Using interferometric synthetic aperture radar (InSAR) we constrain the deformation sources for the July–August 2001 flank eruption of Mt. Etna volcano, Italy. InSAR data from ascending and descending passes of the ERS2 satellite reveal a pattern of deformation that cannot be explained by a dike intrusion alone. In addition to a vertical dike beneath the south rift zone, the spatially large (10 km scale, 15–20 cm in range) negative range displacement lobes across the western (descending data) and eastern (ascending data) flanks require a nearly symmetric set of shallowly dipping normal faults to each side of the central dike. Complexity in the observed InSAR surface displacements constrains an additional dike intrusion beneath its NE flank. Long-term deformation of Etna's eastern and southern flanks is well established through field and InSAR observations. Therefore, the relative symmetry of motion beneath both the western and eastern flanks during the 2001 eruption is surprising. Our model of symmetric flank motion suggests that on the short time scales of a large dike intrusion volcanoes can deform differently from their long-term deformation.
 Mt. Etna volcano is one of the most active and intensely studied volcanoes in the world (Figure 1). It is a large (3300 m high) stratoshield volcano that produces large lava flows from both its summit and flanks. It has two well-developed rift systems extending to the north and south from the summit craters that connect through a series of fault systems as its eastern and southern flanks move away from the volcano center [Rasà et al., 1996; Rust and Neri, 1996; Borgia et al., 1992, 2000; Froger et al., 2001]. Whether this flank motion is deep seated [Borgia et al., 1992, 2000] or takes place higher within the edifice, or a combination of both [Tibaldi and Groppelli, 2002] is still a matter of debate, although recent observations of surface deformation support the former interpretation [Lundgren et al., 2003]. In either case, Etna is viewed as buttressed to the north and west against the Maghreb-Appenine chain in the interior of Sicily, so any motion of its western flank is unexpected.
 From July 17 to August 9, 2001, Mt. Etna experienced a large flank eruption along its southern rift zone, extending from near the SE crater to the south for approximately 6 km [The Research Staff, 2001]. This was preceded from July 12–16 by intense seismic activity and surface deformation consistent with emplacement of a dike [Patane et al., 2002; Bonaccorso et al., 2002].
 The 2001 flank eruption culminated the eight-year period since its last flank eruption that ended in 1993. After that eruption Etna had a two-year quiescence, during which its magmatic system recharged [Lanari et al., 1998; Puglisi et al., 2001; Lundgren et al., 2002]. A new phase of summit magmatic eruptions started in the second half of 1995. From 1995–2001 Mt. Etna was in an extended phase of summit activity with several periods of large lava fountaining, effusive eruptions, and large explosive eruptions.
 We investigate the mechanics of the 2001 flank eruption by analyzing its surface deformation measured with interferometric synthetic aperture radar (InSAR). Interferograms from ascending and descending satellite passes give very different images of the surface deformation. Together the InSAR data constrain the type, geometry, and slip of subsurface structures.
2. InSAR Data
 InSAR is a technique for computing relative surface deformation maps by differencing the coherent phase of two SAR images [Gabriel et al., 1989; Rosen et al., 2000]. We use SAR data from the European Space Agency's (ESA) ERS2 satellite, a C-band SAR with a wavelength of ∼5.6 cm. Raw ERS 2 data were analyzed using the ROI_PAC software package developed at the Jet Propulsion Laboratory and Caltech. Our analysis used the Shuttle Radar Topography Mission (SRTM) digital elevation model to remove topographic effects.
 The shortest time separation interferograms that were possible for the ascending and descending tracks are shown in Figure 2. The ascending interferogram has the minimum time separation possible for ERS (35 days), while that of the descending data is one year. The two interferograms each are characterized by a bi-lobate structure, with one lobe featuring positive surface displacement in the radar line-of-sight (LOS) and one featuring negative LOS surface displacement. For the descending interferogram (Figure 2a) the broad lobe over the western flank of Etna is negative and the tight, dense fringed positive lobe lies over the upper east flank. In contrast, the ascending interferogram (Figure 2b) has a broad negative LOS lobe over the east flank and a more compact positive LOS lobe over the western flank. There is also motion of the east flank bounded by the Pernicana fault expressed as a series of NS parallel fringes that are visible in the descending interferogram. The observed fringe patterns cannot be the product of atmospheric effects based on past ERS observations at Mt. Etna [Lundgren et al., 2003].
3. Data Modeling
 Volcano models are commonly computed with analytic solutions for simple pressure sources (in the case of simple inflation or deflation due to a central magmatic source) or tensile dislocations (in the case of a dike) in an elastic half-space. For large volcanoes effects due to topography are expected to be observable for deeper inflationary sources [McTigue and Segall, 1988; Cayol and Cornet, 1998; Williams and Wadge, 2000], although they should be less significant for sources located within the volcano edifice.
 Using elastic half-space dislocation sources [Okada, 1985] we found we could not fit the observed InSAR data (Figure 2) with a simple NS striking tensile dislocation (Figure 3a) as expected [Bonaccorso et al., 2002]. The effect of a dike extending from the surface to 3 km depth is to produce positive LOS displacements in each interferogram that are much greater in magnitude and spatial extent compared to the negative LOS displacement lobes. We find that to fit the general pattern of tight large amplitude positive lobes and broader slightly lower negative lobes in each interferogram requires adding a shallowly dipping normally slipping dislocation beneath the eastern and western flanks of Mt. Etna; extending from the base of the dike for approximately 10 km down-dip (Figure 3b–3c).
 To understand whether the observed InSAR data could be affected by topographic effects we have computed three-dimensional finite element models (3D FEM) matching the models shown in Figure 3 in which the modeled structures were located relative to the summit of the volcano. The effects of topography were found to be small and not significant to the main interpretations in this study. However, the close similarity of the deformation caused by a dike in a halfspace and one within the edifice, suggests that the depths of the modeled structures should be referenced relative to the summit of the volcano. Thus, the dike in this case extends from the summit to sea level and the tops of the normal faults extend from sea level to ∼5 km depth. This conclusion has been argued theoretically by McTigue and Segall .
 To better match the data, we compute a preliminary forward model (Figure 4) that seeks to limit the number of structures yet match the major features of the data. This model was computed on a trial and error basis, starting from a simple dike extending from the SE crater in line with the surface fractures [The Research Staff, 2001], with two laterally extending normal faults. The positions and orientations of the normal faults and the complexity in the dike shape and tensile slip were added to improve the fit to the data. Despite apparent visual improvement to the fit the actual variance reduction for the complex forward model is only about 30% compared to an inverse solution for a single dike. The reason for the modest improvement in model fit compared to their large apparent differences lies in the large number of points located in areas of relatively low deformation that provide the greatest contribution to the variance.
 The complex model is composed of two simple normal fault dislocations extending to the SW and ESE beneath the western and eastern flanks of Etna, a central dike system beneath the S rift zone, and a shallow, NW dipping dike extending to the NE from the S rift zone. Model parameters can be found in Table A11. Although more complex than the first-order model shown in Figure 3c, the variations in the extent and magnitude of the dike opening change progressively. Schematically the central dike extends from the central feeder system up to a depth of ∼1–2 km below sea level where it extends to the south. The extension required at depths greater than the up-dip edges of the normal faults is less (by approximately the outward motion of the material above these faults) than the 1.5 m of opening in the upper central portion of the dike. In spite of their small opening, the NE striking 60° NW dipping dikes are added to fit the northern positive LOS lobe on the ascending interferogram.
 Several structures are not included in this model and are left for a future, more complete numerical analysis. They include the NE rift/Pernicana fault system, which separates the eastern flank motion from the northern flank of Etna; the SW striking fault that cuts the southern portion of the negative lobe in the descending interferogram. In addition the eastern flank motion, as seen in the ramp of LOS motion across the eastern portion of the descending interferogram (south of the Pernicana fault) requires a basal detachment that is not easily modeled by a simple rectangular dislocation.
4. Discussion and Conclusions
 We have sought to strike a balance between model simplicity and fitting the main components of the surface deformation. Some features of the shape and displacement gradients within individual lobes in the InSAR data could be refined, either through tapering of the slip and/or through complexity in the fault geometry (i.e. the shape of the western, negative lobe in the descending interferogram). Improved fit could be achieved by adding additional structures that are not included (see previous section), and through greater complexity of the included structures (i.e. possible listric shape to the western normal fault). It should also be noted that the interferograms spanning the 2001 eruption over both 35 day and nearly one-year time spans do not show any significant deep (i.e. below the edifice) deflationary signal.
 In the current model the depths to the tops of the normal faults could be moved up to 2 km beneath the surface, though this tends to degrade the shape of the lobes, a possible trade-off with slip complexity or tapering. This is a general feature of geodetic modeling, with the smooth fringes better explained by deeper sources. Given the required widths (down-dip extent) of these faults it is harder to imagine a source depth much less than 2 km in which the hanging wall block moved coherently. The depths of these structures have direct bearing on the material structure of the volcano. If they extend higher into the volcanic edifice this might support the interpretation that, at least in this eruption, the clay layer that is inferred to extend into the edifice [LaBaume et al., 1990; Rust and Neri, 1996; Tibaldi and Groppelli, 2002] could act to decouple the upper flanks of Etna. At 3 km (i.e. sea level) they would extend into pre-existing basement rocks, possibly to a deeper detachment at ∼5 km depth [Tibaldi and Groppelli, 2002]. The smooth shape of the negative LOS lobes suggests that the normal faults extend to several kilometers below sea level.
 InSAR observations require that the central dike caused nearly symmetric spreading of Mt. Etna over shallowly dipping faults. Geologic field and geophysical (notably InSAR) observations demonstrate well-developed structures that accommodate motion of Etna's eastern and southern flanks [Borgia et al., 1992, 2000; Froger et al., 2001; Tibaldi and Groppelli, 2002]. This symmetry is in agreement with other dike-fault interaction observations where motion of the shallow normal faults connecting the dike relieves the stress caused by the dike intrusion [Toda et al., 2002]. The apparent lack of normal fault seismicity [Patane et al., 2002], might be explained by aseismic creep following emplacement of the dike structure. Delayed motion on the flank faults might also explain the differences between the InSAR observations and GPS displacements preceding the eruption [Bonaccorso et al., 2002]. The symmetry of our model for the 2001 eruption, especially its western flank motion, is in stark contrast with its long-term structural deformation. This suggests that over the short time-scales of the dike emplacement (a few days) the volcano deformed in a way that was significantly different from the long-term spreading of its eastern and southern flanks.
 We thank G. Wadge and an anonymous reviewer for their insightful comments. This work was conducted at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.