We investigate crustal deformation along the Eurasia-Nubia plate boundary in Calabria and Sicily revealed by the GPS velocity field obtained by the combination of continuous site velocities with previous results from episodic campaigns. We recognize two distinct crustal domains characterized by different motions and styles of deformation. Convergence in Sicily is taken up by crustal shortening along the former Tyrrhenian back arc passive margin, in agreement with seismological data and geological evidence of recent cessation of deformation along the Plio-Pleistocene subduction front. The analysis of the GPS data and the consistency between earthquake slip vectors and convergence direction suggest that Eu-Nu convergence in Sicily does not require intermediate crustal blocks. Significant Eurasia (∼3 mm/yr to NNE) and Nubia-fixed (∼5 mm/yr to ESE) residual velocities in Calabria suggest instead the presence of an intermediate crustal block which can be interpreted as a forearc sliver or as an independent Ionian block. According to the first hypothesis, subduction is still active in the Ionian wedge, although we find no evidence for active back arc spreading in the Tyrrhenian Sea. The N115°E oriented Sicily-Calabria GPS relative motion is consistent with the extension observed during the 1908 Mw 7.1 Messina earthquake. We suggest that up to 3 mm/yr (∼80%) of this estimated relative motion between Sicily and the Calabrian Arc may be taken up in the Messina Straits.
 The Mw 7.1 earthquake that struck the Messina Straits on 28 December 1908, is one of the largest events [Boschi et al., 1995] ever recorded in the central Mediterranean (Figure 1). Previous studies based on geodetic leveling, seismic waveforms inversion and long-term geological features have converged in the identification of a NNE-SSW trending active normal fault parallel to the strike and controlling the morphology of the Straits, as responsible for the 1908 earthquake [Boschi et al., 1989; Valensise and Pantosti, 1992; Pino et al., 2000; Amoruso et al., 2002]. More controversial is the geodynamic framework in which active extension in the Messina Straits occurs. The evolution of this sector of the Eurasia-Africa plate boundary (Figure 1) over the Neogene and Quaternary times is generally interpreted in terms of slow relative plate convergence [Argus et al., 1989; DeMets et al., 1994] and fast subduction and roll-back of the Ionian lithosphere beneath the Calabrian Arc associated with back arc spreading in the Tyrrhenian Sea [Malinverno and Ryan, 1986; Patacca et al., 1990; Faccenna et al., 2001a]. For the understanding of the present-day pattern of deformation this model provides only limited insights because (1) direct geological and seismological evidence of active shortening in the Ionian Sea and contemporaneous back arc spreading in the Tyrrhenian Sea are lacking and (2) the position and direction of extension of the active fault in the Messina Straits, above the western edge of the Ionian slab, conflicts with its interpretation as a back arc extensional structure. The regional kinematic framework in which active extension in the Messina Straits occurs remains, therefore, poorly understood together with other issues such as (1) the influence of the Wadati-Benioff plane beneath the Tyrrhenian Sea on the crustal velocity field and (2) the distribution and style of active deformation along the plate boundary between Africa and Eurasia. Although geodetic studies aimed at monitoring crustal motion across the Messina Straits started as early as the 1970s, evidence of active extension based on trilateration and GPS across the Messina Straits have since remained elusive [Anzidei et al., 1998].
 In this paper we use GPS data to estimate the crustal motion in the central Mediterranean plate boundary zone between the Eurasia and Nubia (e.g., the African plate west of the East African Rift) plates and propose a geodynamic framework for active extension in the Messina Straits. Inclusion of GPS continuous stations from the “stable” parts of the Eurasian and Nubian plates allows us (1) to define the kinematic boundary conditions imposed by the relative motion between the two plates and (2) to obtain site velocities in the deforming zone in both reference frames. We integrate our permanent site velocity solution with published denser survey-mode GPS results to obtain a combined velocity solution which allows us to estimate the active strain across the Messina Straits and discuss the implications for the regional seismic hazard.
2. Seismotectonic Setting
 The Calabrian Arc and the Sicily Island are located within the complex, articulated plate boundary between the Eurasian and African plates [McKenzie, 1972]. Figure 2 shows the instrumental seismicity recorded by the Italian National Seismic Network [Chiarabba et al., 2004] in the period 1981–2003 and the events with hypocentral depths larger than 50 km extracted from the catalogue of relocated earthquakes of Engdahl et al.  for the period 1964–1983. Beside the seismicity below the Etna volcano and along the Calabrian Arc, the most evident pattern of crustal seismicity is the east-west continuous belt north of Sicily in the Southern Tyrrhenian Sea showing thrust focal mechanisms with NNW-SSE and NW-SE compressive axes (Figure 1). The belt bends toward southeast in the proximity of the Aeolian Islands [Pondrelli et al., 2004] where geological, geodetic, and seismological data [Mazzuoli et al., 1995; Bonaccorso, 2002] suggest that the dominant deformation is right-lateral shear on NW-SE trending faults (Figure 1). Magnitude of the earthquakes along the E-W offshore seismic belt and in the Aeolian Islands has barely exceeded 5.5–6.0 during the instrumental period (the last being the Mw 5.7, 6/9/2002, Palermo earthquake). The region northeast of this continuous belt is dominated by the intermediate and deep events in the Southern Tyrrhenian Wadati-Benioff zone. Here, the depth distribution of earthquakes, as deep as 600 km, fairly well delineates the steeply (∼70°) dipping Ionian lithosphere subducted beneath the Calabrian Arc [Selvaggi and Chiarabba, 1995]. Fault plane solutions within the slab show pervasive downdip compression at all depths beneath 100 km, while at shallower depths downdip extension is observed [Selvaggi, 2001]. Deep and intermediate earthquakes in the Southern Tyrrhenian Wadati-Benioff zone abruptly terminates westward beneath eastern Sicily and the Aeolian Islands (Figure 2). The Hyblean Plateau shows a low level of instrumental seismicity in the past 20 years, although three destructive large earthquakes occurred in historical times in the years 1169, 1542 and 1693 A.D. [Boschi et al., 1995]. The 1693 earthquake is probably one of the largest events that occurred in Italy with estimated magnitude of 7.2.
 The subduction of the Ionian lithosphere has been characterized during the Neogene and Quaternary by trench retreat and back arc extension, in response of the retrograde motion of the subduction hinge [Malinverno and Ryan, 1986]. Geological estimates of 50–70 mm/yr of back arc extension, averaged over the Neogene and Quaternary [Patacca et al., 1990; Faccenna et al., 2001a], provide significant evidence for vigorous trench retreat. Although direct seismological and geological evidence of the present-day subduction activity in terms of shortening in the accretionary wedge and spreading in the Tyrrhenian back arc basin are lacking, claims have been recently made for vigorous post-700 kyr trench retreat [Gvirtzman and Nur, 1999] and active plate divergence and lithosphere formation in the youngest part of the Tyrrhenian Sea, the Marsili seamount [Marani and Trua, 2002]. Active tectonics in Calabria and eastern Sicily is dominated by normal faulting along the Tyrrhenian side of Calabria and in the Messina Straits, describing a continuous extensional belt initially parallel (in Calabria) and subsequently cutting at high-angle (in the Messina Straits) across the front of the Apennine-Maghrebian thrust and fold-belt [Tortorici et al., 1995]. In Sicily cessation of the active convergence along the Plio-Pleistocene nappe thrust front north of the Hyblean Plateau is suggested by Pleistocene deposits onlapping both the thrust front and the flexed north dipping foreland lithosphere in the Hyblean Plateau [Butler et al., 1992; Torelli et al., 1998]. The clear contractional character of the active deformation in the Southern Tyrrhenian Sea (Figure 1), contrasts with late Pliocene-early Pleistocene normal fault growth and subsiding basins, imaged in offshore seismic lines [Pepe et al., 2000]. These extensional structures represent the southern boundary of the Tyrrhenian extensional back arc domain. Thrust fault plane solutions and the lack of clear evidence of inverted structures on the seismic lines [Pepe et al., 2000] imply that a recent tectonic change in the tectonic regime may have occurred along this margin in relation to the cessation of activity along the thrust front north of the Hyblean Plateau and northward migration of the locus where the Eu-Nu convergence is accommodated. It is not clear whether the thrust fault earthquakes reactivate preexisting north dipping extensional faults, related to the Tyrrhenian back arc extension, or rupture south dipping faults accommodating the overthrusting to the north of the Sicilian continental block over the thinned Tyrrhenian crust.
3. Data Analysis
3.1. Continuous GPS Measurements (CGPS)
 Continuous GPS measurements were analyzed using the GIPSY software developed at the Jet Propulsion Laboratory. Nonfiducial satellite orbits, clock files and transformation parameters from free-network to ITRF2000 were imported from JPL (ftp://sideshow.jpl.nasa.gov). In order to define the velocity field with respect to the Eurasian and Nubia plates, we included in our analysis stations located on the stable parts of these plates. In a first processing step daily solutions in the ITRF2000 reference frame for individual station position and corresponding matrices of the covariance among the three position components, spanning the time interval 1996.0 to 2004.5, were initially determined using a precise point positioning strategy [Zumberge et al., 1997]. We then obtained a “regional” ITRF2000 velocity solution for selected stations with the longest observation interval and minimum antenna-receiver operations (ZIMM, GRAS, GRAZ, TORI, GENO, CAGL, MATE, and NOTO), by least squares fitting their weekly averaged Cartesian positions. In the following step, sites located within the area of interest (Figure 3) were brought together for local-network processing and ambiguity resolution. The free-network ambiguity-fixed daily solutions were then transformed into the ITRF2000 by estimating a seven-parameter similarity transformation for each. We estimated the transformation for each day, on the basis of selected common stations that are present both in the ITRF2000 “regional” solution and in the daily ambiguity-fixed solution. On average seven common stations were present both in the ambiguity-fixed and “regional ITRF2000” solution (a minimum of three common stations were required to evaluate daily transformation parameters). From the ITRF2000 daily positions rotated in a local NEU frame we simultaneously estimated, with a weighted least squares approach for each station, a linear station velocity, an annual, and a semiannual periodic signal. Antenna offsets were estimated at discontinuities in the time series related to known equipment change or where a clear discontinuity was observed in the time series. Only time series with an interval longer than 2.5 years have been considered for velocity estimation to mitigate the effect of the bias introduced by a periodic annual signal [Blewitt and Lavallée, 2002]. Only the PHLW site with a slightly shorter time series was retained for its interest. The velocity uncertainties presented in this paper for CGPS measurements account for both noncorrelated (white noise) and temporally correlated site positions effects. We adopted a simple empirical model for estimating the GPS rate error [Mao et al., 1999; Dixon et al., 2000] that includes white and flicker noise. Weighted root mean squared from time series of position for individual stations was used to scale these contributing processes. Using ITRF2000 site velocities (Table 1) and associated errors we estimated the Euler poles of rotation for both the Eurasian and Nubia plates by minimizing, with a weighted least squares inversion, the adjustments to horizontal velocities of 26 stations within the stable Eurasia and 10 stations within the stable Nubia (Tables 2 and 3). The RMS of residuals and the fitting procedure result in a reduced chi-square of about 1 indicating that a rigid plate is an adequate approximation given the uncertainties in the GPS velocities. The relative Eu-Nu pole of rotation (Figure 4) is then derived by differencing the ITRF2000 angular velocities with appropriate error propagation. The convergence vector derived by our GPS Eu-Nu pole of rotation (5 ± 0.28 mm/yr to N45W ± 3° at 17.50°E, 36.65°N) significantly differs from the prediction obtained by the closure of the Africa-North America-Eurasia plate motion circuit [Argus et al., 1989; Calais et al., 2003]. Our solution displays (Figure 1) a 35% smaller and 30° more westerly directed convergence direction in Calabria and Sicily, similarly to others recently published GPS Eu-Nu poles of rotation [Sella et al., 2002; McClusky et al., 2003; Fernandes et al., 2003; Calais et al., 2003]. Whether this discrepancy reflects a real change in Nubia-Eurasia plate motion or unrecognized systematic errors in both geodetic and geological data is discussed in detail by Calais et al.  and will not be discussed here, since the Eu-Nu GPS pole of rotation derived in the present study and in the former quoted papers are similar and not statistically different at the 95% confidence interval.
Table 1. GPS Site Positions, Velocities, 1-Sigma Uncertainties, Correlation Between the East and North Components of Velocity, Time Span, Number of Observing Days, and Residual Velocitiesa
ITRF2000 Velocity, mm/yr
Definitions are as follows: ΔT, time span; ntot, number of observing days; corr, correlation; E(±) and N(±), 1-sigma uncertainties.
Residual velocities are relative to the respective plate.
Residual velocities in the plate boundary are relative to Eurasia.
NOTO velocity is obtained from the composite time series of NOTO prior to 6 September 2000, and NOT1 thereafter.
These six columns give the elements of the covariance matrix (10−9 rads2/Myr2).
3.2. Combination With Episodic GPS Measurements
 A recent study by Hollenstein et al.  presents the results of periodic GPS surveys in southern Italy and provides supplemental data to integrate our permanent sites velocity solution. The distribution of the geodetic sites (Figure 3) analyzed by Hollenstein et al.  allows us to obtain a dense station coverage in those parts of Sicily and neighboring islands not covered by permanent stations and one more site in Calabria (PORO). Site velocities were estimated by Hollenstein et al.  in the ITRF97 reference frame and then rotated in a Eurasia reference using the ITRF97-Eurasia REVEL pole of rotation of Sella et al. . Our permanent sites (Figure 5) and the survey sites velocity solutions are thus nominally in the same Eurasian reference frame, but minor discrepancies are shown at common stations, producing an RMS of the differences in east and north velocities of 0.48 and 0.71 mm/yr. We minimized these discrepancies by inverting with a weighted least-square algorithm, for the rigid rotation that minimizes the differences in velocity at common stations (CAGL, NOTO, MATE, UNPG, LAMP) using the CGPS velocities as the reference solution. The RMS of the velocity differences after combination is reduced to 0.18 and 0.23 mm/yr in the eastern and northern directions, respectively. The nearby MILO and TRAP sites (which were not used to estimate the rigid-rotation parameters) permit to evaluate the consistency between the CGPS and episodic velocities data sets (Figure 6). The ∼2 mm/yr difference is not statistically significant at the 95% confidence interval. The factors which may contribute to this difference may include the relatively long time interval covered by the survey-style measurements of TRAP (7 years) compared with the short time interval of CGPS measurements (2.8 years) at MILO which may still be contaminated by periodic signal.
4. Combined GPS Velocity Field
Figure 6 shows the combined GPS velocity field in the Eurasian and Nubia reference frames. GPS sites lying on one of the two stable plates within 95% confidence interval are CAGL and AJAC, lying on the Eurasian plate, and LAMP lying on the Nubia plate. The velocity of NOTO site, previously considered to lie on the African plate [Ward, 1994], is similar to the motion of Nubia but shows a small but significant 2 mm/yr residual toward N77°E that may be explained either by active extension in the Sicily channel or by elastic strain accumulation on nearby active faults. The last hypothesis, supported by the severe damage (I > X MCS) suffered during the 1693 M 7.2 earthquake [Boschi et al., 1995], still considers southern Sicily and NOTO as lying on the Nubia plate, but allows some minor contamination of the NOTO velocity by elastic strain accumulation on a still poorly documented N-S off-shore active fault along the Malta escarpment. A similar fault geometry has been used by Piatanesi and Tinti  to model the tsunami of the 1693 earthquake. The small Nubia-fixed residuals of NOTO and LAMP, the small velocity gradients between NOTO and TRAP, the significant velocity gradient between TRAP and USTI together with the continuous band of seismicity in the Southern Tyrrhenian, all suggest that the Nubia motion is partly transferred across western and southern Sicily and absorbed by shortening in the Southern Tyrrhenian at a rate of ∼4 mm/yr. This observation argue for accommodation of the Eu-Nu convergence somewhere between Sardinia to the north and the Hyblean Plateau and Sicily channel to the south in agreement with seismological and geological evidence of recent cessation of activity along the subduction front in mainland Sicily. On the basis of a smaller number of CGPS sites, McClusky et al.  reached a similar interpretation placing the Nubia-Eurasia boundary north of station NOTO and south of Sardinia.
 In Calabria the two permanent sites COSE and TGRC, and the survey site PORO, have similar velocities of ∼5 mm/yr to ESE (Nubia-fixed), and ∼3 mm/yr to NNE (Eurasia-fixed). This pattern shows that Calabria has an independent motion with respect to both the Eurasia and Nubia plates. Two hypotheses can be proposed. (1) Assuming that the Ionian lithosphere is attached to the south to the Nubian lithosphere and considering that CAGL and AJAC have very small Eurasia-fixed velocities (i.e., they lie on the overriding plate of the Calabrian subduction), the Nubia-fixed velocities can be used to evaluate the subduction rate of the Ionian lithosphere beneath Calabria whereas the Eurasia-fixed velocities can be used to evaluate the rate of back arc extension in the Tyrrhenian Sea. The eastward directed Nubia-fixed velocities suggest that convergence between Calabria and Nubia is absorbed in the Ionian submerged wedge by the underthrusting of the Ionian lithosphere beneath Calabria. The insignificant elongation of the baselines between the Calabrian Arc (TGRC, PORO, COSE) and the Sardinia-Corsica block (AJAC-CAGL) shows, on the other hand, that although subduction may still be active, the Tyrrhenian basin is not actively spreading. Our results thus do not support recent claims of active plate divergence in the most recent part of the Tyrrhenian basin, the Marsili seamount [Marani and Trua, 2002] and vigorous trench retreat [Gvirtzman and Nur, 1999], which is generally associated with active back arc extension. (2) If, alternatively, the Ionian lithosphere is not part of Nubia, the GPS velocities in the Calabrian Arc are probably representative of deformation and strain accumulation at the margins of a rigid block located in the Ionian Sea (a more complete discussion is given in section 6.1).
 The combined velocity solution shows significant N-S velocity gradient between NOTO and the sites in northeastern Sicily (Figure 6b), resulting in a northward acceleration and an approximately N-S directed extension. The sites located at the extreme northeastern corner of Sicily, move toward NE with respect to Sicily, perpendicularly to the Eu-Nu convergence, and progressively rotate from NE to ESE above the western edge of the Ionian slab (Figure 6b) where deep and intermediate earthquakes track the slab at a depth between 100 and 250 km (Figure 2). Geodetic velocities projected along N30°E (Figure 7) show decreasing values from Sicily to the Aeolian Islands and statistically constant values along the Calabrian Arc. These observations are consistent with (1) buttressing of northeastern Sicily and the Aeolian Islands against the Tyrrhenian basin, and (2) rigid motion of the Calabrian block toward NE. How the motion of the Calabrian Arc toward NE is related to the kinematics of the Adria microplate and the Apulian block is not clear but increasing velocities of MATE and CADM suggest that the Apulian block does not buttress the Calabrian Arc.
 If Calabria and Sicily are characterized by different crustal motions it is important to define the style and kinematics along their common boundary with the aim of characterizing and quantifying the strains that might be released in large earthquakes. Since the site velocities in Calabria are not statistically different, we assume that TGRC is not significantly affected by interseismic elastic effects and postseismic transients after the 1908 Messina earthquake and can be considered as representative for the motion of the Calabria Arc. If we further assume that NOTO is representative of the motion of the southern and western Sicily, we can use a velocity diagram to obtain an estimate of the relative motion between Sicily and Calabria (Figure 8). Because this estimate relies on only two sites, the results of this simple calculation should be considered as a first-order estimate before more complete geodetic data are available. The diagram shows a relative TGRC-NOTO motion of 3.6 ± 06 mm/yr to N115°E ± 7°, approximately perpendicular to the strike of the Messina Strait fault and close to the slip vector of the 1908 Messina earthquake [Boschi et al., 1989]. The results of this simple analysis thus suggest that the direction of crustal extension from geological data and from the slip vector of the 1908 Messina earthquake are consistent with the relative motion between Sicily and Calabria and provide a preliminary upper bound on the total extension taken up in the Messina Straits. In the next section the combined velocity solution will be used to characterize the pattern of active deformation between Sicily and Calabria and evaluate the rate of strain accumulation in the Messina Straits.
5. Strain Rate Analysis and Active Extension in the Messina Straits
 From the combined velocity solution we proceed to evaluate the strain and rigid-rotation field in the boundary zone between the Calabrian Arc and Sicily (Figure 9). We have made discrete estimates of the velocity gradient tensor at points on a 30 km regular grid, using the velocities of a minimum of four sites. If only three such sites exist we do not compute a velocity gradient tensor because such an estimate, being even determined, is liable to instability, particularly if two of the three sites are much closer to each other than to the third. The covariance matrix used to estimate the velocity gradient tensor starting from site velocities is a weighted version of the velocity covariance matrix where the weighting factor [Shen et al., 1996] is exp(2 * d2/s2), where d is the distance of the geodetic site to the point to be estimated and s is a distance-decaying constant, taken as 60 km. This weighting means that a station at 50 km away from the spot to be evaluated will contribute only one-half its unweighted value to the velocity solution. The advantage of this algorithm is that strain rate estimates are less biased because of proper weighting and they are stable because they are evaluated weighted averages over the region. The geodetic sites at Stromboli (STRM) and Panarea (PANA) have not been considered because suspected of being contaminated by local magmatic deformation that culminated in the volcanic activity of December 2002. The area is characterized by rapid variation of the principal axes of the strain rate tensor changing from NNE trending compressional strain rates in the Aeolian Islands (∼50 nanostrain/year) to SE extensional strain rates in the Messina Straits (∼100 nanostrain/year). An homogeneous pattern of instantaneous rotation rates of about 2°–3°/Myr characterizes the area under study. The strain distribution is in agreement with geological observation pointing for NW right-lateral strike slip in the Aeolian Islands [Mazzuoli et al., 1995] and WNW-ESE directed extension in the Messina Straits [Valensise and Pantosti, 1992], and with the focal mechanisms of the largest earthquakes in the area (Figure 9a).
 Beyond a simple comparison between the distribution of strain rates and geological and seismological observations, we can ask what pattern of uniform faulting (e.g., the single set of parallel faults) can accommodate the observed velocity field. Holt and Haines  show that within a three-dimensional strain rate field, in which the magnitude of shear strain rates exceeds the magnitude of dilatational strain rate, there are two possible strike directions of faulting that can produce the horizontal components of the strain rate tensor. These are directions of zero length change in the velocity field, and they correspond to the strikes of the two nodal planes in a fault plane solution. In terms of the elements of the strain rate tensor, these strikes are
where xy, xx, yy are the three horizontal strain rate tensor components and θ is the strike angle of the fault measured anticlockwise from the x (east) axis. In Figure 9d we plot the strike directions of planes of shear (zero length change) derived from the components of the strain rate tensors shown in Figure 9b. The lengths of the lines shown in Figure 9d are proportional to the difference between the two principal strain rates. Purely strike-slip faulting is expected where the two directions are perpendicular. Dip-slip faulting is expected where the two directions are parallel, otherwise oblique faulting is implied. The fault strikes in Figure 9d provide a check on the consistency between the predicted set of faults that can uniformly accommodate the GPS velocity field and the pattern of faulting shown by geological and seismological data. The predicted strikes are similar to the NW trending right-lateral faulting observed in the Aeolian Islands, and NNE trending normal faults (with a right-lateral component) in the Messina Straits. A remarkable parallelism is observed between one of the predicted directions and the direction of the coast in the Messina Straits, which is probably controlled by faulting [Valensise and Pantosti, 1992]. Equation (1) implies that there may be more than one way in which distributed deformation can be accommodated by faulting. The main results of this comparison is that, despite the apparent complexity of the velocity field, one of the predicted set of faults (by which the velocity field can be theoretically accommodated by distributed deformation) is consistent with fault patterns independently observed by the geology and seismology. Furthermore these predictions are indicative of strain accumulation on seismically active faults.
 Given the observed and predicted patterns of faulting, the comparison of Figures 9d and 9c suggests that instantaneous rotations are likely to be released by right-lateral slip on NW striking faults in the western part of the analyzed network, so that no secular block rotation is expected. In the eastern part (Messina Straits) the observed pattern of faulting suggests instead that deformation may be associated with rotations of NNE striking normal fault blocks with a right-lateral component of slip. Our results do not support the existence of a single throughgoing NNW trending active faults cutting from the Aeolian Islands to the Malta escarpment, as proposed by Lanzafame and Bousquet . Our results favor the view in which an area of distributed deformation separates the Ionian lithosphere and the Calabrian Arc to the east from the Sicilian block to the west which transmits most of the Nu motion to the E-W trending offshore seismic belt in the Southern Tyrrhenian. The dense GPS network covering the region between Sicily and Calabria allows us to estimate the amount of relative motion between Sicily and the Calabrian Arc taken up in the Messina Straits. The increase in velocity and extension in the Messina Straits are best shown by projecting site velocities (Figure 10) onto a single profile constructed parallel to the estimated relative motion direction. A sudden increase in velocity occurs across the Messina Straits, where localized deformation may accommodate up to 3 mm/yr and approximately 80% of the relative motion estimated in Figure 8.
6.1. Slip Partitioning in the Calabrian Arc or Independent Ionian Block?
 Slip partitioning of the oblique convergence into its orthogonal trench-parallel and trench-perpendicular components is frequently observed in many convergent settings [Jarrard, 1986; DeMets et al., 1990]. Figure 11 shows the site velocities along the Calabrian Arc compared with the Eu and Nu-fixed velocities predicted for a complete partitioning of the Eu-Nu convergence. Rather than the poor bathymetric expression of the trench we use the average N20°E trend of the Calabrian Arc as a proxy for the trench direction, corresponding to a 25° obliquity angle between the plate convergence vector and the normal to the trench. The comparison between the observed and predicted velocities provides good evidence for slip partitioning of the oblique Eu-Nu convergence. On the other hand this hypothesis relies on the assumption of the Ionian block as a part of the Nubia plate. An alternative hypothesis, proposed by Westaway , suggests the existence of an independent Ionian block, rotating counterclockwise with respect to Nubia and intermediate between such plate and the Adriatic microplate. The pole of rotation of the Ionian block (with respect to Nubia) has been located by Westaway  in Libya resulting in normal faulting along the eastern coast of Sicily and oblique extension in Calabria. The lack of islands in the rigid Ionian Sea precludes the use of GPS to test the existence of such block. Further information on the active deformation of the Ionian wedge is thus likely to be provided by seismological data and reflection profiling experiments in the Ionian wedge and densification of the GPS network along the Calabrian Arc.
 Active tectonics and seismicity in Calabria are generally attributed to N-S to NE-SW trending active normal faults that dominate the geomorphology and follow the axis of the Calabria peninsula [Tortorici et al., 1995; Jacques et al., 2001; Galli and Bosi, 2002, 2003]. Because of the lack of geodetic sites along the eastern side of Calabria, we are not able to detect strain accumulation across these active faults, which may be loaded by increasing southeastward velocities. The NNE directed motion of the TGRC, COSE and PORO sites, predicts that a component of left-lateral motion on north to NNE trending fault should allow transport of the Calabrian Arc toward NNE with respect to the Tyrrhenian Sea and the Sardinia-Corsica block. Over a period of about 250 years (1659–1908) almost the whole active fault system along the Calabrian Arc ruptured through M ∼ 6–7 earthquakes [Jacques et al., 2001; Galli and Bosi, 2002, 2003]. Although the slip vectors of these earthquakes are poorly constrained, significant evidence of left-lateral slip have been observed by Galli and Bosi  along the 1683 earthquake fault ruptures.
6.2. Eu-Nu Plate Boundary and Style of Deformation
 With respect to the style of Eu-Nu convergence we recognize distinct styles of deformation taking up the shortening in the plate boundary. The southern and western part of Sicily can be kinematically considered as the northernmost edge of the Nubia plate. Geological, seismological and geodetic evidence support the hypothesis that active Eu-Nu convergence has probably shifted from the front of the Plio-Pleistocene nappes to the Southern Tyrrhenian Sea, that previously worked as the transtensional passive margin of the back arc basin [see also Pondrelli et al., 2004]. Site velocities in Calabria suggest that underthrusting of the Ionian lithosphere may still occur, although back arc extension in the Tyrrhenian Sea is no longer active. An explanation for these distinct behaviors probably resides in the different structure of the lithosphere approaching the plate boundary zone, well imaged by the difference in bathymetry across the Malta escarpment (Figure 1). The deep and negatively buoyant Mesozoic oceanic lithosphere [Catalano et al., 2001] can be easily subducted beneath the Calabrian Arc. The lack of evidence for active spreading, when compared with Neogene-Quaternary estimates of 50–70 mm/yr of back arc extension, is indicative of a substantial recent deceleration of the rollback of the Ionian lithosphere. The reason for the rollback slowing down is not clear but one can speculate that the episodic nature of back arc extension, the interaction with the 670 km discontinuity or the progressive narrowing of the incoming oceanic lithosphere may have been responsible for rollback deceleration [Faccenna et al., 2001b]. The continental nature and the progressive buoyancy of the lithosphere approaching the subduction front in Sicily, together with the resistance exerted by the rigid Hyblean block, may have induced the Eu-Nu convergence to jump north of Sicily in the Southern Tyrrhenian zone. In this area the contrast in elevation with the low-lying Tyrrhenian basin and the preexisting structures may have favored the concentration of deformation taking up the Eu-Nu convergence. We observe that this pattern is common to other regions along the Eurasia-Nubia plate boundary, such as the northern African coast where active convergence is currently taken up along the former Neogene passive margin separating the Algero-Provencal basin in the north from the Tell and Atlas thrust and fold belt to the south [Morel and Meghraoui, 1996]. To the first-order, the style of Eu-Nu convergence in western Sicily does not require intermediate crustal blocks or widely distributed deformation, since the earthquake slip vectors in the Southern Tyrrhenian zone may directly reflect the Eu-Nu relative kinematics. An additional intermediate crustal block, that can be interpreted in terms of slip partitioning of the oblique convergence in the Calabrian Arc or as an independent Ionian block, is instead required by GPS data in Calabria. Following the first interpretation we speculate that the oblique convergence forces the Calabrian sliver to move semirigidly toward NE with respect to Eurasia, inducing extension at its trailing edge in the Messina Straits. We also speculate that deceleration of slab roll-back may have forced slip partitioning in the overlying forearc, producing NNE motion of the Calabrian Arc and rifting in the Messina Straits, cutting at high angle across the preexisting thrust structures [Monaco et al., 1996]. Stratigraphical constraints on the initiation of rifting in the Messina Straits may provide important data to verify this hypothesis.
 Strain analysis and style of faulting show that NW-SE right-lateral shear in the Aeolian Islands and extension in the Messina Straits dominate the tectonic regime between Sicily and the Calabrian Arc. These observations, however, do not explain the NE motion of the Sicilian corner away from its mainland, perpendicularly to the Eu-Nu convergence vector. This motion is consistent with normal-faulting low-magnitude earthquakes in northeastern Sicily, with NE-SW directed extension direction [Pondrelli et al., 2004]. A first hypothesis may include the lateral extrusion and sideways expulsion perpendicular to the relative plate motion, of the northeastern part of Sicily away from the collision zone. We also speculate that others possible mechanisms may also include the perturbing effects exerted by the mantle anomaly beneath the Etna volcano on crustal motion. Mantle upwelling beneath eastern Sicily has been suggested to explain the intense magmatism and the evolution of the Etna volcano [Tanguy et al., 1997], and has been proposed from seismic tomography analyses [Montelli et al., 2004].
 Geological data [Bordoni and Valensise, 1998] show that the 125 kyr shorelines (stage 5e of the oxygen isotopic scale commonly called “Tyrrhenian”) is regionally uplifted in an area centered over the Messina Straits with elevations decreasing radially away from eastern Sicily and the Etna volcano. Since no evidence support active crustal thickening and magmatic underplating which may have contributed to isostatically raise the surface, and seismic evidence point for a shallower Moho beneath Etna [Nicolich et al., 2000], we speculate that the recent uplift in eastern Sicily may be of dynamic origin and related to mantle upwelling at the western edge of the Ionian slab. Convective flow in the mantle produces vertical deformation of the surface induced by the normal stresses applied at the base of the lithosphere. This dynamic uplift may have increased the gravitational potential energy of the elevated region. The most favorable direction in which the gradient of potential energy is likely to drive deformation is perpendicular to the convergence vector and away from the rigid Sicilian block. Our hypothesis thus emphasizes a combination of factors related to the kinematics and geometry of the Eu-Nu boundary and to the contribution of upper mantle dynamics to crustal motion. The understanding of the relative importance of these factors is currently difficult to evaluate and will require additional seismic and geodetic information on the upper mantle structure and the crustal velocity field.
6.3. Implications for Seismic Hazard
 Our results suggest that up to 80% of the 3.6 ± 0.6 mm/yr relative motion between the Sicilian and Calabrian blocks may be absorbed in the Messina Straits loading the fault responsible for the 1908 Mw 7.1 earthquake. Geological estimates of horizontal slip rate along the Messina fault of about 1 mm/yr [Valensise and Pantosti, 1992] may indicate that (1) a significant deformation occurs by aseismic slip, (2) other faults are required to accommodate this relative motion, or (3) biases in the geological and/or geodetic estimates. Beyond the 1908 Messina seismic event large historical earthquakes are documented in eastern Sicily, such as the devastating 1693 earthquakes [Boschi et al., 1995], whose fault geometry and location is still a matter of debate [Bianca et al., 1999; Piatanesi and Tinti, 1998]. Our kinematic reconstruction allows us to recognize two possible mechanisms for large earthquakes in this region. The first is the ESE motion of the Calabrian Arc with respect to Sicily. We consider this motion responsible for active extension in the Messina Straits and expect it to extend south of the Straits inducing extensional strain along the eastern coast of Sicily [Monaco et al., 1997]. The other mechanism is the progressive tearing of the subducting Ionian oceanic lithosphere from the adjacent continental lithosphere in the Hyblean Plateau [Doglioni et al., 2001]. Seismic reflection lines in the Ionian Sea offshore eastern Sicily [Argnani et al., 2002] show that the fault system forming the Malta escarpment is frequently associated with deformation of recent sediments and disruption of the ocean floor, suggesting the possible reactivation of this structure during catastrophic earthquakes such as the large 1693 seismic event. GPS velocities in eastern Sicily are consistent with low strain rates across the area of the Mw 5.9 1968 Belice earthquake (between NOTO and TRAP-MILO) and higher strain rates north of Sicily where most of the Eu-Nu convergence seems to be accommodated. An important issue for seismic hazard assessment is connected to the mode of seismic release of the Eu-Nu shortening taken up in north of Sicily and in the Ionian wedge where rates of 4–5 mm/yr have been estimated. Given the low-level of the seismic deformation in northern Sicily and in the Ionian Sea, a considerable deficit of seismic release is found in both areas, which can be released in future events.
 The results of this work suggest that two different styles of deformation accommodate the Eurasia-Nubia convergence in Sicily and in the Calabrian Arc. In Sicily geodetic and seismological evidence suggest that 4–5 mm/yr of Eu-Nu convergence are principally accommodated to the north in the Southern Tyrrhenian zone with recent cessation of outward (south) motion of the thrust sheets along the Plio-Pleistocene subduction front. In the Calabrian Arc residual velocities with respect to both the Eurasia and Nubia plates are consistent with the presence of a forearc sliver in the Calabrian Arc partitioning 5 mm/yr of oblique convergence in 2 mm/yr of trench-parallel and 3 mm/yr of trench-perpendicular velocities. An area of diffused deformation, which includes the northeastern corner of Sicily, the Aeolian Islands and the Messina Straits, accommodates the relative motion between Sicily and the Calabrian Arc by right-lateral shear, normal faulting and clockwise rotations. We show that (1) available information on the source geometry of the Mw 7.1 1908 Messina earthquake are compatible with the accommodation of this relative motion and that (2) up to 3 mm/yr (80% of the relative motion) can be taken up in the Messina Straits. Statistically insignificant elongation of the baselines crossing the Tyrrhenian Sea do not support active back arc spreading consistently with the low level of crustal seismicity in the Tyrrhenian basin. The alternative hypothesis of an independent Ionian block, intermediate between the Nubia plate and the Adriatic block is currently difficult to be tested, and awaits further information to be evaluated.
 We suggest that the different style of deformation along the plate boundary are caused by the different lithospheric structures: the Mesozoic oceanic lithosphere may still be subducted beneath the Calabrian Arc or behaving as an independent Ionian block, whereas the continental lithosphere approaching the plate boundary in Sicily resists subduction and transmits the Nubia motion to the Southern Tyrrhenian zone where most of the convergence is accommodated. Future geodetic studies and densification of the permanent network are needed to provide information on specific topics of the active deformation that have significant implications for the seismic hazard. One of these topics is the understanding of the mode of active shortening in those regions (such as the Ionian wedge and the Southern Tyrrhenian zone) that, presenting a significant deficit of the seismic release, may potentially accommodate part of the convergence by aseismic creep or slow deformation.
 We thank the following institutions for facilitating the availability of GPS data: ASI (http://geodaf.mt.asi.it), SOPAC (http://sopac.ucsd.edu), UNAVCO (http://www.unavco.org) and EUREF (http://www.epncb.oma.be). We also thank Giuseppe Casula, Pino Colucci, Salvatore Gabriele, Alessandro Galvani, Enzo Mantovani and Marco Mucciarelli for Rinex data. N.D. acknowledges the Fulbright Commission for fellowship support at the Jet Propulsion Laboratory in Pasadena in 2000–2001. Fruitful discussions with James Jackson, Stefano Sylos Labini, Fabrizio Antonioli, and Gianluca Valensise have been found useful. We also thank the Associate Editor Yehuda Bock and the reviewers Rob Reilinger and Shimon Wdowinski for useful comments and criticisms. This work is partially funded by the Civil Protection GNDT project “Probable earthquakes in Italy between the year 2000 and 2030: guidelines for determining priorities in risk mitigation”. Figures were produced using GMT version 3.4.