Broadband Ground Motion Synthesis via Generative Adversarial Neural Operators: Development and Validation

TL;DR

The paper tackles the challenge of generating broadband, three-component ground motions conditioned on magnitude, distance, site velocity, and faulting style by introducing cGM-GANO, a conditional Generative Adversarial Neural Operator that is discretization-invariant and learns function-valued ground-motion waveforms. Trained on SCEC BBP simulations and KiK-Net recordings, cGM-GANO learns mean scaling and aleatory variability across a broad frequency range, with some misfits at near-field distances and soft soils due to data sparsity and the limited ability to capture stochastic high-frequency components. The approach offers a fast, scalable alternative to physics-based simulations and traditional empirical models, potentially bridging the gap between them, and shows potential to generate hundreds of realistic, coherent ground motions in under a second while highlighting areas for improvement such as nonlinear site effects and azimuthal radiation patterns. Ongoing work aims to enrich training data (especially for near-field and soft-site regimes), incorporate filtering effects, and extend conditioning to additional physical variables to further enhance fidelity and applicability in engineering practice.

Abstract

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

Figures (17)

• Figure 1: Schematic of GANO input-output structure used in this work: given a set of conditional variables (left), the trained GANO (middle) produces three-component acceleration time series (right).
• Figure 2: Schematic of cGM-GANO architecture, and generator - discriminator interaction. The input conditional variables are $\mathbf{M}$, $R_{rup}$, $V_{S30}$ and $f_{type}$. The input function for the generator is finite duration Gaussian random field sample while the output is the normalized three-component acceleration time series and three PGAs. The input function of the discriminator are the real or synthetic three-component acceleration time series while the output is score evaluating the realism of the input ground motions.
• Figure 3: Data distribution of BBP dataset. Subfigures (a) and (b) show crossplot between $log_{10}$(PGA) and magnitude and rupture distance, respectively
• Figure 4: Scenario event comparisons for cGM-GANO trained on the BBP dataset. Each vertical sub-panel depicts the Fourier Amplitude Spectrum (FAS), aleatory standard deviation of FAS, Rotd50 Pseudo-spectral acceleration (PSA) spectrum, and aleatory standard deviation of Rotd50 PSA. The title of each plot corresponds to the midpoint of each bin, and $N_{obs}$ is the number of observed records within each bin. The lines represent the mean in log space of the horizontal components of the synthetic (solid line) or BBP (dashed line) data, while the shaded zone represents the $16^{th}$ to $84^{th}$ percentile range.
• Figure 5: Comparison of cGM-GANO magnitude and distance scaling with BBP dataset for PSA RotD50(T=1 s). Solid lines represent the mean scaling in log space. The shaded region shows the $2^{nd}$ to $98^{th}$ percentile of aleatory variability. Solid dots correspond to the training data.