Geophysical Research Letters

Insights into the dynamic processes of the 2007 Stromboli eruption and possible meteorological influences on the magmatic system



[1] The unrest of Stromboli volcano leading to the February 27–April 2, 2007 eruptive period and to the March 15 paroxysm is constrained by combining broad-band seismic data and 1-Hz GPS (High Rate GPS – hereinafter HRGPS) measurements. During the pre-eruptive stage, the simultaneous examination of seismic and HRGPS data, together with weather parameters, suggests the possible influence of external perturbations on the magmatic system, which evolved toward a critical state after January 2007. Some days after the onset of the eruption, a sudden change of the seismic and eruptive behaviour was recognized, while ground deformation began to show a deflation. The March 15 paroxysm was preceded, about two days before, by a peak of the HRGPS spectral power densities (a small inflation) and by the occurrence of a few VT earthquakes located at depths down to 3.5 km b.s.l. These findings constrain, for the first time at Stromboli volcano, the deep origin of a fast rising magma batch, rich in gas, that led to a strong explosive event.

1. Introduction

[2] Stromboli island, one of the most active volcanoes in the world, is located in the Tyrrhenian sea (Figure 1). Persistent strombolian activity, suggesting an efficient gas–magma transport system, has been documented for over 2000 years. Its continuous explosive activity is sometimes interrupted by lava effusions (last events in 1985, 2003 and 2007) or major explosions [Barberi et al., 1993].

Figure 1.

Map of Stromboli reporting the location of the seismic (yellow triangles) and GPS (red triangles) stations. Inset map shows locations of meteorological stations (Palinuro and Cetraro) used in this study.

[3] Although seismic and ground-deformation monitoring of Stromboli has been performed continuously, respectively since 1985 and 1992 [Falsaperla et al., 2003; Bonaccorso, 1998], a notable improvement of the monitoring system (Figure 1) started only during the 2002–2003 eruptive crisis.

[4] Notwithstanding the monitoring upgrade and the most advanced technologies for ground deformation studies [e.g., Mattia et al., 2004], no short-term signals of an impending eruption were recognized before the 2007 eruption.

[5] After the 2002–2003 eruptive crisis, the volcanic activity returned to normal strombolian explosions. In the following years, the signals related to the volcano-tectonic activity recorded by these networks were very weak, compared with analogous signals collected at other volcanoes.

[6] Moderate volcano-tectonic activity (MLmax = 3.3) began in April 2006 and continued until December 11, 2006, when the last of the VT earthquakes (ML = 2.5) was recorded (INGV, Sezione di Catania Reports; The occurrence of VT earthquakes cannot be considered a usual feature at Stromboli [e.g., Falsaperla et al., 2003] and suggests a first possible indicator of volcanic unrest.

[7] In this paper, the limitations of the Global Positioning System (GPS) analysis, normally applied for volcano monitoring purposes, are overcome by computing the spectral analysis of the HRGPS data. We also performed a careful analysis of seismic activity, focusing on the period encompassing the March 15 paroxysmal explosion. Finally, we demonstrate the usefulness of the comparison of geophysical data with weather parameters in order to investigate any possible influence of external perturbations on the magmatic system.

2. Seismic Data and Signal Characteristics

[8] To investigate the seismo-volcanic signals at Stromboli in the time-frequency domain, a standard spectral analysis was used together with wavelet transform and wavelet coherence analysis. The comparison of both spectrograms and wavelet scalograms (Figure S1 in the auxiliary material) was found to be useful for a fast classification of Stromboli's seismic events (very long period, VLP; long period, LP; volcanic explosion, EXQ; Hybrid, Hy; and volcano-tectonic, VT, events). After an accurate analysis of many different types of events, in terms of spectral content, we devised an automatic detection procedure based on the calculation of signal amplitudes in limited frequency bands (see Text S1 for details).

[9] The most relevant seismic features (Figures 2a–2g), recognized during the pre-eruptive stage are the patterns of the RMS tremor amplitude in the LP band (Figure 2d) and the energy of LP + EXQ events associated with the explosive activity at the summit craters (Figure 2c). On January 2–3, 2007 a strong peak in the RMS tremor amplitude in the VLP band (Figure 2g; first yellow line) marked the onset of these variations in the LP tremor band, which progressively increased until the onset of the eruption. Conversely, the VLP source during this period showed no meaningful variation in terms of locations (INGV, Osservatorio Vesuviano;, while minor fluctuations both in the number of recorded events (Figure 2e) and in their RMS amplitude (Figure 2f) were observed. Regarding the peaks in the VLP tremor band, it is well known that, in this frequency range, sea microseismic noise affects recordings. This is the result of a combination of atmospheric processes (e.g. wind waves and sea waves). Microseismic noise is mainly composed of Rayleigh waves, with the most significant spectral peak around 0.2 Hz. However, seismic signals can be corrupted over almost any frequency [e.g., Correig and Urquizu, 2002, and references therein].

Figure 2.

Comparison between seismic, ground deformation HRGPS, and weather observations in the period October 2006–April 15, 2007. The seismic patterns include the counting of the number (a) of Hy + VT, (b, c) LP + EXQ, and (e, f) VLP events, the related RMS values (energy content) and the RMS amplitude of the volcanic tremor in the (g) VLP and (d) LP bands. HRGPS data refer to SPLN station, while weather observations refer to Palinuro and Cetraro stations (Figure 1). Red lines indicate the vent opening (VO) on March 09 and the March 15 paroxysm (PE). The shadow area marks the eruptive period. The yellow lines mark the wide peak in the signals in the VLP band (Figures 2f and 2g), which coincide with the strongest variations (l, m) in the meteorological conditions.

[10] After the onset of the eruption, on February 27, when lava flows poured from a vent opened in the Sciara del Fuoco at ca. 400 m of altitude, both the tremor amplitude and the explosive activity at summit craters returned to the levels previous to January. Afterward, a sudden change of both the eruptive behaviour and seismic activity was observed. An increase in the number of LP events (Figure 2b) was recorded, together with the appearance of a higher number of Hy + VT events (Figure 2a). In particular, after the swarm of strong Hy events on March 7, a new effusive vent opened on the Sciara del Fuoco on March 9.

[11] The most clear variation in the amplitude of the VLP events occurred after the March 15 paroxysmal explosion (Figure 2f), contemporaneous with the increasing amplitude of the LP events (Figure 2c). In this period, three strong seismic swarms of LP and Hy events occurred on March 20, 22 and 27 (Figures 2a and 2b). The eruption stopped on April 2, after a declining period in the lava effusion rate.

3. Source Location

[12] The location of earthquakes and seismo-volcanic events at Stromboli volcano is an extremely difficult task, due to (1) the large number of small magnitude events recorded daily (300 in average); (2) the high level of noise (volcanic tremor) and (3) the network station geometry, that has mainly been designed to locate VLP events within the volcanic edifice. In this study, we determined the hypocenters of the most energetic LP and Hy events and of the few shallow VT earthquakes, that exhibited the best signal to noise ratio and an impulsive onset. Approximately, 500 events were considered between February 27 and March 20, 2007. P- and S-wave arrival times were picked and verified by the polarization analysis, based on the Covariance Matrix Decomposition polarization filter. The analysed events show predominantly body-wave energy. With the exception of deeper shocks, the transversal energy is usually confined within 1 s after the first onset. Chouet et al. [1997] suggested that near-field observations indicate that most of the transverse wave is radiated directly by the source.

[13] These selected events have been located considering two steps. First, we computed absolute locations by using the HYPOELLIPSE program with the 1D velocity model by Barberi et al. [2007], slightly modified in its shallow part according to Petrosino et al. [2002]. Then we considered the 270 best located events (hypocentral errors: Erh ≤ 0.8 km and Erz ≤ 1.6 km; rms ≤ 0.24 s; azimuthal GAP ≤ 330°, ≤ 180° in 84% of cases), in order to perform a precise relative relocation by using the TomoDD code [Zhang and Thurber, 2003]. We used the TomoDD code because it uses both absolute and relative arrival time data. The final locations show a considerable reduction of hypocentral errors (Erh ≤ 350 m and Erz ≤ 500 m in 88% of cases) and a more reliable distribution of the focii.

[14] In Figure 3, an evident stability of the seismic sources until March 13 is observed. Afterward, few VT earthquakes occurred, between 0 and ca. 3.5 km of depth, just one day before the March 15 strong paroxysm. On March 16 and 17 a deeper seismicity, down to about 4.5 km b.s.l., was observed. It disappeared on March 18.

Figure 3.

Epicentral map and W-E cross section (A–B) of the 270 best located events (Hy, LP and VT). A comparison between the temporal variations of the seismic foci and of HRGPS data (SPLN station) and SPLB-STDF baseline is also shown.

4. Ground Deformation by High-Rate GPS

[15] HRGPS data at Stromboli are processed in real time once per second using the Epoch-by-Epoch algorithm (for details, see Mattia et al. [2004]) and daily using the GAMIT-GLOBK software in a regional framework. Then HRGPS signals have been analysed in the frequency domain by computing spectrograms. The investigations of these spectrograms suggested to focus the analysis on the 2 to 5 minute period (0.0033–0.0083 Hz). Frequencies under 2 minute were discarded because of the possible influence of Rayleigh waves and sea-waves. To emphasise the contribution of the signal components in this range of frequencies, we calculated, at time t, the cumulate energy spectral densities in a fixed time window T. As output we obtained a time series sampled at 4 hours which represents the estimated power spectral density of the signal in the considered frequency range.

[16] Figure 2h shows this series for the east component of SPLN station (baseline SPLN-SVIN) in the investigated period. We note that a series of wide peaks coincide with weather parameter variations. For this reason we perform a coherence analysis between these time series, in order to distinguish peaks “free” of bad weather conditions from peaks influenced by severe weather (see next paragraph). The peak of March 14, occurring about two days before the paroxysm of March 15, can be considered totally free of any external perturbation (Figure 2h–2m). In Figure S2 we show the bi-dimensional azimuthal dispersion of the 1 Hz planar displacement data (E-W versus N-S component) on March 14 at the different stations. Furthermore, we compare these displacements with those ones computed during two periods characterized by different weather condition (calm and strong perturbation). It is noteworthy, however, that other observed peaks could be related to volcanic activity such as that occurring just when the eruption started.

5. Wavelet Coherence Analysis Between Geophysical and Weather Parameters

[17] Besides large earthquakes, several processes and events external to volcanoes are known to influence volcanic activity [Manga and Brodsky, 2006, and references therein]. Figure 2 shows that some anomalies in the geophysical and weather time series coincide. Therefore, we investigated the possible relationships using the cross-wavelet analysis. We chose this type of analysis in order to overcome the problems of classical cross-spectrum analysis, where the window width and thus the time resolution is constant for all investigated frequencies. However, since the wavelet cross-spectrum alone may appear unsuitable for a test of the relationships between different processes, we calculated the wavelet coherence between the analysed time series [Maraun and Kurths, 2004].

[18] The resulting coherence-checks, performed between ground deformation HRGPS displacement for the SPLN station time series and the considered weather parameters (barometric pressure, sea amplitude and wind velocity), reveal that a link seems to exist between the HRGPS signal and the wind velocity time series (Figure S3a and S4). Instead, for the seismic parameters the strongest link exists between the sea spectral amplitude and the RMS tremor amplitude in the VLP band (Figure S3b) and number of VLP events.

[19] Furthermore, we can observe (Figure 2, yellow lines) how, during the investigated period, severe weather conditions (low atmospheric pressure, high wind velocity and high sea spectral amplitude) would seem to coincide with (1) the onset of the increasing tremor amplitude in the LP band (January 2–3), (2) the increase of amplitude of LP events (January 23–24); (3) a new increasing phase of the tremor amplitude in the LP band (February 13–14) and (4) the increase in the amplitude of VLP and LP events and a further change in the LP band tremor pattern (March 18–24). We suggest that these clear correlations could represent the seismic response to a possible meteorological influence on the magmatic system. It is clearly necessary to investigate this possibility over a longer time period, in order to define better the influences that meteorological conditions have on volcanic activity.

6. Conclusions

[20] The eruptive activity at Stromboli is normally related to the explosive activity at the summit craters. The lack of clear geophysical signals preceding volcanic activity (lava eruption and strong paroxysmal events) can be related to the low energy involved in magma upraise mechanisms. The low energy may be due to the small quantities of magma periodically injected in the shallow magma chamber, located above the sea level in the body of the volcano. This is also confirmed by petrological and geochemical data [e.g., Landi et al., 2004], suggesting how the volcano uniform eruptive regime may be associated with continuous refilling, mixing, crystallisation and eruption of magma.

[21] The evolution toward a critical state at Stromboli leading to the 2007 eruption is likely related to the overpressure of the magmatic system occurring during 2006, when unusual VT seismicity was recorded at 4–6 km of depth beneath the island. Since January 2007, this pressure also affected the shallow reservoir within the subaerial edifice. Moreover, the severe weather conditions recorded on January 2–3 and 23–24 and on February 13–14 (yellow lines in Figure 2), that induced fluctuating stresses on the magmatic system, seem to be related to the eruptive activity variations. Nevertheless, further investigations over longer time periods are necessary to validate this hypothesis.

[22] Finally, the March 15 paroxysm was preceded about two days before by a peak of the HRGPS spectral power densities (small inflation) and by the occurrence of VT earthquakes located down to 3.5 km b.s.l. of depth. This is an indication of a stress field variation induced by pressure changes in the shallow plumbing system and probably in a magma chamber located at a depth of 3 ÷ 5 km b.s.l., as also suggested by the deeper seismicity following the paroxysm. Our findings constrain, for the first time, the deep origin of a fast rising magma batch, rich in gas, that led to a paroxysm at Stromboli volcano. This interpretation matches the recent geochemical observations [Burton et al., 2007].


[23] We would like to thank to the two anonymous reviewers for helpful suggestions and to our colleagues M. Martini and L. D'Auria at the INGV-Osservatorio Vesuviano for maintaining BB seismic stations as well as providing helpful discussions.