Villarrica volcano (Chile) is one of the most active volcanoes in South America. Its activity is currently characterized by continuous degassing from a summit lava lake/vent punctuated by explosive events. During November 2004 a multidisciplinary experiment was deployed for a 10-d period to define the style of emission and infer shallow conduit dynamics at this basaltic center. This involved collection of thermal, seismic and infrasonic data to describe the background activity confined inside the crater, and use of samples to texturally and chemically characterize the ejecta from more energetic explosions able to attain the crater rim. The background activity was characterized by gas bursting with a frequency of ∼9 events per minute. This involved emission of gas puffs fed by bubble bursting, with larger bursts emplacing sheets of magma onto the lower crater walls. The ejecta population from the more energetic events was characterized by the coexistence of both scoriae and golden pumice. These two types of clasts have different textures but identical glass compositions, suggesting that they underwent different conduit histories. The golden pumice is interpreted as the expanding inner part of a short-lived jet fed by a rapidly ascending, magma batch. The scoria forms the outer portion of the jet and comprises degassed material entrained during passage of the fresh batch through material residing in the upper-most portion of the conduit. We thus have a largely degassed upper column that feeds persistent bubble bursting, through which fresh batches occasionally rise to feed events of relatively higher energy.
 At the two case-type locales for Hawaiian and Strombolian activity, Kilauea and Stromboli, activity tends to be either Hawaiian or Strombolian. However, even at these locations activity can span a range of energies. At Stromboli activity spans time-averaged discharge rates ranging from zero, for low energy gas bursting [Harris and Ripepe, 2007a], through 2–1700 kg s−1 during normal Strombolian activity [calculated from data by Chouet et al., 1974; Blackburn et al., 1976; Ripepe et al., 1993; Patrick, 2005] to 3–4 × 106 kg s−1 during paroxysms [Rosi et al., 2006]. Likewise, fountaining at Kilauea shows a range of time-averaged discharge rates from 102 to 106 kg s−1 [Kilauea Iki 1955, B. Houghton, unpublished data, 2008; Pu'u 'O'o, episodes 15 to 24, Greenland et al., 1988]. In addition, activity at a single basaltic system can span from weakly to moderately energetic Hawaiian and Strombolian activity to violent Strombolian, Subplinian and Plinian explosions as, for example, at Mt. Etna (Italy) [e.g., Polacci et al., 2006a; Behncke et al., 2006; Coltelli et al., 2000] or Villarrica (Chile) [BGVN, 1971, 1972; Clavero and Moreno, 1994]. Recent eruptions at Etna have also shown that rapid alternation can occur between explosive phases of Strombolian and Hawaiian character [Allard et al., 2005; Andronico et al., 2005; Polacci et al., 2006a]. Even at Stromboli, low energy gas bursting can occur side-by-side with higher energy normal Strombolian activity [Harris and Ripepe, 2007a]. There may thus be limitations to the extent to which simple classifications such as Strombolian and Hawaiian are helpful in describing activity, where the physical variables involved in basaltic degassing and explosive eruptions at a single system may be wider than can be accommodated by such a simple classification.
 During November 2004 we deployed a multidisciplinary experiment at Villarrica (Figure 1b) to define the style of emission and infer the shallow conduit dynamics. During this experiment we were able to collect ejecta samples from the most energetic explosions that we could not observe or record with the geophysical instruments, but which caused ejecta to fall on the crater rim. We could, instead, collect seismic, infrasonic and thermal data for the extremely low energy (gas bursting) activity that we could observe, but not sample, because they remained confined inside the crater.
2. Geological Setting and Experiment Setup
2.1. Geological Setting
 Villarrica has been active since the Middle Pleistocene and is one of the most active volcanoes in South America [Clavero and Moreno, 1994]. Villarrica is a stratovolcano, formed by basaltic and basaltic andesite lavas and pyroclastic deposits, with a south-eastern sector that is covered by a glacier [Clavero and Moreno, 1994] (Figure 1a). The summit hosts a 200 m diameter crater that ranges in depth from <100 to ∼150 m, containing (depending on the magma level) either a small (30–40 m diameter) active lava lake or an open, degassing, vent [Witter et al., 2004].
 Postglacial eruptive activity has been characterized by both explosive and effusive activity. The first historic eruption was recorded in 1558 [Petit-Breuilh and Lobato, 1994]. Since this time at least 59 small to moderate explosive eruptions have been documented [Petit-Breuilh and Lobato, 1994; Simkin and Siebert, 1994]. Following the last effusive eruption in 1985, activity has been characterized by persistent degassing. Witter et al.  estimated that, to account for gas fluxes during 1985–2002, degassing of 1.2 km3 of magma was necessary at a time-averaged rate of 2.2 m3 s−1.
 During our November 2004 experiment, the funnel-shaped summit pit was ∼145 m wide at the rim, ∼80 m deep, and hosted a ∼35 m wide vent on its floor. The “lava lake” had receded from view, but the magma free-surface remained shallow because regular bubble bursts were visible ejecting fragments that fell within the crater itself. Larger events plastered sheets of incandescent fluid lava onto the inner crater walls. These sheets then drained back into the vent. Infrequent more energetic events (which we were unable to observe) emplaced lobes of scoria bombs and lapilli around the crater rim. The limited dispersion of these products, no scoria being found beyond ∼100 m from the rim, confirms the low energy character of all events considered here.
2.2. Experiment Setup
 The products of the more energetic events were collected for textural analysis during two sampling expeditions made to the summit on 9 and 17 November 2004, respectively. At these times fresh bombs and lapilli formed a discontinuous cover on new snow (Figure 1d). On 9 November two separate collections were made along the western and eastern crater rim, respectively; each location representing a discrete ejecta lobe (one west directed, the other east directed). The textural features of the bombs/lapilli confirmed that these were the products of at least two separate explosions. Twenty three clasts were collected from the eastern lobe (sample VI_E) and 57 from the western lobe (sample VI_W). A single sample of more than 300 clasts (sample VII_W) was collected from the western rim on 17 November. These ejecta overlapped in distribution with the western lobe of 9 November but their presence on newly fallen snow indicated that they were the products of an explosion since that date. The first two sets of samples collected on 9 November comprised lower numbers of clasts because weather conditions were poor and deteriorating rapidly, resulting in safety concerns that limited time available at the summit.
 The bulk densities of all 1–7 cm diameter scoria clasts were measured following the technique of Houghton and Wilson  and converted to bulk vesicularities using an average bulk density of 2640 kg/m3 (from Witter et al. ). The density distribution of each sample was then used as a filter to select a subset of 4–12 clasts including representative examples of maximum, minimum, and modal vesicularities. Following the methodology of Gurioli et al.  we next quantified 2- and 3-D vesicle size distributions, i.e., number per area (Na), and stereologically derived number per volume (Nv) using the conversion coefficient for spheres of Sahagian and Proussevitch . Although some of the samples display bubbles with complex (or elongated) shapes, the spherical geometry is the one that most closely approximates the entire bubble population in our samples. For Plinian eruptions, the uncertainty introduced by this assumption usually causes the summed vesicle volume fraction to be ∼10% higher than the measured vesicularity [Adams et al., 2006]. However, seven of our samples show the summed volume fraction to be less than 5% higher than the measured vesicularity; five samples are 6–9% higher and only two (which both show clear evidence of post fragmentation expansion) have >20% differences. Finally, the abundances of phenocrysts (diameter >200 μm) plus microphenocrysts (200–30 μm) and microlites (<30 μm) were measured (Table 1).
Table 1. Summary of Density, Vesicle and Crystal Data for Each Clasta
Nv corr, cm−3
The table shows, for each sample: density; derived-density vesicularity (Ves); number of images processed for each sample (Ni); vesicle-free percentage of phenocrysts and microphenocrysts (Pheno); vesicle-free percentage of microlites (Micro); total area number density of vesicles (Na); total volumetric number density of vesicles referenced to whole clast (Nv); total volumetric number density of vesicles referenced to melt only (Nvcorr).
6.8 × 103
1.4 × 106
1.4 × 107
5.5 × 103
1.0 × 106
5.5 × 106
7.0 × 103
1.6 × 106
6.4 × 106
8.5 × 103
2.1 × 106
5.6 × 106
4.1 × 103
7.4 × 105
9.4 × 106
3.3 × 103
4.5 × 105
2.6 × 106
3.1 × 103
6.2 × 105
3.4 × 106
3.3 × 103
4.2 × 105
1.6 × 106
2.6 × 103
5.1 × 105
1.6 × 106
9.3 × 102
6.4 × 104
7.4 × 105
3.3 × 103
3.8 × 105
2.4 × 106
6.1 × 103
8.3 × 105
3.2 × 106
5.8 × 103
9.3 × 105
3.3 × 106
2.5 × 103
4.7 × 105
1.0 × 106
 Two larger scoria bombs from the 9 November sample suite were processed with synchrotron X-ray computed microtomography; the only high-resolution, non-destructive, technique available to reconstruct the internal structure of these porous materials. The goal was to provide 3-D views of the scoria textures and to compare textural parameters directly measured in 3-D with those obtained via conventional 2-D analysis. The experiments were run using the SYRMEP beamline of the Elettra Synchrotron radiation facility in Basovizza (Trieste, Italy) using a 2048 × 2048 pixel CCD camera with a pixel size of 14 μm, generating a field of view of 28 × 28 mm2. On the basis of this, samples were roughly cut into cubes with 2 cm sides. Small and larger vesicles were counted together by processing the same tomographed volume. The reconstructed tomographic slices were then stacked to provide a 3-D volume of each sample that was further processed to retrieve individual vesicle volumes. Details of the experimental conditions and procedure can be found by Polacci et al. [2006b].
 Electron microprobe analyses of glass compositions were determined using the University of Hawaii five-spectrometer Cameca SX-50 electron microprobe using SAMx automation. The operating conditions were a 1 μm spot size, 15 kV and 10 nA beam current. Peak counting times were 30 s for all elements except for P, which was measured for 90 s. Na was analyzed first to minimize its loss. Accuracy was checked on glass standards VG-2 and A-99 (basalts). The raw data were corrected using ZAF-PAP procedures [e.g., Reed, 1993] and reported analyses (Table 2) are averages of 12 spot analyses.
Table 2. Electron Microprobe Analyses of Matrix Glasses, Sample Density (in kg m3), Phenocryst Content (ϕ), Melt Viscosity [(ηm)] and Melt-Crystal Mixture Viscosity [η(ϕ)] in Pa sa
Microprobe results are reported in oxide wt.%, where FeOT = total Fe as FeO, and η(ϕ) is calculated for a range of maximum crystal concentrations, from 60% [to give maximum expected η(ϕ)] to 70% [to give minimum expected η(ϕ)]. In addition, the analysis of the matrix glass for reticulite sampled in 1999 is given [Witter et al., 2004] for comparison.
 A geophysical array was also installed on the eastern rim of the summit pit. This consisted of a geophone, a microphone and a thermal infrared thermometer targeting the vent and sampling at 54 Hz, in an equipment set-up identical to those described for Stromboli by Ripepe et al.  and Harris and Ripepe [2007b]. Data were collected over a ∼2 h period spanning 15:15 to 17:15 (local time) on 17 November. At the same time a thermal camera sampling at 30 Hz was aimed at the active vent to allow characterization of events following Patrick et al. . These data comprise 320 × 240 pixel images, giving pixel-integrated temperatures corrected for atmospheric effects using atmospheric temperature (14°C) and humidity (25–35%) measured at the time and location of image collection. The camera was located on the west rim and angled downward by ∼43° to target the deepest visible portion of the crater over a line of sight distance of ∼110 m. All angles and distances were measured using a LaserTech laser range finder and, given a thermal image pixel size of 1.3 mrad, mean that pixel temperatures were integrated over a 15 cm wide area. However, given the pointing angle of the instrument, the effective pixel width for a particle ascending vertically through the field of view will be 11 cm. We thus use a pixel dimension of 13 ± 2 cm.
3. Low Energy (Background) Activity Characterized Using Thermal Image Data and the Geophysical Array
 The thermal camera and geophysical data allowed us to document the low energy (gas bursting) activity that persists between the rarer, more energetic, events which we sampled. We hence term the persistent low energy activity “background activity”.
3.1. Thermal Image Data
 We recorded thermal image data for a total of 693 low energy events allowing us to identify 3 subtypes of emission (Figure 2). Type 1 events comprised gas-only emissions (Figure 2a) characterized by ascent of a rolling cloud of high temperature (600–700°C) gas and aerosol. Such gas bursts had a starting form of a thermal and rose convectively, slowing as they cooled. This form of emission has a style identical to the gas bursting activity described for Stromboli by Harris and Ripepe [2007a]. Type 2 emissions involved the emission of gas and ejecta, and could be split into types 2a and 2b. Type 2a events involved less heavily loaded ejecta clouds, with each clast being identifiable as a discrete thermal anomaly up to 0.3–1.5 m in maximum dimension (Figure 2b). Type 2b events involved more heavily loaded ejecta clouds, with individual clasts being almost 3 m across (Figure 2c). In some cases individual blebs coalesced and underwent rheomorphic flow upon landing (Figure 2c). Type 3 events involved the ejection of a coherent sheet of magma, and some detached blebs (Figure 2d), with the sheet showing drain back flow into the vent for a period of several minutes after emplacement. Type 3 events were capable of emitting magma sheets 2–17 m in width (mean = 7 m, σ = 2 m).
 The total of 693 events recorded over the 78.5-min-long measurement period gives an event frequency of 8.8/min. Of these, Type 1 events were the most common followed by types 2a, 2b, and 3 (Table 3 and Figure 3). No patterns were observed in the event type distribution, although pauses in the quasi-continuous type 1 activity, lasting up 2.5 min, were apparent (Figure 3). Durations, emission velocities and launch angles were similar for all events, having means (±1σ) of 0.7 ± 0.5 s, 12 ± 6 m/s and 69 ± 16°, respectively (Table 3). Maximum ejecta heights were also similar, although type 2a events were the only events to attain heights of greater than 20 m (27 type 2a events reached >20 m). Maximum temperatures (∼980°C) were obtained for type 3 events (Table 3).
Table 3. Thermal-Image-Extracted Parameters for Each Observed Event Typea
Duration is defined by the time difference between the appearance of the first and last clast within the image field of view. Heights and velocities are given for a 13 cm pixel, and have an error of ±15 % due to uncertainty in the pixel dimension. Velocity is the maximum velocity obtained for the fastest moving clast, for which the launch angle is also given.
Launch Angle, °
Max Temp, °C
 Thus although type 3 events were the most infrequent, type 2a events were capable of sending ejecta to the highest levels. However, more energetic and less frequent explosive events, capable of sending scoria beyond the rim (as sampled by us), must also occur. These events had timescales of days and ejecta heights that were at least 80 m (allowing them to attain the crater rim).
3.2. Geophysical Array
 The seismic, acoustic and thermal infrared thermometer data recorded the background activity as repeated waveforms with impulsive onsets, of which an example is given in Figure 4. Seismic and infrasonic pulses were strongly coupled and similar in form to those recorded at Stromboli during persistent gas bursting [see Ripepe and Gordeev, 1999; Ripepe et al., 1996]. This implies that the seismic source was very shallow (within meters of the free surface) and controlled by gas bubbles in the magma column. At Villarrica, type 1 gas bursting has a typical periodicity of ∼4 s, longer than the very stable 1–2 s periodicity recorded for Stromboli [Ripepe and Gordeev, 1999; Ripepe et al., 1996; Harris and Ripepe, 2007a].
 Thermal transients typically follow the seismoacoustic signals by up to 1.7 seconds. Following the method of Ripepe et al. , and given a typical emission velocity of 10–13 m/s, this delay implies that the free surface was a further 20–30 m deeper than the line of sight distance into the crater (i.e., 80 m). Thus the free surface was 100–110 m below the rim and an offset of 20–30 m must be added to the ejecta heights given in Table 3 to obtain height above source. The thermal sensor also revealed higher amplitude events every 90–100 s, which is the timescale of the larger events (types 2b and 3) recorded by the thermal camera.
 We thus record two styles of gas burst emission. First, persistent gas bursting is recorded infrasonically and seismically, proceeding at a rate of ∼15 bursts/min. This is recorded as type 1 events in the thermal camera data. Second, larger explosive events are recorded thermally. These are the type 2 and 3 events recorded in the thermal camera data and have a frequency of between 3 and 0.3 events/min. However, some of the type 2 and 3 events had no acoustic (or seismic) signals associated with them. This observation implies that some events were fed by arrival of large gas pockets with very small overpressures.
4. Higher Energy Events Characterized by Clast Density and Vesicularity
4.1. Clast Morphology and Macroscopic Texture
 Sampled ejecta included centimeter- and decimeter-sized scoria, centimeter-sized golden pumice [Sharp et al., 1987], and fresh glassy coarse-ash shards. Bombs were round or elongated in shape and included spindle bombs as well as flattened spatter forms. Scoriae were black and iridescent with a metallic surface luster and rich in phenocrysts of olivine and plagioclase. They were all characterized by heterogeneous vesicularities comprising a mixture of small, mm-sized, spherical vesicles and large (up to decimeter) convoluted or spherical vesicles. In addition low density zones, characterized by numerous vesicles with thin glass walls, and denser zones, characterized by broader glassy regions separating more widely dispersed vesicles, were apparent in close proximity to each other. The golden pumices had a yellowish color and more homogeneous mm-sized vesicles.
4.2. Density Data
 Two density modes are apparent among the clasts, reflecting the two mesoscopic clast types (golden pumice and scoria) observed in the field (Figure 5). While the golden pumice is represented by a relatively well-defined low-density mode at 100–300 kg m−3, the scoria component is represented by a broad density range spanning 500–1600 kg m−3. The two clast types show no overlap in density values; a gap at 400 kg m−3 separating the two groups (Figure 5).
 The three samples can be distinguished on the basis of the relative proportions of these two modal populations. Sample VI_E, from 9 November, lacks the golden pumice component and also the densest scoria clasts (Figure 5a). It is thus marked by a broad, flat, density distribution with no obvious mode. The lack of golden pumice and dense clasts could reflect the small sample size or a bias in the sampling due to the brief collection period available to us and the poor visibility conditions. In contrast, sample VI_W, also from 9 November, comprises both subpopulations: golden pumice and subordinate scoria. The golden pumice is marked by a well defined and uniform distribution with a prominent mode at 300 kg m−3. The scoria subpopulation is marked by a broad, flat distribution ranging from 500 to 1100 kg m−3 (Figure 5b). The relative abundances of clast types are reversed in sample VII_W, from 17 November, with the scoria subpopulation having a well-defined mode at 900 kg m−3 and a dense tail extending to 1600 kg m−3 (Figure 5c). The differences between samples VI_W and VII_W are too pronounced, and the samples too large, to represent some form of sampling bias. The T-test value calculated for these two samples confirm our observations. The probability of the two populations being the same is less than 0.0001. We thus believe that they reflect a primary variation in the nature of the ejecta between two different explosions (or groups of explosions).
4.3. Qualitative Observations of Vesicles
 The juvenile populations of the three samples are dominated by crystal-rich scoria (25–56% phenocrysts, Table 1) with vesicularities of 42–80%, and a mode between 63 and 75% (Table 1). In contrast the golden pumice has a narrower, but higher, range of vesicularities (89–96%, Table 1). This range is slightly lower than that (95–99%) found in reticulite from fountaining events at Kilauea by Mangan and Cashman .
 We can define three density groups for the scoriae: low (400–600 kg m−3), middle (600–1100 kg m−3) and high (1100–1800 kg m−3), with a fourth extremely low density group (100–300 kg m−3) representing the golden pumice (Figure 6). On the scale of an individual scoria clast two different textures, as defined by the intermediate to small vesicles, can be observed sometimes intermingled with each other. The first texture, type p, consists of a subpopulation of relatively closely spaced vesicles of predominantly intermediate size which, in extreme cases, take on the polygonal form described for reticulite by Mangan and Cashman . The second texture, type s, consists of a larger subpopulation of smaller vesicles with thicker glass walls. The type p texture characterizes the golden pumice (Figure 6a) and some domains in the low-to-middle density scoria (Figures 6b and 6c). The type s texture is typical of the bulk of the scoria clasts, and is especially dominant in the high density scoria (Figures 6d and 6e).
 Clasts of both types may, or may not, contain mm-to-cm sized vesicles (Figure 6). These larger vesicles are characterized by complex and irregular shapes (Figures 6c and 6e), or by spherical shapes surrounded by deformed vesicles (Figure 6d). The former are interpreted as being produced by coalescence of spherical-to-slightly deformed smaller vesicles (Figure 6c) or, if located at the rim of the clast, as air entrapped in the clast during spinning and in-flight deformation (Figure 6e). The latter (Figure 6d) are consistent with post-fragmentation expansion [e.g., Mangan and Cashman, 1996].
 Increasing glass wall thickness and decreasing vesicle size seems to be the main factors controlling the transition from low to high dense clasts, as can be seen by looking down the first column of Figure 7. However, all clasts are characterized by the presence of very thin (up to 1 micron) glass walls (Figure 7). For all clasts, vesicles are generally surrounded by melt, but a few vesicles wet phenocrysts (Figure 7a). Expansion of smaller vesicles into larger vesicles, to generate the doughnut-like features of Klug et al. , is common in the middle density scoria (e.g., Figure 7c).
 Although each sample is characterized by a mixture of type p and s textures, in detail each sample shows peculiar and contrasting features. These differences are mainly related to clast textures and to heterogeneity and/or mingling in the clast populations as seen in the varying proportions and distributions of the type p and s regions within single clasts from each sample. Clasts from sample VI_E contain the greatest abundance of small vesicles and, locally, show elongation (Figures 7e, 7f and 7g).
 Vesicularities measured (using X-ray microtomography) in the two scoriaceus bombs range from 71 to 75%, placing these ejecta at the upper end of the density range described above. This can reflect the fact that the samples chosen for the tomography were larger than the clasts that were analyzed above, so that they may have experienced significant postfragmentation expansion. However, the tomography procedure allowed us to observe that the two bombs were dominated by a population of large, highly interconnected vesicles (interconnectivity > 90%), with complex and irregular shapes produced by coalescence of spherical-to-slightly deformed smaller vesicles (Figure 8). The high interconnectivity of the vesicles and their complex, often tortuous, shapes resemble textural features found in scoria clasts ejected by the mild to moderate explosive activity at Stromboli. At Stromboli such textures have been interpreted as preferred pathways for permeable gas flow [Polacci et al., 2007]. We do not observe this high permeability in the 2D samples (Figure 6). This can be due to two factors. First, as described above, it may result from the difference in size of the samples used for tomography versus those used for image analysis. Second, it is possibly an artifact caused by the resolution of tomography which is not high enough to resolve thinner (micron-sized) bubble walls observed during image analysis.
4.4. Quantitative Measurements of Vesicle Populations
 The vesicle sizes range in diameter from 0.02 to 10 mm, with the scoriae from VI_E showing the smallest sized vesicles (Figures 7 and 9). For the two larger bombs used for the tomography, the vesicles were in the volume range 1.6 × 10−5 mm3 to 18.2 mm3. This converts to an equivalent diameter range of 0.03–3.2 mm. Because the larger vesicles are not perfect spheres, the upper limit probably spans 4–5 mm. If we ignore the coarse population, all samples show a slightly asymmetrical, positively skewed and unimodal population (Figure 9). The golden pumices show the coarsest modes at 0.40–0.75 mm, with the low-, middle-, and high-density clasts all showing less coarse, but similar, modal positions (typically around 0.25 mm). Most clasts also show isolated coarse modes between 4 and 10 mm (Figure 9). This corresponds to the large vesicles described in section 4.3. Because of the interpretation given in section 4.3, these large vesicles seem not to belong to the prefragmentation population. We therefore do not consider these largest bubbles any further.
 Our samples show vesicle number densities ranging from 6 × 104 to 2 × 106 cm−3 (Nv in Table 1). The number density displayed by the vesicle population of the 3-D investigated scoria bombs ranges between 1 and 2 × 104 cm−3. The full Nv range obtained by us for Villarrica is comparable with published data for both Strombolian and Hawaiian ejecta (Figure 10). The golden pumice show higher Nv when compared with reticulite from Hawaiian lava fountains analyzed by Mangan and Cashman , as well as from Kilauea Iki (Figure 10). However, some of our Villarrica scoriae also show the highest number densities so far recorded for Strombolian/Hawaiian activity (Figure 10). Although our golden pumice have the lowest Nv of any of our samples (see VIIc_104 in Table 1), they also have values comparable to those of the scoriae (see VI_54 in Table 1). The variability in the golden pumice Nv depends on its degree of expansion. In contrast, all of the scoriae have similar Nv, with little variation between the low and high density scoriae (Table 1). The only exception is sample VI_E, which has the highest Nv. However, Nv corrected for vesicularity shows more consistent values across all samples (Table 1). An explanation for such a pattern is that the batch of melt associated with each sample had, in each case, a similar early history of vesicle nucleation with similar decompression rates. The diversity in terms of vesicle size, shape, etc. reflects a subsequent (later) divergence in degassing patterns, during conduit ascent.
4.5. Crystal Data and Glass Chemistry
 Both the golden pumice and scoriae contain abundant phenocrysts (Figures 6, 7 and 8) and minor microphenocrysts, both of olivine and plagioclase, as well as rare microlites of plagioclase and chromian spinel. Olivines are usually euhedral and plagioclase crystals are euhedral to subhedral. The matrix of the golden pumice comprises a clear, light-brown, glass with only rare plagioclase microlites (Figure 11a). The matrix of the scoriae is a brown-glass with sparse plagioclase microlites and dendritic magnetite microlites (Figure 11b). However, the percentage of microlites is very low in all samples, typically less than 2% (Table 1), similar to Strombolian ejecta at Stromboli [Lautze and Houghton, 2005, 2007; Polacci et al., 2006a]. We also do not find the dense networks of plagioclase and magnetite microlites observed by Witter et al.  in their 1999 samples from Villarrica.
 The vesicle-free volume percentage of phenocrysts plus microphenocrysts ranges from 13 to 56% (Table 1), with a mean (±1σ) of 33 ± 11%. Each sample is characterized by differing crystal abundances and relationships between abundance and density (Table 1). Sample VI_E contains the highest abundance of crystals, with an average of 40 ± 11%, and shows an abundance that decreases with increasing density. Sample VI_W has an average of 33 ± 14%, and shows an increasing abundance with increasing density and vesicularity. Sample VII_W contains the lowest and most homogeneous crystal abundance, with an average of 27 ± 2%. The lowest crystal content (13%) is found in the golden pumice of sample VI_W. If this is removed from the total population, the lower bound of the crystallinity range increases to 25%, and the mean increases to 35 ± 10%.
 Although our samples have differing densities and crystal contents, the glasses for all samples show very similar compositions (Table 2). According to the classification scheme of LeBas et al.  these glasses are basaltic andesite, and are very similar to those presented by Witter et al.  for samples collected in 1999 (Table 2). These data confirm the fact that, in terms of the major elements, Villarrica's magma has not changed since 1984 [Witter et al., 2004].
 Melt phase viscosity (ηm) can be calculated following Shaw , where the glass composition of Table 2, with the shallow system melt temperature (1150 ± 15°C) and water content (0.1 wt%) given by Witter et al. , yields 100–160 Pa s. These values are in good agreement with the viscosities calculated using Whittington et al.  and Giordano and Dingwell  which, although being derived for tephrite and alkali basalt, respectively give 80 ± 25 and 120 ± 40 Pa s. Increasing the water content to that obtained by Witter et al.  from glass inclusions (1.4 wt%) decreases ηm to 45 ± 10 Pa s. However, given that we are considering shallow system dynamics, we prefer to use the lower (0.1 wt%) value for water content to take into account the degassed nature of the shallow system magma.
in which ϕ is the crystal concentration and R is 1/ϕmax, ϕmax being the maximum concentration that can be attained by the crystals. In cases where ϕ < 0.4, equation (1) with an exponent of 2.5 appears valid [Lejeune and Richet, 1995; Costa, 2005]. At higher crystallinities modifications are required, where Costa  provides an empirically derived model for silicate melts at higher crystallinities. Given that our sample crystallinities have averages that are typically ≤40%, we use equation (1). For magma, R of 1.67 has been proposed [Marsh, 1981]. This corresponds to ϕmax of 60% at which the mixture has infinite viscosity. Values as high as 71% (R = 1.41) are considered realistic by Pinkerton and Stevenson , with ϕmax of 50–70% giving reasonable fits to field-derived viscosities for active lavas at Etna and Mauna Loa [Harris and Allen, 2008]. Because we observe a maximum crystallinity of 56% in one sample (VI_34) we apply equation (1) to obtain η(ϕ) using ϕmax of 60–70% (Table 2).
 The chemistry and crystal content of each of the samples yields bubble-free mixture viscosities of 210–740, 400–1630, and 470–4410 Pa s, for the golden pumice, middle density and high density clasts, respectively (Table 2). The full viscosity range (210–4410 Pa s) spans that expected for Hawaiian magmas upon eruption with 0 to 30% crystals (i.e., 100–1400 Pa s [Moore, 1987; Crisp et al., 1994; Harris and Allen, 2008]) and Stromboli magma with 10 to 40% crystals (2000–5000 Pa s [Lautze and Houghton, 2007]). While viscosities calculated for Villarrica's golden pumice overlap the Hawaiian range, those calculated for the high density clasts overlap with Stromboli's range. Low density clast VI_34 has a particularly high phenocryst content. This yields η(ϕ) ≥ 104 Pa s (Table 2); thereby exceeding the upper bound obtained for Stromboli.
 These calculations are for a mixture of melt and crystals, and do not account for the presence of bubbles. Results will thus likely be a maximum bound for the bulk viscosity of the complete melt-crystal-bubble mixture. Low density clast VI_34, for example, has a high vesicularity with sheared bubbles being present. The presence of a sheared vesicle population will reduce mixture viscosity [Manga et al., 1998], as was suggested by Crisp et al.  to explain the anomalously low viscosities obtained for Mauna Loa's 1984 lava by Moore .
5.2. Implications of Geophysical Measurements
 The persistent background activity recorded in our 17 November thermal camera and geophysical data is consistent with the arrival of a persistent stream of large bubbles at the magma free-surface, which arrive and burst at a rate of ∼530 per hour (Table 2). These feed persistent, but weakly explosive, gas bursting. If the width of the sheets thrown out as part of the type 3 events (2–17 m) mimics that of the bursting bubble, this gives the largest bubbles as attaining radii of up to 10 m and volumes (assuming a spherical form) of 200–2500 m3.
 At Villarrica we thus observe what Slezin  describes as two phase flow with discrete gas separation (to allow slug formation and bubble bursting). In the model of Slezin , decreased viscosity or ascent rate favors such a regime, rather than a dispersion regime in which there is continuous eruption of a gas-pyroclastic suspension, leading to “catastrophic” (Plinian) eruptions. Similarly, in the rise-rate model of Parfitt and Wilson , if large bubbles ascend sufficiently fast (in relation to the ascent velocity of the surrounding magma) bubble coalescence will occur to form larger bubbles or gas slugs. These, upon reaching the surface, burst to generate an explosive emission. If the rise speed of the magma is such that bubbles cannot ascend far before the magma itself is erupted, slug formation is suppressed, with the bubbles instead being coupled to the magma [Parfitt and Wilson, 1995]. Rapid expansion of the “locked-in” bubble population at shallow depths then generates a fountain [Parfitt, 2004]. Thus a characteristic set of viscosity and rise rate conditions appear to determine whether, and how, gas slugs can develop to feed discrete bubble bursting as opposed to Plinian eruptions at one end and fountaining at the other. Viscosities at Villarrica span those of Hawaii and Stromboli, but drive a style of activity that is dominated by weakly explosive gas bursting activity, with rarer stronger (but still mild) Strombolian eruptions.
5.3. Implications of Vesicles and Crystal Analysis
 Complex shapes are mainly found for only the largest (>3 mm) vesicles (Figures 7 and 8). These features can be explained by coalescence and expansion, which was continuous right up until fragmentation. In contrast, the spherical nature of the small vesicles suggests short and simple growth histories with minimal shearing, or sufficiently long residence times for vesicles to relax.
 Of the three samples, VI_E has the highest vesicle number density and greatest degree of vesicle elongation. Because vesicle number density is a strong indicator of eruption intensity in mafic magmas [Polacci et al., 2006b; J. E. Sable et al., Eruption mechanism during the climax of the Tarawera 1886 basaltic Plinian eruption inferred from microtextural characteristics of the deposits, submitted to Geological Society of London Special Volume, 2007], these features could indicate that sample VI_E experienced the most rapid ascent rates at the time of vesicle nucleation, implying that vesicles had limited time to grow by decompression and coalescence before the magma was ejected. In contrast, VII_W has the widest range of vesicle number density (including the lowest values of any sample) and the most uniform texture. These features indicate that VII_W may have experienced lower ascent rates at the time of vesicle nucleation, allowing vesicles to expand to the maximum extent in the golden pumice clasts. Characteristics of VI_W are intermediate between those of VI_E and VII_W indicating moderate ascent rates. All samples are poor in microlites, which can indicate ascent of magma from depth under near-isothermal conditions and the maintenance of high temperature up until the time of fragmentation (i.e., limited or no cooling).
 In our Villarrica samples the golden pumices have been able to mingle with the scoriae, as indicated by the presence of the type p domains in the low and medium density scoriae (Figure 6). This textural diversity is purely a physical feature, and all samples have identical matrix chemistries; consistent with continuous replenishment from a chemically stable magma source. However, the presence of these two different textures is indicative of different degassing and conduit histories for the associate melt.
 In our model we propose that the golden pumice represents the expanding inner part of a short-lived jet fed by a rapidly ascending, magma batch. In this scenario, the scoria forms the outer portion of the jet and comprises material entrained during passage of the fresh batch through the relatively degassed material residing in the upper-most portion of the conduit. The presence of the densest material also indicates that each explosion incorporated magmas residing in the uppermost portion of the conduit, where degassed magma is always being mixed and stirred by almost constant bubble passage, bursting, and drain back.
 Our integrated experiment allowed us to define a system characterized by mild bubble bursting in a largely degassed shallow-system, interrupted by rarer (very-short-lived) events that involved rapidly ascending, less-degassed, batches, able to mingle with the dense magma. However, the system can also transition into Hawaiian fountaining, as in 1971 [BGVN, 1971, 1972] or violent Strombolian to Plinian activity [Clavero and Moreno, 1994]. Clearly there is a great range in activity style at Villarrica. This no doubt reflects an equally wide range in magma supply rates, ascent velocities, volatile contents and/or viscosities experienced by this system; ranges that are wider than for systems that tend to experience more stable activity styles as at Stromboli and Kilauea.
 We thank D. R. Baker and L. Mancini for help in running the synchrotron at Basovizza and E. Hellebrand for help in running the microprobe at University of Hawaii. M. Patrick for help with FLIR data collection, E. Calder, R. Carey, C. Cigolini, and J. Hammer for ejecta sampling, and N. Yakos for FLIR data processing. We also thank the Italian Guardia di Finanza, A. Cristaudo, and N. Leo, for their invaluable help and support during deployment and sampling, as well as S. Carn. J. Witter and E. Calder are thanked for their field support, discussions and aid with logistical arrangements without which this study would have been impossible. We thank S. Sparks and L. Pioli for their very helpful comments which greatly improved the content of this manuscript. We are also grateful to K. Taylor for excellent editorial handling.