Late on the night of July 17, 2001, a lateral eruption started from the slopes of Mt. Etna. A 7 km long field of ground fractures opened between 13 and 20 July. The eruption ended on August 9, 2001 after emitting a lava volume of approximately 48 · 106 m3. A strong seismic swarm earthquake was recorded between July 12 and 17. The evolution leading up to the July crisis was monitored through continuous tilt and GPS measurements, which constrained the intrusion preceding the eruption in time, and inferred the position and geometry of the uprising dike. We modeled the marked ground deformation changes recorded in the days before the eruption onset. The result shows that a tensile crack with an opening dislocation of ca. 3 m, crossing the volcano edifice slightly southeast of the crater area, can explain the recorded deformation pattern. The location of the modeled tensile source fits the zone of the seismic swarm occurring during the magma uprising. The ground deformation pattern associated with the final uprising and its modeling suggest a very fast dike emplacement which appears different, both in terms of rapidity (only a few days) and source position, with respect to the sources modeled for the other lateral eruptions in the last twenty years.
 In the last 30 years eruptive activity at Mt. Etna has occurred nearly annually. During the last decade, the last lateral eruption was in 1991–93, which represented the most important lateral eruption in the last three centuries both in terms of duration (472 days) and volume of lava erupted (ca. 235 · 106 m3). It was accompanied by an evident deflation which inferred a depressuring intermediate storage zone 2–3 km b.s.l. [Bonaccorso, 1996]. After the 1991–93 eruption, the volcano did not show its activity through lateral eruptions but, since 1995, through frequent summit phenomena such as strombolian explosions, lava overflows from summit craters, a ten month long summit eruption (February–November 1999) and more than a hundred spectacular lava fountains [GVN, 1998, 2000; La Delfa et al., 2001]. In particular, the positive gravity anomaly, the ground deformation inflation and the seismicity pattern all indicated that from 1994 a new intrusive process involved a crust volume at 3–5 km b.s.l. under the crater area [Budetta et al., 1999; Bonaccorso et al., 2000; Spampinato et al., 2000]. From the second half of 1996 to 1998 the recharge process of the volcanic system was particularly intense as evidenced by the continuous increase in the areal dilatation and the rise in the cumulative seismic strain release pattern. A progressive increase in the activity occurred during 1998–99. In particular, a strong seismic swarm occurring in the western flank in January 1998, was interpreted as a shear response to an intrusive episode inside the volcano edifice [Bonaccorso and Patanè, 2001]. These mechanisms, analogous to the ones preceding the 1991–93 eruption, led to the 1999 summit eruption. During 2000, 66 lava fountain events with small lava flows occurred, while during the first half of the 2001 we witnessed again a lava flow from the summit. The high seismicity level, recorded under the summit area in the first months of 2001, culminated in April with an intense seismic swarm in the western flank. The last lateral July–August 2001 eruption occurred after this intense and varied summit activity about ten years after the previous 1991–93 lateral eruption. The eruption was heralded by an intense seismic swarm (more than 2600 events, Mmax 3.9) starting on the late night of July, 12 [INGV-CT, 2001; Patanè et al., 2002] and continued without any fall in energy release till the late night of July 17 when the lava poured out from the lower fissure in the middle-upper (2100 m a.s.l.) south flank (Figure 1). The July–August 2001 Etna eruption represents an event of great scientific interest for several reasons, the most evident of which are the areal extent of the surface fractures field (7 km long field of fissures), and the rapidity of the dike emplacement (a few days). The lateral eruption finished on August 9, 2001, having a preliminary estimated lava volume of ca. 48 million m3 [INGV-CT, 2001] and a flow of over 6.5 km, which approached the village of Nicolosi, fortunately, without reaching it.
 In the paper we will show that the ground deformation data obtained from the permanent integrated networks (tiltmeter and GPS) allowed us both to constrain the time evolution of the intrusion and to infer the source position and mechanisms during the days before the eruption onset.
2. Ground Deformation Permanent Networks
 Continuous ground deformation monitoring is presently carried out by permanent tilt and GPS networks (Figure 1). The continuous tilt monitoring is realised by a permanent network of eight bi-axial electronic borehole tiltmeters installed at a depth of about 3–5 m. The complete configuration began operating more than ten years ago. The stations are equipped with Applied Geomechanics Inc. tiltmeters model 722 (CDV, MSC, MEG, SPC) with 0.1 μrad precision and model 510 (DAM, MDZ, MMT) with 0.01 μrad precision. In 1997 a high precision long base fluid tiltmeter (PDN) was installed at the Pizzi Deneri observatory in the high (2850 m a.s.l.) northeastern volcano flank (Figure 1). The instrumentation is positioned along two partially underground tunnels, and is composed of two 80 m long orthogonal tubes filled with mercury, whose vertical changes at the extremities are measured by laser sensors, which provide a real sensitivity of 0.01 μrad [Bonaccorso et al., 1998]. The data of all tiltmeters, recorded each half-hour for the borehole stations and every 10 minutes for the long-base one, are transmitted to the INGV-CT via radio-link.
 The permanent network of continuously recording GPS is made up of eleven stations (three in the summit area of the volcano, and eight distributed on the flanks of the volcano). The installation of the continuous GPS permanent network began in September 2000, and was gradually upgraded the following year. Data collected at the remote stations (powered with solar panels) are transmitted through GSM cellular telephones to the master station at the INGV in Catania. A dedicated and modular software, called EOLO [Amore et al., 2001], which can be used with different modules for data processing, automatically downloads data from the remote stations, controls the analysis processes, and manages their visualization and analysis in terms of strain parameters.
 During the first days of the seismic crisis, 12–14 July, tilt variations were recorded at almost all the stations. In detail, tilt variations were very small or not revealed (≈ 0–1 μrad) in the northern sector (MNR, DAM, MMT), were of few μrad (≈ 5 μrad) in the western sector (MEG, MSC), and were much more marked at the three closest stations to the fracture field: MDZ in the SSW sector, CDV in the southern flank close to the final part of the fracture, and PDN in the high north-eastern flank close to the summit crater area. CDV station, interrupted because of technical problems, was promptly repaired on 13 July. CDV and MDZ are located above two shallow clustered zones of the seismic swarm [INGV-CT, 2001] and seems to be affected by co-seismic offset of the more energetic events (Figure 2, left). This aspect was clearly seen at MDZ station, where the tilt variation accelerated in coincidence with the strongest earthquakes located under this station. PDN station did not show any problems, and proved to be the site better showing the time evolution of the process and evidenced that the dyke emplacement took place between 12–14 July just a few days before the eruption onset (Figure 2, right).
 The GPS data have been managed by EOLO modular software, using Trimble Geomatics Office software as the data processing module. The networks have been adjusted using the reference point NICO, situated on the low south flank of the volcano. Data reduction using 24 hour sessions and fixed points external to the volcanic edifice, estimates typical errors of a few millimeters for the horizontal components and 10–12 mm for the vertical components [e.g., Dixon et al., 1997]. In our conditions, without fixed reference points outside the volcanic edifice and with limited time sessions, we calculate that the vertical precision is about 20 mm.
 Daily measurement sessions recorded by the GPS permanent network also showed marked variations (50 to 120 mm) in line lengths starting from July 13 (comparison between 12 and 13 of July) and continuing over the following two days (Figure 3). This aspect, in agreement with tilt data, confirmed that the main surface deformation was recorded coincident with the beginning of the seismic swarm and the following two days, i.e. between 12 and 15 of July. Astride this time interval the GPS flank stations recorded very small vertical changes (from a few mm to 20–25 mm) much smaller than the horizontal changes. This deformation pattern unequivocally displays a response to a tensile mechanism coherent with an intrusion crossing the volcano edifice along a ca. N–S direction in the southern flank. Excepting ETDF benchmark, all points cumulated the displacement between 12–15 July, i.e. they did not record significant variations between 16 and 18 July. ETDF displacement is influenced by the summit part of the ground fracture system which, during the days preceding the eruption, departed from craters southwards passing very close to ETDF [INGV-CT, 2001].
 In order to model the observed deformation pattern we estimated the surface deformation expected by an opening dike through a tensile dislocation in a homogeneous half-space [Yang and Davis, 1986] considering the Poisson coefficient equal to 0.25 (i.e. with λ = μ as Lamè constants). We used the GPS line changes and the PDN tilt recorded in the time interval 11–16 July before the eruption onset. These measurements have much larger variations than can be accounted for experimental error. Exploiting that period, we have the evident advantage of using a very short time interval where the primary source was undoubtedly the initial dike uprising and emplacement. We did not take into account the lines connecting ETDF since this benchmark is strongly affected by the near field of the surface fracture system and cannot be considered truly elastic. We inverted for the eight parameters characterizing the dislocation (3 center coordinates, length, width, direction, dip, opening) using a nonlinear least squares algorithm based on the Marquardt  method. In all the trials performed, even using arbitrary starting models, we always obtained the near-same final solution. The dike is located roughly southeast of the summit crater area and extends vertically in the last 2.4 km with an estimated opening of 3.5 m (Table 1). The goodness of fit is very promising and we have a reduced chi-squared χ2 equal to 1 considering the a-posteriori standard deviations of 0.03 m for the line changes and 1.5 μrad for the tilt. In the inversion procedure we assumed that measurement errors are Gaussian and normalized all measurements by dividing by the standard deviations. The misfit, higher than the measurement error, is probably due to the approximations of a simple rectangular tensile dislocation model and a simple elastic medium, which do not take into account heterogeneity and topography, and therefore cannot completely explain all the details in the observations. The observed and calculated deformation patterns (displacements plus tilt) are reported in Figure 4 together with the position and geometry of the modeled dike. The vertical changes calculated by the model in the flank GPS stations are very small (from few mm to 15 mm) in agreement with the recorded changes. Almost all the earthquakes of the swarm were clustered in the last kilometers below the area around the eastern rim of the Valle del Bove along an approximately N–S direction with the foci clustered in the last shallow kilometers. The location of the modeled tensile source evidenced that for this fast dike emplacement the swarm seismicity zone is well fitted by the position of the dislocation source which provoked the main static deformation at the surface.
Table 1. Estimated Model Parameters and Uncertainties
The reference surface (z = 0) has been assumed at 1500 m a.s.l., namely the mean elevation of points belonging to the permanent networks. The origin of the coordinate system is coincident with Long. 15° 00′ E , Lat. 37° 36′ N.
Xcentre = 1.06 ± 0.17 Km
dip = 90° ± 5°
Ycentre = 14.66 ± 0.16 km
length = 2.2 ± 0.12 km
Zcentre = 1.25 ± 0.13 km
width = 2.3 ± 0.14 km
strike = N − 7° ± 2°
opening = 3.5 ± 0.4 m
 The ground deformation pattern suggested that the modality of the dike emplacement is different from the previous eruptions modeled through the deformation recorded in the last twenty years. A résumé of recent eruptions modeled through ground deformation data at Mt. Etna and their associated mechanisms is provided in Bonaccorso . In general, the dikes have represented shallow fracturing radially propagating from the craters area, usually along preferential directions. They can be caused by magma loading in the upper conduits (e.g., 1989 eruption case) or by final fracturing associated with a magma uprising process (e.g., 1991). There are also cases (e.g., 1981), where the dike vertically crossed the volcano edifice, interpreted as the final emplacement of a magma that ascended rapidly from a deep level region [Bonaccorso, 1999]. Discrete deformation measurements (EDM and leveling) showed large horizontal displacements (several to tens of centimetres) and much smaller vertical changes in the volcano flanks, analogously to the 2001 eruption. In this case, modeling indicates that a dike penetrated inside the volcano edifice below the summit craters area, along an approximately N–S direction, provoking the observed pattern [Bonaccorso, 1999]. Then, when the dike approached the surface, in the last hundreds of meters, a final shallow eruptive fracture propagated along the northern sector. But also the 1981 eruption dike, even though it ascended through the volcano edifice similarly to the 2001 event, is different from this last case. In fact, as evidenced by the continuous tilt variation cumulated in the months before the eruption, the intrusion formed inside the volcano edifice during 3–4 months preceding the eruption. The volume of the 1981 modeled tensile crack (∼21 · 106 m3) fits the volume of the estimated magma output. Therefore, these aspects confirmed that in the 1981 case the magma output was stored in this dike which formed during the 3–4 months before the eruption [Bonaccorso, 1999].
 In the case of the July 2001 eruption, the emplacement was instead much faster and the volume of the modeled dike (∼16 · 106 m3) resulted smaller than the estimation of the erupted volume (∼48 · 106 m3). The discrepancy between expected injection volume and the lava-discharged volume could indicate a supply volume from a deeper region. In any case, this point does not fall within the scope of this work and represents a topic to be investigated in a further investigation. The surface fracture field in the summit area, as evidenced by structural observations [INGV-CT, 2001], suggested a precise behavior. In the days before the eruption, brittle deformation was observed north of the eruptive fracture under a compression consistent with a stress field characterized by a horizontal σ1 oriented approximately N–S. Furthermore, during July, 18, the opening and propagation patterns of the eruptive fissure, indicated a mechanism controlled by a shortening axis oriented, once again, N–S (G. Lanzafame, personal communication) which is the same as the orientation of the regional tectonic field [e.g., Cocina et al., 1997; Lanzafame et al., 1997].
 The July 2001 eruption occurred 10 years after the previous flank eruption, i.e. the 1991–93 eruption, and several years after a continuous recharging of the volcano system. A continuous tension accumulation is well testified by the near constant positive areal dilatation recorded by EDM and GPS and cumulated from 1994 to 2001 [Bonaccorso et al., 2001]. This tension accumulation was partially released by a strong explosive activity during 1998–2001, and by two summit eruptions (February–November 1999, and January–June 2001). In this frame, the July 2001 eruption appears as a transition from a phase of spectacular summit activity to one of lateral energy discharge. The main open problem concerns understanding why this transition occurred so rapidly. In any case, this eruption could represent the interaction of two different aspects: a long term tension accumulation due to a process of new magma supply and recharging over the last years, superimposed by the tectonic field, whose action is supported by the geometry of the dike emplacement having N–S direction in agreement with the σ1 regional field direction. All these aspects suggest that in the future the stress state of regional field should also be investigated through high-resolution instruments, such as deep strainmeters installed also in the low flanks and/or surrounding areas of the volcano. However, as clearly underlined by Larson et al. , a combination of tilt and GPS seems to be the best geodetic volcano monitoring system.
 The July 2001 Etna lateral eruption was preceded by marked deformation recorded at the tilt and GPS permanent networks. The deformation occurred during a few days preceding the eruption onset, and was different from previous modeled lateral eruptions. We modeled the deformation changes through a rapidly propagating dike, which well fits the recorded pattern and the zone of the seismic swarm occurring during the magma uprising. Our work demonstrates the potentiality of integrated (tilt and GPS) continuous deformation measurements to image the spatial-temporal evolution of propagating dikes before the eruption, even in situations of very fast emplacement. Furthermore, it testifies to the utility of the tilt and GPS permanent networks coupled with automatic data analysis system in order to provide an effective early warning.
 We thank M. Bonafede and P.M. Davis for reviews that led to improvements in the manuscript. We are indebted to all the personal of Ground Deformation Unit of National Institute of Geophysics and Volcanology (Catania Section) who guarantee the regular working of the permanent networks. We also thank M. Burton for improvements in the English of this paper.