By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
 We present the analysis of ~4 million infrasonic signals which include 39 episodes of lava fountains recorded at 5.5 km from the active vents. We show that each eruptive episode is characterized by a distinctive trend in the amplitude, waveform, and frequency content of the acoustic signals, reflecting different explosive levels. Lava fountain starts with an ~93 min long violent phase of acoustic transients at ~1.25 Hz repeating every 2–5 s. Infrasound suddenly evolves into a persistent low-frequency quasi-monochromatic pressure oscillation at ~0.4 Hz. We interpret this shift as induced by the transition from the slug (discrete Strombolian) to churn flow (sustained lava fountain) regime that is reflecting an increase in the gas discharge rate. We calculate that infrasonic transition can occur at a gas superficial velocity of ≤76 m/s and it can be used to define infrasonic-based thresholds for an efficient early warning system.
 Among basaltic explosive volcanoes, Etna is one of the most active and well studied with about 150 short-lived (from 20 min up to 10 h long) lava fountain (LFN) episodes recorded in the last 25 years. Geodetic and geochemical observations of LFN episodes indicate that they are triggered by a supply of CO2 gas-rich magma pressurizing a deep (~1.5 km) reservoir [Allard et al., 2005; Aiuppa et al., 2010]. Each LFN episode appears to be systematically anticipated by the abrupt increases of seismic tremor [e.g., Alparone et al., 2003]. Visual and video observations evidence an initial phase dominated by Strombolian-type activity [e.g., Bertagnini et al., 1990], which finally merges into a sustained emission of lava, ash, and gas up to 800–1000 m height.
 Infrasound measured at ~1 km of distance during two episodes in 2001 [Vergniolle and Ripepe, 2008] shows that LFN starts with a series of pressure transients typical of Strombolian activity, which are initially sparse (every few minutes) and weak (~1 Pa) and progressively become more frequent (every ~8 s) and stronger (up to 100 Pa) during the fountaining. Almost continuous infrasonic signal (called “infrasonic tremor”) with a frequency content between 1 and 4 Hz was observed during the paroxysmal stage of the 10 May 2008 lava fountain episode [Cannata et al., 2009].
 Textural and compositional analysis of basaltic scoria presents different vesiculation distinctive of two degassing regimes [Polacci et al., 2006]. Scoriae collected during the Strombolian phase evidence a prolonged coalescence and outgassing of magma prior to fragmentation, while lava fountain scoriae indicate a fast ascent and expansion of a volatile-saturated magma. Large-scale experiments show how the transition between different regimes mainly depends on gas volume flow, which in turn controls magma vesicularity [Pioli et al., 2012].
 Thus, LFN at Etna appears characterized by a transition between different gas flux regime [Parfitt, 2004], responsible for the change in eruptive style [e.g., Calvari et al., 2011]. However, no evidence for this transition has been found in the geophysical (i.e., seismic tremor) signal yet.
 Here we present a new and unique data set of infrasonic records collected by a small-aperture infrasonic array, which provides unprecedented evidence on this flow transition during LFN episodes at Mount Etna. Infrasound reveals that conduit processes during LFN always follow similar patterns and provide evidence on the gas discharge dynamics.
2 The Small-Aperture Infrasonic Array
 Infrasonic activity at Etna volcano is recorded with a small-aperture (250 m) four-element infrasonic array, deployed in September 2007 at Lapide Malerba on the south margin of the Valle del Bove, at an elevation of 2010 m above sea level and at a distance of 5500 m from the summit craters (Figure 1a). The array is equipped with a differential pressure transducer, with sensitivity of 25 mV/Pa in the 0.01–20 Hz frequency band, and installed within aluminum boxes buried ~1 m underground to reduce wind noise. Connection among the elements of the array is realized with 150 m long fiber-optic cables. The array threshold of infrasonic detection is as small as 10−2 Pa. Digital data are collected at 16 bits with a sampling rate of 50 Hz and radio transmitted to the town of Nicolosi, at a distance of ~16 km. Here data are preprocessed and broadcasted in real time to the Department of Earth Science in Firenze and the Department of Civil Protection in Rome. The total power requirement of the array (<1 W) and the fiber optics technology allow an operating efficiency of >99% with only 16 days of interruption over a total period of 2069 days from September 2007 to April 2013.
3 Real-Time Array Location Procedure
 Array analysis is performed by multichannel semblance applied to a grid-searching procedure [Ripepe and Marchetti, 2002], which identifies signals from noise in a sliding 5 s long window both in terms of propagation back azimuth and apparent velocity. Considering the 250 m aperture of the array, at a frequency of 1 Hz, the expected azimuth resolution is of ~2° [Ulivieri et al., 2011], which corresponds to ~190 m at a distance of 5500 m.
 Following the procedure described by Ulivieri et al. , coherent volcano infrasound at the array is searched for 1° back azimuth step intervals, and assuming an apparent sound speed range between 320 and 360 m/s, which accounts both for ambient temperature variations (from −20°C to +30°C) and source-to-receiver elevation difference (1300 ± 100 m). These constraints in terms of back azimuth (from 320°N to 350°N; Figure 1) and apparent velocity allow us to separate in real time the volcano-acoustic sources from other natural and artificial events.
4 Infrasonic Activity at Mount Etna Volcano
 During ~6 years of observation (September 2007 to April 2013), array analysis provided ~4 million detections of infrasound consistent with the summit area (Figure 1b). This corresponds to a continuous infrasound emission from Etna volcano for 11% of the recording time. Infrasonic detections show a wide peak pressure range from 0.04 Pa up to 75 Pa (Figure 1b). Pressure and numbers of detections (Figure 1c) were relatively low during the 2007–2010 period, mainly characterized by mild Strombolian activity with only small LFN episodes (23 November 2007 and 10 May 2008) at the SEC and pit crater (PIT) [e.g., Cannata et al., 2009]. Back azimuth indicates that during the observation period, most of the activity concentrated at the summit craters (324°N–333°N), with minor detections from 337°N to 346°N during May 2008 and July 2009, when explosive activity occurred from the eruptive fissure (EF; Figure 1a) located in the Valle del Leone [e.g., Cannata et al., 2009].
 Since December 2010, peak pressure and number of the infrasonic detections increased (Figure 1c) as a result of a period of intense volcanic activity that was punctuated by a sequence of 40 LFN episodes within 28 months (from 12 January 2011 to 27 April 2013) and resulting in a five times higher rate of infrasonic activity (Figure 1c).
5 Acoustics of Lava Fountaining
 The LFN episode of 12 January 2011 is one of the best documented events from both geophysical and geochemical point of view [Calvari et al., 2011]. Visible and thermal cameras show that the LFN episode started between 19:15 and 20:49 UT on 12 January 2011 with a vigorous Strombolian activity launching bombs and scoriae outsize the PIT crater. At 21:15 UT, explosive activity becomes almost continuous suggesting a transitional eruptive phase, which at 21:50 UT eventually gave rise to a sustained lava column rising ~800 m above the rim and producing an ~6 km high ash column [Calvari et al., 2011].
 The infrasonic array located this LFN episode with a stable back-azimuth of 332° ± 2°N pointing toward the PIT crater (Figure 2d). The multichannel semblance location procedure allows us to isolate infrasound radiated by the LFN event providing a reliable tracking of the amplitude and waveform signature of the event (Figure 2a). The mean pressure amplitude (Pmean) of the acoustic signals detected by the array in 5 min long time window shows a first change at 19:57 UT associated with a gradual increase of the peak pressure (from 0.3 to 2.5 Pa) and the number of detections (Figure 2c) until 21:46 UT. When the lava fountain started, Pmean experienced an abrupt increase, with peak pressures reaching 23 Pa. The frequency content shifted from ~0.55 Hz to lower values of ~0.4 Hz (Figure 2b). After that, Pmean remained above 5 Pa gradually decreasing during the ~1.7 h long fountaining, then finally dropped to pre-eruption values of <1 Pa. At 23:33 UT on 12 January after 3.6 h, the LFN episode ceased.
 Based on amplitude variation, shift of the frequency content, and number of detections of the acoustic signal, we divide the LFN episode into three main phases, PHs, PHf, and PHe (Figure 2a), each of them showing a different infrasonic signature (Figure 2b) associated with a change in eruptive style. It is noteworthy that the seismic signal mimics the amplitude variations of the acoustic wavefield (Figure 2a) suggesting a source dynamics very well coupled with ground and atmosphere.
5.1 PHs Phase (Strombolian)
 The initial PHs phase starts at 19:49 UT and it is punctuated by moderate (1–5 Pa) discrete pressure transients (Figure 3a) with a mean duration of ~0.8 s and peak frequency content of 1.25 Hz (Figure 3c, dashed line), repeating with a mean rate which increases from 18 (at 20:00) to ~30 events per minute (Figure 2c). Frequency component of the single transient is masked by the stationary occurrence rate of one event every ~2 s, which is responsible for a persistent spectral peak at 0.5 Hz (Figure 3c, bold black line). This PHs phase (characterized by more than 2000 explosions in ~1.8 h) is consistent with the almost continuous explosive activity observed with visible and thermal cameras and interpreted as a transitional eruptive style [Calvari et al., 2011]. However, infrasound reveals that after 21:15 UT, the explosive rate has increased ~4 times faster than that possibly observed by cameras.
5.2 PHf Phase (The Lava Fountain)
 Lava fountain activity is marked at 21:46 UT by a transition from this discrete and rhythmic acoustic wavefield to a regular low-frequency oscillation (Figure 3b) centered at ~0.4 Hz (Figure 3d). This infrasonic oscillation is associated with the ~800 m high lava fountain erupting from the PIT crater [Calvari et al., 2011]. Infrasound retains this oscillatory characteristic over the whole ~1.7 h long fountaining episode (Figure 2c) and it represents the novel constraint on the dynamics of LFN events at Mount Etna volcano.
5.3 PHe Phase (The End of Lava Fountain)
 The eruptive episode ends at 23:33 UT with a final short-living (<5 min) phase of discrete, high-amplitude infrasonic transients (PHe), similar in shape to the Strombolian phase (PHs) preceding the fountain but with a larger pressure of >10 Pa. This last phase is associated with a spectral shift toward higher (0.6 Hz) frequency values (Figure 2b) and it marks the abrupt end of infrasonic activity.
6 Transition From Slug to Churn Flow Discharge Regime
 Each of the 39 LFN episodes detected by the array presents a highly variable duration (Figure 4a), spanning from 3 to 1507 min, but with the same recurrent three eruptive phases, with the main features of the acoustic wavefield being preserved. This results into an infrasonic Pmean trend (Figure 4a, red line) that shows a stable pattern closely matching amplitude change of seismic tremor (Figure 2a). Although amplitude of the seismic signal increases ~4 times (Figures 2a, 3a, and 3b), its wavefield shows the same spectral content between 1 and 1.7 Hz (peaked at ~1.3 Hz) during the LFN episode (Figures 3c and 3d).
 Unlike seismic tremor (Figure 3), infrasound wavefield evidences the shift from violent and repetitive Strombolian to sustained and oscillatory lava fountain dynamics. This transition is explained in terms of slug to annular flow regime [e.g., Bertagnini et al., 1990; Calvari et al., 2011]. However, recent laboratory experiments indicate that annular flow conditions in magma are unlikely to be met unless conduit diameter is smaller than a few meters [Pioli et al., 2012]. We then assume here that the infrasonic monochromatic signal associated to oscillations of the eruptive column is probably reflecting a churn flow degassing regime.
 In slug flow, the liquid between two Taylor bubbles (liquid slug) moves at a constant velocity, and its front and tail have a constant speed. With increasing superficial gas velocity USG (i.e., the volume gas flux divided by the cross-section area of the conduit), void fraction in the liquid slug increases, the distance between two bubbles decreases, and the slug flow is destabilized [Jayanti and Hewitt, 1992]. The downcoming liquid film surrounding the Taylor bubble is then flooded by the gas flowing upward and the oscillatory motion of the liquid is observed [Taitel et al., 1980]. The upward and downward oscillatory motion of the liquid is thus considered as the distinctive dynamics of churn flow.
 Assuming that the oscillatory wavefield of the infrasound is reflecting the oscillatory motion of the liquid (i.e., magma) during the churn flow regime induced by the increase in the gas discharge rate, we can estimate the critical superficial gas velocity (USG) necessary to reach the transition using the equation derived for large pipes [Pushkina and Sorokin, 1969]:
where ρM is the magma density (2700 kg/m3) and σ (0.36 N/m) is the surface tension of the magma [Vergniolle and Ripepe, 2008]. Considering that volcanic gas is mainly (~80%) H2O [Allard et al., 2005], we use a gas density ρG = 0.17 kg/m3 (H20 at 1 atm and 1300°K) to calculate that transition from the rapid Strombolian (slug flow) to sustained lava column (churn flow), when occurring near the vent and for a stagnant magma, which requires a superficial gas velocity USG of ≤76 m/s. This velocity represents the upper limit for the critical superficial gas velocity and it reduces to 3.8 m/s if transition is likely to occur at the reservoir depth of 1500 m [Allard et al., 2005]. However, if we assume a near-the-vent transition we calculate for a vent radius R=15 m [Calvari et al., 2011] a gas volume flux Q=pR2USG of 5.4×104 m3/s. This gives for the LFN episode of January 12, 2011 a total gas volume of 3.6×108 m3, which is in good agreement with the 2.4×108 m3 of gas estimated for the same episode by thermal image analysis [Calvari et al., 2011].
7 Implication for Eruption Onset
 The ash-rich plume of LFN episodes generates a large and widespread tephra fallout, with strong effects on nearby communities and local air traffic. For this reason, monitoring LFN episodes at Mount Etna has a strong socioeconomic impact and it requires an efficient and reliable monitoring system. Alerts based on seismic tremor amplitude thresholds [e.g., Alparone et al., 2007] are not considered to be fully reliable (tremor increases also during lava intrusion, or earthquake swarms) requiring for more complex unsupervised classification [Langer et al., 2011].
 We suggest that the persistency of the LFN dynamic features and the ability of infrasound to detect in real time each distinctive explosive phase can provide a simple but effective early-warning system for LFN eruptions at Etna volcano. We consider the Strombolian phase (PHs) as a distinctive precursor and we use it to mark the beginning of the LFN episode. Accordingly, we define an infrasonic parameter (IP = Pmean × Nd) as the product between the mean infrasonic amplitude (Pmean) recorded at the array and the mean number of detection per minute (Nd) calculated in real time within 5 min long time interval.
 Based on the statistical distribution of ~4 million infrasonic events (Figure 4b), we defined two threshold levels (A1 = 15 and A2 = 30) for IP, which correspond to three classes of alert. When acoustic pressure Pmean is <1.25 Pa and number of detection is Nd < 12, IP is below the first threshold (IP < A1) and infrasonic activity is within the ordinary level (Figure 4b). Above the first A1 threshold (IP > A1), rapid Strombolian activity of PHs phase is occurring (Ts in Figure 4) at a rate as high as one event every ≤5 s and with a mean pressure Pmean of ≥2.5 Pa. The onset (Tf in Figure 4) of the LFN episode (IP > A2) is defined when the array is continuously detecting infrasonic signal (Nd = 12) with mean pressure Pmean ≥ 2.5 Pa.
 In the last ~6 years, IP was above A2 39 times and above A1 38 times over the 40 (including one eruption not detected for technical problems) of the well-documented lava fountain episodes (see Table S1 in the supporting information) resulting in 97.5% of true alerts and 2.5% of false alerts (Ip was above A1 but no LFN occurred). Time difference Ts − Tf (Figure 4c) ranges between 3 and 580 min, with an average of 93 min. This means that the infrasonic-based early-warning procedure is able to automatically deliver a pre-alert (IP > A1) more than 1 h before the ash-plume-rich lava fountaining is erupted.
8 Summary and Conclusions
 Mount Etna volcano is an efficient source of infrasound with ~4 million acoustic signals in ~6 years (97.4% of the total detections) confined within the crater sector. Most of this infrasonic activity (64%) is concentrated in the SEC and PIT craters and at the beginning of 2011 experienced a major (five times) increase associated with an intense eruptive period characterized by 40 lava fountains. All these LFN episodes share common amplitude and spectral features associated to changes in the magma discharge rate and reflecting the transition from a slug/Strombolian to a churn/fountaining regime. In large pipes, churn flow is independent of conduit diameter and it is reached at high superficial gas velocity [Pushkina and Sorokin, 1969], which in a near-the-vent stagnant magma condition is ≤76 m/s.
 There is no evidence of this transition in the seismic signal, which only shows an amplitude increase at the same ~1–1.7 Hz frequency range. The efficiency of infrasound to detect in real time this sharp eruptive dynamics change allows us to build an early-warning system, which has 97.5% of success to deliver an alert around 1 h before lava fountain is erupting kilometer-high ash plumes in the atmosphere.
 We are grateful to Salvo Caffo of the Etna National Park and the Guardia Di Finanza of Nicolosi for making the installation of the array possible and to Giorgio Lacanna, Dario Delle Donne, and Riccardo Genco for their continuous support in the field. The paper has been improved by the critical comments of Mie Ichihara, Laura Pioli, and the Associate Editor. This work has been supported by the Department of Civil Protection.
 The Editor thanks Mie Ichihara and Laura Pioli for assistance evaluating this paper.