Item

Abstract

Despite the continuous increase in the number of ground motion stations and the amount of recorded ground motion data in recent years, we still face observational gaps in earthquake records for large-magnitude and small-distance events. Physics-based simulations of recent earthquakes have been capable of reproducing the ground shaking recorded by near-field accelerometric stations but also predicting ground motions in other locations. In this work, we first develop a hybrid database combining calibrated ground shaking simulations and real-world observed ground motions. This hybrid dataset consists of 83864 real and near-source events records (from 1200 events) and physics-based simulations (from 20 events). In the second step, a generative model is trained based on this hybrid dataset. The model simulates nonstationary ground-shaking recordings. It 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, and a random vector called 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. We finallly compare the generative model predictions with the data of the 2023 Turkey-Syria seismic sequence.