A three-dimensional P wave velocity model of the crust and upper mantle down to 400-km depth beneath eastern Tibet is obtained using many temporary seismic stations of the ASCENT project and the Namche Barwa Broadband Seismic Network. We collected 16,508 arrival times of P, Pn and Pg phases from 573 local and regional earthquakes and 7,450 P wave arrivals from seismograms of 435 teleseismic events. Our high-quality data set enables us to reconstruct the 3-D velocity structure under eastern Tibet in more detail than the previous studies. In the shallow depth, our results show that the low velocity zones are not interconnected well (no wide-spread low velocity zones), which may reflect the complex pattern of material flowing in the study region. Below the Moho, we find that the Indian lithospheric mantle underthrusts sub-horizontally under eastern Tibet, and the extent of the northward advancing Indian lithosphere decreases from west to east. In the north, the Asian lithospheric mantle is detected under the vicinity of the Qaidam Basin. Between the Indian and Asian lithospheric mantles, there is an obvious low-velocity anomaly which may reflect an upwelling mantle diapir.
 Despite many geophysical and geological studies have been made on the Tibetan plateau, controversy still remains on some basic topics, such as the nature and extent of the northward dipping Indian lithosphere under the Tibetan plateau and whether the Asian plate has underthrusted southward directly beneath the plateau. There is no doubt that the subduction of the Indian plate (and/or the Asian plate) plays an important role in the collision dynamics. A lot of geophysical results were derived from the INDEPTH and Hi-CLIMB project in the Tibetan plateau [e.g.,Kosarev et al., 1999; Chen et al., 2010; Kind et al., 2002; Rapine et al., 2003; Unsworth et al., 2004; Zhao et al., 2010]. Some results of seismic tomography show high-velocity (high-V) anomalies under the Tibetan plateau south of the Bangong-Nujiang suture (BNS), which was interpreted as the subducting Indian lithosphere [Tilmann et al., 2003]. This interpretation has been supported by some receiver-function studies [Nabelek et al., 2009]. A recent study of body wave tomography [Li et al., 2008] showed that the extending distance of the north-dipping Indian plate decreases from west to east. In contrast, some researchers suggested that the north-dipping plate underthrusts beneath the whole plateau [Priestley et al., 2006; Zhou and Murphy, 2005] or further north in eastern Tibet [Liang and Song, 2006]. In addition to the Indian plate that was detected under the Tibet, the Asian plate is also found under central and eastern Tibet [Chen et al., 2010; Kind et al., 2002; Kumar et al., 2006; Nabelek et al., 2009; Zhao et al., 2010]. In contrast, both the surface wave dispersion analysis [Rapine et al., 2003] and magnetotelluric result [Unsworth et al., 2004] showed that the central-northern Tibet is underlain by a hot upper mantle.
 Geological studies of southeast Tibet suggested that the uplift of the eastern Tibetan plateau has occurred with only slight internal deformation, and the feature is explained as an influx of lower crustal flow [Royden et al., 1997; Shen et al., 2001]. Several geophysical studies have provided evidence for the feasibility of the crustal flow. Strong radial anisotropy in the middle-to-lower crust of Tibet suggests the existence of crustal material flow beneath the Tibetan Plateau [Shapiro et al., 2004]. In a recent study, low Rayleigh-wave phase velocity features across most of the Tibetan middle crust at depths between 20 and 40 km further illustrated the existence of crust flow, but the particular mechanism (partial melting or alignment of anisotropic minerals) is still unclear [Yang et al., 2012]. In southeast Tibet, body wave tomography [Huang and Zhao, 2006; Huang et al., 2009; Li and van der Hilst, 2010; Li et al., 2008; Pei et al., 2007], Rayleigh-wave tomography [Fu et al., 2010] and receiver-function studies [Wang et al., 2010; Xu et al., 2007] revealed widespread low-velocity (low-V) anomalies at the middle or lower crust. Although the crustal flow model has been widely accepted, the detailed pattern of flow is still on debate. It is possible that such a large-scale material flow beneath eastern Tibet or even the whole plateau would be interrupted by the region which is not adaptive for flowing (e.g., the eclogited Indian crust with higher density and low temperature). It is difficult to explain the existence of strong lateral heterogeneity of the low shear speed [Wang et al., 2010; Yao et al., 2008] if the flow is not interrupted. GPS and seismic anisotropy results suggest that the extrusion may also occurs in the deep Tibetan lithosphere [Wang et al., 2008]. Recently, two major zones or channels of high electrical-conductivity at a depth of 20–40 km are revealed by a magnetotelluric study [Bai et al., 2010].
 Most of the above mentioned studies were focused on central or southeast Tibet due to the lack of data in northeast Tibet. The new seismic data from the ASCENT project have provided important new information on eastern Tibet. In this work we have used the new ASCENT data to determine a 3-D P wave velocity model of the crust and upper mantle under eastern Tibet. Our results show that both the Indian and Asian lithospheric mantles exist under the study region, and a strong low-V anomaly is visible between them, which may reflect an upwelling mantle diapir. These results shed new light on the deep structure and dynamics beneath eastern Tibet.
2. Data and Method
 We used two sets of data: one is from the Namche Barwa Broadband Seismic Network that had been operated in 2003 and 2004 [Sol et al., 2007]; the other is from the ASCENT project that was conducted from 2007 to 2009 (Figure 1). The data of Namche Barwa Broadband Seismic Network has been used in many previous seismic works and most of them used the teleseismic events (e.g., the seismic tomography [Fu et al., 2010; Li et al., 2008; Ren and Shen, 2008] and receiver function [Singh and Kumar, 2009]). In contrast, only a few results with the data of ASCENT have been published [e.g., Yang et al., 2012].We used the hypocentral parameters of local, regional and teleseismic events determined by the United States Geological Survey (USGS), but we relocated the local and regional events that were poorly located and fixed at 33 km depth by USGS (Figures 1 and S1). These events have magnitudes (Mb) >3.0 and epicentral distances <90 degrees. In the tomographic inversion we used the teleseismic events that have over 6 P wave arrivals, and used local and regional earthquakes that have over 8 P wave arrivals. Clear arrivals of P, Pg and Pn waves were picked up from the original seismograms (Figures S2 and S3). The Pg wave usually appears as the first arrival at the epicentral distance of <1.5°, but in Tibet it becomes the first arrival even at 3° due to the thick crust. Nearly half of the arrival times were collected manually, in particular, the Pg and Pn phase data. The time difference between the machine-picked arrivals and our collected arrivals is ∼0.2–0.3 s. Our final data set contains 16,508 P wave arrivals from 573 local and regional events and 7,450 P wave arrivals from 435 teleseismic events. The picking accuracy of our data is estimated to be 0.1–0.2 s.
 To determine the 3-D P wave velocity model beneath eastern Tibet, we applied the tomographic method ofZhao et al. [1992, 1994]to invert the arrival-time data from local and regional earthquakes and relative travel-time residuals from the teleseismic events. For details of this method, seeZhao et al. [1992, 1994]. For each teleseismic event, we first construct relative travel-time residuals that are obtained by subtracting the mean travel-time residual from the raw residuals. The theoretical travel times are calculated using the 1-D iasp91 Earth model [Kennett and Engdahl, 1991] for the mantle and the CRUST2.0 model [Bassin et al., 2000] for the crust (Figure S4). The data with absolute travel-time residuals <3.0 s and relative residuals <2.0 s were actually used in the tomographic inversion.
 The study area is parameterized with an optimal grid interval of 0.5° to 1° laterally and 30 to 50 km in depth after we conducted several checkerboard resolution tests. That is, in the central part of the study area, the grid interval is 0.5° in the horizontal direction and 10 to 50 km in depth, while in the edge portions the grid interval is 2° in the horizontal direction and 50 km in depth. The extent of our study area is 14° in longitude, 12° in latitude and 400 km in depth (Figure S5).
 P wave velocity perturbations at the 3-D grid nodes from the 1-D starting velocity model are taken as unknown parameters. The velocity perturbation at any point in the model is calculated by linearly interpolating the velocity perturbations at the eight grid nodes surrounding that point. The LSQR algorithm [Paige and Saunders, 1982] with damping and smoothing regularizations [Zhao, 2001] is used to solve the observation equation that relates the unknown parameters to the observed data. We conducted a number of inversions with different values of damping and smoothing parameters, and found the optimal values of damping and smoothing parameters to be 15.0 and 0.03, respectively.
 To confirm the reliability of our result and the resolution scale of the 3-D velocity model obtained, we conducted the checkerboard resolution [Humphreys and Clayton, 1988; Zhao et al., 1992]. In this test, we assign alternatively negative and positive velocity perturbations to the grid nodes, and calculate synthetic travel times for the checkerboard model. Random noises in the range of −0.5 to +0.5 s are added to the synthetic travel times to simulate the picking errors contained in the observed data. Then we conduct inversion of the synthetic data using the same algorithm as for the observed data. The resolution is considered to be good for the area where the checkerboard pattern is well recovered. The test results show that the central part of the study area is well resolved, especially the upper 200-km depths because many regional and local rays pass through there (Figures 2 and 3).
Figures 4, 5, and 6 show the tomographic images obtained. At depth of 40 to 70 km (Figures 4a and 4b), some low-V zones appear in the study region. However, the low-V zones are interrupted by some fragmented high-V zones. The distribution of the low-V and high-V anomalies exhibits some differences from those revealed by Pn-wave tomography [Liang and Song, 2006; Pei et al., 2011a]. At depth of 100 km (Figure 4c), an obvious high-V anomaly exists under the western part of the study region. Such a wide extent of high-V zone is consistent with the interpretation about the underthrusting of the Indian and/or Asian lithospheric mantle [Kind et al., 2002; Nabelek et al., 2009; Zhao et al., 2011]. In the depth range of 150 to 200 km (Figures 4d and 4e), a broad high-V zone is visible in southeastern Tibet, and its extent becomes smaller with depth. In addition, a prominent low-V zone exists under the Cona rift. At greater depths (Figures 4f–4i), two distinct high-V anomalies are visible around 92°E and 100°E. The western one is related to the subducting Indian slab [Ren and Shen, 2008; Zhang et al., 2011], while the eastern one is associated with the subducting Burmese plate [Huang and Zhao, 2006; Li and van der Hilst, 2010; Li et al., 2008].
 The most conspicuous features in the N-S cross-sections (Figures 5b and 6b) are high-V anomalies beneath the Tibetan plateau and the Qaidam Basin, and the high- zones are separated by a strong low-V cylinder. Considering the results of the resolution tests (Figures 3and S6d), we can say that the low-V cylinder is a reliable feature, not due to vertical smearing. In the E-S cross-section (Figure 5a), the main feature is consistent with previous results [Fu et al., 2010; Ren and Shen, 2008], but there are some differences in details. A strong, dipping low-V zone is detected from 100 to 400 km depth, which is sandwiched by two distinct high-V anomalies (Figure 5a).
 In central Tibet, many studies have been made to trace the southward dipping Asian lithosphere [Kind et al., 2002; Nabelek et al., 2009; Tilmann et al., 2003]. However, in northeast Tibet, there have been few studies about the Eurasian plate because few seismic stations exist there. Using Pn-wave tomography,Liang and Song detected two prominent high-V anomalies under the Qaidam Basin and southeastern Tibet, respectively. The high-V zone under southeastern Tibet is also visible in our image with the same range, but the extent of the high-V zone under the Qaidam Basin is not similar to that determined by the Pn-wave tomography [Liang and Song, 2006]. We consider that the two high-V zones in our image reflect the lithospheric mantle, because a high-V anomaly in the upper mantle usually indicates relatively strong and cold materials. A high-V anomaly is detected which extends to about 34.5°N in the north-south vertical cross-section along 91°E (Figure 6a). In the vicinity of this region, a previous tomography also detected this high-V anomaly by using the teleseismic data recorded by INDEPTH project [He et al., 2010]. In contrast, two high-V zones are revealed in the north-south vertical cross-sections along 96°E and 97°E (Figures 5b and 6b), and the two high-V zones are separated by a low-V anomaly extending from the surface down to 400-km depth. This feature is very similar to the previous studies beneath central Tibet [Kumar et al., 2006; Wittlinger et al., 1996], but slightly different from a recent receiver-function result [Zhao et al., 2011]. An interpretation of this low-V anomaly is the presence of fluids and partial melts [Sato et al., 1989] because the underthrusting lithosphere can bring fluids to the upper mantle. In our images (Figures 5b and 6b), the depth of the two lithospheres is not greater than 200 km, but the low-V anomaly extends down to ∼400-km depth. Another possibility for the low-V zone is that it reflects upwelling of hot mantle materials. A viewpoint of subduction-induced upwelling has been proposed byTilmann et al. . If this is true, then the high-V anomaly under the study region may be interpreted as the Indian lithosphere, and a mantle upwelling passes through the Indian lithosphere. This discontinuous subduction model has been proposed byLiang et al. for central-southern Tibet. Although this interpretation is reasonable, there is no obvious high-V zone under 200-km depth, which indicates that there is no steep subduction or delaminated slab under eastern Tibet. Therefore, we consider that the low-V zone between the two lithospheres may reflect a mantle diapir, and that the southern and northern high-V zones represent the Indian lithospheric mantle and the Asian lithospheric mantle, respectively. In addition, the extent of the northward advancing Indian slab decreases from west to east (Figures 5b, 6a, and 6b), being consistent with a recent tomographic result [Li et al., 2008].
 In Figure 5a, a low-V anomaly, D, is located beneath the Cona rift, and two high-V zones, which are interpreted as the subducted and delaminated slabs, are imaged. This result is consistent with the finite-frequency tomography [Ren and Shen, 2008] and a recent travel-time tomography [Zhang et al., 2011]. Ni et al. suggested that the Burmese microplate descends eastward beneath the Eurasian plate while the Indian plate underthrusts northward. To confirm the reliability of the high-V features, we conducted a synthetic test (Figures S6a and S6b). Although smearing occurs in the area between the two high-V zones (Figure S6b), the main features are recovered. Taking into account the previous tomographic results [Huang and Zhao, 2006; Li and van der Hilst, 2010], we prefer to interpret the high-V zone at 100°E as the Burmese microplate.
 In eastern Tibet, a high Poisson's ratio anomaly was revealed in the crust [Xu et al., 2007; Pei et al., 2011b], which suggests that the crust is deformable, and a lower S-wave velocity indicates that the temperature of the crust is high [Zhao et al., 2010]. Although the pervasive low-V anomalies at shallower depths (Figures 4a and 4b) may suggest the existence of crustal flow in the study region, it is not clear whether the flow is continuous or not, even if it exists. For investigating the pattern of crustal flow, Bai et al. used broadband magnetotelluric data from the eastern Himalaya syntaxis, and they revealed two distinct zones of reduced resistivity at the shear zones in the middle and lower crust. Their result is consistent with our image (the low-V anomalies “A” and “B” inFigures 5a and 5b). Many large faults exist in Tibet, which may influence the pattern of flow and, hence, the style of regional deformation [Yao et al., 2010]. In another word, shearing may occur between the low-V and high-V zones in the crust. To the contrary, a recent result of seismic ambient noise tomography in the Tibet found that low-V zones do not conform to surface faults and are most prominent near the periphery of Tibet [Yang et al., 2012]. Besides the low-V zones “A” and “B,” another two low-V anomalies are revealed (“C,” “D” inFigure 5a). An experimental study shows that shearing of the middle and lower crust may form interconnected fluid-filled cracks [Tullis et al., 1996]. Considering the high attenuation of seismic waves [Brown et al., 1996; Zhao et al., 1993] and low Q in northeastern Tibet [Bao et al., 2011], we suggest that the pervasive low-V zones in the crust may be induced by the mantle diapirs (as discussed above), while further studies are needed to clarify the detailed processes.
 A high-resolution 3-D P wave velocity model of the crust and upper mantle beneath eastern Tibet is determined using a large number of high-quality data recorded by many temporary stations of the ASCENT project and the Namche Barwa Broadband Seismic Network. We carefully collected many P wave arrival times from the original seismograms, nearly half of which were manually picked. Our results show that widespread low-V anomalies exist in the crust but they are interrupted by high-V anomalies, suggesting that complex crustal flow may exist in the crust under eastern Tibet. In a vertical cross section along 30°N, a mantle upwelling is revealed beneath the Cona rift in southeastern Tibet. The delaminated Burmese microplate and the Indian slab are also revealed. At 70∼200 km depths, two gently dipping high-V zones are clearly visible. The southern high-V zone may reflect the northward dipping Indian lithospheric mantle, and its extent decreases from west to east in the study region. The high-V zone in the north of the Jinsha River suture may represent the Asian lithospheric mantle. Between the Indian and Asian lithospheres, a prominent low-V anomaly exists and it extends down to over 200-km depth, which may be a mantle diapir. Widespread low-V anomalies are revealed in the crust of northeastern Tibet, some of which may be induced by the mantle diapir.
 This research is supported by the National Science Foundation of China (grants 40930317 and 41104055), the SinoProbe-02 project and the NSFC Innovation Research Group Fund (grant 41021001). The data used in this study were recorded by the Namche Barwa Broadband Seismic Network and the ASCENT Project, and the original data were downloaded from the IRIS Data Management Center. We thank the anonymous reviewers who provided constructive suggestions that have improved the manuscript. The Seismic Handler software package was used for the arrival-time picking. All the figures were made using the Generic Mapping Tools packages ofWessel and Smith .