# Bulletin of the Seismological Society of America

## Abstract

Moment magnitudes differing by up to 0.5 units have been published for the same events of the 2012 Ferrara seismic sequence. With respect to the mainshock that occurred on 20 May 2012, results by Malagnini *et al.* (2012) and Pondrelli *et al.* (2012) represent opposite extremes: although the former used model Padania, a region‐specific velocity structure based on all the available geological and geophysical information from local studies, the latter used a global crustal model with a set of phase corrections calibrated over the central Apennines by Ekström *et al.* (1998). Model Padania well reproduces the observed dispersion of surface‐wave group velocities in a band of shorter periods, between 33 and 100 s, whereas Pondrelli *et al.* (2012) performed their inversions in the 50–150 s period band. Here, we show that because surface waves generated within the thick sediments of the Po river floodplain dominated the seismograms, the source excitation terms that came out of a regression scheme performed on the ground motions recorded during the sequence were systematically affected by a broadband increase of the spectral amplitudes at frequencies below 0.4 Hz (frequency range of the regressions: from 0.1 to 22.5 Hz). As a consequence, the two largest events of the sequence share a common true moment magnitude *M*_{w}∼5.6, even though their enhanced spectral level from 0.1 to 0.4 Hz is consistent with *M*_{w}∼6.0.

*Electronic Supplement:*Figures of hypocentral distances, moment tensor (MT) solutions, and site terms from the ground‐motion regression; and tables of velocity models and MT solutions.

## Introduction

The first mainshock of the Ferrara/Mirandola seismic sequence (*M*_{L} 5.9, *M*_{w} 5.6, see Malagnini *et al.*, 2012) struck on 20 May 2012 and was preceded by a foreshock (*M*_{L} 4.1, *M*_{w} 3.8). The sequence went on with six more events with *M*_{L}≥5, the latest one on 3 June 2012. On broadband seismological data collected from the sequence, Malagnini *et al.* (2012) calibrated the Padania velocity–density structure (see Ⓔ Table S1, available in the electronic supplement to this article—the reader should be aware that Padania is not an acronym; it is in fact a name that is used to indicate a region that coincides with the Po river floodplain) with the help of the available seismic profiles for the region (Scrocca *et al.*, 2007; Carminati *et al.*, 2010). In its deepest layers, model Padania is identical to model Central Italy, Apennines (CIA, by Herrmann *et al.*, 2011, see Ⓔ Table S1); however, due to the presence of the thick sediments of the Po river floodplain, the seismic velocities in the upper 5 km of model Padania are much slower than in the CIA model.

Malagnini *et al.* (2012) validated model Padania by fitting the group velocity dispersion curves observed outside of the sediments, roughly 100 km north of the sequence, at stations MAGA and SALO. As a result of their effort, Malagnini *et al.* (2012) obtained stable moment tensor (MT) solutions down to *M*_{w} 3.24. The largest event of the Ferrara/Mirandola sequence occurred on 20 May 2012 at 02:03:53 UCT (*M*_{L} 5.9), although damage and loss of lives were mostly a consequence of the second main event of 29 May 2012 at 07:00:03 UCT, *M*_{L} 5.8 (for details on the 20 earthquakes with an MT solution, see Ⓔ table S2 by Malagnini *et al.*, 2012).

A different approach is the one described in Pondrelli *et al.* (2012): using a preliminary reference Earth model (PREM)‐like Earth structure (defined by Dziewonski and Anderson, 1981) without a thick sedimentary coverage, they obtained a suite of MT solutions for some events of the sequence. Ⓔ Table S1 also shows the ak135 model, a travel‐time global model proposed by Kennett *et al.* (1995), very similar to PREM, which we used for a calculation presented in the MT Solutions section. Pondrelli *et al.* (2012) used such a seemingly inadequate velocity model together with the phase corrections provided by Ekström *et al.* (1998). The latter approach allows a decent fit in the time domain between PREM‐based synthetics and observed seismograms, even within the Mediterranean region.

Although an *ad hoc* phase correction of PREM synthetics allows the calculation of meaningful focal solutions in the Apennines, Giardini *et al.* (1994) demonstrated the substantial differences between PREM and the actual velocity structure of the lithosphere of the Mediterranean region; they did so by sampling the dispersion of Love‐wave group velocities along a randomly chosen path connecting the MedNed station VSL to a seismic source in Iran. Specifically, for frequencies higher than 0.01 Hz, the PREM group velocities resulted much faster than the observed ones. However, PREM adequately reproduced the group velocity dispersion curves across the Mediterranean region for periods longer than 100 s. Regardless of phase corrections (i.e., even if the phasing issue can easily be overcome), a faster Earth requires stronger sources to match the observed amplitudes, resulting in moment magnitudes biased high.

Saraò and Peruzza (2012) used the Venetian Plains velocity model by Vuan *et al.* (2011, see Ⓔ Table S1) with the inversion code implemented by Dreger (2002). They obtained MT solutions for many events of the Ferrara sequence, including the two largest shocks. Even though broadband seismograms recorded by the entire Istituto Nazionale di Geofisica e Vulcanologia (INGV) national network were easily available through the European Integrated Data Archive infrastructure that is active within Observatories and Research Facilities for European Seismology), Saraò and Peruzza chose to use only data from the northeast (NE) Italy seismic network managed by the Istituto Nazionale di Oceanografia e Geofisica Sperimentale (OGS). A few INGV stations from NE Italy were added to the OGS set of instruments to better constrain the inversion results, but the meaning of the quoted statement is not clear in their paper.

The Venetian Plains crustal model features a 1.5‐km‐thick layer of loose sediments sitting on top of a layer of consolidated sediments, and its overall characteristics are not very dissimilar from those of model Padania by Malagnini *et al.* (2012). However, the *M*_{w}s from Saraò and Peruzza (2012) are up to 0.5 magnitude units larger than the same estimates obtained by Malagnini *et al.* (2012). No list of used waveforms was indicated in their paper, whether from stations of the OGS network or from elsewhere.

Here, we perform MT inversions using detailed velocity structures of various kinds: from two regional models specific for the Po river floodplain to a global model like ak135. We show that discrepancies of up to 0.5 *M*_{w} units may be obtained if the Earth structure does not include the existing low velocities at shallow depths. As an independent proof of our statement, we analyze the excitation levels of a set of regressions performed on the high‐frequency ground motions between 0.1 and 22.5 Hz and observe that it is possible to distinguish the contribution of the thick sediments at low frequencies that cause an apparent amplitude enhancement of the source‐related excitation terms, even on rock sites located outside the valley at ∼100 km distance from the seismic sources.

## MT Solutions

In this section, we provide a comparison between different MT solutions calculated for the mainshock of the Ferrara sequence that occurred on 20 May 2012. To pursue the task, we compute the Green’s functions for three different Earth models: the ak135 global model (Kennett *et al.*, 1995), the Venetian Plains crustal model (Vuan *et al.*, 2011), and the CIA model (Herrmann *et al.*, 2011). We then calculate the MT solutions for the cited main event, trying to reproduce the results described by Pondrelli *et al.* (2012) and by Saraò and Peruzza (2012). Finally, we compare all solutions with the MT published for the same event by Malagnini *et al.* (2012) that was computed using the Padania model. The parameters of all solutions, including the ones obtained by Malagnini *et al.* (2012) with model Padania, are summarized in Ⓔ Table S2. Ⓔ Figures S2–S5 show waveform fits and other information for the four best solutions calculated here.

We chose ak135 because it is a global Earth model that is very much similar to PREM. Although the velocity structure of ak135 below 120 km depth came from the work of Kennett *et al*. (1995), its density and *Q* structures came from Montagner and Kennett (1996). ak135 provides a good representation of the full set of smoothed empirical travel times and should be very suitable for predicting the arrival times of a wide variety of seismic phases for use in event location or phase association procedures. The crustal structure proposed for the Venetian Plains by Vuan *et al.* (2011) was reconstructed from oil exploration data, deep seismic soundings, and seismic reflection measurements; it was validated by comparing recorded waveforms from medium‐sized earthquakes in the 1–40 s period range. Finally, the CIA model (Herrmann *et al.*, 2011) was calibrated in a region centered around the city of L’Aquila during the 2009 seismic sequence; it was constrained by deep crustal profiles, by fitting surface‐wave dispersion curves, and by the inversion of teleseismic *P*‐wave receiver functions. The CIA model was tested through waveform fit.

As for the choice of the bandwidth to be used in this study for the MT inversions, we note that Saraò and Peruzza (2012) worked in the 20–50 s period range, whereas Pondrelli *et al.* (2012) chose a bandwidth in the 50–150 s period range, and Malagnini *et al.* (2012) used the 33–100 s range. No complete waveform lists were provided either by Pondrelli *et al.* (2012) or by Saraò and Peruzza (2012), and a few examples of waveform fits are shown only in Pondrelli *et al.* (2012). To reproduce their results, we were forced to redo the computation in both cases.

After the calculation of an MT solution with the Venetian Plains model in the 20–50 s bandwidth used by Saraò and Peruzza (2012), we noted that the obtained synthetic seismograms were dominated by a strong ringing developed within the shallow sedimentary layer (see Ⓔ Fig. S5). The reason for this phenomenon is the inadequate attenuation structure for this model, and we decided to apply the 50–150 s band‐pass filter used by Pondrelli *et al.* (2012) also for this inversion (see Ⓔ Fig. S3). For consistency, and because the CIA model has no sedimentary coverage, the same bandwidth between 50 and 150 s was chosen also in the case presented here (see Ⓔ Fig. S4).

What we have just described demonstrates that the different solutions are not easily comparable, either against one another, or against the one by Malagnini *et al.* (2012). Because Padania is the only model that correctly takes into consideration the contribution of the shallow sedimentary layer to the Green’s functions at periods shorter than 50 s, our preferred MT solution for the mainshock of 20 May 2012 is the one published by Malagnini *et al.* (2012).

## High‐Frequency Ground Motions

In this section, we show that the seismic moments of the two main earthquakes of the Ferrara sequence are almost equal and that the amplitudes of the high‐frequency ground motions are consistent, in both cases, with an *M*_{w} 5.6. However, because of the systematic action of the surface waves excited within the sediments of the Po river floodplain, the amplitudes of the shaking below 0.4 Hz are more consistent with an *M*_{w}∼6.0. Details of the technique used here may be found in Malagnini and Dreger (2016) and in Munafò *et al.* (2016).

We analyze waveforms from 81 earthquakes, *M*_{L} ranging from 2 to 5.9; locations of events and stations are indicated in Figure 1, whereas the distance sampling at each station is plotted in Ⓔ Figure S1. Each time history (ground velocity) was corrected for the instrument response, visually examined to eliminate multiple events and noisy signals, and handpicked for *P*‐ and *S*‐wave arrivals. The 2867 selected waveforms were then band‐pass filtered around a set of central frequencies *f*_{i} using an eight‐pole Butterworth high‐pass filter with corner at , followed by an eight‐pole low‐pass Butterworth filter with corner at (a symmetric band‐pass filter in the log space).

If *a*_{n}(*f*_{i},*r*_{jk}) indicates the peak value carried by the *n*th waveform (*j*th event, *k*th sites) filtered around frequency *f*_{i}, its log_{10} value may be written as: *A*_{n}(*f*_{i},*r*_{jk})=log_{10}*a*_{n}(*f*_{i},*r*_{jk}). Throughout this article, the word “site” does actually mean “individual component of the ground motion.” Based on this definition, if we talk about *K* as the total number of sites in our data set, we actually mean the total number of individual stations multiplied by the three components of the ground motion. In our application, however, *K* refers only to the good sites (i.e., to all the individual components of the ground motion that are not affected by systematic anomalies of any kind, either related to instrumental issues, peculiar characteristics of the ambient noise, or strong site responses).

As described in detail by Malagnini and Dreger (2016), the random vibration theory (Cartwright and Longuet‐Higgins, 1956), the convolution theorem, and the Parseval equality can be used all together to separate the contributions of excitation, propagation, and site. After a formal rearrangement, the entire set of observations may be casted into a matrix using the following linear representation: (1)in which *n* indicates the *n*th seismogram and goes from 1 to *N*, the total number of waveforms in our data set (if *J* is the total number of sources and *K* is the total number of good individual components of the ground motion in our data set, *N*≤*JK*), the frequency subscript *i* indicates the *i*th sampling frequency and goes from 1 to *N*_{F}=30.

Using equation (1), we carried out a regression at each sampling frequency (*f*_{i}). The term SRC_{j}(*f*_{i},*r*_{0}) in equation (1) represents the excitation due to the *j*th source, referred to an arbitrary hypocentral distance *r*_{0}=120 km (Figs. 2 and 3) and to an average site response. The term SITE_{k}(*f*_{i}) represents the contribution of *k*th site term to the shaking at the specific distance *r*_{jk}. Finally, the term *D*(*r*_{jk},*r*_{0},*f*_{i}) describes the effect of the crustal wave propagation (geometrical spreading, *g*(*r*), and anelastic attenuation, *Q*(*f*), see equations (2) and (3)). By modeling *D*(*r*_{jk},*r*_{0},*f*_{i}) using equation (2) and a trial‐and‐error procedure, we obtain the parameters described in equation (3). (2)in which (3)Regressions are stabilized by forcing constraints (4) and (5) onto the system (1): (4)(5)On the one hand, by forcing equation (4) onto our system of equations, we totally decouple the combined effect of excitation and site from the contributions of crustal propagation; on the other hand, the effect of constraint (5) is to provide a physical meaning for the excitation terms of equation (1): they now represent the source peak spectra that would be observed at the (arbitrary) reference hypocentral distance *r*_{0}=120 km by the average reference site.

We decided to force equation (5) on a small subset of reference rock sites that are located outside the sedimentary coverage of the Po river floodplain, which are thought to have minimal site responses. In fact, we chose as reference sites the horizontal component of the ground motion from three stations on firm rock named MAGA, SALO, and ROVR, all characterized by significant elevations over sea level (1265, 600, and 1316 m, respectively). Based on our definition of the word site, *K*_{REFSITE}=6 in equation (5), whereas our total number of sites (including all horizontal and vertical components of the ground motion) is *K*=60. After the application of constraint (5), each site term represents the anomaly with respect to the average horizontal site term calculated on the six horizontal components of the ground motion of stations MAGA, SALO, and ROVR.

It must be clear that because of the effect of constraint (5), all the empirical excitation terms SRC_{j}(*f*_{i},*r*_{0}) implicitly contain the average site term calculated on the *K*_{REFSITE} reference sites (as described by equations 7 and 8). Because in constraint (5) we average only the horizontal components of the ground motion at three reference sites, our excitation terms represent the average horizontal motion that would be observed at the same set of rock sites, if the earthquakes were all located at the reference hypocentral distance *r*_{0}=120 km chosen for the implementation of constraint (4). All site terms (either vertical or horizontal) that do not belong to the reference set now represent anomalies with respect to the average reference specified by constraint (5).

The *j*th site term in equation (5) can be described in terms of a local anomaly with respect to the average network site: (6)in which *ν*(*f*_{i})exp(−*πκ*_{0EFF}*f*_{i}) is the average reference site (averaged on the set of channels that enter constraint 5 (see equation 7). The two terms *ν*(*f* ) and exp(−*πκf* ) are not separable in our approach; they trade off with each other, and with the source model *s*(*f* ) (see equation 8). The site terms described in equation (6) are plotted in Ⓔ Figure S6. The term *ν*_{k}(*f* )exp(−*πκ*_{k}*f* ) refers to the specific *k*th site and contains the amplification due to contrasts, or gradients, in seismic impedance and the high‐frequency anelastic attenuation.

Because we do not know the true site amplification, we choose it equal to a certain function *ν*(*f* ) (e.g., the generic or hard‐rock site by Boore and Joyner, 1997, or their C or D site, or any site of the National Earthquake Hazards Reduction Program classification, or even a constant term equal to unity) and determine an effective parameter (*κ*_{0EFF}), which may approximately be written as: (7)in which *K*_{REFSITE}=6 refers, as explained earlier, to the terms that enter constraint (5).(8)in which *ρ* and *β* are the rock density and the shear‐wave velocity, respectively, and Δ*σ* is the Brune (1970, 1971) stress drop.

Figure 3 contains the empirical excitation terms obtained from our data set, modeled using equation (8) and the attenuation parameters described in equation (3). We arbitrarily choose the unknown function *ν*(*f* ) in equations (7) and (8) to be equal to the generic rock site by Boore and Joyner (1997); coherently with their article, we couple it to a high‐frequency filter whose coefficient is (9)(see Akinci *et al.*, 2001).

We use the Brune source spectrum with a magnitude‐dependent stress drop Δ*σ* parameter that is quantified as follows: (10)It is important to point out the unresolvable trade‐off existing between the stress drop Δ*σ* in the Brune source model (8) and the site‐related term *ν*(*f* )exp(−*πκ*_{0EFF}*f* ) of equation (7). These two terms are in competition with each other in the definition of the seismic spectrum at high frequencies: if we assume that *ν*(*f* ) does not depend on the event magnitude, the same fit of Figure 3 may be obtained by keeping a constant stress drop (e.g., Δ*σ*=12 MPa) and by varying the parameter *κ*_{0EFF} as a function of magnitude.

A discussion on source scaling is beyond the scope of the present article. Nevertheless, we think that stress drop as a function of magnitude confirms non‐self‐similarity in this region. We however point out that choosing one approach or the other is immaterial to a discussion focused entirely on the low‐frequency part of the seismic spectrum, which is not affected by these parameters.

Nevertheless, to avoid the contributions of any nonlinear behaviors of the soft sediments to the excitation terms, we refer our results to the average taken on the horizontal components of the ground motions recorded by stations outside the sediments. To be consistent with the work by Malagnini *et al.* (2012), we apply constraint (5) to the horizontal motions observed at rock stations MAGA, SALO, and ROVR. Using rock stations ∼100 km away from the seismic sources, we eliminate issues (doubts) related to possible nonlinear behaviors of the sediments, which would have physically justified a magnitude‐dependent site effect. Recall that the excitation terms are to be referred to the average reference site defined by constraint (5). It is worth mentioning that inversions run with different choices of the average reference site yielded practically indistinguishable results.

## Discussion

Accurately evaluating the size of an earthquake is crucial for many seismological applications; for example, a product like ShakeMap provides useful maps of ground shaking only when the magnitudes of the recorded earthquakes are correctly calculated and when the correct excitation terms are applied. In the case studied here, the systematic deformation of the excitation terms observed in Figure 3 must be taken into account when we assess the characteristics of the earthquake‐induced ground motions.

Strong analogies may be found with other regions of the world with similar crustal structures (e.g., the New Madrid Seismic Zone [NMSZ], see Julià *et al.*, 2004, or the Kachchh region, struck by the *M*_{w} 7.6 Bhuj earthquake of 26 January 2001, see Bodin *et al.*, 2004). More specifically, figure 3 of the paper by Bodin *et al.* (2004) shows a spectacular systematic deformation of the excitation terms of the Bhuj seismic sequence at low frequency, similarly to what is observed in the excitation terms of our Figure 3.

The broadband effect of a thick sedimentary coverage on seismic spectra originated within the basin itself is the outcome of three competing phenomena: (1) the amplification of the ground‐motion amplitudes due to the impedance contrast between the basement rocks, where the earthquakes originated, and the sediment layer at the surface (including the vertical gradient of the seismic velocities due to sediment compaction); (2) the strong anelastic attenuation of seismic waves traveling within loose sediments, which is especially efficient at high frequency; and (3) possible 3D effects that may be induced within the sediments of the Po river floodplain (such as the ones described by Malagnini *et al.*, 1996).

In case of wide and thick sedimentary bodies (e.g., the Po river floodplain, or the Mississippi embayment), the cumulative effect resulting from the three phenomena just described needs to be incorporated into the current hazard maps. A good example of the described effects (with the exception of the one by the 3D geometry) was published by Julià *et al.* (2004) on the area around Memphis, in the Mississippi embayment. To understand the peculiarity of the geological situation in the region, it suffices to recall that Pieri and Groppi (1975) estimated depths as large as 8.5 km for some depocenters of the Po river floodplain.

A fourth effect that could possibly affect the shallow sediments of the Po river floodplain around Ferrara is the occurrence of nonlinear responses. However, addressing nonlinearity is beyond the scope of this study; there is no clear evidence of nonlinear behavior of the soils in our data set, with an exception made for a few deformed seismic recordings collected during the strongest earthquakes of the sequence. Such seismograms were eliminated by our data set and did not enter our regressions.

A nonlinear behavior of soils would contribute to a further reduction of the amplification of the ground‐motion amplitudes through energy dissipation occurring in hysteretic loading cycles, as well as to a reduction of the rigidity parameter at shallow depths. In turn, the effect on rigidity would increase the impedance contrast and positively contribute to the site amplification while shifting resonances toward lower frequencies.

## Conclusions

Studying regional earthquakes requires the calibration of a crustal model over the target region. This is especially true in cases of peculiar crustal structures like the Po river floodplain in northern Italy, or the Mississippi embayment in the central United States, where the NMSZ is located. For these regions, we argue that the *M*_{w}s from MT solutions calculated using a velocity model lacking a thick sedimentary coverage are significantly overestimated.

Without providing comments on the used Earth model, Pondrelli *et al.* (2012) published MT solutions for the events of the Ferrara seismic sequence obtained from PREM‐based Green’s functions. Such a lack of information calls for a clarification: Pondrelli *et al.* (2012) used central Apennines region‐specific phase corrections to work around misfits related to differences between the observed and the synthetic dispersions of group velocities, due to an inadequate crustal model that led to severe overestimations of moment magnitudes for earthquakes in the Po river floodplain.

The CIA model would lead to a magnitude overestimation of about 0.2 *M*_{w} units with respect to the determinations obtained by Malagnini *et al.* (2012) with model Padania (the solution officially provided by INGV indicates *M*_{w} 5.83 for the mainshock of 20 May 2012, see Data and Resources). However, a larger overestimation (*M*_{w} 6.1 for the same event) characterizes the Regional Centroid Moment Tensor solution published by Pondrelli *et al.* (2012), which was obtained via phase correction from PREM‐like Green’s functions.

With a regionally calibrated crustal model, Malagnini *et al.* (2012) obtained *M*_{w}∼5.6 for the mainshock of the Ferrara seismic sequence that occurred on 20 May 2012, an estimate of *M*_{w} that is quite smaller than the *M*_{w} 6.1 given by Pondrelli *et al.* (2012) and by Saraò and Peruzza (2012). Considering the widespread damage observed in the epicentral areas of the two largest shocks of the Ferrara sequence, we must assume that, historically, similar patterns of damage must have originated from the occurrence of earthquakes having sizes similar to the ones of the 2012 Ferrara/Mirandola mainshocks. As a consequence of our results, contributions of the seismic catalog to the seismic hazard in the region needs to be reassessed after re‐evaluating the proper sizes of all the historical earthquakes in the region.

Even though a discussion on the best way of doing a probabilistic seismic‐hazard analysis (PSHA) in a region covered by thick sediments is beyond the scope of this article, we still point out some issues. For PSHA applications, the maximum magnitude for the seismogenic area where the sequence took place is *M*_{w} 6.14 (Stucchi *et al.*, 2011), which is practically equal to the moment magnitude *M*_{w} 6.1 estimated by Saraò and Peruzza (2012) and by Pondrelli *et al.* (2012). The size of one of the 2012 events was almost unanimously thought to be equal to *M*_{wMAX} for the region (Meletti, personal comm., 2012). However, for the same event, Malagnini *et al.* (2012) calculated *M*_{w}∼5.6, and an *M*_{wMAX}=5.6 would be way too small for the role. Thus, we argue that the definition of *M*_{wMAX} for the Ferrara region needs to be re‐discussed, re‐defined, and re‐estimated.

Using the results of our regressions on high‐frequency ground motions, we show that the effective source terms from the two largest events of the Ferrara seismic sequence have almost equal moment magnitudes and spectral characteristics (the ones that occurred close to Ferrara on 20 May 2012 and around the city of Mirandola on 29 May 2012). We observe that all the empirical excitation terms of the sequence are affected by a broadband spectral enhancement at frequencies below 0.4 Hz. We argue that such a feature has more to do with the regional propagation through a crustal structure characterized by a thick sedimentary coverage than with the source excitation, and the structure of equation (1) is such that any systematic feature related to either crustal wave propagation and/or site effects is erroneously mapped by the regressions onto the source‐related terms.

Below 0.4 Hz, both the apparent spectral levels of the two largest events would be compatible with an *M*_{w} 6.0 earthquake, had their epicenters been outside a deep sedimentary valley. At higher frequencies, however, they follow the theoretical level expected for an *M*_{w}∼5.5 earthquake, in agreement with the moment magnitudes estimated by Malagnini *et al.* (2012). This is because the scheme of our ground‐motion regressions cannot isolate systematic effects acting on all waveforms, whether they are site‐ or propagation‐related, because it maps them onto the excitation terms. On the contrary, if the proper crustal structure is used for calculating the Green’s functions, the MT approach performs a correct discrimination between the various contributions to the ground motion.

## Data and Resources

Seismograms and earthquake catalogs used in this study were collected by the Italian National Network run by Istituto Nazionale di Geofisica e Vulcanologia (INGV). The raw waveforms are available at the European Integrated Data Archive (EIDA) repository at http://www.orfeus-eu.org/eida/eida.html (last accessed June 2016). The Time‐Domain Moment Tensor (TDMT) solution of the main event of 20 May 2012 is provided by INGV at http://cnt.rm.ingv.it/event/772691/?tab=MeccanismoFocale#TDMTinfo (last accessed July 2016). Some figures were made using the Generic Mapping Tools v.4.2.1 (Wessel and Smith, 1998).

## Acknowledgments

We thank three anonymous reviewers for their thoughtful comments that helped us improve this article. This work was supported by “Progetto di ricerca MISE-DGRME (cod. 0752.010), per la determinazione di parametri fisici della sorgente sismica per i terremoti della sequenza dell’Emilia 2012; lo studio di possibili markers sismologici del ruolo dei fluidi nella sismo genesi e lo studio delle leggi di scala dei terremoti” and by “Progetto Terremoti 2016: Reconciling Differences Between Spectral‐ and Recurrence-Interval-Based Earthquake Source Characterization, and Scaling.”

- Manuscript received 25 July 2016.