Probabilistic and Scenario-Based Seismic Hazard Assessment on the Western Gulf of Corinth (Central Greece)

Abstract

The Gulf of Corinth (Central Greece) is one of the most rapidly extending rifts worldwide, with its western part being the most seismically active, hosting numerous strong (M ≥ 6.0) earthquakes that have caused significant damage. The main objective of this study was the evaluation of seismic hazard through a probabilistic and stochastic methodology. The implementation of three seismotectonic models in the form of area source zones via a logic tree framework revealed the expected level of peak ground acceleration and velocity for return periods of 475 and 950 years. Moreover, PGA values were obtained through the stochastic simulation of strong ground motion by adopting worst-case seismic scenarios of potential earthquake occurrences for known active faults in the area. Site-specific analysis of the most populated urban areas (Patras, Aigion, Nafpaktos) was performed by constructing uniform hazard spectra in terms of spectral acceleration. The relative contribution of each selected fault segment to the seismic hazard characterizing each site was evaluated through response spectra obtained for the adopted scenarios. Almost all parts of the study area were found to exceed the reference value proposed by the current Greek National Building Code; however, the three urban areas are covered by the Eurocode 8 regulations.

1. Introduction

The Gulf of Corinth, located in Central Greece, is a continental rift with an extension rate of 6–16 mm/yr [1], one of the highest values known worldwide for this type of tectonic structure. Its main axis is aligned in an approximately E–W direction, flanked by major normal faults at its southern and northern shoulders. Major earthquakes (M_w ≥ 6.0) have occurred in the past, both at the eastern [2] and western parts of the gulf. However, the Western Gulf of Corinth (WGoC) is the most active in terms of observed microseismicity [3,4], as it exhibits a higher extension rate—i.e., ~10.8 mm/yr near Aigion—than its eastern counterpart (~5.5 mm/yr) [5]. This has led to the WGoC being studied extensively during past decades, and it is closely monitored by the Corinth Rift Laboratory (CRL) local seismological network [6]. The CRL has also obtained the status of an international EPOS Near Fault Observatory (NFO), providing high-resolution multidisciplinary data and products [7,8,9].

The WGoC is bordered by major north-dipping normal faults in the south, with measured dipping angles on the surface ranging from 50° (e.g., Psathopyrgos fault [10]) to ~60° (e.g., Aigion fault [11]) and up to 65–70° (e.g., Lakka fault [12]), with average slip rates in the Late Quaternary of the order of 2.4–5.0 mm/yr for the Aigion and East and West Helike faults and 2.0–3.5mm/yr for the Psathopyrgos fault [10]. On the northern coast, steep south-dipping antithetic normal faults have been mapped, such as the Marathias fault, dipping at 55° [13], and the Trizonia fault, dipping at 64–72°, the latter with an average slip rate of 0.36–0.44 mm/yr over the last ~130 ka [14]. Offshore E–W-trending normal faults, along with smaller structures, oblique to the main axis of the rift with significant strike-slip components, have also been documented [10,14,15]. Average slip rates during the Quaternary for major offshore faults in the WGoC—i.e., the North and South Eratini and West and East Channel faults—are estimated to be between 0.4 mm/yr and over 1.4 mm/yr [10].

Seismicity in the WGoC is mainly concentrated at focal depths between 5 and 10 km [6], with the vast majority of fault plane solutions indicating dominant E–W normal faulting [4,16]. However, oblique-normal slip has also been reported, usually related to seismic swarms [7,17] or to the activation of older structures [18]; i.e., the External Hellenides, which crosscut the gulf in a roughly N–S direction. Although such faults are not optimally oriented to the regional crustal stress field [19], slip can be facilitated by the intrusion of high pore-pressure fluids in the fault network, which play a significant role in the evolution of swarms in the area [20,21]. Towards the western edge of the rift, the crustal stress field changes [19], favoring a dextral strike-slip regime along the SW–NE-trending Rion–Patras fault. The relocated seismicity presented in Figure 1 shows that the bulk of activity mainly occurs along the rift axis, between Trizonia Island and Aigion, with hypocenters deepening towards the north, developing a north-dipping low-angle detachment zone in which the major faults of the WGoC are rooting [22]. Several earthquakes have been located at intermediate depths of 50–60 km, with resolved focal mechanisms representing reverse faulting (Figure 1). These are indicative of seismic activity related to the oceanic slab at the northwestern end of the Hellenic subduction zone [23], with ascending fluids due to its dehydration possibly playing a role in the triggering of earthquake swarms at shallow depths [22,24].

Despite the constant presence of microseismicity, certain parts of major faults in the WGoC appear to be locked, accumulating stress that is released in major events. The last destructive earthquake in the area was the 15 June 1995 M_s = 6.2 event, which severely damaged the town of Aigion [29]. The source of this earthquake was a north-dipping, low-angle (33°) normal fault, with more recent data excluding its association with the East Helike fault [22]. The rupture surface of the 1995 event appears to be locked, with the observed microseismicity stopping abruptly at the eastern end of the study area (Figure 1). Other major earthquakes have also occurred in the past at the eastern end of the WGoC, with epicenters towards the northern coasts. Significant earthquakes have struck the WGoC in the historical era (i.e., before 1900). One of the more notable was the 373 BCE earthquake, with an estimated minimum magnitude of M = 6.6 [36], which caused a tsunami that destroyed the ancient city of Helike [37]. An earthquake of similar macroseismic characteristics struck the same area on 26 December 1861 (M_w = 6.5 according to Boiselet [38]), triggering a tsunami and causing a strip of plain to submerge under the sea [37]. Both earthquakes are attributed to the north-dipping Helike normal fault, most likely its eastern branch [22]. Recently, the contemporary town of Helike was in the epicentral area of a swarm that started in May 2013, persisting for several months and involving a few earthquakes with magnitudes ranging from 3.3 to 3.7, but which was attributed to activity near the root of Pirgaki fault [17,39]. Other historical earthquakes that struck the town of Aigion occurred in 1748, 1817, and 1888, with estimated magnitudes of 6.3–6.6 [33]. The area of Patras was affected by historical earthquakes in 1785, 1804, and 1806, with magnitudes between 6.1 and 6.4, while the town of Nafpaktos was struck by earthquakes in 1462, 1703, 1714, 1756, and 1831, with magnitudes in the range from 5.9 to 6.5 [33]. In the instrumental era (1900 to today), before the 15 June 1995 earthquake, the previous major event with a magnitude ~6 occurred in the WGoC on 30 May 1909, while intermediate-depth events with magnitudes of 6.4 and 5.9 occurred in 1965 and 1993, respectively; the epicenters of these events were located at the eastern end of the study area.

Although no M_w ≥ 6.0 earthquake has struck the WGoC since 1995, moderate magnitude events (M_w ≈ 5) have occurred in recent years. On 14 July 1993, an M_w = 5.5 event with a strike-slip focal mechanism [30,40] struck near Patras, causing considerable damage. Although the locally mapped structures trend SW–NE, it has been suggested, based on the aftershock distribution, that the earthquake ruptured a NW–SE sinistral-slip fault [41]. A notable case is the earthquake doublet of January 2010 near Efpalio, with an M_w = 5.3 event being followed by an M_w = 5.2 four days later, both related to blind E–W-trending normal faults [32,42,43]. This activity likely played a role in the triggering of another M_w = 5.1 earthquake on 7 August 2011 to the west, near Nafpaktos, characterized by a similar normal faulting style. Later, in 2013–2014, the WGoC presented strong signs of microseismic activity, with an earthquake swarm initiating in September 2013 at the offshore region between Nafpaktos and Psathopyrgos that slowly migrated eastwards, triggering earthquake clusters and culminating in an M_w = 5.0 event, offshore of Aigion, on 7 November 2014 [21,44]. A second seismic swarm excitation, characterized by bilateral spatiotemporal migration in the offshore area between Nafpaktos and Psathopyrgos, involved an M_w = 4.9 earthquake on 21 September 2014, activating a patch related to the westernmost edge of the Psathopyrgos fault, near its junction with the Rion–Patras fault zone [3]. The most recent significant activity in the WGoC was the 2020–2021 seismic sequence [7,45,46], which started on 23 December 2020 with a moderate event near Marathias and evolved in three stages, first triggering earthquake clusters towards Trizonia Island involving some strike-slip events and later producing an M_w = 5.3 earthquake offshore of Psathopyrgos that exhibited peculiarities in terms of its complicated rupture characteristics [47].

In this study, we reassess the seismic hazard in the WGoC utilizing a well-established methodology and incorporating recent data. The term “seismic hazard” describes the expected ground motion level for potential earthquake occurrences within a study area. The methodologies used to evaluate seismic hazard are grouped into two major categories: probabilistic and scenario-based. The first exploits data from historical and instrumental earthquake catalogues and integrates a seismotectonic model for its application. Subsequently, given an earthquake occurrence model, results for the maximum expected ground motion for selected return periods are generated. The second methodology is based on individual earthquake scenarios, with which synthetic waveforms are produced and analyzed. Strong ground motion may be a catalyst for other secondary hazards, the most important of which are rock falls, landslides, and liquefaction phenomena. Therefore, the visualization of the spatial distribution of peak ground motion parameters can be of great importance for scientists and decision makers. The advantage that the study area offers is the wealth of earthquake data. The latter is critical because seismic hazard can be assessed without adopting algorithms that cope with this difficulty, such as the one proposed by Meirova et al. [48]. A plethora of seismic hazard studies have been conducted for the Greek territory, either as a whole or for specific areas, such as [49,50,51,52,53,54,55,56,57,58]. The applied methods for the assessment of seismic hazard are described in detail in Section 2.

Patras, the third largest city in Greece and one of its largest harbors, is situated at the western edge of the study area, on the southeastern shores of the Patras Gulf. The nearby town of Rion hosts important infrastructure, such as the University of Patras and the University General Hospital of Patras, while the Rion–Antirrion bridge connects Peloponnese with Central Greece. Other important towns on the shores of the WGoC include Aigion and Nafpaktos. As the seismic risk strongly depends on the exposure to seismically prone areas, the aforementioned city, towns, and infrastructure may suffer increased economic losses from potentially destructive earthquakes, both in terms of human lives and structural damage. As a consequence, the proper assessment of seismic hazard in the WGoC area, both probabilistic and with specific earthquake scenarios, is required for the subsequent estimation of seismic risk and consideration of means for its mitigation.

2. Materials and Methods

Seismic hazard in the study area was evaluated using two approaches: (a) a Probabilistic Seismic Hazard Assessment (PSHA) and (b) a stochastic approach to worst-case seismic scenarios for known active faults. PSHA remains the primary framework for assessing seismic hazard and is based on prior knowledge of seismicity at a given place. On the other hand, during past decades, finite-fault stochastic ground motion simulations have been proven to be a powerful tool to reliably estimate strong ground motion parameters, such as the Peak Ground Acceleration (PGA) and Peak Ground Velocity (PGV). The combination of these methodologies can provide a holistic evaluation of seismic hazard, which in turn can contribute significantly to the assessment of seismic risk and its mitigation at a given area.

2.1. PSHA

Concerning PSHA, the approach proposed by Cornell and McGuire [59,60] was followed, utilizing the instrumental earthquake catalogue from Makropoulos et al. [25]. Given that this catalogue ends in 2009, it was deemed necessary to extend the period up to 2019 in a homogenous manner, meaning that only reviewed events from the International Seismological Centre (ISC) earthquake catalogue were taken into consideration. Therefore, we refrained from expanding the database for post-2019 events in order to retain homogeneity. It should be noted that only few M_w ≥ 4 magnitude earthquakes have been recorded in the study area since 2019, the largest being the M_w = 5.3 earthquake of 17 February 2021 [7,47]. Thus, we considered that the annual rate of exceedance of the magnitude of completeness would not be altered significantly and would not impact the results of PSHA. According to the Cornell–McGuire approach to PSHA, the results follow a normal distribution, being essentially time-independent. Given that there are both aftershocks and foreshocks included in the utilized earthquake catalogue, a declustering procedure was conducted to retain only the mainshocks. However, it must be noted that ground motion exceedances can be caused randomly by any earthquake occurrence and a significant amount of data could be omitted by applying a declustering procedure. This would have an impact on the obtained results because annual rates of exceedance for the magnitude of completeness would be lower, resulting to the underestimation of the maximum expected ground motion parameters [58,61]. The latter is strengthening the implementation of a non–declustered earthquake catalogue in a Poissonian process, as ground motion at a site caused either by foreshocks or aftershocks may exceed a certain level [61].

Three seismotectonic models were considered herein; namely, the Euro-Mediterranean Seismic Hazard Model 2013 (ESHM13) [62,63], the updated Euro-Mediterranean Seismic Hazard Model 2020 (ESHM20) [64], and the model from Vamvakaris et al. [65] proposed for the Greek territory. The boundaries of the Area Source Zones are selected to include areas with similar seismological and tectonic characteristics, avoiding dividing fault systems, whenever possible. Each model divides the study area into polygons with common seismotectonic characteristics (Area Source Zones, ASZ). Individual faults were not implemented in the PSHA, as a significant part of the total seismicity of the WGoC is related to offshore faults for which certain characteristics are not known; i.e., mean slip rates. The decision to employ more than one seismotectonic model was made in order to acquire differently parameterized earthquake occurrence models so that the epistemic uncertainties would be reduced in a qualitative manner.

The Modified Gutenberg–Richter (MG-R) earthquake occurrence model was applied to characterize the seismic potential of every area source zone for each seismotectonic model [66]. The MG-R was parameterized by importing the following seismicity data for each zone: (a) the b-value, (b) the threshold magnitude (M₀; also considered as the magnitude of completeness (M_c)), (c) the average annual exceedance rate of M_c (λ(M_c)), and (d) the maximum expected magnitude (M_u). The M_c and b-values were estimated for each zone using the maximum curvature (MAXC) method from Wiemer and Wyss [67] and the maximum likelihood function from Aki [68], respectively, with ZMAP software [69]. This method has been proved to be stable, even in cases when a small number of earthquakes within each zone are used [70,71]. The λ(M_c) was calculated through the analysis of the earthquake catalogue and M_u was estimated via the Robson–Whitlock–Cooke (R-W-C) procedure [72,73]. Regarding the geometry data, the depth of each seismic zone was estimated through depth histograms constructed using the focal depths reported in the earthquake catalogue. It must be noted that ESHM13, ESHM20 and VAM16 provide values concerning the aforementioned parameters, each using a different earthquake catalogue. However, in the present study the b-value, M_c, λ(M_c) and M_u were recomputed for each ASZ, im-plementing the updated earthquake catalogue of Makropoulos et al. [25] to retain ho-mogeneity regarding the seismicity and geometry data. Consequently, only the ASZ were adopted from the three zonation models. The computed seismicity and geometry data are depicted in Table S1.

Ground Motion Prediction Equations (GMPEs) proposed for the Greek territory were adopted for the calculation of the maximum expected ground motion in terms of peak ground acceleration and velocity (PGA, PGV). In particular, for PGA, the GMPEs from [74,75,76,77,78] were employed. Sakkas [77] has not proposed a GMPE for PGV estimation. Furthermore, the model from Skarlatoudis et al. [75] was replaced with an upgraded GMPE for PGV, which was taken into account. As area source zones were herein adopted, it was only feasible to use epicentral distances. As a result, more recent GMPEs [79] that utilize different distance metrics were not considered. In addition to PGA and PGV, Spectral acceleration (S_a) values for different periods (T) were calculated for the three most important urban areas of the WGoC—namely, Patras, Nafpaktos, and Aigion—using the GMPE from Danciu and Tselentis [76]. All GMPEs (except for the one from Margaris et al. [74]) include a term related to the type of focal mechanism, taking a value of 0 or 1 for normal and strike-slip/thrust ruptures, respectively. As a seismic zone could potentially include any type of focal mechanism, it was deemed necessary to compute their respective participation rates. Thus, each GMPE was calibrated with the exact percentage of focal mechanism types in each zone, generating weight-specific GMPEs (hereafter, w-sGMPEs). The final results (PGA, PGV for return periods of 475 and 950 years, and S_a for a return period of 475 years) were computed via an equal-weighted logic tree, where each secondary branch was a w-sGMPE (except [74]) and each primary branch was a different seismotectonic model (Figure 2). The reason for assigning equal weights to the primary branches was that none of the three zonation models outweighs the others. There are uncertainties that would be increased if a higher weighting factor would be assigned to the ASZ of ESHM13, ESHM20 or VAM16. For example, there are zones including both the western and eastern Gulf of Corinth in ESHM13 and ESHM20 (zones 13 and 1 in Figures S1a and S1b, respectively). In addition, certain small zones of the VAM16 model (zones 5, 13 in Figure S1c), result in an insufficient number of earthquakes, not adequate to reliably compute the seismicity parameters. The software used for PSHA was R-CRISIS and, specifically, its newest version (V20.0) [66].

2.2. Seismic Scenarios

We compiled a set of seismic scenarios to assess the seismic hazard in the broader WGoC area in terms of maximum expected strong ground motion from specific seismic sources. These scenarios were generated through stochastic simulations for a predicted maximum magnitude (M_max) for well-studied known faults in the area. Simulations were performed using a stochastic finite-fault model based on a dynamic corner frequency approach with the EXSIM code [80,81]. The modeling strategy was based upon the discretization of the earthquake fault plane into smaller subfaults, each of which was considered as a potential earthquake source. The point-source stochastic method [82] was employed to generate synthetic time series for each subfault. The summation of the individual contributions of each subfault, along with a suitable time delay, led to the final ground motion parameters at the sites of interest. This approach has been widely implemented worldwide [83], as well as in Greece [84,85,86,87,88].

The anticipated strong ground motion at a given site is a result of a complex physical process that includes the relative contribution of source, path, and local site effects. Source effects describe the characteristics of the accumulated strain release from the fault, which, when released, results in the generation of an earthquake. These include, among others, the source dimensions (length, width), the moment magnitude (M_w), the slip distribution for the causative fault, and the stress parameter (Δσ). After the earthquake nucleation and rupture, seismic waves travel through the Earth’s crust; therefore, path properties must also be taken into account. Their propagation is most strongly affected by two path parameters; i.e., the geometrical spreading and the anelastic attenuation, which is controlled by the quality factor (Q). Ultimately, the local site properties play an important role in the resulting surface ground motions, given that the impact of the shallower layers may lead to significant amplification of the seismic waves. These effects are treated by the EXSIM code using the kappa (κ₀) parameter [89] and user-defined soil amplification factors.

As a first step to obtain high-resolution model parameters for the desired earthquake scenarios, we performed a stochastic simulation of the most recent strong (M_w ≥ 6.0) earthquake in the study area; namely, the 1995 Aigion M_S = 6.2 (M_w = 6.5 according to the Global Centroid Moment Tensor (CMT) project) mainshock [29]. Modeling of past earthquakes in the study area can significantly help in constraining the path and site components of strong ground motion through comparison with GMPEs, as well as with real strong motion data. In the case of the Aigion 1995 mainshock, 17 recordings from seismic stations up to an epicentral distance of 140 km are available, allowing the calibration of the synthetic results through an iterative procedure to achieve the best fit. The final model parameters are summarized in Table 1. The causative fault’s dimensions and geometry (strike and dip) were adopted following [29] and the upper edge of the fault plane was set according to the model from Console et al. [90], which is in compliance with the geometry of the seismogenic layer of the study area. The slip distribution onto the fault plane was modeled through a random slip pattern, given that detailed finite-fault slip inversions are not available for this case. This approach has been also followed by other authors for this mainshock [86,88]. Finite-fault discretization into smaller subfaults was performed using the empirical relationship proposed by Beresnev and Atkinson [91] (Equation (1)):

logΔl = 0.4M − 2,

(1)

where Δl is the length of each subfault and M is the moment magnitude of the mainshock.

The Δσ is a parameter closely related to the actual stress drop and slip velocity due to an earthquake occurrence, but it does not include the natural context of stress drop [92]. It is generally considered to weakly scale with moment magnitude, but in a more regional extent their interconnection appears to be more profound [93]. It was determined by employing an iterative procedure, comparing the synthetic ground motions with recorded ones, as well as with estimated peak values from well-established GMPEs [94,95] appropriate for the study area. The Δσ value of 56 bars, which is routinely used as a mean value for earthquakes in Greece [96,97], was used initially and, afterwards, various other values were tested to find the best fit. Ultimately, Δσ = 30 bars was deemed appropriate and was adopted in the simulations. This deviation from the mean value was foreseen, given that low stress drop has been documented for this earthquake [98], as well as for the study area, especially in comparison with the eastern part of the gulf where stress drop values appear to be larger [99]. Regarding the path effects, the geometrical spreading model, as described by Atkinson [100], was adopted, which divides the slope of the attenuation relation in three distance intervals (<70 km, 70–130 km, and >130 km), reflecting the dominant type of wave in the seismic signal. The anelastic attenuation, on the other hand, is described by the quality factor proposed by Hatzidimitriou [101] for the broader Aegean region (Equation (2)):

Q(f) = 100f^0.8,

(2)

Lastly, to account for the local site conditions for synthetic ground motions, the average κ₀ value of 0.044 for class C (NEHRP) sites in Greece [102] was adopted, along with the corresponding amplification factors. The selection of class C soil characterization was based on local soil data (e.g., Vs30 profiles) for the major cities in the area and geology data from the locations of the available seismic stations. In order to calibrate and validate our results, the GMPEs of A14 and B14 were employed. Both were derived using strong motion datasets from Italy, Greece, and Turkey.

Table 1. Modeling parameters used for the stochastic ground motion simulation of the M_w = 6.5 Aigion 1995 mainshock performed with the EXSIM code.

Parameter	Value	References
Strike	277°	[29,90]
Dip	33°
Burial depth	5 km
Fault dimensions (Length × Width)	15 km × 9 km
Mw	6.5
Stress parameter (Δσ)	30 bars	This study
Kappa parameter (κ0)	0.044	[102]
Quality factor	100f0.8	[101]
$\mathrm{Geometrical} \mathrm{spreading} \mathrm{as} \mathrm{a} \mathrm{factor} \mathrm{of} \mathrm{distance} (R^n$)	N = −1.0, R < 70 km n = 0.0, 70 km ≤ R < 130 km n = −0.5, R ≥ 130 km	[100]
Site amplification	Empirical amplification factors from Klimis et al. [102]

Table 1. Modeling parameters used for the stochastic ground motion simulation of the M_w = 6.5 Aigion 1995 mainshock performed with the EXSIM code.

Parameter	Value	References
Strike	277°	[29,90]
Dip	33°
Burial depth	5 km
Fault dimensions (Length × Width)	15 km × 9 km
Mw	6.5
Stress parameter (Δσ)	30 bars	This study
Kappa parameter (κ0)	0.044	[102]
Quality factor	100f0.8	[101]
$\mathrm{Geometrical} \mathrm{spreading} \mathrm{as} \mathrm{a} \mathrm{factor} \mathrm{of} \mathrm{distance} (R^n$)	N = −1.0, R < 70 km n = 0.0, 70 km ≤ R < 130 km n = −0.5, R ≥ 130 km	[100]
Site amplification	Empirical amplification factors from Klimis et al. [102]

After validating the basic stochastic simulation parameters for the study area, three well-studied seismogenic faults (Psathopyrgos Fault (PT); Helike Fault (HF); Trizonia Fault (TF)) were selected to assess the worst-case earthquake scenarios (

Table 2. Stochastic modeling source parameters for the three selected seismic scenarios. Fault orientation and dimensions were adopted from various sources [6,38,90]. Associated past strong events were adopted from Boiselet [38].

Source Parameters	Psathopyrgos Fault (PF)	Helike Fault (HF)	Trizonia Fault (TF)
Strike	270°	281°	95°
Dip	40°	34°	65°
Dimensions (length × width)	16 km × 8 km	22 km × 12.5 km	10.5 km × 8.5 km
Mmax	6.3 ± 0.3	6.4 ± 0.3	6.0 ± 0.4
Associated strong events	1462 (Mw = 6.4) 1703 (Mw = 6.1) 1714 (Mw = 6.1) 1756 (Mw = 5.9)	373 B.C.E. (Mw ≈ 6.6) 61 C.E. (Mw = 6.3) 1758 (Mw = 5.9) 1817 (Mw = 6.4) 1861 (Mw = 6.5) 1888 (Mw = 6.3)	23 C.E. (Mw = 6.3)

, Figure 1). Given the limited extent of the area under study and the consistency in the seismotectonic regime, we assumed that path and local site effects were properly constrained from the Aigion 1995 simulations. It was, therefore, only necessary to define the seismic source parameters and, particularly, the fault dimensions and geometry, as well as the M_max. Fault characteristics were adopted from various studies (see Table 2 caption) and the M_max was calculated following the approach described by Kourouklas et al. [103], which is a slightly different version of the method proposed by Pace et al. [104]. The adopted approach combined various scaling relationships between magnitude and rupture length with the maximum observed magnitude obtained from historical data. The finally adopted M_max (Table 2) was weighted and acquired through the combination of all relative M_max values, along with their standard deviations. In our assessment, the scaling laws from [105,106,107,108] were used.

Furthermore, the definition of the hypocenter position and the slip distribution onto the fault plane is a crucial step in stochastic simulations. For the purposes of the present study, we divided each fault segment into subfaults using Equation (1), and the hypocenter position was placed randomly among them (minimum of 10 iterations). Different slip rupture patterns (minimum of 5 iterations) were examined, applying random slip weights to each subfault.

3. Results

3.1. PSHA

The spatial distribution of the PGA, computed via the logic tree approach (Figure 2) for a return period of 475 years, is illustrated in Figure 3a. The highest values were identified in close proximity to the coastline, near the towns of Nafpaktos and Aigion. The PGA decreased towards the north and reached its minimum of about 250 cm/s² at the NE edge of the study area. A maximum of ~325 cm/s² was obtained approximately 5 km SSE of Aigion. The difference between the highest and the lowest values was 75 cm/s², which indicates that their variation was low to intermediate. The spatial distribution pattern of the PGA for a return period of 950 years was nearly identical (Figure 3b). A maximum of ~400 cm/s² was found SSE of Aigion, meaning that there was an increase of 75 cm/s² compared to RP = 475 yr. The lowest value was about 300 cm/s², increased by 50 cm/s² from the respective value obtained for the return period of 475 years.

Regarding the seismic hazard intensity measure of the PGV, the spatial distribution was similar to the PGA for a return period of 475 years (Figure 4a). A maximum of ~18.5 cm/s was found toward the southern edge of Aigion, where the highest PGA was also determined. The lowest value persistently remained at the NE end of the study area, with an approximate PGV of 14 cm/s. In Nafpaktos, the PGVs were lower than in Aigion but higher than in Patras. The variation in the values was small, as the difference between the highest and lowest PGV was only 4 cm/s. In Figure 4b, the spatial distribution of PGVs for a return period of 950 years is depicted. The highest PGV was about 24 cm/s, south of Aigion, and the lowest was 18 cm/s at the NE tip. The increase between PGVs for the two return periods was only 5 cm/s for the highest and 4 cm/s for the lowest values.

Hazard curves were also generated for the three main WGoC city and towns; namely, Patras, Nafpaktos, and Aigion (Figure 5a). PGA values were calculated for a range of exceedance probabilities in 50 years to observe the increase rate for the PGA as a function of the return period. The hazard curves were proximal to each other, indicating that the level of seismic hazard was similar among the three sites. Patras presented a slightly lower PGA than Nafpaktos and Aigion. It could not be determined whether Aigion or Nafpaktos exhibited higher seismic hazard, as the PGAs were almost identical between the two cities. It is worth noting that the value of 1000 cm/s² was exceeded only for very small probabilities of exceedance, which implies that such extreme ground motions are unlikely to be reached. The last output of the PSHA calculations was the computation of Spectral acceleration (S_a) values for periods ranging from 0.1 s to 2.0 s in the three aforementioned sites in order to construct uniform hazard spectra (Figure 5b). The S_a curves for Patras, Nafpaktos, and Aigion were very similar to each other throughout all periods. This was reasonable in light of the close geographical distance among them. As illustrated in both PGA and PGV maps, the city and the two towns were adjacent to regions of comparable maximum expected ground motions. The maximum of the three spectra was about 500 cm/s² at a period of 0.25 s. The curves seemed to be nearly identical for periods above 0.9 s. Thus, at high periods, there was almost no difference between the three sites. Minor deviations were detected in the period range [0.3, 0.5] s, for which Nafpaktos seemed to have slightly lower S_a values than Patras and Aigion. The elastic design spectra proposed by the Current National Building Code (EAK2003) [109] and Eurocode 8 (Ec8) [110] were also plotted to investigate their relation with our results. The seismicity was defined as high (type I) and the soil as bedrock (type A) to match the input data of the herein proposed model. The Ec8 spectrum overlay the spectra of the three urban areas for all periods, while that of EAK2003 covered the spectra for the period range [1.4, 2.0] s.

3.2. Seismic Scenarios

3.2.1. Validation of the Aigion 1995 Mainshock Simulation

In order to assess the ground motion variability caused by the Aigion 1995 mainshock, PGA values were computed on a grid enclosing the area under study, with nodes at a spacing of 0.03° in latitude and longitude. The spatial distribution of synthetic PGA values is depicted in Figure 6a, along with the surface projection of the activated fault [29] and the available recorded PGA. The highest values estimated onshore exceeded 500 cm/s², whereas in the city of Aigion, where the highest observed PGA was reported, they were close to 300 cm/s². The spatial distribution of simulated values was similar to those presented in other studies [88,111], despite using diverse input parameters. The validation of the final parameters used in the simulation is shown in Figure 6b, where simulated PGA values are plotted against those derived from the selected GMPEs and the recorded ones as a function of Joyner and Boore distances (R_jb). The functional form of the GMPEs used corresponded to a normal faulting style and type C (NEHRP) soil condition. As shown in Figure 6b, the PGAs obtained from EXSIM were in good agreement with the GMPE curves throughout the entire R_jb range, except for the points that lay very close to the surface projection of the fault plane (R_jb ≤ 2–3 km). This was, however, anticipated, given that GMPEs are generally not fully capable of reliably reproducing the ground motion in very short distances due to the limited availability of near-fault strong motion datasets, which affects their formulation. Moreover, directivity effects may have a strong influence on near-fault ground motion that is not fully captured by the GMPEs used. In this case, for example, the high PGA value recorded in Aigion (AIGA; Figure 6b) has been attributed to forward rupture directivity, in addition to local soil and topographic characteristics [29,111]. Nevertheless, the overall PGA variability lay inside the ±1σ range for both GMPEs in the entire R_jb range. Regarding the attenuation pattern, PGAs prescribed by the A14 and B14 models appeared to decay slightly faster than the simulated ones; however, taking into account the fact that the same finding also applied to the recorded ones (red triangles; Figure 6b), this highlights the importance of retrieving region-specific parameters that can be incorporated into GMPEs. The finally adopted synthetic PGA values were obtained based on the desire to achieve the right balance between keeping the lowest misfit among synthetic, GMPE-derived, and real values. Consequently, as shown in Figure 6b, our simulations were capable of reproducing the PGAs recorded from the two closest stations (AIGA, AMIA) more closely than the GMPEs, whereas at larger distances the general trend of the recorded PGAs (which presented a relatively high variability in certain cases) was captured by both GMPEs and synthetic values.

3.2.2. Spatial Distribution of Simulated PGA for Selected Faults and Site-Specific Analysis

In the finite-fault stochastic scenarios of the present study, a worst-case seismic scenario was adopted in contrast to the PSHA, where peak ground motion parameters were obtained from a large number of possible earthquakes with a certain probability of exceedance. Figure 7 presents the spatial distribution of the simulated PGA values generated by the three selected faults (Table 2). The color scales retain a common, fixed range of PGA values to enable a comparison between each scenario. Simulated acceleration time series for Patras, Nafpaktos, and Aigion are also illustrated along with the resulting PGA values.

In all cases, the maximum PGAs, located at the surface projection of the faults or nearby, appeared to be relatively close, ranging from approximately 350 to 450 cm/s². The HF may have had larger dimensions among the three cases, but PF produced the highest acceleration values onshore (Figure 7a), even though it was assigned a slightly lower M_max (6.3) than HF (6.4; Table 2).

Similar to the PSHA framework, hazard response spectra for Aigion, Nafpaktos, and Patras for hypothetical future earthquakes originating from the selected faults (PF, HF, TF) were constructed for 5% damping (Figure 8). In this case, however, it became plausible to gain insight into which seismic sources are more capable of causing damage to each urban area. In all cases, the response spectra of simulated ground motions exhibited a spike behavior at short periods of [0.1–0.3] s, followed by a relatively sharp decline. As shown in Figure 8a, the town of Aigion is mostly threatened by HF and TF, whereas PF poses the greatest threat to Nafpaktos and Patras. The city of Patras (Figure 8b) is exposed to a lower level of seismic hazard compared to the other cases when taking into account the major active faults of the WGoC. Maximum S_a values exceeded 300 cm/s² for a hypothetical rupture of PF, whereas HF produced the lowest expected values (approximately half of the corresponding ones for PF). Moreover, the town of Nafpaktos, located in between the aforementioned sites, is highly susceptible to high levels of seismic hazard due to the PF, with maximum expected S_a values approaching almost 900 cm/s². TF and HF, which are located quite far from the town, do not pose significant hazards, given that the predicted maximum S_a values were within the ranges appointed by the National Building Code [109].

4. Discussion

The high seismicity of the study area, both onshore and offshore, is due to the high average extension rate of the tectonic rift [5]. It is expressed through both large earthquakes and seismic swarms [7,17,20,39,112,113,114], which confirm the great importance of seismic hazard assessment as a means to quantify anticipated levels of ground shaking.

When performing PSHA, there are two types of uncertainties, i.e., aleatory uncertainty and epistemic uncertainty [115]. The first one accounts for random variations in PGA, PGV and Sa values due to the implementation of a GMPE, whereas the second one accounts for the accuracy of the values [115]. Epistemic uncertainty is usually handled by the incorporation of a logic tree approach to account for the implementation of more than one GMPE [115,116,117]. This is the procedure followed in the present study, given that by taking into account alternative models the uncertainties can be reduced [115].

The PGA results for a return period of 475 years, obtained through the Cornell–McGuire approach, can be directly compared with the reference value proposed by the current Greek Building Code [109]. EAK2003 divides Greece into three zones and defines the maximum expected ground acceleration for a return period of 475 years. The Western Gulf of Corinth is within zone II, with a reference value of 0.24 g (~240 cm/s²). However, in the herein proposed model, the entire study area exceeded 240 cm/s². The computed PGA results in the northeastern part of the area were close to the EAK provisions, even though they still slightly exceeded them. The highest value was about 325 cm/s², surpassing EAK2003 by 74%. The calculated PGA for a return period of 950 years can be utilized for the construction of buildings of greater importance, such as schools and medical centers. The spatial patterns of PGA and PGV for both return periods were similar; i.e., the highest values were found close to the Gulf, where seismicity is higher, whereas the lowest values were onshore, far from the coastline.

A comparison of the PGA results obtained to those of Banitsiotou et al. [49] and Tselentis et al. [54] was attempted, both of which are for the whole Greek territory. Banitsiotou et al.’s [49] computed PGA values were for certain cities, one of them being Patras, which was assigned a PGA of 0.26 g for a return period of 475 years. In this study, the corresponding PGA was ≈300 cm/s², which is slightly higher. Tselentis et al. [54] computed PGA and PGV for the same return period for Greece. PGA values varied between 0.40 g and 0.50 g close to Patras and Nafpaktos and between 0.50 g and 0.60 g close to Aigion. Concerning PGV, Tselentis et al. [54] proposed values in the broad range of [20.0–105.0] cm/s. In the present study, PGA and PGV values were in the range of [0.25–0.33] g and [15.0–18.5] cm/s, respectively. The existing deviations can be attributed to the different seismic and geometry data used in each case.

Hazard curves were constructed for the most populated sites of the study area, the city of Patras and the towns of Nafpaktos and Aigion. All three are located close to the coastline and are characterized by high expected ground motion values. The distances between them are relatively small and they belong in the same or neighboring (according to the model) area source zones, parameterized with similar geometry and seismicity data. For this reason, their corresponding hazard curves were quite close to each other, whereas strong ground motions (>1 g) occurred only at low exceedance probabilities.

Uniform hazard spectra were obtained by adopting the GMPE from Danciu and Tselentis [76]. The final results were derived using the same logic tree but without the minor branches, as they were replaced with this GMPE. However, the adopted model was, again, parameterized to take into account all types of focal mechanisms for each zone. The values for the spectral acceleration were almost the same for all three cases, as was also observed in the hazard curves. Differences from the elastic design spectra of EAK2003 [109] and Ec8 [110] were evident. The fundamental periods of interest had ranges of [0.1, 0.5] s (Figure 5) for the majority of buildings in Greece. In this range of values, the spectra for Patras, Nafpaktos, and Aigion significantly surpassed EAK2003. However, they were fully overlaid by the design spectra from Ec8. Moreover, given that the probability of exceeding the respective S_a value was equal for all periods, and since all future earthquake events were taken into account, it is reasonable that higher values were computed. Hence, it is suggested that the Ec8 regulations should be respected for the majority of buildings in the study area.

In addition to the traditional PSHA approach, finite-fault stochastic simulations were performed by generating earthquake scenarios for a predicted maximum magnitude (M_max), calculated by taking into account past seismicity and fault data. The three modeled active faults were the Psathopyrgos, Helike, and Trizonia Faults. Final results were obtained by testing both random earthquake nucleation points and co-seismic slip distribution on the surface. Prior to the computations, stochastic modeling parameters regarding the path and local site effects were calibrated by performing strong ground motion simulation of the largest recent event; namely, the Aigion earthquake that struck the study area in 1995. Comparison with real strong motion data and GMPEs proved that this methodology is reliable for the reproduction of the ground motion, regardless of the uncertainties that are involved, which can be ascribed mainly to source complexity and local site condition variability. As a result, this methodology can be treated as a powerful tool for the simulation of the expected strong ground motion of future strong earthquakes, especially for cases where recordings are sparse or not available. It, therefore, provides a unique opportunity to comprehensively evaluate seismic hazard with the synergy between PSHA outcomes and specific scenarios.

Site-specific analysis at the three selected sites was performed through a comparison between response spectra for each seismic scenario. S_a values constitute a good indicator of seismic loading for a variety of structures, as they describe the absolute maximum response of a single degree-of-freedom oscillator to an enforced ground motion. The mean response spectra exceeded the current EAK2003 in some cases, as in Nafpaktos for the hypothetical rupture of the Psathopyrgos Fault. The city of Patras was revealed to be the site less exposed to seismic hazard among those examined for the selected scenarios. However, it should be noted that other faults outside the WGoC may be more dangerous for Patras, such as the Rion–Patras fault zone (Figure 1); the local NW–SE sinistral-slip fault, which was related to the 1993 earthquake [41]; or even the Andravida fault, further southwest, which has caused strong earthquakes in the past [118]. The latter two faults were not taken into account, as the present study assessed seismic hazard in the WGoC area.

Overall, the PGA variability obtained from the specific fault scenarios lay inside the PGA values calculated through the PSHA. Figure 3a,b indicate that the western part of the study area exhibited a similar level of seismic hazard under the PSHA evaluation. The expected peak ground motions in this part, however, were considerably affected by the nearby seismic sources to the west (the Ionian Islands and the westernmost end of the Hellenic Arc), which are among the most active and productive in terms of seismic potential [119]. PSHA provided valuable information reflecting the combined effects of all potential seismic sources, making it possible to assess seismic hazard in terms of statistical likelihood of occurrence. On the other hand, the employment of seismic scenarios made it possible to acquire more realistic results concerning specific strong events through the definition of local site effects and path properties but without taking into account the time frame of occurrence.

5. Conclusions

A holistic seismic hazard assessment, including both probabilistic and scenario-based methods, was conducted for the highly active Western Gulf of Corinth area in Central Greece. Initially, seismic hazard was evaluated using a probabilistic approach, taking into account all the M_w ≥ 4.0 earthquakes recorded during 1900–2019, in order to determine values for PGA, PGV, and S_a in a dense grid. For this purpose, the Cornell–McGuire method was utilized. Aiming to incorporate a multitude of seismotectonic models in order to qualitatively cope with epistemic uncertainties, three models—namely, ESHM13, ESHM20, and VAM16—were considered, and GMPEs were introduced for each zone. The PGA and PGV results were obtained through an equal-weight logic tree approach in which each major branch was a seismotectonic model and each minor branch was a modified GMPE. The logic tree procedure incorporated with the five GMPEs that were developed using Greek data qualitatively reduced epistemic uncertainties, which simple PSHA approaches may carry. In addition, PGA values were computed for the most populated urban areas (Patras, Nafpaktos, and Aigion) for various exceedance probabilities in 50 years.

The results obtained herein highlight the great significance of seismic hazard assessment using a combined approach that takes into account not only the evaluation of ground parameters in terms of probabilities of occurrence in a given time frame but also the anticipated effects of deterministic worst-case scenarios. Future work could include disaggregation of PSHA results so that the parameter pair of magnitude and epicentral distance that contributes most to seismic hazard can be identified. In this way, scenarios could be generated based on this result. Furthermore, a study of the Peak Ground Rotational Acceleration (PGRA) and Velocity (PGRV) values could be undertaken, as it has been proven that their results provide aid for engineering purposes [58,77,120]. In addition, the incorporation of a comprehensive microzonation scheme could provide valuable insight into the impact of future strong earthquakes on the major cities of the study area by identifying the possible amplification trends with respect to the structural response. Lastly, the results of the present seismic hazard study can be exploited in the future to assess seismic risk at urban centers in the WGoC area after incorporating structural vulnerability data.