ARTICLESPUBLISHED ONLINE: 19 SEPTEMBER 2016 | DOI: 10.1038/NGEO2806

Large-scale subduction of continental crustimplied by India–Asia mass-balance calculationMiquela Ingalls1*, David B. Rowley1*, Brian Currie2 and Albert S. Colman1

Continental crust is buoyant compared with its oceanic counterpart and resists subduction into the mantle. When twocontinents collide, the mass balance for the continental crust is therefore assumed to be maintained. Here we use estimatesof pre-collisional crustal thickness and convergence history derived from plate kinematic models to calculate the crustal massbalance in the India–Asia collisional system. Using the current best estimates for the timing of the diachronous onset ofcollision between India and Eurasia, we find that about 50%of the pre-collisional continental crustal mass cannot be accountedfor in the crustal reservoir preserved at Earth’s surface today—represented by the mass preserved in the thickened crust thatmakes up the Himalaya, Tibet and much of adjacent Asia, as well as southeast Asian tectonic escape and exported erodedsediments. This implies large-scale subduction of continental crust during the collision, with a mass equivalent to about 15%of the total oceanic crustal subduction flux since 56 million years ago. We suggest that similar contamination of the mantle bydirect input of radiogenic continental crustal materials during past continent–continent collisions is reflected in some oceancrust and ocean island basalt geochemistry. The subduction of continental crust may therefore contribute significantly to theevolution of mantle geochemistry.

Collision-related crustal mass balance is critical tounderstanding how continental crust behaves in large-scale collision systems, how continental crustal composition

evolves over time, and the degree to which direct input ofcontinental crustal materials into the mantle may impact mantlegeochemistry. It has long been assumed that crustal mass balanceis maintained during collisions1,2, because continental crust is toobuoyant to be subducted. Previous analyses of collision-relatedmassbalance in the India–Asia system have yielded results that rangefrom balanced3 to unbalanced4, to potentially balanced5, reflectingdiffering approaches, ages of initial collision, and masses/volumesof contributing components. We reassess the India–Asia collisionalmass balance by integrating the most recent estimates of collisionage, plate kinematic constraints on India–Asia convergence, and thecorrelation of palaeoelevation and crustal thickness distribution,all with associated uncertainties. We compare these with estimatesof the present crustal mass within the collision system: integratedcrustal thickness within the collision domain, escaped crustal mass,and detrital and dissolved sediment fluxes. We review and justifyeach aspect of the mass-balance calculation below.

Mass balance in an active collision zoneCrustal mass balance in the Himalaya–Tibet system integrates themass at the onset of the collision relative to crustal mass measuredtoday6. If continental crustalmass is conserved at the Earth’s surface,the continental crustal mass at the onset of collision (Mconvergence)must balance the net mass reflected in crustal thickening andexported from the system:

Mconvergence=Mxs+Merosion+Mescape (1)

Mconvergence is the mass of Greater Indian and Greater Tibetancontinental crust (Fig. 1b) reconstructed at the onset of collision,Mxsis the mass added to the Himalaya–Tibet system by collision-related

crustal thickening integrated over the deformation domain(Fig. 1b),Merosion is the eroded sediment flux3,7,8 (including dissolvedsediment), andMescape is lateral ‘tectonic escape’9.

We have compiled the most complete set to date of tectonicand geochronologic constraints on the pre-collisional geography ofIndia and Eurasia. Additionally, recent palaeoaltimetry results fromthe Linzizong Arc along the southern margin of Eurasia providean explicit estimate of the crustal thickness beneath the arc at∼55million years ago (Ma).

The crustal mass at the onset of collision can be constrained bythe initial crustal thickness distribution and the integrated amountof Indian and Eurasian plate convergence since collision onset,assuming a constant mean crustal density. Dating the age of onsetof the collision of India with Asia has been a particularly difficultgeological exercise. Recent estimates are greatly improved and implya fairly complex and diachronous history from west to east alongsutures separating India from Eurasia.

Age of continent–continent collisionHere we summarize the age of collisional onset of the Indianand Eurasian continents, starting along the western margin andproceeding eastwards along the suture (full discussion of thedata in Supplementary Information). In the northwest, on theZanskar Shelf (Fig. 1b) the Ypresian-age (51 ± 1.5Ma) collisionof the Kohistan-Ladakh Arc (KLA) with the Indian passivemargin is well constrained by stratigraphic relations10–12, isotopicdating of ultrahigh-pressure metamorphism from the Kaghan13 andTso Morari14,15, and transition from mantle-derived to crustallyinfluenced magmatism16. We accept these data and assign a likelyage of collision initiation at 51 ± 1.5 (1σ)Ma between India andEurasia along the∼1,000-km-long western segment.

Farther east, provenance data constraining the onset of collisionderive primarily from the vicinity of Sangdanlin and Zhepure Shan(Fig. 1b). Interbedded ash near the top of the Sangdanlin section

© Macmillan Publishers Limited . All rights reserved

1The University of Chicago, Department of the Geophysical Sciences, 5734 South Ellis Avenue, Chicago, Illinois 60637, USA. 2Miami University,Department of Geology, 620 East Spring Street, Oxford, Ohio 45056, USA. *e-mail: ingalls@uchicago.edu; drowley@uchicago.edu

848 NATURE GEOSCIENCE | VOL 9 | NOVEMBER 2016 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2806 ARTICLESLa

titud

e

Latit

ude

Latit

ude

50° N

50° N

50° N55 Ma

IY Suture

b c

Longitude Longitude

Longitude

Greater India

CR FZ

W-Z FZ

GB

CB

PB

EP

Antarctica

Australia

a

40° N40° N

40° N40

40

40750 ± 250 km

30° N 30° N

30° N

20° N20° N

20° N

10° N

10° N

10° N

0°

0°

10° S

10° S

20° S70° E

−6 −4 −2 0Elevation (km)

2 4 6 8

80° E 90° E 100° E

60° E 70° E 80° E 90° E 100° E 110° E 60° E 70° E

Crustal thickness (based on Crust 1.0)Contour interval 15 km

50 km3010 70

80° E 90° E 100° E 110° E

Indus-Yarlung Suture

60P/LZS

ZS

Greater Tibet

Greater Indiaat 56 Ma

56 Ma

33.0639.46

42.7747.23

49.4351.90

53.8156.66

58.38

61.65

56.66

53.8151.90

49.4347.23

42.77

20.1320

56 Ma coast

45

45

4560

45

453015

3040

10.95

10.95

25.8220.13

39.46

33.06

25.82

Figure 1 | Palaeo-geographic and -thickness reconstructions of the Himalaya–Tibet orogen. a, Gondwanan reconstruction with Etopo1 topography.Divergent and transform boundaries (white–grey dashed line) are indicated by Perth Basin (PB), Wallaby–Zenith Fracture Zone (WZFZ), Cuvier Basin (CB),Cape Range Fracture Zone (CRFZ, dotted white line) and interpreted extension of the possible northern edge of Greater India, Gascoyne Basin (GB) andExmouth Plateau (EP). Plate boundaries are indicated using grey lines or a white dashed line. b, Reconstructed geometry at age of collision. Modern geog-raphy and crustal thickness are indicated by dark grey outlines and grey-scale shading. Colour is used to indicate colliding plates and collisional domains.Palaeo-positions of the western and eastern syntaxes (ages in Ma are indicated in blue text) relative to fixed Eurasia are shown with 95% confidenceellipses. The double-ended arrow marks the potential leading edge of Greater India when assigned collisional ages of 58 and 54 Ma. Yakovlev & Clark’s5

Greater India (slanted lines) and ‘retrodeformed Asian margin’ (grey dashed line across modern India); present positions of the Penbo/Linzhou Basin (P/L),Sangdanlin (S), Zanskar Shelf (ZS) and Zhepure Shan (Z). The orange and yellow dashed polygons outline the excess crustal mass domains of this studyand Yakovlev & Clark5, respectively. Purple outline marks the region of tectonic escape relative to the orange polygon outline. c, Reconstruction at 55 Ma ofCRUST1.0 (ref. 23) crustal thicknesses with inferences regarding pre-collisional thicknesses of Greater India and Tibet, and a thick Linzizong Arc margin.

provides a minimum date of deposition of Asian-derived detrituswith deep-water Indian passive margin sedimentary rocks at∼58Ma (ref. 17). We assign the age of initial collision of thickercontinental crust of IndiawithAsia at 56± 2Ma, acknowledging thehigh mean convergence rate of >150 kmmy−1 during this intervaland ∼300 km width of continental slope and rise environmentsalong modern passive continental margins. This requires that theyounger suturing along the KLA–India suture terminate west ofthe Xigaze forearc (∼84.8◦ E) thus limiting the length of theyounger 51 ± 1.5Ma suture to <1,000 km. There are no datathat directly constrain the age of the India–Asia collision eastof the Zhepure Shan18, so we assume collision synchronous withSangdanlin as far east as∼93.0◦ E.

We infer a suturing age of ∼40 ± 4Ma (ref. 19) along the east-facing passive margin corresponding to the Gascoyne, Cuvier, andPerth Basin conjugates of Greater India.

Gondwanan reconstruction of Greater IndiaGreater India, a pre-collisional northern continuation of themodernIndian continent, has long been discussed20 as an explanation forthe double thickness of Tibetan continental crust. We use therifting relations of eastern Gondwana between East Antarctica,India and the western Australian margin to define the geometryof eastern Greater India (Fig. 1a). Greater India as postulated hereextends approximately 2,400 km north–south (see SupplementaryInformation). This is compatible with the existing Late Cretaceouspalaeomagnetic data from the Tethyan Himalayas21.

The fit adopted here explicitly constrains the geometry of onlythe eastern passive margin of Greater India with the westernmargindrawn to match the timing of collisions from Sangdanlin (Fig. 1)and farther west. Greater India occupies an area of∼4.2×106 km2

in our reconstruction.

Bootstrap estimate of areal, volume and mass convergenceWe estimate area, volume and mass convergence associatedwith the diachronous India–Asia collision. We incorporate platekinematic constraints on the convergence history and theiruncertainties in our calculations (Supplementary Tables 1 and 2).The effective convergence zone length (CL) as a function of time(Supplementary Table 3) is defined as the difference (D1–D2)between the suture zone end points (EP1 and EP2) measuredrelative to each of the stage rotations (Pt) times the radius of theEarth (REarth)

D1=Pt ∗EP1 (2)

D2=Pt ∗EP2 (3)

CL12= (D1–D2)∗REarth (4)

The areal convergence, A12, over time (1t= ti− ti−1) is given by

A12=|cos(D2)−cos(D1)|�G (5)

where G= π/180R2Earth, and � is the rotation angle about Pt. The

length-weighted mean convergence rate is then

R12=A12/CL12/1t (6)

where 1t is the time interval between constraining rotations.The cumulative area (C(t)), volume (V (t)), and mass (m(t))convergence as a function of age (t) are

C (t)=t∑0

A12 (7)

NATURE GEOSCIENCE | VOL 9 | NOVEMBER 2016 | www.nature.com/naturegeoscience

© Macmillan Publishers Limited . All rights reserved

849

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2806

V (t)=t∑0

A12× z̄ (8)

m(t)=t∑0

A12× z̄× ρ̄ (9)

where z is the area-weightedmean crustal thickness and ρ is density(using 2,800 kgm−3 as a mean crustal density). We treat theseintegral quantities as summations to reflect the discrete nature ofthe rotation data.

The only inputs required are the convergence zone end pointlocations (EP1 and EP2) as a function of age (SupplementaryTable 3) and corresponding rotations. As both the end pointsand rotations have associated uncertainties, we use a bootstrap toestimate uncertainties resulting from these quantities (see below).We treat the age of collision and interval rotation uncertainties asGaussian-distributed estimates. We allow suture and convergencezone lengths to vary up to their maximum length. Plate convergencedata between India and Eurasia are based on a compilation ofglobal rotations with associated 95% confidences (SupplementaryTables 1–3). Random lengths of the Asia+KLA to India suture(up to 1,000 km) and eastern syntaxis sutures are subtracted fromthe total length of the suture zone such that suture zone lengthgrows to its total length by the age of elimination of interveningoceanic lithosphere at the end of suturing. This end is defined bythe Gaussian-distributed age specified at the start of each iteration.For ages younger than this, the end points are maintained at the fullconvergence zone length.

To summarize, the mass-balance calculations use collision agesof 51 ± 1.5Ma (1σ ) along the <1,000 km western end of thesuture10,11, 56± 2Ma along the main Indus-Yarlung Suture22 (IYS)from∼79◦ E to∼95◦ E, and 40± 4Ma along the≤500 km suturein the far east19.

Total converged area is constrained in our reconstruction,as is the trajectory of motion and associated rotation-relateduncertainties (Fig. 1). We use a 50,000 iteration bootstrap approachto integrate area flux along the IYS, incorporating uncertainty in theage of suturing, length of sutures and rotations as a function of age.

The resulting best estimate of the total areal convergenceof continental crust since continent–continent collision began is9.1+2.4/−2.0 × 106 km2 (2σ ) using plate reconstructions and thetiming of juxtaposition of Greater Indian and Greater Tibetancrust (Fig. 2)). Figure 2b,c summarizes local and areal convergence,incorporating both age- and rotation-related uncertainties.

Assignment of crustal thicknessesWe employ a second bootstrap to estimate the correspondingvolume and mass convergence. We input the total area convergencewith associated uncertainty. We apportion the collisional area intofour crustal thickness domains: Indian coastal plain, Tethyan shelf,Linzizong Arc, and remaining deformed Asia. We use Crust1.0’s23relationship between crustal thickness and elevation, averaged atthe same spatial resolution as the crustal thickness grid, to assigncrustal thicknesses and associated standard deviations to each of theelements in this analysis (Fig. 2).We exclude areas where the currentcrustal thickness is less than the estimated pre-collisional averagecrustal thickness.

Greater India represents ∼50% of the total convergence. Wetherefore assign 25 ± 5% (1σ ) of the converged area to each theIndian coastal plain and Tethyan shelf, which comprise GreaterIndia (Fig. 1c). There is no preserved evidence of a Cretaceous orearly Cenozoic northern source of sediments in the LesserHimalayaprior to the Eocene Subathu and Bhainskati, and clear evidencefor a southern Indian cratonic source in the early Paleocene JidulaFormation of the Tethyan Himalayas17,24. We therefore assign acrustal thickness of 35 ± 5 km (1σ ) to the Indian coastal plain

assuming an average elevation of 0m, and 31 ± 5 (1σ ) km tothe Tethyan shelf assuming an average elevation of−250m. Theseare conservative elevation estimates and therefore conservativecrustal thicknesses.

We use recent palaeoelevation estimates of 4,100 ± 550m(refs 25,26) and the areal extent of Gangdese granitoids andLinzizong volcanics at ∼55Ma to assign a crustal thickness of61± 6 km to the LinzizongArc. The present LinzizongArc occupies∼4 ± 2% (1σ ). The bootstrap allows the areal extent of highelevations in the Linzizong Arc to be: all; a small fraction (but≥0.5%); or a much broader area than implied by preserved outcropof the Linzizong magmatism. The remaining ∼3.85 × 106 km2,which we refer to as Deformed Asia, presumably had a broaddistribution of elevations reflecting the complex prior tectonichistory. We ascribe a mean elevation of 500 ± 500m to DeformedAsia with a crustal thickness of 38 ± 4 km (Fig. 2a) becausesignificant areas had been at or below sea level in the Cretaceous27.This elevation and crustal thickness range encompasses the meanelevation of areas above sea level and mean crustal thickness of thecontinental crust as a whole.

We assume Gaussian fractional area and crustal thicknessuncertainties in this bootstrap. We then solve for volume andmass by summing the crustal thickness (±uncertainty) multipliedby fractional area (±uncertainty) multiplied by area convergence(±uncertainty). The resultant total volume and mass convergencesare 3.3+1.0/−0.8×108 km3 (±2σ ) and 915+290/−237×103 Gt (±2σ ),respectively. These correspond with a mean pre-collisional crustalthickness over the entire domain of ∼36 km, ∼21 km less thancurrently within the deformed region.

Contributions of the mass-balance componentsExcess crustal thickness. We assess the current excess crustal massof the India–Asia system using two data sets of crustal thickness,Crust1.023 (Fig. 2) and that used by Yakovlev and Clark5, appliedto two different area domains of potentially thickened crust duringIndia–Asia convergence and the area of southeastAsia thatmayhavebeen transported laterally via tectonic escape (Fig. 1). We performall of the calculations using both domains and crustal thickness datasets to ensure that our results are insensitive to both domain size ofpotential crustal thickening and the crustal thickness distribution.We incorporate a ±5 km uncertainty in estimates of the presentcrustal thickness25.

Excess crustal mass is defined as:

Mxs=ρc

∫A(ztoday− zCollisionOnset)dA (10)

where z is crustal thickness, A is the area of potential crustalthickening (Fig. 1b), and ρc is mean crustal density. The estimateof excess crustal mass represented by the current distribution ofcrustal thickness, independent of the crustal thickness source anddefinition of the domain, is ∼280 ± 100 × 103 Gt or about30± 10% of the mean pre-collisional crustal mass.

Erosional mass flux. We use a volume of 5.0 ± 1.0 × 107 km3

(±1σ ) for the erosional mass transported out of the collisiondomain, or a mass of ∼140 ± 28 × 103 Gt (±1σ ). This volumeis comparable to two previous estimates of the erosional rockequivalent volume flux. Richter et al.3 estimated erosional lossto be ∼5 × 107 km3. Metevier et al.8 estimated removal of∼3± 1× 107 km3 rock equivalent volume, but their estimatedid not include the Makran (>1.1 × 107 km3), Indo-Burman-Andaman-Sumatran accretionary complex (∼2.5× 106 km3), orthe associated dissolved flux (∼1.0× 107 km3)3. Our erosionalmassflux is nearly three times greater than that assumed by Yakovlevand Clark5.

850

© Macmillan Publishers Limited . All rights reserved

NATURE GEOSCIENCE | VOL 9 | NOVEMBER 2016 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2806 ARTICLES

0

1,000

2,000

3,000

4,000

5,000

0102030405060

Indi

a−A

sia

cum

ulat

ive

conv

erge

nce

(km

)

Age (Ma) (CK′95)

Age (Ma) (CK′95)

−6

−4

−2

0

2

4

6

0 20 40 60 80

Elev

atio

n (k

m)

Below sea level

Crustal thickness (km)

a

b

c

Deformed AsiaTethyan Shelf

Indiancoastal plain

Gangdese Arc

0

200

400

600

800

1,000

1,200

0

2

4

6

8

10

12

0102030405060

Mass (in Et)

Are

a co

nver

genc

e (m

illio

n km

2 )

Figure 2 | Continental thickness and mass convergence. a, Relationship of continental crustal thickness to elevation derived from CRUST1.0 (ref. 23) with1σ thickness standard deviations. b, Mean (dark line) cumulative India–Asia convergence (1σ ) as a function of time together with the east (upper grey line)and west (lower grey line) ends of the Indus-Yarlung Suture. c, Areal convergence with 1σ uncertainties and equivalent mass convergence (withoutuncertainties) associated with India–Eurasia relative motions incorporating only rotation- and age-related uncertainties.

Tectonic escape. We define escape as the transport of the entirecrustal column from within the Deformed Asia domain to alocation outside that domain3. We use the boundaries shownin Fig. 1b of both the deformed region and the region ofpotentially escaped crust. We compute the volume and mass ofthe potentially escaped region as a function of latitude because themass transport is largely north to south. We calculate displacementalong the bounding faults using both CRUST1.0 (ref. 23) andcrustal thickness from Yakovlev and Clark5, separately. We usethe estimated 750± 250 km displacement28 to derive an escapedvolume and mass of ∼1.55± 0.85× 107 km3 and 54 ± 27 × 103Gt (±1σ ), respectively.

Half of pre-collisional continental crust absent from Earthsurface. By this accounting, the currently reconstructable excessmass of the Himalayan orogenic system is 465+192/−176× 103 Gt or∼51% of the syn-collisional mass. This leaves∼450+338/−308×103Gt, or essentially 49+37/−33% (2σ ) of the total syn-collisional mass ofGreater India and Greater Tibet unaccounted for. Previous attemptsto make this calculation have relied on shortening estimatesderived from surface structures relative to convergence4, area5 andvolume3. We have improved this body of work by using rigoroustectonic constraints on the collisional system and uncertainties, themost recent age estimates for collision along the IYS, and recentestimates of pre-collisional palaeoelevation and corresponding

crustal thickness along Eurasia’s southern margin. We concludethat the system is not in mass balance and that essentially half ofthe collisional crustal mass is missing. Wholesale subduction ofcontinental crust into the mantle appears to be the only reasonableexplanation for this missing mass.

Subduction of continental crust. These results imply large-scalesubduction of continental crust into the mantle. This is expectedwhere the lower continental crust is largely transformed to eclogiteduring underthrusting (Fig. 3). Detailed tomography29 implies theexistence of ongoing subduction of seismically fast material into theupper mantle today. Eclogitized lower crust beneath southern Tibethas been inferred both seismically30,31 and from fitting of gravityanomalies32, and it is expected from petrological modelling32,33.Current estimates indicate that eclogitized lower crust extends∼150–200 km in a north–south direction and is ∼19 km thick30,31in the HiCLIMB section. This material would represent 4–5 Myrof residence time at modern convergence rates of ∼40 kmMyr−1.When combined with both shallow and deeper mantle tomographicevidence for subduction of mantle lithosphere, we surmise that thiseclogitized lower crust is being recycled into the mantle. We inferthat this has been a general process that has subducted lower crustinto the mantle throughout this collision.

Existing images29 imply a discontinuous ‘deblobbing’ of litho-spheric mantle and eclogitic crust rather than a continuous tabular

NATURE GEOSCIENCE | VOL 9 | NOVEMBER 2016 | www.nature.com/naturegeoscience

© Macmillan Publishers Limited . All rights reserved

851

ARTICLES NATURE GEOSCIENCE DOI: 10.1038/NGEO2806

Distance from MFT (km)0

Indian upper crust

Himalayanorogenic prism

Indian lower crust

0MFT

Penbo

c 27°

38

28° 29°

25

30° 31°

20

32° 33° 34° N

a

b

d

e50

Platevelocity

at surface(mm yr−1)

100

Dep

th (k

m)

150

200100

Mantlelithosphere

Uncertain/unimaged moho

Rheologically weak zone

200 300 400

Eclogite window

Lowercrustal

subduction

500 600

Eclogite/mafic granulite

Tibetan crust

Mantlelithosphere

700

1050

Altitude (km

)

Moho

Indus-Yarlung Suture

Figure 3 | Schematic tectonic cross-section of India, the Himalayan orogenic prism north of the Main Frontal Thrust (MFT) and southern Tibet duringlower crustal subduction30. a, Placement of the dipping and flattening Indian lower crust and Moho is based on receiver functions30. Seismic doubletrepresents the seismic Moho and the eclogite transition zone (‘eclogite window’). b, Li et al.29 image a steeply dipping channel of seismically fast material(interpreted as subducting mantle lithosphere and eclogitized lower crust) in the upper mantle beneath southern Tibet. c, The relative length of arrowsrepresents plate velocity using surface GPS data from an 85◦ E transect34. d, Crust–mantle boundary in the subduction zone (31◦–33◦ N) is poorly resolved(hashed area). e, A mid-crustal, rheologically weak zone (wavy pattern) allows northward motion of upper crust, while the eclogite/mafic granulite Tibetanlower crust remains stationary relative to fixed Eurasia.

structure typical of subduction zones. Existing interpretations mapmost of the eclogite to Indian lower crust31, but we note that approx-imately half of the present India–Eurasia convergence is accommo-dated north of 32◦ N34 (Fig. 3) requiring detachment of Tibetanmidand upper crust from underlying mantle lithosphere and probablyeclogitic lower crust. Subduction of the lower Tibetan crust probablycontributes to crustal mass loss (Fig. 3).

We infer that loss of eclogitized crust has been ongoing sincethe beginning of collision and is responsible for the missingcrustal mass. The presence of seismically fast material below Indiaattributed to the ‘Tethyan’ slab35,36, and particularly with depthsshallower than 1,300 km, could well represent eclogitized crust andmantle lithosphere derived from both India and Tibet, assuminga mean radial sinking rate of ∼30 kmMyr−1 and ∼3,900 kmof post-56Ma convergence. The eastward continuation of theseseismically fast anomalies extending beneath southeast Asia36 wouldrepresent the along-strike oceanic lithosphere continuation of thismaterial.

To gain some appreciation for the magnitude of the missingcrustal mass, we compare it with the total mass flux of subductedoceanic crust over the same 56 Ma history. The annual oceaniccrustal volume flux is about ∼20.1 km3 yr−1 at the current oceanicsubduction rate of ∼3.1 km2 yr−1 and with a mean oceanic crustalthickness of ∼6.5 km. This flux corresponds to ∼56Gt yr−1. Ifwe assume a constant rate of collision-related mass loss from theIndia–Asia collision, the resulting annual rate of continental masssubduction would be about 7.6Gt yr−1, or∼14% of the total oceancrust subduction mass flux.

Crustal mass loss. Lee37 discusses the petrological consequencesof lower crustal subduction with small-scale examples relative towhat we present here. Although not derived from the India–Asiacollision, geochemical signatures of lower crustal eclogite have beenrecognized in mid-ocean ridge basalts along the Southeast IndianRidge south of Australia38, and possibly in ocean island basalts39 aswell. Previous large-scale collisions, such as the Variscan of centralEurope40, the Pan-African41, and the Grenville42, may have beenresponsible for subduction of lower crust and associated mantlecontamination, as seen in various radiogenic systems39. The scale ofthis phenomenon in the Himalaya–Tibet orogen is compatible withtwo-stage models for continental crustal evolution, which include

significant loss of a more mafic lower crust29. We propose thatsubduction of lower continental crust andmantle lithosphere duringcollision-related convergencemay serve as amore general model forcrustal recycling during continent–continent collisions throughoutEarth’s history, producing radiogenic isotope anomalies in mantle-derived magmas, and influencing the long-term evolution of thecontinental crust.

Data availability. The data that support the findings of this studyare available within the publication and/or from the correspondingauthors on request.

Received 18 April 2016; accepted 8 August 2016;published online 19 September 2016

References1. Dewey, J. F. & Bird, J. M. Mountain belts and the new global tectonics.

J. Geophys. Res. 75, 2625–2647 (1970).2. McKenzie, D. P. Speculations on the consequences and causes of plate motions.

Geophys. J. R. Astron. Soc. 18, 1–32 (1969).3. Richter, F., Rowley, D. & DePaolo, D. Sr isotope evolution of seawater: the role

of tectonics. Earth Planet. Sci. Lett. 109, 11–23 (1992).4. van Hinsbergen, D. J. J. et al . Restoration of Cenozoic deformation in Asia and

the size of Greater India. Tectonics 30, TC5003 (2011).5. Yakovlev, P. & Clark, M. Conservation and redistribution of crust during the

Indo-Asian collision. Tectonics 33, 1016–1027 (2014).6. England, P. & Houseman, G. Finite strain calculations of continental

deformation: 2. Comparison with the India–Asia collision zone. J. Geophys. Res.91, 3664–3676 (1986).

7. Clift, P. D. Controls on the erosion of Cenozoic Asia and the flux of clasticsediment to the ocean. Earth Planet. Sci. Lett. 241, 571–580 (2006).

8. Métivier, F., Gaudemer, Y., Tapponnier, P. & Klein, M. Mass accumulation ratesin Asia during the Cenozoic. Geophys. J. Int. 137, 280–318 (2002).

9. Tapponnier, P., Peltzer, G., Le Dain, A. Y., Armijo, R. & Cobbold, P. Propagatingextrusion tectonics in Asia: new insights from simple experiments withplasticine. Geology 10, 611–616 (1982).

10. Gaetani, M. & Garzanti, E. Multicyclic history of the northern Indiacontinental margin (northwestern Himalaya) (1). Am. Assoc. Petrol. Geol. Bull.75, 1427–1446 (1991).

11. Green, O. R., Searle, M. P., Corfield, R. I. & Corfield, R. M. Cretaceous-Tertiarycarbonate platform evolution and the age of the India–Asia collision along theLadakh Himalaya (northwest India). J. Geol. 116, 331–353 (2008).

12. Roddaz, M. et al . Provenance of Cenozoic sedimentary rocks from theSulaiman fold and thrust belt, Pakistan: implications for the palaeogeographyof the Indus drainage system. J. Geol. Soc. Lond. 168, 499–516 (2011).

852

© Macmillan Publishers Limited . All rights reserved

NATURE GEOSCIENCE | VOL 9 | NOVEMBER 2016 | www.nature.com/naturegeoscience

NATURE GEOSCIENCE DOI: 10.1038/NGEO2806 ARTICLES13. Wilke, F. D. H. et al . The multistage exhumation history of the Kaghan Valley

UHP series, NW Himalaya, Pakistan from U-Pb and 40Ar/39Ar ages. Eur. J.Mineral. 22, 703–719 (2010).

14. Leech, M., Singh, S., Jain, A., Klemperer, S. & Manickavasagam, R. The onset ofIndia–Asia continental collision: early, steep subduction required by the timingof UHP metamorphism in the western Himalaya. Earth Planet. Sci. Lett. 234,83–97 (2005).

15. St-Onge, M. R., Rayner, N., Palin, R. M., Searle, M. P. & Waters, D. J. Integratedpressure-temperature-time constraints for the Tso Morari dome (northwestIndia): implications for the burial and exhumation path of UHP units in thewestern Himalaya. J. Metamorph. Geol. 31, 469–504 (2013).

16. Bouilhol, P., Jagoutz, O., Hanchar, J. M. & Dudas, F. O. Dating theIndia–Eurasia collision through arc magmatic records. Earth Planet. Sci. Lett.366, 163–175 (2013).

17. DeCelles, P., Kapp, P., Gehrels, G. & Ding, L. Paleocene-Eocene foreland basinevolution in the Himalaya of southern Tibet and Nepal: implications for the ageof initial India–Asia collision. Tectonics 33, 824–849 (2014).

18. Rowley, D. Age of initiation of collision between India and Asia: a review ofstratigraphic data. Earth Planet. Sci. Lett. 145, 1–13 (1996).

19. Aikman, A. B., Harrison, T. M. & Lin, D. Evidence for early (>44Ma)Himalayan crustal thickening, Tethyan Himalaya, southeastern Tibet. EarthPlanet. Sci. Lett. 274, 14–23 (2008).

20. Argand, E. Proc. XIIIth Int. Geol. Congr. vol. 1, pt. 5, 181–372(Hafner Press, 1924).

21. van Hinsbergen, D. J. J. et al . Greater India Basin hypothesis and a two-stageCenozoic collision between India and Asia. Proc. Natl Acad. Sci. USA 109,7659–7664 (2012).

22. Orme, D. A., Carrapa, B. & Kapp, P. Sedimentology, provenance andgeochronology of the upper Cretaceous-lower Eocene western Xigaze forearcbasin, southern Tibet. Basin Res. 27, 1–25 (2014).

23. Laske, G., Masters, G., Ma, Z. & Pasyanos, M. Update on CRUST1.0 -A 1-degree global model of Earth’s crust. European Geosciences Union(Geophysical Research Abstracts, 2013).

24. Zhu, B., Kidd, W. S. F., Rowley, D. B., Currie, B. S. & Shafique, N. Age ofinitiation of the India–Asia collision in the East-Central Himalaya. J. Geol. 113,265–285 (2005).

25. Ding, L. et al . The Andean-type Gangdese Mountains: paleoelevation recordfrom the Paleocene–Eocene Linzhou basin. Earth Planet. Sci. Lett. 392,250–264 (2014).

26. Ingalls, M., Rowley, D. & Colman, A. Paleocene-Eocene Lhasaplanopaleoaltimetry: implications for mass balance in the India–Asia collision.Goldschmidt Conf. (Cambridge Publications, 2015).

27. Zhang, K. J. Cretaceous palaeogeography of Tibet and adjacent areas (China):tectonic implications. Cretac. Res. 21, 23–33 (2000).

28. Leloup, P. H. et al . The Ailao Shan-Red River shear zone (Yunnan, China),tertiary transform boundary of Indochina. Tectonophysics 251, 3–84 (1995).

29. Li, C., van der Hilst, R. D., Meltzer, A. S. & Engdahl, E. R. Subduction of theIndian lithosphere beneath the Tibetan Plateau and Burma. Earth Planet. Sci.Lett. 274, 157–168 (2008).

30. Nabelek, J. et al . Underplating in the Himalaya–Tibet collision zone revealed bythe Hi-CLIMB experiment. Science 325, 1371–1374 (2009).

31. Wittlinger, G., Farra, V., Hetenyi, G., Vergne, J. & Nabelek, J. Seismic velocitiesin southern Tibet lower crust: a receiver function approach for eclogitedetection. Geophys. J. Int. 177, 1037–1049 (2009).

32. Hetenyi, G. et al . Density distribution of the India plate beneath the TibetanPlateau: geophysical and petrological constraints on the kinetics oflower-crustal eclogitization. Earth Planet. Sci. Lett. 264, 226–244 (2007).

33. Bollinger, L., Henry, P. & Avouac, J.-P. Mountain building in the NepalHimalaya: thermal and kinematic model. Earth Planet. Sci. Lett. 244,58–71 (2006).

34. Zhang, P.-Z. et al . Continuous deformation of the Tibetan Plateau from globalpositioning system data. Geology 32, 809–812 (2004).

35. Hafkenscheid, E., Wortel, M. J. R. & Spakman, W. Subduction history of theTethyan region derived from seismic tomography and tectonic reconstructions.J. Geophys. Res. 111, B08401 (2006).

36. van der Voo, R., Spakman, W. & Bijwaard, H. Tethyan subducted slabs underIndia. Earth Planet. Sci. Lett. 171, 7–20 (1999).

37. Lee, C. in Treatise on Geochemistry 2nd edn (eds Holland, H. & Turekian, K.)423–456 (Elsevier, 2014).

38. Hanan, B. B. et al . Pb and Hf isotope variations along the Southeast IndianRidge and the dynamic distribution of MORB source domains in the uppermantle. Earth Planet. Sci. Lett. 375, 196–208 (2013).

39. Hofmann, A. W. in Treatise on Geochemistry 2nd edn (eds Holland, H. &Turekian, K.) 67–101 (Elsevier, 2014).

40. O’ Brien, P. J. in Orogenic Processes: Quantification and Modelling in theVariscan Belt (eds Franke, W., Haak, V., Oncken, O. & Tanner, D.) 369–386(Geological Society Special Publications 179, The Geological Society, 2000).

41. Fritz, H. et al . Orogen styles in the East African Orogen: a review of theNeoproterozoic to Cambrian tectonic evolution. J. Afr. Earth Sci. 86,65–106 (2013).

42. Dewey, J. F. & Burke, K. C. A. Tibetan, Variscan, and Precambrianbasement reactivation: products of continental collision. J. Geol. 81,683–692 (1973).

AcknowledgementsWe thank P. Yakovlev for generously sharing both crustal thickness and domain polygonboundary with us. Fieldwork was funded by the National Science Foundation Grant EARno. 0609782 and EAR no. 1111274 awarded to D.B.R. and EAR no. 0609756 awarded toB.C. Supporting geochemical analyses were enabled by NSF EAR no. 0923831 awarded toA.S.C. and D.B.R.

Author contributionsD.B.R. and M.I. completed mass-balance calculations. M.I., D.B.R., B.C. and A.S.C.contributed to the manuscript.

Additional informationSupplementary information is available in the online version of the paper. Reprints andpermissions information is available online at www.nature.com/reprints.Correspondence and requests for materials should be addressed to M.I. or D.B.R.

Competing financial interestsThe authors declare no competing financial interests.

NATURE GEOSCIENCE | VOL 9 | NOVEMBER 2016 | www.nature.com/naturegeoscience

© Macmillan Publishers Limited . All rights reserved

853