Focused Mid-Crustal Magma Intrusion During Continental Break-Up in Ethiopia

Significant volumes of magma can be intruded into the crust during continental break-up, influencing rift evolution by altering the thermo-mechanical structure of the crust and its response to extensional stresses. Rift magmas additionally feed surface volcanic activity and can be globally significant sources of tectonic CO 2 emissions. Understanding how magmatism may affect rift development requires knowledge on magma intrusion depths in the crust. Here, using data from olivine-hosted melt inclusions, we investigate magma dynamics for basaltic intrusions in the Main Ethiopian Rift (MER). We find evidence for a spatially focused zone of magma intrusion at the MER upper-lower crustal boundary (10–15 km depth), consistent with geophysical datasets. We propose that ascending melts in the MER are intruded over this depth range as discrete sills, likely creating a mechanically weak mid-crustal layer.

depths in active rifts obtained through petrology and geochemistry remain limited.While geophysical observations can infer depths of intrusion and crustal melt storage (e.g., through seismicity concurrently triggered during emplacement; Keir et al., 2006;Ebinger et al., 2008), only petrological observations, obtained from basaltic materials derived directly from the intruding melts themselves, can provide first-hand evidence of the magmatic conditions associated with crustal emplacement.
The Main Ethiopian Rift (MER), comprising the northernmost sector of the East African Rift system (EARS), provides a natural laboratory to examine the interplay between rift geodynamics and magmatic intrusion.This late-stage continental rift, which bridges the large fault-bound grabens of the Kenyan Rift and inferred incipient seafloor spreading in Afar (Figure 1a), has been extensively studied through multiple geophysical approaches (e.g., Bastow et al., 2011;Lavayssière et al., 2018;Wright et al., 2016).These studies suggest that significant magma intrusion has occurred in the MER lithosphere, focused under ∼20 km-wide and ∼60 km-long magmatic-tectonic segments (e.g., Bastow et al., 2011), where as much as half of the crustal volume may comprise new igneous material (Daniels et al., 2014;Maguire et al., 2006).The compositional and thermal effects of magma intrusion may modify the response of the Ethiopian crust to extension, controlling where and how strain is localized as rifting proceeds (e.g., Bastow & Keir, 2011;Lavecchia et al., 2016).Furthermore, degassing of intruded melts during and after emplacement contributes to the significant diffuse CO 2 fluxes measured in the MER (Hunt et al., 2017).
In this study we use petrological methods to investigate the storage depths and compositional diversity of intruded basaltic magmas in the northern MER.Our constraints on magma intrusion conditions are derived from analysis of olivine-hosted silicate melt inclusions (MIs), which are small pockets of quenched magma trapped within growing crystals during crustal magma storage (e.g., Wallace et al., 2021).Unlike erupted lavas, MIs can preserve magmatic volatile contents (e.g., CO 2 , H 2 O etc., Wallace et al., 2021), allowing volatile saturation pressures, and therefore magmatic storage depths, to be determined (e.g., Ghiorso & Gualda, 2015).Of particular importance is the volatile species CO 2 , which degases strongly with decreasing pressure in basaltic magmas (e.g., Dixon et al., 1995).Continental rifts, including the MER, are known to be significant sources of passively degassing magmatic CO 2 (Foley & Fischer, 2017;Hunt et al., 2017;Lee et al., 2016).By considering the total CO 2 in MIs, entrapped within both glass and bubble, we provide new well-constrained petrological estimates of basaltic intrusion pressures in the MER.
Olivine crystals from two cones (Figure 1) were picked from disaggregated scoria, individually polished to 0.25 μm grade on glass slides to expose MIs, and mounted in epoxy resin.We have measured the compositions of 40 MIs (full methods in Supporting Information S1).27 of these MIs contain CO 2 -rich vapor bubbles, which form from post-entrapment changes in pressure, volume and temperature (e.g., Maclennan, 2017;Moore et al., 2015).Bubbles can host a significant fraction of the MI CO 2 budget (e.g., Hartley et al., 2014;Wieser et al., 2021).To estimate the total CO 2 in MIs, essential for accurate barometry, 18 MIs were additionally assessed for bubble CO 2 density using Raman spectroscopy.Our approach differs from previous studies considering MIs from the EARS in this regard, which have opted to either (a) experimentally rehomogenize the bubble (Head et al., 2011;Hudgins et al., 2015), (b) use CO 2 equation of state methods (Rooney et al., 2022), or (c) select MIs without vapor bubbles wherever possible (Field, Barnie, et al., 2012;Field, Blundy, et al., 2012;A. Donovan et al., 2017;Iddon & Edmonds, 2020).Benefiting from recent developments in the calibration of the Raman method through the use of standard materials (e.g., Lamadrid et al., 2017;Wieser et al., 2021), the primary advantage of our approach is the direct measurement of bubble CO 2 without making assumptions concerning post-entrapment processes or experimentally modifying MI glass compositions, which will introduce uncertainties that are difficult to assess and quantify (Rasmussen et al., 2020;Wieser et al., 2021).In addition, by selecting bubble-hosting MIs we avoid biases towards magmatic conditions that favor bubble-free MIs, which may not be representative of crustal melt storage.By doing so, we provide robust estimates of total CO 2 in MIs, which can be used to determine crustal melt storage pressures.
After Raman spectroscopy, all MIs were analyzed for trace and volatile elements in the glass phase by secondary ion mass spectrometry, followed by electron probe microanalysis to assess major element compositions of MI glass, carrier melt, and host olivine crystals.MI compositions were corrected for post-entrapment crystallization (PEC) using Petrolog3 software (Danyushevsky & Plechov, 2011, see Supporting Information S1).The total CO 2 of MIs is calculated by mass balance using the CO 2 measured in the MI bubble and glass (e.g., Hartley et al., 2014).To complement the MI compositional data set, we have additionally assessed major and trace element whole-rock compositions of basalts collected from several scoria cones and fissure flows in the region using x-ray fluorescence and solution inductively coupled plasma mass spectrometry respectively.All standards and geochemical data are presented in Data Set S1.

Magma Intrusion Depths in the Main Ethiopian Rift
Our key barometric and geochemical results are presented in Figures 2 and 3 (additional figures in Supporting Information S1).MIs are entrapped within olivine crystals of composition Fo 76−88 , and there are no systematic differences in major, trace, or volatile element concentrations between MIs collected from the two cones in this study (Data Set S1).CO 2 concentrations range from 35 to 5,770 ppm in MI glass only; MIs with CO 2 measurements in both the glass and vapor bubble have total combined CO 2 contents of 1,895-3,248 ppm, with 15%-46% of the CO 2 residing within the bubble (Data Set S1).Where an unanalyzed shrinkage bubble is present, CO 2 contents are assumed to be minima and we estimate the plausible range of total CO 2 using our bubble CO 2 density measurements (see Supporting Information S1).H 2 O concentrations display less variability: discounting the three MIs that have clearly degassed (containing ≤0.4 wt% H 2 O), MIs have mean H 2 O of 1.1 ± 0.2 wt% (Figure S6 in Supporting Information S1), which is comparable to H 2 O concentrations obtained from other MER and EARS MIs (Iddon & Edmonds, 2020;Rooney et al., 2022).
Volatile saturation pressures of MIs are calculated using the fully thermodynamic MagmaSat volatile solubility model (Ghiorso & Gualda, 2015) via the Python library VESIcal (Iacovino et al., 2021;Wieser et al., 2022).Other volatile solubility models are considered in Supporting Information S1.Entrapment pressures for MIs for which total CO 2 contents are known (vapor bubble and glass), determined at a magmatic temperature of 1,200°C (Iddon et al., 2019;Wong et al., 2022), vary over a relatively narrow range from 2.5 to 4.5 kbar (Figure 2a).In  the MER these pressures correspond to depths of ∼10-15 km (assuming a constant upper-mid crustal density of 2.79 g cm −3 ; Cornwell et al., 2006), among the deepest recorded volatile saturation depths for continental rift magmas (Figures 2b-2d).Pressures recorded by MIs without bubbles overlap partially with those that do have analyzed bubbles; however, the average CO 2 concentration and therefore pressure of MIs without a bubble is typically lower than those with a bubble.Two MIs for which only inclusion glass CO 2 is known record higher pressures in excess of 5 kbar (∼20 km), corresponding to the MER lower crust.Overall, our barometric results show a relatively limited distribution of magma storage depths with a narrowly focused zone of intrusion centered at ∼12 km depth, coincident with the seismically imaged boundary between the upper and lower crust in the MER (Maguire et al., 2006), and in close agreement with MI volatile saturation pressures from the Turkana Depression south of the MER (Figure 2c; Rooney et al., 2022).

Melt Inclusion Trace Element Compositions
The major element compositions of MIs overlap with carrier basalt compositions and whole-rock compositions of erupted lavas (Data Set S1; Tadesse et al., 2019;Nicotra et al., 2021).Incompatible trace element concentrations vary considerably in both MIs and lavas, but nonetheless still overlap (Figures 3a and 3b).Greater primary compositional variability is preserved in MIs over whole rocks, for example, the ratios La/Yb and Dy/Yb (Figures 3c and 3d), which persists to lower MgO in MIs.
By comparing CO 2 concentrations with trace elements with similar behavior during mantle melting (e.g., Ba, Rb), CO 2 degassing from mantle melts can be assessed (e.g., Le Voyer et al., 2018).While primary magmatic CO 2 ).MIs are categorized by analyzed components.MIs analyzed for glass CO 2 but not bubble CO 2 (green diamonds) show trails to pressures assuming the mean bubble CO 2 density of our sample set (0.188 g cm −3 ; see Supporting Information S1).Error bars on pressures calculated from MIs for which bubble and glass are analyzed are 1σ.(b-d) Violin plots of volatile CO 2 -H 2 O saturation pressures recorded by mineral-hosted MIs from the EARS and Afar calculated using MagmaSat (Ghiorso & Gualda, 2015).Saturation pressures are individually determined for each MI using their recorded major and trace element composition and magmatic temperatures.Where FeO t is provided without Fe 2 O 3 all Fe is assumed to be Fe 2+ .Subfigure B shows distributions of silicic MIs (SiO 2 > 60 wt%), subfigure C shows basaltic MIs (SiO 2 < 55 wt%), and subfigure D shows the basaltic MIs of this study.Subfigures B and C consider CO 2 and H 2 O in melt inclusion glass only, including experimentally rehomogenized inclusions (triangle markers).The blue line and shaded area across all subfigures marks the mean and 1σ of the MI subset of this study with combined vapor bubble and glass CO 2 .References: 1. Field, Blundy, et al. (2012); 2. Iddon and Edmonds (2020); 3. Field, Barnie, et al. (2012) contents are not known for MER magmas, the highest observed CO 2 /Ba and CO 2 /Rb ratios approach the mantle values inferred from undegassed MORB and Icelandic lavas (Rosenthal et al., 2015;Hauri et al., 2017;Le Voyer et al., 2018).Assuming that initial CO 2 -trace element ratios are somewhat similar to other mantle-derived melts, CO 2 /Ba systematics for MER MIs clearly show evidence for significant degassing of CO 2 prior to entrapment, with ∼50%-95% of initial CO 2 likely having been exsolved during ascent to mid-crustal pressures (Figure 3e).

Depths of Intrusion in the East African Rift
Our total CO 2 saturation pressures determined from vapor bubble and glass are in broad agreement with maximum pressures of melt storage estimated from MI volatiles at other EARS sectors (see Figures 2b-2d).Applying the same volatile solubility modelling performed on our MIs to literature datasets, we determine that our proposed 10-15 km depth range for basalt storage coincides with the deepest MIs at other parts of the EARS and Afar Rift (Figure 2c; e.g., A. Donovan et al., 2017;Rooney et al., 2022).Geophysical observations of crustal melt movement suggests that melt focusing at these pressures may be ubiquitous within the Eastern Branch of the EARS in addition to the MER (Reiss et al., 2021(Reiss et al., , 2022;;Weinstein et al., 2017).
The lack of evidence for significant melt storage within the upper crust in our data set contrasts with the depth distributions for magma storage obtained from suites of MIs collected at large caldera-forming volcanic centers found along the MER (Figure 2; Iddon & Edmonds, 2020).Under these silicic centers, melt storage appears to extend upwards into the upper crust, where evolved magmas are generated via low pressure fractionation (Iddon & Edmonds, 2020).Notably, the maximum storage depths under caldera complexes in the EARS identified both from MI volatile saturation barometry (Figure 2; Iddon & Edmonds, 2020;Rooney et al., 2022) and from mineral barometry (Iddon et al., 2019;Rooney et al., 2005) coincides with the 10-15 km depth range observed in our data set.This depth range may therefore be the locus of initial basaltic melt emplacement along the MER, with important implications for heat distribution within the rifting crust and therefore crustal strength profiles (Buck, 2006;Daniels et al., 2014;Lavecchia et al., 2016), such as the creation of a mid-crustal weak layer (Muluneh et al., 2020).With the exception of those below caldera complexes/silicic volcanoes (e.g., Biggs  Tadesse et al., 2019;Nicotra et al., 2021).Liquid lines of descent with crosses denoting 10% fractionation intervals are determined from our three highest MgO melts using Rhyolite-MELTS v1.2.0 (Gualda et al., 2012, see Supporting Information S1), assuming Rayleigh fractionation with the partition coefficients collated by Iddon and Edmonds (2020).PEC corrections are detailed in Supporting Information S1.E. Olivine-hosted MI CO 2 plotted against Ba.Degassing lines plotted from primary CO 2 /Ba of MORB (81.3;Le Voyer et al., 2018), with shaded area representing range of mantle CO 2 /Ba (48.3-133Rosenthal et al., 2015;Hauri et al., 2017Hauri et al., ). et al., 2011)), upper crustal melt bodies (<10 km depth) in the MER are likely to be ephemeral, perhaps forming during periodic intrusive-eruptive episodes (e.g., Ebinger et al., 2013).
In contrast to the extensive MI data corresponding to mid-crustal pressures, very few MIs from our data set and the MER data set of Iddon and Edmonds (2020) record pressures corresponding to the lower crust or Moho (Figure 2; e.g., Maguire et al., 2006;Lavayssière et al., 2018).Considering the evolved compositions of our olivines (mean Fo 80 ) relative to Fo 90 olivines in other MER volcanic materials (e.g., Rooney et al., 2005), we posit that an initial stage of fractionation near the Moho prior to ascent to mid-crustal pressures is necessary.This hypothesis is supported by low wavespeeds observed at Moho depths from the presence of melt in the heavily intruded lower crust (Chambers et al., 2019;Keranen et al., 2009), and numerical models suggesting that the lowermost crust is weak, hot and underlies a lower-crustal brittle-ductile transition at 20-25 km (Lavecchia et al., 2016;Muluneh et al., 2020).Melts pooling and fractionating at the base of the crust may bypass the ductile lowermost crust entirely if both density differences between melt and crust and lower crustal strain rates are sufficiently high (Muluneh et al., 2021).
Deep CO 2 degassing in the MER, likely derived from degassing of magmas as they ascend towards the weak mid-crust, is focused along discrete fault zones (Hunt et al., 2017;Raggiunti et al., 2023).By making assumptions on the volumes of melt intruded into the crust (e.g., Iddon & Edmonds, 2020), we determine that the difference between expected CO 2 concentrations in primary mantle melts and those recorded in MIs is sufficient to generate the CO 2 fluxes measured from surface degassing (Figure 3e Hunt et al., 2017, see Supporting Information S1).The restriction of significant degassing to localized regions in the MER (Hunt et al., 2017) may suggest that some regions are subject to active intrusion at the present day whereas other portions are not.Future studies should aim to constrain this periodicity of melt emplacement.

Compositional Heterogeneity in Melt Inclusions
Variability in absolute trace element concentrations (Figures 3a and 3b) could result from fractional crystallization of distinct parental melts and/or mixing between variably fractionated melts with distinct origins.In contrast, the broader distribution of trace element ratios observed in MIs relative to whole rocks (Figures 3c and 3d) can only be inherited from the compositional heterogeneity of parental mantle-derived melts.Such variability, derived from the melting of multiple source lithologies (e.g., Shorttle & Maclennan, 2011) and/or unmixed fractional mantle melts (e.g., Gurenko & Chaussidon, 1995), is preserved at lower MgO contents.
Physical interactions between intrusive bodies therefore appear to be restricted, and we infer that intruded magmas reside in a series of discrete bodies emplaced over a relatively narrow depth range.The slightly lower degree of compositional diversity observed in erupted lavas (Figures 3c and 3d), even at higher MgO, suggests that some mixing does occur prior to eruption and that dyke intrusion into the upper crust may involve partially homogenized melts sourced from multiple mid-crustal sills.Erupted melts extend to lower MgO than the MIs (after PEC corrections), and pre-eruptive mixing and homogenization therefore likely occurs during a final stage of differentiation within upper crustal magma bodies.

Basaltic Melt Focusing in the Main Ethiopian Rift
The barometric and compositional data from our MIs suggest that intrusion in the MER crust is characterized by emplacement of multiple discrete magma bodies over a relatively narrow depth range coincident with the seismically identified upper-lower crustal boundary.Although our results do not directly constrain the geometry of intrusive bodies, we suggest, based on the narrow range of MI entrapment depths, that these are likely mid-crustal sill complexes.This model is supported by geophysical observations of the MER crust.Strong horizontally oriented seismic anisotropy observed in the MER at depths of 5-15 km is consistent with the presence of sills (Bastow et al., 2010;Chambers et al., 2021;Kendall et al., 2006).Low seismic moment earthquakes in northern MER magmatic segments are distributed within a narrow depth band between 8 and 16 km and have been interpreted as being triggered by movement or emplacement of mid-crustal melts (Daly et al., 2008;Keir et al., 2006).High-Vp, high-Vp/Vs and high-density bodies are inferred to be present at these depths under Boku and other MER segments (Cornwell et al., 2006;Daly et al., 2008;Keranen et al., 2004;Nigussie et al., 2022), as are high-conductivity crustal anomalies (Whaler & Hautot, 2006), all indicative of partially molten mid-crustal intrusions.Our results are also in good agreement with empirical observations relating MER cone clustering to intrusion depths (Mazzarini et al., 2013).To summarize, the melt storage depths resolved directly using petrological methods concur closely with the deepest intrusion pressures determined using geophysical techniques.
Focusing of ascending basaltic melts at this depth range can, to a first order, be attributed to MER crustal density structure, as the mean density of the lower crust exceeds that of our MIs (mean of 2.708 g cm −3 , calculated after PEC corrections using DensityX, Iacovino and Till (2019); cf.e.g., Cornwell et al. (2006)).Driven by density differences, basaltic melts will rise to mid-crustal depths before they achieve neutral buoyancy, stall and crystallize.The upper crust, comparatively less dense than the lower crust, will limit the ascent of basalt melts beyond the focusing zone (Cornwell et al., 2006;Mickus et al., 2007).
Melt focusing in the mid-crust could also be attributed to the rheological structure of the crust.Numerical models based on seismic observations suggest that the 10-15 km depth range resolved using our MIs coincides with the weakest part of the Ethiopian crust, which is sandwiched between two strong brittle layers in the upper and mid-lower crust (Muluneh et al., 2020).The strong, lower-density brittle crust above this ductile zone, combined with the density limitations discussed above, likely inhibits further ascent of the buoyant melt (Cornwell et al., 2006;Muluneh et al., 2020).Melt may only progress directly to the surface through the breaking of dyke-induced faults (e.g., Casey et al., 2006), by exploiting pre-existing crustal weaknesses (e.g., Le Corvec et al., 2013), or after extensive fractionation to form lower-density silicic melts (e.g., Gleeson et al., 2017).
We therefore hypothesize that the intrusion and emplacement of melts into this weak, ductile mid-crust will have a strong effect on the overall rheology of the rifting crust, which in turn may govern how the crust locally accommodates strain in response to far-field extensional stresses.Ductile stretching may accommodate crustal deformation at a different rate or manner relative to the brittle layers above and below this weak zone, in turn possibly dictating that future batches of melt are focused in the same region.Indeed, the development of crustal sill systems in the MER may arise from pulsed emplacement of magmas from the lower crust or mantle (e.g., Annen et al., 2015).Stacked sills formed in this manner may maintain high localized temperatures in the crust, which can facilitate further intrusion of melt at shallower pressures, or may themselves contract during cooling to generate accommodation space for further intrusions (Magee et al., 2016).Future numerical or analog models of rift deformation in Ethiopia must account for the development of a hot, ductile, weak layer in the crust, and the influences such a layer may have on overall crustal rheology.

Summary
The results of our study are summarized in Figure 4. Through the careful analysis of major, trace, and volatile elements in olivine-hosted MIs, we propose that stacked mid-crustal sills in the depth range of 10-15 km are the dominant form of magmatic storage in the MER (Figures 4a and 4b).Intrusions are known to be discrete and horizontally oriented from trace element variability (Figure 3) and seismic anisotropy respectively (e.g., Chambers et al., 2021), and develop as a consequence of repeated magmatic intrusion into the mid-crust during the progression of late-stage continental rifting.Initially crystallizing at or near the Moho, mantle-derived magmas bypass the ductile lowermost crust to arrive at the Ethiopian mid-crust, heralded by seismic activity during emplacement (Figure 4c).These melts, stored as discrete sills in the weak, ductile mid-crust, are blocked from further ascent by a strong, lower density upper crust (Figure 4d).The diverse range of trace element ratios observed in MIs gives evidence to limited melt mixing in the crust; partial mixing of magmas between sills may occur in the shallow crust prior to eruption (Figure 3).Petrological evidence for mid-crustal sills in the MER presented in this study is in agreement with geophysical observations (e.g., Keranen et al., 2004), and the volatile composition of basalts comprising these bodies are consistent with CO 2 degassing rates measured at the rift floor (Hunt et al., 2017).The presence of hot sills in the MER mid-crust has important implications for how intruding melts in late-stage rifts affect and are affected by the rheological structure of the crust, and should be considered a key element in future development of continental rifting models.KW and PW are funded by NERC DTP studentships NE/L002574/1 and NE/ L002507/1 respectively.KW acknowledges additional support from the Geologists' Association New Researchers Award.SIMS analyses are funded through NERC grant IMF694/1119 awarded to DF. ME acknowledges the support of COMET via NERC.We thank Cristina Talavera Rodriguez and Lesley Neve for performing geochemical analyses on our behalf during the COVID-19 pandemic.Yared Sinetebeb, Harri Wyn Williams, Emilie Ringe, and Richard Walshaw are thanked for assistance with fieldwork, sample preparation, Raman spectroscopy, and EPMA respectively.We acknowledge thought-provoking discussions with Ian Bastow, Emma Chambers, Tim Craig, Sophie Hautot, Derek Keir, Tyrone Rooney, and Kathy Whaler, and constructive comments from James Muirhead and Ameha Muluneh on an earlier version of this manuscript.Finally, we acknowledge the Ethiopian Ministry of Mines and Oromia State Administration for sampling and field permissions.

Figure 1 .
Figure 1.(a) Overview map of the MER, highlighting the location of the Boku Volcanic Complex.Inset figure shows area presented in subfigure (a) within Ethiopia.(b) Simplified geological map of Boku within the Gedemsa magmatic segment, with melt inclusion (MI) and whole-rock (W) sample localities shown.Digital elevation models are GTOPO30 (a; Gesch et al., 1999) and SRTM (b; Farr et al., 2007).Volcano locations in subfigure (a) are obtained from the Global Volcanism Program, Smithsonian Institution (https://volcano.si.edu/).Faults in subfigure b of Agostini et al. (2011).

Figure 2 .
Figure 2. (a) Volatile CO 2 -H 2 O saturation pressures of olivine-hosted MIs from the MER plotted against MI olivine host composition (olivine Fo = 100⋅Mg/ [Fe + Mg]).MIs are categorized by analyzed components.MIs analyzed for glass CO 2 but not bubble CO 2 (green diamonds) show trails to pressures assuming the mean bubble CO 2 density of our sample set (0.188 g cm −3 ; see Supporting Information S1).Error bars on pressures calculated from MIs for which bubble and glass are analyzed are 1σ.(b-d) Violin plots of volatile CO 2 -H 2 O saturation pressures recorded by mineral-hosted MIs from the EARS and Afar calculated using MagmaSat(Ghiorso & Gualda, 2015).Saturation pressures are individually determined for each MI using their recorded major and trace element composition and magmatic temperatures.Where FeO t is provided without Fe 2 O 3 all Fe is assumed to be Fe 2+ .Subfigure B shows distributions of silicic MIs (SiO 2 > 60 wt%), subfigure C shows basaltic MIs (SiO 2 < 55 wt%), and subfigure D shows the basaltic MIs of this study.Subfigures B and C consider CO 2 and H 2 O in melt inclusion glass only, including experimentally rehomogenized inclusions (triangle markers).The blue line and shaded area across all subfigures marks the mean and 1σ of the MI subset of this study with combined vapor bubble and glass CO 2 .References: 1.Field, Blundy, et al. (2012); 2.Iddon and Edmonds (2020); 3. Field, Barnie, et al. (2012); 4. Hudgins  et al. (2015); 5.Head et al. (2011); 6. (a)Donovan et al. (2017); 7. Rooney et al. (2022), without bubble corrections.
Figure 2. (a) Volatile CO 2 -H 2 O saturation pressures of olivine-hosted MIs from the MER plotted against MI olivine host composition (olivine Fo = 100⋅Mg/ [Fe + Mg]).MIs are categorized by analyzed components.MIs analyzed for glass CO 2 but not bubble CO 2 (green diamonds) show trails to pressures assuming the mean bubble CO 2 density of our sample set (0.188 g cm −3 ; see Supporting Information S1).Error bars on pressures calculated from MIs for which bubble and glass are analyzed are 1σ.(b-d) Violin plots of volatile CO 2 -H 2 O saturation pressures recorded by mineral-hosted MIs from the EARS and Afar calculated using MagmaSat(Ghiorso & Gualda, 2015).Saturation pressures are individually determined for each MI using their recorded major and trace element composition and magmatic temperatures.Where FeO t is provided without Fe 2 O 3 all Fe is assumed to be Fe 2+ .Subfigure B shows distributions of silicic MIs (SiO 2 > 60 wt%), subfigure C shows basaltic MIs (SiO 2 < 55 wt%), and subfigure D shows the basaltic MIs of this study.Subfigures B and C consider CO 2 and H 2 O in melt inclusion glass only, including experimentally rehomogenized inclusions (triangle markers).The blue line and shaded area across all subfigures marks the mean and 1σ of the MI subset of this study with combined vapor bubble and glass CO 2 .References: 1.Field, Blundy, et al. (2012); 2.Iddon and Edmonds (2020); 3. Field, Barnie, et al. (2012); 4. Hudgins  et al. (2015); 5.Head et al. (2011); 6. (a)Donovan et al. (2017); 7. Rooney et al. (2022), without bubble corrections.