Data from high-density seismic networks deployed between 2000 and 2007 in the north-central Apennines (Italy) yield unprecedented images of an active orogenic wedge. Earthquake foci from the northern Apennines define a Benioff zone deepening westward from the Adriatic foreland down to ~60 km depth below the chain. The seismicity shows that only the lowermost ~10 km of the Adriatic foreland crust is subducted, whereas the uppermost ~20 km is incorporated into the orogenic wedge. Farther west, an aseismic mantle with markedly negative P-wave-velocity (Vp) anomalies is interpreted as asthenosphere flowing toward an Adriatic slab in retrograde motion. Three and seismicity characteristics are imaged below the northern Apennines: an uppermost 10-km-thick fast crustal layers with different Vp layer affected by extensional faulting, a slow layer with diffuse seismicity down to ~15 km depth, and a lowermost fast and aseismic layer resting directly above the asthenosphere. We interpret the latter layer as having formed by anhydrous crust undergoing granulitization, whereas trapped CO2 (either from the underlying granulites or from the subducting Adriatic crust) is inferred to have been responsible is released along the easternmost normal fault systems breaking for both low Vp and diffuse seismicity in the middle crust. Trapped CO2 the Apennine upper crust, consistent with geochemical and seismotectonic evidence. Compressive earthquakes at 20–25 km depth along the external front suggest offscraping of the subducting foreland crust and show that asthenospheric flow represents the primary source of ongoing shortening along the belt front.


Earthquake depth distribution has classically represented the main geophysical proxy to infer the structure and rheology of the continental crust. About three decades ago a consensus developed that the continental lithosphere is composed of a weak lower crust sandwiched between the strong (and seismogenic) mantle and upper crustal layers (“jelly-sandwich model,” Chen and Molnar, 1983). Recently, Maggi et al. (2000) reanalyzed the depth distribution of continental earthquakes and showed that the seismogenic layer consisted of the entire crust, with no convincing evidence of uppermost mantle seismicity. Accordingly, they suggested that the strength of the continental lithosphere resided entirely in its uppermost crustal shell (see Jackson et al., 2008, for a review).

Crustal rheology plays a key role in the geometry of orogenic wedges during plate collision and continental subduction. Seismicity distribution and seismic-wave-velocity anomalies from several mountain belts (i.e., the Himalayas, New Zealand) have undoubtedly imaged the progressive subduction of the continental lithosphere beneath orogens (Schulte-Pelkum et al., 2005). Yet, crustal and mantle imaging has not been able so far to discriminate the behavior of the various crustal layers descending into the mantle. Still unresolved key questions are: How much of the continental crust does remain coupled to the mantle and undergoes subduction? At what depth is the basal detachment yielding the progressive offscraping of the lower plate and incorporation of crustal slices into the orogenic wedge?

The joint interpretation of high-resolution earthquake hypocenters and velocity models may significantly contribute to resolving such issues. We present here original evidence for the north-central Apennine fold and thrust belt (Italy), which represents an ideal study area owing to the large, high-quality seismicity data recently collected by the Istituto Nazionale di Geofisica e Vulcanologia (INGV), using a large number of permanent and temporary seismic stations (Fig. 1). These data offer the opportunity for imaging the deep structure of a mountain belt with a previously unattainable resolution, and to verify the coupling between crustal and mantle continental material entering a subduction system.


The Apennine belt has developed by the eastward migration of an extensional-compressive pair starting in late Miocene time (e.g., Patacca et al., 1990). This migration has been associated with the eastward rollback of an Adriatic slab (Malinverno and Ryan, 1986; Faccenna et al., 2001; Rosenbaum and Lister, 2004), yielding subcrustal earthquakes (down to 70–90 km depth, according to Selvaggi and Amato, 1992) and positive P-wave anomalies traceable in the upper mantle beneath the northern Apennines (Lucente et al., 1999; Piromallo and Morelli, 2003).

At the rear of the migrating wedge the Liguro-Provençal Basin and the Tyrrhenian Sea backarc basins opened in Oligocene–mid-Miocene and Tortonian–Pleistocene times, respectively. The Tyrrhenian margin of the northern-central Apennines (Tuscany and Latium, Fig. 1) was similarly characterized by Pliocene–Pleistocene extension, a thinned crust, high heat flow, and widespread magmatism (Martini and Sagri, 1993; Serri et al., 1993; Jolivet et al., 1998). P- and S-wave attenuation in the mantle suggests the rising of the asthenosphere to as high as 30 km depth beneath the western-central side of the northern-central Apennines (Mele et al., 1997).

The central Apennines display a thicker crust (Mele et al., 2006; Di Bona et al., 2008; Piana Agostinetti and Amato, 2009) and a higher relief than the arc-shaped northern Apennines. In the central Apennines, thrust-sheet emplacement ceased at about the Pliocene-Pleistocene boundary (1.5–2 Ma ago; Cavinato and De Celles, 1999; Patacca et al., 2008), whereas in the external (Adriatic) side of the northern Apennines, shortening is still active today, as testified by compressive earthquakes recorded during the last thirty years (e.g., Chiarabba et al., 2005).

The trend of the Apenninic orogenic wedge and the subduction zone is oblique with respect to the overall E-W trend of the Africa-Eurasia margin, and this may explain both the complexity of the Apennine orogenic architecture and the presently low subduction rate in the northern Apennines, as suggested by the mild deformation of Pleistocene strata (Tozer et al., 2006) and recent Global Positioning System (GPS) data (Devoti et al., 2008). The reported amount of shortening accommodated by the northern Apennine belt varies greatly between several hundreds and a few tens of kilometers, depending on the assumed thrusting style (thin- or thick-skinned tectonic models of Bally et al., 1986; Coward et al., 1999; Butler et al., 2004).

As a consequence of the poor resolution of middle-lower crustal features imaged by deep seismic profiles (CROP03 and CROP11 crossing the northern and central Apennines, respectively; e.g., Barchi et al., 1998a, 1998b; Patacca et al., 2008), the depth of the basal detachment and the amount of shortening are still strongly debated (e.g., Bigi et al., 2003). In thin-skinned tectonic models this basal décollement is placed within the Triassic evaporites, and only the shallow Mesozoic-Cenozoic sedimentary cover is incorporated in the orogenic wedge (Bally et al., 1986). Conversely, thick-skinned models postulate thrust faults that cut the whole Apennine crust (see Barchi et al., 1998a; Coward et al., 1999), consistent with the modeling of magnetic anomalies from the northern Apennines (Speranza and Chiappini, 2002).

The internal sectors of both the northern (Tuscany) and central Apennines (Latium, Fig. 1) were affected by extensional tectonics since late Miocene time. Jolivet et al. (1998) and Collettini et al. (2006) proposed that northern Apennines extension is mainly accommodated by a set of subparallel east-dipping, low-angle normal faults, of which only the easternmost (the Altotiberina fault) is presently active. This east-dipping, low-angle (~20°) fault is traceable down to ~12 km depth by seismic reflection data and microseismicity (Barchi et al., 1998b; Boncio et al., 2000; Chiaraluce et al., 2007). The present-day extension of the northern Apennines is accommodated by both the Altotiberina fault and an antithetic array of west-dipping normal faults (Fig. 1). Destructive earthquakes (e.g., the 1997 Umbria-Marche sequence, Chiaraluce et al., 2004) have occurred along the Apenninic divide on the antithetic steps.

The presence of high-pressure crustal fluids is considered a primary factor that has controlled the anomalously low dip of the Altotiberina normal fault, as well as unusual characteristics of the 1997 Umbria-Marche seismic sequence (Miller et al., 2004). This is consistent with the diffuse CO2 measured in the inner and axial northern Apennines (Chiodini et al., 2004). Relying on geochemical evidence, Chiodini et al. (2000) argued that the thermometamorphism of shallow limestones is not enough to explain the nature of the very high CO2 flux and that a metasomatized mantle source is needed.


In this study we used P- and, subordinately, S-wave arrival times from ~8000 local earthquakes recorded by a total of 357 permanent and temporary seismic stations. Body wave arrival times from the local seismicity recorded by both the regional Marche-Abruzzi seismic network (see De Luca et al., 2009) and the national seismic network are complemented by temporary data from the aftershock sequence of the 1997 Umbria-Marche sequence and the Città di Castello 2000–2001 and Abruzzi 2003–2004 passive experiments (see Chiaraluce et al., 2004; Piccinini et al., 2003; Bagh et al., 2007). The high-quality merged data set results in an unprecedented resolution for the seismicity in this sector of the Apennines.

We first located the earthquakes with the Hypoellipse code, a one-dimensional (1-D) velocity model taken from the literature and a Vp/Vs ratio equal to 1.8 (Chiarabba and Amato, 2003; Chiarabba et al., 2005). Then we applied a relative-location procedure (HypoDD, Waldhauser and Ellsworth, 2000) to a selection of data: 4816 events, which have hypocentral errors <2 km, an azimuthal gap <180°, and a minimum of 20 phases. We tuned the hypoDD parameters to achieve the best convergence toward a global minimum, following Waldhauser and Ellsworth (2000). At the end of the location procedure, the residual weighted root mean square (rms) is equal to 0.04 s with an improvement of 60%. Both in the 1-D absolute and relative locations we used S-wave arrivals to constrain the hypo-central parameters.

Figure 2 shows that most earthquakes are concentrated in the uppermost 10 km and along the normal-faulting seismic belt close to the Apennine divide (see Chiarabba et al., 2005). In the middle crust (10–15 km depth) the seismicity is sparse, whereas it is clustered in the lower crust, at 20–25 km depth, in the eastern side of the Apennines. A few earthquakes with depth >30 km are represented, and they deepened westward down to ~60 km depth. This hypocentral distribution ensures an adequate sampling of the crust and the uppermost mantle with vertical seismic rays.

Finally, we used the selected earthquakes for the three-dimensional (3-D) P-wave tomography, using the inversion technique developed by Zhao et al. (1992, 1994). This technique is highly appropriate to image velocity heterogeneity at a local/regional scale, thanks to the use of an approximate ray tracer that can model the refracted first arrivals by the Moho (see Zhao et al., 1994, for details). To better constrain the inversion, we use the hypocenters obtained by the relative location procedure as starting locations. In the inversion we used 111,376 P-wave arrival times. Because of the small number of available S-wave arrivals, we did not invert for the Vs model. The model is parameterized with a continuous velocity, defined at nodes of a 3-D grid, and with a flat Moho discontinuity at a constant depth of 30 km; i.e., the inferred average depth of the Adriatic Moho before undergoing subduction (e.g., Scarascia et al., 1994). Nodes are spaced 0.15° in latitude and longitude. Four layers are located in the crust at 4, 8, 12, and 20 km depth, and three in the uppermost mantle at 32, 42, and 52 km depth. Velocity values are laterally homogeneous and taken from the 1-D velocity model used in the first location step.

After three iterations we found a final unweighted rms of 0.3 s with a variance improvement of 61%. Hypocentral location variations from the 1-D starting values are small, ensuring that the choice of not using S-wave arrivals has negligible effects on the Vp model determination.


The reliability of the velocity model was verified with synthetic tests. We traced seismic rays through previously known structures: a classical checkerboard and large-scale positive or negative velocity bodies. In the checkerboard test, positive and negative anomalies of 5% are alternated every two nodes in latitude and longitude, and every layer in depth.

Synthetic data are inverted, after the addition of a Gaussian noise with standard deviation equal to the final variance of the inversion, using a laterally homogeneous velocity model. Figure 3 shows how well the synthetic features are reproduced by our data. Good resolution is observed in the crustal layers, whereas the paucity of NE-directed rays partially degrades the resolution in the uppermost mantle, predominantly sampled by Pn arrivals. In the mantle the prevalent ray geometry may generate spurious NW-elongated structures. To further investigate the resolution power, and to test for the observed features specifically, we used synthetic low or high Vp volumes, located in layers at 22 and 42 km depth.

In the second test we introduced a high and a low Vp body in the lower crust and uppermost mantle, respectively. Figure 4 shows that our data resolve these large-scale anomalies well, at least in the inverted portion. In the lower crust, the synthetic high Vp feature is well reproduced for almost the entire layer, even with incomplete amplitude recovery, indicating the reliability of both the presence and geometry of the high Vp anomalies imaged by the real inversion (see next section). The synthetic anomaly in the mantle is well reproduced both in amplitude and in geometry.


For the time period considered, a diffuse crustal seismicity characterizes the northern Apennines, whereas the central Apennines are virtually aseismic (Fig. 1). The deep (>30 km) seismicity also occurs along a narrow northern Apennine sector, at about 43°–44°N latitudes. Most of the shallow seismicity (and specifically the foci of the 1997 extensional Umbria-Marche sequence) occurs within a ~10-km-thick upper-most crustal layer displaying positive Vp anomalies (Figs. 5 and 6). The shallow fast layer overlies markedly slow crust, showing diffuse seismicity at ~10–15 km depth. This slow crust rests in turn above a second, poorly seismogenic and fast lower crust layer, retrievable down to ~25 km depth. Both lower layers are most prominent in the northern Apennines, whereas the internal central Apennines invariantly display negative anomalies below 10 km depth. The lower fast layer shows an interruption along the axial belt, where the overlying slow layer displays the lowest Vp values of 5.8 km/s, as well as a hypocenter concentration. Finally, deep crustal (20–25 km) hypocenter clusters (yielding predominantly compressive focal mechanisms with E-trending P-axes; see next section) are solely observed at 43°N, along the most external belt sector.

The deepest (30–60 km) seismicity occurs in a narrow, low Vp zone deepening westward below the northern Apennine chain (Fig. 6), and located up to ~10 km above the westward-plunging Adriatic Moho, as documented by Piana Agostinetti et al. (2002); Mele et al. (2006), Roselli et al. (2008), and Piana Agostinetti and Amato (2009). Farther west the northern Apennine mantle is virtually aseismic and is characterized by low (7.5 km/s) P-wave velocities. In the central Apennines the negative Vp anomaly at 30–40 km beneath the belt is mostly contained in the crust.


The existence of a narrow and shallow domain undergoing normal faulting along the axial part of the Apennines is well documented by focal mechanisms, borehole break-outs (Montone et al., 2004), and the analysis of recent seismic sequences (Chiaraluce et al., 2004). Conversely, the kinematics of the earthquakes from the deep (>15–20 km) Apennine-Adriatic crust is still poorly understood. Although deep crustal events are abundant, their small magnitude results in a small number of events for which we have a significant number of P-wave polarities. For a total set of 80 events, selected with small hypocentral errors and having more than 15 observations, focal mechanisms were computed using the P-wave polarity and the Fpfit program (Reasemberg and Oppenheimer, 1985).

We obtained well-constrained solutions for 68 events (Fpfit quality factors, Qf and Qp = A or B). At 20 km depth the P-axes are prevalently subhorizontal and E to NE trending (Fig. 7), indicating compression at the front of the deep northern Apennine wedge. Conversely, the focal solutions of earthquakes occurring at depths >25 km within the subducting Adriatic lower crust show subhorizontal NW-trending P-axes and E-NE–trending subhorizontal T axes. This stress regime is similar to that observed throughout the Adriatic foreland and within sectors of the Adriatic lithosphere lying below the Apennine belt (e.g., Di Luccio et al., 2005; D'Agostino et al., 2005; Serpelloni et al., 2007). This evidence indicates that a rather uniform regional stress is transmitted to the Adria plate, even to some fragments of the Adriatic slab now undergoing subduction beneath the peri-Adriatic orogen.


The deep hypocenters from the northern Apennines trace a Benioff seismic zone that dips westward beneath the chain at 43°–44°N. The seismicity is entirely concentrated in the lower crust of the subducting Adriatic plate, within a zone of moderately negative Vp anomaly (Fig. 6). The negative Vp anomaly is consistent with the presence of subducting crustal material, as observed in the shallow parts of other subducting systems (see Eberhart-Phillips and Bannister, 2002). If the deep foci are assumed to document the thickness of the crust undergoing subduction and progressive eclogitization, it follows that ~10 km of the Adriatic lower crust is subducted along with the underlying mantle lid (Fig. 8). Geochemical and petrological evidence from upper Miocene to Pleistocene igneous bodies from the northern Tyrrhenian Sea and Tuscany similarly requires the subduction of crustal material within the mantle (Serri et al., 1993). Thus the Adriatic lower crust may have undergone subduction since at least late Miocene time—i.e., during the entire northern Apennine–Tyrrhenian Sea genesis.

The hypocenters from the Benioff zone do not exceed ~60 km in depth. Their shallow disappearance, when compared to other subduction zones, may be related to both (1) the low subduction rate, as inferred by both poor (if any) compressive deformation recorded by Pleistocene strata from the external northern Apennines (Tozer et al., 2006) and recent GPS evidence (Devoti et al., 2008), and (2) the subduction of only the lowest (and hottest) 10 km of the Adriatic crust. By considering the average thickness of the Adriatic crust in the foreland (~30 km), and a conservative geothermal gradient of 20°/km, the hot and slowly subducting lower crust could well reach temperatures of 600–700 °C at ~60 km depth. At such temperatures ductile creeping is expected to overcome friction, justifying the seismicity cutoff.

The Adriatic lower crust is definitely not a weak decoupling layer, but it seems to be rigid and dense enough to remain coupled to the underlying subducting mantle. This is suggestive of anhydrous granulite-facies rocks of predominantly mafic composition, as put forward by Rudnick and Fountain (1995) and Maggi et al. (2000). The focal mechanisms for events deeper than 25 km in the subducting Adriatic lower crust indicate NW shortening, suggesting a uniform regional stress regime acting along both foreland and subducted fragments of the Adria plate (Serpelloni et al., 2007).

At ~43°N, compressive earthquakes at 20–25 km depth indicate shortening in the external Apenninic belt and the progressive offscraping of the middle to upper Adriatic crust and its incorporation into the orogenic wedge (Figs. 1, 5, and 6). Similarly, deep compressive earthquakes also have been documented for a broader external northern Apennine domain (Chiarabba et al., 2005), suggesting ongoing orogenic accretion along the whole northern Apennine arc.

West of the subducting plate the virtually aseismic and low Vp northern Apennine mantle documents the asthenosphere flowing toward the Adriatic slab in retrograde motion (Fig. 8). Thus the mantle beneath the Apennines is not made of stacked tectonic slices of Adriatic mantle but rises from an asthenospheric upwelling. Consequently, a new Moho is forming beneath the Apennines, in agreement with Doglioni (1992). According to Piana Agostinetti et al. (2002), Mele et al. (2006), and Bianchi et al. (2008), the crust here is 20–25 km thick. This thickness is fully consistent with the evidence that only ~10 km of the 30-km-thick Adriatic crust is subducted, whereas the uppermost ~20 km is incorporated into the Apenninic wedge.

The poorly seismogenic and high Vp lower crust of the northern Apennines directly overlies the hot (likely ~1200 °C) asthenosphere; thus the anomaly necessarily must be associated with anhydrous rocks undergoing progressive granulitization (the presence of water would necessarily imply widespread melting). Within the slow layer at 10–15 km depth, diffuse seismicity concentrates in the domain with the lowest Vp values, and this in turn is just below the normal faults that yielded the 1997 Umbria-Marche seismic sequence (Figs. 5 and 6). For this fault system, as well as for the east-dipping, low-angle Altotiberina fault, CO2 at near-lithostatic pressure was recognized to have played a key role in the control of both fault geometry and seismicity characteristics (Miller et al., 2004; Collettini et al., 2006). Relying on fault analogues in evaporitic rocks, Collettini et al. (2008) proposed a quantitative model of fault weakening in the northern Apennines by CO2 overpressure episodes. Low Vp/Vs anomalies identified by local earthquake tomography are consistent with the presence of a dehydrated rock volume around the Altotiberina fault that acts as an impermeable barrier to the uprising fluids (Moretti et al., 2009).

We therefore interpret the low Vp crustal layeras containing trapped CO2 yielding seismicity related to hydrofracturing. We cannot rule out the additional presence of other fluids, though again “wet” crust at an estimated temperature of 300–500 °C should not release seismicity. We suggest that CO2 trapped at 10–15 km depth may reach the surface only along the normal faults that cut the upper crust in Tuscany and along the axial northern Apennines as far east as the presently active normal fault system. Here, the most vigorous gas upwelling occurs, as testified by lowest Vps and highest mid-crustal seismicity (Figs. 6 and 7). CO2 is locally trapped beneath the presently stretching seismic belt and helps in promoting seismicity (see Miller et al., 2004; Moretti et al., 2009), whereas the absence of crustal extension inhibits the rise of CO2 farther east, consistent with geochemical evidence (Chiodini et al., 2004).

In principle, CO2 from the northern Apennines might be released either by the heated middle-lower crust undergoing granulitization or the subducting Adriatic crust (Fig. 8). In the first case, CO2 would be released by Mesozoic carbonates incorporated in the lower crust of the northern Apennine wedge and heated by the underlying asthenosphere, whereas in the second it would arise from the metamorphism of the subducting lower crust of Adria, the nature of which is, however, speculative (Vai, 1994; Patacca et al., 2008). In any case there is a clear correspondence between the areas exhibiting high CO2 release (e.g., Chiodini et al., 2004) and the existence of an ongoing subduction, as indicated by hypocenters deeper than 30 km.

The central Apennines turn out to be completely different for both seismicity and Vp layering (Figs. 1, 5, and 6). Background seismicity is very low and is completely absent at depths >30 km, consistent with geological evidence of cessation of shortening at 1.5–2 Ma ago. The internal chain displays an aseismic, thick (30–40 km), and low Vp crust. The crustal structure of the central Apennines may represent a more evolved tectonic stage with respect to the northern Apennines, implying that the progressive heating from the underlying asthenosphere has eventually led to partial melting of the crust. This interpretation is consistent with the lateral fragmentation of the high Vp slab in the mantle and the hypothesis of slab tear as proposed by Rosenbaum et al. (2008).

In the northern Apennines, frontal orogenic wedge accretion occurs at 20–25 km depth, partly above the asthenospheric layer (Figs. 6 and 8). Thus the asthenosphere must play a primary role in inducing shortening along the external Apennines by flowing northeastward and exerting a drag on the overlying crust. This is compatible with the eastward-flowing asthenosphere being the primary engine of the system (Doglioni, 1992), or the downward drag produced by the subducting slab yielding a corner-flow cell, and hence trenchward stress at the base of the lithosphere (Cavinato and De Celles, 1999). Furthermore, the retrograde motion of the Adriatic slab is likely the sum of both passive subduction of the Adriatic slab and northeastward drift of Adria with respect to the internal Apennines as documented by GPS data (D'Agostino et al., 2006).

The CROP-03 seismic reflection profile (Barchi et al., 1998b) suggests a possible connection between ongoing extension along the Altotiberina fault and shortening in the frontal thrust structures, implying a gravitational origin for (at least part of) the shortening occurring along the external structures. Conversely, our data and interpretation (Fig. 8) support the idea that shortening is related to the mantle drag induced by the eastward flowing of the asthenosphere.


The analysis of seismic data recorded by high-density networks deployed in the central-northern Apennines reveals the deep geometry of an orogen developing above a continental subduction zone with a resolution not attained so far. In the northern Apennines, mainly between 43° and 44°N, a 60-km-deep Benioff zone, along with independently estimated Moho depths, indicates that only the lowermost ~10 km of the Adriatic foreland crust is subducted, mechanically coupled with the underlying mantle lid. The compressive focal mechanisms of earthquakes from the subducting Adriatic crust, coherent with those occurring elsewhere in the Adriatic lithosphere, suggest that a rather uniform stress is transmitted along both the foreland and slabs of the Adria plate presently subducting beneath the Mediterranean Alpine belt. Offscraping of the uppermost 20 km of the foreland crust and its incorporation into the northern Apennine wedge are testified by the compressive foci with E- to NE-trending P-axes clustering at 20–25 km depth beneath the external northern Apennines. Farther west an aseismic northern Apennine anomalies is mantle with markedly negative Vp interpreted as asthenosphere flowing toward the Adriatic slab in retrograde motion.

Three crustal layers with different Vp and seismicity characteristics are imaged below the northern Apennines: an uppermost 10-km-thick fast layer affected by extensional faulting, a slow layer with diffuse seismicity down to ~15 km depth, and a lowermost fast, aseismic layer. Thus seismic evidence yields the following insights on both the rheology and characteristics of the active continental subduction zone:

  • The lower crust of the Adriatic foreland is not a decoupling layer, but it is rigid enough to remain coupled to the underlying mantle undergoing subduction.

  • The lowermost high-Vp crust of the orogen is likely composed of anhydrous crust undergoing granulitization, resting directly above the asthenosphere.

  • None of the earthquakes occurs in the continental mantle (of either the subducting plate or the orogenic wedge). The whole pattern of seismicity distribution, Vp geometry, and inferred location of the basal orogenic detachment indicates that a weak layer prone to tectonic decoupling may reside in the middle rather than the lower crust, as suggested in the past.

  • The broad, low-Vp region associated with sparse seismicity documented in the middle crust is interpreted as a CO2-saturated layer. The lowest Vp values are observed just below the active normal fault belt of the uppermost 10 km, suggesting a close relation between fluid-filled volumes and the clustering of earthquakes in the uppermost crust. This is consistent with the model of the 1997 Umbria-Marche seismic sequence proposed by Miller et al. (2004), where the unusual chronology and magnitude of the aftershocks is related to the migration of high-pressure CO2 at depth. It is unclear whether CO2, trapped in the mid-crustal layer, comes from the subducting lower crust of Adria (which is speculative) or from Mesozoic carbonate slices detached from the foreland and stacked in the high-Vp lower crustal layer of the orogen. In any case the close correspondence between the domain undergoing high CO2 surface degassing (Chiodini et al., 2004) and the sectors of the orogen characterized by ongoing subduction is clear, and points to a first-order control of the CO2 generation and rise by the continental subduction process.

We thank Editor Ray Russo for his careful evaluation of this work. E. Sandvol and two anonymous reviewers provided thoughtful comments on this manuscript. Finally, we thank all those who worked hard in the management of temporary and permanent networks, and also A. Amato, N. Piana Agostinetti, and R. Di Stefano for discussion.