Strain Signals Before and During Paroxysmal Activity at Stromboli Volcano, Italy

In the last decades, Mt. Stromboli produced four vulcanian eruptions, in 2003 and 2007 and July and August 2019, recorded by INGV monitoring network. Specifically, last three events are studied through records from borehole strainmeters, which allow to infer details on source dynamics. These events are preceded by a slow strain buildup, starting several minutes before the paroxysms, which can be used in future for early warning. Eruptions consist of two or more strain pulses, with oscillations ranging from several seconds, as in 2007, to some minutes, as in 2019, and lasting from several minutes to 1 hr after the explosions.


Foreword
Stromboli can be considered one of the most active volcanoes in the world. Its persistent but moderate explosive activity, termed "Strombolian," is interrupted by occasional episodes of more vigorous activity accompanied by lava flows, as seen in 1975, 1985, 2002-2003, 2007, 2014, and more recently in July and August 2019. Persistent eruptive activity and ease of access make this volcano an ideal laboratory for detailed study of source processes associated with magmatic activity.
The dynamics of the sources associated with the explosions has been investigated by Chouet et al. (1997Chouet et al. ( , 1999Chouet et al. ( , 2003 by using seismic arrays and a dense network of broadband seismometers. These authors investigated the very-long period (VLPs) source mechanisms, revealing a volumetric component attributed to mass transport through the magma conduits. Long period components of volcanic tremor (10-50 s) have been also documented (De Lauro et al., 2005;. The volcanic tremor of Stromboli is a continuous permanent signal with a spectrum covering traditionally a 1-3 Hz band. Just after the 2002 extensive phase of activity, the latter studies indicated the need to complement the monitoring system with some borehole strainmeters, instruments capable of recording the volcano behavior at low frequencies. Two such instruments were installed during 2006, with the support of Italian Civil Protection Department, by INGV, Università degli Studi di Salerno (Italy) and Carnegie Institution of Washington DC (USA): the TDF instrument, located near the Ginostra village in the western side of the island, lying at a radial distance of 1.5 km from the main eruptive vents, and the SVO instrument, situated in the village of Stromboli, about 2.5 km northeast from the summit craters ( Figure 1).
The Sacks-Evertson strainmeters are long stainless steel cylinders of about 7 cm in diameter and 4 m in length filled with degassed silicone oil, which provide two signal outputs, obtained by two different hydro-mechanical amplification systems: A bigger sensing volume is connected with a small bellow, whose length changes in proportion to the volume of oil entering or leaving it. The position of the top of the bellow is measured through a linear variable differential transformer (LVDT). The dynamic range of the instrument is about 140 dB. A second, less sensitive, bellows-displacement transducer-valve assembly is connected with the ©2020. The Authors. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. • Supporting Information S1

10.1029/2020GL088521
• Data Set S1 • Data Set S2 • Figure S1 • Figure S2 • Figure S3 • Figure S4 Correspondence to: P. Romano, pierdomenico.romano@ingv.it first one. The high-sensitivity output integrates the volumetric change in the strained reservoir. The low-sensitivity one is connected to the strained reservoir only when the instrument is sensing a rapid and strong strain change, thus measuring strain. Usually, the low-sensitivity channel measures the pressure in a closed cell, that is proportional to local temperature. The temperature resolution is~10 microdegrees. Air pressure is also measured. Nominal resolution of the Sacks-Evertson strainmeter is about 10 −11 , nominal dynamic range is 10 −11 -10 −3 . Low-frequency calibration of installed strainmeters is obtained by comparison with Earth tides (Hart et al., 1996). Seismic signal, high-and low-sensitivity strain signals are continuously recorded and sampled at 50 sps. Sacks-Evertson strainmeters are installed in boreholes by using expansive cement to couple them to the surrounding rocks. Once installed, the strainmeters are not adjustable nor recoverable.
The two instruments installed in Stromboli are each cemented at the bottom of a borehole at 120 m depth. Data recorded at both sites are sent to Vesuvius Observatory and are retrieved and processed at Vesuvius Observatory and University of Salerno. After the installation, comparison of data from both strainmeters for low-and high-frequency signals (earth tides and summit explosions, respectively) revealed that the TDF instrument is not well coupled with the surrounding rocks, while the second (SVO) shows a sensitivity 1 × 10 −11 per digital count.
In order to obtain the change in rock strain, data are processed by removing atmospheric pressure, Earth tides, and ocean loading by using a Bayesian approach (Hart et al., 1996).
Here we show a correlation among the previous paroxysmal eruption that occurred in 2007 and those during summer 2019: The strainmeter data show time-histories that are similar for all recorded paroxysms. This observation suggests a common source mechanism.

The 2007 Paroxysm
Initial technical and logistic issues delayed the start of reliable continuous data recording until 20 January 2007. After a few weeks, an extensive phase of activity of the volcano, with a lava flow outpouring from This activity produced an ash column reaching a height of 3-3.5 km above the craters. The paroxysmal activity lasted about 7 min. This extensive phase of activity ceased with the start of the flank effusive activity, which started on 2 April and continued through 7 April 2007. From the middle of February until around middle of April, data recorded from the SVO strainmeter show the highest strain rate during the year that high strain rate corresponded to the increase of the Strombolian activity. The extensive activity was not preceded by an increase of seismicity and only the largest explosion on 15 March shows clearly, in the short period before the occurrence of this event, a change in strain rate, while the data recorded during the effusive phase on 27 February show a decrease in the strain rate~2 hr after the lava flow.
Very interesting signals have been recorded during a more intense explosion, on 15 March 2007, which extruded massive lava blocks to >1 km distant from the active craters (Pistolesi et al., 2011). The data recorded at both strainmeters (see Figure S0 in the supporting information) indicate that a. before the explosion 1. for~660 s before, there is a slow source pressure buildup (green background in Figure S0); 2. for 45 s, source pressure does not increase or decrease but exhibits an 8 s oscillation: this Observation suggests that the conduit is still sealed (yellow background in Figure S0); b. after the explosion: 1. there is a 20 s period with decaying oscillations, mainly due to gas emissions (orange background in Figure S0); 2. pressure and strain observations indicate an oscillation in the air lasting 1,100 s at SVO and 800 s at TDF: This effect is driven by the collapse of the eruptive column, creating a gravitational load on the volcanic edifice.

The 2019 Paroxysms 2.2.1. The 3 July Event
During the weeks and months preceding the paroxysmal eruption on 3 July 2019, no significant changes in the strain and other monitored volcanic parameters were observed. During the period from December 2018 to January 2019, the volcano had a phase of higher activity, followed by an activity level classified as "low to moderate." The activity increased from June 2019, keeping a "moderate" level from 12 June until the day before the eruption. This kind of behavior in volcano dynamics has been very common in recent years.
Almost 1 hr before the paroxysm, at 3:59 p.m. on 3 July, a new vent appeared on the upper break in slope of the Sciara del Fuoco (SdF), a few 100 m northwest from the crater terrace: Hot rocks broke off and, shortly after, a small lava flow started and slowly traveled downslope (retrieved from http://www.ingv.it/it/stampae-urp/stampa/comunicati-stampa-1/2019/37-comunicato-straordinario-stromboli-04-07-2019-09-00-utcaggiornamento-sul-fenomeno-in-corso/file). The appearance of this new vent was the only precursor so far known: The magma in the conduit was experiencing a pressure rise from underneath and opened a small side-wise vent. Immediately before the onset of the paroxysm, lava flows emerged from all vents. The speed of the magma being pushed out from the conduit dramatically increased along with seismic activity which, then, quickly started to increase. One minute later, an extremely large gas bubble arrived at the surface and generated two lateral blasts (as reported by INGV in surveillance camera videos-see www.ingv.it) and a few seconds later the paroxysmal explosion started. In the time-lapse videos recorded by INGV thermal cameras, the giant lava bubble bursting out from the crater is clearly visible. Most of the summit area was then under rain of lava bombs of all sizes.

The 28 August Event
This paroxysmal event produced an eruptive column, which reached a height of 4 km above the crater summit. The paroxysm was preceded by an increase of the strombolian activity starting a day before. On 28 August, the paroxysmal sequence began, consisting of three explosive events, two of which were located in the central-southern crater area, while the third-the least energetic of the sequence-occurred 20 s later in the northern area, causing a lateral blast whose products spilled out over the SdF. The products generated by the collapse of the eruption plume, as well as those produced by the lateral blast, contributed to the generation of a pyroclastic flow: it traveled down the SdF, reached the sea level and traveled several hundred meters out to sea. The full explosive sequence produced several morphological modifications of the northern area of the crater terrace overlooking the SdF (see for further descriptions INGV website).

Data Interpretation
The 2007 source pressure variations have been related to overpressurization and depressurization of a shallow magmatic chamber, located at 1.5 km depth, responsible for the vulcanian blast (Bonaccorso et al., 2012). On Mt. Stromboli, a detailed analysis of volcanic tremor leads to the hypothesis of nonlinear coupling of gas fluctuations within surrounding rocks as the main source of observed wavefield and other signal features of volcanic tremor during the steady state behavior of the volcano (De Lauro et al., 2008). The dilatometer data ( Figure S0) reveal that the process of pressurization of the conduit starts about 11 min before the explosion, as indicated by the microbarograph signal.
Initially, there is an inflation at SVO, whereas TDF strain change has the opposite sign. This observation has been modeled in terms of a source inflation located at about 1.5 km depth, just before and following the explosion (Bonaccorso et al., 2012). Then, some minutes later, we observe contraction at both stations, corresponding to a shallower source, linked to a gas-magma slug along the upper portion of the conduit. The 20 s oscillations of strain after the explosion are similar to the observations made by Chen et al. (2018) at Montserrat volcano, suggesting an interaction between a shallow and a deeper magma chamber.
After the paroxysm occurred in 2007, Stromboli entered a new phase, characterized by persistent activity, sometimes accompanied by major explosions and/or overflow of lava from the summit craters. This phase culminated in an increase of the most energetic episodes, which led to an effusive eruption in August 2014 and a major eruption in October of the same year (Di Traglia et al., 2018). These events, although significant, have not been analyzed in the current paper because no data were available from the strainmeters.  Figure S2a shows that the change in strain starts 10 to 15 min before the events. Figure S2b shows that the two 2019 explosions are quite similar and larger than the 2007 event. Earth tides and atmospheric pressure are removed. A high pass filter (for periods shorter than 5,000 s) has been applied.

Geophysical Research Letters
DI LIETO ET AL.
The strainmeter signals recorded before, during, and after the paroxysmal activity of 3 July 2019 are surprisingly similar to the 2007 event but larger in amplitude and time duration (Figure 2). Due to malfunctioning of the recording system TDF, only records from the site SVO are available. The strain signal however shows a slow strain increase clearly evident at least 10 min before the explosion. During this build up, a VT event occurred, presumably triggered by the strain increase, as evident from the high frequency dilatometer signal. Two explosions correlate with two sudden strain drops and with the subsequent high frequency signals due to air shock and related air pressure fluctuations (E1 and E2 in Figure 3). The first air pressure pulse has been used as marker of the eruption onset, considering the delay time (almost 6 s from the main vent of the crater  The strain recorded at SVO strainmeter shows many features similar to the 3 July event. The overall strain pattern looks very similar to the one recorded during the previous paroxysm, showing a very slow strain buildup, which becomes clearly observable several minutes before the explosion. Two major explosions are visible from the air pressure data, occurring soon after the decrease of the strain ( Figure 4). As for the two previous paroxysms from 2007 and earlier in 2019, the strain pattern shows an abrupt depressurization starting about 21 s before the air shock wave-defining the beginning of the paroxysm-that is recorded by the barometer installed near the strainmeter.

Discussion
Stromboli eruptions shows many similarities with Sakurajima vulcanian eruptions, where a cyclic inflation-deflation pattern of ground deformation has been associated with each eruption (Iguchi et al., 2008). This cycle is used to forecast eruptions, as an inflationary radial tilt is observed for periods of 10 min to 7 hr prior to eruptions, enabling automated warnings to be issued (Kamo & Ishihara, 1989). The complexity of the magma feeding system, due to the presence of multiple vents at the surface, does allow a reliability of the eruption prediction scheme only at a confidence level of 70%. The gas leakage phenomena introduce further mechanisms, providing a more complicated pattern of precursors (Yokoo et al., 2013).
As related to the mechanisms of the paroxysmal activity of Stromboli, a general model for the dynamics of paroxysmal explosions at Stromboli is not yet available and remains a matter of debate. Since the magma plumbing system of the Stromboli volcano is widely studied from geophysical, vulcanological, and geochemical perspectives, it has been recognized that the feeding system can be depicted as a vertical column (see, e.g., Métrich et al., 2010), connecting the open vents on the surface with a shallow ponding zone located at 2-4 km bsl (Bonaccorso et al., 2012). The deeper reservoir is located at 7-10 km depth (Morelli et al., 1975;Ubide et al., 2019): It is now widely recognized that these deep, low porphyricity This phase lasts about 10 min. Orange shows the depressurization activity linked to the initial phase of the paroxysm and red the last phase associated to the blast, to the air shock and to refilling from a deeper magma chamber.
10.1029/2020GL088521 magma reservoirs play an important role in triggering paroxysmal activity. These eruptions are triggered by a larger supply of deep magma or gas-magma as demonstrated from the petrological analysis of products of the 1930 and XVI century eruptions (Bertagnini et al., 2011). The mechanism proposed by Calvari et al. (2011) suggests instead that the triggering is generated by the emptying of the upper part of the conduit due to prolonged lava activity. This process is based on the daily measurements of effusive activity and from the consideration that the threshold in the decompression rate is proportional to the height of the bubbly LP magma, as It has already been shown, from the strain data recorded during the 2007 vulcanian activity by Bonaccorso et al. (2012), the strain field changes at r ¼ ffiffi ffi 2 p d where r is the radial distance and d is the source depth, as a result of the Mogi model. The source depth d has been found at Stromboli to be at 1.4 km bsl, which is realistic also for 2019 sources due to the comparable size of these two explosions. By using the simple source model proposed by McTigue (1987), the observed strain during the 2019 events can be simply attributed to a source located at 1.4 km depth with volume loss amounting to 10 6 m 3 , which is comparable with the ejected volume of tephra. This approach is very simple and this depth may be considered also as a lower limit value.
The duration of the signals and the rate of strain build up and drop are compatible with a gas-magma recharge rate of ≈10 3 m 3 /s from a shallow reservoir. Pistolesi et al. (2011) estimated a comparable value for the 2007 paroxysm. The 20 s strain oscillations soon after the paroxysms are probably related to magma ascent fluctuations from a shallower source, located in the upper portion of the magma conduits. The cyclic strain changes can be interpreted as induced by the forced flow of melt and the magmatic fluids into a narrow dike which form the upper magma conduit. The values of the oscillations during the vulcanian explosions are compatible with a source of size of 200 meters. In the supporting information, there is a more detailed discussion on the approach to derive these last parameters in analogy with the longer period oscillations (20 min) recorded at SHV, Montserrat, modeled by Chen et al., (2018).
A schematic of the mechanism for the 2019 paroxysms is illustrated in Figure 5, in which the normal volcanic activity (preparoxysmal phase) is shown in green. The yellow phase can be interpreted as the pressure growth in the upper chamber due to the accumulation of the gas-magma slug. In this phase, strain builds up, until a threshold value is reached. From the time-lapse video published on the INGV website, it is possible to verify that once the maximum value of the strain is reached, all the crater vents start a more vigorous degassing phase, concurrently with the stress drop recorded by the strainmeter and depicted in orange in Figure 5: This is a usual habit for other open conduit volcanoes, like Sakurajima in Japan (Kazahaya et al., 2016). The final, red phase, occurs once the paroxysm begins: higher frequencies are observed as a consequence of the complex activity occurring in the summit craters area. This activity is very similar to the 1930 blast, as suggested by Bertagnini et al. (2011), which is the largest explosive eruption of the twentieth century at Stromboli. The 2007 mechanism is more complicated by the presence of the higher decompression rate due to the lava flow.

Conclusions
The installed borehole strainmeters contribute to the knowledge of the magma feeding system and deformation mechanism at Stromboli.
The high resolution of deformational changes in the subsurface allows borehole strainmeters uniquely to resolve signals that are typically invisible to other geodetic/geophysical monitoring instruments. Numerous studies on borehole strainmeter records from active volcanoes worldwide contributed already significantly toward (i) the detailed resolution of complex magmatic systems in the subsurface (Hautmann et al., 2013;Linde et al., 2016), (ii) the understanding of long-term (Hautmann et al., 2017) and short-term eruption dynamics (Hautmann et al., 2014), and (iii) the identification of early warning signals preceding the onset of eruptive events (Bonaccorso et al., 2014;Sturkell et al., 2013). Our study presented adds to this list, as we show three cases of high quality strain signals and some relatively long period oscillations, during the explosions, which help in putting constraints on the size of conduits. In addition, the analyzed data are of high value for early warning purposes, as the records unveil for the first time a distinct pressure increase over 15 min preceding discrete, short-term explosion events at Stromboli. An early warning system for civil protection will be implemented in future by using these signals and transmitting in real-time variable alert levels (see also Giudicepietro et al., 2020). Further contribution for a better and more quantitative interpretation of these data will be derived from the planned installation of an additional borehole strainmeter and a two long baseline high sensitivity tiltmeters and by the inclusion in the data base of all the geophysical and geochemical data collected by the present monitoring networks.