High CO2 content in magmas of the explosive andesitic Enco eruption of Mocho-Choshuenco volcano (Chile)

Mocho-Choshuenco volcano has produced several highly explosive eruptions during its history, which make it one of the most hazardous volcanoes in the southern volcanic zone of Chile, although it is still relatively little studied to date. We present a geochemical study of the products of the sub-Plinian, andesitic, Enco eruption that occurred about 1600 years ago. We determined the major and trace elements compositions, as well as the volatile (H2O, CO2, Cl, and S) contents of melt inclusions trapped in minerals (olivine, plagioclase, and pyroxene) using electron microprobe, ion microprobe (SIMS), and 3D confocal Raman mapping. Though the whole-rock composition of the Enco magma is andesitic (60.2 ± 1.1 wt.% SiO2), the melt inclusions have SiO2 contents ranging from 50.3 to 67.3 wt.%, following the magmatic series of Mocho-Choshuenco, and the compositions of the most mafic melt inclusions are close to those of the most mafic erupted magmas. Geochemical modeling indicates that mixing occurred between a mafic magma and an andesitic-to-dacitic magma. Glass analysis revealed typical parental arc magma values for H2O (2.6–3.8 wt.%), S (116–1936 ppm), and Cl (620–1439 ppm). However, CO2 contents are very high in some melt inclusions with concentrations above 4000 ppm (measured in the glass), suggesting trapping depths >  ~ 17–22 km. Presence of solid carbonates inside inclusion-hosted bubbles clearly indicates that the CO2 contents measured in the glass phase were minimum values. We conclude that a CO2-rich basaltic magma ascended and mixed with a shallower andesitic magma. The magma cooled and exsolved high amounts of CO2, which may have dramatically increased the pressure and triggered the highly explosive Enco eruption.

The relatively poorly studied, and dangerous Mocho-Choshuenco (Chile) represents another case of a volcano presenting explosive eruptions fed by basaltic to andesitic magmas. To better disentangle the origin of its explosive behavior and bring new data on the local magmatism, we focused our investigations on the volatile content of the products of a recent explosive eruption: the Enco-MC15 (as defined by Rawson et al. 2015) eruption. We considered this event because it generated the most mobile pyroclastic flows of the recent eruptive history of Mocho-Choshuenco. In addition, field observations do not provide evidence of phreato-magmatism (Rawson et al. 2015, and our own work), hence excluding a crucial role of magma-water or magma-ice interaction in the explosive character of this eruption. We characterized major, trace, and volatile element compositions of olivine-, pyroxene-, and plagioclase-hosted melt inclusions in order to determine the pre-eruptive conditions of the magmas that fed the Enco-MC15 eruption at Mocho-Choshuenco. Finally, we present a possible scenario of the magmatic storage conditions and the key factors controlling the explosive behavior of the eruption.

Geological background and sampling
Mocho-Choshuenco volcano (2422 m a.s.l.) is located in Chile (39° 55′S 72° 2′W) in the Southern Volcanic Zone (SVZ) of the Andes (Moreno and Lara 2007; Fig. 1). This volcano has been characterized by 75 post-glacial (< 18 ky) mainly andesitic to dacitic eruptions, including Plinian events that occurred during the Neltume (11 ka-5.3 km 3 fallout deposit), the Pirehueico (9 ka-2.0 km 3 fallout deposit), the Enco-MC15, and the 1864 AD eruptions (Rawson et al. 2015). The Neltume and Pirehueico eruptions probably led to a partial collapse of the edifice, forming a 4-km-wide caldera that was later filled by successive tephra falls and pyroclastic flow deposits forming the El Mocho cone (Rawson et al. 2015). The total magma volume emitted during the last 18 ky was estimated to ~ 20 km 3 , making this volcano the most productive of the SVZ (Rawson et al. 2015). It represents a major hazard as more than 400 000 people are living in a 100 km area from the volcano (Rawson et al. 2015). The Enco-MC15 sub-Plinian eruption (Rawson et al. 2015) emitted ~ 1.7 km 3 of andesitic magma and generated pyroclastic fall and flow deposits. This event generated the most highly mobile pyroclastic flows of the recent eruptive history of Mocho-Choshuenco, with run-out distances of at least ~ 27 km from the vent in the eastern sector (Clavero et al. 2008).
We sampled an outcrop of a pyroclastic flow deposit of the Enco-MC15 eruption which belongs to unit UM4-5 as defined by Rawson et al. (2015). This outcrop, named VMJ32 in our study, is located on the Eastern flank of Mocho-Choshuenco volcano, 13 km away from the summit vent ( Fig. 1). It is composed of several flow units whose tops are outlined by large (10-30 cm) scoria. The material is composed of coarse (mean grain size, ϕ = − 3.35 to − 2.54) and well-sorted (sorting,σ ϕ = 1.16 to 1.24) juvenile pyroclasts, and rare (< 5 vol. %) lithic clasts. The upper part of these flow units was sampled in the framework of this study. We found charcoal whose 14 C dating gave an age of 1750 ± 40 BP (performed by Beta Analytic Inc., USA). After calibration using OxCal and the southern hemisphere calibration curve SH13 (Ramsey 2009), the age obtained is 1622 ± 86 BP. This is overlapping and consistent with calibrated ages of 1580 ± 115 BP given by Rawson et al. (2015) and Fontijn et al. (2016). Major element compositions of the whole rock samples were analyzed by ICP-AES at Laboratoire Magmas et Volcans, Clermont-Ferrand, France (Table 1, ± 10% relative error at the 2σ level). We focused our study on melt inclusion analysis, and the methods we used are presented below.

Preparation of the melt inclusions
The lapilli-sized (main fraction between 8 and 16 mm with few larger (up to > 32 mm) lapilli) pyroclasts yielding the most suitable phenocrysts to extract for investigation were selected. After crushing and sieving, the > 300 µm fraction was retained to separate phenocrysts from the glass. Olivines and pyroxenes were separated by density using Bromoform (CHBr 3 ), whereas the remaining plagioclases and glass were separated magnetically using a FRANTZ magnetic separator.
The selected crystal-hosted melt inclusions (hereafter denoted as MIs) were glassy, ellipsoidal, with shrinkage bubble(s) and no cracks. Inclusions aligned within fracture planes in the host minerals, due to late melt infiltration, were excluded.

Electron microprobe
Following the method of Moune et al. (2012), the major and volatile (S, Cl, and F) element compositions of 41 MIs, of the host crystals, and of the groundmass glasses were analyzed using a SX-100 CAMECA electron microprobe at Laboratoire Magmas et Volcans, Clermont-Ferrand (France) with a 15 kV accelerating voltage. Analyses of glass samples were done using a probe current of 8 nA for major elements and 40 nA for sulfur, chlorine, and fluorine volatile species, with a defocused beam of 5, 10, or 20 µm according to the size of the MIs. For each sample, prior to sulfur concentration measurement, the S Kα peak maximum was Enco-MC15 considered in this study is located in unit UM4-5 and is called VMJ32 (modified from Rawson et al. 2015). c Photograph of outcrop VMJ32 showing a coarse-grained pyroclastic flow deposit of the Enco eruption between finely grained surge units. The material analyzed was sampled from juvenile material located in the upper part of the outcrop. The measuring tape is two meters long first located by using the automated peak-scan routine, and the value of the peak position was recorded in the analysis setup (as described in Rose-Koga et al. 2020). Variations in the wavelength of sulfur Kα X-ray as a function of its oxidation state in silicate glasses (Wallace and Carmichael 1994;Métrich and Clocchiatti 1996) were taken into account during S analysis. For minerals analyses, the beam current was 15 nA, and the beam size was 1 µm. Spots were both acquired at the edge of the inclusions and at the mineral rim. The precision of the electron microprobe (2σ) was less than 10% for major elements, except for MnO, TiO 2 , and P 2 O 5 , and it was 6% for S and Cl. Fluorine data were excluded as they were all below the detection limit (< 370 ppm).

Mass spectrometry (SIMS and LA-ICPMS)
MIs larger than 40 µm in diameter were selected for mass spectrometry analysis. H 2 O and CO 2 contents of 16 MIs (7 olivine-and 9 plagioclase-hosted) were analyzed using the CAMECA IMS 1270 Ion Probe (SIMS) at the Centre de Recherches Pétrographiques et Géochimiques, Nancy (France), with a 15-μm beam size for all analyses. Pyroxenehosted MIs were not selected for SIMS and LA-ICPMS analysis being their size lower than the beam size. The selected crystals were transferred directly to indium sample holders, and no electron microprobe analyses were performed before using the SIMS in order to avoid carbon-coating contamination. Samples were gold coated and held in a gas chamber for degassing and cleaning ("pre-sputtering"). The glasses were sputtered with a 10 kV Cs + primary beam of 10-15 nA. When a high CO 2 (> 1500 ppm) content was detected, the analysis was performed three times to check for any possible carbon contamination. Before and after each analytical session, a series of international reference standard basaltic and andesitic glasses were used for calibration of water and carbon contents (measurements on standard material and calibration curves are available in Supplementary Material S1). The error (2σ) was 1.5% for H 2 O and 15% for CO 2 .
In situ trace element analyses of 10 MIs (4 olivine-and 6 plagioclase-hosted) were performed by laser ablation inductively coupled-plasma mass-spectrometry (LA-ICPMS) at Laboratoire Magmas et Volcans using a Resonetics M50 Excimer Laser with a 193-nm wavelength coupled to an ICPMS Agilent 7500cs. Raw data were processed with Glitter Software using CaO content as an internal standard (see method details and analytical errors in Venugopal et al. 2016).

Raman spectrometry
Raman measurements were carried out at the Laboratoire Magmas et Volcans using a Renishaw inVia confocal microspectrometer equipped with a 532-nm diode laser, a Peltier-cooled CCD detector, and a Leica DM 2500 M optical microscope with a motorized XYZ stage. Scattered light was collected in a back-scattered geometry. Laser power on the sample was reduced to 8-16 mW to avoid damaging the samples. A 2400 lines/mm grating, a 20-μm slit aperture (high confocality setting), and a 100 × microscope objective were used for the analyses. These analytical conditions resulted in spectral resolution of ~ 0.5 cm −1 and lateral spatial resolution of ~ 1 μm. Vertical resolution was ~ 2-3 μm near the surface but deteriorated with depth mainly due to light refraction at the air/glass/bubble interfaces. Daily calibration of the spectrometer was performed based on the 520.5 cm −1 peak of Si.
Two-or three-dimensional Raman mapping was performed on four MI-hosted bubbles. The 2D maps were acquired in the plane of the maximum bubble diameter with a step size of 1 µm. For 3D maps, we first performed depth profiles to define the vertical dimension of the map. The step size between acquisition points was 1 or 2 µm on the x-and y-axis and 2 µm on the z-axis. The acquisition time ranged from 20 to 60 s/point, and the spectra were continuously collected over 3 days in the ~ 200-1400 cm −1 wavenumber range. Each 3D mapping produced a single hypermap of several thousand of spectra that was processed using WiRE™ 4.2 software. We applied a polynomial baseline correction to the entire dataset. For each phase present in the bubble, we selected a characteristic peak, whose intensity variations in the analyzed volume informed about the spatial distribution of that phase in the bubble. The intense peak located between 1085 and 1097 cm −1 (Frezzotti et al. 2012) was used to map solid carbonates (Mg-calcite, magnesite, dolomite, and siderite), while a few grains of sulfides showing peaks near 320, 340, and 370 cm −1 were identified as either pyrrhotite or pyrite. The Fermi doublet peaks of fluid CO 2 at ~ 1286 and 1388 cm −1 were barely recognizable in some spectra as their intensities remained very weak, most likely due to low fluid density. The calibration by Lamadrid et al. (2017) was used to determine CO 2 density.
Several 2D sections along the z-axis were processed using Image J software to obtain the area fractions of each phase present inside the bubble. To estimate the volumetric fraction of each phase, the area fraction in each section was multiplied by the vol% of the associated spherical segment (i.e., the volume between adjacent x-y sections), and then the volumetric fractions of each phase in all spherical segments were summed (for further details, the reader is referred to Schiavi et al. 2020). The mass of carbonates was estimated from the measured volume and assuming a mean carbonate density value of 3.2 g/cm 3 . The total CO 2 concentration in a MI was given by the sum of the CO 2 dissolved in the melt and the CO 2 contained in the bubble. The inclusion volume was calculated assuming that inclusion thickness is equal to the arithmetic mean of the two inclusion axes visible in the polished plane, as it seems to be the most accurate method (Tucker et al. 2019). The errors associated with the estimated inclusion volume and CO 2 concentration are relatively high because they include the uncertainties related to the spatial resolution of the technique, the decreasing intensity of the Raman signal with depth, the approximation of the applied geometric model, and the assumption about phase densities (more details on the method can be found in Supplementary Material S2 and in Schiavi et al. 2020).

Petrography and textures of the melt inclusions
Samples contain plagioclase (mainly), clinopyroxene, olivine, rare orthopyroxene phenocrysts, and bubbles (see Supplementary Material S3). The phenocrysts represent ~ 5-10 vol.% of the whole rock. Finely crystallized plagioclase xenoliths are rare. Amphibole is absent, suggesting no excess of water. The microlites represent ~ 15-25 vol.% of the groundmass, whereas the vesicularity is ~ 30-40 vol.%. Figure 2 shows some of the textures commonly observed in MIs analyzed in this study. Three MI textures are identified in all studied host phases: (1) the typical rounded glassy case with one shrinkage bubble occupying less than 10% of the MI volume (Fig. 2a), (2) a shrinkage bubble occupying more than 50% of the inclusion volume (Fig. 2b), and (3) tens to hundreds of small bubbles ~ 1 to 10 µm in diameter, distributed along the rim or inside the whole MI volume, and occupying 10-50 vol. % of the MI, which are generally associated with a larger shrinkage bubble (Fig. 2c). The two last aforementioned textures represent 55% of the analyzed MIs.

Post-entrapment modifications
Following entrapment, MIs can undergo modifications that include bubble formation, host crystallization, chemical diffusion (e.g., Fe and Mg diffusion in olivine-and pyroxenehosted MIs), formation of daughter minerals, and volatile leakage Le Voyer et al. 2017). All analyzed MIs lacked daughter minerals and were not fractured. Crystallization of the host mineral, however, commonly leads to the formation of an overgrowth rim which may also modify the MI composition (see also , Schiano 2003). The presence of an overgrowth at the MI rim is assumed based on the disequilibrium between the melt and host-crystal given by K D values. This rim must be accounted for in order to determine the original melt composition. We used the well-established Fe-Mg distribution coefficient between olivine and melt to correct MIs compositions (K Dol-melt = [X Fe ol /X Mg ol ] / [X Fe melt /X Mg melt ] = 0.30 ± 0.03; Roeder andEmslie 1970, Toplis 2005). The Fe 3+ /Fe tot ratio was calculated to be ~ 0.2 based on the fO 2 (from NNO-0.5 to NNO + 0.5, Rawson et al. 2015). For the Enco-MC15 Mocho-Choshuenco eruption, the uncorrected K Dol-melt values obtained from measurements range between 0.10 and 0.24, indicative of post-entrapment crystallization (PEC). The oxide contents of the MIs were corrected by adding olivine (i.e., with the starting composition being the measured composition of olivine to which the MI belongs) incrementally to the inclusion composition until the equilibrium was reached, assuming that these corrected values represent the equilibrium compositions (PEC ranging from 2.5 to 9%). Uncorrected and corrected data for major, trace, and volatile elements composition, K D and PEC values are reported  Correction of pyroxene-and plagioclase-hosted inclusions from host crystallization is more challenging due to the uncertainty associated with their respective distribution coefficients. The Fe-Mg distribution coefficient between clinopyroxene and MIs measured in the studied samples is K D = 0.20-0.25, which is slightly lower, but within the uncertainty of the equilibrium K D of 0.27 ± 0.08 provided in literature (Putirka 2008). This suggests that host crystallization of pyroxene within the MIs was minimal. The measured K D between orthopyroxene and MIs is equal to 0.12-0.14. Unlike the clinopyroxene-hosted MIs, the calculated K D using Putirka (2008) where X Si is the fraction of silica cation in the melt) is ~ 0.30, indicating that the orthopyroxenehosted MIs are not at equilibrium. Consequently, these data were not used for modeling. The K D for the plagioclase (albite-anorthite)-liquid equilibrium found in literature is 0.10 ± 0.05 for a magmatic liquid with a temperature less than 1050 °C (Putirka 2008). In our samples, the temperature of the plagioclase-hosted MIs (andesitic composition) is estimated at 1035 ± 60 °C (see the "Petrography and textures of the melt inclusions" section), and the observed K D is between 0.07 and 0.17, suggesting that the melt may be at equilibrium with the host mineral. Taken as a whole, the major element compositions of most of the pyroxene-and plagioclase-hosted MIs follow the trend defined by the bulkrock samples (Fig. 3), suggesting negligible modification of the original melt composition by post-entrapment crystallization. As no correction was applied to the MIs hosted in pyroxene and plagioclase, the concentrations of volatile and incompatible elements in these inclusions could be slightly higher than the original concentrations in the trapped melt. In this regard, we point out that even a large amount of PEC (i.e., 10%) has a very low impact on concentration of incompatible elements.

Major and trace element compositions
The whole rock composition is andesitic, with 60.2 wt.% SiO 2 , 2.22 wt.% MgO, and 6.08 wt.% total alkalis (Table 1). The groundmass glass that represents the last stage in the melt evolution is andesitic and homogeneous, with 61.5 wt.% SiO 2 and 6.24 wt.% total alkalis representing the average of eight measurements. The MIs, however, have compositions ranging from basaltic to dacitic depending on the nature of the host mineral. The host olivines have a narrow forsterite range, Fo 75 -81 (Table 1). Melt inclusions from the same olivine host do not vary considerably in major element contents. The olivine-hosted MIs have well constrained oxide contents and are basaltic to basaltic-andesites in composition with: 50.3-55.9 wt.% SiO 2 , 3.08-5.15 wt.% MgO, and 0.51-1.28 wt.% K 2 O (Table 1, Fig. 3). These compositions are more primitive than the whole rock and groundmass glasses. The pyroxene (En 42-67 ) and plagioclase (An 44-51 and one An 76 ) hosted MIs are more evolved, being andesitic to dacitic in composition, with 56.3-67.3 wt.% SiO 2 , 0.78-3.39 wt.% MgO, and 1.43-2.08 wt.% K 2 O. Moreover, these compositions are generally more evolved than those of the groundmass and whole rock. These data can be compared with the eruptive glass compositions at Mocho-Choshuenco that range from ca. 55 wt.% to ca. 73 wt.% SiO 2 (Rawson et al. 2015) and the erupted whole rock compositions that range from ca. 50 wt.% to ca. 68 wt.% SiO 2 . The most mafic MIs compositions, in particular, are close to that of the most mafic erupted magmas, whereas the most evolved MIs are following the general trend of Mocho-Choshuenco erupted magmas.
Trace element compositions in the inclusions show typical arc-like patterns with low concentrations of HFSEs such as Nb, and enrichment in U, Pb, and Ba, as shown in the spider diagram (Supplementary Material S4). The compositional variability of the observed trends is related to the composition of the host minerals, so that the plagioclasehosted MIs are more enriched in trace elements, whereas more primitive MIs in olivines display lower trace element concentrations.

Volatile element compositions
The different volatile species show contrasting patterns as a function of the degree of differentiation (i.e., increasing K 2 O content, Fig. 4). The H 2 O contents in olivine-hosted inclusions range from 2.62 to 3.10 wt.%, except for an inclusion with a large bubble (Ol6-13_MI1, 0.48 wt.%). The plagioclase-hosted MIs present a large variability in H 2 O content, ranging from 1.21 up to 3.84 wt.%. The highest H 2 O concentration is found in a MI with hundreds of small bubbles hosted in An 76 plagioclase (Pl2_MI1). The plagioclase-hosted MIs show a general trend of decreasing H 2 O content with increasing K 2 O (Fig. 4c), which is consistent with crystallization accompanied by degassing of a H 2 O-saturated magma.
The CO 2 content measured in the MI glass is below 1100 ppm in most cases, independent of the host mineral, but four inclusions show much higher values of 2836 ppm in an olivine-hosted MI (Ol6-5_MI1), and 2710 ppm (Pl6_MI1), 4213 ppm (Pl1_MI1), and 4430 ppm (PlM1_MI1) in plagioclase-hosted MIs. Two of the inclusions mentioned above (Ol6-5_MI1 and PlM1_MI1) have several bubbles inside the inclusion (hundreds of them in Ol6-5_MI1, see Fig. 2c). The two other inclusions with high CO 2 content have a regular shape with a shrinkage   1% from 1 to 5% FC, 5% from 5 to 30% FC, and 10% from 30% to the end of the FC. Error bars are 2σ. See Table 1 for values bubble similar to the MI shown in Fig. 2a. Furthermore, it is worth noting that the two other MIs containing more than a hundred of small bubbles (Ol6-4_MI1 and Pl2-MI1) have relatively high CO 2 contents in the glass: 873 ppm and 1075 ppm, respectively. Raman analysis of olivinehosted inclusions containing a shrinkage bubble revealed precipitation of solid carbonates at the rim of the bubble. In the analyzed bubbles (Ol6-4_MI1, Ol6-5_MI1, Ol6-14_MI1, and Ol6-15_MI1 MIs), the volume of the solid carbonates ranges from 5 ± 2 to 26 ± 7% of the bubble volume (supplementary material S2). Consequently, the quantity of CO 2 within the bubble represents 41 to 93% of the total amount, excepted for Ol6-5 with only 10 to 20% of the CO 2 inside the bubble (see supplementary material S2). Carbonates were observed only in the bubbles of olivine-hosted MIs (Fig. 5). When CO 2 stored in carbonates is taken into account, the corrected values of CO 2 for Ol6-4_MI1, Ol6-5_MI1, Ol6-14_MI1, and Ol6-15_MI1 (Fig. 5) Allison et al. 2021). These results suggest that the magma was originally very rich in CO 2 , with contents possibly larger than 4000 ppm. The S content decreases with increasing degree of differentiation (i.e., increasing K 2 O; Fig. 4b). It ranges from 605 to 1936 ppm in olivine-hosted MIs. It is in general lower in pyroxene-and plagioclase-hosted MIs (excepted two inclusions with 712 and 1184 ppm). Considering oxidation conditions at Mocho-Choshuenco between NNO-0.5 and NNO + 0.5 (Rawson et al. 2015), the S content at saturation in basaltic melts is estimated at 1500 ± 500 ppm at 200 MPa (Wallace and Carmichael 1992;Lesne et al. 2015) and is expected to vary from 1500 to 4500 ppm as fO 2 increases at ~ 1 GPa (Jugo 2009). The large variability in the S content of the olivine-hosted MIs (Fig. 4b) suggests saturation of the basaltic magma. In contrast, the Cl content generally  The yellow area corresponds to sulfur; S in the bubble is not discussed here; however, it confirms that the bubble is also a reservoir for S, as suggested in Venugopal et al. (2020). A photograph of the bubble is shown below the map increases with differentiation (Fig. 4a), which is consistent with its incompatible behavior during differentiation. It ranges from 620 to 1118 ppm in olivine-hosted MIs and from 773 to 1439 ppm in pyroxene-and plagioclase-hosted MIs. Considering the Cl solubility in a hydrated magma (less than 1 wt.% Cl; Balcone-Boissard et al. 2016), the basaltic and andesitic melts of the Mocho-Choshuenco Enco eruption were undersaturated in Cl.

Magmatic processes and storage conditions
The MI compositions permitted us to estimate their trapping temperature and pressure. In order to estimate MIs trapping temperatures, olivine-liquid thermometry was used for basaltic MIs. Rb/Co versus Rb, c Sr/Y versus Sr, and d Zr/V versus Zr. The blue curve represents output of mixing calculation model (Günduz and Asan 2021) between the most primitive and the most evolved measured MIs. The diamonds represent increment of 10% mixing by addition of the evolved end-member. The modeling fits very well with the measurements, supporting magma mixing as a main process in melt evolution. In contrast, modeled fractional crystallization (FC) paths (black dashed line) obtained using Petrolog3 (Danyushevsky and Plechov 2011) do not fit with the entire range of measured compositions, excepted in the case of Sr/Y versus Sr content. The black diamonds represent increments with steps of: 1% from 1 to 5% FC, 5% from 5 to 30% FC, and 10% from 30% to the end of the fractional crystallization. Distribution coefficients of the selected trace elements were obtained from LaTourrette and Burnett (1992), Hart and Dunn (1993), Forsythe et al. (1994), Bindeman et al. (1998), andZanetti et al. (2004) hydrous conditions, yields temperatures of ~ 1050-1150 °C (note that the standard error of estimate (SEE) is 29 °C). It is well-known that water loss can occur due to H + diffusion through the host olivine (Chen et al. 2011). In the set of studied MIs, there is no clear evidence of post-entrapment water loss (except for Ol6-13_MI1); however, we cannot exclude its occurrence. In this regard, lapilli smaller than 29 mm in size are expected to have similar cooling behavior to ash, with limited water loss (Lloyd et al. 2013). Thus, for the olivine-liquid thermometry, the calculation was based on all MIs except Ol6-13_MI1. For andesitic and dacitic MIs, plagioclase-liquid thermometry, using Eq. 24a from Putirka (2008), yields temperatures of ~ 975-1095 °C (with SEE of 36 °C). Additionally, clinopyroxene-liquid thermometry, using the pressure-independent parametrization of Putirka et al. (1996), results in temperatures of ~ 1045-1105 °C (with SEE of 69 °C).
We estimated trapping pressure of the MIs using the models of Newman and Lowenstern (2002), Papale et al. (2006) and Witham et al. (2012) for basaltic MIs, and the experimental study of Botcharnikov et al. (2006) for andesitic melts. When considering this approach, loss of volatiles (H 2 O, CO 2 , S) from the melt into the bubble and/or to the host crystal must be taken into account. Diffusion of volatiles from the melt to the bubble can be important, as shown by Raman investigations (e.g., Esposito et al. 2016;Robidoux et al. 2018;Venugopal et al. 2020;Schiavi et al. 2020); therefore, the amounts measured in the glass are minimal estimates. We calculated that a basaltic liquid containing 0.3 wt.% CO 2 and 3.0 wt.% H 2 O is in equilibrium with a fluid phase containing a mole fraction of H 2 O ≤ 0.1 at a temperature of 1100 °C and a pressure of 600 MPa. The very low amount of H 2 O that is predicted to be present inside the bubble would most likely be in the form of a very thin fluid film on the bubble wall ), which may not be detected during the analyses, in agreement with our observations. Moreover, significant amounts of CO 2 and S can be stored in the bubble as mineral precipitates (carbonates, sulfides, or sulfates; Venugopal et al. 2020;Schiavi et al. 2020); thus, MIs volatile contents can be underestimated if their presence is neglected. Diffusion of volatiles from the melt to the bubble can partially explain the variability of CO 2 contents in the glass (Fig. 4d). Assuming melt saturation in both H 2 O and CO 2 and considering the inclusions with the highest volatile contents, we obtain the following trapping pressure range: 560-720 MPa (17-22 km depth) for the two most enriched basaltic olivine-hosted MIs (i.e., Ol6-15_MI1 with 3077 ± 744 ppm CO 2 and Ol6-5_MI1 with 3155 ± 190 ppm CO 2 ) based on the corrected CO 2 content (melt + bubble, see Supplementary Material S2 for details on the correction calculations); 490 ± 10 MPa (14.5-15 km depth) for the andesitic plagioclase-hosted MIs (i.e., PlM1_ MI1 and Pl1_MI1 with 4430 and 4213 ppm CO 2 , respectively). The latter is a minimal pressure estimate because the CO 2 content of the bubbles in plagioclase-hosted MIs is not known.
The major element compositions of MIs define two partially overlapping clusters: (1) with most olivine-hosted MIs being basaltic and representing parental magma and (2) plagioclase and pyroxene-hosted MIs (and olivine-hosted Ol6-12_MI1) exhibiting more evolved compositions, i.e., mainly andesitic. The MI major element contents seem to follow a differentiation trend from the parental to the more evolved magma, with a compositional gap between the two groups of MIs. The presence of this compositional trend suggests that the two groups of MIs share the same origin. To investigate the role played by fractional crystallization, we modeled the evolution of melt compositions using Petrolog3 software (Danyushevsky and Plechov 2011). Details on the models selected are provided in Supplementary Material S5. The starting composition is defined by the less evolved MI composition (i.e., M16 MI), and starting pressure is fixed at 720 MPa, corresponding to the maximal estimated trapping pressure, while the starting temperature is fixed at 1100 °C. The water was treated incompatibly as the effect of CO 2 on H 2 O solubility is not taken into account in the models used. The two fractional crystallization models that better fit the major element data are shown in Fig. 3. In both models, the composition of the melt evolves by fractional crystallization of olivine, clinopyroxene, plagioclase, and magnetite in presence of H 2 O. The first model (1) set at  O contents diagram for the studied MIs. The green circles represent the corrected CO 2 content for the basaltic MIs Ol6-4_MI1, Ol6-5_MI1, Ol6-14_MI1, and Ol6-15_MI1. Closedsystem (with and without vapor in excess) and open-system degassing paths starting from the composition of the CO 2 -and H 2 O-richest basaltic MI have been calculated using VolatileCalc (Newman and Lowenstern 2002). Most of data are well explained by degassing under closed-system conditions. In addition, the high CO 2 content measured in two plagioclase-hosted MIs suggests the occurrence of CO 2 flushing from depth. One sample (Pl2_MI1, 3.84 wt. % H 2 O) is well outside this path and could be an indicator of a volatile-supersaturated melt (Pichavant et al. 2013) constant pressure (P = 720 MPa) partially reproduces the compositional variability of the mafic olivine-hosted MIs but stops at 56% crystallization, due to depletion of TiO 2 , before reaching more evolved andesitic and dacitic compositions. Instead, the second model (2) of fractional crystallization associated to a fast decompression rate of the magma (− 5 MPa.°C −1 ) can reproduce the entire compositional range of major element data within about 75% of fractional crystallization. Despite the quite good fit obtained for the major element composition, fractional crystallization cannot be the only process responsible for the evolution from the basaltic end-member to the andesitic-dacitic one because of at least three main reasons. First, the final mineral compositions obtained in both models are not in agreement with the measured compositions in the investigated samples, i.e., Fo 64 -An 61 -En 54 and Fo 45 -An 43 -En 27 for models (1) and (2), respectively, against measured mineral compositions of Fo 75-81 -An 44-51 /An 76 -En 42-67 . Possibly, as a significant CO 2 content in the melt and the presence (rare) of orthopyroxene in the investigated samples are not considered in the models used, the accuracy of the output may be low. Thus, some care should be taken with these modeling results. Second, simple calculations demonstrate that variations in absolute trace element concentrations as well as in trace element ratios (Fig. 6) cannot be explained by a process of fractional crystallization alone. A good fit is only observed for Sr/Y versus Sr contents (Fig. 6c), probably due to the fact that Sr has a compatible behavior with plagioclase (Bindeman et al. 1998). Instead, fractional crystallization can partially account for the compositional variability (major and trace elements) of the olivine-hosted basaltic melts. Third, as the mafic and evolved end-members in the MIs were coerupted, it is not likely that they were directly related by only fractional crystallization, with the host minerals being most probably phenocrysts. In fact, physical models taking into account the evolution of parameters such as magma density, viscosity, and crystal nucleation/growth rates (Brophy 2009) cannot explain the occurrence of an extreme degree of fractional crystallization (i.e., up to 75%) coupled to rapid magma ascent.
We consider trace elements to assess if the observed compositional trends are the result of a process of magma mixing. In the case of major elements, mixing lines between averaged compositions of olivine-hosted MIs and (1) (2) (1) Mocho-Choshuenco (Enco), Chile, this study. Uncorrected (MI glass only) (2) Mocho-Choshuenco (Enco), Chile, this study. Corrected (MI glass + MI bubble) Corresponding MIs Fig. 8 Comparison of CO 2 contents in MIs of the Enco magma (red diamonds on the left side of the diagram, showing exceptionally high contents) with data of other arc volcanic systems. The diagram shows studies giving "uncorrected" CO 2 contents (measured only in the MI glass) and "corrected" CO 2 contents (i.e., taking into account the CO 2 content of both MI glass and bubble). The mean associated error with these data is ~ 15% plagioclase/pyroxene-hosted MIs were calculated (but not shown in Fig. 3 for clarity) and may explain the clusters observed. However, as fractional crystallization possibly occurs, especially in the case of basaltic melts, major elements do not show well-defined linear trends. PetroGram (Gündüz and Asan 2021) was used to model magma mixing for trace elements (Fig. 6). Highly to moderately incompatible elements like Th, Rb, Ce, Y, and Zr have low mineral-melt partition coefficients, while Sr is compatible in plagioclase, Co has strong affinity with olivine (and to a lesser extent, clinopyroxene), and V is slightly incompatible in clinopyroxene under slightly oxidizing conditions (e.g., see Laubier et al. 2014;Geochemical Earth Reference Model 2018). The magma mixing appears as a hyperbolic curve in incompatible/incompatible or incompatible/compatible versus incompatible trace element plots (Fig. 6). Incremental mixing from the most primitive melt measured (Ol6-14_MI1) to the most evolved end-member (represented by Pl5_MI1 sample) reproduces successfully trace element data. At the scale of the lapilli, mingling-like textures were observed with clearer portions in contact with darker portions. Thus, the compilation of sample observations, major, trace element data, and modeling suggests that the Enco eruption occurred as a result of mixing between a hot basaltic magma (represented by olivine-hosted MIs), which was evolving by fractional crystallization at depth, and a colder, more evolved, andesitic to dacitic magma (represented by plagioclase and pyroxene-hosted MIs).

Volatiles in the Enco magma
The different textures in the MIs reflect various formation and/or post-entrapment processes. While the majority of the analyzed MIs show a regular shape and exhibit only one shrinkage bubble, the olivine-hosted inclusion Ol6-13_MI1 with a single large bubble (Fig. 2b) has the lowest water (0.48 wt.%) and CO 2 (225 ppm) contents, which suggest fast decompression of the inclusion (depressurization and then vesiculation or decrepitation) (Lowenstern 1995), and important volatiles loss. Among the studied inclusions, the olivine-hosted inclusion Ol6-5_MI1 (Fig. 7) has the highest measured CO 2 content in the glass (2836 ppm) and the highest corrected total (glass + bubble) CO 2 concentration of 3155 ± 190 ppm for this mineral phase. However, when taking into account the correction incertitude, the inclusion Ol6-15_MI1 may display the highest corrected CO 2 content with 3077 ± 744 ppm. Higher CO 2 contents have been measured in plagioclase-hosted MIs with up to 4430 ppm in PlM1_MI1 (uncorrected data, glass content only). A peculiar texture has been observed in this sample and also in Ol6-5_MI1 (Fig. 2c), Ol6-4_MI1, and Pl2_MI1, with numerous small bubbles evenly distributed that suggest fast exsolution of the volatiles and quenching of the small bubbles hence formed. The presence of a large number of very small bubbles preserved in the glass of several inclusions indicates volatile supersaturation causing a high nucleation rate and is possible evidence of a very rapid syneruptive decompression followed by rapid quench, as shown by theory and experimental results (Mourtada-Bonnefoi and Laporte 2004;Toramaru 2006;Le Gall and Pichavant 2016). It is worth noting that the two olivine-hosted MIs showing this texture are those with the highest CO 2 contents in the glass, whereas high CO 2 concentration in the glass is not systematically associated to a large number of bubbles in plagioclase-hosted MIs. Moreover, both textures (i.e., bubble-rich, and single bubble) are observed in the same tephra unit. These observations can be explained by the high sensitivity of bubble nucleation to small variations in initial conditions (e.g., Mourtada-Bonnefoi and Laporte 2004;Toramaru 2006;Le Gall and Pichavant 2016), including the internal pressure and volatile concentration in the melt, but also to post-entrapment crystallization processes, thermal contraction, and elastic deformation (volumetric re-equilibration) (Schiavi et al. 2016;Drignon et al. 2020) that affect MIs before quenching. An early degassing before the MIs entrapment could explain the large variability of CO 2 , H 2 O, and S contents observed at a given K 2 O value (Fig. 4). Additionally, mixing between the mafic magma and the andesitic-dacitic one is able to affect the volatile contents, as observed in the case of S (Fig. 4b). Most data plotted in the CO 2 vs. H 2 O contents diagram (Fig. 7) can be explained by a closed-system equilibrium degassing of the melt in presence of excess vapor. This is consistent with the strong explosivity behavior observed for the Enco eruption, as a closed-system equilibrium degassing would keep the volatiles under pressure (see also, Cassidy et al. 2018). The high CO 2 content measured in some MIs, in particular in two plagioclase-hosted MIs, also suggests the occurrence of CO 2 flushing from depth, as observed in several other volcanic systems (Blundy et al. 2010). We do not have a clear explanation for the high H 2 O content (3.84 wt.%) of the inclusion Pl2_MI1 hosted in the Ca-rich plagioclase (An 76 ) that plots above the main trend, but we cannot exclude the possibility that it is a previously formed xenocryst or that it registered supersaturation of the magma in a deep portion of the system (Pichavant et al. 2013).
The H 2 O, S, and Cl contents of the Enco magma are within the ranges given for other arc volcanic systems (Wallace 2005). The CO 2 contents in some of our MIs, however, are exceptionally high (up to 4430 ppm in the glass of PlM1_MI1, Table 1). The four most enriched MIs have CO 2 contents higher than those given by studies on other arc volcanoes in South and Central America (generally no more than 2000 ppm; see Fig. 8). In those studies, CO 2 in the bubbles was generally not taken into account. Two exceptions are the MIs from San Cristobal (Nicaragua), for which corrected CO 2 contents are 718-1889 ppm versus 71-527 ppm for uncorrected data (Robidoux et al. 2016), and the MIs from Fuego (Guatemala) with corrected CO 2 contents of 513-2598 ppm versus 189-511 ppm for uncorrected data (Moore et al. 2015). Recent work by Rasmussen et al. (2020) gave corrected CO 2 contents for Fuego (Guatemala) ranging from 370 to 3409 ppm.
A possible mechanism of CO 2 enrichment at upper levels could be a gas flushing event (Caricchi et al. 2018), which seems in agreement with our observations (Fig. 7). According to the trace elements analysis, the magma of the Enco eruption of Mocho-Chochuenco volcano derived from a mixing between a deep and primitive basaltic magma rising from more than 22 km in depth and an evolved andesitic to dacitic magma stored in a hypothetic, shallower reservoir at a minimum estimated depth of 15 km. These estimates, as already stated previously, are minimal. However, only few studies in the literature report the composition of magmas erupted at Mocho-Choshuenco volcano. Moreover, no geophysical study, showing the position of a magmatic reservoir is, until now, available at Mocho-Choshuenco in order to confirm the minimal depth estimates obtained with MI investigations. Thus, our data are also important in providing this first vital information for future investigations that will better define magma storage conditions below Mocho-Choshuenco. The recharge of the magma chamber by this primitive CO 2 -rich magma has possibly resulted in the enrichment in CO 2 and the triggering of the eruption. Hence, the high CO 2 content could have been the main cause of the explosivity of the Enco eruption event that generated highly mobile pyroclastic flows. In fact, other possible causes of explosivity seem not significant. The high temperature (> 1000 °C) and the low crystallinity (< 30 vol.%, including macrocysts and microliths) of the erupted magma both suggest that the magma had a low viscosity and was not prone to an explosive regime (Lavallée et al. 2007). From the model of Vetere et al. (2006) for andesitic melts, with a water content of ~ 3.0 wt.%, the viscosity of the magma that fed the Enco eruption is estimated to be ~ 10 2 -10 3 Pa.s, which is in the low viscosity range, while the high viscosity range is between 10 8 and 10 13 Pa.s (Vetere et al. 2006). As a comparison, the range of viscosities estimated at Mocho-Choshuenco for the Enco eruption is at least three orders of magnitude lower than the effusive eruption of Quizapu volcano (Chile), which emitted a 5 km 3 andesitic-dacitic mingled lava flow in A.D. 1846-1847, with pre-eruptive H 2 O contents of ~ 4-6 wt.% (Ruprecht and Bachmann 2010). The effusive behavior of this eruption, despite the relatively high viscosity of its magma, was explained by a late reheating following a mixing of dacite with andesite recharge; leading to an acceleration of volatile diffusion; enhancing bubble nucleation, growth, and coalescence; and inhibiting brittle fragmentation (Ruprecht and Bachmann 2010). In the case of the Quizapu, the andesitic-dacitic mixing has prevented an explosive eruption to occur; as generally, silicic magma explosivity is strongly controlled by the viscosity (Cassidy et al. 2018). In contrast, the magma mixing which occurred at Mocho-Choshuenco seems to have rather favored the overpressure, possibly due to the high CO 2 contents, leading to the sub-Plinian Enco eruption. The CO 2 contents measured at Mocho-Choshuenco are also significantly higher than those recorded in some other Chilean volcanoes such as Llaima or Villarica, with CO 2 contents measured in MI glass below 500 ppm (see Fig. 8; Witter et al. 2004;Bouvet de Maisonneuve et al. 2012).
Finally, H 2 O, S, and Cl contents are within the range of other arc volcanic systems around the world, with no exceptional concentrations measured (Wallace 2005). In this context, the low solubility of CO 2 (Bottinga and Javoy 1990;Le Gall and Pichavant 2016;Laporte et al. 2016) could have controlled the eruption dynamics. Moreover, CO 2 is known to significantly decrease the solubility of H 2 O (Lowenstern 2001 and references therein). Water contents are also high; hence, high amounts of exsolved volatiles (CO 2 and H 2 O) could have increased the magma buoyancy, generated high overpressures, and favored fast ascent of the magma and explosive fragmentation (Papale and Polacci 1999;Blundy et al. 2010;Freda et al. 2010;Aiuppa et al. 2011;Dallai et al. 2011;Allison et al. 2021). Owing to the exceptionally high CO 2 and fairly high H 2 O contents we report, it appears that the magma of the Enco-MC15 eruption at Mocho-Choshuenco was particularly prone to explode violently despite its moderate silica content.

Conclusions
We measured the volatile contents of MIs in minerals extracted from juveniles of a pyroclastic flow deposit formed during the andesitic Enco-MC15 eruption of Mocho-Choshuenco volcano. Though H 2 O, S, and Cl contents are in ranges given for other arc volcanic systems, the CO 2 contents are exceptionally high compared to those given in published studies, with concentrations up to 4430 ppm. The initial CO 2 content was probably even higher, as suggested by the MI textures (i.e., presence of hundreds of small bubbles) as well as precipitated carbonates inside shrinkage bubbles. 3D Raman mapping allowed us to quantify the volume of solid carbonates trapped in inclusion bubbles and, thus, correct the CO 2 content in the MIs. As a whole, the volatile contents indicate that the basaltic MIs hosted in olivines were trapped at minimal depths of ~ 17-22 km, whereas trapping of andesitic to dacitic melts in plagioclase crystals occurred at ~ 15 km depth.
We conclude that the exceptionally high content and the low solubility of CO 2 , combined with the fact that this volatile phase enhanced exsolution of H 2 O, could have caused high overpressure in the magma reservoir and led the crystal-poor Enco andesitic magma of low viscosity to erupt explosively. A possible scenario considers that the deeper basaltic CO 2 -rich magma ascended and recharged a shallower reservoir where the andesitic-dacitic magma was ponding. Once there, the mafic magma began to cool, exsolved volatiles, and mixed with the evolved magma, which resulted in rising of gas pressure (similar to gas sparging of Bachmann and Bergantz 2006); possible rapid bubble expansion associated with fast magma ascent finally triggered the explosive eruption. We note that the wide range of glass compositions and textures (i.e., scarce presence of phenocrysts and microliths, hundreds of bubbles preserved in the inclusion glass) of the studied samples seem to testify to a mild mixing process associated with a very rapid ascent of the recharging magma through the shallower storage levels, most likely caused by the strong enrichment in CO 2 of the basaltic melt. Further investigations on other explosive eruptions of Mocho-Choshuenco volcano are needed to confirm if the CO 2 enrichment is a typical feature throughout the life of the edifice or if it concerned only the Enco eruption. Our study supports the conclusions of earlier works that pointed out the possible correlation between high CO 2 contents and explosivity of poorly to moderately differentiated magmas. It suggests also that initial CO 2 contents in arc magmas may be commonly underestimated and that further work is required to better constrain the role that CO 2 may have in promoting explosive eruptions of basic to intermediate magmas.

Acknowledgements
The authors thank the Associate Editor, Tobias Fischer, as well as Tom Shea, and an anonymous reviewers for their constructive and helpful reviews. We thank Jean-Luc Devidal (LMV) for assistance with the electron microprobe and LA-ICP-MS and Etienne Deloule for his help with the ion probe at CRPG Nancy.
Funding Open access funding provided by University of Vienna. This research was financed by the French National Research Institute for sustainable Development (IRD), the French Government Laboratory of Excellence initiative No. ANR-10-LABX-0006, the Region Auvergne and the European Regional Development Fund, and the Chilean Science Commission (Fondecyt projects 1070162 and 1090387). This is Laboratory of Excellence ClerVolc contribution number 533.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.