A NEW APPROACH TO STATISTICAL ANALYSIS OF PALEOSEISMIC DATA CONFIRMS THE FUNDAMENTAL PERIODICITY OF MOST LARGE EARTHQUAKES

We present a new statistical approach that confirms the fundamental periodicity of most large earthquakes observed in paleoseismic records. This simple, robust methodology allows for more objective inferences on the nature of earthquake recurrence from paleoseismic records, overcoming significant limitations of previous qualitative analyses. We show that this method can be used to reliably infer earthquake recurrence behavior using only two pieces of information in a paleoseismic record – the coefficient of variation (COV) of the interseismic intervals and the number of recorded events. Stochastic numerical modeling indicates that that this method can be used to test hypotheses related to earthquake recurrence behavior whether a paleoseismic record is assumed to represent a complete fault history, an uninterrupted subsequence of events from a longer fault history, or events randomly sampled from a fault history. When applied to 19 previously published paleoseismic records, our analysis indicates that the majority (68% at 95% confidence; 75% at 90% confidence) support an interpretation of quasi-periodic, time-dependent earthquake recurrence. All records containing more than 13 events support an interpretation of time-dependent recurrence. No records support an interpretation of clustered recurrence. The paleoseismic records analyzed by our method include strike-slip, normal, and reverse faults from a wide range of tectonic settings. Thus, although many of these records exhibit substantial variability in their interseismic intervals, the majority are statistically non-random, suggesting that stress-renewal is the fundamental process controlling large earthquake rupture in the crust. These results argue strongly against the use of probabilistic seismic hazards models based on the Poisson distribution – the preferred model for random, time-independent recurrence.