The NW submarine portion of Stromboli volcano has been investigated by deep-towed sidescan sonar, bathymetric surveys, video camera runs and dredging during two research cruises in 2002 and 2004. The surveys resulted in the identification of an extensive pillow lava field (106-107m3) at about 2300 m of water depth and 9 km from the shoreline of Stromboli Island. The pillow lavas have a unique composition that does not match any known subaerial product, although a limited affinity exists with those erupted during the Neostromboli eruptive cycle of the island (13–6 ka). This is the first finding of a submarine eruption on the northern side of Stromboli and improves the knowledge of its flank activity and volcanic hazards. This eruption is interpreted as marking the onset of a new volcanic cycle from the edifice periphery fed by a new, distinct magma mixed with traces of the previous magma that survived the emptying of the Neostromboli magma chamber.
 Flank activity is common in basaltic and andesitic volcanoes where it is concentrated in rift zones (e.g., Kilauea, Etna, Mauna Loa) or fed by radial dykes (Etna, Piton de la Fournaise). Additionally flank eruption can be fed by the direct ascent of magma along conduits or fissures not connected to central vent plumbing system. Ascent of magma takes place along paths that are oriented normal to the local or regional direction of least principal stress [Valentine and Krogh, 2006].
 Recent models [Takada, 1997] suggest that basaltic and polygenetic volcanoes can undergo cycles of flank and summit activity. Cycles start with the development of flank activity on the periphery and terminate with central vent activity and relaxation of the stress within the volcanic edifice.
 Stromboli, the northernmost volcanic island of the Aeolian Archipelago, is located in the Southern Tyrrhenian Sea, a few tens of km offshore from the north coast of Sicily and the Italian peninsula (Figure 1). The island, with an elevation of 924 m, is a cone with steep slopes and represents the subaerial part of a larger volcanic edifice extending to a depth of 1500–2600 m below sea level. The geological evolution of Stromboli volcano is the result of seven main periods of activity covering a time span of ∼100 ka, i.e., Paleostromboli I, II, and III (100–35 ka), Scari (≈35 ka), Vancori (25–13 ka), Neostromboli (13–6 ka), and Recent Stromboli (<6 ka) [Hornig-Kjarsgaard et al., 1993; Gillot and Keller, 1993]. Transitions between each period are marked by significant structural modifications of the edifice (caldera collapses during the early periods and flank collapses after the Vancori period [Pasquarè et al., 1993; Tibaldi, 2001]) and changes in magma compositions [Hornig-Kjasgaard et al., 1993; Francalanci et al., 1993]. The composition of Stromboli eruptive products is typical of subduction-related magma series and ranges from calc-alkaline and high K-calcalkaline to shoshonitic, with the distinctive exception of the K-alkaline magmas produced exclusively during the Neostromboli period.
 The volcano is famous for its persistent activity that began, sometimes between the third and seventh centuries AD [Rosi et al., 2000]. This activity consists of mild intermittent bursts and continuous gas streaming from the crater area located at ∼750 m a.s.l. inside the Sciara del Fuoco (SdF), a horseshoe-shaped collapse scar that occupies the NW sector of the island.
 Since the seventh century AD [Rosi et al., 2000] volcanic activity has been mainly located in a restricted summit area or within the SdF. The latest flank eruption outside the SdF, occurred between 360 BC–100 AD [Arrighi et al., 2004; Speranza et al., 2008] and produced the San Bartolo lava flow exposed in the NE flank of the volcano from about 500 m a.s.l down to the shore (Figure 1). Flank activity was particularly significant from 8 to 6.2 ka, during the Neostromboli period, as witnessed by eruptive fissures, tumuli and exposed dykes, recently dated on a paleomagnetic basis, and exposed in the northeastern and western sectors of the island. In contrast, information is absent regarding the type and composition of volcanics erupted in the early stage of Recent Stromboli activity (6 ka and 400 BC [Speranza et al., 2008; Rosi et. al., 2000]).
 In the submarine portion of the edifice, two lateral vents have been identified and sampled: Cavoni and Casoni seamounts, at a depth of 700 and 1200 m respectively, about 3–4 km SW from Stromboli, the latter with evidence of very recent volcanic activity [Gabbianelli et al., 1993; Gamberi et al., 2006].
 Some authors suggest that magma ascent paths and opening of fissures during past eruptive activity at Stromboli volcano are controlled by the regional stress field and in recent times (<13 ka) also by the unbuttressing of the NW flank [Tibaldi, 2003; Corazzato et al., 2008]. Dike intrusion on the SdF shoulders has additionally been considered as a cause for subsequent flank instabilities.
 Magma intrusion and endogenous growth have also been invoked to adjust mass balance calculations related to continuous degassing [Francis et al., 1993; Allard et al., 1994]. In fact, the emplacement of a shallow intrusion at about 1200 m b.s.l. and extending as far as 4.5 km NE of the island, has been proposed based on geophysical signals from December 1994 to March 1995 [Bonaccorso, 1998].
 Subaerial flank eruptions outside the SdF represent rare events in the known history of Stromboli and represent a hazard to inhabited areas. Submarine flank eruptions, if connected to the destabilization of a wide sector of the volcano, could trigger tsunamis, that could impact the whole southern Tyrrhenian Sea.
 For this reason, a growing research effort aimed at detailing hazard assessment including information from the submarine portion of volcano has been carried out. This paper reports the first evidence of an important and well preserved lava flow-field, discovered on the northern foot of the Stromboli edifice and related to its flank activity.
2. Seafloor Characterization
 Information on the whole submerged portion of the Stromboli was provided by a complete coverage of the region by multibeam swath bathymetry, acquired in 1996 and 1999 by the CNR Institute of Marine Geology of Bologna. During the 4th and 14th IOC-UNESCO-ESF Training-Through-Research cruises in 1994 and 2004, respectively, the NW submarine flank offshore of the SdF was characterized.
 Seafloor characterization involved deep-towed MAK sidescan sonar (sss) and bathymetric surveys. During the regional survey an irregular backscatter, hummocky outcrop was imaged on the NW slope between 2300 and 2400 m depth, about 9 km to the N of Punta Labronzo (Figure 1). The sss data reveal downslope-directed high backscatter swathes representing the active coarse sediment flows sourced mainly from the SdF area (Figure 1a). However, close to the base of slope, between 2200 and 2300 m water depth and 9 km to the north of Stromboli Island, the SSS reflectivity exhibits numerous high backscatter, closely-spaced, hummocky outcrops, sharply contrasting with the SSS characteristics of the surrounding seafloor (Figures 1a and 1b).
 In the multibeam data this corresponds with a linguoid, ∼2 km2 wide area with isobaths that protrude down-slope (Figure 1b).
 The sss imagery was ground-truthed during the 2004 cruise by 3 video camera runs performed across-slope (W–E) at 1800 m, 2300 m and 2550 m b.s.l. on the NW flank of Stromboli. The high backscatter targets on the sss lines were shown by the TVMS19 video camera run (2320–2340 m b.s.l) to consist of an extensive pillow lava outcrop (Figure S1 in the auxiliary material). Moving from the margins towards the inner part of the outcrop the camera imaged single, isolated, round-shaped to tongue-shaped pillows and several meters thick mounds of tongue-shaped pillows. Chaotic pillow breccias and angular lava fragments are abundant on the entire outcrop. In places pillows are capped by a few mm-thick yellowish hemipelagic mud. The overall thickness of the pillow deposit is unknown since the basal contact is always masked; however a thickness of several meters can be deduced by TV camera runs where vertical failure surfaces are produced by pillow pile instability. On the basis of these observations, a minimum volume on the order of 106-107 m3 can be estimated for the whole outcrop.
3. Rock Sample Description
 Pillow lava samples were collected by dredging (auxiliary material) and mainly consist of coalescing lobes. The outer surface of each lobe is corrugated, displays abundant spreading wrinkles and is always bordered by a ∼1 cm-thick, black, glassy rind. The outer surface of the pillows is reddish brown in color.
 Fresh spreading cracks often cut the early quenched crust resulting in multiple rind structure, as possible evidence of pillow inflation by lava resupply or volatile expansion in the pillow interior [Yamagishi, 1985; Kawachi and Pringle, 1988].
 Overall vesicularity is very low (<4 vol.%). A few sub-spherical vesicles occur in the glassy rinds; inward, cm-sized vesicles are flattened below the pillow margin. Millimeter-sized sub spherical vesicles are present in the pillow interior together with several cm-long pipe vesicles. In cross section the pillow lava samples display roughly radial joints.
 Alteration is limited to greenish/yellowish coatings on open fractures, exposed crack surfaces, and in pipes and vesicles. Yellow-gray, indurated, hemipelagic inter-pillow sediments are also present.
4. Petrographic and Mineralogic Characterization
 The pillow lava samples are porphyritic rocks (28–30 vol.% of crystals) with phenocrysts of plagioclase (12–14 vol.%), olivine and pyroxene (sum of mafic phases = 16 vol.%).
 On the basis of the groundmass texture, three main zones can be distinguished from the margin to core of the pillows: i) a ∼10 mm-thick, yellow to light brown glass rim with dispersed olivine and pyroxene microlites, plagioclase microlaths and very minor Ti-Magnetite; ii) a 2–3 cm-thick band characterized by an intersertal groundmass with interstices between phenocrysts occupied by cryptocrystalline material and very minor glass; iii) an inner portion characterized by an intergranular groundmass with microlites of K-feldspar and pyroxene, thin laminae of biotite and minor grains of olivine and Ti-magnetite.
 These concentric arrangements are related to the cooling of the pillow and are in agreement with observations of other authors [Marescotti et al., 2000, and references therein]. Plagioclase crystals (up to 3.5 mm in size) occur as elongated single phenocrysts with rounded rims and as radiated crystal clots. Both species often show evidence of large bands with sieve texture surrounding more homogeneous cores. Oscillatory zoning is widespread and covers a large range of composition (An87-An65) regardless of the crystal size; outer rims tend to be represented by the most sodic compositions (An67-An65).
 Plagioclase phenocrysts are bordered by up to 15 μm-thick rims of K-feldspar with the exception of those in the glassy rinds.
 Clinopyroxene (up to 8 mm in size) and olivine occur as isolated, euhedral to subhedral phenocrysts and in orthocumulates with interstitial subhedral plagioclases. Olivine show frequently skeletal textures and abundant glassy inclusions. Clinopyroxene has prevalent compositions in the range Fs12–17Wo44–47, and often show variously-sized bands with Fs6–8Wo45–46 composition. Microcrystals in the groundmass are slightly more evolved (Fs14–22Wo46–50). The majority of olivine crystals are homogeneous (Fo72–75). However two distinct direct zoning patterns are also present with respectively Fo89–91 and Fo81–85 cores and Fo73–74 rims. The Mg-rich cores are often associated with tiny euhedral Mg-Cr spinels.
5. Chemical Characterization
5.1. Bulk Rock
 Whole rock analysis plots in the trachybasalt field of the total alkali-silica (or TAS) classification diagram of Le Maitre . Thus the pillow lava falls well above the shoshonitic field of the SiO2-K2O diagram which is traditionally employed to classify subduction-related magmas (Calc-alkaline and High K calcalkaline to shoshonitic) erupted from Stromboli during its volcanic history [Hornig-Kjarsgaard et al., 1993; Francalanci et al., 1993]. Though K2O content is comparable with Neostromboli volcanics, the pillow lava shows lower silica content and does not match any known composition of rocks forming the subaerial part of the volcano (Figure 2a and Table S1).
 The trace elements spidergram (Figure 2b) shows a close resemblance between the pillow composition and the products of subaerial fissure eruptions of Neostromboli and striking differences (in particular for LILE enrichment) with Vancori, Present-day products and the most recent flank eruption (S.Bartolo lavas).
 Analysis of the groundmass glasses measured on the pillow rim yielded trachyandesitic compositions (SiO2 = 53.7–55 wt%; K2O = 6.4–6.6 wt% MgO = 2–2.5 wt%). These glasses appear to retain an appreciable quantity of volatiles (H2Otot = 1.65 ± 0.1 wt%, CO2 = below the detection limit) due to their emplacement below a significant water column. FT-IR spectra showed that these hydrous glasses include both molecular water and hydroxyl species (OH-). The latter species are predominant (≈70% of the total water) in agreement with water solubility and speciation models in basaltic glasses at eruptive temperature (≈1100°C) [Dixon et al., 1995]. Thus the possibility of significant post-emplacement hydration processes can be discarded.
 Assuming a maximum water content of 1.65 wt% in the quenched glassy rim and conditions of water saturation of magma at the eruptive vent, we can infer on the basis of available solubility models [Newman and Lowenstern, 2002] a pressure of emplacement of about 27 Mpa corresponding to a water depth of 2700 m. This depth is slightly overestimated, if compared with the actual depth of the submarine outcrop (2320–2340 m b.s.l).
 Despite this small discrepancy, attributable to analytical errors and/or solubility models, these results exclude the possibility that a vent located at significant shallower water depth fed the flow, and that appreciable subsidence or slumping phenomena took place after its emplacement.
 Marine geology data coupled with volcanological observations and volatile content in the glass, demonstrate that the lava flow-field found at about 2300 m b.s.l. on the northern flank of the Stromboli edifice is related to a submarine effusive eruption originating from a nearby eruptive fissure. This is the first finding of a submarine eruption on the northern side of the volcano and contributes to improve the knowledge of its flank activity.
 The small ridge in the bathymetric map just upslope from the lava flow-field can be interpreted as a N-trending eruptive fissure (Figure 1b). The inferred N–S direction differs from the main N and NW-striking regional tectonic trend [Falsaperla et al., 1999]. In the last 13 ka dykes with NNE orientation occur around the SdF depression, apparently favored by local unbuttressing associated with the sector collapse [Tibaldi, 2003]. However we interpret that this local modification of the stress field induced by the presence of the collapse scar does not extend down to the foot of the volcano edifice as far as 9 km from the shoreline.
 The preservation of primary flow structures on the pillow surface, the lack of hydration, and the limited cover of hemipelagic sediments all suggest a relatively young age for this lava flow. Bulk rock composition of the lava differs significantly from present day summit products and from those erupted during the historical flank eruption of S. Bartolo (360 BC–100 AD). Whole rock and glass compositions of pillow lavas show a high K2O content and a general geochemical pattern more similar to the less evolved K-alkaline products of the Neostromboli period erupted between 13 and 6 ka. However, the silica content is outside of the known compositional range of Neostromboli, hence the pillow lava could represent a not yet documented stage of the structural evolution of the volcano.
 The compositional evolution within the subaerial Neostromboli products is relatively well constrained, including the early stages of the period, represented by the explosion breccias of Breccia Frontone [Hornig-Kjarsgaard et al., 1993]. On the other hand, the early stages of development of the Recent Stromboli period after the Neostromboli collapse are still poorly known [Rosi et al., 2000; Speranza et al., 2008]. In this perspective, the flank vent of pillow lavas could be the “missing link” of early Recent Stromboli evolution by representing the beginning of this new intrusive cycle occurring between 6 and 2.4 ka. Similar to the model of Takada , initiation of the Recent Stromboli period, may have started with a submarine lateral eruption at the periphery of the edifice. The unique composition of the pillow lavas would may have resulted from the interaction between magmas feeding the new cycle and batches of previous Neostromboli magmas, still present after the emptying of the Neostromboli magma chamber.
 On the whole, deep submarine flank eruptions have a very different structural significance from that of flank eruptions occurring in the subaerial part of the volcano. In fact the latter are related to intrusions radiating from very shallow magma reservoirs or conduits and are induced by an overpressure of the shallow reservoir or by the modification of the local stress field associated with flank instabilities. Recent examples, though at small scale, are the 2002–2003 and 2007 lava flows fed by eruptive fissures connected at very shallow levels with the central vent plumbing system [Landi et al., 2006]. Vice versa the newly discovered submarine, lateral eruption can be associated to an intrusion not connected to the central conduit.
 These scenarios have different hazard implications: while flank eruptions on the subaerial portion could have an obvious direct impact on inhabited areas, submarine eruptions have the potential to trigger slope instability. Processes of flank inflation, opening of the eruptive fissures and gravitational loading due to the lava flows piling-up, are all able to induce sediment destabilization and possible flank failure.
 In a polygenetic island volcano like Stromboli, both structural evolution models and hazard assessments cannot overlook information related to the submarine portion. Thus additional systematic surveys aimed at documenting volcanic features linked to submarine flank activity are clearly important.
 We thank the captains and crew of R/V Urania and Logachev. M. Ivanov and the TTR14 cruise scientific parties are also thanked for their help in data acquisition. K. Putirka and B. Chadwick are acknowledged for their constructive reviews. This work was funded by INGV and the Italian Civil Protection Department (Research project V2). Topolin topolin!