Domniki Asimaki

This study presents the development and application of a scalable non-ergodic ground motion model (NGMM) for the Los Angeles area. The NGMM is trained and validated on physics-based simulated ground-motion data from a recent Statewide California Earthquake Center (SCEC) CyberShake study. The NGMM is formulated as a Gaussian Process (GP) regression model, where the prior median is defined as the ASK14 ergodic ground-motion model and the posterior median is obtained by learning the non-ergodic effects embedded in the training data. These non-ergodic effects include systematic site and path effects, which are represented in the GP using Matérn and specialized covariance kernels that explicitly characterize path vectors. Implementing the NGMM requires hyperparameter tuning and inference on large datasets (on the order of one million data points or more), posing significant computational challenges for conventional GP approaches. To address this scalability issue, this paper presents a suite of computational strategies, including sparse Cholesky inversion, parallel computing, GPU acceleration, and stochastic gradient descent minimization. Despite these advances, the full CyberShake dataset (on the order of hundreds of millions of data points) remains computationally prohibitive. Therefore, aleatory variability is modeled separately using a mixed-effects formulation to represent within-event and between-event variability. The developed NGMM has two primary applications: interpolation of partially observed ground-motion fields and predictive modeling for ground motions in unobserved earthquake scenarios. Validation results on independent datasets demonstrate accurate performance in both applications. A case study of power transmission network assessment in an Mw 6.7 Puente Hill scenario further demonstrated that the developed NGMM closely reproduces physics-based simulation results.

This study presents the development and application of a scalable non-ergodic ground motion model (NGMM) for the Los Angeles area. The NGMM is trained and validated on physics-based simulated ground-motion data from a recent Statewide California Earthquake Center (SCEC) CyberShake study. The NGMM is formulated as a Gaussian Process (GP) regression model, where the prior median is defined as the ASK14 ergodic ground-motion model and the posterior median is obtained by learning the non-ergodic effects embedded in the training data. These non-ergodic effects include systematic site and path effects, which are represented in the GP using Matérn and specialized covariance kernels that explicitly characterize path vectors. Implementing the NGMM requires hyperparameter tuning and inference on large datasets (on the order of one million data points or more), posing significant computational challenges for conventional GP approaches. To address this scalability issue, this paper presents a suite of computational strategies, including sparse Cholesky inversion, parallel computing, GPU acceleration, and stochastic gradient descent minimization. Despite these advances, the full CyberShake dataset (on the order of hundreds of millions of data points) remains computationally prohibitive. Therefore, aleatory variability is modeled separately using a mixed-effects formulation to represent within-event and between-event variability. The developed NGMM has two primary applications: interpolation of partially observed ground-motion fields and predictive modeling for ground motions in unobserved earthquake scenarios. Validation results on independent datasets demonstrate accurate performance in both applications. A case study of power transmission network assessment in an Mw 6.7 Puente Hill scenario further demonstrated that the developed NGMM closely reproduces physics-based simulation results.

This data repository contains the code and data produced for the manuscript titled "Scalable Gaussian Process for Learning Non-Ergodic Ground Motion Model from Physics-Based Simulations with Application to Power Infrastructure Assessment". Below is the abstract of the manuscript. This study presents the development and application of a scalable non-ergodic ground motion model (NGMM) for the Los Angeles area. The NGMM is trained and validated on physics-based simulated ground-motion data from a recent Statewide California Earthquake Center (SCEC) CyberShake study. The NGMM is formulated as a Gaussian Process (GP) regression model, where the prior median is defined as the ASK14 ergodic ground-motion model and the posterior median is obtained by learning the non-ergodic effects embedded in the training data. These non-ergodic effects include systematic site and path effects, which are represented in the GP using Matérn and specialized covariance kernels that explicitly characterize path vectors. Implementing the NGMM requires hyperparameter tuning and inference on large datasets (on the order of one million data points or more), posing significant computational challenges for conventional GP approaches. To address this scalability issue, this paper presents a suite of computational strategies, including sparse Cholesky inversion, parallel computing, GPU acceleration, and stochastic gradient descent minimization. Despite these advances, the full CyberShake dataset (on the order of hundreds of millions of data points) remains computationally prohibitive. Therefore, aleatory variability is modeled separately using a mixed-effects formulation to represent within-event and between-event variability. The developed NGMM has two primary applications: interpolation of partially observed ground-motion fields and predictive modeling for ground motions in unobserved earthquake scenarios. Validation results on independent datasets demonstrate accurate performance in both applications. A case study of power transmission network performance assessment in an Mw 6.7 Puente Hill scenario further demonstrated that the developed NGMM closely reproduces physics-based simulation results at a substantially reduced computational cost, and that neglecting non-ergodic effects in ground motion modeling can bias damage estimates by nearly a factor of two.

Earthquake hazard analysis and design of spatially distributed infrastructure, such as power grids and energy pipeline networks, require scenario-specific ground-motion time histories with realistic frequency content and spatiotemporal coherence. However, producing the large ensembles needed for uncertainty quantification with physics-based simulations is computationally intensive and impractical for engineering workflows. To address this challenge, we introduce Ground-Motion Flow (GMFlow), a physics-inspired latent operator flow matching framework that generates realistic, large-scale regional ground-motion time-histories conditioned on physical parameters. Validated on simulated earthquake scenarios in the San Francisco Bay Area, GMFlow generates spatially coherent ground motion across more than 9 million grid points in seconds, achieving a 10,000-fold speedup over the simulation workflow, which opens a path toward rapid and uncertainty-aware hazard assessment for distributed infrastructure. More broadly, GMFlow advances mesh-agnostic functional generative modeling and could potentially be extended to the synthesis of large-scale spatiotemporal physical fields in diverse scientific domains.

Earthquake hazard analysis and design of spatially distributed infrastructure, such as power grids and energy pipeline networks, require scenario-specific ground-motion time histories with realistic frequency content and spatiotemporal coherence. However, producing the large ensembles needed for uncertainty quantification with physics-based simulations is computationally intensive and impractical for engineering workflows. To address this challenge, we introduce Ground-Motion Flow (GMFlow), a physics-inspired latent operator flow matching framework that generates realistic, large-scale regional ground-motion time-histories conditioned on physical parameters. Validated on simulated earthquake scenarios in the San Francisco Bay Area, GMFlow generates spatially coherent ground motion across more than 9 million grid points in seconds, achieving a 10,000-fold speedup over the simulation workflow, which opens a path toward rapid and uncertainty-aware hazard assessment for distributed infrastructure. More broadly, GMFlow advances mesh-agnostic functional generative modeling and could potentially be extended to the synthesis of large-scale spatiotemporal physical fields in diverse scientific domains.

This is a collection of all data and data processing and plotting scripts used in the manuscript "Investigating a subsurface model for the 2.4 Hz resonance in InSight data" in preparation for submission to the journal Journal of Geophysical Research: Planets. It consists of processed data from the InSight SEIS VBB seismometer, as well as the Pressure Sensor and TWINS wind sensor, as well as Matlab and Python codes to produce the processed data and plot results. A full description is available in the file README.md which is in the base directory (InSightSubsurface2.4) of the zip file InSightSubsurface2.4.zip

ABSTRACT We analyze the seismic response of a class of fragile geologic features (FGFs), referred to as rock towers (RTs) at the Trona Pinnacles, a group of RTs in southern California that suffered strong shaking during the 2019 Ridgecrest earthquake sequence. FGFs, including RTs, may provide maximum constraints on past earthquake shaking intensity, and thereby support probabilistic seismic hazard assessments (PSHAs). In a rare case study, we explore the hypothesis that RT structural integrity is time dependent, as damage accumulates progressively. We develop finite-element method (FEM) models of the RTs using photogrammetric shape models. We validate the models by comparing numerical simulations of their response to broadband ground shaking with low-intensity seismic recordings obtained at the Pinnacles. Results of our simulations are in good agreement with the seismic recordings of actual earthquake aftershocks. We next use the results of the FEM models to analyze the response and evolution of RTs. Our analyses elucidate the influence of geometry over their seismic response, providing a rationale that may explain the rarity of slender RTs at Trona: high-aspect-ratio structures that respond in bending develop detrimental tensile stresses that crack the rock, whereas low-aspect-ratio ones’ response also includes shearing, which does not compromise material integrity as much as tension. Field measurements with a rebound hammer support this finding, suggesting that the material around the base of slender rocks has been weakened relative to other parts of the RT. We also study how to define simplified mechanical models (“archetypes”) to predict the natural frequencies of RTs. Results from our work illuminate the fundamental mechanisms of seismic response and progressive failure of RTs, and open new avenues of research to potentially incorporate these geologic features as long-return period constraints on PSHA, in ways analogous to those of the widely used precariously balanced rocks.

Abstract Strike-slip faults—where tectonic plates grind past each other horizontally—are a defining feature of many densely populated continental seismic zones worldwide, including the San Andreas fault system in California, the North and East Anatolian faults in Türkiye, and the Sagaing fault in Myanmar (Burma). Although their lateral motion has long been recognized, a growing body of global evidence is now highlighting a more hazardous aspect of these systems: supershear earthquakes—fast propagating ruptures that exceed the speed of shear waves and can cause disproportionately intense shaking and destruction. Four of the last six Mw 7.0+ earthquakes on strike-slip faults have been recognized as supershear events, including the damaging Mw 7.7 Myanmar and the Mw 7.8 Pazarcik earthquakes, highlighting the need to confront the potential implications of such future events.

Abstract Observations of the 2019 magnitude 7.1 Ridgecrest, California, earthquake indicate a relatively slow rupture (2 km/s). The fault is surrounded by sedimentary rocks and low‐velocity damage zones, which can amplify ground motions but also slow down rupture. Here, we develop 3D dynamic rupture models to elucidate the significance of such effects on the Ridgecrest earthquake. We find that: (a) sedimentary rocks and damage, being shallow, do not explain the slow rupture but amplify slip and ground motion by more than a factor of 3; (b) accounting for ground motion amplification by sedimentary rocks improves the agreement with empirical predictions; (c) damage zone contributions to surface slip are minor (5%) for this event but could reach 25% in future southern California earthquakes. Our study corroborates the significance of source and site effects due to heterogeneous near‐fault materials during the Ridgecrest earthquake, and provides insights for future rupture and source‐to‐site hazard modeling efforts.

This study presents the formulation of two new sedimentary velocity models (SVMs) applied to the San Francisco Bay Area (SFBA) to improve the near‐surface representation of shear‐wave velocity () for large‐scale, broadband numerical simulations, with the ultimate goal of enhancing the representation of the sedimentary layers in community velocity model. The first velocity model is stationary and is based solely on the time‐average shear‐wave velocity of the top 30 m (); the second velocity model is spatially varying and has location‐specific adjustments. They were developed using a dataset of 200 measured profiles. Both models were formulated within a hierarchical Bayesian framework, using a parameterization that ensures robust scaling. The spatially varying model includes a slope adjustment term modeled as a Gaussian process to capture site‐specific effects based on location. Residual analysis shows that both models are unbiased for values up to 1000 m/s. Along‐depth variability models were also developed using within‐profile residuals. The proposed models show higher in the South Bay, East Bay, and Livermore Valley compared to the USGS SFBA velocity model by a factor of two or more in some cases. Goodness‐of‐fit (GOF) comparisons using one‐dimensional (1D) linear site response analysis at selected sites demonstrate that the proposed models outperform the USGS SFBA velocity model in capturing near‐surface amplification across a broad frequency range. Incorporating along‐depth variability further improves the GOF scores by reducing over‐amplification at high frequencies. These results underscore the importance of integrating data‐driven models of the shallow crust, like the ones presented here, in coarser regional community velocity models to enhance regional seismic hazard assessments.

Expanding on neural operators, we propose a novel framework for stochastic process learning across arbitrary domains. In particular, we develop operator flow matching (OFM) for learning stochastic process priors on function spaces. OFM provides the probability density of the values of any collection of points and enables mathematically tractable functional regression at new points with mean and density estimation. Our method outperforms state-of-the-art models in stochastic process learning, functional regression, and prior learning.

Abstract Field investigations have long been an important component of the scientific response to surface-faulting earthquakes. However, in light of advances in remote data and models, the question arises whether field-based observations of surface rupture remain useful for understanding rupture processes and seismic hazards. We approach this question using a field-based study of the central 2023 Mw 7.8 Pazarcık, Turkey (Türkiye), earthquake rupture, at the intersection of the east Anatolian fault (EAF) and Narlı fault. Our field observations include the surface rupture expression and extent of the central EAF and northernmost Narlı fault in generally forested and steep terrain and 68 measurements of left-lateral surface displacement. These data improve our understanding of the Pazarcık rupture complexity, resolve the surface geometry of the Narlı fault–EAF connection, and exhibit a clear (>2 m) change in surface displacement across this intersection zone that confirms remote-based coseismic slip models. Our study shows that focusing field efforts in areas of obscured or low-resolution remote data can yield essential data for refining rupture extent, documenting perishable on-fault displacement, and improving postearthquake situational awareness. A comparison of similarly large-magnitude continental surface-rupturing earthquakes indicates that displacement uncertainties relate to a complex set of factors, including measurement methods, rupture complexity, and displacement magnitude. Our study validates the need for postearthquake field observations, which, when driven by clear motivating questions and knowledge of methodological strengths and limitations, provide high-resolution rupture data that complement remote-based models.

This study presents the development of two new sedimentary velocity models for the San Francisco Bay Area (SFBA) to improve the near-surface representation of shear-wave velocity ($V_S$) for large-scale, broadband numerical simulations, with the ultimate goal of enhancing the representation of the sedimentary layers in the Bay Area community velocity model. The first velocity model is stationary and is based solely on $V_{S30}$; the second velocity model is spatially varying and has location-specific adjustments. They were developed using a dataset of 200 measured $V_S$ profiles. Both models were formulated within a hierarchical Bayesian framework, using a parameterization that ensures robust scaling. The spatially varying model includes a slope adjustment term modeled as a Gaussian process to capture site-specific effects based on location. Residual analysis shows that both models are unbiased for $V_S$ values up to 1000 m/sec. Along-depth variability models were also developed using within-profile residuals. The proposed models show higher $V_S$ in the San Jose area and Livermore Valley compared to the USGS Bay Area community velocity model by a factor of two or more in some cases. Goodness-of-fit (GOF) comparisons using one-dimensional linear site-response analysis at selected sites demonstrate that the proposed models outperform the USGS model in capturing near-surface amplification across a broad frequency range. Incorporating along-depth variability further improves the GOF scores by reducing over-amplification at high frequencies. These results underscore the importance of integrating data-driven models of the shallow crust, like the ones presented here, in coarser regional community velocity models to enhance regional seismic hazard assessments.

Abstract The analysis of soil–structure interaction (SSI) problems has been established successfully in recent decades. In particular, the solution in the frequency domain provides an exact and efficient method for computing the response of the coupled system. Despite this, the state of practice as a first attempt to incentivize time domain analyses compatible with standard finite element packages introduces the so‐called dimensionless flexible‐base frequency. This frequency, which depends solely on the structure‐to‐soil‐period ratio, allows transforming the frequency domain analyses into time domain analyses using frequency‐independent soil impedance values. However, if such frequency exists for the combined system, it must depend on several physical variables. In this work, we propose a supervised approach to obtain the flexible‐base dimensionless frequency at which the frequency‐independent soil impedance should be used. The analysis is carried out using five dimensionless parameters, and the importance of each one to the estimation of the dimensionless flexible‐base frequency is investigated. We use an inverse problem based on ensemble Kalman inversion (EnKI) to obtain the optimal frequency of the interaction. The data obtained are then employed in a machine‐learning framework to map a set of dimensionless parameters to such a frequency. The generated mapping is finally verified, and a significant improvement in time‐domain simulations is shown compared to the state of practice.

ABSTRACT We present a data-driven framework for ground-motion synthesis that generates three-component acceleration time histories conditioned on moment magnitude (M), rupture distance (Rrup), time-average shear-wave velocity at the top 30 m (VS30), and style of faulting. We use a Generative Adversarial Neural Operator (GANO)—a resolution invariant architecture that guarantees model training independent of the data sampling frequency. We first present the conditional ground-motion synthesis algorithm (cGM-GANO) and discuss its advantages compared to the previous work. We next train cGM-GANO on simulated ground motions generated by the Southern California Earthquake Center Broadband Platform (BBP) and on recorded the Kiban–Kyoshin network (KiK-net) data, and show that the model can learn the overall magnitude, distance, and VS30 scaling of effective amplitude spectra (EAS) ordinates and pseudospectral accelerations (PSA). Results specifically show that cGM-GANO produces consistent median scaling with the training data for the corresponding tectonic environments over a wide range of frequencies for scenarios with sufficient data coverage. For the BBP dataset, cGM-GANO cannot learn the ground-motion scaling of the stochastic frequency components (f > 1 Hz); for the KiK-net dataset, the largest misfit is observed at short distances (Rrup<50 km) and for soft-soil conditions (VS30<200 m/s) due to the scarcity of such data. Except for these conditions, the aleatory variability of EAS and PSA are captured reasonably well. Finally, cGM-GANO produces similar median scaling to traditional ground-motion models (GMMs) for frequencies greater than 1 Hz for both PSA and EAS but underestimates the aleatory variability of EAS. Discrepancies in the comparisons between the synthetic ground motions and GMMs are attributed to inconsistencies between the training dataset and the datasets used in GMM development. Our pilot study demonstrates GANO’s potential for efficient synthesis of broadband ground motions.

Realistic strong-motion accelerograms of earthquakes are required for various earthquake engineering tasks, including modeling structural and site response to large and near-source events. In this work, we introduce a data-driven framework for three-component ground motion synthesis intended for engineering applications. Leveraging the increase of ground-motion data from seismic networks and recent advancements in machine learning, we train a generative adversarial neural operator (GANO) to produce realistic three-component acceleration time histories conditioned on moment magnitude (M), rupture distance (Rrup), time-average shear-wave velocity at the top 30 m (Vs30) based on a California dataset compiled from PEER NGA-West2 database, and a public DesignSafe California database. The results show that the framework can efficiently recover the magnitude, distance, and Vs30 scaling of Fourier amplitude and pseudo-spectral accelerations. Through a comprehensive residual analysis using empirical data, we have verified that our model accurately captures both the mean values and aleatory variability of the evaluated ground-motion parameters.

Characterization of earthquake ground motion for engineering applications varies from place to place, and the degree of variability, which depends on the application and on the place, is not always intuitive. One would expect, for example, that parts of the globe with frequent earthquakes and dense strong-motion networks (e.g., areas with rich ground motion datasets, such as California) would have developed ground motion models with lower uncertainty than the other parts of the world, where moderate and large earthquakes occur infrequently, and monitoring networks are sparse or have a short history of use. Following that line of thought, one can also expect that stakeholders in metropolitan areas, where seismic networks are more dense than in rural regions, would be able to use the additional ground motion data to refine the ground motion models around these metropolitan areas and reduce the uncertainty further.

In the fields of structural and geotechnical engineering, improving the understanding of soil–structure interaction (SSI) effects is critical for earthquake-resistant design. Engineers and practitioners often resort to finite element (FE) software to advance this objective. Unfortunately, the availability of software equipped with boundary representation for absorbing scattered waves and ensuring consistent input ground motion prescriptions, which is necessary for accurately representing SSI effects, is currently limited. To address such limitations, the authors developed Seismo-VLAB (SVL v1.0-stable) an open-source software designed to perform SSI simulations. The methodology considers the integration of advanced techniques, including the domain decomposition method (DDM), perfectly matched layers (PMLs), and domain reduction method (DRM), in addition to parallel computing capabilities to accelerate the solution of large-scale problems. In this work, the authors provide a detailed description of the implementation for addressing SSI modeling, validate some of the SVL’s features needed for such purpose, and demonstrate that the coupled DRM–PML technique is a necessary condition for accurately solving SSI problems. It is expected that SVL provides a significant contribution to the SSI research community, offering a self-contained and versatile alternative. The software’s practical application in analyzing SSI and directionality effects on 3D structures under seismic loading demonstrates its capability to model real-world earthquake responses in structural engineering.

We present a data-driven framework for ground-motion synthesis that generates three-component acceleration time histories conditioned on moment magnitude, rupture distance , time-average shear-wave velocity at the top $30m$ ($V_{S30}$), and style of faulting. We use a Generative Adversarial Neural Operator (GANO), a resolution invariant architecture that guarantees model training independent of the data sampling frequency. We first present the conditional ground-motion synthesis algorithm (cGM-GANO) and discuss its advantages compared to previous work. We next train cGM-GANO on simulated ground motions generated by the Southern California Earthquake Center Broadband Platform (BBP) and on recorded KiK-net data and show that the model can learn the overall magnitude, distance, and $V_{S30}$ scaling of effective amplitude spectra (EAS) ordinates and pseudo-spectral accelerations (PSA). Results specifically show that cGM-GANO produces consistent median scaling with the training data for the corresponding tectonic environments over a wide range of frequencies for scenarios with sufficient data coverage. For the BBP dataset, cGM-GANO cannot learn the ground motion scaling of the stochastic frequency components; for the KiK-net dataset, the largest misfit is observed at short distances and for soft soil conditions due to the scarcity of such data. Except for these conditions, the aleatory variability of EAS and PSA are captured reasonably well. Lastly, cGM-GANO produces similar median scaling to traditional GMMs for frequencies greater than 1Hz for both PSA and EAS but underestimates the aleatory variability of EAS. Discrepancies in the comparisons between the synthetic ground motions and GMMs are attributed to inconsistencies between the training dataset and the datasets used in GMM development. Our pilot study demonstrates GANO's potential for efficient synthesis of broad-band ground motions

ABSTRACT We demonstrate the effects of soil–structure interaction (SSI) for three idealized building typologies on a regional scale, using a simulated earthquake scenario of the 2008 Mw 5.4 Chino Hills mainshock in southern California as an example. All the three buildings lie on shallow foundations, and they are subject to three-component simulated ground motions. To carry out this task, we develop a reduced order model (ROM) for each building typology that accounts for the effects of SSI on the building system in the time domain. We specifically use ensemble Kalman inversion (EnKI) to extract the soil impedance values from fully coupled soil–foundation–structure interaction simulations; and we interpolate the EnKI results to derive analytical functions that span the range of applicability of the soil impedance model. We then verify our ROMs by comparing results to fully coupled soil–foundation–structure interaction simulations, also known as direct modeling methods. We finally populate the simulation grid across southern California with the verified building ROMs, and interpret the responses in the form of maps that represent urban-scale effects of SSI on the seismic demand parameters such as maximum displacement, acceleration, and interstory drift. We also identify areas where the effects of SSI, given the resonant characteristics of a specific building, the foundation typology, and the local site conditions, lead to higher seismic demand relative to the fixed-base response.

This paper presents a model for distributing zones of liquefaction and nonliquefaction for use in regional liquefaction risk analysis. There are two broad methodologies that have been used to evaluate liquefaction risk on the regional scale: (a) application of site‐specific procedures using soil properties inferred from geology, or (b) application of geospatial proxies for liquefaction. The first approach will tend to predict similar liquefaction probabilities across broad areas with similar geology, water table depths, and shaking intensities. The second approach yields the probability of liquefaction, which can be interpreted as the portion of the area affected by liquefaction (). Neither approach, however, gives an informed prediction of the spatial distribution of liquefaction and the resulting displacements, which are particularly important for assessments of seismic risk for spatially distributed infrastructure systems. We propose a methodology for incorporating spatial correlation into a geospatial proxy for liquefaction to create maps of liquefaction and nonliquefaction for a given earthquake scenario. First, we describe a latent Gaussian process that is assumed to govern the spatial distribution of liquefaction. Next, a database of empirical observations of liquefaction is used to obtain the coefficients that describe that latent Gaussian process. The proposed model yields random realizations of maps of liquefaction and nonliquefaction conditioned on a map of . Such maps can be used to constrain the area over which displacements are estimated using soil properties inferred from geology and are therefore a critical component in reducing bias in assessments of liquefaction risk at the regional scale.

Next article Free accessTechnical Breakthrough AbstractsMar 7, 2023The 2022 Chihshang, Taiwan, Earthquake: Initial GEER Team ObservationsAuthors: Trevor J. Carey, H. Benjamin Mason https://orcid.org/0000-0003-4279-2854 [email protected], Domniki Asimaki, A.M.ASCE, Adda Athanasopoulos-Zekkos, M.ASCE, Fernando E. Garcia, A.M.ASCE, Brian Gray, Grigorios Lavrentiadis, and Chukwuebuka C. Nweke, M.ASCEAuthor AffiliationsPublication: Journal of Geotechnical and Geoenvironmental EngineeringVolume 149, Issue 5https://doi.org/10.1061/JGGEFK.GTENG-11522 PDF