Abstract
When earthquakes from a single source govern a structure's design, decisions are sensitive to the probability distribution of times between characteristic earthquakes and correlations between time elapsed and magnitude. Scarcity of data prevents a purely statistical study of these matters while absence of trustworthy theories prevents a purely analytical approach. Based on simulation using a physically founded mathematical model and on data from various parts of the world, we postulate a slip-predictable model and four candidate distributions with random parameters for the distribution of interoccurrence times in subduction zones: lognormal, gamma, inverse gaussian and Weibull. Probabilities are updated using data from three groups of subduction earthquakes. In all cases the lognormal distribution minimizes the disutility due to choosing a single model. The same distribution with fixed parameters gives almost equally good results.

This publication has 0 references indexed in Scilit: