Continental flood basalts derived from the hydrous mantle transition zone

It has previously been postulated that the Earth’s hydrous mantle transition zone may play a key role in intraplate magmatism, but no confirmatory evidence has been reported. Here we demonstrate that hydrothermally altered subducted oceanic crust was involved in generating the late Cenozoic Chifeng continental flood basalts of East Asia. This study combines oxygen isotopes with conventional geochemistry to provide evidence for an origin in the hydrous mantle transition zone. These observations lead us to propose an alternative thermochemical model, whereby slab-triggered wet upwelling produces large volumes of melt that may rise from the hydrous mantle transition zone. This model explains the lack of pre-magmatic lithospheric extension or a hotspot track and also the arc-like signatures observed in some large-scale intracontinental magmas. Deep-Earth water cycling, linked to cold subduction, slab stagnation, wet mantle upwelling and assembly/breakup of supercontinents, can potentially account for the chemical diversity of many continental flood basalts. The Earth’s mantle transition zone may play a key role in large-scale intraplate magmatism and plate tectonics. Here, the authors provide evidence for the origin of continental flood basalts in this zone, by combining oxygen isotope and geochemical evidence from the late Cenozoic Chifeng volcanics of East Asia.

C ontinental flood basalts (CFBs) form part of large igneous provinces (LIPs) erupted onto continental crust and are characterized by anomalously high rates of mantle melting, representing the largest volcanic events in Earth's history. Such events have significant implications for understating Earth dynamics 1,2 and global environmental catastrophes 3 . CFBs are conventionally attributed to decompressional melting of hot mantle, driven by mantle plumes-hot upwellings originating at the core-mantle boundary 2 . However, the chemical compositions of most CFBs are dissimilar to ocean island basalts (OIBs) 4 that are associated with hot spots-the surface expressions of mantle plumes 5 . Furthermore, although CFBs are generally several hundred kilometres from plate boundaries (subduction zones), some exhibit subduction-like geochemical signatures 4,6 . Petrological and geochemical evidence from the Karoo 7 , Etendeka 8 , Siberian 9 and Central Atlantic LIPs 6 suggests that generation of CFBs is ultimately related to fluid/melt derived from subducted slabs. However, the question of how fluid is released from deep subducted oceanic slabs and effects or controls partial melting of the mantle in an intracontinental setting is unclear.
The mantle transition zone (MTZ) has been considered to act as a gigantic water tank in deep-Earth fluid cycling [10][11][12][13][14][15][16][17] due to water-enriched wadsleyite and ringwoodite at depths between 410 and 660 km (refs 11,13,16,17), which can host enough water to comprise up to 2.5% of the weight of these minerals 13,14 . This implies that the MTZ may play a key role in terrestrial magmatism and plate tectonics 12,18 . As an oceanic plate descends into the upper mantle, any fluid retained within the subducting plate beyond the depths of magma generation constitutes a return flux to the Earth's interior 15 . Subducted oceanic slabs tend to be subhorizontally deflected or flattened as 'stagnant slabs' above, across or below the 660-km discontinuity 11 . Hydrous phases in the MTZ would eventually decompose into anhydrous phases by releasing fluids to produce wet upwellings that pool within the MTZ 11,19 . However, direct observations for the water content of the MTZ are spare and mainly inferred from geophysics 16,17 . With the exception of a single B10 mm ringwoodite grain from one location 13 , there is no geochemical evidence that the MTZ is water rich. Thus, finding confirmatory evidence of the origin of primary mantle melts from a hydrous MTZ is an important step in understanding deep-Earth water cycling and identifying contributions from non-plume processes in the generation of CFBs.
This study suggests that oxygen isotopes, 39 Ar/ 40 Ar ages, and other geochemical features of the Chifeng CFBs of East Asia are correlated with distance to the western edge of the stagnant Pacific slab within the MTZ. We demonstrate that the lavas immediately above the western edge of the stagnant Pacific slab are characterized by being oversaturated in silica, having the oldest eruptive ages, the lowest olivine d 18 O values and possibly the highest fluid contents. The combination of clearly recycled oceanic crust-derived signatures and their correlation with the inferred position of stagnant Pacific slab within the hydrous MTZ therefore provides an important constraint on the depth of the mantle source. These observations lead us to propose an alternative thermochemical model that may be more widely applicable, whereby slab-triggered wet upwelling produces large volumes of melt that may rise from the hydrous MTZ. This model links deep-Earth water cycling, slab stagnation, wet mantle upwelling, large-scale intraplate magmatism and assembly/ breakup of supercontinents into a self-organised system. large intracontinental scale, because Cenozoic volcanic rocks in this region (Fig. 1) are associated with stagnant slabs in the MTZ 20,21 . The subducted slab of the Pacific plate extends horizontally over a distance of 800 to 41,000 km in the MTZ at the 660-km discontinuity 11,21,22 . Such a stagnant slab has the ability to add a large amount of subducted fluid into the mantle, with the potential for subsequent triggering of wet mantle upwelling 21,[23][24][25] .

Hydration
The Chifeng CFBs are located far from plate boundaries, but were erupted directly above the inferred edge of the Pacific stagnant slab (red dashed line in Fig. 1), and extend about 200 km further to the west. They were mainly (490% volume) erupted between 23.8 and 6.1 Myr ago ( Supplementary Figs 1 and 2), producing 100-450-m-thick sheet-like lava flows that cover a total area of 3,000 km 2 (ref 26). Eighty-five samples were collected in this study from across the Chifeng CFB province to investigate the role of deep water cycling in their generation. The rocks include quartz tholeiite, olivine tholeiite and alkali basalt, all with high-sodium contents (Supplementary Figs 3 À 5) and depletion of heavy rare earth elements ( Supplementary Fig. 5). All the samples show prominent positive spikes for Ba, K, Pb and Sr in the primitive mantle normalized diagram ( Supplementary Fig. 5), a typical characteristic of hydrated mantle melts 6,20,27,28 .
The most striking observation is that oxygen isotopes, 39 Ar/ 40 Ar ages, and other geochemical features of the Chifeng CFBs (SiO 2 , Nb, Sm/Nd, Ce/Pb, K/Ce, Tb/Yb and Sr/Ce) are correlated with distance to the western edge of the stagnant Pacific slab ( Fig. 2; Supplementary Fig. 6). The western termination of the Pacific slab is located at B119°30 0 E (ref 22) (red dashed line in Fig. 1), with the Chifeng CFBs extending westward to around 117°E. In a westward direction, the lavas decrease in age, are depleted in silica, and have lower Sm/Nd ratios, whereas they show progressive enrichment in Nb and have higher Ce/Pb ratios. Importantly, the olivine d 18 O values of the basalts increase westward away from the stagnant slab (Fig. 2a). Thus, the lavas immediately above the western edge of the stagnant Pacific slab are characterized by being oversaturated in silica, having the oldest eruptive ages, the lowest olivine d 18  Variations in oxygen isotopes in the mantle source are ultimately related to surface water-rock interaction [29][30][31] . The oxygen isotopes of primary melts are therefore sensitive to contributions from hydrothermally altered recycled components. Partial melts of unaltered mantle peridotites record a relatively uniform value of d 18 Fig. 2a). Because the above-mentioned features are tightly correlated with the position of the stagnant paleo-Pacific slab within the MTZ, this implies that slab stagnation, dehydration and wet upwelling may have played an important role in the generation of the Chifeng CFBs.
Differentiation of the Chifeng CFBs is dominated by olivine and clinopyroxene fractionation ( Supplementary Fig. 9). Crystallization of olivine will increase both the silica content 32 and the d 18 O 29 value in differentiated magmas and thus would generate positive d 18 O-SiO 2 and negative d 18 O-Fo correlations, respectively. This is in stark contrast with our observations (Fig. 3a,c). Crustal materials are generally characterized by high SiO 2 contents and high La/Sm, but low Sm/Nd, ratios, whereas asthenospheric mantle-derived melts generally have typical mantle oxygen isotopes (olivine d 18  , the OIB-like end-member melt has a chemical composition similar to that of the incipient partial melt of garnet peridotite 33 , with SiO 2 ¼ 44 À 46 wt%, TiO 2 ¼ 2.0 À 2.5 wt%, Al 2 O 3 ¼ 9.4 À 10 wt%, MgO ¼ 17 wt% and FeO total ¼ 10 À 11 wt%; Sm/Nd ¼ 0.20, Zr/HfZ44, Nb/LaZ1.4, Lu/Hfr0.04, Al 2 O 3 / TiO 2 ¼ 4-5 and La/SmZ4.0. The evidence from low 3 He/ 4 He ratios shows that they are similar to melts generated in the shallow mantle, but significantly lower than those generated in the primordial lower mantle 28 . The lack of lower mantle-rooted low-velocity seismic structure 21 , the alkalic nature of the dominant phase and no evidence for pre-eruption uplift 34 argue strongly against a deep plume origin for the OIB-like endmember melt 35 . The large-scale doming in the Hangai-Hövsgöl region of central Mongolia has been cited as key evidence for the presence of a mantle plume below Central Asia in the late Cenozoic 36 . However, a combined gravity-and topography-based geophysical investigation 37 demonstrated that a high-heat-flux mantle plume was unlikely to be involved because upper mantle    (Fig. 3a), depleted incompatible trace element compositions (Figs 2c and 3b) and subduction slab-like Ce/Pb ratios (Fig. 3d). Furthermore, the depletion of 18 O is coupled with depletion in high-field-strength elements (for example, Nb; Fig. 2c) and other incompatible elements, as evidenced by decreasing La/Sm ratios (Fig. 4b). More importantly, the low d 18 O end-member melt also has characteristics of melts typical of recycled oceanic crust that is dominated by gabbroic components, such as high-positive Sr anomalies (Fig. 4a), high Al 2 O 3 (Fig. 4c) and SiO 2 (Fig. 3a) contents, and high Fe/Mn ( Supplementary Fig. 8f), Sm/Nd (Fig. 3b) and Lu/Hf (Fig. 4d) ratios, but low La/Sm (Fig. 4b) and Tb/Yb ( Supplementary Fig. 6b) ratios. This is confirmed by the studies of olivine-hosted melt inclusions 39 from the south-eastern extremity of the Chifeng CFBs (119°15 0 E) erupted directly above the western edge of the stagnant Pacific slab. The melt inclusion data imply that these CFBs were mainly derived from an olivine-free pyroxenite-dominated source with high water content (4450 p.p.m.). These observations show that the low d 18 O end-member melt most likely originated from hydrothermally altered oceanic gabbros 30,32 . Furthermore, from the systematic regional trend presented in Fig. 2, hydrothermally altered oceanic gabbro was most likely present in the stagnant Pacific slab. Thus, the combination of clearly recycled oceanic gabbro-derived signatures and their correlation with the inferred position of stagnant Pacific slab within the hydrous MTZ provides an indirect, but important, constraint on the depth of the mantle source.
For the Chifeng CFBs, the hydrous melts considered to have originated in the MTZ do not record any high d 18  In summary, the source of the CFBs is dominated by two endmember reservoirs: the stagnant Pacific slab and fertile peridotite. Mixing of melts from these two end-members can explain the major element, incompatible trace element, and oxygen isotope compositions of the Chifeng CFBs (Fig. 3a,b,d). The two-endmember mixing calculation shows that the source of low Ce/Pb lavas may also contain minor subducted sediments ( Fig. 3d; Supplementary Figs 11 and 12). This is consistent with the coexistence of low Ce/Pb and normal olivine d 18  MTZ wet upwellings produce the Chifeng CFBs. These observations lead us to propose an alternative model for the Chifeng CFBs, whereby slab-triggered wet upwelling and upward percolation/hydration, which originated at the MTZ, produced large volumes of melt (Fig. 6). The slow kinetics of the pyroxenegarnet transformation 41 and slab viscosity 42 within the MTZ enables the Pacific slab to reside for a long time in this zone (over 10 8 year) and hence lose water by dehydration reactions within the MTZ 11 . Because feedbacks between trench backward migration and slab deformation are integrated during the slab stagnation 42 , the declining rate of Pacific-Eurasia convergence from a late Cretaceous convergence rate of 120-140 mm year to a minimum in the Eocene of 30-40 mm year 43 may also have contributed to the Pacific slab stagnation. After completion of the pyroxene-garnet transformation 41 , the stagnant Pacific slab is likely to eventually fall into the lower mantle (slab avalanche) and release most of its water by dehydration melting when it penetrates the 660-km discontinuity 19 . The viscosity reduction due to slow grain growth may also induce slab penetration 42 . Combining geophysical and experimental studies has therefore demonstrated the presence of extensive dehydration within the MTZ beneath this region 25 . Experiments 44 and seismic investigations 45 indicate that hydrous melt or supercritical fluid is gravitationally stable atop the 410-km discontinuity and may ARTICLE possibly spread laterally. The density contrast between melts and surrounding solids determines the direction of material transport in the Earth and therefore the density of silicate melts plays a very important role in controlling the chemical differentiation of the Earth 46-50 . In this model, the density contrast between hydrous melts and coexisting minerals is a critical parameter that affects hydrous melting and consequent upwelling. Hydrous melt is denser than the associated minerals if the water content is low, whereas when the water contents increase to higher than the threshold values, the hydrous melts will be buoyant 50 . This aspect has been investigated 48,49 and it is confirmed that a hydrous melt formed at relatively low temperatures will be buoyant because the melt should have a large water content, but that a hydrous melt formed at high temperature will be denser. Thus, large-scale wet upwelling may be caused by a combination of slab avalanches and the positive buoyancy generated by accumulation of less-dense fluid phases. The wet upwellings will fertilize the shallow mantle (highlighted by yellow colour in Fig. 6) and decrease its solidus temperature, resulting in large-scale partial melting 51 . The temporal-spatial distribution of the Chifeng CFBs may thus be controlled by variations in slab stagnation and its final penetration of the 660-km discontinuity into the lower mantle.
Initial subduction of the Pacific oceanic plate under this part of Asia was inferred to take place at 130 À 120 Myr ago by integrating the drift history of the Pacific plate with tectonic evolution and gold mineralization in eastern China 52 . Combining the temporal distribution of volcanism and calculating the cumulative amount of total subduction within the MTZ with geophysical tomography in this region shows that collapse of stagnant slabs may have started about 25 Myr ago 12 , which coincides with the onset of eruption of the Chifeng CFBs (ca. 24 Myr ago; this study). Furthermore, because the completion of the pyroxene-garnet transformation within the MTZ will take B100 Myr ago 41 , the age of the stagnant Pacific slab is inferred to be B125 Myr ago, consistent with previous estimates 52 . Our study therefore leads to the proposition that a MTZ water-filtering model may have controlled water in the MTZ via stagnation of the Pacific oceanic plate. The contributions from water filtering 19 and high water solubility 18 result in hydration of the MTZ 10-12 .

Discussion
The hydrous MTZ model may also apply to CFBs associated with the breakup of supercontinents. Geological evidence shows that both the Pangea 53 and Rodinia 54 supercontinents were surrounded by circum-supercontinent subduction zones, a feature that seems to be true for other supercontinents and is likely to have important geodynamic implications 53 . Assembly of supercontinents therefore potentially leads to accumulation of a large volume of hydrated oceanic slabs within the MTZ 11,12 and may possibly change the mode of mantle convection 11 . The time delay of 140-150 Myr between the final assembly and initial breakup of Pangea 53 and Rodinia 54 is in good consistent with the lifespan of stagnant slabs within the MTZ constrained by slab viscosity 42 and phase transformation 41 . Such similarity may hint at a possible causal relationship between slab avalanche (and/or mantle overturn) and breakup of supercontinents. Slab avalanches within the hydrated MTZ would generate large-scale wet upwellings that hydrate the subcontinental lithospheric mantle (Fig. 6), resulting in an instability of the lowest part of the continental lithosphere 55,56 . More importantly, hydrous basaltic melts would prefer to pond at the boundary between the lithosphere and asthenosphere 57,58 , providing a lubricating mechanism for speeding-up plate motion. The slab avalanchedriven upwelling and mobility of hydrous basaltic melts would hence promote the breakup of supercontinents and may in turn facilitate slab avalanches by adjusting trench motions and changing the geometry of subducted slabs 59 . Recent studies have proposed that the strictures of classical physics combined with modern seismic imaging, require that convective circulation in the upper mantle is plate driven and that the principal response to plate subduction and delamination is large, wide, slowly upwelling regions of mantle rooted within the MTZ 60 . Thus, plate-and slab avalanche-driven upwelling originating from the hydrous MTZ may have played an important role in the breakup of supercontinents and the circulation of water (and possibly other fluid phases) between Earth's surface and the MTZ.
The arc-like geochemical features of CFBs are traditionally attributed to continental lithosphere (both crust and mantle) input. The presence of strong arc-like geochemical signatures and hydrous phenocrysts in some high-magnesium basaltic   Figure 6 | Effects of slab stagnation and water cycling on the upper mantle thermochemical state. Water cycling refers to wet upwelling, upward percolation and refertilization. This model is mainly based on the strong spatial correlation of geochemical features with distance of eruptive lavas relative to the edge of the stagnant slab. This is based on water partitioning in the Earth's mantle 18 , behaviour of slab-triggered wet upwelling and upward percolation 23,40 , hydrous mantle melting 51 , the mantle wedge model 21 , and upwelling from the hydrated mantle transition zone 24 . rocks from the Siberia 9 and Karoo 7 LIPs show that crustal contamination cannot be the controlling factor. Detailed geochemical and isotopic analyses of basaltic rocks from the Siberia 9 , Karoo 7 and Central Atlantic 6 LIPs demonstrate that their generations involved water and possibly other fluid phases released from subducted slabs and that their source reservoirs were both isolated and hydrated over a considerable period of time. Because the hydration processes would result in instability of subcontinental lithospheric mantle 56 , hydrated subcontinental lithospheric mantle cannot remain isolated from the convective upper mantle. Geochemical studies and thermochemical modelling suggest that the massive degassing of volatiles (CO 2 and HCl) in some CFBs mostly originates from recycled crust in the deep Earth 61 . These observations show that continental lithosphere cannot be the dominant reservoir that produces the observed hydrous and arc-like geochemical characteristics of these CFBs. Thus, finding a mechanism to explain the long-term hydration and isolated nature of the mantle reservoir beneath the lithosphere is crucial for understanding the origin of CFBs.
The deep-Earth cycling of water thus requires a fresh perspective to explain long-term hydration and its isolation within the mantle. For example, water and other fluid phases released from stagnant slabs carry large-ion lithophile elements such as Ba, Pb, K and Sr, and light rare earth elements (La to Nd), but are not able to carry high-field-strength elements such as Nb, Ta and Ti 9 . Long-term hydration and isolation within the MTZ therefore provide a mechanism to produce a hydrated reservoir that is characterized by both low high-field-strength and high large-ion lithophile element abundances and enriched radiogenic isotopes-a characteristic of arc-like CFBs 4,6-9 . Water-driven wet mantle will rise by its own buoyancy and melt on crossing its wet solidus beneath supercontinents. Such a water cycle, linked to cold subduction, slab stagnation and downward fluid transfer in the MTZ by water filtering, is expected to produce arc-like CFBs 4,6-9 with chemistry similar to that of island arcs, but at greater distances from the trench. These processes may account for the geochemical diversity of CFBs that makes them elementally and isotopically distinct from OIBs 4 . In this model, mantle upwelling is driven by deep-Earth water cycling rather than excess temperature, such as in a classical thermal plume model, which can cause about 0.8-1 km of broad surface uplift per 100°C of plume excess temperature 62 . Because the wet upwelling of mantle requires no excess temperature in this model, there will be no thermal anomaly to result in pre-volcanic eruption surface uplift 62 . Also, some parts of the subducted slab may undergo dehydration melting and release water when it penetrates the 660-km discontinuity into the lower mantle 19 , whereas other subducted slab fragments may be entrapped in wet upwellings and returned to the upper mantle 10,51 . The latter provides a variant of the model that also accounts for the involvement of deep subducted slabs in the generation of large-scale intracontinental magmatism. In addition, numerical simulations 23 and geophysical investigations 11,21,22 have demonstrated that volcanism produced by wet upwelling of the mantle originating from a hydrated MTZ can be located between 600 and over 1,000 km from the trench. This can lead to CFB provinces being located at the margins of Precambrian cratons 1 . The evidence for a craton-accreted terrane setting is particularly well established for the Central Atlantic LIP, where Paleozoic island arcs and continental fragments were accreted onto the eastern margin of Laurentia during the assembly of Pangea 4 . Subsequently, the Central Atlantic LIP was extruded directly onto these terranes during the breakup of Pangea. The dominant arc-like, low-Ti-type basalts 4,9 within the Siberian LIP have also been attributed to wet upwelling, driven by dehydration of the Mongolia-Okhotsk oceanic slab at the MTZ 9 . Finally, water and other fluid phases released from the hydrated MTZ may also lead over time to melt accumulating at the lithosphere-asthenosphere boundary 57,58 . Accumulated melts can escape to the surface through weak zones, such as pre-existing suture zones, within the lithosphere. This explains why CFB provinces are commonly located over suture zones 1 . Accumulated hydrous melts at the LAB have the potential to behave as a lubricant to continent rifting and thus enhance the breakup of supercontinents. This means that arc-like CFBs, such as the Karoo, Central Atlantic and Siberian LIPs, were prominent in the initial breakup of Pangea. Here we emphasize that the presence of a plume that originated at the core-mantle boundary is not a necessary condition for producing arc-like CFBs, especially those associated with rifting and initial breakup of supercontinents.

Methods
Whole-rock major and trace element analysis. Seventy-three fresh basalts samples were collected from the Chifeng-Keshiketengqi area (119°30 0 E À 117°0 0 0 E, 42°00 0 N À 43°20 0 N; Supplementary Fig. 1) for major element analysis (Supplementary Dataset 1), 45 of which were further analysed for whole-rock trace elements (Supplementary Dataset 2). Rock samples were sawn into slabs and the central parts (4200 g) were used for bulk-rock analyses. The slabs were crushed into small fragments (o0.5 cm in diameter) in a jaw crusher before being further cleaned and reduced to powder in a corundum mill. Bulk-rock major element oxides were analysed using X-ray fluorescence (Rikagu RIX 2,100) with analytical uncertainties better than 3% for SiO 2 , Al 2 O 3 , Fe 2 O 3 , MgO, CaO, Na 2 O and K 2 O and better than 5% for TiO 2 , MnO and P 2 O 5 . Trace elements were analysed using inductively coupled plasma mass spectrometry at the Key Laboratory of Continental Dynamics, Northwest University, Xi'an, after acid digestion of samples in Teflon bombs. Repeated runs gave o3% relative standard deviation for most trace elements analysed.
Mount making. Eleven samples which were analysed for whole-rock major and trace elements were chosen to separate olivine grains for electron microscope chemical and in situ secondary-ion mass spectrometry (SIMS) oxygen isotope analyses (Supplementary Dataset 3). The separated olivine grains were mounted together with reference San Carlos olivine (d 18 O ¼ 5.35%; Eiler 29 ) in the centre of epoxy resin mounts. The mounts were then ground and polished using different grain size Diamond Films to minimize any relief difference.
Electron microscope analysis. Chemical compositions of olivines were determined using a JEOL JXA-8,100 Superprobe at the Institute of Geology and Geophysics, Chinese Academy of Sciences (IGGCAS), Beijing. The operating conditions were: 15 kV accelerating voltage, 20 nA beam current, 3 mm beam diameter and 10 s peak counting time for most elements (30 s for Ca and Ni, 7 s for Na and 8 s for K). The data reduction was carried out using the atomic number, absorption, and characteristic fluorescence (ZAF) correction.
In situ SIMS olivine oxygen isotope analysis. The sample mount was gold coated and oxygen isotope compositions of olivines were analysed with a CAMECA IMS-1,280 ion microprobe at the IGGCAS. The Cs þ primary beam was accelerated at 10 kV with an intensity of ca. 2 nA. The spot size was about 20 mm in diameter (10 mm beam diameter þ 10 mm raster). The normal incidence electron flood gun was used to compensate for sample charging during analysis. Secondary ions were extracted with a À 10 kV potential. Oxygen isotopes were measured in multi-collector mode with two off-axis Faraday cups with a mass resolution of 2,500 (exit slit #1) in a circular focusing mode. Each analysis consisted of 20 cycles Â 4 s counting time. Typical ion intensities of 2-4 Â 10 6 counts per second obtained on the 18 O peak yielded an internal 2s uncertainty of better than ± 0.2%. Three unknown and two standard d 18 [63][64][65] . However, olivine with Fo number o65 has been proven to show about 0.6 permil bias IMF with San Carlos olivine as the standard 65 . Considering the unknown olivine grains have Fo number ranging from 60 to 91, we selected only grains with Fo 470 for oxygen isotope analysis. Thus, the potential matrix effect need not be considered further. 40 Ar/ 39 Ar analysis. Twelve groundmass samples were dated using the stepheating 40 Ar/ 39 Ar method (Supplementary Dataset 4). 40 Ar/ 39 Ar measurements were performed at the IGGCAS, Beijing. To constrain the eruption age and avoid excess argon, the groundmass grains with grain size ranging from 0.2 to 0.3 mm were purified carefully under a binocular microscope to remove the visible phenocrysts and xenocrysts. The samples were cleaned with acetone followed by further cleaning with deionized water in an ultrasonic bath for three times, each for 40 min. The cleaned samples were then dried at B100°C for 20 min. Groundmass wafers weighing 3-16 mg, and multiple samples of the 18.6 ± 0.4 Ma neutron fluence monitor mineral Brione muscovite, were irradiated in vacuo within a cadmium-coated quartz vial for 5 h in position H8 of the Beijing Atomic Energy Research Institute reactor . The argon isotopes were analysed on a MM5400 mass spectrometer.