The understanding of shallow intrusive processes during 2002–2004, as well as the causes of the volcano-tectonic seismicity, has been improved at Mt. Etna by comparing the inversion results from GPS data with accurate 3D hypocentral locations. Our findings indicate that short periods of deflation (about six months) were followed by recharging phases after the end of both the 2001 and 2002–2003 flank eruptions. During the last recharging phase (June 2003–August 2004), modeling results and seismic observations suggest a composite mechanism of re-injection of magma into the rift-zones (S and NE), similar to that leading to the 2002–2003 flank eruption, which could have triggered the summit eruption started on September 7, 2004.
 In the last decade the increasing number of seismic stations installed at Mt. Etna (Figure 1) and the recent definition of 3D velocity models [e.g., Patanè et al., 2003], which more adequately represent the internal structure of the volcano, have allowed a good characterization of the variety of seismic signals recorded, as well as an accurate picture of their space-time evolution. Surface deformation measurements by EDM, tiltmeters, GPS and InSAR [Bonaccorso et al., 2002; Puglisi and Bonforte, 2004; Lundgren and Rosen, 2003; Aloisi et al., 2003] have been widely used to constrain the size, shape and evolution of magma intrusion during the latest eruptive activity at Mt. Etna (1989, 1991–1993, 1999, 2001, 2002–2003). The installation (November 2000) of a permanent GPS network (Figure 1) has improved the knowledge of the most recent dynamic processes acting on the volcano, as already demonstrated during the last two eruptive episodes [Bonaccorso et al., 2002; Aloisi et al., 2003].
 In this paper observations from the GPS and seismic permanent networks have been used to improve the understanding of magma transport through the shallow brittle crust (depth ≤ 10 km) and the causes of volcanic seismicity, before and after the October 2002–January 2003 flank eruption.
2. Eruptive Activity During 2001–2004
 After nearly 10 years without any major flank eruption, volcanic activity resumed at Mt. Etna on July 17, 2001, giving rise to the first of the two most striking flank eruptions on this volcano in recent times. The July–August 2001 eruption was heralded by several days of intense seismicity, ground deformation and fracturing [Patanè et al., 2002; Bonaccorso et al., 2002]. Fifteen months after the end of this eruptive episode, a new eruption started abruptly with only a few hours of premonitory seismicity accompanying the opening of eruptive fissures along a bi-radial direction (Figure 1). For both eruptions the analytical inversion of the ground deformation patterns clearly indicates the propagation of dike intrusions at shallow levels [Bonaccorso et al., 2002; Aloisi et al., 2003]. Dike emplacement concurs with the hypocentral distribution of the recorded seismicity and with field evidence of the eruptive fracture systems (Figure 1).
 Finally, after a period of weak volcanic activity at the summit craters on September 7, 2004 a summit eruption occurred along a WNW-ESE to NW-SE oriented fracture system at the base of the South East summit crater (Figure 1), without measurable short-period variations in either seismicity or ground deformation.
3. Ground Deformation and Seismicity
 The time series of a geodetic network and the slope distance variations between two GPS stations are useful and simple tools to visualize trends and highlight the deformation pattern of a volcanic edifice. Starting from June 2001, in Figure 2 we compare Mt. Etna's GPS time series (error estimation is of about 5–10 mm for the horizontal components and 20 mm for the vertical ones) with the daily occurrence of earthquakes and the related seismic strain release. The values of the seismic strain have been derived from the energy values, calculated using the duration magnitude of the earthquakes and Gutenberg and Richter's empirical relationship, log E = 11.8 + 1.5 Md [Gutenberg and Richter, 1956]. The selected GPS time series (planar coordinates of EMFN station, NICO–ESLN and NICO-EMGL baselines) have been chosen in order to investigate three distinct sectors of Mt. Etna's edifice, which in the recent past had generally shown a different geodynamic behaviour: 1) the eastern flank (EMFN planar coordinates, Figure 2a), characterized by a higher degree of faulting (Figure 1), where the major displacements are observed, 2) the western sector of Etna (slope distance NICO-EMGL, Figure 2b), showing a “quasi-elastic” response to magmatic intrusion and the periodic activation of the NE-SW fault system in the western flank [Bonaccorso et al., 2004] and 3) the southern flank (NICO–ESLN, Figure 2c), characterized by the presence of the SSE Rift zone and of the NNW-SSE fault system, where the emplacement of dikes occurs more frequently.
 The 2001 eruption presented a different intrusive mechanism with respect to the 2002–2003 eruption. This can be clearly recognized in Figures 2a and 2c by comparing the ground deformation patterns at the onset of each eruption. Conversely, if we compare the temporal patterns of both the seismicity and ground deformation after the 2001 and 2002–2003 flank eruptions, some common characteristics can also be found. Albeit with a different intensity, we can suggest that deflation phases (Figure 2b) of similar duration (about six months), accompanied by a decreasing rate of seismicity (Figure 2d), continue through recharging phases (grey shadow areas in Figures 2a and 2c). In particular, starting from February 2002 and June 2003 a similar increase in the distance between NICO and EMGL (Figure 2b), a decrease in the distance between NICO and ESLN (Figure 2c) and a weak acceleration in the seismic strain release (Figure 2d) can be observed. The comparison of planar coordinates recorded at EMFN station (Figure 2a), instead reveals a greater displacement of the eastern sector after the 2002–2003 eruption.
 In order to investigate the space-time evolution of seismicity, we performed 3D hypocentral locations of earthquakes between October 26, 2003 and August 2004 and compared the results with those of Gambino et al. , obtained for the pre-eruptive 2002–2003 period. A total of about 600 earthquakes, with Md ≥ 2.0, have been re-located by using the SIMULPS14 code [from Thurber, 1993] and the 3D velocity model of Patanè et al. . For our analysis, we selected hypocentral locations (Figures 3b and 3c) with RMS residuals and errors in the hypocentral parameters less than 0.15 s and 0.5 km, respectively. During the pre-eruptive 2002–2003 period, the overall pattern of seismicity mainly affected the southern and eastern flanks of the edifice and ruptures occurred along ca. ENE-WSW and NW faults (ENEF and NWF in Figure 3a [Gambino et al., 2004]). During the 2002–2003 eruptive period (Figure 3b), if we exclude the seismicity related to the intrusive mechanisms, the location of earthquakes is mainly restricted to the eastern part of ENEF, to the NWF and to the Pernicana fault (PF in Figure 3). This behaviour emphasize (Figure 3c) between the end of the 2002–2003 eruption and the onset of the September 7, 2004 summit eruption.
4. Pre- and Post-Eruptive 2002–2003 Modelling of GPS Data
 We inverted ground deformation GPS data collected by permanent stations (Figure 1) during the two recharging phases (February–October 2002 and June 2003–August 2004). The modelling of the deflation phases has been reserved for further studies. In this study we preferred to invert data measured only by the permanent network. This because GPS data from annual discrete surveys could show a possible superimposition of different deformation phases (i.e. post-eruptive deflation and pre-eruptive inflation). We have used a procedure based on two main steps: 1) the modelling of a point-source of pressure under the south-western upper flank using the Mogi  formulae; 2) the modelling of the residual displacements using a generalized least square inversion approach and the Okada  rectangular dislocation model. The application of Mogi's model, during the pre- and post 2002–2003 eruption, is justified if we consider: i) the almost radial orientation of the displacement vectors in the western flank (Figures 3a and 3c), ii) the elastic behaviour shown by this flank (baseline NICO-EMGL in Figure 2) in response to inflation and deflation processes and iii) the recent findings of a high attenuative volume located south to south-west of the central craters at 0–3 km depth [De Gori et al., 2005]. Instead, the large measured displacements, trending SE, in the eastern sector and the seismicity mainly affecting this flank of the edifice, may also reflect a combination of: i) stress redistribution along Etnean local and regional structures, in response to magma-induced edifice inflation and ii) stress induced by other magmatic deformative sources, such as intrusions of magma in the form of dikes along the S and NE rifts. In the inversion procedure we considered the reference surface (z = 0) at 1500 m and assumed that measurement errors are Gaussian. The goodness of fit has been verified by the χ2 test. It is noteworthy that even though GPS stations used to constrain the sources provide sufficient constraints, some parameters have been fixed (Table 1) in order to decrease the number of unknowns. However, further tests on the best fitting models have also been performed in order to avoid problems of local minima during the inversion. For each best solution achieved (Table 1) a trial-and-error procedure with a grid-search technique in the space of the parameters of the tabular dislocation models was adopted together with a wide range of parameters variability. Then we calculated the deformation effects of all possible solutions in this grid and verified the results by the χ2 test. Considering all the solutions that satisfy the χ2 test, within the significance level of 5% and for the considered degrees of freedom, we calculated that the proposed models are affected by an uncertainty on the source parameters within 2σ where σ is the computed analytic uncertainty (Table 1). Because this uncertainty is however greater than the uncertainty of the individual source elements (Table 1) the remaining misfit is related to the simplicity of the adopted Okada model. Therefore the use of simple models derived from one or more acting magmatic sources of deformation and/or structures under the boundary conditions of a homogeneous, isotropic and elastic half-space, cannot always explain all the measured deformations. This is the case of the large deformative effects measured at the summit stations ETDF and EPDN (Figure 1), affected by the eruptive fracture fields during the 2002–2003 eruptions. For this reason these stations have not been included in the inversions. To explain the large deformative effects measured on the eastern flank of the edifice between 1995 and 1998, Puglisi and Bonforte  invoked the presence of two wide sub-horizontal planes (sliding planes). Lundgren and Rosen , using InSAR data, model two faults dipping west and east under the summit area of Etna to explain the bi-lobate structure of the interferograms. In this study, in order to fit the deformative effects in the eastern sector, we considered a fault plane located in the upper eastern flank, striking ENE-WSW and with a dip of 30° in SE direction (ENEF in Figure 3). Our choice is supported by the distribution of earthquakes along this fault during the investigated period (Figures 3a and 3c). This fault may belong to the well known Messina-Fiumefreddo alignment or to the ca. NE-SW fault system controlling the tectonic evolution of the northern margin of the Hyblean Plateau [Torelli et al., 1998]. NE-SW and ENE-SSW faults can be clearly recognized at south of the volcano along the Catania foredeep. Strain release along these active faults beneath the Mt. Etna was also confirmed by the earthquakes distribution between 1988 and 2001 [Bonaccorso et al., 2004; Patanè et al., 2004].
 Although we are sure that the use of the ENE-WSW fault plane is a further simplification of the more complex structural framework in the eastern flank, this allows us to set a physical constraint on an otherwise undefined displacement surface.
 For the pre-eruptive period of the 2002–2003 eruption, the observed deformation pattern can be well explained by a Mogi-type source ca. 1 km beneath the southwest flank that inflated and by an uprising dike in the upper southern flank (Figure 3a). The dike, ca. NNW-SSE oriented, is located roughly south of the summit craters and extends vertically for 2.9 km, is 3.0 km long with an estimated opening of 1.0 m. It is noteworthy that a dike intrusion along the NNW-SSE fault system (S rift) occurred during both the 2001 [Bonaccorso et al., 2002] and 2002–2003 eruptions (Figure 3b) [Aloisi et al., 2003]. Also for the recharge phase started in June 2003 we need to consider a composite mechanism (Figure 3c), that includes a pressure point-source located in the south-western flank (ca. 1 km depth), a vertical NNW-SSE oriented uprising dike in the upper southern flank and a radial intrusion in the NE sector. The location of these two dikes is very similar to those of the 2002–2003 intrusion. The S dike extends vertically 2.0 km, is 3.2 km long with an estimated opening of 0.3 m, whereas the NE dike is elongated 5.2 km from the summit craters, in the NE direction and extends vertically for 5.0 km with an estimated opening of 0.6 m. The observed and calculated horizontal displacement vectors are reported in Figure 3 together with the position and geometry of the modelled dikes. Table 1 reports the parameters modelled during the inversion process and the relative uncertainties. In this table the model of the 2002–2003 intrusive mechanism considering the ENEF fault is also proposed.
5. Conclusive Remarks
 Magma intrusion in the shallow crust (or in the upper part of the volcanic pile) may be followed by both lateral and summit eruptions, after several years/months of internal replenishment. A repeatedly observed feature at Mt. Etna is that the summit eruptions are not generally preceded by significant variations in the pattern of deformation and volcano-tectonic seismicity [Patanè et al., 2004]. On the other hand, lateral eruptions, such as the two most recent 2001 and 2002–2003 eruptions, are forerun by a few days/hours of seismic crisis just before the eruptive fissures open.
 Ground deformation data acquired by the GPS permanent network show that active deformation of Mt. Etna has been continuous, and that after both the end of the 2001 and 2002–2003 eruptions recharging phases followed brief periods of deflations. Our findings suggest that during these recharge phases a replenishment of the shallow reservoir [De Gori et al., 2005] located south to south-west of the central craters occurs. Moreover, a progressive northward shift of the deformation source seems to be testified by the location of the modelled dikes before (Figure 3a), during (Figure 3b) and after (Figure 3c) the 2002–2003 eruption and by the strong displacements and seismicity recorded on the eastern and northern flank, along the Pernicana fault. Presumably a common deeper source [Patanè et al., 2003] fed both the region of shallow storage in the south-western flank and the dikes.
 Finally, the last summit eruption started on September 7, 2004 (still in progress) could be inserted in the more general recharging phase affecting the volcano since June 2003, which is still ongoing, and then triggered by a small pressure increase in the central feeding system.
 This work was supported by grants of the Gruppo Nazionale per la Vulcanologia-INGV. Comments and critical suggestions by anonymous reviewers helped to improve the manuscript.