Regression models for predicting the probability of near-fault earthquake ground motion pulses, and their period

Near-fault earthquake ground motions containing large velocity pulses are known to cause severe demands on structures and geotechnical systems, but the probability of occurrence of these pulses in future earthquakes is not well understood. Using a database of past ground motions that have been classified as containing or not containing velocity pulses, this paper develops calibrated regression models to predict the occurrence of velocity pulses in future ground motions, as well as the nature of the pulses if they do exist. The regression model selection procedure indicates that useful predictors of pulse occurrence include source-to-site geometry variables such as the length of rupture between the epicenter and location of interest, and the closest distance from the location of interest to the fault rupture. It is observed that the period of any resulting velocity pulse is related primarily to the earthquake magnitude, but other predictive parameters are also considered and discussed. Both empirical regression tests and theoretical seismology explanations are given as to why the chosen predictor variables are important and meaningful. Some comparisons with previous similar models are also presented. The resulting predictive models can be incorporated into probabilistic seismic hazard analysis calculations. These results demonstrate the potential importance of quantitatively considering occurrence of near-fault pulses, and facilitate seismic reliability calculations that explicitly consider near-fault ground motion pulses.