On the mechanical behaviour of a low angle normal fault: the Altotiberina fault (Northern Apennines, Italy) system case study

Geological and seismological observations have been used to parameterize 2D numerical models to simulate the interseismic deformation of a complex extensional fault system located in the Northern Apennines (Italy). The geological system is dominated by the presence of the Altotiberina fault (ATF), a large (60 km along strike) low-angle normal fault 20° 10 dipping in the brittle crust (0-15 km). The ATF is currently interested by a high and constant rate of microseismic activity and no moderate-to-large magnitude earthquakes have been associated to it for the past 1000 years. Modelling results have been compared with GPS data in order to understand the mechanical behaviour of this fault and a suite of minor synand antithetic normal fault segments located in the main fault hanging-wall. The results of the simulations demonstrate the active role played by the Altotiberina fault in accommodating the on going 15 tectonic extension in this sector of the chain. The GPS velocity profile constructed through the fault system cannot be explained without including the ATF's contribution to deformation, indicating that this fault although misoriented has to be considered tectonically active and with a creeping behaviour below 5 km of depth. The low angle normal fault also shows a high degree of tectonic coupling with its main antithetic fault (the Gubbio fault) suggesting that creeping along the ATF may control the observed strain localization and the pattern of microseismic activity. 20

LANFs using positively discriminated slip planes from the focal mechanisms (Jackson andWhite 1989, Collettini andSibson 2001). On the contrary, observations of large displacements (Lister and Davis, 1989;John and Foster, 1993;Hayman et al., 2003;Collettini and Holdsworth, 2004;Jolivet et al., 2010;Mirabella et al., 2011), and the well-identified microseismic activity (Chiaraluce et al., 2007;Rietbrock et al., 1996) associated with these structures suggest that the LANFs are tectonically active accommodating crustal extension and possibly formed at low angle. For these reasons, the 5 LANFs cannot be excluded from the assessment of seismic hazard, although their inclusion still represents a debated issue.
The LANFs mechanical paradox could be solved if one of the two basic hypotheses of the Anderson's theory is not verified: that is, if the orientation of the maximum principal stress on the fault plane is not vertical and/or if µ s is much smaller than 0.6. Laboratory experiments on fault rock samples indicate that friction drops to very small values (µ s ≈ 0.2) at high sliding velocity suggesting that large portions of crustal faults are significantly weak (e.g. San Andreas Fault; Zoback et al., 1987). 10 Moreover, depending on the amount of clay minerals in the fault core, the coefficient of friction may be significantly lower than 0.6 (Saffer et al., 2001;Brown et al., 2003;Collettini et al., 2009a;Lockner et al., 2011). Laboratory experiments on slipping zones rocks sampled on exhumed low-angle normal fault zones (e.g. Zuccale fault, Italy), have shown very low values of sliding friction coefficients, thus a prevalent velocity strengthening behaviour of the fault (Smith and Faulkner, 2010;2009b). For these reasons, some authors proposed that a stable sliding regime might characterize the 15 LANFs behaviour (Chiaraluce et al., 2007;Hreinsdottir and Bennett, 2009;Collettini, 2011). If this frictional behaviour would be a common characteristic for the majority of such misoriented structures, it would explain the lack of moderate-tolarge magnitude earthquakes observed to nucleate on these tectonic structures.
The frictional behaviour of faults is usually modelled in terms of rate-and state-dependent constitutive laws (Dieterich, 1994) in which friction depends on slip rate, state variables and normal stress. Spatial variations of frictional properties can 20 explain the heterogeneity of crustal faulting and the frictional response to tectonic loading (both permanent and transient changes in loading conditions, Boatwright and Cocco, 1996). Velocity weakening and velocity strengthening frictional regimes are commonly used to identify dynamic instabilities (i.e., earthquakes) and stable sliding (aseismic or creeping), respectively.
In this paper we integrate geological, seismological and geodetic data with 2D numerical simulations of faulting to model the 25 state of stress and strain of an active normal fault system located in the Northern Apennines of Italy. The system is dominated at depth by the presence of a large low angle normal fault (dip angle 15°-20°), named the Altotiberina fault (ATF; Barchi et al. 1998;Boncio et al. 2000; Figure 1). The 2D model is constrained by the geological and structural cross section published by Mirabella et al. (2011) based on the interpretation of seismic profiles, deep boreholes data and geological observations at the Earth surface. We will use and interpret the on-going deformation inferred from GPS to understand the 30 tectonic role played by this LANF and to study the effects of the interseismic tectonic loading on the state of stress and deformation of the whole ATF system together with possible strain partitioning and coupling in between the distinct active faults segments. Earth Discuss., doi:10.5194/se-2016-48, 2016 Manuscript under review for journal Solid Earth Published: 8 March 2016 c Author(s) 2016. CC-BY 3.0 License.

The Altotiberina fault system
The investigated fault system is located at the Tuscany-Umbria-Marche regions boundary within the Northern Apennines ( Figure 1a), a NE-verging thrust-fold belt under-going NE-trending extension at a rate of about 3 mm/yr (Serpelloni et al. 2005). The system is dominated by a low angle normal fault: a ≈ 60 km long and 40 km width NNW-trending fault plane dipping at 15°-20° toward ENE named Altotiberina fault (ATF; see fault trace at surface in Figure 1a). The depth range of 5 the whole fault system is 0-15 km (Figure 1b).
In the ATF hanging-wall block, synthetic and antithetic structures dipping at higher angles have generated moderate magnitude earthquakes; the largest one was a M W 5.1 occurred in 1984, currently named the Gubbio earthquake ( Figure 1a; Westaway et al., 1989;Collettini e al., 2003). Several authors suggest that this seismic event, originally named as the Perugia earthquake (Haessler et al., 1988), has nucleated on the Gubbio fault plane (GuF; Haessler et al., 1988;Pucci et al. 2003) 10 while Collettini et al. (2003) by interpreting a set of multidisciplinary data, argued against this hypothesis.
Following Chiaraluce et al., 2007, only micro seismic events (< 2.3 M L ) have been located along a 500-1000 m thick fault zone cross cutting the upper crust from 4-5 km down to 14-16 km depth and coinciding with the geometry and location of the ATF (Figure 1b) as derived from geological observations and interpretation of depth-converted seismic reflection profiles (Mirabella et al., 2011). While in the hanging wall of the ATF the distribution of earthquakes highlights the presence 15 of higher angle (40°-60° of dip) synthetic and antithetic minor normal faults (4-5 km long) that sole into the detachment.
The seismicity nucleating along the ATF is characterized by a nearly constant rate of earthquake production r = 7.30e−04 eqks/day*km 2 (Chiaraluce et al. 2009), corresponding at about 3 events per day with M L < 2.3. This seismic activity is cinematically consistent with the local tectonic setting characterized by the ATF low-angle plane and shows a higher b-value than the seismicity located in the hanging-wall block (antithetic structures) showing seismic sequences like behaviour. 20 Chiaraluce et al., (2007) interpreted these features as the presence within the fault system of fault zones possessing different rheology and/or frictional properties. It is also worth noting that the microseismicity nucleating on the ATF is not able to explain the amount of deformation associated with the short and long-term slip rate inferred by geological (Collettini and Holdsworth 2004) and geodetic studies and data (D'Agostino et al. 2009). These observations together with the lack of largemagnitude (M > 7) historical earthquakes that ruptured the whole ATF in the past 1000 years (Rovida et al. 2011, Chiaraluce 25 et al., 2014 suggest the occurrence of aseismic deformation or creeping as proposed by Hreinsdóttir and Bennett (2009) by interpreting regional GPS data. This hypothesis is supported by laboratory experiments performed on fault rock samples of the Zuccale low-angle normal fault considered the (older) exhumed analogue of the ATF. Talc minerals, characterised by a very low friction coefficient over a wide range of environmental conditions (0.05< µs<0.23; Moore and Lockner, 2008), has been in fact observed to form interconnected foliated networks within the Zuccale fault core resulting in a velocity 30 strengthening behaviour (Smith and Faulkner, 2010;Collettini et al, 2009b

Numerical simulations
We perform 2D finite element numerical simulations with plain-strain approximation by means of the commercial software COMSOL Multiphysics (http://www.comsol.com/). In order to parameterize the numerical model, we use a NE-SW geological cross-section cutting the central part of the ATF system ( Figure 1a) thus considering the whole set of associated faults segments defined by Mirabella et al. (2011;see Figure 4b). 5 The mesh consists of approximately 270.000 triangular elements with a finest resolution of 25 m near the fault zones and decreases in resolution down to 2000 m along the boundaries. The crust is characterized entirely by an elastic rheology and we include those layers representing the main lithological units characterized by homogeneous competence (Pauselli and Federico, 2003;Mirabella et al., 2011; Table 1). We prolong the layers in order to avoid boundary effects during the simulation. Hence in proximity of the boundaries, the layers are maintained flat because no information is available about 10 their realistic setting ( Figure 1b).
In order to facilitate the convergence of the solution, the simulations were performed in two subsequent stages. In the initial stage, the model was subject only to the gravity load; this means that no velocity at the boundaries was imposed to simulate extension. In this way, the model compacts under the weight of the rocks and it is brought in a stable equilibrium with gravity. In this first step, the boundary conditions, applied to all models, are the following: (a) the upper boundary of the 15 mesh is free to move in all directions; (b) the lateral boundaries of the mesh and the bottom are kept fixed in the perpendicular direction. This means that slip parallel to these boundaries is allowed. In the second stage (interseismic phase) we start stretching the model (e.g. crustal extension) for 10 years, applying a constant horizontal velocity of 0.5 mm/yr and 3.5 mm/yr on the SW and NE lateral boundaries, respectively, in according to the present-day strain rate and kinematics of the region (Serpelloni et al., 2005). The stress field resulting from the first stage is defined as uniaxial strain reference frame 20 following Engelder (1993). This state of stress is characterized by vertical stress where ρ is the density, g is the gravity acceleration and z is the depth, and horizontal stress where ν is the Poisson's ratio. In this way for ν = 0.25, the vertical stress is three times larger then the horizontal stress, 25 mimicking the extensional tectonic regime of the study area.
In the simulations the faults (see cartoon in Figure 2a) are defined as 100 m thick shear zones ( Figure 2b). Only the ATF zone is represented by a thickness of 500 m as proposed by Chiaraluce et al. (2007). The weaker rheology of the fault zones is modelled by considering a variation of the elastic properties respect to the hosting intact rocks (e.g. Gudmundsson, 2004;Faulkner et al., 2006). In this way, when a fault is considered locked the Young modulus of the corresponding fault zone is 30 set to 10 GPa (Figure 2b) meaning that the deformation associated with that structure will depend mainly on the intensively fractured rocks typically, for instance, of the damage zone. Otherwise, when creeping faults are considered, the Young modulus of the corresponding fault zone is reduced by three orders of magnitude lower then the intact rocks (0.01 GPa, see Solid Earth Discuss., doi: 10.5194/se-2016-48, 2016 Manuscript under review for journal Solid Earth Published: 8 March 2016 c Author(s) 2016. CC-BY 3.0 License. Figure 3). In this case the deformation associated with the causative fault will depend mainly on the creeping layer and on minor contributes from the damage zone. This abrupt mechanical contrast between the fault zones and the surrounding intact rocks along the hanging wall and footwall blocks simulates a slip similar to those observed when creep motion is modelled along fault planes. However, the adoption of the creeping fault equivalent method results in reaching more easily a numerical convergence in solving the problem and a reduction of the computation costs. 5 Through the numerical simulations we aim to model the present day deformation of the ATF system ( Figure 1b Observatory, devoted to the understanding of the physics of earthquakes and faulting . To retrieve the 2D displacement profile we use 11 GPS stations whose distance from the cross-section is less then 10 km (from SW to Ne they are: SIO1, REPI, CSSB, VALC, UMBE, ATLO, MVAL, PIET, ATFO, ATBU, FOSS; Figure 1a).
Thus, in order to quantify the interseismic stress build-up in the last 10 years we compute the Von Mises stress, a parameter 15 that expresses the difference between the principal components of stress giving indication of the amount of shearing (e.g. Pauselli and Federico, 2003) and four different types of simulations have been performed (see Figure 3). In the first model all faults are considered locked (Model 1). In the second model (Model 2), creeping is simulated only along the ATF considering a (5 km) locking depth we derived from our tests (see next paragraph). In the third model (Model 3), we simulate creeping only along the GuF, whereas the ATF is considered locked. While in the fourth model (Model 4A) 20 creeping behaviour is simulated both along the ATF and GuF. This last model has been further complicated by considering within the ATF fault zone a set of multi-creeping-layers, as proposed by .

Fault segments locking depth
In this section we discuss the effects of assuming different locking depths identifying the depth at which creeping begins along synthetic and antithetic faults of the ATF system. The tests have been performed relying on the results of Vadacca 25 (2014) and the following settings were considered for the analysis. First, we investigated the ATF locking depth by considering four characteristic depths whence the ATF is in creeping: 2, 5, 8 and 11 km, respectively. In this first test case, we do not consider the contribution of the other fault minor segments. In order to identify the optimal model parameterization, the model results are compared with the available GPS data (Figure 4a). The best fitting was obtained by assuming an ATF locking depth of 5 km (Model-a in Fig. 4). In a second set of models, we considered the influence of 30 modelling creeping along both the synthetic and antithetic faults of the ATF system by assuming different configurations for each of the ATF locking depths previously measured. Initially we considered that all the synthetic and antithetic faults were in creeping. Successively we locked these faults one-by-one from west to east (Figure 4b setting is shown in Figure 4a. Models ATFb and ATFc correspond to an ATF locking depth of 2 and 5 km, respectively and creeping only along the Gubbio fault. Whereas the ATFd and ATFe models correspond to an ATF locking depth of 8 and 11 km, respectively and creeping simulated along all the synthetic and antithetic faults. From the analysis of the out coming results an ATF locking depth of 5 km leads to a better fit of the GPS data. This hypothesis is in agreement with the evidence of microseismic activity occurring along the ATF plane starting from the same 4-5 km of depth (Chiaraluce et al. 2007) and 5 the model resulting from regional geodetic data by Hreinsdottir and Bennett, 2009. In this best model we cannot exclude the creeping deformation interesting also a portion of the Gubbio fault. For this reason this model configuration will be analysed more in detail in the next section. Figure 5 show the interseismic stress build-up for all the tested models represented in terms of Von Mises stress patterns. We 10 observe that, when no creeping is included in the simulations (like in Model 1 of Fig. 3), the response of the medium to the imposed interseismic tectonic extension is completely controlled by the adopted lithology. In other words, the stress build up is only controlled by the mechanical contrast of the different lithology and the geometry of the fault system ( Figure 5a).

Modelling results
When creeping is simulated along the ATF, starting from 5 km of depth like in Model 2, the pattern of the stress build up changes completely. Stress is mainly localized around 5 km of depth and in the ATF hanging wall (Figure 5b). This model 15 also predicts a large stress accumulation along the deepest (3-5 km) portion of the GuF and low values of stress for the ATF footwall block where currently we observe a lack of seismic activity (see Fig. 1b). This deeper sector of the GuF shows a quite flat geometry contrary to its shallower portion where it is at high angle. Our model resolution does not allow separating the contribution of these two diverse fault portions even if in reason of its flat attitude, the deepest segment is the one being more efficient. Coherently with this we observe that when we simulate creeping only along the GuF (Model 3), we observe 20 that the stress pattern (see Figure 5c)  The combined effect of creeping along ATF and GuF is finally simulated through the Model 4A and is shown in Fig. 5d.
Here we observe a wedge of higher values of stress located between the ATF hanging wall and the GuF footwall in proximity of the assumed locking depth (5 km). This model predicts stress accumulation in the footwall of the GuF where micro-seismicity is currently observed (Valoroso et al., 2014). The last model differs from the previous one for the presence along the ATF creeping sector of a fault zone composed by a multi-layer including alternate creeping and stick-slip layers 30 (Model 4B in Fig. 3). The interseismic stress build-up decreases with respect to previous models and stress accumulation is mainly located at the intersection between the ATF and the GuF (Figure 5e). Earth Discuss., doi:10.5194/se-2016-48, 2016 Manuscript under review for journal Solid Earth Published: 8 March 2016 c Author(s) 2016. CC-BY 3.0 License. Figure 5 provides a clear picture of the role played by the ATF presumably creeping along its deepest portion. At the same time we observe signatures of tectonic coupling between the ATF and the GuF antithetic fault in a context where the ATF is clearly dominating the mechanical behaviour of the whole system. Figure 6 shows the trend of the Von Mises stress and cumulative displacement computed across the ATF and the GuF fault, 5 zones after 10 years of tectonic extension respectively at 9 and 3 km of depth. When no creeping is simulated (Model 1; pink dashed line in Fig. 6), the same stress threshold is reached on the hanging wall and footwall rocks both across the ATF

Discussion
The results of the 2D simulations performed in this study have important implications on the role played by the ATF in 10 accommodating the tectonic extension in this sector of the Northern Apennines. The first main outcome is that the GPS observations cannot be explained without considering the ATF contribution to deformation, indicating that this fault although misoriented has to be considered tectonically active. Our findings also indicate the need for creeping behaviour along the ATF below 5 km of depth At the same time our simulations suggest that using only the ATF does not exhaustively explain the GPS data deserving the contribution of at least another active segment located in the ATF hanging-wall volume 15 such as the Gubbio fault (Figure 7). In addition, the best fit of the GPS velocity data is obtained considering some degree of creeping also along a portion of the GuF. The mechanical behaviour of the west-dipping GuF is still debated in the literature (Boncio et al., 2000;Barchi et al., 1999;Mirabella et al., 2004Mirabella et al., , 2008Barchi and Ciaccio, 2009). It is considered a seismogenic fault in active fault databases (DISS Working Group Database of Individual Seismogenic Sources, version 3.1.0, A compilation of potential sources for earthquakes larger than M 5.5 in Italy and surrounding areas,available at http://diss. 20 rm.ingv.it/diss) and the last moderate-magnitude event (M w 5.1; Westaway et al., 1989) occurred in 1984 in the area (namely, the Gubbio earthquake) occurred few kilometres apart (see location in Figure 1a). According to Collettini et al. (2003), the main shock has nucleated on a different fault (see location in the map of Fig. 1a). However, the spatial pattern of the micro-seismic activity of the Gubbio area is quite complex and it does not allow the identification of the GuF as a locked active structure (Chiaraluce et al., 2007;Valoroso et al., 2014). We rely on the interpretation of the mechanical behaviour of 25 the Gubbio fault as a mixed mode seismic/aseismic characterized by a relevant creeping component during the interseismic phase of the seismic cycle for both geometrical (flat attitude in the deeper portion and high-angle to ward the surface) and mechanical reasons. Our results are in fact in in agreement with the results of recent friction laboratory experiments performed on outcrop rocks from the Gubbio fault core (Finocchio et al., 2013). The experiments have shown that the presence of layers with coupled velocity strengthening-velocity weakening behaviour (Bullock et al., 2014). normal fault (the western exhumed analogue of the ATF) suggesting low friction and a velocity strengthening behaviour (Smith and Faulkner, 2010;Collettini et al, 2009b).
Model-4A (fault zone with one creeping layer) and Model-4B (fault zone with multi creeping layers) present very similar values of weighted root mean squares (Figure 7). This complicates the interpretation of the structural setting and the mechanical behaviour of the ATF system. Complex fault zones with multiple strands are widely documented in strike slip 5 regimes (e.g. Carboneras fault in southeastern Spain). These large fault zones present a distributed deformation over several metres of multiple active phyllosilicate-rich fault gouges. The derived mechanical behaviour is a prevalent creeping with small repeating earthquakes (Faulkner et al., 2003;Faulkner et al., 2008). However no evidence of multicores in exhumed low-angle normal fault zones are reported in literature (e.g. Collettini et al., 2011); for instance, a single-fault core zone, a few meters thick, characterizes the Zuccale fault. For this reason, a fault zone with a single core can be considered a 10 plausible model for the ATF, where the stable sliding along this layer loads adjacent stick patches that can fail generating earthquakes.
The results of this study are preparatory for a 3D modelling of the Altotiberina fault system in which the lateral variations of frictional properties (Boatwright and Cocco, 1996) can be modelled to explain the observed seismicity pattern.
Notwithstanding, the outcomes of these 2D simulations contain original indications and allow the definitive indication for an 15 active LANF as well as for a tectonic coupling between the two major structures of this fault system generating a negligible stress accumulation in the ATF footwall, coherently with the lack of microseismicity, and a redistribution of the stress build up consistent with the high seismicity rate in hanging wall of the ATF, including the footwall of the GuF (Chiaraluce et al., 2007;Figure 1b).

Conclusions 20
In this work we integrate the results of 2D numerical models with geological, seismological and geodetic observations in order to understand some of the key features on the mechanical behaviour of the Altotiberina fault and its associated fault system. Our results show that: • the tectonic loading of this sector of the Apenninic chain is mainly accommodated by and active LANF (the Altotiberina fault); 25 • such a LANF is creeping from about 5 km of depth; • although the deformation caused by creeping along the Altotiberina fault is a first order condition to explain the strain concentration revealed by GPS data, a tectonically coupled model is needed to explain observations thus including a contribution of a partially creeping antithetic fault (the Gubbio fault); • the stress redistribution within the fault system caused by creeping on both the ATF and GuF together with the 30 presence of heterogeneous frictional properties on the ATF fault zone volume can explain the observed microseismicity pattern and the lack of moderate earthquake in the last 1000 years nucleating on this fault. Earth Discuss., doi:10.5194/se-2016-48, 2016 Manuscript under review for journal Solid Earth Published: 8 March 2016 c Author(s) 2016. CC-BY 3.0 License.

Solid
We speculate that the stable sliding along the creeping sections of the ATF may load adjacent stick patches within and around the fault zone that can fails in small seismic events. This mechanism, could explain the microseismic activity detected all along the ATF fault plane and within the ATF hanging wall.  (Pauselli and Federico, 2003;Mirabella et al., 2011). Earth Discuss., doi:10.5194/se-2016-48, 2016 Manuscript under review for journal Solid Earth  (Westaway et al., 1989;Collettini et al., 2003). The red lines represent the projection at surface of the Altotiberina faults system (after 5 Mirabella et al., 2011). The yellow arrows represent GPS velocities vectors in the last 10 years . Blue line represents the cross-section used for the 2D models. (b) Cross section used for the numerical simulations on the basis of seismic profiles (after Mirabella et al., 2011). The grey points represent the micro-seismicity in a range of 10 km from the cross section (dashed blue line in Fig.   1a).