Seismic scenario simulation and ANN-based ground motion model development on the North Tabriz Fault in Northwest Iran

Vol.: (0123456789) J Seismol (2025) 29:147–169 https://doi.org/10.1007/s10950-024-10264-x RESEARCH Seismic scenario simulation andANN‑based ground motion model development ontheNorth Tabriz Fault inNorthwest Iran CaglarTemiz· S.M.SajadHussaini· ShaghayeghKarimzadeh· AysegulAskan· PauloB.Lourenço Received: 23 July 2024 / Accepted: 25 October 2024 / Published online: 20 November 2024 © The Author(s) 2024 Abstract Earthquakes pose significant seismic hazards in urban regions, often causing extensive damage to the built environment. In regions lacking robust seismic monitoring networks or sufficient data from historical events, ground motion simulations are crucial for assessing potential earthquake impacts. Yet, validating these simulations is challenging, leading to notable predictive uncertainty. This study aims to simulate four scenario earthquakes with moment magnitudes of 6.8, 7.1, 7.4, and 7.7 in Iran, specifically investigating variations in fault plane rupture and earthquake hypocenter. The North Tabriz Fault (NTF), located within the seismic gap in northwest Iran, is selected as the case study due to the lack of well-recorded ground motions from severe earthquakes, despite historical evidence of large-magnitude events. Simulations are conducted using a stochastic finite-fault ground motion simulation methodology with a dynamic corner frequency. Validation of the simulations is performed by comparing estimated peak ground motions and pseudo-spectral ordinates with existing ground motion models (GMMs), supplemented by inter-period correlation analysis. Simulation results reveal high hazard levels, especially in the northeastern area near the fault plane. Intensity maps in terms of the Modified Mercalli Intensity (MMI) scale underscore the urgency for comprehensive preparedness measures. Finally, a region-specific GMM is developed using Artificial Neural Networks (ANN) to predict peak ground motion parameters with an online platform accessible to end-users. Keywords Stochastic ground motion simulation· Seismic hazard maps· Modified mercalli intensity (MMI) map· North Tabriz Fault (Tabriz Iran)· Artificial neural networks (ANN)-based ground motion model (GMM) 1 Introduction From ancient times to the present day, earthquakes have remained one of the most devastating natural C.Temiz Departments ofCivil Engineering, Middle East Technical University, Ankara, Türkiye e-mail: caglar[email protected] S.M.S.Hussaini· S.Karimzadeh(*)· P.B.Lourenço Civil Engineering Department, Institute forSustainability andInnovation inStructural Engineering (ISISE), ARISE, University ofMinho, Guimaraes, Portugal e-mail: [email protected] S. M. S. Hussaini e-mail: [email protected] P. B. Lourenço e-mail: [email protected] A.Askan Departments ofCivil Engineering andEarthquake Studies, Middle East Technical University, Ankara, Türkiye e-mail: [email protected] 148 J Seismol (2025) 29:147–169 Vol:. (1234567890) disasters, leading to significant casualties and the destruction of infrastructures in seismically active areas. However, recent advancements in technology and computational power have improved the ability to assess potential earthquake hazards, enabling the design of safer structures that help mitigate the effects of earthquakes. In regions with well-established seismic networks that are prone to frequent and severe earthquakes, recorded ground motions can be used as inputs for precise risk assessment studies, thereby improving the safety of the built environment. However, in areas with seismic gaps and limited recorded ground motions, particularly from strong earthquakes at short source-to-site distances, the availability of region-specific ground motion data is often insufficient for defining a design scenario (Rezaeian etal. 2024; Yamamoto and Baker 2013). Additionally, the practice of selecting and scaling ground motions can alter the correlation between the modified motions and their original physical conditions (Hancock etal. 2006; Watson-Lamprey and Abrahamson 2006). In such cases, ground motion simulations offer an alternative approach for generating representative time series tailored to the region’s seismological features (Arora etal. 2020). These simulations can be used to evaluate the anticipated hazard levels and the resulting potential seismic damage and risk levels in seismic gap regions worldwide. The literature outlines a range of ground motion simulation techniques, each with varying levels of accuracy and requiring different input-model parameters and computational approaches. These techniques can be broadly categorized into deterministic, stochastic, and hybrid methods (Rezaeian and Sun 2014). Deterministic methods use representative seismic source rupture and velocity models alongside numerical methods to solve the partial differential equations governing wave propagation (Frankel 1993). These methods are most effective for simulating lowfrequency contents in regions lacking high-resolution earth models due to the minimum wavelength constraints. In contrast, stochastic methods are better suited for simulating higher-frequency content that is random and incoherent. In the Stochastic method, a far-field S-wave spectrum is enhanced by random phases to generate a mean horizontal ground motion component (David M. Boore 1983). Hybrid methods combine both deterministic and stochastic approaches to accurately simulate the full frequency spectrum and leverage their benefits (Kamae etal. 1998). Ground motion simulations provide valuable insights into the potential impacts of earthquakes, particularly in regions with limited recorded seismic activity, enabling more accurate risk assessments and the development of targeted mitigation strategies. Ground motion simulations have been widely used in various studies to replicate historical events, evaluate different earthquake scenarios, assess structural demands across various types of structures, and develop ground motion models (GMMs). Within this context, Karimzadeh and Askan (2018) simulated the historical 1939 Erzincan earthquake in Türkiye, comparing their results to empirical intensity maps. Similarly, Tanırcan and YelkenciNecmioğlu (2020) simulated the 2017 Bodrum-Kos earthquake, validating their results through comparison with empirical GMMs and generating seismic hazard maps. Later, Can etal. (2021) simulated the 2002 Cay earthquake in Türkiye and validated their findings by comparing goodness-of-fit scores between observed and simulated motions in addition to comparing their results to empirical intensity maps. These studies underscore the reliability of simulations, when properly validated, in generating robust seismic hazard estimates comparable to empirical GMMs and observed intensities. In addition to historical event modeling, large suites of simulations have contributed to the development of GMMs, supporting the estimation of peak ground acceleration (PGA) and spectral ordinates (Karimzadeh etal. 2023a; Karimzadeh etal. 2023b; Withers etal. 2020). Simulations have also been used to estimate seismic demands for different structural systems, ranging from single-degree-of-freedom systems (Atkinson et al. 2011) to more complex reinforced concrete and steel structures (Ozsarac et al. 2021; Tsioulou etal. 2019; Xu and Feng 2018). Building on these developments, Smerzini et al. (2024) developed the BB-SPEEDset, a dataset of near-source, broadband physic-based simulated accelerograms, to bridge knowledge gaps caused by sparse ground motion recordings. They validated the simulations from both seismological and engineering perspectives to ensure reliability for structural assessments. In a related study, Hariri-Ardebili and Rezaeian (2024) explored the applicability of 149 J Seismol (2025) 29:147–169 Vol.: (0123456789) stochastic ground motion simulations for probabilistic seismic performance assessments of geo-structures, such as concrete dams, by developing fragility curves and identifying optimal intensity measures. More recently, Karimzadeh etal. (2024) simulated the 1998 Faial earthquake in the Azores, focusing on the structural demands of masonry buildings, which are often vulnerable to the high-frequency components of seismic motions. Recent efforts have emphasized the importance of region-specific simulations to enhance pre-earthquake risk assessments, especially in seismically active areas. For instance, Arslan Kelam et al. (2022) and Askan etal. (2024) applied ground motion simulations to evaluate seismic risks in vulnerable regions, emphasizing the importance of scenario modeling for hazard identification and mitigation strategies. The significance of such proactive assessments was highlighted by the February 2023 earthquakes in Kahramanmaras (Mw = 7.8) and Elbistan (Mw = 7.5) in Türkiye, which resulted in devastating casualties, underscoring the need to evaluate seismic risks before major events occur. Collectively, these studies showcase the growing importance of ground motion simulations in seismic hazard and risk assessments, as well as in analyzing seismic demands for different structural systems, assuming proper calibration. However, the precision of these assessments largely depends on the selected simulation methodologies and the quality of their calibration and validation processes. This study aims to simulate ground motions in an important seismic gap located on the North Tabriz Fault (NTF) in Northwest Iran while considering the uncertainty related to rupture effects. Tabriz is selected as the study area due to its designation as one of the world’s hazardous regions with a seismic gap. The NTF, characterized by a sharp trace with a North-West to South-East trend, traverses the northern part of the city center. Ghasemi etal. (2021) have examined the anticipated levels of damage resulting from the rupture of two scenario events with moment magnitudes (Mw) of 7.4 and 7.3. Previous studies have suggested that the NTF has the potential to experience an earthquake with a maximum Mw of 7.7 (Solaymani Azad et al. 2015). Despite historical earthquakes that have caused significant loss of life and widespread destruction, there is a lack of recorded strong earthquakes in recent history. This highlights the need to evaluate the potential hazard level of forthcoming seismic events. Although several studies have focused on seismic hazard, vulnerability, and risk assessment in the region (Eslami and Taghizadeh-Farahmand 2020; Karimzadeh etal. 2014; Mohammadi etal. 2020a, b), there remains a significant gap in the investigation of large-magnitude earthquake scenario simulations specific to this area and the development of region-specific GMM. This study particularly addresses a unique problem by investigating the aleatory uncertainty associated with the fault plane rupture and the location of the hypocenter on the NTF. Through ground motion simulation, and validation against various empirical ground motion models, the seismic hazard in the region is also investigated. To estimate the seismic hazard level in the city center, we use the stochastic finite-fault ground motion simulation method with a dynamic corner frequency approach proposed by (Motazedian and Atkinson 2005). To model the most catastrophic earthquake scenarios, simulations are conducted for potential earthquakes with varying magnitudes, including Mw = 7.7, 7.4, 7.1, and 6.8. The simulation results are presented and analyzed through both visual and statistical methods. Initially, to validate the simulations, four ground motion models (GMMs) are used, denoted as BA08 (Boore and Atkinson 2008), AC10 (Akkar and Cagnan 2010), ASB14 (Akkar etal. 2014), and KAAH15 (Kale etal. 2015). BA08 is a globally applicable GMM, while AC10 and ASB14 focus on the Middle Eastern and European regions, and KAAH15 specifically involves data from Eastern Türkiye and Iran. For further validations of the simulated dataset, this study delves into the inter-period correlation of the residuals of pseudo-spectral ordinates. Following the validation steps, anticipated hazard measures are presented using deterministic seismic hazard maps that display various scenario events and their associated uncertainties. Additionally, a mean intensity map based on the Modified Mercalli Intensity (MMI) scale is generated. Finally, a regional artificial neural network (ANN)-based GMM is developed to predict peak ground motion parameters for anticipated scenarios in the region. Ultimately, the study produces an extensive compilation of ground motion time series along with the ANN-based GMM intended for future applications in engineering practice in an online platform. 150 J Seismol (2025) 29:147–169 Vol:. (1234567890) 2 Study area Tabriz is a prominent city in Iran with a rich historical heritage. The archaeological excavations indicate a history of urbanization spanning approximately 5000 years. Presently, Tabriz serves as the capital of EastAzerbaijan Province and is situated in the northwest region of Iran. Geographically, Tabriz lies between eastern Anatolia, the western Caspian Sea, the southern Caucasus thrust belt, and the northern Zagros Mountains range. According to the Statistical Centre of Iran (2020), the population of Tabriz slightly exceeds 3 million. The NTF is a prominent geological feature in the region, consisting of northwestern and southeastern segments that stretch from Mishu Mountain in the west to Bostanabad in the east (Farahani 2018). This fault has an average strike of 115°, a length of approximately 240 km, a dip of predominantly vertical, and a minimum slip rate of 2 mm/year (Karakhanian etal. 2004). The urban area of Tabriz comprises young and unconsolidated deposits from rivers and glacial sediments of Cenozoic and Quaternary formations. Additionally, Tabriz is located in an alluvial basin (Azarafza and Ghazifard 2016). The alluvial thickness in Tabriz varies by region. In southern Tabriz, it reaches up to 250–300 m. However, in the inhabited areas, including the city center (the actual study area), the alluvial thickness changes from 50 to 100 m (Karimzadeh et al. 2014). Moreover, excavations in the area have revealed the presence of conglomerate, sandstone, and marl layers beneath these alluvial sediments (Azarafza and Ghazifard 2016). The urbanized area is mostly on clay and silt, and marl and sandstone layers (Karimzadeh etal. 2014). The southern and eastern parts of the city are characterized by conglomerate and marl layers, while sandstone formations are found in the northern and northeastern mountainous areas. Berberian and Arshadi (1976) have investigated the seismic activities of the NTF, including both historical events and those in the twentieth century. Previous studies indicate that Tabriz suffered significant devastation from multiple earthquakes between 634 A.D. and the late 1800s. Notably, three major earthquakes in 1042 (Mw = 7.3), 1721 (Mw = 7.3), and 1780 (Mw = 7.4) caused severe damage through coseismic surface faulting, as documented in reports from the nineteenth century and earlier (Berberian and Yeats 1999; Hessami etal. 2003). However, the existing descriptions of these events lack sufficient detail to accurately assess the extent of destruction and ground deformations (Berberian and Arshadi 1976). Notably, Tabriz City experienced substantial impacts from the ruptures along the northwestern segment of the fault in 1721 and 1780, as documented by K. Hessami etal. (2009). While seismic activity records became more reliable in the twentieth century, these events did not result in major damage to the city. Figure1 presents the regional tectonic map with the associated earthquakes. Given the long interval since the last large-magnitude earthquake (Mw ≥ 7.5) and the ongoing activity of the NTF, the potential for future strong earthquakes in the area should not be ignored. Therefore, the study area is centered on Tabriz city center. Kheirizadeh Arouq etal. (2020) conducted a study based on catastrophe theory, revealing that over a quarter (35%) of urban areas are classified as highly vulnerable to seismic activity. Additionally, Tabriz is identified as a very highly susceptible region on the earthquake susceptibility map developed by Mohammadi etal. (2020a) This study simulates potential strong earthquakes and obtains simulated ground motions from 22 designated dummy stations within Tabriz city center, as illustrated in Fig.2. These stations are selected based on high levels of urbanization and human population. The NTF trace is shown with the red line, while the blue dots indicate the location of dummy stations within the city center. These dummy stations are positioned in areas with important structures such as residential buildings, airports, hospitals, and other essential facilities. 3 Ground motion simulation methodology In regions lacking well-studied local source and velocity models, the stochastic methodology is considered effective and preferred for simulating ground motions. This method is particularly advantageous for capturing the aleatory uncertainty associated with earthquakes, thereby accounting for the inherent variability in ground motions. However, it primarily focuses on simulating the random intermediate and mediumto high-frequency components of ground motions (Beresnev and Atkinson 1997; David M. Boore 1983a, 1983b; Motazedian and Atkinson 2005). While this is suitable for many engineering applications, such as for structures sensitive to these frequency ranges, the approach does not fully capture low-frequency components. Low-frequency ground motions, which are important 151 J Seismol (2025) 29:147–169 Vol.: (0123456789) for the response of tall buildings and other long-period structures, require more complex deterministic physics-based simulations or a hybrid simulation scheme. Despite these limitations, stochastic techniques are known to accurately simulate ground motion amplitudes within the frequency range of primary interest to engineers (Boore 2009; Karimzadeh 2019; VahidiFard etal. 2017). Fig. 1 Epicenter of earthquakes within a 20 km radius around the North Tabriz Fault (NTF) (adapted from (Farahani 2018) and (Ghorbani 2021)) Fig. 2 Fault line and selected stations for ground motion simulations 152 J Seismol (2025) 29:147–169 Vol:. (1234567890) For the simulation of the scenario events in this study, the stochastic finite-fault ground motion simulation technique developed by (Motazedian and Atkinson 2005) is employed using the EXSIM platform. This approach represents the fault plane using a rectangular geometry and models it as a collection of smaller sub-faults that act as point sources (David M Boore 1983a, 1983b). The acceleration response spectrum in the frequency domain is modeled for each point source, taking into account the contributions of source, path, and site effects. Subsequently, the responses for all sub-faults are summed in the time domain to obtain the final response of the main fault as follows: where a(t) is the total seismic signal at a time t, nl and nw are the number of sub-faults along the length and width of the main fault, respectively, and Δtij is the relative delay time for the radiated wave from the ijth sub-fault to reach the observation point. For each subfault (David M Boore 1983a, 1983b), the acceleration spectrum is computed as: (1) a (t)= ∑ nl i=1 nw ∑ j=1 aij(t + Δtij) , (2) Aij(f)=CM0ij Hij ⎡⎢⎢⎣ (2𝜋f)2∕ ⎡⎢⎢⎣ 1+ � f fcij � 2 ⎤⎥⎥⎦⎤⎥⎥⎦ e−𝜋fRij Q(f)𝛽G � Rij � A(f)e−𝜋Kf Where C = R 𝜃𝜑FV 4𝜋𝜌𝛽 2 is a constant, with R𝜃𝜑 as the radiation pattern (average value of 0.55 for shear waves, F as free surface amplification (2.0), V as partition onto two horizontal components (0.71), ρ as the density, and β as the shear-wave velocity. M 0ij = M 0 S ij ∑ nw k=1∑ nl l=1 S kl is the seismic moment, Sij is the relative slip weight and f c ij (t) is the dynamic corner frequency of the ijth subfault where f ij(t)=N R(t)−1∕34.9 ×106β( Δσ M0 ave ) 1∕3 . Here, Δσ is the stress drop, NR(t) is the cumulative number of ruptured sub-faults at time t , and M 0 ave = M0 N is the average seismic moment of subfaults. Sij is the distance from the observation point, Q(f) is the quality factor, G( R ij) is the geometric spreading factor, A(f) represents the site amplification term, and e−πKf is a high-cut filter included to provide the spectral decay at high frequencies described with the Kappa factor of soils (Anderson and Hough 1984). Rij is a scaling factor introduced to conserve the high-frequency spectral level of the subfaults. The term H ij =(N∗( ∑ (f2∕[1+ � f f0 � 2 ])∕ ∑ ((f2∕[1+ � f f0 ij � 2 ]))) 1∕2 corresponds to a scaling factor for conversation of the spectral shape at higher frequencies. To simulate the scenario events for each magnitude, the length and width of the ruptured fault are estimated using the study by Wells and Coppersmith (1994). The entire length of the NTF is approximately 240 km (Farahani 2018; Solaymani Azad etal. 2015). Alternative fault planes are considered to capture the variability of the rupture plane. For the largest event with Mw = 7.7, the NTF is divided into 17 alternative ruptured fault lines, each spanning a length of 140 km, to generate various possible scenarios. Similarly, for the other magnitudes, the NTF is divided into more than 13 events for each case. Specifically, for Mw = 7.4, the fault plane is divided into 12 segments, each with a length of 84 km; for Mw = 7.1, it is divided into 13 segments with a length of 52 km, and for Mw = 6.8 it is divided into 15 segments with a length of 32 km. The respective widths of these ruptured fault planes are determined to be 24 km, 20 km, 16 km, and 12 km for Mw = 7.7, 7.4, 7.1, and 6.8, respectively. It is noted that for all scenarios with events of the same magnitude, the epicenter is shifted along the fault plane by an equal distance. In addition to the variability of the ruptured fault plane, the study also investigates the randomness associated with the hypocenter location for alternative scenario events. To achieve this, the focal depth and epicenter location are treated as random variables. Moradi etal. (2011) found that the median focal depth for earthquakes in the Tabriz region is approximately 12.1 km, with a standard deviation of around 4.9 km. In this study, to incorporate the uncertainty related to the focal depth, a range of 6.0 km to 18.0 km is considered. Additionally, the epicenter location is assumed to change along the length of the ruptured fault plane. 153 J Seismol (2025) 29:147–169 Vol.: (0123456789) Excluding the random parameters discussed above, the other input-model parameters are determined in a deterministic manner across all magnitudes and are essential for completing the simulations. These parameters are sourced from various studies. The strike and dip angle values, derived as 115° and 90°, respectively, are obtained from the studies by Farahani (2018) and Ghayamghamian and Rajool (2012). The quality factor is taken from the study conducted by Jafarian Vernosfaderani et al. (2019), while the kappa factor is derived as 0.035 based on the research conducted by Amiranlou et al. (2016). The duration model follows the work of Herrmann (1985) as applied in Hoveidae etal. (2021) for the simulation of scenario earthquakes in the Tabriz region. This study uses site amplifications from Boore and Joyner (1997) for a generic soil class with a Vs30 value of 310 m/s. This value is selected because the dummy stations are mainly located in areas with Eurocode soil Class C (Code Price 2005), with Vs30 values ranging from 300 to 360 m/s, as indicated by GIS and DEM-based seismic site condition studies by Karimzadeh et al. (2017, 2019). Yaghmaei-Sabegh and Hassani (2020) further corroborate this by reporting an average Vs30 value of approximately 300 m/s for the region, reflecting minimal soil variability based on analyses of seismic events in Iran. Although soil variability is important in improving ground motion simulations, this study adopts a constant Vs30 value to focus on uncertainties related to the fault rupture plane and hypocenter location, which limits the results to a generic soil classification. The stress drop is calculated using the empirical equation given by Mohammadioun and Serva (2001). Geometrical spreading, pulsing percent, rupture velocity, density, and window type are assumed to be the same as in the research conducted by Amiranlou etal. (2016) and Hoveidae etal. (2021). A summary of all this information can be found in Table1. 4 Simulated ground motion dataset A total of 206,382 simulated ground motion time series have been generated for Tabriz City at the selected stations. The dataset encompasses earthquake scenarios with Mw of 6.8, 7.1, 7.4, and 7.7, and Joyner-Boore distances (RJB) ranging from 5.4 km to 87 km with focal depths (Fd) between 6 and 18 km. Given that soil class C predominates in the region, we used a representative Vs30 value of 310 m/s under a strike-slip fault mechanism. The seismological parameters of the dataset are illustrated in Fig.3. It is noteworthy that a higher number of simulations have been generated for larger Mw values, shorter RJB distances, and shallower focal depths to address the scarcity of strong near-fault recordings. Figure4 presents sample simulated ground motions for scenario events with Mw of 7.7, 7.4, 7.1, and 6.8, at a RJB distance of 12 km, and a Fd of 10 km along with their respective peak ground acceleration (PGA) values. 5 Assessment ofsimulations This section focuses on validating the simulations through comparison with GMMs, analyzing attenuation of the simulated data, and developing seismic hazard and MMI maps. Table 1 Deterministic input parameters for ground motion simulation of scenario events with different magnitude values Parameters Value Mw7.7 7.4 7.1 6.8 Length (km) 140 84 52 32 Width (km) 24 20 16 12 Sub-fault length (km) 4 4 4 4 Sub-fault width (km) 4 4 4 4 Stress Drop (bars) 110 95 80 65 Fault Mechanism Strike-slip Strike (o) 115 Dip (o) 90 Shear Wave Velocity (km/s) 3.3 Rupture Velocity/Shear Wave Velocity 0.8 Density (g/cm3) 2.8 Pulsing Percent 35 Quality Factor 103f0.88 Geometrical Spreading R−1R≤85 R−0.5R>85 Duration model T = T0 + 0.05 (T0: Source duration) Kappa Value 0.035 Window Type Saragoni-Hart Site Amplifications Generic soil Sampling Time (s) 0.005 154 J Seismol (2025) 29:147–169 Vol:. (1234567890) 5.1 Validation of simulations It is important to validate the simulated motions before they can be used for engineering applications. Validation methods vary depending on the availability of recorded motions. In cases where recorded motions are absent, a commonly employed approach involves using empirical GMMs for validation purposes. The Fig. 3 Histograms of seismological parameters of the simulated dataset Fig. 4 Samples of simulated time series for the study area 155 J Seismol (2025) 29:147–169 Vol.: (0123456789) literature offers numerous parametric and non-parametric GMMs. However, selecting appropriate GMMs for comparison is crucial, especially those that exhibit similar seismological and tectonic characteristics to the region under study. In this study, four distinct GMMs are used for the validation process, namely BA08 (Boore and Atkinson 2008), AC10 (Akkar and Cagnan 2010), ASB14 (Akkar etal. 2014), and KAAH15 (Kale et al. 2015). These four selected GMMs fulfill different purposes. The BA08 model is recognized as a global GMM due to its development, incorporating recorded data from diverse global locations. In contrast, the AC10 model serves as a local model, specifically tailored to the seismic characteristics of Türkiye. The ASB14 model is a regional one developed to capture the seismic attributes of the Middle Eastern and European regions, encompassing multiple countries. Lastly, the KAAH15 model falls under the category of a local GMM, customized to account for the seismic behavior observed in Iran and Türkiye. The validation primarily focuses on five essential intensity measures (IMs): PGA, Peak Ground Velocity (PGV), and Pseudo-Spectral Acceleration (PSA) with a 5% damping ratio for periods of T = 0.2 s, 0.5 s, and 1.0 s. These parameters serve as representative indicators for evaluating the accuracy and reliability of the simulated dataset. Figure 5 illustrates the distance attenuation of simulated ground motion IMs (i.e., PGA, PGV, and PSA at T = 0.2 s, 0.5 s, and 1.0 s) in comparison to estimates from the GMMs. The results indicate that the simulated IMs generally fall within ± 2 standard deviations (std) of the GMM estimates. For distances less than 20 km, the simulated intensity levels closely align with all GMMs. However, the attenuation patterns for simulated IMs are noticeably faster at distances exceeding 20 km, particularly for longerperiod PSA. These discrepancies can be attributed to various parameter settings in the finite fault stochastic simulation method. Key factors, such as fault geometry (length, width, and depth), stress drop, and rupture velocity, influence how seismic energy is released and propagated. The choice of geometrical spreading model significantly affects how IMs decay with distance, while the quality factor and kappa value represent energy dissipation in the medium and site effects, respectively. Overall, these parameter settings contribute to the observed differences between simulated IMs and those from empirical GMMs, underscoring the importance of conducting region-specific studies to refine these parameters. The present research attempts to assess not only the attenuation of the representative IMs but also to determine the anticipated seismic hazard level within the study area. As a further step in the validation process, the correlations between normalized residual ( ϵ ) of PSA values for various vibration periods are investigated. Recently, Altindal and Askan (2023) validated a simulated ground motion dataset generated for the entire Türkiye with this approach yielding a robust assessment. Given a PSA from an earthquake scenario with magnitude Mw at a recorded distance RJB, ϵ value can be calculated as (Baker 2011): where μlnPSA(Mw,RJB,T) and σlnPSA(T) are the mean value and standard deviation of lnPSA(T) estimated from GMM, respectively. lnPSA(T) is the natural logarithm of observed PSA values, referred to herein as the simulation. In this study, the ϵ values are calculated for all the simulated motions in the dataset for a period range of 0.05 s to 2 s, regarding each of the selected GMMs. In seismic demand assessment of multi-degree-of-freedom (MDOF) systems that are subjected to various periods of excitation, the correlation coefficient of ϵ across the period range ( ρϵ ) is a crucial indicator (Burks and Baker 2014). Therefore, ρϵ of PSA, obtained using Pearson product-moment correlation coefficient, regarding each selected GMM is illustrated in Fig. 6 over the considered period range. These contours indicate that the correlations calculated based on BA08 exhibit the highest interperiod correlation, whereas those derived from AC10 and KAAH15 yield the lowest correlations. However, ρϵ values regarding ASB14 exhibit high inter-period correlations towards longer periods. Nonetheless, a consistent pattern is observed across all GMMs for both inter-period correlations and attenuations of considered intensity measures, affirming the validity of the simulated dataset. 5.2 Seismic hazard and MMI maps To provide a comprehensive understanding of ground motion intensity levels for different scenario events, seismic hazard maps are developed for the region. (3) ϵ(T) = lnPSA(T)−μ lnPSA (M w ,R JB ,T) σ lnPSA (T) 162 J Seismol (2025) 29:147–169 Vol:. (1234567890) respectively. The between-event residuals fluctuate within the range of –0.4 to 0.4, whereas the withinevent residuals exhibit a wider span between –1.0 and 1.0, consistent with findings from previous studies (Akkar et al. 2014; Kale et al. 2015; Karimzadeh etal. 2023a; Mohammadi etal. 2023). It is important to note that the model’s predictions are most reliable within the training data range (6.8 ≤ Mw ≤ 7.7, Fig. 14 Distribution of between-event (τ) and within-event (σ) uncertainties, along with total uncertainties (ϕ), for ln(PGA), ln(PGV), and ln(PSA) at periods ranging from 0.05 s to 2 s Fig. 15 Distribution of the between-event residual (η) with respect to Mw for ln(PGA), ln(PGV), and ln(PSA) at periods of 0.2 s, 0.5 s, 1.0 s, and 2.0 s 163 J Seismol (2025) 29:147–169 Vol.: (0123456789) 4 km ≤ RJB ≤ 54 km) and can be less accurate when applied outside of that range. The red lines in Figs.15 and 16 represent the mean of residuals with respect to the independent variables. The absence of discernible trends in the mean residuals for both between-event and within-event categories across all IMs indicates the unbiased nature of model errors. The extremely narrow confidence intervals, rendering them invisible on the plot, add further weight to this observation. Therefore, the model shows no systematic bias towards Mw and RJB. 7.2 Validation of the ANN-based GMM The model adequacy is attained by evaluating the attenuation pattern of IMs across a range of Mw and RJB values with a Vs30 of 310 m/s, a strike-slip mechanism, and a mean Fd of 10 km. Figure17 illustrates the variation of selected IMs with respect to Mw for different RJB values of 5.0, 20.0, and 50.0 km. IMs acquired from simulated earthquake events are represented by filled gray dots. Additionally, a comparison with KAAH15, a parametric GMM tailored to Iran, is presented for further analysis. Results indicate that higher Mw and lower RJB values correlate with elevated levels of all selected IMs. Similarly, Fig.18 demonstrates the variation of these IMs with RJB, for different Mw values of 7.0, 7.3, and 7.6, alongside a comparison to the KAAH15 GMM. It is observed that as RJB increases, there is a consistent decrease in PGA, PGV, and PSA across all periods, a trend effectively captured by the proposed GMM, highlighting Fig. 16 Distribution of the within-event residual (ϵ) with respect to RJB for ln(PGA), ln(PGV), and ln(PSA) at periods of 0.2 s, 0.5 s, 1.0 s, and 2.0 s 164 J Seismol (2025) 29:147–169 Vol:. (1234567890) Fig. 17 Variation of PGA, PGV, and PSA at periods of 0.2, 0.5, 1.0, and 2.0 s with respect to Mw for a strike-slip fault mechanism, Vs30 = 310 m/s, and RJB values of 5.0, 20.0, and 50.0 km. The proposed model is shown within two standard deviations 165 J Seismol (2025) 29:147–169 Vol.: (0123456789) Fig. 18 Variation of PGA, PGV, and PSA at periods of 0.2, 0.5, 1, and 2 s with respect to RJB for a strike-slip fault mechanism, Vs30 = 310 m/s, and Mw values of 7.0, 7.3, and 7.6 km. The proposed model is shown within two standard deviations 166 J Seismol (2025) 29:147–169 Vol:. (1234567890) its ability to reflect distance-dependent attenuation. Moreover, the relationship between higher Mw and amplified IMs remains consistent with prior observations, as the magnitude increase correlates with a greater energy release during seismic events (Karimzadeh et al. 2023a). The analysis reveals a strong alignment between the proposed ANN-based GMM using simulated data and the empirical GMM derived from real data, notably in PGA and PSA at the period of 0.2 s. Nonetheless, discrepancies become more pronounced for other IMs, particularly noticeable in PSA at longer periods, where the disparity increases. Some of these discrepancies might be attributed to the selection of the model parameters in the simulations as well as the potential limitations of the stochastic finite-fault approach in accurately capturing the longer period content of the records and basin effects. 8 Conclusions This study employs stochastic finite-fault ground motion simulation based on the dynamic corner frequency to simulate alternative scenario events at different magnitude values within the city center of Tabriz, Iran. The effects of diverse hypocenter locations and rupturing fault planes for specific magnitudes are evaluated statistically in detail. The analyses focused on the city center, allowing for the assessment of variations in ground motion intensity resulting from different hypocenter and fault plane configurations. The results are validated against empirical groundmotion models (GMMs), namely BA08 (Boore and Atkinson 2008), AC10 (Akkar and Cagnan 2010), ASB14 (Akkar etal. 2014), and KAAH15 (Kale etal. 2015). The findings reveal the validity of the simulations from a seismological point of view, exhibiting considerable variability in ground motion intensity measures (IMs) across each site of interest. This variability is attributed to uncertainties in the earthquake hypocenter and the fault plane rupture. Moreover, the evaluation of the inter-period correlation of the residual pseudo-spectral acceleration (PSA) serves as an additional step in the validation process. The highest values of all ground motion IMs are consistently observed in the immediate vicinity of the North Tabriz Fault (NTF) line, underscoring the significance of this area in terms of potential hazards. The findings suggest that higher earthquake magnitudes result in higher values of seismic IMs such as peak ground acceleration (PGA), peak ground velocity (PGV), and PSA across different periods. Specifically, comparing the moment magnitude (Mw) of 7.7 with that of 6.8, 7.1, and 7.4 reveals a significant reduction of approximately 65%, 45%, and 15%, respectively, based on the mean average of the IMs. Furthermore, the considerable standard deviations concerning the mean values across various Mw and IMs indicate a broad spectrum of potential groundshaking levels. The varying seismic intensity experienced during earthquakes of similar magnitudes within the region underscores the significant hazard faced by the city center, especially for the highest Mw of 7.7, stressing the importance of preparedness. Estimations of Modified Mercalli intensity (MMI) values across various scenarios further highlight the increased vulnerability of Tabriz city center to significant potential hazards. A local ground motion model (GMM) is developed by training artificial neural networks (ANN) on simulated ground motions. The developed ANNbased GMM incorporates moment magnitude (Mw), Joyner-Boore distance (RJB), and focal depth (Fd) as input parameters, enabling the estimation of ln(PGA), ln(PGV), and ln(PSA) at various periods. Through residual analysis, the uncertainty of the GMM is quantified in terms of between-event and within-event variations. Comparison with an empirical GMM tailored to Iran reveals the robust performance of the proposed GMM in accurately capturing observed patterns in ground motions despite some discrepancies for spectral ordinates at longer periods. This might be attributed to the potential limitations of the stochastic finite-fault approach in accurately capturing the longer period content of the records and basin effects. Moreover, an online platform is offered to ensure a user-friendly application of the developed GMM for end-users. The analyses in this study are performed with uniform soil conditions. Future studies should aim to address the uncertainty stemming from input parameters, particularly site parameters along with the simulation methodology further to enhance our understanding of seismic hazards in the region. Additionally, future studies could explore the integration of these findings with vulnerability and exposure data to develop a comprehensive risk assessment framework. 167 J Seismol (2025) 29:147–169 Vol.: (0123456789) Acknowledgements This work was partly financed by FCT / MCTES through national funds (PIDDAC) under the R&D Unit Institute for Sustainability and Innovation in Structural Engineering (ISISE), under reference UIDB / 04029/2020 (doi. org/https:// doi. org/ 10. 54499/ UIDB/ 04029/ 2020), and under the Associate Laboratory Advanced Production and Intelligent Systems (ARISE) under reference LA/P/0112/2020. This study has been partly funded by the STAND4HERITAGE project that has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (Grant Agreement No. 833123), as an Advanced Grant. This work is partly financed by national funds through FCT (Foundation for Science and Technology), under grant agreement UI/BD/153379/2022 attributed to the second author. Author contribution C. Temiz: Formal analysis, Data curation, Investigation, Resources, Visualisation, Writing—original draft, Writing—review & editing. S. M. S. Hussaini: Formal analysis, Investigation, Visualisation, Writing—original draft, Writing—review & editing. S. Karimzadeh: Conceptualisation, Data curation, Formal analysis, Investigation, Methodology, Resources, Supervision, Validation, Writing—original draft, Writing—review & editing. A. Askan: Resources, Writing—review & editing. P. B. Lourenço: Funding acquisition, Resources, Supervision, Writing—review & editing. Funding Open access funding provided by FCT|FCCN (b-on). Data availability This study uses the Streamlit package in Python to construct a user-friendly graphical interface tool, facilitating convenient access to the ANN-based ground motion model (GMM). The code can be found at https://github.com/SM-S–H/Tabriz-GMM-ANN, while the interface tool itself is accessible via https://tabriz-gmm-ann.streamlit.app/. Users are prompted to input the parameters of a scenario earthquake, such as moment magnitude (Mw), Joyner-Boore distance (RJB), and focal depth (Fd). The software then generates outcomes in terms of intensity measures (IMs), encompassing peak ground acceleration (PGA), peak ground velocity (PGV), and pseudo spectral acceleration (PSA). Finally, all data supporting this paper will be made available upon request to the corresponding author. Declarations Competing Interests The authors declare no competing interests. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. References Abrahamson NA, Youngs RR (1992) A stable algorithm for regression analyses using the random effects model. Bull Seismol Soc Am 82(1):505–510 Akkar S, Cagnan Z (2010) A local ground-motion predictive model for Turkey, and Its comparison with other regional and global ground-motion models. Bull Seismol Soc Am 100(6):2978–2995 Akkar S, Sandıkkaya MA, Bommer JJ (2014) Empirical ground-motion models for pointand extended-source crustal earthquake scenarios in Europe and the Middle East. Bull Earthq Eng 12(1):359–387 (Kluwer Academic Publishers) Altindal A, Askan A (2023) Construction and validation of a simulated ground motion dataset for Turkey. J Seismol 2023:1–19 (Springer) Amiranlou H, Pourkermani M, Dabiri R etal (2016) The stochastic finite-fault modeling based on a dynamic corner frequency simulating of strong ground motion for earthquake scenario of north Tabriz fault. Open J Earthq Res 05(02):114–121 Anderson JG, Hough SE (1984) A model for the shape of the fourier amplitude spectrum of acceleration at high frequencies. Bull Seismol Soc Am 74(5):1969–1993 Arora S, Joshi A, Kumari P etal (2020) Strong ground motion simulation techniques—a review in world context. Arab J Geosci 13(14):1–12 (Springer) ArslanKelam A, Karimzadeh S, Yousefibavil K et al (2022) An evaluation of seismic hazard and potential damage in Gaziantep, Turkey using site specific models for sources, velocity structure and building stock. Soil Dyn Earthq Eng 154:107129 Askan A, Altindal A, Aydin MF etal (2024) Assessment of urban seismic resilience of a town in Eastern Turkiye: Turkoglu, Kahramanmaras before and after M7.8 Kahramanmaras earthquake, Earthquake spectra.https:// doi. org/ 10. 1177/ 87552 93024 12747 15 Atkinson GM, Goda K, Assatourians K (2011) Comparison of nonlinear structural responses for accelerograms simulated from the stochastic finite-fault approach versus the hybrid broadband approach. Bull Seismol Soc Am 101(6):2967–2980 (GeoScienceWorld) Azarafza M, Ghazifard A (2016) Urban geology of Tabriz City: environmental and geological constraints. Adv Environ Res 5(2):95–108 (Techno‑Press) Baker JW (2011) Conditional mean spectrum: tool for groundmotion selection. J Struct Eng 137(3):322–331 Berberian M, and Arshadi S (1976) On the eviderne of the youngest activity of the North Tabriz Fault and the seismicity of Tabriz City, Geological Survey of Iran, Tehran. Iran Rep 39:397–418 168 J Seismol (2025) 29:147–169 Vol:. (1234567890) Berberian M, Yeats RS (1999) Patterns of Historical Earthquake Rupture in the Iranian Plateau. Bulletin of the Seismological Society of America89(1):120–139.https:// doi. org/ 10. 1785/ BSSA0 89001 0120 Beresnev IA, Atkinson GM (1997) Modeling finite-fault radiation from the ω n spectrum. Bull Seismol Soc Am 87(1):67–84 (The Seismological Society of America) Bilal M, Askan A (2014) Relationships between felt intensity and recorded ground-motion parameters for Turkey. Bull Seismol Soc Am 104(1):484–496 Boore DM (1983a) Stochastic simulation of high-frequency ground motions based on seismological models of the radiated spectra. Bull Seismol Soc Am 73(6A):1865–1894 Boore DM (1983b) Strong-motion seismology. Rev Geophys 21(6):1308–1318 Boore DM (2009) Comparing stochastic point-source and finite-source ground-motion simulations: SMSIM and EXSIM. Bull Seismol Soc Am 99(6):3202–3216 Boore DM, Atkinson GM (2008) Ground-motion prediction equations for the average horizontal component of PGA, PGV, and 5%-damped PSA at spectral periods between 0.01 s and 10.0 s. Earthq Spectra 24(1):99–138 (Earth‑ quake Engineering Research Institute) Boore DM, Joyner WB (1997) Site Amplifications for Generic Rock Sites. Bulletin of the Seismological Society of America 87(2):327–341. https:// doi. org/ 10. 1785/ BSSA0 87002 0327 Burks LS, Baker JW (2014) Validation of ground-motion simulations through simple proxies for the response of engineered systems. Bull Seismol Soc Am 104(4):1930–1946 (GeoScienceWorld) Can G, Askan A, Karimzadeh S (2021) An assessment of the 3 February 2002 Cay (Turkey) earthquake (Mw=6.6): modeling of ground motions and felt intensity distribution. Soil Dyn Earthq Eng 150:106832 Dhanya J, Raghukanth STG (2018) Ground motion prediction model using Artificial Neural Network. Pure Appl Geophys 175(3):1035–1064 (Birkhauser Verlag AG) Code Price (2005) Eurocode 8: Design of structures for earthquake resistance-part 1: general rules, seismic actionsand rules for buildings. European Committee for Standardization, Brussels, p 10 Eslami M, Taghizadeh-Farahmand F (2020) Evaluation of the deterministic seismic hazard by using fuzzy inference system, case study: Tabriz City, Iran. Iran J Geophys (IJG) 14(4):19–28 Farahani JV (2018) Statistical properties of aftershocks for Ahar-Varzeghan twin earthquakes on 11 August 2012, NW Iran, and investigation of seismicity of north Tabriz fault. Int J Geosci 09(02):106–118 Frankel A (1993) Three-dimensional simulations of ground motions in the San Bernardino Valley, California, for hypothetical earthquakes on the San Andreas fault. Bull Seismol Soc Am 83(4):1020–1041 Ghasemi M, Karimzadeh S, Matsuoka M etal (2021) What would happen if the M 7.3 (1721) and M 7.4 (1780) historical earthquakes of Tabriz City (NW Iran) occurred again in 2021? ISPRS Int J Geo-Inf 10(10):657 Ghayamghamian MR, Rajool A (2012) Long-period ground motion simulation for NTF fault near-source energy released. Proc Fifteenth World Conf Earthq Eng, Lisbon, Portugal 2012:23–25 Ghorbani M (2021) Faults and tectonic phases of Iran, pp. 81–149. https:// doi. org/ 10. 1007/ 978-303071109-2_3 Hancock J, Watson-Lamprey J, Abrahamson NA etal (2006) An improved method of matching response spectra of recorded earthquake ground motion using wavelets. J Earthq Eng 10(sup001):67–89 (World Scientific) Hariri-Ardebili MA, Rezaeian S (2024) Utilization of stochastic ground motion simulations for scenario-based performance assessment of geo-structures. Reliab Eng Syst Saf 251:110375 (Elsevier) Herrmann RB (1985) An extension of random vibration theory estimates of strong ground motion to large distances. Bull Seismol Soc Am 75(5):1447–1453 Hessami K, Pantosti D, Tabassi H et al (2003) Paleoearthquakes and slip rates of the NorthTabriz Fault, NW Iran: preliminary results. Ann Geophys 45(5). https:// doi. org/ 10. 4401/ ag3461 Hessami K, Pantosti D, Tabassi H et al (2009) Paleoearthquakes and slip rates of the North Tabriz Fault, NW Iran: preliminary results. Ann Geophys 46(5). https:// doi. org/ 10. 4401/ ag3461 Hoveidae N, Fathi A, Karimzadeh S (2021) Seismic damage assessment of a historic masonry building under simulated scenario earthquakes: a case study for Arge-Tabriz. Soil Dyn Earthq Eng 147:106732 (Elsevier) JafarianVernosfaderani SL, Heidari R, Mirzaei N etal (2019) Coda wave attenuation in the northwestern Iran, using short time Fourier transform. J Seismol 23(5):1085–1095 (Springer Netherlands) Kale Ö, Akkar S, Ansari A etal (2015) A ground-motion predictive model for Iran and Turkey for horizontal PGA, PGV, and 5% damped response spectrum: investigation of possible regional effects. Bull Seismol Soc Am 105(2A):963–980 Kamae K, Irikura K, Pitarka A (1998) A technique for simulating strong ground motion using hybrid Green’s function. Bull Seismol Soc Am 88(2):357–367 Karakhanian AS, Trifonov VG, Philip H etal (2004) Active faulting and natural hazards in Armenia, eastern Turkey and northwestern Iran. Tectonophysics 380(3–4):189–219 Karimzadeh S, Askan A (2018) Modeling of a historical earthquake in Erzincan, Turkey (Ms~7.8, in 1939) using regional seismological information obtained from a recent event. Acta Geophys 66(3):293–304 Karimzadeh S, Miyajima M, Hassanzadeh R etal (2014) A GIS-based seismic hazard, building vulnerability and human loss assessment for the earthquake scenario in Tabriz. Soil Dyn Earthq Eng 66:263–280 Karimzadeh S, Feizizadeh B, Matsuoka M (2017) From a GIS-based hybrid site condition map to an earthquake damage assessment in Iran: methods and trends. Int J Disaster Risk Reduction 22:23–36 Karimzadeh S, Feizizadeh B, Matsuoka M (2019) DEMbased Vs30 map and terrain surface classification in nationwide scale—a case study in Iran. ISPRS Int J Geo Inf 8(12):537 Karimzadeh S, Mohammadi A, Hussaini SMS etal (2023) ANN-based ground motion model for Turkey using stochastic simulation of earthquakes. Geophys J Int 236(1):413–429 (Oxford Academic) 169 J Seismol (2025) 29:147–169 Vol.: (0123456789) Karimzadeh S, Funari MF, Szabó S etal (2024) Stochastic simulation of earthquake ground motions for the seismic assessment of monumental masonry structures: sourcebased vs site-based approaches. Earthq Eng Struct Dyn 53(1):303–330 (John Wiley & Sons) Karimzadeh S, Mohammadi A, Salahuddin U et al (2023) Backbone ground motion model through simulated records and XGBoost machine learning algorithm: an application for the Azores plateau (Portugal). Earthq Eng Struct Dyn. John Wiley & Sons, Ltd. https:// doi. org/ 10. 1002/ EQE. 4040 Karimzadeh S (2019) Seismological and engineering demand misfits for evaluating simulated ground motion records. Appl Sci (Switzerland) 9(21). MDPI AG. https:// doi. org/ 10. 3390/ app92 14497 Kheirizadeh Arouq M, Esmaeilpour M, Sarvar H (2020) Vulnerability assessment of cities to earthquake based on the catastrophe theory: a case study of Tabriz city, Iran. Environ Earth Sci 79(14). Springer. https:// doi. org/ 10. 1007/ s1266502009103-2 Khosravikia F, Clayton P (2021) Machine learning in ground motion prediction. Comput Geosci 148:104700 (Pergamon) Marquardt DW (1963) An algorithm for least-squares estimation of nonlinear parameters. J Soc Ind Appl Math 11(2):431–441 Mohammadi A, Karimzadeh S, Valizadeh Kamran K et al (2020a) Extraction of land information, future landscape changes and seismic hazard assessment: a case study of Tabriz, Iran. Sensors 20(24):7010 Mohammadi A, Karimzadeh S, Kamran KV et al (2020) Extraction of land information, future landscape changes and seismic hazard assessment: a case study of tabriz, iran. Sensors (Switzerland) 20(24):1–27 (MDPI AG) Mohammadi A, Karimzadeh S, Amir S et al (2023) The potential of region-specific machine-learning-based ground motion models: application to Turkey. Soil Dyn Earthq Eng 172(January):108008 (Elsevier) Mohammadioun B, Serva L (2001) Stress drop, slip type, earthquake magnitude, and seismic hazard. Bull Seismol Soc Am 91(4):694–707 (GeoScienceWorld) Moradi AS, Hatzfeld D, Tatar M (2011) Microseismicity and seismotectonics of the North Tabriz fault (Iran). Tectonophysics 506(1–4):22–30 Motazedian D, Atkinson GM (2005) Stochastic finite-fault modeling based on a dynamic corner frequency. Bull Seismol Soc Am 95(3):995–1010 Ozsarac V, Karimzadeh S, Askan A et al (2021) Seismic demands of bare and base-isolated steel frames for real against simulated records of a past earthquake. 18(9):1266–1281.https:// doi. org/ 10. 1080/ 15732 479. 2021. 18952 27 Taylor & Francis Pedregosa F, Varoquaux G, Gramfort A et al (2011) Scikitlearn: machine learning in Python. J Mach Learn Res 12(85):2825–2830 Rezaeian S, Stewart JP, Luco N etal (2024) Findings from a decade of ground motion simulation validation research and a path forward. Earthq Spectra 40(1):346–378 (SAGE Publications Inc.) Rezaeian S, Sun X (2014) Stochastic ground motion simulation. Encyclopedia of Earthquake Engineering. Berlin, Heidelberg: Springer Berlin Heidelberg: 1–15. https:// doi. org/ 10. 1007/ 978-364236197-5_ 239-1 Smerzini C, Amendola C, Paolucci R etal (2024) Engineering validation of BB-SPEEDset, a data set of near-source physics-based simulated accelerograms. Earthq Spectra 40(1):420–445 (SAGE Publications Inc.) Solaymani Azad S, Philip H, Dominguez S etal (2015) Paleoseismological and morphological evidence of slip rate variations along the north tabriz fault (NW Iran). Tectonophysics 640:20–38 (Elsevier) Tanırcan G, Yelkenci-Necmioğlu S (2020) Simulation of the strong ground motion for the 20 July 2017 (Mw. 6.6) Bodrum-Kos earthquake. Bull Earthq Eng 18(13):5807–5825 Tsioulou A, Taflanidis AA, Galasso C (2019) Validation of stochastic ground motion model modification by comparison to seismic demand of recorded ground motions. Bull Earthq Eng 17(6):2871–2898 (Springer Netherlands) VahidiFard H, Zafarani H, Sabbagh-Yazdi SR etal (2017) Seismic hazard analysis using simulated ground-motion time histories: the case of the Sefidrud dam, Iran. Soil Dyn Earthq Eng 99:20–34 Watson-Lamprey J, Abrahamson N (2006) Selection of ground motion time series and limits on scaling. Soil Dyn Earthq Eng 26(5):477–482 Wells DL, Coppersmith KJ (1994) New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement. Bull Seismol Soc Am 84(4):974–1002 (GeoScienceWorld) Withers KB, Moschetti MP, Thompson EM (2020) A machine learning approach to developing ground motion models from simulated ground motions. Geophys Res Lett 47(6):e2019GL086690 (John Wiley & Sons, Ltd) Xu J, Feng DC (2018) Seismic response analysis of nonlinear structures with uncertain parameters under stochastic ground motions. Soil Dyn Earthq Eng 111:149–159 (Elsevier) Yaghmaei-Sabegh S, Hassani B (2020) Investigation of the relation between Vs30 and site characteristics of Iran based on horizontal-to-vertical spectral ratios. Soil Dyn Earthq Eng 128:105899 Yamamoto Y, Baker JW (2013) Stochastic model for earthquake ground motion using wavelet packets. Bull Seismol Soc Am 103(6):3044–3056 Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.