On July, 12, 2001, Mt. Etna experienced a sudden increase of seismic activity heralding one of the most intense eruptions of the past 30 years. Between July 12 and July 18, when the eruption started, thousands of small magnitude earthquakes occurred and were recorded by a dense seismic network run by the Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania (INGV-CT). Hypocentral depths of earthquakes were very shallow, mostly located above 3 km b.s.l. and clustered near the summit area. The high quality seismic dataset gives us the unique opportunity to study the process of magma migration before the eruption. In this study we present the three-dimensional earthquake locations and the velocity structure obtained by a tomographic inversion. The shallowness of seismicity allowed us to enhance the details of the structure beneath the summit craters, in a volume poorly defined by previous tomographic studies. The presence of a high Vp-body previously observed at Mt. Etna is confirmed at shallow depth beneath the southeastern part of the summit area. The earthquakes preceding the eruption onset concentrated at its western border. A low Vp/Vs anomaly is found at 0–1 km depth, just at the top of the volume where the magma intruded before the eruption. This anomalous zone can be considered as molten material wealthy in gas. The relocated seismicity occurs in a cylinder below the vents activated along the fracture system and exhibited an upward migration until the eruption. All these results show evidence for the emplacement of a near-vertical dyke striking about N-S and a few kilometres south of the summit craters.
 In the past decades, different seismological techniques have provided useful information on the physical state of volcanic structures, and on the shape and dimension of magma chambers beneath a number of volcanoes. Seismological observations complete and integrate the models of magmatic systems derived from other disciplines. Among seismological techniques, seismic tomography represents the most common method to generate three-dimensional images of velocity and attenuation [e.g., Sanders et al., 1995]. The main limitation of this approach is related to the uneven distribution of earthquakes and seismic stations and to the coupling between hypocentral and velocity parameters [see Thurber, 1993]. The use of a dense seismic array and the availability of a large number of earthquakes recorded at 3-component stations allow us to reduce such problems and to resolve velocity variations in highly heterogeneous media to a scale of some hundreds of metres.
 In the past years, numerous tomographic studies have focused on Mt. Etna using different techniques and data sets. Hirn et al.  obtained a 3D velocity model for the central portion of the volcano by the inversion of arrival times from local earthquakes recorded during a seismic crisis in 1984. Subsequently, Cardaci et al.  and De Luca et al.  obtained a 3-D Vp model within the volcanic edifice down to depth of about 20 km. In all these studies a high Vp body is postulated under the volcanic edifice. A further extension of these last two works is reported in Villasenor et al. , in which a non-linear inversion procedure is used to obtain the three-dimensional P-wave velocity structure. The chief limitation of all these studies is related to the data set of local earthquakes, since most of the first arrivals of earthquakes were read from paper recordings at only a few stations.
 Recently, Chiarabba et al.  and Laigle et al.  furnished a more detailed tomographic reconstruction of the upper crustal structure beneath Mount Etna, by using digital data from a large number of permanent stations and from a dense temporary array of three-component stations, respectively. The main structural feature observed is a high-Vp body embedded in the continental crust covered by thick sedimentary layers. Large high-Vp bodies are commonly found in several volcanic areas and interpreted as solidified intrusions [see Chiarabba et al., 2000, and references therein]. The high-Vp body at Mt. Etna shows a large contrast with the surrounding rocks. It is centered under the southern part of the Valle del Bove and extends also under a part of the summit region. Laigle et al.  also showed that this massive body presents a high degree of heterogeneity of the Vp/Vs ratio. In this work we present Vp and Vp/Vs models and 3-D located seismicity occurring a few days before the most important flank eruption of the past 30 years. The shallowness of the seismicity allowed us to enhance the details of the structure in the upper 3 kilometres of crust beneath the volcano edifice, a volume poorly defined by previous tomographic studies. The eruption occurred six days after the onset of seismicity, clarifying any ambiguity in the interpretation of velocity anomalies and seismic occurrence. For the first time, seismological observations concur to unravel the existence of a rising dyke which fed a lateral eruption at Mt. Etna.
2. Data and Method
 Between 12 and 18 July, 2001 a total of 2645 earthquakes were recorded by the dense permanent seismic network (Figure 1) run by INGV-CT. The seismic swarm preceded and accompanied the opening of a system of fractures along the western rim of the Valle del Bove (VDB in Figure 1).
 P- and S-wave arrival times have been read on digital waveforms recorded at 25 stations among which are 8 three-component stations. Over 350 earthquakes were located using HYPOELLIPSE code [Lahar, 1989], in order to take into account the difference in seismic station altitudes and correctly detect the hypocentre spatial patterns in the volcanic cone. These preliminary hypocentral locations led us to identify three main seismogenetic zones (depth range between −1 and 5 km). The first, located at −1 to 3 km b.s.l. beneath the summit craters and Mt. Calcarazzi, is characterized by strongly clustered epicenters, aligned in a N-S direction, following the eruptive fissure system. The second zone lies a few kilometers to the west of Zafferana, and has hypocentral locations reaching 5 km b.s.l. The last seismogenic zone is located in the western flank of Mt. Etna, and was characterized by shallow foci mainly up to 1 km b.s.l.
 To perform the 3D inversion, we selected 286 best located events of this swarm, which satisfy the following criteria: hypocentral errors ≤1.0 km, at least 15 P- and 4 S- phases recorded at 3-C stations, azimuthal gap ≤120 degrees.
 The inversion method used is that developed by Thurber  and modified by Eberhart-Phillips  and Eberhart-Phillips and Reyners , which solves for Vp and Vp/Vs parameters on a 3D grid of nodes, where the velocities are continuously defined within the volume. Hypocentral locations are simultaneously determined and updated at each iteration step, while Vp and Vp/Vs parameters are computed by inverting P and S-P times with damped least-squares algorithm. The model has been parameterized with five layers between −1 and 3 km depth and nodes spaced every 2 km horizontally. Vp values assigned to each node are from the 1D velocity model by Chiarabba et al.  and a Vp/Vs of 1.73 is used, taken from Laigle et al. .
 A total of 4152 P- and 796 S-P times have been inverted to determine 968 velocity parameters. After 4 iteration steps, we found a variance improvement of 37% and a final rms of 0.11 s. Relocated earthquakes have hypocentral errors less than 0.2 and 0.4 km, horizontally and vertically respectively.
 The resulting three-dimensional P-wave velocity structure between 0 and 3 km depth is shown in Figure 2. Results generally show low velocities (2.5–3.6 km/s) extending from the surface to depths of about 3 km, which can be correlated both with the rocks of the volcanic edifice and with thick sedimentary deposits. A main high-Vp body is found from 1 km b.s.l. to 3 km b.s.l. beneath the southeastern part of the summit area (Figures 2 and 3). Earthquakes occur at the western border of the high Vp body and are mostly elongated in a N-NW direction. The Vp/Vs model shows a positive and a negative anomaly at −1 km a.s.l., interpreted as regions with different degrees of saturation in hydrothermal fluids. The most interesting anomaly is a low Vp/Vs anomaly at 0 and 1 km depth, even if the resolution below 1 km depth is poor. This anomaly coincides with the region where the magma intruded before the flank eruption.
 The resolution of Vp and Vp/Vs models has been verified by a complete analysis of the resolution matrix, visually inspecting diagonal elements, averaging vectors and the spread function [see Chiarabba et al., 2000 and references therein]. We have recovered all the information on the density and ray coverage within the modeled volume. The central part of the Vp model has diagonal elements higher than 0.5 (see Figure 2), indicating a good resolution of the main revealed features in all the inverted layers. Resolution decreases at 3 km depth and below. Instead, the resolution of the Vp/Vs model is fair at −1 and 0 km depth and degrades downwards, due to the small number of S-P times. Thus, the central low Vp/Vs found at 1 km depth (see Figure 2), even if very attractive, is not well resolved by our data.
 The cross-sections of Figure 3 clearly show that the central high-Vp body is centered under the southern part of the Valle del Bove. This result confirms and integrates previous studies [e.g., Chiarabba et al., 2000; Laigle et al., 2000], because its top is found to reach about 1 km b.s.l.
 Comparing Vp and Vp/Vs anomalies, we observe: i) that the low-Vp medium in which the high-Vp massive body is embedded has high Vp/Vs, in agreement with the characteristics of sedimentary rocks where water-filled pores and cracks exist; ii) an anomalous wide region of low Vp/Vs with minimum values located near the top of the central-western part of the high-Vp massive body.
 The relocation of the seismic swarm indicates that the earthquakes mainly occur in the upper part of the high-Vp body, defining two main sub-cylindrical clusters. The western cluster is N-NW elongated and slightly bent toward the summit craters and is located below the vents activated along the fracture system. At 1 km depth, the earthquakes surround the low Vp/Vs anomaly (Figure 3).
4. Concluding Remarks
 Tomographic images obtained in this study allow us to get a better resolved velocity structure in the upper three kilometers of the crust beneath Mt. Etna volcano and to define a model for the very shallow part of the plumbing system. We found that the use of a detailed 3D Vp and Vp/Vs model improves earthquake locations and produces significant changes of hypocenters due to the high velocity heterogeneity of the shallow crust. This is testified by the relocation of the 12–18 July seismic swarm. Hypocenters are better clustered with respect to the preliminary locations, and well define a cylinder located below the vents activated along the eruptive fracture system, exhibiting an upward migration until the eruption (Figure 4).
 Apart from the improvements on the shape and geometry of the upper portion of the high-Vp body observed in previous tomographic studies, an interesting result obtained in this work is the presence of a low Vp/Vs anomaly at the top of the volume where the magma intruded before the July–August 2001 flank eruption. Seismic velocity variations depend on the characteristics of the rock (fracture, porosity, saturation) and on its physical conditions (temperature and pressure). In the literature an increase of Vp/Vs is related to increases in temperature, fracture and especially partial melt, while a decreasing ratio can be associated with the presence of gas or supercritical fluids [see Sanders et al., 1995 and references therein]. Laboratory experiments found that for igneous rocks, such as basalt or granite, the Vp/Vs generally decreases as temperature increases, while it increases as pressure increases [see Christensen, 1989; Muller and Raab, 1997; Lees and Wu, 2000]. In our case, the evidence that the expulsion of a significant volume of products occurring during the July–August eruption supports the interpretation of the shallow low Vp/Vs anomaly (values as small as 1.65) as a region of molten material wealthy in gas. Temperatures computed for the erupted basaltic lavas are higher than 1100°C. The sedimentary rocks surrounding the main volcanic feature, which present low-Vp, need to have even lower Vs because of lithology, pores, cracks and their fluid content, so that Vp/Vs is relatively high.
 Seismicity and tomograms point to the existence of an uprising dyke, located ca. 1–2 km b.s.l, 300–500 m east of the fracture system. Such a dyke presents a consistent Vp/Vs reduction at its top and is located at the eastern border of a broad high Vp intrusion, with lateral dimension of more than 6 km. This shallow high Vp intrusion, extending beneath Valle del Bove, is the main feature of the very shallow plumbing system of Mt. Etna and may play a significant role in the magma ascent. The intrusive mechanism invoked in this study is in agreement with all the available geophysical evidences, ground deformation and gravimetry, regarding the July–August 2001 eruption. In particular, ground deformation inversion indicates that the location of a final tensile crack, at depth between −1.5 to 0.7 km b.s.l., with an open dislocation of ca. 3.5 m and crossing the edifice between La Montagnola and Belvedere (see blow up area in Figure 1), in a roughly N-S direction, can explain the recorded deformation pattern during the period preceding the eruption onset [Bonaccorso et al., 2002].
 The results discussed here are certainly of relevance for future works aimed at modeling seismicity and stress field associated with the July–August 2001 flank eruption, and more generally for a better understanding of the physics of magma migration and storage.
 We thank P. Dawson and the anonymous reviewer for helpful comments. Financial support from INGV-Gruppo Nazionale per la Vulcanologia is acknowledged.