DOI: 10.5176/2251-3353_GEOS14.60
Authors: Vasily Pavlenko, Prof Andrzej Kijko
Abstract: Probabilistic seismic hazard analysis became a standard practice that precedes construction of engineering structures of high importance. The most frequently used method of the probabilistic analysis is Cornell-McGuire procedure (Cornell, 1968). Ground-motion variability is a very important issue of this method (Bender, 1984). The common assumption is that this variability can be described by a random variable with a lognormal distribution (Joyner, 1981). However, this hypothesis has not been robustly tested. The evidence for a lognormal distribution was confirmed by a Kolmogorov-Smirnov test at the 90{6e6090cdd558c53a8bc18225ef4499fead9160abd3419ad4f137e902b483c465} significance level (Campbell, 1981). Such test is not sensitive to the tail region of the distribution. Furthermore, a right tail of the lognormal distribution is not limited, so the use of the lognormal distribution for calculation of hazard curves results in non-zero probabilities of exceedance of unrealistic values of peak ground accelerations. In this study, the random variable that represents the ground-motion variability is modeled by a number of parametric distributions. One parametric law that gives the best approximation to the distribution of the residuals is chosen by statistical criteria. The results of the analysis show that the best fit for ground motion variability is achieved with Generalized Extreme Value Distribution, and the tail of the distribution of residuals is modeled in the best way by the Generalized Pareto Distribution.
Keywords: probabilistic seismic hazard analysis ground-motion variability hazard curves
Updating...