The 7 September 2008 Vulcanian explosion at Stromboli volcano: Multiparametric characterization of the event and quantification of the ejecta

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 117,, doi: /2011jb009048, 2012 The 7 September 2008 Vulcanian explosion at Stromboli volcano: Multiparametric characterization of the event and quantification of the ejecta Sonia Calvari, 1 Ralf Büttner, 2 Antonio Cristaldi, 1 Pierfrancesco Dellino, 3 Flora Giudicepietro, 4 Massimo Orazi, 4 Rosario Peluso, 4 Letizia Spampinato, 1 Bernd Zimanowski, 2 and Enzo Boschi 1 Received 24 November 2011; revised 12 March 2012; accepted 16 March 2012; published 2 May [1] On 7 September 2008 a major ash explosion occurred from the SW summit crater of Stromboli volcano. This explosive event lasted 2 min and consisted of three discrete eruptive pulses, forming an eruptive ash cloud m high and 300 m wide, rising with speed of m s 1. The event was recorded by our camera and seismic networks, as well as by two electric stations installed at a 500 m mean distance from the SW crater. The electric signals recorded by the two stations during this event were 10 6 times greater than signals recorded during the persistent Strombolian activity, and the seismic trace had a bigger amplitude and a longer duration. Camera image analysis allowed us to infer that a partial obstruction took place at the SW crater three days before the explosive event, suggesting that a constriction within the upper conduit could have likely led to magma overpressure. Data analysis, combined with previous experimental investigations, revealed that the higher energy output of the ash explosion, when compared to the persistent Strombolian activity, resulted in a greater magma fragmentation and erupted mass. Integration of the different parameters allowed us to classify the event as a Vulcanian type, and electric signal analysis enabled retrieval of the total volume of erupted ash and of the amounts of the juvenile, phreatomagmatic, and lithic components. Citation: Calvari, S., R. Büttner, A. Cristaldi, P. Dellino, F. Giudicepietro, M. Orazi, R. Peluso, L. Spampinato, B. Zimanowski, and E. Boschi (2012), The 7 September 2008 Vulcanian explosion at Stromboli volcano: Multiparametric characterization of the event and quantification of the ejecta, J. Geophys. Res., 117,, doi: /2011jb Introduction [2] Stromboli is an island less than 4 km wide and 924 m high characterized by persistent, mild explosive activity from the summit craters, which are located at 750 m elevation within a depression 300 m long, 50 m wide, 50 to 100 m deep, and elongated toward NE-SW. The position of the summit craters has been remarkably constant over the years [Washington, 1917], resulting probably from the 1 Istituto Nazionale di Geofisica e Vulcanologia, sezione di Catania Osservatorio Etneo (INGV-OE), Catania, Italy. 2 Physikalisch-Vulkanologisches Labor, Universitaet Wuerzburg, Wuerzburg, Germany. 3 CIRISIVU, Dipartimento di Scienze della Terra e Geoambientali, Università di Bari, Bari, Italy. 4 Istituto Nazionale di Geofisica e Vulcanologia, sezione di Napoli Osservatorio Vesuviano (INGV-OV), Napoli, Italy. Corresponding Author: S. Calvari, Istituto Nazionale di Geofisica e Vulcanologia, sezione di Catania, Piazza Roma 2, I Catania, Italy. (sonia.calvari@ct.ingv.it) Copyright 2012 by the American Geophysical Union /12/2011JB steady state supply from the source region during the last two millennia [Rosi et al., 2000], combined with paucity of local tectonic earthquakes [Falsaperla and Spampinato, 1999]. Only the number and size of vents and their temperature have varied with time, and this is apparently due to changing magma level within the conduit [Calvari et al., 2005a, 2005b; Burton et al., 2008]. Following Patrick [2007] and Patrick et al. [2007], persistent explosions at Stromboli can generally be classified into two groups: the normal Strombolian activity (or Type 1), and ash-rich explosions (Type 2). The former are dominated by rhythmic gas bubble bursts and in-flight ductile magma fragmentation [Walker and Croasdale, 1971; Wright et al., 2007] that emit coarse ballistic particles, whereas the latter generate optically thick, ash-rich plumes with (Type 2a) or without (Type 2b) ballistic particles that can display plume rise rates covering both gas thrust (>15 m s 1 ) and buoyant (<15 m s 1 ) regimes [Patrick, 2007]. [3] At a mean rate of twice per year until 2002, more powerful major explosions occur [Bertagnini et al., 1999; Corsaro et al., 2005; Andronico et al., 2008; Landi et al., 2008; Andronico and Pistolesi, 2010], with ejection of 1of17

2 Figure 1. (a) Location of the Aeolian Volcanic Arc (AVA) and Stromboli in Southern Italy. (b) Island of Stromboli with the Sciara del Fuoco (SdF) depression and the position of the seismic stations (blue dots). The area enlarged in Figure 1c is also located. (c) The NE-SW summit crater depression, with the location of the SPI (Stromboli Infrared thermal camera at Il Pizzo) and SQV (Stromboli Visible camera at 400 m elevation) cameras. (d) The position of the active vents within the crater terrace as could be seen from the SPI camera located at Il Pizzo, where S area, C area and N area represent the SW, Central and NE craters, as of August Figures 1a 1c are courtesy of M. Neri [modified after Neri and Lanzafame, 2009]. lithic and juvenile material well outside the crater terrace [Calvari and Pompilio, 2001]. Much more powerful explosions, named paroxysms, occurred several times during the last century [Rittmann, 1931; Barberi et al., 1993], and recently in 2003 and 2007 [Calvari et al., 2006; D Auria et al., 2006; Rosi et al., 2006; Harris et al., 2008; Calvari et al., 2010, 2011], often causing severe damage to the settled area. Compared to the general extent of explosive events in volcanology, the scale of major explosions and paroxysms at Stromboli is quite small, and even in the case of paroxysms the volume of erupted pyroclasts rarely exceeds 10 6 m 3 [Bertagnini et al., 1999; Pistolesi et al., 2008]. However, given that the short distance between craters and villages is less than 2 km, and that between craters and the Il Pizzo summit viewpoint is less than 300 m (Figure 1), and considering the thousands of tourists climbing the summit of the island every month for at least 7 months per year, it is crucial to raise a prompt alarm in case of strong explosions. Although a number of studies focused on the distinction between major explosions and paroxysms at this volcano, the nature of their diversity is still a matter of debate [e.g., Bertagnini et al., 1999; Calvari et al., 2006; Andronico et al., 2008]. From a volcanological point of view, the main difference is the involvement/non-involvement of the gas-rich, low-porphyritic (LP) magma during paroxysms, rising fast from 6 10 km depth. This is erupted commonly during paroxysms, whereas only gas-poor, high-porphyritic (HP) magma, residing in the upper part of the conduit, is normally erupted during major explosions [Métrich et al., 2005, 2010; Calvari et al., 2011]. Major explosions and paroxysms also differ by the size of the eruptive column, which reaches a few hundred meters during major explosions, and a few kilometers during paroxysms [e.g., Calvari et al., 2006; Andronico et al., 2008; Harris et al., 2008; Andronico and Pistolesi, 2010; Calvari et al., 2010, 2011]. A Vulcanian eruption style was recognized for the two most recent paroxysms at Stromboli in 2003 and 2007 [Calvari et al., 2006, 2010], based essentially on the high exit velocity of the eruption plume, on the size of the lithic components, and on the obstruction of the crater before the explosive event [Morrissey and Mastin, 2000; Formenti et al., 2003]. [4] Major explosions and paroxysms, at Stromboli, generally produce volcanic convective plumes. These are threephase mixtures composed of variable proportions of solid particles (comprising fragments of rock, crystals, glassy shards and vesicular particles), volcanic gases, aerosols, and droplets of condensed volcanic gases and atmospheric water vapor [Sparks et al., 1997]. The presence of large electric potential gradients in volcanic plumes is well known [Sparks 2of17

3 et al., 1997; James et al., 1998], especially for plumes rich in solid silicate particles [James et al., 2008], and is often demonstrated by the occurrence of spectacular lightning [Anderson et al., 1965; Porarinsson, 1976]. Certain electrical effects during volcanic explosions can be qualitatively and quantitatively explained by fragmentation processes occurring in the melt [Büttner et al., 1997; James et al., 1998]. Low-viscosity basaltic magmas do not generally produce large plumes because they allow exsolving gases to escape easily and less explosively than high-viscosity silicic magmas. Thus, basalt fragmentation is often considered similar to a ductile liquid spray process rather than of brittle failure, producing mostly relatively large liquid clots instead of ash [James et al., 2008]. However, violent explosive basaltic explosions can occur when the ascending magma encounters water (either from the sea, or an aquifer, ice, etc.), giving rise to violent phreato-magmatic eruptions [Büttner and Zimanowski, 1998; Zimanowski, 1998], or even when the conduit is obstructed, causing Vulcanian explosions [Morrissey and Mastin, 2000; Calvari et al., 2006, 2010]. Phreato-magmatic eruptions generate significant electrification, by fine ash brittle fragmentation that is caused by an increased strain rate and water cooling on the magma [Büttner and Zimanowski, 1998; James et al., 2008]. However, laboratory and field experiments carried out at Stromboli volcano during Strombolian-type explosions have shown that effective electric charge generation, related to fragmentation of magma and generation of pyroclasts, can be detected and measured on a short timescale [Büttner et al., 1997, 2000]. The surface area that is generated during fragmentation can be estimated with the use of grain-size particle analysis, and increases linearly with explosion intensity [Büttner et al., 1997]. Büttner et al. [2000] found that the voltage-time ratio of electrostatic field gradients reflects different physical mechanisms of magma fragmentation and expansion. Measurements of the electrostatic potential gradient (i.e., maximum voltage) have been found to be independent on the shape and position of the array, provided that an upwind position is chosen, in order to avoid any interference with the eruption cloud [Büttner et al., 1997]. The maximum voltage U max varies with the distance D, and can be described by U max 1/D [Büttner et al., 2000]. Comparison of experimental data showed that the typical and reproducible delay time between the onset of electric signals and onset of seismic signals ranges between 2 and 3.5 ms for the phreatomagmatic type explosions, and between 12 and 14 ms for Strombolian-type bursts, where fragmentation is delayed as it follows the expansion process [Büttner et al., 2000]. [5] With the aim of monitoring Stromboli s summit explosive activity also during poor weather conditions, i.e., when the craters are completely obscured by clouds and even the fixed thermal cameras are blind, on 4 September 2008 we installed two 5 m-spaced-electric stations on the volcano summit at a 500 m mean distance from the western crater rim and close to the STR9 seismic station (Figure 1). These devices were built for Stromboli following the procedure described by Büttner et al. [2000], and were installed a few days before the occurrence of a major explosive event on 7 September 2008, for which also time-lapse images collected by the camera network, and seismic data, were available. Thus, we classify this ash explosion on the basis of data gained from a multiparametric monitoring system, including (i) real-time recording of explosive activity by videos from two view points, (ii) broadband seismic data, and (iii) traces of the two electric stations (Figure 1). For comparison purposes, we also use data that were collected during the volcano typical persistent Strombolian activity, and also retrospectively, with the available data of the 5 April 2003 and 15 March 2007 paroxysms [Calvari et al., 2006, 2010, 2011; D Auria et al., 2006]. Additionally, using results from experimental investigations on magma fragmentation and transport mechanisms during explosive eruptions [Dellino et al., 2010], we classify the 7 September explosion as a major explosive event displaying a Vulcanianlike behavior. Finally, we provide quantification of the volume of the erupted material and its relative components, and tentatively assess the fragmentation processes driving ash explosions at Stromboli. 2. The Monitoring Network [6] The Istituto Nazionale di Geofisica e Vulcanologia (INGV) monitoring system on Stromboli has been greatly improved after the effusive eruption [e.g., Martini et al., 2007; Bertolaso et al., 2008; De Cesare et al., 2009; Salerno et al., 2009; Zanon et al., 2009]. In addition, on 4 September 2008 two electric stations for the measurement of atmospheric electrical potential gradient were installed close to the summit craters in order to monitor the summit explosive activity. [7] In this paper, we use data from the monitoring camera and seismic networks, and the electric stations. The camera network is maintained by INGV - Osservatorio Etneo and consists of four cameras: one installed at Il Pizzo at 918 m a.s.l., one of the topographic highs of Stromboli 170 m above the summit craters, and three located along the eastern margin of the Sciara del Fuoco (SdF; Figure 1). For this study we used images from the summit infrared camera (SPI) located at Il Pizzo, and from a visual (SQV) camera located at 400 m (Figure 1). The SPI infrared camera field-of-view (FOV) covers the entire crater terrace from South and from an inclined distance of m, depending on the elevation of the crater floor (Figure 1). It enables imaging of the explosive vents and tracking of the height reached by the products up to a maximum of 160 m above the vents, as well as discrimination between the predominant types of ejecta (ash, lapilli, bombs or blocks). SPI is an OPGAL EYE-M320B thermal camera sensitive to the 8 to 14 mm wave band. The detector consists of an uncooled microbolometer recording a pixel frame every 2 s. The camera has a FOV, corresponding to a targeted pixel size of 0.9 m over a distance of 250 m. This camera is not calibrated for temperature retrieval. The visual SQV camera gives an oblique view of the NE flank of the summit cone from below and from 1 km distance (Figure 1). It has a favorable position, covering all the eruptive fissures that opened after 1985 and breached the NE flank of the crater terrace [De Fino et al., 1988; Calvari et al., 2005a, 2005b, 2010]. The SQV camera allows observation of the eruptive plume from a greater distance (1 km) and a different perspective (from NE rather than from South) when compared to SPI, thus enabling detection of the eruptive cloud up to a maximum height of 350 m above the vents. Video sequences are recorded 3of17

4 Figure 2. Graph of explosive activity. (a) Mean total daily number of explosions h 1 between January and September 2008, obtained from SPI infrared camera located at Il Pizzo, with a two-point running mean (red line). The 7 September point is displayed in yellow. (b) Details of the mean daily number of explosions h 1 between January and September 2008 for each of the three crater zones, North (N, red), South (S, blue), and Central (C, green). with a 2 s time interval, and transferred to the INGV monitoring room, where they are visualized and stored [e.g., Andò and Pecora, 2006; Behncke et al., 2009]. Each of the cameras provides continuous, real-time monitoring of the eruptive activity. Images recorded from the cameras show date (dd/mm/yy) and time (hh:mm:ss) expressed in UTC, and are synchronized through a Global Positioning System time-code. [8] The seismic network is maintained by INGV - Osservatorio Vesuviano and consists of 13 telemetric broadband stations (Figure 1). Data are acquired and analyzed at the Napoli and Catania recording centers [De Cesare et al., 2009]. This network allows monitoring of the seismicity associated with the eruptive activity of the volcano and of the landslides that affect the cliffs of the island. Usually, the network records a few hundred seismic events per day related to the persistent Strombolian explosions [Martini et al., 2007]. These events contain a Very Long Period (VLP) component with a frequency in the range Hz, which relate to the migration of gas slugs in the 4of17

5 Figure 3. (a) Grain-size distribution and (b) electric calibration used for the recalculation of the ash mass erupted. MFCI = melt fuel coolant interaction. Voltage represents the horizontal electrical potential gradient. The calibration factors result from the voltage measured in the fragmentation regimes. See text for further explanations. conduit, and is consistent with volumetric components within the seismic source of the VLP signals [Chouet et al., 2003, 2008; James et al., 2006]. During Strombolian activity, the location of VLP is clustered in a small volume below the shallow portion of the SdF (Figure 1) at a surface elevation of m a.s.l. When effusive phases occur, the location of the VLP sources undergoes small but significant spatial changes both laterally and vertically [Martini et al., 2007; Giudicepietro et al., 2009]. [9] In order to characterize the explosive activity and its products, on 4 September 2008 two permanent electric stations were installed at the summit of Stromboli. They were located at 5 m distance between each other, 500 m West of the SW crater, and close to the STR9 seismic station (Figure 1). Each electric station consists of a calibrated electrometer, recording the local electrical potential gradient between a sensor grid mounted on a 2 m high pole and the local ground, and a linear dc signal amplifier [Büttner et al., 2000, 2002]. The analog signals of the electrostatic sensors are recorded at a rate of 250 sps (samples per second) by a high resolution (24-bit), low power (below 850 milliwatts) data logger (GILDA) [Orazi et al., 2006]. Data are transmitted to a collection hub using the WiFi infrastructure deployed since 2005 to improve the monitoring systems of Stromboli [De Cesare et al., 2009]. The collection hub is located at the Observatory of S. Vincenzo, which is the Centro Operativo Avanzato (COA) of the Italian Civil Protection. From there, data are transmitted to the INGV recording centers of Napoli and Catania. 3. Methods [10] The SPI camera allows a 24 h, complete view from South of the summit crater zone. From this view point, we distinguish three sectors (Figure 1d): the NE crater zone (N area), the Central crater zone (C area), and the SW crater zone (S area). Using images from the SPI camera, we have manually counted the daily number of explosions occurred at the whole summit zone and obtained the averaged daily values per hour (Figure 2a), as well as the averaged daily number of explosions occurring at each of the three crater areas from January to the end of September 2008 (Figure 2b). The values are approximated to the next integer, and variability is 20%. [11] For the electric signals, we found that at Stromboli the short-time fluctuations of the atmospheric electric field are caused by the formation and ejection of ash and produce a strong signal (signal-to-noise ratio S/N > 10 6 ). This is because, on explosion, the formation and rapid decoupling of new solid surfaces perturb the charge distribution of the electrostatic field [Büttner et al., 1997]. The extent of perturbation is proportional to the new surface that generates per time, and thus it is proportional to the size, amount, type, and kinetic energy of the particles. The electric signal provides therefore information on the nature of the eruptive mechanism responsible for ash emission (either magmatic, phreato-magmatic, or by passive collapse), on the release of mechanical energy, and on the amount of ash produced [Büttner et al., 2000]. For a first approximation, data from experiments using basaltic melts are used as references to estimate the amount of charged surface that formed the eruptive column. Measurements during fragmentation experiments showed that kinetic energy release, new surface area, and electrical potential gradient are linearly correlated [Büttner et al., 1997]. Data obtained from the fragmentation experiments by phreato-magmatic fragmentation, magmatic fragmentation, and passive ejection of ash from a conduit result in three linear ratios, which can be used to derive the kinetic energy release and total new surface area from the reconstructed horizontal electrical potential gradient within the eruption cloud (Figure 3 and Table 1). The partitioning of the total surface area into phreato-magmatic (MFCI, melt fuel-coolant interaction), magmatic, and passively ejected pre-existing particles can be estimated (if data from particle analysis by Scanning Electron Microscope is not available) by extracting the value of the potential gradient at the characteristic voltage-time ratio, thus estimating the partial surface area and finally the partial kinetic energy release. To approximate energy and mass of ash, in the case of the 7 September 2008 explosion, a typical grain-size distribution for ash erupted from Stromboli was used [Andronico et al., 2008]. The calculated particles surface was distributed 5of17

6 Table 1. Parameters of Electrical Signals at Stromboli on 7 September 2008 a Parameter Value Total Kinetic Energy Explosive Event at 07:49:00 Voltage time ratio 600 V/s Local field gradient 1.8 kv/m 0.8 kg MFCI-particles 750 kj 80 kg juvenile magmatic ash 3,500 kj kg lithic ash 15,000 kj Total kinetic energy kj 4.6 kg TNT Explosive Event at 11:53:00 Voltage time ratio 76 V/s Local field gradient 0.3 kv/m MFCI-particles 13 kg juvenile magmatic ash 560 kj kg lithic ash 2,400 kj Total kinetic energy kj 0.7 kg TNT Explosive Event at 12:25:00 Voltage time ratio 76 V/s Local field gradient 0.25 kv/m MFCI-particles 11 kg juvenile magmatic ash 480 kj kg lithic ash 2,070 kj Total kinetic energy kj 0.6 kg TNT Explosive Event at 19:17:00 Voltage time ratio 182 V/s Local field gradient 0.76 kv/m MFCI-particles 34 kg juvenile magmatic ash 1,470 kj kg lithic ash 6,300 kj Total kinetic energy kj 1.6 kg TNT a The estimated particle mass and energy values are in the order of 10%. MFCI = melt fuel coolant interaction. to cubes of 125 mm 3, and a density of kg m 3 (for nonvesicular magma) was assumed. On this basis we obtained the values reported in Table The Explosive Activity Prior to the 7 September Event [12] The graph in Figure 2a shows the variability in the number of explosions at the three crater areas from January to late September The values reported are daily average approximated to the next integer, and variability is 20%. The explosive activity general pattern is characterized by marked fluctuations describing cycles of increasing and decreasing of the intensity of the eruptive activity. These fluctuations over time scales of tens of minutes and vertical distances of tens of meters are common at Stromboli, and are related to changes in the magma free surface level or in the gas jet velocity [e.g., Ripepe et al., 2002; Burton et al., 2008]. In detail, from January until mid-march the number of events ranged between 14 and 5 events h 1 ; then, it rapidly declined down to a minimum of 3 explosions h 1 on 23 March. This was followed by a new trend that increased until mid-august and reached events h 1. Between 15 and 22 August, the trend of explosions decreased again fluctuating from 7 to events h 1. Figure 2b shows the number of events at each of the three crater areas providing details on the relationship between the three sites. In particular, the graph reveals that the C zone was not active during most of the period, and that the N and S zones displayed generally contrasting behavior, with the explosive activity at the N zone having greater variability (average of 1 13 events h 1, Figure 2b) than that at the S zone (average of 3 10 events h 1, Figure 2b). From early July to early August, the explosive activity increased at both the N and S areas, then showed an opposite trend until 19 August. From this date onward and for about one week, explosions at the N and S crater areas displayed both pulsating decreasing trends (Figure 2b). In early September, the explosive activity at the S area climaxed prior to the 7 September event with a peak of 10 events h 1. Conversely, the N crater area activity declined to a minimum of 4 explosions h 1 (Figure 2b). The 7 September event occurred from the S area while the N zone was characterized by a decreasing explosive activity. It is worth noting that a few days after the 7 September explosion, the C area, that had a quite rare explosive activity before this date on the timescale of the observations, showed an increasing trend in the mean daily number of explosions (Figure 2b). [13] In early September, when the average daily rate of the total explosive events was 14 events h 1 (Figure 2a), the volcanic tremor was within the typical value of Stromboli, and six vents were active within the summit depression, two for each of the summit crater zones (Figure 4a). The eruptive activity was characterized by puffing from the two central vents (bc1 and bc2, Figure 4a); ballistic coarse scoria ejection (Type 1 Strombolian explosions of Patrick et al. [2007]) from vents at the N crater zone (bn1 and bn2, Figure 4a), where the height of explosions was up to 60 m from the crater rim; and by sporadic explosions, producing ash (Type 2a and 2b Strombolian explosions) [Patrick et al., 2007], from the S vents (bs1 and bs2, Figure 4a), where ash plumes were rising a few tens of meters from the vent rim. Small morphological changes occurred within the crater depression starting from 28 August, when small rockfalls from the NW crater wall occurred, and an explosion from the N cinder cone widened the bn2 vent (Figures 4a and 4b). On 29 August, explosions from the S vents were mostly of Type 1, and their height increased to 50 m, with small amount or no ash involved. Puffing from the central vents moved toward the southernmost one. At the southern vents, the increase of height of ejecta, which reached 100 m from the vent rim, together with a better collimation of ejecta, suggested increasing magma surface depth within the upper conduit [Taddeucci et al., 2012]. On the early morning (02:00:00 03:00:00 A.M.; all times here reported are in UTC) of 30 August, a new decrease of the ejecta height suggested an uprise of the magma level. This upward migration of the magma level produced a change in the eruptive style observable at the C vents, with an increasingly brighter (hotter) puffing that eventually evolved to a mild, continuous spattering. The rising of the magma column resulted also in a transition of activity at the S crater zone, where explosions passed initially from Type 2b to Type 2a Strombolian types, and then from Type 2a to Type 1 [Patrick et al., 2007]. Additionally, a new vent (bs3) opened at the tip of the cinder cone where bs1 was located (Figure 4c). This was characterized by mild spattering. In the meanwhile, the number of explosions at the N crater zone decreased. [14] On 31 August puffing was observed at vents bc2 and bs1, and at bc1 in the second half of the day, whereas 6of17

7 Figure 4. (a e) Images taken from the SPI camera looking at the crater terrace from South-East and from a distance of m, showing the whole crater area. NE is on the right, SW on the left (see Figure 1 for orientation). The active vents within each crater zone are shown, with bn1 and bn2 belonging to the N area, bc1 and bc2 indicating the two closely spaced vents of the C area, and bs1 and bs2 the active vents in the S area. Figures 4a and 4b show erosion of vent bn2 due to explosions blowing out the western upper flank. In Figure 4c bs3 appeared in the afternoon of 30 August between bs1 and bs2 in the S area. (f ) Thermal image collected from Il Pizzo with a portable thermal camera showing the map of temperatures within the summit craters as on 4 September 2008, and (g) corresponding visible photo taken from the same position and on the same date. Type 1 Strombolian explosions were produced by bs2, bn1 and bn2. Vent bs1 showed the most variable intensity and eruptive regime, passing, over a time-span of minutes, from no activity to weak, passive degassing, puffing, mild spattering, and Type 1 explosions. Increases in the activity at bs1 were usually accompanied by high puff frequency at vents bc2 and bc1, suggesting a shallow connection between them. On the morning of 1 September, ash explosions of Type 2a resumed at bs2 with high frequency of the events, while Type 1 explosions were occurring at the other vents. The variable level of the eruptive activity, and thus the amount of erupted volumes of pyroclasts at individual vents, resulted in a different height of the cinder cones, that on early September were higher on the N zone, where the rim of bn2 was 20 m higher than the rim of bs2, and with the lowest bc vents 10 m below the S vents (Figure 4d). This resulted in a difference in elevation between bn2 and bc rims of 30 m. Mild spattering and Type 1 explosions started by noon of 1 September at bs1 and bs3, accompanied by puffing and sometimes by mild spattering also 7 of 17

8 Figure 5. Explosive sequence on 7 September 2008 recorded by the (a f) SPI camera, and by the (g l) SQV camera located on the east flank of the SdF. from bc1 and bc2. In the afternoon of the same day, Type 2a explosions were sometimes occurring also at bn1, while Type 1 events were observed at bn2, and puffing at bc1, bc2, bs1 and bs3, alternating to spattering at the last three vents. Type 1 explosions and spattering increased on early 2 September at the bc and bs vents (Figure 2b), with almost vertical jets of spatter and two additional vents (bs4 and bs5) opened in the S crater zone (Figure 4d). The gradual opening of a great number of small vents suggested deepening of magma level within the conduit [Spampinato et al., 2008]. During the afternoon of the same day, magma level became shallower, with very frequent and intense spattering at bs1 and bs3 accompanied by ash emissions, suggesting erosion and widening of the vents [Spampinato et al., 2008]. This activity caused the merging of bs1 and bs3, forming a wider bs1 (bs1+3, Figure 4e) vent, and was punctuated by Type 1 explosions, with bombs reaching m in height from bs1, and up to 150 m at bs2. Later on the same day also bs4 and bs5 merged together (Figures 4d and 4e). [15] Images between 06:30:00 of 3 September and 08:15:00 of 4 September are lacking because of a failure in the transmission system of the camera network. The first images available on 4 September showed that the three S vents had widened, with the previously merged bs1+3 (Figure 4e) and a new vent (bs6) opened to the NW of bs2 (Figure 4e). This latter vent did not show any explosion. After noon on the same day, spattering became increasingly more common at bs1 with ejecta up to 50 m in height, spreading all around the vent (Figures 4f and 4g) and causing a wide dispersion of products and producing the fast growth of the S cinder cone. In the meantime, also bc2 fed spattering, while bs2 and bs5 were displaying Type 2 explosions up to 100 m in height, suggesting erosion of the conduit and deeper magma level. The explosive activity increased between the evening of 4 September and early 8of17

9 12:30:00 of 6 September and 06:11:00 of 7 September, when bs1 looked widened toward bc2 (Figure 5a). Figure 6. Photos courtesy of Tullio Ricci (INGV-Rome), taken from Il Pizzo, and comparing the crater morphology (a) before and (b) after the 7 September explosive event. The dotted yellow line shows the rim of bs1 vent. Note in Figure 6b the much wider vent depression, and the accumulation of pale rocks (lithics) around bc2, which became also higher than bc1 by deposition of fallout from the 7 September explosion. 5 September at all vents, and especially at bc2 and bs1, with Type 1 explosions producing fast cinder cone growth. Activity decreased again by the end of 5 September. On 6 September, explosions at the S crater zone did not rise vertically but were inclined eastward, thus having a greater horizontal component and expanding toward the inner crater, leading to a fast accumulation of spatter on the slope between bc2 and bs1. Spatter accumulation resulted in a partial obstruction of the bs1 vent, whose diameter decreased significantly between 3 and 6 September, as observed directly via the camera network (Figures 4c 4e). The change of eruptive jet spreading direction, although observed on other occasions [e.g., Zanon et al., 2009], is rather unusual, and did not occur during the previous days, when the explosion jets were always directed outwards from the center of the crater depression. This change thus suggested modifications of the inner structure of the S shallow conduit or partial obstruction of the vent. The camera network did not transmit between 5. The 7 September Explosive Event [16] On 7 September 2008 at 07:49:10 a strong ash explosion started at the S area of Stromboli (Figures 5a 5l). This was recorded by SPI and SQV. The cameras FOV allowed description of the event from both a close-up and frontal perspective (SPI, Figures 5a 5f), and from a 1 km distance and inclined view (SQV, Figures 5g 5l). The view from SPI permitted the characterization of the initial phases of the explosive event. At 07:49:02, bs1 fed a low-energy spattering, and suddenly, at 07:49:10 two diverging explosive jets made of hot, juvenile material (Figure 5a) originated from bs1 and from bs3, the new vent just-opened between bs1 and bc (Figure 4c). The jet from bs1 was inclined toward SW, whereas the jet from bs3 was inclined toward NE, with the latter slightly bigger than the former (Figure 5a). The two jets expanded outwards the crater with maximum estimated speeds of 20 m s 1 for the SW jet, and 27 m s 1 for the NE jet, calculated on the basis of the time-lapse frames (Figure 5b). The eruption cloud from bs3 spread also laterally, and 2 s later (Figure 5c) displayed convective movements and formed a vigorous vortex ring giving rise to a rooted thermal plume [Patrick, 2007]. A few seconds later, the ring appeared also at the top of the eruption column developed above bs1 (Figures 5d 5f). The dark color (in the infrared image) of this cloud suggested that it consisted essentially of lithic (cold) ash. The two plumes became cold where diverging (dark portion in Figures 5c 5f), implying likely fragmentation to ash size and involvement of lithic, cold, country rock material, with the juvenile, hot (white in Figures 5a 5f) portions confined to the initial and outer parts of the spreading clouds. Two seconds after the blast, coarse ballistics were evident (Figures 5b 5e), being more abundant and bigger from the cloud spreading NE (Figure 5d, right). [17] The view provided by SQV showed that the explosive episode consisted of three major pulses (Figures 5g 5l). The first corresponded to the sequence just described as observed from SPI and occurred 1 min earlier, then the ash cloud obscured the sight from SPI. The second was at first visible from SQV at 07:50:03 and became more evident a few seconds later (Figure 5j). The third pulse started at 07:50:27 and formed a plume that expanded after a few seconds (Figure 5k). By 07:51:07 the eruption cloud was already rising up and detaching from the crater (Figure 5l) forming a thermal plume [Patrick, 2007]. Overall, the explosive event lasted 2 min, and formed an ash plume that merged the two initial jets (Figure 5j), slowed down to an average rise rate of 18 m s 1, and reached an estimated maximum height of m and width of 300 m. The ash cloud drifted SW with the wind, and disappeared from SPI camera view at 07:50:20, and at 07:51:11 from that of SQV. [18] A few minutes later, mild Type 1 explosions resumed from bn2, while bs2 fed Type 2b events, and bc produced puffing. Direct field observations made at the summit crater zone (T. Ricci, personal communication, 2008) reported that no golden pumice (gas-rich, LP magma) [Bertagnini et al., 1999] were erupted, and that visible morphological changes at the summit vents had occurred (Figures 6a and 6b) 9of17

10 Figure 7. Images recorded on 7 September 2008 by the SPI camera at (a) 07:49:10, (b) 07:49:36, and (c) 07:50:02. (d) Electric (red and black lines) records of the eruption. (e) Comparison between the seismic (green line) signal, and seismic filtered signal for highlighting the VLP component (blue), and the electric signals (red and black lines). The intensity for the electric signals is rescaled by 6 E10 3 and 6 E10 1 for a better comparison with the seismic signal. 10 of 17

11 Figure 8. Comparison between (a) the location of the seismic source of the VLP event associated with the 7 September 2008 major explosion (red diamond: Latitude ; Longitude ; elevation 502 m a.s.l.) and (b) the locations of the VLP events recorded on the same day (red area, comprising 360 events). The blue dots indicate the positions of the INGV-OV broadband seismic stations. with significant widening of the bs1 vent within the S cone, heightening of bc2 by fallout accumulation, and involving also lithic material from the conduit walls still visible around the base of the bc cinder cone (Figure 6b). No blocks or coarse ejecta reached the Il Pizzo zone, and most of the fallout material fell inside the crater depression, with only ash being drifted west by the wind. 6. Signals Recorded During the 7 September Explosive Event [19] Our observations of the 7 September event are summarized in Figure 7, where: in Figures 7a 7c we report three SPI images representing the main important phases of the eruptive cloud development; in Figures 7d and 7e the electric signals (red and black), and in Figure 7e the seismic signal (blue) filtered at Hz, highlighting the VLP event associated with the explosion (light blue line). The timing of the electric signals can be accurately linked to the optical detection of the eruption cloud, because no electrical effects are recorded before the ash explosion (Figures 7a 7e). This was also observed during previous electrical measurements carried out at Stromboli [Büttner et al., 2000]; thus, we can exclude the effect of the electrical variations in the soil, possibly connected to the presence of a shallow aquifer [Finizola et al., 2003, 2009], especially given that these signals display a different shape [Crespy et al., 2008]. Over the whole trend of the electrical signals, at least four main high-amplitude, short-lasting spikes are superimposed (Figure 7d). Considering the high amplitude and short duration, we tentatively interpret these as possibly due to lightning discharges produced within the ash cloud [e.g., Lane and Gilbert, 1992; Hoblitt, 1994; Bennett et al., 2010]. [20] The electrical potential gradient generated within the eruption cloud during the major explosion of 7 September 2008 was recalculated using data from both laboratory (fragmentation experiments using remelted volcanic rock) and field (using volcanic ash) experiments on a larger scale [e.g., Zimanowski et al., 1997; Zimanowski, 1998; Büttner and Zimanowski, 1998; Büttner et al., 2006; Dellino et al., 2007, 2010]. The decay of the horizontal potential gradient with distance from the source was determined, and a model of the electrical field was obtained. In detail, the best fit for the geometry of the field is a line charge model, and the decrease of the potential gradient U max with distance D is well approximated by U max 1/D [Büttner et al., 2000]. Thus, in the case of the ash explosion, considering a distance of 500 m from the source, and a U max measured at the site of 3.6 V/m, the maximum horizontal potential gradient in the eruption column of 1.8 kv/m is obtained. The overall voltage-time ratio that was found to characterize the main eruption mechanism is 600 V/s. This value is still in the range of magmatic fragmentation (10 2 V/s). However, during small increments of time, maximum local voltage time ratio values of 7 kv/s can be detected, approaching the typical range found for phreato-magmatic fragmentation (10 4 V/s) [Büttner et al., 2000]. [21] Using the estimated horizontal potential gradient of 1.8 kv/m and the voltage/surface area diagram of Figure 3, three total surface areas can be calculated for MFCI (11.3 m 2 ), magmatic ( m 2 ), and lithic particles ( m 2 ). These are distributed to the model grains (cubes), and result in the masses reported in Table 1. The calculations on particle surface using a grain-size distribution typical for Stromboli, as earlier described, result in 0.8 kg of MFCI particles (produced by phreato-magmatic fragmentation), 80 kg of juvenile ash (produced by magmatic fragmentation), and kg of passively ejected, lithic ash-sized particles. Although the estimated mass of MFCI particles is small, due to the small grain-size and the high specific surface, it contributes to the energy of the eruption. From specific energy-surface ratios taken from fragmentation experiments also the kinetic energy release can be calculated, being kj - i.e., 4.6 kg TNT equivalent (Table 1). By comparison, the same calculations have been performed on the electric signals recorded during three of the strongest events belonging to the persistent Strombolian activity, which occurred at 11:53:00, 12:25:00 and 19:17:00 on the same day (Table 1). They show a range of total kinetic energy between 0.6 and 1.6 kg TNT. These 11 of 17

12 Figure 9. Seismic trace recorded during the major explosion occurred at Stromboli on 7 September 2008, and a typical VLP waveform that occurred 10 min before. Counts (arbitrary units, y axis) versus time in sec (x axis), starting from 07:30:00. The recording seismic station is STR9, see Figures 1b and 1c for location. explosive events occurred at the S area and involved shortlasting ash clouds. It is interesting to observe that each explosive event included 1% of juvenile material, with most of the ejecta consisting of lithics. This is also evidenced by the low temperature (dark color) of the eruptive clouds detected from SPI camera (Figures 5c 5f). Only the major explosive event occurred at 07:49:00 erupted a small amount of MFCI particles, resulting in 0.01% of the total erupted mass (Table 1). [22] The correlation between the seismic and electrical signals is shown in Figure 7e. Comparison between electric and seismic signals is essential to distinguish electric signals produced by weather clouds from those produced by volcanic plumes. This comparison encouraged us to elaborate the data further, in order to interpret the mechanisms of major explosions and compare them with other explosion categories at Stromboli. Thus, given that the 5 April 2003 paroxysm displayed a deeper source origin when compared to the persistent explosions [D Auria et al., 2006], in order to compare the seismic signal of the 7 September major ash explosion with that of the persistent Strombolian activity, we have located the source of 360 VLP events recorded on 7 September 2008 by using the semblance technique [Martini et al., 2007]. This exploits the polarization properties of the signal (Figure 8a), with a particle motion pointing in the direction of the source. The comparison between the 7 September major event (red diamond in Figure 8a and yellow star in Figure 8b) and the persistent Strombolian activity (red area in Figure 8b) recorded on the same day is shown in Figure 8b by the red area around the yellow star. It is worth noting that the major event associated with the ash explosion, similarly to the other events of the persistent Strombolian activity, is located along the SdF slope at an elevation of 500 m a.s.l. On the other hand, by comparing the waveform of the VLP associated with the major ash explosion with that of a common VLP event that occurred a few seconds earlier (Figure 9), it is evident that the waveform of the major ash explosion is more complex, being characterized by a greater amplitude and a longer oscillation duration. This is indicative of a difference of both the time history of the source mechanism, and the energy of the signal. Major explosions occur frequently at Stromboli [e.g., Corsaro et al., 2005; Métrich et al., 2005; Andronico and Pistolesi, 2010], and show a seismic waveform similar to that of paroxysms, suggesting similar eruptive mechanisms, although their intensity is much lower [D Auria et al., 2006]. In addition, paroxysms display also an Ultra-Long-Period (ULP) signal that has never been recorded during the persistent summit activity [D Auria et al., 2006]. 7. Discussion and Comparison With Paroxysms Data [23] The 7 September 2008 major explosion at Stromboli occurred after several months of increasing explosive activity observed at the summit craters since the first half of March (Figure 2a). This general growing trend was essentially produced by increasing explosivity at the N area (Figure 2b), displayed at least until the end of June. Thereafter, the N and S areas showed an opposite behavior, with reduced explosivity at the N area being balanced by a corresponding increase at the S area. A few days before the major event, a sudden drop in the number of explosions at the N area was followed by a relative increase at the S area (Figure 2b). This opposite behavior at the two crater zones has been observed also in the temperature of the eruptive plumes measured during the 5 April 2003 paroxysm [Calvari et al., 2006], and was interpreted as revealing upper conduit instability, as well as shallow connection between the vents. A few days before the 7 September 2008 major 12 of 17

13 event, some important morphology changes occurred at the summit vents, leading to a visible reduction of the diameter of the S area vents (Figures 4a 4g) suggestive of partial obstruction of the S area. Many papers [e.g., Harris and Stevenson, 1997; Chouet et al., 2003; Burton et al., 2008] have already pointed to a very shallow connection among the Stromboli summit vents. Thus, it is possible that a decrease in the activity at one of the summit zones (in this case, the N area) might have resulted in the formation of a cooler crust that reduced vent diameter and acted as a partial obstruction of the upper feeder conduit. As a consequence, considering that Stromboli is a steady state system, to reequilibrate the total mass eruption rate of the system, eruptive activity increased at one of the other summit crater zones (i.e., bs1). However, the increased explosive frequency was too high to be sustained by only a small vent, which was then affected by inner collapses. These were displayed by changes in the ejecta direction [e.g., Cannata et al., 2009; Zanon et al., 2009], which spread inside the crater on 6 September. The vent constriction increased gas pressure within the magma column, until this was eventually suddenly released through a major explosion, blowing out part of the crater rim, as happened during the 7 September major explosive event (Figures 6a and 6b), and also during other previous strong explosions at Stromboli [e.g., Bertagnini et al., 1999; Andronico et al., 2008]. The fragmentation of cold crusted material during the 7 September event is revealed by the growing dark ash cloud visible between the two eruption plumes in the infrared images recorded by the SPI camera (Figures 5c 5f), and by the accumulation of lithic material around the base of the bc vents (Figure 6b). The contribution of lithic material to the formation of the ash cloud is supported by the results obtained from the electric signals (Figure 5 and Table 1). The same proportion between juvenile and lithic material is maintained during the persistent, mild explosive activity, the only difference with the major explosion being the total amount of ash erupted (Table 1), and the involvement of a small amount of MFCI ash by the major Vulcanian event. [24] The removal of the cold plug obstructing the crater during the initial phase of the explosive event (Figures 5c 5f), as well as the ejection of lithic material deposited around the vent (Figure 6b) and the erosion of the vent rim, are consistent with a Vulcanian-like eruption mechanism [Cas and Wright, 1988; Morrissey and Mastin, 2000; Formenti et al., 2003], that was invoked also for the 5 April 2003 paroxysm [Calvari et al., 2006]. Like previous major explosions [Bertagnini et al., 1999; Andronico and Pistolesi, 2010] and paroxysms [Calvari et al., 2006, 2010], also the 7 September event was characterized by an eruption column consisting of several pulses, although in this case only the S area was interested by the blast rather than the whole crater area. When compared to recent paroxysms, the 7 September 2008 event displayed lower muzzle velocity (20 30 versus 170 m s 1 ), smaller eruption plume ( versus m high), shorter duration (2 versus 7 min), and did not involve eruption of the deep-seated LP magma commonly erupted during paroxysms [Métrich et al., 2005; Calvari et al., 2006, 2010; Rosi et al., 2006; Harris et al., 2008]. The different scale of major explosions and paroxysmal events relates also to the different volume of erupted ejecta (10 4 versus 10 6 m 3, respectively), that during paroxysms produces lava fountains and pyroclastic flows [Calvari et al., 2006, 2010]. In addition, paroxysms cause ULP seismic events recorded several minutes before eruption onset, likely produced by the deformation of the volcanic edifice during magma ascent and eruption [D Auria et al., 2006; Martini et al., 2007; Casagli et al., 2009]. In addition, before paroxysms, a high frequency seismic signal is recorded and apparently related to magma fragmentation, but this was never observed during major eruptions [D Auria et al., 2006]. All these data together point out to a different source for paroxysms and major explosions, with the much deeper and fast rise of LP magma triggering paroxysms, whereas a shallow, partial vent obstruction is causing major explosive events. An alternative explanation is given by Del Bello et al. [2012], who suggest that paroxysms result from slug flow, occurring just on a much larger scale than the normal, persistent Strombolian activity. In this context, the rise of LP magma might increase the size of rising slugs, promoting paroxysms rather than mild Strombolian explosions. [25] It seems also possible that there is a major difference in the fragmentation process and gas decoupling style between paroxysms and major explosions. In fact in paroxysms, the presence of LP magma suggests that some gas was coupled with the magma as it rose, and fragmentation was internal. In addition, recent experiments have proved that the LP magma is more prone to explosively fragment than is the HP magma [Dürig et al., 2012]. In the 7 September 2008 major explosion instead, where apparently only HP magma was involved (T. Ricci, personal communication, 2008), the gas accumulated at the top of the conduit and was then decoupled from magma, thus suggesting again some kind of upper conduit obstruction or morphology change. Therefore, any fragmentation of magma in the conduit was probably driven by gas near, but not within, the magma. [26] The analysis of video sequences of the 7 September eruptive activity clearly shows that the ash explosion, and the associated plume, is to be attributed to a Type 2a event, which has a much smaller grain size when compared to the Type 1 persistent activity [Patrick et al., 2007]. This is clearly visible by the dark cloud developing in between the two jets of Figures 5c 5e, which is produced by lithic (cold) ash due to the strong fragmentation of the bs1 rim blown away during the explosion (cfr. Figures 6a and 6b). Interestingly, a recent petrologic analysis of the erupted ejecta by the different summit vent zones of Stromboli has revealed that the material erupted by the S vent zone is normally 10 C cooler and 6 8 wt% more crystallized than that erupted by the C vents, whereas the N vents display features that are intermediate between the two, and variable in different times [Landi et al., 2011]. The S area is also the one that commonly displays ash-dominated Type 2b eruptions, and it is thus possible that a cooler and more crystallized magma erupted by this site is also more prone to give rise to brittle rather than plastic fragmentation, resulting in a greater amount of ash released during the explosive activity with respect to coarse ballistic ejecta [Dürig et al., 2012]. [27] By combining seismic and electric data with observations, it is therefore evident that the major explosion of 7 September (occurred at 07:49:00), although not associated with the emission of a large amount of pyroclasts, is characterized by a smaller particle grain-size and a much higher 13 of 17