Answers are grounded in this paper and cite their sources.
Fourier Amplitude Spectrum (FAS) plays a central role in stochastic ground motion simulations and seismic hazard assessments, particularly in regions with limited data. This study develops a Ground Motion Model (GMM) for predicting the FAS of earthquake ground motions in New Zealand using data-driven methods. A comprehensive strong-motion database of New Zealand earthquakes is compiled and used to train and validate advanced machine learning models, including a Feedforward Artificial Neural Network (ANN), Long Short-Term Memory (LSTM) networks, and a Conditional Variational Autoencoder (CVAE). These models are designed to capture the nonlinear relationships between ground motion intensity and key seismological and site predictors such as magnitude, source-to-site distance, site conditions, and faulting style. Model performance is evaluated using cross-validation and residual analysis. All three approaches yield nearly the same aleatory uncertainty as well as median predictions, demonstrating robust performance. Among them, the LSTM network performs best in terms of evaluation metrics, indicating superior deterministic predictive capability. The CVAE model have the advantage of capturing epistemic uncertainty. By generating ensembles of conditional FAS realizations, the CVAE framework allows for the simulation of realistic ground motions tailored to specific input scenarios. A weighted averaging framework is adopted to combine the predictions of the CVAE, LSTM, and ANN models, assigning equal weights to each component. This integrated approach enables the final model to leverage the complementary strengths of all three models, combining robust deterministic prediction with probabilistic uncertainty quantification. The framework enhances regional GMMs by supporting probabilistic simulations, improving seismic hazard and risk assessment.
No takes yet. Share an insight, caveat, or question.
Hati et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: