Are Aftershock Sequences Pertinent to Long‐Term Seismic Hazard Assessments? Insights From the Temporal ETAS Model

Aftershocks are commonly removed from observed earthquake catalogs in probabilistic seismic hazard analyses, using declustering techniques. We use stationary and temporal Epistemic‐Type Aftershock Sequence (ETAS) models to generate aftershocks from background seismicity and preceding aftershocks. We assume that the mainshocks equal the background seismicity, to divide the synthetic earthquake catalogs into mainshock (declustered) and complete (nondeclustered) versions. Only mainshocks follow a Poissonian distribution. We then evaluate how accurately we can forecast the recurrence of the largest events based on the simulated catalogs. A single b value is derived from the simulated catalogs and used in the magnitude‐frequency forecast. When the b value of the mainshocks is considerably smaller than the b value of the aftershocks, the information derived from the mainshock catalog leads to accurate predictions of the occurrence of the largest events. Conversely, when the mainshock and aftershocks have comparable b values, only the complete catalog produces representative estimates for the occurrence statistics of the largest events. We also show that using Poisson statistics leads to representative assessment of long‐term recurrences, even if aftershocks have a non‐Poissonian distribution. Finally, we analyze a recent case of induced seismicity, Oklahoma, USA, where the complete catalog displays a kink in the magnitude‐frequency distribution. Declustering removes this kink, leading to better b value estimations for the largest magnitude events. We conclude that temporal declustering for seismic hazard assessment is only recommended in catalogs with a large number of earthquakes and in catalogs where the b values of the mainshocks are significantly different from the b values of the complete catalog.