Basin-effects observed during the 2012 Emilia earthquake sequence in Northern Italy

Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

Contents lists available at ScienceDirect

Soil Dynamics and Earthquake Engineering journal homepage: www.elsevier.com/locate/soildyn

Basin-effects observed during the 2012 Emilia earthquake sequence in Northern Italy Jetson Ronald Abraham a,n,1, Carlo G. Lai b, Apostolos Papageorgiou c a

Aon Benfield, 4th Floor, Xchanging Tower, Whitefield, Bangalore 560066, India Department of Civil Engineering and Architecture, University of Pavia and European Centre for Training and Research in Earthquake Engineering (EUCENTRE), Via Ferrata 1, Pavia 27100 Italy c Department of Civil Engineering, University of Patras, 26500 Patras, Greece b

art ic l e i nf o

a b s t r a c t

Article history: Received 9 February 2015 Received in revised form 30 June 2015 Accepted 19 August 2015

During the 2012 Emilia earthquake sequence, prolonged shaking associated with long-period motions were observed in the Po Plain basin in Northern Italy. Such anomalous characteristics of ground-motion were unnoticed at the rocky sites outside the Po Plain basin. To explain these phenomena, a series of detailed analyses were carried out using the strong motion records from May 20 to May 29 main events. The observed amplification, calculated using the spectral ratio method indicates a fundamental resonance period at 5 s. Well-dispersed surface Rayleigh waves of periods between 3 and 10 s were noticed and the contribution of surface waves to the total motion was significant despite the source being located beneath the Plain. The late arriving long-period surface waves significantly increased the duration of ground shaking. The envelope delay spectrum shows that the duration lengthening of ground motion could be well correlated with the dispersion of surface waves. The greatest lengthening of the records was observed around the fundamental period of the basin. Large peak ground-motions were observed in the near-field region, especially in the vertical component which is attributed to source effects (predominantly vertical movement of the causative fault) while the prolonged duration of motions seems caused by surface waves. & 2015 Elsevier Ltd. All rights reserved.

Keywords: Emilia earthquake Po Plain Basin-effects Surface waves

1. Introduction On May 20, 2012, Northern Italy was struck by a MW 5.9 (ML 5.9) earthquake which severely damaged the downtown area of Emilia region. The mainshock was followed by several aftershocks and among them, the MW 5.65 (ML 5.8) event which occurred on May 29, 2012 was relevant because of the large damage it caused2 [24]. The Emilia earthquake sequence was monitored by the RAN (Rete Accelerometrica Nazionale, http://www.protezionecivile.gov.it/ jcms/it/ran.wp) Italian strong motion network managed by the Italian Department of Civil Protection and RAIS (Rete Accelerometrica Italia Settentrionale, http://rais.mi.ingv.it/) seismic network managed by INGV (Istituto Nazionale di Geofisica e Vulcanologia), the n

Corresponding author. E-mail addresses: [email protected] (J. Ronald Abraham), [email protected] (C.G. Lai), [email protected] (A. Papageorgiou). 1 Present address: Karisalpatti, Tirunelveli (Dist), Tamil Nadu 627414, India. 2 The seismotectonic structure responsible for the May 29 event is different from the one that generated the May 20 earthquake. Thus it is still questioned whether the May 29 event may be technically considered an “aftershock” of the May 20 sequence [20]. http://dx.doi.org/10.1016/j.soildyn.2015.08.007 0267-7261/& 2015 Elsevier Ltd. All rights reserved.

Italian equivalent of USGS. A visual inspection of the records displayed well developed, long-period waveforms with prolonged durations which seem to suggest that the basin structure of the Po Plain has significantly influenced the ground shaking. This phenomenon is known in the literature as basin-effects which are due to an unfavorable combination of a particular geological configuration and the direction of wave propagation causing focalization of wave energy and generation of surface waves. It is a geometric process caused by interference of seismic waves due to geological irregularities and mechanical impedance contrasts. There are several well documented examples of basin-effects observed from past earthquakes from the September 19, 1985 Mexico City earthquake to the recent September 4, 2010 Darfield and February 22, 2011 Christchurch earthquakes in New Zealand [10,16,4]. Understanding how seismically-induced ground-motion is modified by complex basin geometry has been a fertile area of research for quite a long time. Basin related ground-motion can be studied from a seismological as well as engineering perspective. Seismologists are interested in basin-effects from the point of view of the conditions and physical mechanisms causing them, whereas earthquake engineers are more concerned with the characteristics of basin-related

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

ground-motion that are detrimental to structures and engineering facilities. Both these perspectives have played a vital role in understanding the importance of basin-effects. The consequences of Emilia earthquake resulted in 27 fatalities, hundreds were injured and approximately 16,000 were left homeless from the two events of May 20 and May 29 [19]. This study concentrates on assessing basin-effects observed over the Po Plain during the May 20, 2012 mainshock as well as during the May 29 aftershock(1), and it covers the following aspects: (1) explaining the typical ground-motion features observed in the Po Plain, (2) evaluating the amplification observed in the Po Plain using the spectral ratio method, (3) identification of surface waves by analyzing the basin records, and (4) quantifying the duration lengthening due to surface waves, using the envelope delay spectrum method.

2. Seismicity of the region According to the latest issue of the Italian seismic catalog [23], the area mostly affected by May–June 2012 Emilia sequence has centuries-old seismic history of comparatively low magnitude events. This region is on the outer front of the Northern Apennines which is characterized by low to medium seismic hazard. This can be understood from Fig. 1 which shows earthquakes with magnitude above 4.5 which occurred in the Emilia-Romagna region over the latest 200 years. The data were retrieved from the Italian earthquake catalog CPTI11 (http://emidius.mi.ingv.it/CPTI11/). Fig. 1 contains some important historical as well as instrumental earthquakes in the region. It is evident from the past seismicity that the eastern section of the Southern Alps is characterized by a relatively large frequency of earthquake occurrence as well as energy released per event. This defines a seismic belt at the foothills of the chain, where several active faults are located [2]. Most of the earthquakes in this region are concentrated South of the Po River along portions of the pede-Apennines thrust front and on some buried Apennines outer fronts [7]. Earthquakes are more infrequent north of the Po River and West of Milan [11]. The most destructive historical event in this area is the November 17, 1570 earthquake, which struck the town of Ferrara, and the March 17, 1574 event that yielded damage in Finale Emilia [23]. Both events

231

have caused widespread phenomena of soil liquefaction [17]. The other important historical events within or close to the Emilia region occurred in 1624 and 1796. Other historical earthquakes with magnitudes up to 6 have occurred in the Southern part of the Emilia-Romagna Region, close to the Apennine chain (e.g. the September 10, 1781 earthquake). Furthermore, a M5.7 event occurred on May, 12, 1802, located West of Garda Lake in the central northern part of the Po Plain, which caused serious damage in Northern Italy, in the provinces of Brescia, Bergamo, Lodi, and Cremona [2]. The most recent large earthquakes of the region occurred in 1971 and 1983 near Parma. They were characterized by MS 5.7 and MS 5.0, respectively. On October 15, 1996 a MS 5.1 earthquake occurred near the town of Reggio Emilia on the southern edge of the Po Plain which caused moderate damage in unreinforced masonry structures in Reggio Emilia and other small towns in the Po Plain [27]. The DISS (Database of Italian Seismogenic Sources) [5] lists many seismic sources within the Po Plain basin as well as in the foothills of the SouthernAlps.

3. Seismotectonic setting of Northern Italy The Italian seismicity is mainly due to the movement of the African plate in the North direction, and the consequent (continent to continent) collision with the Euro-Asiatic plate. In the NorthEast, the Adriatic (micro) plate is a geologic structure playing a key role in the observed seismicity of that part of the Italian Peninsula. This is the remains of a large African promontory that in ancient times occupied most of the actual Central-Western Mediterranean basin [21]. In its Northern part, the Adriatic plate collides with the Euro-Asiatic margin along the oriental Alps, giving rise to systems of inverse faults leading to moderate to high seismicity. Some of the most significant earthquakes in this zone (e.g. Friuli earthquake, 1976–77) are due to the subduction of the Adriatic plate beneath the oriental Alps. The Central area of Northern Italy is characterized by the presence of the Po Plain which is a wide flexural basin covered by thick Quaternary sediments. It stretches East–West (EW) across Northern Italy for more than 40,000 km2, the widest part being the alluvial basin of the peninsula. The Plain is locally up to 100 km wide, and is drained axially by the 652 km long Po River, the longest in Italy [11]. Fig. 2 shows a topographical map of

Fig. 1. Seismicity map of Northern Italy containing the following data: (1) distribution of epicenters with MW 44.5 over the last 200 years (http://emidius.mi.ingv.it/CPTI11/), (2) major historic events (diamonds), (3) recent major events (triangles), and (4) the 2012 Emilia sequence (stars).

232

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

Fig. 2. Topographic map of the Po Plain in Northern Italy. The figure also shows the epicenters of the mainshock and the aftershock(1) along with strong motion recording stations located in the Plain. Table 1 Main source parameters of the 2012 Emilia mainshock and its aftershock [24]. Event

Epicenter Origin time Lon (UTC: dd/ Lat mm/yy, h:m:s)

Mainshock 20/05/ 2012, 02:03:53 Aftershock 29/05/ 2012, 07:00:03

Depth (km)

44.89 11.23

6

44.85 11.09

10

ML

MW

Strike Dip Rake

5.90 5.90 282

51

93

5.80 5.65

64

89

274

the Po Plain area. It is delimited at the South and North by the topographic highs of the Apennines and the Alps respectively, and to the East by the Adriatic coastline. The entire Northern Apennines and the Po Plain belong to a chain-foredeep-foreland system developed from the collision between the European and African continents [12]. It is a low-relief area characterized by active crustal shortening due to the convergence between the Adria micro plate and the European plate which is accommodated by blind thrust faulting. It is an EW trending basin which slightly dips towards East, filled with Quaternary alluvium. Sediments are several kilometers deep in the eastern part of the basin, toward the Adriatic Sea which reaches a thickness of 2–8.5 km of Tertiary and Quaternary sediments [11]. Hence, such complex geological features are bound to influence the ground-motion characteristics of earthquakes occurring in this area. 3.1. May–June, 2012 earthquake sequence An earthquake located beneath the Po Plain basin struck the Emilia region on May 20, 2012. According to the Database of Individual Seismogenic Sources [14] the event was originated by the composite source of Novi-Poggio Renatico constituted by the individual structures of Mirandola, Canalazzo di Finale Emilia and Concordia [26]. The Emilia sequence of May 2012 reactivated the basal thrust in the central portion of the Ferrara-Romagna arc, which resulted in two important events of ML 5.9 and ML 5.8 that

occurred on May 20 and 29 respectively. Both were located close to the buried front of Ferrara northward-verging active thrust belt (Bonini et al., 2014). The mainshock (May 20) occurred in between the small towns of San Felice sul Panaro and Finale Emilia, involving the central part of the Ferrara arc whereas the aftershock(1) (May 29) was located South Westward with respect to the mainshock, close to the San Felice sul Panaro town. The related aftershocks cover the Western-Central part of the thrust front. The seismic sequence covered a large area extending in the EW direction between the towns of Mirandola and Ferrara. The seismicity spread along a more than 50 km long fault system, the area being shaken by 68 earthquakes with ML 43, including 13 shocks with ML Z4.0. The available fault plane solutions of that area from the previous earthquakes show a prevalent reverse and reverse oblique mechanism. The authors in Ref. [24] inferred the fault plane solutions of the Emilia sequence earthquakes and the results are consistent with the past inferences. The estimated source mechanisms are reported in Table 1. Both events have reverse slip mechanism with predominant vertical movements as well as higher dip. The strikes of the events are along the EW direction. 3.2. Recorded strong motion data The strong motions of Emilia earthquake and its aftershocks were recorded by more than 100 stations located at epicentral distances ranging from 15 km up to more than 800 km. This earthquake generated not only a large number of recorded motions in the Po Plain, but also complicated wave patterns at sites inside the Po Plain basin. The locations of the seismic stations are shown in Fig. 2. Records analyzed in this article were provided by the Italian Department of Civil Protection. All stations recorded 3-component seismograms. For few records when the acceleration trace was integrated twice to obtain displacement time history, the resulting trace drifted considerably. To eliminate these drifts, the data were processed using simple baseline corrections, and lowcut filtering with a second-order acausal 0.05 Hz ‘Butterworth filter’. Furthermore, the analyses demanded several filtering and integration operations, which introduced undesired long-period motions, as discussed in [8]. To avoid these long-period errors, zero pads of adequate length were appended at the beginning and

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

at the end of each time series [8]. Furthermore, the short-period waves being associated with short-wavelength carry information on shallow sediments whereas long-period waves (above 1 s) represent oscillations that are influenced principally by large-scale averages of the seismic velocity structure. Therefore, periods longer than 1 s are not expected to be influenced by the velocity profile of top 30 m because of their longer wavelength. Hence, in this study the local site characteristics, as these are characterized by VS30 (average shear-wave velocity of the upper 30 m) have been ignored.

4. Waveform analysis Observed ground-motions in the Po Plain displayed several distinct features. Stations located closer to the source, recorded signals with strong near-fault pulses, while the far-field records are marked with long-period surface waves with extended durations. In order to view the energy distribution of ground-motion in space and time the vertical component velocity time histories of the mainshock are superimposed on the location of the station on the map (Fig. 3). The Eastward dipping (i.e. depth increasing towards East) basin shows a correlation of amplitude with the depth. Larger amplitude with longer duration was observed in the Southeastern part of the Plain where the basin is considered to be deeper. Furthermore, the propagation of long-period surface waves is apparent, suggesting that the long-period motion grows in amplitude and duration towards the Eastern part of the basin. The closest station to the epicenter (14 km) of the mainshock is Mirandola (MRN). Due to the close proximity to the fault, the recorded ground-motions in this station differ significantly from observations at greater distances from the fault. Fig. 4 shows acceleration and velocity time histories of the three components with their corresponding pseudo acceleration response spectra (PSA) as well as pseudo velocity response spectra (PSV). The record has a very large vertical acceleration as well as large horizontal velocity pulses. The maximum peak ground acceleration (PGA) resulted from the vertical component, while the maximum PGV was

233

obtained from the horizontal components. Besides, both PSA and PSV spectra for the vertical component indicate a larger response at high frequencies, whereas horizontal components have a larger response in the low-frequency range. It has been observed in the past that the peak ground-motion amplitude of the vertical-tohorizontal (V/H) ratio may significantly exceed unity in the nearfault region in the high-frequency range (Bradley, 2011). Hence, the larger PGA observed in the vertical components could be attributed to source effects, and could be well justified from the source mechanism, which had a predominantly reverse fault movement. Also it is interesting to highlight that the North–South (NS) component of ground-motion resulted in a large velocity pulse which caused a large spectral response. The fault normal component is observed to have stronger near fault effects. Since the NS component is almost in the fault normal direction, larger pulses were observed compared to the EW component. Three stations namely ZPP (41 km), CSP (63 km) and ALF (76 km) located at different epicentral distances in the Po Plain were selected for the analyses. These records have a significant content of longperiod waves. The recorded acceleration and the corresponding velocity time histories of these stations are illustrated in Fig. 5 together with a highlight of their peak values. It is interesting to observe that the PGA occurs in the early part of ground-motion record (body wave) while PGV is vividly seen to occur during the latter part of the signal (very likely surface waves). Furthermore, the time of peak occurrence shifts significantly to the right with the distance. This is due to dispersion. Therefore, the influence of the basin resulting in generation of surface waves (which generally appear more evident in velocity records rather than in the corresponding accelerograms) is consistent with the dispersion characteristics of the waveforms. It is easy to visualize the surface wave arrival at about 25 s at ZPP station. The difference in arrival time is apparent as the waveforms propagate from the ZPP station towards the ALF station. Strong surface waves are observed beyond 50 s at the CSP station while at the ALF station strong surface waves manifest themselves beyond 60 s. The waveform becomes more complex as the wave travels from ZPP to ALF station. ZPP station has a single pack of waves with a single peak arrival time around 40 s, while at

Fig. 3. Map showing the velocity time series of the vertical component observed in the Po Plain during the mainshock. The epicenter is indicated with a yellow star. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

234

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

Fig. 4. Example of a near-fault ground-motion recorded in the Po Plain during the May 20, 2012 Emilia earthquake in Northern Italy. Top: acceleration and velocity time histories of the MRN station from the mainshock. Bottom: the corresponding PSA and PSV response spectra.

Fig. 5. Acceleration (left panel) and velocity (right panel) time histories recorded at different station in the Po Plain basin (vertical components) are plotted along with the time of peak occurrence (circle). The epicentral distance of the station is given in parentheses. These records belong to the mainshock.

the ALF station, several wave packs arrive at different intervals, hence the motion becomes more complicated. The peak arrival time seems to suggest that the long-period vertical component of ground-motion (which could be associated with a Rayleigh wave) travel at about 1 km/s which is verified in the sequel along with a more rigorous treatment of surface waves. 5. Amplification The amplification caused by the presence of unconsolidated soil deposits has long been recognized as a subject of paramount

importance in explaining the intensity of observed ground-motion. When a body wave (particularly an S wave) enters a slower velocity material, such as a loose sediment layer, seismic amplitudes will generally increase due to the mechanical impedance contrast. Amplifying effects of sediments can be approximately evaluated based on the weighted average of the shear wave velocity of the top 30 m (VS30). This approach is typically used for instance, in codebased design of structures. However 1D lithostratigraphic amplification alone is unable to explain the intensity of ground-motion that sometimes is measured in large sedimentary basins. These basins

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

235

have thicknesses ranging from hundred meters to few kilometers and they may have a strong influence in amplifying the intensity of ground-motion, especially at periods larger than about 1 s. To measure the observed amplification, the conventional spectral ratio method is followed. This technique consists of dividing the horizontal component of the Fourier amplitude spectrum of the seismogram recorded at the basin site by the corresponding component of a seismogram recorded, for the same earthquake, at a reference hard rock site. The Fourier amplitude spectrum at a basin site is denoted by B (f ), which can be factorized as B (f ) = S (f )· Pi (f )· Gi (f ), where S (f ) , Pi (f ) and Gi (f ) are the source, path and site factors, respectively. The term R (f ) denotes the corresponding Fourier spectrum of ground-motion recorded at a hard rock site of the same earthquake. If ground amplification is assumed negligible at the rock station, then R (f ) = S (f )· Pref (f ) where Pref (f ) is the path factor associated with the reference rock site. Then the spectral ratio can be expressed as:

Ai (f ) =

S (f ) Pi (f ) Gi (f ) S (f ) Pref (f )

(1)

If the distance between sites where the ground-motions B (f ) and R (f ) are recorded, is small compared with the epicentral distance, then the path's factors, Pi (f ) and Pref (f ) can be assumed to be approximately the same. Thus the spectral ratio given by Eq. (1) provides an estimate of site amplification G i (f ) which depends on the geological characteristics of the site. It may possibly include basin effects. In the Po Plain case, the reference stations are located outside the Plain and the distance between the reference and the basin stations is more than 30 km. Hence, the spectral amplitudes were corrected for the geometric spreading factor 1/R (R is the hypocentral distance) and the spectral ratio is then given by:

Ai (f ) =

Ri S (f ) P (f ) Gi (f ) Rref S (f ) P (f )

(2)

Since the amplification of interest is below 1 Hz and the frequency-dependent material attenuation due to soil inelasticity is expected to be mild at low frequencies, the latter is ignored. Prior to Fourier analysis, horizontal earthquake recordings were cosinetapered i.e. time-windowed from the onset of S waves. The window lengths covered later-arriving long-period waves. The spectral amplitude was determined from the root mean square of the Fourier spectral amplitudes of two horizontal components. The estimated spectral ratio for the mainshock is shown in Fig. 6. A large amplification is noted between 0.15 and 0.4 Hz, and a consistent peak around 0.2 Hz seems to suggest that the fundamental period of the basin is around 5 s. A similar trend was also observed for the event of May 29. Furthermore, the amplification beyond 1 Hz approaches unity. From Fig. 6 the maximum spectral amplification was observed at MDC (epicentral distance, 56 km) and CSP (epicentral distance, 64 km) stations, despite their large epicentral distances. Surface waves generated at different locations overlap along their propagation path leading to the superposition of multi-pathing wavetrains, which might cause large amplification even at relatively large epicentral distances. The mean spectral ratio in Fig. 6 shows a peak whose amplitude is greater than 10. Furthermore, it is remarked that the geometric spreading factor 1/R is strictly not valid for surface waves. Since the observed large spectral values in the low-frequency range are most likely due to the contribution of surface waves (which have different geometric spreading characteristics than body waves), the amplitudes of the spectral ratios shown in Fig. 6 need to be carefully interpreted.

Fig. 6. Spectral ratios obtained at different stations overlapped to the mean estimate (thick line) for the mainshock of May 20, 2012.

6. Surface waves It is well known that long-period ground-motions observed at the free-surface of sedimentary basins have a significant contribution from surface waves which are generated due to the lateral heterogeneities of the medium (multidimensional effect). According to Brad et al. [4], the multidimensional effects become important in two cases: when there exists a steep velocity gradient or when the thickness of the sediments is considerable (say  1 km). The Po Plain meets both of these criteria. Furthermore, surface waves are clearly seen in the records; hence, the observed motion should be carefully examined for their influence. A typical record from a basin would exhibit a strong time evolving frequency composition due to the dispersive nature of surface waves, and the time varying amplitude occurs largely due to the contribution of body waves [9]. In order to illustrate the groundmotion characteristics of such records (for instance amplitude, frequency content and duration) highlighting their nonstationarity, neither the time domain nor the Fourier domain representations of a signal would be appropriate. A time-frequency representation, on the contrary, would be best suitable. In Fig. 7, a displacement time history recorded at the MDN station is displayed and the arrival of long-period surface waves is fairly visible between 30 and 65 s. The same record is also represented in a time-frequency domain using the continuous wavelet transform adopting the Daubechies of order 4 as mother wavelet [18]. The dispersion is well illustrated in the wavelet spectrum. Waves with periods between 2 and 10 s arrive in the time frame between 30 and 65 s and this is a strong zone of dispersion in period and time domains respectively. Geometric dispersion (i.e. the dependence of the speed of propagation upon frequency) is an intrinsic characteristic of surface waves propagating in layered media, thus the plot indicates that there is a significant amount of surface waves in the records from the Po Plain. Therefore, it is worthwhile to explore the characteristics of surface waves in more detail and such exercises are carried out in the following sections. 6.1. Group velocity dispersion curves Surface waves traveling in layered media are dispersive in nature which implies that the speed of propagation is frequencydependent. Dispersion curves are usually adopted to represent this

236

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

property. The displacement time series of three components of the ZPP station records are shown in Fig. 8. Phases at periods of 2, 4, 6, 8, and 10 s were extracted using a ‘Gaussian filter’. It is easy to visualize that the envelope peaks of different phases arrive at different times. Long-period wave groups arrive before shorter periods as expected for these types of dispersive signals. Furthermore, the extracted phase at 2 s is composed of coupled wavetrains, which could have been generated at different locations. The wavetrains are not easily identifiable because the longer-period phases propagate faster so that the travel time among different wavetrains are smaller and harder to be seen in the record. These filtered wave packs can be used to estimate the dispersion curve following a procedure that is briefly described in the following.

The dispersion curve provides an idea about the frequency content of surface waves and their velocity of propagation. Dziewonski et al. [15] has demonstrated that the Multiple Filtering Technique (MFT) is a fast and efficient method to analyze dispersed signals. It uses an array of narrow-band Gaussian filters to isolate phases and amplitudes to measure dispersion parameters. A very complex signal can be effectively dealt using this method. The method performs well even when several dominant periods arrive almost simultaneously. This technique has been used widely in seismology to determine the shear-wave velocity distribution of the earth's crust from the inversion of surface Love and Rayleigh waves. In this study, a computer program based on MFT developed by Corchete [13] was used to estimate the group velocity as a function of frequency. The program applies a Gaussian filter with peak amplitude centered on the desired period to the displacement traces in the frequency domain. The peak of the envelope of the corresponding time domain signal is considered to be the arrival time of the corresponding phase, then the group velocity, Distance UG (f ) is estimated from the relation UG (f ) = T (f ) . Distance indiG

Fig. 7. Displacement trace of the MDN station record (vertical component) and its time-frequency representation via wavelet transform.

cates the distance between the receiver and the surface wave origin, and TG (f ) is the arrival time. Since the source is located beneath the basin, surface waves were more likely to be originated upon entering the basin from below. Hence, the hypocentral distance was used for the estimation of UG (f ). The estimated group velocities of the fundamental mode Rayleigh and Love waves using the MFT technique are displayed in Fig. 9. We used records from the same stations which were used to estimate amplification in the previous section (refer Fig. 6 for stations used). For Rayleigh waves, the group velocities varies roughly from 0.70 km/s to 2 km/s for periods from 3 to 10 s, while for Love wave between 1 km/s and 1.5 km/s for periods from 3 to 10 s. A relatively larger dispersion is noted for Rayleigh waves which may be attributed to the type of faulting mechanism (reverse faulting) which strongly favored the generation of Rayleigh waves as opposed to Love waves. Hence, larger dispersion was observed for Rayleigh waves. By assuming the average group velocity to be 1 km/s for periods between 3 and 10 s, T×β we can make a rough estimate of the average depth (H ≈ 4 ,

Fig. 8. Three normalized components of ZPP station records. Phases with periods from 2, 4, 6, 8 and 10 s are extracted from original displacement traces (top subfigure) using a Gaussian filter.

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

where H is the depth; T is period and β is the average shear wave velocity) of the basin based on Eq. (3). The estimated average depth turned out to be on the order of 1.5 km (depth was estimated between periods 3 and 10 and average was taken) which is in agreement with the actual sediment depth measured from geological–geophysical investigations [28]. Furthermore, it also indicates that deeper layers (above 2 km) do not have significant

Fig. 9. Average group velocity dispersion curves (fundamental mode) with associated 7 standard deviation for Love and Rayleigh waves obtained using the MFT for the Po Plain records.

237

influence the on the observed ground-motion since the dispersion curves constrain depth largely below 2 km. 6.2. Separation of body and surface waves Strong motion observations are typically recorded within the epicentral distance of 300 km, where at such distances the separation between different types of waves due to dispersion, especially in the time domain, is usually not significant [29]. Therefore isolating surface waves from body waves is a challenging task. The three components (radial, vertical and transverse) velocity records and the corresponding Fourier spectra at the ARG station are shown in Fig. 10. The arrival of surface waves is well noticeable. The surface wave arrives after about 40 s and overlaps with body waves for about 10 s, then the motion is dominated predominantly by surface waves. In the frequency domain, the low-frequency motion below 0.33 Hz is distinctively dominant. Therefore, identification in basin records of surface waves would be more easily performed in the frequency rather than in the time domain. Band-pass filters can be used to isolate surface waves. Such filtered motion would still be contaminated by early arriving long-period body waves, however their contribution to the overall motion usually is minor. Surface waves are characterized by a frequency band (primarily low frequencies) which seldom overlaps with body wave frequencies (primarily large frequencies). The difficulty in designing a band-pass filter is to find the best-suited lower and upper cutoff frequencies capable of isolating surface waves. Recently Meza-Fajardo et al. [22] have proposed a new method for the identification and extraction of surface waves from three-component seismograms which seems to be very promising. From dispersion analysis (Fig. 9), it becomes evident that surface waves can be dominant in medium to large-periods such as between 3 and 10 s. A fair dispersion was well noticed in this band. Furthermore, larger dispersion was observed for PSV for periods above 1 s. Hence, the lower and the upper cutoff frequencies of the band-pass filter were set to be 0.10 and 0.33 Hz, respectively. Thus, a bidirectional ‘Butterworth’ band-pass filter with the above mentioned lower and upper cutoff frequencies was used for the analysis of the Po Plain records.

Fig. 10. Example showing separation in both the time and the frequency domains of surface and body waves. Three component velocity time series and their corresponding Fourier amplitude spectra for the mainshock recorded at the ARG station are displayed. The separation of surface waves is more evident in the frequency domain (0.1– 0.33 Hz).

238

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

Fig. 11. Particle motion plots (hodograms) of body and Rayleigh waves at the CST station for the earthquake of May 29. (a) Band-passed (3–10 s) displacement time series of horizontal (normalized) and vertical components (normalized to the PGV of horizontal component), (b) hodogram of displacement from 40 to 100 s. The bottom two panels contain hodograms of 10 segments (10–40 s, 40–50 s, 50–60 s, 60–70 s, 70–80 s, 80–90 s, 90–100 s, 100–110 s, 110–120 s, 120–130 s) of ground-motions. The dark and gray circles refer to the start and the end of particle motion respectively.

It is a common practice, to facilitate the isolation of surface waves, to rotate the mutually perpendicular EW, NS and vertical components of ground-motion with respect to the radial and transverse directions of the causative fault. It is a well-known fact that Rayleigh waves are rich in the radial component while Love waves are polarized in the transverse component. Thus, the filtered waves from the transverse component were assumed to be Love waves while the vertical and the radial components of motion were considered as the vertical and horizontal components of Rayleigh waves respectively. 6.3. Hodograms To view the particle motions produced by surface waves, Fig. 11 shows the hodograms plotted for the in-plane Rayleigh waves retrieved from the records of the CST station of the earthquake of May 29. The hodogram plotted for the motion between 40 and 100 s shows a well displayed elliptical particle orbit (Fig. 11b). Furthermore, hodograms are illustrated for 10 time frames starting from 10 s at 10 s time interval, except for the first segment (10– 40 s). Body waves are dominant between 10 and 40 s, which resulted in a complex particle motion. The arrival of Rayleigh waves at 40 s is well reflected in the hodogram (40–50 s segment) as the motion follows an elliptical path from 40 s, which continues

up to 100 s. A well displayed ellipticity was observed for segments such as 50–60 s, 60–70 s, 80–90 s and 90–100 s. Furthermore, the particle motion between 40 and 100 s is consistently retrograde as expected for a Rayleigh wave. 6.4. Direction of surface wave propagation To investigate the spatial origin of surface waves, the backazimuth (measured clockwise from north) of Rayleigh waves was estimated using the method proposed by Baker and Stevens [3] which was adopted by Wang et al. [29]. The basic idea is to find an azimuth for which the vertical and Hilbert transformed radial component particle motions form a straight line [29]. The first step is to rotate the two horizontal components into assumed radial and transverse directions, with a trial backazimuth ranging from 0° to 360°. Then, the radial component is Hilbert transformed. The Hilbert transformation has the effect of shifting the horizontal waveform by a phase of 90°, which converts the elliptical polarization of the Rayleigh wave into a linear polarization. The next step is to calculate the cross correlation of the Hilbert transformed radial component with the vertical component. The largest value of the cross correlation of these two components of groundmotion corresponds to the optimum backazimuth of the Rayleigh wave being examined. The procedure is illustrated in Fig. 12. It

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

239

Fig. 12. An illustration of the process used to calculate the backazimuth of Rayleigh wave propagation:(a) band-pass filtered (3–10 s) displacement traces of NS, EW and vertical components, (b) comparisons of the vertical and Hilbert transformed radial displacements, and (c) cross correlation for trial backazimuth.

Fig. 13. A map showing directions of Rayleigh wave propagations in the Po Plain as deduced by the procedure proposed by [3] applied to various seismic stations located inside the Po Plain.

shows the original three components (NS, EW, and vertical) bandpass filtered (3–10 s) displacement traces (Fig. 12a). The two horizontal components are a mixture of Rayleigh and Love surface waves. To separate Rayleigh and Love waves, the two horizontal components have been rotated 249° (measured clockwise from north) since this rotation provides the largest value of cross correlation (Fig. 12c). A fairly well matching Hilbert transformed radial component with the vertical component can also be observed (Fig. 12b). Since a surface wave trace usually includes different propagation modes and multi-pathing waves with different polarizations, the calculated backazimuth represents an average value. Hence, the calculated direction of surface wave propagation should be regarded as an average of the waves propagating in the dominant direction.

Fig. 13 shows the estimated average backazimuth of Rayleigh waves reaching the stations from the mainshock of May 20. Interestingly the overall direction from all stations seems to point towards the source. The Rayleigh waves seem to propagate in the radial direction from the source, which implies that the basin structure is characterized by strong lateral velocity variation resulting in a large amount of surface waves generated despite the fact that the source is located inside the basin. It is important to remark that surface waves are off-source waves, i.e. they originate due to the propagation effects, and generally basin margins are the main source of generation. In this case, since the source is located within the basin, not much diffraction is expected. However, in addition to the influence of lateral heterogeneities inside the Po Plain, other influencing factors include (a) width of the basin which is quite large (greater than 100 km), hence the modal

240

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

conversion of body waves could take place as they propagate, (b) low velocity sediments, and (c) shallow hypocentral source. Such factors strongly favor the generation of surface waves in the radiation path from the source. Thus the backazimuth would be similar to the source-station azimuth.

7. Envelope delay spectrum The complex basin structure under a site causes both ground amplification due to the effect of soft layers and an increase of duration due to the presence of surface waves. Spectral ratio methods have been developed and used to estimate ground amplification, however, the duration lengthening due to the latearrival of surface waves is rarely estimated. To this end, the frequency-dependent envelope delay spectrum can be potentially useful. One of the early proponents of the use of envelope delay (or group delay) spectrum is [25]. The application of envelope delay spectrum on weak and moderate motion recordings from the European test site near Volvi (Greece) has been found to yield satisfactory results [6]. The envelope delay has the advantage over phase difference that it can be interpreted more directly in terms of duration. Taking advantage of this attribute, an approach to extend the duration of ground-motion generated using the stochastic simulation approach for basin sites was attempted by Boore [9]. The calculation of envelope delays involves taking the derivative of phase with respect to frequency. This was carried out using the method suggested by Boore [9] which does not require unwrapping and ill-posed finite-difference differentiation. Smoothing of the envelope delay spectrum was carried out as suggested by Beauval et al. [6]. The adopted approach has been successfully tested by Abraham [1]. Formally, the envelope delays is computed as follows:

Tgr (f ) =

1 dΦ (f ) 2π df

(3)

where Tgr (f ) is the envelope delay spectrum, Φ is the phase, f is the frequency. The estimated envelope delay spectra of the radial component records from the CSP station of mainshock and aftershock are illustrated in Fig. 14. In order to facilitate comparison with the time series, the envelope delay amplitude is plotted on the abscissa even though it is a dependent variable. The spectra from both earthquakes recorded at a same station show remarkable similarity although their waveforms look different. The amplitude of envelope delay has nothing to do with the waveform characteristics: if the record is shifted by an amount, then the envelope delay spectrum will also be shifted by the same amount. The envelope delay spectrum indicates a strong nonstationarity for frequencies less than about 2 Hz, with the lower frequencies arriving later than the higher frequencies. This is consistent with the velocity waveforms shown in Fig. 14. Furthermore, they show a peak towards low frequencies with maximum amplitude around the fundamental resonance frequency determined by the spectral ratio method, which is in line with the expectations. The consistent peak in the spectra over low frequencies (primarily due to surface waves) seems to suggest that the envelope delays associated with surface waves are largely dependent on the basin characteristics rather than on the source parameters. Although, the source parameters (e.g. focal mechanism, hypocentral depth) could influence the generation of surface waves, the latter may be due to the geometrical characteristics of the basin. Thus, this similarity of the envelope delay spectra for the mainshock of May 20 and the aftershock(1) of May 29 was in a way expected. The larger envelope delays amplitudes observed below 2 Hz denotes stronger frequency dependency due to surface waves, whereas envelope delays amplitudes above 2 Hz are relatively insensitive to frequency. Most of the energy associated with the envelope delay spectrum occurs near 40 s. Although no attempt is

Fig. 14. The radial component envelope delay spectra of two May 20 and May 29 earthquakes recorded at the CSP station in the Po Plain (Northern Italy).The top panels show the normalized velocity time series. The gray line indicates the smoothed envelope delay spectra. These show strong dependence on the frequency, with lower frequencies corresponding to greater values of the amplitudes as would be predicted from the time histories.

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

made to isolate the envelope delay spectrum resulting from the basin-induced surface waves, inspection of the spectra suggests that the observed large amplitudes below 2 Hz are caused by surface waves. The peak values of the spectra occurring around 80 s is again due to surface waves and it causes a duration lengthening of the signal.

8. Discussion and conclusions The Po Plain in Northern Italy is an important geological feature where its deep and wide basin structure is expected to influence ground-motion characteristics. During May 20–29, 2012 Emilia sequence, the basin response was well displayed in the groundmotion records in the Po Plain. Well distinguished long-period motion arriving at the later part of the seismogram, which eventually increased the overall duration of the motion was systematically observed. The present study focused on the ML 5.90 mainshock of May 20, and ML 5.80 possible aftershock of May 29 which produced high quality broad-band recordings at the basin. Distinctive features in the ground-motion records were noticed in the Po Plain. In the near-field region, large horizontal velocity pulses were observed, while in the far-field, dispersive, long-period motion, prolonging the duration of ground shaking was observed. Below is a summary of the main observations made after analyzing the strong motion records.

241

basin-effect. The increased durations and late arrival of long-period motions are believed to have had a significant contribution in the damages observed in the Emilia region and in the widespread liquefaction phenomena that occurred. The large amplitude of long-period motions could be attributed to the geometric structure of the basin (depth), while longerdurations could be attributed to the dispersion characteristics of the basin. The concluding remarks from the current study are that the Po Plain basin structure in Northern Italy is unique in terms of its geomorphology and buried geometry. Furthermore, the seismicity affecting the region is characterized by shallow hypocentral earthquakes. These two unfavorable circumstances increase the seismic hazard of the basin. To mitigate the seismic risk associated with the building stock in the Po Plain from future earthquakes, expected basin-effects need to be carefully assessed.

Acknowledgments We thank Dr. Corchete for supplying MFT program for calculating group velocity curves, and also Dr. Baker for sharing with us the matlab code to estimate the back azimuth of Rayleigh waves. We acknowledge the Italian Department of Civil Protection (DPC) for providing us the strong-motion data of the Emilia May 20–29, 2012 earthquakes.

 The intrinsic nature of surface waves when traveling in layered





media is geometric dispersion. The estimated frequency-dependence of group velocity of surface waves from the observed ground-motion obtained using the MFT record station shows evidence of dispersion beyond 3 s. Rayleigh waves were more dispersive than Love waves. A significant separation of waves in the frequency domain was also noticed. Therefore a band-pass filtering of well dispersed long-period waves in the range between 3 and 10 s was carried out. The filtered Rayleigh waves displayed a strong elliptical particle motion reflecting the nature of Rayleigh waves and their unequivocal presence in the observed ground-motion record. It is important to highlight the role of the earthquake source. Usually strong-amplitude surface waves are generated when the source is located outside the basin. Yet, despite the hypocenters of the mainshock and aftershock of the Emilia earthquakes were located inside the basin, the amount of generated surface waves was relevant mainly because of the shallow hypocentral depth. Furthermore, the basin is highly heterogeneous both in the vertical and lateral directions. Since the information of ground-motion duration is reflected in the phase Fourier spectrum, the derivative of the phase with respect to frequency yields the arrival time of energy as a function of frequency. This quantity is commonly referred to in the literature as the envelope delay spectrum. The increase of envelope delay amplitudes appears closely related to the local geology and the envelope delay spectrum gives an insightful explanation for the long duration of ground shaking. Prolonged motions were caused by the late arrival of surface waves and it occurred for the most part around the fundamental frequency of the basin.

From the foregoing discussion it is rather evident that basineffects cannot be simply reduced to amplification of groundmotion amplitude. Further modifications of the characteristics of ground-motion are also introduced like enrichment of low-frequency energy content and also prolonging the duration of the seismogram. Yet, the engineering community strongly identifies the spectral amplification as the main, if not the only, possible

References [1] Abraham AJ. Study of basin-induced surface waves and their numerical simulation (Ph.D. thesis). Pavia, Italy: ROSE School, IUSS; 2003. [2] Albini P, Rovida A. The 12 May 1802 earthquake (N Italy) in its historical and seismological context. J Seismol 2010;14(3):629–51. [3] Baker GE, Stevens JL. Backazimuth estimation reliability using surface wave polarization. Geophys Res Lett 2004;31(9):1–4. [4] Bard PY, Campillo M, Sánchez-Sesma F, Chávez Garcia FJ. The mexico earthquake of september 19, 1985-a theoretical investigation of large-and small scale amplification effects in the mexico city valley. Earthq Spectra 1998;4 (6):609–33. [5] Basili R, Valensise G, Vannoli P, Burrato P, Fracassi U, Mariano S, Tiberti MM, Boschi E. The database of individual seismogenic sources (DISS), version 3: summarizing 20 years of research on Italy's earthquake geology. Tectonophysics 2007;453(1–4):20–43. [6] Beauval C, Bard PY, Moczo P, Kristek J. Quantification of frequency-dependent lengthening of seismic ground-motion duration due to local geology: applications to the volvi area (Greece). Bull Seismol Soc Am 2003;93(1):371–85. [7] Bonini L, Toscani G, Seno S. Three-dimensional segmentation and different rupture behavior during the 2012 Emilia seismic sequence in Northern Italy. Tectonophysics 2004;630:33–42. [8] Boore D. On pads and filters: processing strong-motion data. Bull Seismol Soc Am 2005;95(2):745–50. [9] Boore D. Phase derivatives and simulation of strong ground motions. Bull Seismol Soc Am 2003;93(3):1132–43. [10] Bradley BA. Strong ground motion characteristics observed in the 4 September Darfield, New Zealand earthquake. Soil Dyn Earthq Eng 2012;42:32–46. [11] Burrato P, Ciucci F, Valensise G. An inventory of river anomalies in the Po plain, Northern Italy: evidence for active blind thrust faulting. Ann Geophys 2003;46 (5):865–82. [12] Ciaccio MG, Chiarabba C. Tomographic models and seismotectonics of the Reggio Emilia region, Italy. Tectonophysics 2002;344(3–4):261–76. [13] Corchete V. A methodology for filtering and inversion of surface waves: an example of application to the lithospheric structure determination of the South Iberia. Tecnociencia 2006;8(1):91–112. [14] DISS Working group. Database of individual seismogenic sources (version 3.0.4). A compilation of potential sources for earthquakes larger than M 5.5 in Italy and surrounding areas. Available at 〈http://www.ingv.it/DISS〉; 2004. [15] Dziewonski A, Bloch S, Landisman M. A technique for the analysis of transient seismic signals. Bull Seismol Soc Am 1969;59(1):427–44. [16] Furumura T, Hayakawa T. Anomalous propagation of long-period ground motions recorded in Tokyo during the 23 October 2004 MW 6.6 Niigata-ken Chuetsu, Japan, earthquake. Bull Seismol Soc Am 2007;97(3):863–80. [17] Galli P. New empirical relationships between magnitude and distance for liquefaction. Tectonophysics 2000;324:169–87. [18] Kumar P, Georgiou E. Wavelet analysis for geophysical applications. Rev Geophys 1997;35(4):385–412.

242

J. Ronald Abraham et al. / Soil Dynamics and Earthquake Engineering 78 (2015) 230–242

[19] Lai CG, Bozzoni F, Mangriotis M, Martinelli M. Soil liquefaction during the May 20, 2012 M5.9 Emilia earthquake, Northern Italy: field reconnaissance and post-event assessment. Earthquake Spectra 2015 (In press). [20] Luzi L, Pacor F, Ameri G, Puglia R, Burrato P, Massa M, Augliera P. Franceschina G, Lovati S, and Castro R. Overview on the Strong-Motion Data Recorded during the May–June 2012 Emilia Seismic Sequence. Seismol Res Lett 2013;84 (4):629–44. [21] Mantovani E, Viti M, Babbucci D, Cenni N, Tamburelli C, Vannucchi A, Falciani F, Fianchisti G, Baglione M, D’Intinosante V, Fabbroni P. Sismotettonica dell’Appennino settentrionale Implicazioni per la pericolosità sismica della Toscana Regione Toscana; 2011. [22] Meza-Fajardo KC, Papageorgiou AS, Semblat JF. Identification and Extraction of Surface Waves from Three-Component Seismograms Based on the Normalized Inner Product. Bull Seismol Soc Am 2015;105(1):210–29. [23] Rovida A, Camassi R, Gasperini P, Stucchi M. CPTI11, parametric catalogue of Italian earthquakes, 2011 version, Milano, Bologna. Availabe at: http://emi dius.mi.ingv.it/CPTI; 2010. [24] Saraò A, Peruzza L. Fault-plane solutions from moment-tensor inversion and preliminary Coulomb stress analysis for the Emilia Plain. Ann Geophys 2012;55(4):647–54.

[25] Sawada S. Phase characteristics on the site amplification of layered ground with irregular interface. Rotterdam: Balkema; 1998. [26] Scognamiglio L, Margheriti L, Mele FM, Tinti E, Bono A, Gori PD, Lauciani V, Lucente FP, Mandiello AG, Marcocci C, Mazza S, Pintore S, Quintiliani M. The 2012 Pianura Padana Emiliana seimic sequence: locations, moment tensors and magnitudes. Ann Geophys 2012;55(5):549–59. [27] Selvaggi G, Ferulano F, Di Bona M, Frepoli A, Azzara R, Basili A, Chiarabba C, Ciaccio MG, Luccio F, Lucente FP, Margheriti L, Nostro C. The Mw 5.4 Reggio Emilia 1996 earthquake: active compressional tectonics in the Po plain, Italy. Geophys J Int 2001;144:1–13. [28] Vuan A, Klin P, Laurenzano G, Priolo E. Far-source long-period displacement response spectra in the Po and Venetian plains (Italy) from 3D wavefield simulations. Bull Seismol Soc Am 2011;101(3):1055–72. [29] Wang G, Tang G, Boore D, Burbach G, Jackson C, Zhou X, Lin Q. Surface waves in the western Taiwan coastal plain from an aftershock of the 1999 Chi–Chi, Taiwan, earthquake. Bull Seismol Soc Am 2006;96(3):821–45.