. Glass inclusions can also retain the isotopic composition of the preeruptive magma at the moment of entrapment, which have been used to infer petrogenetic processes and contributions from subducting slabs, mantle sources, or crustal assimilates (e.g., Hauri, 2002;Saal et al., 1998;Shaw et al., 2008;Walowski et al., 2015). Most important for this contribution is that the initial preeruptive hydrogen isotope composition of a magma can be used to characterize the source of the magmatic water, and infer the tectonic processes that provide it. However, this preeruptive hydrogen isotope composition may be overprinted by fractionation caused by degassing (e.g., Newman et al., 1988;Taylor et al., 1983).
During degassing, the hydrogen isotope composition of a magma will move to more negative δD values as D preferentially partitions into the exsolved vapor, leaving a residual melt enriched in H (e.g., Dunbar & Kyle, 1992;Newman et al., 1988;Taylor et al., 1983). The systematic decrease in H 2 O wt.% and D/H ratios documented in dense, glassy silicic pyroclasts is one of the best established petrologic records of eruptive degassing (Chaitén and Cordón Caulle [Castro et al., 2014]; Little Glass Mountain, Newberry Crater, Glass Creek [Taylor et al., 1983]; Mazama [Mandeville et al., 2009]; and Mono Craters [J. D. Barnes et al., 2014a;Newman et al., 1988]). This record has been explored using closed-and open-system degassing models to better understand the behavior of volatiles in the conduit environment, from the explosive onset to the effusive waning stages of an eruption (e.g., Rust et al., 2004). What more can we learn by looking deeper? Preeruptive degassing in the magma reservoir is known to occur, but the style and magnitude are difficult to constrain. The hydrogen isotope composition of glass inclusions may be used to discern if closed-or open-system behavior of the vapor operates in the reservoir. Because closed-system degassing produces little isotopic fractionation, an additional opportunity may exist to characterize the source of water in the magma reservoir using the relatively unaltered hydrogen isotope composition of the melt.
To investigate the source, behavior, and composition of water in a rhyolitic reservoir, we measured the hydrogen isotope composition of a suite of quartz-hosted rhyolitic glass inclusions from the early Bishop Tuff, Long Valley Caldera in eastern California, United States. The combined analyses of D/H (‰) and H 2 O contents (wt.%) for a population of inclusions is the first such data set reported for a rhyolitic system, and the first report of D/H in quartz-hosted glass inclusions. We calibrate our measurements with a variety of natural and synthetic rhyolitic materials designed to mitigate matrix effects and analytical artifacts. Bishop Tuff inclusions are unmodified by magmatic degassing, diffusive loss, or assimilation-fractional crystallization processes. The hydrogen isotope composition of the rhyolitic Bishop Tuff is most comparable to basaltic glass inclusions from subduction zones worldwide.
Quartz crystals were handpicked from a sample of loose, disaggregated pyroclastic fall collected from the lower 8 cm of the basal horizon of the Bishop Tuff at Chalfant Valley pumice quarry, Long Valley Caldera (Figure 1) (e.g., Wilson & Hildreth, 1997). The quartz was then submerged in mineral oil and inspected BEFUS ET AL.
2 of 22 10.1029/2020GC009358
Figure 1. a) Regional geologic map of Long Valley Caldera in eastern California, United States, showing the sample location (star, at 37°27'38" N, 118°21'60" W) (modified from Wilson and Hildreth [1997]). (b) Schematic measured section of the Bishop Tuff at Chalfant Valley pumice quarry. Quartz crystals were collected from the lowest 8 cm of the basal fall (star). with a binocular microscope. We selected a separate of quartz crystals containing polyhedral-shaped glass inclusions ≥40 μm in size, avoiding those containing vapor bubbles, crystallites, or intersected by cracks.
The position and dimensions of each inclusion were measured optically. Analyzed inclusions were located 50-300 μm from the closest crystal face. The horizontal dimensions were measured using the binocular eyepiece reticle on an optical microscope, whereas the thickness of each inclusion (e.g., vertical dimension) was measured by focusing on the top and bottom of the inclusion using a petrographic microscope equipped with a Heidenhain focus linear drive encoder. Individual crystals that met the above specifications were mounted in Crystalbond, adhered to glass slides, and carefully polished to produce thin crystal wafers holding one or more doubly exposed, doubly polished inclusions (Figure 2). The inclusion-bearing quartz wafers were removed from the glass slides and the Crystalbond was removed by soaking in acetone and ethanol.
We selected one synthetic and six natural rhyolitic glasses as calibration material for the hydrogen isotope measurements. Throughout the remainder of the manuscript we refer to these glasses as "standards," BEFUS ET AL.
3 of 22 10.1029/2020GC009358 although we acknowledge that they have not been analyzed by multiple, independent laboratories, and that uncertainty remains regarding their spatial homogeneity. Iceland Deep Drilling Program (IDDP) is a hydrous rhyolite glass actively quenched during drilling of the Krafla, which contains ∼2 wt.% H 2 O and δD -114‰ ± 2‰ (Martin et al., 2017;Zierenberg et al., 2013). The North Mono Craters, CA, eruptions at 1325-1340 AD is the source for the remaining natural rhyolitic samples (Sieh & Bursik, 1986). The samples were collected and analyzed using Fourier-transform infrared (FTIR) and Thermal Conversion Elemental Analyzer (TC/EA) by J. D. Barnes et al. (2014a). P2N, P2D, and P10I are pyroclastic obsidian chips recovered from fall deposits associated with Panum Crater Dome, containing 1.9 ± 0.1, 1.2 ± 0.1, and 0.7 ± 0.1 wt.% H 2 O and δD values of -62 ± 2, -69 ± 2, and -79‰ ± 2‰, respectively. PCD5 and NWC23 are obsidian collected from the effusive Panum Crater Dome and Northwest Coulee lavas. PCD5 contains 0.4 ± 0.1 wt.% H 2 O and δD value of -101‰ ± 5‰, whereas NWC23 has 0.2 ± 0.1 wt.% H 2 O and δD value of -97‰ ± 5‰ (J. D. Barnes et al., 2014a).
B24 is an experimental glass composed of hydrated rhyolite that was synthesized for this project. The starting material was an 86-mg cylinder of glass cored from a block of holohyaline, rhyolitic obsidian from North Coulee, Mono Craters. The glass initially contained ∼0.2 wt.% H 2 O and a hydrogen isotopic composition of ∼-100‰. The glass cylinder was loaded into a 3-mm outside diameter Au capsule along with 6 mg of H 2 O with a known isotopic composition of 0.8‰ (J.D. Barnes and T. Larson standard at University of Texas at Austin Stable Isotope Lab named "Kona"). The Au capsule was welded, shut and placed within a larger 5-mm O.D. Au capsule. The space between the capsules was packed with finely powdered dry rhyolite glass and then the larger capsule was welded shut. The pressurizing medium for the experimental system was water, which was continuously in contact with the Au capsule during the experiment. H diffuses orders of magnitude faster through gold than rhyolite melt. The purpose of the packed, outer dry rhyolite powder was to act as a sponge to absorb and prevent H from the pressurizing water penetrating into the inner experimental capsule. The experiment was performed by loading the capsule and a steel filler rod into an externally heated, cold-seal pressure vessel, and then inserted into a furnace. The experiment ran at 800°C and 270 MPa for 6 days. Such conditions were chosen to ensure the melt was water undersaturated and to allow sufficient time for the D and H to diffusively equilibrate throughout the melt. The sample was quenched in ∼1 min by removing the pressure vessel from the furnace, blowing on it with compressed air, and then submerging it in a bucket of water. The capsule was then removed and weighed. Finally, the experimental glass was removed and prepared for petrographic, spectroscopic, and isotopic analyses.
The bulk hydrogen isotope ratio of B24 glass was measured using a TC/EA coupled to a ThermoElectron MAT253 Isotope Ratio mass spectrometer in continuous flow mode at UT-Austin following the methods of Sharp et al. (2001) and Cassel et al. (2012). Approximately, 7-10 mg of finely crushed material was enclosed in Ag foil capsules, dried under vacuum at 70°C for 24 h, and flushed with dry He gas within a zero-blank auto-sampler prior to analysis. Replicate aliquots of the sample were analyzed over the analytical session. The sample was calibrated using contemporaneous analyses of three internationally distributed hydrogen isotope standards: USGS 57, USGS 58, and NBS-22, as well as an internal volcanic glass standard, SN09052RW. Water content from the results of these TC/EA analyses is calculated from a linear calibration using the known wt.% H 2 O contents of standards, USGS 57 (biotite) and USGS 58 (muscovite), the mass of materialized analyzed, and the corresponding signal intensities.
We measured the dissolved H 2 O and CO 2 contents (wt.% and ppm, respectively) of the glass inclusions and rhyolite glass standards by FTIR spectroscopy using a Thermoelectron iN10. Spectra were collected in both the mid-IR and near-IR range. Concentrations of molecular and hydroxyl H 2 O in the water-rich glass inclusions and rhyolite standards were determined from absorbances at ∼5,200 and ∼4,500 cm -1 , using the calibration of Zhang et al. (1997). CO 2 contents were determined in the glass inclusions using the absorbance at ∼2,350 cm -1 with the modified Beer-Lambert law and an absorption coefficient of 1,214 ± 16 L cm -1 •mol -1 (Behrens et al., 2004). Each inclusion measurement consisted of 60 scans collected at a resolution of 4 cm -1 and a 40 × 40 μm spot size. Water in the water-poor rhyolite standards was determined using the Beer-Lambert law, absorbance at ∼3,500 cm -1 , an absorption coefficient of 71 L cm -1 •mol -1 , and a density of 2,350 kg•m -3 . Rhyolite standards were analyzed using targeted spots and area maps. Spot analyses consisted of 60 scans collected at a resolution of 4 cm -1 . Maps were collected using a step size of 50 or 100 μm to march across the sample, and then compiling the hundreds of individual analyses into a map. Map analyses used a spectral resolution of 8 cm -1 , 60 scans, and a 50 × 50 μm spot size.
After collection of FTIR spectra, the quartz and rhyolite glass wafers were mounted in indium metal for secondary-ion mass spectrometry (SIMS) analyses . The wafers were carefully arranged into the center hollow of a 1-inch aluminum round holder and pressed firmly with a vice to produce a flat surface. The surface of the sample and holder were gold coated prior to analysis.
The hydrogen isotope compositions of rhyolite standards and quartz-hosted glass inclusions were determined by analyzing D, H, and 16 O using the Cameca IMS 6F secondary-ion mass spectrometer at Arizona State University. All D/H ratios are reported in standard δD (‰) notation relative to Vienna Standard Mean Ocean Water. A focused primary beam of Cs + was accelerated to +10 keV at a current of 1-1.6 nA. The beam was focused to a spot ∼25 µm in diameter and rastered over a square area of 35 × 35 µm 2 . The sample was held at -5,000 V. Transfer optics and a small-field aperture (400 µm) were adjusted to allow collection of negative secondary ions solely from the central, circular 15-µm-diameter area of the sputtered crater. Sample charging was corrected with a normal-incidence electron gun, tuned using the technique described in Chen et al. (2013). Each analysis consisted of first obtaining a secondary-ion image for oxygen. Next, the position of the Cs + beam was adjusted to be centered on the secondary-ion optical axis. The field aperture was then inserted and the target area was presputtered for 4 min to remove surface hydrogen. The secondary magnet was cycled between the two hydrogen isotopes, counting H for 1 s and then D for 10 s. After 100 to 600 cycles the secondary magnet was re-set to count 16 O -for 5 s. A typical analysis of 100 cycles took ∼30 min to complete and attained a precision of ∼10‰ (two standard errors of the mean). The 2σ (standard errors of the mean) uncertainties on rhyolite inclusions ranged from 3‰ to 13‰. Analysis of (presumably) dry quartz in the indium mount demonstrates the background chamber counts contribute 1% to the analyses of D/H values in the glass inclusion. Background contributions to the isotopic ratios are ∼2‰, and are thus less than the 2σ uncertainty of the measurement and are not a significant source of uncertainty (e.g., J. J. Barnes et al., 2014b;Tartese et al., 2013).
The glasses and inclusions were measured during a 5-day analytical session. Standards were repeatedly analyzed throughout the session to calibrate and monitor any drift in calibration. Standards B24, P2N, P2D, and IDDP were used to establish the isotopic instrumental mass fractionation (e.g., "alpha," defined as (D/H) SIMS /(D/H) BULK ) because their elevated water contents were most similar to the water-rich inclusions (Figure 3). The calibration shifted after the third day, after a minor adjustment to the electron gun settings. We report the "first" and "second" calibrations for D/H and H 2 O. The instrumental mass fractionation factors for D/H were 1.1836 ± 0.0079 and 1.0831 ± 0.0055 for the first and second calibrations, respectively (Figures 3b and3d). To convert SIMS H -/O -measurements to H 2 O, a nonlinear second-order polynomial provides the best fit (Figures 3a and3c). Projecting our best-fit polynomial calibrations to the more elevated water contents expected in Bishop Tuff glass inclusions generates unavoidable uncertainties of +0.3 and -0.4 wt.% H 2 O in the first calibration and +0.3 and -0.2 wt.% H 2 O in the second (Figure 3).
The quartz-hosted rhyolitic glass inclusions contain 4.2-5.8 wt.% H 2 O and up to ∼70 ppm CO 2 (see Table 2, Figure 4). Those volatile contents correspond with previously published values and are thus considered representative of the early Bishop Tuff melt. Neither the H 2 O nor CO 2 contents correlate with the position of the inclusion within the host quartz.
The average hydrogen isotope composition of 28 quartz-hosted rhyolitic inclusions is -65‰ ± 15‰ (see Table 2, Figure 7). The inclusions present a spread of ∼55‰, with δD ranging from -38‰ to -92‰. Values ranging from -40‰ to -60‰ are considered most representative for the early Bishop Tuff. δD does not correlate with the inclusion's position within the quartz host, nor does it show any relationship with FT-IR-derived H 2 O content (Figure 7a). SIMS-derived H 2 O displays a negative correlation with δD in both the first and second calibrations (Figure 7b). The analytical effects that produce these differences are further discussed in Section 4.1. (b)
The reliability of the glass inclusion isotopic compositions directly depends on the rhyolite calibration standards. As such, we carefully documented the textural, volatile, and isotopic homogeneity of each. B24, IDDP, P2D, and P10I are bubble-free, dense glass with sparse microlites and microphenocrysts (Figure 5). P2N, PCD5, and NWC23 contain vesicles that occupy 5-10 vol.% of the sample. Enriched and depleted volatile haloes generated by degassing or regassing may extend hundreds of micrometers from bubble margins in obsidian chips. The diameter of the modified halo corresponds closely to the size of the bubble (Watkins et al., 2012). To avoid potential isotopic fractionations associated with late-stage, eruptive processes, we targeted SIMS spots far from vesicles (approximately hundreds of micrometers).
The FTIR-derived H 2 O content in each rhyolite glass matches the bulk measurement from TC/EA. The rhyolites are internally consistent in H 2 O, with maximum relative variability ranging from 1% to 10% (Table 1).
Where minor heterogeneity is observed, it presents spatially as gradients (P10I and B24), patches (P2D and IDDP), or bands (P2N) (Figure 5). SIMS-derived H 2 O contents have much greater scatter than independent measurements with FTIR and TC/EA, but when the analyses are considered en masse, SIMS statistically reproduces the established H 2 O contents (Figure 6a).
The bulk isotopic composition of the seven standards was established by TC/EA, and encompasses a wide range of δD values from +43‰ ± 2‰ to -114‰ ± 2‰ (J. D. Barnes et al., 2014a;Martin et al., 2017). However, the internal isotope homogeneity of these "standards" had not yet been explored. To mitigate this uncertainty, we spent more time analyzing the rhyolite glasses than quartz-hosted inclusions. Indeed, 48 out of the 76 SIMS analyses were performed on the rhyolite reference standards (Table 1). Of those, we focused primarily on the higher water content glasses (>1 wt.%) because they (B24, P2N, P2D, and IDDP) were most comparable to the inclusions. Those rhyolites display internal D/H variability <2% (by SIMS). Such consistent results, collected in dense glassy domains across the samples, suggest the rhyolites are reliable references that can be used to establish instrumental mass fractionation (particularly in the context of the samples that are the subject of this study). Conversely, our limited SIMS exploration of the rhyolites with <1 wt.% H 2 O (P10I, PCD5, and NWC23) indicate that they fail to reproduce bulk δD determined by TC/EA. These drier rhyolites may be internally heterogeneous or require a different instrumental calibration. We do not consider them further. SIMS measurements of B24, P2N, P2D, and IDDP reproduce the bulk δD established by TC/EA when multiple analyses are considered as a population (Figure 6b). Single analyses can present variability ranging up to ∼30‰.
We include experimental glass B24 as one of our rhyolite standard materials.
The Bishop Tuff is a meticulously studied series of interbedded pyroclastic fall and density current deposits (>600 km 3 ) produced by the 760 ka supereruption from Long Valley Caldera (Hildreth, 1979(Hildreth, , 2004;;Wilson & Hildreth, 1997). The units are commonly separated into an "early" and "late" petrologic classification that is based on rigorous temporal and compositional constraints, although that model has recently come under scrutiny because of assumptions regarding thermodynamic equilibrium, or lack thereof (e.g., Chamberlain et al., 2015;Evans et al., 2016;Gardner et al., 2014;Gualda & Ghiorso, 2013). derson et al., 1989;Dunbar & Hervig, 1992;Peppard et al., 2001;Roberge et al., 2013;Skirius et al., 1990;Wallace et al., 1999). Major and trace element compositions of phenocrysts and glass from the early Bishop Tuff indicate that quartz-hosted inclusions were likely entrapped at 150-225 MPa and 720°C-820°C in the presence of a preeruptive exsolved fluid phase that comprised 1-6 wt.% (≤20 vol.%) (Bindeman & Valley, 2002;Gualda & Ghiorso, 2013;Wallace et al., 1995Wallace et al., , 1999)). Unlike the late-erupted Bishop Tuff quartz, early-erupted quartz crystals seldom preserve bright cathodoluminescent rims with high CO 2 inclusions (Gualda, 2007;Peppard et al., 2001;Roberge et al., 2013). Neither bright rims nor enriched CO 2 concentrations were observed in our inclusions or host quartz. The volatile contents of our sample suite correspond with previous studies, giving confidence that our analyzed inclusions are representative parcels of early Bishop Tuff melt (Figure 4, Table 2).
The early Bishop Tuff glass inclusions indicate that the Long Valley rhyolitic melt was a hydrogen isotopic reservoir with δD value of ∼-40‰ to -60‰. The observed δD within the data set of 28 individual inclusions ranges over ∼55‰. Long-duration SIMS analyses, comprised of 400-600 cycles, generate 2σ uncertainties <5‰ (Table 2). Such precision on individual inclusions, yet large range of δD, presents a challenging data set to confidently interpret. We use the following sections to explore the data set in the context of analytical uncertainties, viability of varied isotopic fractionations, and petrogenetic processes.
Water contents as determined by FTIR represent the foundational observation for the glass inclusion data set. FTIR spot and map analyses of the standard rhyolite glasses correspond to the bulk water contents BEFUS ET AL. 9 of 22 10.1029/2020GC009358 Table 2) (Dunbar & Hervig, 1992;Lu et al., 1992;Skirius et al., 1990;Wallace et al. 1999).
Although the FTIR and SIMS produce the same overall water content when the inclusions are considered as a group, the correspondence fails when studied on an inclusion-by-inclusion basis. Many inclusions are reproduced within analytical uncertainty; however, other inclusions differ by +0.8 to -2.1 wt.% (e.g., H 2 O FTIR -H 2 O SIMS ) (Table 2). This ambiguity is well displayed when FTIR and SIMS water contents are compared to the hydrogen isotopic compositions of the inclusions (Figure 7). A prominent negative correlation between δD and SIMS-derived H 2 O is observed, implying that water-rich inclusions have the most negative δD, with values ranging from -80‰ to -90‰. Conversely, there is no discernable relationship between δD and FTIR-derived water contents.
The incongruous trends displayed by SIMS and FTIR measurements are more likely to be an analytical artifact than a result of natural processes. We acknowledge it is possible that the high-H 2 O inclusions (by SIMS) have more negative values of δD resulting from preferential in-diffusion of H as the evolving chamber becomes richer in H 2 O. However, this relationship is not observed in H 2 O contents measured by FTIR in the same inclusions. Instead, the correlative relationship is more likely caused by a matrix effect during SIMS analyses. Hauri et al. (2006) showed that as the H 2 O content of glasses increased, the measured D/H ratio became enriched in H. This outcome was a result of changing sputter yield and H ionization efficiency with increasing H 2 O content. These controls on the calculated δD values were most pronounced in Fe-poor rhyolites, like those studied here. The matrix effects generate a decrease of 1‰-5‰ for each increase in wt.% H 2 O (Hauri et al., 2006). The D/H ratios enriched in H, as measured here with higher SIMS-derived H 2 O content, are thus considered to result from a matrix effect that we could not account for in our calibration. We conclude that the early Bishop Tuff preserves no relationship between δD and H 2 O.
Rhyolite glasses with similar H 2 O contents and known D/H ratios are needed to confirm the effect of H 2 O on the relative yields of the two hydrogen isotopes. Our results suggest that the D/H ratio changes by ∼50‰ as H 2 O increases by 2.5 wt.%. Support for increasing matrix effects with increasing H 2 O content is provided by the curvature in Figures 3a and3c. A similar H 2 O-dependent relationship was observed by Hauri et al. (2002) in basaltic glasses because water impacts the microstructure of silicate glass. In a follow-up study investigating basaltic to rhyolitic glass compositions, Hauri et al. (2006) produced a linear calibration and showed that as the glass H 2 O content increases, the ratio of D to H decreases (after normalizing to bulk D/H). The implication for the isotopic composition of the inclusions is that the most trustworthy D/H measurements from our data set are from inclusions containing near 4.2 wt.% H 2 O (the H 2 O content of the D/H standard closest to glass compositions), with δD value of -40‰ to -60‰.
The difference between FTIR-and SIMS-derived H 2 O contents for a specific inclusion cannot be explained by a matrix effect like D/H. Instead, we attribute the differences to the spatial domain analyzed by each technique. FTIR measures a 40 × 40 µm area through the full thickness of the inclusion wafer (30-130 µm, or an analytical volume of ∼48,000-208,000 µm 3 ), whereas SIMS samples a 15-µm-diameter spot that sputters <5 µm depth. Hence, SIMS analyzes a much smaller volume of the inclusion (<1,000 µm 3 and <2% of the FTIR volume), providing a more spatially specific measure of its composition. (Liu et al., 2005). Previously published volatile contents from early to middle Bishop Tuff quartz-hosted glass inclusions are shown in gray to demonstrate that our data set is representative (diamonds for Skirius et al. [1990]; triangles for Dunbar and Hervig [1992], who only measured H 2 O; squares for Lu et al. [1992]; and circles for Wallace et al. [1999], n = 68).
Our analyses demonstrate that early Bishop Tuff inclusions are internally heterogeneous with respect to H 2 O content. This result is consistent with other studies. Glass inclusions are known to be compositionally heterogeneous in major and trace elements (e.g., Kent, 2008;Newcombe et al., 2014;Saper & Stolper, 2020).
Volatile heterogeneity in glass inclusions has been rarely demonstrated in the literature (e.g., Seaman et al., 2006), but it should be expected. Every glass inclusion study has shown that volatile contents vary from one inclusion to the next. The volatile contents of a magma are estimated using population averages or maximum values. This is the accepted practice even for pristine, glassy materials. Partially crystallized or bubbly inclusions are experimentally rehomogenized to better reflect initial entrapment conditions (e.g., BEFUS ET AL. 12 of 22 10.1029/2020GC009358 Skirius et al., 1990). The variability within a population is attributed to post-entrapment processes such as crystallization and open-or closed-system degassing (Lowenstern, 1995;Newman & Lowenstern, 2002). Individual inclusions in the early Bishop Tuff may be heterogeneous, even those that appear physically faultless. The internal variability of a single inclusion likely results from the profound structural effects water exerts on melt at the molecular level. Optically homogenous glasses preserve medium-range disorder on scales <100 Å, generating a mixture of water-rich and water-poor domains (Sato et al., 2018;Urakawa et al., 2020). Such structural domains subsequently influence nanolite crystallization and bubble nucleation, which will amplify nano-heterogeneities (e.g., Di Genova et al., 2018;Gonnermann & Gardner, 2013;Mujin & Nakamura, 2014). Water speciation within the inclusion will also change in response to decom-BEFUS ET AL. 13 of 22 10.1029/2020GC009358 pression and cooling rates (Wallace et al., 2003;Zhang et al., 1995). We reiterate that SIMS analyses of quickly quenched B24 glass does not display such variability. Thus, the internal variability in the Bishop Tuff glass inclusions is a coarse record of the complex interplay of processes kinetically arrested during ascent, cooling, and transformation of liquid to glass. Building on this observation will likely present rewarding opportunities for future microanalytical studies.
To further interpret the D/H data, we consider the isotopic fractionation pathways that would be produced by open-and closed-system degassing, diffusive loss, country-rock assimilation, and crystallization.
Energetic degassing of magma drives volcanic eruptions. In many magmatic systems, water is the most abundant volatile. During degassing the ratio of D to H in the residual melt changes toward more negative values as D preferentially partitions to the exsolved vapor. The fractionation follows a speciation-dependent bulk isotopic fractionation factor α vapor-melt . Importantly, α will change as a function of the amount of degassing (F) and the relative abundance of the mole fraction of H 2 O and OH -dissolved in the melt (X H2O and X OH ) (Dobson et al., 1989;Newman et al., 1988). We calculate the mole fraction of H 2 O and OH -using the speciation model of VolatileCalc assuming a rhyolite melt degassing from 6 wt.% H 2 O at 800°C (Newman & Lowenstern, 2002). The calculation for the bulk vapor-melt fractionation factor is (Rust et al., 2004) (
where the fractionation factors for α H2Ovapor-H2Omelt and α H2Ovapor-OHmelt are 0.9896 and 1.0415, respectively (after Dobson et al., 1989). Using the bulk value, fractionation during degassing can be modeled using either closed-or open-system assumptions (e.g., Batch and Rayleigh, respectively). Closed-system degassing means the exsolved H 2 O vapor stays in equilibrium with the melt:
(2)
Closed-system degassing can be visualized as bubbles of exsolved fluid maintaining chemical exchange with melt. Closed-system degassing results in a near-linear decrease in δD with a total isotopic shift of a few tens of ‰ (Figure 8). As permeability increases, H 2 O vapor may escape from the system along connected bubble pathways, conduit walls, or fumaroles. This results in open-system degassing, which can be modeled as follows:
Open-system degassing results in a pronounced decrease of δD in the residual melt (Figure 8).
Obsidian clasts in pyroclastic air-fall deposits have been shown to contain an array of volatile contents.
Those clasts are typically interpreted to represent syn-eruptively quenched parcels of magma, whose individual volatile contents record snapshots of eruptive degassing at variable depths (Castro et al., 2014;Dunbar & Kyle, 1992;Giachetti et al., 2020;Newman et al., 1988;Rust et al., 2004;Taylor et al., 1983). Populations of such clasts from single eruptions typically display open-system Rayleigh degassing trends marked by decreasing D/H values with decreasing H 2 O concentrations (e.g., Taylor et al., 1983) (compilation of all published obsidian chips in Figure 8). In some cases, a period of closed-system degassing preludes the open-system degassing signaling a change in the eruptive style from explosive eruption to quiescent dome formation (Newman et al., 1988).
Preeruptive degassing in the Long Valley reservoir must have occurred because the early Bishop Tuff is estimated to have contained significant exsolved gas, equivalent to 5-20 vol.% vapor (Wallace et al., 1995). The
style of such degassing was previously unconstrained. Open-system degassing of the reservoir should have produced a systematic isotopic shift of tens of ‰ (Equation 3). Instead, our best estimate for the composition of the early Bishop Tuff is a δD value of -40‰ to -60‰, which projects "upstream" along the trend established by obsidian clasts from pyroclastic air-fall deposits (Figure 8). The absence of an open-system degassing trend in the glass inclusions indicates the reservoir degassed as a closed system. The vapor solely existed as bubbles trapped in a closed reservoir. This suggests that exsolved gases in some magmas may be prevented from escaping to the surface. In such examples, surface fumaroles do not provide direct evidence of subsurface conditions, leaving geodetic measurements of inflation as the only monitoring tool to determine the state of magmatic degassing.
When molten, melt inclusions become physically isolated from the continually evolving bulk magmatic system. Melt inclusions are far from perfectly closed systems, however. Prior to eruption and quenching at the Earth's surface, inclusions can re-equilibrate with their host crystals and surrounding melt through diffusion. The diffusion of most major and trace element cations is limited by low partitioning behavior and slow diffusion rates in host phases like quartz and olivine. In contrast, hydrogen can diffuse on geologically rapid timescales (e.g., Hauri et al., 2002). Experimental results demonstrate that at magmatic temperatures, olivine-hosted melt inclusions can equilibrate H with the external melt via diffusion in days (e.g., Bucholz et al., 2013;Gaetani et al., 2012), and theoretical calculations suggest quartz-hosted melt inclusions would reach equilibrium with respect to hydrogen in years (e.g., Qin et al., 1992). This post-entrapment diffusive loss causes isotope fractionation, as H protons diffuse more readily than D. Diffusive loss produces inclusions with higher hydrogen isotope ratios. In olivine-hosted glass inclusions, experimental studies and natural samples impacted by post-entrapment H loss display strong negative correlations between δD and H 2 O, with >100‰ δD variation in a single suite of glass inclusions from a single sample (Bucholz et al., 2013;Gaetani et al., 2012;Hauri et al., 2002;Walowski et al., 2015).
The early Bishop Tuff glass inclusions does not display strong negative correlations between δD and wt.% H 2 O (Figure 8), suggesting that high-temperature diffusive loss is not a primary driver of measured δD variability. Selection of glassy inclusions from a pyroclastic fall deposit suggests geologically rapid quenching occurred in our samples, and thus, reduces the likelihood of diffusive H loss (Lloyd et al., 2013). Further, there is no relationship between D/H and inclusion size, position relative to crystal faces, or water speciation, suggesting negligible low-temperature diffusive loss of H 2 O m that would not fractionate D from H (Hudak & Bindeman, 2020).
Meteoric waters infiltrate and modify the composition of volcanic glass following eruptive emplacement on timescales of <10-10 4 years (Cassel & Breeker, 2017;Giachetti et al., 2020;Nolan & Bindeman, 2013;Seligman et al., 2018). The meteoric water modifies the primary magmatic hydrogen isotope composition initially preserved in volcanic glass because meteoric waters are enriched in D relative to H, with the ratio controlled by precipitation, temperature, latitude, and altitude (e.g., Cassel et al., 2009;Giachetti et al., 2020Giachetti et al., , 2015;;Hudak & Bindeman, 2018;Ingraham & Taylor, 1991;Jackson et al., 2019;Seligman et al., 2016 values enriched in H relative to D as expected on the rain shadowed, high-altitude eastern flank of the Sierra Nevada (Friedman et al., 2002;Ingraham & Taylor, 1991). Indeed, rehydrated Bishop Tuff glass has very low δD value of -134‰ (Mulch et al., 2008).
The rhyolitic glass inclusions are fully encased within a protective crystalline quartz vessel, which were each extracted from rapidly quenched pumices. Diffusive rehydration would initially enrich H over the relatively slower D, thus producing more negative δD with increasing wt.% H 2 O (e.g., Seligman et al., 2016). However, over longer time periods, the D would also have sufficient time to diffuse, thus reducing fractionation. The ∼760 ka since eruption of the Bishop Tuff certainly presents a long timescale for diffusion. Total H 2 O contents appear unmodified. In addition, the diffusive exchange of hydrogen is very slow in quartz at ambient conditions, <10 -17.5 m 2 •s -1 , although better calibrations for both D and H await future experimental constraints (e.g., Biró et al., 2017). Nevertheless, the inclusion data do not plot along a rehydration evolution trend (Figure 8). Nor does δD correlate with the position of the inclusion within the quartz host. We conclude that it is unlikely that the hydrogen isotope compositions of the glassy inclusions were modified by rehydration.
Despite debate regarding the preeruptive storage and evolution of the Bishop Tuff reservoir (e.g., Chamberlain et al., 2015;Evans et al., 2016;Gardner et al., 2014;Gualda & Ghiorso, 2013), mineralogically and geochemically differences are evident between the early and late Bishop Tuff (e.g., Hildreth & Wilson, 2007).
The early Bishop Tuff fall deposit studied here is a high-silica rhyolite with a mineral assemblage dominated by quartz, sanidine, and plagioclase (3:3:1 ratio, respectively, Hildreth, 1979), with minor amounts of biotite and magnetite, and no pyroxene (which is present in all late-erupted units). Most mineralogical arguments using those phases agree that the early-and late-erupted materials did not interact (e.g., Gualda & Ghiorso, 2013;Hildreth & Wilson, 2007), and that late-erupted materials alone show evidence for a recharge event that may have triggered eruption (Evans & Bachmann, 2013, and references therein).
The isotopic composition of the reservoir was not affected by anhydrous mineral phases. The presence of a hydrous phase, biotite, necessitates consideration of δD fractionation during crystallization. At magmatic temperatures, crystallization of biotite is likely to drive residual melt to more positive δD (biotite → H 2 O, 1,000 ln α biotite-H2O = -25.3 @ T = 700°C, Suzuoki & Epstein, 1976). Biotite is found in small concentrations in the early Bishop Tuff (<4% of the phenocryst population, which itself is <5% of the bulk, Hildreth, 1979).
It is unlikely to have changed δD values by more than a few ‰, less than the measured analytical uncertainty. Interestingly, Gualda (2007) suggested that the millimeter-scale melt heterogeneity that may have produced the bright cathodoluminescence (CL) quartz rims could be explained by decompression-driven biotite breakdown. Although local (millimeter-scale) variability in biotite stability could drive variability in δD values, the location of analyzed inclusions in crystal interiors, far from the bright CL rims, rules out this process.
Another important consideration for measured δD compositions is the impact of assimilation processes that commonly occur in the crustal reservoirs. Although some previous work suggests that assimilation of crustal material did not play a role in contributing to the compositional variability observed between the early-and late-erupted units of the Bishop Tuff (e.g., Hildreth & Wilson, 2007), contamination of the magma by surrounding country rock is likely to be significant during the assembly of any large-volume magmatic system in the upper crust. Here, the contributions are wall rocks of Mesozoic Sierran granitoids and reservoir roof of Paleozoic marine shales (Hildreth et al., 2018). Masi et al. (1981) found that Sierran granitoids display a large range in δD (-131‰ to -46‰), whereas shales typically have δD between -60‰ and -100‰ (Bindeman et al., 2016). Assimilation of these materials would lead to a magmatic composition with unchanged or isotopically lighter hydrogen.
Previous investigations of other stable isotope systems and radiogenic isotopes can provide additional insight into the role of assimilation. The surrounding country rocks have distinctly radiogenic isotope compositions, but perhaps surprisingly, Sr isotope studies find little evidence for significant crustal assimilation in the early Bishop Tuff (Halliday et al., 1984;Knesel & Davidson, 1997). Oxygen isotope studies demonstrate higher δ 18 O values than the mantle, indicative of fractionation and assimilation of a crustal component. However, oxygen isotopes rule out assimilation of materials hydrothermally altered by meteoric H 2 O, which would drive melts to compositions more enriched in H relative to D (Bindeman & Valley, 2002).
Taken together, fractionation and assimilation processes that occurred during the storage and evolution of the early Bishop Tuff could result in both positive and negative excursions in δD from initial mantle values. However, the previous mineralogical and isotopic investigations outlined above suggest that the impacts of differentiation are unlikely to have an impact on δD greater than the uncertainty and measured spread in our data set, such that the presented hydrogen isotope compositions likely represent estimates of the initial δD of the mantle-derived melts from which the rhyolitic magmas evolved.
Long Valley Caldera is not located in a tectonic regime traditionally associated with abundant magmatism (e.g., a hotspot or subduction zone). The complexity of tectonics in California throughout the past 30 Ma has led to numerous hypotheses related to the generation of magma in the Sierra Nevada and Walker Lane in the Cenozoic (Putirka et al., 2012). Various studies have concluded that initiation of the San Andreas Fault system, migration of the Mendocino Triple Junction, and sinking of the remnant Farallon slab have contributed to asthenospheric upwelling and lithospheric degradation beneath the Sierra Nevada (Putirka et al., 2012, and references therein). The petrogenesis of the large-volume Bishop Tuff reservoir thus remains a key question. The answer is inherently tied to mantle process, which may be informed by our D/H measurements. The hydrogen isotope composition of the mantle is heterogeneous and varies globally. Variability is controlled by plate tectonic environment and mantle dynamics. Analyses of glassy MORB suggest that the δD of the mantle source varies with degree of enrichment (e.g., mantle reservoirs like DMM, HIMU, EM1, and EM2) and geography, with recent estimates of δD ranging from -40‰ to -100‰ (Dixon et al., 2017, and references therein). Typical arc basalts display more positive δD than MORB. Arc basalts δD values regularly range from -10‰ to -60‰ because the mantle may be metasomatized by D-enriched slab fluids (Dixon et al., 2017;Giggenbach, 1992;Shaw et al., 2008;Walowski et al., 2015). Mantle reservoirs sampled by ocean island basalts are also diverse and have δD values that vary significantly (Figures 9 and 10) (e.g., Dixon et al., 2017;Kyser & O'Neil, 1984;Kingsley et al., 2002;Loewen et al., 2019).
Hotspot magmatism that impinges upon overriding continental crust is further complicated by assimilation and fractional crystallization. The Yellowstone hot spot is thought to have an initial magmatic δD in the range of -70‰ to -90‰ based on D/H in amphibole from the Lava Creek Tuff (Loewen & Bindeman, 2015;Martin et al., 2017). Because Long Valley Caldera shares some aspects with Yellowstone, the Bishop Tuff has been inferred to have a δD value of -80‰ to -90‰ (Martin et al., 2017). Important similarities that may generate or modify the isotopic composition of magmas do exist between these caldera systems, including the formation of a large-volume, rhyolitic reservoir by fractional crystallization of basaltic melts and assimilation within a 40-to 50-km-thick crustal filter (Chulick & Mooney, 2002;Hill, 1976;Metz & Mahood, 1991;Yuan et al., 2010). Differences also exist. Radiogenic isotope studies of Bishop Tuff magmas suggest evolution by extensive fractional crystallization of basaltic magmas derived from an enriched lithospheric mantle source (Cousens, 1996;Knesel & Davidson, 1997), whereas Yellowstone magmas are plume derived. At Long Valley, trace element enrichments in regional mantle-derived melts are thought to have been imparted to the lithosphere by dehydration of the Laramide-age Farallon slab. The hydrogen isotope composition of the Bishop Tuff presents evidence for such a persistent slab signature (e.g., characterized by δD relatively enriched in D), suggesting petrologic modifications contributed by the Farallon slab are present in the rhyolite. No estimates exist for the hydrogen isotopic composition of the mantle in the Sierra Nevada/Basin and Range region of western North America.
To further explore the significance of the hydrogen isotope composition of the Bishop Tuff glass inclusions, we compare our measurements to a global, terrestrial data set of δD measured specifically in basaltic glass inclusions (Figures 9 and 10). To date, no studies have explored more silicic compositions. Despite higher water contents, we find that the early Bishop Tuff inclusions are most similar to a compilation of arc magmas, including those from the southern Cascades (Walowski et al., 2016). δD in early Bishop Tuff inclusions is consistent with interpretations of an enriched mantle source with modifications contributed by the Farallon slab, as has been suggested for regional quaternary Sierra Nevada and Walker Lane basaltic magmas (Putirka et al., 2012). 2010), Antang, China, basalts from Wu et al. (2020, p. 105337), Aoba and Vanuatu alkali basalts from Métrich and Deloule (2014), Marianas arc units from Shaw et al. (2008), Lassen cinder cones from Walowski et al. (2015), and Cerro Negro basalt from Bucholz et al. (2013) and Gaetani et al. (2012).
Quartz-hosted, rhyolitic glass inclusions from the early Bishop Tuff display a range of δD, spanning from -38‰ to -92‰. The most negative values arise from analytical effects. δD values ranging from -40‰ to -60‰ are considered to be the most representative of the early Bishop Tuff magma reservoir. Comparisons between FTIR-and SIMS-derived wt.% H 2 O indicate the glass inclusions are internally heterogeneous with respect to water content. Such heterogeneity is attributed to structural changes during quenching. Glass inclusions preserve no systematic decrease in δD values with decreasing water contents, as would be expected with open-system degassing processes in the reservoir. This suggests that the Long Valley reservoir behaved as a closed system and exsolved gases remained trapped in the reservoir as bubbles. There was negligible degassing of volatiles to the surface, which may inform modern monitoring efforts of fumarolic systems at Long Valley and elsewhere. There is a similar lack of evidence for isotopic fractionation by rehydration, fractional crystallization, or assimilation. The D/H values of early Bishop Tuff glass inclusions thus represent an isolated isotopic reservoir. Through a comparison with a global compilation δD measured in basaltic glass inclusions, our results support a subduction-related source, perhaps modified by contributions from the Farallon slab.
10.1029/2020GC009358