Pliocene-Quaternary crustal melting in central and northern Tibet and insights into crustal flow

There is considerable controversy over the nature of geophysically recognized low-velocity–high-conductivity zones (LV–HCZs) within the Tibetan crust, and their role in models for the development of the Tibetan Plateau. Here we report petrological and geochemical data on magmas erupted 4.7–0.3 Myr ago in central and northern Tibet, demonstrating that they were generated by partial melting of crustal rocks at temperatures of 700–1,050 °C and pressures of 0.5–1.5 GPa. Thus Pliocene-Quaternary melting of crustal rocks occurred at depths of 15–50 km in areas where the LV–HCZs have been recognized. This provides new petrological evidence that the LV–HCZs are sources of partial melt. It is inferred that crustal melting played a key role in triggering crustal weakening and outward crustal flow in the expansion of the Tibetan Plateau.

T he Tibetan Plateau is an area of anomalously thick (B50-90 km) continental crust and it is the highest and largest topographic feature on Earth [1][2][3] . Three main mechanisms have been proposed to account for crustal thickening and the development of the high topography: thinning of thickened mantle lithosphere 4,5 , intracontinental subduction 2,6 and crustal (channel) flow [7][8][9][10][11][12] . The dispute stems partially from the sparse data on the thermal evolution of the Tibetan deep crust and mantle lithosphere, and inconclusive interpretations of the geophysically determined low-velocity zones (LVZs) and high-conductivity zones (HCZs) within the Tibetan crust. These LV-HCZs have been interpreted as weak layers within the crust that resulted from sub-horizontal orientation of mica crystals in a matrix of isotropic crystals 13 , the presence of mantle-derived melts 14 , aqueous fluid 15,16 , crustal shear zones 2 , and the presence of melt derived from within the zones 17 . In the crustal channel flow model, the LV-HCZs are considered to be partially molten layers within the crust based on geological evidence of anatectic melts 7,18 , numerical models 11,19 , and magnetotelluric (MT) and seismic models 1,8,[20][21][22] . The diversity of models involving the LV-HCZs highlights that resolving their nature and origin requires petrological evidence on samples from the deep crust.
Volcanic rocks and entrained xenoliths derived from the deep crust or the upper mantle provide important information about the thermal regime of the Tibetan crust. It has been argued that the Tibetan crust is too dry to trigger crustal melting, based on granulite xenoliths entrained in Cenozoic volcanic rocks of central Tibet and a limited seismic wavespeed data set from central Tibet 23 . This caused some studies to call into question the role of crustal flow in the growth of the Tibetan Plateau 2,24 , and especially for the central and northern Tibetan Plateau 18 . However, other studies suggested that crustal melting played a key role in triggering crustal weakening and flow (refs 1,8,11,20,25,26).
Here we report on 4.7-0.3 Myr ago old felsic volcanic rocks from the Qiangtang, the Songpan-Ganzi, and the Central Kunlun Blocks in central and northern Tibet, some of which contain granulite xenoliths. The volcanic rocks were predominantly generated by partial melting of mid-to-lower crustal rocks at temperatures of 700-1,050°C and depths of 15-50 km, and as such they provide important new petrological evidence for the nature of LV-HCZs within the crust.
Cenozoic felsic volcanic lavas in central and northern Tibet. Cenozoic volcanic rocks occur widely on the Tibetan Plateau 30 .
Ages. Zircon U-Pb, and whole-rock and biotite 40 Ar- 39 Ar and K-Ar ages indicate that the Pliocene-Quaternary lavas reported here were generated at 4.7-0.3 Myr ago ( Fig. 1b (Figs 2 and 4a).
Geochemistry. The Pliocene-Quaternary felsic lavas from central and northern Tibet plot in the fields of trachyandesite, dacite and rhyolite in SiO 2 versus (K 2 O þ Na 2 O) diagrams (Fig. 5a). Except for some more evolved rocks that plot in middle-and low-K calcalkaline fields, the Pliocene-Quaternary felsic lavas in central and northern Tibet are predominantly shoshonitic/high-K calc-alkaline (Fig. 5b). They have high SiO 2 (58-76 wt.%), low MgO  Table 2 (Fig. 5a,b) and light rare-earth-element contents, and enriched Sr-Nd isotope compositions. Hacker et al. 23 pointed out that if the lower crust beneath Tibet is partly metasedimentary, it is likely to have interacted with hot mantle-generated melts. Indications of in situ melting of biotite, feldspar, and quartz in the metasedimentary xenoliths, and of resorption of biotite, feldspar and quartz in the volcanic rocks hosting the xenoliths, imply that partial melts of xenolith material or other metasedimentary rocks did contribute to the mantle-derived magmas 23 . The widespread occurrence of fine-grained, undigested xenocrysts suggests that the unusual chemical patterns of some Tibetan lavas might be a mixture of lower Tibetan crustal fragments with mantle-derived melt 23 . However, we suggest that models of partial melting of enriched mantle 5 , and of mixing between crust-and mantlederived magmas or crustal assimilation of mantle-derived magmas 23 cannot account for the formation of the Pliocene-Quaternary felsic lavas reported here. First, the Pliocene-Quaternary felsic lavas reported here are not potassic-ultrapotassic lavas, which typically contain clinopyroxene, only rare biotite and no amphibole 5,31 . Potassic rocks should have  Fig. 5c) of the Pliocene-Quaternary felsic lavas, and the absence of contemporary mantle-derived potassic-ultrapotassic magmatic rocks in the area (Fig. 1b), suggest that these lavas were not directly derived from the upper mantle and do not reflect crustal assimilation of mantle-derived magmas. Miocene (18-13 Myr ago) potassic-ultrapotassic mafic rocks are exposed in the Songpan-Ganzi and the Central Kunlun Blocks (Fig. 1b), and some Pliocene-Quaternary (5.0-0.07 Myr ago) potassic-ultrapotassic mafic rocks occur in western Kunlun area 5,30,31 , highlighting the absence of contemporary mantlederived potassic-ultrapotassic magmatic rocks in the studied area. The lavas in the Dongyue Lake and Wulanwula areas contain minor small fine-grained granulite xenoliths or xenocrysts 23 , indicating that they might be mixtures of lower Tibetan crustal fragments with mantle-derived melt 23 . However, these small finegrained granulite xenoliths or xenocrysts may also be residues after high-temperature crustal melting 38 although this needs further work. The groundmass glass of the Dongyue dacites has 70-75 wt.% SiO 2 and o0.5 wt.% FeO þ MgO 23 , and as a whole the lavas in the Dongyue Lake and Wulanwula areas exhibit dacitic or trachyandesitic compositions (Fig. 5a), which are markedly different from the mantle-derived northern Tibetan potassic-ultrapotassic lavas with lower SiO 2 and higher Mg# values (Fig. 5c).
Third, the low-Mg# values, whole-rock Nd-Sr isotope and trace element compositions of the Pliocene-Quaternary felsic lavas support a crustal melting model. Their low-Mg# values are similar to those of experimental melts of metabasaltic rocks or eclogites (Fig. 5c). Their whole-rock Nd-Sr isotope ratios are similar to those of other crustal samples (granulite and amphibolite xenoliths from Cenozoic magmatic rocks) in the area (Fig. 5f) and the high-Th/La ratios of the rhyolites indicate a significant sediment contribution from their source rocks 32,36,39 . The central-northern Qiangtang lavas have Sr-Nd isotope compositions similar to garnet-bearing mafic granulite and amphibolite xenoliths from B28 Myr ago intrusive rocks in the Hoh Xil area of the Songpan-Ganzi Block (Fig. 5f), and low-Th/ La ratios (Fig. 5d), suggesting crustal source rocks with a greater contribution of mafic material 36 .
The mineral assemblages in the crustal source rocks of the Pliocene-Quaternary felsic rocks can be further constrained by their geochemical characteristics. Given that plagioclase is strongly enriched in Sr and Eu, and garnet is strongly depleted in LREEs and enriched in HREEs and Y, the distinct negative Sr and Eu anomalies, and the high-La/Yb and low-Sr/Y ratios of most of the Pliocene-Quaternary lavas (except for a few adakitic rocks) ( Fig. 5e; Supplementary Fig. 2; Supplementary Data set 1), reflect the presence of residual plagioclase and garnet in their sources 36,40,41 . Moreover, given that rutile is strongly enriched in Nb, negative Nb anomalies in felsic adakitic rocks commonly indicate crustal melts derived from eclogitic rocks in the stability field of rutile 42 . Thus, the samples from the Henglianghu and Zhaixinshan areas with adakitic characteristics (small variable Eu and Sr and negative Nb anomalies, and high-La/Yb and -Sr/Y ratios and low Yb and Y) ( Fig. 5e; Supplementary Fig. 2; Supplementary Data set 1), may contain a greater contribution from eclogitic rocks in their source regions (Fig. 5e), that is, residual garnet þ rutile and little or no plagioclase [40][41][42][43][44] .  Fig. 4b). The low-T zr (700-844°C) Pliocene-Quaternary lavas from the Songpan-Ganzi-Central Kunlun area appear to have been derived by partial melting of sedimentdominated crustal rocks (they have elevated Th/La, Fig. 5d) in some cases with contributions from eclogitic source rocks (for example, adakitic trachyandesites with low Th/La but high La/Yb and Sr/Y, Fig. 5d,e). In contrast, the high-T zr (896-983°C) lavas from the central-northern Qiangtang and Songpan-Ganzicentral Kunlun areas are thought to reflect higher temperature melting of more mafic crustal source rocks (Figs 4b and 5d). The Henglianghu adakitic rocks from the northern Qiangtang Block have the highest T zr (968-983°C) (Supplementary Data set 1; Fig. 4b), consistent with partial melting of eclogite-facies crustal rocks (Fig. 5d) at4960°C. Pressure-temperature conditions for partial melting of crustal rocks based on experimental data are summarized in Fig. 6a. They indicate that garnet forms at pressures of 0.5-0.6 GPa and temperatures of 750-900°C during partial melting of metasedimentary rocks 36,46,47 , and that the lower limit of garnet stability is 0.5 GPa (Line 9 in Fig. 6a). Plagioclase is a common residual mineral during partial melting of metasedimentary and igneous rocks (such as tonalites and basalts), but it disappears at pressures 41.2-1.5 GPa 40,42,43,48 (Fig. 6a), and the lower limit of rutile stability is typically 1.5 GPa 42 (Fig. 6a). Given the evidence for residual garnet and plagioclase and the zircon saturation temperature data, we interpret the Songpan-Ganzi-central Kunlun rhyolites to have been generated by dehydration partial melting of metasedimentary source rocks in the temperature and pressure ranges of 700-844°C and 0.5-1.2 GPa (Fig. 6a) (corresponding to depths of 15-40 km 36 (Supplementary Data set 1; Fig. 4c)). The more adakitic Zhaixinshan magmas were derived by partial melting of eclogitic rocks with residual garnet þ rutile and little or no plagioclase at temperatures and pressures of 741-825°C and 41.2-1.5 GPa 32,33 (Fig. 6a) (corresponding to depths of 40-50 km (Supplementary Data set 1; Figs 4c and 5e)).
The combination of residual garnet and plagioclase in the source, and the high-T zr of lavas, except for the Henglianghu adakitic rocks from central-northern Qiangtang and Songpan-Ganzi-central Kunlun (Supplementary Data set 1; Fig. 4b), suggests that they were generated by fluid-absent melting of granulite-facies crustal rocks at 4900°C (Fig. 6a). This is consistent with the occurrence of titaniferous magnetite and ilmenite and F-Ti-rich mica in the Dongyue Lake dacites (Supplementary Table 3), suggesting that their magmas were generated in H 2 O-poor and high-temperature (4950°C) conditions given the stability of titaniferous magnetite in extremely H 2 O-poor 49 and F-Ti-rich mica in high-temperature 17,23 (Fig. 6a 23 , possibly indicating young granulite-facies metamorphic and/or melting events before entrainment in the dacitic lavas. Pressure-temperature calculations suggest that the Dongyue Lake granulite xenoliths formed at pressures of 0.8-1.5 GPa and temperatures of 800-1,100°C 23 (Fig. 6b), consistent with fluid-absent melting of metasedimentary and igneous rocks (0.7-1.5 GPa and 900-1,050°C) 36,40,43,[46][47][48] . In summary, most of the high-T zr lavas from the central-northern Qiangtang and Songpan-Ganzi-Central Kunlun areas are attributed to fluidabsent melting of granulite-facies rocks at temperatures and pressures of 900-1,050°C and 0.8-1.5 GPa (Fig. 6a) (corresponding to crustal depths of 26-50 km (Supplementary Data set 1; Fig. 4c). The Henglianghu adakitic rocks in contrast appear to have been derived by partial melting of eclogitic rocks with residual garnet þ rutile and little or no plagioclase at temperatures and pressures of 968-983°C and 41.2-1.5 GPa (Fig. 6a) (corresponding to crustal depths of 40-50 km (Supplementary Data set 1; Fig. 4c)).

Discussion
There is some west-east variation in the thermal state of southern Tibet 50 Fig. 6b. Moreover, the estimated temperatures and pressures for the generation of the Pliocene-Quaternary felsic magmas are consistent with the present crustal geotherms for central and northern Tibet (Fig. 6b), indicating that high temperatures in the mid-lower crust of central and northern Tibet were responsible for the fluidabsent partial melting. These elevated temperatures in central and northern Tibet have been attributed to upwelling asthenosphere in response to lower lithosphere delamination 4,5 , mantle counterflow coupled with the northward downwelling of the Indian mantle lithosphere or the southward downwelling of the Asian mantle lithosphere 1,28 , and to squeezing between the northward advancing Indian and resisting Qaidam and Tarim lithospheres 30,50 . In all these models, heat conducted from the underlying hot lithospheric mantle heated the mid-to-lower crust, which melted to form felsic magmas in central and northern Tibet (Fig. 2). Radioactive isotopes may possibly have been introduced to the mid-to-lower crust beneath central and northern Tibet in sediments during India-Eurasia convergence or pre-Cenozoic subduction 2,6,23 , and they would have provided an additional heat source for crustal melting and the generation of felsic magmas 56,57 . The presence of crust-derived Pliocene-Quaternary (4.7-0.3 Myr ago) felsic rocks in central and northern Tibet (Figs 1b, 2 and 4) provides new evidence as to the nature of the LV-HCZs within the crust beneath Tibet. The LV-HCZs occur at depths of 15-50 km, similar to the depths at which the Pliocene-Quaternary crust-derived magmas were generated (Figs 1b, 2 and 4), and the   felsic magmas are restricted to the areas of the largest negative values ( À 12 to À 6%) for the amplitude of the Vs perturbation associated with the LVZ (Fig. 1a) and the lowest shear wave speeds (V s o3.35 or 3.25 km s À 1 ) 17,22 (Fig. 4c). The MT data also indicate that the lowest resistivities occur in the middle crust beneath the northern Qiangtang Block 14,15,21,26,27 , where the Pliocene-Quaternary crust-derived felsic rocks occur. These results are consistent with experimental results and model calculations of the seismic properties of partially molten rocks that strongly suggest that the electrical and seismic anomalies measured beneath the Tibetan Plateau are best explained by the presence of partially molten rocks 17,26,58,59 .
In this study, simple batch melting models, constrained by estimates of Rb/Sr in the source from Nd and Sr isotopes 60 and REE contents of the Pliocene-Quaternary lavas of central and northern Tibet, indicate that the crustal melts from central and northern Tibet reflect 8-22% partial melts (Supplementary  Tables 4-11). Such degrees of partial melting are those present at the time the volcanic magmas were generated, and the youngest of those is 0.3 Myr ago old. In contrast, the amounts of melt at depth at the present day is constrained geophysically and, based on MT data from Tibet, the estimates for the amounts of melt required to explain the HCZs in Tibet range from 5 to 23% 20,25,26 . Melting and numerical experiments also suggest that the melt fractions required to explain the HCZs in Tibet were 8-23%(refs 58,59).
While such estimates of 8-23% are similar to those from the geochemical data in this study and the MT data in Tibet, both are higher than the melt fractions (B1-5%) suggested by a number of seismic studies 17,61 . In practice, melt fractions estimated from seismic and MT data are not always in agreement, and seismic studies in Tibet have generally given lower estimates of melt fractions compared with those derived from MT studies 26 . Le Pape et al. 26 provided a detailed discussion of what factors might affect these values and how they might be reconciled. Seismic velocity varies by a factor of o3 for the melt range discussed, whereas the resistivity can vary by 3 orders of magnitude 26,62 . Thus, electrical resistivity is more sensitive to the size of the melt fractions.
The inferred melt fraction estimates from the resistivity models are also consistent with values predicted from fluid-absent melting petrological models at the observed P/T conditions and for similar compositions 26  These are in good agreement with those (8-22% and 5-23%) calculated in this study, and estimated from MT and numerical experiment data 20,25,26,58,59 . In Tibet, the lower resistivities in the middle crust beneath the northern Qiangtang Block, indicating where the higher melt fractions occur 27 , are in the area where the Pliocene-Quaternary felsic lavas outcrop (Figs 1, 2 and 4c).
Taken together the geochemistry and MT data indicate that B10-20% partial melt is present today in the high-conductivity zones (LV-HCZs), and that similar degrees of melting were required to generate the volcanic rocks in the period 4.7-0.3 Myr ago. This indicates that conditions in the mid-to-lower crust would have facilitated tectonic movements in this region for at least the last 5 Myr ago.
In summary, three main mechanisms (intracontinental subduction, lithosphere thinning and crustal flow) have been proposed to account for the crustal thickening and high topography in Tibet. The first two emphasize changes in lithospheric mantle structure 2,4-6 ( Fig. 2), and they may have occurred at relatively early stages in the development of the Tibetan Plateau 1,36 , or at the margin of the Tibetan Plateau (for example, crustal brittle thickening in northern part of western Kunlun 63 or Qilian areas 2,6 ( Fig. 1a)). In contrast, crustal flow involves a decoupling of movement in the upper crust from those in the high-temperature mid-to-lower crust, probably during the relatively late heating and eventual melting of the crust in response to the continuous convergence between the Indian and Eurasian plates 1,36 (Fig. 2).
Crustal flow requires a layer with a viscosity less than that of the adjacent rocks and an effective viscosity below an absolute threshold that is dependent on layer thickness 7,11,20 . In a reevaluation of the experimental data, Rosenberg and Handy 64 suggested a melt fraction of B7 vol% as the 'melt connectivity transition', marking the increase of melt-interconnectivity that causes the dramatic strength drop. The mechanical response of crust containing a layer with 8% melt may be very different from one that contains only 2% melt, but it will not differ markedly from that of crust containing a layer with 50% melt, despite their different microstructures and compositions 65 . If only 1-2% and 3-5% melts now occur in the crust of the Qiangtang and Songpan-Ganzi Blocks, as suggested by the seismic data 17,61 , the strength of the mid-lower crust would not have been significantly changed 64 . Alternatively, if there are B10-20% melts in the mid-lower crust of the Qiangtang and Songpan-Ganzi Blocks, as suggested by MT and the new Pliocene-Quaternary felsic lava data, the strength of the mid-lower crust beneath central and northern Tibet will have been markedly changed 59,64 . This in turn would facilitate northward and eastward flow of melt-weakened mid-lower crust of central and northern Tibet 12,25,26,36,66 .
Crustal melt-enhanced ductile flow in the high-temperature, partially molten, mid-to-lower crust makes it easier to maintain a uniform elevation in the Tibetan Plateau 1,19,36 and it accounts for the present expansion and frequent earthquakes along its northern and eastern margins 3,12,25,26,36,66 . Similar examples of crustal melt-enhanced ductile flow are likely to have occurred elsewhere (for example, the modern Andean and Anatolian plateaus), shaping the deep structure of the Earth's continental crust and restraining the thickness and elevation of mountain belts 19,36 .

Methods
Zircon U-Pb age analyses. Zircon U-Pb analyses for sample 5133-1 and samples 11WL59-2 and 11WL60-3 were conducted using the Cameca IMS-1280 SIMS (CASIMS) at the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS) and the State Key Laboratory of Isotope Geochemistry, Guangzhou Institute of Geochemistry, Chinese Academy of Sciences (SKLaBIG GIGCAS), respectively. The O 2 À primary ion beam with an intensity of ca. 10 nA was accelerated at À 13 kV. The ellipsoidal spot is about 20 Â 30 mm in size. The aperture illumination mode (Kohler illumination) was used with a 200 mm primary beam mass filter aperture to produce even sputtering over the entire analysed area. In the secondary ion beam optics, a 60 eV energy window was used, together with a mass resolution of ca. 5,400. Rectangular lenses were activated in the secondary ion optics to increase the transmission at high-mass resolution. A single electron multiplier was used on ion-counting mode to measure secondary ion beam intensities by peak jumping sequence: 196 ( 90 Zr 2  .24 s, 2.08 s, 1.04 s, 2.08 s, 2.08 s, 2.08 s and 0.24 s, respectively. Each measurement consisted of seven cycles, and the total analytical time is ca. 12 min. Calibration of Pb/U ratios is relative to the zircon standard TEMORA 2 (417 Myr ago) based on an observed linear relationship between ln( 206 Pb/ 238 U) and ln( 238 U 16 O 2 / 238 U). U and Th concentrations of unknowns were determined relative to the standard zircon 91500 (1,065 Myr ago) with Th and U concentrations of ca. 29 p.p.m. and 81 p.p.m., respectively. Measured compositions were corrected for common Pb using non-radiogenic 204 Pb. Uncertainties on individual analyses are reported at 1s level; mean ages for pooled U-Pb analyses are quoted at 95% confidence.
Geochemical and mineral composition analyses. Whole-rock geochemical and mineral composition analyses were carried on at the SKLaBIG GIGCAS. Major elements were measured on individual minerals using a JEOL JXA-8100 Superprobe with an accelerating potential of 15 kV and sample current of 20 nA. Wholerock major element oxides (wt.%) for whole-rock powders were determined using a Varian Vista PRO ICP-AES using wavelength X-ray fluorescence spectrometry with analytical errors better than 2%. Whole-rock trace elements, including the REEs, were analysed using a Perkin-Elmer ELAN 6000 inductively-coupled plasma source mass spectrometer (ICP-MS). Analytical precision for most elements is better than 3%. Whole-rock Sr and Nd isotopic compositions of selected samples were determined using a Micromass Isoprobe multi-collector mass spectrometer (MC-ICP-MS). The 87 Sr/ 86 Sr ratio of the NBS987 standard and 143 Nd/ 144 Nd ratio of the Shin Etsu JNdi-1 standard measured were 0.710288 ± 28 (2s) and 0.512109±12 (2s), respectively. All measured 143 Nd/ 144 Nd and 86 Sr/ 88 Sr ratios are fractionation corrected to 146 Nd/ 144 Nd ¼ 0.7219 and 86 Sr/ 88 Sr ¼ 0.1194, respectively.
Partial melting calculations. The calculated degrees of partial melting for the crust-derived magmas are based on simple batch melting models. These are constrained by comparing estimates of Rb/Sr in the source from combined Nd and Sr isotope ratios with the Rb/Sr ratios of the rocks analysed, and the REE contents of the felsic magmatic lavas together with appropriate distribution coefficients for the different minerals. The detailed methods and equations are presented in Supplementary Methods.