Families of similar events and modes of oscillation of the conduit at Yasur volcano (Vanuatu)

Journal of Volcanology and Geothermal Research 322 (2016) 196–211

Contents lists available at ScienceDirect

Journal of Volcanology and Geothermal Research journal homepage: www.elsevier.com/locate/jvolgeores

Families of similar events and modes of oscillation of the conduit at Yasur volcano (Vanuatu) Jean Battaglia a,⁎, Jean-Philippe Métaxian b, Esline Garaebiti c a b c

Laboratoire Magmas et Volcans, Université Blaise Pascal - CNRS - IRD, OPGC, 5 rue Kessler, 63038 Clermont Ferrand, France ISTerre, IRD R216, CNRS, Univ. de Savoie, Le Bourget du Lac, 73376, France Vanuatu Meteorology and Geohazards Department, P.M.B 9054 Port Vila, Vanuatu

a r t i c l e

i n f o

Article history: Received 22 July 2015 Accepted 4 November 2015 Available online 10 November 2015 Keywords: Yasur Strombolian activity Volcano seismology Explosion quakes Similar events

a b s t r a c t We examined one year of seismic recordings collected in 2008 during a temporary experiment at Yasur volcano (Vanuatu). The volcano has a permanent Strombolian activity that was at a relatively high level during most of our experiment with commonly at least one explosion per minute. Associated with this activity, the network recorded intense seismicicity with hundreds of transients per day. Video recordings indicate that most of the high frequency transients are directly related to the strombolian explosions. They also outline the presence of fewer signals which are not accompanied by any surface activity. The classification of transient events recorded at a station close to the summit indicates that a significant part of the events exhibit waveform similarity. This technique allows the identification of characteristic repeating events among the hundreds of thousands of transients recorded during the experiment. Most of the families of similar events are groups of explosion quakes (EQs) but a few are groups of Long Period events related to deeper processes. By scanning the 9 months of continuous data available at the summit station with master events extracted from these families we reconstruct their temporal evolution. Our results show that several families dominate the activity with a few of them lasting for several months. We show that their temporal evolutions can be used to probe changes in the structure or activity of the volcano. We observe that a major change was induced by a M = 7.3 subduction earthquake which occurred on April 9, 2008 about 80 km from the volcano. While this event did not change significantly the surface morphology of the volcano nor the intensity of the eruptive activity, it interrupted the families as none of them is present both before and after the event. This change in the waveforms can be explained by a drop in the seismic velocity of the volcano caused by the distal event. Numerous other transitions between families are observed, sudden or progressive. These can be interpreted as representative of changes in the eruptive dynamics. The presence of similar EQs, especially for impulsive explosions, indicates that the source mechanism is reproducible and has a stable location for some periods. This favors a source process based on the oscillation of the conduit or oscillation of the edifice in response to the explosive decompression of gas slugs at the free surface of the conduit. Our results suggest that the seismic activity of Yasur is characterized by the presence of dominant modes of resonance of the conduit which may be influenced both by external and internal factors. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Yasur volcano is located in the south-eastern part of Tanna island, in the south of Vanuatu (South-West Pacific). It is part of an island arc related to the subduction of the Australian plate below the Pacific plate. Yasur is a small basaltic scoria cone (360 m a.s.l.), which is part of a larger structure: the Yenkahe resurgent dome. The dome is characterized by a discontinuous uplift with a mean rate estimated by coral dating of 15.6 cm/year over the last 1000 years (Carney and Macfarlane, 1979; Chen et al., 1995). The dome is located inside the 24 km2 Siwi caldera (Coulon and Maury, 1981; Robin et al., 1994; Allen, 2005). Yasur volcano ⁎ Corresponding author. E-mail address: [email protected] (J. Battaglia).

http://dx.doi.org/10.1016/j.jvolgeores.2015.11.003 0377-0273/© 2015 Elsevier B.V. All rights reserved.

has a permanent Strombolian to Vulcanian type of activity since at least 300 years (Simkin et al., 1981), which makes it one of the rare open vent systems with permanent activity. A crater, 400 m in diameter occupies its summit, with 2 sub-craters including 3 vents named A, B and C (Fig. 1). Frequent explosions are observed with commonly at least one explosion per minute. The seismicity observed at Yasur is comparable to other volcanoes with open vent conditions with Strombolian or Vulcanian activity, permanent like Stromboli (Italy) or Erebus (Antarctica) or more discontinuous like Etna (Italy) or Tungurahua (Ecuador) for example. This activity is accompanied by a wide variety of seismic signals generally recorded especially at a close distance (several kilometers) to the vents. Explosion Quakes (EQs) are observed directly associated with the explosions (Ripepe, 1996). In the short period range (N0.5 Hz) most of their energy

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

d

197

Y32

Y05 Y31 Y30

Yasur b a

e

c

Fig. 1. Geological setting and layout of the seismic network. (a) Location of Yasur volcano and epicentral location of a magnitude 7.3 subduction earthquake, which occurred on April 9, 2008 (longitude–latitude coordinates in degrees). (b) Location of the seismic stations in the Southern part of Tanna island and (c) around Yasur volcano (UTM coordinates in km). Seismic antennas are indicated as red circles and broadband stations as blue circles. (d) Summit area of Yasur volcano with the approximate location of the crater rim is indicated by a dotted line as well as the rough location of the 3 active vents A, B and C. (e) Typical sensor distribution for a 7-sensor (9-component) seismic station.

is often found between 1 and 5 Hz. They frequently also include a high frequency (N 5 Hz) acoustic air wave (Braun and Ripepe, 1993; Ripepe and Braun, 1994) but this phase is sometimes unclear or not visible. At frequencies below 0.5 Hz, Very Long Period signals (VLP) are often observed with EQs (Neuberg et al., 1994; Rowe et al., 1998; Chouet et al., 2003). The source of VLP signals is typically attributed to the coalescence (Ripepe et al., 2001) or ascent (James et al., 2006; Chouet et al., 2008) of gas slugs but no clear consensus is found in the literature about their source mechanism. In the short period frequency range more or less permanent tremor (Wassermann, 1997; Gordeev, 1999) accompanies degassing, fountaining, sloshing and bubbling in the eruptive vents. In between tremor and clear short duration EQs related to impulsive explosions, volcanoes emit various intermediate signals related to jetting and longer duration material emissions. Short period signals described in the previous paragraph can be included in the larger family of Long Period seismicity related to fluid transfers and resonances of fluid filled volumes commonly observed on volcanoes. Long Period (LP) transients are commonly interpreted as the response of a fluid-filled resonator to the excitation of a triggering mechanism (Chouet and Matoza, 2013). Various resonating sources have been proposed such as a spherical source (Crosson and Bame, 1985), a cylindrical pipe (Chouet, 1985) or a crack (Chouet, 1988). Various triggering mechanisms have also been proposed. They are summarized by Chouet and Matoza (2013) into 5 categories: (1) self-sustained fluid oscillations, (2) magma-hydrothermal interactions, (3) magmatic degassing, (4) brittle fracture of melt, and (5) solid extrusion dynamics and plug stick-slip. LP events are observed at active volcanoes as well as quiescent ones. They may be very shallow when related to surface processes like degassing for example but deeper LPs have also been observed, sometimes precursory to eruptions (White, 1996). Repeating similar LP events

have been observed on volcanoes like Kilauea (Battaglia et al., 2003; Matoza et al., 2014), Etna (Lokmer et al., 2007) and Shishaldin (Petersen, 2007). Seismicity of Yasur has been previously studied by Nabyl et al. (1997) who found a good similarity with the seismicity observed at Stromboli volcano. Kremers et al. (2013) located and inverted the source mechanism of 24 EQs using a multi-parameter dataset and found mostly isotropic source mechanisms. In present paper we examine the seismicity recorded during a one-year experiment carried in 2008–2009. We first examine the temporal evolution during the entire experiment of parameters such as the number of detected transient events, the average seismic RMS amplitude and the amplitude of individual events. We next correlate several hours of video recordings with the corresponding seismicity to identify the origin of the seismic transients. Then, to simplify the analysis of the hundreds of thousands of events recorded during the experiment, we search for characteristic repeating events as they might have specific importance for understanding the eruptive dynamics. We recompose the temporal evolution of such repeating events and use them to probe changes in the volcanic structure or eruptive dynamics. 2. Data We use data collected by a temporary network, which was installed around Yasur volcano in 2008 for a duration of one year. The network included up to 22 stations with an installation done in 2 parts. At the end of January 2008, 12 seismic stations were installed. Nine of which had a starlike geometry (Fig. 1e) and included 9 short period components with one three component sensor surrounded by 6 1-component sensors distributed every 60° at 20 or 40 m distances from the center. The peripheral components are named clockwise C00 to C05 with C00 being roughly oriented

198

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

toward the eruptive crater. The components of the central sensor are named C06, C07 and C08 respectively for the vertical (Z), North-South (NS) and East-West (EW) components. Two additional antennas (Y11 and Y02) had only 6 short period components with a comparable starlike geometry except for the absence of the 3 shorter branches. One station (Y12) was only equipped with a 3-component sensor. All of these stations were equipped with Agecodagis CDJZ short period sensors with a 2 Hz natural frequency connected to an Osiris or a Kephren Agecodagis digitizer. In May 2008, 10 broadband stations equipped with CMG-40T broadband Guralp sensors with natural periods of 30 and 20 s (components named C00, C01 and C02 respectively for Z, NS and EW) and Kephren Agecodagis digitizers were installed. All of the stations provided continuous recordings with a 100 Hz sampling frequency and a storage on local USB disks. In October 2008, 10 stations were removed to be installed on the island of Ambrym (Vanuatu) and the network continued working with a reduced station coverage (4 broadband stations and 6 antennas). All stations were removed at the beginning of February 2009.

In this paper we use mostly data from station Y05 for classification purpose as it is one of the closest to the summit and was installed at the beginning of the experiment in January 2008. Possibly because of its location on an ash plain near the summit, this station also displays simple waveforms as compared to Y31 located on the cone itself or Y32, which was installed on a ridge near the summit. However, Y05 station was vandalized mid-October 2008 and therefore for waveform classification we only consider the period between January 24 and October 14, 2008, while for the seismicity overview we consider the period from January 24, 2008 to February 6, 2009. 3. Overview of seismicity Our network recorded intense seismicity with hundreds of short period transients per day and significant variations in the seismic energy (Fig. 2). An intense VLP activity below 1 Hz was also recorded by the broadband stations but in the present paper we focus on the

Fig. 2. Helicorder presenting seismic recording at station Y05 (vertical component C03) for the entire day 26/4/2008. Each line shows 30 min of data, beginning of day is in the upper left corner of the figure with onset hours of the different lines indicated on the left side of the plot.

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

short period (N 0.5 Hz) signals. At the stations near the volcano, most of the recorded seismicity is clearly related to the volcanic processes with events generated by the Strombolian activity such as explosion quakes (EQs) and tremors. Additionally some tectonic events have been recorded, with mostly events related to the subduction zone including a magnitude 7.3 subduction earthquake, which occurred about 80 km from the volcano on April 9, 2008 (Fig. 1a). Tectonic earthquakes are, however, rare as compared to the hundreds of daily volcanic signals. To detect and count transient events, an STA/LTA (Short Term Average over Long Term Average) algorithm has been applied to continuous data from one of the vertical components of antenna Y05 and to the vertical component of broadband station Y32. We use a LTA of 1 min and a STA of 1 s and apply the algorithm after filtering the data between 1 and 6 Hz with a 2-pole Butterworth filter as our goal is to detect mostly short period transients. We consider a minimum of 8 s between two successive events to avoid counting twice the same event. This means we count a maximum of 1 event per window of 8 s. Both stations detect almost exactly the same number of events and therefore to get continuous time series in Fig. 3a we use station Y32 to

199

complement measurements from Y05 when the station was not working. Fig. 3a presents time series as number of events counted per windows of 2 h for two different STA/LTA thresholds. Counting all events with an STA/LTA higher than 2.5 includes small events and provides a time series rather constant for the first 200 days of the experiment with about 140–150 events per 2 h. Starting about the end of August (day 243) the number of detected events decreases slowly until the end of the year. A sharp increase in the number of small events is observed at the beginning of 2009 followed by a decrease and increase toward the end of the experiment. Times series for events with an STA/LTA higher than 4.0 display comparable variations to those for the threshold 2.5. The number of events detected is, however, nearly twice smaller indicating that many of the detected events at threshold 2.5 are very small and/or poorly visible out of the background tremor. To quantify the amplitude of seismic activity we first calculated the RMS amplitude using 10 min sliding windows and a 50% overlap (5 min shift). RMS amplitude provides an averaged measurement of the energy radiated by the volcano, including both transient events like EQs as well as permanent background tremor. The RMS has been

Fig. 3. (a) Number of events per window of 2 h with a STA/LTA higher than 2.5 (black and red) and 4.0 (blue and green). Measurements from station Y05 (black and blue) have been completed with those from station Y32 (red and green) to get continuous time series. (b) RMS amplitude calculated over 10-minute sliding windows. Individual measurements are shown in the background and have been averaged using a 20-point sliding window to obtain the black and red thick curves. Measurements from station Y05 (black and brown) have been completed with those from station Y31 (red and gray). (c) Peak to peak amplitudes in the 1–6 Hz frequency band for detected events with an STA/LTA higher than 2.5. Measurements from station Y05 (black) have been completed with those from station Y31 (red). The completions of event counting (Y32) and amplitudes (Y31) are done using the stations with the most similar temporal evolutions to Y05.

200

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

calculated for one of the vertical components of antenna Y05 and for the vertical component of broadband station Y31. Both of those stations are located near the summit of the volcano. RMS amplitudes have been calculated in the spectral domain in the frequency range between 0.5 and 25 Hz. Both stations show almost identical temporal evolutions and have same amplitudes after scaling the time series of Y31 by an empirical factor of 0.9. To gain complete time series for the entire experiment, we use in Fig. 3b station Y31, which was installed 110 days after Y05, to fill the gaps of Y05 and provide RMS amplitude after Y05 stopped working in mid October 2008. The obtained temporal series display a strong short term variability mostly related to the presence of transient events such as EQs. To enhance longer term variations a 20-point sliding averaging window has been applied to the raw RMS data. The averaged time series display, however, quite large variations, especially during the first months of the experiment. Sharp increases or decreases of the mean amplitude on the timescale of a few hours are observed outlining sharp variations in the intensity of the eruptive activity. This means that the volcano can quickly shift from a quiet activity toward a period with strong explosions possibly throwing bombs outside the crater. These variations are superposed on longer term changes. Toward the end of the experiment lower and smoother seismic amplitudes are observed. Finally to quantify the amplitude of individual transient events we calculated the peak to peak amplitude of the events detected in the STA/LTA detection process detailed above after filtering between 1 and 6 Hz. Fig. 3c displays the individual peak to peak amplitude of 530,000 events detected at Y05 and Y31 for an STA/LTA threshold of 2.5. In the same way as for the RMS amplitude, we use station Y31 to fill the gaps of station Y05 and extend the time series until the end of the experiment in February 2009. Also in the same manner, amplitudes for Y31 are multiplied by 0.9 to be consistent with those from Y05. Amplitudes are expressed in the same arbitrary units as the RMS. The histogram displaying the number of events as a function of the amplitude has a smooth mono-modal log-normal distribution peaked at 140 a.u. (same units as in Fig. 3b and c) indicating that most of the detected events have small amplitudes. This is in good agreement with the strong decrease in the number of detected events between a threshold of 2.5 and 4.0 in Fig. 3a. We observe strong and sharp fluctuations in the amplitude of the larger events. These are usually correlated with peaks in the RMS. This confirms that strong explosions may appear very quickly after periods of low activity. At stations near the volcano the seismic activity is dominated by signals related to the volcanic activity and therefore all of the 3 parameters examined here are representative of this activity. Large numbers of transient events are observed as well as large variations in the amplitude of the global energy (RMS) and amplitude of the strongest volcanic signals. The origin of the different short period transients and their possible link with the activity at the different craters needs to be established. 4. Video identification of events and similarity of EQs To correlate the seismic transients with surface activity, hours of video recordings have been made with a camera. The recordings have been synchronized with Universal Time by filming the time indicated on the screen of a portable GPS at the beginning and end of each video sequence. These sequences last between 10 and 60 min. This simple technique allows us to add a timer on each movie and use it to identify the transients directly related to explosions in the different craters (EQs) and also possibly other transients not directly related to any surface activity. Filming the GPS at the beginning and end of each recording allows us to check for possible time drifts caused by the use of magnetic tapes. The precision of the synchronization is probably below +/−0.2 s and no specific drift could be determined. During our experiment all 3 vents have been active. Sometimes independently but sometimes vents A and B or even A and C had simultaneous explosions suggesting

possible connections between the vents. The presence of such links at depth between the different vents is comparable to the common feeding conduits suggested by Landi et al. (2011) for the different vents at Stromboli. Explosions from vent C were often less common during the experiment. Fig. 4 shows an example of the precise correlation between a 19-second seismic waveform from station Y05 (component C03) and screen-shots extracted from a video. The screen-shots focus on the outlet of vent A and show a Strombolian explosion and its subsequent expanding plume. The explosion plume appears on the video recordings (Fig. 4c) about 0.3 s before the impulsive start of the EQ (Fig. 4a). According to Perrier et al. (2012) the P-wave velocity model below antenna Y05 reaches 2000 m/s 25 m below the station that is located about 600 m from crater A. Therefore, waveform propagation could explain the 0.3 s delay and this suggests that the EQ is directly generated by the decompression of the slug at the free surface of the conduit, probably very shallow (b 10 m) below the vent outlet. Considering a propagation velocity of sound of 330 m/s and including topography, this shallowness is also in good agreement with the acoustic phase that is observed about 2 s after the beginning of the EQ. This correlation favors the interpretation of the short-period component of the EQ as being a resonance or oscillation of the conduit and/or the edifice in response to the violent decompression of the gas slug at the free surface of the conduit. The EQ is in such a case comparable to an LP event (Chouet, 1996) in which the triggering mechanism required to initiate the oscillation (Chouet, 1988) is the surface slug decompression and the oscillating crack is replaced by the magma-filled conduit. We note that our timing between video and seismic differs from a similar correlation proposed for Stromboli by Ripepe et al (2001) as their seismic signal starts before the appearance of eruptive material at the vent bottom. At Sakurajima volcano, Ishihara (1985) also showed the explosion quakes are observed prior to the generation of shock waves and emissions of ejectas at the bottom of the crater. These differences may be explained by greater depths of the explosion quakes in those two cases as compared to the case of shallow explosion we present for Yasur in Fig. 4. Fig. 5 shows seismic signals at the same station Y05 and component C03 between 20:04 and 20:34 UT on April 19, 2008 with superposed the onset times of several explosions from vents A and C deduced from video observations. For this time frame, the camera was filming straight toward vent A. with vent B (inactive at that time) being hidden by the crater ridge in between the camera and vent A. Vent C was on the right of the field of view (Fig. 4) and its explosions are visible because of the shade created by the ejected plumes and projected blocks. The obtained correlation indicates that most of the observed seismic transients are directly related to the surface explosions. Also, data for 19/04/2008 (Fig. 5) outline that waveforms for explosions coming from a same vent (A or C) share a good waveform similarity (Fig. 6). A strong difference is observed between waveforms coming from vents A and C. This is confirmed by the classification of waveforms using cross-correlation. A high degree of waveform similarity is obtained for vent A, which generates simple oscillatory waveforms. On the other hand for vent C a poorer similarity is obtained with waveforms having a large acoustic high frequency phase with poorly developed long period oscillations. These results, as well as other correlations with video recordings, suggest explosion activity at a given vent, or at least part of it, may correspond to a given family of similar signals. Reversely each family of similar EQs may reflect activity at a given vent. These results are comparable to results from McGreger et al. (2004) who compared displacement waveforms for EQs at Stromboli volcano recorded during 63 h of enhanced activity. They identified clusters of events with high waveform similarity coming from one of the eruptive craters while other displayed lower similarity. They used these results for vent discrimination. Our comparisons also outline the presence of a reduced percentage of signals not being directly caused by surface phenomena or not accompanied by any surface process

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

c

b

g

e

f

201

a

d

Fig. 4. Timing of seismic recordings at Y05 (C03) with GPS-synchronized video recordings. (a) Seismic raw velocity recording with the high frequency acoustic air wave starting at second 21. (b) Only persistent degassing is observed at crater A. (c) First appearance of explosion plume in the vent. (d) to (e) Plume expanding in the crater at regular time intervals. (e) Larger ejected blocks falling on the crater floor. Pictures were taken from the Eastern part of the crater rim looking toward the West.

visible on the videos. This observation is comparable to the distinction made at Sakurajima volcano between EQs and B-type events (Minakami, 1974; Iguchi, 1994) which can be assimilated to LP events unrelated to surface activity. These results suggest the possibility of using waveform classification to identify families of characteristic repeating events and therefore simplify the analysis of the hundreds of thousands of transients recorded during our experiment.

5. Identification of characteristic repeating events In order to search on a larger time scale for the presence of similar events and determine the different characteristic repeating events we proceed to waveform classification at a reference station (Y05, vertical component C03). This station was chosen because of its temporal coverage including the beginning of our seismic deployment, proximity to the summit allowing good signal to noise ratios for smaller events and

20:04 20:06 20:08 20:10 20:12 20:14 20:16 20:18 20:20 20:22 20:24 20:26 20:28 20:30 20:32 0

30

60 Time (s)

90

120

Fig. 5. Helicorder showing the identification of the crater at the origin of the EQs by video observations and matched filtering. Data displayed are for station Y05 (C03) for the time frame between 20:04 and 20:34 on April 19, 2008. The onset time of each line is indicated on the left part of the plot. Red vertical bars indicate the onset time of explosions from crater A according to video observations and blue lines are for crater C. Superposed waveforms in red color are EQs from crater A having a cross-correlation higher than 0.70 with the reference template which is indicated with a red circle. Similarly EQs from crater C similar to the template that is in the blue circle are shown in blue color.

202

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

Vent A (P01)

0

2

4

6 8 10 Time (Seconds)

12

14

Vent C (P03)

16 0

2

4

6 8 10 Time (Seconds)

12

14

16

Fig. 6. Example of similar unfiltered waveforms for explosions from vent A and C recorded on April 19, 2008 (Fig. 5).

similar events is mostly due to the type of activity observed at the vents. Impulsive explosions tend to generate simple waveforms that are reproducible with a good level of waveform similarity. During periods dominated by such type of activity as shown in the example in Fig. 5, high percentages of grouped events are observed. In contrary periods dominated by jetting, complex multiple explosions or ash rich explosions generating low amplitude signals, percentages of similar events are low. We note that decreasing the similarity threshold to 0.70 increases significantly the percentage of similar events that can reach up to 78% (Fig. 8). Our daily classification outlines a significant number of DFs for the 247 days we examined. However, similar events may possibly occur over several days, weeks or months and several DFs issued from different days may correspond to similar waveforms. In other words several DFs may be part of a same family. Our goal with this classification procedure is mostly to identify families corresponding to significantly different waveforms over the whole experiment and for this purpose we applied an iterative procedure. (1) We first randomly select several DFs and calculate stacked waveforms by summing the waveforms from the 25 most similar events. These stacks have a 12-second duration and start roughly 1.5–2.0 s before the onset of the signal, which is often emergent. The obtained stacks are compared to each other in order to eliminate stacks with a similarity greater than 0.75. (2) We next identify in the continuous data of station Y05 (C03) the time windows similar to each reference stack. This is done by correlating the reference stacks with the continuous data filtered between 1 and 6 Hz. This scanning

simplicity of the observed waveforms as compared to other stations of the network (Fig. 7). Our analysis covers the period from January 24 to October 14, 2008 when the station stopped providing data. Because of the high number of transients we proceed with a waveform classification on a daily basis. We use the detections obtained for a threshold higher than 2.5 provided by the STA/LTA technique detailed above and extract for each detection a 20.48 second window starting 6 s before the triggering time. All events detected on a same day are compared to each other by calculating for each pair the normalized cross-correlation coefficient after filtering the waveforms between 1 and 6 Hz. The obtained correlation matrix is used to determine families grouping events similar above a given threshold. An important point is that our procedure provides open clusters what means that an event is part of a family as soon as its correlation coefficient is higher than the defined threshold with one of the over events, not all of them (closed cluster). This classification allows chain similarity as explained in the discussion hereafter. The number of grouped events and number of families depends on the similarity threshold. To avoid having too large families but rather smaller homogeneous ones including events with a high level of similarity, we use a threshold of 0.80, which is rather high considering the relatively great length of the compared signal windows (20.48 s). Fig. 8 shows the percentage of events grouped in daily-families (DFs) for threshold 0.80 and the corresponding number of DFs of more than five events. At this level of similarity there are between 1 and 8 DFs per day grouping between 2 and 61% of the events. This great variability in the daily number of

2008/06/01 23:29

0

2008/06/16 21:06

Y32

Y32

Y31

Y31

Y05

Y05

Y30

Y30

Y07

Y07

Y38

Y38

Y33

Y33

Y36

Y36

Y35

Y35

Y37

Y37

Y10

Y04

Y04

Y34

Y34

Y06

Y06

Y39

Y39

Y01

Y01

Y08

5

10 Time (s)

15

0

5

10 Time (s)

15

Fig. 7. Example of waveforms recorded at the different stations of the network for an EQ (left) and for a LP event (right). Waveforms were high pass filtered above 1.0 Hz. They are normalized and ordered from top to bottom according to increasing distance to the summit.

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

80

% evts.

0.70 60 40 20

% evts.

0 60

0.80

40

N_Fam

20 0 8 6 4 2 0

N. Families (0.80)

30

60

90

120

150

180

210

240

270

Day of Year (2008) Fig. 8. Daily percentage of events involved in families of similar events according to the single-station classification at station Y05, component C03. Percentages are presented for two different classification thresholds assuming events are similar for a cross-correlation higher than 0.70 (upper plot) and 0.80 (middle plot). Lower plot shows the daily number of families for threshold 0.80.

procedure or “matched filtering” provides a list of signal windows similar to each of the reference stacks with the corresponding correlation coefficients. We consider as reasonably similar all signal windows with a correlation higher than 0.7. (3) Finally the obtained lists are used to assign the different DFs to one of the reference stacks, i.e. to one of the already identified families. In practical terms a DF is assigned to one of the reference stacks when more than 80% of its members have been detected by matched filtering. The iterative procedure is then started again by searching for new reference stacks among the unassigned DFs. At this time we identified 42 different reference stacks (Fig. 9) with each of them (except for two pairs) having a correlation lower than 0.75 with all the other ones. These events are characteristic of the observed seismicity in the sense that many transients are similar to them. Our search is, however, not exhaustive and additional repeating events may exist. Most of the family group EQs but at least 2 of them group events related to a deeper process. We examine hereafter the temporal evolution of the identified families. 6. Temporal evolution of the families To recompose the temporal evolution of the identified families we use a matched filtering procedure as mentioned above. The continuous data of station Y05 (C03), filtered between 1 and 6 Hz, are scanned with the identified reference stacks using linear cross-correlation. The number of signal windows being identified as similar to the reference events depends on the considered correlation threshold. We consider here a threshold of 0.7. The temporal evolutions of the different families are represented in number of events per 2 h (Figs. 10 and 11). In order to identify families corresponding to EQs we use video recordings, we visualize data searching for the presence of acoustic high frequency phases and examine the amplitude ratios between a station located near the summit of the volcano (Y05 – C03 located 650 m from center of crater) and a station located at 1800 m (Y07 – C06). As explained above, video observations have been used to identify EQs accompanying the Strombolian explosions and some eventual events not accompanied by any surface activity. This way, we could determine that families B02 and B05 accompanies explosions from vents A or B. Indeed on March 5, 2008 movies show that only those 2 vents were

203

active. They were exploding individually but also often simultaneously with EQs from family B02 or B05 accompanying these explosions independently of the occurrence of explosions at a single or multiple vents. This suggests that these two vents may correspond to a common conduit. In a same way, video recording for April 19, 2008 show that family P01 accompanied explosions from vent A and family P03 accompanied explosions from vent C. Additional recordings from March 5 show that events from families L01 and L02 occur without any surface explosion. In order to distinguish such events from EQs, we name such events LP events and corresponding families are named Lxx where xx takes on numerical values solely for identification purposes. Similarly on April 19 we observe events from L04 occurring without any explosion. Since our video recordings only allow us to identify the origin of few families, we also visually examined the waveforms in search for high frequency acoustic phases that are characteristic of EQs (Fig. 12). This procedure allowed us to identify several families of EQs. We name Bxx the families corresponding to reference stacks from DFs before April 9 and Pxx those for DFs after this date, which corresponds to the occurrence of a major M = 7.3 subduction earthquake about 80 km from the volcano. We, however, note that acoustic phases are not always clearly visible and explosions in a same family may or may not display such high frequency phases as shown in the examples in Fig. 12. Finally we calculated for individual events the amplitude ratios between stations Y05 (C03) and Y07 (C06). At the time of writing this paper only preliminary locations were available for seismic sources (Perrier, 2011) not allowing a precise distinction between shallow and deeper events. Therefore, assuming sources located below the summit, we use the amplitude ratio between Y05 and Y07 as a proxy for the depth of the sources. We calculate amplitudes peak to peak after filtering the signals between 1 and 6 Hz. This calculation was repeated for 30 to 500 of the most well detected events of each family and average values were calculated. Despite some random variability as well as a few progressive temporal evolutions, average values are quite stable. Interestingly for the families identified by video, we get ratios of 2.30 and 2.22 for families of EQs B02 and P01 respectively and ratios of 0.61 and 0.72 for LP-event families L01 and L04. L02 has a mean value of 1.36 and B05 a value 1.91 suggesting the presence of deeper EQs and shallower LPs. More confusingly, the family P03 including EQs from vent C has a mean ratio of 0.73 that would suggest a rather deep source for these explosions. Despite this variability, the technique outlines clearly shallow sources, which may be considered as EQs. We note that this variability of source depths at Yasur is, however, comparable to the results obtained at Sakurajima where sources of EQs are found at depths comparable to those of BH-type and BL-type (LP) events (Ishihara and Iguchi, 1989; Iguchi, 1994). The combination of the 3 analysis techniques allows us to identify potential families of EQs and possible deeper LPs. In addition to the families identified on video, we consider as families of EQs those with an amplitude ratio higher than 2.0 and/or having clear high frequency acoustic phases. On the other hand, in addition to those identified by video, we consider as families of LPs those having an amplitude ratio lower than 1.0 and/or having no acoustic phases on the examined waveforms. This process identifies 29 families of EQs and 6 of LPs. Ambiguities remain for 7 families (B06, L03, L06, P13, P04, L05 and P11). Fig. 10 presents temporal evolutions for families of EQs. These families have complex time histories, spread in some cases over several months. For most of them, peaks of activity alternate with periods of quiescence. Video recordings suggest each family is associated with explosions at a given vent or at least given crater as distinction between vents A and B may be impossible. In this context the families of EQs may be seen as modes of oscillation of parts of the conduit in response to the violent slug decompression at the free surface. The fact that some peaks of activity are shared by several families, corresponding to significantly different master events, means that there are intermediate events, which can be identified by several different master events. This suggest that gradual waveform changes may link some of the master

204

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

Fig. 9. Left plot shows waveforms of the 42 reference events identified by our classification procedure. Right plot shows the corresponding dendrogram joining the different events. Numbers (distances) at the junction of the different branches are 100 × (1 − Xcorr) where Xcorr is the minimum correlation value between the events at the extremities of the branches.

events. For example families P02, P05, P06 and P21 identify common peaks of activity but also different ones. These may be seen as submodes of a same mode of oscillation. A striking feature is the major and durable change observed on April 9, 2008 at 12:46 UT (T0), the time of occurrence of a major M = 7.3 regional earthquake 80 km from the volcano as all families are mostly active before or after T0 but not during both periods. On several other occasions we observe abrupt transitions between families suggesting alternations between the modes of oscillation possibly between different vents or at a same vent. Fig. 11 shows temporal evolutions for several families of LP events. L01 corresponds to a master stack before April 9 and L04 after. Each family mostly detects events during the corresponding period but also a few events, with a poorer similarity, during the other period. Battaglia et al. (submitted) showed that these events are short term

precursors of EQs with a variable delay whose distribution is peaked near 12 s. Battaglia et al. (2012) noted that these 2 families are basically a single one with a distortion of the coda wave induced by the regional earthquake. L02 and L10 share also some similarity in the beginning of their waveforms and could also correspond to the same process. This could also be attested by their similar sawtooth pattern temporal evolution (Fig. 11). L05 comes interestingly in replacement for a few days when L10 dies out but its LP origin (Vs. EQ) is more questionable because of a higher amplitude ratio. L03 has an interesting smooth temporal evolution and is suddenly replaced by L06. For these two families, however, some ambiguities also remain about their LP nature. Two additional LP families detect events during the period before April 9 and include a few events (L07) or detect events at a rather steady rate of 2–4 events per 2 h (L08).

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

50

0 60

B03 2008.03.31 B07 2008.04.05

30 0 40

B04 2008.04.06

20 0 40

Number of Events / 2 Hours

Number of Events / 2 Hours

B01 2008.02.17 B08 2008.04.05

50

40

60

80

100

120

140

160

P01 2008.04.19 P20 2008.08.17

100 50 0 40

P04 2008.05.15 P19 2008.07.03

20 0 40

P08 2008.07.30 P10 2008.04.15

20 0 40

B10 2008.02.17 B09 2008.03.18 B11 2008.02.28 B13 2008.01.31

20 20

P02 2008.05.03 P05 2008.06.24 P21 2008.07.10

0

0 100

0

100

B05 2008.01.26 B02 2008.03.05

50

205

P03 2008.04.16 P14 2008.05.21 P09 2008.07.29

20 0

180

40

60

80

100

120

140

160

180

200

220

240

260

280

Day of the Year (2008)

Day of the Year (2008)

Fig. 10. Time distributions for various families of EQs for a detection level of 0.70. Families for master events before April 9 (left plot) and after (right plot). On each plot, the 4 upper subplots show the main families with in some cases variations in the time history depending on the master event. Lower subplots show short duration families or those grouping few events. Origin day of the master events is indicated by filled circles. Vertical red dashed line indicates the time of occurrence of the April 9 M = 7.3 regional earthquake.

family L01 into L04. They applied coda wave interferometry (Poupinet et al., 1984; Snieder et al., 2002) to these signals and showed that the distal earthquake induced a drop in the medium velocity of the volcano, followed by a slow partial recovery process. In the present work we illustrate that this disturbance affected practically all the families of similar events detected at Yasur volcano during this period including EQs as well as LPs. In order to examine the effect of the distal earthquake on EQs, we used the results from the daily classification of triggered events detailed above. The classification of all events detected on 9/4/2008 at station Y05 shows that for a threshold sufficiently high EQs separate into two distinct groups of events recorded before and after T0. Lowering the similarity threshold leads all the events to merge into a single family. This suggests that the distal earthquake induced a limited waveform change into a group of events coming from the same vent. Therefore we selected 107 EQs recorded during

The presence of characteristic events with their respective temporal evolutions indicates the presence of specific points of the conduit oscillating with specific modes, which may last for months and may be influenced by external factors such as the occurrence of a major regional earthquake. 7. Discussion 7.1. Changes on EQs induced by a M = 7.3 regional earthquake Battaglia et al. (2012) showed that the occurrence of a M = 7.3 subduction earthquake 80 km from Yasur volcano had a major effect on the occurrence of LP families L01 and L04. They showed that despite the apparent lack of impact on the eruptive dynamics, the distal earthquake induced a distortion in the coda wave of these events, transitioning

30 L01 2008.01.27 L04 2008.08.07

20

Number of Events / 2 Hours

10

0 L02 2008.03.05 L10 2008.04.27 L05 2008.05.30

20

10

0 L03 2008.03.22 L06 2008.03.24

30 20 10 0

30

60

90

120

150

180

210

240

270

Day of the Year (2008)

Fig. 11. Time distributions for various families of LP events for a detection level of 0.70. L01, L02 and L04 are deeper events unrelated directly to any surface phenomena as attested by video recordings. L01 and L04 have similar waveforms with only distortions in the coda induced by the April 9 M = 7.3 regional earthquake (vertical red dashed line). L03, L06, L05 and L10 lack the high frequency acoustic phases. Origin day of the master events is indicated by filled circles. Vertical red dashed line indicates the time of occurrence of the April 9 M = 7.3 regional earthquake.

206

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

B01

0

2

4

6 Time (s)

8

10

L04

0

2

4

6 Time (s)

8

B02

0

2

4

6 Time (s)

8

10

10

L02

0

2

4

6 Time (s)

8

10

Fig. 12. Unfiltered waveforms at station Y05 for two families of similar explosion quakes (B01 and B02) and for two families of LP events (L02 and L04).

the hours around T0, including 44 before T0 and 63 after. We compared waveforms from before T0 to those from after using a cross-correlation moving window technique (CMW) (Poupinet et al., 1984). We used a 512-point sliding window. The technique provides for each calculation window a time shift and the corresponding cross-correlation value. Fig. 13 shows a comparison between average correlation values calculated between a reference event before T0 and the 106 other selected EQs for signal windows in the early and late parts of the signals. It shows a drop of the waveform similarity in the later part of the signals while the earlier few seconds keep a good similarity. This drop occurs exactly at T0. Looking at time delays as a function of time (Fig. 14), we observe a progressive dephasing along the waveforms with an almost linear trend of about 2.4% for the presented example. However, we note that the correlation values for the sliding windows are very low when moving toward the coda of the EQs providing low quality linear fits. Obtaining stable results is extremely difficult for most of event pairs. Fig. 14 shows slope values obtained for a limited number of selected pairs from 4 different vertical components of station Y05. They correspond to reasonably linear trends. Quite variable values are observed with absolutes values mostly spread between 1.5 and 3.5%, with a mean value of 2.3% +/− 0.83. This value is lower than the value of 3.57% +/− 0.31 obtained at Y05 for LP events by Battaglia et al. (2012). For stations farther away from the summit, waveforms before and after T0 are too different to properly apply the CMW technique. Our results show that the separation between families active before and after T0 can be explained by waveform distortions caused by a drop

in the medium velocity of the volcano. This drop is comparable to the sharp decreases induced at Colima volcano by regional earthquakes (Lesage et al., 2014) or to the drops imaged under volcanic regions of Japan by Brenguier at al. (2014) following the Tohoku-Oki earthquake in 2011. However, while all pre-T0 families detect no events after T0, two post-T0 families (P05 and P21) detect some events during the 12 h (P05) or 2.5 days (P21) prior to T0. These two families are submodes of a same mode of oscillation as shown in Fig. 10, but have rather complementary temporal evolutions around T0 as shown in Fig. 15. These evolutions suggest that families P21 and P05 are a continuation of B04 after T0 with a waveform change induced by the velocity change in the volcanic shallow structure. Battaglia et al. (2012) showed that the abrupt velocity drop observed at T0 was followed by a slow partial recovery process. Hence because P21 and P05 correspond to master stacks from more than 2.5 months after T0, these may have lost part of the waveform distortion and share a slightly higher similarity to B04. However, the fact these master stacks identify almost no events earlier that 2.5 days before T0 also suggests a waveform evolution of EQs precursory to the distal earthquake. This is in agreement with a velocity change imaged at stations Y03 and Y01 by Battaglia et al. (2012) during the days prior to T0. 7.2. Transitions between families Additionally to the major change in the temporal evolution of families observed on 9/4/2008, we note on numerous occasions rapid or

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

207

Y033

0

5

10

15 30

dt (0.01*s)

Time (s) 1

Correl. Fact.

Correlation coeff.

0.8 0.6 0.4

20

+2.7 %

10 0 -10 -20 0.8 0.4 0.0 0

0.2

5

10

Early Window Late window

15

20

Time (s)

4 0

10

12

14

C00 C01 C02 C03

16

Hours (9/4/2008)

/ %)

2 Fig. 13. Upper plot: stacked EQ waveform showing the position of the early and late windows used for calculating the mean correlation values. Lower plot: Mean correlation coefficients for comparisons between an EQ occurring before T0 and other similar explosion quakes. The time of the reference EQ is indicated by red dashed line and T0 by vertical black line. The comparison is done using two different signal windows: at the beginning of the events (black) and in the coda (red). The plot shows a drop of the similarity in the coda caused by the regional earthquake.

0

-2

-4 10

slow transitions between families. These transitions may be representative of changes in the eruptive dynamics. We illustrate here several of these transitions and examine the corresponding evolutions of amplitude ratios between stations for individual events. These may be indicators of changes in the depth (Y05/Y07) or vent origin of the explosions (Y32/Y31 after mid-May 2008). The period around T0 detailed in Fig. 15 involves numerous families and is rich in transitions beside those occurring at T0, which have been discussed in the previous section. Before T0 we note the partial and short term replacement on day 96 of B07 by B08 and later its replacement by B04. Y05/Y07 amplitude ratios suggest a progressive deepening of the source of B07 before the short activation of B08 with a rather good continuity in the amplitude ratios. On the other hand the transition between B07 and B04 is accompanied by an increase of the amplitude ratios suggesting a rapid change in the source depth. After T0 smooth changes in amplitude ratios are observed for P05/P06, which are common sub-modes. P10, which replaces at some point P05/P06 comes in the continuity of these families with slightly lower amplitude ratios. Toward the end of the sequence we observed rapid transitions between P10, P03, P10 and P01, which are each time accompanied by rapid changes in the corresponding amplitude ratios. According to video recordings, P03 corresponds to explosions from vent C and P01 from vent A. This sequence illustrates that transitions between families may have various origins: drops or rises in the depth of sources, smooth transition without any clear change in source location, shift from an activity at vent A toward activity at vent C or changes in the medium (9/4/2008). Fig. 15 additionally shows the significantly different waveforms belonging to the different families recorded during the 20-day period. Fig. 16 shows two additional examples of transitions in February and July–August 2008. In February we observe an alternation between families B01 and B05 with first a sudden transition from B05 to B01 corresponding to a rather steady Y05/Y07 ratio. A few days later we

12

14 Hours 9/ /

)

Fig. 14. Upper plot: Correlation Moving Window technique applied to two EQs recorded at station Y05 (vertical component C03) before and after T0. The upper part shows the waveforms aligned on their onsets, middle part shows delays along the waveform calculated using 512-point windows, with linear interpolation and evaluation of the slope dt/t, lower part shows corresponding correlation value. Lower plot: Slopes for selected pairs of EQs. Events from before T0 are compared to those after with corresponding slope values plotted at the time of occurrence of the “before event” independently of the event used for comparison. Similarly “after events” have been compared to “before events”. Various components of station Y05 are used.

observe a progressive transition from B01 to B05. The progressive disappearance of B01 is accompanied by an increase of the Y05/Y07 ratio while progressively appearing B05 events have a lower ratio. During the July–August sequence 4 families are identified. For this sequence we also examine the amplitude ratio between stations Y32 and Y31, located respectively to the north-east and south-west of the summit, as it provides rough insights about the origin crater of the explosions. Higher ratio would suggest explosions closer to Y32, at vent C while lower ratio could correspond to explosions from vents A or B closer to Y31. Interestingly we note that P08 tends to have deeper sources similarly to P03 identified as corresponding to crater C, it also tends to have higher Y32/Y31 ratio despite having quite a lot of dispersion possibly in relation with the greater depth of the sources. P02 and P07 are common sub-modes and have the same behavior for amplitude ratios. P09 appears to be spread over variable depths but would rather be issued from vents A/B. The anti-correlation observed between P08 and P09 could therefore be interpreted as a shift of the eruptive activity from vents A/B to vent C, inversely to the transition from P08 to P02/P07. While the interpretation of transitions between families remains in some cases uncertain and keeping in mind that our families do not represent all of the EQs activity, our results outline several types of behaviors. In some cases the observed alternations outline transitions

208

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

80

P03

c

B04 B07 B08

a

P10

60

B08 Number of Evts / 2 Hours

40

B07 20

P01 0

P05

80

P06 P01 P10 P03 P21 P05

60 40

P06 B04 P21

20

Amp. Ratio Y05/Y07

0

90

95

100 Day of the Year (2008)

105

110

95

100 Day of the Year (2008)

105

110

0

5 ime (s)

10

4

2

b 0

90

Fig. 15. (a) Focus on the temporal distributions of EQs families around the time of occurrence of the M = 7.3 subduction earthquake (vertical dashed line). Upper part shows families for master stacks before 9/4/2008 and lower for stacks after. (b) Amplitude ratios between peak-to-peak amplitudes at stations Y05 and Y07 for families shown on plot (a). Color codes are same as on plot (a). (c) Waveforms for events detected between days 90 and 110 belonging to the different families.

in the activity of vents A/B to C or vice versa. This clearly suggests that these craters are somehow inter-connected. In some other cases, transitions are rather corresponding to a sudden drop or rise of the source depths. In the context of EQs being interpreted as oscillation of the conduit, the source location may correspond to the centroid of the oscillation (Battaglia et al., 2003). Therefore sudden changes in the source depth could correspond to changes in the shape and depth of the oscillator caused by sudden drops or rises of the free surface in the conduit for example. More progressive changes in this depth could lead to progressive transitions from one mode of oscillation to another, i.e. one family to another. Transitions not accompanied by any source depth change could be related to changes in the properties of the magma such as viscosity or bubble concentration and distribution (Garcès et al., 1998). 7.3. Continuum of waveforms and chain similarity The ability of an event to detect numerous similar windows by matched filtering depends on its waveform. Therefore we use stacks of events calculated based on the DFs to have simplified averaged waveforms. Despite starting from 42 significantly different events we note that some families obviously share common peaks of activity (Fig. 10), we name them sub-modes of a same mode of oscillation. On the other hand some families like P01 and P02 appear complementary as they replace each other systematically in their temporal evolutions. However, according to amplitude ratios between stations Y05 and Y07

and Y31 and Y32, they share a spatial continuity. Fig. 17 illustrates possible connections by chain similarity that can link the P01 and P02 reference stacks. These reference events were obtained by stacking similar events for DFs respectively on 19/4/2008 and 3/5/2008. They have a very low correlation value of 0.35. Using matched filtering both of these reference stacks identify events on 2/8/2008. Stacking separately the events identified by the 2 initial reference events we generate two new stacks for P01 and P02 for day 2/8/2008. These new stacks still have high correlation values of 0.87 and 0.85 with their respective “parents”. The correlation value between them is now at 0.70. Searching among the individual EQs recorded on 2/8/2008 for the most similar event to both new stacks we identify an intermediate EQ that has correlation values of 0.87 and 0.86 with them (shown in blue in Fig. 17). This process illustrates that it is possible to identify highly similar intermediate events, which link step-by-step the two initial reference events. This result suggests that a large part of the EQs waveforms, especially for EQs from a same vent, may be part of a continuum of waveforms in which our reference events are only endmembers. Transitions between these end-members may be progressive or sudden as imaged in the previous section. 8. Conclusions We examined the seismicity recorded by a temporary network that was installed around Yasur volcano between January 2008 and February 2009. This network recorded daily hundreds of transients with significant

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

Nb of Evts / 2 Hours

120

209

B01 B05

80

40

Amp. Ratio Y05/Y07

0 4 3 2 1 0

47

45

49

51

53

Amp. Ratio Y32/Y31 Amp. Ratio Y05/Y07

Nb of Evts / 2 Hours

Day of the Year (2008)

40

02 07 08 09

20

0 5 4 3 2 1 0 2

1

0 207

209

211

213

215

217

219

Day of the Year (2008) Fig. 16. Temporal distributions of families of EQs and corresponding amplitude ratios between peak-to-peak amplitudes at stations Y05 and Y07. Upper plot shows a period of 9 days in February 2008 and lower plot a period of 13 days in July–August 2008. On each plot color codes are the same for temporal evolutions of families and ratios. Lower plot also includes amplitude ratios between peak-to-peak amplitudes at stations Y32 and Y31.

variations in the daily number of events and radiated energy. Since the exhaustive analysis of all transients is impossible, we searched for characteristic repeating events. We identified 42 families corresponding to significantly different reference events and recomposed their temporal evolutions using matched filtering. This approach simplifies the analysis of an intense seismicity as it helps focusing on repetitive events which have a specific signification for the volcanic system. A few families group signals related to deeper processes, i.e. not directly linked with surface activity. These include two families of LP events corresponding to short time precursors of the Strombolian

explosions (Battaglia et al,, submitted). They may be interpreted as resonances of deeper segments of the volcanic conduit possibly triggered by irregularities in the fluid flow at these points. However, most of the identified families group EQs. Video observations show that the EQ families are assigned to given vents or at least craters since activities at vents A and B may be sometimes impossible to distinguish. The families of EQs may be seen as modes of oscillation of the conduit with the rapid slug decompression at the free surface being the triggering mechanism of the oscillations. The temporal evolution of the EQs families indicates that some families may last for months with often peaks of activity

210

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211

P02 (3/5/08)

0.35

P01 (19/4/08)

link between the activity at those vents. Other transitions between families occur for activity at a same crater. In some cases these transitions are sudden and accompanied by a possible change in the depth of sources. Such transitions could correspond to rapid drops or rises of the free surface in the conduit, which would lead to changes in the depth of the centroid of the oscillations, i.e. in the source depth. Other transitions not accompanied by any change in source locations could be related to changes in the properties of the magma such as viscosity or bubble concentration and distribution. In general, families illustrate changes in the eruptive behavior whose interpretation needs to be further investigated by using multi-parametric studies and high precision source locations. Our results show that the Strombolian seismicity of Yasur is characterized by the activation of specific modes of oscillations of the conduit. Acknowledgments

0

5

10

Time (s)

P02 (3/5/08) 0.85 P02 (2/8/08) 0.86

All the data used in this study were collected during a temporary experiment carried with seismic stations from the French seismic networks IHR and RISC (Isterre, Grenoble, France). We are grateful to David Nakedau, Svetlana Byrdina and Richard Yatika who helped significantly in field work. We thank Mikhail Zhizhin and Jean-Luc Got for providing codes, which are at the origin of some of those used for this study. We are grateful to two anonymous reviewers, to the editor Lionel Wilson and to the associate editors Nicole Métrich and Sylvie Vergniolle for constructive comments. This work has been supported by the ANR (France) contract ANR-06-CATT-02 Arc-Vanuatu.

0.70 References

0.87 P01 (2/8/08) 0.87

0

P01 (19/4/08)

5

10

Time (s) Fig. 17. Waveforms illustrating the continuum of waveforms between families. Upper plot shows reference stacks for families P01 and P02 calculated respectively for DFs from 3/5/2008 and 19/4/2008. Normalized cross-correlation between the 2 events is 0.35. Lower plot shows additionally two stacks calculated for day 2/8/2008 using events identified by matched filtering with these 2 reference events and an intermediate waveform corresponding to a single EQ observed on 2/8/2008. Corresponding cross-correlation values between pairs are indicated.

separated by periods of quiescence. This means that the conduit and volcanic structure, as well as source mechanism, may retain and regain quasi-identical properties over long durations. Temporal evolutions of the modes of oscillation display complex patterns. Transitions between families may occur in progressive or sudden modes, providing possible information about changes in the eruptive dynamics or volcanic structure. A major change in the families was induced by a M = 7.3 subduction earthquake, which occurred about 80 km from the volcano as it interrupted all the existing modes of oscillation, shallow (EQs) as well as deeper ones (LPs). In this case the transition can be explained using coda wave interferometry by a change in the medium velocity induced by the distal event. Surprisingly families also illustrate a possible precursory change starting 2.5 days prior to the distal event. On numerous other occasions, transitions between families of EQs are observed. In some cases the transitions outline changes in the activity between vents A/B and C indicating the presence of a strong

Allen, S.R., 2005. Complex spatter- and pumice-rich pyroclastic deposits from an andesitic caldera-forming eruption: the Siwi pyroclastic sequence, Tanna, Vanuatu. Bull. Volcanol. 67, 27–41. Battaglia, J., Got, J.-L., Okubo, P., 2003. Location of long-period events below Kilauea Volcano using seismic amplitudes and accurate relative relocation. J. Geophys. Res. Solid Earth 108 (B12) (1978–2012). Battaglia, J., Métaxian, J.-P., Garaebiti, E., 2012. Earthquake–volcano interaction imaged by coda wave interferometry. Geophys. Res. Lett. 39 (11). Battaglia, J., Métaxian, J.-P., Garaebiti, E., 2015. Short term precursors of Strombolian explosions at Yasur volcano (Vanuatu). Geophys. Res. Lett. (submitted). Braun, T., Ripepe, M., 1993. Interaction of seismic and air waves recorded at Stromboli Volcano. Geophys. Res. Lett. 20 (1), 65–68. Brenguier, F., Campillo, M., Takeda, T., Aoki, Y., Shapiro, N.M., Briand, X., Emoto, K., Miyake, H., 2014. Mapping pressurized volcanic fluids from induced crustal seismic velocity drops. Science 345 (6192), 80–82. Carney, J., Macfarlane, A., 1979. Geology of Tanna, Aneityum, Futuna and Aniwa. New Hebrides Gov. Geol. Surv. Rep. 5–29. Chen, J.K., Taylor, F.W., Edwards, R.L., Cheng, H., Burr, G.S., 1995. Recent emerged reef terraces of the Yenkahe resurgent block, Tanna, Vanuatu—implications for volcanic landslide and tsunami hazards. J. Geol. 103, 577–590. Chouet, B., 1985. Excitation of a buried magmatic pipe: a seismic source model for volcanic tremor. J. Geophys. Res. Solid Earth 90 (B2), 1881–1893 (1978–2012). Chouet, B., 1988. Resonance of a fluid-driven crack: radiation properties and implications for the source of long-period events and harmonic tremor. J. Geophys. Res. Solid Earth 93 (B5), 4375–4400 (1978–2012). Chouet, B., 1996. Long-period volcano seismicity: Its source and use in eruption forecasting. Nature 380, 309–316. Chouet, B.A., Matoza, R.S., 2013. A multi-decadal view of seismic methods for detecting precursors of magma movement and eruption. J. Volcanol. Geotherm. Res. 252, 108–175. Chouet, B., Dawson, P., Ohminato, T., Martini, M., Saccorotti, G., Giudicepietro, F., De Luca, G., Milana, G., Scarpa, R., 2003. Source mechanisms of explosions at Stromboli Volcano, Italy, determined from moment-tensor inversions of very-long-period data (1978–2012) J. Geophys. Res. Solid Earth 108 (B1) (ESE-7). Chouet, B., Dawson, P., Martini, M., 2008. Shallow-conduit dynamics at Stromboli Volcano, Italy, imaged from waveform inversions. Geol. Soc. Lond., Spec. Publ. 307 (1), 57–84. Coulon, C., Maury, R.C., 1981. Petrology of tholeiitic lavas from Tanna Island (New Hebrides): importance of cumulative processes in island arc magmatism. Bull. Volcanol. 44 (4), 661–680. Crosson, R.S., Bame, D.A., 1985. A spherical source model for low frequency volcanic earthquakes. J. Geophys. Res. Solid Earth 90 (B12), 10237–10247 (1978–2012). Garcés, M.A., Hagerty, M.T., Schwartz, S.Y., 1998. Magma acoustics and time‐varying melt properties at Arenal Volcano, Costa Rica. Geophys. Res. Lett. 25 (13), 2293–2296. Gordeev, E., 1999. Gas bubble dynamics model for shallow volcanic tremor at Stromboli. J. Geophys. Res. 104 (B5), 10–639. Iguchi, M., 1994. A Vertical Expansion Source Model for the Mechanisms of Earthquakes Originated in the Magma Conduit of an Andesitic Volcano: Sakurajima, Japan.

J. Battaglia et al. / Journal of Volcanology and Geothermal Research 322 (2016) 196–211 Ishihara, K., 1985. Dynamical analysis of volcanic explosion. J. Geodyn. 3 (3), 327–349. Ishihara, K., Iguchi, M., 1989. The Relationship Between Micro-Earthquake Swarms and Volcanic Activity at Sakurajima Volcano. Disaster Prevention Research Institute Annuals. James, M.R., Lane, S.J., Chouet, B.A., 2006. Gas slug ascent through changes in conduit diameter: laboratory insights into a volcano-seismic source process in low-viscosity magmas. J. Geophys. Res. Solid Earth 111 (B5) (1978–2012). Kremers, S., Wassermann, J., Meier, K., Pelties, C., van Driel, M., Vasseur, J., Hort, M., 2013. Inverting the source mechanism of Strombolian explosions at Mt. Yasur, Vanuatu, using a multi-parameter dataset. J. Volcanol. Geotherm. Res. 262, 104–122. Landi, P., Marchetti, E., La Felice, S., Ripepe, M., Rosi, M., 2011. Integrated petrochemical and geophysical data reveals thermal distribution of the feeding conduits at Stromboli volcano, Italy. Geophys. Res. Lett. 38 (8). Lesage, P., Reyes‐Dávila, G., Arámbula‐Mendoza, R., 2014. Large tectonic earthquakes induce sharp temporary decreases in seismic velocity in Volcán de Colima, Mexico. J. Geophys. Res. Solid Earth 119 (5), 4360–4376. Lokmer, I., Bean, C.J., Saccorotti, G., Patanè, D., 2007. Moment‐tensor inversion of LP events recorded on Etna in 2004 using constraints obtained from wave simulation tests. Geophys. Res. Lett. 34 (22). Matoza, R.S., Shearer, P.M., Okubo, P.G., 2014. High‐precision relocation of long‐period events beneath the summit region of Kılauea Volcano, Hawai 'i, from 1986 to 2009. Geophys. Res. Lett. 41 (10), 3413–3421. McGreger, A.D., Lees, J.M., 2004. Vent discrimination at Stromboli volcano, Italy. J. Volcanol. Geotherm. Res. 137 (1), 169–185. Minakami, T., 1974. Seismology of volcanoes in Japan. Phys. Volcanol. 6, 1–27. Nabyl, A., Dorel, J., Lardy, M., 1997. A comparative study of low‐frequency seismic signals recorded at Stromboli volcano, Italy, and at Yasur volcano, Vanuatu. N. Z. J. Geol. Geophys. 40 (4), 549–558. Neuberg, J., Luckett, R., Ripepe, M., Braun, T., 1994. Highlights from a seismic broadband array on Stromboli volcano. Geophys. Res. Lett. 21 (9), 749–752. Perrier, L., 2011. Apport de l'étude des Sources Sismo-Volcaniques à la Connaissance des Processus Éruptifs du Volcan Yasur, Vanuatu. Université de Grenoble, France (Phd thesis).

211

Perrier, L., Métaxian, J.-P., Battaglia, J., Garaebiti, E., 2012. Estimation of the near-surface velocity structure of the Yasur–Yenkahe volcanic complex, Vanuatu. J. Volcanol. Geotherm. Res. 227, 50–60. Petersen, T., 2007. Swarms of repeating long-period earthquakes at Shishaldin Volcano, Alaska, 2001–2004. J. Volcanol. Geotherm. Res. 166 (3), 177–192. Poupinet, G., Ellsworth, W.L., Fréchet, F., 1984. Monitoring velocity variations in the crust using earthquake doublets: an application to the Calaveras Fault, California. J. Geophys. Res. 89, 5719–5731. Ripepe, M., 1996. Evidence for gas influence on volcanic seismic signals recorded at Stromboli. J. Volcanol. Geotherm. Res. 70 (3), 221–233. Ripepe, M., Braun, T., 1994. Air-wave phases in Strombolian explosion-quakes seismograms: a possible indicator for the magma level. Acta Vulcanol. 5, 201–206. Ripepe, M., Ciliberto, S., Della Schiava, M., 2001. Time constraints for modeling source dynamics of volcanic explosions at Stromboli. J. Geophys. Res. Solid Earth 106 (B5), 8713–8727 (1978–2012). Robin, C., Eissen, J.-P., Monzier, M., 1994. Ignimbrites of basaltic andesite and andesite compositions from Tanna, New Hebrides Arc. Bull. Volcanol. 56 (1), 10–22. Rowe, C.A., Aster, R.C., Kyle, P.R., Schlue, J.W., Dibble, R.R., 1998. Broadband recording of Strombolian explosions and associated very‐long‐period seismic signals on Mount Erebus Volcano, Ross Island, Antarctica. Geophys. Res. Lett. 25 (13), 2297–2300. Simkin, T., Siebert, L., McClelland, L., Bridge, D., Newhall, C., Latter, J.H., 1981. Volcanoes of the World: A Regional Directory, Gazetteer, and Chronology of Volcanism During the Last 10,000 Years. US Hutchinson Ross Publishing. Snieder, R., Grêt, A., Douma, H., Scales, J., 2002. Coda wave interferometry for estimating nonlinear behavior in seismic velocity. Science 295, 2253–2255. Wassermann, J., 1997. Locating the sources of volcanic explosions and volcanic tremor at Stromboli volcano (Italy) using beam-forming on diffraction hyperboloids. Phys. Earth Planet. Inter. 104 (1), 271–281. White, R.A., 1996. Precursory deep long-period earthquakes at Mount Pinatubo: spatiotemporal link to a basalt trigger. In: Newhall, C.G., Punongbayan, R.S. (Eds.), Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. Univ. of Wash. Press, Seattle, pp. 307–327.