Supplementary material 1 – Analytical techniques

Fourier transform infrared spectroscopy (FTIR)

H2O and CO2 contents of groundmass glasses and some melt inclusions were determined by Fourier Transform Infrared (FTIR) spectroscopy. The thickness of each doubly polished ground mass glass wafer was measured with a Mitutoyomicrometer whereas glass thickness in melt inclusions was measured optically using a universal stage. FTIR measurements were carried out using a Bruker IFS 66 FTIR spectrometer at the University of Iceland and a Bruker IFS 120 HR spectrometer at BayerischesGeoinstitut, Bayreuth, Germany. Both spectrometers are equipped with an infrared microscope. The resolution of each measurement was 4 cm–1 and spot size varied from 30 to 100 µm. Four hundred scans were collected for each spectrum. Spectra were collected on both the melt inclusions, and the host crystals immediately adjacent to the inclusions. Glass spectra were then corrected for contamination by the host mineral by subtracting the host crystal spectrum from the mixed spectrum of the melt inclusion analysis. Following this subtraction, a flat background was fitted below the OH- and CO3- bands. The maximum heights of the 3500 cm-1H2O+OH- and 1420 cm-1 carbonate bands were used to determine H2O and CO2 concentrations. Extinction coefficients determined for MORB by Shishkina et al. (2014) were applied.

Raman spectroscopy

Fluid inclusions within minerals, and melt inclusion-hosted bubbles, were analysed by confocal Raman spectroscopy using the Horiba JobinYvonLabRAM HR800 Raman instruments at the BayerischesGeoinstitut, Bayreuth, Germany, and the ELTE TTK FFI Instruments Center of the Eötvös University Budapest, Hungary (e.g.: Ferrero et al. 2014). Measurement conditions were identical in both labs: we used a 514 nm Ar laser with 200 mW output power, 1800 mm grating, 500 µm confocal pinhole, and a microscope equipped with a 50x long-distance objective. All analyses were carried out at room temperature. Spectra were collected between 1000 and 4200 cm-1, integrating 2-4 repetitions of 30-200 s measurements.Peak positions were calibrated according to the Rayleigh line (0 cm–1).

Electron microprobe analysis (EPMA)

Electron microprobe analyses (EPMA) were acquired with the JEOL JXA-8230 Superprobe at the Institute of Earth Sciences, University of Iceland. Major and minor element concentrations were determined in glasses, melt inclusions and their host macrocrysts. Volatiles S and Cl were also measured in melt inclusions and glasses. For all analyses the operating potential and the beam current were 15 keV and 10 nA respectively. A spot size of 10 µm was used for all analyses except the smallest (<12 µm diameter) melt inclusions, where the spot size was reduced to 5 µm. Internal data reduction was performed using the CITZAF quantitative correction program (Armstrong, 1991). Pyrite (Astimex Standards Ltd.) and Meionite (NMNH R6600) were used as standards for S and Cl, respectively. For both S and Cl, the counting time was 100 s on the peak and 100 s for each of the lower and higher backgrounds.We used basaltic glasses A99 and VG2 as a secondary standards. The 2  uncertainty in the Figures 3-8 indicates the variation of 240 analyses for major and minor elements and of 35 analyses for S and Cl. These glass standards contain an order of magnitude higher Cl than some of the melt inclusions studied here. Therefore, we added an additional 2 estimate in Figure 6 which was calculated based on the error determined for Cl-poor (<40 ppm) melt inclusions by the microprobe software.

EPMA analyses at the University of Cambridge were performed on a Cameca SX100 instrument, using an operating potential of 15 keV, a beam current of 10 nA and a spot size of 10 µm. Major, minor and volatile (S, F, Cl) element concentrations were determined in WDS mode with peak counting time between 20 and 60 seconds depending on element abundance. The background count time was half the peak count time, or equal to the peak count time for minor elements such as phosphorous. Internal data reduction was performed using the inbuilt Cameca X-Phi PeakSight software.

Secondary ionization mass spectrometry (SIMS)

Volatile (H2O, CO2, F) and trace element analyses of Holuhraun melt inclusions and glasses were performed by high resolution secondary ion mass spectrometry on the Cameca ims-4f ion microprobe at the University of Edinburgh. Melt inclusions were selected for analysis after careful examination under transmitted and reflected light, taking care to select only naturally glassy inclusions with no cracks. Exposed melt inclusions were mounted in Epo-thin epoxy resin, and a 1 µm diamond suspension was used for the final polish. Mounts were gold-coated, and allowed to outgas at high vacuum (1 x10-7torr) in the instrument airlock for ~24 hours. The sample chamber was maintained at <6 x10-9torr during analyses. The analytical procedure followed the method described by Hartley et al. (2014). Carbon was measured separately from other volatile and trace elements, with subsequent measurements of H2O, F and trace elements made in the same pits during a second round of analyses.

Carbon analyses were performed using a 15 kV primary O- beam, with positive secondary ions accelerated to 4500 V with an offset of -50 V to suppress molecular ion interference. The beam current was maintained at 5±1 nA. The primary beam was rastered over a 20x20 µm area for 3 minutes prior to each analysis, with measurements performed using a focused beam of 15-20 µm diameter. The following isotopes were measured in each cycle of a 15-cycle run, with counting times in parentheses: 24Mg2+ (5), 12C (10), 40Ca2+ (3), 30Si (2). The first 7 cycles were discarded to minimise the effects of surface contamination. The mass resolving power of ~1600 was sufficient to resolve interference by 24Mg2+ on the 12C peak, and produced typical 12C+ ion yields of 0.1 counts per second (cps) per ppm C. The CO2 detection limit was taken to be 20 ppm, equivalent to ~2.0 cps of 12C. Background counts were monitored by analysing CO2-free standards GSA-1G, BCR-2G and matrix glasses known to be almost completely degassed. Background counts were averaged over each analytical session and subtracted from the raw data. The average background correction was equivalent to 10.6 ppm CO2.

Trace elements, H2O and F were analysed using a 15 kV primary O- beam with accelerating voltage of 4500 V minus an offset of -75±5 V, a 5±1 nA beam current and a 20±5 µm image field. Measurements were made with a spot size of ~20 µm centred in the pit made during the preceding C analyses. Isotopes were analysed in a 10-cycle run, including 1H (4), 19F (4) and 30Si (2). Masses 0.7 and 130.5 were measured as background during each cycle and were always zero. Ion yields were assessed by repeat analyses of NIST-SRM610. Repeat analyses of standard glasses BCR-2G, GSA-1G, GSD-1G and KL2-G were used to monitor analytical drift, which was shown to be insignificant. For H2O, the first 5 cycles of each analysis were discarded to minimise the effects of surface contamination. Background counts were monitored by analysing anhydrous standards GSD-1G, BCR-2G and matrix glasses known to be almost completely degassed. The average background correction was equivalent to 0.05 wt.% H2O.

Background-corrected counts on five basaltic glass standards of known composition were used to establish calibration curves for 1H/30Si and 12C/30Si versus the respective volatile component (FigS1). Similar calibration curves were established for 19F and trace elements. New calibration curves were established for each analytical session to take into account minor changes to the instrumental set-up or background conditions. Counts were normalised to 30Si to account for variations in the SiO2 content of the standards. Calibration curves were then used to calculate absolute volatile and trace element concentrations in unknown samples, first normalising 30Si counts to the SiO2 content of each inclusion or glass as determined by electron microprobe.

The 1σ reproducibility of the standard glass compositions across all sessions was 8% or better for CO2 and H2O, and 10% for F.

FIGURE S1: Calibration curves for SIMS analyses of CO2 and H2O, derived from five basaltic glass standards of known composition with volatile contents measured by FTIR (Shishkina et al., 2010). Dashed grey lines show linear fits to the standard measurements.

Figure S2: Comparison of H2O and CO2 measurements in fiveHoluhraunmelt inclusions by FTIR spectroscopy and SIMS. H2O in wt.%, CO2 in ppm.

Figure S3:Representative Raman spectra of CO2-SO2 and pure CO2 fluid inclusions.

Supplementary material 4 – Modelling of fractional crystallisation

Fractional crystallisation was modelled by Petrolog3 program package DanyushevskyPlechov (2011), using the H2O and pressure sensitive mineral-melt models of Danyshevsky (2001) and the pressure sensitive models of Langmuir et al. (1992). In both cases forward modelling was applied where starting composition was the average composition of the four most primitive melt inclusions analysed. H2O contents of primitive melt inclusions are modified by post-entrapment diffusion (see main text and Hartley et al. this issue), therefore the initial melt H2O contents were set to 0.15 wt.%. Oxygen fugacity was fixed equal atQFM+0.5 which is estimated based on olivine-spinel pair by Halldórsson et al. (this issue) and based on the fO2 sensitive sulphur solubility model of Ariskin et al. (2013) (see details below). Variation of Nb, Ce, Nd and Dy was calculated along the liquid line of descent using the partition coefficients of O'Neill and Jenner (2012) for the rare earth elements and White (2013) for Nb.To calculate fractional crystallisation paths for volatiles in Figures 3a to 8a we multiplied the trace element contents of each melt composition with the volatile/trace element ratios deduced from melt inclusions and groundmass glasses. The Dy content of the most primitive melt inclusions showed significant scatter due to low data quality. Therefore, the parental Dy content was estimated by fitting the calculated trajectories to the measured Sconcentrations, which follow a tight trend. The difference between the results of the various models is shown in Figure 3 in the main text.

Sulphur solubility in melt inclusions was calculated by the COMAGMAT5.2.1. program package computing the fO2 sensitive model of Ariskin et al. (2013). Pressure was fixed at 1.2 kbar which was deduced as potential sulphur saturation pressure based on fluid inclusions. The temperature was set to be calculated according to liquidus T. As the model is sensitive for Ni content in the melt, Ni content of each melt inclusions were calculated first (microprobe measurements had very high uncertainty). Ni contents were calculated along a liquid line of descent (LDD) by Petrolog3 (using the same conditions as above) applying partition coefficients of Laubier et al. (2014). The Ni content of the bulk rock is well characterised (Halldórsson et al., this issue), and Nicontents in some Barðarbunga glasses are well known (Óladóttir et al., 2011). Therefore, the starting Nicontents were varied such that these melt compositions should fit to the calculated LDD as they fit in case of major and other trace elements.

Fig. S4:Relationship between post entrapment crystallisation (PEC in %) and F/Nd ratios in Holuhraun melt inclusions. PEC correction described in details in Hartley et al. (this issue).Plag-plagioclase hosted, Ol-olivine hosted, Cpx-clinopyroxene hosted.

Figure S5: Sulphur concentration as a function of FeO(t) content in melt inclusions and groundmass glasses. Compositions are compared to sulphur saturated compositions (SCSS) calculated at oxygen fugacity of 0.5 logarithmic units above the fayalite-magnetite-quartz (FMQ) buffer. Note that S content increases in melt inclusions up to 12 wt.% FeO and it does not increase any more due to saturation.

References:

Armstrong JT (1991) Quantitative elemental analysis of individual microparticles with electron beam instruments. In: Heinrich, K.F.J. & Newbury, D.E. (Eds): Electron Probe Quantitation, Plenum Press, 261-317.

Ariskin AA Dayushevsky LV, Bychkov KA, McNeill AW, Barmina GS, Nikolaev GS (2013) Modeling solubility of Fe-Ni sulfides in basaltic magmas: The effect of Nickel. Economic Geology, 108, 1983-2003.

Danyushevsky LV,PlechovP. (2011) Petrolog3: Integrated software for modeling crystallization processes. Geochemistry Geophysics Geosystems, 12, Q07021.

Danyushevsky LV (2001) The effect of small amounts of H2O on crystallisation of mid-ocean ridge and backarc basin magmas. Journal of Volcanology and Geothermal Research, 3-4, 265-280.

Ferrero S, Braga R, Berkesi M, Cesare B,LaridhiOuazaa N (2014) Production of metaluminous melt during fluid-present anatexis:an example from the Maghrebian basement, La GaliteArchipelago, central Mediterranean. Journal of Metamorphic Geology, 32, 209–225.

Hartley ME, Maclennan J, Edmonds M, Thordarson T (2014) Reconstructing the deep CO2 degassing behaviour of large bassaltic fissure eruptions. Earth and Planetary Science Letters, 393, 120-131.

Hartley ME, Bali E, Halldórson SA, Maclennan J, Neave DA (this issue): Petrology and geochemistry of the 2014-2015 eruption at Holuhraun, North Iceland: Melt inclusion constraints on petrogenesis.

Langmuir CH, Klein EM, Planck T (1992) Petrological Systematics of Mid-Ocean Ridge Basalts: Constraints on Melt Generation Beneath Ocean Ridges. In: Mantle Flow and Melt Generation at Mid-Ocean Ridges. Geophysical Monograph, 71, American Geophysical Union, pp. 183-280.

Laubier M, Grove TL, Langmuir CH (2014) Traceelementmineral/meltpartitioningforbasalticandbasalticandesitic melts:AnexperimentalandlaserICP-MSstudywithapplication totheoxidationstateofmantlesourceregions. Earth andPlanetaryScienceLetters, 392, 265-278.