Introduction

In the twentieth century, at least a dozen of worldwide strato-volcanoes generated high-magnitude eruptions that killed thousands of people and caused extensive damage with severe social and economic impacts. Some of these events interrupted long lasting periods of quiescence (e.g., Pinatubo, Philippines, 1991), while others followed a short volcanic rest (e.g., Colima, Mexico, 1913) or occurred with only little warning (e.g., Chaiten, Chile, 2008). Hence, understanding the trigger mechanisms as well as the syn-eruptive dynamics of these eruptions is crucial for their forecasting. Combined quantitative textural and geochemical studies on volcanic rocks have proven to be a valuable approach in exploring the conditions related to magma ascent in volcanic conduits (e.g., Klug et al. 2002; Adams et al. 2006; Mastrolorenzo and Pappalardo 2006; Gurioli et al. 2008; Polacci et al. 2009, 2014; Shea et al. 2010a, b; Rust and Cashman 2011; Gonnermann and Houghton 2012; Pappalardo et al. 2014; Rotella et al. 2014), improving our ability to interpret volcano-monitoring signals and perform hazard evaluations. Particularly, 3D textural data have played a key role in assessing nucleation, growth, and coalescence of gas bubbles and magma fragmentation that in turn influence the style and intensity of explosive eruptions. In fact, 3D data allow the direct observation and quantification of the dimension, shape, and orientation of the vesicles, as well as of their degree of interconnectedness and related permeability, difficult to determine using the more conventional 2D techniques (e.g., Song et al. 2001; Okumura et al. 2008, 2012; Polacci et al. 2009, 2012, 2014; Degruyter et al. 2010; Giachetti et al. 2011; Baker et al. 2012). X-ray computed microtomography (micro-CT) is a powerful, non-destructive method to carry out 3D textural studies of igneous rocks (e.g., Jerram and Higgins 2007; Baker et al. 2012; Cnudde and Boone 2013). Nevertheless, only a few studies of volcanic pyroclasts using this tool are available especially on alkaline rocks (e.g., Polacci et al. 2009; Hughes et al. 2017).

The Somma–Vesuvius alkaline volcanic complex, located east of the metropolitan area of Naples, is one of the most dangerous volcanoes in the world (e.g., Mastrolorenzo et al. 2008; Lirer et al. 2010; Mastrolorenzo and Pappalardo 2010 and the references therein), Fig. 1. The volcanic activity has been characterized by at least four Plinian eruptions, interposed to minor events covering a large range of magnitude and intensity (Cioni et al. 2008; Santacroce et al. 2008; De Vivo et al. 2010), the latest occurred on March 1944 (e.g., Cole and Scarpati 2010; Pappalardo et al. 2014; Cubellis et al. 2016). The Pomici di Base eruption (Andronico et al. 1995; Delibrias et al. 1979; Bertagnini et al. 1998; Landi et al. 1999; Siani et al. 2004; Santacroce et al. 2008; Klebesz et al. 2012, 2015; Scarpati et al. 2016), occurred about 22 ka, is generally acknowledged as the first and most intense Vesuvian Plinian event. It marks the transition, after a period of prevalent effusive activity, to the explosive behaviour of the volcano as well as the beginning of the caldera collapse events. The eruption was characterized by the emission, during a sustained-column Plinian phase, of at least 4.4 km3 (bulk volume) pyroclastic products dispersed towards the E–NE, followed by a phreatomagmatic phase during which fallout activity has alternated with minor pyroclastic density currents confined to the volcano slopes. The Plinian fallout deposit is characterized by a strong compositional variation from white trachytic pumices (0.34 km3 DRE; Landi et al. 1999) to black latitic–shoshonitic scoriae (0.96 km3 DRE); during this phase, the mass discharge rate remained stable in the range of 2–2.5 × 107 kg/s corresponding to a column height of 16–17 km (Bertagnini et al. 1998), despite the change towards mafic composition.

Fig. 1
figure 1

On the left: location map for Somma–Vesuvius volcano (Italy), situated within Campanian Potassic Volcanic Province (gray field). On the right: distribution of Pomici di Base fallout unit (Bertagnini et al. 1998)

Full size image

In this case study, the micro-textural features of clasts from the whole Pomici di Base Plinian fallout succession have been quantitatively investigated by generating high-resolution three-dimensional digital maps via X-ray computed microtomography. The obtained textural information (i.e., density measurements, bubble and throat number density and size distributions, bubble interconnectivity, etc.), combined with geochemical and Sr- and Nd-isotopic data on both separated groundmass and phenocrysts, has contributed to reconstruct the trigger mechanisms and conduit dynamics that controlled this Plinian eruption fed by a less evolved magma.

Sampling and methods

Representative juvenile pumice and scoria clasts of the entire Plinian fallout succession of the Pomici di Base eruption were collected from two key stratigraphic sections located at about 15 km NE of the Somma–Vesuvius (Palma Campania, near Naples), Fig. 1. The Plinian deposit is composed of a basal white pumiceous layer, overlain by a transitional gray pumice layer and an upper black scoria bed (Fig. 2). The three beds have an approximately constant relative thickness of 2:1:5 (Bertagnini et al. 1998). The sampling interval was dictated by changes in type, grain size, and color of the pyroclastic products, according to the different stratigraphic units recognized by Bertagnini et al. (1998), Fig. 2. Samples location within eruptive units is reported in Fig. 2.

Fig. 2
figure 2

Representative photos and schematic stratigraphic column for the Pomici di Base Plinian deposits and sample number for clasts collected for this study. Samples were collected from two stratigraphic sections (a and b–c, respectively, on the right). Sampling interval was dictated by changes in type, grain size, and color of the pyroclastic products, according to the different stratigraphic units recognized by Bertagnini et al. 1998 (on the left): U-1—ash and pumice fall deposits; U-2—Plinian fallout deposit composed by a basal white pumiceous layer (U-2a and U-2b), a transitional layer (U-2c), an upper thick black scoria bed (U-2d and U-2e); U-3—U-6 lithic-rich fallout and PDCs (pyroclastic surge and flow) deposits

Full size image

Bulk density

To account for possible density variations with size, we used juvenile clasts within a − 5 to − 2 phi size range for density measurements. Sets of 100 clasts for each granulometric class (where present) were weighted and coated with a thin film of paraffin wax, and their density was determined using a pycnometer. We calculated bulk vesicularity for each clast following the procedure described in Houghton and Wilson (1989), using a dense-rock equivalent (DRE) density of 2700 (latite–shoshonite) and 2400 (trachyte) kg/m3, obtained by pycnometry measurements on fine-grained whole-rock powder as described in ASTM (2007).

No differences in clast-density distributions have been observed between the analyzed granulometric classes. Moreover, juvenile clasts with different sizes and same density, accurately observed under optical microscope, show comparable textural features. However, to avoid any possible influence of post-fragmentation expansion, we have selected clasts within 4 and 8 mm size range, as larger clasts do not necessarily preserve the vesicularity of the magma immediately prior to fragmentation (e.g., Thomas et al. 1994; Gardner et al. 1996; Kaminski and Jaupart 1997).

Modal density clasts were selected from the base to the top of the deposits to investigate micro-textural variations with stratigraphic height (see also Gurioli et al. 2005; Balcone-Boissard et al. 2008, 2012; Houghton et al. 2010).

Microtomographic investigation

The microstructure of the samples was investigated by X-ray computed microtomography (micro-CT) using a Carl Zeiss Xradia Versa-410 3D X-ray microscope at the Istituto Nazionale di Geofisica e Vulcanologia—Sezione di Napoli “Osservatorio Vesuviano” (INGV-OV, Naples).

X-ray computed microtomography is a non-destructive analysis technique that offers the opportunity to visualize and quantify the internal structure of rock samples by generating three-dimensional digital maps with a very high resolution (down to submicron). Particularly, the result of microtomographic investigation is a three-dimensional gray-level image proportional to the X-ray attenuation coefficient of the sample (that for the same energy is a function of the density and atomic number of the interested material), which allows the observation and measurement of the properties of objects (e.g., shape, size, distribution, and orientation of fractures, pores, crystals, etc.) entirely avoiding the stereological corrections needed for measurements carried out in two dimensions.

Particularly, Xradia Versa architecture uses a two-stage magnification technique. First, a geometric magnification, as with conventional micro-CT, is obtained. In the second stage, a scintillator converts X-rays to visible light, which is then optically magnified. Reducing dependence upon geometric magnification enables Xradia Versa instruments to maintain submicron resolution at large working distances.

In this study, cylinders of maximum diameter 0.5 cm were cut from the samples and the scan was performed over a 360° rotation using 4001 projections, 80 KV voltage, and 10 W power. The resulting nominal voxel (volumetric pixel) size ranges from 0.9 to 2 µm depending on the optical magnification used (10X and 20X, see Table 1). However, imaging using any method (optical, SEM, XMT, etc.), where the smallest feature (vesicle or vesicle wall) is less than three pixels/voxels in diameter, is subject to significant uncertainty (Hughes et al. 2017). Therefore, we consider that the minimum measurable voxel size ranges from ≈ 3 to 6 µm (corresponding to 3 voxel size) as also confirmed by accurate visual inspection of slices.

Table 1 Textural parameters for Pomici di Base rocks
Full size table

Reconstruction of the attenuation data was performed through the filtered back-projection algorithm using XRMReconstructor Xradia proprietary software producing a stack of 967 cross-sectional, gray-scale digital images, Fig. 3.

Fig. 3
figure 3

Examples of volume rendering showing 3D microstructure of Pomici di Base trachytic (a–d) and latitic–shoshonitic (e–f) rocks. Cylinders diameter: 1000 µm. Vesicles are black, melt/feldspars/pyroxenes are gray, and oxides are white

Full size image

3D and 2D textural measurements

Vesicle and throat size distributions and number densities were obtained by processing the 3D tomographic images. First, the obtained 3D micro-CT images were filtered using grayscale-to-grayscale filters available in Blob3D software (Ketcham 2005) to improve the brightness and contrast and to obtain an edge enhancement of the vesicles. Then, the 3D images were segmented to isolate vesicles from matrix glass and crystals on the basis of their gray-level values (thresholding), which are related to the X-ray attenuation coefficient. In particular, the automatic Otsu algorithm (Otsu 1979) was adopted, manually adjusting the threshold when necessary using ImageJ (Schneider et al. 2012) software (e.g., Okumura et al. 2008; Caricchi et al. 2011; Zandomeneghi et al. 2010; Voltolini et al. 2011; Berg et al. 2016). A further step involved the extraction of a volume of interest (VOI) with dimensions suitable for the available computing resources but preserving sample representativeness to separate the connected pores and restore the pre-fragmentation conditions. The VOI has to be larger than the Representative Elementary Volume (REV; Bear 1972), which corresponds to the smallest portion of the sample that statistically represents all features of the entire sample (e.g., Degruyter et al. 2010; Zandomeneghi et al. 2010; Berg et al. 2016; Kennedy et al. 2016). The determination of REV is estimated by an iterative process based on the calculation of porosity on increasing 3D image volumes until reaching a plateau value. Potential cutting effects from sample preparation and artifacts of the cone–beam reconstruction were avoided by selecting VOIs in the central parts of the imaged volumes.

Connected vesicles have been separated reconstructing the thin glass films lost during image acquisition and processing or partially retracted in a late-stage of coalescence (Shea et al. 2010a; Giachetti et al. 2011; Rust and Cashman 2011; Hughes et al. 2017). Separations of connected vesicles were obtained using the Separate Objects tool available in Avizo FEI software based on a combination of watershed, distance map, and H-maxima. To quantitatively and graphically characterize pore interconnection, we adopted Pore Network Statistics Extension in Avizo FEI software. The obtained Pore Network Model (PNM, Fig. 4d) is composed of branching or endpoints of the network called pores (or vesicles for volcanic rock) and lines connecting pores called throats. The results of the different analyzed components (vesicles and throats) and their distributions are reported in Table 1. Volume rendering was obtained using Dragonfly ORS software.

Fig. 4
figure 4

Cumulative Vesicles size distributions (CVSDs) for trachytic (a) and latitic–shoshonitic (b) Pomici di Base rocks and Cumulative Throats size distributions (c). CVSDs of trachytic pumices show a curved continuous trend characterized by exponential distribution for the small bubbles and power-law distribution for the large bubbles, indicating a continuous nucleation process (see text). The presence of pre-existing bubbles in the upper gas-rich level of the chamber could justify the large bubble population observed in G80 sample. On the contrary, CVSDs of scoria latitic–shoshonitic samples show irregular trends formed by multiple curved segments suggesting discrete vesiculation events, attributed to nucleation pulses driven by fast CO2 release during the ongoing decarbonation process (see text for further explanation). Example of Image of Pore Network Model (sample G10) obtained by Avizo FEI software (d). A PNM is composed of labeled branching or endpoints of the network called pores (or vesicles for volcanic rocks) and lines connecting pores called throats. Pores are displayed using spheres, and throats are displayed using cylinders. Cube side: 300 µm (pixel size = 1 µm)

Full size image

Moreover, permeability measurement was carried out using Avizo FEI software (Petrophysics Extension), which contains the Absolute Permeability Experiment Simulation tool. Permeability has been calculated along the three orthogonal directions assuming a gas viscosity of 10−5 Pa s (Rust and Cashman 2011; Okumura et al. 2012).

Small density contrast between microlites (clinopyroxene and feldspar) and matrix glass has prevented a 3D quantitative investigation of microlites, for which conventional 2D analytical methods have been adopted.

Microlite crystallinity and size distributions were obtained by 2D Back-Scattered Electron images. For each sample, at least 4–5 BSE images (270 × 200 µm) were acquired by a Field Emission Scanning Electron Microscope (FE-SEM) JEOL JSM-6500F (Istituto Nazionale di Geofisica e Vulcanologia, Rome, Italy) and a Scanning Electron Microscope (SEM) Jeol JSM 5310 (DiSTAR, Università di Napoli Federico II) equipped with energy dispersive system and operating at 15 kV. BSE images were processed and analyzed using ImageJ software. They were first reduced to binary images and then manipulated to reduce noise and separate individual microlites. Microlite size distributions were obtained using CSD Corrections 1.5 program (Higgins 2000, 2002, 2006; Higgins and Chandrasekharam 2007) that includes corrections for both the intersection probability and the cut section effects.

Electron microprobe analyses (EMPA)

Analyses of major and volatile elements in groundmass glasses were performed with a JEOL-JXA-8200 electron microprobe (WD/ED combined microanalyzer) at the laboratories of the Istituto Nazionale di Geofisica e Vulcanologia (Rome). Element concentrations were measured with a 10 μm beam at 15 keV, a beam current of 10 Na, and a counting time of 10 s. For each analysis, a defocused beam was used to minimize losses of alkalis and volatiles, which were counted first to avoid diffusion effects. The following standards have been adopted for the various chemical elements: jadeite (Si and Na), corundum (Al), forsterite (Mg), andradite (Fe), rutile (Ti), orthoclase (K), barite (Ba), celestine (S), fluorite (F), apatite (P and Cl), and spessartine (Mn). Data reduction was carried out using ZAF4/FLS software by Link Analytical. The analytical uncertainty was about 1 wt% for most elements. In a first attempt, the water content of all analyzed glass was estimated using the “volatile by difference” method based on EMPA analyses (Devine et al. 1995; King et al. 2002). EMPA data are reported in Table 2.

Table 2 EMPA analyses of selected Pomici di Base volcanic rocks
Full size table

H2O and CO2 determination

Carefully selected natural pumice and scoria fragments were stored overnight in an H2O2 (20% in aqueous solution) bath to remove organic materials. Successively, the samples were left to dry in the air for 24 h and lately stored overnight in a drying oven at 110 °C to release water possibly adsorbed on the glass surface.

After this treatment, the H2O–CO2 contents in the scorias and in the pumices were analyzed with different methods at the Institute of Mineralogy at the University of Göttingen.

Water contents on bulk samples were measured by thermogravimetric analyses (TGA) using a SetaramTM TGA92. During a typical analyses, 10–20 mg of coarsely powdered material is loaded into a Pt crucible (4 mm diameter and 10 mm height) and covered with a Pt lid. The sample is heated from ambient temperature to 1200 °C with a ramp rate of 10°C/min. After 30 min dwell time, the sample is cooled to room temperature again with a cooling rate of 30 °C/min. In the case of iron poor samples, the measurements are performed in the air, while iron rich samples (FeO > 5 wt%) are measured in helium to avoid the oxidation of Fe2+ to Fe3+ by reaction with the dissolved water. During the analyses, the weight loss of the material is constantly recorded. Once a day, a final heating and cooling cycle is performed after a simple run to account for buoyancy changes dependent on the temperature of the sample and, therefore, to correct the measured sample weight loss (Schmidt and Behrens 2008).

For each sample, three-to-six thermogravimetric analyses were performed.

The determination of the CO2 content on bulk samples was performed with an ElementarTM Inductar CS Cube Carbon–Sulfur Analyser (CSA). About 100–150 mg of coarsely crushed sample mixed with 0.5 g of Fe and 2 g of W chips are filled in a corundum crucible. After inductive firing, the mixture is burned in an oxygen stream releasing the CO2 which is then measured with an IR cell. According to the manufacturer, a temperature of approx. 2000 °C is reached within 1 min.

Several steel samples with known carbon contents are used as standard calibration (Behrens et al. 2009).

Radiogenic isotopes

Isotopic analyses for Sr and Nd via thermal ionization mass spectrometry (TIMS) were carried out at the Istituto Nazionale di Geofisica e Vulcanologia—Sezione di Napoli “Osservatorio Vesuviano” (INGV-OV, Naples), using a ThermoFinnigan Triton TI multi-collector mass spectrometer. Samples were processed through conventional HF–HNO3–HCl dissolution before Sr and middle REE (MREE) were separated by standard cation exchange column chemistry, and Nd was further purified on an anion column. Sr and Nd were then loaded onto Ta and Re filaments, respectively. Sr and Nd blanks were negligible for the analyzed samples during the periods of measurements. Measured 87Sr/86Sr ratios were normalized for within-run isotopic fractionation to 87Sr/86Sr = 0.1194, and 143Nd/144Nd ratios to 143Nd/144Nd = 0.7219. The mean measured value of 87Sr/86Sr for NIST–SRM 987 was 0.710215 ± 0.000008 (2 sigma, n = 36) and of 143Nd/144Nd for LaJolla was 0.511843 ± 0.000006 (2 sigma, n = 11). The quoted error is the standard deviation of the mean (2 sigma) for n = 180. Sr and Nd isotope ratios have been normalized to the recommended values of NIST SRM 987 (87Sr/86Sr = 0.71025) and La Jolla (143Nd/144Nd = 0.51185) standards, respectively. Results are reported in Table 3.

Table 3 Isotopic ratios of separated glass and minerals of Pomici di Base rocks
Full size table

Results

Clast density, vesicularity, and crystallinity

All samples show less than 5 wt% content of phenocrysts that are present both as isolated crystals and in aggregate and are made up of sanidine > plagioclase > clinopyroxene > biotite in order of decreasing abundance, together with minor amount of amphibole, magnetite, and garnet.

Clast densities as well as the degree of vesicularity and microlite crystallinity vary progressively upward in the stratigraphic sequence, Table 1, similar to what previously observed by Bertagnini et al. (1998). In particular, the basal and intermediate (white to gray) trachytic pumices show low-density modal values (0.57–0.61 g/cm3), high vesicularity (69–87%), and absence of microlites, while the upper (black) latitic–shoshonitic scoriae are characterized by higher density values (1.1–1.5 g/cm3) and poorly vesiculated (45–59%) microlite-rich groundmass. Bulk vesicularities calculated by pycnometry are generally consistent with those obtained by the 3D images at lower 10× magnification (see Table 1), while the values slightly increase in high-resolution three-dimensional images; however, the dissimilarity is always less than 15% as also observed by Giachetti et al. (2011) using a similar procedure for Vulcanian pyroclasts erupted from Montserrat.

Two vesicle populations characterize highly vesiculated trachytic pumices: small (diameter < 25 µm) spherical bubbles and irregular-shaped large bubbles (> 25 µm) showing many stages of coalescence (Fig. 3a–d and supplementary movie 1). Particularly, large bubbles have smooth pore apertures or interstitial filaments between coalesced cavities. In these cases, melt films between bubbles can be very thin, reaching a 1 µm minimum thickness. This corresponds to the inferred critical thickness of liquid film rupture (Cashman and Mangan 1994) that, in the case of equally sized bubbles, is suggested to be caused by approximately equal pressure acting on the film from inside each bubble (Klug et al. 2002). Concurrently, small bubbles present wrinkled melt retraction films, possible in response to release of gas overpressure (Adams et al. 2006). Poorly vesiculated latitic–shoshonitic scoriae have different textures that are predominantly characterized by small bubble population (< 40 µm) and subordinately by large polylobate, amoeboid bubbles separated by thick (> 10 µm) microlite-bearing glass (Fig. 3e, f, and supplementary movie 2).

Microlite types are small (< 5–150 µm) blocky clinopyroxene in the lower part of scoriaceous level (15 vol%, N200 sample) joined to elongated plagioclase in the upper scoria samples (30 vol%, S1 sample).

Vesicle number density and size distributions

Textures of pyroclasts, specifically the Vesicle Number Density (VND) and Size Distribution (VSD), can be quantified from image analyses (see Methods) and used to constrain processes and conditions of magma decompression during the course of the eruption (e.g., Cashman and Mangan 1994; Mangan and Cashman 1996; Hammer et al. 1999). Particularly, VND (number of vesicles per volume unit) is potentially a powerful tool to explore rates of magma ascent because of the relationship between number density and decompression rate observed in laboratory experiments as well as numerical models, while VSDs are largely used to constrain the processes of bubbles nucleation, growth, and coalescence and then to investigate the relationships between magma degassing and eruptive behavior (e.g., Gaonac’h et al. 1996; Blower et al. 2002; Toramaru 2006; Proussevitch et al. 2007a, b).

Table 1 includes VND values calculated by using all the micro-CT images acquired at different resolutions; however, as suggested by other authors (Shea et al. 2010a, b; Rotella et al. 2014), we consider the VND data measured from the higher resolution micro-CT images more representative as small bubbles dominate the bubble density in silicic pyroclasts. Moreover, following the method proposed by Adams et al. (2006) for the 1912 eruption of Novarupta, VND values have been extended to 1 µm vesicle diameter size by extrapolating the exponential curve described by the small bubbles. Particularly, following the last authors, we fitted the trend of smaller bubbles with an exponential curve and extended it to 1 µm vesicle diameter size. As in the case of 1912 eruption of Novarupta, for our trachytic pumice samples, the exponential size distribution corresponds to vesicles ≤ 30–25 µm, while for our scoria samples, the exponential trend fits vesicles smaller than 10–15 µm.

VNDs are quite similar for all trachytic pumice clasts collected at different stratigraphic levels ranging from 3.1 to 3.6 × 1014 m−3, while increase in latitic–shoshonitic scoria samples at the top of the deposit that show VND values up to 2.8 × 1015 m−3, Table 1. The obtained VND values are consistent with those produced by heterogeneous nucleation during decompression experiments on Neapolitan trachytes (4.8 × 1013–2.9 × 1014 m−3; Mastrolorenzo and Pappalardo 2006) and phonolites (4.3 × 1013–3.8 × 1014 m−3; Larsen 2008; 8 × 1013–9 × 1015 m−3; Shea et al. 2010b), as well as those produced by homogeneous nucleation on phonolites (5.7 × 1014–7.7 × 1014 m−3; Iacono-Marziano et al. 2008).

Cumulative vesicle size distributions (CVSDs) of trachytic pumices show a continuous trend characterized by an exponential distribution for small bubbles (less than 25 µm) and a power-law distribution for large bubbles (Fig. 4a). Similar trends have been observed in other Plinian events and widely interpreted as due to expansion and coalescence of earlier formed vesicles (large bubbles’ pattern) during a continuous nucleation process (Gaonac’h et al. 1996; Blower et al. 2002), while the small bubbles reflect the last nucleation event in the shallow conduit (e.g., Baker et al. 2012; Gonnermann and Houghton 2012; Rotella et al. 2014).

On the contrary, CVSDs of scoria samples show irregular trends formed by multiple curved segments suggesting discrete vesiculation events (Fig. 4b). These sorts of distributions are generally not typical of Plinian clasts; however, they have been previously recognized in experimental samples that suffered carbonate interaction and explained as due to vesiculation pulses triggered by fast release of CO2-rich fluids (Blythe et al. 2015). Interestingly, it is the first time that these trends have been observed in natural Somma–Vesuvius rocks, possibly thanks to the high-resolution of the applied 3D microtomographic technique (examples of 3D micro-CT data for trachytic and latitic–shoshonitic rocks are in supplementary movies 1 and 2, respectively).

Number density and size distribution of throats, connectivity, and permeability

Total Throats Number Densities (TNDs) range from 8.6 × 1013 to 4.4 × 1014 m−3, not showing substantial differences between latitic–shoshonitic scoriae and trachytic pumices, Table 1. The ratio between numbers of throats/total pores (#th/#por in Table 1) and throats/connected pores (#th/#con) ranges, respectively, from 2.1 to 3.6 and 3.4 to 5.2 in trachyte and 1.6 to 1.8 and 3 to 4.5 in latite–shoshonite. Their cumulative distributions (CTSDs, Cumulative Throats Size Distributions) follow broadly power-law trends for large sizes, while seem to follow exponential trends for the small sizes, similar to those observed in the vesicle size distributions of trachytic pumice, Fig. 4c. These observations are consistent with the fact that bubble growth in pumices is better developed than in scoriae, resulting in thinner melt films, which facilitate the occurrence of coalescence process between neighboring vesicles (e.g., higher #th/#por in trachyte with respect to latite). Moreover, they reflect the high number of small isolated vesicles contained in scoriae, characterized by different CVSDs’ trends. The degree of bubble interconnection has been evaluated based on bubble connectivity by the procedures described in Okumura et al. (2012). Connectivity is defined as Ψ = Nmax/Ntotal (where Nmax and Ntotal represent the voxel numbers belonging to the largest bubble and to all of the bubbles, respectively) and reaches the very high value of about 0.99 in all samples. Permeability values (see Methods) range from 1.42 to 4.09 × 10−12 m2 for trachytic pumices and from 0.45 to 0.89 × 10−12 m2 for latitic–shoshonitic scoriae, without showing significant variations among the three orthogonal directions (Table 1).

Microlite size distributions

Following the pioneering work of Marsh (1988) and Cashman and Marsh (1988), the size of microlites and their abundance can be represented as Crystal Size Distributions (CSDs), generally shown as a semi-logarithmic plot of population density (number of crystals per unit volume) vs crystal size (maximum length) with the slope equal to 1/(growth rate × residence time). Thus, if the growth rate is known, the crystallization time can be computed.

CSDs for clinopyroxene present in the glassy groundmass of our latitic–shoshonitic samples were obtained by textural analyses of 2D back-scattered electron images; however, it was not possible to analyze plagioclase crystals as their average atomic mass is close to that of the surrounding matrix glass; thus, they cannot be quantitatively resolved in the BSE images.

These distributions show intercept and slope values of 17.69 and 19.31 mm−4 and − 63 and − 93.2 mm−1 (for N200 and S1 samples, respectively, Fig. 5a). Because decompression experiments indicate that the growth rate of microlites can be highly dependent on the decompression rate, in our timescale calculations, we used the entire range of growth rates (from 1 × 10−6 to 1 × 10−8 mm/s, Brugger and Hammer 2010) available for microlites. By applying this range of growth rates, we calculated a crystallization time ranging from hours to days, in agreement with clinopyroxene texture displaying elongate, tabular, and swallowtail morphologies that indicate rapid crystallization close to the surface (Fig. 5).

Fig. 5
figure 5

a Crystal size distributions for clinopyroxene microlites (white microcrystals in BSE images). Insert: microlite number volume (Nv) against crystallinity showing microlite formation depths. These must be considered gross estimates as the equilibrium lines are calibrated on the basis of experimental rhyolitic samples (Blundy and Cashman 2008). Back-scattered electron (BSE) images of groundmass textural features in Pomici di Base trachytic (b) latitic–shoshonitic (c and d) samples

Full size image

Radiogenic isotope

To explore the potential influence of limestone assimilation in the evolution of the Pomici di Base magmas, we performed Sr–Nd-isotopic analyses on separated phenocrysts (feldspar and clinopyroxene) and groundmass in all studied samples. Radiogenic systems are in fact a powerful tool to spot magma contamination, due to the different Sr- and Nd-isotopic signatures of primary magmas with respect to crustal rocks.

The obtained results show a marked increment in 87Sr/86Sr ratios from trachytic (0.70753–56) to latitic–shoshonitic melts (0.70760) (Fig. 6a; Table 3), thus suggesting a prominent involvement of crustal contamination in the petrogenesis of the hotter mafic melts (Tliquidus = 1200 °C, calculated using MELTS program; Ghiorso and Sack 1995; Asimov and Ghiorso 1998), as also supported by the abundance of carbonate–metamorphic clasts present as lithic fraction in the Pomici di Base fallout deposits and included as fragments inside juvenile products (Bertagnini et al. 1998; Landi et al. 1999). Particularly, the highest 87Sr/86Sr ratios have been measured in separated latitic–shoshonitic glassy groundmass (0.70760) with respect to sanidine (0.70750–55) and clinopyroxene (0.70747–50) phenocrysts implying that contamination was a later process occurred possibly at shallow level after precipitation of phenocrysts, Fig. 6a.

Fig. 6
figure 6

a 87Sr/86Sr vs 143Nd/144Nd compositions for separated matrices and minerals. Lines represent the results of energy-constrained assimilation and fractional crystallization (EC–AFC; Spera and Bohrson 2001) simulations (see text for further explanations). Ta = initial T assimilant. b H2O* (water in residual melts) vs Vg/Vl (ratio of the volume of gas (Vg) to the volume of melts (Vl), after Balcone-Boissard et al. 2008). Lines correspond to theoretical closed-system degassing evolution for initial water content of 5 wt% (latite–shoshonite), 6 and 8 wt% (trachyte), representative of saturation conditions (Di Matteo et al. 2004, 2006)

Full size image

In other volcanic contexts, partial/selective assimilation has been advocated to explain isotopic disequilibrium between groundmass and feldspars (e.g., Duffield and Ruiz 1998); however, in the case of Pomici di Base rocks, this feature is restricted to mafic melts, while it is not significant in evolved trachyte, thus excluding that fractionation was coupled to a continuous selective assimilation process.

Nd-isotopic compositions are much less variable and cluster around 0.51243–45 both in matrix glasses and in phenocrysts (Fig. 6a), this in agreement with the low Nd content of sedimentary carbonates that leave a very slight fingerprint in the contaminated magmas.

The effects of assimilation on isotope variations have been modeled using the Energy-Constrained Assimilation and Fractional Crystallization (EC-AFC) model of Spera and Bohrson (2001), Table 4. We consider an early intrusion of a primitive shoshonitic magma at a liquidus temperature of 1200 °C into the upper crust at 350–600 °C ambient temperature (De Lorenzo et al. 2006) and 6–8 km depth supposed to be the top of the magma reservoir (Scaillet et al. 2008; Pappalardo and Mastrolorenzo 2010, 2012; Balcone-Boissard et al. 2016), possibly developed inside carbonatic sequences as suggested by borehole (Brocchini et al. 2001) and geophysical investigations (e.g., Berrino et al. 1998; Improta and Corciulo 2006). The model shows that the observed Sr- and Nd-isotopic variation is justified by 2–4% of carbonate rocks contaminating a magma already crystallized for about 55% of its initial mass, Fig. 6a.

Table 4 EC-AFC parameters used for modeling isotopic trend
Full size table

Dynamics of magma ascent in the volcanic conduit

The Plinian fallout phase of the Pomici di Base eruption was fed in its early beginning by sialic (0.34 km3 DRE trachytic) melts followed by a larger volume of mafic (0.96 km3 DRE latitic–shoshonitic) magmas for the entire course of the sustained-column phase. This chemical variation towards mafic composition was not associated with any changes in eruption style, as flow rate as well as column height remained firmly high (Bertagnini et al. 1998) for tens of hours during the whole Plinian phase.

However, our 2D and 3D quantitative textural data and isotopic ratios show a significant difference between sialic and mafic rocks suggesting contrasting degassing as well as crystallization regime during magma ascent and fragmentation in the volcanic conduit.

Bubble nucleation and growth recorded in microlite-free trachytic pumice

Trachytic pumices have high degree of vesiculation (69–87%) and are characterized by large coalescent bubbles separated by thin glass walls, whose thickness approaches the critical rupture threshold. These features, commonly observed in pumice samples of Plinian eruptions from worldwide volcanoes (e.g., Rotella et al. 2014 and reference therein), allow us to hypothesize that during decompression, bubble growth occurred for exsolution up to the achievement of a vesicularity threshold that for silicic melts generally ranges from 70 to 80% in function of magma characteristics (expansion rate, melt viscosity, shear stresses, and presence of different phases). Above this value, experimental data (Namiki and Manga 2008; Takeuchi et al. 2009; Rust and Cashman 2011) indicate the existence of an abrupt increase in permeability due to bubbles coalescence. High throat number densities, power-law exponents > 3 in CVSDs as well as high bubble connectivity and permeability values of trachytic pumice also support extended bubble coalescence during magma ascent.

On the basis of the numerical model developed by Toramaru (2006), we estimated the decompression rate during this eruptive stage using the VND values calculated for pumice samples, surface tension of 0.1 N/m (Iacono Marziano et al. 2007; Mangan and Sisson 2000; Moutada Bonnefoi and Laporte 2004), 6 wt% pre-eruptive H2O content at saturation pressure of 150–200 MPa (Di Matteo et al. 2004), and water diffusivity of 2.41 × 10−12 m2/s. This last value has been calculated using the empirical equation proposed by Fanara et al. (2013) at 1100 °C (minimum experimental temperature adopted by the authors) and a correction factor of 100 to take into account the influence of the temperature (scaling from 1100 to 900 °C; following Watson 1994; Larsen 2008; Shea et al. 2010b). The resulting decompression rate is in the order of 5.5–6.1 MPa/s; however, these values must be considered maximum values as the used formulation does not take into account the role of pre-eruptive magmatic CO2 in the melt that may influence water solubility at saturation.

The obtained high decompression rate implies that during the climax of this early phase of the eruption, magma degassing occurred rapidly under closed-system regime (see also Fig. 6b) from a storage depth of 150–200 MPa (Pappalardo et al. 2004; Scaillet et al. 2008; Pappalardo and Mastrolorenzo 2010, 2012; Balcone-Boissard et al. 2016) up to the fragmentation surface localized at ≤ 40 MPa, as inferred on the basis of residual water content of glass (Fig. 6b). At this point, the viscosity of trachytic magmas shifted from 102.9 to 104.91 Pa s (average values assuming saturation conditions for pre-eruptive stage and T = 900 °C, calculated using the model of Giordano et al. 2010).

Moreover, some pumice clasts show evidence of expansion without further water exsolution (Fig. 6b), a degassing regime that may correspond to the expansion of pre-existing bubbles in the chamber, despite we cannot completely exclude post-fragmentation vesiculation. We consider, however, the first hypothesis more plausible, because this feature is limited to the very first erupted trachytic products that represent the upper gas-rich level in the chamber possibly at critical conditions of volatile oversaturation; the occurrence of hydrous minerals (such as biotite) in the phenocryst assemblage supports crystallization under hydrous melt conditions; and in fact, crystallization experiments on alkaline trachytic melts (e.g., Fabbrizio and Carroll 2008) indicate that biotite instead of amphibole is stable as hydrous phase in alkaline trachytic compositions due to the high K2O/Na2O ratios of these melts. We speculate that the presence of melts in near-saturation conditions at the top of the chamber could act as an eruption trigger generating an excess pressure (5–25 MPa, following Blake 1984) sufficient to cause the wall-rock rupture.

Bubble and microlite nucleation and growth recorded in microlite-rich latitic–shoshonitic scoriae

Latitic–shoshonitic scoriae erupted in the late Plinian fallout stage, on the counterpart, are characterized by poorly vesicular texture (45–59%) as well as thick microlite-rich (crystallinity between 15 and 30 vol%) interstitial glass.

These features imply low magma ascent rate favoring open-system degassing regime (outgassing) and in turn bubble collapse as well as degassing-induced microlite crystallization (see also Figs. 5, 6b). On the other hand, a peculiar aspect of these rocks is their high vesicle number density and multiple CVSDs patterns, in contrast with their low-porosity and high-crystallinity nature. Similar behavior (high VND values and CVSDs distributions) has been previously observed in experimental mafic samples from Somma–Vesuvius and Merapi volcanoes exposed to limestone contamination (Blythe et al. 2015) and presumed due to fast (hours to days, Sottili et al. 2009, 2010; Deegan et al. 2010, 2011; Freda et al. 2011; Troll et al. 2012; Jolis et al. 2013) CO2 nucleation pulses. This last process can be possibly driven by magma digestion of skarn blocks detached by contact aureole formed during an early magma-limestone metamorphic stage (skarn recycling, Jolis et al. 2015). Although we did not observe clear evidence of metasomatic nodules in our juvenile fragments, possibly due to their rare occurrence; Landi et al. (1999) reported the presence, within juvenile clasts, of metasomatic carbonates/skarn nodules (< 0.2 wt% of the total rock) commonly associated with interstitial trachytic glass interpreted as portions of the metasomatized carbonate walls of the upper part of the Pomici di Base magma chamber. In any case, we cannot exclude that magma/wall-rock interaction would involve also CO2-rich pristine limestone.

Hence, we suggest that a mechanism of carbonate/skarn dissolution can reconcile all the 2D and 3D textural as well as isotopic parameters documented in Pomici di Base mafic rocks. We propose that the evacuation of the more evolved trachytic liquids from the cap of the reservoir has caused the inception of caldera collapse with the consequent digestion/assimilation of carbonate/skarn blocks into the remnant latitic–shoshonitic liquids, Fig. 7.

Fig. 7
figure 7

Schematic model for conduit processes during the Plinian phase of Pomici di Base eruption. The fast evacuation of the more evolved trachytic liquids from the cap of the reservoir, caused the inception of the collapse of the deeper part of the plumbing system and the digestion/assimilation of skarn blocks into the remnant hotter latitic–shoshonitic liquids

Full size image

Specifically, we suggest that the widening of the lower part of the plumbing system causes a reduced magma ascent rate giving more time both to gas to escape (outgassing) and to microlite to grow, with a consequent increment of melt viscosity (104.43−7.48 Pa s, average values calculated on the basis of the model proposed by Vona et al. (2011), assuming an equilibrium temperature of 1000 °C). In fact, decompression experiments (e.g., Rutherford and Gardner 2000; Hammer and Rutherford 2002; Couch et al. 2003; Martel and Schmidt 2003) demonstrated that number density and modal content of microlites change accordingly to decompression rate and that microlite crystallization requires timescales of days to weeks, while it can be absent in the case of fast magma ascent.

This stage of conduit collapse as well as the associated phase of slow magma ascent through the conduit, possibly lasted from hours to days as indicated by our CSDs data on clinopyroxene microlite. During this time, the consequent digestion/assimilation of carbonate/skarn blocks triggered CO2 bubble formation and growth, forcing rising and fragmentation of the more viscous magma. Dallai et al. (2001) reported that the occurrence of CO2 degassing due to the interaction of hot magma with carbonate has the potential to change the overall volatile solubility of magmas, thus justifying their ability to rapidly rise and explosively erupt to the surface. Recently Carr et al. (2018) by applying a numerical model of magma ascent to 2006 Merapi eruption, found that the addition of 1000 ppm of CO2 can reduce water solubility in the melt, forcing vesiculation and generating overpressure at the top of the storage region in a short time (1–2 days).

Our EC-AFC simulations indicate the ingestion of 2–4 wt% of limestone by 0.96 km3 (Landi et al. 1999) of latitic–shoshonitic melts that corresponds to the release of a maximum of 0.04 Gt of CO2 during the eruption (considering that for 1 kg of limestone, 439 g of CO2 are liberated for complete decarbonation, Deer et al. 1992). Similar values of CO2 liberation are reported during the AD 79 Pompeii (0.31–0.56 Gt) and the AD 472 Pollena (0.04–0.07 Gt) eruptions at Somma–Vesuvius (Jolis et al. 2015).

The fragmentation mechanisms

Trachytic magma

In high-viscous magma, fragmentation occurs when (1) a critical viscosity-dependent strain rate is exceeded (strain-rate criterion; Papale 1999); (2) gas overpressure overcomes the tensile strength of the melt (stress criterion; Alidibirov 1994; Zhang 1999); or (3) expanding magma exceeds a critical vesicularity (critical volume fraction criterion; Sparks 1978).

The strain-rate criterion is based on the view that rapid acceleration may cause the melt to cross the glass transition and, therefore, fail brittlely. Papale (1999) indicates that magmas fragment in a brittle fashion when a critical, viscosity-dependent strain rate is exceeded. The minimum bulk viscosity (µ) required for strain-induced fragmentation is defined as µ ≥ (CG∞πr3/Q)(1/0.9), where r is the conduit radius (m), Q is the volume flux (m3/s), G∞ is the elastic modulus at infinite frequency (10 GPa), and C is a fitting parameter (0.01 (Pa s)−0.1) (Gonnermann and Manga 2003). Thus, to verify this criterion for the studied trachyte, we used in the above formulation the known mass flux for Pomici di Base eruption of Q = 2.5 × 107 kg/s (Bertagnini et al. 1998) (equivalent to a volume flux of Q = 104 m3/s) and the calculated average viscosity for trachytic melt (104.91 Pa s); however, the obtained conduit radius results around 1 m and thus geologically unrealistic.

The stress criterion establishes that fragmentation takes place when volatile overpressure, ΔPfr, exceeds the tensile strength of the melt and ruptures bubble walls (Alidibirov 1994; Zhang 1999). Spieler et al. (2004) provide a formulation with good fit to a broad range of experimental data: ΔPfr = σm/φ, where σm is the effective tensile strength of the melt (0.995 MPa) and φ is the porosity. This formulation has been modified by Mueller et al. (2008) to account for permeability (k) as follows: ΔPfr = (a k½ + σm)/φ, with a = 8.21 × 105 MPa/m and σm = 1.54 MPa, respectively. Thus, applying in the above formulations the measured porosity and permeability for trachyte melt (0.75 and 2.7 × 10−12 m2), a bubble overpressure ranging from 1.33 to 3.85 MPa is required to cause fragmentation. As the calculated melt viscosity gives relaxation times (τs) of 8.13 × 10−6 s using the expression τs = µs/G∞ (Dingwell and Webb 1989), then the onset of non-Newtonian, un-relaxed, viscoelastic behavior can be fixed at 8.13 × 10−4 s (2 orders of magnitude below, Webb 1997). The above calculated ΔPfr (1.33–3.85 MPa) and timescale (8.13 × 10−4 s) implies unrealistic huge decompression rates (comprised between 1.63 and 4.74 × 103 MPa/s) needed to initiate fragmentation.

The critical volume fraction criterion is thought to arise from some form of instability within the thin bubble walls, once ϕ ≈ 0.75 is reached (Verhoogen 1951; Sparks 1978). At high bubble-interconnectivity condition, the fragmentation efficiency strongly depends on the balance between rate of magma decompression and rate at which gases escape from the rising magma (outgassing). Okumura et al. (2012) have estimated the rate of outgassing from magmas ascending in volcanic conduits on the basis of Darcy’s law and using the calculated gas permeability for silicic melts. This author reports that when the pressure gradient driving the permeable gas flow is assumed to be lithostatic (0.03 MPa m−1), as can be postulated on the basis of the results of numerical models for silicic magma ascent during Plinian eruptions (Papale and Dobran 1993), the gas velocity is estimated to be less than 10−5 m/s at vesicularities up to 70 vol%. As this velocity is much smaller than the decompression rate (5.5–6.1 MPa/s) calculated by VND values in the previous “Bubble and microlite nucleation and growth recorded in microlite-rich latitic–shoshonitic scoriae”, the degree of outgassing can be considered inefficient on the timescale of the eruption, thus allowing the reaching of the 0.75% porosity threshold necessary for fragmentation also in presence of high bubbles connectivity.

We can hence suppose that fragmentation of trachytic magma can occur after bubble expansion when a fixed gas volume threshold is reached (bubble packing state) as also suggested by Mastrolorenzo and Pappalardo (2006) on the basis of a compositional and textural study on both experimental and natural trachytes from Campania volcanoes.

Latitic–shoshonitic magma

In low-viscous liquids, fragmentation can be controlled by inertia due to inertial stretching and hydrodynamic breakup during rapid bubble growth.

Actually, the examined scoria clasts lack the textural features typical of low-viscous inertia-driven fragmentation products (fluidal shapes, very low VND values etc), possibly due to the large increment in melt viscosity caused by outgassing and microlite precipitation processes that affected mafic melts during magma rise in the conduit. On the other hand, the calculated high viscosity values suggest that brittle fragmentation can be a possible mechanism for latitic–shoshonitic melts.

To verify this hypothesis, we applied the mass flux measured for Pomici di Base eruption and the calculated viscosity values for latitic–shoshonitic compositions (104.43−7.48 Pa s) in the expression of Gonnerman and Manga (2003) for strain-induced brittle fragmentation (strain-rate criterion); a wide range of conduit radius values from 0.31 to 171.73 m results from calculation in function of viscosity variation depending on microlite content (ranging from 15 to 30 vol%). Indeed, our 3D data show only localized bubble deformation in mafic scoriae, implying that strain localization (Wrigth and Weinberg 2009) can occur during magma rising in the conduit, and force melts to cross the glass transition and achieve the fragmentation.

Moreover, calculations taking into account other fragmentation mechanisms (stress criterion and critical volume fraction criterion) give a wide range of decompression rates that can reach also extreme values (from 6.68–15.95 to 0.63–1.32 × 104 MPa s−1) and vesicularity values significantly larger compared with those measured in the studied natural samples. It is noteworthy that in these calculations, we have neglected the contribution of CO2-oversaturation condition possibly derived by the postulated limestone assimilation.

Finally, we speculate that latitic–shoshonitic magmas cannot reach classically defined fragmentation conditions and complex mechanisms such as bubble overpressure driven by CO2 oversaturation (see also Pichavant et al. 2013) and strain localization can concur.

Conclusions

In this paper, we show that a combination of textural and petro-chemical quantification of the eruptive products can be a powerful tool for reconstructing volcanic conduit dynamics during volcanic eruptions. Our 2D and 3D quantitative textural data combined with Sr- and Nd-isotopic investigations demonstrated that explosive behavior during the Pomici di Base Plinian eruption was first controlled by rapid decompression under closed-system degassing regime of the uppermost trachytic liquids; subsequently, the inception of the caldera collapse triggered the digestion of detached skarn blocks into the remnant latitic–shoshonitic liquids and in turn the occurrence of rapid vesiculation pulses that contributed to magnify the intensity of the eruption during the evacuation of the mafic liquids.

To conclude, these results highlight the importance of magma/limestone interaction as a syn-eruptive process able to produce vigorous gas liberation, thus accelerating magma ascent and amplifying eruption intensity. This mechanism of CO2 fast liberation warrants more detailed consideration as a mechanism driving explosive basaltic volcanism.