Time-frequency domain hybrid dimension-reduction simulation for fully non-stationary stochastic ground motion field

This study extends the filtered white noise model by proposing a time-frequency hybrid dimensionality reduction model for fully nonstationary seismic ground motion random fields, thereby overcoming the limitation of simulating only ground motion processes without capturing spatially distributed ground motion fields. Specifically, to address the difficulty in directly representing the spatial coherence of seismic random fields within the impulse response function of the filtered white noise model, a proper orthogonal decomposition (POD)-based dimensionality reduction simulation method is introduced. This approach enables a frequency-domain representation of spatially coherent white noise random vector processes. By applying the impulse response functions and modulation functions corresponding to different locations within the seismic random field to filter and modulate the respective white noise components, an efficient time–frequency hybrid dimensionality reduction representation of fully nonstationary seismic random fields is achieved. Numerical examples validate the accuracy and engineering applicability of the proposed model by comparing mean values, standard deviations, auto-/cross-correlation functions, as well as response spectra and coherence functions.