Magmas Erupted during the Main Pulse of Siberian Traps Volcanism were Volatile-poor

The eruption of the Siberian Traps Large Igneous Province (SLIP) at the Permo-Triassic boundary was synchronous with environmental degradation and the largest known mass extinction in the geological record. The volatile emissions associated with these eruptions have been linked to the environmental change yet we understand little of their source and magnitude and how they varied with time. There are a number of possible sources for the volatiles that were emitted during the eruptions: the mantle (including metasomatized lithosphere), volatile-rich sediments (through metamorphism or direct assimilation) and the crustal basement. To assess the relative importance of these sources (with the exception of the metamorphic outgassing source), we have conducted a geochemical study of melt inclusions hosted by clinopyroxene in Siberian Traps low-Ti tholeiitic lavas and sills of the Khakanchansky, Ayansky and Khonnamakitsky Formations. The magmas were not emplaced into or erupted onto evaporite deposits, in contrast to samples studied previously. The trace element compositions of the melt inclusions are highly variable compared with the uniform whole-rocks, exhibiting a wide range of La/Yb ratios from 0 7 to 9 5. The melt geochemistry is consistent with relatively large degrees of partial melting of a dominantly peridotite mantle source. A negative Nb anomaly indicates a degree of crustal contamination, but there is no evidence for contamination by volatile-rich evaporites. Enrichment of some of the melts in large ion lithophile elements (Ba, Sr) indicates their interaction with a fluid. We suggest that, consistent with the observed depletion in other incompatible trace elements in the melt inclusions, the volatile concentrations in the melts were relatively low, and that subsequently the melts underwent variable degrees of degassing in the crust. Overall, the melts are more volatile-poor than those reported previously from the SLIP and were erupted after the first “pulse” of more volatile-rich magmas described by Sobolev et al. (2015). These volatile-poor magmas may have been widespread across the region during the Siberian Traps eruptions once a pyroxenite component in the mantle source had been exhausted.


INTRODUCTION
The eruption of the Siberian Traps Large Igneous Province (SLIP) (Fig. 1) took place at the Permo-Triassic boundary and was synchronous with the largest known mass extinction (Erwin, 1990;Renne & Basu, 1991;Kamo et al., 2003;White & Saunders, 2005;Reichow et al., 2009;Burgess et al., 2014).It is likely that the enormous quantities of volatiles outgassed to the atmosphere during the prolonged volcanism were an important causal factor in severe environmental degradation, which may have led to climate change and ocean anoxia (Campbell et al., 1992;Renne et al., 1995;Wignall, 2001), although the processes and feedbacks are likely to have been complex (Self et al., 2005;Mather, 2008).To understand this link, records of the volatile budget of the magmas involved in the Siberian  Malich et al. (1974)].The distribution of Devonian and Cambrian evaporites is from Zharkov (1984) and Petrychenko et al. (2005).Sampling areas indicate the locations of samples from which melt inclusions were analysed.
Siberian Traps lavas crop out on the Siberian Craton and form a discontinuous subcrop in the West Siberian Basin (Zolotukhin & Al'Mukhamedov, 1988;Reichow et al., 2002Reichow et al., , 2005)).Estimates of the total maximum area of the magmatic province range from 4Á3 million km 2 (Vasiliev et al., 2000) to 5 million km 2 (Reichow et al., 2009), which makes the SLIP the largest amongst all continental flood basalt provinces.The prevailing hypothesis for the origin of the SLIP involves the impact of a mantle plume with temperatures up to 1500-1580 C at the base of 130-180 km thick lithosphere (Richards et al., 1989;Renne & Basu, 1991;Basu et al., 1995;Sobolev et al., 2009).The lack of expected uplift that the upwelling of hot asthenosphere should produce has been explained by: edge-driven convection accompanied by lithospheric melting and delamination (Anderson, 1994;Czamanske et al., 1998;Elkins-Tanton & Hager, 2000); the presence of $15% recycled oceanic crust in the head of a superplume, which would have led to only minor regional uplift owing to the presence of high-density eclogite (Sobolev et al., 2009); or flux melting owing to the dehydration of a cold, stagnant subducted slab (Ivanov & Litasov, 2014).Alternatively, Saunders et al. (2005) proposed that the uplift occurred in the West Siberian Basin and the evidence for it was buried beneath a thick sediment pile.
It has been shown that the total volatile budget of the SLIP was complex: volatiles were probably sourced from the mantle itself (Sobolev et al., 2009(Sobolev et al., , 2011) ) and from the crust, including from volatile-rich sediments (Svensen et al., 2009).Sourcing volatiles from the crust requires either bulk assimilation and melting of the assimilated material (Black et al., 2012) or outgassing of the country rocks through contact metamorphism (Svensen et al., 2009;Aarnes et al., 2010Aarnes et al., , 2011)).In the former scenario, the geochemical signature of the assimilation would be preserved in both erupted magmas and sills.The analysis of whole-rock basalts (Grinenko, 1985;Ripley et al., 2003) and melt inclusions (Black et al., 2014) shows that sulfur isotope compositions are enriched to different extents in 34 S over 32 S, which might be explained by the assimilation of isotopically heavy anhydrite-dominated evaporite deposits (Black et al., 2014), which are prevalent in much of the region (Fig. 1).Extensive outgassing owing to contact metamorphism is believed to have significantly augmented the total Siberian Traps volatile budget (Svensen et al., 2009).This process is unlikely to have preserved any geochemical signature in the magmas, although the flow of metamorphic fluids into sills has been proposed for intrusions at Duluth, Minnesota, on the basis of osmium isotopic variations (Ripley et al., 2001).The sediment devolatilization mechanism could also explain the extreme negative excursions in the carbon isotope record in contemporaneous sediments (Baud et al., 1989) by means of outgassing of isotopically light methane derived from metamorphism (Ganino & Arndt, 2009;Svensen et al., 2009).Such excursions may also be attributed to decreased light carbon removal as a consequence of less active productivity of marine species (Brasier et al., 1996;Kaufman et al., 1996;Isozaki, 1997;White, 2002).
Previous studies of Siberian Traps magmatic rocks have demonstrated enormous geochemical diversity (Lightfoot et al., 1990(Lightfoot et al., , 1993;;Wooden et al., 1993;Hawkesworth et al., 1995).The early high-Ti basaltic series of the SLIP (Ivakinsky, Syveminsky, Gudchikhinsky; Fig. 2) are thought to have formed from melting of a pyroxenite mantle source and represent the first pulse of the eruption (Sobolev et al., 2011); low-Ti magmas (e.g.Tulkonsky, Mokulaevsky, Samoedsky Formations; Fig. 2), sourced from a peridotite mantle, erupted later (Fedorenko et al., 1996).The most primitive melt compositions found, as inclusions hosted in Fo 79Á1-82Á8 Fig. 2. Schematic stratigraphy of the Lower Tunguska, Putorana, Norilsk and Maymecha-Kotuy regions of the Siberian Traps (see Fig. 1 for locations).Modified from the Soviet regional stratigraphic scheme for Triassic deposits of the Tungusskaya syncline and Kuznetsk Basin.Melt inclusions were analysed in samples from formations whose names are highlighted in blue.
olivines from the Gudchikhinsky Formation (Fig. 2), are thought to be derived from melting of a pyroxenite mantle source and are not contaminated by continental crust, in contrast to the lavas of the Ivakinsky and Syverminsky Formations, which show evidence for contamination by continental crust (see below; Wooden et al., 1993;Sobolev et al., 2009Sobolev et al., , 2011)).Low-Ti Siberian Traps magmas form the main magmatic series and make up the largest volume of emplaced sills and lavas (Wooden et al., 1993;Reichow et al., 2005).They have undergone variable degrees of crustal contamination, varying from high (Nadezhdinsky Formation) to insignificant (e.g.Mokulaevsky and Kumingsky Formations) (Fig. 2).
The primary melt reconstructed from melt inclusions from the Gudchikinhsky Formation (Fig. 2) is poor in sulfur (400 6 200 ppm) but rich in chlorine ($350-400 ppm) compared with typical mantle-derived primary melts (Sobolev et al., 2009).It has been proposed that these melts were derived from a mantle source containing a significant proportion of chlorine-and carbon-rich recycled oceanic crust (i.e.pyroxenite).The degassing and eruption of such melts would have caused substantial HCl and CO 2 outgassing prior to the main magmatic pulse, thereby perhaps triggering the mass extinction (Sobolev et al., 2011).The hypothesis, initially proposed by Hofmann & White (1982), that the pyroxenite component of a heterogeneous mantle source melts out first, generating a change in the character of melts with time, is not unique to the SLIP.Recent studies have demonstrated that a certain proportion of enriched recycled oceanic crust, variably enriched in volatiles, could be present in the mantle source regions of the Karoo Province (Heinonen et al., 2014), Cook Islands (Cabral et al., 2013), Hawaii (Sobolev et al., 2005) and Iceland (Shorttle & Maclennan, 2011).
Melt inclusions in olivines in both ore-bearing (Talnakh) and barren (Zelenaya Griva and Nizhniy Talnakh) intrusions in the Norilsk region are rich in chlorine (from 0Á01 to 0Á61 wt %) and sulfur (from 0Á01 to 0Á14 wt %) (Krivolutskaya, 2001).High halogen contents are also observed in melt inclusions in basalts from south and NE of the Siberian Craton: plagioclaseand olivine-hosted melt inclusions from the Ust-Ilimsk and Bratsk dolerite sills (Fig. 1) have chlorine contents of 0Á01-0Á78 wt %, sulfur contents of 0Á01-0Á17 wt % and fluorine contents of 0Á02-1Á95 wt % (Black et al., 2012).Such high volatile contents have been linked to the assimilation of Cambrian evaporites, which are predominant in the south of the platform, into which the magmas were intruded (Black et al., 2012).The origin of high fluorine concentrations in those melt inclusions is inconsistent with salt assimilation as seawater does not contain sufficient fluorine.Fluorine may have come from sulfate evaporites, sedimentary phosphorite deposits or fluorite mineralization, which is found locally on the Craton.Variable assimilation of evaporites is supported by a range in Cl/K ratios at a constant K 2 O content of the melt inclusions and by the circumstantial evidence of thick salt layers in the south of Tunguska Basin, from which these samples originated (Black et al., 2012).Volatile abundances are similarly high in melt inclusions in meimechites located in the NE of the region (Fig. 1; Black et al., 2012).These highly magnesian alkaline rocks are thought to have formed at the lowest degrees of harzburgite melting at a depth of >200 km (Arndt et al., 1995(Arndt et al., , 1998)).Melt inclusions in olivines from these rocks contain 0Á02-0Á51 wt % S, 0Á01-0Á15 wt % Cl and 0Á02-0Á16 wt % F on average (Black et al., 2012).Low degrees of peridotite or harzburgite melting at high pressures of 6-8 GPa (Arndt et al., 1998) and delaminated lithosphere mixed with metasomatized harzburgite (Sobolev et al., 2009) have been proposed to contribute to the volatile budget of the meimechite melts.Overall, however, these extremely volatile-rich magmas (meimechites and sills intruded into Cambrian evaporites), constitute less (perhaps much less) than 10% of the total magma volume of the SLIP (Vasiliev et al., 2000) and are not representive of the vast volumes of tholeiitic Siberian magmas.
In this study we report whole-rock and melt inclusion compositional data for the Khakanchansky, Ayansky and Khonnamakitsky Formations erupted in the region SE of Norilsk (Figs 1 and 2).This represents the first melt inclusion study of the magmas from the low-Ti series.The lavas and sills sampled for this study were not directly emplaced into or erupted onto evaporite deposits (Fig. 1); however, we acknowledge that the magmas could have potentially encountered evaporites at depth.As a result, we would not expect evaporite assimilation to have played an important role in their petrogenesis.The magmatic rocks analysed here are representative of the vast volumes of tholeiitic lavas erupted as part of the main magmatic pulse, emplaced after the pyroxenite-derived melts analysed by Sobolev et al. (2009).A major aim of this work is to assess whether these magmas were as volatile-rich as the melts inferred to be derived from a mantle source containing recycled oceanic crust that erupted earlier in the SLIP sequence (Sobolev et al., 2009) or whether they have a more depleted signature, consistent with large degrees of melting of peridotite.We use the trace element and volatile geochemistry of the melt inclusions to deconvolve mantle-derived, primary melt compositions and heterogeneity from the effects of crustal contamination, fluid signatures and degassing.Understanding the role of heterogeneous mantle plumes in supplying volatiles to the atmosphere and the time-evolution of large and sustained periods of volcanism is of great importance for explaining the contemporaneous environmental degradation.

Sampling area
For this study, we use samples of sills and lava flows from Khantayskoe lake, Dupkin lake, and the Kureika and Lower Tunguska river regions south of Norilsk (Fig. 1).The samples were collected from surface outcrops (see Supplementary Data A1 for geographical coordinates) during a 2010 expedition led by the University of Oslo.The basalts and dolerites belong to the Khakanchansky, Ayansky and Khonnamakitsky Formations in the Norilsk and Putorana regions (Figs 1 and 2) (see also Supplementary Data A1 for more details about the samples; supplementary data are available for downloading at http://www.petrology.oxfordjournals.org).The crystals hosting the melt inclusions were picked from two lava flows (samples S10-3 and S10-19) and one sill (sample S10-44) from three locations (Khantayskoe lake, Dyupkin lake and the Lower Tunguska river).

Bulk-rock analysis
Nineteen samples were analysed for bulk-rock major and trace elements by inductively coupled plasma atomic emission spectrometry (ICP-AES) and inductively coupled plasma mass spectrometry (ICP-MS) at the University of London, Royal Holloway.Powders for geochemical analyses were prepared from about 10 g of fresh rock chips using a tungsten Tema mill; 0Á2 g of powdered sample was weighed into a graphite crucible and 1Á0 g of LiBO 2 was added.The powders were carefully mixed and fused at 900 C for 20 min, then dissolved in 200 ml of cold 5% nitric acid.Ga was added to the flux to act as an internal standard.This solution was then analysed for Si, Al and Zr by ICP-AES using a Perkin Elmer Optima 3300R system calibrated with natural and synthetic standards.The solution was also used to analyse for Cs, Nb, Rb, Ta, Th, Tl, U, Y, and rare earth elements (REE) by ICP-MS on a Perkin Elmer Elan 5000 system calibrated with natural and synthetic standards.Another batch of 0Á2 g of powdered sample was dissolved in 6 ml of HF and HClO 4 (2:1 mixture), evaporated to dryness, cooled and dissolved in 20 ml of 10% HNO 3 .This solution was analysed by ICP-AES for Fe, Mg, Ca, Na, K, Ti, P, Mn, Ba, Co, Cr, Cu, Li, Ni, Pb, Sc, Sr, V, and Zn.The details on accuracy and precision of the measurements are provided in Supplementary data A2.

Micro-analytical methods
Clinopyroxene and olivine grains were hand-picked from crushed and sieved whole-rock material in the size fraction 250-500 mm.All the melt inclusions in the selected clinopyroxenes and olivines were wholly crystalline.Rehomogenization was carried out in a 1 atm vertical gas-mixing furnace.Clinopyroxene and olivine grains were wrapped in a platinum foil envelope and suspended in the furnace with a platinum wire.In the furnace, the samples were heated to the target temperature (1190-1200 C), held there for 20 min and then quenched rapidly in water.The oxygen fugacity was maintained at FMQ -1 (where FMQ is fayalite-magnetitequartz) with a mixture of CO and CO 2 gases.The grains did not oxidize and crystalline melt inclusions became fully homogeneous and crystal-free, although in some melt inclusions a bubble remained after rehomogenization (Fig. 3b).Single grains were mounted in stubs with epoxy resin and polished to expose the melt inclusions.
To acquire a complete geochemical analysis, including volatile, major and trace elements for the same melt inclusions, which were largely 10-20 mm in diameter, trace elements and volatiles were measured first by secondary ion mass spectrometry (SIMS) using an oxygen beam, followed by electron probe microanalysis (EPMA) for major elements and chlorine.Microanalysis of 43 rehomogenized melt inclusions hosted by clinopyroxene was carried out.
Melt inclusions were analysed for sulfur, fluorine and carbon dioxide, on a Cameca IMS-1270 system, followed by trace element analysis on a Cameca IMS-4f system at the University of Edinburgh.The stubs with mounted grains were gold-coated to remove the extra surface charge and loaded into the chamber 24 h in advance to outgas and create sufficient vacuum.For the measurements of volatile elements melt inclusions were bombarded with a 15-20 mm diameter beam of oxygen O 2 -with 12Á2 kV primary voltage and 3Á3 nA current.To remove any surface contamination the beam was rastered over the surface for 60 s with a 10 mm raster and only the last 10 cycles out of the 20 measured were used.The same pits in the melt inclusions were then used for the analysis of trace elements.Negative oxygen O -was used as a primary ion beam of 10Á8 kV and a beam current of 6Á6 nA.The secondary ions were accelerated at a voltage of 4Á5 kV minus a 75 eV offset.The calibration of volatiles was performed using the standards VG-2, VG-A99, NIST-610 and BCR-2G reported by GeoRem (http://georem.mpch-mainz.gwdg.de).The detection limits for carbon dioxide, fluorine and sulfur were less than 10 ppm.Thus, the errors for sulfur and fluorine averaged 2Á41% and 9Á59%, respectively, for glasses with less than 200 ppm of each volatile (see Supplementary Data A3 for calibration details, accuracy and precision of S, F and CO 2 measurements).The analyses of standards indicated different accuracies for the trace elements varying from 15Á4% for Ba to 0Á75% for Yb (Supplementary Data A4).The analytical precision is better than 3% for all trace elements except Sm, Dy, and Tm, for which it is up to 4Á8% based on repeat measurements.
Major element compositions of host clinopyroxenes and melt inclusions were determined by electron microprobe (Cameca SX100) at the University of Cambridge.Glass analyses were performed with a spot size of 5 mm, an operating potential of 15 kV and a beam current of 10 nA.Mineral analyses were performed with an operating potential of 15 kV, beam current of 20 nA and a beam size of 1 mm.Counting times for clinopyroxene analyses for different elements were 10 s for Na and K, 20 s for Fe and Si, 30 s for Mg, Al, Cr and Mn, 60 s for Ni and Ca, and up to 120 s for Ti.Melt inclusions were analysed for a longer time: 150 s for Mg and Al, 60 s for Ti and Ca, 120 s for Fe, 90 s for Cr and Ni, 30 s for P and Mn, 20 s for Si, 10 s for Na and K, and up to 260 s for Cl.Calibration standards were as follows: jadeite for Na, periclase for Mg, Si glass for Si, K-feldspar for K, rutile for Ti, fayalite for Fe, corundum for Al, apatite for P, pure metals for Cr, Mn and Ni, and halite for Cl.Data reduction was performed using the inbuilt Cameca X-Phi PeakSight software for glass analyses and PAP corrections for mineral analyses.Analyses with totals outside the range 98Á5-100Á5 wt % were discarded.For major elements the error of the measurement did not exceed 5%.For chlorine the detection limit was 150 ppm and the precision was 8Á55%.

Post-entrapment modification of melt inclusion geochemistry
Rehomogenization in a furnace allows analysis of glassy melt inclusions for major and trace elements, as well as for volatiles.We neglect the effect of postentrapment crystallization (PEC) and Fe loss (Danyushevsky et al., 2000(Danyushevsky et al., , 2002) ) that affects only melt crystallization on the inclusion rims.The possible congruent melting of the host mineral owing to overheating during homogenization of the melt inclusions in the furnace may lower the concentrations of some elements.This factor, if taken into account, would result in lower absolute concentrations of volatile and trace elements, depending on the amount of host mineral added to the melt.A second process will have an identical consequence: owing to the size of the inclusions (10-20 mm) being similar to the beam size, overlapping of the beam onto the host mineral during SIMS microanalysis should be considered.The effect of clinopyroxene host dilution (during reheating) or beam overlap is greater for those elements that are highly incompatible in clinopyroxene [e.g.light rare earth elements (LREE), Ba, Nb]; the decrease in content of these elements is proportional to the dilution factor.For those elements with high clinopyroxene-melt partition coefficients [e.g.heavy rare earth elements (HREE)], the dilution effect is less significant, especially for melt inclusions with lower enrichment in trace elements (Supplementary Data A5).The absolute difference in concentrations is more significant for highly enriched melts (e.g.S10-3 mi6).Thus, if 30%, which is considered to be a maximum probable proportion, of host clinopyroxene is overlapped or diluted into the inclusion, the relative decrease in concentrations of LREE is about 24-29Á5%, whereas for HREE it ranges from 3 to 20% for depleted and enriched melt inclusions, respectively.The changes in trace element ratios are less significant: La/Yb is 13-16%, La/Sm 8-15% and Gd/Yb is 1-2% lower for highly and moderately enriched melt inclusions and 8% higher for depleted melt inclusions.Such variations owing to overlapping or dilution of the host clinopyroxene are substantially lower than the variability observed in the analysed melt inclusions.If the actual analyses demonstrate overlapped or diluted inclusions, then the true melt compositions would have been even more enriched in highly incompatible trace elements.Moreover, rehomogenization experiments did not influence the geochemistry of the inclusions beyond a thin boundary layer at the edges, according to the profiles through the melt inclusion (Supplementary Data A6).
During rehomogenization of the melt inclusions at liquidus temperatures, loss of volatiles by diffusion through the host crystal may occur.Although research has been carried on the diffusion rates of volatiles in silicate melts (Baker et al., 2005;Freda et al., 2005;Baker & Balcone-Boissard, 2009;Balcone-Boissard et al., 2009), very few studies have investigated the diffusion of S, Cl and F in silicate minerals.It seems likely that fluorine, similar to hydrogen, may diffuse from the melt through olivine (and probably also clinopyroxene), whereas chlorine does not (Portnyagin et al., 2008;Gaetani et al., 2012;Le Voyer et al., 2014).Although the diffusivity of sulfur in silicate melts is intermediate between that of chlorine and fluorine (Baker et al., 2005), it has a relatively large ionic radius and so we assume that sulfur will not diffuse through the host crystal to any large degree on rehomogenization timescales.
All of the melt inclusions were examined thoroughly for the presence of cracks and fractures running through them.Despite inclusions affected by cracks being excluded from the reported dataset, the possibility of breaching cannot be eliminated.Contraction of magma during cooling may result in microfractures inside the crystals at the margins of the inclusions (Tait, 1992;Kent, 2008).Such cracks could further serve as pathways for hydrothermal fluids inside the inclusion, which may then change the concentrations of mobile (i.e.Na, K) and large ion lithophile elements (LILE; i.e.Ba, Rb, Sr).If these breached inclusions are then subject to rehomogenization in a furnace, volatiles may be lost from the inclusion.Low volatile contents may be a sign of breaching, but they might also be explained by preentrapment melt degassing.Trace elements in breached inclusions might show elevated concentrations of LILE, whereas REE and high field strength elements (HFSE) would not be significantly affected.However, principal component analysis performed both including and excluding volatiles did not show a component driving LILE, K and Na up whilst driving volatiles down (See Supplementary Data).We, therefore, conclude that breaching is not significant in influencing the composition of the melt inclusions described here.

Petrographic description and bulk compositions
Whole-rock compositions correspond to tholeiitic to subalkaline low-titanium basalts (Table 1).The mineral assemblage of the lava flows includes 70% clinopyroxene microphenocrysts set in a groundmass consisting of 25% plagioclase and 5% accessory minerals, such as biotite and ilmenite.Laths of plagioclase $0Á06-0Á3 mm long are partly enclosed in elongated clinopyroxene microphenocrysts, which are up to 0Á6 mm long, defining a sub-ophitic texture (Fig. 3a).Sills, as distinct from lava flows, contain $5% of euhedral olivine phenocrysts up to 0Á4 mm long with a forsterite content ranging from 46Á5 to 51Á2 mol %.The anorthite content of the plagioclase ranges from 52Á2 to 63Á8 mol % in lava flow S10-3, from 54Á73 to 74Á17 in S10-19 and from 46Á92 to 77Á36 in sill S10-25.(Supplementary Data A7).The samples from the Khakanchansky, Ayansky and Khonnamakitsky Formations are characterized by an average TiO 2 content of 1Á09 (1Á05-1Á13), 1Á05 (1Á02-1Á12) and 1Á28 (0Á92-1Á73) wt % respectively (Table 1).They belong to the low-titanium upper basaltic sequence (see stratigraphy in Fig. 2).Rare earth and trace element patterns of the analysed sills and lavas (Fig. 4) demonstrate the overall geochemical variation of the studied formations.The bulk-rocks have Mg-numbers of 53Á4-55Á9 (Fig. 5).
Most of the analysed sediments that are associated with the studied lavas and sills (Supplementary Data A7) have a geochemical composition between average continental crust (Rudnik & Gao, 2003) and extremely  enriched Bolgokhtokhsky granodiorite (Hawkesworth et al., 1995).This continental crust composition is used in the modelling described in the section below.

Melt inclusion compositions
The Mg# (Mg-number) values of the host clinopyroxenes [defined by the molar ratio 100MgO/(MgO þ FeO), assuming all iron is ferrous] vary from 69Á2 to 82Á5, with an average of 78Á7 (Table 2, Fig. 5).Magnesium numbers of clinopyroxene microphenocrysts are significantly higher than forsterite contents in olivines (Table 2, Supplementary Data A7), as such clinopyroxenes likely have trapped more primitive melts.Only a very few melt inclusions were found in olivines, which were serpentinized, therefore we further only report the analysis of clinopyroxene-hosted melt inclusions.The major and trace element compositions of the melt inclusions are reported in Table 3.The compositions of the melt inclusions as a function of their host clinopyroxene Mg# are shown in Fig. 6.More primitive clinopyroxenes, with the higher Mg#, have higher MgO and FeO contents (Fig. 6a  and b), whereas TiO 2 , K 2 O and SiO 2 , as well as incompatible trace element concentrations, increase with Oxides and total are in wt %; trace elements are in ppm.LOI, loss on ignition.decreasing host crystal Mg# (Fig. 6d-i).Volatile contents (Fig. 6j-l) are independent of clinopyroxene Mg#.REE and trace element profiles for both whole-rocks and melt inclusions are shown in Fig. 7, normalized with respect to chondritic abundances (Sun & McDonough, 1989).The compositions of local granodiorite (from Hawkesworth et al., 1995), average continental crust (from Rudnick & Gao, 2003), the clinopyroxene hosts and "primitive uncontaminated" melts in olivine-hosted melt inclusions from the Gudchikhinsky Formation (Sobolev et al., 2009) are also plotted for comparison.Compared with the uncontaminated Gudchikhinsky melt inclusions (which have La/Yb $ 5), the melt inclusions described in this study are, in general, more depleted.Figure 7a illustrates the remarkably flat REE patterns of the uncontaminated Gudchikhinsky melts in the LREE part of the spectrum, with a significant decline towards the HREE.In contrast, the melt inclusion compositions reported here have, in general, higher La/Sm and lower Sm/Yb ratios.Some of the melt inclusions are more enriched than the bulk-rock composition (shown in black), whereas others are more depleted.La/Yb in the melt inclusions ranges from 0Á7 to 9Á5, with a median of 2Á4 versus 4Á1 in the bulk dolerites (Fig. 7a).Principal component analysis (PCA) (see Discussion and Supplementary Data for more details) demonstrates that the first principal component is related to the uniform dilution or enrichment of trace elements, whereas the second principal component is associated with variation in the slope.
The melt inclusions are highly variable in their LILE contents, such as Ba, Rb, Sr and K, relative to Primitive Mantle (McDonough & Sun, 1995) (Fig. 7b).The multielement patterns of the melt inclusions demonstrate distinctive negative Nb and P anomalies, which do not appear in the Gudchikhinsky melt inclusion patterns.
The volatile concentrations in the melt inclusions are mostly low, but variable, with <890 ppm CO 2 , <510 ppm Cl, <850 ppm F and <510 ppm S (mostly < 300 ppm) (Table 3).The absolute concentrations of CO 2 are uniformly low, about 180 ppm on average (Table 3).The melt inclusions from our study are considerably poorer in volatiles than the primary Gudchikhinsky melts (Sobolev et al., 2009).We consider the volatile systematics using volatile/refractory trace element ratios.Owing to their similar compatibility in the melt, changes in these ratios indicate either mantle source composition variability, crustal contamination or magma degassing, which depletes the volatile concentration.Ratios of the volatile elements to similarly compatible trace elements show a considerable range.F/Nd ratios range from 9 to 79; Cl/K from <0Á005 to >0Á07 and S/Dy from 6 to 126.Overall, the F/Nd ratio is significantly higher than expected for primitive mantle and mid-ocean ridge basalt (MORB), whereas S/Dy is considerably lower than expected.The melt inclusions are poor in sulfur, which does not correlate with any trace element.Uncontaminated Gudchikhinsky melts are also poor in sulfur, ranging from 300 to 500 ppm for a MgO content of 9Á70-10Á05 wt % (Sobolev et al., 2009).The Cl/K ratios of the melts are mostly higher than for MORB (Cl/K $0Á03; Saal et al., 2002), even though the absolute concentrations of Cl are relatively low (Table 3).Enrichment in Cl relative to K has also been observed in both olivine-and plagioclase-hosted melt inclusions from Ust-Ilimsk and Bratsk sills (Black et al., 2012) and in the Gudchikhinsky melt inclusions (Sobolev et al., 2009).The F/Nd ratio of the analysed melt inclusions extends up to 80 (Fig. 8b), which is far beyond the uniform value of 21 defined by oceanic basalts (Workman et al., 2006) and observed for Siqueiros MORB (Saal et al., 2002).Recent experiments by Rosenthal et al. (2015) suggest that partition coefficients for fluorine are more similar to those of La than Nd.The F/La ratios of the melt inclusions range up to 315, which much higher than for MORB.

DISCUSSION
Geochemistry of the Khakanchansky, Ayansky and Khonnamakitsky Formations REE systematics of the whole-rock compositions are summarized in Fig. 9.The Khakanchansky Formation forms part of the first low-titanium sequence, together with the Tulkonsky and Nadezhdinsky Formations (Reichow et al., 2005).The La/Sm ratios of the Khakanchansky lava flows vary from low (2Á25) to high (4Á92), whereas Sm/Yb is <1Á92, in common with all magmas of the low-Ti magmatic series.Magmas with higher La/Sm, such as those of the Nadezhdinsky Formation, are thought to have experienced a high degree of crustal contamination (Reichow et al., 2005), which is also reflected in high 87 Sr/ 86 Sr (0Á70672-0Á70872) and low eNd (-11 to -8Á2 for the lower parts of the Nadezhdinsky Formation) (Wooden et al., 1993).The Khakanchansky basalts, based on their relatively low La/Sm, appear to have experienced low or variable degrees of crustal contamination.The basalts of the Fig. 5. Clinopyroxene compositions (Mg#) compared with the whole-rock Mg#, calculated with FeO set to 90% of the total iron in the whole-rock.The grey band shows the range of equilibrium clinopyroxene compositions for a given Mg-number (Mg#) when K D ¼ 0Á275 6 0Á067 (Putirka et al. 2003).The 2r errors are smaller than the symbol size.
Ayansky and Khonnamakitsky Formations belong to the upper low-titanium sequence (Reichow et al., 2005).Samples from these formations are characterized by La/Sm ranging from 2Á19 to 2Á97 and Sm/Yb ranging from 1Á21 to 1Á33, similar to those from the Morongovsky Formation (Fig. 9).Most of the Khonnamakintsky sills have low La/Sm and Sm/Yb ratios, similar to those of the Mokuaevsky, Khraelakhsky and Kumingsky Formations of the upper series and the Samoedsky Formation, which is thought to represent the top of the magmatic sequence (Hawkesworth et al., 1995;Reichow et al., 2005).Two sills (samples S10-48 and S10-62) were probably contaminated by continental crust, with La/Sm ratios in the field of those for the Nadezhdinsky Formation, which is highly contaminated by continental crust.Notably, samples targeted here for their melt inclusions (encircled in red on Fig. 9) are clearly not contaminated significantly by continental crust.

Petrogenetic relationships between crystals and whole-rocks
Melt inclusions were analysed in clinopyroxene macrocrysts.The macrocrysts may be phenocrysts (i.e. in equilibrium with the whole-rock) or antecrysts (entrained from disaggregating crystal mushes and out of equilibrium with the whole-rock).To assess whether the Oxides and total are in wt %.Average of 3-4 points analysed around each melt inclusion.
crystals were produced from the melts during equilibrium crystallization, we plot the Mg-number of the whole-rocks (which may approximate the bulk melt composition) versus the Mg-number of the clinopyroxene microphenocrysts contained within them (Fig. 5).The Fe-Mg distribution coefficient between clinopyroxene and liquid in equilibrium [K D Fe-Mg ¼ (Mg Liq Fe Cpx )/ (Mg Cpx Fe Liq )] is 0Á275 6 0Á067 (Putirka et al., 2003), which is temperature independent.The Mg-number of the whole-rock is defined as the molar ratio 100MgO/ (MgO þ FeO), assuming FeO/FeO tot ¼ 0Á9.The differences between the Mg-number of the bulk-rocks (53Á4-55Á9) and those of their clinopyroxene phenocrysts (69Á2-82Á5; Fig. 5) broadly reflects equilibrium, suggesting that the pyroxenes are not antecrysts.Most of the analysed clinopyroxene hosts to melt inclusions are in equilibrium with the whole-rock, whereas those lying below the curve can be attributed to closed-system differentiation (Putirka, 2008) and accumulation of mafic minerals, the latter resulting in elevated whole-rock Mg#.

Melt inclusion geochemistry
Below we discuss the major processes responsible for the compositions of the melt inclusions.We begin by summarizing the main features of their geochemistry and discuss the provenance of the melts with respect to their crystal hosts and the whole-rock composition.We then discuss the following as potential factors in their petrogenesis: (1) the role of mantle heterogeneity in the generation of these melts, evaluating the role of mixed peridotite-pyroxenite sources in generating some of the distinctive trace element chemistry; (2) crustal contamination; (3) evaporite assimilation; (4) the potential involvement of a fluid phase.Lastly, we discuss our findings in the context of the volatile output of the Siberian Traps volatile output and associated environmental impacts.
Principal component analysis (PCA) reveals the main controls on melt inclusion geochemistry.The first principal component elevates all the incompatible trace elements and volatiles to the same extent (see Supplementary Data), with variation related to crystal fractionation (75% of total variance; see Supplementary Data for a more detailed explanation).The major element geochemistry is dominated by trends arising from the fractionation of pyroxene and plagioclase.The melt inclusions span a large range of compositions, from primitive basalt to andesite; the latter is considerably more evolved than the bulk-rock composition, which is basaltic.
The second most important factor, according to the PCA, is a mantle melting signature that controls the slope of the REE pattern.Together these two principal components account for more then 90% of the variance in the melt inclusion geochemistry, which suggests that crustal contamination (and post-entrapment modification) is not a dominant geochemical control, because this process would generate distinctive covariances between sets of trace elements that are not observed in the dataset.The trace element compositional profiles of the melt inclusions show overall more depleted REE abundances relative to the "primary" olivine-hosted melt inclusions reported by Sobolev et al. (2009), suggesting that these melts are derived by large degrees of melting of a source composed of dominantly peridotite, with a small amount of halogen-rich pyroxenite.The primary melts were relatively depleted in incompatible elements, including volatiles, with the absolute concentrations of sulfur, chlorine and fluorine being lower than in primary melts derived by smaller degrees of melting of peridotite.Halogens are relatively enriched with respect to similar refractory trace elements, however, perhaps as a result of halogen-rich pyroxenite in the source.

Relationship between melt inclusions, the carrier liquids and their crystal hosts
The diversity of melt inclusion compositions relative to that of the bulk magma and the carrier liquid (matrix glass) is shown in Fig. 7.This is a feature observed in many ocean island basalts, MORB and continental basalts worldwide (Maclennan et al., 2003;Danyushevsky, 2004; Kent, 2008).Magma mixing diminishes the compositional variations during or after melt entrapment and therefore a larger range of melt compositions is retained in the melt inclusions (Sobolev & Shimizu, 1993;Nielsen et al., 1995;Kent, 2008).The trapped melts have also undergone variable amounts of crystal fractionation, as illustrated by major element trends versus decreasing host clinopyroxene Mg# (Fig. 6).All these factors combined lead to geochemically diverse melt inclusion compositions.

Mantle heterogeneity and degree of melting
Previous studies have considered the influence of lithological variations in mantle source regions on the geochemical composition of the melts being produced (Sobolev et al., 2005(Sobolev et al., , 2007;;Herzberg, 2006;Shorttle & Maclennan, 2011).To track the influence of SLIP mantle source heterogeneity (Sobolev et al., 2009(Sobolev et al., , 2011) ) on the trace element characteristics of the melts we have modelled the melting of two potential mantle source components: peridotite and pyroxenite.The pMELTS model was used to calculate the major and trace element composition of accumulated fractional melts produced by isentropic, adiabatic decompression of peridotite and pyroxenite with modelling performed with the alphaMELTS software (Ghiorso et al., 2002;Smith & Asimow, 2005).Peridotite compositions KLB-1 (Takahashi, 1986) and depleted MORB mantle (DMM) (Workman & Hart, 2005) were used for the starting major and trace element compositions, respectively.The pyroxenite major element starting composition in the model was hybrid Px-1, which has been proposed to guarantee a stable pyroxenite lithology (Sobolev et al., 2007).Estimates of trace element concentrations were based on the approach described by Rudge et al. (2013); that is, binary mixing of DMM (Workman & Hart, 2005) and bulk subducted igneous crust in equal proportions.The subducted oceanic crust was assumed to contain 25% altered MORB, 25% fresh normal (N)-MORB and 50% gabbro, as proposed by Stracke et al. (2003).The starting temperature and pressure were assumed to be 1500 C and 4 GPa, which are consistent with previous mantle potential temperature and depth of melting assessments (i.e.1500-1580 C and 130-180 km; Sobolev et al., 2009).
To track volatile/trace element ratios during melting we used partition coefficients for F and Cl in clino-and orthopyroxene from Dalou et al. (2014), and for olivine and garnet from Hauri et al. (2006), which were also used by Rosenthal et al. (2015).We assume initial F and Cl contents in peridotite of 16 and 0Á83 ppm, respectively, based on MORB concentrations (Saal et al., 2002).Pyroxenite is assumed to contain 140 ppm Cl, based on the assumption that the uncontaminated purely pyroxenite-derived melts of the Gudchikhinsky Formation represent a 0Á4 melt fraction, and 280 ppm F, based on a F/Cl ratio of around two in EM2-sourced Society Islands melts (Kendrick et al., 2014).We illustrate the paths of liquid composition evolution during melting based on the parameters mentioned above (Fig. 8).
To address the specifics of the Siberian Traps melts we plot Gd/Yb versus Ti/Y (Fig. 10).The high-Ti melt inclusion compositions (Ti/Y > 500) with high Gd/Yb > 3 are believed to be controlled by residual garnet in the source and represent melts of the pyroxenite component, whereas low Ti/Y and Gd/Yb ratios are attributed to higher degrees of melting of shallower peridotite (e.g.Xu et al., 2001;Kamenetsky et al., 2012).Pyroxenitederived high-Ti Gudchikhinsky melts have high Gd/Yb of $3 (Fig. 10; Sobolev et al., 2009).Most of the analysed melt inclusions have low Gd/Yb (<2) and low Ti/Y (<300), defining a positive correlation between these two ratios.Compositions plotting outside this trend suggest that other processes exert a control on dominantly peridotite-sourced melts.Continuous mixing of peridotite (low Gd/Yb and Ti/Y) and pyroxenite (high Gd/Yb and Ti/ Y) components is not capable of explaining the extreme values of La/Yb (>9 and $1) in the melts.The assimilation of siliceous crustal material, such as the extremely enriched Bolgokhtoksky granodiorite (Hawkesworth et al., 1995), is required to explain high Gd/Yb ratios at low Ti/Y (Fig. 10) This may also explain the very high La/ Yb ratios in a few of the samples (e.g.S10-3 mi 6, 8, 15).The concentrations of HREE and HFSE in evaporites are negligible (Pang et al., 2013) and their assimilation could not produce the observed compositions.
Variable degrees of partial melting of pyroxenite (>30%) and peridotite (>10%) can broadly account for the chlorine and fluorine variations in the melt inclusions, according to our model predictions, as indicated by the arrows in Fig. 8b-d.The arrows show that the ratios change with progressive partial melting of peridotite and pyroxenite, respectively.Variations in Cl/K, F/ Nd and F/La at a given value of La/Yb lie close to the model peridotite melting curves.The divergence in these ratios gravitates towards the pyroxenite melting curves.The sulfur concentrations in the melts are low and dominated by a degassing signature.

Crustal contamination and the role of a fluid-rich component
As demonstrated above, some of the geochemical diversity in melt inclusion compositions may be explained by mantle heterogeneity, except for the LILE enrichment, Nb minimum and volatile content variations.Fractionation and original heterogeneity within the primary melts of the mantle source (see above) have undoubtedly controlled a large amount of the diversity in melt inclusion geochemistry.Crustal contamination is perhaps almost inevitable during prolonged storage of large volumes of basaltic magma in the continental crust and has been shown to be a distinctive and important feature of many other continental flood basalts (Hawkesworth et al., 1995;Devey & Cox, 1987;Brandon et al., 1993;Baker et al., 1996;Ewart et al., 1998).Melt inclusions with a distinctive negative Nb anomaly probably represent melts contaminated by the continental crust (Fig. 7b), consistent with previous studies that have shown that Nb and Ti anomalies correlate with high concentrations of radiogenic Sr (Naldrett et al., 1992;Lightfoot et al., 1993;Fedorenko et al., 1996;Reichow et al., 2005).
The melt inclusions exhibit a range of Nb/Y ratios from 0Á02 to 0Á52, increasing towards higher La/Yb (Fig. 11).Some of the melt inclusion compositions may be accounted for by heterogeneous source melting (trend to high Nb/Y, high La/Yb), consistent with mixing between peridotite-and pyroxenite-derived melts, whereas others show a smaller increase in Nb/Y with increasing La/Yb, which trends towards continental crust.Some melt inclusions show high Ba/La and Sr/La uncorrelated with any other elements (e.g.Cl, K, REE).These melts may have interacted with a component enriched in these elements, which could be a crustal fluid or a low-degree metasomatic melt (Fig. 12).The role of crustal contamination in the formation of the SLIP has been previously addressed by a number of researchers (Lightfoot et al., 1993;Fedorenko et al., 1996;Reichow et al., 2005).It is crucial to discriminate here between evaporite and siliceous crustal components.Assimilation of evaporites, both sulphate-and salt-dominated, will not influence melt REE enrichment (e.g.La/Yb ratio) owing to the very low concentrations of most REE in evaporites (Fig. 13a), but will substantially affect the volatile budget of the magmas (Fig. 13b-d).A siliceous crustal component, such as the extremely enriched Bolgokhtoksky granodiorite and also average continental crust, in contrast, will not contribute to the volatile content of melts (and will in fact dilute volatile concentrations) but could promote considerable melt enrichment in LILE and LREE.Crustal assimilation, if it was a key factor in the formation of these melts, would inevitably increase La/Yb ratios, which is not observed in most of the analysed melt inclusions (Fig. 8).The observed enhancement of Cl over K in the melt inclusions might be consistent with minor interaction with evaporites, as proposed by Black et al. (2012) for elsewhere in the province, but is also consistent with a small amount of chlorine-rich pyroxenite in the source (Fig. 8c), independently corroborated by the relatively high Ti/Y and Gd/Yb in a subset of the samples (Fig. 10), as previously proposed by Sobolev et al. (2009).

Implications of these results for the volatile budget of the SLIP
The overarching conclusions of our study are that those SLIP magmas erupted 400-800 kyr after the *not measured or below detection limit Gudchikinsky Formation magmas (which have been interpreted to be generated by pyroxenite melting) were the result of melting a source that had been depleted substantially in its pyroxenite component.The overall low concentrations of volatiles in the melts, which are consistent with the low concentrations of other incompatible trace elements, are a result of the melts having been formed by relatively large degrees (>15%) of peridotite melting.Relative enrichments in Cl and F with respect to refractory trace elements with similar partition coefficients suggest either that the pyroxenite component of the source, although minor in amount, was extremely halogen-rich or that the magmas interacted with evaporites in the subsurface (evaporites do not form an outcrop or subcrop in the region).Other trace element indicators of pyroxenite involvement (e.g.elevated Ti/Y, Gd/Yb) suggest that this is the most likely explanation.Crustal contamination is relatively unimportant for these melts, but the overall effect of a small amount of contamination with siliceous crustal material is to dilute the incompatible elements in the melt, including the volatiles, further.The volatile contents of the studied melt inclusions are low; furthermore, they are lower than those in the primary melts reported by Sobolev et al. (2009), which were interpreted to have been dominated by pyroxenite melting earlier in the Siberian Traps eruptive sequence.For sulfur and carbon, the low concentrations may be explained by pre-entrapment degassing, which is extremely common in melt inclusions in continental magmas.Chlorine and fluorine, on the other hand, are not expected to be modified significantly by degassing [owing to their greater solubilities in silicate melts (e.g.Webster, 2004, but their concentrations have certainly been modified by it (eradicating any correlations with other trace elements; Fig. 8).These magmas would contribute only around 20-50% of the magmatic gases sulfur dioxide, hydrogen chloride and carbon dioxide to the atmosphere compared with those supplied by the pyroxenite-derived magmas of Sobolev et al. (2009).This equates to $70 ppm Cl in the analysed melt inclusions versus 350-400 ppm Cl for the primary melt inclusions of Sobolev et al. (2009) and, similarly, $270 ppm S versus 350 ppm in the primary melts.The volatiles supplied to the atmosphere from large melt fractions of peridotite are also relatively small in magnitude Fig. 8. Volatile/trace element ratios vs La/Yb in SLIP melt inclusions (MI) from this and other studies.Data sources: Gudchikhinsky Formation uncontaminated primitive melt inclusions, Sobolev et al. (2009); olivine-hosted MI from ore-bearing intrusions, Krivolutskaya (2001).Shown for comparison are data for Siqueiros MORB (Saal et al., 2002) and average continental crust (Rudnick & Gao, 2003).Model melting curves for peridotite (continuous line) and pyroxenite (short dash-dot line) are colour-coded for pressure, varying from 4 to 1 GPa (see text for details).compared with those outgassed from magmas intruded into volatile-rich sediments (e.g.Svensen et al., 2009;Black et al., 2012) or from alkaline meimechite magmas (e.g.Black et al., 2012).The reduced gaseous output from the magma does not, however, diminish the importance of degassing of volatile-rich country rocks owing to contact metamorphism and volatile release through venting structures (Svensen et al., 2009;Aarnes et al., 2010Aarnes et al., , 2011)).
According to the most accurate existing estimates of the regional distribution of intrusions, lava flows and volcaniclastic material, the total volume of Siberian Traps magmas emplaced on the Siberian Craton is 1Á75 million km 3 (Vasiliev et al., 2000).Taking into account the lavas and sills that occur as subcrops in the West Siberian Basin, the total volume of the SLIP may be up to 4 million km 3 (Courtillot et al., 1999;Fedorenko et al., 2000).The volume of magmas not intruded directly into evaporites (but that could have interacted with them at depth) is up to about 590 000 km 3 [estimated based on the regional distribution of magmatic rocks in the SLIP reported by Vasiliev et al. (2000) compared with the known distribution of evaporites; Fig. 1], forming nearly one-third of the total magma volume on the Craton.The total volume of volatile-rich "'pyroxenite-derived'" melts of the high-Ti series is less than 25% of the total volume.The magmatic contribution to the volatile budget of the SLIP, if it is assessed based on extrapolating the compositions of volatile-rich melts from Norilsk (Sobolev et al., 2009), the Maymecha-Kotuy Province (Black et al., 2012) and the southern Cambrian evaporite region in the province (Black et al., 2012;Tang et al., 2013), may thus be highly overestimated.Assimilation of volatile-poor siliceous continental crustal material may have affected large volumes of the low-Ti series magmas, which were dominantly derived from a peridotite source (Fig. 2; Reichow et al., 2005), which would serve to further dilute their volatile concentrations.It is presumed that the magmas enplaced in the West Siberian Basin were largely peridotite-derived, similar to the samples studied here.This means that about two-thirds of the Siberian Trap magmas, taking into account those in the West Siberian Basin, may have been poorer in volatiles compared with the volatile-rich melts analysed by Sobolev et al. (2009) and Black et al. (2012), which should be taken into account in volatile budget estimations.
Figure 14 shows a schematic illustration of the gaseous magmatic output of the SLIP over time, incorporating our interpretation of mantle source evolution.The first intensive volatile output, which occurred  (1990,1993); Wooden et (1993); Hawkesworth et al. (1995).Model curves for partial melting peridotite (continuous line) and pyroxenite (short dash-dot line) are colour-coded for changing pressure from 4 to 1 GPa.Melt fractions are Samples analysed for melt inclusion composition are highlighted.
before the main pulse of magmatism, came from purely pyroxenite-sourced magmas, which were rich in volatiles (particularly chlorine and carbon species; Sobolev et al., 2009Sobolev et al., , 2011)).With time, the pyroxenite component in the heterogeneous mantle source became exhausted and the source eventually became dominated by peridotite.The melt inclusions reported in our current study preserve a signature suggesting a small proportion of pyroxenite component in the source (see discussion above).According to the latest geochronology studies (Kamo et al., 2003;Song et al., 2012;Burgess et al., 2014;Burgess & Bowring, 2015),   (Sobolev et al., 2009), average continental crust (Rudnick & Gao, 2003) and average N-MORB (Hofmann, 1988).Model curves showing partial melting of peridotite (continuous line) and pyroxenite (short dash-dot line) are colour-coded for pressure, varying from 4 to 1 GPa.
both the latest Permian and earliest Triassic mass extinctions occurred after the emplacement of magmas derived largely from a pyroxenite mantle source.The volatile output of these early high-Ti magmas (Fig. 2) was much greater than that of the meimechites and alkaline magmas erupted later in the sequence, which became volatile-rich owing to their emplacement in evaporites (Fig. 14).The low-Ti basaltic magmas that make up most of the SLIP were derived from a depleted peridotite mantle source and emitted considerably smaller mass burdens of gases.Thus mantle source heterogeneity created a sequence of volcanism that began with immense emissions of climate-altering gases into the atmosphere (associated with melting of a pyroxenite source component) and, with time, gradually evolved into eruptions of more depleted melts with a much lower volatile output.Mapping of mantle composition directly onto atmospheric volatile fluxes during volcanism has been suggested recently for modern-day eruptions (e.g.Iceland, Kilauea) and may be highly significant for our understanding of the environmental impact of long periods of intense volcanism in the geological past.

CONCLUSIONS
The present study was designed to determine the effect of intrinsic mantle source heterogeneity on the  compositions of SLIP magmas and their volatile outputs.It has been shown that sills and lava flows erupted to the SE of Norilsk were formed from melts that were subject to differentiation, a small amount of crustal contamination, interaction with crustal fluids and degassing (of carbon and sulfur species).Despite all of these superimposed processes, evidence for the mixing of melts derived from differing degrees of melting of a heterogeneous mantle source is preserved.Moreover, this is the second most important parameter controlling the compositional variability of the melts after fractional crystallization.Mixing of melts derived from variable degrees of melting of peridotite containing a small amount of halogen-rich pyroxenite can account for most of the geochemical features observed in the melt inclusion dataset.Overall, the melts are more volatilepoor than those reported from the Gudchikhinsky Formation by Sobolev et al. (2009) and from the Bratsk and Ust-Ilimsk sills by Black et al. (2012).In the former case it is the result of both an increased proportion of peridotite in the source and a significant amount of dilution by assimilation of volatile-poor continental crust, whereas in the latter it is due to a lack of contamination by evaporites.We suggest that these volatile-poor magmas may have been widespread across the Siberian Craton and that the overall volatile budget of the Siberian Traps Large Igneous Province may have been previously overestimated.More accurate estimates of the magmatic contribution to the total volatile budget require careful asessment of the overall volumes of volatile-rich melts, including those sourced from pyroxenite and those emplaced into volatile-rich sediments.(green lines and pale green shaded fields).Ages for the onset and the end of magmatic activity are from Kamo et al. (2003) and for meimechite emplacement from Arndt et al. (1998).High-Ti series magmas make up less than 1 million km 3 , which is less than a quarter of total magma volume.The onsets of both the latest Permian and the earliest Triassic mass extinctions occur after the emplacement of high-Ti pyroxenite-derived chlorine-and carbon-rich magmas.Two-thirds of SLIP magma volume were emplaced before or during the extinction (Burgess & Bowring, 2015).

Fig. 1 .
Fig. 1.Schematic geological map of the Siberian Craton and the Tunguska Basin, showing the distribution of outcrops of sills, lava flows and volcaniclastic deposits [modified fromMalich et al. (1974)].The distribution of Devonian and Cambrian evaporites is fromZharkov (1984) andPetrychenko et al. (2005).Sampling areas indicate the locations of samples from which melt inclusions were analysed.

Fig. 6 .
Fig.6.Major (a-f) and trace (g-i) element and volatile (j-l) contents of melt inclusions vs Mg# of host clinopyroxene.

Fig. 10 .
Fig. 10.Gd/Yb vs Ti/Y for analysed cpx-hosted melt inclusions (MI) colour-coded for La/Yb variations.A model curve showing melting of peridotite is colour-coded for pressure, varying from 4 to 1 GPa.Melt fractions are indicated.Shown for comparison are the average composition of continental crust and the Bolgokhtokhsky granodiorite.

Fig. 11 .
Fig. 11.The variation of Nb/Y vs La/Yb in the analysed cpx-hosted SLIP melt inclusions (MI) from this study compared with MI from the Gudchikhinsky Formation(Sobolev et al., 2009), average continental crust(Rudnick & Gao, 2003) and average N-MORB(Hofmann, 1988).Model curves showing partial melting of peridotite (continuous line) and pyroxenite (short dash-dot line) are colour-coded for pressure, varying from 4 to 1 GPa.

Fig. 12 .
Fig.12.Nb/La vs Ba/La, colour-coded according to Sr/La.Variability of large ion lithophile and volatile elements is subtracted from fractionation by dividing Nb, Ba and Sr by La.

Fig. 14 .
Fig. 14.Magmatic contribution to the total volatile output and mantle source evolution over time.The ages of the mass extinctions are from Song et al. (2012) (orange vertical lines with pale orange shaded field representing error bars) and Burgess et al. (2014) (green lines and pale green shaded fields).Ages for the onset and the end of magmatic activity are from Kamo et al. (2003) and for meimechite emplacement from Arndt et al.(1998).High-Ti series magmas make up less than 1 million km 3 , which is less than a quarter of total magma volume.The onsets of both the latest Permian and the earliest Triassic mass extinctions occur after the emplacement of high-Ti pyroxenite-derived chlorine-and carbon-rich magmas.Two-thirds of SLIP magma volume were emplaced before or during the extinction(Burgess & Bowring, 2015).

Table 1 :
Major and trace element composition of basalt flows and intrusions

Table 2 :
Host clinopyroxene major element composition

Table 3 :
Major and trace element compositions of clinopyroxene-hosted melt inclusions and host clinopyroxenes