Validation of the Epidemic-Type Aftershock Sequence (ETAS) Models for Simulation-Based Seismic Hazard Assessments

Abstract

Abstract Moderate-to-large-magnitude earthquakes induce considerable short-to-medium-term increases in seismic hazard, due to the subsequent occurrence of aftershocks. Most studies on aftershock occurrence models focus on their performance as part of Operational Earthquake Forecasting (i.e., the real- or near-real-time predictions of ongoing sequences). Aftershock model validation on a longer timescale is scarcely dealt with in the literature, however. This study addresses this research gap by validating the medium-term (i.e., months to years) performance of an advanced formulation of the epidemic-type aftershock sequence (ETAS) model for potential future implementation in simulation-based probabilistic seismic hazard analyses (PSHAs). The aim is to determine whether the considered ETAS model can forecast adequate numbers of aftershocks, and consistent spatial and magnitude–frequency distributions, for three years after a given moderate-to-large-magnitude mainshock. Two different calibration procedures (region wide and sequence averaged) are used to investigate the ability of resulting ETAS models to describe common characteristics of the considered sequences. The region-wide methodology is based on the entire catalog of a large geographic area and is the conventional approach for calibrating ETAS models. The second calibration methodology is instead based on sequence-specific data. The validation procedure employs only out-of-sample (i.e., retrospective) testing, and its evaluation metrics are the catalog-based scores recently proposed for the Collaboratory for the Study of Earthquake Predictability framework. It is demonstrated using the New Zealand catalog (1990–2020), including seven sequences in the period 1990–2020, in two case studies. Resulting test scores indicate that the conventional region-wide calibration approach is not suitable for building an ETAS model to use in simulation-based PSHA. Sequence-averaged ETAS models are generally more acceptable, in terms of both numbers of aftershocks and consistent spatial and magnitude–frequency distributions. The findings of this study can be used to guide future implementations of the considered ETAS formulation, possibly in combination with a simulation-based mainshock PSHA.