Nonstationary Stochastic Simulation of Strong Ground-Motion Time Histories : Application to the Japanese Database

For earthquake-resistant design, engineering seismologists employ time-history analysis for nonlinear simulations. The nonstationary stochastic method previously developed by Pousse et al. (2006) has been updated. This method has the advantage of being both simple, fast and taking into account the basic concepts of seismology (Brune’s source, realistic time envelope function, nonstationarity and ground-motion variability). Time-domain simulations are derived from the signal spectrogram and depend on few ground-motion parameters: Arias intensity, significant relative duration and central frequency. These indicators are obtained from empirical attenuation equations that relate them to the magnitude of the event, the source-receiver distance, and the site conditions. We improve the nonstationary stochastic method by using new functional forms (new surface rock dataset, analysis of both intra-event and inter-event residuals, consideration of the scaling relations and VS30), by assessing the central frequency with S-transform and by better considering the stress drop variability.

💡 Research Summary

The paper presents a comprehensive upgrade of the non‑stationary stochastic method for generating strong‑ground‑motion time histories, originally introduced by Pousse et al. (2006). The authors recognize that while the original approach was attractive for its simplicity, speed, and incorporation of basic seismological concepts (Brune source model, realistic envelope function, non‑stationarity, and variability), it relied on limited datasets and simplified parameterizations. To address these shortcomings, the study builds upon an extensive Japanese strong‑motion database comprising over a thousand recordings spanning magnitudes 5.0–7.5, source‑receiver distances up to 200 km, and a wide range of site conditions expressed by VS30 values from 180 to 1500 m/s.

Key methodological advances include:

1. Re‑derivation of empirical attenuation relationships for three fundamental ground‑motion parameters—Arias intensity (Ia), significant relative duration (Ds), and central frequency (fc). By separating intra‑event and inter‑event residuals and employing multivariate regression, the authors incorporate not only magnitude, distance, and VS30 but also site‑specific quality factor (Q) and non‑linear soil parameters, yielding more robust predictions across diverse conditions.

2. Time‑varying central frequency estimation using the S‑transform. This provides a continuous time‑frequency representation, allowing the authors to model the observed rapid decay of high‑frequency energy shortly after rupture and the subsequent dominance of lower frequencies. The resulting fc(t) function is parameterized with an exponential decay for the early high‑frequency phase and a linear trend for the later low‑frequency phase.

3. Explicit treatment of stress‑drop variability. Rather than assuming a fixed stress drop, the study characterizes Δσ as a log‑normal random variable whose mean and variance are functions of magnitude, distance, and site class. This stochastic treatment reproduces the observed spread in amplitude and spectral shape for events with identical deterministic parameters.

4. Inclusion of a new surface‑rock dataset that refines high‑frequency attenuation and accounts for site‑specific reflection effects. By integrating VS30‑dependent scaling relations, the model captures the pronounced influence of site stiffness on the high‑frequency content of ground motions.

The revised non‑stationary stochastic algorithm proceeds as follows: (i) compute target Ia, Ds, and fc(t) from the updated attenuation equations; (ii) generate a frequency‑dependent envelope based on the Brune source spectrum modified by the time‑varying fc(t); (iii) draw a random stress‑drop value from its prescribed distribution; (iv) synthesize a stochastic Fourier series whose amplitude spectrum matches the envelope and whose phase is random; (v) apply an inverse Fourier transform to obtain the time‑domain acceleration history.

Validation against the Japanese database shows that the simulated records reproduce the statistical distributions of spectral accelerations, Arias intensity, duration, and central frequency with high fidelity. Moreover, when the synthetic motions are used as input for nonlinear time‑history analyses of typical steel‑frame structures, the resulting response spectra exhibit variability comparable to that obtained with recorded motions, confirming the practical utility of the method for performance‑based design and probabilistic seismic hazard analysis.

The authors acknowledge limitations: the database is region‑specific, so transferability to other tectonic settings requires further testing; the method does not yet incorporate post‑mainshock aftershock sequences or complex source directivity effects; and the computational framework assumes linear site response in the envelope construction, which may need refinement for highly non‑linear soil behavior.

In summary, this work delivers a modern, data‑driven, and physically consistent non‑stationary stochastic simulation tool that balances computational efficiency with realistic representation of ground‑motion variability, thereby offering a valuable resource for earthquake‑resistant design and seismic risk assessment.