We performed a series of fieldwork campaigns, at 1:25000 scale, in the study area and surrounding sectors, to collect fault-slip data to be integrated with the geological–structural observations reported in Brozzetti et al. (2017a). In addition to traditional survey methods, we used the Fieldmove App (PetEx Ltd., version 2019.1) installed on a tablet computer to acquire the data in the field (e.g., Allmendinger et al., 2017; Novakova and Pavlis, 2017; Testa et al., 2019; Brozzetti et al., 2020; Cirillo, 2020), and we managed them in ArcGIS v.10.8 (ArcMap©). Figure 2 shows the location of the survey sites, considered structurally homogeneous outcrops falling within a maximum distance of 500 m (see also Fig. S2 in the Supplement). The overall fault-slip dataset was first subdivided into minor and local homogenous kinematic subsets, the latter represented as pseudo-focal mechanisms using FaultKin 8 software (Marrett and Allmendinger, 1990; Allmendinger et al., 2012; Fig. 3). The fault-slip data were subsequently inverted (see following Sect. 4.3).

To better characterize the 3D features of the tectonic structures located in the study area, we performed a high-quality hypocenter location. We enlarged, with respect to previous works by Totaro et al. (2013, 2015) and Brozzetti et al. (2017a), the time window for earthquake analyses (i.e., January 2010 and October 2018), selecting earthquakes with local magnitude greater than 1.0 and hypocentral depth range 0–30 km from the INGV and the University of Calabria database (https://www.ingv.it/, last access: 19 April 2021; http://www.sismocal.org, last access: 19 April 2021). Automatic and manually revised P- and S-wave arrival time picks have been selected for this dataset. The recording network, including both temporary and permanent stations managed by the University of Calabria and INGV (D'Alessandro et al., 2013; Margheriti et al., 2013), consisted of 61 stations with a maximum epicentral distance of 150 km (Fig. S1 in the Supplement). We computed accurate absolute hypocenter locations by applying first the non-linear Bayloc earthquake location algorithm (Presti et al., 2004, 2008) and subsequently the double-difference relative location method HypoDD (v.2; Waldhauser, 2001) and using the 3D velocity model by Orecchio et al. (2011). The Bayloc algorithm gives for each earthquake a probability density cloud with shape and size related to the main factors involved in the location process (e.g., network geometry, picking errors) and allows a generally more accurate estimate of hypocenter parameters and location uncertainties with respect to the more commonly used linearized location methods (e.g., Lomax et al., 2000; Husen and Smith, 2004; Presti et al., 2008). The application of the Bayloc algorithm provides, on average, horizontal and vertical errors of the order of 1.0 and 1.5 km, respectively, allowing us to obtain a well-constrained database. As the second step, we apply the HypoDD algorithm, which minimizes phase delay-time residuals between pairs of events recorded at common stations (Waldhauser and Ellsworth, 2000). We compute the delay times from each event to its 30 nearest neighbors within 10 km distance, and to further ensure the robustness of the double-difference inversion, only event pairs with at least eight phases observed at common stations were used. The final relocated dataset consists of 3109 events (Figs. 4 and S1). During the decade before the 2010–2014 Pollino sequence, the instrumental data, available within a range of nearly 75 km from the Mercure basin, referred to background seismic activity (Frepoli et al., 2005; Castello et al., 2006; Brozzetti et al., 2009). A significant seismic activity which affected the region was the moderate-magnitude 1998–1999 Mercure sequence that developed in the northern part of the homonym Quaternary basin (Fig. S1; Guerra et al., 2005; Arrigo et al., 2006; Brozzetti et al., 2009) and showed some similarities to the recent Mercure-Pollino sequence (e.g., prevalent kinematics of focal mechanisms and hypocentral depth range). We explored the data available for this seismic activity to compute a high-quality earthquake location, following the procedure described above for the 2010–2018 earthquake dataset. Since the recording network operating during the 1998–1999 seismic phase was significantly different from today, in terms of the number of stations deployed in the region and their spatial distribution, the available data do not allow the high level of constraint needed to perform the 3D structural model reconstruction to be reached.

To investigate the coherence between the geological and the seismological stress fields, we applied stress tensor inversions to the available fault-slip data (Figs. 2, 3) and focal mechanisms (Fig. 4). We used the “TENSOR” program and the inversion procedure proposed in Delvaux and Sperner (2003). We applied it separately on the different datasets. The procedure computes the orientation of the three principal axes of the stress ellipsoid (σ1, σ2, σ3) and the stress ratio Φ=(σ2-σ3)/(σ1-σ3) that optimize the misfit function (i.e., F5 in the “TENSOR” program, described as f3 in Delvaux and Sperner, 2003). The latter is built (i) to minimize the slip deviation between the observed slip line and resolved shear stress (30∘ misfit value is not expected to be exceeded) and (ii) to favor higher shear stress magnitudes and lower normal stress to promote slip on the plane. The inversion procedure provides for the preliminary (kinematic) analysis of data using an improved version of the Right Dihedron method (Angelier and Mechler, 1977) to determine the starting model parameters (e.g., the reduced stress tensor). The stress ellipsoid is then computed through a 4D grid-search inversion involving several runs during which the reduced tensor is rotated around each stress axis with a decreasing range of variability (from ±45 to ±5∘), and the full range of Φ values (0–1) is checked. Each step attempts to find the parameters that minimize the misfit function and that are used as a starting point for the next run (see for details Delvaux and Sperner, 2003).

The geological data input consists of 268 quality-controlled fault-slip data measured in the study area (Figs. 2, 3). During the formal inversion, the same weight value was assigned to each fault. The seismological data input is represented (initially) by both nodal planes of each focal mechanism; afterwards, the plane that is best explained by the stress tensor in terms of the smallest misfit function is considered as the actual fault plane (Delvaux and Barth, 2010). The inverted seismological data are represented by focal mechanisms from Totaro et al. (2015, 2016) and reported in Fig. 4. An exponential weighting factor (corresponding to the earthquake magnitudes) has been assigned to account for the prevailing kinematics of the most energetic events. The final inversion (Fig. 5) includes only the fault and focal planes that are best fitted by a uniform stress field (Gephart and Forsyth, 1984).

Following the methodology defined by the Community Fault Model of Southern California (Nicholson et al., 2014, 2015; Plesch et al., 2014), also applied for recent Italian earthquakes (Lavecchia et al., 2017; Castaldo et al., 2018; Bello et al., 2021a), we obtained the 3DFM of the Pollino area by integrating Quaternary fault mapping (Brozzetti et al., 2009, 2017a; this paper) with a high-quality seismicity dataset (2010–2018), and by using the Move suite software v. 2019.1 (Petroleum Experts Ltd).

In particular, we created several sets of closely spaced transects (distance: 2 km) to cross and sample the seismogenic fault zones in different directions (Fig. 6). The set oriented SW–NE is approximately perpendicular (e.g., Fig. 6a and b) to the ROCS (VCT and RSB). Differently, the set oriented NW–SE is approximately sub-parallel to MPR active faults (e.g., Fig. 6f). A further NNE–SSW-striking set of transects was traced approximately perpendicular to the active fault alignment bounding the study area to the east, which includes the CSPT and VPP faults (Fig. 6g and h).

The 3DFM building was carried out following three steps graphically depicted in Fig. 7 and synthetically described below.

Step 1 – Extrusion of fault traces to shallow depth

The traces of the Quaternary faults are “extruded” to a pre-set depth of 2 km b.s.l., according to the fault-plane dip measured in the field. In the absence of measured dip angles, we assumed a fixed value of 60∘. The obtained so-called “fault ribbons” are rimmed upward by the topographic surface (a 10 m resolution DEM; Tarquini et al., 2012).

The computed geological stress tensor (Fig. 5) shows a relevant percentage of fault-slip vector pairs (∼53 %) consistent with a uniform extensional stress field which is characterized by a N244-trending and sub-horizontal σ3. The stress ratio Φ=0.22±0.13, and the rank quality is QRw = A (ranking as in Sperner et al., 2003). Nearly all the kinematic axes related to the inverted data belong to a normal-fault regime, as also pointed out by the triangle in Fig. 5 (Frohlich, 2001).

The seismological stress tensor (Fig. 5b) obtained from inverting 50 actual fault planes (nt: 124 nodal planes) shows a normal fault regime with an ENE-WSW-trending and sub-horizontal σ3 (N062/01 ± 19). The stress ratio Φ=0.52±0.3 and the quality ranking is QRfm = A (ranking as in Heidbach et al., 2010). Most of the nodal planes show normal-fault kinematics (see Fig. 5b).

In both the inversions, a normal-fault regime with sub-horizontal and collinear (∼ SW–NE-trending) σ3 axis has been obtained. This result points out the coherence between the geological (long-term) and the present-day stress field and the persistence of this extensional regime at least since the Middle Pleistocene (Brozzetti et al., 2017a).

In addition, it is worth noting that 76 % of the successfully inverted fault-slip vector pairs are related to the active fault planes belonging to the E- and W-dipping domains (Fig. 5a), while the remaining 24 % include data related to the S-dipping system (CVN and POL). The evidence together with the similarity between the computed stress tensors is consistent with the prevalent activation, in the Late Quaternary, of the E- and W-dipping fault systems.

The 2010–2018 seismic activity in the Pollino–Mercure area followed a peculiar evolution over time (Fig. 4) with epicenters concentrated in a series of neighboring clusters, numbered as Cluster 0, 1, 2, and 3, from west to east, according to their activation time. Such clusters, independent and unconnected to each other, are related to fault segments that are not in an along-strike continuity.

Cluster 0 (30 January 2010–31 July 2011) includes low magnitude (1.0≤ML≤2.9) activity located in an NNE–SSW-oriented sector at the western boundary of the epicentral area. It is delimited westward by the more external segment of the E-dipping CRFS.

Cluster 1 started after 5 October 2011 and lasted for the entire 2011–2014 seismic activity. It extended continuously, both northward and southward, reaching a NW–SE length of ∼12 km (Fig. 4a–c). It comprehends the higher number of earthquakes and is largely the major cluster as regards the wideness (∼60 km2) and energy release. It includes 30 events with ML≥3.0 besides the strongest event of all the Pollino seismic activity (25 October 2012). During the 2015–2018 interval, Cluster 1 area was affected by low seismic activity, mostly distributed in its northern and southern portions; conversely, its central part, where epicenters were particularly dense between 2011 and 2014, became less active. Overall, the surface extent of Cluster 1, which partly overlaps with Cluster 0, is limited eastward by the W-dipping RSB and VCT faults. Its southern boundary nearly coincides with the southeastern continuation of the AVN fault (PAC, Fig. 4c).

Cluster 2 started in May 2012 in the sector between the two WSW-dipping RSB and the MPR faults. It elongates in the N–S direction, for ∼7 km to the northwest of the Morano Calabro town. Afterward, it was nearly continuously active, particularly during the periods May 2012–October 2014 (Fig. 4b, c); in the period 2015–2018, significant seismicity also persisted (Fig. 4d). Cluster 2 includes mainly low-magnitude events besides the strongest ones of 28 May 2012 and 6 June 2014 and three other earthquakes with 3.0≤ML≤3.5.

Further east, in the sector comprised between MPR and the alignment VPP-SDD-CAS faults, a minor cluster of seismicity (Cluster 3) developed since December 2011 (Fig. 4a). Since then (2011–2018), it was affected by poor and low-magnitude seismicity, which, however, was clearly above the threshold of background seismicity, with two ML=3.0 events (Fig. 4a–d).

The obtained 3DFM (Fig. 8), which includes the seismogenic fault system involved during and after the 2010–2014 Pollino seismic activity (CRFS, ROCS, and MPR), also encompasses those faults (GCG, PPS, AVN, BAT, CSPT, VPP, SDD, CAS) that, while showing no direct evidence of recent seismic activity, play a significant role in the seismotectonic frame of the area.

The westernmost fault structures (i.e., GCG and PPS), whose deep geometry is not strictly constrained by subsurface data, have been interpreted according to the structural extensional style proposed by Brozzetti et al. (2017a). The latter is coherent with the reconstructions of the active extensional belt of the southern and central Apennines described in the literature (Barchi et al., 2007; Amicucci et al., 2008; Brozzetti et al., 2011, 2017a, b; Lavecchia et al., 2017). Overall, this style is characterized by an asymmetric extension driven by a low-angle (20 to 35∘) E-dipping detachment fault, which represents the basal decollement of all the other extensional structures. In the model, all the faults are traced at the surface with their dip angle as measured in the outcrop and evolve downward with nearly listric geometries to join the detachment at increasing depth from west to east. The latter represents the structurally controlled base of the seismogenic layer. The GCG (Figs. 1b, 8), which crops out at a low angle and overcomes all the other east-dipping faults (in terms of both slip and associate extension), is the currently inactive break-away zone of such a detachment. The AVN and BAT (Figs. 2, 8), which are the easternmost E-dipping splays, are suggested to be active and seismogenic, being possibly the causative structures of the Cluster 0 of hypocenters (Fig. 4a). Cluster 1 and Cluster 2, which are confined downward by the E-dipping detachment, confirm the activity of the WSW-dipping ROCS and MPR faults, so that we consider them the main geological structures involved during the 2010–2014 seismic activity (Figs. 4 and 8a, a1). Further east, the 3DFM has been widened to include the W-dipping CSPT and VPP faults, considered the outer seismogenic front of the extensional system. The along-strike continuity of POL and CVN is interrupted by the W-dipping ROCS and MPR faults (Fig. 8c, d), coherently with the cross-cut relationships observed in the field (Fig. 2). The deep geometry of POL and CVN is interrupted by the NNE-dipping AVN (Fig. 8d), which acts as the southern and basal boundary of the entire active fault system.

Finally, the 3DFM shows that almost all the 2010–2018 seismicity correlates with the W-dipping structures but without affecting their southern termination zones. In other words, no or very few events are located south of the intersection with POL and CVN faults. This latter observation suggests that although the POL and CVN did not play an active role in causing the considered seismicity, they play a significant role in influencing its distribution.

Coherently with what is observed in most Apennine chains (D'Agostino et al., 2001; Ferranti et al., 2014; Montone and Mariucci, 2016; Mariucci and Montone, 2020), the upper crustal Pollino seismicity develops in response to WSW–ENE-oriented extension. This is well constrained by the focal solutions of the strongest events (Mw5.2, 25 October 2012; Mw4.3, 28 May 2012; and Mw4.0, 6 June 2014 earthquakes) and of all the Mw≥3.5 earthquakes that occurred during 2010–2014, and with the results of the geological and seismological inversion (Fig. 5). Such consistency suggests that the present stress field is in continuity with the long-term one, which has been set up at least since the Early–Middle Pleistocene, as already suggested by previous works (Papanikolaou and Roberts, 2007; Brozzetti et al., 2009, 2017a).

Comparing the distribution of all 2010–2018 seismic activity with the Late Quaternary structures mapped at the surface, we maintain that the ROCS and the MPR faults are suitable as the seismogenic sources for the Mormanno (2012, Mw5.2) and Morano Calabro (2012, Mw4.3 and 2014, Mw4.0) earthquakes, respectively. In addition, our 3DFM allows a parameterization of the sources and their seismogenic potential assessment. The map view of the W-dipping faults (Fig. 9a) depicts irregularly shaped seismogenic boxes which are delimited to the east by the fault traces (at the surface) and to the west by the branch line of each fault with the base of the seismogenic layer. Some of these boxes include historical or instrumental earthquakes (Fig. 9b), while others are not associated with any significant event.

The performed 3D reconstruction allowed us to estimate the effective area extent of all the fault segments (Fig. 9c), that, when inserted in the appropriate scaling relationships, provide the expected magnitude possibly releasable in case of entire rupture (Fig. 9c).

We also computed the magnitude values obtained using the regressions as a function of the surface fault length (Fig. 9c). Using six different empirical relations (Wells and Coppersmith, 1994; Wesnousky, 2008; Leonard, 2010; Stirling et al., 2013), we compared the values determined for all the investigated active normal faults (Fig. 9d, e).

It is evident that, for each fault, the expected magnitude computed using fault area is lower than the one calculated by using fault length. The range of variation is narrower for the values calculated on the ground of fault-area regressions (yellow bars in Fig. 9d, e).

Given the significant difference in the magnitude values computed using area- or length-based scaling relationships, we suggest that (where possible) the reconstruction of a 3D-fault geometry should be pursued and preferred in order to derive more reliable parameters to be used (Table S1). This is even more essential in complex extensional systems such as the one we investigated along the Calabrian–Lucanian border.

In fact, the 3DFM highlights that the areal extension of the W-dipping faults depends on their position within the hanging wall of the detachment (see Sect. 5.3). This implies that faults with comparable length at the surface may have significantly different areas, depending on the depths reached. The CSPT, VPP, and CAS crop out at the greatest distance from the GCG break-away zone. Consequently, they intersect the basal detachment at the higher depth and have the maximum area extent among the W-dipping fault set (Fig. 9a, d).

By applying the aforementioned scaling laws (Fig. 9) to the W-dipping faults identified to be involved during the 2010–2014 seismic activity, we calculated the expected magnitude of ∼Mw=6.1 for the VCT and the RSB, and of ∼Mw=6.2 for the MPR. Since the two faults (RSB + VCT) of the W-dipping ROCS have been interpreted to join at hypocentral depth to form a single structure (thus a unique seismogenic patch was reconstructed – Fig. 10a), a value of ∼Mw=6.4 could be reached in the case of complete and concurrent ruptures on both the segments. The aforesaid values are sensibly higher than the magnitudes of the earthquakes recorded to date in the Mercure–Campotenese area (Figs. 1b, 9b), thus suggesting that the considered faults may have only partially released their seismogenic potential during historical times.

This inference also agrees with the distribution and evolution of the 2010–2018 seismic activity. The clusters of the relocated hypocenters concentrated in the deepest parts of the ROCS and MPR faults (Fig. 6) confirm that only a portion of such faults ruptured during the sequence, without the rupture reaching the surface.

The seismogenic patches activated on the ROCS and MPR faults during the 2010–2014 seismic sequence are considered as the reasonable approximation of the actual portion of the faults which broke during the mainshock and the sequence of the early aftershocks. We obtained them by projecting the relocated hypocenters on the reconstructed fault surface and depicting their distribution using the Kernel density geostatistical analyst, available as a tool of the ESRI ArcGIS software package. The delimitation of each seismogenic patch and its parameterization allowed us to verify the correlation between its dimensions and the magnitude released by each fault during the mainshocks.

The temporal analysis of the sequence shows that their overall extent was already well defined within the first 72 h after the major events. In any case, inside the surrounding volumes, some seismicity had started before the mainshock and continued to persist constantly throughout the development of the entire sequence so that they include ≥ 70 % of all hypocenter locations. The along-strike elongation and area extent of the patches obtained over the VCT and MPR fault surfaces can be assumed respectively as the effective subsurface rupture length and rupture area (RLD and RA in Fig. 10b, and c, respectively, according to Wells and Coppersmith, 1994) associated with the Mw5.2 Mormanno (on VCT fault) and Mw4.0 and 4.3 Morano Calabro (on MPR fault) earthquakes.

The parameters obtained for the VCT fault are RLD = 4.9 km and RA = 8.3 km2, while RLD = 1.2 km and RA = 3.6 km2 are assessed for the MPR fault. Introducing the aforesaid parameters in the appropriate scale relationships (Fig. 10b, c), we observe a good agreement between the theoretical magnitudes based on the subsurface rupture length and the magnitudes of the mainshocks. The value obtained for the VCT fault (causative of the Mw5.2 Mormanno earthquake) is Mw5.3, whereas for the MPR fault (causative of the Mw4.0 and 4.3 Morano Calabro earthquakes) it is Mw=4.5. The magnitude calculated using the RA-based relationships provides values slightly lower than expected for the VCT (4.9<Mw<5.0) and slightly higher for the MPR (4.5<Mw<4.6). In both cases, however, the magnitude values obtained using the scale relationships differ from those observed by an amount <0.3.

The seismological dataset we used demonstrates that the two main clusters of earthquakes of the 2010–2018 seismicity were generated by as many independent sources related to the sub-parallel, 10 to 15 km long, ROCS and MPR faults.

Brozzetti et al. (2017a) highlighted that the above seismogenic style, characterized by a perpendicular-to-fault strike evolution of the seismic activity, is unlike from those which followed the major instrumental earthquakes recorded in the Apennine Extensional Belt of Italy in recent years, such as the Colfiorito 1997 (Mw6.0), L'Aquila 2009 (Mw6.3), and Norcia 2016 (Mw6.5) events (Chiaraluce et al., 2011, 2017; Lavecchia et al., 2011, 2012a, 2016). They also speculated that this peculiar behavior could have been controlled by the geometric fault pattern of the area, which is characterized by WSW-dipping faults bounded southward by nearly E–W pre-existing structures. These latter are genetically related to the regional-scale, long-lived, “Pollino lineament s.l.” (Bousquet, 1969, 1971; Ghisetti and Vezzani, 1982, 1983; Knott and Turco, 1991; Van Dijk et al., 2000) and determine the abrupt contact between the Apennine carbonate platform unit and the San Donato metamorphic core complex (Grandjaquet, 1962; Servizio Geologico Nazionale, 1970; Amodio Morelli, 1976). The cross-cut relationships detected in the field between the ROCS-MPR set and POL-CVN, highlighted in our 3D model, lead us to exclude the latter fault to have a present seismogenic role, as also supported by the distribution of the instrumental earthquakes which clusterized along with N–S-striking crustal volumes. However, this significant structural–geological boundary could exert an influence on the southward propagation of the currently active seismogenic faults, driving the eastward transfer of the active extensional deformation belt. This inference is confirmed by the spatial distribution of the hypocenters of all 2010–2018 relocated seismicity which is confined within the CVN footwall (Fig. 8d).