The 30 October 2020 Samos (Eastern Aegean Sea) Earthquake: effects of source rupture, path and local-site conditions on the observed and simulated ground motions

On 30 October 2020 a M W 7.0 earthquake occurred in the eastern Aegean Sea, between the Greek island of Samos and Turkey’s Aegean coast, causing considerable seismic damage and deaths, especially in the Turkish city of Izmir, approximately 70 km from the epicenter. In this study, we provide a detailed description of the Samos earthquake, starting from the fault rupture to the ground motion characteristics. We rst use Interferometric Synthetic Aperture Radar (InSAR) and Global Positioning System (GPS) data to constrain the source mechanisms. Then, we utilize this information to analyze the ground motion characteristics of the mainshock in terms of peak ground acceleration (PGA), peak ground velocity (PGV), and spectral pseudo-accelerations. Modelling of geodetic data shows that the Samos earthquake ruptured a NNE-dipping normal fault located offshore north of Samos, with up to 2.5-3 m of slip and an estimated geodetic moment of 3.3 ⨯ 10 19 Nm (M W 7.0). Although low PGA were induced by the earthquake, the ground shaking was strongly amplied in Izmir throughout the alluvial sediments. Structural damage observed in Izmir reveals the potential of seismic risk due to the local site effects. To better understand the earthquake characteristics, we generated and compared stochastic strong ground motions with the observed ground motion parameters as well as the ground motion prediction equations (GMPEs), exploring also the ecacy of the region-specic parameters which may be improved to better predict the expected ground shaking from future large earthquakes in the region.

between the main fault systems surrounding Izmir, as well as the attenuation of waves propagation through the Earth and the site speci c characteristics in the metropolitan area to generate realistic estimates of seismic hazard through scenario based ground motion modeling for the future possible earthquakes.
In this study, we investigated the main features of earthquake source rupture and mechanism together with observed ground motion variability in different sites. In this respect, rst we utilized Interferometric Synthetic Aperture Radar (InSAR) data acquired by the Sentinel-1 satellites and Global Positioning System (GPS) measurements to study the ground displacement eld and to derive, by using elastic dislocation modelling, the fault geometry and slip distribution of the causative fault segment, that represent crucial input parameters for the ground motion simulations, especially for the areas in the near-source region. In regard to earthquake ground motion, we describe the seismic wave attenuation pattern as a function of distance and frequency from source to site taking into account earthquake source complexity as well as the local soil condition properties on ground motions. To do so we generated high-frequency strong ground motion recordings obtained during the 2020 Samos earthquake using a stochastic finite-fault simulation approach. Then we compared those with observed ground accelerations and velocities to bring detailed understanding of the damage caused in the city of Izmir (Motazedian and Atkinson 2005;Boore 2009). Several approaches such as the horizontal to vertical ratio (H/V) (Nakamura 1989), the standard spectra ratio (SSR) (Borcherdt 1970) and the frequency domain transfer function (Haskel 1960) method have been applied to estimate the fundamental soil frequency as well as the site ampli cations for the stations located in the soft soils around the Izmir Bay. Both the observed and the simulated pseudo-acceleration response spectra are compared with the EC8 design spectrum and the new Turkish building code (TBSC-2018) for engineering applications.

Tectonical And Seismological Setting Of The Study Area
The Greek island of Samos is located in the eastern part of the Aegean Sea, a few kilometers west of the Turkish coast ( Fig. 1). The Aegean region and western Turkey are among the most seismically and tectonically active areas in the Euro-Mediterraenan region, therefore characterized by a signi cant seismic hazard. The current stress regime of the Samos area is associated with regional Aegean Sea NNE-SSW extension. In particular, this area represents an extensional back-arc region located behind the Hellenic subduction zone (e.g. McKenzie 1978), as a result of the ongoing movement between the Eurasian, African, Arabian and Anatolian plates (Şengör 1987;Yılmaz et al. 2000). In fact, northeast motion of the African plate underneath Greece and western Anatolia (Turkey) along the Hellenic Arc resulted in a E-W oriented rift and graben systems in the Aegean Sea, as well as in western Turkey (McKenzie 1978;Taymaz et al. 1991). This extensional crustal deformation in the region (about 7.4 mm/yr of crustal extension between Samos and western Anatolia as suggested by Vernant et al. 2014) results in various complex fault systems (a combination of normal-slip and strike-slip faults). Based on eld evidence for onshore faults and on effects of faulting on the sea oor from offshore faults, different active tectonic structures have been identi ed in the study area (e.g. Chatzipetros et al. 2013 In historical and recent times, Samos and Turkey's west coast have experienced multiple damaging earthquakes. In the early 20th century, on 11 August 1904, a M 6.8 earthquake struck the south coast of Samos causing signi cant damage in the Greek islands and destroying many cities in western Anatolia (Macropoulos et al. 2012). In addition, during the historical period, around 201-197 BC, 46-47 AD and between 1700 and 1799, circa ten damaging earthquakes occurred in and around Samos (Stiros et al. 2000). Particularly, large earthquakes occurred in the region on 10 July 1688, in 1739 and 1788 with catastrophic consequences in the city of Izmir (Soysal et al. 1981;Ambraseys 2009). Recently, a series of moderate magnitude earthquakes, including the 17 October and 20 October 2005 M 5.7 and M 5.9 earthquakes, respectively, together with the intense seismic activity observed adjacent to the southern part of the Gülbahçe Fault (Aktar 2007), caused damage to buildings in the region (Sözbilir et al. 2009). Detailed information and knowledge on these and many other events in terms of magnitude and the geometry and kinematics of the activated faults can be found in Tan et al. (2014), Chatzipetros et al. (2013), Ambraseys (2009), Kouskouna and Sakkas (2013).

Geodetic dataset
We used InSAR data acquired by the Sentinel-1 satellites in TOPS (Terrain Observation by Progressive Scans) mode and continuous GPS observations to measure the ground displacement due to the 30 October 2020 M W 7.0 Samos earthquake. In particular, we used Sentinel-1 ascending interferogram made with the 23 October and 10 November 2020 scenes acquired on track 029, and the descending interferogram made with the 24 October and 11 November 2020 scenes acquired on track 136 (Figs. 2 and S1), because these 18-days pairs provide better quality interferograms than the previous pairs. Other interferograms of different tracks (i.e. ascending track 131) or with different temporal acquisition scenes were also processed, but the extracted information was of a lower quality.
The interferograms were produced using the Sentinel Application Platform SNAP software developed by the European Space Agency. The topographic phase component was removed exploiting the DEM (Digital Elevation Model) provided by Shuttle Radar Topography Mission 1 arcsec (Jarvis et al. 2008). We enhanced the signal to noise ratio of the resulting interferograms by applying the Goldstein ltering algorithm (Goldstein and Werner 1998) with a coherence threshold of 0.3. Both the interferograms show a clear fringe pattern of the coseismic displacements (Fig. S1), corresponding to ground deformation onshore on the west of Samos.
The interferograms were then unwrapped using Statistical-Cost Network-Flow Algorithm for Phase Unwrapping (SNAPHU) (Chen and Zebker 2001) and nally geocoded to obtain the ground deformation maps. In the displacements pattern, we observed a major lobe of increasing LOS (Line-Of-Sight) displacements (with maximum value of about 10-15 cm) towards the satellite in the NW Samos, except from a small sector of the NE coast of Samos, where the motion is away from the satellite (Figs. 2, S2 and S3).
In addition, we also used the GPS measurements of coseismic displacements from Ganas et al. (2020), which reveal a general relative normal motion E-W to ESE-WSW oriented between Samos and Turkey's west coast (Fig. 3), in agreement with the moment tensor solutions of both regional and teleseismic data (e.g. CSEM-EMSC). The largest horizontal and vertical static offsets were measured at SAMO site, where about 37 cm of movement towards S-SW and an uplift of about 9 cm were observed (Fig. 3), in agreement with InSAR data.

Inversion for fault geometry and coseismic slip
To image the fault geometry and slip distribution of the 2020 Samos mainshock, we simply inverted the static deformation for distributed slip on a number of different fault geometries, corresponding to a NNE-dipping fault rupture scenario. We adopted the same inversion scheme as in Cheloni et al. (2019), which used rectangular dislocations embedded in an elastic, homogeneous and isotropic half-space (Okada 1985). Before modelling, the InSAR interferograms were downsampled to reduce the number of data points from several millions to a set of about some hundreds of points, using a resolution-based resampling algorithm (Lohman and Simons 2005;Figs. S2 and S3). The geodetic data is thus inverted for slip magnitude on each fault patch of constant rake and size (2.5 km ⨯ 2.5 km), inferring the optimal fault geometry iterating by grid searching over, locations, strikes, dips and rake angles of the fault plane jointly inverting InSAR LOS and GPS displacements, while using relative weights to properly combine the different data sets (Fig. S4). We also used positivity constraints in the inversions using a bounded-values, weighted leastsquares algorithm (Stark and Parker 1995) to impose positivity constraints on the estimated slip and we regularized the linear inversion by applying spatial smoothing (Fig. S5). Additional terms consisting of a linear ramp for each InSAR interferogram are also included in the inversion to minimize the effect on the solution of any residual long-wavelength orbital signal in InSAR images.
The inversion of the geodetic displacements satisfactorily reproduces both the observed InSAR LOS displacement maps (Fig. 2) and the GPS offsets (Fig. 3). The best-tting fault model is obtained with N285°ESE-WNW striking and 38°NNE dipping normal (rake angle equal to -80°) fault plane in good agreement with focal solutions and with the distribution of aftershocks (Fig. 4). The coseismic slip distribution model on the preferred fault plane (50 km ⨯ 25 km; including 20 ⨯ 10 sub-faults) shows a single major asperity with peak slip of about 2.5-3 m, located WNW respect to the epicentre, for an along-strike length of about 40 km. Although a detailed ne slip distribution cannot be obtained from the available geodetic dataset (due to the lack of measurements at sea), nevertheless, the obtained slip distribution is compatible with aftershocks location, which is characterized by the almost complete absence of seismic events in the area where we retrieved most of the slip, suggesting therefore that the largest slip may have occurred just in this region. The resulting seismic moment is 3.3 ⨯ 10 19 Nm, corresponding to a M W 7.0 earthquake, in agreement with seismological results. Our retrieved slip distribution located westward of the epicenter, implies therefore a predominantly WNW rupture directivity. Our preferred fault plane geometry and the dimension and positions of the main slip asperity agree also with the recent study of Ganas et al. (2020), who suggest that rupture during the Samos earthquake occurred on a 37 km NNE dipping normal fault located offshore of the north coast of Samos. These source  Table 1 the strong ground motion stations are given, along with the station name, location, site class (according to the EC8 classi cation), sourceto-site distances, and the observed peak ground accelerations and peak ground velocities at three components.

Generation of synthetic strong ground motions
In this section, to better understand the ground motion characteristics and the physical processes and parameters underlying seismic wave propagation of the 30 October 2020 Samos earthquake, we composed a stochastic ground motion model to generate predictions of peak ground acceleration and velocity, and 5%-damped response pseudospectral acceleration for the M W 7.0 mainshock. These simulations account for source, path and site effects related to the fault rupture, the seismic wave propagation as well as the complex site responses in the case of the sedimentary basin, providing a spatial variation of ground motion, which may be essential to assess the seismic hazard and reduce the seismic risk in the populated urban areas.
The strong ground motion synthetics were completed by using a stochastic nite-fault simulation model, based on dynamic corner frequency (Motazedian and Atkinson 2005;Boore 2009). Stochastic simulation methods contemplate the physics-based rupture process and call for well-de ned source and slip distribution, together with path, and local site conditions within the region of interest.
The nite-fault source model with slip distribution (Fig. 4a) determined in the present study is intended as an input to the ground motion calculations (see previous section and Table 2). The region speci c crustal S waves attenuation parameters are selected as characterized by Akinci et al. (2012), using several hundred regional earthquakes (M 2.5-5.8) recorded during the Western Anatolia Seismic Recording Experiment (WASRE) between November 2002 and October 2003 (Akyol et al. 2006;Zhu et al. 2006). The anelastic attenuation for the entire region is given as the quality factor, Q s (f) = 180f 0.55 and the geometrical spreading coe cient, g(r), occurring at a hypocentral distances of 20 and 40 km which changes from a body-wave-like function g (r) ∝ r −1.0 to g (r) ∝ r −0.8 and g (r) ∝ r −0.5 expected for surface waves for distances larger than 100 km. These results are quite similar to that obtained by Kurtulmus and Akyol (2013) as Q s (f) = 190f 0.64 with a rapid decay of g(r) ∝ r −1.0 both for short and long distance ranges in the same region. The stress drop parameter was determined from the residuals between the observed and the simulated peak ground accelerations which were calculated using different Δσ, ranging from 50 to 200 bars (Fig. S6). As observed in Figure S6, a stress drop value of Δσ = 80 bars was considered as the most reasonable value to provide a satisfactory comparison, with lower residuals. This value is also in good agreement with those determined with earthquakes of similar magnitude in Turkey.
The level of the acceleration spectrum at high frequencies is estimated between 8-20 MPa by Akinci et al. (2006Akinci et al. ( , 2014, Cheloni and Akinci (2020) and Malagnini et al. (2010) for the 1999 Kocaeli M w 7.2 and 2020 Elazig (Doganbey) M w 6.8 earthquakes with strike-slip faulting, respectively, and for the 2011 Van Lake M w 7.1 earthquake with reverse faulting in Turkey.
The parameters employed to quantify the impact of the shallower sedimentary layers to the seismic ground motion are the soil ampli cation factor D(f) and the kappa parameter ( ). The former is characterized as an exponential decay to produce the diminution factor, e -πfκ (Anderson and Hough 1984). In the present study, we used frequency-dependent generic site ampli cations and kappa parameters for different site classes from Boore and Joyner (1997) and Pischiutta et al. (2020). The parameter was set as 0.035 s for the site class A (average V s30 1150 m/s, Pischiutta et al. 2020) and B (V s30 between 620 m/s, Boore and Joyner 1997), while for the site class C and D (average V s30 255 m/s and 190 m/s), it was favored as 0.055 s.

Page 7/19
The physical parameters that characterize the earthquake source rupture and the seismic wave propagation together with the site related parameters considered to generate synthetic ground motions are listed in Table 2, while the generic site ampli cations are given in Table S1. We note that the adopted generic site ampli cations, calculated through the quarter wavelength approach (Joyner and Fumal 1985;Boore and Joyne, 1997) using the averaged velocity over the uppermost 30 m of the soil, and are characterized by velocity gradients which are not sensitive to impedance contrasts between the layers (Boore et al. 1994(Boore et al. , 2011. In the following section, to give more attention to this topic, we evaluated local site responses in terms of spectral ampli cation by using several different techniques, particularly for sites located in the Karsiyaka and Bayrakli districts where heavy damage was observed during the 30 October 2020 earthquake (Erdik et al. 2020).

Spatial distribution of horizontal peak ground accelerations and velocities
In order to investigate the spatial distribution of the ground motions we simulated the M W 7.0 Samos earthquake at 1055 virtual stations distributed on a regular grid with 5-km spacing together for the sites where strong ground motion stations were actually located. In Figures 5a and b we presented the spatial distribution of the PGA and PGV values obtained from the simulations using the main fault rupture and the spectral parameters (Table 2) within the selected area using a uniform rock site ampli cation with V s30 = 760 m/s referring to the engineering bedrock. The spatial distribution of simulated PGA and PGV values for the 30 October 2020 mainshock holds explicit effects of the earthquake source rupture since the soil condition is assumed to be uniform across the region. The strongest ground shaking is observed in areas close to the main asperity with a larger slip located roughly in the central part of the fault and calculated around 0.5 g in PGA and 50 cm/s in PGV in the northwest of Samos. Simulations at the nearest stations, GMLD at R jb = 11 km and 0905 at R jb =21 km resulted in PGA of 0.23 g, and 0.15 g while observed values range from 0.19-0.27 g, and 0.15-0.18 g, respectively. There is good consistency found between observed and predicted ground motions close to the source rupture, which con rms the well constrained source parameters (fault geometry and slip distribution) described in the present study. Simulated ground motions are then somewhat validated by the observed ground motion parameters and compared with two selected ground motion prediction equations (GMPEs), in terms of PGA and PGV values, for epicentral distances up to 250 km.

Comparisons between observed and simulated time histories and Fourier Spectra
In this section to demonstrate the e cacy of our ground motion model both in time and frequency domain, we compared some selected recordings of acceleration and velocity time histories and their Fourier Amplitude spectra with synthetic ones. In our simulations we examined different geological conditions for each strong motion station according to the network classi cation (Table 1). As shown in Figure 6a the synthetic time series are in rather good agreement with those of the selected recordings considering only generic site ampli cation factors were used. Although the synthetic spectra match quite well with those of the recorded data, especially at most of the stations located on rock and stiff soil, reproducing acceptable amplitudes and frequency content, they underestimate spectral amplitudes between 0.5 and 2.0 Hz for those recordings observed at stations located in the Bornova Basin, Izmir (Fig. 6b). Stations 3519, 3513, 3521 and 3518 are located on alluvial sites close to the shoreline of the Bayrakli and the Karsiyaka district whereas stations 3514, 3524, 3521 are deployed on the stiff soil site in Izmir. As can be observed in Figure 6a the stations 3519, 3513, 3521 and 3518 have longer periods and higher amplitudes due to their local site characteristics. In fact, a considerable part of the metropolitan area of İzmir is located on sedimentary alluvial deposits. Furthermore, the notable uvial degradation takes place in the delta of the Gediz River in the northern section of the İzmir Bay (Uzel et al. 2013). Therefore, the use of the generic site ampli cations may not be enough to properly quantify the site ampli cation factors since they disregard speci c geological features of the soil stratigraphy such as the impedance contrasts in sedimentary basins and the soil depths. Although the generic soil ampli cations have been used commonly in many applications, Boore (2013) has exposed their limitations inferred by the method, the quarter wavelength approach, which smooths so that underestimates of the fundamental resonant peaks generated by the presence of the strong bedrock/soil seismic impedance contrasts. In addition, several studies have also indicated the importance of the soil/bedrock impedance contrast, thickness of soil, and soil properties on characterizing the site Both the recorded and simulated peak accelerations and velocities at distances up to 250 km from the fault are compared with the two selected empirical GMPEs (Figs. 7a and b). We also gathered the simulation ground motion parameters performed for the generic soil site class at the 1055 virtual stations to those simulated for the 116 strong motion stations of the AFAD network on different geological conditions. Generally, the ground motions obtained from simulations agree well with the observed ground motion parameters although the recordings present larger nonconformity over greater distances R jb > 100 km. The observed PGAs and PGVs are mostly in the range of the empirical relationships for the BSSA14 and the AC10 yet, the BSSA14 model better predicts the observed peak accelerations and velocities at distances over 100 km with respect to the AC10. At the closest R jb distances (up to 20 km), simulated PGA and PGV values present a large variability and have a more scattered data distribution compared to the variations provided by the empirical relationships with one standard deviation. Because the simulated PGAs and PGVs acquired from normal faulting mechanism represent large variation between the hanging-wall and foot-wall section of the fault plane. Maximum acceleration values at the closest distances range between 0.5 and 0.8 g on the hanging-wall of the fault plane.
Moreover, we observed a clear trend of the observed data due to the EC8 soil type classi cation up to distances of 100 km. The consequences of site effect become more evident over longer periods on PGV parameters with respect to PGAs around the 50 km R jb distances. Most of the highest PGA and PGV values correspond to stations belonging to C site class located on the Bornova basin in Izmir and are remarkably underestimated by the GMPEs. The generic ampli cation factors, used in our simulations, were also not able to capture properly locally ampli ed ground motions at particular frequencies.

Site ampli cations
In order to completely understand the ground motion characteristics of the earthquake throughout synthetics, we investigated site characteristics of strong motion stations located on the shorelines of the city of Izmir. Particularly we examined the stations 3519, 3513, 3518, 3521 with site classi cation C and D, and the stations 3514, 3524, 3511 and 3520 with site classi cation A and B type. Site ampli cation factors are estimated over commonly used methods: 1) horizontal-to-vertical component spectral ratios, HVSR (e.g. Nakamura 1989;Lermo and Chavez-Garcia 1993); 2) standard spectral ratio, SSR (Borcherdt 1970); and 3) the propagator-matrix method, PMM, of Haskell (1960) and the quarter wavelength, QWL techniques (Joyner and Fumal 1985;Boore and Joyner 1997).
To determine the frequency-dependent ampli cations, rst of all, the ratio of the horizontal-to-vertical (H/V) components of ground motion are calculated between the horizontal Fourier spectrum (N-S and E-W spectra) and the vertical one over the S-wave; starting 1.0 s before the manually picked S-wave arrivals and over around 15-20 s time windows.
The SSR technique is applied to the recordings of the mainshock as a secondary approach to investigate the soil ampli cations at the selected sites. The spectral ratio is obtained by dividing the Fourier spectrum of the acceleration for the S wave at the selected stations by the spectrum of the S wave at the reference stations on the hard-rock sites. The station 3514 is chosen as a reference station being deployed on a geologic bedrock (V s30 = 836 m/s).
In order to interpret the site ampli cations based on shear wave velocity, density and attenuation as a function of depth, as a tertiary approach we computed the theoretical ampli cations for three sites using the propagator-matrix method of Haskell (1960). To do so we used the site_amp and nrattle computer program, a modi ed version of C. Mueller's program rattle by R. B. Herrmann (part of the SMSIM computer codes of Boore 2003). We prototyped shear-wave velocities and attenuation as a function of depth from those pro les provided by Pamuk et al. (2018Pamuk et al. ( , 2018b and computed ampli cations in the frequency-domain. In these velocity models, engineering bedrock (V s > 760 m/s) depth changes from 200 to 400 m in the Bornova basin while the seismic bedrock (V s > 3000 m/s) ranges between 1200 to 1400 m. The velocity pro les S2, S3, S5 and the deeper one S4, with greater proximity to the 3519 station, are adopted from Pamuk et al. (2018). Figure 8a shows adopted 1D prototyped V s -depth pro les (see Figure 5 of Pamuk et al. 2018) and Figure 8b shows ampli cations with a vertical incidence wave of geometrical effect calculated using the PPM (continuous lines) and QWL (dashed lines) methods. Attenuation effect at the closest sites is included using kappa parameters 0.055 s. As can be seen clearly in Figure 8b, the QWL method underestimates the resonance peaks generated by the large impedance contrasts. The resonant frequencies are different based on the four models and the shear wave pro les considered for the site ampli cations. The site responses derived from the PPM are double those obtained from the QWL one. There is also clear evidence that the maximum peak value is shifted towards lower frequencies when the thickness of the layer increases. The S4 model with a strong impedance variation at roughly 120 m and 600 m presents larger peaks around 0.7 and 1.6 Hz yet the frequency of resonant peak decreases, 0.3 s, with increasing thickness, having a relevant impedance contrast around 450 m, in the case of the velocity model S3.
In Figures 9a and b we compare the resultant site ampli cations from HVSR (left six panels) and SSR methods (right six panels): it is evident the strong ampli cations at stations 3519, 3513, and 3521 with a clear peak at low frequencies between 0.3 and 1.5 Hz, while stations located on rock and stiff soil do not show any evidence for strong ampli cations at least for frequencies lower than 3-4 Hz. These clear and sharp peaks may imply the presence of rm velocity contrast at depth between unconsolidated deposits and underlying bedrocks. From Figure 9, we can conclude that the surface/bedrock ground motion felt at the station 3519 was 5-8 times larger in respect to that at station 3514 at the fundamental period of 0.7 s. Ampli cation values obtained from the HVSR method are slightly greater, roughly at low frequencies, than the SSR results. Site ampli cations from theoretical experiments of the quarter-wavelength approximation con rm the suggestion that the maximum amplitude of spectral ratios is sensitive to relevant impedance contrasts.
Although the three methods were able to closely reveal the fundamental frequencies, all being ranged between 0.3 and 2.0 Hz, they provide relatively different results concerning the amplitude of ratios. This could be due to several factors such as the ampli ed vertical component by the multiple re ected and converted waves by basin geometry or the selected reference station which does not have a perfectly at spectrum and may be explored in future studies. Finally, in Figure 10 we demonstrated the effect of site ampli cations obtained from the PPM and the QWL approximations throughout our simulations. We computed spectral accelerations at three sites 3519, 3513 and 3520, located on the soft soils and compared those with the observed spectral accelerations. Simulated spectral accelerations calculated using the site response spectra obtained from the PPM t well with the observed data reproducing resonant peaks and corresponding amplitudes. However, these resonance peaks are always underestimated by the QWL ampli cations since the method is insensitive to the important impedance contrasts between layers and the velocities are averaged over depth. In the Supplementary Material we further provided some synthetic time series and their complementary Fourier spectrum calculated using site responses with the resonance peaks (Fig. S7).

Acceleration response spectra and observed damage distribution in the Metropolitan area of Izmir
In general, the damage potential of the ground motions on the structures are measured by the amplitudes and shape of the response spectrum. Buildings are designed withstand forces no higher than these spectral values. To examine the effect of the earthquake on the Izmir Metropolitan area, we compared code compiled spectrum with the spectra computed from the recorded ground motions. Figure 11 shows the spectral accelerations (response spectra, 5% damping) computed from the NS and EW components recorded at eleven stations, with ve and six being located on soft and rocky soil sites within the study area. In the same gure, previous (TSC-2007)  values at 1s period for soft and hard soil stations are 0.37 g and 0.10 g, respectively) are far below the design-level spectra (0.57 g and 0.22 g for soft and hard soils, respectively) of the recent earthquake building codes. Therefore, buildings would have neither suffered from any excessive damage, nor collapsed if they had been constructed according to the engineering practice and design guidelines.
Following the 30 October 2020 Samos earthquake, the Ministry of Urbanization and Environment completed on-site assessment of the building inventory in Izmir to determine the condition of the structures and share the assessment reports on the website (MUE 2020) for possible reclamations by the property owners. The spatial representation of the distribution of the damaged buildings acquired from the reports is plotted in Figure 11 in order to correlate the soil condition and computed spectral values of the records. Two important morphological features are clearly observable, associated with the high bedrock outcrops and with the low-lying basin structure with thick sedimentary deposits. A considerable part of İzmir's metropolitan area is located on sedimentary alluvial deposits. Furthermore, the notable uvial degradation takes place in the delta of the Gediz River in the northern section of the Bay of Izmir. During the 30 October earthquake, site response played an important role in the ampli cation of ground motions on soft soil sites such as 3513, 3518, 3519, 3521 and 3522 sites (Fig. 12).
Izmir is one of Turkey's major cities, where migration in the last couple of decades increased signi cantly. Thus, rapid housing demand has been a critical issue in which the construction quality and proper engineering design has been compromised. Catastrophic results of the soil ampli cation and the inadequate construction techniques were the main issues in the Izmir Earthquake Master Plan (MMI 2000) where the areas have already been highlighted as having the potential for signi cant building damage (Fig. 13).
Similarities between the predicted and observed distribution maps reveal the necessity of considering the local site conditions and also housing demand in accordance with population growth must be explicitly included in the urbanization plans. On the other hand, the assessment reports and observations on the site indicate that the damaged buildings during the earthquake were constructed with low quality material and workmanship which are crucial defects of the construction sector in the developing countries. To avoid further damage, the Ministry of Environment and Urbanization decreed to demolish all buildings with major to moderate damage (as of December 2020, 652 and 778 major and moderate damaged buildings, respectively) as well as the surrounding buildings if constructed by the same contractors. The damage distribution due to the earthquake clearly reveals all engineering steps including soil examination, soil-structure interaction in terms of resonance frequency, code-complied design and construction must be substantially considered in urbanization policies.

Conclusion
In this study we provided a complete description of the strong ground motions induced by the 30 October 2020 M W 7.0 Samos (eastern Aegean Sea) earthquake taking into account three principal components that characterize the observed ground motions (earthquake source complexity, seismic wave propagation and site effects) through the earthquake simulations. Although we studied each of the physical parameters and the consequences on the predicted ground motion, particular attention was given to interpretation of the observed ground motions in terms of the site effects that played a particular role in the Bay of Izmir area.
With regards to the seismic source, the Samos earthquake was caused by the rupture of a NNE-dipping normal fault located offshore of north coast of Samos. The geometry and location of our preferred model is therefore consistent with the rupture of the offshore North Samos Fault during the main seismic event, which was believed to be active mainly on the basis of its effect on the sea oor (Chatzipretros et al. 2013). The slip distribution obtained from the inversion of geodetic data shows a main asperity (peak slip of about 2-5-3 m), located WNW from the epicenter, suggesting therefore a westward directivity during the mainshock. The seismic moment release is 3.3 ⨯ 10 19 Nm (equivalent to a M w 7.0), in agreement with magnitude estimates provided by different research Institutes.
Our results indicate that the detailed knowledge of the regional crustal and near-site speci c attenuations as well as the earthquake source rupture characteristics, are essential for a complete understanding of the earthquake induced seismic hazard and to better predict the earthquake ground motions in a region. Region speci c frequency-dependent attenuation together with the geometrical-spreading coe cient as described by Akinci et al. (2013) are successfully used to simulate ground motions in the study area.
We demonstrated that the generic soil ampli cations used for the simulations were not adequate for the soil and the site conditions in Izmir Bay where simulations systematically underestimate the observed ground motions at the period range of 0.5 to 1.5 s. Stations located within the Bornova basin are characterized by the largest estimates of the ampli cation amplitudes at that particular period range that are commonly detected by the three different approximation methods (HVSR, SSR and PPM). Conversely, stations located on the rock units show the least ampli cation level, ranging between 2 and 4. We have obtained substantial improvements on our simulations by considering the strong resonance effects particularly at lower frequencies (f < 1.5 Hz).
Damage to the city of Izmir exhibits that small-to-moderate ground motions can be locally ampli ed modifying the intensity of ground shaking. Since these ampli ed waves may cause structural damage, a proper calibration of the site response is fundamental not only for ground motion simulations, but also for site classi cations throughout the Izmir Bay area. The peculiar resonant frequencies of ground shaking this need to be taken into consideration when undertaking the seismic-resistant design of civil engineering structures in the Izmir Metropolitan area, where the annual rate of construction of residential buildings has increased ve fold (from 1500s to 7500s per year) over the last 18 years (TSI 2021). Furthermore, active tectonic structures and historical seismicity of the area suggest that earthquakes with similar size may occur close to the city again in the future, and could produce much greater damage in the area.

Declarations
Zandieh A, Pezeshk S (2011) A Study of Horizontal-to-Vertical Component Spectral Ratio in the New Madrid Seismic Zone. Bulletin of the Seismological Society of America 101:287-296. doi: 10.1785/0120100120 Zhu L, Mitchell BJ, Akyol N, et al (2006)