Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering

journal homepage: www.elsevier.com/locate/soildyn

Ground shaking scenarios at the town of Vicoforte,

L. Scandella a,n, C.G. Lai a, D. Spallarossa b,1, M. Corigliano a a European Centre for Earthquake Engineering (EUCENTRE), via Ferrata 1, Pavia, Italy b Dipartimento per lo Studio del Territorio e delle sue Risorse (DIPTERIS), University of Genoa, Viale Benedetto XV 5, 16132 Genoa, Italy article info abstract

Article history: Vicoforte is a small town in Northern Italy, which hosts a Cathedral with the world’s largest elliptical Received 11 February 2010 dome. The name of the Basilica is ‘‘Regina Montis Regalis’’ and it is of extraordinary architectural and Received in revised form structural importance. The main objective of this study is the definition of the seismic hazard at the site of 25 November 2010 Vicoforte following a deterministic approach. Although Vicoforte is located in an area of moderate Accepted 1 December 2010 seismicity, the calculation of the most unfavourable seismic ground shaking scenarios is of great interest due to the importance of the Basilica and its vulnerability to even a moderate seismic excitation. The closest active faults to Vicoforte were identified in order to simulate the potentially most severe ground shaking scenarios compatibly with the tectonic and seismic setting of the region. Subsequently, numerical simulations were conducted through finite faults numerical models using two different approaches: the extended kinematic source model of Hisada and Bielak [24] and the stochastic method of Motazedian and Atkinson [38]. They, respectively, simulate the low and high frequency ranges of predicted ground motion. The numerical models used for the simulations were calibrated by a comparison between synthetic results and recorded data. A parametric study was finally carried out to identify the most critical fault rupture mechanisms. & 2010 Elsevier Ltd. All rights reserved.

1. Introduction Crack patterns due to the foundation settlements and to the structural configurations of the Cathedral are currently present on This paper illustrates the steps that were undertaken to estimate the the dome-drum system, but they seem stabilized. From the historical worst rock shaking scenarios at the site of Vicoforte, a municipality of records at the Sanctuary, it is known that cracking phenomena began to the city of in Northern Italy, where the ‘‘Regina Montis Regalis’’ occur in the early stages of the construction, in particular in the zone Basilica sits. Although Vicoforte is located in an area of moderate between the drum windows and at the base of the buttresses. In 1985 seismicity, it was selected as case study due to the presence of the such severe cracking prompted the decision to undertake monitoring Cathedral with the world’s largest elliptical dome (Fig. 1). and strengthening works. A system of 56 active tie-bars, slightly The internal axes of the dome of the Sanctuary are 37.15 m stressed by jacks, was installed in 1987 within the masonry at the top of 24.8 m, making it the fifth largest dome in the world (after the the drum along 14 tangents around the perimeter, and a complex Pantheon and Saint Peter in Rome, S. Maria del Fiore in Florence, monitoring system was set up to measure movements of the structure Italy and Gol Gumbaz Mausoleum in India) and by far the largest and propagation of cracks, as well as stresses in the reinforcing tie-bars. elliptical structure. In recent years a new project was started for a thorough renovation of The Basilica was first conceived by Duke Charles Emanuel I of Savoy the monitoring system [15] and the updating of the structural models as the mausoleum of the family. The original architectural composition adopted to explore the static configuration and integrity of the Basilica was an idea of engineer Ercole Negro di , Cont of Sanfrount, but and to define the characteristics of the strengthening system [13]. architect Ascanio Vitozzi implemented the project. The construction The present study was developed in the framework of a more wide- started in 1596 and since the early beginning the building was affected ranging research aimed to define the seismic input for dynamic by differential settlements due to inhomogeneity of foundation soil. analyses of the Basilica. The study was carried out in two phases: Due to the large settlements (of the order of 25–30 cm) in 1615 the the first part concerned with site-specific Probabilistic Seismic Hazard project was abandoned for more than a century, till 1692 when the Analysis (PSHA, [28]), while the second part focused on a Deterministic construction works were entrusted to the architect Francesco Gallo. Seismic Hazard Analysis (DSHA) at the site, both under the assumption The dome was completed in 1731. of stiff ground and level topographic condition. The choice of whether using PSHA or DSHA for an adequate analysis constitutes a vivid debate in the scientific community. Fiercely

n discussions whether PSHA or DSHA or a possible combination of Corresponding author. Tel.: +39 0382 516925; fax: +39 0382 529131. E-mail address: [email protected] (L. Scandella). the two should be used show how debated this topic is [9,26,27]. 1 Tel.: +39 010 3538 038; fax: +39 010 352 169. Nevertheless, both methods present advantages and limitations,

0267-7261/$ - see front matter & 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.soildyn.2010.12.004 758 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

Fig. 1. ‘‘Regina Montis Regalis’’ Basilica of Vicoforte, Northern Italy and detail of the world’s largest elliptical dome.

as summarized by McGuire [34]. The PSHA study is performed to determine the seismic hazard at the site as resulting from the historical seismicity. On the other hand, a deterministic approach is more suitable when the hazard comes from a single fault which is dominating the overall hazard. As a consequence, DSHA illustrated in the present paper was carried out to evaluate ground shaking scenarios, compatibly with the regional seismotectonic setting. As a first phase of the DSHA, the closest active faults to Vicoforte were identified in order to simulate the potentially most severe ground shaking scenarios. Numerical analyses were performed using two different approaches which simulate the low and high frequency ranges, respectively. The low frequency analyses allowed identifying the most critical fault rupture mechanisms, while the high frequency simulations were performed with the aim to obtain a valid result of comparison with the PSHA in terms of Peak Ground Acceleration (PGA), which is associated to the high frequencies. Calibration of synthetic seismograms was performed through a comparison with recorded data of recent seismic events. Finally, a parametric study was carried out to identify the potential most critical rupture scenarios, thus the most unfavourable ground Fig. 2. Map of the area of interest with the surface projections of the planes of the shaking scenarios, which represented the input for the evaluation main faults identified in the region: Monferrato, Western Alps and Western Liguria faults. of local site effects. This aspect was investigated in [28] through 1D stochastic and 2D deterministic approaches. On the basis of historical and recent events which struck and the surrounding regions, the seismic activity in North-Western Italy was identified in three main regions: Western Liguria, Western 2. Tectonic and seismological setting of Vicoforte area Alps and Monferrato (Fig. 2). Three potential seismogenic sources were associated to these areas, characterized by the seismological para- The definition of ground-shaking scenarios in North-Western meters listed in Table 1. Given the uncertainty of the magnitude and the Italy clashes with the strong uncertainty on geometry and kine- insufficient knowledge of the seismic potential of each source due to matics of the sources of earthquakes occurred in the past. Due to lack of geologic data, a range of possible maximum magnitudes was the large return periods of earthquakes on individual seismogenic specified instead of a single value. faults in Italy (often greater than 1000 years), it is not usually possible to identify such sources on the basis of recent seismolo- 2.1. Western Liguria gical data, thus data from pre-instrumental earthquakes based on macroseismic databases (e.g., DBMI04 database, [47]) and seismo- The Ligurian Sea originated from the counter-clockwise rotation tectonic interpretations are generally used. of the Corsican-Sardinian block relative to the European plate, due From a detail study of the seismotechtonic setting in the region, to the convergence of the European and African plates. The the main active faults of interest for the site could be identified. In continental slope is characterized by shallow normal faults parallel France the only region characterized by the presence of active faults to the coast which intersect a more recent fault system NW–SE is in Provence [31], at a distance of more than 100 km from oriented [14]. The geological and tectonic features would seem to Vicoforte and characterized by a very low seismic potential. confirm the extensional origin of the basin, but recent seismolo- The seismotectonic setting of North-western Italy, associated to gical observations depict a present-day compression. Major recent the complex geodynamic process of the Western Alps evolution earthquakes in the Ligurian Sea, indeed, show compressive focal (e.g., [8,43]) was studied in detail to identify the main seismogenic mechanisms (Fig. 3) with NW–SE principal horizontal stress sources, which may potentially affect Vicoforte. vectors [20,29].Be´thoux et al. [6] suggest that the Ligurian Sea is L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772 759

Table 1 Main parameters of the identified seismic sources. L and M indicate length and width of the faults.

Seismogenic source Mw Fault centre Depth (km) Strike, Dip Type L (km) W (km)

Western Liguria 6.326.7 Lat.: 43.741N 6212 601,601 Reverse 10220 8212 Long.: 8.131E Western Alps 5.726.0 Lat.: 44.841N 328601,451 Normal 8 4 Long.: 7.261E Monferrato 5.125.6 Lat.: 44.821N 8215 501,801 Strike-Slip 6 3 Long.: 8.421E

Fig. 3. Focal mechanism solutions and earthquake magnitudes for the Massif and the Northern Ligurian margin (after [29]). currently closing and a compression has been reactivating due to the lateral expulsion of the south-western Alps along the Apulian indenter. Chaumillon et al. [14], instead, propose that this area is subject to a superposition of two strain regimes: an extensional one Fig. 4. Map of Ligurian seismicity from year 1000 to 2003. For the 1818, 1819, 1854, and 1887 earthquakes both offshore [4] and onshore [51] epicentral solutions are near the surface and a deeper compressional one. displayed (after [4]). This area is currently characterized by a low-to-moderate seismic activity but some destructive earthquakes occurred in the past. The strongest one was on February 23, 1887 Ligurian Sea 2.2. Western Alps earthquake with MCS intensity I0¼IX (MwE6.3). This event may be associated with a release of stress along a normal fault system The area encompassing the Western Alpine belt has for many oriented parallel to the coast. Ferrari [21], analysing the distribu- years been the subject of studies aimed at defining its complex tion of the seismic effects produced by this earthquake, estimated a tectonics (e.g. [20,44,49,30,17]). Based on the analysis and the magnitude of around 6.2–6.5. The event was felt over a wide area, interpretation of the present-day geodynamics and active tec- encompassing Northern Italy, Southern France and Corsica and it tonics, seismological data (e.g. focal mechanism solutions) and generated tsunami waves, which were observed from Cannes to Global Positioning System (GPS) observations, the Western Alps Genoa along 250 km of the coast [21,19,39,40,48]. can be divided into an internal and an external sectors. The former The location and geometry of the Western Liguria fault (see Table 1) is a continuous zone of extension that follows the topographic crest was inferred based on the damage pattern produce by the earthquake line of the Alpine arc (Fig. 5). This area is characterized by low-to- and was constrained by seismological and geologic data. moderate seismicity, however relatively frequent earthquakes

Other relevant seismic events occurred in 1818 (MSE5.4), 1819 occur with hypocentral depth ranging up to around 13 km. The (MSE4.7) and 1854 (MSE5.7), whose epicentral parameters were larger events are located in the Western Swiss Alps where exten- recently revised [4] on the basis of macroseismic fields (DOM 4.1 sion is associated with a notable present-day uplift (e.g., [25]). The database, [35]), the seismotectonic background, and the distribution of external sector, which is located at the Po Plain border, is char- recent seismicity. All epicentres are localized 15–20 km offshore from acterized by a compressive-transpressive regime. Focal mechanism the coast at the base of the continental slope, where the recent offshore solutions indicate that the direction of maximum compression seismic activity is mainly concentrated (Fig. 4). The 1818 and 1819 (i.e., P-axis) rotates counter-clockwise from north (where P-axis is events were localized in the same area of the 1887 earthquakes (Fig. 4) ENE–WSW oriented) to south (where they are approximately N–S while the 1854 one was set further west on the fault system that directed), following the contour of the Alpine belt [17]. Earthquakes caused the December 26, 1989 (ML¼4.2) and April 15, 1990 (ML¼3.9) are mainly concentrated in the western part of this sector at the earthquakes [4]. border with the internal extensional sector. 760 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

ground shaking scenarios at Vicoforte site assuming outcropping rock conditions. In order to cover the whole frequency range of engi- neering interest, numerical modelling was carried out following two different approaches, which simulate the low and high frequency ranges, respectively:

the extended kinematic source model by Hisada and Bielak [24] implemented by the GRFLT12S code for the low frequency range (o1–2 Hz); the stochastic method implemented in EXSIM code [38] for the high frequency range (40.5–1 Hz).

The Hisada-Bielak [24] model is a semi-analytical method suitable for extended domains and it simulates the complete 3D wave propagation field induced by an extended kinematic source. This approach is based on the computation of static and dynamic Green’s functions of displacements and stresses for a viscoelastic horizontally layered half space. It takes advantage of the stress discontinuity representation for boundary and source conditions (i.e. kinematic model of the source) and of an analytical form derived from the generalized method of reflection and transmission coefficients (R/T) to asymp- totically solve the integrands of the Green’s functions. This approach allows investigating the effects of source, fling step, rupture directivity and strong motion in near-fault conditions, which may play a relevant role in the ground shaking prediction. Since the cases under study are neither in near-field condition Fig. 5. Seismotectonic map of the Western Alps showing the distribution of nor with surface fault rupture, only the dynamic contribution of the extension (black) and compression (white) as derived from focal mechanism Green’s functions was taken into account. solutions. Earthquake location is superimposed (after [17]). EXSIM code [38] is one of the several available stochastic finite-fault simulation methods widely used at high frequency The Italian Database of Individual Seismogenic Sources—DISS3.1 (40.5–1 Hz), i.e. above the frequency range commonly simulated ([5] http://diss.rm.ingv.it/diss/) indicates in the area of interest for by deterministic methods. This programme implements an exten- Vicoforte only one fault, with prevalent inverse mechanism (rake¼ sion of the point source stochastic model developed by Boore 1 135 ), associable with the April 2, 1808 earthquake (M¼5.7), the (SMSIM code, [12]). It simulates the finite fault as a plane divided largest event occurred in the Western Alps foothill. This earthquake into a series of subfaults, each one modelled as a stochastic point occurred at the transition between the extensional and compressive source, using a Brune (o2) source spectrum. To simulate the motion zones, as defined by Delacou et al. [17]. As for the other faults, the from an extended rupture, the radiation at an observation point is location and geometry of this source (see Table 1) were constrained by computed by the sum of the contributions due to each subfaults the damage pattern and geological data. with appropriate delays and amplitude scaling. The EXSIM code recently revised by Boore [11] is available at the 2.3. Monferrato web site http://quake.usgs.gov/boore/software_online.htm. It is important to point out that the Hisada-Bielak method This region represents a transition zone between the Alps and the simulates the complete wave field, whereas EXSIM simulates the Apennines. It is characterized by a low seismic activity mainly S-wave field only. Furthermore, GRFLT12S programme computes concentrated in the southernmost part, near the Liguria border, where velocity time histories in the two horizontal and vertical directions, two significant seismic sequences occurred in August 2000 and July while EXSIM provides an average horizontal component only.

2001 [32] with main shocks of magnitude ML¼5.1 and ML¼4.8, Both the approaches require the definition of suitable models respectively. Massa et al. [32] investigated the spatial and temporal of the seismic source and of the crustal structure along the propa- evolution of the two seismic sequences, providing constraints of the gation path. causative seismogenic source. The seismicity is concentrated at depths up to around 20 km. Focal mechanism solutions relative to the 2000– 2001 sequences suggest the prevalence of a strike-slip regime and a 4. Definition of crustal models SW–NE prevalent orientation of one of the nodal planes, in agreement with the distribution of the events. These observations point out the In order to carry out the numerical simulations with the presence of a vertical active structure that may be associated with methods previously described, it is necessary to define a crustal pre-Burdigalian tectonic structures of the Tertiary Piedmont Basin. profile and attenuation model suitable for the propagation path in Furthermore, it is worthwhile noting that, on April 2003 another the region of interest. This is a difficult task due to the large sequence, with main shock of magnitude ML¼4.9, occurred near the uncertainties in the definition of a geological model, which makes Monferrato area, specifically near Novi Ligure and Tortona. most of the assumptions based on subjective judgments rather than on objective data. The three identified faults (Western Alps, Western Liguria and 3. Numerical simulations Monferrato) lie in different crustal settings since each of them is far from the others. Therefore, three different crustal profiles were Deterministic seismic hazard analysis was performed by finite- assumed on the basis of available 1D crustal propagation models source numerical simulations, aiming at computing the most severe generally used for the localization of seismic events in the area L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772 761

under study [45]. The rock profiles adopted for the inland regions transmission. In the formula f0 is a reference frequency and Q0 is the (Western Alps and Monferrato) are rough in the shallow part. In quality factor at f0 (Q0,S for S-waves or Q0,P for P-waves). Morasca fact only one layer characterizes the uppermost 2 km of the et al. [37]derived for North-West Italy average over depth values of

Monferrato profile and the uppermost 1 km of the Alpine profile. Q0,S equal to 400 and n equal to 0.2 at a reference frequency f0 of Consequently, they were adapted to fit a shallow rock model 1.0 Hz. These parameters were used for the high frequency following the formula proposed by Cotton et al. [16], which was simulations. Values of the quality factors related to both S and developed and calibrated on rock sites in West Central Europe, a P-waves (Q0,S and Q0,P) of the layered crustal models adopted for region of moderate seismicity, comprising eastern France, south- low frequency simulations are listed in Tables 2–4 for the Mon- west Germany and northern Switzerland, thus appropriate also for ferrato, Western Alps and Western Liguria models, respectively. the area under study. This model is based on the knowledge of the EXSIM also applies a high frequency decay parameter k0 [2] to VS30 parameter. In this case a VS30 of 600 m/s was adopted to take into account the exponential high frequency anelastic gradually connect the deep profile to the VS value of the marlstone attenuation. A k0 value of 0.012 s was adopted in agreement with layer, which constitutes the bedrock in the proximity of the Morasca et al. [36], who evaluated this coefficient in the region Basilica, as it was identified by in situ geophysical surveys [28]. under study. The S and P-wave profiles of the assumed layered models are Concerning the geometrical spreading, EXSIM code requires as shown in Fig. 6. The crustal properties of the Monferrato, Western input parameter a tri-linear geometrical spreading function g(r)in Alps and Western Liguria models are listed in Tables 2–4, respec- terms of distance (r), while GRFLT12S automatically accounts for tively. It is important to remark that for the Monferrato case various this aspect within its formulation. Following Morasca et al. [36], assumptions were adopted during the calibration as illustrated in who estimated the attenuation parameters for the Western Alps Section 5. The model described in Table 2 and shown in Fig. 6 was region on the basis of recordings of the RSNI (Regional Seismic the one which produces the best results in terms of comparison network of North-Western Italy, http://www.dipteris.unige.it/geo between synthetic and recorded seismograms (see Section 5). fisica/), the g(r) function was assumed, with r the hypocentral While the Hisada-Bielak approach assumes a layered profile, distance, as follows: EXSIM adopts a homogeneous model. As a result, high frequency r0:9 r r40km simulations were carried out using average properties between the ð1Þ r0:5 r 440km layers where the faults lie, depending on their depth location. Concerning the crustal attenuation model, both methods ana- Eq. (1) describes in the log–log space a bilinear function with a lyse the earth crust as a linear viscoelastic medium with a body-wave-like spreading function (g(r)1/r) within 40 km, and n 0.5 frequency-dependent quality factor Q(f)¼Q0(f/f0) of the propaga- cylindrical spreading (g(r)1/r ) beyond 40 km, at frequencies tion medium in terms of both S-waves (QS) and P-waves (QP) larger than 0.5 Hz.

Fig. 6. VS and VP profiles adopted for the layered crustal models used for the Western Liguria (a), Monferrato (b), and Western Alps (c) faults ground motion simulations.

Table 2 Material properties assumed for the layered crustal model used for the Monferrato fault ground motion simulations.

3 Layer no. q (kg/m ) VP (m/s) VS (m/s) Q0,P Q0,S Depth (m) Thickness (m)

1 2200 1992 1150 300 150 from 0 to 130 130 2 2200 3204 1850 400 200 from 130 to 330 200 3 2300 3724 2150 500 250 from 330 to 630 300 4 2300 4174 2410 600 300 from 630 to 1100 470 5 2500 4625 2670 700 350 from 1100 to 2000 900 6 2700 5023 2900 800 400 from 2000 to 8000 6000 7 2700 5543 3200 800 400 from 8000 to 15 000 7000 8 2700 7000 4041 800 400 from 15 000 to 30 000 15 000 10 2700 8000 4619 800 400 from 30 000 to N – 762 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

Table 3 Material properties of the layered crustal model used for the Western Alps fault ground motion simulations.

3 Layer no. q (kg/m ) VP (m/s) VS (m/s) Q0,P Q0,S Depth (m) Thickness (m)

1 2200 953 550 200 100 from 0 to 15 15 2 2200 1386 800 300 150 from 15 to 40 25 3 2300 1905 1100 300 150 from 40 to 80 40 4 2300 2598 1500 400 200 from 80 to 160 80 5 2300 3204 1850 400 200 from 160 to 300 140 6 2400 3637 2100 500 250 from 300 to 550 250 7 2500 4105 2370 600 300 from 550 to 1000 450 8 2700 5500 3175 800 400 from 1000 to 3000 2000 9 2700 5800 3349 800 400 from 3000 to 5000 2000 10 2700 6100 3522 800 400 from 5000 to 11 000 6000 11 2700 6400 3695 800 400 from 11 000 to 16 000 5000 12 2700 6600 3811 800 400 from 16 000 to 25 000 9000 13 2700 7600 4388 800 400 from 25 000 to 30 000 10 000 14 2700 8000 4619 800 400 from 25 000 to N –

Table 4 Material properties of the layered crustal model used for the Western Liguria fault ground motion simulations.

3 Layer no. q (kg/m ) VP (m/s) VS (m/s) Q0,P Q0,S Depth (m) Thickness (m)

1 2700 4000 2309 800 400 from 0 to 2000 2000 2 2700 6000 3464 800 400 from 2000 to 10 000 8000 3 2700 6300 3637 800 400 from 10 000 to 18 000 4000 4 2700 7000 4095 800 400 from 18 000 to 25 000 4000 5 2700 8000 4619 800 400 from 25 000 to N –

5. Calibration of the numerical model for the Monferrato fault

Prior to numerical simulations of future earthquakes scenarios, efforts were made to calibrate the numerical models through a comparison between synthetic seismograms and recordings of recent seismic events. Unfortunately, records were available only from the RSNI network and among these data few records of events larger than Mw 5 are available, since the site of Vicoforte belongs to a region of low seismicity. Thus, it was possible to calibrate only the Monferrato model, using the event of August 21, 2000 of magni- tude Mw¼4.9. Despite the low magnitude, it was proper since the epicentre was located on the plane of the Monferrato fault and data recorded on outcropping rock were available. Among the rock stations of the RSNI network, the GENL station was chosen for the calibration (Fig. 7), since it is the nearest station to the fault, with an epicentral distance of about 65 km. This is similar to the distance between the fault and Vicoforte site. Fig. 7. Location of the August 21, 2000 Monferrato earthquake, modelled for the It is well known that point-source and finite-fault models calibration of numerical simulations. From GENL station records on rock were used should predict the same ground motions for moderate earthquakes for the comparison with synthetic seismograms. The variation of the hypocentral and at large distances. At this purpose a comparison was made location on the fault plane is shown in the box in the top right corner. A total of 12 different nucleation points were considered. between results obtained by the adopted finite fault methods and the double-couple point dislocation source by the PUNCH code [41]. Due to the better agreement with recordings obtained with Fault dimensions were determined following the empirical rela- the Hisada-Bielak method, if compared with the results provided tions proposed by Wells and Coppersmith [50], which are function by the double-couple simulation, particularly in the low frequency of Mw and on the prevalent source mechanism (normal, reverse or range (fo1 Hz), finite fault simulations were performed. This strike-slip). aspect was recently investigated by Atkinson et al. [3] comparing The fault was subdivided into 42 subfaults (7 6, 7 along the stochastic point-source and finite fault models. length L and 6 along the width W), following the criterion of Due to the uncertainties in the seismogenetic characterization maximum subfault area recommended by Mavroeidis et al. [33] to of the Monferrato fault, parametric analyses were performed have accurate results. varying both the crustal model in the uppermost part (in the phase Hypocentral location is one of the most critical parameter for of discretization of the continuous profile provided by the formula the kinematic definition of the source model. A parametric analysis of Cotton et al. [16] (see Section 4)), and the fault parameters which was carried out, varying the nucleation point along the fault plane mostly influence ground motion, namely fault size, hypocentral as shown in Fig. 7. location, rake angle, and rise time. The parameters of the crustal Due to lack of data, both the fault slip and the rupture velocity and fault models that provide the best agreement with the recorded were assumed homogeneous over the fault. The slip rate Du was data are listed in Tables 2 and 5, respectively. evaluated as function of the seismic moment and fault dimension L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772 763

Table 5 Source parameters of the rupture model for the August 21, 2000 earthquake adopted for the simulations using both the low and the high frequency methods.

Mw M0 Fault centre Strike Dip VR LWMin. depth Max. depth Du Rise time rake Stress drop – (N m) (deg.) (deg.) (m/s) (km) (km) (km) (km) (m) (s) (deg.) (bar)

GRFLT12S o Lat.: 44.82 N 0.0647 0.4 220 – 4.9 2.19E+16 Lon.: 8.42oE 50 80 2700 2.7 3.5 13.3 16.7 Depth: 15 km EXSIM

0.0666 1/f0sb –7

(M0 ¼AmDu with M0 the seismic moment, A the fault area, and m the shear modulus), while the rupture velocity VR was assumed equal to 0.75 of the average shear wave velocity of the crustal layers where the fault lies. It is remarked that the slip rates Du adopted in the GRFLT12S and EXSIM simulations were slightly different due to adoption of different rock models (a layered crustal profile for the former, a uniform medium for the latter). Analyses in the low frequency range were performed adopting different values for the rise time, in the range from 0.4 to 1.2 s. The value which gives the better results is equal to 0.4 s, as listed in Table 5. This is in agreement with the empirical relation proposed by Geller [22], which calculates the rise time on the basis on the fault area and the average shear wave velocity. EXSIM code computes this parameter as 1/f0sb,wheref0sb is the subfault dynamic corner fre- quency, namely a time dependent corner frequency, which decreases as the rupture area grows, changing the frequency content of the ground motion as the rupture progresses [38]. The Hisada-Bielak extended kinematic source model requires the definition of further parameters which are the rake angle and the slip distribution. The rake angle was varied as well as the hypocentral location in order to achieve the best fit with the recorded time histories. Fig. 8. Comparison between recorded (black) and synthetic (red) seismograms 1 1 1 1 Values from 40 to 40 and from 140 to 220 were assumed, in simulated by the low frequency GRFLT12S code for the August 21, 2000 Monferrato agreement with the prevalent strike-slip mechanism of the Mon- earthquake. Velocity time histories and Fourier spectra at GENL station are plotted: ferrato fault (see Table 1). The commonly used smoothed triangular NS (a), EW (b), and UD (c) components. velocity function was adopted for the slip distribution. A stress parameter has to be defined for the EXSIM simulations. source and the variability of propagation path. The source effect is Due to the lack of regional data, a Brune magnitude-dependent correctly simulated, particularly in the EW component, where the stress drop of about 7 bar was assumed for the Monferrato fault P-wave arrival is well captured. The amplitude peaks are reason- simulation case. Finally, the high frequency simulations were ably reproduced as well. Concerning the signal duration, the performed using a homogeneous medium with shear wave velocity agreement is also satisfactory in terms of source contribution, of 3600 m/s and mean density of 2700 kg/m3. These are average while the path duration is strongly limited by the assumption of a properties of the strata where the Monferrato fault lies in the 1D layered rock profile which neglects propagation path irregula- crustal model defined for the low frequency range analyses (see rities, local discontinuities, site and topographic effects. The Table 2 and Fig. 6b). contribution of recorded signals beyond 40 s is probably due to a For the Liguria and Western Alps faults no strong motion data characteristic layer that generates a wave-guide effect. This effect originated by these lineaments were available. cannot be reproduced by a 1D layered soil profile, as adopted in the simulations. In terms of frequency content, frequencies up to 1.5–1.7 Hz are simulated with a satisfactory agreement along the 5.1. Numerical results in the low frequency range horizontal directions, while the frequency contribution is under- estimated for the vertical component. Note that, both recorded and From the parametric analyses performed with the GRFLT12S simulated velocities shown in Fig. 8 were filtered prior to compar- code it turns out that a rake of 2201 coupled with nucleation point ison with a low-pass filter up to 1.7 Hz. number 3 (see Fig. 7) provides the results which best fit the records, as shown in Fig. 8 in terms of velocity time histories and Fourier spectra. 5.2. Numerical results in the high frequency range Velocity was used for comparison because it is the best simulated ground motion parameter by an extended kinematic source model as Stochastic, high frequency simulations were carried out using that implemented in GRFLT12S. Moreover, records of ground velocity EXSIM. They were performed assuming all the fault parameters are available at the GENL station of RSNI network. fixed except the hypocentral location, which was considered to The agreement between measured and predicted signals is vary as previously shown in Fig. 7 and also randomly. Fig. 9 shows satisfactory although this was the simulation of a small earthquake the comparison between recorded and simulated seismograms at at a large distance which is a difficult case to simulate, due to the the GENL station for a ‘‘random’’ epicentre. Since EXSIM provides an large uncertainties associated to both the definition of seismic average horizontal acceleration component only, in the figure the 764 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

Fourier spectrum of the synthetic accelerogram (c) is superimposed numerical derivation of the synthetic accelerogram. The Fourier on both the NS (a) and EW (b) recorded components. It is remarked spectrum of the synthetic velocigram (f) is superimposed for com- that ‘‘recorded’’ acceleration traces were obtained by numerical parison on both the NS (d) and EW (e) recorded components. integration of recorded velocity time histories, while the ‘‘simu- A good agreement is obtained in terms of both velocity and acce- lated’’ velocity trace, shown in the same figure, was calculated by leration at frequencies higher than 1 Hz since these are properly simulated by a stochastic method. As expected, lower frequencies are poorly predicted. As illustrated for the results obtained with GRFLT12S programme, the duration of synthetic seismograms is shorter if compared with the recorded traces, due to source dura- tion effect only. Furthermore, it is important to remark that, due to lack of data, simulations with EXSIM were performed keeping the duration equal to the source duration. Although duration is a parameter of fundamental importance for reliable synthetic accel- erograms, this aspect was not thoroughly investigated, since the aim of high frequency analyses was limited to a comparison with the PSHA in terms of PGA, which is mainly associated to the source effect. Comparing the simulated velocity trace (Fig. 9f) with the recorded time histories in the horizontal directions (in Fig. 9d, e), it turns out that the S-wave direct arrival is correctly reproduced. Fig. 9f compares the horizontal velocity Fourier spectra com- puted in the low (NS and EW components) and high (average horizontal component) frequency ranges, showing that they are complementary with a frequency window of connection between 1 and 2.5 Hz.

6. Worst ground shaking scenarios at Vicoforte

As mentioned in the introduction the main scope of this study is the identification of the most severe ground shaking scenarios due to a potential reactivation of the faults identified in the region.

With reference to the expected average ranges of Mw listed in Table 1 for each fault, three events were simulated representing the most likely critical situation:

Monferrato fault: event of Mw¼5.5; Western Alps fault: event of Mw¼5.7; Western Liguria fault: event of Mw¼6.5.

On the basis of the bounded Gutenberg and Richter [23] relations developed in the framework of the PSHA conducted at Vicoforte site [28], recurrence times of 690, 475 and 2575 years can be associated respectively to the Monferrato, Western Alps and Western Liguria events listed above. Fig. 9. Comparison between recorded (black) and synthetic (gray) seismograms simulated for the August 21, 2000 Monferrato earthquake with the high frequency Parametric analyses were performed by studying the influence EXSIM code. Acceleration and velocity time histories are plotted together with the of the hypocentral location and slip direction to identify the respective Fourier spectra at GENL station: recorded NS (a) and EW (b) acceleration combination of parameters which would lead to the worst ground components; synthetic average horizontal acceleration (c), whose Fourier spectra is shaking scenarios. In the following sections the results obtained for also overlapped on the two recorded components in (a) and (b); recorded NS (d) and EW (e) velocity components; synthetic average horizontal velocity (f), whose the events of the Monferrato and Western Liguria faults will be Fourier spectra is also overlapped on the two recorded components in (d) and illustrated, leaving aside for brevity the case of the Western Alps (e). In (f) the horizontal spectra obtained by GRFLT12S are also overlapped. fault, which represents an event of intermediate severity.

Table 6

Source parameters of the rupture model for the Monferrato Mw 5.5 ground shaking scenarios.

Mw M0 Fault centre Strike Dip VR LWMin. depth Max. depth Du Rise time rake Stress drop - (N m) (deg.) (deg.) (m/s) (km) (km) (km) (km) (m) (s) (deg.) (bar)

GRFLT12S Lat.: 44.82oN 0.31 0.6 0–180 – 5.5 2.0E+17 Lon.: 8.42oE 50 80 2400 6 3 10 13 150–210 Depth: 11.5 km EXSIM

0.32 1/f0sb –65 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772 765

6.1. Monferrato fault ground motion parameter for a comparison between low and high frequency results. The Fourier spectra obtained by the two proce- The parameters defining the Monferrato source model for the dures are complementary. GRFLT12S is able to simulate frequen- prediction of ground motion scenarios were identified with the cies up to about 2 Hz; EXSIM simulates frequency higher than 2 Hz same criteria illustrated for the calibration phase and they are listed and underestimates frequencies between 0.6 and 2 Hz if compared in Table 6. Since a prevalent strike-slip mechanism was identified with the GRFLT12S spectra. The arrivals of direct P and S-waves for this fault (see Table 1), rake angles related to a pure strike slip, i.e. can be clearly distinguished in the GRFLT12S velocity traces 01 and 1801, and to a combined mechanism from 1501 to 2101 were (Fig. 11a–c), while EXSIM simulates the S-waves only (Fig. 11d). assumed for the low frequency simulations. The location of the Both the low and high frequency methods simulate a Peak Ground hypocentre was varied with the same geometric distribution used Velocity (PGV) of about 0.02 m/s as the most severe ground motion for the calibration and shown in Fig. 7, however taking into account at Vicoforte. EXSIM predicts a Peak Ground Acceleration (PGA) of the different size of the rupture area. The crustal model adopted in 0.15 m/s2 with nucleation point n. 4 (see Fig. 12) and a maximum the low frequency analyses was characterized by the rock proper- PGA of 0.21 m/s2 assuming the hypocentre n. 12. It is important to ties listed in Table 2 and the velocity profiles shown in Fig. 6b. remark that the hypocentre location in EXSIM has not the same The high frequency simulations were performed assuming a ‘‘physical’’ meaning as in GREFLT12S. However, its variability was homogeneous medium with average properties of the layers analysed with the aim of computing various stochastic ground corresponding to where the fault lies in the crustal model assumed motion scenarios for the same source model. for the low frequency analyses, namely shear wave velocity of To validate the results, horizontal (PGVh) and vertical (PGVv) 3200 m/s and mean density of 2700 kg/m3. Peak Ground Velocities computed with GRFLT12S and horizontal A total of 48 low frequency simulations (12 hypocentres by Peak Ground Accelerations (PGA) calculated with EXSIM were 4 rake angles) were performed. In the high frequency range, 14 compared with the values predicted by suitable Ground Motion simulations were carried out: 12 hypocentres with homogeneous Prediction Equations (GMPEs). After a careful examination, the slip rate, one with random hypocentre and homogeneous slip rate, GMPEs adopted for the region under study are as follows: and finally one with random hypocentre and random slip rate. It is interesting to analyse the effect of the rake angle on the i. Sabetta and Pugliese, 1996—abbreviated as SP96 [42]; ground motion response. Fig. 10 shows the velocity time histories ii. Bindi et al., 2009—Bi09 [7]; simulated with GRFLT12S and caused by a rupture process on the iii. Ambraseys et al., 2005—Am05 [1]; Monferrato fault with nucleation point n. 7 and three rake angles: iv. Boore and Atkinson, 2008—BA08 [10]. 1501, 1801 and 2101. All these mechanisms were chosen to generate forward directivity effects toward the Vicoforte site, as this would SP96 and Bi09 relations were specifically developed for Italy. In correspond to the most severe scenarios. Due to the large epicentral particular, SP96 is based on a strong motion database, which distance to the site (about 65–70 km), not relevant differences includes 95 records from 17 Italian earthquakes, while Bi09 is were observed in the waveforms and in the amplitude peaks as based on the recently compiled ITACA strong motion database well. However, from the parametric analysis it turns out that a (http://itaca.mi.ingv.it). Am05 GMPE was calibrated for Europe and kinematic rupture with a rake angle of 1501 and nucleation point Middle East, while BA08 was developed in the framework of the number 4 provides the worst shaking scenario, as illustrated in ‘‘PEER NGA’’ (Pacific Earthquake Engineering Research Center’s Fig. 11 in terms of particle velocity. This is the most appropriate Next Generation Attenuation Ground Motion) project and on the

Fig. 10. Effect of the rake angle on ground motion predicted at Vicoforte site in terms of velocity time histories simulated by GRFLT12S for the Mw 5.5 Monferrato event, assuming the nucleation point number 7: NS (a), EW (b), and UD (c) components. 766 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

Fig. 11. Synthetic seismograms in terms of velocity time histories and corresponding Fourier spectra simulated for the Mw 5.5 Monferrato event, assuming the nucleation point number 4 and a rake angle equal to 1501: NS (a), EW (b), and UD (c) components computed by GRFLT12S (black); (d) horizontal average component computed EXSIM (gray), whose Fourier spectrum is also overlapped on the two low frequency horizontal components in (a) and (b).

Fig. 12. Acceleration time history and corresponding Fourier spectrum of the horizontal average component computed by EXSIM for the Mw 5.5 Monferrato event, assuming the nucleation point number 4.

basis of an extensive strong-motion database of thousands of A seismic event of Mw 6.5 was simulated adopting the fault records from shallow crustal earthquakes in active regions world- parameters listed in Table 7. Parametric analyses were performed wide. BA08 is shown to be representative of the NGA models [46] varying the rake angle and the hypocentral location on the fault and although it was not calibrated for Italy or Europe, it may be plane as shown in Fig. 15. In order to simulate forward directivity applicable to Europe as shown by Stafford et al. [46]. effects, rake angles of a prevalent reverse mechanism varying from Fig. 13 shows that the PGVs obtained from numerical simula- 451 to 1351 were assumed. tions, assuming a fixed rake angle and the hypocentral locations of The crustal model of Fig. 6a with the associated material Fig. 7, are in agreement with the GMPEs developed for rock sites parameters listed in Table 4 were used for the low frequency

(VS,304760 m/s). The large range of variability of the numerical simulations, while homogeneous rock with shear wave velocity of results for different hypocentres is due to the directivity effect, not 3550 m/s and mean density of 2700 kg/m3 were adopted for the taken into account by the adopted GMPEs. high frequency simulations. In terms of acceleration, EXSIM provides an average horizontal A total of 60 numerical simulations (20 hypocentres by 3 rake PGA calculated stochastically and thus ignoring directivity effects. angles) were performed in the low frequency range, while 22 For this reason, the variability of the results is more limited if simulations were carried out in the high frequency range: 20 compared with that of the GMPEs, as shown in Fig. 14. hypocentres with homogeneous slip rate, one with random hypo- centre and homogeneous slip rate, and finally one with random hypocentre and random slip rate. 6.2. Western Liguria fault The influence of the rake angle on ground motion is shown in Fig. 16, where velocity time histories computed with GRFLT12S The seismic hazard assessment associated to the Western were calculated assuming hypocentre number 2 (Fig. 15) and Liguria fault was carried out following the same scheme as for different rake angles. Computed waveforms are similar in all spatial the Monferrato fault case. directions, while the amplitude peaks vary by a factor up to 2, L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772 767

Fig. 13. Comparison between GMPEs and horizontal (a) and vertical (b) PGVs predicted by GRFLT12S for the Mw 5.5 Monferrato event, assuming various hypocentral locations and a rake angle equal to 1501.

Fig. 14. Comparison between GMPEs and average horizontal PGAs predicted by EXSIM for the Mw 5.5 Monferrato event, assuming various hypocentral locations.

Table 7

Source parameters of the rupture model for the Western Liguria Mw 6.5 ground shaking scenarios.

Mw M0 Fault centre Strike Dip VR L W Min. depth Max. depth Du Rise time rake Stress drop – (N m) (deg.) (deg.) (m/s) (km) (km) (km) (km) (m) (s) (deg.) (bar)

GRFLT12S Lat.: 43.74oN 1.24 1.4 45–90 – 6.5 6.31E+18 Lon.: 8.13oE 60 60 2600 17 9 4.1 11.9 135 Depth: 8 km EXSIM

1.18 1/f0sb –81

shaking scenario obtained with the assumption of a rake angle of 1351 and hypocentre n. 2 (Fig. 15). The velocity time histories computed by the two methods in the horizontal direction are in satisfactory agreement in terms of arrival time and peak amplitude of S-waves. Particularly, the average horizontal component com- puted by EXSIM (Fig. 17d) compares well with the EW component (Fig. 17b) calculated by GRFLT12S. In terms of frequency contribu- tion the results provided by the two methods are in agreement at low frequencies, below 0.6–0.7 Hz. At higher frequencies GRFLT12S results start to decay significantly. Fig. 15. Variation of the hypocentral location on the Western Liguria fault projected The maximum PGAs are those obtained by EXSIM with values on the surface. A total of 20 different nucleation points were considered. of 0.58 and 0.48 m/s2 calculated for hypocentre n. 11 and n. 2 (see Fig. 18), respectively. The last nucleation point provides also depending on the rake angle. A maximum PGV of 0.047 m/s was the most severe scenario in the low frequency range (see Fig. 17). predicted in the EW direction with a rake angle of 1351 and in the Concerning the validation of numerical results through GMPEs, vertical direction with a rake angle of 901. the same comments made for the Monferrato case still apply. Fig. 17 shows a comparison between seismograms computed A good agreement is achieved both in terms of PGV and PGA, in the low and high frequency ranges for the most severe ground as shown in Figs. 19 and 20. 768 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

Fig. 16. Effect of the rake angle on ground motion predicted at Vicoforte site in terms of velocity time histories simulated by GRFLT12S for the Mw 6.5 Western Liguria event, assuming the nucleation point number 2: NS (a), EW (b), and UD (c) components.

Fig. 17. Synthetic seismograms in terms of velocity time histories and corresponding Fourier spectra for the Mw 6.5 Western Liguria event, assuming the nucleation point number 2 and a rake angle equals to 1351: NS (a), EW (b), and UD (c) components computed by GRFLT12S (black); (d) horizontal average component computed by EXSIM (gray) whose Fourier spectrum is also overlapped on the two low frequency horizontal components in (a) and (b).

7. Comparison of results assuming different rake angles, according to the prevalent focal mechanism of each fault, are shown. As expected, the extended A total of 144 low frequency and 50 high frequency numerical kinematic source model implemented in GRFLT12S correctly simulations were performed to define the worst ground shaking simulates the directivity effects. In fact, referring to the Western scenarios. Liguria fault, which is characterized by a prevalent reverse mecha- Fig. 21 summarizes the results determined in the low frequency nism, maximum values of PGV occur for the deepest hypocentres range in terms of PGV along the three spatial directions. The values (numbers 1–5, see Fig. 15), since they radiate the seismic energy in of PGV obtained by the activation of various nucleation points and the direction towards the site of Vicoforte. It is also readily apparent L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772 769

Fig. 18. Acceleration time history and corresponding Fourier spectrum of the horizontal average component computed by EXSIM for the Mw 6.5 Western Liguria event, assuming the nucleation point number 2.

Fig. 19. Comparison between horizontal (a) and vertical (b) PGVs predicted by GRFLT12S for the Mw 6.5 Western Liguria event, assuming various hypocentral locations and a rake equal to 1351.

Fig. 20. Comparison between average horizontal PGAs predicted by EXSIM for the Mw 6.5 Western Liguria event, assuming various hypocentral locations.

the influence of the rake angle on computed PGV amplitudes in the of 901 yields the highest PGV in the NS and UD directions, three spatial directions (see Fig. 21a). A rake angle of 1351 yields to while an angle of 451 yields the EW largest values (see Fig. 21b). the maximum PGV in the EW direction, a rake angle of 901 yields It is important to remark that rake angles of 451 and 1351 (as the peak values in the UD direction, while along the NS direction indicated by DISS3.1 database [5]) yield the same values of PGV rake angles of 901 or 451 yield maximum PGV depending on the since they define the same slip direction. In fact, for the same hypocentral location. slip direction, the different sign (e.g. normal vs. reverse fault Concerning the prevalent normal Western Alps fault, values of mechanism) determines only the inversion of polarity of the time PGV increase for the shallowest nucleation points (numbers 9–12, histories which, as expected, preserve the same waveforms and Fig. 21b). The combination of these hypocentres with a rake angle amplitudes. 770 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

Fig. 21. Comparison between the results obtained through the low frequency simulations at the Vicoforte site in terms of PGVs along the three spatial directions (NS, EW, UD), assuming different hypocentral locations (abscissa) for each fault: (a) Western Liguria, (b) Western Alps, and (c) Monferrato.

Fig. 22. Comparison between the results obtained by the low and high frequency simulations at the Vicoforte site in terms of average horizontal PGAs and PGVs, assuming different hypocentral locations (abscissa) for each fault: (a) Western Liguria, (b) Western Alps, and (c) Monferrato.

The hypocentres located in the Eastern part of the Monferrato Velocity time histories by EXSIM were computed by derivation fault (numbers 3, 4, 7, 8, 11, 12 in Fig. 7) radiate the greatest amount of acceleration time histories, while accelerations by GRFLT12S of seismic energy in the direction towards Vicoforte, thereby were calculated by integration of velocity time histories. providing the largest PGV. In agreement with the prevalent It is important to remark that directivity effects cannot be strike-slip mechanism, a rake angle of 1501 yields the most severe simulated by EXSIM, since this programme is based on a stochastic ground shaking (see Fig. 21c). approach and it does not account for the slip direction. For this Fig. 22 compares the numerical results computed using reason the hypocentral location has not the same effect as in GRFLT12S and EXSIM in the low and high frequency ranges, GRFLT12S where it influences the radiation of energy from the respectively. Particularly, the values of average horizontal PGV source. However in EXSIM the hypocentral location was varied in and PGA are shown for various hypocentral locations and assuming order to assess the sensitivity of the results to the variation of this the rake angles that provide the maximum ground shaking. parameter, as shown in Fig. 22. L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772 771

Table 8 Comparison between peak ground motion parameters associated with the most severe ground shaking scenarios due to the reactivation of the three main faults of the area and that predicted by the PSHA [28].

GRFLT12S EXSIM PSHA

Mw H (km) TR (years) PGV PGA PGV PGA PGA (m/s) (g) (m/s) (g) (g)

Western Liguria 6.5 10 2475 0.047 0.009 0.065 0.059 0.160 Western Alps 5.7 3.5 475 0.010 0.004 0.020 0.023 0.096 Monferrato 5.5 12 690 0.020 0.013 0.019 0.021 –

As expected, high frequency seismograms predict larger values Western Alps fault). In fact, the PSHA predicted PGA values of of PGA if compared with the low frequency ones, as listed in Table 8. 0.096g and 0.16 g for the corresponding return periods of 457 and For this case study larger values of PGV are provided by EXSIM as 2475 years, which can be related to the recurrence periods of well, except for the Monferrato event, for which the maximum reactivation of the Western Alps and Western Liguria faults, PGV value estimated by both methods is in average the same respectively (see Table 8). (0.02 m/s). This difference in the results between deterministic and prob- The difference in ground motion parameters estimated by the abilistic methods was in a way expected. It is due to a complete two methods increases by increasing the magnitude of the event. different approach to seismic hazard of the two methods, although The Western Liguria source provides the most severe ground epicentral distances and magnitudes associated with the control- shaking scenario, yielding the largest values of PGA (0.058 g) and ling earthquakes yielded by deaggregation analyses of PSHA are PGV (0.065 m/s). Similar values of maximum PGV (0.02 m/s) comparable with the distance and magnitude of seismic fault and PGA of (0.02 g) were calculated for both the Western Alps ruptures from the site of Vicoforte [28]. The fact that the PSHA and Monferrato sources. provides higher values of ground motion parameters if compared with DSHA is expected, as shown by Bommer [9] and Krinitzsky [27]. Nevertheless, PSHA and DSHA are both important as they 8. Conclusions provide complementary information to the predicted hazard.

The present paper focused on deterministic seismic hazard assessment at Vicoforte, where the ‘‘Regina Montis Regalis’’ Acknowledgements Basilica, the world’s largest elliptical dome, sits. The main objective of the study was the computation of the most severe ground The present study was developed in the framework of the shaking scenarios, compatibly with the tectonic and seismic setting research contract undersigned with the Administration of Vicoforte of the region. To this aim three main seismic sources were identi- Sanctuary for the monitoring and survey of the ‘‘Monte Regalis fied in the area of interest: the Monferrato, Western Alps and Basilica’’, with the support of the Fondazione Cassa di Risparmio di Western Liguria faults. Finite fault numerical analyses were per- Cuneo. The authors would like to extend their gratitude to Prof. formed using two different approaches implemented in the codes Mario Alberto Chiorino, Coordinator of the Program (Politecnico di GRFLT12S [24] and EXSIM ([38,11]), which generate synthetic Torino) and to the Administration of the Sanctuary for their seismograms in the low and high frequency ranges, respectively. continuous and unstinted support during various stages of the The recent earthquake of August 21, 2000 of magnitude Mw¼4.9 project. The authors would also like to thank Prof. R.J. Archuleta was modelled with the aim to calibrate the results of numerical (University of California, Santa Barbara) for his scientific support in simulations. A satisfactory agreement was obtained between syn- the simulation of synthetic seismograms at Vicoforte. thetic seismograms and recorded data. Afterwards, a parametric study was performed for a total of 144 low frequency and 50 high References frequency simulations in order to identify the most critical rupture mechanisms and thus ground shaking at Vicoforte site. [1] Ambraseys NN, Douglas JJ, Sarma SK, Smit PM. Equations for the estimation of From the comparison of numerical results, it turns out that high strong ground motions from shallow crustal earthquakes using data from frequency simulations predict larger values of ground motion Europe and the Middle East: horizontal peak ground acceleration and spectral parameters if compared with low frequency simulations. This is acceleration. Bull Earthquake Eng 2005;3:1–53. [2] Anderson JG, Hough SE. A model for the shape of the Fourier amplitude true, as expected, for PGA but also for PGV. The low frequency spectrum of acceleration at high frequencies. Bull Seismol Soc Am 1984;74: analyses were used to identify the fault rupture models yielding the 1969–93. worst ground shaking scenarios. In fact, they account for directivity [3] Atkinson M, Assatourians K, Boore DM, Campbell K, Motazedian D. A guide to differences between stochastic point-source and stochastic finite-fault simu- effects, which are correctly simulated by GRFLT12S, but ignored lations. Bull Seismol Soc Am 2009;99(6):3192–201. by EXSIM programme. The fundamental mode of vibration of the [4] Barani S, Spallarossa D, Bazzurro P, Eva C. Sensitivity analysis of seismic hazard Basilica is about 0.6 s [28], which corresponds to a frequency of for Western Liguria (North Western Italy): a first attempt towards the 1.67 Hz. This frequency range can be correctly captured by both understanding and quantification of hazard uncertainty. Tectonophysics 2007;435:13–35. low and high frequency approaches. [5] Basili R, Valensise G, Vannoli P, Burrato P, Fracassi U, Mariano S, et al. The deterministic ground shaking scenarios confirm the low The Database of Individual Seismogenic Sources (DISS), version 3: summariz- seismicity of the area with a maximum PGA of 0.065 g and PGV of ing 20 years of research on Italy’s earthquake geology. Tectonophysics 2008, doi:10.1016/j.tecto.2007.04.014. 0.059 m/s (these are associated to the Western Liguria fault). [6] Be´thoux N, Frechet J, Guyoton F, Thouvenot F, Cattaneo M, Eva C, et al. A closing Comparing the results yielded by DSHA and PSHA [28], it turns Ligurian Sea? Pageoph 1992;139:179–94. out that the most severe ground shaking is that predicted by the [7] Bindi D, Luzi L, Massa M, Pacor F. Horizontal and vertical ground motion prediction equations derived from the Italian Accelerometric Archive (ITACA). PSHA, with PGA values which exceed the ones predicted by EXSIM Bull Earthquake Eng 2009, doi:10.1007/s10518-009-9130-9. Published online: by a factor from 2 (for the Western Liguria fault) to 5 (for the 16 June. 772 L. Scandella et al. / Soil Dynamics and Earthquake Engineering 31 (2011) 757–772

[8] Blundell D, Freeman R, Mueller S, editors. A continent revealed: the European [29] Larroque C, Bertrand Delouis B, Godel B, Nocquet JM. Active deformation at the geotraverse. New York: Cambridge University Press; 1992. southwestern Alps–Ligurian basin junction (France–Italy boundary): evidence [9] Bommer JJ. Deterministic versus probabilistic seismic hazard assessment: an for recent change from compression to extension in the Argentera massif. exaggerated and obstructive dichotomy. J Earthquake Eng 6 Special Issue Tectonophysics 2008;467:22–34. 2002;1:43–73. [30] Lippitsch. R, Kissling E, Ansorge J. Upper mantle structure beneath the Alpine orogen [10] Boore DM, Atkinson GM. Ground-motion prediction equations for the average from high-resolution teleseismic tomography. J Geophys Res 2003;108:2376–91. horizontal component of PGA, PGV, and 5%-damped PSA and spectral periods [31] Marin S, Avouac J-P, Nicolas M, Schlupp A. A probabilistic approach to seismic between 0.01 s and 10.0 s. Earthquake Spectra 2008;24(1):99–138. hazard in Metropolitan France. Bull Seismol Soc Am 2004;94(6):2137–63. [11] Boore DM. Comparing stochastic point-source and finite-source ground- [32] Massa M, Eva E, Spallarossa D, Eva C. Detection of earthquake clusters on the motion simulations: SMSIM and EXSIM. Bull Seismol Soc Am 2009;99: basis of the waveform similarity: an application in the Monferrato region 3192–216. (Piedmont, Italy). J Seismol 2006;10:1–22. [12] Boore DM, SMSIM; Fortran Programs for Simulating Ground Motions from [33] Mavroeidis GP, Dong G, Papageorgiou AS. Near-fault ground motion, and the Earthquakes: Version 2.3A Revision of OFR 96-80-A, U.S. Geological Survey response of elastic and inelastic single-degree-of-freedom (SDOF) systems. Open-File Report, 2005 (A modified version of OFR 00-509, describing the Earthquake Eng Struct Dyn 2004;33:1023–49. program as of 15 August, 2005 (version 2.30)). [34] McGuire RK. Deterministic vs. probabilistic earthquake hazards and risks. Soil [13] Calderini C, Chiorino MA, Lagomarsino S, Spadafora A. Non linear modelling of Dyn Earthquake Eng 2001;21:377–84. the elliptical dome of Vicoforte. New Delhi: Structural Analysis of Historical [35] Monachesi G, Stucchi M. DOM4.1 un database di osservazioni macrosismiche / Constructions; 2006. ISBN:972-8692-27-7. di terremoti di area italiana al di sopra della soglia del danno, 2000, http:// S [14] Chaumillon E, Deverchere J, Rehault JP, Gueguen E. Re´activation tectonique et emidius.mi.ingv.it/DOM/ . flexure de la marge continentale ligure (Mediterranee Occidentale). CR Acad [36] Morasca P, Malagnini L, Akinci A, Spallarossa D, Herrmann RB. Ground-motion Sci Paris 1994;319:675–82. scaling in the Western Alps. J Seismol 2006;10:315–33. [15] Chiorino MA, Aoki T, Calderini C, Spadafora A. Monitoring and modelling [37] Morasca P, Mayeda K, Gok¨ R, Phillips WS, Malagnini L. 2-D coda and direct wave strategies for the world’s largest elliptical dome at Vicoforte. New Delhi: tomography in Northern Italy. Bull Seismol Soc Am 2008;98:1936–46. [38] Motazedian D, Atkinson G. Stochastic finite-fault modeling based on a dynamic Structural Analysis of Historical Constructions; 2006. ISBN:972-8692-27-2. corner frequency. Bull Seismol Soc Am 2005:995–1010. [16] Cotton F, Scherbaum F, Bommer JJ, Bungum H. Criteria for selecting and [39] Pelinovsky E, Kharif C, Riabov I, Francius M. Study of tsunami propagation in adjusting ground-motion models for specific target regions: application to the Ligurian Sea. Nat Hazards Earth Syst Sci 2001;1:195–201. central Europe and rock sites. J Seismol 2006;10:137–56. [40] Pelinovsky E, Kharif C, Riabov I, Francius M. Modelling of tsunami propagation [17] Delacou B, Sue C, Champagnac JD, Burkhard M. Present-day geodynamics in the in the vicinity of the French coast of the Mediterranean. Nat Hazards bend of the Western and central Alps as constrained by earthquake analysis. 2002;25:135–59. Geophys J Int. 2004;158:753–74. [41] PUNCH A. Computer program for the determination of the Green’s functions in [19] Eva C, Rabinovich AB. The February 23, 1887 tsunami recorded on the Ligurian layered media by E. Kausel. USA: Massachusetts Institute of Technology; 2005. coast, western Mediterranean. Geophys Res Lett 1997;24:2211–4. [42] Sabetta F, Pugliese A. Estimation of response spectra and simulation of nonsta- [20] Eva E, Solarino S. Variations of stress directions in the western Alpine arc. tionary earthquake ground motion. Bull Seism Soc Am 1996;86(2):337–52. Geophys J Int 1998;135:438–48. [43] Scafidi D, Solarino S, Eva C. P wave seismic velocity and Vp/Vs ratio beneath the [21] Ferrari G. The 1887 Ligurian earthquake: a detailed study from contemporary Italian peninsula from local earthquake tomography. Tectonophysics scientific observations. Tectonophysics 1991;193:131–9. 2009;465:1–23. [22] Geller RG. Scaling relations for earthquake source parameters and magnitudes. [44] Schmid SM, Kissling E. The arc of the Western Alps in the light of geophysical Bull Seismol Soc Am 1976;66(5):1501–23. data on deep crustal structure. Tectonics 2000;19:62–85. [23] Gutenberg B, Richter CF. Earthquake magnitude, intensity, energy and accel- [45] Spallarossa D, Ferretti G, Augliera P, Bindi D, Cattaneo M. Reliability of eration. Bull Seismol Soc Am 1956;46:105–45. earthquake location procedures in heterogeneous areas: synthetic tests in [24] Hisada Y, Bielak J. A theoretical method for computing near fault ground the South Western Alps Italy. Phys Earth Planet Inter 2001;123(2–4):247–66. motion in a layered half-spaces considering static offset due a surface faulting, [46] Stafford PJ, Strasser FO, Bommer JJ. An evaluation of the applicability of the with a physical interpretation of fling step and rupture directivity. Bull Seism NGA models to ground-motion prediction in the Euro-Mediterranean region. Soc Am 2003;93(3):1154–68. Bull Earthquake Eng 2008;6:149–77, doi:10.1007/s10518-007-9053-2. [25] Kahle HG, Geiger A, Buerki B, Gubler E, Marti U, Wirth B, et al. Recent crustal [47] Stucchi M, Camassi R, Rovida A, Locati M, Ercolani E, Meletti C, et al. DBMI04, il movements, geoid and density distribution; contribution from integrated database delle osservazioni macrosismiche dei terremoti italiani utilizzate satellite and terrestrial measurements. In: Pfiffner OA, Lehner P, Heitzmann P, per la compilazione del catalogo parametrico CPTI04. Quad Geofis 2007;49:38. Mueller S, Steck A, editors. Deep structure of the Swiss Alps: results of NRP20. /http://emidius.mi.ingv.it/DBMI04S. Basel: Birkha€user Verlag; 1997. p. 251–9. [48] Tinti S, Maramai A, Graziani L. The new catalogue of Italian tsunamis. Nat [26] Klugel¨ J-U. Problems in the application of the SSHAC probability method for Hazards 2004;33:439–65. assessing earthquake hazards at Swiss nuclear power plants. Eng Geol [49] Waldhauser F, Kissling E, Ansorge J, Mueller S. Threedimensional interface 2005;78:285–307. modelling with two-dimensional seimic data: the alpine crust-mantle bound- [27] Krinitzsky EL. How to combine deterministic and probabilistic methods for ary. Geophys J Int 1998;135:264–78. assessing earthquake hazards. Eng Geol 2003;70(1–2):157–63, doi:10.1016/ [50] Wells DL, Coppersmith KJ. New empirical relationships among magnitude, S0013-7952(02)00269-7. rupture length, rupture width, rupture area, and surface displacement. Bull [28] Lai CG, Corigliano M, Sa´nchez Liza´rraga H, Scandella L. Definition of seismic Seism Soc Am 1994;84:974–1002. input at the ‘‘Regina Montis Regalis’’ Basilica of Vicoforte. Northern Italy: IUSS [51] Working Group CPTI, Catalogo Parametrico dei Terremoti Italiani, version 2004 press; 2009. ISBN:978-88-6198-039-6. (CPTI04), INGV, Bologna, /http://emidius.mi.ingv.it/CPTI/S.