MODELING THE CASCADIA PALEOSEISMIC RECORD OF LARGE EARTHQUAKE CLUSTERS USING LONG-TERM FAULT MEMORY

A challenge for earthquake hazard assessment is that geologic records often show large earthquakes occurring in temporal clusters separated by long periods of quiescence. This behavior is seen in Cascadia’s paleoseismic record, derived from turbidites, which has 4 to 5 distinct clusters of events. These are important because if we are still in the most recent cluster that began about 1700 years ago, a large earthquake is likely to happen soon. If we are between clusters, a great earthquake is less likely. Earthquake hazard assessments typically employ one of two models, neither of which directly incorporate clustering. In one, earthquake probability is time-independent and modelled as Poissonian, so an earthquake is equally likely at any time. The fault has no “memory” because when a prior earthquake occurred has no bearing on when the next will occur. The other common model is a time-dependent earthquake cycle in which the probability of an earthquake increases with time until one happens, after which the probability resets to zero. Because the probability is reset after each earthquake, the fault “remembers” only the last earthquake. This approach can be used with any assumed probability density function to sample the recurrence times. We propose an alternative, Long-Term Fault Memory (LTFM), a modified earthquake cycle model where the probability of an earthquake increases with time until one happens, after which it decreases, but not necessarily to zero. Hence the probability of the next earthquake depends on the fault’s history over multiple cycles, giving “long-term memory”. Physically, this reflects an earthquake releasing only part of the elastic strain stored on the fault. We use the LTFM to simulate earthquake clustering in Cascadia as well as the nearby San Andreas Fault. In some portions of the simulated earthquake history, events appear quasiperiodic, while at other times, the events can appear more Poissonian. Hence a given paleoseismic or instrumental record may not reflect the long-term seismicity of a fault, which has important implications for hazard assessment.