No abstract

Despite the exponential growth of the amount of ground-motion data, ground-motion records are not always available for all distances, magnitudes, and site conditions cases. Given the importance of using time histories for earthquake engineering (e.g., nonlinear dynamic analysis), simulations of time histories are therefore required. In this study, we present a model for simulating nonstationary ground-motion recordings, which combines a conditional generative adversarial network to predict the amplitude part of the time–frequency representation (TFR) of ground-motion recordings and a phase retrieval method. This model simulates the amplitude and frequency contents of ground-motion data in the TFR as a function of earthquake moment magnitude, source to site distance, site average shear-wave velocity, and a random vector called a latent space. After generating the phaseless amplitude of the TFR, the phase of the TFR is estimated by minimizing all differences between the observed and reconstructed spectrograms. The simulated accelerograms produced by the proposed method show similar characteristics to conventional ground-motion models in terms of their mean values and standard deviations for peak ground accelerations and Fourier amplitude spectral values.

In this article, we address the question of how observed ground-motion data can most effectively be modeled for engineering seismological purposes. Toward this goal, we use a data-driven method, based on a deep-learning autoencoder with a variable number of nodes in the bottleneck layer, to determine how many parameters are needed to reconstruct synthetic and observed ground-motion data in terms of their median values and scatter. The reconstruction error as a function of the number of nodes in the bottleneck is used as an indicator of the underlying dimensionality of ground-motion data, that is, the minimum number of predictor variables needed in a ground-motion model. Two synthetic and one observed datasets are studied to prove the performance of the proposed method. We find that mapping ground-motion data to a 2D manifold primarily captures magnitude and distance information and is suited for an approximate data reconstruction. The data reconstruction improves with an increasing number of bottleneck nodes of up to three and four, but it saturates if more nodes are added to the bottleneck.