Comparison and reliability of declustering methods evaluated using an ETAS framework

Declustering of earthquake catalogs is a fundamental preprocessing step in seismicity analysis and probabilistic seismic hazard assessment (PSHA), as it aims to separate background, approximately Poissonian seismicity from dependent events such as foreshocks and aftershocks. The choice of declustering method can significantly influence estimated seismicity rates, b-values, spatial source models, and ultimately seismic hazard results. Despite its widespread use, there is no consensus on the most reliable declustering approach, and different algorithms often produce substantially different background catalogs for the same dataset. This study presents a systematic comparison of commonly used declustering techniques, including the window-based methods of Gardner and Knopoff, Uhrhammer, and Grünthal; the interaction-based Reasenberg algorithm; the nearest-neighbor clustering method of Zaliapin; and Epidemic-Type Aftershock Sequence (ETAS) based stochastic declustering. All methods are applied to the same regional earthquake catalog with consistent magnitude completeness and spatial coverage to ensure a fair comparison. The resulting declustered catalogs are evaluated in terms of the fraction of events classified as background, their temporal and spatial distributions, and their impact on magnitude-frequency relationships. To assess the reliability of each declustering approach, we use the ETAS model as a reference framework. The comparison reveals pronounced method-dependent variability, particularly at short inter-event times and distances, with window-based methods generally removing a larger proportion of clustered events and interaction-based methods showing sensitivity to user-defined parameters. The Zaliapin method offers a data-driven alternative but may be influenced by spatial heterogeneity, while ETAS-based stochastic declustering provides a probabilistic and internally consistent representation of seismicity at the cost of higher computational and data-quality requirements. The results highlight the need for careful method selection and uncertainty-aware declustering in seismic hazard applications and demonstrate the value of ETAS-based diagnostics as an objective benchmark for evaluating declustering performance.