1 Introduction

The uplift and expansion of the Tibetan plateau (TP) are the result of the Indian-Eurasian collision since ~50 million years ago (e.g., Yin and Harrison 2000; Royden et al. 1997, 2008). Despite many studies in the past (e.g., Molnar and Tapponnier 1975; Tapponnier et al. 1982; Royden et al. 1997; Liu et al. 2014), the uplift and deformation of the TP are still controversial. The southeastern (SE) margin of the TP is one of the key areas for understanding the tectonic evolution of the plateau. Adjacent to the Sichuan basin, the Longmenshan (LMS) rises about 5000 m above the Sichuan basin within a range of 50 km and is the steepest margin, whereas the SE margin toward Yunnan is gentle. Across the LMS, the crustal thickness also greatly varies from ~60 km in the west to ~40 km in the east (Zhang et al. 2009, 2010). However, the eastern margin lacks large-scale young crustal shortening structures (Shen et al. 2005). Two models have been proposed to explain how the LMS is formed: (1) crustal shortening and thickening model, in which extrusion along the lateral strike-slip faults caused by the subduction of Asian lithospheric mantle results in crustal shortening and uplift of the LMS (Tapponnier et al. 2001); (2) crustal channel flow model, in which eastward ductile mid-crustal flow is obstructed by the rigid Sichuan basin (Royden et al. 1997). An important difference between these two models is whether crust and mantle of the eastern plateau are decoupled. The high-resolution velocity models from tomography can provide more constraints on these different models.

According to Shen et al. (2005), the SE margin of the TP can be divided into several blocks. The lateral variation in the crustal strength of the blocks and interaction between blocks make this area one of the most active seismic regions in China. The 2008 Wenchuan earthquake and the 2013 Lushan earthquake occurred on the Longmenshan fault (LMSF) system. Notably, there is an obvious aftershock gap of about 35 km in length between the Wenchuan and Lushan earthquakes (Xu et al. 2013). Previous tomography results have shown a clear low-v anomaly in the gap in the crust (e.g., Lei and Zhao 2009; Lei et al. 2014; Pei et al. 2014; Wang et al. 2015).

Previously, extensive geophysical imaging studies have been conducted in the SE margin of the TP, including magnetotelluric imaging (Bai et al. 2010; Zhao et al. 2012), seismic body-wave tomography (Pei et al. 2010, 2014; Wang et al. 2009, 2015; Zhang et al. 2012), surface-wave tomography (Li et al. 2009; Huang 2014; Yao et al. 2008; Yao 2012), receiver function analysis (Xu et al. 2007) and joint inversion of receiver function and surface waves (Liu et al. 2014; Sun et al. 2014). However, in most cases, the models from different studies are not consistent with each other, especially in the LMSF zone (Zhang et al. 2012; Wang et al. 2009, 2015; Liu et al. 2014).

In this study, we aim at conducting the first joint inversion of body- and surface-wave data to resolve a more reliable velocity model in the SE margin of the TP. Because two data types are complementary to each other, the joint inversion is able to reduce the non-uniqueness of the velocity model inverted using only a single data type. The body-wave arrival times can better resolve the model in the zones with dense ray coverage and can have good depth resolution. In comparison, the short-period surface-wave dispersion data can provide good resolution in the shallow part of model, where body-wave arrival times generally have poor resolution because of lacking crossing rays (Zhang et al. 2014). In addition, surface waves extracted from ambient noise analysis are not dependent on the distribution of earthquakes. By using the jointly inverted velocity model, we hope to elucidate the structural features of the aftershock gap and further to explain why the gap exits and understand the earthquake generation mechanisms in this area.

2 Data and method

For the joint inversion, we used two data sets: the first arrivals of P and S waves and dispersion curves for Rayleigh waves. For the body waves, we obtained 65,525 P- and 36,866 S-wave first arrival times from 7190 events recorded by 102 stations for the period of 2001–2004, and from May to August 2008 and April to May 2013 in the SE margin of the TP (Fig. 1). These selected arrival times are carefully checked to follow the major trend of travel-time curves (Fig. 2). From these absolute arrival times, we also constructed 362,105 P- and 231,609 S-wave differential times. Rayleigh wave dispersion curves at periods from 4 to 40 s are extracted from ambient seismic noise data recorded by the densely deployed array, containing 298 broadband seismometers with average inter-station distance of about 15 km in the SE margin of the TP (Huang 2014; Liu et al. 2014). The phase velocity maps for Rayleigh waves  from Huang (2014) are shown in Fig. 3. These maps show the basic velocity variation patterns of the study area. At short periods, the Sichuan basin is a low-phase velocity region indicating thick sediments, whereas the SE margin of the TP shows high-velocity anomalies, which are related to Songpan–Garze flysch (Burchfiel et al. 2008). In comparison, at longer periods, the Sichuan basin is associated with high-velocity anomalies, suggesting cold and rigid Yangtze Craton and the SE margin of the TP shows obviously low-velocity anomalies. It is clear that the boundary between the high- and low-velocity zones follows the LMSF in both short and long periods in the SE margin of the TP.

Fig. 1
figure 1

Map of the SE margin of the TP. Stations (black triangles) and earthquakes (gray dots) used in this study and faults (black lines) overlaid on a topographic map of the SE margin of the TP. Gray lines show block boundaries from GPS observations (Shen et al. 2005). The black box shows the location of the study area. Inset: the black arrows represent the approximate surface motion relative to the south China block (Shen et al. 2005). The black box shows the location of the study area. Acronyms: LMSF Longmenshan fault, XSHF Xianshuihe fault, LTF Litang fault, MLF Muli fault, LJF Lijiang fault, ANHF Anninghe fault, ZMHF Zemuhe fault, XJF Xiaojiang fault

Full size image
Fig. 2
figure 2

P and S travel-time curves

Full size image
Fig. 3
figure 3

Phase velocity maps for the Rayleigh waves at 4, 8, 12, 16, 20, 26, 32 and 38 s

Full size image

We jointly invert surface-wave dispersion data and body-wave arrival times using the new joint inversion method of Zhang et al. (2014), which combines the regional-scale version of double-difference (DD) tomography algorithm (tomoFDD) and surface-wave inversion. DD tomography can use both absolute times and more accurate differential times to relocate earthquakes and invert velocity models simultaneously (Zhang and Thurber 2003). TomoFDD uses a finite-difference travel-time algorithm and can calculate travel times and ray paths in the spherical earth that is converted into a Cartesian grid system (Zhang and Thurber 2006). LSQR (Paige and Saunders 1982) is used to solve the system of equations. Damping and first-order spatial smoothing are applied to regularize the inverse problem.

Compared with only body-wave inversion which does not have a good resolution in the shallow part of the crust, the v S model constrained by body-wave arrival times and surface-wave dispersion data can be more accurate because short-period surface waves are sensitive to shallower crustal structures (Yao et al. 2008). In the traditional surface-wave inversion, v P model can be obtained from v S model by postulating a constant v P/v S or according to empirical formulas to calculate the dispersion curves (Li et al. 2009). Instead, in the joint inversion, the v P model inverted by body-wave times can be directly used to calculate the dispersion data, and therefore, the v S model inverted by joint inversion is more reliable.

3 Inversion details and model resolution analysis

A three-dimensional (3D) velocity model was inverted based on a minimum one-dimensional (1D) velocity model from the regional 1D velocity model of Zhao et al. (1997). The inversion grid intervals are 0.5° in both latitude and longitude. And in the depth direction, the inversion grid nodes were placed at 0, 5, 10, 17.5, 25, 35, 45, 65 and 90 km, respectively.

The damping and smoothing parameters used to regularize the inversion system as well as the relative weighting between body-wave and surface-wave data are carefully selected by the following processes. Firstly, we determined the optimal damping and smoothing parameters for the inversion system with only body-wave data through a trade-off analysis (Aster et al. 2013). The same damping and smoothing parameters are applied to the inversion parts related to v P and v S. The final absolute P- and S-wave travel-time root mean square (RMS) residuals from only body-wave inversion are 0.218 and 0.298 s, respectively. Based on the selected optimal damping and smoothing parameters, we then select the optimal data weighting parameter to balance surface- and body-wave contributions to the joint inversion objective function. We keep the body-wave weight fixed at 1 and change the surface-wave weight in the range 1–100 because the number of the surface-wave dispersion data is less than that of the body waves. Based on the trade-off analysis, the optimal body- and surface-wave data weights are selected as 1 and 10, respectively. Using these optimal parameters, we obtained the final v P and v S models that reduced the absolute body-wave travel-time RMS residual from 1.286 to 0.239 s, an 81% reduction, and surface-wave dispersion data error from 0.130 to 0.018 km/s, an 86% reduction. Specifically, in the joint inversion, the absolute P-wave travel-time RMS residual is reduced from 1.324 to 0.213 s and from 2.377 to 0.295 s for S-wave arrival times, which are around the same level as the separate body-wave-only inversion. The histograms of P- and S-wave travel-time residuals corresponding to the initial models, and the velocity models obtained by body waves only and by joint inversion are shown in Fig. 4. We can see that travel-time residuals from body-wave-only inversion and joint inversion are more concentrated near 0 s than those corresponding to the initial models.

Fig. 4
figure 4

Histograms of (top) P-wave and (bottom) S-wave arrival time residuals for different models. a and d: Initial model; b and e: Body-wave-only inversion; c and f: Joint inversion. Residual misfits for velocity models obtained from body-wave inversion only and from joint inversion are more concentrated near 0 s than those related to the initial model

Full size image

For Rayleigh wave dispersion data fitting, it can be seen that the RMS residuals for the dispersion data at most of inversion nodes are small (~0.003–0.02 km/s) except for some inversion nodes (~0.05 km/s) (Fig. 5). At some selected inversion nodes, the Rayleigh wave dispersion curves calculated using the velocity model inverted from joint inversion can fit those obtained dispersion curves from ambient noise tomography very well (Fig. 6). In comparison, the predicted dispersion curves according to the velocity model inverted by body-wave data only cannot fit well with those obtained dispersion curves (Fig. 6).

Fig. 5
figure 5

RMS residuals for Rayleigh wave phase velocity dispersion data at each inversion grid node. The residual scales are shown on the right

Full size image
Fig. 6
figure 6

Comparison of predicted surface-wave dispersion curves from the velocity models inverted using body-wave arrival times only (square) and from joint inversion (circle) and from real surface-wave dispersion curves (cross) at selected grid nodes. The node locations are shown in Fig. 10

Full size image

We use the checkerboard resolution test (Aster et al. 2013) as well as the semblance test to evaluate the resolution of the tomography models. For the checkerboard test, we add velocity perturbations of ±5% to the initial model alternately to adjacent grid nodes in three directions and calculate synthetic travel times with the same event-station distribution as in the real data using the perturbed model. Figure 7 shows the horizontal slices of recovered checkerboard patterns at different depths. The semblance test is an alternative way of assessing the model quality. In the semblance test, we first add random variations on the initial 1D velocity model in three dimensions up to ±7.5%. Then we calculate the P-wave, S-wave travel times and surface-wave dispersion curves with the same data coverage as the real data. The synthetic data are then inverted using the same inversion parameters as the real data. This process is repeated for 100 times and we calculate the semblance between the input model and the recovered model based on Zelt (1998). The semblance values for 100 test models are averaged to obtain the semblance distribution for the model. A comparison of the semblance distribution of v P and v S models for the body-wave inversion only and v S for joint inversion of body and surface waves is shown in Fig. 8. The regions with semblance values above 0.8 are regarded as well-resolved regions.

Fig. 7
figure 7figure 7figure 7

Recovered checkerboard patterns at each depth for different models. The models are from left to right are v P inverted using body-wave arrival times only, v S inverted using body-wave arrival times only, v S inverted using surface waves only and v S jointly inverted by body and surface waves

Full size image
Fig. 8
figure 8

Horizontal slices of the semblance values for (left) v P and (middle) v S inverted using body waves only and v S (right) jointly inverted by body and surface waves. The contour line indicates the semblance value is 0.8

Full size image

The checkerboard test and semblance test results are basically consistent. The two tests indicate that v P obtained using body-wave arrival times only is well resolved for the model region except for the margin of the study area at each depth (Figs. 7 and 8). However, the checkerboard test shows that the v S obtained using body-wave data is only well resolved in the central part of the model region down to 17.5 km but is poorly resolved for deeper model zones (Fig. 7) because the ray coverage for S waves is poor in the deeper depths. The v S obtained using surface waves only is well resolved in the shallow depths, especially at 5 km but have poor resolution at deeper depths (Fig. 7). As expected, the resolution for the v S model from joint inversion is improved, especially at shallow depths (Figs. 7 and 8). When incorporating the surface-wave data, the outer part and deeper part of the model are better resolved down to depths of 35 km (Figs. 7 and 8). However, at greater depths of 45 and 65 km, the resolution of the v S model from joint inversion is poorer than that of v S from only body-wave inversion (Figs. 7 and 8). This may be because in these deeper depths, the ray coverage of S waves is poor and the surface waves are dominated. Because longer period surface waves have broader depth sensitivity kernels at the deeper region (Yao et al. 2008), the joint inversion cannot resolve such size of grids, resulting in poorer resolutions in the deeper depths.

4 Results

The horizontal slices of the v P and v S models inverted from body-wave data only, as well as the v S model jointly inverted by body- and surface-wave data at 5, 10, 17.5, 25 and 45 km depth are shown in Fig. 9, respectively. The vertical profiles through the velocity models along the lines AA′, BB′, CC′, DD′ and EE′ are shown in Fig. 10, respectively. The lines AA′, BB′ and CC′ are perpendicular to the LMSF and are through the Wenchuan earthquake, the aftershock gap and the Lushan earthquake from north to south, respectively. The line DD′ crosses the gap from the southwest to the northeast while the line EE′ goes along the gentle-topography margin of the SE Tibet.

Fig. 9
figure 9figure 9

Horizontal slices of v P (left) inverted by body waves only, v S (middle) by body waves only and v S (right) inverted jointly by body and surface waves at depths of 5, 10, 17.5, 25 and 45 km. The black circles at depth of 10 km depict the big earthquakes occurring from 2008 to 2015 in the study area. Three red dashed ellipses show three high-velocity patches (HV1, HV2, HV3) along the LMSF and two red solid ellipses show the two low-velocity zones (LV1, LV2) in the LMSF zone

Full size image
Fig. 10
figure 10

Vertical cross sections through the (top) v P and (middle) v S models inverted from body waves only and the (bottom) v S model inverted jointly by body and surface waves along the lines (A) AA′, (B) BB′, (C) CC′, (D) DD′ and (E) EE′, respectively. The surface topography is shown on the top of each profile. Black dots represent the relocated earthquakes within 5 km from each profile. Red stars in profile AA′ and CC′ represent the Wenchuan and Lushan earthquakes, respectively. The profile locations are shown on the inset map. The v P and v S color scales are shown on the bottom. N 1, N 2, N 3, N 4, N 5 and N 6 are node locations shown in Fig. 6. F1: Wenchuan–Maowen fault; F2: Yingxiu–Beichuan fault; F3: Guanxian–Jiangyou fault

Full size image

At shallow depths (5 km), the patterns of velocity anomalies coincide with the local geology (Fig. 9a). For the Sichuan basin, it is clearly associated with low-velocity anomalies, which are caused by primarily Mesozoic and Paleozoic sediments of >10 km thick in the upper crust (Burchfiel et al. 2008). Among v P and v S models inverted separately or jointly, the Quaternary sediments are better delineated by the jointly inverted v S model (Fig. 9a). Along the LMSF, there exist obvious and variable high-velocity anomalies, which correspond well with the Middle to Upper Triassic flysch with 10 km thick in the LMSF block (Burchfiel et al. 2008). The Yajiang block shows variable and high velocities and the Central Yunnan block shows variable but normal velocities (Fig. 9a).

In the upper crust (10 km), the velocity map shows obvious crustal heterogeneity (Fig. 9b). It is noted that the Sichuan basin is not shown as an overall low-velocity anomaly zone (Fig. 9b). Instead, a high-velocity anomaly extending parallel to the strike of the LMSF separates two low-velocity anomaly zones from northwest to southeast in the Sichuan basin (Fig. 9b). The Sichuan basin can be divided into three units: the Northwestern Depression, the central uplift and Southeastern Depression (Ma et al. 2007). The Northwestern Depression bounded by the LMSF to the northwest and the Longquanshan fault (LQSF) to the southeast is a foreland basin containing late Proterozoic (Sinian) to Quaternary sediments whose thickness decreases from up to 12 km near the LMSF to about 8 km in the LQSF (Wang et al. 2016). The central uplift bounded by the LQSF to the west and the Huayingshan fault (outside our study area) to the east consists of sediments about 5 km thick (Ma et al. 2007; Wang et al. 2016). In the southern end of Southeastern Depression, the low-velocity anomalies correspond to the sediments of about 10 km thick (Ma et al. 2007; Wang et al. 2016). The velocity anomalies vary in the LMSF, Yajiang and Central Yunnan blocks (Fig. 9b). The high-velocity anomaly in the north of Central Yunnan block at the depth of 10 km (Fig. 9b) is related to the inner core of the Emeishan flood basalt which is caused by materials intruding upwards from the mantle (He et al. 2003). This high-velocity anomaly in the crust is also imaged in previous tomographic studies (Huang 2014; Liu et al. 2014; Li et al. 2009). In the central part of the study area, our model shows a high-velocity stripe extending southeastwards from the intersection of the LMSF and Xianshuihe fault (XSHF) to the location around longitude 103.5° and latitude 29°. It is also noted that on the depth slices of v P and v S models from separate body-wave inversion and joint inversion at 10 km depth, there exist three high-velocity anomaly zones from north to south along the LMSF (HV1, HV2 and HV3 in Fig. 9b). Between these high-velocity zones, there exist low-velocity anomalies along the LMSF (LV1 and LV2 in Fig. 9b).

For the major faults in the study area, they are generally associated with velocity contrasts in the upper crust (Fig. 9a, b). For example, the LMSF is located on the boundary between low velocity in the Sichuan basin and high velocity in the LMSF block (Fig. 9a, b). Similarly, Xianshuihe, Muli, Anninghe and Zemuhe faults separate low-velocity anomaly on one side and high-velocity anomaly on the other side (Fig. 9a, b).

At the depth of 17.5 km, it is noted that there exist relatively large discrepancies between velocities inverted from separate body-wave inversion and joint inversion (Fig. 9c). In the lower crust (25–45 km), the v P and v S models inverted by body waves only are still complex, but the v S model inverted by joint inversion is simpler and mostly shows large-scale features (Fig. 9d, e). This is because longer period surface waves have larger resolution length (Huang 2014). The v S model from joint inversion is similar to previous results (Liu et al. 2014; Li et al. 2009) although the checkerboard test shows that the resolution is poorer than that corresponding to body-wave-only inversion. In the joint inversion v S model, the most notable feature is the ubiquitous high-velocity anomaly in the Sichuan basin and low-velocity anomaly in the Songpan–Garze block (Fig. 9d, e), which reflects the cold and rigid craton root to the east and active Cenozoic deformation to the west of the LMSF, respectively (Royden et al. 2008).

For the cross sections along profiles AA′, BB′, CC′, DD′ and EE′ of the joint inversion v S model, it can be seen that there clearly exist the mid-crustal low-velocity anomalies (Fig. 10). Similarly, the v P model from separate body-wave inversion also shows low-velocity layers in the middle crust but vague. In comparison, the v S model from body-wave-only inversion does not show this low-velocity anomaly at all, indicating the v S model has poorer resolution in this depth range (Fig. 10).

5 Discussion

5.1 The relation between large earthquakes and velocity anomalies

Figure 9b shows the distribution of large earthquakes (M ≥ 5) that occurred from 2008 to 2015 and velocity variations in the SE margin of the TP. We can see that most large earthquakes occurred around velocity contrasts between the low- and high-velocity zones (Fig. 9b). This phenomenon was also noticed in the previous studies (e.g., Zhao et al. 2002; Mooney et al. 2012). In Japan, it was found that most large earthquakes are generally located along the boundary between the low- and high-velocity zones in the crust and uppermost mantle (Zhao et al. 2002). It was interpreted that in the volcanic areas the low-velocity anomalies are due to high-temperature anomalies caused by magma upward intrusion, and in the forearc areas under the Japan Islands they are caused by fluids originated from sea water filtration through faults or from slab dehydration (Zhao et al. 2002). In continental regions, it was found that majority of the earthquakes occur around the edges of positive velocity anomalies and a few small magnitude events occur in the regions with positive velocity anomalies (or within continental interior) (Mooney et al. 2012). These studies showed that the process of earthquake generation may be closely related to the velocity distribution or rigidity of the material. The high-velocity zones correspond to stronger blocks and have higher lithospheric strength (rigidity) and a lower strain rate within the seismogenic upper crust (Mooney et al. 2012). As a result, the stress can be more easily accumulated on high-velocity zones, which can thus act as asperities for generation of large earthquakes (Pei et al. 2014). For low-velocity areas, mechanical strength of materials is weak and is not apt to accumulate strain and stress. Around the velocity contrast between high and low velocities, it is weak and thus the earthquake can be easily nucleated there. This study shows that by determining the detailed velocity structure for one region we may be able to delineate the zones where large earthquakes may occur.

5.2 The relation between the earthquake rupture and velocity distribution

At 10 km depth, along the LMSF, there exist three strong high-v P and v S bodies from northeast to southwest (HV1, HV2 and HV3 in Fig. 9b), which correspond to metamorphic rocks of Xueshan plateau, Pengguan Massif, and Baoxing & Kangding Massifs, respectively (Burchfiel et al. 2008; Xu et al. 2008). In particular, the shape and size of the “L-shaped” high-velocity body at the intersection of the LMSF and XSHF coincide well with the Kangding Massif (Burchfiel et al. 2008; Xu et al. 2008). There are two obviously low-velocity anomalies between the three high-velocity bodies along the LMSF zone in the upper crust (LV1 and LV2 in Fig. 9b). The LV2 is situated in the aftershock gap of the Wenchuan and Lushan earthquakes. The LV1 is located from Xiaoyudong to Beichuan where less coseismic slip and fewer aftershocks of the Wenchuan earthquake occurred than the neighboring high-velocity regions (Nishimura and Yagi 2008; Shen et al. 2009). These two low-velocity anomalies are also imaged in the previous tomographic studies (Pei et al. 2010, 2014; Wang et al. 2014, 2015), which mainly consist of Paleozoic marine sedimentary rocks (Burchfiel et al. 2008; Xu et al. 2008). The LV1 is also associated with high-v P/v S ratio anomalies (Pei et al. 2010). For the aftershock gap (LV2), two-dimensional Pg imaging revealed that the gap is shown as a low-velocity anomaly (Pei et al. 2014). The 3D velocity model inverted using body-wave arrivals also revealed that the gap is associated with low-velocity and high-conductivity anomalies (Wang et al. 2014, 2015). The two low-velocity anomalies (LV1 and LV2 in Fig. 9b) with high-conductivity or high-v P/v S ratio anomalies are weak in mechanical strength and may behave more ductile than the surrounding high-velocity bodies. They could be caused by aqueous or partial molten materials extruded from the lower crust and/or the upper mantle of the eastern Tibet into the crust along the permeable faults in the LMSF (Wang et al. 2015). In the two low-velocity regions, it is hard to accumulate enough stress for large earthquakes to occur. For the aftershock gap, both the 2008 Wenchuan and 2013 Lushan earthquakes did not rupture there. Therefore, this low-velocity zone, with more ductile behavior, may act as barriers for rupturing of both earthquakes. On the other hand, the high-velocity bodies around the Wenchuan and Lushan earthquakes act as asperities for the two earthquakes.

5.3 The distribution of mid-lower crustal low-velocity zones

The mid-lower crustal low-velocity zones (LVZs) are clearly visible in the SE margin of the TP (Figs. 9d, e,  10), which are also imaged in previous studies (Yao et al. 2008, 2010; Li et al. 2009; Yang et al. 2012; Zhang et al. 2012; Liu et al. 2014). These LVZs are also generally associated with high Poisson’s ratio (Xu et al. 2007; Wang et al. 2010), high heat flow (Hu et al. 2000) and low electrical resistivity (Bai et al. 2010; Zhao et al. 2012). These anomalies could be caused by partial melt and aqueous fluids in the crust, suggesting that the middle and lower crust in the SE margin of the TP is mechanically weak. The LVZs in the middle and lower crust may be interpreted as the weak mid-lower crustal channel flow (Liu et al. 2014), which might be localized due to the complex 3D geometry of the crustal low-velocity zones (e.g., Yao et al. 2010; Liu et al. 2014). The surface-wave tomography study also showed the strong positive radial anisotropy (v SH > v SV) associated with the crustal LVZs (Huang et al. 2010), indicating subhorizontal alignment of crustal minerals, which is consistent with the channel flow model. From the velocity images, it can be seen that the thickness of the mid-lower crustal low-velocity channels varies in the SE margin of the TP (Fig. 10). The LMSF, ANHF and ZMHF mark the main eastern boundary of the LVZs in the middle and lower crust (Fig. 9d, e).

6 Conclusions

New three-dimensional v P and v S models are obtained for the SE margin of the TP by joint inversion of the body and surface waves to explore the crustal deformation and earthquake generation mechanism in this region. Compared with separate body-wave-only inversion, joint inversion improves the resolution for the v S model. At 10 km depth, the three strong high-v P and v S bodies along the LMSF are consistent with Precambrian massifs. Two low-velocity anomalies between the three high-velocity bodies along the LMSF are situated in the gap of the Wenchuan and Lushan earthquakes and are associated with less coseismic slip from Xiaoyudong to Beichuan, respectively. The two low-velocity anomalies with high conductivity and high Poisson’s ratio imply that the anomaly zones are weak and ductile and thus are hard to accumulate stress enough to cause large slip and earthquakes. The velocity distribution of the Sichuan basin at the depth of 10 km coincides with the thickness of the sediments. The high-velocity anomaly in central Yunnan block at the depth of 10, 17.5 km is related to the inner core of the Emeishan flood basalt. The big earthquakes (M ≥ 5) occurred in regions between high and low-velocity bodies, which are the suitable locations to accumulate stress and generate large earthquakes. The low-velocity layers in the mid-lower crust may suggest that the mid-lower crust is mechanically weak, which may facilitate channel flow of crustal materials. However, the distribution of mid-lower crustal low-velocity channel is not homogeneous and is bounded by LMSF, ANHF and ZMHF to the east.