Earthquake Ground Motion Conditioned Simulation Using Sparsely Distributed Observed Motions for Analysis and Design of Lifeline Structures

The spatial variation of ground motions affects the response of the distributed lifeline structures such as bridges and natural gas pipelines within a seismic region. At present, the number of earthquake ground motion recording stations is generally sparse. Therefore, ground motion simulation methods are deployed to estimate the input motions at locations with no available instruments. In this study, a Gaussian Process Regression model is presented to interpolate the surrounding observed ground motion time series to estimate the motion at an adjacent unobserved location. The interpolation process is done using a Gaussian Process Regression, which models the Fourier spectrum's real and imaginary parts as random Gaussian variables. The physics-based simulated motions for the 1906 San Francisco earthquake is employed to train the model. The M7.0 Hayward fault fully deterministic physics-based simulated earthquake ground motions are used to evaluate the trained model's performance. The preliminary evaluations illustrated that the trained Gaussian Process regression model could decently predict the ground motion time series.