Abstract
The effects of temporal and magnitude dependence among seismic recurrences, which are ignored in the conventional Poisson earthquake model, are studied. The potential impact of non-Poissonian assumptions on practical hazard estimates are considered. A broad set of recurrence models with memory are analyzed using convenient second-moment time-magnitude statistics to parameterize a general class of semi-Markov models. The conventional time- and slip-predictable models are included and studied as special cases. Conditions are identified under which the Poisson model provides a sufficient engineering hazard estimate (i.e., either conservative or unconservative by a factor of no more than three). Cases in which the Poisson estimate is insufficient are limited practically to those in which the hazard is controlled by a single feature for which the elapsed time since the last significant event exceeds the average time between such events. Moreover, this situation creates a problem only if events on the fault show strongly regular, “characteristic time” behavior.
