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

Creators: Shi, Yaozhong; Lavrentiadis, Grigorios; Asimaki, Domniki¹; Ross, Zachary E.¹; Azizzadenesheli, Kamyar

1. California Institute of Technology

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 (⁠R_(rup)), time‐average shear‐wave velocity at the top 30 m (⁠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 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 V_(S30) 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 (⁠R_(rup) < 50 km⁠) and for soft‐soil conditions (⁠V_(S30) < 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.

Copyright and License

Acknowledgement

The authors would like to thank Dr. Weiqiang Zhu and Dr. Mahdi Bahrampouri for their help during the curation and preprocessing of the Kiban–Kyoshin network (KiK‐net) dataset. The authors would also like to thank Dr. Jian Shi and Dr. Robert W. Graves for providing the Broadband Platform (BBP) dataset. This material is based upon work supported in part by the U.S. Geological Survey (USGS) under Grant Number G22AP00231. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the USGS. Mention of trade names or commercial products does not constitute their endorsement by the USGS. The authors would also like to thank Dr. Kim Olsen, an anonymous reviewer, and associate editor Luis Angel Dalguer for the review and constructive comments that helped improve the final article.

Data Availability

Kiban–Kyoshin network (KiK‐net) time histories can be downloaded through National Research Institute for Earth Science and Disaster Resilience at https://www.bosai.go.jp/e/index.html, whereas the kik‐net flatfile of the selected events is summarized in Bahrampouri et al. (2021) and can be downloaded through DesignSafe at https://www.designsafe-ci.org/data/browser/public. The code for conditional ground‐motion Generative Adversarial Neural Operator (cGM‐GANO) is published at https://github.com/Caltech-geoquake/GM-GANO. All websites were last accessed in February 2023. The supplemental material contains a graphical example supporting the selection of velocity time series as our training data (Fig. S1); a realization of fault and station locations for the Broadband Platform (BBP) simulations, depicting the sparsity of the recorded ground motions (Fig. S2); two scenarios of M 7.0 events at 10 km simulated using the BBP for reverse and strike‐slip faults in the time and frequency domains, and their comparison with cGM‐GANO‐generated ground motions for the same conditional variables (Fig. S3); the interfrequency correlation of the cGM‐GANO ground motions for the model trained on BBP, clearly depicting the separation between the deterministic and stochastic regimes of the algorithm (Fig. S4); residual plots of the KiK‐net trained model based on the validation (hold‐off) dataset (Fig. S5); velocity time histories from observations and cGM‐GANO‐generated ground motions using the model trained on the KiK‐net data (Fig. S6); and displacement time histories from observations and cGM‐GANO‐generated ground motions using the model trained on the KiK‐net data (Fig. S7).

Supplementary data

bssa-2023207_supplement- pdf file

Conflict of Interest

The authors acknowledge that there are no conflicts of interest recorded.

Files

bssa-2023207_supplement.pdf

Files (3.9 MB)

	All versions	This version
Views	0	0
Downloads	0	0
Data volume	0 Bytes	0 Bytes