1.1. Vulcanello magmatic source and products

Based on the analysis of fluid inclusions in quartz xenoliths, Zanon et al. [2003] provided evidence that the deepest level of magma storage at Vulcanello is located between 17 and 21 km, close to the Moho [21–25 km; Falsaperla et al. 1985] and excluded significant magma ponding at mid-crustal levels. This is in accordance with the aeromagnetic data at Vulcano pointing to a magma reservoir between 18 and 21 km depth [De Ritis et al. 2013]. Moreover, Zanon et al. [2003] provided and estimate for Vulcanello magmas temperature of 1083 ± 40 °C.

The formation of Vulcanello occurred during the last Eruptive Epoch of the island of Vulcano [Eruptive Epoch 8, De Astis et al. 2013], between AD 1100 to 1250 [Arrighi et al. 2006], simultaneously to effusive and explosive activity in the La Fossa crater [Fusillo et al. 2015].

The lava platform comprises several superimposed shoshonitic lava flows, which, together with the products of explosive activity that form a scoria cone, were the first subaerial products of the peninsula (Vulcanello 1). The rocks composing the lava platform are mainly shoshonitic and contain large (up to 1.5 cm) xenocrysts of plagioclase [Davì et al. 2009, De Astis et al. 2013, Fusillo et al. 2015]. After a short time break, the activity resumed with explosive eruptions (Vulcanello 2) which led to the formation of a second cone during a single eruptive unit, showing 3-m-thick fallout deposit having shoshonitic composition [Fusillo et al. 2015]. Following Vulcanello 2, no activity took place during ∼500 years. Subsequently, lava emission resulted in the emplacement of the latitic Roveto lava flow, that was followed by three explosive phases leading to the edification of Vulcanello 3 cone. These rocks are latitic in composition with the presence of clinopyroxenes megacrysts [ca. 1.5 cm; Fusillo et al. 2015]. The activity of Vulcanello 3 ended with the emplacement of a final lava flow (Valle dei Mostri latitic lava flow), which flowed in the same direction than the Roveto latitic lava flow [Fusillo et al. 2015, Nicotra et al. 2018, Davì et al. 2009].

2.1. Starting material and experiments

The starting material of our experiments is a natural scoria lava from Vulcanello 1. The bulk rock composition (Table 1) is characterized by SiO₂ = 53.40 wt% and Na₂O + K₂O = 8.76 wt%. In the total Alkali-Silica (TAS) diagram this composition plots in the field of shoshonite (Table 1). The modal abundance of phenocrysts, mainly composed of clinopyroxene (Cpx) and plagioclase (Pl), is 8–11 vol%. The phenocrysts often contain inclusions of Fe–Ti oxides. As reported in Davì et al. [2009] the composition of clinopyroxene and feldspar phenocrysts ranges between Wo_49–44En_36–42–36Fs_9–20 and An_5–46Ab_51–52Or_3–43, respectively. Feldspar microphenocrysts show a similar composition to that of phenocrysts (An_3–35Ab_51–61Or_4–46). Olivine has a compositional range Fo_42–62 and the Fe–Ti oxides have Usp content close to 14 mol%. The groundmass consists of glass, leucite (Lc), Fe–Ti oxides, Cpx, Pl and small amount of olivine (Ol).

Sample preparation prior to high-pressure experiments was performed at the Petro-Volcanology Research Group laboratories of the University of Perugia (PVRG; http://pvrg.unipg.it). About 100 g of crushed bulk rock was melted in a Pt crucible in air at 1600 °C for 4 h in a Pt₈₀Rh₂₀ crucible. Melting was performed in a Nabertherm HT 04/17 MoSi₂-heated box furnace. The melt was quenched to a glass by pouring it on a brass plate. The glass was crushed, re-melted and quenched again using the same procedure. This technique ensured compositional homogeneity of the glass and avoided crystallization [Vetere et al. 2015a]. The composition of the glass and the absence of crystals were controlled by electron microprobe imaging and measurements with a CAMECA SX 100 at the Institute of Mineralogy University of Hannover (see details below; Table 1).

The NBO/T parameter corresponds to the number of non-bridging oxygens (NBO) per tetrahedrally coordinated cation (T) and is a good proxy to depict the structure of silicate melts in terms of polymerization [Mysen and Richet 2005]. This ratio can be calculated as follows:

In order to prepare volatile-bearing charges, we used the following procedure:

The capsules (Au₈₀Pd₂₀; 5 mm diameter, 3 cm in length, previously annealed at 1100 °C for 10 min and then welded on one end) were filled with the following procedure: (a) the capsule was charged with the desired amounts of ±H₂O ± Ag₂C₂O₄ + glass powder (Ag₂C₂O₄ is the source for CO₂ as the silver oxalate decomposes during heating and generates the carbon dioxide); (b) a steel piston was used to gently compact the charge; (c) to avoid any volatiles release, the capsule was tightly squeezed on top, rolled up in a wet tissue, and frozen by placing it into a bath of liquid nitrogen; (d) the upper end of the capsule was welded shut with a conventional graphite-arc welder. In agreement with literature data on the Vulcanello magmatic source (21–17 km depth and 1083 ± 40 °C, see above) we performed experiments at 500 MPa and 1080 °C for 70 h, in order to shed light on processes relevant to the magmatic source of Vulcanello magma.

A total of 13 experiments with variable volatile contents (see Figure 1) were performed at 500 MPa and 1080 °C (Tables 2 and 3). Experiments were done in an internally heated pressure vessel (IHPV) equipped with a rapid-quench sample holder at the Institute fur Mineralogy, Leibniz University of Hannover (Germany). Details for the IHPV are provided in Berndt et al. [2002]. Temperature was measured using four unsheathed S-type thermocouples (precision of temperature was ±5 °C). The IHPV was pressurized with argon. Pressure was monitored   using digital pressure transducers having an uncertainty of about 1 MPa. The variation of pressure during the experiments was ⩽5 MPa. Some experiments were performed at the intrinsic oxygen fugacity (fO₂) conditions of the pressure vessel which was found to be close to QFM+4 (the intrinsic fO₂ corresponding to H₂O-saturated conditions), with QFM being the quartz–fayalite–magnetite equilibrium [Berndt et al. 2002]. At H₂O-undersaturated conditions, fO₂ is lower, depending on the prevailing water activity [QFM−0.8 to QFM+3; Botcharnikov et al. 2005]. In our experiments fO₂ was calculated using the following relation:

2.2. Electron microprobe and image analyses

All run-products were mounted in epoxy, ground flat and polished for textural and chemical analyses. Glasses and minerals produced during the experiments were analysed using a Cameca SX100 microprobe at the Institute of Mineralogy. An accelerating voltage of 15 kV was used with a beam current of 15 nA for silicate minerals and oxides and 6 nA for glasses. The beam was defocused to at least 15 μm for glass analyses and focused to <2 μm for the analysis of crystal phases. All elements were analysed with a counting time of 10 s on peak. Standards used for calibration were Fe₂O₃, MgO, MnTiO₃ (Mn and Ti), albite (Na), wollastonite (Si for mineral and glass and Ca for glass), apatite (P and F for glass and F, Ca and P for apatite), orthoclase (K), anhydrite (Ca and S). Standard deviations reported in Table 2 were calculated based on 10 to 20 analysed spots on the same sample. Raw data were corrected with the software “Peak Sight” and “PAP” matrix [Pouchou and Pichoir 1991]. Precision and accuracy were determined by measuring reference glasses VG-568 (rhyolite) and VG-2 (basalt) from Smithsonian standards collection [Jarosewich 2002].

Due to the presence of crystals in experimental charges, the water content of the glasses was estimated following the “by-difference” method described in Devine et al. [1995]. Following Parat et al. [2008], we estimate that the error on determination of H₂O content is ∼0.5 wt%. CO₂ in glasses were not determined but estimated from the solubility curve in Figure 1.

Back-scattered electron (BSE) images were collected with a Philips FEG (Field Emission Gun) Quanta 200F equipped with a Si/Li-SUTW detector (EDAX, Philips Electronics) installed at the University of Calabria (Italy). Representative BSE images were collected at different magnifications (400× up to 1600×) depending on the crystal size. Surface percentage of the different phases was determined for all images using grey levels with Image-ProPlus 6.0 [see details in Vetere et al. 2015a]. Seven to ten BSE images obtained in different parts of the same section were evaluated for homogeneity. We quantified surface fractions of glass, clinopyroxene, spinel, plagioclase and gas bubbles. In total, 110 BSE images were analysed.

4.1. Effect of volatiles on phase equilibria

In order to test for equilibrium condition in our experiments we can refer to models proposed by Putirka [2008]. Tests for equilibrium between Cpx and a coexisting liquid can be made by comparing observed and predicted values for Fe–Mg exchange, or K_D(Fe–Mg)^cpx–liq, which should be 0.28 ± 0.08 for Cpx (Putirka,2008; considering FeO =FeO_tot). Our results show values ranging from 0.236 and 0.248 when using compositions in Table 2, indicating compositions close to equilibrium.

The glasses of our isothermal experiments show geochemical trends that are entirely controlled by the degree of crystallinity that itself depends on the composition and amount of volatiles in the system. The evolution trends for most of the elements (Ca, Mg, Si, Al) are consistent with the crystallization of Cpx (Figure 2), which predominates the solid phase. Ti is almost constant in experiments performed at reduced conditions whereas its abundance is lower in the experiments performed under oxidized conditions, due to the crystallization of Fe–Ti oxides in the latter. The composition of melts from two experiments (VL46 and VL47 at very low H₂O content; Tables 2 and 3) deviates slightly from the general trend for some elements and can be explained by the presence of Pl in the run products (Figure 5).

The amount of CO₂ that can be dissolved in melts is lower than that of H₂O and the solubility of both volatiles has been investigated as a function of pressure in a variety of melt compositions [e.g., Brooker et al. 2001, King and Holloway 2002, Holloway and Blank 1994, Dixon and Stolper 1995, Dixon et al. 1995, Jakobsson 1997, Morizet et al. 2010, Baker and Alletti 2012, Schanofski et al. 2019]. Carbon species (CO₂, CH₄, and CO) have low solubility in basalt and shoshonite melts [at 500 MPa < 4500 ppm CO₂, Shishkina et al. 2014, Vetere et al. 2014, Behrens et al. 2009, Botcharnikov et al. 2006] and carbonate minerals are not stable in mafic melts. It could therefore be anticipated that (for low CO₂ melt concentrations) CO₂ has no effect on phase compositions and proportions when compared to CO₂-free dry systems. Indeed, samples with high water contents do not show any important differences in terms of crystallinity between CO₂-bearing and CO₂-free samples (Figure 3). This indicates that water plays the most important role on controlling the crystallization temperatures and that the role of minor to moderate CO₂ concentrations is subordinate. However, for systems with low water and relatively high CO₂ levels, this observation seems not to be valid. The difference of 10% in crystallinity found between nominally dry samples (fluid absent and pure CO₂ fluid) implies that CO₂ does not behave as a fully inert component, only reducing water activity. Following the determinations of Husen et al. [2016], it is emphasized that the melts synthesized at nominally dry fluid-absent conditions in the internally-heated pressure vessel used at Hannover probably contain at least 0.5 wt% H₂O. Absolutely dry conditions can never be realized because entrapment of adsorbed water during capsule preparation cannot be avoided and because hydrogen can diffuse through the capsule material at high temperature. This process is partially related to reduction of ferric iron to ferrous iron in the melt during the high T–P experiments following the reaction

Although experimental conditions are very different, our observations are in agreement with observations of Fiege et al. [2015] showing that the addition of ∼2000 ppm of CO₂ to a hydrous basaltic melt (5 wt% H₂O) has a small effect on crystal content (Cpx, Spl) and glass fractions (during decompression). Nevertheless, when evaluating in detail those data, we can see that samples containing only H₂O show relatively higher abundance of Cpx compared to those containing both H₂O and CO₂. Experiments simulating the effects of CO₂ fluxing conducted by Giuffrida et al. [2017] also indicate that the addition of CO₂ to a partially crystallized basalt can influence the phase proportion and result in a partial dissolution of Cpx, in contrast to the common assumption. Vetere et al. [2015b] noted that the crystallinity of Etna basalt decreases as CO₂ is added to the system, which is in agreement with the data from this study on a shoshonitic system. In particular, in those experiments, Pl was the most abundant mineral phase, but Pl was absent when only CO₂ was added to the experimental charge. In addition, CO₂ appeared to lower Cpx abundance [Vetere et al. 2015b]. Thus, CO₂ may not act as a complete inert phase in mafic systems in which it is mainly incorporated as carbonate species.

A closer look on data presented in Figure 3 allows us to further discuss the effect of the CO₂ in magmatic systems. In fact, Figure 3 shows, at least for added water content higher than 2 wt%, a nearly linear correlation between crystallinity and added water content, independently on fO₂. In Figure 6 we show that, if the amount of added CO₂ is larger than 0.65 wt% and added water content is close or lower than 2 wt%, the crystallinity of the run products becomes lower than what would be predicted from the nearly linear trend between crystallinity and added water, as highlighted by the dashed black and red lines (Figure 6).

More in details, the flushing experiment performed by Giuffrida et al. [2017], showed that a CO₂-rich fluid phase leads to an increase of the amount of Cpx and a decrease of the abundance of Pl at 300 MPa. This decrease of Pl proportion is associated with a change in An content. Although the conditions are very different, as described above, this proves an effect of CO₂ on phase stability. Even if small, we also observed a change in An content from CO₂ free to CO₂ bearing (VL46 to VL47) as observed in Giuffrida et al. [2017] from high to low pressure experiments on Etna basalt. The glass compositions presented in Table 2 also highlight some differences between VL46 and VL47 samples. The SiO₂ and Al₂O₃ concentrations are lower in VL46 whereas FeO and CaO concentrations are higher in VL46. This leads to a small variation of NBO/T (0.33 to 0.28) but could mirror the small mineral chemistry variations that we observed. Finally, a liquidus temperature depression, caused by dissolved CO₂ in melt, is expected and this depression may range from negligible to several hundred degrees depending on silicate melts composition and pressure [Eggler 1975, 1976, Mysen and Boettcher 1975, Eggler and Kadik 1979, Eggler and Rosenhauer 1978, Boettcher et al. 1987, Boettcher 1984]. The extent of this depression is linked to the solubility and solubility mechanism(s) of CO₂ in the melt. For instance, the liquidus temperature depression of the CaMgSi₂O₆–CO₂ system compared to the volatile free system illustrates this mechanism [Eggler and Rosenhauer 1978, Mysen and Richet 2005].

Finally, it is emphasized that the difference in liquidus temperature for the CO₂ bearing and CO₂ free compositions (samples VL46 and VL47) observed experimentally at 500 MPa could not be reproduced with alphaMELTS, with T_L of 1223 °C and 1231 °C, respectively. Thus, the application of thermodynamic models to predict the role of high CO₂ concentrations on phase stability in water-poor systems needs to be reconsidered. This appears to be particularly relevant in Italians volcanic systems, which are often reacting with carbonates.

4.2. Implications for magma dynamics

The data presented above (literature data and our new experimental results) can have profound implications upon the rheological behavior of magmas. In particular, viscosity is highly influenced by the amount of solid phases. This, in turn, might influence the way magma ascends towards the Earth surface and eventually erupts.

The Vulcanello peninsula formed by effusions from a 600 m long, 0.5 to 1.5 m wide eruptive fissure (dike) [Vetere et al. 2007, and reference therein]. Based on the analysis of fluid inclusions in quartz xenoliths we assume for the storage of the shoshonitic magma a trapping depth between 21.5 and 17 km (corresponding to the Moho depth) [Zanon et al. 2003, Peccerillo et al. 2006]. As such, a possible deep storage reservoir could be hypothesised from which, during the eruptive phase of Vulcanello 1, the magma moved directly to the surface, without intermediate storage into the crust [Davì et al. 2009, Nicotra et al. 2018]. Taking into account the water free samples VL46 and VL47 (Figure 5), having a crystal content of about 20 and 30 area%, respectively, we can calculate the effective viscosity of the crystal bearing system using the models proposed by Sato [2005] and Vona et al. [2011] [𝜂_r =𝜂 _eff∕𝜂_m where 𝜂_r is the relative viscosity, 𝜂_eff is the effective viscosity of the liquid containing a fraction of crystals, and 𝜂_m is the viscosity of the melt using the model of Vetere et al. 2007]. Moreover, as required by the Vona et al. [2011] model, using image analysis technique on BSE images collected after experiments, one can extract and characterize the crystal shapes from the input grey-scale slides, without considering the possible presence of nanometric crystals [Cassetta et al. 2023]. Procedures are described in Dellino and La Volpe [1996] and Loncaric [1998]. The aspect ratio (AS = major axis/minor axis) parameter providing information on the particles elongation was determined following [Cox and Budhu 2008]. Results indicate that for our experiments the average AS for Pl is 2.5 and for Cpx is 1.7 while Fe–Ti oxides (when present) have AS close to 1. By taking into account these results, viscosity values of 10^4.0–10^4.4 Pa⋅s are obtained for system containing 20 and 30 area% of crystals, respectively by using the Vona et al. [2011] model. These values are comparable to those obtained using the model proposed by Sato [2005, i.e. 10^3.9 and 10^4.2 Pa⋅s].

Figure 7 shows the variation of viscosity of the shoshonitic magma as a function of the crystal and volatile contents. The increase in viscosity is clearly related to the decrease of the volatile content and the relative increase of solid phases.

Considering that magma ascent in dykes is strongly conditioned by dyke width, magma ascent in narrow dykes could be obstructed by magma crystallization. Critical dyke widths were calculated by Petford et al. [1994] for magma with felsic composition showing similar viscosities as those investigated here [see also Scaillet et al. 1996]. Results show that dykes thinner than 1 m can hardly propagate from very large depths. We are aware that narrow dykes  (<1 m) cannot be approximated to a smooth tabular shape with a constant dyke width, due to the natural roughness of rock fracture planes. However, by doing the identical exercise as in Petford et al. [1994], and taking into account (a) basaltic magma temperature, (b) freezing point, (c) rock country temperature (1200 °C, 700 °C and 300 °C respectively), (d) the latent heat for basalt (400 J/g) and (e) the specific heat of basaltic magma (1.0 J⋅g⁻¹⋅°C⁻¹), the critical size for dikes reduces to 0.3 m. Thus, we have some constrains for the following discussion.

In order to calculate a possible ascent velocity (u; m⋅s⁻¹) of these magmas we can use the relation of Lister and Kerr [1991] for laminar flow:

As stated above, the eruptive fissure at Vulcanello has width ⩽1.5 m. If now we consider a density contrast between magma and country rock of 𝛥𝜌 = 275 kg/m³ as reported in Vetere et al. [2007], and 𝜂 = 10^3.5 Pa⋅s for a system containing melt with 0.3 wt% H₂O (see also Table 3 samples VL46, VL47), we can estimate a possible rise velocity of such magmas, assuming that the width of the conduits varies from 0.1 to 1.5 m. Results are presented in Figure 8a where the ascent velocity is found to vary in the range u(CO₂ free) = 1.5 × 10⁻⁴–3.5 × 10⁻² m/s, and u(CO₂ bearing) = 5.7 × 10⁻⁴–1.3 × 10⁻¹m/s. The estimated velocity is subject to an abrupt increase if we consider H₂O–CO₂ bearing systems. Considering that at Vulcanello the water content is estimated between 0.3 and 1.9 wt% [analysis of melt inclusions from shoshonitic samples of Vulcanello in Gioncada et al. 1998 and Clocchiatti et al. 1994a, b] and considering a temperature of 1080 °C [the estimated temperature for Vulcanello magmas is 1083 ± 40 °C, Zanon et al. 2003], a crystallinity of 17 area% can be expected for H₂O- and CO₂-bearing samples (e.g. VL37; Table 3). At these conditions the magma viscosity is about 10^2.8 Pa⋅s and its ascent velocity (1.5 × 10⁻²–3.3 m/s as w varies from 0.1 to 1.5 m, respectively, Figure 7) would be faster than in a volatile-free magma (see u(CO₂ free) given above). Thus, in case of a nominally dry magmatic system rising in conduits having width 1.5 m, the time to reach the surface starting from a depth of 20 km is slightly lower than 4 years. However, if CO₂ is present, this duration decreases to ∼300 days, and <2 h are estimated assuming that 2.0 wt% H₂O + 0.6 wt% CO₂ are dissolved in the melt. Although these short timescales cannot be applied to the Vulcanello magma, which has lower volatile [Clocchiatti et al. 1994a, b], they could be relevant to other shoshonitic volcanic systems.

As a further example, by making identical considerations as above, we can consider a hypothetic scenario with a viscosity 𝜂 = 10^3.5 Pa⋅s (for a system containing melt + 0.3 wt% H₂O see Table 3, samples VL46, VL47). Thus, if dyke widths are ⩽3 m and the density contrast between magma and country rock is 𝛥𝜌 = 200 kg/m³ and adopting Equation (4), we can have an estimate of the possible rise velocity as presented in Figure 8b. In this case, results are: u(CO₂ free) = 1.1 × 10⁻⁴–1.0 × 10⁻¹ m/s, and u(CO₂ bearing) = 4.2 × 10⁻⁴–3.7 × 10⁻¹ m/s. Finally, if we consider H₂O–CO₂ bearing systems with concentrations similar to the experiments VL32 or VL33 (Table 3) and a temperature of 1080 °C, the lower crystallinity allows magma to have viscosity in the order of 10² Pa⋅s and, as consequence, to rise much faster in the range 4.7 × 10⁻²–42.9 m/s as w varies from 0.1 to 3 m, respectively (Figure 8). It is important to keep in mind that in dynamic magmatic systems (during ascent) the CO₂ release is relatively rapid and possibly can locally accelerate magma ascent.