Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 Contents lists available at ScienceDirect Soil Dynamics and Earthquake Engineering journal homepage: www.elsevier.com/locate/soildyn Simulated shaking maps for the 1980 Irpinia earthquake, Ms 6.9: Insights on the observed damage distribution Maria Lancieri a,, Aldo Zollo a,b a Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Napoli, Via Diocleziano 328, 80124 Napoli, Italy b Dipartimento di Scienze Fisiche, Università di Napoli Federico II, Via Diocleziano 328, 80124 Napoli, Italy article info Article history: Received 6 June 2006 Received in revised form 10 June 2008 Accepted 25 January 2009 Keywords: Irpinia 1980 earthquake Strong motion simulation Shake maps Damage scenario abstract The region stricken by the Irpinia earthquake (Ms ¼ 6.9) on 1980, along the Southern Appenninc chain, is one of the highest seismic hazard areas of Italian peninsula. This event produced vast damaging and strong amplitude shaking on a wide area. This is mostly related to the occurrence of a multiple fracture process during which three different segments of a subparallel normal fault system have been activated. The principal aim of this study is to investigate how the source complexity can have influenced the ground motion recorded at the earth surface and the areal distribution of strong motion parameters and observed building damage. We use a deterministic approach to simulate the ground motion scenario in the whole Campania region generated by the Irpinia event, based on the present knowledge of its rupture history and multiple faulting geometry. The kinematic source model has been calibrated by comparing the observed and synthetic data in frequency and time domain. Maps of the main strong motion parameters and damage maps for different class of buildings and levels of damage have been computed and quantitatively compared to the observations of damages and macroseismic intensity data. & 2009 Elsevier Ltd. All rights reserved. 1. Introduction A deterministic earthquake scenario describes the ground motion and the effects generated by a large earthquake in a given target area. It represents a useful tool for hazard assessment and risk mitigation. In the case of the occurrence of a strong event, the rapid estimation of the strong ground shaking can be an advantageous for planning and coordinating emergency response and for alert notifications to civil protection agencies. Based on the fast availability of event location and magnitude, a method called TriNet ShakeMap [1] provides post-event characterization of strong motion, combining instrumental measures of shaking with attenuation laws and information on local geology. If a dense strong-motion network is deployed on the epicentral area, the ShakeMap methodology provides an accurate assessment of shaking level, while in case of sparse coverage of instrument is suitable to estimate ground motion using a simple source model able to include the effect of earthquake source finiteness and rupture directivity. Corresponding author. Now at: Ecole Normale Supérieure, 24 rue Lhomond, 75231 Paris, cedex 05, France. Fax: +39 08124203 34. E-mail addresses: maria.lancieri@na.infn.it (M. Lancieri), aldo.zollo@na.infn.it (A. Zollo). Dreger and Kaverina [2], proposed a methodology to fast determine the earthquake source process and providing strong motion information. Their technique is based on the inversion of the representation theorem for finite source parameters [3]. Fault length and area are estimated using the Wells and Coppersmith [4] relationships, fault orientation is based on the available moment tensor solution and the rupture velocity is assumed to be constant and equal to the 80% of S-wave velocity. This methodology is difficult to apply to near source strong motion data and in near-real time since the accelerogram inversion at high frequencies can be very unstable [5]. More recently Yamada and Heaton [6] developed a near-real time inversion technique that allows to estimate within 20 s after the event occurrence, the fault strike orientation length assuming a bilateral rupture model. This technique requires a dense azimuthal coverage distribution of the earthquake source by near source stations. In this paper we introduce a technique similar to Yamada and Heaton s [6] for rapid simulation of strong motion accelerograms, where the earthquake source is modelled using a line source oriented along the strike. The methodology is based on convolution of several point sources having a Haskell type source time function. The rapid simulation of ground shaking after the occurrence of a destructive event is only the first step to build an earthquake scenario. The second task is to evaluate the earthquake severity by using the relationships between the ground motion parameters 0267-7261/$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.soildyn.2009.01.007
M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 1209 and the damage level. Our aim is to demonstrate that it is possible to obtain a fast and reliable prediction of the damage pattern including the effect of rupture length finiteness and directivity. The damage can be expressed in terms of Mercalli Intensity maps or in terms of structural damage for different building classes. In the first case, specific empirical relationships are required to correlate the Mercalli Intensity and ground motion quantities as peak ground acceleration (PGA), velocity (PGV) and/or response spectra [1,7,8]. An alternative strategy would be to use the deterministic assessment of the ground shaking jointly with vulnerability relationships which express the probability of a class of structures to reach or to exceed various limit states at given levels of ground motion. Few ground motion damage relationships have been developed for European countries, based on damage surveys [9,10]. The Orsini [10] vulnerability curves characterize the shaking level in terms of intensity evaluated from observations of earthquake effects, while Sabetta et al. [9] evaluated the distribution of the damage as a function of the seismic input (PGA, Arias intensity and effective peak acceleration) from damage surveys carried out after some destructive Italian earthquakes (Abruzzo 1984, Irpinia 1980). Rossetto and Elnashai [11] recently proposed the vulnerability curves for reinforced concrete building based on a dataset of 99 post-event damage distribution, correlating the probability to exceed a given damage level and peak ground acceleration, spectral acceleration and displacement for 5% of critical damping. In this paper we apply our numerical simulation technique to produce the ground shaking and damage scenario of the Mw ¼ 6.9 Irpinia (Southern Italy) 1980 event. 2. The 1980 Irpinia earthquake The maximum instrumental earthquake for the Irpinia region (Southern Italy) occurred on November 23rd, 1980. The Irpinia earthquake (M ¼ 6.9) was produced by a complex normal faulting event, characterized by three main rupture episodes nucleating at about 0, 20 and 40 s along different fault segments [12]. The source complexity determined the anomalous shaking motion (the near source station recorded more than 70 s of signal see Fig. 5), whose consequence was a dramatic effect on the population: more than 3000 people killed, an epicentral intensity of X of Mercalli scale, an isoseismal area VII of about 100 km length. This event marked the beginning of quantitative seismic hazard assessment in southern Italy, and the wide documentation and data-set available for it provide a good opportunity for examining the details of a complex normal faulting for which the fault segment geometry, dimension and mechanisms have been widely studied using different approaches and a multi-disciplinary data set. Westaway and Jackson [13] first discovered more than 10 km of breakage surface faulting. They jointly used geological and geodetic observation as well as seismometric and accelerometric recordings to propose a reconstruction of fault geometry and source time function. Bernard and Zollo [12] constrained the timing, geometry and kinematics of the multiple segment rupture process through a detailed analysis of near-source strong motion and levelling data [15], integrating the teleseismic waveform modelling of Westaway and Jackson [14], aftershock locations and mechanisms [16] (Fig. 1) Following Westaway and Jackson [13,14], Pantosti and Valensise [17] performed an extended geological survey of the surface breakages produced by the event and provided measurements of fault slip at the surface. Cocco and Pacor [18] performed the slip inversion for the zero sec event using strong motion data. In present study the source parameters of the kinematic rupture model developed by Bernard and Zollo [12] (see Fig. 2 and relative source parameters Table 1) are used. 3. The strong motion simulation method The basic idea of our approach is to simulate the ground shaking of a large event describing its rupture process with a along-strike, line source model and complete wave field Green s function [19,20]. The evaluation of the complete wave field Green s function makes it possible to correctly estimate the PGA at sites where this is brought by surface waves and/or secondary arrivals, even if it can result very expensive in term of computing time. Engineering applications often make use of the point source model to predict earthquake ground motion parameters; however this model is unable to reproduce the correct signal duration, and it does not take in account the effects of rupture directivity and of fault length finiteness that can significantly affect the seismic records and the inferred damage, not only for large event but also for event of moderate size [21,22]. To represent the rupture process over an extend fault, various kinematic approaches have been developed [23 25]. However, because of the expensive computational efforts required, for the immediate post-event shake-map computations simple source/ propagation models are generally adopted [1]. For instance, the stochastic method [26] allows to rapidly simulating accelerograms for a given earthquake, characterized by seismic moment, corner frequency and duration, where the high frequency content is purely stochastic, or alternatively the peak ground motion can be predicted using empirical attenuation relationships. In those cases the resulting peak ground motion pattern is isotropic. 3.1. The source model Between a point source and an extended fault model we propose and intermediate source approximation, i.e. the line source model. Based on Haskell [27] model, the line source model takes into account the fault directivity, rupture velocity and length, and assumes a uniform slip distribution along the line. The main advantage of this simplified source model is that only few parameters are needed to describe the faulting process: line length, orientation, depth, rupture velocity, average slip, so that it can be used for a rapid computation of peak ground shaking maps. The line source model is built by positioning a series of equally spaced, double couple point sources along the line, each of them having the same source duration, mechanism and moment. The sum of point source seismic moments is set to be equal to the event seismic moment. If a uniform dislocation model is used, the synthetic seismograms are dominated by the stopping and nucleation phases that give rise to anomalous peaks on synthetic records. To reduce this effect, a 10% tapering of the moment distribution at the line edges may be applied, and the moment value for each point source along the line may be re-normalized accordingly. In order to set parameters for the line source model we assume that the event magnitude (or seismic moment), location and focal mechanism are a priori known from previous estimation. A rough estimation of the fault length can be also derived from the event magnitude using the Wells and Coppersmith
1210 M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 Real Data Acceleration (m/s 2 ) Shallow source line (3 Km depth) Time (s) Real Data Acceleration (m/s 2 ) Deep Source line (8 Km) Deep Source line (10 Km) Time (s) Fig. 1. Effect of the source depth on relative amplitude of the phases contributing to the signal. The synthetic signal is simulated in (0.1 8 Hz) frequency range. Top comparison between the observed and the synthetic signals for a shallow line source (3 km of depth). The surface waves amplitude generates a peak ground acceleration 3 times greater than the actual one. Bottom comparison between the observed and the synthetic signals for deeper line sources (8 and 10 km). The acceleration amplitude is comparable with the observed and the peak is correctly associated with the S waves. The amplitude does not show significant differences between the 8 and 10 km depth. relationship [4] L ¼ 10 2:44þ0:59 dmw (1) where L is the length of the rupture, and Mw is the moment magnitude. The rupture velocity (V r ) is assumed to be constant and equal to the 80% of the shear waves velocity in the investigated area. The elementary point source spacing and their rise-time are fixed on the basis of the maximum frequency chosen for the numerical accelerogram computation. The aliasing effect is avoided by positioning the elementary sources at a distance smaller than the minimum wavelength given by [28] where fny is the selected Nyquist frequency, and v r the fixed rupture velocity. We decided to use at least 6 point sources for the minimum wavelength, and consequently the source spacing is given by v r Ds ¼ l 6 ¼ (3) 6f ny In addition the signals emitted from each elementary source must overlap at the receiver, which means that the source duration (rise-time) of each point source (rt) has to be greater than the time needed by the rupture front to reach each single point source l ¼ v r f ny (2) rtb v r Ds (4)
M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 1211 M 40 =3e18 Nm 40 s (70,124,-90) M 0 =13.2 e18 Nm SW 1 NE (60,315,-90) 0 s 1 20 s 2 2 M 20 =4e18 Nm (20,300,-90) Fig. 2. Source model. Comparison between the Irpinia 1980 earthquake fault model described by Bernard and Zollo (1989) [12], and the source model used in this work. The event was characterized by three main rupture segments, nucleating at 0, 20 and 40 s. Grey stars on left panel represent the nucleation points, the grey lines are the modelled line sources placed at 8 km (0 s segment) and 5 km (20 and 40 s segments.) The bold black lines are the fault top. The 0 sub-event fault is modelled as a 40 km long line source (red line in left panel) characterized by a bilateral fracture propagating 5 km towards SE direction and 35 km towards NW direction. The 20 s (20 km line source) and 40 s (10 Km line source) fractures are unilateral and propagating towards SE direction. The focal mechanism is also reported: the 20 s fault plane has a small dip angle (201), and the 40 s fault plane is antithetic and dips in the NE direction. Table 1 Source parameters for the 1980 Irpinia earthquake. After Bernard and Zollo (1989). Sub-event (s) Hypocentral coordinates 3.2. The propagation model Focal mechanism Dip strike rake Seismic moment (Nm) 0 151E 19.92 401N 46.68 60, 315, 90 13.2e19 20 151E 26.00 401N 40.00 20, 300, 90 4.0e18 40 151E 12.10 401N 51.00 70, 124, 90 3.0e18 In our approach the Green s functions are evaluated in a 1D layered velocity model, using the discrete wave number method [19]. The Green s functions are computed using the discrete wave number method introduced by Bouchon and Aki in 1977 [19], and numerically implemented in the Axitra code by Coutant [20]. This method introduces a spatial periodicity of sources to discretize the radiated wave field, and it is based on the Fourier transform in the complex frequency domain to calculate the Green s functions. Hisada [29,30] describes the problems that occur in wavenumber integration codes when: source and observer have the same depth; source is close to the boundary between two layers; source lies on a boundary. These critical conditions have to be carefully analysed when deciding the line source position within the 1D velocity model. 3.3. Modelling of the Irpinia, 1980 fault planes We describe the rupture process of the Irpinia earthquake by a combination of line sources activated along the three fault segments. In this application we use the Bernard and Zollo (1989) model [12] to constrain the following parameters: constant velocity rupture of 3 km/s for all segments fault length of 0 s sub-sources equal to 40 km fault length of 20 s sub-sources equal to 20 km fault length of 40 s sub-sources equal to 10 km bi-lateral rupture on 0 s sub-sources propagating for 5 km toward SE and 35 km in NW direction (see Fig. 2); and uni-lateral rupture propagation on both the ruptures associated to 20 and 40 s events assumed to be uni-lateral and to propagate toward SE. (see Fig. 2) On the basis of a series of trial simulations with different frequency, we found that a Nyquist frequency of 8 Hz gives us the
1212 M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 best compromise between source spacing (wavelength sampling) and computational time. From Eqs. (2) and (3) the source spacing is equal to 60 m on each sub-fault with a rise time 0.05 s. 0s Fault Depth 18s 39s Faults Depth Fig. 3. Comparison between the Bernard and Zollo (1989) [12] velocity model (dashed line) used for event location, and the 1D velocity model used in this work [30]. Table 2 D velocity model synthetized for the Southern Apenninic chain. The Q p and Q s are the quality factors for P and S waves respectively. h (m) V p (m/s) V s (m/s) r (g/cm 3 ) Q p Q s 0 2000 1150 2100 40 20 250 3200 1850 2300 40 20 1000 4500 2600 2500 100 50 2500 6200 3600 2700 250 150 15,000 7400 4300 2900 900 300 The 1D velocity model used for Green s function evaluation, shown in Fig. 3, has been synthesized from the geological proposed by Improta et al. [31] for the Irpinia region. The very shallow layers reproduce the sedimentary sequence of the Bradano Trough Quaternary deposits, and the Lagonegro basin, characterized by relatively low P and S velocities. The main seismic discontinuity at 2.5 km depth represents the top of the Mesozoic limestone corresponding to the Apulian Carbonate Platform. The deepest discontinuity represents the basement, characterized by high P and S velocities. The values of Vp, Vs, Qp and Qs are shown in Table 2. We use low Q factor values for shallow layer, while higher Q factor values are used for the deepest layer [32]. The depths of the line sources have been mainly constrained based on the location depths of the main rupture events of the 1980 Irpinia earthquake as inferred by Bernard and Zollo [12]. We decided to place the line source inside that Apulian Carbonate platform layer, where the three sub-events have been located (see Table 1). The line source depth also affects the relative amplitude of the phases contributing to the signal. In Fig. 1a we show three different simulations for the 0 s sub-fault line, corresponding to a depth of 3, 8 and 10 km. For the shallowest source position, the maximum acceleration peak is six times greater than the actual peak and the maximum acceleration is carried by surface waves, while, on the observed trace, the peak is brought by direct S waves (e.g., Bernard and Zollo, 1989 [12]). Moving the line source deeper (Fig. 1b), the acceleration peak is correctly associated to the direct S waves. The peak amplitude does not significantly vary with a line source at 8 or 10 km depth. Consequently, we decided to fix the depth of the line source at 8 km, at about half the depth of the layer. Following similar arguments we concluded that a line source placed at 5 km depth better justify the contribution of the surface waves to the observed amplitude for the 20 and 40 s subevents at closest stations. The chosen source depths are consistent with location of maximum slip patches on the fault slip inversion [18] and with depth distribution of aftershock distribution [16]. 4. Validation of the source model In order to validate the simulation method and calibrate the source parameters, we computed peak quantities, spectral 10 0 10 3 PGA (g) 10-1 Synthetic data Real Data Abrahamson 1987 Boore 1997 Campbell 1997 Sabetta 1996 Distance (km) Distance (km) Fig. 4. Peak ground acceleration and velocity measured on synthetic and real data. The circles represent the peak values measured on the synthetic data, the black filled diamonds represent the peak measures on real data, both filtered in the 0.1 8 Hz range. The PGV and PGD, that are controlled by the intermediate and low frequencies, coincide at all the station. The dashed curves are the Abrahamson attenuation law (M ¼ 6.9 normal fault, no hanging wall effects) [34]; Campbell attenuation law (M ¼ 6.9 strike-fault) [35], Boore attenuation law (M ¼ 6.9, shear-wave velocity 30 m ¼ 1150 m/s) [36]. The black solid curve is the Sabetta and Pugliese attnuation curve for PGA and PGV [37]. PGV (cm/s) Synthetic data Real Data 10-2 10 0 10 0 10 1 10 2 10 0 101 10 2 10 2 10 1
ARTICLE IN PRESS M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 ordinates and waveforms from synthetic data and compared them with the accelerograms recorded at the national network owned by the Italian electric agency (ENEL). The ENEL network consisted of SMA-1 Kinemetrics accelerographs. These instruments work in trigger mode without recording any pre-event time window. In many cases recordings are triggered on high amplitude first S waves and P phases of first recorded event do not appear. Sometimes one of the three sub-events is not contained in the recorded signal: this is the case of records at Auletta, Brienza and Tricarico stations which do not show the zero seconds event, while at Benevento station the 40 s event has not been recorded [12]. In our calibration study we selected 12 records from near-fault stations (Fig. 5). 4.1. Peak motion quantities At first, we compare peak ground acceleration and velocity (Fig. 4) with the observed data and with existing attenuation relationships. All data are plotted as a function of minimum 1213 distance from the fault top projections at the earth surface. The observed and synthetic data have been filtered in the 0.1 8 Hz frequency range. Synthetic data are simulated at bedrock condition (i.e. with any site response correction), while most of the recording sites are placed on deep soil deposits [33]. The goal of such a comparison is to understand if the simulated amplitudes are compatible with those deduced from the attenuation laws and if the discrepancies between simulated and observed amplitude are justifiable in terms of site response amplification. Significant PGA discrepancies are observed at station Tricarico (54 km distant from the 0 s fault) where the synthetic PGA is overestimated by a factor 4, and at station Sturno (19 km distant from to 0 s fault) where the PGA is underestimated by a factor 3. Such discrepancies are likely due to site amplification effects. In fact, Tricarico is situated on a calcarenite outcrop, lying above an argillaceous terrain with lower shear wave velocity, while Sturno station lies on silty clays and marls, interlayed by marly limestones and quartz sandstone. Line sources Nucleation points 1 SW NE 2 Fig. 5. Acceleration, velocity and displacement waveforms. The grey lines are the recorded data, the black lines are the synthetic waveforms. Since the origin time are not available, data are aligned using the Bernard and Zollo (1989) S waves timing [12].
1214 M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 We compare data trends with several attenuation laws used to predict PGA as a function of minimum distance from the fault projection at the earth surface: [34 37]. The black solid line is the Sabetta and Pugliese [37] attenuation curve that was obtained using a data base of accelerometric records of Italian earthquakes which included also the Irpinia event, and it provides the better fit to the data. We note that both filtered PGA and PGV values show a significant decay with distance, with PGA well consistent with the Sabetta and Pugliese attenuation relationship. 4.2. Waveforms and spectral shapes We further proceed on the comparison between real and synthetic waveforms records, both filtered in the 0.1 8 Hz frequency range, in the acceleration, velocity and displacement domains (Fig. 6). Due to instrument characteristics, the origin time of observed data was not available, so the 40 s picking of Bernard and Zollo [12] has been used in order to align synthetic and real data. Despite of the use of a flat-layered velocity model and a composite line-source model, the synthetic signals show rather complex waveforms. The whole duration of the signals is well reproduced at different sites for the different sub-events, and the time spacing between phase arrivals from the 0, 20 and 40 s events on synthetic accelerograms at all the stations is reproduced with an uncertainty of less than 0.5 s (Fig. 5). Such an observation confirms the cogency of the adopted line source model to reproduce the relative timing of the rupture propagating along the three fault segments. The relative amplitude of three sub-events is well reproduced at Sturno, Bovino, Bisaccia, Mercato, Bagnoli, Torre del Greco sites, where the first sub-event dominated the recorded shaking, and for Rionero where the signal maximum amplitude is related to the 40 s event. This condition is not verified at the Auletta station. At this site the maximum amplitude on observed seismogram corresponds to the 20 s event, while on synthetic waveforms it is dominant the contribute of the 40 s segment. In order to better assess the reliability of the used source model, we compare the low-frequency signals, i.e. the displacement and velocity waveforms. In many cases, the simulated velocity and displacement are able to reproduce the duration and amplitude variation on observed records; in particular the station of Sturno, located Line sources Nucleation points SW 1 NE 2 Fig. 6. Horizontal Response Spectra 5% damped grey lines are the recorded data, the black lines are the syntetic waveforms. In many stations synthetic and real PSV appear in good accordance around 1 Hz. The high frequency asymptote is the PGA value.
M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 1215 As it is expected, moving at higher frequencies, our simple source/propagation model is not able to fully reproduce the complexity of the observed accelerograms. This relies on the line source approximation that prevents to reproduce the high frequency radiation related to the variation of velocity rupture and slip asperities. But it also depends on the 1D propagation model and on the absence of effects of shallow layers reverberations. In Fig. 6 the response spectra, PSV (with 5% damping) are also shown for several stations. As general remark the shape of real and synthetic PSV show a good accordance in the investigated frequency range, with the exception of Brienza and Bagnoli stations. The synthetics PSV are not able to reproduce the amplification effect observed at Calitri and Bisaccia, for frequency below 1 Hz, and at Torre del Greco and Brienza, for frequency greater than 1 Hz. Such amplification effects are probably related to the sites whose effects are not included in the synthetic. 5. Shaking maps Due to the relative simplicity of the source/propagation model, the proposed source model can be used to produce a rapid evaluation of deterministic ground shaking scenarios over a large area of interest [6]. Using the kinematic model described above, a massive strong ground motion simulation has been performed at the nodes of a 20 km spaced grid covering the whole Campania region on an area of 160 120 km 2. The values of PGA, PGV and PGD (geometric means of the horizontal components) have been evaluated from synthetic seismogram and then interpolated obtaining the ground shaking maps plotted in Fig. 7. The rupture direction and the fracture complexity control the spatial behaviour of the ground motion level, giving rise to asymmetrical distribution of peak values in shake maps. A comparison between the observed and computed Mercalli intensity maps is also shown in Fig. 8. The observed map is expressed in terms of Mercalli Cancani Sieberg (MCS). For the computation of the intensity the Faccioli and Cauzzi [38] PGA MCS intensity relationship has been used I ¼ 1:83 d log PGA þ 6:57 The rose diagram in Fig. 8 is centered at the epicentral area, and shows the azimuthal variation of the root mean square (RMS) of I- residuals evaluated for 201 ranges. The RMS is strongly reduced in correspondence of the rupture direction. The higher RMS value in the east direction can be attributed to the poor station coverage since only few sites determine the RMS value, while in the west direction we cannot reproduce the macroseismic intensity over the south-west peninsula. 6. Relationship between ground shaking and structural damages Fig. 7. Shake maps for the Campania region. The maps are obtained interpolating the peak values simulated at a regular grid of station, 20 km spacing, that covers an (160 120) km 2 area. The peak values have been interpolated using a surface filter. 10 km far from the zero seconds event top, in directive position, shows a good fit of both the displacement and velocity waveforms. The strong-motion simulation method has been applied to investigate the relationship between the peak ground acceleration and the structural damage produced by the 1980 Irpinia event. The existing motion damage relationships for the Italian territory are expressed in terms of fragility curves that make use of macroseismic intensity as ground shaking parameters [10,39]. Rossetto and Elnashai [11] offer an exhaustive discussion on the disadvantages of using the intensity as ground motion characterization. In particular they point out that the intensity has the
1216 M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 41 00' 40 30' 10 9 8 7 6 5 4 3 Macroseismic Intensity Synthetic Data NORTH km 0 50 40 00' 14 00' 14 ' 1 00' 30 5 15 30' 6 1 00' WEST EAST 41 00' 40 30' 10 9 8 7 6 5 4 3 Macroseismic Intensity Synthetic Data 0.2 km 0 50 40 00' 14 00' 14 ' 1 00' 30 5 15 30' 6 1 00' Fig. 8. Intensities distribution. The maps show the observed intensities distribution (after G.N.D.T.) (upper) and the simulated intensities using synthetic PGA and the Faccioli, Cauzzi (2006) [38] relationship (lower). The intensity distribution appears to be strongly influenced by the fracture complexity. The line source model adequately reproduces the spacial variation of the damage as confirmed by the RMS azimuthal distribution shown in the rose diagram. The error is always under 11, and it is strongly reduced in correspondence of rupture direction. disadvantage to be a subjective and discrete scale, which cannot be associated to reliable attenuation relationships. However, to better correlate the structural damage with ground-motion quantities it is necessary to have, for each damaged site, both a ground motion record and a damage survey, which is not the case in Italy for the last destructives earthquakes (Friuli, 1976; Irpinia, 1980; Abruzzo, 1984; Umbria Marche, 1997). Alternatively, Sabetta et al. [9] correlated the peak ground acceleration derived from attenuation law and the structural damage based on the surveys performed after the Irpinia 1980 and Abruzzo 1984 earthquakes. They obtained empirical curves which are derived from the regression of a mean damage index (p) [10,9] as function of PGA estimated from attenuation relationships. According to the MSK macroseismic scale, Braga et al. [40] simplified the available 13 constructive typologies and eight levels of damage of the original Irpinia data-set, in six damage levels and three vulnerability classes of building A, B and C with decreasing vulnerability [41]. Following the Sabetta et al. [9] approach, we evaluate the mean normalised damage (p), for each structural class defined as p ¼ 1 n X N i¼1 d i N i N where n is the number of damage levels; N i is the number of buildings with damage d i per municipality and vulnerability class; N is the total number of buildings per municipality and vulnerability class. In this section we use the PGA simulated by the line source model to investigate the correlation between damage and synthetic strong motion parameters. In Fig. 9 we show the regression results for vulnerability classes A and B, which are predominant in the damage data-base. In column A the observed damage is correlated with the PGA inferred from strong motion data, the regression line has been obtained using a third order polynomial regression, as indicated by Sabetta et al. [9]. In the same figure, the observed damage is correlated with the PGA evaluated using the Sabetta and Pugliese [37] attenuation law, the polynomial function is the one recovered by Sabetta et al. [9]. In each panel we report the root mean square sum of residuals (RMS). Most of the points on Fig. 9b are clustered around a PGA value of 0.1 g. This is a consequence of the use of an attenuation relationship, where the estimated PGA only depends on the epicentral distance. Nevertheless the corresponding observed mean normalized damage shows a great variability (0.1 0.5), which suggests that there is a complexity in the damage
M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 1217 PGA inferred from Strong Motion Simulation PGA inferred from Attenuation Law Mean Normalized Damage 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.5 Class A buildings RMS=0.05 Mean Normalized Damage 0.1 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 PGA (g) PGA (g) 0.5 0.7 0.6 0.5 0.4 0.3 0.2 Class A buildings RMS=0.18 Class B buildings Class B buildings Mean Normalized Damage 0.4 0.3 0.2 0.1 Mean Normalized Damage 0.4 0.3 0.2 0.1 RMS=0.09 RMS=0.14 0 0 0.1 0.2 0.3 0.4 PGA (g) 0 0 0.1 0.2 0.3 0.4 PGA (g) Fig. 9. Regression results for the predominant vulnerability classes A and B. In column A the observed damage is correlated with the PGA inferred from strong motion data, the regression line has been obtained using a third order polynomial regression, as indicated by Sabetta et al. (1998) [9]. Column B the observed damage is correlated with the PGA evaluated using the Sabetta and Pugliese (1996) [37] attenuation law, the polynomial function is the one recovered by Sabetta et al. (1998) [9]. In each panel we report the root mean square sum of residuals (RMS). pattern which cannot be explained only in terms of epicentral distance. When we correlate the observed mean normalized damage with the PGA estimated from our strong motion modelling, this clustering effect disappears. Such an observation suggests that the PGA inferred from a ground motion simulation which includes the fault length finites, source duration, directivity of fracture and focal mechanism better correlates with damage respect the PGA evaluated only in function of epicentral distance. Of course, to give a statistical value to these observations is necessary to extend such procedure to greater damage database. 7. Discussion and conclusions In this work we have proposed a simplified approach for rapid simulation of ground motion shaking and damage scenario based on non-complex source-propagation models. In order to include the rupture directivity effect on the ground motion level, we model the seismic source using an Haskell-type source model (line source, with uniform slip amplitude). The complete wave-field Green s functions helped to better estimate the peak ground motion quantities (PGA, PGV and PGD) even in cases where they are brought by secondary body wave arrivals or surface waves. The method has been validated on and applied to the data recorded during the 1980 Irpinia earthquake in southern Italy. There are several reasons that determine the choose of this event: it is the maximum instrumental earthquake for southern Italy and it is often used as reference earthquake for the seismic risk analysis in the Campania area. Moreover it represents a valuable and comprehensive case study due to the contemporary availability of high quality ground motion, intensity and damage observations. Due to the earthquake complexity we adopted a composite line source model, according to geometries and mechanisms given by Bernard and Zollo [12].
1218 M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 Green s functions for a flat-layered velocity model have been computed using the discrete wave-number method of Bouchon and Aki [19]. The adopted source model for the Irpinia has been validated through the matching between observed and synthetic peak, waveforms and spectral shapes. Our methodology gives consistent estimate of peak ground quantities for acceleration, velocity and displacement time series in the 0.1 8 Hz frequency range. The mean error on the PGA evaluation is around a factor 2, while the peak trends are consistent with several attenuation relationships, in particular with Sabetta and Pugliese [37] (inferred on the Italian event) and with Boore [36] laws. As a general remark we note that such a combination of source/ propagation model provide rather complex waveforms well reproducing the areal distribution of most of the strong motion parameters. In some near source station (as Sturno, Bagnoli, Auletta, Mercato San Severino) the synthetic displacement waveforms reproduce the shape of the observed signal. We further use this methodology to evaluate shake maps and an a posteriori earthquake scenario for the Irpinia region. The evaluated shake maps show a dominant source effect since the adopted composite line source model is the main cause for spatial variation of strong motion parameters. The source effects are also evident in the structural damage distribution, and we have demonstrated how the damage prediction can be improved using massive strong-motion simulations based on well-constrained rupture models with respect to ground motion quantities obtained using classical attenuation relationships. The main advantage of the presented method is that it allows for computing a preliminary set of synthetic seismograms over a vast area, using few parameters to describe the event source. Many methods developed for the rapid determination of source parameters use information on event magnitude and location few seconds after its occurrence; while focal mechanism, source length and slip distribution can be obtained few minutes later [2,6]. Such a methodology can be integrated in Yamada and Heaton [6] procedure, to fast estimate broad-band frequency accelerograms, while in case of sparse network, the fault length, causative plane and velocity rapture can be estimated following the approach by Dreger and Kaverina [2]. A limit of the described approach is that synthetic records do not include the site response. As a first order approximation, the site response effect can be accounted by a convolution of the synthetic signal with the a priori determined site response curve as proposed in Zollo et al. [42]. The theoretical estimates of strong ground motion quantities obtained by using a finite length source model can be therefore usefully integrated with observed measurements at an existing, dense accelerometric array to rapidly produce ground shaking maps soon after a destructive event. Moreover we believe that the use of ground motion simulations for shake map evaluation can be useful when the event is located outside the monitoring network and few or no near source records are available. We have shown that the synthetic seismograms obtained from a relatively simple source model can be used to create reliable maps for all the strong motion engineering oriented parameter (PGA, PSV, Arias intensity, etc.) and to obtain an early, approximate damage distribution over an interest area or in some strategic sites (hospitals, factories or critical facilities), which is a relevant information for emergency actions and rescue operations to be taken in the first hours after a destructive earthquake. References [1] Wald et al. ShakeMap manual, 2006. [2] Dreger D, Kaverina A. Seismic remote sensing for earthquake source process and near-source strong shaking: a case study for October 16, 1999 hector mine earthquake. Geophys Res Lett 2000. [3] Hartzell S, Heaton T. Inversion of strong ground motion and teleseismic waveform data for the fault rupture history of the 1979 Imperial Valley California. Bull Seism Soc Am 1983;73(6):1553 83. [4] Wells DL, Coppersmith KJ. New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement. Bull Seism Soc Am 1994;84(4):974 1002. [5] Cohee BP, Beroza GC. A comparison of two methods for finite-fault inversion using strong-motion data. Ann Geofis 1994;37(6):77 101. [6] Yamada M, Heaton T. Real-time Estimation of Fault Rupture Extent using Envelopes of Acceleration 2007. [7] Murphy JR, O Brien LJ. The correlation of peak ground acceleration amplitude with seismic intensity and other physical parameters. Bull Seism Soc Am 1997;67(3):877 915. [8] Atkinson GM, Sonley E. Empirical relationships between modified Mercalli intensity and response spectra. Bull Seism Soc Am 2000;90(2):537 44. [9] Sabetta F, Goretti A, Lucantoni A. Empirical fragility curves from damage survey and estimated strong ground motion. In: 11th European Conference on Earthquake Engineering, 1998. [10] Orsini G. A model for buildings vulnerability assessment using the parameterless scale of seismic intensity (PSI). Earthquake Spectra 1999;15(3): 463 83. [11] Rossetto T, Elnashai A. Derivation of vulnerability functions for European type RC structures based on observational data. Eng Struct 2003;25:1241 63. [12] Bernard P, Zollo A. The Irpinia (Italy) 1980 earthquake: detailed analysis of a complex normal faulting. J Geophys Res 1989;94(B2):1631 47. [13] Westaway R, Jackson J. Surface faulting in the southern Italian Campania Basilicata earthquake of 23 November 1980. Nature 1984(312):436 8. [14] Westaway R, Jackson J. The earthquake of the 1980 November 23 in Campania Basilicata (southern Italy). Geophys J R Astron Soc 1987;90:375 443. [15] Pingue F, De Natale G, Briole P. Modeling of the 1980 Irpinia earthquake source: constraints from geodetic data. Ann Geofis 1993;XXXVI(1):27 40. [16] Deschamps A, King GCP. Aftershocks of the Campania-Lucania (Italy) earthquake of 23 November 1980. Bull Seismol Soc Am 1984;74:2483 517. [17] Pantosti D, Valensise G. Source geometry and long-term behavior of the 1980, Irpinia earthquake fault based on field geologic observations. Ann Geofis 1993;XXXVI(1):41 9. [18] Cocco M, Pacor F. Space-time evolution of the rupture process from the inversion of strong motion waveforms. Ann Geofis 1993;XXXVI(1):41 9. [19] Bouchon M, Aki K. Discrete wavenumber representation of seismic source wave filed. Bull Seismol Soc Am 1977;67:259 77. [20] Coutant O. Program de simulation numerique AXITRA. In Rapport LGIT. Universitè Joseph Fourier, Grenoble, 1989. [21] Bouchon M, Hatzfeld D, Jackson JA, Haghshenas E. Some insight on why Bam (Iran) was destroyed by an earthquake of relatively moderate size. Geophys Res Lett 2006;33:L09309. [22] Favreau P, Archuleta RJ. Direct seismic energy modeling and application to the 1979 Imperial Valley earthquake. Geophys Res Lett 2003;30(5):1198. [23] Herrero A, Bernard P. A kinematic self-similar rupture process for earthquakes. Bull Seismol Soc Am 1994;84(4):1216 28. [24] Irikura K, Kamae K. Estimation of strong ground motion in broad-frequency band based on a seismic source scaling model and empirical Green s function technique. Ann Geofis 1994;XXXVII:1721 43. [25] Gallovic F, Brokesova J. Hybrid k-squared source model for strong ground motion simulations: Introduction Phys. Earth Planet Int 2007;160:34 50. [26] Boore DM. Stochastic simulation of high-frequency ground motion based on seismological models of the radiated spectra. Bull Seismol Soc Am 1983;73(6):1865 94. [27] Haskell NA. Total energy and energy spectral density of elastic wave radiation from propagating faults. Bull Seismol Soc Am 1964;54(6A):1811 41. [28] Bouchon M. A review of the discrete wavenumber method. Pure Appl Geophys 2003;160:445 65. [29] Hisada Y. An efficient method for computing Green s functions for a layered half-space with sources and receivers at close depths. Bull Seismol Soc Am 1994;84(5):1456 72. [30] Hisada Y. An efficient method for computing Green s functions for a layered half-space with sources and receivers at close depths: part II. Bull Seismol Soc Am 1995;85(4):1080 93. [31] Improta L, Bonagura M, Capuano P, Iannaccone G. An integrated geophysical investigation of the upper crust in the epicentral area of the 1980, Ms ¼ 6.9 Irpinia earthquake (Southern Italy). Tectonophysics 2003;361:139 69. [32] Malagnini L, Hermann RB, Di Bona M. Ground-motion scaling in the Apennines (Italy). Bull Seismol Soc Am 2000;90(4):1062 81. [33] Sabetta F, Pugliese A. Attenuation of peak horizontal acceleration and velocity from italian strong-motion records. Bull Seismol Soc Am 1987. [34] Abrahamson NA, Silva WJ. Empirical response spectral attenuation relations for shallow crustal earthquakes. Seismol Res Lett 1997;68(1):94 127. [35] Campbell KW. Empirical near-source attenuation relationship for horizontal and vertical components of peak ground acceleration, peak ground velocity, and pseudo-absolute acceleration response spectra. Seismol Res Lett 1997.
M. Lancieri, A. Zollo / Soil Dynamics and Earthquake Engineering 29 (2009) 1208 1219 1219 [36] Boore DM, Joyner WB, Fumal TE. Equations for estimating horizontal response spectra and peak acceleration from Western North American earthquakes: a summary of recent work. Seismol Res Lett 1997;68(1):128 53. [37] Sabetta F, Pugliese A. Estimation 414 of response spectra and simulation of nonstationary earthquake ground motion. Bull Seismol Soc Am 1996. [38] Faccioli E, Cauzzi C. Macroseismic intensities for seismic scenarios, estimated from instrumentally based correlations. First European Conference on Earthquake Engineering and Seismology, Geneva, Switzerland, 3 8 September 2006, Paper no. 569. [39] Bramerini F, Di Pasquale G, Orsini G, Pugliese A, Romeo R, & Sabetta F. 1995. Rischio sismico del territorio italiano: proposta di una metodologia e risultati preliminari. In: Proceedings of the 7 Convegno Nazionale L Ingegneria Sismica in Italia, Siena September 1995, Vol. 3, p. 1099 108. [40] Braga F, Dolce M, Liberatore D. A statistical study on damaged buildings and an ensuing review of the msk-76 scale. In: Proceedings of the Seventh ECEE Athens, 1982. [41] Liberatore A. Statistical models of damage to buildings and the MSK scale. In: Proceedings of the Tenth World Conference on Earthquake Engineering, 1992. [42] Zollo A, Emolo A, Herrero A, Improta L. High frequency strong ground motion modelling in the Catania area associated with the ibleo-maltese fault system. J Seism 1999;3(3).