Abstract
Ground-motion prediction equations are developed by regression analysis based on datasets of recorded ground-motion parameters at multiple stations during different earthquakes and in various source regions. The present probabilistic seismic hazard analysis applies the standard deviation of these equations to evaluate the hazard at a specific site (the ergodic assumption). The standard deviation has a strong influence on the results of probabilistic seismic hazard analysis at long return period. In the ground motion prediction equations, there is inevitably some mixing of epistemic uncertainty into the model of aleatory uncertainty. To improve the results of hazard analysis, it is important to quantify the aleatory uncertainty by removing the ergodic assumption. In this paper, the semivariogam procedure is adopted to separate of aleatory and epistemic uncertainty from the inter-event residuals in ground motion prediction equations for small-to-moderate earthquakes. The standard deviations of aleatory uncertainty in inter-event residuals are 0.315 for PGA and 0.354 for PGV. These values are about 40% smaller for PGA and about 20% smaller for PGV when compared the standard deviations of the whole data set.
