Introduction

Sudden transitions on eruptive style and intensity are frequently observed during volcanic eruptions with an immediate impact on the surrounding territory and its population. A recent severe example is the eruption of the Mount Sinabung in Indonesia, started on September 2013 and culminated on February 2014 with the death of at least 14 people as a result of the rapid increase of explosive behavior, which occurred just after residents, that lived more than five kilometers from the mountain, had been allowed to return home following a temporary decline of volcanic activity1.

This behavior is also known to occur in mafic volcanoes2. The Stromboli volcano (Tyrrhenian Sea), known for its persistent moderate activity, has undergone in the last decades sudden vulcanian explosions, abruptly interrupting effusive eruptions3,4. Such events have been attributed to the rapid ascent of gas-rich batches of magma, similar in composition to the ongoing lava flows5,6,7.

The trigger for such sudden shifts in the eruptive regime has been generally attributed to various reasons (such as magma chamber and/or conduit processes; vent geometry; magma/water interaction etc.), however it remains poorly understood for many eruptions and hence, difficult to predict. Fundamental information on subsurface magma behavior and its influence on eruptive style can be obtained by the compositional and textural studies on volcanic rocks. These studies, applied to the rocks emitted during recent well-monitored eruptions, have revealed crucial relationships between the petrography of emitted rocks and the pre-/syn-eruptive geophysical signals8,9,10.

However, most studies have focused on intermediate to felsic magma eruptions, while mafic magmas remain less explored. The last mafic magma eruption of Vesuvius occurred on March 1944, consisting of an initial effusive phase (supplementary table 1) culminated in a paroxysmal explosive activity, provides an unusual and unique for Vesuvius, opportunity to investigate this key issue. The eruption in fact, lasted about 10 days, was well documented both for volcanological and seismological record, due to the detailed work of Giuseppe Imbò11 that was then director of the Vesuvius Observatory. Recent studies have shown that the intensity of this eruption and its effects on territory and population have long been underestimated, perhaps due to the Second World War going on12. After an initial effusive phase I (18–21 March), during which two towns in the west of the volcano were partly destroyed by the passage of lava flows, followed by a phase II characterized by the succession of eight lava fountains (21–22 March), the eruption culminated quickly in a violent explosive phase III (22 to 23 March) in which the eruption column exceeded 10 km altitude on the crater and ash carried by winds reached considerable distances from Vesuvius, up to Albania and Yugoslavia13. This paroxysmal phase lasted only 24 hours, since later the explosions became discontinuous and the ash cloud reached altitudes not exceeding 2 km from the crater (23–29 March - phase IV). The examination of new accounts of witnesses has shown that probably the number of victims (both humans and animals) associated to this paroxysmal phase may be greater than previously considered, due to the dispersion of the volcanic ash over wide areas14. Moreover, the mafic (phonotephrite-tephrite) character of the volcanic products makes more difficult the identification of the reasons that led to this sudden and drastic increase in the Volcanic Explosivity Index (VEI) that generally is instead associated with the emission of sialic magmas. We have conducted a detailed geochemical and textural study on the full succession of erupted products and compared the obtained results with the geophysical data recorded throughout the course of the eruption by Giuseppe Imbò.

The obtained results contribute to shed light on the evolution of conduit dynamics that triggered a sudden increase in the VEI, associated to the emission of a large amount of volcanic ash, that covered wide areas with a strong impact on territory and population.

Moreover our results suggest that the main phase of the volcanic paroxysms can be preceded by important changes in petrological indicators, correlated to geophysical signals that if monitored during volcanic crisis could provide fundamental clues about the ongoing eruption dynamics. This was also observed during recent well-monitored eruptions (e.g. the 2011–2012 El Hierro, Canary Islands, submarine eruption), for which the time correlation between magma evolution and the monitored geophysical signals has been evidenced and used to track the geological processes accompanying the magma movement towards the surface15.

Results

Petrological indicators

All the analyzed rocks are porphyritic characterized by euhedral phenocrysts (crystals > 0,1 mm) included in a groundmass consisting of microlite (< 100 micron) crystal population and tephritic to phono-tephritic glass (supplementary Table 2).

Phenocrysts comprise plagioclase, clinopyroxene, olivine and leucite as prevalent phases, while titanomagnetite and biotite are subordinate. Apatite is a common accessory crystal. Groundmass glass is composed prevalently of feldspar, clinopyroxene, leucite and magnetite microlites.

In general, none important difference has been observed in the composition of phenocrysts from rocks emitted during the different phases of the eruption (supplementary Table 3). This suggests that their crystallization occurred under similar chemical-physical conditions. Particularly, plagioclase crystals have composition ranging from bytownite (An85-Ab12) to labradorite (An62-Ab34), while pyroxene compositions fall in the diopside field and show a general trend toward Fe-rich composition (En39-36-Fs11-14). The olivine (Fo71-90 Fa10-30) and leucite crystals show moderately homogeneous composition (supplementary figure S1).

In spite of the uniform composition of phenocrysts and groundmass glass, the analyzed volcanic samples show different textural characteristics in microlites, depending upon which stage of the eruption they represent. Thus suggesting that the variations in the conditions of magma ascent and degassing during the growth of microlites in volcanic conduit could be the main cause of the observed changes in the eruptive style during the course of the eruption.

In particular, lava (effusive phase I) and subordinately scoria (lava-fountains phase II) samples are characterized by larger prismatic plagioclase (60–80 micron) associated to smaller blocky clinopyroxene (30 micron). While pumice and ash (sustained-column III phase and vulcanian - IV phase), contain sparse smaller acicular clinopyroxene and in minor amount plagioclase (15-10 micron) (figure 1, table 1). Moreover, number density (number of microlites per unit area) shows the lowest values in lavas (11673) and scorias (10895-25803) while they increase in pumices and ashes from the phase III (40000) and phase IV (35000) (table 1).

Table 1 Textural variations of microlites in samples from the different phases of the eruption
Full size table
Figure 1
figure 1

Back-scattered electron (BSE) images of groundmass textural features in samples from the different stages of the Vesuvius 1944 eruption.

Lava (phase I) and subordinately scoria (phase II) samples are highly microcrystalline with larger prismatic plagioclase associated to smaller blocky clinopyroxene. Pumice and ash (phases III and IV) contain sparse smaller acicular clinopyroxene and in minor amount plagioclase. Crystal size distribution plots display gentle slopes and low intercept values for larger-size plagioclase and higher y-axis intercept and steep slopes for smaller-size clinopyroxene.

Full size image

Following the pioneering work of Marsh16,17 and Cashman and Marsh18, the size of microlites and their abundance can be represented in a semi-logarithmic plot of crystal size distributions (CSD). Generally, CSDs show a linear trend on log–linear plots, in a steady-state crystallization regime16,17. Kinking or curvature in the plot profile has been attributed to a complex crystallization history19 (e.g. crystal accumulation and removal17; mixing of crystal populations19,20,21; compaction22,).

For all the studied samples, larger-size plagioclase and smaller-size clinopyroxene microlites display straight CSD trends (figure 1 and table 1). Particularly larger plagioclase, that characterize prevalently lava and subordinately scoria groundmass, show gentle slopes (−30) and low intercept values (from -11 to -12), while pumice and ash commonly include smaller clinopyroxene and subordinately plagioclase microlites, with steep slopes (-175 to - 608) and a higher y-axis intercept (from -7 to -9) (figure 1).

Seismological indicators

Vesuvius Observatory is among the first institutions performing monitoring of an active volcano using seismological instruments23. In March 1944 two seismometers were operating at Vesuvius Observatory (V.O.): Vicentini (three component) and Omori-Alfani-Malladra (hereinafter O.A.M.) vertical component seismometer. Another three-component Wiechert seismometer was deployed at the “Istituto di Fisica Terrestre” (I.F.T.) of the University of Naples. In the following we compare the recordings of the O.A.M. and of the Wiechert, following the work of Imbò24 (see Supplementary information Quadro IV and Quadro X).

Seismic signals, related to the 1944 eruption, are generally characterized by a background volcanic tremor, sometimes showing highly variable amplitudes. Different transients, emerging from the tremor, can be recognized24. They can be linked to both local earthquakes and volcanic explosions. The visual inspection of paper recordings of the 1944 eruption (supplementary figure S2), does not allow an unambiguous discrimination between these two typologies.

Imbò conducted a detailed analysis of these recordings, summarizing the observations in terms of various indices. He indicated aM as the maximum amplitude of the seismic trace within a given time window and aP as the “prevalent” amplitude of the background tremor within the same window. He performed this analysis over all the available paper recordings of the eruption. We consider here only the data from O.A.M. and the Wiechert seismometers, since they operated almost continuously during the considered interval, providing the most reliable data. The recordings of the Vicentini seismometer, as reported by Imbò24, were often saturated during intervals of more energetic eruptive activity, do not allowing a quantitative estimation of the amplitudes.

In figure 2 we represent both the prevalent and the maximum amplitudes for the O.A.M. seismometer recordings from seismometer from March 21st to 23rd 1944. During the phase I they show a fairly constant value. During the phase II they show marked increments in correspondence of lava fountains24. During the transition between phases II and III, the amplitudes remain on high values, until they drop at the end of the first interval of the phase III at 17:53. This reflect a waning of the eruptive activity which lasts until 20:53 when it increases again, marking the beginning of the second interval (figure 2). This pattern is consistent for both the aM and the aP indices.

Figure 2
figure 2

Amplitudes at the O.A.M. seismometer from March 21st to 23rd 1944.

aM means “maximum” amplitudes, while aP mean “prevalent”. See text for details. Data from Imbò24 The gray shaded area from 17:17 of 21st to 12:48 of 22nd indicates the duration of the eruptive phase II. The hatched pattern marks the interval of the last lava fountain of this phase. The crosshatched areas mark the first and the second intervals of the eruptive phase III.

Full size image

On the other hand, the amplitude ratio between the O.A.M. seismometer at V.O. and the Wiechert seismometer at Naples shows a different behavior. Values are fairly constant during the first and the second phase, showing a progressive drop at the beginning of the 8th lava fountain (figure 3). The decrease is more evident considering the ratio between the vertical component at O.A.M. (Zv) and that of the Wiechert (Zn), but it is also recognizable on the ratio between Zv and the horizontal component of the Wiechert (Hn) (figure 3).

Figure 3
figure 3

Amplitude ratios between the prevalent amplitudes of the O.A.M. seismometer at V.O. and the Wiechert seismometer at Naples.

Red triangles indicate the ratio between the vertical component of O.A.M. and the vertical component of the Wiechert seismometer. Conversely, blue triangles indicate the ratio between the vertical component of O.A.M. and the horizontal component of the Wiechert seismometer (see text for details).

Full size image

The amplitude at O.A.M. remains high until the end of the first interval (figure 2), hence the decrease of the ratios is related to an increase of the amplitude at the Wiechert. During the second phase the amplitude at O.A.M. is tightly correlated to the lava fountain activity24. This has been observed, in modern times, also on other volcanoes like Kilauea25 and Etna26. The seismic source relate to this kind of activity is generally very shallow (few hundred meters), making its amplitude to decay quickly at increasing distances from the eruptive vent. The V.O. is only 2 km apart from the Vesuvius crater, while the I.F.T. in Naples is about 14 km away from it.

Hence a likely explanation of the observed variation in the ratios is an increase in the activity of deeper seismic sources, which can be related to different phenomena occurring at greater depth. This hypothesis is supported by the observed variations in the amplitude ratios among different components at the Wiechert seismometer (figure 4). In particular we observe that, while the E/N ratio oscillates around 1 (with the higher variability during the phase II), the H/Z ratio shows a marked decrease since the beginning of the 8th lava fountain, following a trend very similar to the ratios shown in figure 3.

Figure 4
figure 4

Amplitude ratios between the horizontal and vertical component of the Wiechert seismometer.

Green triangles indicate the ratio between the horizontal and the vertical component, while yellow triangles indicate the ratio between the E-W and the N-S components.

Full size image

In conclusion, since the beginning of the 8th lava fountain, there is an increase in the activity of seismic sources, deeper than those related to the lava fountain activity and possibly related to a different physical mechanism (e.g. earthquakes). These sources become progressively dominant during the first interval of the phase III.

Discussion

The homogeneous composition of phenocrysts and glass in all the studied rocks indicates that the magmas erupted in the different stages of the eruption, have crystallized at similar chemico-physical conditions. To determine temperature, pressure and water content during phenocrysts crystallization in the reservoir, we have used thermodynamic calculations based on the minimization of the free energy of Gibbs (MELTS Code27,28 and PELE code29). In Melts calculations the following parameter ranges were assumed: pressure from 1000 to 4000 bar, QFM (quartz, fayalite, magnetite) oxygen fugacity ± 3 log units and initial H2O content from 1 to 6 wt %.

Among the calculated liquid lines of descent and mineral crystallization sequences produced by MELTS, those under conditions of water saturation, oxygen fugacity buffered along QFM-3 and pressures of ≥ 250–400 Mpa are in general agreement with the phenocrysts assemblage and modal abundance observed in the studied natural rocks (that was olivine ≤2% followed by clinopyroxene ≤20%, apatite ≤1% and magnetite ≤5% and later by plagioclase ≤20% and biotite≤5%30,). Indeed, under lower pressure conditions (<250 Mpa), clinopyroxene always starts to crystallize earlier than olivine, that is present with higher modal abundance for the whole crystallization sequence in contrast to what observed in real rocks, (figure 5). The obtained results indicate that the crystallization occurred from a parental tephritic magma at near water-saturation condition, during cooling (T = 1150°C – 950°C) in a crustal reservoir located at depth between 16 and 10 km (P between 400 and 250 Mpa); the first precipitating phases are olivine and pyroxene, followed by magnetite, biotite and apatite and plagioclase. These results are in agreement with the pressure values obtained by using the volatiles (CO2 and water) content measured on melt inclusions31,32,33 trapped in olivine and diopside (figure 6), compared with the available solubility models. Moreover, in the explored pressure and temperature ranges, leucite never has appeared on the liquidus, this is also consistent with the volatile contents measured on melt inclusions in leucite that record CO2 below detection limit and low H2O content. This suggests the formation of this phase during a later crystallization event, at very low pressure (figure 6), probably during the ascent of the magma in volcanic conduit, also in agreement with experimental results indicating that the crystallization of leucite in Vesuvius magmas occurred at pressure <100 MPa34.

Figure 5
figure 5

Mineralogical assemblage during magma cooling in the reservoir as calculated by Melts and Pele program at different P conditions.

Full size image
Figure 6
figure 6

Behavior of volatile in groundmass glasses (this work, colored symbols) and melt inclusions30,31,32 (data from literature, grey symbols).

a) Isobars (100 to 400 MPa) for H2O and CO2 calculated on the basis of VolatileCalc Newman and Lowenstern50. b) Cl content versus CaO (after Pappalardo and Mastrolorenzo51) and c) Cl content versus H2O. Solubility lines calculated from Signorelli & Carroll34. During initial slow ascent, the composition of magma emitted during the effusive phase I follows solubility model both for Cl and water, thus indicating equilibrium degassing during slow ascent in the conduit. On the contrary, the composition of glass samples of explosive phases departs from equilibrium concentrations due to fast ascent preventing fluid/liquid exchange (disequilibrium degassing).

Full size image

Constraints on the magma ascent rate can be obtained by using the volatile content measured in groundmass glass (figure 6), whose composition corresponds to that of the residual liquid during degassing in the volcanic conduit. In particular, the low water content (H2O wt% ~ 0.5), that characterizes the groundmass glass of lavas emitted at the beginning of the eruption, are compatible with a slow magma decompression, that favors the volatile exolution, bubbles coalescence (that controls magma permeability) and then removal of gas (including H2O) from the liquid. On the contrary, the highest content of water (H2O 1.5–3 wt%), in groundmass glass of scoria and ash of the explosive phases of the eruption, is indicative of a more rapid ascent, that prevents the removal of the gas phase, which then remains in contact with the liquid until the explosive fragmentation. Similar assumptions are suggested by Cl content behavior. Signorelli and Carroll35 have shown that Cl content increases within the liquids saturated with both a H2O vapor phase and a Cl-rich brine, as pressure decreases. The lavas of the phase I follow this Cl solubility model, corroborating the hypothesis of an initial slow ascending magma (figure 6b), while scoria and ash of the following phases show a constant Cl concentration, indicating a fast ascent which prevents fluid/melt Cl exchange.

This scenario is verified by the groundmass microlite textural variations detected in 1944 volcanic rocks, showing an evolution with time toward low values of crystallinity and mean microlite size, as well as a growth in crystal number density, associated to a progression in microlite shape that shifts from prismatic toward more skeletal (acicular, shallow-tail). Actually, decompression experiments36,37,38,39,40,41 have revealed that groundmass microlite texture is strictly related to mechanisms and time of magma decompression in volcanic conduit. Particularly, under conditions of rapid ascent, due to high rate of undercooling (defined as the difference between the liquidus temperature and that of the magma) the crystallization is controlled by nucleation of new sites (nucleation-dominated regime); thus resulting in many tiny skeletal (e.g. hopper, shallow-tail, acicular shape) microlites. On the contrary in low ascending magma, growth of existing crystals prevails (growth-dominated regime) over nucleation due to the reduction over time of the degree of undercooling; thus resulting in fewer but larger (e.g. tabular, prismatic) microlites. Moreover, recent experiments42 have determined that during higher-pressure decompression, crystallization occurs under growth-dominated regime; whereas at lower-pressure decompression crystal nucleation prevails. Therefore, we suggest that the largest prismatic plagioclase, that mainly characterize lavas emitted during the first phase of the eruption, have crystallized slowly under higher-pressure decompression conditions and can be associated to the magmatic intrusion during the widening of volcanic conduit; while the smaller acicular clinopyroxene observed prevalently in scoria and pumice from the subsequent explosive phases of the eruption are indicative of a fast decompression in the shallower part of the unlocked conduit.

Extimation of microlites crystallization time during decompression in the conduit can be obtained from the slope of the CSDs trends (if the crystal growth rate is known) using the following equation16: slope = -1/(G*T), where G is the growth rate and T is the crystallization time. Applying this equation to the CSDs trends of the different sized microlites observed in 1944 volcanic rocks, we have estimated the magma ascent rate at various depth in the conduit during eruption progression. Particularly, the largest plagioclases (characterized by gentle slope in CSDs) represent the earlier magma intrusion in the deeper conduit, while the smaller clinopyroxenes (steep slope in CSDs) reflect a later stage of ascent at shallower depth. Growth rate of groundmass plagioclase and clinopyroxene due to water exolution during magma ascent, are generally estimated to be between 10−7 to 10−8 mm/s42,43,44,45 for mafic composition. By using the above range of crystal growth, we have calculated a minimum and maximum crystallization time for largest plagioclase (gentle CSD slopes) included in lavas, ranging from 38 to 4 days. Thus indicating the possible development of an ephemeral (in the order of months to days) shallow storage zone beneath the volcano, in which the earliest erupted magmas would have been arrested during the enlargement of the conduit. A volume of magma of 0.02 km3 was erupted during the phase I, eventually stored for a short time in this shallow region. Imbò24 reports a variation in the seismicity pattern since the beginning of March, about 20 days before the onset of the eruption, that corresponds to the intrusion of magma at shallow depth inferred on the basis of our petrological data. Moreover, microlite crystallization time calculated for the smaller clinopyroxene (steep CSD slopes) observed in the rocks emitted in the latter stages of the eruption, indicates that decompression became gradually faster, ranging from tens of hours at the beginning of the lava-fountains (phase II) up to few hours during the most violent paroxysm (phase III). This scenario is in agreement with seismological data showing an increase in the amplitude of tremor signals in correspondence of lava fountains, related to an increased gas ejection rate. This relationship between volcanic tremor amplitude and lava fountains has been observed on various volcanoes, as Kilauea25 and Etna26. Remarkably, changes in both petrological and seismic indicators have preceded the most violent paroxysm of the phase III, with the emission of low crystalline rocks characterized by high number density (32000–40000) tiny (15-10 μm) microlites, associated to variations in the amplitude ratios of seismic signals reflecting a variations in depth and nature of the seismicity.

We consider that these changes are due to the partial collapse of the deeper parts of the plumbing system (figure 7) as testified also by the high abundance of lithic fragments (including also skarn, clinopyroxenites, dunites32,33) observed in the corresponding volcanic deposits. This process caused a pressure drop at the top of the underlying chamber and then the fast extrusion of a larger volume of magma (0.2 km3) in a short time, with a mass eruption rate that shifted from 2.8 × 106 up to 9.7 × 107 kg/s.

Figure 7
figure 7

Schematic diagram for conduit processes and eruption phenomena during the recognized four phases of the Vesuvius 1944 eruption, deduced from the evolution of petrological and seismic indicators.

On the top we show examples of textural features representative of products of each eruptive phase. Groundmass texture shows high crystallinity associated to larger prismatic microlites during the first phases of the eruption, while shifts toward low values of crystallinity and size of microlites that became progressively also more skeletal. Phase I (18-21 March) initiated with magma intrusion and stagnation at shallow level during conduit enlargement, the effusive activity lasted about 4 days with a mass discharge rate of 1.4 × 105 kg/s. During phase II (21-22 March) magma experienced an increment in ascent rate due to sudden conduit-vent opening, causing the shift from the effusive to the lava fountains activity, with a mass discharge rate of 2.8 × 106 kg/s. During the phase III (22–23 March) the partial collapse of the deeper parts of the plumbing system caused a pressure drop at the top of the underlying chamber and then the fast extrusion of a larger volume of magma (0.2 km3) in a short time, with a mass eruption rate up to 9.7 × 107 kg/s. The end of the eruption with the passage at the vulcanian phase IV (24–29 March) is due to both the decrement in magma supply (2.7 × 104 kg/s), as well as to the progressive obstruction of the conduit by wall collapse.

Full size image

The subsequent decrement in magma supply (2.7 × 104 kg/s), as well as the progressive obstruction of the conduit by wall collapse, defines the passage at the final vulcanian phase IV. During the end of the eruption it was observed the emission of large amount of free crystals of leucite and pyroxene that could represent the crystal-mush formed along the wall of the chamber/conduit system and transported on surface by the last remaining liquid.

In conclusion our merged petrological and seismic evidences imply that a complex conduit dynamic, rather than different pre-eruptive condition, was the main cause of the evolution of 1944 eruption of Vesuvius towards violent explosive behavior. Particularly, we documented that the increment of the explosive character of the eruption that culminated in the paroxysmal phase III, was due to the progressive collapse of the deeper part of the conduit that was revealed, since its onset (at the end of the lava-fountains phase II), by robust changes of the petrological and seismic indicators. The reduction of crystallinity and sizes of the microlite and the steepening of the CSD slope during the last part of the phase II, indicate clearly an increase in the ascent rate that cannot be univocally explained by using petrological data alone. However, the contemporaneous change in the seismic amplitude ratios, that reflects the increased contribution of a deeper seismic source, suggests that the incipient collapse of the deeper part of the conduit was the cause of the observed variations in both petrological and seismic indicators. Hence, the joint interpretation of these two independent datasets, points out that an early detection of ongoing modifications within the plumbing system, leading to a sub-plinian phase, can be successfully forecasted at least few hours before the onset of the main explosive phase.

For this reason, we suggest that the development of monitoring methods to evaluate in near-real time petrological features of ejected juvenile material, by using new fast analytical methods46, during volcanic crisis and their interpretation alongside conventional geophysical monitoring techniques could allow the prompt reconstruction of the ongoing dynamic of the plumbing system and then could be useful for forecasting the eruption evolution (such as the onset of more violent phases).

This is even more important as our groundmass textural data indicate that during the paroxysm, mafic magmas were transferred directly from the deep-seated magma reservoir to the surface at high decompression rate, while residence in upper chambers was limited at the earlier magmatic intrusion. This scenario excludes a long-lasting shallow storage of magma, as suggested elsewhere for stratovolcanos47, that could act as brake to retard the rapid extrusion of new ascending deeper batches, thus strengthening the chance of hazardous eruptions in the future.

Methods

The volcanic samples were examined under polarized light microscope, scanning electron microscope (JEOL JSM-6500F) and electron microprobe (JEOL-JXA-8200) for textural analysis (microlite shape, number density and size distributions) and the measurements of major and volatile elements (Cl, F, S) concentration in the groundmass glass and in the mineralogical phases. The measurements were carried out at the laboratories of the Istituto Nazionale di Geofisica e Vulcanologia, (Roma).

Textural analysis

Textural analyses of microlite shape, number density and size distribution were carried out on polished thin sections of epoxy-impregnated grains using a standard polarized light microscope for preliminary rock characterization followed by a Field Emission Scanning Electron Microscope (FE-SEM) JEOL JSM-6500F (Istituto Nazionale di Geofisica e Vulcanologia, Roma, Italy) for quantitative measurements. Backscattered electron (BSE) images were processed using Adobe Photoshop and ImageJ software to measure the number, size and area of microlites. Stereological methods have been used to convert the parameters obtained by studying two-dimensional sections into three- dimensional textural values. Specifically, crystal number densities and size distributions have been obtained with the CSD Corrections 1.3 program48 that includes corrections for both intersection probability and cut section effects. An estimation of the rock fabric and crystal aspect ratio are necessary for generate accurate CSD data. In the studied samples the rock fabric was massive and the crystal aspect ratio has been calculated with CSDslice program49. Results are reported in Table 1.

Geochemical analysis

Analyses of major and volatile elements in groundmass glasses and mineral phases were performed on polished thin sections of epoxy-impregnated grains with a JEOL-JXA-8200 electron microprobe (WD/ED combined micro analyzer). Element concentrations were measured at15 keV, a beam current of 4 nA and a counting time of 10 s on peak and 5 sec on both background. For each analysis, a defocused beam was used to minimize losses of alkalis and volatiles, which were counted first to avoid diffusion effects. We attempted to obtain at least ten analyses for each sample. Data reduction was carried out using the ZAF software. The analytical uncertainty was about 1% for most elements.

The H2O content of all analyzed glass was estimated by using the “volatile by difference” method based on EMPA analyses. Results are reported in (supplementary Tables S2 and S3).