Simulation of ground motions using stochastic method with improved temporal nonstationarity

Simulating ground motion (GM) is essential for assessing seismic hazards and evaluating the risks to civil infrastructure in earthquake engineering. The widely used stochastic method achieves temporal nonstationarity in simulated GM by applying a window function to Gaussian white noise. Typically, this window function has a fixed shape due to its constant shape parameters, resulting in uniform waveforms in simulated GMs. This study proposes a novel approach that generates a window function from recorded GMs, rather than one constrained by a specific mathematical form with constant parameters. Here, the relative location of the peak ground acceleration ( P ) and the 5%–75% significant duration ( D 70 ) are selected as shape parameters for the window function. To determine target shape parameters for specific scenarios, this study establishes a probability distribution model for P and applies a well-established predicting model for D 70 . A time series, shaped to meet the target parameters, is generated from recorded GMs, and its normalized envelope, computed using the Hilbert transform, is utilized as the window function. The proposed approach enables the window function’s shape parameters to better align with statistical characteristics, incorporating the temporal nonstationarity of recorded GMs into the simulation. Case studies on the M w 6.9, 2008 Iwate and M w 6.8, 2007 Chuetsu-oki earthquakes illustrate improvements in the envelopes and durations of the simulated GMs.