1891 AD Submarine Eruptive Processes and Geochemical Studies of Floating Scoria at Foerstner Volcano, Pantelleria

On October 17, 1891 a submarine eruption occurred at Foerstner volcano in the Straits of Sicily 4 km northwest of the island of Pantelleria, Italy. The eruption produced floating scoria bombs, or balloons, that discharged gas at the surface and eventually sank to the seafloor. Activity occurred for a period of one week from an eruptive vent located within the Pantelleria Rift at a water depth of 250 m. Remotely Operated Vehicle (ROV) video footage and high resolution multibeam mapping of the Foerstner vent site was used to create a geologic map of the 1891 AD deposits and conduct the first detailed study of the source area associated with this unusual type of submarine volcanism. The main Foerstner vent consists of two overlapping circular mounds with a total volume of 6.3 x 10 m and relief of 60 m. It is dominantly constructed of clastic scoriaceous deposits with some interbedded effusive pillow flow deposits. Petrographic and geochemical analyses of Foerstner samples by X-ray fluorescence and inductively coupled plasma mass spectrometry reveal that the majority of the deposits are highly to extremely vesicular, hypocrystalline tephrite basanite scoria that display porphyritic, hyaloophitic, and vitrophyric textures. An intact scoria balloon recovered from the seafloor consists of an interior gas cavity surrounded by a thin lava shell comprised of two distinct layers; a thin, oxidized quenched crust surrounding the exterior of the balloon and a dark grey, tachylite layer lying beneath it. Ostwald ripening is determined to be the dominant bubble growth mechanism of four representative Foerstner scoria samples as determined by vesicle size distributions. Characterization of the diversity of deposit facies observed at Foerstner in conjunction with quantitative rock texture analysis indicates that Strombolian-like activity is the most likely mechanism for the formation of buoyant scoria bombs. The deposit facies observed at the main Foerstner vent are very similar to those produced by other known submarine Strombolian eruptions (short pillow flow lobes, large scoriaceous clasts, spatter-like vent facies). Balloons were likely formed from the rapid cooling of extremely vesicular magma fragments as a result of a gas-rich frothy magma source. The exterior of these fragments hyperquenched forming a vesicular glassy shell that acted as an insulating layer preventing magmatic gas in its interior from escaping and thus allowing flotation as densities reached less than 1000 kg/m. We believe that lava balloon eruptions are more common than previously thought, as the eruptive conditions required to generate these products are likely to be present in a variety of submarine volcanic environments. Additionally, the facies relationships observed at Foerstner may be used as a paleoenvironmental indicator for modern and ancient basaltic shallow submarine eruptions because of the relatively narrow depth range over which they likely occur (200-400 m).

viii

Introduction
Shallow submarine volcanism that produces floating basaltic scoria bombs is one of the most peculiar and rarely observed eruption styles on Earth (Kueppers et al., 2012). Rapid degassing of mafic magma at shallow depths (< 400 m) can lead to highly vesicular, extremely low-density volcanic products. Owing to their morphology and behavior, the floating scoria bombs have been termed "lava balloons" (Gaspar et al., 2003). Eruptive conditions that generate these products are not well understood due to their rare occurrence and paucity of direct submarine observations. Previous investigations only recovered samples while they were still floating on the sea surface and no attempts to describe the vent site and/or the distribution of the bombs after sinking have been made. This has led to debate as to the style of submarine eruptions that produces these unique deposits (e.g. Gaspar et al., 2003).
There have been only five cases where floating basaltic bombs have been observed during submarine eruptions: the recent 2011-2012 eruption of El Hierro, Canary Islands, Spain (Troll et al., 2012), a 1993 eruption near Socorro Island, Mexico (Siebe et al., 1995), a 1877 eruption west of the island of Hawai'i, USA (Moore et al., 1985), a 1998-2001 eruptive episode of the Serreta Submarine Ridge northwest of Terceira Island, Azores (Gaspar et al., 2003;Kueppers et al., 2012), and the 1891 eruption of Foerstner volcano northwest of Pantelleria island (Washington, 1909: Butler, 1892.
The Foerstner eruption was the most recent volcanic activity in the Straits of Sicily. It occurred within the Pantelleria graben, a large, tectonic depression, 90 km long and 30 km wide, that has a general NW-SE orientation and is bounded by very steep normal faults (Civile at al, 2010). Like other balloon eruptions, it produced ellipsoidal, scoriaceous bombs that rose intact through the water column and floated on the surface of the sea before becoming saturated with seawater and sinking (Washington, 1909, Butler, 1892. Many were greater than 1 m in diameter and extremely vesicular with individual vesicles reaching up to decimeters in size. In this study, we report on remotely operated vehicle (ROV) explorations of the vent area of the 1891 Foerstner eruption carried out during cruise NA-018 of the E/V Nautilus. This is the first detailed study of the vent site of a basaltic balloon eruption and a geologic map of the volcanic products has been created using a highresolution mapping and photographic survey of the area. Geochemical and mineralogical analyses of samples collected by the ROV provide information about magma composition, viscosity, and crystal content of the 1891 products. The MATLAB-based program, FOAMS, was used to investigate the degassing processes of the basaltic balloons based on imaging of bubble size distributions. We compare the Foerstner balloons to the products of other basaltic balloon events and develop an eruption model to describe the vent facies and processes that led to the formation of the balloons during the 1891 eruption. The results from this well constrained example provide an important basis for the facies interpretation of ancient highly vesicular submarine basaltic sequences (e.g. Simpson and McPhie, 2001).

Geologic Setting
The Straits of Sicily is located in the northern part of the African continental plate called the Pelagian block ( Figure 1, Burollet et al., 1978). It is bounded by the Skerki bank to the west and the Malta escarpment to the east and is very shallow (averaging 350 m depth), except in three NW-trending depressions (Civile et al., 2010). The Sicily Channel has been affected by Late Miocene-Early Pliocene continental rifting (Civile et al., 2008), which produced several geologic features. The rifting created the Pantelleria, Malta, and Linosa tectonic depressions, which reach depths of 1350, 1580, and 1720m and are bounded by NW-SE trending normal faults (Calanchi et al., 1989;Civile et al., 2008). It also created two volcanic islands (Pantelleria and Linosa) and a series of magmatic seamounts located in the Adventure Plateau, Graham and Nameless banks (Peccerillo, 2005;Rotolo et al., 2006). The rifting also led to a thinning of the continental crust beneath the troughs up to about 17 km along the Pantelleria graben axis (Civile et al., 2008). The rifting has been interpreted as a result of mantle convections developed during the roll-back of the African lithosphere slab beneath the Tyrrhenian basin (Argnani, 1990). The tectonic depressions have also been interpreted as large pull-apart basins involving deep crustal levels, formed within a large wrench zone in front of the Africa-Europe collisional belt (Cello et al., 1985). The rifting may also be related to the NE-directed displacement of Sicily away from the African continent (Illies, 1981). These diverse interpretations show that the tectonic mechanisms responsible for the Sicily Channel rift zone are still not well understood.
Pantelleria Island has a surface area of 83 km 2 and represents the largest extent of an emerged composite volcano in the Sicily Channel, rising from a depth of 1300 m to 830 m above sea level (Civile et al., 2010;Martorelli et al., 2011). The island is composed wholly of peralkaline trachytes and rhyolites (pantellerites) (Civetta et al., 1988) produced by Late Quaternary volcanic activity. The activity was mainly explosive and characterized by caldera collapses producing large volumes of ignimbrites and pyroclastics (Civetta et al., 1984(Civetta et al., , 1988. Pantelleria is characterized by notable volcano-tectonic features such as caldera rims, emission centers, and dike swarms, most significant of which are two large calderas that developed on the southeastern part of the island (Civile et al, 2008). High-resolution mapping of the Pantelleria submarine flanks shows numerous small volcanic cones concentrated to the NW of the island (Bosman et al., 2007). Seismic profiles and regional magnetic anomalies indicates that volcanic bodies with a clear magnetic signature are exposed at the seafloor as far as 37 km northwest and southeast of Pantelleria, with their elongation consistent with the orientation of the rift (Calanchi et al., 1988). Volcanic bodies southeast of the island are buried beneath undisturbed Upper Pliocene-Quaternary sediments, suggesting major volcanic activity in the early stage of the graben development, which partly filled the rift floor (Calanchi et al., 1988). The bodies located northwest of the island are probably related to volcanic activity associated with the development of the Pantelleria Rift (Calanchi et al., 1988).
Seismic activity of the Sicily Channel is characterized by shallow (<25 km), low magnitude (2 to 4) events (Civile et al., 2008). Seismicity is notably absent along the Pantelleria graben, while some earthquakes have been recorded north of Pantelleria.
In the Sicily Channel, volcanic activity was concentrated mainly on the islands of Pantelleria and Linosa during the Pleistocene. Minor submarine volcanism began during the early Pliocene and lasted until 110 years ago, mainly occurring in Adventure plateau and in Graham and Nameless Banks (Corti et al., 2006;Rotolo et al., 2006). The oldest products have been found in a volcanic seamount located east of Nameless Bank, where dredged samples gave a K-Ar age of 9.5 Ma (Beccaluva et al., 1981). The most recent activity occurred during the nineteenth century, most notably in the Graham Bank in 1831 and Foerstner volcano in 1891, the most recent eruption (Washington, 1909). Additional eruptions occurred in 1801, 1845, 1846, and 1863 but were not observed because they did not give rise to a permanent subaerial island due to short periods of activity (Washington, 1909).

Historical Accounts of the 1891 Foerstner Submarine Eruption
In 1890, many premonitory signals to an eruption were detected on Pantelleria, as summarized by A. Ricco (Butler, 1892). These signals included increased fumarolic activity causing damage to vineyards, increased earthquake activity, and uplift of the island's north coast due to increased tectonic activity (Butler, 1892). On October 14-15, 1891 (three days before the eruption began), stronger earthquakes were accompanied by the drying up of hot springs and a further rise of the north coast (totaling 80 cm), which resulted in surface cracks (Washington, 1909;Butler, 1892).
Large amplitude earthquakes preceded the beginning of the eruption on the morning of October 17, after which all earthquake activity ceased (Washington, 1909).
The 1891 eruption of Foerstner volcano lasted only a week and was not described by any scientific observer. Descriptions of the eruption have been derived from the testimony of fishermen, most notably A. Ricco, whose account was translated by G.
W. Butler in 1892 and provides the basis for the description that follows (Washington, 1909). The first signs of the eruption were deep rumblings and columns of "smoke" protruding from the sea surface 4 km west of the town of Pantelleria at the northwest end of the island. Black, subspherical, scoriaceous bombs up to 1 m in diameter were seen rising to the surface along a NE-SW trending line about 850-1000 m long, initially thought to have been produced by fissure activity (Figure 2). Some of the bombs were still degassing at the surface and as a result, were propelled laterally by horizontal steam jets. Some were thrown up to 20 m in the air as a result of rapid degassing. Bombs collected at the surface were still at very high temperature inside (at least 415º C as indicated by fusion of Zn) and one bomb was noted as being incandescent. After the degassing episodes had ceased, the scoria balloons sank as a result of seawater saturation. Some claim that an ephemeral island was formed (including Foerstner himself), but both Ricco and Butler explicitly deny this suggestion. The highest water temperature recorded at the eruption site as 1.5º C above the ambient temperature. It was noted that there was a strong smell "as of gunpowder" at the site, which was most likely H 2 S and SO 2 gas emissions. The eruption ceased on October 25, 1891.

Methodology
The suspected vent site of the 1891 Foerstner eruption (Bosman et al., 2007) was explored using the remotely operated vehicle (ROV) Hercules during cruise Scoria and lava flow samples were collected at 42 sites at or in close proximity to the submarine vent site using the ROV. Ten samples that encompass the major clast types observed at the vent site were selected for petrographic and geochemical analyses. Bulk samples were cleaned in de-ionized (DI) water, sonicated for 30 minutes to remove foreign particles, rinsed in DI water again, and then dried for 48 hours at 100° C. Powdered whole rocks were analyzed for major elements by X-ray fluorescence (XRF) using the standard BHVO-2 at the Ronald B. Gilmore X-ray Fluorescence Laboratory, University of Massachusetts, Amherst. Trace element compositions in the same powders were analyzed using the New Wave 213 nm Nd-YAG laser, attached to a Thermo X-Series II ICP-MS at Dr. Katie Kelley's lab at URI's Graduate School of Oceanography (GSO) in Narragansett, RI following the methods of Kelley et al. (2003). Standards used were: JB-3, BHVO-1, DNC-1, W-2, and EN026 10D-3. Reproductibility of replicate analyses is <2% rsd. Petrographic descriptions of these samples were completed using a Zeiss Axioscopt petrographic microscope in Dr. Steven Carey's lab at GSO.
Characterization of bubble textures in vesicular samples was carried out following the methods of Shea et al. (2010). Thin sections made from selected samples were imaged using a petrographic microscope and a scanning electron microscope (SEM), using 5x-100x magnifications to image a range of vesicle sizes between 10 µm and 1.58 mm. SEM imaging was done using a JEOL JSM-5900LV SEM at Mike Platek's lab at the University of Rhode Island, Kingston, RI. The nested images were used to derive vesicle volume distributions via the FOAMS program (Shea et al. 2010).
A geologic map of Foerstner volcano was created using high definition video footage recorded by Hercules. A total of 40 hours of video footage of the Foerstner vent site and surrounding area was recorded during dives H1205, H1206, and H1207.
Preliminary viewing of the video was used to identify 17 different facies of volcaniclastic and effusive deposits. The video footage was then systematically viewed and facies type was recorded at one minute intervals and linked to the ROV navigation tracks. The map was created using Adobe Illustrator®.

Mineralogy and Petrography
All samples from the vent area of Foerstner volcano are plagioclase-olivinephyric basanite. The majority are hypocrystalline and display porphyritic, hyaloophitic, and vitrophyric textures. Plagioclase is present mostly as acicular with some tabular and few equant microphenocrysts and phenocrysts (0.03-1.85 mm) as well as microlites. Olivine is mainly present as microlites, but with some euhedral to subhedral microphenocrysts and few phenocrysts (0.03-0.85 mm) that contain melt inclusions. Olivine also occurs within aggregates (glomerocrysts) in some samples.
Subhedral augite is the least commonly found phenocryst.
The groundmass of the samples consists of sideromelane, tachylite, or a mixture of both. Tachylite is cryptocrystalline glass and is the result of confined conditions of crystallization, controlled by local enrichment-depletion of elements adjacent to precipitating minerals (Taddeucci et al. 2004). Tachylite is nearly opaque as it contains abundant microlites leading to an optical isotropy similar to magnetite (Morris et al., 1990). Vesicularity descriptions used in this study are adopted from Houghton and Wilson (1989) and are as follows: 0-5% non-vesicular, 5-20% incipiently vesicular, 20-40% poorly vesicular, 40-60% moderately vesicular, 60-80% highly vesicular, >80% extremely vesicular. Most of the samples from the Foerstner vent area are highly to extremely vesicular.

Whole-rock and Trace Element Compositions
Ten samples were chosen for chemical analyses based on the location and lithology of the submarine deposits (Table 1). Samples analyzed were from the main vent of Foerstner (NA018-019, 021, 022, 023, 026), 125 m south of Foerstner (NA018-025), a northern, deeper vent (NA018-027), in the saddle region between Foerstner and a large seamount to the west (NA018-030), and at the summit of the western seamount (NA018-032, 033) ( Table 1). All scoria samples are classified as tephrite basanite according to IUGS classification, occupying a narrow range in SiO 2 abundance from 43.9-44.9 wt% (Figure 5ab, Table 2). Sample NA018-025 to the south of Foerstner is more evolved than the rest of the Foerstner deposits with SiO 2 content of 47.0 wt%. Washington (1909) carried out the only other chemical analyses of Foerstner scoria deposits. Samples analyzed from the Foerstner vent site in this study are similar in composition to those of Washington (1909), although slightly less evolved ( Figure   5a). Low abundances of SiO 2 (~44 wt%), MgO (~5.5 wt%), and alkalis (~4.5 wt%) are common between both analyses as well as the abundance of CaO. Discrepancies arise when comparing Al 2 O 3 , P 2 O 5 , and TiO 2 . The content of Al 2 O 3 is higher in samples from this study (1.7 wt%), while there is 2.1 wt% and 0.5 wt% more TiO 2 and P 2 O 5 in Washington's samples. Some of the discrepancies may reflect true compositional differences but they more likely represent interlaboratory differences in techniques used for the analyses that span more than 100 years.
Samples from Foerstner are typically less evolved than samples taken from other nearby volcanic centers in the Straits of Sicily, which are basalt and trachy-basalt ( Figure 5a) (Calanchi et al, 1989). An exception are samples from Nameless and Tetide Banks that are less evolved when compared to Foerstner and plot near the boundary of tephrite basanite-foidite and tephrite basanite-picro-basalt ( Figure 5a) (Calanchi et al., 1989).
Whole-rock chemical analyses of the Foerstner samples were compared with all other known submarine lava balloon deposits. Chemical data plotted on the IUGS TAS diagram confirms that each of the four lava balloon deposits are classified as four different types of volcanic rock: Foerstner -tephrite basanite (this study), Socorrotrachy-basalt (Siebe et al., 1995), Hawaii -basaltic andesite (Moore et al., 1985), Azores -basalt (Gaspar et al., 2003) (Figure 5b). There does not appear to be a strong correlation between the composition of magma and the ability to produce lava balloons during eruption, although it is noted that most basaltic balloon eruptions tend to be more alkalic in nature.
The Foerstner basanites show very similar enrichment in the rare earth elements relative to chondrites, with light rare earth elements (LREE) being strongly enriched relative to the heavy rare earth elements (HREE) ( Figure 6, Table 3).
Following the methods of Peace and Norry (1979), trace elements of the basanites are plotted on the Zr/Y-Zr tectonic discrimination diagram (Figure 7, Table 4). According to this diagram, the high Zr-Y Foerstner samples plot in the within-plate (oceanic island) volcanic province domain. Mid-ocean ridge-normalized (N-MORB) trace element patterns were plotted following the methods of Pearce (1983). Trace element patterns show most incompatible elements are enriched relative to MORB as well as significant enrichment of Ta and Nb ( Figure 8) (Pearce, 1983). These patterns are indicative of intra-continental plate basalts and agree with the REE results. Sample NA018-025 shows anomalous geochemical patterns relative to the rest of the Foerstner samples. It is significantly less enriched in rare earth and trace elements suggesting that it originated from a different magmatic source.

Vent Structure and Deposit Facies
A bathymetric map of the suspected Foerstner vent site was created using high resolution ROV surveys and shows that the volcanic edifice has a slight elliptical shape formed by the overlapping of two circular mounds (

General Textural Observations
Thin section -Certain textural features are common throughout most of the Foerstner scoria deposits. Vesicles contained within sideromelane groundmass are typically more abundant but smaller in size, while those found in tachylite are fewer but larger in size. Small vesicles (i.e., L<0.2 mm) are typically round while larger vesicles become increasingly irregular with non-circular outlines. Vesicle walls are smooth, and bubbles are found adjacent to both groundmass and crystals. There is no evidence of shearing as evidenced by the lack of elongated vesicle trains.
Tachylite has a higher content of microlites and a lower vesicularity compared to sideromelane (Figure 17a). This is most likely a result of a slower rate of cooling allowing time for microlite nucleation and growth. The transition from sideromelane to tachylite groundmass observed in the shell of some Foerstner samples can be explained by a variation of cooling rates, as described by Kueppers et al. (2012) for basaltic balloons from the Azores. Upon discharge into seawater, lava instantly cools at a very high rate (up to 1,259 K/s) forming a thin, oxidized crust that consists of rapidly quenched glass (sideromelane) and large vesicles. This crust provides a thermal boundary layer between the incandescent interior of the scoria bomb and cold seawater. The glass beneath the crust is not in direct contact with the seawater and cools at a much slower rate (~30 K/s). This slower rate allows for an extensive nucleation and growth of microlites, leading to the development of tachylite ( Figure   17b). The VVD plot of a scoria bomb representative of those observed at Foerstner recovered from the large western vent (NA018-032) displays a negative skewness with an extended tail towards small vesicles (Figure 23). Vesicles range from 0.01 to 1 mm in diameter. The distribution shows a progressive disappearance of smaller vesicles with a major mode between 0.32 and 0.63 mm (ignoring the significant drop off in volume fraction at 0.5 mm).

Bubble formation and growth within AD 1891 magmas
A comparison of bubble texture data from samples of AD 1891 deposits allows inferences to be made about magma degassing processes. Samples were chosen to cover the major lithofacies found at the Foerstner vent site and surrounding seafloor.
All four samples exhibit generally unimodal distributions on VVD plots (Figures 20-23) suggesting a single distinct pulse of nucleation and growth (Shea et al. 2010). The negative skewness, also observed on all four plots, is interpreted to be a consequence of bubble ripening during the course of the eruption.
The bubble sizes and spatial distributions observed in the Foerstner samples suggest that growth resulted from the steady diffusive transfer of gas between bubbles through films. This transfer process, known as Ostwald ripening, is driven by the pressure excess inside bubbles, which is high for small bubbles and low for large bubbles (Mangan and Cashman 1996). Since the bubble distribution within magma is generally polydispersed, internal pressures will be irregular. As bubbles occupy a greater volume of the melt (usually near the vent surface), the nearest-neighbor distance becomes smaller and gas diffuses from regions of high to low pressure. As a result, large bubbles grow and small bubbles shrink or may disappear altogether (Mangan and Cashman 1996). In this respect, the process differs from coalescence, which is driven by mechanical and molecular interactions resulting in a more favorable thermodynamic condition, but is not actually driven by surface energy (Herd and Pinkerton 1997).
The progressive disappearance of small, and enrichment of large vesicles is observed in all VVD plots (Figures 20-23). As in coalescence, the foam coarsens as magma residence time in the conduit increases and the surface area of the gas-melt interface decreases. However, individual bubbles shrink or grow gradually during ripening, and a normal rather than polymodal bubble size frequency distribution evolves (Mangan and Cashman 1996). Studies have shown that ripening has a significant influence in the bubble textures of basaltic fire-fountain eruptions (Mangan and Cashman 1996). Additionally, it has been observed that Strombolian activity at Heimaey, Iceland was controlled by the bursting of large, individual bubbles (Blackburn et al. 1976) such as those that are formed by ripening. Further discrimination between Hawaiian fire fountaining and Strombolian-type eruption mechanisms cannot be deduced from these data.

Facies Distribution and Eruption Model
Eruption mechanisms that vary from dominantly effusive to explosive have been proposed to explain the production of floating scoria and their subsequent deposition. Gaspar et al. (2003) and Kueppers et al. (2012) proposed that the magma involved in the Serreta submarine ridge eruption was fluid and gas-rich, favoring the segregation and accumulation of gas under a cooler lava crust at vent level. They proposed the development of large gas bubbles within the magma just below the crust.
At a critical point of accumulation, these large gas bubbles form blisters that, in a subaqueous setting, detach from the lava surface as swollen lava balloons and rise by flotation. In contrast, Siebe et al. (1995) proposed that intermittent lava fountaining could be responsible for the production of floating scoria bombs, as well as other pyroclastic deposits and pillow flows. Fountaining results from changes in eruption velocities due to variation in exsolved volatile content. While most of the clasts fall out close to the vent, some gas-charged magma could produce highly vesiculated scoria that rise to the surface (Smith and Batiza 1989). At the summit of Foerstner volcano there was no evidence for a solidified lava lake or extensive pillow flow lobes. Thus, the dominant clastic nature of the vent site points more strongly to a dynamic explosive mechanism for balloon formation.
The dominant ripening degassing process determined by analyzing the bubble size distributions of Foerstner deposits favors either a Hawaiian fire fountaining or Strombolian-type eruption mechanism. Each of these eruption types are characterized by fluctuating magma discharge rates due to changing exsolved volatile contents (Siebe et al. 1995;Pioli et al. 2009). Either submarine fire fountaining or Strombolian activity could explain the deposition of abundant pyroclastic deposits as well as the effusive lava flow deposits observed at the main vent site. Magma eruption velocities are unlikely to remain constant for extended periods of time; instead they would fluctuate as observed at subaerial lava fountains (Head and Wilson 1987). Periods of rapid gas exsolution would promote explosive eruptions and their subsequent deposits, while waning exsolution promotes effusive eruptions (Pioli et al 2009).
The diversity of facies observed at the Foerstner vent site are most likely a result of varying magma rise speed in the feeding conduit, enhanced volatile content of the source alkali basalt, and generally low viscosity of the magma. These factors promote differential rise speeds of melt and bubbles allowing large, early-nucleated bubbles to migrate upward in the conduit as a slug flow and overtake smaller, laternucleated bubbles. The emergence of these large bubbles through magma within the conduit drives Strombolian type eruptions (Head and Wilson 2003). This erupted magma would have contained a lower fraction of small vesicles as they were overtaken by larger, more energetically stable bubbles. The lack of small gas bubbles limits the disruption of erupted lava into smaller fragments, allowing for very large magma clots to be ejected at sizes comparable to the width of the conduit (Head and Wilson 2003). This mechanism could explain the deposition of scoria bombs up to 2 meters in size observed on the flanks of Foerstner. Head and Wilson (2003) made predictions about the resulting deposits and landforms of submarine Strombolian eruptions. The initial stages of a Strombolian eruption are characterized by dike emplacement and extrusion of lavas at very low discharge rates. This leads to the deposition of short flows and pillows rather than extensive lobate sheets (Head et al. 1996;Gregg and Fink 1995), much like the welldefined pillow flow lobes forming linear ridges that emanate radially towards the northwest and west from the main Foerstner vent ( Figure 10). As magma rise rates increase and stabilize, typical Strombolian activity occurs as the gas bubble rise rate exceed that of the magma. Explosive disruption of the magma occurs at the ventwater interface, producing fragmental deposits that typically fall within 10-20 m from the vent as a result of hydrodynamic drag and subsequent deceleration of pyroclastic fragments. The larger blocks and bombs (64 to >256 mm in diameter) that constitute the major facies observed at Foerstner most likely formed from the plug of magma ejected in front of the larger rising gas bubbles. Smaller fragments quickly settle near the vent upon ejection as a result of their low inertia and accumulate as agglutinate.
Deposits of this kind may correspond to spatter-like deposits observed at the summit of Foerstner.
In general, explosive Strombolian activity becomes localized at the cone vent, while lava flows emerge from lateral vents located at the base of the cone (Valentine et al. 2005;Pioli et al. 2009). This type of vent morphology is observed at Foerstner as the main vent site is dominated by explosive pyroclastic deposits and the three smaller mounds observed immediately to the northwest are constructed of pillow flow lobes. Simultaneous eruption of pyroclastics from the cone and lava flows from the lateral vents requires segregation of a low viscosity, exsolved volatile-rich magma into a gas-rich mixture that ascends through the central conduit and gas-poor lava flowing laterally. Observations of Strombolian activity at Paricutin volcano in Mexico showed that the mass eruption rate (MER) controlled the proportion of magma emitted by explosive vs. effusive activity and the initial formation of lateral vents increased the explosivity of eruptions occurring at the cone (Pioli et al. 2009). Dual activity of this type requires a MER of 10 3 to 10 5 kg/s and when this drops to a rate below 10 3 kg/s, degassing dominates producing either lava effusion or mild explosive activity (Pioli et al. 2009).
The pyroclastic deposits that occur on Foerstner can be distinguished from submarine fire fountain eruption products by the dominance of large scoriaceous clasts deposited near the vent, a spatter-like vent facies, lack of extensive pyroclastic flow deposits, and short, pillow textured flows. If Hawaiian-style fountaining was the dominant eruption mechanism, the predicted deposits would include vesicular sheet flows, partly agglutinated distal fragments, relatively small grain sizes, and abundant pyroclastic flows surrounding the cone (Head and Wilson 2003).
Thus, we propose that Strombolian activity was the most likely eruption style Although we use the term Strombolian style to describe the eruption mechanism it is noted that there are fundamental differences between the behaviors of Strombolian events in the subaerial versus the submarine environment. First, the presence of water significantly dampens the dispersal of fragmented clasts in the submarine environment and will likely affect the resulting morphology at the vent area. In the subaerial environment Strombolian activity produces scoria cones with a central crater because fragmented clasts are able to be dispersed over relatively large distances (>100 m) by ballistic trajectories through the atmosphere. Underwater, the drag effect restricts dispersal and likely builds a mound-like cone without a welldefined central crater. Second, in the submarine environment gas-rich magma can attain positive buoyancy relative to seawater prior to fragmentation (Friedman et al. 2012;Rotella et al. 2013). This condition, which can never be attained in the subaerial environment, can dramatically affect the nature of submarine eruptions. Positive buoyancy flux results in a potentially complex globular discharge of highly vesicular magma that then fragment by gas expansion (Friedman et al. 2012). This style of activity was captured by ROV photography on West Mata submarine volcano in the Pacific (Rubin et al. 2012, fig. 2b).
Assuming that Foerstner exhibited Strombolian-type eruptive activity, predictions can be made about the generation of the lava balloons. We suggest that discharge of magma at the vent-seawater interface produced batches of coarse magma "blobs" that rapidly cooled and decelerated upon contact with the surrounding seawater. Since magma erupted from Strombolian activity is typically a gas-rich froth rather than liquid, many of these blobs have the potential to be extremely vesicular and thus buoyant relative to seawater. Maximum bubble sizes for submarine Strombolian eruptions can reach up to 1.5-2 m in size as a result of higher ambient pressure (Head and Wilson 2003). Data from other studies suggest that the exterior of these fragments can hyperquench with a cooling rate of ~1,259 K/s to form a solid lava shell that acts as an insulating layer preventing the magmatic gas in its interior from escaping (Kueppers et al. 2012). These extremely vesicular bombs attained densities less than 1000 kg/m 3 and thus can rise buoyantly to the surface. Magma forming these rising balloons continued to degas as the ambient pressure decreases leading to balloon inflation. This expansion leads to enlargement of the balloons' surface area and formation of new skin. This process is facilitated by the presence of still-molten lava inside the balloon, which prevents complete rupture of the outer solid crust. Some lava balloons collected off the sea surface in 1891 were noted to still have molten, redhot interiors (Butler 1892).
All other known occurrences of floating scoria have been associated with a relatively narrow water depth range between 30 and 1000 m (Siebe et al. 1995;Kueppers et al. 2012;Rivera et al. 2013), but with the best-defined source vents being located within a more limited range of only 200 to 400 m. It is likely that at depths greater than 400 m, confining pressures prevent the degree of volatile exsolution required to generate extremely vesicular magma that is buoyant relative to seawater.
In contrast, at depths shallower than 200m, extremely low pressures promotes rapid volatile exsolution and a high degree of fragmentation preventing the discharge of coarse magma blobs. At such shallow depths highly explosive activity is likely driven by phreatomagmatic explosions in addition to primary degassing (Sigurdsson et al. 1999). The presence of highly inflated scoria bombs in association with a dominantly clastic vent site, as observed at Foerstner, may represent a paleoenvironmental indicator for modern and ancient basaltic shallow submarine eruptions at depths of several hundred meters. producing short, pillow flow lobes that emanate from the center of the vent to the northwest and west. Typical Strombolian activity soon followed as the gas bubble rise rate exceeded that of the magma. Explosive disruption of the magma occurred at the vent-water interface and produced coarse fragmental deposits that fell on the slopes, while smaller fragments quickly settled near the vent and accumulated as spatter-like deposits. We suggest that buoyant scoria bombs were formed from the rapid cooling of extremely vesicular magma blobs. The exterior of these fragments hyperquenched forming a solid lava shell that acted as an insulating layer preventing magmatic gas in its interior from escaping. The bombs attained densities less than 1000 kg/m 3 and thus rose buoyantly to the surface. Rupture of the outer solid crust was prevented by the presence of a still-molten interior that accommodates expansion by progressive thin crust formation.

Conclusions
Lava balloon eruptions may occur more frequently than previously thought.
The eruptive conditions that characterize these products are now better documented and likely occur in a variety of submarine volcanic environments: low silica alkalic magma (43-52% SiO 2 ), high volatile content (CO 2 , H 2 O), and low hydrostatic pressures (water depths between ~200-400 m). Most historical balloon eruptions had durations of days to weeks, providing a very small window of time to allow for their detection by direct observation, seismology, or remote sensing. Only the Serretta Ridge eruption in the Azores lasted for a year or longer (Gaspar et al. 2003;Kueppers et al. 2012). Identifying and monitoring active submarine volcanoes that satisfy the

Fig. 5a
Average composition of Foerstner basanitic scoria in comparison with samples from other major volcanic centers in the Straits of Sicily as plotted in the TAS diagram after the International Union of Geological Sciences (Calanchi et al., 1988;Washington, 1909;Beccaluva et al., 1981). b Average composition of Foerstner basanitic scoria balloons in comparison with all other known submarine lava balloon deposits (Siebe et al., 1995;Moore et al., 1985;Kueppers et al., 2012). Red dashed line discriminates between alkaline and sub-alkaline rocks.  Pearce (1983). Sample NA018-025 shows significant less trace element enrichment relative to the rest of the Foerstner samples indicating it likely originated from a different magma source.      (OL) and inner tachylite (TL) layers labeled. Arrow points to the location of the white horizon layer separating the lava shell from the hollowed interior.

Fig. 22
Vesicle volume distribution of a pillow flow lobe fragment recovered from the northwest mound (NA018-027).

Fig. 23
Vesicle volume distribution of a scoria bomb representative of those observed at Foerstner recovered from the large western vent (NA018-032).