In this work, waveform variations in repeating volcano-tectonic earthquakes occurring from 2001–2009 in the north-eastern flank of Mt. Etna were studied. Changes in waveform were found mainly during 2002–2003; and consisted of a decreasing similarity in the coda of events in earthquake families, as revealed by cross-correlation analysis, and delays, increasing proportionally to the lapse time, detected by coda wave interferometry. Such variations, mainly evident at stations located in the north-eastern flank of the volcano, were likely due to medium changes taking place within this region. Localized medium velocity decreases were inferred to occur in 2002–2003, followed by successive increases. The velocity decrease was interpreted as being caused by the opening or enlargement of cracks, produced by intruding magma bodies, intense ground deformation, and/or VT earthquake activity that accompanied the 2002–2003 Mt. Etna eruption. On the other hand, subsequent velocity increases were interpreted as resulting from healing processes.
 Multiplets, also called repeating earthquakes or earthquake families, are earthquakes with similar waveforms. As a result of their characteristic of having repeatable sources at the same spot but at different times [Schaff and Beroza, 2004], many seismological applications make use of repeating earthquakes, as follows: to highlight seismic structures at depth by highly precise locations [e.g., Waldhauser et al., 2004]; to acquire information on the dynamics of active faults and their slip rates [e.g., Chen et al., 2008]; and to detect temporal variations of attenuation [e.g., Antolik et al., 1996], shear wave splitting [e.g., Zaccarelli et al., 2009], and medium velocity [e.g., Schaff and Beroza, 2004]. The last application can be particularly important in a volcanic environment, where the detection of temporal changes in the subsurface structure plays a key role in both understanding volcanic dynamics and in monitoring activity variations.
 In this work, repeating volcano-tectonic (VT) earthquakes taking place at Mt. Etna during 2001–2009 were analyzed in order to highlight waveform changes that were likely related to crustal structure variations caused by volcanic and/or tectonic activities.
2. Data Analysis
Cannata et al. analyzed the repeating VT earthquakes taking place at Mt. Etna from 1999–2009. Families were detected via cross-correlation analysis. For this work, those families found byCannata et al. , located on the north-eastern flank of Etna (which includes the Pernicana fault system;Figure 1a) and containing a minimum of four events over a span of at least 1 year, were selected. These Pernicana families occurred throughout the entire 1999–2009 period and provided an opportunity to look for changes in material properties over a broad period of time. Here, 14 families with a total of 106 VT earthquakes (see Figure S1 in the auxiliary material for an example of a VT family recorded at the EZPO station) were utilized, with each family having between 4–16 events ranging in magnitude from 1.3 to 3.4. Data from 12 seismic stations, belonging to the permanent network run by the Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Etneo – Sezione di Catania (Figure 1a) were employed.
 Following Chen et al. [2008, 2009], differential S-P times were calculated in order to verify that the events in each family were co-located. To acquire time information at subsample precision, the signals were resampled at 1000 Hz and the differential S-P times for pairs of events belonging to the same family were calculated (Figure S2). Most of the differential times were lower than 0.015 seconds (Figure 2), on the basis of which the hypocentral separations should be at most equal to tens of meters. This confirms the co-location of the VT earthquakes that make up each family.
2.1. Cross-Correlation Analysis
 Prospective time changes in the repeating earthquake waveform similarity were quantitatively evaluated by the cross-correlation coefficient (hereafter referred to as CCC). FollowingGret et al. , two CCC values were calculated for each doublet, the first for a 3-second-long “early” seismogram window containing the first arrival, the second for a 8-second-long “late” seismogram window beginning seven seconds after the P-wave arrival and containing part of the coda. Before calculating the CCC, doublet waveforms were aligned on the first arrival so that the CCC was maximized. The vertical component of the seismic signals was utilized, and doublet event pairs were analyzed in chronological order for all VT pairs in a given family. For each station, the average CCC values of all doublets were calculated for both the early and late seismogram windows. The spatial distribution of these values, as well as of the difference in the average CCC between the early and late seismogram windows, are plotted inFigures 1b–1d. Close to the source area (the north-eastern flank of the volcano) both the highest values of the CCC for the early seismogram windows and the lowest values of the CCC for the late seismogram windows were observed, which is reflected in the high CCC difference values shown inFigure 1d. Station EZPO (Figure 1a) had the longest time series of all of the stations near the multiplet source area on the north-eastern flank.Figure 3 indicates that most of the high CCC difference values calculated at EZPO were related to doublet event pairs that took place from 2002–2003.
 To verify this trend, the analysis was also performed using three-component data and the first event in each family as a reference event for the doublet analysis. The results of this CCC analysis were consistent with those described above.
2.2. Coda Wave Interferometry
 In order to further investigate time variations of the VT family waveforms, the coda wave interferometry technique (CWI) [Poupinet et al., 1984; Snieder, 2006] was applied. The time-windowed CCC was calculated, by comparing two-second-long windows (that overlapped by 1.95 seconds) of the doublets, and time delays corresponding to the maximum CCC were obtained (Figure 4). Prior to comparing the seismic windows, the signals were resampled at 1000 Hz. For changes in medium seismic velocity, time delays should increase for progressively later times in the coda [Snieder et al., 2002; Snieder, 2006]. In order to investigate whether or not the observed CCC changes were caused by changes in the seismic velocity, the time delays were linearly fitted. Following Wegler et al. , only delays with a CCC larger than 0.7 were used. Furthermore, only regression results with a R2 coefficient (goodness of the linear regression fit) higher than 0.8 were taken into account. As for the CCC analysis, doublets in the CWI analysis were constructed either with consecutive VT pairs in a given family, or by using the first event in a family as a reference event (Figures 5a and 5b, respectively). The results indicated that the relative medium velocity changes mainly ranged between −1 and 1%. Schaff and Kim  demonstrated that even “small” event hypocentral separations (e.g. 100 m) can cause biases in velocity change measurements by up to 1%. In this regard, it cannot be excluded that some of the velocity changes (<1%) in Figure 5 are due to small event hypocentral separations. On the other hand, greater velocity changes (decreases up to ∼2%), as obtained for stations EZPO, EPOL, EVNA and ECBD, cannot be justified by these biases and must be considered reliable. The velocity results obtained using consecutive VT pairs indicated that such decreases mainly took place during 2002–2003 and were followed by successive increases (Figure 5a).
3. Discussion and Conclusions
 Waveform variations in the repeating VT earthquakes occurring during 2001–2009 in the north-eastern flank of Mt. Etna were sought. Clear changes, mainly taking place during 2002–2003, were determined; and consisted of a decreasing similarity in the late seismogram windows and of delays, increasing proportionally to lapse time, as detected by the CWI analysis.
 A decreasing similarity in the late seismogram windows was especially evident at stations on the north-eastern flank of the volcano (Figure 1). These changes could not be due to slight source shifts or source mechanism variations, since in both cases the effects on the seismic coda would have been of the same order of magnitude as the effect on the first arriving energy [Gret et al., 2005; Snieder and Vrijlandt, 2005]. Furthermore, if these changes were caused by a source shift, the CCC would be lowered at many stations over a wide area [Yamawaki et al., 2004], instead of localized to a few stations such as at Etna. This suggests that the detected late seismogram waveform variations were due to temporal changes of the crustal structure. Indeed, small medium changes have been shown to produce negligible changes in the early portions of seismograms, in contrast to the coda where multiply-reflected waves are more heavily influenced by localized velocity changes [Snieder et al., 2002; Gret et al., 2005]. Given that the changes were mainly apparent at EZPO and EVNA, it was inferred that the medium changes occurred on the north-eastern flank of the volcano close to these two stations. Also, it is worth noting that such changes were localized not only in space but also in time. Indeed, as shown inFigure 3, the changes mainly took place during 2002–2003.
 The CWI analysis provided independent support for velocity changes. Velocity decreases of up to 2% during 2002–2003 were observed, followed by a recovery at stations EZPO, EVNA, ECBD, and EPOL (Figure 5). The first three stations are located in the north-eastern flank of the volcano, the same area where evidence of velocity changes was seen using CCC variations. Comparable velocity changes have been observed by authors within both volcanic and tectonic areas. For instance, reductions in the phase velocity by 1.5% were detected at Mt. Asama byNagaoka et al.  prior to the 2008 eruption. Battaglia et al.  found a sudden velocity drop of a few percent at the Yasur volcano, following an M = 7.3 earthquake.
 Etna experienced one of the most intense eruptions of the previous decade during 2002–2003, when several eruptive fissures opened on the southern and north-eastern flanks (see the red lines inFigure 1a). This eruption began in October 2002 and ended in January 2003, after three months of continuous explosive activity and discontinuous lava flow output [Andronico et al., 2005]. The 2002–2003 eruption was accompanied and followed by an important eastern flank dynamics, as determined by intense deformation [Palano et al., 2006], ground fracturing [Neri et al., 2004] and strong shallow VT activity along the Pernicana fault system [Alparone et al., 2012]. Indeed, the eastern flank of Mt. Etna is affected by a continuous seaward sliding, and the Pernicana fault system is considered to be the northern boundary of the unstable sector [e.g., Palano et al., 2006].
 Therefore, it seems reasonable to infer that the intruding magma bodies, that fed the north-eastern flank fissures, as well as VT activity and ground deformation along the Pernicana fault system, could have produced the observed temporal velocity changes. All of these phenomena can cause opening or enlargement of cracks, and hence medium velocity decreases [Schaff and Beroza, 2004; Patanè et al., 2006]. The subsequent recovery in velocities, observed during the following months/years, was also in agreement with this interpretation, and can be read as resulting from the healing processes taking place within the rocks of the north-eastern volcano flank following intrusive and tectonic activities. Similar healing processes were inferred bySchaff and Beroza  at the San Andreas fault following the 1989 Loma Prieta earthquake.
 Gruppo Analisi Dati Sismici of Istituto Nazionale di Geofisica e Vulcanologia, Osservatorio Etneo - Sezione di Catania, is kindly acknowledged for providing the VT location information. I thank Stephen Conway and Kimberly Mace for revising and improving the English text. I am grateful to Seth Moran and David Schaff for their useful suggestions that greatly improved the paper.
 The Editor thanks Seth Moran and David Schaff for assisting in the evaluation of this paper.