We present results from the first crustal seismic tomography for the southern Tyrrhenian area, which includes ocean bottom seismometer (OBS) data and a bathymetry correction. This area comprises Mt. Etna, the Aeolian Islands, and many volcanic seamounts, including the Marsili Seamount. The seismicity distribution in the area depends on the complex interaction between tectonics and volcanism. The 3-D velocity model presented in this study is obtained by the inversion of P wave arrival times from crustal earthquakes. We integrate travel time data recorded by an OBS network (Tyrrhenian Deep Sea Experiment), the SN-1 seafloor observatory, and the land network. Our model shows a high correlation between the P wave anomaly distribution and seismic and volcanic structures. Two main low-velocity anomalies underlie the central Aeolian Islands and Mt. Etna. The two volumes, which are related to the well-known active volcanism, are separated and located at different depths. This finding, in agreement with structural, petrography, and GPS data from literature, confirms the independence of the two systems. The strongest negative anomaly is found below Mt. Etna at the base of the crust, and we associate it with the deep feeding system of the volcano. We infer that most of the seismicity is generated in brittle rock volumes that are affected by the action of hot fluids under high pressure due to the active volcanism in the area. Lateral changes of velocity are related to a transition from the western to the central Aeolian Islands and to the passage from continental crust to the Tyrrhenian oceanic uppermost mantle.
 The southern Tyrrhenian Sea is characterized by the transition from an oceanic crust domain found at the Marsili Basin (MB) (Figure 1a) [Marani and Trua, 2002] to a continental crust at the margin offshore Sicily and Calabria [Pepe et al., 2005]. The complex geodynamic framework of this area is governed by sinking and rollback of the Ionian subducting slab [Malinverno and Ryan, 1986; Kasten et al., 1988; Doglioni, 1991; Argnani, 2009]. In its final stage, this process led to the formation of two back-arc sub-basins within the Tyrrhenian Sea: the Vavilov (4.3–3.6 Ma) and the Marsili (<2 Ma) [Marani and Trua, 2002, and references therein]. In only a few hundreds of kilometers, we find oceanic (Ionian plate and Tyrrhenian Basin) and continental (African) lithospheres interacting with the asthenosphere, within a subduction system where the Ionian slab is one of the narrowest subduction zones worldwide. In this uncommon active arc/back-arc system, the expected island arc basalts (IABs; Aeolian Arc and seamounts including Marsili) coexist with ocean island basalts (OIBs; Ustica Island and Mt. Etna), implying complex geodynamic evolution and magma generation [Trua et al., 2004]. There are also significant variations of asthenosphere topography, from 10–30 km to about 150 km depth, and of crustal thickness, going from ∼10 km of the oceanic crust of the Marsili Basin [Marani and Trua, 2002, and references therein] to the ∼30 km thick continental crust of northern Sicily [Langer et al., 2007, and references therein]. There are also very strong topography changes, from ∼−3500 m of the Marsili Basin to the peak of Mt. Etna at ∼3350 m.
 There are tomography models which image the crust of the area at regional [e.g., Di Stefano et al., 1999] or at very local scale [e.g., Scarfì et al., 2009]. Studies at the scale of the present work were published by De Luca et al. , Neri et al. , and Barberi et al. , but the lack of instrumental coverage at sea puts serious limitations on these models. In particular, the regional models can give an image down to the uppermost mantle but with a resolution which is not high enough to resolve finer structures. On the other hand, the local models have high resolution and can resolve finer structures but are limited in depth and lateral extension due to station-event geometry. Because of these limitations, the local tomographies cannot resolve the passages between the continental and oceanic domains and between Mt. Etna and the Aeolian Islands. High-resolution tomographies of Mt. Etna have resolution at most down to the middle crust and cannot image the crust-mantle transition zone [e.g., Patanè et al., 2006; Villaseñor et al., 1998]. The only published tomography study based on data recorded by marine stations is by Montuori et al. , but it is at a much wider scale and is concentrated on the upper mantle of the southern Tyrrhenian. The ocean bottom seismometer (OBS) deployed during the long-term Tyrrhenian Deep Sea Experiment (TYDE EC project) [Dahm et al., 2002] and the SN-1 seafloor observatory [Monna et al., 2005; Sgroi et al., 2007] increased station coverage north (mostly) and east of Sicily, allowing the detection of previously unknown seismicity at sea. Thanks to these new long-term marine experiments, we could cover the gap between the available local and regional tomography models and obtain previously unknown information on this area. The objective of this work is to resolve the P wave velocity structure of the crust, the crust-mantle transition, and to highlight the relation of imaged velocity anomalies to seismicity distribution, volcanism, and tectonics.
2. Seismotectonic and Volcanic Setting
 The southern Tyrrhenian Sea and the Sicilian region are very heterogeneous from a geological, tectonic, and volcanological point of view. The southern Tyrrhenian seismotectonic region is E-W striking and undergoing compression (Figure 1a). Northern Sicily, which comprises the Madonie-Nebrodi and the Peloritani mountains, is subject to tension.
 Following De Astis et al. , we define three distinct domains with different types of seismicity and stress regime in the southern Tyrrhenian area (Figure 1a): southern domain (SD), western domain (WD), and eastern domain (ED). The central Aeolian Islands mark the transition between the WD and ED.
 The Southern Domain includes the Messina Strait (MS), northern Sicily, and southern Calabria. Seismic events are mainly confined to crustal depths. Fault plane solutions of lower magnitude earthquakes (M ≤ 4) are in agreement with a NW-SE extensional stress field [Caccamo et al., 1996; Frepoli and Amato, 2000]. The SD also includes Mt. Etna, the biggest subaerial volcano in Europe, in which volcano-tectonic seismicity is mostly concentrated in the upper 12 km [Murru et al., 2007] but is also found in the lower crust (20–30 km depth), showing typical main shock-aftershock sequences [Castellano et al., 1997]. A strike-slip/extensional regime characterizes both Mt. Etna and the central Aeolian Islands [Pondrelli et al., 2004].
 The Western Domain includes Filicudi and Alicudi Aeolian Islands, Ustica Island, and western seamounts. Earthquakes cluster along the main fault systems at crustal depths: the WNW-ESE striking Sisifo-Alicudi system (SA, Figure 1a). WD is bounded to the east by Vulcano, Lipari, and Salina (the central Aeolian Islands, CAI). Several authors [e.g., Lanzafame and Bousquet, 1997; Billi et al., 2006] have proposed the existence of a NNW-SSE strike-slip fault, the Vulcano-Tindari-Letojanni (TL) connecting the island of Vulcano with mainland Sicily, although its prosecution at sea has been disputed [Argnani et al., 2007]. It is commonly agreed upon that TL, starting from northeast Sicily, continues to the south and is the prosecution of the Malta Escarpment in the Ionian Sea (ME) [Neri et al., 1996; De Luca et al., 1997]. In the WD, stress distribution deduced from seismic tensor inversion is consistent with a compressive stress regime trending NW-SE [Pondrelli et al., 2006], a conclusion also confirmed by GPS data [Serpelloni et al., 2007]. A strong transition is observed across the WD with GPS data as a rapid increase in velocity and change of direction (from NW to N) of crustal deformation, going from Alicudi toward the central Aeolian Islands [Argnani et al., 2007].
 The Eastern Domain includes the eastern Aeolian Islands, Stromboli and Panarea, and the eastern seamounts (ED in Figure 1a). ED is bordered to the east by southern Calabria and to the west by the central Aeolian Islands. The ED includes a lower number of crustal earthquakes (<30 km) with respect to the WD [Falsaperla et al., 1999] and is affected mostly by intermediate and deep seismicity. An abrupt change in the focal mechanism style has been observed across the ED, from thrust faulting along the Calabrian margin to extensional and strike-slip faulting in the central Aeolian Islands [Pondrelli et al., 2004].
 There is evidence for active subduction processes of the Ionian oceanic lithosphere of Mesozoic age [e.g., Catalano et al., 2000]. The close proximity of IAB-type to OIB-type points to multiple mantle magma sources [Peccerillo, 2001]. OIB magmas, found at Mt. Etna and Ustica Island, are uncommon in active volcanic arc systems [Trua et al., 2004, and references therein]. Volcanic activity propagated from west to east along the Aeolian Islands and is now affecting the central-eastern Aeolian Islands. The western islands have not been volcanically active in the past 30–40 ka [Tranne et al., 2002; Lucchi et al., 2008]. De Astis et al.  explained the end of volcanic activity in the WD as being caused by a compressive local stress regime on cracks. In this model, a clamping effect closing the cracks prevents uprising of magmas for the western Aeolian Islands.
 Mt. Etna is positioned in the forearc of the Calabrian arc subduction system, although its magma is not subduction related. Mt. Etna's activity is well known, but despite the intense and continuous monitoring, its origin and anomalous position within the geodynamic context of the area are still controversial [e.g., Doglioni et al., 2001]. To explain Mt. Etna's relation to the subduction zone, Gvirtzman and Nur  proposed that the melting below Mt. Etna is a consequence of the suction of African asthenospheric material through a lateral slab window caused by Ionian slab rollback. The hypothesis of a slab window related to Mt. Etna has been strengthened by recent tomographic models [e.g., Montuori et al., 2007; Monna and Dahm, 2009]. Focal mechanisms computed for earthquakes located at Mt. Etna for depths greater than 10 km (down to ∼30 km) indicate a complex stress regime [Pondrelli et al., 2004; Neri et al., 2005]. There is extension in the eastern and southeastern part of the volcano [Monaco et al., 1997] and compression in the western and central sector [Barberi et al., 2000; Patanè and Privitera, 2001].
 In this work, we used earthquake data recorded by marine (TYDE and SN-1) and land (Italian Seismic Catalogue (ISC)) networks (shown in Figure 1b). The TYDE network was deployed around the Aeolian Islands down to about 3.5 km sea depth (December 2000 to May 2001). All 14 TYDE marine stations housed a hydrophone and six were also equipped with a three-component broadband seismometer (OBS). The OBS/ocean bottom hydrophone (OBH) systems were developed by two German institutions: the Institute of Geophysics at Hamburg University and the Research Center for Marine Geoscience of Kiel University. The sampling rate of both OBS and OBH was 50 Hz [Dahm et al., 2002]. The 20-bit digitizer recorded at a sampling rate of 50 Hz. OBS and OBH clocks were synchronized with GPS clock signal before deployment and just after its recovery. The time delays observed at each single station did not exceed ±0.6 s during the 6 month campaign. A linear time correction for this time delay was applied [Dahm et al., 2002]. A recent analysis based on ambient noise (T. Dahm, personal communication, 2013) confirmed the linear drift. The SN-1 multidisciplinary seafloor observatory was installed in stand-alone mode offshore Catania (Sicily), about 25 km from Mt. Etna, in the Ionian Sea at 2.1 km sea depth (October 2002 to February 2003). SN-1 was equipped with a set of geophysical and oceanographic instruments (gravity meter, hydrophone, conductivity and temperature versus depth, 3-C single-point current meter, and several status sensors) and, in particular, with a three-component broadband seismometer (Guralp CMG-1T, 0.0027–50 Hz bandwidth and 100 Hz sampling rate), that was synchronized by a high-precision rubidium clock (drift ∼0.5 ms/day; see Favali et al.  for details). The Italian Seismic Network (ISN) consists of Kinemetrix S-13 short-period single-component, Lennartz 5s extended band and Trillium 40s broadband three-component stations [Schorlemmer et al., 2010]. Land data span the years 1988 to 2007, while OBS data are from December 2000 to May 2001 (TYDE array), October 2002 to February 2003, and 2005 to 2006 (SN-1 observatory).
 Our data set combines arrival time data from the ISC [Castello et al., 2005] and the Italian Seismic Bulletin (ISB, ) managed by Istituto Nazionale di Geofisica e Vulcanologia, with those we picked on waveforms recorded by land and marine stations. Our manual picking process was performed following the same procedure adopted by the analysts of the ISC and ISB. In the picking procedure, a weight is assigned to each pick to indicate its accuracy. Picks with errors greater than 2 s (weight 4) are excluded. In total, we picked 3241 P and 1439 S arrivals from 388 events recorded both by marine and land stations. P phases were determined on the vertical component for land and marine stations and on the OBH when the first arrival was visible, while the S phases were picked on the horizontal components. For the single-component land stations, the S phase was picked on the vertical component. For land waveforms, the P phase picking procedure was performed without the application of a filter, while for the S phase, a band-pass filter of 2 to 12 Hz was applied. For marine waveforms, the same band-pass filter was always applied for both P and S. Figure 2 shows recordings and picks for the same earthquake recorded by land and marine stations. Higher frequency energy is more visible on the broadband OBS recordings with respect to the short-period land sensors, for both P and S phases. In general, marine seismograms have a more complex form due to multiples coming from the water surface and the sediment layer. The interval 2–12 Hz for the band-pass filter was found to be the optimal frequency range for discriminating the P and S phases of local events. To increase ray coverage, we also added arrival times from ISC and ISB catalogs of crustal earthquakes which occurred between January 1988 and December 2007. The final data set comprised 21,860 P phases from 2193 local crustal earthquakes (1.0 ≤ M ≤ 5.6). We relocated the events adding the available S phases to have a more accurate estimation of depth.
4. Relocation of Seismic Events and Calculation of Initial Tomography Model
 The location of offshore crustal seismicity in the southern Tyrrhenian Sea continues to be affected by large errors [Sgroi et al., 2006] in spite of the improvement of the ISN. Relocated seismicity in the INGV catalog has reasonably small azimuthal gaps when events occur on land (<180°) and larger gaps for events at sea (<240°). Most of the hypocenter locations have RMS < 1 s, and for deep events horizontal epicentral errors are <6–10 km and vertical errors <10 km [Chiarabba et al., 2005]. To overcome these limitations, we used earthquake data recorded both offshore and onshore by the TYDE OBS network, the SN-1 seafloor observatory, and the ISN. Relocation was performed using the HYPOELLIPSE code [Lahr, 1989] by considering three different 1-D velocity models which take into account the strong lateral heterogeneities of the study area [Sgroi et al., 2006, and references therein]. The HYPOELLIPSE software needs a single value for Vp/Vs; we started with a standard Poisson solid value of 1.73. Two values of Vp/Vs for continental and oceanic sectors were derived from Wadati diagrams. For the continental crust we calculated Vp/Vs = 1.74, while for the oceanic crust Vp/Vs = 1.78. We found that our hypocentral locations were not very sensitive to such changes in Vp/Vs. Earthquakes selected for this initial data set were recorded at a minimum of four stations (with a total number of arrival times not less than seven). More than 250 small earthquakes (∼12% of the total event number) were undetected by land stations and located only because of the marine stations. The final relocated data set used for the minimum 1-D model calculation (see below) was composed by a great majority of events having GAP ≤ 200° (∼80%) and rms ≤ 0.6 s (∼97%). The selection for the 3-D inversion was more stringent, and only events with at least five P phases were considered.
 The 3-D inversion procedure we apply requires a reliable initial 1-D model. Taking into account the above-mentioned models used for the preliminary event location, we used the software VELEST by Kissling et al.  to calculate a minimum 1-D model. After many trials, we decided to divide the data set in two subsets: subset A with recordings from land stations only, and subset B with events recorded also by marine stations (Table 1). Subset A was composed by earthquake recordings with at least eight P phases and GAP ≤ 180°, while the requirements for subset B were more relaxed for the OBS recordings with the GAP ≤ 250° and number of P phases ≥ 5. Subset A consisted of 650 events with 7579 P phases, and subset B of 650 events with 7543 P phases. After five iterations, rms reduction was from 0.98 to 0.46 s (∼53%) for subset A and from 1.24 to 0.56 s (∼55%) for subset B. After several VELEST runs, we obtained two minimum 1-D models for the two subsets (Figure 3a) and then calculated a single model by combining the two models (Figure 3b). Thanks to this analysis, and the inclusion of OBS data, we were able to identify two seismic discontinuities at different depths: one for the marine area and one for the land area, at about 23 and 31 km, respectively. These two discontinuities were persistent features in the solution models of the VELEST procedure, so we consider them as “robust.” The existence of two Mohos (onshore and offshore) and their depths are confirmed by the Moho model of Grad et al. , which is based on gravity and seismic data, and also by the model of Di Stefano et al.  (and references therein), based on controlled-source seismology data and receiver function analysis. The two seismic discontinuities are visible as steps found at 23 and 31 km depth in the graphic representation of the P wave 1-D starting model used for the linearized inversion process (Figure 3b). This starting model, which should best represent the average properties of the study volume, is composed by 10 layers and includes the robust features which were identified with the VELEST procedure. From this starting 1-D model, we constructed a 3-D plane-layered model, with 4 km thick layers (the top of the model is at 4 km above sea level). We further modified the layer between 0 and 4 km depth to account for the strong bathymetric variations (down to ∼4000 m b.s.l; Figure 4a) and the presence of water by including bathymetric data [GLOBE Task Team et al., 1999]. After this correction, the velocity varies horizontally in the model within the 0–4 km depth layer (Figure 4a). The effect of this correction on the direct part of the tomography procedure is twofold: (1) the travel time of each wave is corrected by an amount that depends on the effect of bathymetry on the wave and (2) the wave paths are also corrected (i.e., seismic P waves that would go through parts of the model occupied by water in the plane-layered model now do not go through water). The effect of bathymetry on travel time is quite strong for stations near the coast and for the OBS (up to ∼0.5 s). Figure 4b shows how bathymetry affects theoretical travel time calculation for a sample group of stations with respect to the reference plane-layered starting model. Each value on the vertical axis of Figure 4b equals the difference, for a station-event pair, between the travel time calculated in the starting model which includes bathymetry and the travel time calculated in the plane-layered model. On the horizontal axis is the event number. More information on the bathymetry correction can be found in Appendix Appendix.
Table 1. Data Sets Used to Calculate the Two Minimum 1-D Models With VELESTa
Number of Events
The two models mainly represent average velocities of the continental area (A) and of the oceanic area (B). The model used as starting model for the tomography inversion was derived from these two models.
Land + Marine
5. P Wave 3-D Model Calculation and Description
 The calculation of the Vp 3-D model was performed using the code FDTOMO of Benz et al.  starting from the initial model, which includes the bathymetry correction previously described. FDTOMO uses a finite difference scheme for travel time calculation [Podvin and Lecompte, 1991] and the least squares QR decomposition (LSQR routine by Paige and Saunders ) for simultaneous inversion of velocity structure and hypocenters. After performing synthetic tests and real data inversions using different parameterizations, we determined the best cell size for the inversion to be 10 × 10 × 4 km3 (larger in the horizontal directions). The investigated volume is included in ∼400 × 400 × 44 km3. The top of the model is at 4 km above sea level. Starting from the relocated data set, the total number of inverted P phase travel times was ∼18,000, with an rms reduction of ∼55% after three iterations.
 Checkerboard synthetic tests helped us understand which parts of the study area were well resolved, where there could be artifacts, and how much of the anomaly amplitude could be recovered. The input model is composed by alternating high-velocity and low-velocity anomalies, with a maximum anomaly amplitude perturbation of ±10%. The output model shows mostly a good anomaly pattern reconstruction at all depths, especially in the central part of the model (Figures 5a–5f). In general, the anomalies' amplitudes are underestimated, except in parts of the model that are strongly sampled by seismic rays, where amplitude is recovered up to 8% (white and black cells). Underestimation of the anomalies comes from the necessary regularization in the inversion procedure. In the lower part of the model (Figure 5f), strong smearing with a NE-SW direction is present in the Tyrrhenian sector of the ED. This is due to the event-station geometry and lack of coverage by marine stations in that part of the Tyrrhenian. The interpreted anomalies, calculated from the real data inversion, are completely within the part of the model where the checkerboard pattern is well reconstructed and where cells are traversed by at least 10 rays. This well-resolved volume is included by a contour in the 3-D layers and profiles (Figures 6 and 7). This is also confirmed by ray-tracing analysis (supporting information Figures S1 and S2).1
 Below we describe the main anomalies found in the 3-D model (Figure 6). The calculated 3-D model shows velocity anomalies, which appear as positive (blue, high velocity) and negative (red, low velocity) anomalies. Starting from the top two layers (0–8 km depth, Figures 6a and 6b), we observe a high-velocity anomaly (H1) around the Messina Strait and a transition from a low-velocity anomaly (L1) in the central Aeolian sector to positive velocity values (H2) in the western Aeolian sector. Both H1 and L1 extend from 0 to 16 km depth (Figures 6a–6d), attaining maximum values between 8 and 12 km depth. H2 is visible in the western Aeolian sector and around Ustica Island. An E-W elongated low-velocity anomaly (L2) at 8–20 km follows the structural high of the Nebrodi, partially replacing H1 (Figures 6c–6e). From 20 km down to the bottom of the model, the anomaly distribution is more homogeneous (Figures 6e and 6f). Anomaly L2 broadens and gains intensity at 24–28 km depth beneath northern Sicily, stretching northeast into southern Calabria. L2 attains values down to −11% of the starting value (black area in Figure 6f) in the northern part of Mt. Etna. The minimum velocity subvolume of L2 (L2E, Figure 6f) has an areal extension of about 40 km longitude × 30 km latitude. Below 16 km depth, a diffuse high-velocity zone (OL in Figure 6f) covers the Tyrrhenian sector, including Ustica and the Aeolian Islands.
 The relevant features of the model are also evident in the profiles (Figure 7): anomalies L1 and H2 are confined to shallower depths than anomalies L2 and OL (profiles EE′ and FF′); L2E is visible below Mt. Etna (profiles DD′ and EE′) and reaches the base of the crust at 28 km depth. H1, under the Messina Strait area, overlies L2 (profiles BB′ and CC′). In general, the profiles show that the Moho isobaths (from independent data) mark the lateral passage from L2 to OL and agree well with the geometry of OL. Finally, it is noteworthy that the seismicity relocated in the 3-D model is concentrated along the lateral transition zones between low-velocity and high-velocity anomalies.
 The distribution of lateral velocity contrasts correlates well with geological features that are found in the domains previously introduced. The seismicity relocated in the 3-D model follows the main tectonic structures shown in Figure 1a: the Sisifo-Alicudi (SA) system from west to east, the Tindari-Letojanni (TL) from north-northwest to south-southeast, and the Peloritani-Nebrodi-Madonie tectonic high in northern Sicily.
 In the SD, the high-velocity volume H1, which extends about 100 × 50 km2 horizontally and 16 km in depth, is quite visible around the Messina Strait. H1 can be explained by the presence of crystalline basement nappes in southern Calabria and the Peloritani in northeast Sicily [Bonardi et al., 2001]. Seismicity concentrates in the part of H1 found below southern Calabria and the northeastern tip of Sicily (profile BB′). This area has a high seismic hazard potential, as it has been affected by very strong historical earthquakes, such as the M = 7.2 1908 Messina earthquake [Boschi et al., 1997]. Still in the SD, visible from 8 to 12 km depth, there is the strong low-velocity zone L2 in central northern Sicily. L2 could well represent the thick crust underlying the tectonic highs of the Peloritani-Nebrodi-Madonie range in Sicily and continuing toward the Aspromonte Mountains in Calabria. L2 attains a minimum value of about 6.7 km/s (−11%) within subvolume L2E found below the northern sector of Mt. Etna near the base of the crust, at 24–28 km depth in our model (Figure 6f and profiles DD′ and EE′ in Figure 7). Although recent high resolution local earthquake tomography studies of Mt. Etna exclude the existence of a deep active magma chamber [e.g., Patanè et al., 2006], they image the upper-middle crustal structure and are only able to find a main high-velocity anomaly in the southeastern part of the volcano down to 12 km depth [e.g., Villaseñor et al., 1998]. These studies do not show low-velocity volumes that could be associated with the presence of melt accumulation. To have an image of the lower crust or the lithosphere-crust boundary below Mt. Etna, it is necessary to expand the resolved region. The tomography at regional scale by Di Stefano et al.  shows a low-Vp anomaly found at a depth of 38 km (uppermost mantle). We should keep in mind that this model has lateral resolution of 25–50 km and is formed by three layers at depths of 8, 22, and 38 km. Thanks to marine data, we were able to overcome the limits of previous tomography studies at this scale and identify L2E anomaly at the base of the crust below Mt. Etna. The location of the L2E anomaly agrees quite well with the low-velocity volume (suggested melt region) at a depth of ∼20 km imaged by Sharp et al.  with the use of deep sounding techniques (with shots performed at sea), whose study was based on seismic travel time and attenuation measurements. The large volumes of volatiles erupted suggest the existence of a large magma chamber [Allard et al., 1994]. On the other hand, isotopic studies imply short magma transition times, which exclude the existence of a magma chamber [Armienti et al., 1989; Albarkde, 1993]. Hirn et al.  reconcile this conflict by considering the chamber not as a large volume inside the crust but as a melted lens situated on the top of an upwelling mantle. We interpret the strong L2E anomaly as related to Mt. Etna's deep feeding system; in particular, we identify it with a melted lens on top of the mantle, in agreement with Hirn et al. . The existence and location of this strong low-velocity anomaly can also be related to the slab window below Mt. Etna proposed in the geodynamic model of Gvirtzman and Nur .
 Seismicity in the Mt. Etna area relocated in our 3-D model is found in volumes of rock that are adjacent to and overlie L2E (Figure 7, profiles CC′, DD′, and EE′). This seismicity probably originates from the complex interaction between tectonics and volcanism.
 In the WD, the high-velocity zone H2 found below the western Aeolian Arc and Ustica Island corresponds to a well-known tectonic uplift composed of volcanic rock [De Astis et al., 2003] and correlates well with the compressive stress regime [Pondrelli et al., 2006; Serpelloni et al., 2007]. The location of H2 also correlates with the topography of the structural high [Bortoluzzi et al., 2010]. High velocities are found already in correspondence of Alicudi, Filicudi, and Ustica at shallow depths (0–8 km; Figures 6a and 6b). High-velocity anomalies in the western Aeolian Arc very likely correspond to dykes, which are the result of the oldest volcanic activity in the Aeolian Arc. In fact, volcanism in the western Aeolian Islands has not been active in the past 30–40 ka [Tranne et al., 2002; Lucchi et al., 2008]. On the other hand, the L1 anomaly (0–16 km depth) found in the transition zone marking the passage between the WD and the ED agrees with the active volcanism of the central Aeolian Islands. Interaction of intrusions related to active volcanism can explain lateral heterogeneities [e.g., Kühn and Dahm, 2008]. Low velocities can be expected in this area where, due to volcanic activity, temperatures are high and rocks are fractured and rich in fluids. Unfortunately, due to incomplete ray coverage Stromboli and Panarea volcanoes (ED) are not within the resolved part of the model. The distribution of high-velocity and low-velocity anomalies along the Aeolian Arc agrees well with the space-time evolution of volcanic activity, which migrated from west to east. Seismicity below the central Aeolian Islands is found within the low velocity and surrounding rock volumes, which have intermediate velocity values (colored yellow in the maps and profiles). These rock volumes are brittle and they are likely weakened by injection of fluids which increases pore pressure in the rock [e.g., Špičák and Horálek, 2001] and reduces effective modules [e.g., Dahm and Becker, 1998].
 The two low-velocity volumes underlying the central Aeolian Islands and Mt. Etna appear separate in the model. The first is found within the crust and the other at the base of the crust (profiles DD′ and EE′ in Figure 7). This last finding points to the existence of two independent magmatic systems, in agreement with petrographic and geochemical data, which attribute IAB type (Island Arc Basalt) to the central Aeolian Islands and OIB type (Ocean Island Basalt) to Mt. Etna's magmatic products [e.g., Peccerillo, 2001].
 The broad high-velocity zone OL in the southern Tyrrhenian visible at 20 km depth and below (Figures 6e and 6f) can be explained by the presence of oceanic lithosphere. This anomaly distribution in depth and its arcuate geometry are in very good agreement with the Moho isobaths (Figures 6f and 7) and mark the passage from Sicilian continental crust to the Tyrrhenian oceanic uppermost mantle.
 We present a 3-D Vp model of the crust/crust-mantle boundary in the southern Tyrrhenian-northern Sicily area with the use for the first time of marine data. The 3-D inversion is made more accurate by the inclusion of bathymetry, which in this area has very strong variations (from ∼−3500 m of the Marsili Basin to the peak at ∼3350 m of Mt. Etna). Thanks to the use of marine data, the crust and the crust-mantle transition are modeled with higher resolution than previously published. The lateral variations of seismic velocity and the distribution of seismicity shown in our model are strongly correlated with the main geological features and volcanism present in the region. Three different domains can be identified in the study area on the basis of seismicity and stress regime. The domains are separated by the main tectonic structures: Sisifo-Alicudi and Tindari-Letojanni systems and Peloritani-Nebrodi-Madonie structural high. Below the tips of northern Sicily and southern Calabria, there is a high-velocity block (H1) which can be explained by the presence of crystalline basement nappes. This block is found in an area with very high hazard potential, where strong historical earthquakes and tsunamis took place. The thick and deep continental crust below the Peloritani-Nebrodi-Madonie is represented by the elongated, west-east, low-velocity anomaly (L2) in northern Sicily. We identify a subvolume of L2 with minimum velocity values (L2E, about 40 × 30 × 4 km3, horizontal × vertical), which agrees well with the low-velocity zone below Mt. Etna near the base of the crust proposed by Sharp et al. . We interpret this strong L2E Etna anomaly as related to Mt. Etna's deep feeding system; in particular, we identify it with a melted lens on top of the mantle, in agreement with Hirn et al. . Furthermore, the existence and location of this strong low-velocity anomaly could be related to the slab window below Mt. Etna proposed in the geodynamic model of Gvirtzman and Nur . As far as we know, we have a first image of what we interpret to be the deep feeding system of Mt. Etna.
 In the southern Tyrrhenian, the transition from the inactive volcanism of the western sector to the active volcanism of the central sector of the Aeolian Islands is displayed in our model by strong lateral velocity variations in the upper crust. The evolution of volcanic activity agrees with the lateral variation of seismic velocity along the Aeolian Arc. Seismicity below the central Aeolian Islands is found within brittle rock volumes (having low or average velocities). These rock volumes are weakened by injection of fluids, related to the volcanic activity, which increases pore pressure in the rock and reduces effective modules.
 The two main low-velocity volumes which are related to the active volcanism of the central Aeolian Islands and Mt. Etna appear to be separated. The one below the central Aeolian Islands is found above 16 km, while the one related to Mt. Etna is quite visible at greater depths starting from 24 km down to 28 km. Furthermore, the two low-velocity anomalies are separated by a volume of rock having mostly high-velocity values (H1). Our observations agree with studies on magmatic products which point to different sources (IAB type for the central Aeolian Islands and OIB type for Etna) and therefore suggest the existence of two independent magmatic systems.
 Another interesting aspect of our model is that we are able to image the continent-ocean transition in the Tyrrhenian Sea. Tyrrhenian oceanic uppermost mantle north of Sicily is found at depths > 20 km. The depths and patterns of the continental crust-oceanic lithosphere transition reconstructed in our model are in very good agreement with Moho isobaths derived from independent data.
 The model which is used as starting model for the tomography inversion (FDTOMO code) is usually a plane-layered model, in which each layer has a constant velocity value defined by the chosen 1-D starting model. In this study case, given the very strong bathymetry variations, we decided to correct for this effect in the following way. The “corrected” cells are part of the 0–4 km depth layer. The bathymetry was resampled on the tomography grid and given the average value it has in each cell (Figure A1). Then, the velocity attributed to each cell is the average between the velocity of the 1-D model value and the velocity of water (1.5 km/s) weighted by the relative height of bathymetry and water in that cell:
where is layer thickness (in km), Hbat is the average of the absolute bathymetry value in the cell, V1-D is the velocity value in the layer 0–4 km of the plane-layered starting model, and Vwater is the velocity of sound in water. The effect of this correction is to have more realistic wave paths. In fact, if bathymetry is not corrected for, several first arrival waves calculated in the uncorrected (plane layered) model would be, in reality, traveling in water and therefore would not be the first arrivals. Figure 4b shows that this correction matters, especially for marine and island stations. The amount of the correction is given by the difference between travel times calculated, for each station-event pair, for the two models (i.e., the plane-layered model and the model with the bathymetry correction, Figure A2). The time differences between for two ray paths, associated with the same station-event pair, calculated in the two models can be up to ∼0.5 s.
 We thank the partners of TYDE project: INGV, ISMAR-CNR, IfM-GEOMAR, Hamburg University, and EC for its support. We thank Francesco Frugoni, Andrea Argnani, Paolo Favali, Laura Beranzoli, and Dan-Thanh Ton-That for precious comments and discussions. Layers and profiles were produced using GMT [Wessel and Smith, 1991]. We thank the Editor, James Tyburczy, and two anonymous reviewers for their useful comments, corrections, and encouragement.