By continuing to browse this site you agree to us using cookies as described in About Cookies
Wiley Online Library is migrating to a new platform powered by Atypon, the leading provider of scholarly publishing platforms. The new Wiley Online Library will be migrated over the weekend of March 17 & 18. You should not experience any issues or loss of access during this time. For more information, please visit our migration page: http://www.wileyactual.com/WOLMigration/
 We analyzed short period seismograms recorded at Stromboli, searching for a possible signature of the tsunami following the two landslides occurred on the island on 30 December 2002. By taking advantage of their close occurrence in space and time, we recognized the seismic mark of clock- and counterclockwise tsunami waves trapped by the island coast. The most energetic direct arrivals, running around the island in the two opposite directions, traveled with an average velocity of about 12 m/s, while the waves that traveled more than once around the island display a slightly higher velocity. These observations represent the first instrumental evidence of trapped waves around pseudo-conical islands and demonstrate the possibility of using coastal seismic stations for monitoring tsunami propagation. Our results provide important constraints for early warning of tsunami generated at Stromboli in the Sciara del Fuoco area, where considerable landslides are known to have occurred in the past.
 Trapped waves could produce disasters, causing severe damage even in sheltered areas, located on the opposite side of the tsunami source, where the waves traveling around the island in opposite directions superimpose. This effect, previously reported by Bascom , is well documented for the tsunamis following the 1992 Flores (Indonesia) and 1993 Hokkaido Nansei-oki (Japan) earthquakes, which produced unusually large run-up heights on the lee side of Babi [Yeh et al., 1993] and Okushiri Islands [Hokkaido Tsunami Survey Group, 1993], respectively, destroying entire villages and causing loss of lives.
 The theoretical problem of waves refracted by conical islands has been studied, either analytically or numerically, by a large number of authors, starting with the paper published by Arthur . From the conceptual point of view, trapping of tsunami energy around conical island results from sea waves refraction caused by island slope. To a first approximation, in shallow water the wave velocity depends only on water depth d, in the form v = , where g is the acceleration of gravity. Then, due to the dipping sea bottom all around the island, sea waves repeatedly bend toward slower propagation velocity areas, therefore, back to the coast. Trapped waves may continue whirling around the island several times and, for more energetic tsunami, successive arrivals could be observed at the same location. Because of the non-perfect trapping, some energy continuously leaks away and, in the end, the amplitude dies out.
 In addition to theoretical studies, several laboratory experiments have been carried out on island models in small basins, starting with the pioneering work of Synolakis . In spite of numerous simulations, no clear instrumental observation of island trapped waves exists for real tsunami. The quantitative observation of an actual case would provide the chance to check the predictive capabilities of the modeling techniques and define the accuracy of the scenarios used for damage mitigation. In these regards, the detection of tsunami waves traveling clock- and counterclockwise would provide simple and direct indication of island trapped energy. However, great tsunamis come ashore like long lasting sea oscillations and, when small islands are involved, this would prevent the observation of multiple arrivals. Due to (1) the small mass involved in the slumps causing the tsunami, (2) the vicinity of the landslides to the shore, (3) the relatively limited areal extension of the island, and (4) the approximately conical shape of the island with steep slopes, the 30 December 2002 Stromboli Island (Aeolian Islands, southern Italy) tsunami provides a suitable framework for the generation and observation of trapped sea waves.
2. The 2002 Stromboli Landslides and Tsunami
 Stromboli Island is an active volcano located in the southern Tyrrhenian Sea. It stands 924 m a.s.l., with a coastal length of about 15 km, but the whole volcano is about 3000 m, referred to the seafloor, with very steep slopes (Figure 1). Its name derives from the Greek term Strongylos, “round”, because of the pseudo-circular perimeter. All the present eruptive products pour in the Sciara del Fuoco, a large depression characterizing the NW flank of the volcano from the crater terrace, at about 700 m a.s.l., down to almost the submerged base of the apparatus.
 Several important collapse episodes took place in the last 13 ka along the NW flank of Stromboli [Tibaldi, 2001] and, in particular, in the Sciara del Fuoco sector. The last one occurred at the beginning of the 2002–2003 eruptive crisis. Based on eyewitnesses and bathymetric and topographic surveys, two main episodes could be recognized for this latter collapse. A first submarine failure, displacing about 10 × 106 m3 [Chiocci et al., 2008], was followed by a subaerial mass slump totaling a volume of about 5 × 106 m3 [Tommasi et al., 2005] (Figure 1b). Both landslides caused tsunami waves detected all along the southern Tyrrhenian coasts, with a maximum runup of 10 m, measured on Stromboli [Tinti et al., 2006a]. At that time, only very few geophysical instruments were installed on the island and no tide gauge or wavemeter were available. Short period seismograms are the sole quantitative, non-static, recordings that provide useful information on the processes involving the 30 December 2002 landslides and the associated tsunamis. In fact, by using the waveforms recorded at the seismic station SX15 (Figure 1b), Pino et al.  derived a chronological description of the slump episodes, also providing mass estimates compatible with the results of the static investigations, based on analysis of differential topography and bathymetry [Tommasi et al., 2005; Chiocci et al., 2008]. Pino et al.  results indicate that the subaerial slump occurred about 7 minutes after the first submarine one, each of the two landslides being composed by a succession of distinct detachments closely spaced in time. The slump sequence for the first and the second landslide lasted, respectively, about 120 s and 90 s.
3. Seismic Mark of the Tsunami
 In the present study, we analyze the same waveforms used by Pino et al. , focusing on the detection of tsunami waves signature. The recording site was on the eastern side of the island, less than 1 km from the closest coastal area, Ficogrande beach, and about 2 km from the area where the detachments occurred (Figure 1b). The signals produced by the two landslides are very clear in the original recordings and, in between, a couple of short, spindle-shaped phases are also evident (Figure 2a). These latter have frequencies between 2 Hz and 4 Hz, very different from the frequency content of the waves associated with the gravitational mass sliding, which exhibit maximum energy in the range 0–2 Hz (Figure 2b). A large single pulse is also present, at low frequency, in the deconvolved displacement, velocity, and acceleration waveforms (Figure 3). According to Pino et al. , this long period, high amplitude pulse is due to tilt effect caused by the load of the huge amount of water that invaded the coast in Ficogrande beach at the time the tsunami most energetic phase reached the shoreline. In fact, Ficogrande beach is the area where the maximum inundation was measured. Similarly, Yuan et al.  explained the long period signals observed on seismic broadband recordings of the 26 December 2004 Indian Ocean tsunami in terms of tilt due to coastal loading. Both these results evidence that tsunami inundation effects may be recorded by seismic stations.
 The above mentioned spindle-shaped events occur in correspondence with major changes in the large acceleration pulse (Figure 3) and are then to be associated with the ingression and regression, respectively, of the tsunami wave on the rocky beach. Pino et al.  suggested that the dragging and rolling of pebbles and larger stones produced those signals. Following their hypothesis, we searched for similar wave trains, possibly produced by the second subaerial landslide. Moreover, if trapped waves were generated with sufficient energy, further arrivals could be detected, for both episodes, corresponding to the most energetic phase of sea waves traveling along different or/and longer paths.
4. Evidence of Tsunami Trapped Waves
 In order to enhance the high frequency phases and detect the presence of possible later arrivals, we band-pass filtered the original waveforms in the range 2–3.5 Hz, squared and, then, low-pass filtered the result at 0.1 Hz to smooth the signal (Figure 4). Two major peaks (the largest is indicated by 1a in Figure 4) are very well evidenced in correspondence of the spindle-shaped phases associated by Pino et al.  with the arrival at Ficogrande beach of the tsunami produced by the first submarine landslide. A small peak is also visible about 45 s before the one corresponding to sea ingression (highest pulse). As a matter of fact, according to eyewitness accounts, the very first sea movement during the event was characterized by water recession [Tinti et al., 2006a]. Also, unlike what predicted for subaerial landslides, a negative sea movement is expected from modeling of submarine landslides along the Sciara del Fuoco, at Stromboli [Tinti et al., 2006b]. Therefore, the small peak preceding 1a corresponds to the observed regression. In this analysis we focus on the high frequency seismic events, such as the spindle-shaped ones, corresponding to the most energetic phases of the two tsunamis - as seen by a seismic station - which likely correspond to water ingression and are not necessarily the first sea wave arrivals. On this basis, in comparing the sea waves generated by the two landslides, in the following the origin time of the first tsunami is delayed by 45 s in order to account for the different first sea movement, On this basis, the arrival time at Ficogrande of the sea wave associated with phase 1a is about 320 s. Combining this estimate with the length of 3.8 km, which represents the distance measured along the coast from the source area to Ficogrande beach, an apparent sea wave velocity of 11.9 m/s (42.7 km/h) results, well compatible with generally observed velocities of tsunami waves in proximity of the coasts [e.g., Garay and Diner, 2007]. By dropping the assumption of a 45 s shift for the tsunami origin time, a velocity difference of 1.5 m/s would result for this phase and even smaller for later arrivals.
 Several other major peaks are present in the time series. In particular, a distinct spike (2a) is evident at less than 7 minutes from arrival 1a, a time interval corresponding to the lag between the two landslides. This evidence suggests that the phase 2a is likely to be produced by the arrival at Ficogrande of the sea wave originating from the subaerial detachment. The presence of similar phases, displaying the same delay from the two main mass detachments, confirms Pino et al.'s  interpretation in terms of tsunami effects, rather than isolated, later slump episodes occurring with the same delay from the each landslide.
 The phase 1a, corresponding to the first tsunami, displays a series of relatively high amplitude peaks, while a single spike, with smaller amplitude, characterizes the arrival 2a. In our interpretation, this feature reflects the higher energy of the sea waves originated by the submarine landslide, with respect to the subaerial one, due to the larger displaced volume and the better coupling with the water that produce larger and more vigorous water ingression and regression. The apparent absence of any other acceleration peak (Figure 3) testifies the significantly smaller amount of water involved in the ingression caused by the second tsunami at Ficogrande beach.
 A small group of further major peaks (1b) is evident in Figure 4. Its maximum occurs after about 580 s from phase 1a, i.e., 900 s from the origin of the first landslide. By using the distance of 15–3.8 = 11.2 km, this time corresponds to an apparent velocity va = 12.4 m/s (44.8 km/h), very close to what obtained above for the clockwise wave. The small discrepancy could be due either to different average depth along the two paths or to a small uncertainty in the distance evaluation.
 Again, the interpretation of phase 1b as produced by sea waves is strongly supported by the presence of a well defined peak (2b) after about 7 minutes. In addition, the phases 1b and 2b display the same number of peaks as arrivals 1a and 2a, respectively. In this framework, the peak 2b is to be associated with the arrival at Ficogrande of the tsunami wave caused by the second landslide and propagating counterclockwise, similarly to the sea wave generating phase 2a.
 For each tsunami, the superposition of the maximum amplitudes traveling in the two different directions should have occurred after about 10 minutes from the landslide onset, on the opposite side of the island with respect to the Sciara del Fuoco, in Malpasseddu area, a coastal sector fortunately not inhabited.
 Some energy (1c) is visible about 200 s before peak 2b. This time corresponds to 840 s after the first tsunami arrival 1a. Similarly to what evidenced for couples 1a–2a and 1b–2b, a conjugated single peak 2c can be recognized at something less than 7 minutes from 1c, indicating that these phases represent the effect of analogous waves generated by the two landslides. We suggest that these energy bursts are caused by clockwise running sea waves that reached the coast in Ficogrande after whirling around the island more than once. By considering the distance of 15+3.8 = 18.8 km for this path, an apparent velocity va = 16.2 m/s results, which is larger than what derived for the waves corresponding to 1a, 2a, 1b, and 2b. Actually this is not surprising and corresponds to what would be expected for a pseudo-conical island. In fact, as evident from ray patterns of trapped waves [Liu et al., 1995], paths around the island are more spiral-like rather than perfectly circular and, in general, sea waves traveling longer paths spend more time at larger distance from the coast, in areas with larger water depth d, where the propagation velocity va is higher. As a result, the time delay between the arrivals of two waves, whose paths differ by a whole lap, is shorter than the sum of the times needed to go from the source to any place along the coast by traveling one way and the other way around, respectively.
 The average sea floor depth d along the paths traveled by the different phases can be derived by using the above formula for va, obtaining values of 14.4 m, 15.7 m, and 26.8 m for couples 1a–2a, 1b–2b, and 1c–2c, respectively. However, it should be remarked that the apparent average velocities are lower than actual ones, since we used the distance measured along the coast, while trapped sea waves travel along slightly longer paths.
 The close occurrence, in space and time, of two landslides at Stromboli, on 30 December 2002, provided the chance for the detection of the seismic mark of the tsunami generated by each slumps. By analyzing the seismic waveforms recorded close to Ficogrande beach, where the maximum runup was observed, we recognized the effect produced by the arrival to Ficogrande of the tsunami waves trapped by the island coast. In particular, for each of the two tsunamis, we evidenced the seismic signal produced by the arrival of trapped waves traveling in both clock- and counterclockwise directions. We also interpreted a later arrival as due to trapped sea waves hitting the coast after traveling around the island more than once. The resulting apparent velocities, on the order of 10–20 m/s, are in line with what observed for tsunami propagation in coastal areas.
 To our knowledge, this evidence represents the first instrumental observation of tsunami trapped waves around a pseudo-conical island and, in any case, the first obtained by means of seismic recordings. Besides providing important constraints for testing theoretical models, our analysis demonstrates the possibility of using seismic stations located in coastal areas for monitoring tsunami wave propagation, even when only relatively short-period instruments are available.
 At Stromboli, the Sciara del Fuoco - the area where the tsunami originated - is known to have been the locus of several important collapses in the past and the availability of reference times for the development of alert systems are highly desirable. We provide the travel times of the most energetic sea waves generated by landslides along the Sciara del Fuoco to Ficogrande beach, which is located in proximity of the most exposed and inhabited part, in an area where the population totals 400, but it may reach the number of 4000 during summer.