Characterization and location of infrasonic sources in active

volcanoes: Mount Etna, September–November 2007

A. Cannata,1,2 P. Montalto,3,4 E. Privitera,3 and G. Russo3,5

Received 12 August 2008; revised 3 June 2009; accepted 16 June 2009; published 25 August 2009.

[1] The period September–November 2007 was characterized at Mount Etna byexplosive activity and intense degassing. During this time interval, infrasonic signals wererecorded by an infrasonic network. By a triggering procedure, about 1000 infrasonicevents were found, characterized by very high signal-to-noise ratio and grouped into ninefamilies. Successively, the spectral analysis allowed subdividing these nine families intothree clusters based on the peak frequency and the quality factor of the events. Finally,by the location analysis a cluster (cluster 1) was related to the degassing activity of thenortheast crater (NEC), while the other two (clusters 2 and 3) to the explosive activity ofthe southeast crater (SEC). The comparison between the stacked infrasonic waveforms,interpreted as generated by the vibration of large gas bubbles, and the synthetic ones,permitted to calculate radius, length of the bubble, and initial overpressure, by a geneticalgorithm method. The higher overpressure values of cluster 3 compared to the cluster2 values were in good agreement with the stronger intensity of the explosionsaccompanying the infrasonic events of cluster 3. The variation of both intensities andwaveforms is tentatively attributed to the occasional accumulation of lithic clasts (due tomoderate landslides?) on the explosive vent. Indeed, events belonging to cluster 3 were nolonger observed once the landslides had ended. Finally, the daily emitted gas volume,related to the active degassing, was estimated for NEC and SEC by using theinfrasonic data during the studied period.

Citation: Cannata, A., P. Montalto, E. Privitera, and G. Russo (2009), Characterization and location of infrasonic sources in active

volcanoes: Mount Etna, September–November 2007, J. Geophys. Res., 114, B08308, doi:10.1029/2008JB006007.

1. Introduction

[2] New insights into explosive volcanic processes havebeen achieved in the last years by studying infrasonicsignals [e.g., Vergniolle and Brandeis, 1994; Ripepe et al.,1996, 2001a]. In fact, infrasonic activity on volcanoes isclear evidence of open conduit conditions and thus providesimportant indications on the explosive risk. The infrasonicsignal for short distances travels in an almost homogenousatmosphere without structures that may scatter, attenuate orreflect acoustic waves. Then, unlike the seismic signalwhose wavefield is strongly affected by topography[Neuberg and Pointer, 2000] and path effects [Gordeev,1993], the infrasonic signal maintains its features almostunchanged during propagation, allowing obtaining informa-tion concerning source dynamics. The source mechanism of

the sound radiated during eruptions is still open to debate.According to some studies, this signal can be related tothe acoustic resonance of magma in the conduit, triggeredby explosive sources, implying propagation of soundwaves in the magma and atmosphere through an open vent[Buckingham and Garces, 1996; Garces and McNutt, 1997;Hagerty et al., 2000]. Other theories relate the source ofsound to eruption dynamics, such as a sudden uncorking ofthe volcano [Johnson et al., 1998; Johnson and Lees, 2000],local coalescence within a foam [Vergniolle and Caplan-Auerbach, 2004] and Strombolian bubble vibration[Vergniolle and Brandeis, 1994, 1996; Vergniolle et al.,1996, 2004].[3] The location of the source of the infrasonic signals,

generally coinciding with active vents, is of great importancefor volcanic monitoring. Therefore, some techniques, basedon signals recorded by infrasound arrays or networks, weredeveloped to locate the source of this signal [e.g., Ripepeand Marchetti, 2002; Garces et al., 2003; Johnson, 2004;Matoza et al., 2007; Jones et al., 2008]. At multiventsystems, like Stromboli [Ripepe and Marchetti, 2002] andKilauea [Garces et al., 2003], methods based on the com-parison of the infrasonic signals by cross correlation orsemblance functions have enabled monitoring the explosiveactivity of distinct craters.[4] At Mount Etna, the first infrasound investigations

were performed by temporary experiments [e.g., Ripepe et

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 114, B08308, doi:10.1029/2008JB006007, 2009ClickHere

for

FullArticle

1Dipartimento di Scienze Geologiche, Universita di Catania, Catania,Italy.

2Now at Istituto Nazionale di Geofisica e Vulcanologia, Sezione diCatania, Catania, Italy.

3Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania,Catania, Italy.

4Dipartimento di Ingegneria Elettrica, Elettronica e dei Sistemi,Universita di Catania, Catania, Italy.

5Consorzio Cometa, Catania, Italy.

Copyright 2009 by the American Geophysical Union.0148-0227/09/2008JB006007$09.00

B08308 1 of 15

al., 2001b; Gresta et al., 2004]. The summit area of MountEtna volcano is currently characterized by four activecraters: Voragine, Bocca Nuova, southeast crater, and north-east crater (hereafter referred as VOR, BN, SEC, and NEC,respectively; see Figure 1). These craters are characterizedby persistent activity that can be of varying nature andsometimes involve different types simultaneously: degass-ing, lava filling or collapses, low rate lava emissions,phreatic, phreatomagmatic or Strombolian explosions, andlava fountains [Cristofolini et al., 1988]. The greatest riskfor the villages located around the volcano is linked to flankeruptions that take place from fractures cutting the slope ofthe edifice.[5] During the second half of the 20th century, Mount

Etna was characterized by an unusually high level ofactivity, with a clear increase in effusive rates and in thefrequency of summit and flank eruptions in the last decades[Tanguy et al., 1996]. In particular, a dramatic increase inthe frequency of short-lived, but violent eruptive episodes atthe summit craters was observed: between 1900 and 1970about 30 paroxysmal eruptive episodes occurred at thesummit craters, while there have been more than 180 sincethen [Behncke and Neri, 2003].[6] In this work we analyze the infrasonic signal recorded

at Mount Etna during the period September–November2007, in order to investigate possible relationships betweendifferent volcanic activity at distinct craters and signalvariability, and to characterize the source of the infrasonicsignals. Our aim is to find simple and powerful tools able tocontribute to the volcano activity monitoring and give someinsights into explosive eruption dynamics.

2. Data Acquisition and Analysis

[7] We analyzed the signals recorded during September–November 2007 period by an infrasonic network. Thiscomprises four sensors distributed at various azimutharound the volcano and is run by the Istituto Nazionale diGeofisica e Vulcanologia, Sezione di Catania. The sensorsare located at distances ranging between 1.5 and 2.5 kmfrom the center of the summit area (Figure 1) and consist ofa Monacor condenser microphone MC-2005, with a sensi-tivity of 80 mV/Pa in the 1–20 Hz infrasonic band. Theinfrasonic signal is acquired at a sampling rate of 100 Hz.[8] The infrasonic signals, recorded in September–

November 2007, contain a large number of infrasonic tran-sients (hereafter called ‘‘infrasonic events’’), generally short(1–4 s), characterized by impulsive compression onsets andby a spectral content in the frequency range 1–5 Hz(Figure 2). The observed waveforms of these infrasonictransients are very similar to those recorded at othervolcanoes: Stromboli [Ripepe et al., 1996], Klyuchevskoj[Firstov and Kravchenko, 1996], Sangay [Johnson andLees, 2000], Karymsky [Johnson and Lees, 2000], Erebus[Rowe et al., 2000], Arenal [Hagerty et al., 2000], andTungurahua [Ruiz et al., 2006].

2.1. Event Triggering

[9] In order to obtain a large data set of infrasonic events,an automatic triggering procedure was developed and ap-plied on the whole infrasonic time series. The events wereextracted by using a method based on ‘‘standard deviation

squared series’’ (SD2) of the infrasonic signal. This param-eter is calculated as follows:

SD2 ¼ 1

n� 1

� �Xni¼1

xi � xmð Þ2" #

ð1Þ

where n is the number of samples, xi and xm indicate the ithsample and the average value of the considered infrasonictime series, respectively. The SD2 time series was compiledcalculating one SD2 value for each 3-s-long window ofinfrasonic signal. Two different thresholds are fixed andapplied on the SD2 series: the upper threshold is used todetect the infrasonic events, and the lower one indicates theonset and the end of the events.[10] EBEL was chosen as the reference station. We noted

that this was the least affected by noise due to bad weatherand had the longest time series. Then, two different data setswere built by using the aforementioned method. The firstdata set, characterized only by events with very high signal-to-noise ratio at EBEL, was obtained by fixing a high valueof the upper threshold (90th percentile of the consideredinfrasonic window). This data set, composed of 987 eventswith duration ranging between 1 and 4 s, was used for allthe analyses except to estimate the emitted gas volume. Thesecond data set, containing most of the detectable eventsoccurring during the investigated time period, was compiledby using a low value of threshold (10th percentile of theconsidered infrasonic window). This data set comprisesabout 55000 events and was used to quantify the emittedgas volume. It is worth noting that the events occurringduring the lava fountains (4–5 September and 23–24 November, see section 3) were not detectable. In fact, thevery high occurrence rate of events during the paroxysmalepisodes, that gives rise to a continuous signal (the so-calledinfrasonic tremor), did not allow us to detect single events.Moreover, also the events occurring during time periodscharacterized by bad weather and strong wind were notdetectable because of the high noise.[11] Finally, a low value of the lower threshold was

chosen for the first data set (80th percentile of the consid-ered infrasonic trace window) to pick the onset of the eventsas exactly as possible.

2.2. Waveform Classification

[12] A high degree of similarity of the waveforms ofdifferent infrasonic events implies the same source mecha-nism and locations varying roughly within one fourth of thedominant wavelength of the events [Geller and Mueller,1980]. However, the exact fraction of the wavelengthdepends on heterogeneity of velocity structure around thesource [Nakahara, 2004]. In light of these assumptions, weperformed a waveform classification of the infrasonicevents to get an initial idea about the variability of sourcemechanism and location. We used a modified version of themethod of Green and Neuberg [2006], which comprised thefollowing steps: (1) the 987 infrasonic events, belonging tothe first data set (see section 2.1) and recorded at EBELstation were filtered in the frequency range 1–5 Hz; (2) thecorrelation matrix was calculated (Figure 3a); (3) a thresholdvalue of cross correlation coefficient equal to 0.85 waschosen; (4) a master event was selected as the event with

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

2 of 15

B08308

the largest number of correlation values above the chosenthreshold; (5) an average family waveform was found bystacking all events well correlated with the master event;(6) this stack waveform was then cross-correlated with theoriginal infrasonic records, and all events with a correlationgreater than 0.8 were grouped into a waveform family; and(7) steps 4–6 were repeated until the entire matrix wasclassified into distinct groups. In this procedure, overlapbetween clusters was not allowed; in fact, once an event wasassigned to a given group, it was removed from the corre-lation matrix. Generally, the choice of the cross correlationthreshold is a trade-off between classification accuracy andevent detection. A too low threshold allows waveformshaving different structure to be classified into the same

group; conversely, a too high one does not allow eventscharacterized by poor signal-to-noise ratio to be classified atall. For this reason, unlike the method of Green and Neuberg[2006], we fixed two different cross correlation thresholds insteps 3 and 6: thus, we were able to achieve a reasonablenumber of families, each of which contains a significantquantity of events.[13] By using this method nine families, comprising

about 95% of all the considered events, were found. Thisresult suggests the repetitive excitation of stationary sour-ces. The average waveforms and the time distribution ofnine families are shown in Figures 3b and 4, respectively.The great similitude observed among families (Figure 3b)suggested the application of the cross correlation analysisdescribed above on the stacked waveforms of the ninefamilies. Tuning the cross correlation coefficient thresholdto 0.9, three clusters were recognized. Families from 1 to 5were grouped in cluster 1, families 6 and 7 in cluster 2, andfamilies 8 and 9 in cluster 3.[14] Finally, it is noteworthy that unlike the other families

that are spread out over periods of months, the eventsbelonging to the families 8 and 9 (cluster 3) occurred duringthe interval 1–4 September.

2.3. Peak-to-Peak Amplitude, Spectral Analysis,and Features Plane

[15] Several analyses were performed on the 987 eventsbelonging to the first data set (see section 2.1) to charac-terize the infrasonic activity at Mount Etna during thestudied period.[16] At first, we calculated the peak-to-peak amplitude of

the infrasonic events that ranged from 1 to over 100 Pa. Thehighest values are mainly clustered at the beginning ofSeptember and are linked to the Strombolian activity at theSEC (Figure 5a).[17] Successively, we performed two different spectral

analyses on the infrasonic events. First, we took intoaccount a 10-s-long window (1024 samples) for each event

Figure 1. Sketch map of Mount Etna with location of thefour infrasonic sensors (triangles) used in this work. In topleft inset the distribution of the four summit craters (VOR,Voragine; BN, Bocca Nuova; SEC, southeast crater; NEC,northeast crater) is reported.

Figure 2. Examples of infrasonic events recorded by EBEL station and their short time Fouriertransform, obtained by using 2.56-s-long windows overlapped by 1.28 s. The volcanic activity and thefamily (as explained in the text) are reported at the top.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

3 of 15

B08308

recorded at EBEL, comprising the whole waveform, andcalculated a spectrum by fast Fourier transform (FFT).Then, a peak frequency value was obtained for each event(Figure 5b). Most of the calculated peak frequencies rangedbetween 1 and 4 Hz. It is also noteworthy that most of the

highest frequency values (3–4 Hz) are clustered in theinterval 1–4 September.[18] Second, in order to verify the aforementioned spec-

tral variations and to obtain information about the dampingfeatures of these infrasonic signals, the Sompi analysis

Figure 3. (a) Correlation matrix of the 987 infrasonic events. (b) Waveforms obtained by stacking theevents belonging to each family.

Figure 4. Time distribution of the families of the infrasonic events. These families were obtained byfiltering the infrasonic signals between 1 and 5 Hz and following the method of Green and Neuberg[2006] (see text for details). As explained in the text, families 1–5 are linked to activity at NEC, andfamilies 6–9 are linked to activity at SEC.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

4 of 15

B08308

Figure 5. (a) Peak-to-peak amplitude and (b) peak frequency, calculated by FFT, of the infrasonicevents at EBEL. (c and d) Peak frequency and quality factor at EBEL obtained by Sompi method for ARorder of 2.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

5 of 15

B08308

[Kumazawa et al., 1990] was performed. By this method ofspectral analysis, on the basis of an autoregressive (AR)model, a signal is deconvoluted into a linear combination ofcoherent oscillations with amplitudes decaying and addi-tional noise. In this approach we define a complex frequen-cy fc = f � ig, where f is the frequency, g is the growth rate,and i is

ffiffiffiffiffiffiffi�1

p[Kumazawa et al., 1990]. The quality factor Q

is then given by �f/2g. Therefore, 2-s-long windows ofinfrasonic signal, recorded at EBEL and coinciding with thetails of the events, were taken into account, and frequencyand quality factor were calculated for AR orders rangingbetween 2 and 10. The sharply monochromatic nature of theinvestigated signals justifies the choice of these low orders[Lesage, 2008]. The results were very similar for all theused orders. As shown in Figure 5c, the obtained frequencyvalues confirmed the overall results of the FFT analysis. Thequality factor values mainly ranged between 2 and 10(Figure 5d).[19] In order to confirm the results gained with the cross-

correlation method, and try to reduce the size of information(signals, data, etc.), we performed a parametric approachthat is able to speed up classification analyses, especiallyduring monitoring activities. Generally, definition of pecu-liar features or properties by a ‘‘feature extractor’’ may beuseful. In our case, we can use the obtained spectralcharacteristics, computed by Sompi method (our ‘‘featureextractor’’), as features to describe the infrasonic events.Then, in order to investigate prospective similarities ordifferences among these features extracted from the infra-sonic signals, we plotted the frequency and the qualityfactor, in the x axis and y axis, respectively, and obtainedthe so-called ‘‘features plane.’’ As shown in Figure 6, thethree clusters found by cross-correlation method wereconfirmed. Cluster 1 (containing the families 1, 2, 3, 4and 5) showed frequency and quality factor in the range1.2–2.2 Hz and 1.5–6.0, respectively. Cluster 2 (compris-ing the families 6 and 7) was characterized by frequencyrange 2.0–3.5 Hz and quality factor 1.5–4.0. Finally,cluster 3 (composed of the remaining families 8 and 9)

comprises events with frequency range 3.5–4.5 Hz andquality factor 3–9. The average waveforms of the threeclusters and the respective spectra and f � g diagrams areshown in Figure 7. The stacked trace of cluster 1 ischaracterized by the longest duration. The broadest rangeof quality factor of cluster 3 is due to the characteristics ofthis cluster. In fact, the waveforms of the events belongingto this cluster, composed of about 2.5 cycles with slowdecay followed by a sharp amplitude decrease, cannot beconsidered as a perfect superposition of simple decayingsinusoids. Hence, the calculated quality factor values areslightly scattered.

2.4. Source Location

[20] In order to obtain preliminary qualitative informationabout the source location of the infrasonic events, thedifferences between the arrival times of the 987 events ofthe first data set at the four stations were calculated(Figure 8). It is noteworthy that only two different sets oftime lag values were found, suggesting the existence of onlytwo source locations. Moreover, events belonging to thesame family shared the same time lag values, confirming acommon source location.[21] Successively, the location of the infrasonic source

was estimated by a grid searching procedure over a surfaceof 1.5 � 1.5 km, with spacing of 25 m, centered on thevolcanic edifice and coinciding with the topographicalsurface. We assumed a point source and a sound speed of340 m/s in the atmosphere. The infrasonic source wasassumed to be in each node of the grid, and for each nodewe first calculated the theoretical infrasonic travel times atthe four sensors. Then, by these theoretical travel times,infrasonic signals at the different stations were delayed andcompared by the semblance function [Neidell and Taner,1971; Almendros and Chouet, 2003; Ripepe et al., 2004;Patane et al., 2008]. This function is assumed representa-tive of the probability that a node has to be the sourcelocation. Therefore, the source was finally located in thenode where the delayed signals showed the largest sem-blance value. As the events of each family both show verysimilar waveforms and time lag values, they obviously sharethe same location and source mechanism. Therefore, wedecided to calculate the stacked waveforms of each familyat all the stations and to locate them. In order to calculatethe stacked waveforms, we took into account all the wave-forms from the two first events of the family 1 (one set of4 signals for each event) and tried different time lagsbetween these two sets of signals. For each time lag value,we evaluated the similarity between the events by averagingthe cross correlation coefficients calculated for each coupleof corresponding signals (for example, ‘‘EBEL of event 1’’and ‘‘EBEL of event 2’’ are two corresponding signals). Thetime lag value that gave the maximum average crosscorrelation coefficient was chosen and an average signalwas obtained for each station. The other events werecompared to the average event that was ‘‘updated step bystep.’’ Thus, all the events contributed to the stacked signalsrepresenting the family (Figure 3). The same procedure wasfollowed for the other families. Then, the stacked wave-forms were located (Figure 9). The results of this analysis,in good agreement with the position of the most activesummit craters during the studied period, confirm the

Figure 6. Peak frequency of the nine families of infrasonicevents plotted versus the quality factor calculated by Sompimethod (see legend).

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

6 of 15

B08308

existence of only 2 sources, the one coinciding with NECand the other one with the pit crater on the eastern flank ofSEC. Therefore, we can attribute the families 1, 2, 3, 4 and 5to NEC and the families 6, 7, 8 and 9 to SEC. Similarly,cluster 1 is related to the activity of NEC, and clusters 2 and3 are related to the activity of SEC. Similar results werefound at Stromboli by Ripepe and Marchetti [2002] thatrecognized two different groups of infrasonic signals asso-ciated to two different craters. However, in other cases,different clusters of infrasonic events were found with noapparent correlation to location [e.g., Ruiz et al., 2006].

2.5. Source Mechanism

[22] The detected infrasonic events are similar to thesignals described by Vergniolle and Brandeis [1996], RipepeandMarchetti [2002], Vergniolle [2003], and Vergniolle et al.[2004], and explained as generated by the vibration of alarge gas bubble before bursting. The other possible well-developed models give rise to different waveforms to ourrecordings. For example, the local coalescence within afoam [Vergniolle and Caplan-Auerbach, 2004] producesgradually decaying sinusoids. Our detected events aregenerally characterized by a first energetic part roughlycomposed of one cycle and a half, followed by a secondpart with various weaker oscillations. The acoustic reso-nance of magma/fluid in a conduit [Garces and McNutt,1997] should produce signals with different harmonics

components (a fundamental mode and overtones), whereasour recorded signals are mainly monochromatic. Therefore,in order to quantitatively investigate the source mechanismof the infrasonic events on the basis of bubble vibrationmodel, we performed a waveform inversion. Using theequations reported by Vergniolle and Brandeis [1996] andVergniolle et al. [2004] and fixing the values of density,viscosity of the magma, and thickness of magma above thevibrating bubble to 2700 kg/m3, 400 Pa s, and 0.1 m,respectively [Vergniolle, 2003; Vergniolle and Ripepe,2008], we were able to calculate synthetic waveforms.Then, by optimization algorithms, we constrained the valuesof the three unknown parameters (radius, length of thebubble, and initial overpressure) that allow finding the bestfit between synthetic and measured waveforms.[23] The optimization method chosen for parameter esti-

mation is the genetic algorithm (GA) that applies Darwin’sevolutionary theory to general optimization problems. Thismethod is based on individuals, grouped into populationsthat represent the parameters searched in the estimationprocess. The GA method can be reassumed by the followingsteps: (1) an initial set of candidate solutions, called initialpopulation, are generated; (2) the evaluation of the candi-date solutions is based on some fitness criteria; (3) on thebasis of the performed evaluation, some candidate solutionsare kept and the others are discarded; and (4) finally, certainvariants are produced by using some kinds of operator on

Figure 7. (a) Waveforms obtained by stacking the events belonging to the three clusters. (b) Amplitudespectra of the average waveforms, shown in Figure 7a. (c) The f�g diagrams of the average waveformsshown in Figure 7a and obtained for all the trial AR orders (4–60). The gray lines represent lines alongwhich the quality factor (Q) is constant. Clusters of points indicate a resolved dominant mode; scatteredpoints represent noise. The solid ellipses encircle clusters of points representing resolved dominant modewith high spectral amplitude.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

7 of 15

B08308

the surviving candidate solutions [Mitchell, 1996]. Theidentification problem can be formulated as an optimizationtask whose aim is to find a set of parameters that minimizethe prediction error between measured data and the modeloutput.[24] In this case, the model is represented by the vibration

of a large gas bubble before bursting, mathematicallydescribed by the equations of Vergniolle and Brandeis[1996] and Vergniolle et al. [2004], as mentioned above,and the measured data for each cluster consist in the wave-forms of the two events with minimum and maximumvalues of peak-to-peak amplitude, and the stacked wave-form. The comparison between synthetic and observed orstacked waveforms is reported in Figure 10 and shows thatthe fit is good at least for the first cycle of vibration: anexplanation can be that at the end of the first cycle thebubble bursts and then the model ceases to be valid. Theresults of the inversion (Table 1) exhibit smaller bubblesfor clusters 2 and 3 (roughly radius of 3–4m and length of 2–7 m) than for cluster 1 (roughly radius of 5–7m and length of11–16 m). Moreover, the initial overpressure is higher forclusters 2–3 (in the range 0.06–1.15 MPa) than cluster 1 (inthe range 0.02–0.32 MPa). It is also worth noting that cluster3 shows the maximum values of overpressure (up to1.15 MPa). The obtained values of the three unknownparameters are reasonable and in good agreement withpreviously obtained values at Etna [Vergniolle, 2003;Vergniolle and Ripepe, 2008; Cannata et al., 2009], atStromboli [Vergniolle and Brandeis, 1996], and at Shishaldin[Vergniolle et al., 2004]. Moreover, given that clusters 2and 3 shared the same source crater, the similarity of thebubble radii of these clusters was expected. In fact, thebubble radius strongly depends on the size of the source

vent. Finally, the lower overpressure value of cluster 1 isconsistent with the longer duration of the events. Indeed,high gas overpressure produces large pressure perturba-tions in the atmosphere with short duration. Conversely,low regimes of gas overpressure generate long lastingexplosions with small acoustic amplitude [Wilson andHead, 1981; Ripepe and Marchetti, 2002].

2.6. Emitted Gas Volume

[25] Following the method shown by Vergniolle et al.[2004], Vergniolle [2008], and Vergniolle and Ripepe[2008], we quantified the emitted gas volume per day forNEC and SEC. In this case the second data set, comprisingthe detectable events occurring during the studied period,was used. The whole data set was divided into subsets, eachcontaining the events occurring in a day. Then, on the basisof the spectral features of the events, each subset was againdivided into three subsets, containing events with similarspectral content to the ones belonging to clusters 1, 2 and 3.Therefore, assuming that events of all the three clustersoccurred every day during the studied 3 months, 91 times 3subsets should be created. However, only 103 subsets wereformed because of the lack of events in some clusters duringseveral days (especially cluster 3, taking place onlyduring the first days of September). Successively, thenumber of events and the stacked waveform at EBELstation (Figure 11) were calculated for each obtained subset.Then, by performing the waveform inversion, as shown insection 2.5, the bubble volume and the overpressure werecalculated for each stacked waveform. The bubble gasvolume expelled in the atmosphere can then be deducedfrom the gas volume at the vent by using the overpressurevalues and the perfect gas law [Vergniolle and Ripepe,

Figure 8. Difference between the arrival times of the infrasonic events at EBEL and EPDN (bluediamonds), at EBEL and ECPN (pink squares), and at EBEL and EPLC (yellow triangles).

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

8 of 15

B08308

2008]. Finally, the daily emitted gas volume was calculatedfor each cluster by multiplying the bubble gas volume in theatmosphere with the daily number of events. The dailyemitted gas volume of the NEC coincides with the total gasvolume of the subset of events similar to cluster 1, whereasthe one of the SEC is equal to the sum of the gas volume ofthe subsets of events similar to clusters 2 and 3. Moreover,we estimated the daily duration of the time periods charac-terized by infrasonic tremor (recorded at the same time asthe two lava fountain episodes) and bad weather, when wewere not able to detect the events (Figure 12a). Then, for thedays characterized by wind and/or tremor we multiplied theestimated daily emitted gas volume by a factor calculated onthe basis of the percentage of the day when the eventdetection was not possible. For example, assuming thatduring a certain day we estimated a time interval of 12 h of

impossible event detection, the gas volume estimated for thatday would be multiplied by 2. Thus, we could correct thedaily emitted gas volume taking into account also theprospective nondetectable infrasonic events during these‘‘blind time intervals’’ (Figures 12b, 12c, and 12d). Insummary, this method enabled estimating the emitted gasvolume for each active crater. The average and maximumdaily emitted gas volume were equal to about 106 and107 m3, respectively, at NEC, and about 105 and 106 m3,respectively, at SEC.[26] It is worth noting that infrasound measurements can

only detect bubbles with an inner overpressure and thuslikely coming from depth. Surface degassing activity ob-served at vents results from both the bubbles coming fromdepth and those quasi-stagnant in the conduit, which aresilent on acoustic records [Vergniolle and Ripepe, 2008].

Figure 9. Location of the stacked events of the nine families recognized by the waveform classification.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

9 of 15

B08308

Consequently, degassing related to the former is calledactive, while degassing caused by the latter is passive[Vergniolle and Ripepe, 2008]. Therefore, by using theaforementioned method we estimated the daily activedegassing per crater. This was also confirmed by theoverpressure values calculated for each stacked waveform(Figure 12e), much higher than the ones obtained by thecoalescence of two bubbles in a conduit [Vergniolle andCaplan-Auerbach, 2004; Vergniolle and Ripepe, 2008].Finally, NEC has been the most active crater from theactive degassing point of view, although most of theexplosive and effusive eruptions have affected SEC duringthe investigated period.

3. Volcanic Activity

[27] In September–November 2007 the eruptive activityat Mount Etna volcano, described in the reports of IstitutoNazionale di Geofisica e Vulcanologia, Sezione di Catania(INGV Staff, Volcanological and geochemical reports, 2007,available at http://www.ct.ingv.it/Default.asp?Pagina=Report.asp&Anno=2007&UF=UFVG), was characterized by explo-sive activity and degassing at two summit craters (Figure 13a).In particular, SEC was affected by Strombolian activity(resumed on 15 August 2007), ash emission and two lavafountain episodes that took place on 4–5 September and

23–24 November. These two episodes lasted several hoursand gave rise to lava flows on the eastern flank of thevolcano. Strong degassing activity was observed at NEC.Moreover, we used the images recorded by a video cameralocated on the upper southern flank of the volcano (ca. 4 kmaway from SEC) to follow the eruptive activity of SEC indetail. The images of video camera are stored at a rate of30 frames per minute. Unfortunately, only very short time

Table 1. Strombolian Bubble Parameters Obtained by the Wave-

form Inversiona

R (m) L (m) P (MPa)

Min (cluster 1) 6 15 0.02Max (cluster 1) 5 16 0.32Stacked (cluster 1) 7 11 0.08

Min (cluster 2) 3 7 0.06Max (cluster 2) 4 2 0.56Stacked (cluster 2) 4 6 0.13

Min (cluster 3) 3 5 0.08Max (cluster 3) 3 2 1.15Stacked (cluster 3) 3 4 0.23

aR, L, and P indicate radius, length of the bubble and initial overpressure,respectively. Min, Max, and Stacked indicate the waveforms of the eventswith minimum and maximum peak-to-peak amplitude, and the stackedwaveforms, respectively.

Figure 10. Comparison between the observed and stacked waveforms of the three clusters of infrasonicevents (black) and the synthetic ones (gray). Min, Max, and Stacked indicate the waveforms of the eventswith minimum and maximum peak-to-peak amplitude and the stacked waveforms, respectively. Thesource parameters obtained by the waveform inversion are reported in Table 1.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

10 of 15

B08308

periods could be analyzed because of the generally badweather conditions at high altitude. Focusing on the timeperiod 1–4 September, characterized by fairly good visibil-ity of the summit area and by the occurrence of clusters 2and 3, both linked to the activity of SEC (see section 2.4),we noted that the infrasonic events generally occurred at thesame time as explosions taking place at the pit crater openedon the ESE flank of the SEC during the 2004–2005eruption. In particular, the events belonging to cluster 3(that was present only in this time span) were coincidentwith ‘‘more visible’’ explosions than the events of cluster 2.The explosions linked to cluster 3 were characterized by arelevant presence of ash composed by lithic clasts due tomoderate landslides and collapse of the inner walls of thecone (Figure 13b). The different features of the explosionscan be due to (1) variable gas expansion velocity in themagma column; (2) changes in the depth of the bubbleformation; and (3) variations in the gas volume and/or thegas mass fraction in the magma. Because explosions withdifferent features were not separated in time but generallyoccurred during the same time intervals, the hypothesesinvolving continuous and quick variation of the magmacharacteristics are not plausible. In light of this, the differentintensity of events belonging to cluster 3 may be linked tothe occasional occurrence of landslides, as testified byimages gathered. The interpretation of an overload generatedby lithic clasts accumulation that occludes the vent seemsmore reasonable. Moreover, the stronger intensity of theexplosions accompanying the infrasonic events of cluster 3is also supported by the higher values of overpressure ofcluster 3 than cluster 2, obtained by the study of sourcemechanism (see section 2.5). In fact, the overpressure can beconsidered a useful parameter to estimate the ‘‘strength of aneruption’’ [Vergniolle et al., 2004].

4. Discussion and Conclusions

[28] The period September–November 2007 was charac-terized at the summit craters of Mount Etna by explosive

activity and intense degassing. At the same time, a four-sensor network, distributed at various azimuths around thevolcano (Figure 1), continuously recorded the infrasonicsignal (Figure 2). By a triggering procedure, about 1000infrasonic events, occurring in this period and characterizedby high signal-to-noise ratio, were selected and analyzed inorder to investigate possible relationships between differentvolcanic activity at distinct craters and signal variability. Bycross correlation analysis, nine families of events wererecognized (Figures 3 and 4). Successively, on the basisof the cross correlation results and by a parametric approachusing frequency and quality factor values (Figure 5), thesenine families were grouped in three clusters (called clusters1, 2, and 3). Considering frequency and quality factor(estimated by Sompi analysis) as peculiar features of thedetected infrasonic events and by using the so-called‘‘features plane’’ shown in Figure 6, we were able to recoverthe same results obtained using cross correlation method.Cluster 1 showed frequency and quality factor in the range1.2–2.2 Hz and 1.5–6.0, respectively. Cluster 2 was char-acterized by frequency range 2.0–3.5 Hz and quality factor1.5–4.0. Finally, cluster 3 comprises events with frequencyof 3.5–4.5 Hz and quality factor 3–9. Location analysis,performed by a grid searching procedure, allowed us toattribute cluster 1 to the degassing activity of NEC and theother two clusters to the explosive activity of SEC(Figure 9). Consequently, a simple spectral analysis of theinfrasonic signal recorded at a single station can givepreliminary information about location and intensity of theexplosive activity. In any case, in a volcano such as Etna,characterized by almost continuous eruptive activity withdifferent styles and topographical variations of the summitarea, the vent geometry can change and consequently alsothe infrasound spectral features. Therefore, spectral charac-terization and source location have to be considered com-plementary, especially when long lasting periods areinvestigated.

Figure 11. Examples of (a) daily subsets of infrasound events and (b) corresponding stacked waveforms(see section 2.6 for details).

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

11 of 15

B08308

Figure 12. (a) Daily number of hours, when the detection of events was not possible because ofinfrasonic tremor during paroxysmal volcanic activity or noise mainly due to bad weather. (b) Total dailyemitted gas volume of NEC and SEC, calculated by inversion of infrasound events (see section 2.6 fordetails). (c and d) Daily emitted gas volume of NEC and SEC, respectively. (e) Overpressure valuescalculated by waveform inversion of each daily stacked event related to clusters 1, 2, and 3 (see legend).

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

12 of 15

B08308

[29] Given that the detected infrasonic events are similarto the signals described by Vergniolle and Brandeis [1996],they were modeled by the vibration of a large gas bubblebefore bursting. Hence, the waveform inversion of the threeclusters was performed (Figure 10 and Table 1). The sourcemechanism of cluster 1, related to the activity of NEC, wascharacterized by larger bubbles (radius and length of 5–7and 11–16 m, respectively) and lower initial overpressure(0.02–0.32 MPa) than the other two clusters. The longerduration of the events of cluster 1 can be caused by thelower overpressure value. In fact, low regimes of gasoverpressure generate long lasting explosions with smallacoustic amplitude [Wilson and Head, 1981; Ripepe andMarchetti, 2002]. Moreover, cluster 2 showed lower over-pressure values than cluster 3 (0.06–0.56 and 0.08–1.15MPa, respectively). Detailed observations of the eruptive

activity at SEC permitted to relate this difference of theinfrasonic source with different intensity of the explosionscoinciding with the infrasonic signals (Figure 13). Indeed,the events of cluster 3 occurred at the same time as moreintense explosions than cluster 2. It is our view that theoccasional accumulation of the lithic clasts on the bottom ofpit crater of SEC (due to moderate landslides of its walls)causes an overload and a modification of the physicalcharacteristics of the explosive vent (i.e., the thickness ofmagma above the vibrating bubble). These two featuresmay explain both the greater strength of the explosions andthe different waveforms of cluster 3 with respect to cluster2, both observed at the same crater. Other possible explan-ations of the higher overpressure values of cluster 3 can be(1) variable gas expansion velocity in the magma column;(2) small changes in the depth of the bubble formation; and

Figure 13. (a) Eruptive activity at Mount Etna between September and November 2007 shown for twosummit craters (SEC, southeast crater; NEC, northeast crater) (1, degassing; 2, ash emission; 3,Strombolian activity; 4, lava fountains). See Figure 1 for details of the summit area. (b) Video cameraframes showing explosive activity at the SEC recorded at the same time as infrasonic events belonging toclusters 2 and 3.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

13 of 15

B08308

(3) variations of the gas volume and/or the gas mass fractionin the magma. Because events belonging to clusters 2 and 3occurred during the same intervals, the hypotheses involv-ing continuous and quick variation of the magma character-istics are not plausible. After the 4 September lava fountain,landslides were no longer observed and contemporaneouslythe occurrence of cluster 3 stopped. Consequently, webelieve that accumulation of the lithic clasts on the bottomof the crater is a very reasonable hypothesis to explain theorigin of cluster 3.[30] The daily emitted gas volume, related to active

degassing processes [Vergniolle and Ripepe, 2008], wasalso estimated for NEC and SEC by using a different dataset to the one previously used and comprising all thedetectable infrasonic events during the studied period(Figure 12). The average daily emitted gas volumes areroughly equal to 105–106 m3 at SEC and NEC, respectively.Aiuppa et al. [2008] evaluated the total volatile flux fromMount Etna by summing the simultaneously evaluated fluxesof the three main volcanogenic volatiles, measured duringJanuary 2005 to July 2007: H2O, CO2 and SO2. In particular,SO2 flux was determined by routine differential opticalabsorption spectrometry (DOAS) traverse measurements,while H2O and CO2 were evaluated by scaling MultiGAS-sensed H2O/SO2 and CO2/SO2 plume ratios to the UV-sensedSO2 flux. The obtained total volatile flux was equal to about21,000 t/d [Aiuppa et al., 2008]. The density value of thevolcanic gas strongly depends on its composition. On thebasis of the compositions of the total volatile flux at MountEtna during eruptive and noneruptive periods estimated byAiuppa et al. [2008] and of the density of the single gasspecies [e.g., Parfitt, 2004], the emitted daily gas volumesrange between 8 and 9 *107 m3/d. This value is much higherthan the one estimated by infrasound recordings. In fact, atvolcanoes gas is also emitted without detectable infrasoundradiation [Johnson, 2003; Vergniolle and Ripepe, 2008].Further, it is worth noting that NEC proved to be the mostactive crater from the active degassing point of view,although the explosive and effusive eruptions have affectedSEC during the studied period.[31] Finally, it is particularly interesting to note that

activity at SEC is often associated with activity at NEC.This could be due to NEC being open most of the time andonly quietly degassing. However, the investigation of theprospective links among the summit craters at Mount Etnarequires longer time series and is beyond the scope of thiswork.[32] In conclusion, the subdivision of the infrasonic

events into three clusters on the basis of spectral featuresproved to have an important ‘‘physical’’ meaning, depend-ing on the source vent and on the eruptive activity. Thesefindings imply that a simple Sompi analysis of the infra-sonic signal recorded at a single station provides prelimi-nary information about location and intensity of theexplosive activity. Moreover, the infrasound data appearsto be a useful tool for quantifying the emitted gas volumefor each active crater, related to active degassing processes.[33] This work shows that the study of infrasonic signals

at volcanoes is a very powerful tool for monitoring activevolcanoes, since it allows locating explosive activity andestimating its intensity. Both these data are also very useful

to give important contributions to the understanding ofexplosive eruption mechanisms.

[34] Acknowledgments. We thank Sylvie Vergniolle, an anonymousreviewer, and the Associated Editor for their constructive comments. PeggyHellweg is also acknowledged for her useful advice on the waveformstacking method.

ReferencesAiuppa, A., G. Giudice, S. Gurrieri, M. Liuzzo, M. Burton, T. Caltabiano,A. J. S. McGonigle, G. Salerno, H. Shinohara, and M. Valenza (2008),Total volatile flux from Mount Etna, Geophys. Res. Lett., 35, L24302,doi:10.1029/2008GL035871.

Almendros, J., and B. Chouet (2003), Performance of the radial semblancemethod for the location of very long period volcanic signals, Bull. Seis-mol. Soc. Am., 93, 1890–1903, doi:10.1785/0120020143.

Behncke, B., and M. Neri (2003), Cycles and trends in the recent eruptivebehaviour of Mount Etna (Italy), Can. J. Earth Sci., 40, 1405–1411,doi:10.1139/e03-052.

Buckingham, M. J., and M. A. Garces (1996), A canonical model of vol-cano acoustics, J. Geophys. Res., 101, 8129 – 8151, doi:10.1029/95JB01680.

Cannata, A., P. Montalto, E. Privitera, G. Russo, and S. Gresta (2009),Tracking eruptive phenomena by infrasound: May 13, 2008 eruption atMt. Etna, Geophys. Res. Lett., 36, L05304, doi:10.1029/2008GL036738.

Cristofolini, R., S. Gresta, S. Imposa, and G. Patane (1988), Feeding me-chanism of eruptive activity at Mt. Etna based on seismological andpetrological data, in Modelling of Volcanic Process, edited by C. Y. Kingand R. Scarpa, pp. 73–93, Springer, Berlin.

Firstov, P. P., and N. M. Kravchenko (1996), Estimation of the amount ofexplosive gas released in volcanic eruptions using air waves, Volcanol.Seismol., 17, 547–560.

Garces, M. A., and S. R. McNutt (1997), Theory of the airborne sound fieldgenerated in a resonant magma conduit, J. Volcanol. Geotherm. Res., 78,155–178, doi:10.1016/S0377-0273(97)00018-8.

Garces, M., A. Harris, C. Hetzer, J. Johnson, S. Rowland, E. Marchetti, andP. Okubo (2003), Infrasonic tremor observed at Kilauea Volcano, Hawaii,Geophys. Res. Lett., 30(20), 2023, doi:10.1029/2003GL018038.

Geller, R. J., and C. S. Mueller (1980), Four similar earthquakes in centralCalifornia, Geophys. Res. Lett . , 7 , 821 – 824, doi:10.1029/GL007i010p00821.

Gordeev, E. I. (1993), Modelling of volcanic tremor as explosive pointsources in a singled-layer, elastic half-space, J. Geophys. Res., 98,19,687–19,703, doi:10.1029/93JB00348.

Green, D., and J. Neuberg (2006), Waveform classification of volcanic low-frequency earthquake swarms and its implication at Soufriere Hills Vol-cano, Monserrat, J. Volcanol. Geotherm. Res., 153, 51–63, doi:10.1016/j.jvolgeores.2005.08.003.

Gresta, S., M. Ripepe, E. Marchetti, S. D’Amico, M. Coltelli, A. J. L.Harris, and E. Privitera (2004), Seismoacoustic measurements duringthe July August 2001 eruption at Mt. Etna volcano, Italy, J. Volcanol.Geotherm. Res., 137, 219–230, doi:10.1016/j.jvolgeores.2004.05.017.

Hagerty, M. T., S. Y. Schwartz, M. A. Garces, and M. Protti (2000), Ana-lysis of seismic and acoustic observations at Arenal Volcano, Costa Rica,1995–1997, J. Volcanol. Geotherm. Res., 101, 27–65, doi:10.1016/S0377-0273(00)00162-1.

Johnson, J. B. (2003), Generation and propagation of infrasonic airwavesfrom volcanic explosions, J. Volcanol. Geotherm. Res., 121, 1 – 14,doi:10.1016/S0377-0273(02)00408-0.

Johnson, J. B. (2004), Source location variability and volcanic vent map-ping with a small-aperture infrasound array at Stromboli Volcano, Italy,Bull. Volcanol., 67, 1–14, doi:10.1007/s00445-004-0356-8.

Johnson, J. B., and J. M. Lees (2000), Plugs and chugs: Seismic andacoustic observations of degassing explosions at Karymsky, Russia andSangay, Ecuador, J. Volcanol. Geotherm. Res., 101, 67–82, doi:10.1016/S0377-0273(00)00164-5.

Johnson, J. B., J. M. Lees, and E. I. Gordeev (1998), Degassing explosionsat Karymsky Volcano, Kamchatka, Geophys. Res. Lett., 25, 3999–4002,doi:10.1029/1998GL900102.

Jones, K. R., J. Johnson, R. Aster, P. R. Kyle, and W. C. McIntosh (2008),Infrasonic tracking of large bubble bursts and ash venting at Erebusvolcano, Antarctica, J. Volcanol. Geotherm. Res., 177, 661 – 672,doi:10.1016/j.jvolgeores.2008.02.001.

Kumazawa, M., Y. Imanishi, Y. Fukao, M. Furumoto, and A. Yamamoto(1990), A theory of spectral analysis based on the characteristic propertyof a linear dynamic system, Geophys. J. Int., 101, 613–630, doi:10.1111/j.1365-246X.1990.tb05574.x.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

14 of 15

B08308

Lesage, P. (2008), Automatic estimation of optimal autoregressive filters forthe analysis of volcanic seismic activity, Nat. Hazards Earth Syst. Sci., 8,369–376.

Matoza, R. S., M. Hedlin, and M. Garces (2007), An infrasound array studyof Mount St. Helens, J. Volcanol. Geotherm. Res., 160, 249 –262,doi:10.1016/j.jvolgeores.2006.10.006.

Mitchell, M. (1996), An Introduction to Genetic Algorithms, 205 pp., MITPress, Cambridge, Mass.

Nakahara, H. (2004), Correlation distance of waveforms for closely locatedevents—I. Implication of the heterogeneous structure around the sourceregion of the 1995 Hyogo-Ken Nanbu, Japan, earthquake (Mw = 6.9),Geophys. J. Int., 157, 1255– 1268, doi:10.1111/j.1365-246X.2004.02278.x.

Neidell, N., and M. T. Taner (1971), Semblance and other coherency mea-sures for multichannel data, Geophysics, 36, 482–497, doi:10.1190/1.1440186.

Neuberg, J., and T. Pointer (2000), Effects of volcano-topography on seis-mic broadband waveforms, Geophys. J. Int., 143, 239–248, doi:10.1046/j.1365-246x.2000.00251.x.

Parfitt, E. A. (2004), A discussion of the mechanism of explosive basalticeruption, J. Volcanol. Geotherm. Res., 134, 77–107, doi:10.1016/j.jvolgeores.2004.01.002.

Patane, D., G. Di Grazia, A. Cannata, P. Montalto, and E. Boschi (2008),Shallow magma pathway geometry at Mt. Etna volcano, Geochem. Geo-phys. Geosyst., 9, Q12021, doi:10.1029/2008GC002131.

Ripepe, M., and E. Marchetti (2002), Array tracking of infrasonic sources atStromboli volcano, Geophys. Res. Lett., 29(22), 2076, doi:10.1029/2002GL015452.

Ripepe, M., P. Poggi, T. Braun, and E. Gordeev (1996), Infrasonic wavesand volcanic tremor at Stromboli, Geophys. Res. Lett., 23, 181–184,doi:10.1029/95GL03662.

Ripepe, M., S. Ciliberto, and M. Della Schiava (2001a), Time constraint formodelling source dynamics of volcanic explosions at Stromboli, J. Geo-phys. Res., 106, 8713–8727, doi:10.1029/2000JB900374.

Ripepe, M., M. Coltelli, E. Privitera, S. Gresta, M. Moretti, andD. Piccinini (2001b), Seismic and infrasonic evidences for an impulsivesource of the shallow volcanic tremor at Mt. Etna, Italy, Geophys. Res.Lett., 28, 1071–1074, doi:10.1029/2000GL011391.

Ripepe, M., P. Poggi, and E. Marchetti (2004), Small aperture infrasonicarray monitors activity of Stromboli volcano, Inframatics, 7, 1–14.

Rowe, C. A., R. C. Aster, P. R. Kyle, R. R. Dibble, and J. W. Schlue (2000),Seismic and acoustic observations at Mount Erebus Volcano, Ross Island,Antarctica, 1994–1998, J. Volcanol. Geotherm. Res., 101, 105–128,doi:10.1016/S0377-0273(00)00170-0.

Ruiz, M. C., J. M. Lees, and J. B. Johnson (2006), Source constraints ofTungurahua volcano explosion events, Bull. Volcanol., 68, 480–490,doi:10.1007/s00445-005-0023-8.

Tanguy, J. C., G. Kieffer, and G. Patane (1996), Dynamics, lava volume andeffusion rate during the 1991–1993 eruption of Mount Etna, J. Volcanol.Geotherm. Res., 71, 259–265, doi:10.1016/0377-0273(95)00081-X.

Vergniolle, S. (2003), Listening to volcanoes as a tool to understand erup-tion dynamics: Etna (Italy), Shishaldin (Alaska), paper presented atIUGG XXIII General Assembly State of the Planet: Frontiers and Chal-lenges, Sapporo, Japan.

Vergniolle, S. (2008), From sound waves to bubbling within a magmareservoir: Comparison between eruptions at Etna (2001, Italy) andKilauea (Hawai’i), in Fluid Motions in Volcanic Conduits: A Source ofSeismic and Acoustic Signals, edited by S. J. Lane and J. S. Gilbert, Geol.Soc. Spec. Publ., 307, 125–146.

Vergniolle, S., and G. Brandeis (1994), Origin of the sound generated byStrombolian explosions, Geophys. Res. Lett., 21, 1959 – 1962,doi:10.1029/94GL01286.

Vergniolle, S., and G. Brandeis (1996), Strombolian explosions: A largebubble breaking at the surface of a lava column as a source of sound,J. Geophys. Res., 101, 20,433–20,448, doi:10.1029/96JB01178.

Vergniolle, S., and J. Caplan-Auerbach (2004), Acoustic measurements ofthe 1999 basaltic eruption of Shishaldin volcano, Alaska: 2. Precursor tothe Subplinian activity, J. Volcanol. Geotherm. Res., 137, 135–151.

Vergniolle, S., and M. Ripepe (2008), From Strombolian explosions to firefountains at Etna Volcano (Italy): What do we learn by acoustic measure-ment?, in Fluid Motions in Volcanic Conduits: A Source of Seismic andAcoustic Signals, edited by S. J. Lane and J. S. Gilbert, Geol. Soc. Spec.Publ., 307, 103–124.

Vergniolle, S., G. Brandeis, and J. C. Mareschal (1996), Strombolian ex-plosions: Eruption dynamics determined from acoustic measurements,J. Geophys. Res., 101, 20,449–20,466, doi:10.1029/96JB01925.

Vergniolle, S., M. Boichu, and J. Caplan-Auerbach (2004), Acoustic mea-surements of the 1999 basaltic eruption of Shishaldinvolcano, Alaska:1. Origin of Strombolian activity, J. Volcanol. Geotherm. Res., 137,109–134, doi:10.1016/j.jvolgeores.2004.05.003.

Wilson, L., and J. W. Head (1981), Ascent and eruption of basaltic magmaon the Earth and Moon, J. Geophys. Res., 86, 2971–3001, doi:10.1029/JB086iB04p02971.

�����������������������A. Cannata, P. Montalto, and E. Privitera, Istituto Nazionale di Geofisica

e Vulcanologia, Sezione di Catania, Piazza Roma 2, I-95123 Catania, Italy.(privitera@ct.ingv.it)G. Russo, Consorzio Cometa, via Santa Sofia, 64, I-95100 Catania, Italy.

B08308 CANNATA ET AL.: INFRASONIC SOURCES AT MT. ETNA

15 of 15

B08308