Lithospheric delamination and upwelling asthenosphere in the Longmenshan area: insight from teleseismic P-wave tomography

We apply teleseismic P-wave tomography to reconstruct the velocity structure of the Longmenshan area. Our results show possible large-scale delamination beneath the Songpan-Ganzi and Qiangtang terranes, which induced upwelling asthenosphere. Upwelling asthenosphere might have led to lower crust heating, facilitating eastward extrusion of the Songpan Ganzi terrane resulting in localized deformation and uplift along the Longmenshan orogenic belt. We suggest that the eastward extrusion of the Songpan-Ganzi terrane against the rigid lithospheric root of the Sichuan Basin results in stress accumulation and release, leading to large earthquakes in the Longmenshan area.

The Longmenshan thrust (or orogenic) belt is approximately 500 km long and 30-50 km wide 1-7 with a relief of over 5 km 8 . It is the central segment of the north-south-trending major seismic belt in China and lies along a crustal-scale boundary between the younger and possibly weaker Triassic Songpan-Ganzi terrane in the northwest and the stronger lithospheric root of the Yangtze block to the south 9,10 ( Fig. 1).
The deformation and uplift of the Longmenshan orogenic belt were initiated as early as the Middle to Late Triassic and were associated with the Indosinian orogeny and the amalgamation of the North China, South China and Qiangtang continental blocks [11][12][13] . Reactivation during Cenozoic Indian-Asian collision resulted in the formation of one of the steepest continental escarpments in the world [14][15][16] . The Longmenshan orogenic belt has been a key region to evaluate the Mesozoic-Cenozoic tectonic evolution of China [17][18][19][20][21] and has also been termed the "Asian puzzle" 22 with important records of continental collision, magmatism, basin evolution and earthquakes 23 .
The critical tectonic location and frequent earthquakes have made the Longmenshan area the focus of various investigations such as petrology, GPS, numerical simulation, geochronology, geomorphology and geophysics in the last several years 15,21,[24][25][26][27][28] . However, the deformation and uplift mechanism for the Longmenshan orogenic belt remains controversial, and two end-member models have been proposed: (1) Crustal thickening through major slip along thrust faults rooted in the lithosphere [29][30][31] and (2) Channel flow or extrusion of partially molten middleto lower-crustal materials with extremely low viscosity outwards from the Tibetan Plateau since approximately 40 Ma 4,10,14,25,32,33 . Obviously, these works do not take into account the effect of mantle dynamics.
In this work, we collected high-quality seismic data recorded by the China earthquake network and employed teleseismic P-wave tomography to detect the upper mantle velocity structure. The results show three high-velocity perturbations under the Songpan-Ganzi terrane, Ordos Basin and Sichuan Basin. These high-velocity structures are at almost the same depth of ca. 400-500 km, and we suggest that these high-velocity structures may be attributed to ancient lithosphere that could have been delaminated beneath the Songpan-Ganzi terrane, Ordos Basin and Sichuan Basin. Specifically, a large-scale low-velocity perturbation covers almost all the Songpan-Ganzi area in the uppermost mantle; under this anomaly, there is a high-velocity anomaly (plate-like) at 400-600 km depth, 1

Results
The average value of the relative travel-time residual at each station is negative in the western part of the study region or the Longmenshan orogenic belt and positive in the eastern part of the study region (Fig. 2), which suggests the presence of lower-and higher-velocity anomalies beneath the western and eastern parts of the Longmenshan orogenic belt, respectively.
The 60, 110, 200 and 300 km depths reveal a prominent and large-scale low-velocity perturbation (Lv1) in the western part of the Longmenshan orogenic belt (Fig. 3). Huang et al. 57 and Yang et al. 58 also determined similar results. Adjoint tomography 59 , ambient noise tomography 60 , East Asia mantle tomography 61 and multi-scale travel-time tomography 48 also show a low-velocity structure beneath the Songpan-Ganzi terrane. In the eastern part of this study area, a large-scale high-velocity perturbation is observed at 60, 110 and 200 km depths beneath the Ordos Basin (Hv1), and another large-scale high-velocity perturbation is observed at 60, 110, 200 and 300 km beneath the Sichuan Basin (Hv2) (Fig. 3), which is consistent with previous studies 57,58,62-64 . Xin et al. 65 also indicated high-velocity structures beneath the Sichuan and Ordos Basins using double-difference seismic travel-time tomography. Hv1 and Hv2 should be the lithospheric roots of the Ordos Basin and Sichuan Basin, respectively.
At 300 and 400 km depths, there is a small high-velocity perturbation (Hv3) (Fig. 3), which is located within the Ordos Basin. At 400, 500 and 600 km depths, there are three high-velocity anomalies (Hv4, Hv5 and Hv6) beneath the western part of the Longmenshan orogenic belt, the Sichuan Basin and the Ordos Basin (Fig. 3), respectively. Due to the low resolution, we discard the features on the 700 and 800 km depth sections.
Profiles a and b show that Lv1 is approximately 300 km thick and Hv2 is approximately 350 km thick (Fig. 4a,b). A larger high-velocity structure (Hv4) lies beneath the Songpan-Ganzi terrane, and another Huang et al. 57 also revealed two high-velocity structures beneath the south part of the Songpan-Ganzi terrane and the Sichuan Basin at 450-600 km depths, but these depths and the locations of the high-velocity structures are different from our results. Yang et al. 58 showed a high-velocity structure beneath the Sichuan Basin at 300-500 km depths, which is similar to our Hv5 (Figs 3 and 4).
However, most studies did not reveal the high-velocity structures Hv4 beneath the Songpan-Ganzi terrane and Hv6 beneath the Ordos Basin. Moreover, three high-velocity structures (Hv4, Hv5 and Hv6) identified by this study are at the same depth (Figs 4 and 5), and, there is an especially good correspondence between the high-velocity (Hv4) and low-velocity (Lv1) perturbations in the Songpan-Ganzi terrane (Fig. 4). Our results also indicate the Longmenshan orogenic belt is the boundary between the low-and high-velocity perturbations (Fig. 4).

Discussion
Lithospheric delamination. The Indosinian orogeny might have imposed its imprints upon the seismic velocity structure and composition in the crust and upper mantle 66 . A number of studies have demonstrated that the high-and low-velocity relics generated by lower crustal/lithospheric delamination (or a subducting slab) and upwelling asthenosphere can be retained for two billion years and can be detected by seismic techniques [67][68][69][70][71][72][73][74] . The large-scale high-velocity perturbation (Hv4) beneath the Songpan-Ganzi and Qiantang terranes (Figs 3 and 4) is approximately 200 km thick, which is thicker than the crust; accordingly, we suggest this feature might be associated with the delamination of the thickened lower crust/lithosphere mantle. This region has been affected by two major tectonic events: one is the Indosinian orogeny in the Early Triassic, and the other is the Himalayan orogeny in the Cenozoic. Based on the scale and depth of Hv4, we speculate it might either have been induced by the collision and amalgamation among the Yangtze, North China and Qiangtang terranes during the Indosinian orogeny in the Mesozoic or during the Cenozoic India-Asia collision. Previous interpretations of tomographic models have proposed the lithosphere foundering of lithosperic slabs beneath the Tibetan Plateau due to the northward subduction of the Indian Plate during the Himalayan orogeny in the Cenozoic 75 . Therefore, we cannot exclude that the lithosphere mantle of the Songpan-Ganzi terrane may have been removed by the eastward www.nature.com/scientificreports www.nature.com/scientificreports/ extrusion of the Tibetan Plateau or that the high-velocity structure (Hv4) may be linked to the subduction of the lithosphere in the Cenozoic 76,77 .
Orogenies generally lead to lithospheric delamination simultaneously or after collision 78 . Geological studies demonstrated that during the Early Triassic Indosinian orogeny, the crust of the Songpan-Ganzi terrane was strongly thickened through continental collision among the Yangtze, the North China and the Qiangtang blocks [53][54][55][79][80][81][82][83] , which accompanied the closure of the Paleotethys Ocean 84 . The thickened lower crust experienced eclogite-facies metamorphism, leading to increased density that might have resulted in the delamination of the lower crust/lithosphere in the Songpan-Ganzi terrane in the Middle Triassic 79 .
Delamination can result in upwelling asthenosphere that fills voids formed by delamination 85 . Just above Hv4 is a large-scale low-velocity perturbation (Lv1); we therefore suggest that this large-scale low-velocity structure may have been produced by upwelling asthenosphere that filled the voids generated by delamination. Petrological   www.nature.com/scientificreports www.nature.com/scientificreports/ studies have demonstrated that in the Songpan-Ganzi and Longmenshan terranes, magmatic events are recorded throughout the Mesozoic [81][82][83]86 , which should have been associated with the upwelling asthenosphere. This upwelling asthenosphere might have heated the base of the lower crust and promoted partial melting to generate adakitic magmas in the lower crust and/or A-type granitic magma 79 . Geochemical studies show that the adakitic and A-type granite magmatism in the Songpan-Ganzi and Longmenshan terranes occurred during the Late Triassic within a post-collisional setting 79 .
On the other hand, Hv5 and Hv6 (as well as Hv3) are under the Sichuan Basin and Ordos Basin, respectively. The root of the Sichuan Basin lithosphere is thicker than that of the Ordos Basin; however, the scale of Hv6 is larger than that of Hv5 (Figs 4 and 5). Based on this contrast, we suggest that lithospheric delamination to different degrees occurred beneath the Sichuan Basin and the Ordos Basin in the Late Triassic due to the stronger collision among the Qiangtang, Yangtze and North China blocks 86 . Hv5 might have been part of the Sichuan Basin lithosphere, and Hv6 might have been part of the Ordos Basin lithosphere in the Early Triassic.
Deformation and uplift mechanism of the longmenshan orogenic belt. The channel flow model of the middle/lower crust 32,43,[87][88][89] has been employed in some studies to explain the formation of the Longmenshan orogenic belt. Low to moderate Poisson's (or Vp/Vs) ratios in the Songpan-Ganzi and Longmenshan area suggest that the lower crust is dominated by felsic to intermediate compositions 41,90 , which can preserve partial melting 20 to 30 Ma after crustal thickening 91,92 . Geophysical studies indicate the low-velocity zone associated with high conductivity at a depth below 10 km on the southern Tibetan Plateau 93 . Therefore, we cannot exclude the possibility that the crustal thickening was generated by the channel flow of the middle/lower crust.
Furthermore, many workers believe that the crustal thickening that occurred in the Longmenshan area was caused by ductile deformation rather than by thrust faulting or crustal shortening 4,24 . The eastward extrusion of the Tibetan Plateau is considered associated with rapid slip along these faults 94 , and significant relative motion along major strike-slip faults facilitated the eastward extrusion of crustal material out of the Tibetan Plateau [95][96][97] .
Our study shows that the inferred large-scale delamination beneath the Songpan-Ganzi and Qiangtang terranes resulted in the removal of more rigid lithosphere and the upwelling asthenosphere, leading to further heating of the lower crust and producing ductile and easily deformed lower crust or forming channel flow of the middle/lower crust. Moreover, the upwelling asthenosphere directly contacted the lower crust, resulting in a detachment surface between the lower crust and upper mantle, facilitating the easy eastward extrusion of the Songpan-Ganzi terrane and leading to the ductile deformation or thickening of the lower crust of the Songpan-Ganzi terrane due to the collision between the Indian and Eurasian plates in the Cenozoic.
The large-scale high-velocity perturbation beneath the Sichuan Basin (Hv2) (Figs 3-5) is the old, strong lithosphere that underlies the Sichuan Basin 31,46,98 , pushing up the crust above and achieving steep topography through dynamic pressure. This process accounts for the deformation and uplift of the Longmenshan orogenic belt.

earthquakes.
A ductile crust plays an important role in generating earthquakes through coseismic slip and rupture along the Longmenshan fault zone. The lower-crustal material flowing outward from the Songpan-Ganzi block might have been supported by a strong craton-like lithosphere underlying the Sichuan foreland basin 46 . The present-day strain energy related to continental deformation in the central Tibetan Plateau, generated through collision between the Indian and Eurasian plates, is related to strike-slip faulting along active strike-slip faults such as the Kunlun and Xianshuihe faults, generating strong earthquakes that release the accumulated strain energy 99 .
The surface ruptures of the 1997 Manyi, 2001 Kokoxili, and 2010 Yushu earthquakes were dominated by pure left-lateral faulting along the WNW-trending strike-slip faults, representing an eastward motion of the Songpan-Ganzi and Qiangtang blocks bounded by mega-strike-slip faults in the northern Tibetan Plateau 95,99-103 .
Meanwhile, eastward movement of the Songpan-Ganzi block against the lithospheric root of the Sichuan Basin (Hv2) (Fig. 4) results in large stress accumulation along the Longmenshan fault zone. When the stress exceeds the threshold, it is suddenly released through a violent rupture along the faults. The Longmenshan area is likely indicative of cumulative offsets from earthquakes 4,28,104 . Our results support the notion that the eastward extrusion of the Songpan-Ganzi block might have resulted in stress concentration mostly at the bottom of the Longmenshan fault zone, which is thought to be responsible for the generation of the 2008 Mw 7.9 Wenchuan and 2013 Mw 6.6 Lushan earthquakes ( Fig. 1: E1 and E2). Therefore, we suggest that the northward movement of the Indian plate entrains the eastward extrusion of the Songpan-Ganzi terrane, causing stress accumulations and releases as well as large earthquakes in the Longmenshan area.

Conclusions
Our results suggest that there was massive delamination of the lower crust/part of the lithosphere in this area during the Late Triassic due to the stronger collision and amalgamation of the North China, South China and Qiangtang continental blocks. We cannot exclude the notion that the large-scale high-velocity structure beneath the Songpan-Ganzi terrane may be linked to the subduction of the lithosphere or the removal of part of the rigid lithosphere in the Cenozoic due to the eastward extrusion of the Tibetan Plateau induced by the Indian-Asian collision. The large-scale delamination of the lithosphere resulted in upwelling asthenosphere, which enhanced the ductile nature of the Songpan-Ganzi block, resulting in crustal thickening, deformation and uplift of the Longmenshan belt in the Cenozoic due to the eastward extrusion of the Tibetan Plateau. The eastward extrusion of the Songpan-Ganzi block is inhibited by the rigid lithosphere of the Sichuan Basin, generating stress accumulation in the Longmenshan area. When the stress exceeds the threshold, its sudden release along the faults accounts for the violent earthquakes in the Longmenshan region, such as the 2008 Mw 7.9 Wenchuan earthquake and the 2013 Mw 6.6 Lushan earthquake. (2019) 9:6967 | https://doi.org/10.1038/s41598-019-43476-0 www.nature.com/scientificreports www.nature.com/scientificreports/

Data and Method
For this study, we collected 554 teleseismic events recorded by Chinese earthquake networks with 403 seismic stations from July 2007 to March 2014 (Fig. 1) 105 . The seismic events with magnitudes >6.0 and epicentral distances ranging from 30°-85° were selected (Fig. 1, insert figure). Waveforms were cut 15 s before and 50 s after the first P-wave arrival from digital seismograms and filtered between 0.3 and 3 Hz. We selected 66492 first P-wave arrivals from the cut and filtered seismograms of seismic events using the time cross-correlation method 106 (Fig. S1). The range of −2.5 s to +2.5 s in the relative travel-time residuals was used in the tomographic inversion (Fig. S2).
The lateral grid interval was 1° in this optimal Vp model, and vertical grid intervals were designed at 60, 110, 200, 300, 400, 500, 600, 700 and 800 km depths. The computation of ray paths and theoretical travel times was performed by an efficient 3-D ray-tracing method 72 . We adopted a conjugate-gradient inversion algorithm 107 with smoothing and damping regularizations to determine the sparse and large system of observation equations 72 . In the 3-D tomographic inversion, the iasp91 1-D Earth model 108 was used as the starting model.
The correction of the relative travel-time residuals was done with the CRUST1.0 model 109 following the method of Jiang et al. 110 , which removed the effect of crustal heterogeneity generated by crustal thickness and velocity 110,111 . Based on the upper depth limit of tomography, we applied a crustal correction for the upper 60 km of the lithosphere. The results showed that the corrected value induced a large effect on the distribution of the relative travel-time residuals (Figs S3 and S4), and the larger corrected values were mainly distributed in the western part of the study region (Fig. 4). To select a suitable damping parameter for use in the tomographic inversion, we performed numerous inversions with different damping parameters to derive a trade-off or L-shaped curve that could measure the misfit of each solution model with the data [112][113][114] . Eventually, a damping value of 11.0 was taken to invert the final solution model (Fig. S5).
The checkerboard resolution test (CRT) was employed to assess the reliability of the obtained tomographic images and the adequacy of ray coverage. All the grid nodes of the 3-D space were assigned 2.5% negative and positive velocity perturbations. Figure S6 shows the CRT result. In general, most parts of the study region have good resolution (Fig. S6), where the checkerboard pattern and the amplitude of velocity anomalies are well recovered. The resolution is reduced at 60, 700 and 800 km depths and in the western part of the study area, indicating that the seismic rays do not crisscross well in these locations.