Magmatic water content in HIMU basalts from the Cook-Austral Islands: constraints on degassing processes and source composition from clinopyroxene phenocrysts

Abstract

The amount of water recycled during subduction is unclear. Water/Ce estimates in HIMU (high-²³⁸U/²⁰⁴Pb) basalts are variable, ranging from < 100 to > 250 in glasses and melt inclusions. Because clinopyroxene (cpx) is a common early liquidus phase and the cpx/melt partitioning of water is well constrained, cpx phenocrysts provide an additional constraint on magmatic water contents. We present water and trace element concentrations in HIMU-basalt-hosted cpx phenocrysts from the Austral Islands. Calculated melt [H₂O] (up to 4.3 wt.%) and H₂O/Ce ratios (22–825) are higher than in olivine-hosted inclusions, as are estimated equilibration pressures. Correlation between estimated melt [H₂O] and crystallization or entrapment pressure suggests significant water loss during magma ascent. Most ocean island basalts (OIBs) span a limited range in H₂O (~ 1–1.5 wt.%), and low (< 100) H₂O/Ce ratios are primarily observed in melts with unusually high [Ce] (up to 350 ppm). Additionally, [H₂O] and H₂O/Ce in some suites correlate with entrapment pressure despite having quench pressures high enough to prevent significant water loss from open- or closed-system degassing (< 100 MPa). Polybaric “sparging”, whereby low-P melts re-equilibrate with CO₂-rich fluids exsolved at higher pressure, may result in water loss at pressures less than CO₂ saturation. This may more accurately describe OIB degassing processes than open or closed system degassing. After correcting for degassing, primary Australs melts likely have H₂O/Ce of ~ 300–600. If applicable to OIB sources in general, this limits the total water budget of the mantle, including the mantle transition zone, to < 2.4 ocean masses.

Introduction

The abundance and distribution of H₂O and other volatiles in Earth’s mantle has a fundamental influence on mantle physical properties. Trace quantities of water can have a significant impact on mantle viscosity (Hirth and Kohlstedt 1996; Zhao et al. 2004), electrical conductivity (Yoshino and Katsura 2013), and strongly affect melt generation (Asimow and Langmuir 2003). Additionally, the amount of water stored in Earth’s mantle and the fluxing of water in and out of the mantle through magma degassing and plate subduction influence the size and composition of Earth’s surface hydrosphere over geologically long periods (Rüpke et al. 2006). However, many aspects of Earth’s deep water cycle are poorly constrained, with current estimates for the total quantity of water stored in Earth’s mantle ranging from < 1 to > 6 ocean mass equivalents (e.g., Hirschmann 2006; Ohtani 2003; Bercovici and Karato 2003).

Prior to subduction, oceanic lithosphere experiences low-temperature alteration due to circulation of seawater-derived fluids within the crust (and mantle) and becomes enriched in both H₂O and CO₂ over time (e.g., Staudigel et al. 1996). Much, but not all, of this H₂O and CO₂ is removed through dehydration and decarbonization reactions during subduction. However, the amount of H₂O and CO₂ retained in the slab and returned to the deep mantle is still uncertain (e.g., Cabral et al. 2014; Jackson et al. 2015). One way to constrain the amount of H₂O recycled into the mantle via subduction is to analyze [H₂O] variations in basalts that derive from mantle sources containing recycled material, such as enriched mantle (EM1 and EM2) and HIMU (high-²³⁸U/²⁰⁴Pb) mantle reservoirs.

Water and Ce are both highly incompatible during anhydrous mantle melting (Michael 1995; Danyushevsky et al. 2000), so H₂O/Ce ratios should remain relatively unchanged during mantle melting or fractional crystallization, but these species become uncoupled during magma degassing. Olivine-hosted melt inclusions and submarine glasses are often used to constrain primary magmatic [H₂O] and H₂O/Ce (E.g., Dixon et al. 2002; Cabral et al. 2014). Submarine basalt glasses and olivine-hosted melt inclusions are less affected by magma degassing than subaerially erupted basalts, but they can be affected by degassing occurring at depths greater than the depth of entrapment or eruption. H₂O/Ce ratios of glasses and melt inclusions have commonly been used to “see through” magmatic processes and provide constraints on mantle [H₂O] using independent constraints on source [Ce]. Several previous studies have reported low H₂O/Ce ratios (< 90) in melt inclusions and submarine glasses from EM-I and EM-II localities (e.g., Pitcairn, Society, and Samoa; Workman et al. 2006; Kendrick et al. 2014) relative to mid-ocean-ridge basalts (MORB) (~ 150–300; Michael 1995; Danyushevsky et al. 2000). Dixon et al. (2002) argued that the low H₂O/Ce ratios reported from these EM localities suggest high degrees of slab dehydration (~ 95–98% dehydration). Using similar methods, Bizimis and Peslier (2015) estimated [H₂O] in the EM-I mantle source (~ 160 ug/g) only slightly higher than that estimated for MORB-source mantle (~ 54–116 ug/g; Simons et al. 2002; Salters and Stracke 2004; Bizimis and Peslier 2015).

The [H₂O] in HIMU mantle, however, is poorly constrained due to large variations of measured H₂O/Ce in HIMU glasses and melt inclusions, with published ratios ranging from ~ 75 to 84 in submarine glasses from Tuvalu seamount (Jackson et al. 2015), up to ~ 245 in olivine-hosted melt inclusions from Mangaia (Cabral et al. 2014). The variability in measured H₂O/Ce in glasses and melt inclusions from the Cook-Austral Islands and Tuvalu suggest either that HIMU mantle has highly variable [H₂O], or that the measured H₂O/Ce ratios have been modified by magma degassing, brine assimilation, or other processes that can decouple H₂O and Ce abundances.

Clinopyroxene (cpx) phenocrysts have increasingly been used as an independent magma hygrometer. Experimental constraints on the partitioning of [H⁺] between clinopyroxene and basaltic melt allow magmatic water contents to be estimated from cpx [H₂O] measurements (Hauri et al. 2006; Wade et al. 2008; O’Leary and Hauri 2010; Novella et al. 2014). Clinopyroxene phenocrysts typically form at higher pressures (during magma storage) than submarine quenched glasses, reducing the effects of shallow magma degassing. Pyroxenes are often a common early liquidus phase, particularly in alkali basalts, and may crystallize over a range of pressures. Several pyroxene-based thermobarometers allow estimation of crystallization pressures from pyroxene composition, permitting investigation into degassing processes through examination of trends in measured [H₂O] with crystallization pressure. Additionally, several studies have shown that natural pyroxenes usually show less evidence of diffusive H₂O loss during magma or xenolith ascent than olivines (Hirschmann et al. 2005; Warren and Hauri 2014; Bucholz et al. 2013; Tian et al. 2017). Comparison of [H₂O] in olivine-hosted melt inclusions and pyroxene phenocrysts from the same basalts can therefore be used to examine both degassing processes prior to and during eruption, as well as possible post-entrapment diffusive H₂O loss in olivine-hosted melt inclusions.

In this study, we examine structurally bound hydrogen concentrations as well as major and trace element compositions in clinopyroxene phenocrysts from 12 basalts from Raivavae, Rapa, and Mangaia in the Cook-Austral Islands. These data are combined with existing data for melt inclusions from a subset of these samples to examine the role of processes such as magma degassing in controlling the [H₂O] and H₂O/Ce ratios of primitive magmas, and to better constrain the [H₂O] of HIMU mantle.

Background

Basalts from the Cook-Austral Islands falling along the MacDonald age progression display considerable isotopic variation, with ²⁰⁶Pb/²⁰⁴Pb ranging from ~ 19.0 to 21.9 (e.g., Woodhead 1996; Lassiter et al. 2003). In addition to HIMU-type isotopic signatures prevalent in the northern Cook-Austral Islands (Mangaia, Rurutu, Tubuai, and Raivavae), two other isotopic endmembers are apparent, including a depleted-mantle-like component present in northern Cook-Austral Islands and most prevalent in eastern Raivavae samples, and an EM- or FOZO-like component present in islands and seamounts south of the Austral Fracture Zone (Lassiter et al. 2003).

Several lines of evidence suggest that HIMU mantle contains ancient recycled oceanic crust that experienced dehydration during subduction. For example, basalts from Raivavae display a negative correlation between ²⁰⁶Pb/²⁰⁴Pb and K/U ratios, with HIMU basalts having K/U ratios extending to much lower values (< 4000) than typical upper mantle-derived basalts (~ 12–13,000) (Lassiter et al. 2003). Such correlations likely reflect preferential removal of fluid-mobile elements (e.g. K, Pb) relative to immobile elements (e.g. U) during dehydration of subducted oceanic crust, resulting in recycled components with elevated U/Pb and low K/U (Weaver 1991; Lassiter et al. 2003). Additionally, Cabral et al. (2013) reported evidence for MIF (mass independent fractionation) of sulfur isotopes in olivine-hosted sulfides from Mangaia. MIF signatures are common in sediments older than ~ 2.4 Ga, but are absence in younger sediments, reflecting the rise in atmospheric O₂ at that time (e.g., Farquhar et al. 2011), Therefore, MIF S-isotope signatures in Mangaia basalts are consistent with an ancient (> 2.4 Ga) recycled component in the HIMU source.

The origin of the depleted component in northern Cook-Austral Islands is less clear, as is the nature of the EM/FOZO-like component found in basalts from Rapa, Marutiri, and MacDonald seamount. As discussed by Lassiter et al. (2003), the “depleted” component observed in northern Cook-Austral basalts and most apparent from the broad negative correlation in Nd- and Pb-isotopes in basalts from these islands is unusual in simultaneously having slightly higher ⁸⁷Sr/⁸⁶Sr (> 0.7029) than typical for endmember HIMU basalts (~ 0.70285) but more radiogenic ¹⁴³Nd/¹⁴⁴Nd. Based on low Cl/K ratios in melt inclusions from basalts with depleted isotopic signatures, Lassiter et al. (2003) argued against assimilation of altered Pacific lithosphere to account for the depleted isotopic signatures but noted the isotopic similarity in “depleted” Cook-Austral signatures and “enriched” signatures from near-ridge Pacific seamounts. Thus, an origin from enriched (relative to DMM) components in the Pacific asthenosphere or lithosphere is possible.

Finally, basalts from the southern Austral Islands, including Rapa Iti, trend to more enriched isotopic compositions, with less radiogenic Pb-isotopes (²⁰⁶Pb/²⁰⁴Pb from 19.07–19.28) and higher ⁸⁷Sr/⁸⁶Sr (> 0.7036). They also have less radiogenic Nd- and Hf- isotopic compositions, with ε_Nd values ranging from 2.3 to 3.8 and ε_Hf values ranging from 2.1 to 4.2 (Lassiter et al. 2003; Chan et al. 2009). Lassiter et al. (2003) suggested the enriched source could not be generated from the addition of ancient pelagic sediments to HIMU mantle, due to the lower ¹⁸⁷Os/¹⁸⁸Os values reported in these basalts compared to HIMU basalts. They suggested the enriched source sampled at Rapa may derive from subduction-modified sub-arc mantle wedge material that was recycled into the deep mantle through viscous coupling to downgoing slabs. Chan et al. (2009) further supported this hypothesis based on elevated δ⁷Li measured in Rapa basalts relative to fresh MORB.

In the following discussion, we group our samples as having a predominantly HIMU-like signature (²⁰⁶Pb/²⁰⁴Pb > 20), a depleted signature (²⁰⁶Pb/²⁰⁴Pb < 20, ⁸⁷Sr/⁸⁶Sr < 0.7030), or an EM/FOZO-like signature (²⁰⁶Pb/²⁰⁴Pb < 19.5; ⁸⁷Sr/⁸⁶Sr > 0.7035).

Samples and methods

We examined clinopyroxene phenocrysts from 11 Raivavae and Rapa basalts collected by John Lassiter and Hans Barsczus in 1997, including 7 basalts from Raivavae and 4 from Rapa. We also examined phenocrysts from one Mangaia basalt (MG1001) provided by Matt Jackson. Sample descriptions along with whole rock major and trace element and Sr, Nd, Pb, Hf, Os, and Li isotopic data for most samples are presented in Lassiter et al. (2003), Chan et al. (2009), and Herzberg et al. (2014). Based on hand sample estimates, most samples are highly phyric, with > 10% olivine and/or cpx phenocryst abundances. Whole rock Mg#s [100* molar Mg/(Mg + Fe²⁺), assuming Fe²⁺/Fe_tot = 0.85] range from 42 to 78. Trace element abundances and concentrations of H₂O and other volatile species in olivine-hosted melt inclusions from a subset of these samples have previously been reported (Lassiter et al. 2002, 2022; Szramek 2010; Cabral et al. 2014).

Clinopyroxene phenocrysts with no visible alteration or adhered matrix were hand-picked under a binocular microscope from crushed and sieved samples. Pyroxene phenocrysts are generally fewer but larger than olivine phenocrysts in the same samples, and many picked cpx grains appear to be broken fragments from larger phenocrysts with no obvious crystal faces. Because cpx in these samples are quite dark, it was generally not possible to determine if cpx grains contain inclusions of spinel or other mineral or melt inclusions. However, cpx were carefully examined for exposed inclusions after polishing and care taken in subsequent analyses to avoid any exposed inclusions. Sixty-nine phenocrysts were selected and mounted in indium mounts for subsequent analysis of major and trace element and H₂O abundances via EPMA, LA-ICP-MS, and SIMS. Details of sample and mount preparation are provided in the Supplemental Materials. Most spots analyzed were from the center of each mounted grain fragment.

Major elements were analyzed using a JEOL JXA-8200 electron probe microanalyzer (EPMA) at the University of Texas at Austin using analytical procedures outlined in Gao et al. (2017) and Marshall et al. (2018). EPMA analyses were conducted using a 20 nA beam current, 15 kV accelerating voltage, and a 10 μm defocused beam. Count times were 30–40 s on peak and 15–20 s off peak. Cr-Augite 164905 was used as a secondary standard. For elements with concentrations greater than 1 wt%, reproducibility was better than 2%. For elements with concentrations less than 1 wt%, reproducibility ranged from ~ 4% for Cr₂O₃, ~ 11% for Na₂O and, TiO₂, and ~ 12% for MnO. Averaged analyses were accurate within ~ 2% of published values for all elements, except for Cr (~ 3%), Ti (~ 4%), and Mn (~ 8%) (Jarosewich et al. 1980). Details of analytical methods and standard measurements are presented in the Supplemental Materials.

Following major element analysis, cpx phenocrysts were analyzed for hydrogen via the Cameca 6f secondary ion mass spectrometer (SIMS) at Arizona State University using analytical procedures outlined in Marshall et al. (2018). Matrix-matched olivine and pyroxene standards from Mosenfelder and Rossman (2013 A and B) were used to construct calibration curves for each analytical session. Repeated analyses of secondary pyroxene standard PMR-53 indicate an analytical uncertainty of ~ 13% (1σ). Details of the SIMS analytical procedures are presented in the Supplemental Materials.

Following SIMS analysis, cpx phenocrysts were analyzed for trace elements using an Excimer LA-ICP-MS with a New Wave UP-193FX laser system coupled to an Agilent 7500Ce quadrupole (LA-ICP-MS) at the University of Texas at Austin. Repeated analyses of LA-ICP-MS secondary standard BCR-2G were accurate within 6% of published values for all elements except Ti (8%) and Pb (7%) (see Supplemental Materials). Repeated analyses of BCR-2G were reproducible within 5% (1 σ) for all elements other than Tm (13%), Lu (11%), and U (6%). Details of the LA-ICP-MS analytical methods are presented in the Supplemental Materials.

For the EPMA, SIMS, and LA-ICP-MS analyses, the same spot (or as close as possible) was used to minimize possible effects of intra-mineral heterogeneity.

Results

Clinopyroxene major elements

Major element compositions for 88 spot analyses from 69 cpx phenocrysts are reported in Table S1. Diopside is the dominant component of most cpx, with the exception of phenocrysts from RVV-343, which are augitic in composition. Clinopyroxene Mg#s range from 63 to 86, with an average Mg# of 81. Mg# is negatively correlated with Al₂O₃, Na₂O, and TiO₂. Most analyses plot along a well-defined Mg#-Al₂O₃ trend, but a subset of grains are displaced to lower Mg# for a given Al₂O₃ abundance (Fig. S1). Individual basalts may contain grains from both trends.

Seven of the 12 samples have relatively homogenous cpx phenocryst major element compositions (e.g., Mg# standard deviation < 1.5). However, several samples have phenocrysts that span a wider range in Mg#. For example, phenocrysts from RPA-367 have Mg#s ranging from 74 to 86, well outside analytical uncertainty. Phenocrysts from RVV-318 also span a wide range in composition, with Mg#s ranging from 78 to 86. Although we were unable to directly evaluate core-rim variations, examination of intra-grain duplicate analyses from different locations on the same grain suggests that most of the inter-grain variability observed in individual samples does not reflect intra-grain (e.g., core-rim) heterogeneity. Most grains for which multiple spots were analyzed show very limited major element or [H₂O] variations that are close to analytical error.

Clinopyroxene H ₂ O content

H⁺ concentrations reported as ug/g equivalent [H₂O] of individual spot analyses are reported in Table S1. [H₂O] ranges from 72 to 1019 ug/g. Average [H₂O] of cpx from individual basalts ranges from 90 to 756 ug/g. Intra-sample variability (1σ) ranges from 15 to 42%. Most samples have pyroxene [H₂O] spanning a relatively narrow range, within ~ 2σ analytical uncertainty. Two samples (RPA-367 and RPA-414) display greater variability in [H₂O] of ~ 42 and 37% (1σ), respectively. Clinopyroxene [H₂O] does not correlate with Mg#, but is broadly correlated with LREE enrichment (e.g., La/Sm) when considered across all samples (Figures S2 and S3). However, in some but not all individual samples, [H₂O] is negatively correlated with Mg# (e.g., phenocrysts from RVV-318 and MG-1001). Clinopyroxene CaO is positively correlated with cpx [H₂O] and with calculated melt [H₂O] (see Section "Estimation of parental magma H2O"). Taken as a whole, there is a very broad positive correlation between cpx [H₂O] and both Al₂O₃ and [Ce]. Weakly significant correlations are also observed in some, but not all, phenocryst populations from individual samples. For example, [H₂O] is positively correlated with both Al₂O₃ and [Ce] in RVV-318 phenocrysts.

Clinopyroxene trace elements

Individual spot cpx trace element analyses are reported in Table S4. Clinopyroxene phenocrysts have hump-shaped REE patterns, with high MREE/HREE ratios (e.g., (Sm/Yb)_n > 3; where the subscript denotes normalization to primitive mantle values of McDonough and Sun 1995) but low LREE/MREE ratios (e.g., (La/Sm)_n < 1) (Fig. 1). [Ce] ranges from 2.5 to 34.5 ug/g. Clinopyroxene [Ce] is negatively correlated with cpx Mg# (Fig. S1) and positively correlated with cpx Al₂O₃ within the population as a whole as well as within several individual basalt sample phenocryst populations. Correlations within individual samples create sub-parallel trends, which are offset from sample to sample.

Fig. 1

Primitive-mantle-normalized REE abundances in clinopyroxene phenocrysts (solid line, open symbols), calculated equilibrium melt compositions (solid line, closed symbols), and corrected host matrix compositions (dashed line, open symbols). For clarity, the average phenocryst and calculated melt composition for each sample is shown. Matrix compositions and calculated equilibrium melts overlap in composition for the middle to light REEs. However, calculated melt compositions have systematically lower HREE such as Yb and Lu, and thus higher La/Yb, than estimated matrix compositions (wholerock composition corrected for phenocryst accumulation), which may reflect systematic overestimation of HREE cpx/melt K_d values, and thus underestimation of melt HREE abundance. Primitive mantle REE abundances from McDonough & Sun (1995). Green circles = DM-type (Raivavae); Red squares = HIMU-type (Raivavae); Blue triangles = EM-type (Rapa); Diamond = HIMU-type (Mangaia)

Discussion

Numerous studies have examined water contents in submarine quenched glass and in olivine-hosted melt inclusions to constrain variations in water content and H₂O/Ce ratios in different OIB endmembers (e.g., Dixon et al. 2002; Workman et al. 2006; Kendrick et al. 2017, and many others). However, quenched submarine glasses are not available for all locations, and olivine-hosted melt inclusions may be susceptible to post-entrapment water loss through diffusion. Clinopyroxene phenocrysts are common in many OIB lavas, and provide a complementary avenue for estimating magmatic water contents. In the sections below, we use measured cpx phenocryst compositions to constrain both the compositions of the melts from which these phenocrysts derived as well as their crystallization pressures, and examine the roles of magmatic degassing and source enrichment in controlling water contents in Australs melts.

Estimation of parental magma major element composition

Most of the samples analyzed in this study appear to contain significant quantities of cumulous phenocrysts, so that cpx are not in chemical equilibrium with wholerock compositions. This is reflected in the high measured Mg#s of most wholerocks, which are out of equilibrium with the average Mg#s of wholerock-hosted phenocrysts (Fig. 2). We estimated melt compositions in equilibrium with the cpx phenocrysts starting with the assumption that most phenocrysts and basalt hosts are related to the same parental melts. Parental melt compositions can be approximated by adding or (more often) removing appropriate quantities of olivine and clinopyroxene phenocrysts to the basalt wholerock composition to bring the Mg#s of the phenocrysts and wholerock into equilibrium. This correction routine is similar to that outlined by Hauri (1996). Published major element data for basalts from Raivavae, Rapa, and Mangaia reveal that CaO and Al₂O₃ concentrations are well correlated with MgO, with relatively minor differences for the different island suites. The slopes of these correlations are consistent with variable addition/subtraction of a roughly 70:30 mixture of olivine and clinopyroxene. This ratio is similar to the average ol/cpx ratio observed in basalt hand samples for the samples from this study. Because the phenocryst ratio can vary significantly even within hand samples, we use the observed major element correlations to constrain the ol/cpx ratio used in the correction procedure. A 70:30 mixture of olivine and clinopyroxene (using the average major element compositions of all cpx from this study and published Australs olivine data) is added or removed incrementally from the bulk rock composition until the composition is in Fe/Mg equilibrium for a given phenocryst, assuming cpx/melt Fe/Mg K_d = 0.28 (Putirka 2008).

Fig. 2

Comparison of basalt wholerock and cpx phenocryst Mg#s. Cpx Mg# calculated on molar basis as 100*Mg/(Mg + Fe_T). Wholerock Mg# calculated assuming Fe²⁺  = 0.85*Fe_T. Cpx-melt equilibrium field calculated assuming (Mg/Fe)_melt/(Mg/Fe)_cpx = 0.28 ± 0.08 (Putirka 2008). Filled symbols: Cpx Mg# > 80. Open symbols: Cpx Mg# < 80

The required phenocryst subtraction averages over 25% and ranges up to almost 50%. This reflects the fact that we selected samples previously analyzed in earlier studies (Lassiter et al. 2002, 2022, 2003; Cabral et al. 2014), and many of these samples had been previously selected specifically because they were phenocryst-rich, high-MgO samples. Wholerock MgO content ranges from 4.7 to 19.8 wt.%, with all but one sample having MgO > 11 wt.%. In contrast, after correction for phenocryst accumulation, the calculated melts in equilibrium with the phenocrysts have much lower MgO, ranging from 2.8 to 9.6 wt.%. The large fractionation correction therefore introduces uncertainty in the corrected melt composition, particularly for elements such as Na, Ca, and Al that are sensitive to the cpx/ol ratio of the cumulate assemblage. In addition, the large range in trace element abundances observed in different phenocrysts from some samples cannot simply be accounted for by variable fractionation, and instead requires different parental melts, suggesting an important role for magma mixing and hybridization prior to eruption. In the following discussion we therefore only use estimated SiO₂, FeO, and MgO in calculations of cpx/melt REE K_d values, as these components are relatively less sensitive to the exact cpx/olivine ratio used in the correction procedure. However, as discussed in Section "Constraints and controls on original melt volatile contents in Australs melts and global OIB", calculated melt [H₂O] is correlated with inferred melt CaO. As melt H₂O and CaO are inferred using different proxies and assumptions, this correlation suggests that our matrix correction method for inferring melt major element composition is robust to first order.

Whole rock trace element abundances were adjusted for olivine + cpx accumulation in the same manner, using the average measured trace element abundances in cpx. Tables S5 and S6 present the estimated whole rock major and trace element compositions calculated by this method along with the amount of olivine + cpx added or removed from the measured bulk rock composition. However, in the following discussion we do not use the calculated matrix trace element abundances except for comparison with estimated melt abundances calculated directly from cpx measurements and estimates for cpx/melt K_d values.

Estimation of parental magma REE abundances

Rare earth element abundances in the cpx phenocrysts provide an independent constraint on melt REE concentrations. Several models allow estimation of REE cpx/melt K_d values based on cpx and/or melt compositions. The partitioning model of Sun and Liang (2012) estimates REE K_d values using cpx major element composition, melt water content, and temperature as input parameters. Bédard (2014) reported empirical correlations between REE K_d values and several clinopyroxene and melt compositional parameters. For this study, we have taken an agnostic approach and averaged the K_d estimates from several Bédard (2014) parameterizations as well as the Sun and Liang (2012) model. We use cpx-based temperature estimates (see Sect. "Estimation of crystallization pressures") and melt water content estimated from measured cpx [H₂O] (see next section) as model inputs. The Bédard estimates used are based on estimated melt temperature, cpx Al(IV) and cpx Mg#, and melt SiO₂, FeO, and MgO. In general, the different Bédard estimates are relatively consistent, whereas the Sun & Liang estimates span a wider range and extend to both higher and lower estimates as those from Bédard. On average, the Sun & Liang model results in higher LREE-MREE Kd estimates compared to the Bédard estimates, but similar HREE Kd estimates. Table S7 lists the different K_d estimates for each method as well as the standard error of the average value. Average standard errors range from ~ 18% for La to ~ 6% for Yb, but vary from spot to spot. The uncertainties in the calculated K_d values are larger than uncertainties in the measured cpx REE contents, so that this uncertainty is the primary source of uncertainty in the calculated melt composition.

Calculated melt REE contents are given in Table S9. Despite the uncertainties outlined above, there is broad similarity in the melt REE patterns calculated from inversion of the cpx data and the estimated melt compositions derived from correction of wholerock compositions for phenocryst accumulation (Fig. 1). In general, estimated melt LREE and MREE concentrations are similar on average to those from corrected wholerock abundances. For example, calculated melt [Ce] and corrected matrix [Ce] is well correlated (R² = 0.60; p < 0.0001; N = 56). Melt HREEs estimated from cpx inversion are ~ 30% lower on average than those estimated from wholerocks, which may indicate a systematic overestimation of HREE Kd values. Because estimated melt REE abundances are well correlated with measured cpx [REE], errors in estimated K_d values will affect the absolute melt [REE], but will not strongly affect relative variations in REE abundance – cpx with higher REE abundances almost certainly derived from melts with higher REE than for cpx with lower [REE].

Estimation of parental magma H ₂ O

Measured water contents in clinopyroxene phenocrysts have increasingly been utilized to estimate magmatic water contents in settings where submarine quenched glasses are not available (e.g., Wade et al. 2008; Mazza et al 2019). Calculation of magmatic water content requires estimation of the cpx/melt water K_d. Previous studies have shown that partitioning of water into cpx is strongly influenced by cpx Al₂O₃ content and in particular by Al(IV), due to coupled substitution of H⁺ and Al³⁺ for Si⁴⁺ (Hauri et al. 2006; O’Leary and Hauri 2010; Novella et al. 2014). The different parameterizations for cpx/melt K_d(H₂O) produce varying but strongly correlated estimated K_d values, with the Novella et al. (2014) parameterization typically producing the lowest estimated K_d values, and Wade et al. (2008) producing the highest. Following our approach to estimation of REE K_d values, we have averaged the K_d estimates from Novella et al. (2014), O’Leary and Hauri (2010), and Wade et al. (2008) and used this average estimate to calculate melt [H2O] from measured cpx [H₂O]. Average standard error (SE) of the estimated K_d values is ~ 15%, similar to uncertainty from cpx [H₂O] measurements. Estimated cpx/melt K_d values from each model are reported in Table S8. Average estimated melt [H₂O] values using all three Kd estimates are reported in Table S9. Estimated K_d values range from 0.014 to 0.049 and are negatively correlated with cpx Mg#, as is expected given the negative correlation between cpx Mg# and Al₂O_3. Estimated melt [H₂O] varies from 0.35 wt.% to 4.25 wt.%. Estimated melt [H₂O] does not correlated with estimated cpx/melt K_d(H₂O), but correlates strongly with measured cpx [H₂O]. Thus, as with the estimates of [REE], there is some uncertainty in absolute melt [H₂O] introduced by uncertainty in the estimations of cpx/melt K_d(H₂O), but the relative variations in estimated water content appear robust and truly reflective of variations in melt [H₂O]. Factoring in both the uncertainty in the cpx [H₂O] measurements and in the cpx/melt Kd(H₂O), the estimated melt [H₂O] estimates have an uncertainty of ~ 20%.

Estimation of crystallization pressures

Estimated entrapment or closure pressures of olivine-hosted melt inclusions from the Austral Islands based on measured H₂O and CO₂ abundances primarily range between 100 and 200 MPa, with an average of 120 ± 50 MPa (Lassiter et al. 2002, 2022; Cabral et al. 2014). However, there is no a priori reason why clinopyroxene phenocrysts must have equilibrated at the same time and same depth as the melt inclusions. In many samples, clinopyroxene phenocrysts are larger and fewer in number than the olivine phenocrysts, which qualitatively suggests they may have formed from slower cooling (leading to lower nucleation) over a longer period than the olivine phenocrysts.

Clinopyroxene-based thermobarometers have significantly improved in recent years. Because of the uncertainties in calculated melt compositions discussed above, we have chosen to use several cpx-only thermobarometers to estimate the P–T conditions of phenocryst crystallization. Pressure and temperature were estimated using the cpx-only thermobarometers of Putirka (2008) using Eqs. 32b and 32d. Magmatic water content, which is an input into the pressure calculation, was taken from the magmatic water contents estimated using measured cpx [H₂O] (Sect. "Estimation of parental magma H2O"). Pressure and temperature were also calculated using the models of Petrelli et al. (2020) and Jorgenson et al (2022). Calculations were performed using Thermobar (Wieser et al., 2022). Pressures and temperatures from all three models are presented in Table S1. In the following discussion, we use the average P and T derived from the three models. Estimated temperatures (average estimate) range from 1098–1193 °C. Temperature estimates using the Putirka (2008) model are on average ~ 30 °C cooler than for the Petrelli et al. (2020) and Jorgenson et al. (2022) models. Estimated crystallization pressures range from 70–850 MPa (average 362 MPa). Again, the Putirka (2008) model results in lower pressures (average 273 MPa) compared to the Petrelli et al. (2020) and Jorgenson et al. (2022) models (both ~ 430 MPa).

One sample, RVV-343, has a wholerock composition that is close to equilibrium with its clinopyroxene phenocrysts based on Mg#, requiring < 5% phenocryst addition to bring the calculated melt Mg# into equilibrium with phenocryst Mg#s. For this sample, use of the Neave and Putirka (2017) pyroxene-melt thermobarometer can be used to check the validity of the pyroxene-only barometric estimates. The two methods provide similar results, with pressures estimated using the pyroxene-melt barometer of Neave and Putirka (2017) yielding pressure estimates ranging from 453 to 577 MPa for phenocrysts from RVV-343, and the pyroxene-only barometers yielding estimates of 282–371 MPa. Considering all samples, the clinopyroxene-only and pyroxene-melt thermobarometers yield results that are well correlated, with the cpx-only barometers shifted to lower pressures by ~ 180 MPa. All of the cpx-melt and cpx-only thermobarometers considered here suggest crystallization pressures extending to much higher values than recorded in olivine-hosted melt inclusions.

The suggestion that clinopyroxene phenocrysts formed on average at significantly greater depths than the equilibration pressures recorded by olivine-hosted melt inclusions would seem initially surprising, given that in most samples the olivine and clinopyroxene phenocryst populations span similar ranges in Mg#. One possibility is that this reflects the pressure dependance of olivine and clinopyroxene stability in basaltic melts. Although olivine is a common liquidus phase at low pressure, the stability field for olivine shrinks and that of clinopyroxene grows as pressure increases. Previous studies of cognate xenoliths from Hawaiian volcanoes have concluded that dunite xenoliths primarily formed at low pressures (< 1–200 MPa) whereas pyroxenite xenoliths from the same eruptions often formed at greater depths (e.g., Lassiter et al. 2022; Gao et al. 2022). Thus, the different pressures inferred from cpx phenocrysts and olivine-hosted melt inclusions may reflect mixing of melts that ponded at different depths and their respective crystal cargos immediately prior to eruption.

Alternatively, the apparent entrapment pressures of olivine-hosted melt inclusions may not reflect the actual depth of initial olivine growth. Several recent studies have suggested that volatile contents (and thus calculated entrapment pressures) in melt inclusions may be affected by volatile loss during ascent due to either incomplete closure or to decrepitation followed by subsequent healing (e.g., Métrich and Wallace 2008; Longpré et al. 2017; Maclennan 2017). Even in fully closed melt inclusions, rapid hydrogen diffusion through olivine can result in resetting of inclusion melt [H₂O], although inclusion [CO₂] is not affected by this process. As a result, melt inclusions often record final pre-eruptive storage conditions, often in relatively shallow magma chambers. Regardless of whether olivine crystallized at shallower depths than clinopyroxene phenocrysts, or melt inclusion entrapment pressures reflect final equilibration (through decrepitation/healing or diffusion) in shallow magma chambers, comparison of magma volatile contents inferred from clinopyroxene and from olivine-hosted melt inclusions can provide insights into magmatic degassing and primary melt volatile budgets.

Magmatic degassing recorded in melt inclusions and clinopyroxene phenocrysts

Several studies have previously argued that H₂O loss during degassing of most natural non-arc basalt magmas at pressures greater than ~ 100–200 MPa is limited, because exsolved vapor phases at these pressures are typically CO₂-rich and H₂O-poor (e.g., Cabral et al. 2014; Jackson et al. 2015). Water loss may accelerate at lower pressures as the H₂O/CO₂ ratio of exsolved volatile phases increases. Closed system degassing models predict greater H₂O loss during magma ascent than open-system models (e.g., Dixon 1997; Jackson et al. 2015). Anderson et al. (2024) argued that closed-system degassing better describes degassing processes for Samoan melts, and reported significant H₂O loss during magma ascent (e.g., Anderson et al. 2024) recorded in submarine quenched glasses and olivine-hosted melt inclusions. However, closed-system degassing models still predict very limited water loss during degassing at pressures greater than ~ 200 MPa (Fig. 3). Because olivine-hosted melt inclusions can record pressures extending to higher values, they have often been considered less susceptible to water loss due to magma degassing and thus an accurate recorder of the primary water content of OIB melts (e.g., Cabral et al. 2014). However, this conclusion is challenged by [H₂O] variations in melt inclusions from the Azores, which strongly correlate with equilibration pressures ranging from 87 to 468 MPa (MéTrich et al. 2014).

Fig. 3

Calculated cpx crystallization pressures vs. melt [H₂O] based on measured clinopyroxene compositions (larger symbols) and olivine-hosted melt inclusions (smaller symbols). Symbols are as in Fig. 2. See text for details of pressure estimates and melt water content estimates. Degassing path for closed (solid line) and open (dashed line) system degassing calculated for melt with initial [H₂O] = 4 wt.% and [CO₂] = 1.5 wt.% using VESIcal program (Iacovino et al. 2021) and solubility model from Iacono-Marziano et al. (2012). Other solubility models (e.g., Dixon 1997) produce qualitatively similar degassing paths with very limited H₂O loss at pressures > 200 MPa. Calculated degassing paths poorly match the observed broad correlation between calculated pressure and melt [H₂O]. Symbols as in Fig. 2. Representative error bars for melts with 1–4 wt.% H₂O assume an error of ~ 20% for H₂O estimates and ~ 75 MPa for pressure estimates. See supplemental text for details on error estimation

The average entrapment (or equilibration) pressures of melt inclusions from individual Raivavae, Rapa, and Mangaia samples range from 88 to 140 MPa (Lassiter et al. 2002, 2022; Cabral et al. 2014). The estimated crystallization pressures of clinopyroxene phenocrysts extend to much higher pressures than those recorded by olivine-hosted melt inclusions, extending to over 800 MPa. These pressures are much greater than the pressures where significant H₂O loss is expected for either open or closed-system degassing. However, estimated magmatic water content is positively correlated with entrapment/equilibration and crystallization pressures of Austral Island melt inclusions and cpx phenocrysts, suggesting a role for magmatic degassing. Figure 3 shows calculated melt H₂O concentrations versus cpx equilibration pressure. Water contents and equilibration pressures for previously published olivine-hosted melt inclusions are also shown for comparison. There is a very weak positive correlation between cpx equilibration pressure and melt [H₂O] considering all samples (p value = 0.034; R = 0.226; N = 88). Average melt water content inferred from cpx phenocrysts ranges from 1.1 wt.% for phenocrysts with equilibration pressures < 200 MPa, to 1.5 wt.% for phenocrysts with equilibration pressures from 200–300 MPa, to 1.9 wt.% for phenocrysts with equilibration pressures > 300 MPa. Phenocrysts from some (but not all) individual samples record slightly stronger correlations between pressure and [H₂O]. For example, calculated melt P and [H₂O] is positively correlated in phenocrysts from MGA-1001 (p value = 0.018, R = 0.666; N = 12). Olivine-hosted melt inclusions have systematically lower water contents and estimated entrapment pressures. When melt inclusions and melts calculated from cpx phenocrysts are considered together the correlation between equilibration pressure and melt [H₂O] becomes much stronger (p value < 0.001; R = 0.612; N = 261). In addition, for four basalt samples (RVV-310, RVV-318, RPA-502, and MGA-1001), data on magma water content exist from both melt inclusions and clinopyroxene phenocrysts. In all four cases, both the pressures and the magmatic water contents inferred from the phenocrysts exceed those inferred from melt inclusions. The simplest explanation for these observations is that the parental melts to both the clinopyroxene phenocrysts and olivine-hosted melt inclusions have experienced water loss through degassing, with the degree of degassing increasing with decreasing pressure. However, the observed trend is not consistent with open- or closed-system degassing trends predicted using several water solubility models (e.g., Dixon 1997; Iacono-Marziano et al. 2012) (Fig. 3).

Several different processes may contribute to volatile loss that are not fully captured by open- or closed-system degassing models. Average equilibration pressure for olivine-hosted melt inclusions from Raivavae, Rapa, and Mangaia is ~ 120 MPa, suggesting that these inclusions formed or reached final closure or equilibration in relatively shallow magma reservoirs, possibly within the respective volcanic edifices. Magma in such reservoirs may experience significant vertical movement leading up to and following volcanic eruptions. At Kilauea, highly degassed melts have been collected from the submarine east rift flank, requiring that these melts at one point degassed at shallower depths than their final emplacement (Dixon et al. 1991). One possible explanation for the loss of water inferred from the olivine-hosted melt inclusions and shallowly-formed cpx phenocrysts is that gas-rich magmas ascended close to the surface, lost a significant portion of their volatiles, and then the resultant denser melts drained back into the deeper reservoir to mix with less degassed melts. Mixing of degassed and undegassed melts as a result of conduit convection is likely an important process in many ocean island settings, and may provide an explanation for linearly correlated CO₂ and H₂O abundances in some suites (e.g., Witham 2011). For example, CO₂ and H₂O abundances are positively correlated in melt inclusions from the Azores (MéTrich et al. 2014). Although MéTrich et al. (2014) suggested that the correlation between CO₂ and H₂O in melt inclusions from the Azores could result from percolation of CO₂-rich vapor (generated at pressures > 500 MPa) through the volcanic system prior to eruption (see discussion below), such correlations could also reflect mixing of degassed and undegassed melts through vertical mixing in the volcanic plumbing system, similar to what is observed at Kilauea.

An alternative explanation for the mismatch between pyroxene and melt inclusion [H₂O] is post-entrapment loss of H₂O from the inclusions prior to, during, or after eruption. Olivine-hosted melt inclusions are susceptible to H₂O loss through rapid H diffusion in olivine in slow-cooling melts (e.g., Hauri 2002; Cervantes and Wallace 2003). Water loss during and following eruption can be avoided if melt inclusions from scoria samples are examined (Cervantes and Wallace 2003; Kelley et al. 2010). However, on older ocean islands, scoria is rapidly eroded and is also typically more heavily altered than the interiors of massive flows. Therefore, many previous studies have examined melt inclusions from massive flows, where melt inclusions may experience diffusive water loss during cooling (e.g. Lassiter et al. 2002, 2022 and Cabral et al. 2014).

Within natural samples, hydrogen diffusion appears to be more rapid in olivine than pyroxene by nearly a factor of 100 (Tian et al. 2017). Diffusive H₂O loss from melt inclusions during eruption or cooling at the surface should result in a correlation between inclusion size and [H₂O], because large inclusions are less susceptible to diffusive H₂O loss (Chen et al. 2011). [H₂O] in melt inclusions from one Mangaia sample (MGA-B-47) do correlate with inclusion diameter, which likely indicates inclusions from this sample have lost H₂O through post-entrapment diffusion (Cabral et al. 2014). In contrast, [H₂O] in melt inclusions from MG1001 do not correlate with inclusion diameter, which may indicate little diffusive H₂O loss during eruption (Cabral et al. 2014). Inclusions from this sample also have the highest measured [H₂O] of any Australs melt inclusions, and are modestly lower than [H₂O] inferred from cpx phenocrysts from the same sample with the lowest equilibration pressures (average MG1001 MI [H₂O] = 1.5 wt.% vs. 2.1 wt.% in melts inferred from cpx phenocrysts from this sample). Melt inclusions from the Raivavae and Rapa samples are smaller than those studied by Cabral et al. (2014), making them more susceptible to diffusive H₂O loss. Melt inclusions from these samples also have much lower average [H₂O] than cpx-inferred melts from the same samples, with average [H₂O] ranging from 0.25 to 0.53 wt.% vs 0.75 to 3.01 wt.% in average cpx-inferred melts. Based on the lower average [H₂O] in melt inclusions from these samples than calculated for melts in equilibrium with coexisting clinopyroxene phenocrysts from the same samples, we conclude that the Raivavae and Rapa melt inclusions have likely lost significant H₂O via post-entrapment diffusion. In contrast, many of the Mangaia samples appear to be less affected by late-stage diffusive water loss.

An additional process that may result in correlations between melt [H₂O] and pressure is “sparging” by a CO₂-rich vapor phase generated at greater pressures than those recorded by cpx phenocrysts or olivine-hosted melt inclusions (e.g., MéTrich et al. 2014). CO₂ flushing has been proposed to account for magmatic volatile variations in numerous systems (e.g., Iacovino et al. 2016; Allard 2010). The concentration of CO₂ in OIB sources or primary melts is difficult to constrain, but numerous studies have suggested that many OIB sources are carbon-rich, and carbonatitic metasomatism of mantle sources may be common. For example, Saal et al. (1998) reported carbonate globules in melt inclusions from Mangaia and suggested that the presence of this phase directly indicates CO₂-rich magmas in the Cook-Austral chain. This is also consistent with work by Hauri et al. (1993), who suggested that trace element compositions of peridotite xenoliths from Tubuai, Cook- Austral Islands, reflected metasomatism by carbonatitic melts. Saal et al. (2002) estimated a CO₂/Nb ratio for the depleted upper mantle source of MORB of ~ 239. If this ratio is applicable to OIB melts, then the elevated [Nb] in the estimated Australs melts (~ 33–150 ppm) would imply initial CO₂ concentrations ranging from ~ 0.8–3.6 wt.%. High initial [CO₂] in Austral Island primary melts would result in early, deep onset of vapor saturation and production of volatile phases with high CO₂/H₂O, increasing the likelihood that sparging processes could produce H₂O loss in Australs melts. Percolation of CO₂-rich fluids (generated at depth) through shallower melts that have already experienced CO₂ loss results in re-addition of CO₂ to the melt, which causes a significant reduction of H₂O solubility as the melt evolves along the solubility isobar to intersect the high-P H₂O/CO₂ isopleth.

Whether by sparging or by shallow degassing and then vertical migration and mixing with deeper magmas, the observed correlations between crystallization pressure and inferred magma [H₂O] indicates that the parental melts to the Australs cpx phenocrysts have experienced variable degrees of degassing, so that inferred volatile contents should be considered minimum estimates of primary volatile inventories.

Constraints and controls on original melt volatile contents in Australs melts and global OIB

Magmatic water contents in undegassed magmas are influenced by the degree of partial melting during melt generation as well as variations in the water content of different mantle sources. Water is incompatible during partial melting of nominally anhydrous peridotites. Therefore, magmatic water content should correlate with the degree of partial melting. Consistent with this expectation, estimated Australs melt [H₂O] is positively correlated with melt La/Sm (R = 0.3326; p = 0.0016; N = 88) (Fig. 4). This correlation becomes stronger (R = 0.4270, p = 0.0004; N = 64) if phenocrysts with equilibration pressures < 300 MPa are excluded. However, no statistically significant correlation is observed between calculated melt [H₂O] and [Ce], although H₂O and Ce do correlate in the cpx phenocrysts (Figs. 5 and 6). In contrast, melt [H₂O] is positively correlated with both measured cpx CaO and estimated melt CaO (R² = 0.29 and 0.35, respectively; p < 0.0001; N = 66, excluding phenocrysts with equilibration P < 300MPa) (Fig. 7). The significance of the correlation between melt H₂O and CaO is unclear. One possibility is that this reflects prior metasomatism of the source(s) of Austral magmas by water-rich carbonatitic melts. As noted above, several studies have suggested a role for carbonate recycling and carbonatite metasomatism in the generation of HIMU mantle (e.g., Coltorti et al. 1999; Jackson and Dasgupta 2008). The correlation between CaO and H₂O in Australs melts may result from the coupled addition of these phases during carbonatite metasomatism.

Fig. 4

Estimated melt [H₂O] vs. melt La/Sm. See text for details. Symbols as in Fig. 2. MORB and OIB data from quenched glass and melt inclusions. References for compiled data for MORB and OIB are in the Supplemental Material. Bermuda data from cpx phenocryst and wholerock data of Mazza et al. (2019), with melt [H₂O] recalculated using the K_d estimation methods outlined in this paper. Melt [H₂O] estimated from cpx phenocrysts is systematically higher than measured in submarine glasses and melt inclusions. Representative error shown for H2O (~ 20%) and La/Sm (~ 22%) for sample with 4 wt.% H₂O and La/Sm = 10

Fig. 5

Clinopyroxene [Ce] vs [H₂O]. Symbols as is Fig. 2

Fig. 6

Calculated melt [Ce] vs melt [H₂O]. Symbols and data sources as in Figs. 2 and 4. Data sources as in Fig. 4. Most submarine glasses and melt inclusions have maximum [H₂O] of ~ 1.6 wt.% regardless of Ce enrichment. Ce and H₂O are well correlated at low [Ce] (< 50 ug/g), but poorly at higher [Ce]. Melt [H₂O] estimated from cpx phenocrysts is systematically higher than in glasses or melt inclusions for a given [Ce]. Constant H₂O/Ce trendlines shown for reference. Error bars for calculated melt [H₂O] (~ 20%) and [Ce] (~ 16%) shown for representative samples with 1–4 wt.% H₂O and 50–250 ug/g Ce

Fig. 7

Estimated melt CaO vs. melt H₂O. Symbols as in Fig. 2

Because magmatic water content is influenced by the degree of partial melting or fractionation that melts and their source(s) have experienced, many studies have examined H₂O/Ce ratios to “see through” magmatic evolution processes and place constrains on variations in mantle [H₂O] (e.g., Dixon et al. 2002). In many magmatic suites H₂O and Ce are well-correlated, reflecting the incompatible behavior of both species, and H₂O/Ce ratios can be used to estimate the relative enrichment or depletion of water relative to other non-volatile incompatible elements. However, although the Australs samples selected for this study span a wide range in isotopic composition, there is no apparent correlation between any radiogenic isotope ratio in the wholerock samples from which cpx were extracted and either average calculated melt [H₂O] or H₂O/Ce, as illustrated in Fig. 8. For example, taking the average melt composition from each sample inferred from clinopyroxene, samples with “HIMU-like” isotopic signatures (defined here as ²⁰⁶Pb/²⁰⁴Pb > 20) have H₂O/Ce ranging from 22 (RVV-316) to 391 (RVV-318). Excluding RVV-316 (for which only one highly evolved phenocryst was analyzed), HIMU samples have average H₂O/Ce ranging from 146 to 391. Samples with “DM-like” signatures span a similar range, from 83 to 482, whereas Rapa samples with EM/FOZO-like signatures have inferred H₂O/Ce ranging from 77 to 304. It is possible that the degassing processes described above have destroyed any initial correlations that existed between volatile enrichment and isotopic composition in the different primary Australs melts. However, degassing in samples with equilibration pressures > 300 MPa is likely limited, and the observed correlations between [H₂O] and both La/Sm and CaO suggest that primary variations in melt water content have been preserved to some extent. Therefore, the lack of correlation between H₂O/Ce and radiogenic isotopes in the Australs melts suggests a decoupling of these tracers in the different mantle sources involved in Australs melt genesis.

Fig. 8

a) Calculated melt [H₂O] vs. host basalt ²⁰⁶Pb/²⁰⁴Pb. b) Calculated melt H₂O/Ce vs. host basalt ²⁰⁶Pb/²⁰⁴Pb. Error bars denote 1SD of [H₂O] or H₂O/Ce estimates for cpx from an individual sample. Isotopic data for MGA1001 is not available, so average value of published data from Mangaia basalts is plotted. Symbols as in Fig. 2

Although H₂O/Ce in some plume-influenced MORB suites broadly correlate with radiogenic isotopes (e.g. Dixon et al. 2002), global compilations of OIB display poor or absent correlations between H₂O or H₂O/Ce and radiogenic isotopes (see e.g. Figure 10 from Jackson et al. 2015), consistent with the lack of correlation observed in Australs melts. In contrast, H₂O/Ce is negatively correlated with [Ce] in OIB (Fig. 9), and positively correlated with [H₂O] (Fig. 10). Jackson et al. (2015) argued that variations in H₂O/Ce result from variations in source lithology. They suggested that the high [Ce] and low H₂O/Ce in some HIMU and EM glasses could result from low-degree melting of a dominantly pyroxenite component, and that the low [Ce] and high H₂O/Ce ratios reported in MORB reflect higher degrees of melting of a dominantly peridotitic source. This model is based on previous work by Bizimis and Peslier (2015), who note that Ce has a higher partition coefficient than H during crystallization of pyroxene-rich lithologies, resulting in pyroxenites with higher Ce and lower H₂O/Ce than peridotites. Whether this variation in H₂O/Ce would be preserved during mantle melting is unclear – because the K_d(H₂O/Ce) is much lower in cpx than in other mantle phases (e.g., opx or ol), pyroxenite-derived melts will tend to have higher H₂O/Ce than their source, whereas melt/source fractionation of H₂O/Ce is much smaller for peridotite melting. Furthermore, if H₂O/Ce variations in OIB reflect mixing of (high H₂O/Ce) peridotite and (low H₂O/Ce) pyroxenite sources, then H₂O/Ce in OIB should to some degree correlate with other proxies for pyroxenite melting. For example, melts derived from a pyroxenite source are predicted to have higher CaO/Al₂O₃, lower NaO/TiO₂, and more radiogenic Pb-isotopes than ambient mantle peridotite (Hauri 1996; Putirka 1999; Jackson and Dasgupta 2008; Jackson et al. 2012). If H₂O/Ce ratios track variations in source lithology, then they should correlate with other chemical or isotopic signatures consistent with pyroxenite melting, but this is not observed either in the Australs melts or in global OIB for any of the signatures of pyroxenite melting previously proposed.

Fig. 9

Melt H₂O/Ce vs. melt [Ce]. Symbols and data sources as in Figs. 2 and 4. Error bars for calculated melt [Ce] (~ 16%) and H₂O/Ce (~ 26%) shown for representative samples with 50–200 ug/g Ce and H₂O/Ce = 200–600

Fig. 10

Melt H₂O/Ce vs. melt [H₂O]. Symbols and data sources as in Figs. 2 and 4. Error bars for calculated melt [H₂O] (~ 20%) and H₂O/Ce (~ 26%) shown for representative samples with 1–4 wt.% H₂O and H₂O/Ce = 200–600

A simpler explanation for the correlation of H₂O/Ce with both [Ce] and [H₂O] is that it results from variable degassing recorded in OIB glasses and melt inclusions. As shown in Fig. 6, many OIB water estimates derived from submarine glasses or melt inclusions have maximum [H₂O] of ~ 1.5–2.0 wt.%, regardless of degree of Ce enrichment. More than 95% of OIB glasses and melt inclusions have [H₂O] less than 1.5 wt.%, and 98% have < 2 wt.% H2O. In contrast, Australs melt [H₂O] inferred from cpx extend to more than 4 wt.%. The negative correlation observed globally between [Ce] and H₂O/Ce largely reflects decoupling between these species in most OIB suites that results from vapor saturation and degassing. Given the evidence presented previously that many Australs melts experienced significant degassing at P < 300 MPa, which exceeds the depths of emplacement or entrapment for most OIB glasses and melt inclusions, it is likely that many OIB have also experienced water loss through degassing and/or sparging processes, resulting in underestimation of their primary water content. Degassing processes, rather than mixing of pyroxenite- and peridotite-derived melts, is likely the primary source of variation in H₂O/Ce in OIB.

Our results are consistent with those of Mazza et al. (2019), who estimated primary magmatic water content in Bermuda basalts using cpx phenocrysts. Using the same methods outlined above to constrain cpx/melt K_d(H₂O), the Bermuda cpx indicate melt [H₂O] ranging from 2.4 to 4.5 wt.% and, using bulk rock [Ce], H₂O/Ce ratios ranging from ~ 104 to 305. As with the Australs samples, the Bermuda estimates have higher [H₂O] and H₂O/Ce at a given [Ce] than any other OIB melt estimates based on quenched glass or melt inclusions. Although additional studies are needed, the available data suggest that cpx phenocrysts consistently record higher melt [H₂O] than either quenched glasses or melt inclusions, suggesting that magma degassing is a systemic problem in OIB. The problem of degassing is exacerbated in OIB relative to MORB because (1) sampling depths for submarine glasses are typically shallower than for MORB, (2) the presence of volcanic edifices extending to shallow depths or subaerial structures raises the possibility of extensive vertical magma mixing, with volatile-rich magmas ascending to near the surface, outgassing, and then sinking and mixing back with deeper, less-degassed magmas (e.g., Dixon et al. 1991), and (3) because most OIB are enriched in incompatible species, they are likely to be inherently richer initially in H₂O and CO₂ than MORB, resulting in degassing or sparging processes initiating at greater pressures.

Estimated Austral melt H₂O/Ce ratios correlate strongly with melt [H₂O] (Fig. 10). This reflects the decoupling of H₂O and Ce in the Australs melts. This decoupling contrasts with the strong correlation observed between [H₂O] and [Ce] in most MORB glasses. Similar decoupling is observed in many suites of OIB glasses and melt inclusions, which as a whole display significantly weaker correlations between [H₂O] and [Ce] than in MORB. The correlation between [H₂O] and H₂O/Ce is observed in the Australs melts even when samples are filtered for equilibration pressures > 300 MPa or cpx Mg# > 80. A likely explanation for the observed trend between [H₂O] and H₂O/Ce is that it reflects variable magma degassing, which results in the decoupling of H₂O and Ce concentrations. In this case, water concentrations even in the melts inferred from cpx with equilibration pressures > 300 MPa have been variably affected by magmatic degassing or by water loss from the cpx during ascent. In this case, the “primary” magmatic H₂O/Ce ratios in Australs melts may have extended to ~ 500–600, with one sample recording an even higher H₂O/Ce of 825. As previously stated, this suggests that estimated [H₂O] in both the Australs and other OIB should be considered minimum estimates of primary water content – most OIB suites appear to have been affected by degassing processes to variable degrees that are difficult to quantify.

Global water budget of the mantle

Our results suggest, but do not prove, that many OIB glasses and melt inclusions have lost significant water through degassing or sparging processes. This is supported by recent comparison of water contents in Samoan submarine quenched glasses and melt inclusions, which suggest significant water loss during melt ascent from 200 to 400 MPa (recorded in melt inclusions) up to 10–20 MPa (quenched glasses) (Anderson et al. 2024). Whereas previous studies of Samoan quenched submarine glasses suggested a H₂O/Ce ratio for the EM-II source of < 100, Anderson et al. (2024) report H₂O/Ce in olivine-hosted melt inclusions up to 275, and show that [H₂O] differences between melt inclusions and quenched glasses can be explained by closed-system degassing of CO₂-rich melts. As a result, H₂O/Ce ratios in mantle endmembers based on submarine glasses in particular may systematically underestimate mantle H₂O/Ce. This may be especially important for estimates of EM components, for which glass data are mostly limited to highly evolved samples with [Ce] > 100 ppm. Additional studies are needed to compare melt water estimates from cpx phenocrysts with those from glasses or melt inclusions from the same localities to determine whether magmatic water contents in Australs and Bermuda melts are unusually high, or if other OIBs have lost much of their initial water content and thus those estimates are too low.

Several previous studies have suggested that Earth’s mantle may contain the equivalent of many ocean masses of water. For example, Peslier et al (2017) estimate a mantle water content of 1.1–7.4 ocean mass equivalents, with the transition zone alone containing one ocean mass or more and up to 4.5 ocean mass equivalents in the lower mantle. Other studies have noted the higher water storage capacity of transition-zone minerals relative to the upper and lower mantle, and argued for significant water storage in this portion of the mantle (e.g., Bercovici and Karato 2003). Mazza et al. (2019) suggested that the high water contents of Bermuda basalts indicated derivation from the volatile-rich transition zone. However, although the estimated [H₂O] in Bermuda melts is higher than most previous OIB estimates, the H₂O/Ce ratios (~ 100–300) are not elevated relative to most MORB or OIB.

We develop a simplified mantle water budget by considering only two components – the depleted mantle and OIB-source mantle with uniform H₂O/Ce in each reservoir, without assuming either the size or location of these reservoirs. The Bulk Silicate Earth (BSE) is estimated to contain ~ 1.675 μg/g Ce (McDonough and Sun 1995), whereas average continental crust has ~ 42 ug/g Ce (Rudnick and Fountain 1995). Considering the mass of continental crust (~ 0.5% of BSE), mass balance therefore suggests a bulk mantle [Ce] of ~ 1.47 μg/g. The fraction of the mantle represented by the depleted MORB-source mantle is considered to range from 0.25 (the mass fraction represented by the upper mantle above 660 km) up to 0.9 (roughly the mantle mass less either D” or the transition zone). Depleted MORB-source mantle (DMM) is estimated to have ~ 0.55 μg/g Ce (Workman and Hart 2005). The concentration of Ce in the OIB source varies depending on the fraction of the mantle represented by this source, ranging from 1.8 μg/g (for a DMM mass fraction of 0.25) up to 9.8 μg/g (for a DMM mass fraction of 0.9). MORB have an average H₂O/Ce of ~ 250. Many OIB extend to lower H₂O/Ce, but as discussed above, this may in part reflect water loss during degassing or sparging. We therefore assume a H₂O/Ce ratio of the OIB-source mantle between 300 (average Australs melts) and 600 (near the upper limit defined by the global [H₂O]-H₂O/Ce trend).

Given the above constraints, the DMM is estimated to contain ~ 0.1–0.4 ocean mass equivalents of water, whereas the OIB-source mantle contains ~ 0.8–2.3 ocean mass equivalents. For an OIB-source mantle H₂O/Ce of 300, the OIB source contains 0.8–1.1 ocean mass equivalents (for an OIB reservoir mass of 0.1–0.75 BSE). For a H₂O/Ce ratio of 600, the water in OIB-source mantle increases to 1.7–2.3 ocean mass equivalents. The total mantle water content estimate ranges from ~ 1.2–2.4 ocean mass equivalents, somewhat higher than the mantle water content of ~ 0.9 ocean mass equivalents estimated by Bodnar et al. (2013). These estimates are independent of where in the mantle this water resides. Although previous studies have suggested that hydrous melting at the transition zone boundary to the upper and lower mantle could result in a “water filter” that effectively limits water transport out of the transition zone (e.g., Bercovici and Karato 2003), such melting would be equally effective at trapping incompatible trace elements in the transition zone as well. In the transition zone “water filter” model of Bercovici and Karato (2003), wet mantle ascending from the transition zone into the upper mantle experiences hydrous melting, and the generated melts, if denser than the ambient mantle, remain trapped at depth. However, the bulk D(H₂O/Ce) for upper mantle peridotite lithologies approaches unity, so that ascending depleted residues of such melting (the presumed source of OIB and MORB in this model) would preserve H₂O/Ce ratios close to their initial (transition zone) value. This creates a zero-sum equation where water and trace element enrichment of the transition zone needs to be balanced by depletion elsewhere. The transition zone cannot have an arbitrarily high water content without also having high [Ce], because the elevated H₂O/Ce ratios required in this case are not observed in OIB (Fig. 11). For the extreme case where most of the mantle is depleted, and the transition zone contains all of the enriched mantle components, we estimate a maximum water content in the transition zone of ~ 1.7 ocean masses. However, the transition zone can only represent a significant water reservoir to the extent that it also contains a significant fraction of Earth’s Ce and other incompatible trace elements. For the above scenario, the transition zone, which represents ~ 10% of the mantle mass, would need to be enriched ~ 18 × in Ce relative to DMM, and contain close to 10 ppm Ce. Current evidence does not support total mantle water content significantly in excess of ~ 2.4 ocean mass equivalents.

Fig. 11

Histogram of H₂O/Ce in OIB and MORB glasses and melt inclusions and melt compositions inferred from cpx phenocrysts (this study). Data sources as in Fig. 4

Our estimates for mantle water content are significantly lower than those of Peslier, et al. (2017), whose preferred estimate (~ 4 ocean mass equivalents) is nearly double our upper bound estimate. The primary difference between our estimates and those of Peslier et al. (2017) is they propose much higher [H₂O] in the transition zone and lower mantle. These estimates are based on assumptions regarding the degree of saturation of different mantle phases, and do not consider geochemical constraints based on H₂O/Ce ratios in OIBs that sample mantle plumes. However, their estimates require portions of the mantle to possess much higher H₂O/Ce than has been see to date in any plume-derived magmas. For example, if the depleted mantle, whose H₂O/Ce is well constrained from MORB to be ~ 250, represents 25% of the mantle mass, then mass balance requires the rest of the mantle to have a H₂O/Ce ratio of ~ 1023 if the mantle contains 4 ocean mass equivalents of water, with this ratio increasing with increasing size of the depleted mantle. Figure 11 shows histograms for H₂O/Ce ratios in MORB and OIB glasses, melt inclusions, and melts inferred from our cpx measurements. Bulk mantle H₂O/Ce required for different mantle water contents are shown for comparison. Water/Ce ratios in MORB and OIB sufficient to be compatible with more than ~ 2 ocean mass equivalents in the mantle are exceedingly rare. Although the present study suggests that magmatic degassing processes have likely significantly lowered H₂O/Ce ratios in many OIB glasses and melt inclusions, even H₂O/Ce ratios inferred from cpx phenocrysts formed at high pressure fail to extend to such high H₂O/Ce. Mantle water contents significantly higher than our estimate therefore require that mantle plumes somehow fail to sample large regions of the mantle with much higher H₂O/Ce, a prospect that appears unlikely.

Conclusions

Clinopyroxene phenocrysts from Austral Island basalts have [H₂O] ranging from 72 to 1019 ppm, indicating magmatic water contents ranging from 0.35 to 4.25 wt.%. Estimated magmatic water contents correlate with estimated cpx crystallization depths and melt inclusion entrapment/equilibration pressure. Although there is considerable scatter, the average inferred magmatic water content increases from 1.1 wt.% for phenocrysts with crystallization pressures < 200 MPa, to 1.5 wt.% for crystallization pressures of 200–300 MPa, to 1.9 wt.% for pressures > 300 MPa. Estimated magmatic water contents are higher than previously reported for Australs melt inclusions or most OIB estimated from quenched submarine glasses or melt inclusions. The observed correlations between crystallization pressure and magmatic water content suggest a significant role for magma degassing, but closed and open system degassing models fail to capture the observed trends. We suggest that sparging processes whereby CO₂-rich fluids generated by degassing of deeper magmas re-equilibrate with shallower melts with lower CO₂/H₂O ratios results in significant water loss at pressures < 300 MPa.

Melt water content and H₂O/Ce ratios in less-degassed samples (P > 300 MPa) do not correlate with radiogenic isotopes, but do correlate with melt CaO and La/Sm. The lack of correlation with isotopic composition is consistent with generally poor correlations observed in OIB globally, and may reflect the effects of magmatic degassing even at pressures > 300 MPa.

Compared to global compilations of OIB glass and melt inclusions, the magmatic water contents inferred for Australs melts from measurement of cpx phenocrysts extend to much higher [H₂O] at a given degree of melt enrichment (e.g., La/Sm or [Ce]). Similar results were previously reported for basalts from Bermuda (Mazza et al. 2019). Given the evidence that Australs melts experienced water loss at pressures exceeding the depths of quenching or entrapment for most glasses and melt inclusions, it is likely that many OIB suites have also lost significant water through CO₂ sparging and/or closed-system degassing (e.g., Anderson et al. 2024), and that previous estimates of OIB [H₂O] and H₂O/Ce are systematically too low.

Despite these concerns, there is little evidence from either the Australs or Bermuda phenocrysts for mantle reservoirs with H₂O/Ce much higher than ~ 300–600, though some individual samples do extend to higher values. This constrains the total mantle water budget to no more than ~ 2.4 ocean masses, and no more than ~ 1.7 ocean masses in the mantle transition zone.