Effects on Probabilistic Seismic Hazard Estimates That Result from Nonuniqueness in Declustering an Earthquake Catalog

Declustering a seismicity catalog to obtain a background seismicity model for probabilistic seismic hazard analysis is not a uniquely defined process. Zaliapin and Ben-Zion (2020) present a method for randomly thinning a complete catalog by removing nearest-neighbor earthquakes. The number of events in the residual catalog depends on a continuous parameter, α0, called the “cluster threshold.” Varying α0 results in a family of residual catalogs, and when enough events are removed the catalog is nearly Poissonian and can be considered declustered. This family of thinned catalogs is used to generate a corresponding family of background seismicity models, which in turn are used to find the probabilistic seismic hazard on an east–west profile across California. Additional models are developed by renormalizing the thinned Zaliapin and Ben-Zion (2020) catalogs to the catalog rate of earthquakes with moment magnitude Mw≥5 and from the maximum shaking earthquake catalog of Anderson et al. (2021). Adding fault contributions, the simplified estimate of the hazards with probability of exceedance of 2% in 50 yr are comparable to the 2018 National Seismic Hazard Model (NSHM). Where faults dominate the hazard, the method of thinning has little effect. The range of hazard estimates from the set of background models alone illustrate the range of effects of catalog thinning for any location where faults do not dominate. The spread of background hazard estimates at most sites is generally within a multiplicative factor of ∼2 of the hazard estimated from a catalog declustered for the 2018 NSHM. However, the spread in the estimates is very large in the vicinity of aftershock zones of large earthquakes. The family of randomly thinned catalogs, including alternative smoothing parameters and optional rescaling, may span the body and range of background hazard that can be inferred from the known history of earthquakes.