Conditioned simulation of local fields of earthquake ground motion

Abstract

Over distances between supports of typical multi-support structures, high-frequency components of earthquake ground motions tend to be weakly correlated in space and low-frequency components of highly correlated. Random field theory, substantiated by empirical data from dense strong-motion arrays, allows engineering characterization of the stochastic input for seismic analysis of multi-support structures, in terms of frequency-dependent spatial correlation functions. The paper extends the methodology of “kriging” (linear estimation theory applied to random functions) to simulation of a local field of earthquake ground motions, consistent with given information about the motions' frequency content and space-time correlation, at an arbitrary set of locations on the free-field surface. Recorded motions may be specified at one or more locations, and the method produces compatible accelerograms at local-field points where the motion has not been recorded. Methodology is first developed for conditional simulation of quadrant symmetric space-time random fields; it is then extended to nonstationary fields and the application to earthquake ground motion simulation is illustrated through several examples.