Introduction

Stromboli, the northernmost volcano-island of the Aeolian Archipelago in South Italy, is famous in the volcanological literature thanks to the persistence of its explosive activity over the last 1,400–1,800 years (Rosi et al. 2000). It rises 924 m asl and represents the emergent part of a large 2,500-m high stratovolcano. Persistent activity at Stromboli consists of mild explosions every 10–20 minutes, which eject scoriaceous bombs, lapilli and ash from vents where glowing lava stands at a high level in an open conduit. This activity is associated with a continuous streaming of gas from the summit craters with an estimated output of 6–12×103 tons/day (Allard et al. 1994). The source mechanisms of the intermittent explosions have been interpreted as the result of the periodic formation and release of gas slugs in relation to the collapse of unstable foam layers formed by the collection and coalescence of rising bubbles in the conduit (Jaupart and Vergniolle 1988, 1989; Ripepe et al 2001, 2002; Chouet et al. 2003) or bubble coalescence in slowly rising magma (Wilson and Head 1981; Parfitt and Wilson 1995). According to Chouet et al. (2003), the gas slug would be formed at 220–260 m beneath the active vents. Therefore, the scoriae expelled with the gas jets during these explosions most likely arise from similar or shallower depths. This “normal activity” is periodically interrupted by more energetic explosions called either “major explosions” or “paroxysms” depending on the power of the blast. Major explosions consist of a short-lived blast which causes the ballistic fallout of scoriaceous and pumiceous lapilli, spatters and blocks, within a distance of several hundreds of meters from the craters. On average 1–2 “major explosions” occur per year. Paroxysms, the most powerful explosive eruptions, are less frequent, and usually last from several hours to a few days and produce fallout of heavy material up to several kilometers from the craters, sometimes affecting the two villages on the coast (Barberi et al. 1993; Bertagnini et al. 1999). Lava effusions occur about every 10–20 years.

Crystal-rich products are emitted, either as scoriae or lava flows, during the normal activity and effusive episodes, whereas nearly aphyric, light, golden pumices are only emitted during paroxysms and most of the major explosions, together with scoriae with the same characteristics of those erupted during the normal activity (Bertagnini et al. 1999; Rosi et al. 2000). Evidence of sin-eruptive mingling between pumice and scoriae is ubiquitous. This is supported by the occurrence of clasts made up of pumice and scoriae with lobate, sharp contacts and, of pumice containing a fairly high content of crystals (up to 20–30 vol%), most of which are wetted by brown glass with the composition of the glassy matrix of the scoriae. The nearly aphyric pumice and the dense scoriae, rich in plurimillimetric crystals, have a bulk rock composition ranging from HK-basalt, at the limit with shoshonitic basalt, to HK-basaltic andesite (Rosi et al. 2000; Métrich et al. 2001). Such contrasting texture, in apparent disagreement with the limited chemical variability of the bulk rocks, reflects a magmatic system mainly controlled by water content and its kinetics of exsolution (Métrich et al. 2001). The dense scoriae represent a crystal-rich magma that resides in the upper part of the volcano and is derived from the volatile-rich magma via crystallization mainly driven by decompression and volatile loss at low pressure (Métrich et al. 2001). This shallow magma body that feeds persistent strombolian activity represents a rarely documented case where highly viscous, crystallized magma is refilled by low viscous, nearly-aphyric, volatile-rich melt blobs associated with only slight bulk chemical variation. Moreover, it is percolated by a high flux of gas and has maintained the same character for 1,400–1,800 years (Rosi et al. 2000). The main question addressed here is how the crystal-rich body works in order to maintain its persistent homogeneity.

It is recognized that both chemistry and texture of plagioclase vary in response to intensive parameters (temperature, pressure, melt composition, H2O...) and primary zoning is preserved owing to the very slow interdiffusion rate within the crystal (Anderson 1984; Grove et al. 1984; Tsuchiyama 1985; Rutherford and Devine 1988; Nelson and Montana 1992; Hammer and Rutherford 2002). As a result, plagioclase has been largely employed in the literature for the reconstruction of the magma history (Kuo and Kirkpatrick 1982; Singer et al. 1995; Hattori and Sato 1996; Tepley et al. 1999; Tepley et al. 2000; Stewart and Fowler 2001).

We have undertaken a systematic study of plagioclase phenocrysts from scoriae and lavas plus plagioclase xenocrysts from pumice emitted at Stromboli during different episodes and types of activity over the period 1985–2000. Plagioclases provide the opportunity to assess the effect of water and its degassing on the crystallization process during dynamic mixing between magmas mainly differing by their volatile contents. Crystal stratigraphy of plagioclase reveals repetitive core-to-rim zoning with large variations in An content and obvious textural discontinuities. Using these compositional and textural features, together with the chemistry of both whole rocks and matrix glasses, we present a detailed picture of the mechanisms of mixing, degassing and crystallization at the micrometer scale and discuss their inference on the magmatic system as a whole.

Petrochemical background

Pumice and scoriae erupted since the persistent activity was established, 1,400–1,800 years ago, range between HK-basalts/shoshonitic-basalts and HK-basaltic andesites (SiO2~49–53 wt%; K2O~1.6–2.2 wt%) and, as a whole, have maintained similar textural and mineralogical characteristics (Rosi et al. 2000; Métrich et al. 2001). Pumice range from 1.62 to 2.07 K2O wt% (CaO/Al2O3=0.52–0.69; Rb=52–73 ppm; La=39–48; Th=11.1–18.2), are nearly aphyric with microphenocrysts of clinopyroxene (Fs5–8 Wo 45–48) and olivine (Fo82–91) in HK-basaltic/shoshonitic glassy matrix (CaO/Al2O3 =0.52–0.64). Olivine-hosted melt inclusions yield high volatile content with 1.8–3.4 wt% H2O, 894–1,689 ppm CO2, 1,660–2,250 ppm S and 1,660–2,030 ppm Cl (Métrich et al. 2001; Bertagnini et al. 2003).

Black dense scoriae cover a narrow compositional range from 1.92 to 2.07 K2O wt% (CaO/Al2O3=0.59–0.62; Rb=65–68 ppm; La=44–46; Th=14.3–17.2). They contain nearly 50 vol% crystals of plagioclase An62–90, the most abundant mineral phase, clinopyroxene Fs6–14 Wo42–47 and olivine Fo70–75 in glassy to hypocrystalline matrices with shoshonitic composition (CaO/Al2O3=0.45–0.50). Melt inclusions in olivine recorded low H2O content (0.05–0.6 wt%), CO2<100 ppm, but some variability with respect to major elements (CaO/Al2O3 =0.6–0.4), S (<dl-1,300 ppm) and Cl (1,000–2,600 ppm) (Métrich et al. 2001). Temperature of scoria has been assessed at 1,115±10 °C compared to 1,145±10 °C for pumice, on the basis of optical thermometry on MI (Clocchiatti 1981; Métrich et al. 2001). Finally, a mixing process was proposed on the basis of isotopic data (Francalanci et al. 1999) and crystal zoning of megacrysts of clinopyroxene (Clocchiatti 1981).

Sample selection

In this study, we focus on products erupted over the period 1985–2000. Samples were collected to be representative of the various types of activity and different parts of the crystal-rich magma body. During this span of time, the strombolian activity at the summit crater threw out black scoriae every 10–20 min and a total of at least 23 major explosions were documented (Coltelli, personal communication). The latter emitted a volume of golden pumice, on the order of 103 - 104 m3 (Bertagnini et al. 1999), associated with minor scoriae. The major lava flow episode between December 1985 and April 1986 produced 5–6x106 m3 of lava (De Fino et al. 1988). After this effusive event, the volcano maintained its persistent strombolian activity.

The selected samples consist of (1) 26 crystal-rich scoriae emitted during the normal strombolian activity and major explosions, (2) 3 samples of lava flows and pyroclastics erupted in the initial and final phases of the 1985–1986 lava flow, and (3) fragments of crystal-rich material and plagioclase xenocrysts embedded in three mingled pumice emitted during the 1998 and 1999 major explosions.

Analytical techniques

Major elements in whole rocks were analyzed by wavelength dispersive X-Ray fluorescence, except for MgO, Na2O, determined by atomic absorption spectrometry, and FeO, by titration at Dpt. di Scienze della Terra, Pisa University. The routine precision is 0.5% for MgO, 1% for SiO2 and Na2O, 3% for Al2O3, FeO, Fe2O3, CaO and K2O, 8% for P2O5 and 15% for MnO. Major elements of mechanically separated minerals were analyzed by Emission-ICP at Centre de Recherches Pétrographiques et Géochimiques (Nancy-France). Trace elements in bulk rocks were analyzed by ICP-MS at CRPG and Dpt. di Scienze della Terra, Pisa University. The differences between the analyses performed at the two laboratories are within the analytical error. ICP routine precision on trace elements is <5%, apart from Co, Cs, Cu, Ge, Ho, Lu, Pb, Sm, Tm, W (<10%) and As, Be (<20%).

Modal analyses were performed using an optical microscope equipped with a point counter. In each sample 600–700 points, excluding bubbles, where counted. Major element analyses in minerals and glassy matrices together with backscattered electron images of plagioclases were obtained using Energy-Dispersive X-ray Analysis using EDAX X-4I on a Philips XL30 scanning electron microscope (Dpt. Scienze della Terra, Pisa). The analytical error is 1% for concentrations higher than 15 wt%, 2% for 5–15 wt%, 5% for 1–5 wt%, and 30% for <1 wt%. The chemical traverses in plagioclase were performed using a Cameca SX50 electron microprobe at Istituto di Geologia Ambientale e Geoingegneria – CNR Roma with a 5 μm defocused probe, and 10-μm step intervals. The relative error on plagioclase composition is 0.5 mole% An. The S and Cl content in melt inclusions and glassy matrices were determined using the SX50 (CAMPARIS, Paris 6, France) with experimental conditions described in Métrich et al. (2001).

Petrography and geochemistry

Major elements were analyzed on 30 samples of scoriae and lavas, 16 of which were selected for additional trace elements analyses as reported in Table 1. Scoriae and lavas are HK-basalts/basaltic shoshonites with SiO2 clustering around 49.2–50.9 wt% and K2O 1.96–2.18 wt%.

Table 1 Whole rock composition of selected crystal-rich scoriae and lavas erupted over the period 1985–2000. For comparison, whole rock analyses of one pumice sample is reported
Full size table

Trace elements indicate no systematic variations between scoriae emitted during normal activity and major explosions, but a slight chemical evolution with decreasing of such elements as Rb and Ba through the time (Fig. 1). They also show that the primitive magmas that refilled the shallow system and erupted as pumice in 1998 and 1999 have lower incompatible element (particularly Th) content and could be less evolved. Following this line of reasoning the scoriae would derive from the pumice by crystal fractionation. However, the range of variation in incompatible elements (e.g. Th from 11 to 14 ppm) cannot be explained only by this process since the compatible elements such as Sr, Co, Cr, Ni do not vary extensively (Table 1). Therefore, crystal fractionation could be a minor process considering that an open system periodically refilled by primitive magma and continuously tapped tend to be enriched in incompatible with respect to compatible elements (O’Hara 1977). Alternatively, the pumice erupted in 1998 and 1999 would represent magma batches slightly different by their content in alkalis (and Ba, Sr) with respect to scoriae. Indeed, the chemical variability of the primary melts that have sustained the persistent activity since its beginning has been demonstrated on the basis of melt inclusions (K2O varying from 1.3 to 1.9 wt%, Bertagnini et al. 2003). This hypothesis is in agreement with input of fresh magma batches with different Sr isotopic signature (Francalanci et al. 1999).

Fig. 1
figure 1

Variation of selected trace elements with time, of scoriae and lava samples erupted over the period 1985–2000. Filled circle: lava; filled triangle: scoriae from normal activity; filled square: scoriae from major explosion. As comparison the composition of two pumice is also plotted (open square)

Full size image

Modal analyses were carried out on 12 samples of scoriae emitted during strombolian activity and major explosions. They yield 47–55 vol% of euhedral phenocrysts (Table 2). Plagioclase is the dominant phase (57–71 vol%), associated with clinopyroxene (20–34 vol%) and olivine (4–12 vol%). Microprobe analyses were performed on each mineral phase and reported in histograms (Fig. 2). All samples have seriate texture with plagioclase ranging from 150 μm to 2.5 mm in size and clinopyroxene and olivine from 150 μm up to 3–4 mm. Plagioclase, shows oscillatory zoning, and covers a large range An62-An88 regardless of the crystal size, with a well defined mode at An68. Clinopyroxene varies from Fs5 to Fs17, the less evolved composition Fs6-Fs10 occurring as small cores and/or thin bands. Olivine is fairly homogeneous from Fo70 to Fo74. Compositions of the crystal rims in equilibrium with the groundmass are An63-An68, Fs12-Fs14 and Fo71-Fo73. Phenocrysts from one lava sample (ST85) and one scoria (ST56) were also mechanically separated and analyzed as bulk crystals (Table 3). In both samples plagioclase, clinopyroxene and olivine have the same compositions (An68.7-An69.5, Fs13.3-Fs14.1, Fo71.2-Fo70.8). Comparison between the bulk and microprobe analyses indicates that compositions not in equilibrium with the groundmass are subordinate.

Table 2 Modal analyses of selected crystal-rich samples
Full size table
Fig. 2
figure 2

Plagioclase, clinopyroxene and olivine composition as An mol%, Fs mol% and Fo mol% respectively. Analyses were performed on lava samples (ST85-Dec 85, ST86–6-Apr 86), scoriae erupted during normal activity (ST44-Jul 94; ST56-Sept 94, ST212–18 Jul 00) and major explosions (ST155–8 Sept 98; ST160–24 Nov 98, ST177–26 Aug 99; ST182–26 Aug 99), and fragments of crystal-rich scoriae included in pumice (ST130–23 Aug 98; ST168–24 Nov-98; ST178–26 Aug 99). -a: analyses of plagioclase crystals <1 mm; -b: histogram of An content of plagioclase drafted with 1,300 analyses obtained from rim-to–core compositional traverses on 25 selected phenocrysts (1–2 mm)

Full size image
Table 3 Major elements ICP analyses of mechanically separated crystals of plagioclase, clinopyroxene and olivine
Full size table

Glassy matrices free of microlites are also analyzed in nine samples, two of them are from scoria fragments in pumice erupted during two different major explosions, two from 1985–1986 effusive activity, two from scoriae of normal activity and three from scoriae erupted during major explosions. They yield shoshonitic compositions clustering around SiO2=52.9–53.6 wt%, K2O=4.1–4.6 wt% and CaO/Al2O3=0.44–0.48 (Table 4). Based on least-square mass balance calculations, glassy matrices derived from whole-rocks by subtracting of 52–56 wt% solid, made of 59–61 wt% PL, 22–23 wt% CPX and 17–18 wt% OL. These results match the average modal distribution, after converting volume in weight percentage using mean crystal density, which points out 54±2 wt% crystal content (with 59±4 wt% PL, 29±5 wt% CPX, 12±2 wt% OL). As a whole these results suggest that the liquid/solid ratio and the relative proportions of minerals have not significantly changed in the scoriae and lava produced over the period 1985–2000.

Table 4 Composition of glassy matrices of selected crystal-rich scoriae (a), inclusions of crystal-rich magma in pumice (b) and pumice (c)
Full size table

A comparison with data obtained from products erupted during older activity, as presented in the “Petrochemical background” section, indicates that the black dense scoriae have maintained a crystal contents ~50 vol%, associated with poorly variable shoshonitic groundmasses, and virtually unchanged mineral chemistry since the beginning of the persistent activity. The variability of the bulk-rock compositions, as displayed by different contents of some trace elements (e.g. Th~14.1–17.1), possibly mirrors geochemical changes of the deep magma which feeds the crystal-rich, shallow body.

Plagioclase composition and morphology

In every sample, plagioclase crystals are commonly euhedral and tabular with dimensions up to 2.5×0.5 mm in sections ⊥(010). After crushing and sieving, individual crystals were hand picked from the grain size 1–2 mm. About 500 crystals from scoriae and 200 xenocrysts from pumice were embedded in epoxy resin, cut //(010) or ⊥(010) through the approximate geometric center and polished. Because of the difficulty in extracting non-fractured crystals from lava samples, plagioclases of the 1985–1986 lava flow were studied in thin sections. Crystals were observed using backscattered electron images and analyzed by EDS on a SEM. A total of 25 crystals were selected for compositional rim-to-core profiles from which about 1,300 major element analyses of plagioclase were obtained.

Plagioclase from scoriae. Almost all plagioclase crystals are complexly zoned. In most cases they consist of alternating, concentric layers of bytownitic (An70–88) and labradoritic (An62–70) plagioclase (Fig. 3a), which display highly variable thickness from <10 to 100 μm. The labradoritic layers are free of glass inclusions and exhibit ubiquitous, small-scale (1–5 μm) oscillatory zoning with limited compositional variations 3–4 An% (Fig. 3b). The bytownitic layers are patchy zoned and show sieve texture with abundant micrometric glass inclusions and voids, particularly well illustrated in the broad bytownitic bands whose width achieves 100 μm (Fig. 3b, c). The outer boundaries of bytownitic layers are usually crenulate surfaces, roughly parallel to the small-scale oscillatory zoning of the labradoritic bands. These bytownitic layers overgrow on dissolution surfaces characterized by angular discordance and embayments cross-cutting the labradoritic, oscillatory zoned layers (Fig. 3c, d). Large glass pockets are commonly present at the contact to the dissolution surfaces. They are either elongated against the inner margin of the sieve-textured zones or rectangular, cutting the oscillatory-zoned layers (Fig. 3a, b). In some cases, dissolution phenomena have partially or completely corroded the labradoritic layers. As a result, some crystals show an inner region dominated by bytownitic composition, associated with relicts of labradoritic plagioclase and abundant, elongated glass pockets, both arranged to form concentric layers broadly parallel to the growth surface of the crystals (Fig. 3e). Many crystals show a bytownitic core hosting scattered, up to 150 μm, irregularly shaped, glass pockets (Fig. 3f).

Fig. 3a–f
figure 3

Back-scattered electron images of plagioclase phenocrysts. Crystals are cut nearly parallel to (010) through the approximate geometric center. a) Zoned phenocryst consisting in alternating labradoritic (light gray) and bytownitic (dark gray) concentric layers. Large glass pockets are commonly present at the inner boundaries of the bytownitic layers; b) the labradoritic layers are characterized by small-scale, oscillatory zoned texture, the bytownitic by patchy-zoned, sieve textures, including micrometric glassy inclusions and voids; c) thin bytownitic layers and their compositions are shown. Thin layers have composition An74–79, whereas An contents >80 are typical of thick layers (>30 μm); d) dissolution surfaces at the inner boundary of the bytownitic layers are marked by angular discordances and gulfs cross-cutting the labradoritic, oscillatory-zoned layers. The outer boundary are crenulate surfaces (the crystal rim is on the top) (see also b and c); e) phenocryst showing an inner part with dominant bytownitic composition and remnants of labradoritic zones resulting from partial dissolution; f) coarse sieve-textured, patchy-zoned core rimmed by relicts of labradoritic layers

Full size image

Compositional profiles show patterns which are commonly composed of zoned Ab62–70 base-line, corresponding to small-scale oscillatory zoned layers, interrupted by An70–88 spikes whose size varies from several tenths of a μm to a few μm (Fig. 4a, b, c). The thin bands have, on average, a composition of An74–79, whereas layers with thickness >10 μm are usually more calcic (Fig. 3c). Most of the compositions An71–73 are found on the boundary between two layers and we cannot exclude that they represent average compositions. Frequency and amplitude of the zoning are highly variable in each crystal. Based on backscattered electron images, a maximum of 5–6 bytownitic layers >30 μm are counted in a single crystal. Compositional discontinuities associated with thin (usually <10 μm) bytownitic bands, are more difficult to document, but they are abundant, at least 10–15 in a single crystal. These are minimum values, however, since the dissolution episodes of labradoritic layers have often removed an important part of the growth history of the crystals. Accordingly, the most complex textures associated with intensive dissolution episodes, result in very irregular spike patterns ranging from An65 to An85, but with only scarce labradoritic zones (Fig. 4d, e). The patchy zoned cores with irregular large inclusions give an uneven pattern covering the whole range of compositions (Fig. 4d, f). The whole spectrum of variations is also observed in crystals from a single sample of scoria. Finally, the rims of phenocrysts are 20 to 100 μm thick. As a whole, they have the same composition and texture of the labradoritic layers.

Fig. 4a–f
figure 4

Representative compositional traverses from the rim to the approximate core of plagioclase phenocrysts cut // (010) (b, d, e, f) and ⊥(010) (a, c)

Full size image

Plagioclase xenocrysts from pumice. Euhedral xenocrysts of plagioclase, inherited from crystal-rich magma are commonly present. They can be wetted either by the shoshonitic matrix glass of the crystal-rich scoriae (CaO/Al2O3=0.45–0.50 and K2O=4.1–4.6 wt%) or HK-basaltic glassy matrix of the pumice (CaO/Al2O3=0.63–0.64; K2O=2.2–2.3 wt%). Figure 5a provides a typical example of the plagioclase xenocrysts wetted by pumice glass. They display the same compositional and textural zoning observed in the phenocrysts from scoriae and are surrounded by a reaction and growth corona. The corona is composed of patchy-zoned, skeletal to sieve-textured plagioclase, rich in voids and melt inclusions commonly <10 μm. Voids can be either randomly distributed or settled along the inner part of the corona. They are rimmed by glass that reflects entrapment of gas bubbles most likely generated during melt degassing that accompanied the skeletal growth of the corona (Fig. 5b, c). The enlarged SEM image demonstrates a continuum from fine-sieve to evident skeletal texture from the inner to the outer part of the corona associated with normal compositional zoning from An90 to An75. The sieve textures derive from initially skeletal textures developed during rapid crystal growth as will be discussed further (Fig. 5b, c).

Fig. 5a–d
figure 5

Back-scattered electron images of plagioclase xenocrysts in pumice. a) Typical crystal mantled by sieve textured rim ≈100 μm large. The glass adhering to the crystal (ST130 pl4) has the composition of the glassy matrix of the hosted pumice (CaO/Al2O3=0.64; K2O=2.2 wt%); b) close up of the ≈100 μm corona (ST130 pl10). The characteristic skeletal texture with abundant micrometric glass inclusions and voids is shown. Voids are also settled along the inner boundary. Corona is directed zoned from An80–90 to An75–80; c) dissolution surface, cutting the oscillatory zoning layers, at the boundary between the internal part of the crystal and the corona. Skeletal texture evolves and produces sieve texture; d) the labradoritic layer is completely resorbed in the central part of the figure resulting in the superposition of two bytownitic layers. On the rim a microphenocryst of olivine Fo72 with a thin rim (≈10 μm) Fo84 is present. The composition Fo84 is in equilibrium with the glassy matrix of the pumice

Full size image

The An-rich coronas grew on resorption surfaces as shown by evident angular discordance cutting the pre-existing labradoritic layer (Fig. 5c). Dissolution processes can be strongly efficient and cause the total removal of the labradoritic zones to build up crystals with successive sieve textured, bytownitic bands (Fig. 5d).

Chemistry of glasses trapped in plagioclase

Glassy inclusions trapped in bytownitic, sieve-textured layers were not analyzed, because their dimensions are usually smaller than 10 μm. The large glass pockets, preferentially located on discontinuity surfaces, are comparable in composition to the totally degassed, residual glass matrix of scoria. They contain 4–5.5 wt% K2O (K2O/Na2O=1.2–1.7); 1,000–1,300 ppm Cl; 90–530 ppm S contents (Fig. 6). The most evolved compositions are rich in K2O (up to 5.5 wt%), possibly due to crystallization of the host mineral. It seems unlikely that the large glass pockets represent remnants of residual glasses preserved through different dissolution-crystallization events without any evidence of post-trapping evolution, either dissolution or crystallization. Their unaffected residual composition suggests that these large glass pockets were derived from late infiltration of melts through fractures, which refilled dissolution-related cavities.

Fig. 6
figure 6

Plots showing Na2O/K2O, sulfur and chlorine variations in the large glass pockets hosted in plagioclase. Filled circle: plagioclase phenocrysts of crystal-rich scoriae and xenocrysts of pumice. Open circle: glassy matrix. S and Cl were measured in two selected crystals (square: ST182 pl2; diamonds: ST160 pl5). The line at S=90 ppm indicates the detection limit

Full size image

Discussion

Mechanisms inducing textural and chemical zoning in plagioclase

The small scale oscillatory zoning with amplitudes less than 3–4% An, in the labradoritic layers, are commonly attributed to a local kinetic effect at the crystal-melt boundary layer, that does not imply variations in the chemico-physical conditions of the whole system, but indicates conditions close to equilibrium crystallization (Allegre et al. 1981; Ortoleva 1990; L’Heureux 1993; Holten et al. 2000). Conversely abrupt compositional and textural changes in plagioclase phenocrysts are usually ascribed to mixing processes in the magma chamber, associated with recharge of the system during which changes in temperature and major element contents play the main role (Singer et al. 1995; Hattori and Sato 1996; Tepley et al. 1999; Tepley et al. 2000; Stewart and Fowler 2001). However, variations of H2O content, due to either influx of external fluids or a volatile gradient in the magma chamber, also affect the zoning of the plagioclase (Hattori and Sato 1996; Ginibre et al. 2002). Discriminating the respective effect of the different parameters is delicate. At Stromboli, mixing occurs in a volcanic conduit between deep, volatile-rich, HK-basaltic melts and residing degassed magma with 45–50 wt% of shoshonitic liquid, within a range of temperature from 1,145 °C to 1,115 °C.

We performed MELTS computations (Ghiorso and Sack 1995), in order to delineate the stability field of the plagioclase in a HK-basaltic melt akin to bulk compositions of the pumice, at 1,145 °C, with H2O contents varying from 3 to 0.2 wt%, and pressure ≤1 kbar (Fig. 7). We started with an average H2O content which represents that of the HK-basaltic primitive melts at the origin of the pumice, before major degassing process (Bertagnini et al. 2003). For H2O content >1 wt%, the phases in equilibrium with the melt are olivine (Mg# 83–84) and clinopyroxene (Mg# 0.85–0.86), at 1,145 °C, in agreement with the paragenesis of the pumice (Métrich et al. 2001; Bertagnini et al. 2003). The mass of the solid which crystallizes, does not exceed 8 wt.%. Plagioclase An83-An85 becomes stable when the H2O concentration of the melt achieves ~1 wt% (PH2O <Ptotal), with pressure varying from 0.1 to 1 kbar. For H2O decreasing from 1 to 0.2 wt%, the mass of solid ranges from 10 to 40 wt%, and the plagioclase in equilibrium is in the range An85 and An73(Fig. 7). In an anhydrous HK-basaltic system a low crystal content, akin to that of the pumice (~5 wt%), can be reproduced only at T≥1,180 °C. Even at this high temperature, the plagioclase in equilibrium with the anhydrous HK-basalt is <An80. Plagioclase An86 may also be obtained in equilibrium with a HK-basaltic melt containing 1.5 wt% H2O at 1 kbar PH2O, but for a lower temperature of 1,115 °C. These calculations thus indicate that any plagioclase at contact with a melt whose H2O concentrations exceed 1 and 1.5 wt%, will be dissolved at 1,145 and 1,115 °C, respectively. It places the upper and lower limits on the water concentrations of the HK-melt, which may generate An-rich plagioclase. MELTS computations also suggest that plagioclase An>75 cannot be obtained in anhydrous HK-basalt within the range of the measured temperatures.

Fig. 7
figure 7

Stability field of plagioclase in a HK-basaltic melt, within the range H2O 0.2–3 wt% and pressure between 0.1 and 1 kbars, according to MELTS calculations (Ghiorso and Sack 1995). Water saturation boundaries according to MELTS calculations (solid line) and to Papale (1997) (dotted line) are shown

Full size image

A key to understanding the mechanism of water degassing and its effect on plagioclase, can be retrieved from the texture and composition of the corona of the plagioclase xenocrysts in the pumice. Based on dissolution phenomena affecting the pre-existing labradoritic plagioclase and the high concentration of voids (gas bubbles) in the coronas, we propose a mechanism involving dissolution of the pre-existing labradoritic plagioclase at contact with a gas (H2O-CO2) - melt emulsion and fast crystallization of An90–85 plagioclase (Fig. 5b). The composition An>85 which begins to grow on the dissolution surface is not reproduced by computations (Fig. 7). It is possibly related to crystal-melt boundary kinetics involving dissolution, discharge of Al2O3 and nucleation of highly calcic plagioclase. As degassing proceeds, the liquidus moves towards higher temperatures and plagioclase progressively changed its composition down to An80–75 at the contact with the strongly degassed matrix glass of the pumice (Fig. 5b).

One important point that arises is that the water exsolution should be highly dynamic and the dissolution-crystallization processes could occur in a system over-saturated with respect to water. Dynamic and fast H2O exsolution accounts for the development of skeletal textures of the plagioclase during undercooling. These skeletal textures evolve and produce fine sieve-textured plagioclase rich in voids as occurs in the innermost part of the corona (Fig. 5c). Many authors have interpreted fine sieve-textures in plagioclase as the result of dissolution episodes (Kawamoto 1992; Nelson and Montana 1992; Castro 2001). Conversely, at Stromboli, the bytownitic coronas demonstrate without ambiguity that their fine sieve-textures have originated by skeletal growth during gas exsolution as envisaged by Anderson (1984) for plagioclase at Fuego volcano, although the effect of gradients in temperature cannot be totally ruled out.

In the light of the above observations, we propose that the thick, sieve-textured bytownitic layers growing on dissolution surfaces in phenocrysts from the crystal-rich magma involve successive dissolution-crystallization processes during repetitive inputs of non-erupted, volatile-rich HK-basaltic melt blob and rapid degassing. The case of major explosions, during which volatile-rich magma is emitted as pumice, most likely documents the most extreme conditions.

The recurrent thin oscillations with intermediate composition (on average An75) would be mainly controlled by water exsolution and crystallization of HK-basaltic melts that progressively tend to reach the composition and the temperature of the most residual evolved shoshonitic melt. Alternatively, these intermediate plagioclases may result from the degassing and crystallization of a hybrid melt involving mixing between the HK-basaltic melt and the residual shoshonitic melt of the crystal-rich magma, in various proportions. However, as described above, the dominant and representative plagioclase of scoriae is An68, in equilibrium with olivine Fo71–73, clinopyroxene Fs12–14 and the shoshonitic residual melt (Fig. 2). The large dominance of this low temperature paragenesis in equilibrium with H2O-poor shoshonitic residual melt, likely mirrors a low volumeinput magma / volumeresident magma ratio during a mixing event. Actually, the volume of volatile-rich melt blobs erupted as pumice during the major explosions is on the order of 103 m3 (Bertagnini et al. 1999), and subordinate compared to the estimates for the crystal-rich body (4×107 to 3×108 m3; Francalanci et al. 1999). Intrusion of non-erupted volatile-rich magma likely involves smaller volumes.

Time scale of the breakdown events

Abrupt compositional changes, observed at the boundary bytownitic/labradoritic layer, suggest that after crystallization of bytownitic plagioclase, the system rapidly shifts back to the chemical and physical equilibria pertaining to the crystal-rich magma. As a result, the duration of the destabilization event can be evaluated by estimating the growth time of a bytownitic layer 30–100 µm thick. A large range of growth rates of plagioclase is proposed in basaltic systems. In Hawaiian basalts, they were calculated, by CSD method, to be 11−10–10−11 cm/s (Cashman and Marsh 1988; Cashman 1993). According to Cashman (1993), undercooling increases the growth rates up to 10−8 cm/s. Faster growth rates, about 10−7–10−9 cm/s, have been determined for microphenocryst crystallization of a Mauna Loa basalt lava flow (Crisp et al. 1994). Jambon et al. (1992) calculated, in tholeiitic basalts, values from 4×10−8 to 8.7×10−6 cm/s from undercooling ranging from 15 to 150 °C. The skeletal texture of the plagioclase would suggest an undercooling >100 °C, assuming that we can extrapolate the experimental data on plagioclase carried out at 5 kbars by Lofgren (1974). According to these data the time for crystallizing a sieve-textured bytownitic layer 100 μm thick (Fig. 3b), is ~1 h. The growth rate could be much higher during rapid decompression and degassing events as illustrated by the corona of plagioclase xenocrysts in pumice, resulting in a high density of bubbles randomly distributed in the bytownitic layer (e.g. Fig. 5c).

Magma mixing processes in the crystal-rich body

Texture and zoning of the plagioclase indicates that the mixing processes play a fundamental role on the evolution of the whole system. The sharp boundary between sieve-textured bytownitic layers and labradoritic zones, and rather short span of time of the destabilization events would imply efficient mixing during the refilling episodes of the shallow body. We cannot exclude that crystallization is enhanced on the walls of the conduits because of temperature gradient. According to Chouet et al. (2003), the shallowest part of the conduit would be a dike-like system, ~60° dipping to the NE and NE-SW striking, along dominant structural directions. The existence of a superficial hydrothermal system has been also proposed as the result of infiltration of meteoric water interacting with the volcanic heat sources (Finizola et al. 2002). Such a situation strongly favors crystallization along the conduit margins. In this hypothesis, a part of the crystals wrapping the dike boundaries could be involved in the mixing processes and would explain the extensively evolved paragenesis (plagioclase <An68, olivine <Fo70, clinopyroxene >Fs15), as previously proposed by Métrich et al. (2001).

We address here the question of what mechanisms enable an efficient mixing process to occur in a mushy, highly viscous, shallow magma body. The magmas which mix together have contrasting viscosity and different density, in response to their distinct crystal and volatile contents, of 20–30 Pa s and 2,500 kg/m3 for the volatile-rich magma, and >104 Pa s and 2,700 kg/m3 for the degassed crystal-rich one (following Marsh 1989 and Lange 1994).

We considered the model proposed by Huppert et al. (1986) where a fluid system involving a dense magma replenished by an influx of light magma is described by Reynolds numbers

$$ \begin{array}{*{20}l} {{\operatorname{Re} _{{\text{i}}} = ({\text{g}}'{\text{Q}}^{3} /\nu ^{5}_{{\text{i}}} )^{{1/5}} \;{\text{for}}\;{\text{input}}\;{\text{magma}}\;({\text{i}})} \hfill} \\ {{\operatorname{Re} _{{\text{r}}} = ({\text{g}}'{\text{Q}}^{3} /\nu ^{5}_{{\text{r}}} )^{{1/5}} \;{\text{for}}\;{\text{resident}}\;{\text{magma}}\;({\text{r}})} \hfill} \\ \end{array} $$
(1)

where Q is the volume flow rate, νi is the kinematic viscosity of the input magma, νr is the kinematic viscosity of the resident magma and g’ = g (ρi−ρr )/ρr is the reduced gravity where g is the gravity acceleration and ρi and ρr are the density of input and resident magmas respectively.

The estimates of the average magma flux (Q), presently available at Stromboli, are deduced from gas (Allard et al. 1994) and heat (Giberti et al. 1992; Harris and Stevenson 1997) fluxes, between 0.01 and 0.004 km3/yr. This range of Q values provides very low Reynolds numbers (Re i<40 and Re r<<1) which infer a varicose flux, at the limit with laminar regime, and very little, if any, mixing. However, these calculations are based on an average Q value, most likely smaller than input influx due to discrete events. We attempted to estimate the flow rate assuming discrete magma batch input by considering the magma volume produced as pumice during recent major explosions (~103 m3) and the duration of the event (as short as ~30 seconds (Bertagnini et al. 1999). They yield Q~30 m3/s and Reynolds numbers Re input =637 and Re resident =2 which involve a form of motion at the limit of varicose instability and unsteady flux with possible incorporation of resident magma into the input plume. It represents the limit conditions implying the emission of very light bubble-rich magma blobs due to rapid decompression and almost instantaneous volatile release and gas expansion at low pressure. Change in the density of the input magma, resulting from volatile exsolution, does not imply significant variations of the Reynolds number, whereas a possible increasing of the viscosity of the input magma associated with bubbles formation moves the system towards the laminar regime. We cannot exclude a change in the flux regime from laminar towards unsteady during the volatile rich magma ascent through the shallow magma body. All together these results suggest that influx of volatile-rich magma blobs into the shallow and denser system infers a laminar or varicose regime without significant mixing between the two magmas. Such a conclusion is inconsistent with (1) the emission of extensively mingled pumice during major explosions that requires unsteady to turbulent plume, at least in the uppermost part of the conduit, and (2) the chemical and mineralogical characteristics of the crystal-rich magma that imply efficient mixing. Disagreement between calculations and observations could be explained by the fact that the resident fluid was static in the experiments of Huppert et al. (1986), as noted by the authors themselves. Indeed, Stromboli is a steady state volcano producing an average gas flux of 6–12 103 tons/day associated with a low output rate of magma. As reviewed by Ripepe et al. (2002), normal activity at Stromboli has been related to ascending gas slugs in the shallowest part of the conduit related to unstable gas foam formed by collection of rising bubbles. Such a process was ascribed either to the presence of physical boundaries such as the chamber roof (Jaupart and Vergniolle 1989), a constriction at the dyke to conduit transition (Ripepe et al. 2001), the upper wall of an inclined dike-like conduit (Chouet et al. 2003), a change in the magma viscosity (Thomas et al. 1993), or to bubble coalescence in slowly rising magma (Wilson and Head 1981; Parfitt and Wilson 1995). Ripepe et al. (2002) also proposed a model where magma supply and bubble concentration are variable with degassing cycles through periods of high and low magma supply rate. As a result, the shallow magma body is continuously percolated by various proportions of gas bubbles that could drastically affect the physical parameters and the rheological behaviour of the magmatic system. For instance, experimental measurements suggest that the presence of bubbles can decrease the fluid viscosity (Lejeune et al. 1999; Bagdassarov and Dingwell 1992), leading to higher Reynolds numbers and promoting turbulent flow regimes. Moreover, the rising of a low-viscosity, gas-rich melt plume in higher viscosity magma induces sinking of the resident material and overturning (Kazahaya et al. 1994; Stevenson and Blake 1998). Convection can therefore disrupt the laminar flow and cause efficient mixing of the two fluids. In summary, discrete magma input events, recurrent bubble accumulation and foam disruption at shallow levels and influx of gas bubbles, possibly generated during slow ascent of deeper and more primitive melts, enhance convection at different scales in space and time. Such considerations reconcile the intense, and recurrent oscillatory zoning of plagioclase (and other minerals) and the petrochemical homogeneity of the crystal rich magma (bulk scoriae, and lava) produced by a steady state, open system basaltic volcano.

Finally, we have attempted to place a limit on the vertical extension of the crystal-rich body, in which mixing and convection are effective. Based on the plagioclase stability field, labradoritic plagioclase is in equilibrium with a shoshonitic strongly degassed melt (H2O<0.6 wt%), whereas bytownitic plagioclase is computed to be stable in HK-basaltic magmas containing ≤1–1.5 wt% H2O at 1,145–1,115 °C, which imply, for the highest H2O content, PH2O ~240 bars, assuming a CO2-free system (Papale 1997 and MELTS calculations). By assuming PH2O =Plithotastic we obtain a depth around 700–800 m. Harris and Stevenson (1997) previously proposed that the shallow magma at Stromboli resides at a depth <1 km, and rather limited volume and vertical extension have also been suggested, on the basis of thermal flux (Giberti et al. 1992).

Conclusions

Data presented in this paper and in previous works entail that the shallow crystal rich magma body has preserved the same general arrangement over a period of 1,400–1,800 years. Textural and compositional variations in the plagioclase phenocrysts indicate, however, that equilibrium is maintained through frequent changes in the physical-chemical factors pertaining to the magma body, which destabilize the labradoritic plagioclase, in equilibrium with the residual melt of the crystal-rich magma, and stabilize bytownitic compositions. Skeletal and sieve texture with abundant voids of bytownitic layers clearly indicate that they grow in undercooling conditions due to fast degassing. Accordingly, textural and chemical zoning of plagioclase is due to repetitive dissolution and crystallization events related to successive input of small amounts of volatile-rich melt.

The intense, and recurrent oscillatory zoning of plagioclase (and other minerals), the predominance of plagioclase An68 together with the overall petrochemical homogeneity of the shallow crystal-rich body suggest mixing events and convection, possibly at different space and time scales, in relation to discrete magma input events, recurrent bubble accumulation and foam disruption at shallow levels and input of gas bubbles possibly generated during episodes of slow ascent of deeper and more primitive melts. These conclusions drawn from mineralogical and chemical studies of the scoriae (and some pumice) at Stromboli, a steady state basaltic volcano, most likely apply to many volcanoes whose shallow systems are periodically refilled by fresh, volatile-rich deeper magmas.