Traditional seismic hazard analyses underestimate hazard levels when compared to observations from the 2023 Kahramanmaras earthquakes

On February 6, 2023, two significant earthquakes with moment magnitudes (Mw) of 7.8 and 7.5 struck Southeastern Turkey near Pazarcik and Elbistan. The region's high seismic hazard was well-known due to historical seismicity and numerous active faults, including segments of the East Anatolian Fault Zone (EAFZ) and Dead Sea Fault Zone (DSFZ). While many studies indicated high seismic potential in the area, particularly a seismic gap near Kahramanmaras, the estimated maximum magnitudes were generally lower than the observed values. The earthquakes ruptured multiple fault segments, an occurrence not fully anticipated by previous studies that often provided separate maximum magnitude estimates for individual segments. This study aims to quantitatively assess whether this earthquake sequence was an exceptionally rare event by focusing on parameters with reliable empirical prediction models, specifically estimating exceedance probabilities for earthquake magnitudes and ground motion amplitudes.

Numerous studies highlighted the high seismic potential of the southern EAFZ, suggesting a seismic gap near Kahramanmaras. However, estimated maximum magnitudes were typically lower than the observed Mw 7.8 and 7.5. Studies like Emre et al. (2018) provided detailed fault segmentations and maximum magnitude estimates for each segment, while others utilized GPS data (Aktuğ et al., 2016) or focused on seismotectonics (Güvercin et al., 2022) to assess seismic potential. Probabilistic seismic hazard assessments (PSHA) for the region (Gülerce et al., 2017) considered multi-segment ruptures but often underestimated the combined effect of the observed ruptures. Existing models primarily focused on individual segment ruptures rather than the combined multi-segment rupture that occurred.

This study utilizes a dataset of recorded ground motions from Gülerce et al. (2023), derived from data published by the Disaster and Emergency Management Presidency of Turkey (AFAD). After quality control, the dataset included 51 records for the Pazarcik event and 25 for the Elbistan event. Three ground motion intensity measures were calculated: peak ground acceleration (PGA), and 5% damped elastic spectral acceleration at periods of 0.5 and 1.0 seconds (SA(T=0.5) and SA(T=1.0)). Probabilistic Seismic Hazard Assessment (PSHA) was performed using the open-source OpenQuake platform and inputs from the European Seismic Hazard Model 2020 (ESHM20). ESHM20's source model, including area and fault sources, was employed, considering sources within a 200km radius. The logic tree of ESHM20 accounts for epistemic uncertainty in magnitude-frequency distributions (MFDs), using double truncated Gutenberg-Richter and tapered Pareto distributions. Ground motion characterization in ESHM20 uses a scaled backbone logic tree approach with the Kotha et al. (2020) partially non-ergodic GMM. Mean return periods for observed ground motion levels were calculated from the hazard curves generated for each seismic station. Recorded ground motions were also compared to the Boore et al. (2014) GMM to analyze residuals and assess the extremity of observed ground motions relative to model predictions. Multivariate normal distribution was used to calculate the probability of observing ground motion as extreme as observed at specific periods. Finally, the recorded ground motions were compared to the most extreme ground motions from the NGA-West2 database.

The study found that the mean return period for the Mw 7.8 Pazarcik earthquake was estimated to be approximately 1000 years using the fault source model (TRCF002), while the area sources (TRAS481 and LBAS341) yielded estimates of 4000 and 3500 years, respectively. For the Mw 7.5 Elbistan earthquake, the fault source model (TRCF03H) estimated a return period of about 7800 years, while the area source model (TRAS434) estimated 2600 years. The maximum estimated return periods for exceeding the observed ground motion levels during the Pazarcik earthquake were approximately 6900, 4100, and 6900 years for PGA, SA(T=0.5), and SA(T=1.0), respectively. For the Elbistan earthquake, the maximum estimates were 1200, 1700, and 6900 years for PGA, SA(T=0.5), and SA(T=1.0). Analysis of residuals using the Boore et al. (2014) GMM revealed no apparent trend for the Pazarcik earthquake, indicating general agreement with empirical models, except for some stations with very high residuals across various periods. For the Elbistan earthquake, a positive trend was observed for residuals at longer periods, indicating significant underestimation. Two stations recorded spectral accelerations exceeding the maximum values in the NGA-West2 database. The probability of observing ground motion at least as extreme as that observed at station 3135 (Arsuz, Hatay) was calculated to be approximately 0.0005 (periods 0.3 and 3.0 s) and 0.0086 (periods 0.5 and 1.0 s), indicating unusually high amplitudes.

The study demonstrates that while the seismic potential of the region was known, traditional seismic hazard analyses underestimated the observed magnitudes and ground motion intensities. The combined rupture of the Amanos and Pazarcik segments, not considered in previous studies, contributed significantly to the event's severity. The higher return periods estimated for Hatay might be attributed to both directivity effects and denser seismic network coverage. The discrepancies between return periods estimated from fault and area sources highlight the challenges in modeling complex fault systems accurately. The high return periods, exceeding those for which structures are designed, highlight the significant demands placed on buildings. However, these high return periods should not fully justify the damage observed, as these represent local peaks rather than average shaking.

This study underscores the limitations of traditional PSHA in accurately predicting the severity of complex, multi-segment ruptures. The observed ground motions, particularly the high amplitudes recorded in Hatay, highlight the need for improved ground motion models that incorporate physics-based simulations of earthquake sequences and non-ergodic effects. Future research should focus on enhancing ground motion models, particularly for near-fault, large-magnitude events, and incorporate physics-based earthquake simulations to better account for complex fault interactions and triggering. The limitations of the PSHA methodology, particularly in validating long return periods, also necessitate ongoing refinement.

The study's reliance on existing ground motion models derived from datasets lacking sufficient large-magnitude, near-fault recordings is a key limitation. The relatively short instrumental record in the region contributes to uncertainty in return period estimates. The assumption of statistical independence between the Pazarcik and Elbistan earthquakes in the PSHA analysis is also a simplification given the likely triggering mechanism. The high variability and sensitivity of results to different input combinations (logic tree branches) also emphasize the inherent uncertainty.