Uncertainty in the estimates of peak ground acceleration in seismic hazard analysis

Abstract

Probabilistic seismic hazard analysis has become a standard procedure preceding the antiseismic construction. An important component of the relevant calculations is the allowance for the uncertainty in the strong motion parameters (e.g., peak ground acceleration (PGA)). In the present-day approaches of probabilistic analysis, this uncertainty is modeled by a random variable (a residual) which has a lognormal distribution. With this model, the extrapolation into the area of long return periods yields nonzero probabilities of unrealistically high PGA. In the present work, the distribution of the logarithmic PGA residuals is modeled by different parametric distributions. From the set of these distributions, the one which provides the closest approximation of the empirical data is selected by the statistical criteria. The analysis shows that the generalized extreme value distribution (GEVD) most accurately reproduces the residuals of the logarithmic PGA, and the tail of the distribution is approximated by the generalized Pareto distribution (GPD).