. Xu and Ma [1992] summarized previous results from deep seismic soundings and magnetotelluric studies [Sun et al., 1988], and found the widespread presence of low-v anomalies and high-conductivity layers under the rift in the crust. The present tomographic model illustrates the existence of obvious low-v anomalies along the rift extending down to 300 400 km depth (Figures 13 and 14). Such a structural feature confirms what was already observed by previous tomographic studies [e.g., Zhao et al., 2009; Tian et al., 2009], suggesting the existence of hot material upwelling in the upper mantle. This is somewhat similar to that under the Baikal rift zone [e.g., Petit et al., 1998; Zhao et al., 2006]. 5.5. Causes of Large Earthquakes [39] Previous studies revealed that the hypocenters of strong earthquakes in North China are underlain by obvious low-v anomalies in the crust [e.g., Huang and Zhao, 2004; Qi et al., 2006; Lei et al., 2008, 2011]. A similar structure is found under the source areas of the 17 January 1995 Kobe earthquake (M 7.2) in Japan, the 26 January 2001 Bhuj earthquake (M 7.6) in India, the 12 May 2008 Wenchuan earthquake (M 8.0) in China, and some strong earthquakes around the central and western Tien Shan orogenic belt [e.g., Zhao et al., 1996; Mishra and Zhao, 2003; Lei and Zhao, 2009; Lei, 2011]. These low-v anomalies may reflect an overpressurized, fluid-filled, fracture rock matrix, which might have contributed to the initiation of these large earthquakes. The existence of overpressurized fluids under the source area has been extensively demonstrated around the world [e.g., Johnson and McEvilly, 1995; Gupta et al., 1996; Miller, 1996]. [40] Various possible sources of fluids in the crust have been advocated by several researchers, including the dehydration of hydroxyl-bearing minerals in the crust, fluids trapped in pore spaces, and meteoric water, but mantle sources of fluids may be different. For example, the 1995 Kobe earthquake may be related to the dehydration of the oceanic crust on the top of the subducting Philippine Sea slab [Zhao et al., 1996]. In the present study, clear low-v anomalies are visible under the source areas of most strong earthquakes in the upper mantle, and some high-v anomalies appear in the mantle transition zone under eastern and central NCC. Furthermore, under Datong volcano and Bohai Sea a continuously Y-shaped low-v structure is imaged down to the lower mantle (Figures 13 and 14). These results suggest that low-v anomalies in the lower crust revealed by previous studies [e.g., Huang and Zhao, 2004; Qi et al., 2006; Lei et al., 2008; Lei, 2011] could be related not only to the hot material upwelling from the upper mantle due to the dehydration of the stagnant Pacific slab in the mantle transition zone [Zhao, 2004; Lei and Zhao, 2005, 2006; Lei et al., 2008; Huang and Zhao, 2006; Li and van der Hilst, 2010] but also to the lower mantle plume. Huang et al. [1996] once investigated the relationship between Neogene volcanism and Holocene earthquakes in the Tanlu fault zone, and concluded that the upwelling magma may promote the fault to slip to generate an earthquake. 5.6. Interpretation [41] Some low-v anomalies in the upper mantle extend down to 200 400 km depth under the Zhang-Bo seismic zone, Tanlu fault zone, and Shanxi rift, while the low-v anomalies under Datong volcano and Bohai Sea are connected at 200 km 24 of 29

Figure 19. A schematic west-east cross section showing the dynamic process of the NCC destruction, origin of Datong volcano, formation of the rift, and seismotectonics in the region. See the text for details. The triangle represents Datong volcano (DT). White circles and crosses denote large earthquakes (M 7.0) and small earthquakes (M < 7.0) in the crust within 35 km off the profile, respectively. Dashed lines denote the Moho (extracted from Bassin et al. [2000]), 410-km and 660-km (directly adopted from the IASP91 model [Kennett and Engdahl, 1991]) discontinuities, respectively. The curve at the top shows the topography along the profile as shown in Figure 14b. There is no vertical exaggeration except for the surface topography on the top. White thicker arrows denote the directions of the hot material upwelling, while white thinner arrows indicate the possible movement directions of detached lithosphere in the upper mantle and the stagnant Pacific slab in the mantle transition zone. Curvy arrows in the mantle transition zone correspond to the dehydration of the stagnant Pacific slab in the mantle transition zone. CNCC and ENCC, central and eastern NCC; BHS, Bohai Sea; SXR, Shanxi rift (two double dipping lines with a V-like shape in the crust); TLF, Tanlu fault zone (two double vertical lines in the crust). depth to a broad low-v anomaly ascending from the lower mantle (Figures 13 and 14). Most large crustal earthquakes are located right above these low-v anomalies in the mantle. A prominent north-south trending high-v anomaly ( Re ) representing the remainder of detached lithosphere is visible under the west of eastern NCC. Some high-v anomalies ( De ) representing the detached lithosphere are visible at 200 400 km depth under eastern and central NCC, while those ( In ) under the Ordos block are clearly observed above 400 km depth, reflecting intact lithosphere (Figures 13 and 14). These structural features confirm what was already observed by previous tomographic studies [e.g., Huang and Zhao, 2009; Tian et al., 2009; Zhao et al., 2009]. Receiver function analyses showed that the lithospheric-asthenosphere boundary (LAB) of the NCC gradually becomes deeper from the east to the west [Chen, 2010], and that the Moho and LAB are sharply uplifted in a very small area right under the Tanlu fault zone [Chen et al., 2006]. [42] The stagnant Pacific slab in the mantle transition zone under eastern China has been detected as high-v anomalies by several global tomographic models [e.g., Fukao et al., 1992; Bijwaard et al., 1998; Zhao, 2004; Lei and Zhao, 2006; Li et al., 2008]. The regional tomography of Huang and Zhao [2006], Li et al. [2006], and Li and van der Hilst [2010] demonstrates that the western edge of the Pacific slab arrives at about 119 120 E. The present tomographic model shows broad high-v anomalies representing the stagnant Pacific slab in the mantle transition zone under 25 of 29