Localized foundering of Indian lower crust in the India–Tibet collision zone

Significance

Although plate tectonics explains subduction of oceanic crust, the areally distributed deformation observed in continental collision requires more complex explanations of the geometric interaction of two colliding continents and the mechanics of plateau formation. Our seismic images reveal variations of the India–Tibet collision parallel to the Himalaya. Our observation of localized thinning of Indian lower crust, that is sufficiently extreme as to require material loss presumably into the underlying mantle, is an advance in explaining the apparent lack of mass balance in the India–Asia collision that requires loss of continental material. These zones of lower-crustal thinning and likely foundering represent slab segmentation (slab tears) that can geometrically accommodate the curvature of the Himalayan arc.

Abstract

The deep structure of the continental collision between India and Asia and whether India’s lower crust is underplated beneath Tibet or subducted into the mantle remain controversial. It is also unknown whether the active normal faults that facilitate orogen-parallel extension of Tibetan upper crust continue into the lower crust and upper mantle. Our receiver-function images collected parallel to the India–Tibet collision zone show the 20-km-thick Indian lower crust that underplates Tibet at 88.5–92°E beneath the Yarlung-Zangbo suture is essentially absent in the vicinity of the Cona-Sangri and Pumqu-Xainza grabens, demonstrating a clear link between upper-crustal and lower-crustal thinning. Satellite gravity data that covary with the thickness of Indian lower crust are consistent with the lower crust being only ∼30% eclogitized so gravitationally stable. Deep earthquakes coincide with Moho offsets and with lateral thinning of the Indian lower crust near the bottom of the partially eclogitized Indian lower crust, suggesting the Indian lower crust is locally foundering or stoping into the mantle. Loss of Indian lower crust by these means implies gravitational instability that can result from localized rapid eclogitization enabled by dehydration reactions in weakly hydrous mafic granulites or by volatile-rich asthenospheric upwelling directly beneath the two grabens. We propose that two competing processes, plateau formation by underplating and continental loss by foundering or stoping, are simultaneously operating beneath the collision zone.

Knowledge of the structure of the Himalayan collision zone is the key for understanding all geological processes––geodynamics, tectonics, magmatism, metallogeny––occurring during continental collision. However, the lack of constraints on the geometry and extent of the Indian plate beneath the Tibetan plateau has fueled speculation whether the Tibetan plateau was built by underthrusting of Indian crust beneath the entire plateau (1), by discrete subduction events between coherent lithospheric blocks (2), or by homogeneous thickening of the lithosphere of the entire plateau (3) and widespread viscous flow in the crust (4). Knowledge of the collision zone has been improved by seismic studies in the past two decades (5–11), which suggest that Indian lower crust attached to lithospheric mantle has underplated to ∼31°N (7), that a west–east transition from underplating to steep subduction of the Indian plate has occurred in the central–eastern collision zone (9, 10), and that a large portion of the Indian crust has been transferred from the lower plate to the upper plate via crustal-scale duplexing (11). However, the amount of underplating (∼400 km from the Main Frontal Thrust to ∼31°N (Fig. 1 and ref. 7) and crustal-scale duplexing accounts for only ∼50 ± 17% of the Indian crust underthrust beneath southern Tibet since onset of the India–Asia collision (12). The fate of the Indian crust (especially the lower crust) beneath southern Tibet during ∼57 My since collision remains largely speculative.

Fig. 1.

Topographic map of southern Tibet showing location of seismometers. Yellow-filled diamonds were used to construct RF Profiles A and B in Figs. 2 and 3, and blue diamonds mark our Gangdese seismic stations deployed from 2011 to 2019 (9, 10). Black diamonds indicate other seismic stations (5–8, 13). Dotted cyan line: northern limit of underthrust ILC (mantle suture) (7, 8, 10). Cyan rectangles: locations of maximum thinning of ILC on Profile B, and likely correlative points on Profile A as interpreted on Fig. 3. Dashed red lines: Yarlung-Zangbo (YZS) and Banggong-Nujiang (BNS) sutures. Solid red lines: MFT, dextral Jiali fault (JF), and normal faults in the Tangra-Yumco (TYG), Pumqu-Xainza (PXG), Yadong-Gulu (YGG), and Cona-Sangri (CSG) grabens (14). Cyan quadrant circles: focal mechanisms of earthquakes with epicentral depths ≥50 km (SI Appendix, Fig. S4) verified by waveform modeling (15). (Inset) Map shows full extent of the Tibetan plateau.

Receiver Function Imaging of the Himalaya–Tibet Collision Zone

Since summer 2011 we have deployed 411 seismic stations in southern Tibet, equipped with either Guralp 3ESPCD or Guralp 3TD broadband seismometers. Most of them operated for ∼12 mo. These stations fill the gaps between the Himalayan-Tibetan Continental Lithosphere During Mountain Building (Hi-CLIMB) (7), Himalayan Nepal Tibet Seismic Experiment (6), International Deep Profiling of Tibet and the Himalaya (5), and Namche Barwa (8) passive seismic experiments, and provide much improved coverage for the India–Tibet collision zone (Fig. 1).

Here we present results from two high-resolution receiver-function (RF) profiles (Figs. 2 and 3) paralleling the India–Tibet collision zone, constructed with data we recorded from 2015 to 2019 on two densely spaced 800- and 600-km-long linear arrays. These arrays extend east–west from the eastern Himalayan syntaxis to the Tangra-Yumco graben (TYG) in the central-western Gangdese belt, with Profile B along the Yarlung Zangbo suture (YZS) and Profile A located ∼75 km north (Fig. 1).

Fig. 2.

RF images with no vertical exaggeration, showing the principal contrasts within the entire crust and uppermost mantle along Profiles A and B. Red and blue colors represent interfaces with increasing and decreasing impedance with depth, respectively. The “d” indicates the doublet conversion, or the conversion from the top of the ILC (5–8). The “m” denotes the conversion from the Indian crust–mantle boundary (or Moho). Focal mechanisms from 28.5–30.5°N (15) in A and B are projected onto the profiles along the strike of nearby grabens (SI Appendix, Table S3). Depths are referred to sea level. Thin gray lines at 65-km depth are for reference. Representative RFs (stacked every 5° of back-azimuth with time transformed to depth) are shown in C. Note that the Moho has eluded imaging beneath the station YS91 for some northern azimuthal arrivals (where marked “?”), suggesting the thinning of the ILC may have also occurred on the northern side of the station. (Lower Right) Small blue and green crosses show the piercing points at 65-km depth of all RFs used in Profiles A and B, respectively. (Inset) Maps show the corresponding piercing points for the three representative stations (YS91, YS78, and LM47) identified in A and B and shown in C.

Fig. 3.

(A and B) show RF profiles with 5× vertical exaggeration, and gravity data and models. Superimposed on RF profiles are approximate top (green lines) and bottom (black lines) of the ILC as used for gravity data modeling. Stars are published depths to top ILC and Moho at ∼85°E (7), ∼87°E (6), ∼90°E (5), and ∼93°E (8). Cyan rectangles on Profile B show locations of maximum thinning of ILC, and on Profile A our interpreted correlative points. Thin gray lines are Moho depths from the CRUST1.0 model. Black curves above each RF profile are relative satellite gravity anomalies at 8,500-m altitude (SI Appendix, Fig. S3D), and other colors are modeled anomalies for different hypothetical densities of the ILC (brown: 2,850 kg/m³, same as overlying crust; thick green: best fit with 3,050 kg/m³; blue: 3,300 kg/m³, same as underlying mantle). (C) Interpretive geologic cartoon along Profile B. Solid lines: clearly observed interfaces; dashed lines: inferred boundaries. Pink lines: proposed direct contact of Indian middle/lower crust with the asthenosphere or the LAB (lithosphere-asthenosphere boundary). Dashed black lines: interpreted Main Himalayan Thrust (MHT) where not following top of ILC. Arrows show relative motions. Focal mechanisms (15) for earthquakes 28.5–30.5°N are projected onto the profile along-strike of nearby grabens (SI Appendix, Table S3). Depths are relative to sea level. Nomogram in A (87–88°E) shows appearance of true dips of 0–20° after 5× vertical exaggeration as used throughout this figure.

Previous RF profiles south to north across the collision zone identified an RF “doublet” interpreted as representing the top and bottom (Moho) of Indian lower crust (5–10, 16), allowing us to tie our observations to the previous interpretations (Fig. 3 and SI Appendix, Fig. S1A). The (Indian) crust–mantle boundary (or Moho), is the deepest prominent positive conversion and is seen along most of both profiles (“m” in Figs. 2 and 3). The Moho converter is at 65–70-km depth at the west end of Profile B (Fig. 3) as previously determined along the Hi-CLIMB profile at ∼85°E (7, 17), then deepens dramatically to ∼90 km––the largest depth in the study region––at 87.5°E. The Moho conversion is absent immediately west of the Pumqu-Xainza graben (PXG), whether due to difficulty in imaging a steep structure or due to the lack of a seismic impedance contrast, and reappears at <65-km depth just 30 km further east. From the PXG east to the Cona-Sangri graben (CSG) the Moho on Profile B is subhorizontal at ∼75 km depth for ∼280 km between ∼89° and ∼91.5°E. A second Moho disruption occurs beneath the CSG (see ref. 10), then the Moho shallows gently to only ∼55 km in the vicinity of the eastern Himalayan syntaxis at the east end of Profile B. Profile A shows the same two zones of Moho disruption as Profile B, less clearly beneath the PXG where data coverage is incomplete, but more obvious at 91.5–92°E where the Moho conversion is locally absent (see ref. 10), ∼50 km west of the northern mapped extent of the CSG (18). On Profile A the two zones of Moho disruption bound a subhorizontal section of Moho, just as on Profile B, that on A is narrower (spans only ∼220 km from west to east) and a few kilometers deeper (implying ∼5° northward dip).

The top of the RF doublet, normally interpreted as the top of (Indian) lower crust (ILC) (5–10, 16), is the prominent positive conversion lying above the Moho. It is well-observed along both of our profiles (marked “d” in Figs. 2 and 3) especially where the doublet reaches ∼20-km thickness from ∼89–91.5°E in the region where it was first identified (5). Immediately west and east on Profile B, two depressions in the top-ILC converter are obvious, of up to 5 km west of PXG and of up to 10 km beneath CSG. In these regions and at the western and eastern limits of our profiles the top-ILC converter is arguably too thin to be resolved from the Moho as a separate converter, and our data are easily modeled by a single converter (SI Appendix, Fig. S2C). At the largest scale from west to east the top ILC appears to mirror the Moho (i.e., deeper [shallower] when the Moho is shallower [deeper]), although locally, where best observed from ∼91.5–89°E, it is clear that the top and bottom of the doublet both dip north, both deepening ∼5 km from beneath Profile B to beneath Profile A.

Common conversion-point (CCP) images of the P-RF multiple phases, and of the S-RFs (SI Appendix, Fig. S1 C and D), corroborate our interpretations of the simple P CCP images shown in Figs. 2 and 3.

Gravity Characteristics of the Indian Lower Crust in the Collision Zone

Previous work not only identified the doublet layer between “d” and “m,” but also showed that portions of the ILC >300–350 km north of the Main Frontal Thrust (MFT), somewhat north of the YZS, have density and wavespeed characteristic of partially eclogitized mafic crust (7, 16, 17). We next use satellite gravity observations (SI Appendix, Fig. S3) to test this hypothesis. In opposition to simple expectations of lower gravity over thicker crust, the relative gravity anomaly reaches +50 mGal (+70 mGal on Profile B) from ∼89–91.5°E where the Moho is significantly deeper than adjacent areas (Fig. 3, black line; SI Appendix, Fig. S3D). Simple modeling with a range of densities finds acceptable fits if the doublet layer has a density ∼3,000–3,100 kg/m³ (Fig. 3; rms misfit ∼24 mGal, SI Appendix, Table S2), clearly higher than the density of granulitic rocks (2,850 kg/m³, rms misfit = 34 mGal), but lower than the density of the mantle (3,300 kg/m³, rms misfit = 34 mGal), or mafic eclogite (3,450 kg/m³) (values taken from ref. 16). The average density of the ILC along Profiles A and B thus matches mafic granulites with ∼30% eclogitization (6), or metastable gabbro (16, 19, 20). We stress that we have modeled the gravity anomaly from a smooth interpretation of the RF image using constant density contrasts at the top and bottom of the ICL. Because our focus is our seismic data we prefer to show an acceptable gravity fit from a simple model rather than a perfect fit from a very complex model. Our modeling approach and result agrees with previous models (16) that permit gravitationally stable but chemically metastable mafic ILC beneath Profiles A and B, incompletely eclogitized but with the potential to rapidly chemically equilibrate leading to gravitational instability.

Localized Foundering of the ILC

Southern Tibet stands out globally for its unusual lower-crustal and upper-mantle earthquakes (15, 21, 22) (Figs. 1 and 2 and SI Appendix, Fig. S4). These earthquakes require a strong lithospheric mantle (15, 21) capable of brittle failure, and may be associated with eclogitization of the lower crust (23). When we project earthquakes with depths well-determined by waveform modeling (15) to be ≥50 km, along-strike of the grabens onto our profiles (SI Appendix, Table S3), the earthquakes cluster in two small areas where the ILC becomes thinner and the Moho is less well-imaged (we suggest due to strong crust–mantle interaction producing a diffuse Moho) (Figs. 3C and 2). These earthquakes show a consistent pattern of east–west extension so are likely related to the rifting process that produces PXG and CSG. However, the degree of thinning observed in the ILC in our profiles, locally >60% beneath PXG and CSG, and regionally [>15% where confidently observed and independently corroborated from 87° (6) to 93°E (8)], far exceeds the 3–6% extension observed at the surface (24). Hence, in addition to focused extension of the ILC achieved by decoupling between upper and lower crust (13), material must be physically removed from the ILC to produce the observed thinning. Because the earthquakes occur close to the bottom of the ILC and the upper-mantle low shear-wavespeed zones (Fig. 4 and SI Appendix, Figs. S6 and S8), we suggest they enable eclogitization (23) and stoping of foundering of newly densified ILC into the mantle (Figs. 3C and 4) as the mechanism of this mass removal. Rapid eclogitization to exceed the mantle density may be enabled by dehydration reactions in weakly hydrous mafic granulites (16) or by volatile-rich asthenospheric upwelling directly beneath the rifts (25). The absence of the Indian Moho converter beneath the CSG and PXG could then be attributed to local juxtaposition of eclogitized ILC against asthenospheric mantle with similar seismic impedance (Fig. 3C). Loss of dense lower-crust foundering into the mantle requires replacement by an equal volume of material. We speculate that this replacement is by mantle flow from below (to create observed Moho shallowing) but also by channel flow of the weakest crustal layer, felsic/intermediate Tibetan middle crust (26), into the region of delamination, thereby deepening the top of the ILC.

Fig. 4.

Interpretive cartoon (with exaggerated topography) of processes currently operating in the India–Tibet collision zone. Beneath the Moho we show mantle S-wave velocity structure from our teleseismic tomography (SI Appendix, Figs. S6 and S8). ILC is simultaneously underplating and foundering beneath the CSG and PXG grabens, and foundering then subducting at the northern end of the collision zone. West–east cuts in the three-dimensional block model are along our Profiles A and B. South–north cut along 85°E from ref. 7. Arrows are inferred directions of lithospheric foundering (dark blue) and asthenospheric upwelling (purple) in the upper mantle.

Localized foundering of Indian lower crust into upwelling asthenospheric mantle beneath the CSG and the PXG is consistent with the observation in these rifts of primordial mantle ³He in thermal springs that is a signature of incipiently melting (i.e., asthenospheric) mantle directly beneath the crust (27, 28). Adjoint-tomographic images suggest lithospheric foundering of high-wavespeed upper-mantle “blobs” as small as ∼300 km across, north of the YZS at 88° and 92° (29), but lack the resolution to show smaller features. Our own travel-time tomographic models show ∼100-km length-scale high-velocity bodies beneath the two grabens (Fig. 4 and SI Appendix, Figs. S6 and S8) provide additional support for localized foundering of the underthrusting Indian continental lithosphere.

The inefficient conversion at the top of ILC beneath the YZS near the west ends of our profiles (Figs. 2 and 3 and refs. 7, 17) lies directly above low-wavespeed anomalies in the upper mantle (Fig. 4 and SI Appendix, Figs. S6 and S8 and ref. 30), therefore is likely causally linked to in situ high mantle temperature, which may have suppressed eclogitization (19) in the ILC and thereby decreased the seismic impedance contrast at its top boundary.

Non-coaxial Rifting Deformation in Southern Tibet

Delamination of the lower lithosphere (25) and channel flow in the middle crust (13) have both been linked to west–east extension of Tibet manifested at the surface by the NNE-trending extensional grabens in southern Tibet. Our two regions of steeply dipping and disrupted Moho along Profile B are broadly beneath the CSG and PXG (Fig. 1) but because the widths of disruption are comparable with the rift spacing in southern Tibet and because we see no evidence of disruption beneath the Yadong-Gulu graben (YGG) or TYG (Fig. 3) we have insufficient data to prove a direct link between surface faulting and Moho faulting (13). Nonetheless the two foci of crustal thinning on Profile B are so close to the CSG and PXG that we speculate their origins are linked. Tearing of subducting Indian mantle lithosphere (31) along inherited Indian basement faults (32) could focus lower-crustal delamination or foundering (25) thereby weakening the crust and focusing surface extension (Fig. 4). The existence of a decoupling midcrustal channel means that surface extension is not required to be vertically above the foundering lower crust (13). Such non-coaxial rifting can explain both the presence of Moho disruption at ∼91.5°E on Profile A ∼50 km west of the CSG (10), and the absence of Moho disruption beneath the YGG on either profile. We note that the lateral extent of regions of shallow Moho and thinned ILC are broader west-to-east on Profile A than on our southern Profile B. Tomographic images suggest this may be associated with processes at the northern limit of the ILC at the Moho, where the ILC likely subducts (7, 9, 10) rather than underthrusts Tibetan crust, allowing Tibetan (noncratonic) mantle at the Moho to infiltrate above ILC (Fig. 4 and SI Appendix, Figs. S6 and S8).

Our results suggest that two competing processes, underplating and localized foundering, are operating simultaneously in the collision zone, with the former maintaining or increasing crustal thickness and the latter reducing crustal thickness. The thickness and northern limit of ILC underplating southern Tibet is the net result of these opposing processes. The zones of lower-crustal foundering represent slab segmentation (slab tears) that can geometrically accommodate the curvature of the Himalayan arc.

Materials and Methods

We focus our images on lower-crustal and upper-mantle structures (Fig. 3). We obtained the images using 12974 P-wave RFs derived from 168 seismic stations (yellow-filled diamonds in Fig. 1) and the CCP stacking method. The RFs were calculated with the time-domain iterative deconvolution technique (33), in which the Gaussian filter factor was set to 1.5 to mitigate high-frequency conversions from the upper crust and retain signal from the lower crust and upper mantle. All of the RFs were migrated from time to depth by tracing the rays from the location of each seismic station through a layered reference model (SI Appendix, Table S1). No lateral or vertical smoothing was applied to the images, to best image east–west variability. We also used teleseismic travel-time tomography to study upper-mantle processes associated with the localized foundering of ILC (Fig. 4 and SI Appendix, Figs. S5–S9). In total, 21,731 S-wave (S, sS, ScS, SKS, and SKKS) arrival times from 1,128 earthquakes and 725 seismic stations were used to invert the upper-mantle S-wave velocity structure. More details are given in SI Appendix, Text.

Data Availability.

All RF and travel-time data used in this study are available (34) (https://purl.stanford.edu/jz904sc7304).