Introduction

Rare-earth elements (REEs) are essential for many high-technology applications, new energy industries, and defense systems. Thus, understanding the processes by which they are concentrated and trapped in the crust is currently a focus of investigations. REE deposits come in a wide variety1,2, and carbonatite-associated REE deposits (CARD) are among the most significant types, which supply half of the world’s REE + Y(yttrium)1,3,4 and thus are prime targets for focused exploration5,6.

About 527 carbonatite occurrences are exposed on all of Earth’s continents7,8,9, and 75% of them are emplaced within 600 km of craton edges10. These carbonatites may have been directly erupted from their mantle sources or frequently emplaced after protracted modification within the lithosphere9. They mainly occur as circular to ovoid complexes in continental rifts (e.g., East Africa) and orogenic belts (e.g., southwestern China) but are also associated with large igneous provinces (e.g., Deccan) and/or alkaline provinces (e.g., Greenland and the Kola Peninsula)10,11. A commonly held view suggests that carbonatites originate from an enriched or metasomatized mantle source10, but the relative roles of the lithosphere and asthenosphere in carbonatite formation have been controversial8,12,13,14,15,16. A crucial assessment of these controversies requires a close examination of the underlying mantle structure and its plausible evolution.

Explorations show that only a few carbonatites host giant and large-sized CARDs2,4,5,6,17. These CARDs mainly form in continental rift settings (e.g., Bayun Obo in north China)18 and syn- to post-collisional settings (e.g., Maoniuping in southwest China)17,19,20. The associated magmas are regarded to have been derived from the subcontinental lithospheric mantle (SCLM)21,22, enriched by plume activity or previous subduction2,8,22,23,24. Recent studies indicate that the total resources of such CARDs are controlled by the volume of the parent magmas and the availability of enriched mantle sources2 and that primary REE-poor orogenic carbonatites can be enriched by early processes of metasomatism22. However, how the mantle architecture influences lithosphere focusing and how REEs are fertilized in the metasomatized SCLM and later transported out of the mantle for the formation of CARDs remain unclear. We demonstrate that the key to understanding these issues is deciphering the lithospheric architecture and deep processes beneath REE-rich young orogens at cratonic edges.

The eastern Tibetan orogen, formed by the Cenozoic Indo-Asian collision19, hosts a REE ore system—the Cenozoic Mianning–Dechang (MD) REE belt20 (Fig. 1a). This orogen is an ideal place to investigate the factors controlling CARD formation because of its young mineralization age (27–12 Ma) without later reworking25, a clearly defined post-collisional setting for their occurrence22, and the current lithospheric structure without noticeable modification after the CARD formation. The MD REE belt was developed along the western margin of the Yangtze craton, where the exposed remnants of 1000–740 Ma arc volcanic rocks and granitoids recorded the Neoproterozoic accretion related to subduction26, and Cenozoic reworking by the Indo-Asian collision produced numerous Cenozoic strike-slip fault systems (e.g., Red-River fault [RRF], Xiaojiang fault [XJF], and Xianshuihe fault) in the orogen19 (Fig. 1a). This collision also caused extensive Eocene potassic–ultrapotassic magmatism along the RRF (Fig. 1a)21,27 and formed a NS-trending 200-km-long REE-rich complex belt (i.e., MD REE belt) along the XJF at the cratonic edge20,21. This belt is composed of numerous carbonatite–syenite complexes (30–12 Ma), which host several CARDs, such as the giant Maoniuping (22 Ma), large-sized Dalucao (12 Ma), and medium Lizhuang (27 Ma)20,25, constituting a Cenozoic giant REE ore system along the cratonic margin (Fig. 1a).

Fig. 1: Topographic-geological map and P-wave velocity image of the Yangtze craton and the Tibetan orogen.
figure 1

a shows a tectonic framework of the Tibetan orogen and craton (modified from ref. 19 and ref. 39), in which the Cenozoic CARDs (red hammers), including the giant Maoniuping (22 Ma), the large Dalucao (12 Ma), the medium Lizhuang (27 Ma), and the small Muluozhai (26 Ma)20,25, are shown. Black dotted lines are the boundaries of the Indian block and Yangtze craton. Black lines are the Cenozoic strike-slip faults (KLF: Kara–Kunlun fault; XSHF: Xianshuihe fault; RRF: Red-River fault; XJF: Xiaojiang faults) and thrust fault (LMT: Longenshan fault) with ages labeled. The Cenozoic magmatic rocks are shown as follows: Tengchong basalts (red, 2.7–0.2 Ma), Eocene–Pliocene ultrapotassic (green, 37–30 Ma) and potassic rocks along the RRF (orange, 43–32 Ma), Oligocene-Miocene carbonate-syenite complexes in the MD REE belt (yellow, 30–10 Ma), and Miocene ocean-island basalt (OIB) (mauve triangle, 13–1.7 Ma). Pink star: mantle xenolith locality; yellow stars: crustal xenolith localities. The positions of three seismic profiles across the MD REE belt are shown. b shows a P-wave velocity slice at depth of 160 km in the study area. The domains of mantle-derived He gas in modern springs (purple dotted line) are also shown.

Full size image

Here, we reveal the lithospheric architecture and deep processes of the MD REE belt by integrating a P-wave tomography model with geochemical data over an area covering the Tibetan orogen to the Yangtze craton. Our results provide a framework to probe the genesis of CARDs and their controlling factors, and the resulting genetic model provides a useful reference for the exploration of such CARDs.

Results and discussion

Lithospheric architecture imaged through seismic tomography

Seismic images with different resolutions have revealed the crust- and uppermost-mantle velocity structures in Tibet28,29. In this study, the teleseismic double-difference seismic tomography method was used to image the lithospheric architecture that controls the CARD formation using seismic arrival times from earthquakes within and surrounding the China continent recorded by local, regional, and teleseismic stations (see “Methods” and Supplementary Figs. 1–8).

Our P-wave images clearly show both the Indian and Yangtze blocks as high-Vp anomalies at depths of 160 and 80 km, respectively, with boundaries around longitudes 98–99°E (Fig. 1b), consistent with previous result29. The two high-velocity bodies are contiguous around latitudes 28–30°N (Fig. 1b), indicating that the two blocks collided at this location, thus controlling lithospheric deformation in the orogen. We interpret the high-Vp boundary of the Indian block as the northeastern front of the Indian lithospheric mantle (Fig. 1b)30. There are two large regions of low-velocity anomalies to the north and south of this convergent zone, again consistent with previous results31. These two low-velocity regions may be connected by a low-velocity channel between the Indian and Yangtze blocks at depths of 80–110 km around longitude 99°E, as evidenced by a high-resolution lithosphere Vs model for continental China (Supplementary Fig. 8b)32. Both regions are also spatially correlated with Cenozoic mantle-derived magmatic provinces at the surface (Fig. 1a), high-temperature mantle domains (1200 °C)33, and mantle He in modern hydrothermal reservoirs (Fig. 1b)34.

The three roughly E‒W or NE‒SW profiles show that the subducted Indian lithosphere (high Vp) reached the mantle transition zone, or even deeper, at a high angle with slight flattening in the north (Fig. 2a) and steepening in the south (Fig. 2b). Along the RRF and the adjacent region at 97–101°E, a remarkable feature of the P-wave image is the occurrence of a large mushroom-shaped low-velocity body. Its stem is rooted at depths >400 km, and the head extends laterally beneath the rigid Yangtze mantle lithosphere (Fig. 2b, c), locally stretching upward to the crust–mantle boundary (Fig. 2a). The overlying lithosphere shows obvious velocity heterogeneities. In the south, a large high-velocity body indicates that the orogenic lithosphere is relatively intact and thick (Figs. 1b and 2c). To the north, the cratonic lithosphere becomes thin (~100 km) and disjointed, as shown by two distinct and separated, relatively weak high-velocity domains beneath the Yangtze craton (Fig. 2b). We interpret this large low-velocity body as a large-scale asthenospheric upwelling focused along the block boundaries, thermally eroding and reworking the overlying lithosphere at the cratonic margin.

Fig. 2: Images of the Vp structures along three profiles.
figure 2

a–c represent profiles A-A′, B-B′, and C-C′ in Fig. 1a respectively. Geological interpretation is based on seismic velocity and geological observations. Sources of Cenozoic magma and faults were shown. White dotted lines with arrows refer to upwelling and lateral flow of upwelling asthenosphere. MTZ mantle transition zone, RRF Red-River fault, XJF Xiaojiang fault.

Full size image

Along the western margin of the Yangtze craton at approximately 102°E, there are widely distributed shallow low-velocity anomalies (Fig. 2a, b), coinciding with the MD REE belt around the XJF system35. They occur along a N‒S-striking Cenozoic strike-slip fault system superposed on a Permian rift21 and become weak southward at 26°N, where the MD REE belt disappears (Fig. 2a). These shallow low-velocity anomalies extend down to an ~100-km depth and connect with a large underlying low-velocity body through a tilted low-velocity band, which separates the Yangtze lithosphere into two segments with distinct thicknesses (Fig. 2a, b). This structure is also imaged by receiver function results36, which can be interpreted as representing the eastward lateral flow of the asthenosphere and its subsequent upwelling along the trans-lithospheric faults. The lateral flow of the asthenosphere is also supported by the dominant E‒W fast polarizations of teleseismic SKS waves from shear wave splitting analysis in this area37,38.

New and previously-published geochemical data further constrain the lithospheric architecture imaged by seismic tomography (Supplementary Data 1–3). The explosive ultrapotassic magmas with eruption ages peaking at 35 ± 2 Ma are extensively developed along the RRF (Fig. 1a), and their geochemical signatures suggest a derivation from partial melting of metasomatized SCLM (Supplementary Figs. 9 and 10)27,39 at an ~80-km depth (Fig. 3a). This process is consistent with our seismic images (Fig. 2a, b) and high estimated temperatures of the upper mantle (Fig. 3a), confirming that the cratonic SCLM was disrupted, thinned, and partially melted by the asthenospheric upwelling along the RRF. Coeval potassic granitoids in Dali, showing geochemical affinity with adakites, are considered to have originated from the collision-thickened (45- to 50-km) mafic lower crust in the orogen40 (Fig. 3a). Our seismic images also reveal that the orogenic lower crust was heated by high heat flows from the upwelling asthenosphere, resulting in the upward percolation of small-volume mantle melts (Fig. 2a). Minor Quaternary basalts (13–0.3 Ma) locally developed after the main magmatic event (~35 Ma) along the RRF (Fig. 1a). These basalts contain mantle xenoliths and show geochemical affinity with oceanic island basalts (OIBs)19,39, suggesting decompression melting of the asthenosphere during upwelling since the mid-Miocene. All these observations point to a consistent interpretation—that the steep subduction of the Indian continental lithosphere led to continuous asthenospheric upwelling, thermal erosion, partial melting, and interaction with the overlying cratonic lithosphere since ~35 Ma.

Fig. 3: Source depth and geochemical affinities of the Cenozoic magmatic rocks.
figure 3

a shows the relationship between the source depths for the Cenozoic magmas and the lithospheric architecture and thermal status. The bottom interface of the SCLM (i.e., LAB) is inferred from receiver function results36 and our seismic images. The Moho is from ref. 86. The 1000 °C and 1200 °C adiabats of the upper mantle are estimated based on our seismic data following the method of An and Shi33. Note that there has been no intensive magmatic activity in the western Yangtze craton after ~20 Ma (forming age of MD carbonatites), and the mantle thermal states can be preserved. The melting depth of carbonatitic melts is inferred from available experimental data (see text); others were estimated by bulk-rock La/Yb ratios using the methods of ref. 87. Errors are two standard errors. SCLM: subcontinental lithospheric mantle; LAB: lithosphere/asthenosphere boundary; RRF: Red-River fault; XJF: Xiaojiang fault. b shows Sr–Nd isotopic compositions of the MD carbonatites, mafic rocks and Eocene ultrapotassic rocks at the cratonic edge, suggesting their cratonic SCLM sources metasomatized by fluids and marine sediments. Other carbonatites (e.g., Bayun Obo and Mountain Pass) were shown for comparison. The numbers on the orange (fluid from subducted sediments88) and gray (REE-bearing sediments89,90) curves represent the percentage contents of the endmembers in magma mixing models. c shows systematic variation in La/Yb ratio (other ratios see Supplementary Fig. 10) of the ultrapotassic rocks and the MD carbonatites and mafic rocks, reflecting a lateral increase in the intensity of fluid metasomatism toward the interior of the Yangtze craton. d shows the higher Ba/Th of the MD carbonatites, requiring the involvement of CO2-rich fluids derived from recycled sediments and REE-rich sediments. Data from Supplementary Data 1–3 and ref. 21. The numbers on the orange, blue and gray curves represent the percentage contents of the endmembers in magma mixing models. In the mixing calculation, assuming that element A mixes with element B, the binary hybrid values are given by (Ba/Th)A(X%)—B(1-X%) = [ThA*X%*(Ba/Th)A + ThB*(1-X%)*(Ba/Th)B]/[ThA*X%+ThB*(1-X%)] and [(87Sr/86Sr)i]A(X%)—B(1-X%) = [SrA*X%*[(87Sr/86Sr)i]A + SrB*(1-X%) * [(87Sr/86Sr)i]B]/[SrA*X%+SrB*(1-X%)].

Full size image

The partial melting of the SCLM focused at 35 Ma is not limited to the RRF but also extends eastward across the RRF, at times ranging from 35 to 12 Ma (Fig. 2a–c). The eastward lateral flow is supported by the eastward prolongation of the ultrapotassic magmatism (35–30 Ma) and its association with the MD carbonatites (30–12 Ma; Fig. 1a). The westward flow is recorded by Quaternary calc-alkaline lavas (14–0.2 Ma) in Tengchong. Their arc-like signature and the enriched Sr–Nd isotope compositions suggest a derivation from the mantle wedge, previously metasomatized by Neo-Tethyan oceanic subduction (Supplementary Figs. 9–11) but heated by the westward asthenospheric upwelling since 14 Ma (Fig. 3a).

Three lines of evidence further confirm the asthenospheric upwelling in the Cenozoic. First, electron backscattered diffraction studies of mantle xenoliths hosted by OIB-like basalts at Maguan (Fig. 1a) show strong deformation fabrics, recording a vertical flow of mantle material41, and their equilibration temperatures are estimated at >1200 °C, much higher than that predicted by a cratonic geotherm42. Secondly, lower crustal xenoliths from Tengchong and Dali (Fig. 1a) underwent high-temperature granulite- and amphibolite-facies metamorphism at very high equilibration temperatures (1050 °C–900 °C)43. Third, P-T-t paths for the metamorphic belts along the RRF reflect a complex process evolving from heating and ductile flow at 40–35 Ma to brittle shear at 30–23 Ma44. Therefore, we conclude that the subduction of the Indian lithosphere during collision induced the upwelling of the Asian asthenosphere, thus changing the cratonic–orogenic lithospheric thermal regime, triggering its partial melting and controlling crustal deformation in the orogen.

Carbonatite formed by melting of the metasomatized SCLM

Previous modeling, overwhelmingly based on geochemical and Sr, Nd, Pb, O, and C isotope data of carbonatites worldwide, suggested distinct mantle sources for their genesis, such as the asthenosphere related to plume activity8,12, heterogeneous lithospheric mantle15, and mixing source between the lithosphere and asthenosphere14,16. Three lines of evidence suggest that the MD carbonatites originated from the cratonic SCLM with variable degrees of metasomatism22, except for the carbonatite at Dalucao showing crustal contribution21,23,24. First, these carbonatites have more radiogenic Sr–Nd isotope compositions with initial (87Sr/86Sr)i of 0.7056–0.7080 and εNd(t) of −6.0 to −1.619,21,22, overlapping with those of the Eocene ultrapotassic rocks that originated from the metasomatized cratonic SCLM22,45 (Fig. 3b). Second, the associated syenites yielded similar Sr–Nd isotope compositions, with initial 87Sr/86Sr = 0.7043–0.7049 and εNd(t) from −2.4 to −6.539, suggesting a genetic link. Thirdly, the magmatic zircons from these syenites yielded δ18O of +3.8 to +5.4‰ and εHf of −3.7 to +1.2, also indicating a SCLM source with minor marine sediments (Supplementary Fig. 11).

Figure 2a, b clearly shows that the MD carbonatite belt spatially coincides with an underlying shallow low-velocity mantle zone, which could be attributed to the presence of low-viscosity carbonate melts or volatiles (CO2, H2O, Cl, and F) released during magmatic degassing. Our seismic images show that the SCLM at the cratonic edge is cut by a west-dipping XJF system (Fig. 2a, b), along which upwelling and lateral flow of the asthenosphere could trigger partial melting of the SCLM, thus endowing the carbonatitic melts.

Experiments have shown that the partial melting of mantle peridotite with CO2 only produces SiO2-undersaturated basaltic magmas at <80-km depths46. However, with increasing pressure (80–150 km) and temperature (>1200 °C), its melting could yield carbonatitic melts with silicate components46,47,48,49. Recent studies further suggested that incipient melts of peridotites containing both H2O and CO2 at 80–230 km are carbonatite50,51, and these melts progress toward carbonated silicate compositions with increasing temperatures50. When the mantle contains 0.15 wt% of CO2, carbonatites could be generated by very low degrees (<0.5%) of melting49. This means that the formation of the MD carbonatites requires a thick lithosphere, appropriate mantle temperatures, and a carbonated mantle source. These requirements are fully satisfied at the Yangtze cratonic margin: (1) a relatively thick underlying lithosphere (>90 km), thermally eroded by heat from asthenospheric lateral flow (Fig. 2a, b); (2) relatively high mantle temperatures (≥1000 °C) at 80- to 120-km depths (Fig. 3a), enabling partial melting of the carbonated SCLM; and (3) carbonate metasomatism in the mantle source, evidenced by light δ26Mg and heavy δ66Zn values of the associated Cenozoic alkaline rocks along the cratonic margin52,53. Note that the cratonic lithosphere was strongly thinned south of 26°N (Fig. 2c), implying the loss of a lithosphere root, where the MD carbonatites were not developed (Fig. 1a). This reinforces the concept that the carbonatitic melts resulted from the partial melting of a previously metasomatized SCLM54,55.

REE-fertilization of the SCLM by CO2-rich fluids from recycled marine sediments

The extreme enrichment in REEs of the MD carbonatites, through further strengthening by intra-crustal magmatic processes because of their incompatibility2,3,21, must have undergone a refertilization process to have affected the partial-melt source. Modeling of the Sr–Nd (Fig. 3b), Hf–O (Supplementary Fig. 11), and Li–O56 isotopic compositions of the MD carbonatites and associated syenites indicates that their SCLM sources were strongly pre-metasomatized by recycled marine sediments or fluids from a subducted slab22,57. The systematic increases in the ratios of La/Yb (Fig. 3c) coupled with Ce/Sr and La/Ba (Supplementary Fig. 12) for the Eocene mafic ultrapotassic magmas from the margin to the interior of the Yangtze craton probably reflect a lateral increase in the intensity of fluid metasomatism27,58, consistent with the trajectory of continuous subduction of the oceanic lithosphere beneath the Yangtze craton since the Neoproterozoic27.

Metasomatism by H2O-rich fluids from subducted oceanic slabs or recycled marine sediments can hydrate, oxidize, and fertilize the SCLM with ore-forming elements (e.g., Cu and Au)59, generating potassic–ultrapotassic magmas during later melting45. However, it would be difficult for this process to carbonate the mantle enough to generate carbonatite melts60 and to explain the high Ba/Th ratios observed in the MD mafic rocks and carbonatites, because such fluids yield either a low 87Sr/86Sr close to those of midocean ridge basalts and/or have relatively low Ba/Th (Fig. 3d). Direct melting of carbonate sediments recycled into the cratonic mantle theoretically could produce primary carbonatitic melts, but these melts would yield more radiogenic Sr isotope compositions (87Sr/86Sr > 0.706)61 and lower Ba/Th ratios (<100) than those in the MD mafic rocks and carbonatites (Fig. 3d).

Here, we propose that CO2-rich fluids from recycled marine sediments plus REE-rich sediments are most likely responsible for the geochemical and isotopic characteristics of the MD mafic rocks and carbonatites inherited from the metasomatized SCLM (Fig. 3b, d). CO2-rich fluids generally have high large-ion lithophile element (LILE) contents and LILE/high field strength element (HFSE) ratios62. Experiments indicate that CO2-rich fluids could transport and add light REEs (LREEs) to the SCLM at pressures >25 kbar47 because of the high partitioning coefficients of LREEs in CO2-rich fluids/melts48. The SCLM metasomatized by such LREE- and CO2-rich fluids would become a carbonated peridotite enriched in LILEs and REEs. The partial melting of this material would concentrate the alkaline-earth elements (e.g., Ca, Sr, and Ba) and LREEs into carbonate-rich melts48, leaving HFSEs, heavy REEs, and residual Y in refractory rutile because of their high Dmin/melt63. This scenario is also consistent with the thermodynamic modeling of the dehydration of subducted oceanic lithosphere, which releases 60–90% of CO2 with increasing depth64.

Previous studies have demonstrated that REEs are concentrated in metalliferous, Fe-oxyhydroxide-bearing sedimentary muds on the modern seafloor, such as in the southern Pacific (>1000-ppm total REE + Y) and northern Pacific (>400-ppm total REE + Y)65. Recent observations demonstrate that the now-exposed Mesoproterozoic (~1.6 Ga) shallow marine dolomites and phosphorites along the western margin of the Yangtze craton contain sections of exhalative-sedimentary Fe–Cu–REE mineralization66. These rocks provide an analog for Neoproterozoic recycled marine sediments, and they contain abundant LREE-rich bastnäsite and have total REE contents up to 4440 ppm66. We suggest that such REE-rich marine sediments could be the source of the metasomatizing LREE-rich and CO2-rich fluids during Neoproterozoic subduction. These fluids strongly reacted with the overlying SCLM, generating a carbonated peridotite source for the geochemically distinctive partial-melt component of the carbonatitic magmas. This scenario is also consistent with recent Mg- and Zn-isotope studies on Cenozoic alkali basalts in the Tibetan collision orogen52. A simulation calculation shows that binary mixing of the depleted mantle with fluids derived from Mesoproterozoic marine sediments could also explain the high Ba/Th (>100) and 87Sr/86Sr values of the MD mafic rocks and carbonatites (Fig. 3d). We therefore suggest that preexisting REE refertilization of the SCLM produced by the Proterozoic subduction is critical for post-collisional carbonatites and associated REE deposits in the MD belt.

Formation of carbonatite-associated rare-earth element deposits

Our results reinforce three key factors previously recognized for the formation of fertile carbonatites and CARDs in collisional settings: (1) a thick lithosphere with associated high pressure (>25 kbar)47, (2) a carbonated lithosphere mantle source enriched in REEs22, and (3) favorable pathways for magmas ascending through the overlying crust, where additional fractionation promotes the formation of a mobile alkaline “brine-melt” and alkali REE carbonates3,67. We argue that the optimal configuration of all three factors at the activated craton edges is critical to the formation of giant CARDs (Fig. 4). The cratonic lithosphere in orogenic belts (e.g., collisional) is typically relatively thick, partly because it preserves a continental root (Fig. 2a, b), and the SCLM above a fossil subduction zone had previously subjected to subduction-related metasomatism with various degrees (Fig. 4). Continental subduction during collision disturbed the overlying mantle, which caused the asthenospheric upwelling along the convergent block boundaries and subsequent lateral flow beneath the rigid SCLM, as observed in Fig. 2. The head of the upwelling asthenosphere, similar to a mantle plume, can thermally erode and interact with the overlying SCLM and generate ultrapotassic magmas (Fig. 4). Asthenospheric lateral flow over time triggers partial melting of the REE-rich thick carbonated cratonic SCLM, generating carbonate-rich melts ascended along trans-lithospheric faults and/or planes of weakness (e.g., XJF) into the shallow crust near the cratonic edge (Fig. 4). Similar melting may also happen in the REE-refertilized SCLM below some rifts at the cratonic edge, along which the resulting carbonatite melts ascend upward and form CARDs in the crust (Fig. 4).

Fig. 4: A schematic cartoon illustrating the genesis of CARDs at cratonic edges.
figure 4

At cratonic margins, vertical upwelling and lateral flow of hot asthenosphere driven by subduction or mantle convection thermally eroded and partially melted overlying the cratonic SCLM. The SCLM at cratonic margins previously underwent metasomatism by REE- and CO2-rich fluids from recycled marine sediments and thus refertilzed in REEs, which were entrained later by carbonate-rich melts ascending along lithospheric discontinuities (e.g., strike-slip faults; rifts), forming giant CARDs in the orogens and rifts. Melting of the cratonic SCLM with lacking source metasomatism by REE- and CO2-rich fluids could generate carbonatic, ultrapotassic and mafic melts, but they have limited potential for CARD formation. The adiabatic temperatures of 1000 °C and 1200 °C of the upper mantle are shown (see Fig. 3a).

Full size image

We argue that mantle metasomatism with REE refertilization prior to melting is an important prerequisite for giant CARD formation. It is likely that whether this process happened depends on the thermal states of the subduction systems and the thickness and behavior of the recycled sediments. Kinematic modeling and instability calculations indicate that the subducted oceanic crust dehydrates before it reaches subarc depths68, while sediments detach from the downgoing slab to form buoyant diapirs at subarc depths69. Melting of the rising diapirs at >1000 °C would release trace elements (Th, Pb, and LREE) into arc lavas, which thus show the sediment melt signature69 or into the overlying SCLM, which generate ultrapotassic magmas during later melting (see Fig. 4). A greater fraction of carbonate sediments may be transported to deeper mantle depths with ongoing subduction70, where they release REE- and CO2-rich fluids62 or generate near-solidus small-degree partial melts69, then metasomatized the overlying depleted SCLM (see Fig. 4). The continuous subduction processes lead to the homogeneity of the cratonic SCLM and refertilization with ore-forming elements varying from Cu–Au to REE from the margin to the interior of a craton (Fig. 4).

Our genetic model for giant REE ore systems provides a reasonable explanation for other CARDs18,71,72. A typical example is the Proterozoic Mountain Pass REE deposit in the United States72, which is located at the reactivated edge (accretionary orogen) of an Archean craton73. The associated carbonatites originated from a metasomatized mantle source at 1.8–1.6 Ga74. This is also the case for the giant Bayun Obo deposit on the northern margin of the North China Craton18. Although the asthenospheric upwelling and lithospheric thinning strongly modified this craton since the Mesozoic75, its ancient SCLM was metasomatized by subducted sediments at 1.9–2.0 Ga76, and subsequent melting generated the REE-rich carbonatites and associated CARD during mid-Proterozoic rifting2. Some medium–small CARDs, although also located along the cratonic margins2,17, have limited potentials and prospects. This is likely related to other factors, such as the lack of REE refertilization and strong metasomatism in the SCLM, the limited volume of carbonatic melts2, unfavorable intra-crustal magmatic processes for REE mobility and enrichment3, and even geopolitics, environmental concerns, economics and so on. We argue that cratonic margins with REE refertilization most likely satisfy the optimally-configured conditions for generating CARDs and therefore suggest that the reactivated cratonic edges should be favorable targets for exploration of such CARDs.

Methods

To image the mantle structure of the eastern Indo-Asian collision zone, we conducted seismic tomography for continental China and its surrounding regions using teleseismic double-difference tomography method77. We mainly assembled two sets of data from the ISC-EHB Bulletin and China Digital Seismic Network (CDSN) catalog (Supplementary Figs. 1 and 2). The ISC-EHB Bulletin contains a refined version of the methodology described in ISC Bulletin by the EHB algorithm78 to minimize errors in earthquake locations. We collected P-wave arrival times of globally distributed stations in the period of 1964–2017. For the CDSN catalog, we assembled approximately 780,000 events recorded by 1434 stations during the period from 2008/10 to 2018/06. Most events are shallow and of small magnitude. In this study, we carefully selected a limited number of events based on the following criteria. We first divided the study region into 0.5° × 0.5° × 50 km cells and only selected the first 10 events with maximum observations in each cell, and each event had more than 20 observations. As a result, 3105 events were selected from the CDSN catalog. From the two datasets, we constructed 1.69 million absolute arrival times, from which ~7.2 million event-pair differential arrival times were constructed by selecting event pairs having more than 50 common observations for the ISC-EHB Bulletin and more than 20 common observations for the CDSN. The distances between event pairs were between 50 and 500 km.

Compared with available global velocity models, our model better resolves more details in the mantle structure in the study area (Supplementary Fig. 5).

For the inversion, we followed the teleseismic double-difference seismic tomography method (teletomoDD) of ref. 78, which is a global version of the double-difference seismic tomography method (tomoDD)79,80. It can use both regional and teleseismic stations to invert for the velocity structure of the study region. The model is parameterized by nested grids with coarse and global grids outside the study region and finer and regional grids for the study region. For the regional initial model, we choose the AK135 model81. The regional model is from 74°E to 136°E in longitude and from 18°N to 54°N in latitude and extends to 2500 km in depth. The regional grid interval is 2° along longitude and latitude and ~40~300 km in depth. For the global model, we choose the TX2019slab model82 as the initial model, with a grid interval of 5° in the horizontal direction and ~55~225 km in depth.

In the inversion, we applied a hierarchy strategy to invert for the velocity structure79 by first giving absolute arrival times a larger weight and then giving event-pair differential times a larger weight. After seven iterations, the absolute data RMS residual decreases from 1.85 s to 0.73 s, and the weighted data residual decreases from 3.31 s to 0.56 (Supplementary Fig. 3). Compared to the data residuals with respect to the initial model, final travel time residuals are more concentrated near 0 s (Supplementary Fig. 3). We plotted different depth slices of the final Vp model for continental China and surrounding regions (Supplementary Fig. 4).

The checkerboard resolution test was used to check the model resolution. In our checkerboard test, we added ±5% anomalies at the adjacent grid nodes in the regional model. Here, we only show recovered checkerboard patterns for the eastern Indo-Asian collision zone. Supplementary Figs. 5 and 6 show the checkerboard recovery at different depths and along different latitudes. Overall, for the eastern Indo-Asian collision zone, the model has a good resolution in most parts.

Along the three profiles in Fig. 1a, we also plotted four widely used global velocity models, including the GAP_P4 model83 (Supplementary Fig. 7a), the LLNL_G3Dv3 model84 (Supplementary Fig. 7b), the MIT-P08 model85 (Supplementary Fig. 7c), and the TX2019slab model82 (Supplementary Fig. 7d). In comparison, our model (Fig. 2) shows more detailed mantle features, while the other four global models show some fuzzy and inconsistent features. This is because we used both dense regional seismic stations and teleseismic stations. For example, in the GAP_P4 model (Supplementary Fig. 7a), although the subducted Indian slab can be imaged, the Yangtze block is not well resolved. For the LLNL_G3Dv3 model (Supplementary Fig. 7b), both the Indian and Yangtze blocks are not well imaged. For the MIT-P08 model (Supplementary Fig. 7c), the amplitude of velocity anomalies is subdued and although the subducted Indian slab can be identified the Yangtze block is very fuzzy. Lastly, the TX2019slab model also shows low-resolution features for the Indian and Yangtze blocks (Supplementary Fig. 7d). Compared to our Vp model (Fig. 2), these four models do not show clear upwelling asthenosphere from the MTZ (Supplementary Fig. 7).

Because the Vp model mostly focuses on the mantle structure of the eastern Indo-Asian collision zone, to obtain higher-resolution images for depths of 60–110 km, we also plotted Vp and Vs depth slices at 80 km of the USTClitho2.0 model inverted from joint inversion of body and surface wave data (Supplementary Fig. 8)32. It can also be seen that the India block and Yangtze block are in close contact at approximately 101°E, and there is a clear channel shown as low-velocity anomalies between the two blocks. This is consistent with the features observed in the depth slice of 160 km in the Vp model obtained in this study.