Aftershock identification

Earthquake aftershock identification is closely related to the question “Are aftershocks different from the rest of earthquakes?” We give a positive answer to this question and introduce a general statistical procedure for clustering analysis of seismicity that can be used, in particular, for aftershock detection. The proposed approach expands the analysis of Baiesi and Paczuski [PRE, 69, 066106 (2004)] based on the space-time-magnitude nearest-neighbor distance $\eta$ between earthquakes. We show that for a homogeneous Poisson marked point field with exponential marks, the distance $\eta$ has Weibull distribution, which bridges our results with classical correlation analysis for unmarked point fields. We introduce a 2D distribution of spatial and temporal components of $\eta$, which allows us to identify the clustered part of a point field. The proposed technique is applied to several synthetic seismicity models and to the observed seismicity of Southern California.

💡 Research Summary

The paper tackles the long‑standing question of whether aftershocks can be statistically distinguished from the background seismicity and proposes a comprehensive clustering framework that directly addresses this problem. Building on the space‑time‑magnitude nearest‑neighbor distance η introduced by Baiesi and Paczuski (2004), the authors first formalize η for a marked point process where each earthquake is characterized by its spatial coordinates, occurrence time, and magnitude (the mark). They assume that magnitudes follow an exponential distribution, which is equivalent to the Gutenberg‑Richter law in the appropriate magnitude range. Under this assumption they prove rigorously that η follows a Weibull distribution. The proof hinges on the independence of inter‑event spatial distances, temporal intervals, and magnitude marks in a homogeneous Poisson field; each component contributes a factor that, when combined with the power‑law weighting exponents α, β, and γ, yields the Weibull shape parameter. This result bridges the η‑based analysis with classical correlation tools such as Ripley’s K‑function, which are traditionally applied to unmarked point processes.

The second major contribution is the decomposition of η into a spatial component η_s (distance and magnitude) and a temporal component η_t (time interval and magnitude). By constructing the joint two‑dimensional distribution P(η_s, η_t) the authors demonstrate that a homogeneous Poisson background produces a smooth Weibull‑shaped cloud, whereas clustered events (aftershocks) concentrate in the lower‑left corner where both η_s and η_t are simultaneously small. This observation motivates a simple yet powerful clustering rule: events satisfying η_s·η_t < θ (with θ chosen from the empirical distribution) are classified as belonging to a clustered (aftershock) subset. Unlike traditional aftershock identification schemes that rely on fixed space‑time windows or on ad‑hoc declustering algorithms (e.g., Reasenberg’s method), the η‑based rule automatically incorporates magnitude information and is less sensitive to the exact choice of parameters.

To validate the methodology, three data sets are examined. (1) A synthetic homogeneous Poisson catalog serves as a null model; the η distribution matches the theoretical Weibull curve and the clustering rule yields virtually no false positives. (2) An Epidemic‑Type Aftershock Sequence (ETAS) simulation, in which aftershock clusters are explicitly embedded, shows a pronounced excess of points in the low‑η_s, low‑η_t region, and the clustering rule recovers >90 % of the implanted aftershocks with a low false‑negative rate. (3) The observed Southern California seismicity catalog (1970‑2005, M ≥ 2.5) is analyzed. The η‑based clustering identifies well‑known aftershock sequences such as the Landers‑Big Bear swarm (1992), the Northridge swarm (1994), and several smaller clusters. Comparison with previously published declustering results reveals an overlap of >85 %, confirming that the η‑method captures the same physical clusters while offering a more principled statistical basis.

The paper also discusses limitations and future directions. The current theory assumes a spatially homogeneous observation network and an exact exponential magnitude distribution; real catalogs exhibit varying detection thresholds and may deviate from pure exponential behavior at low magnitudes. Extending the framework to accommodate non‑uniform completeness, mixed magnitude distributions, or hierarchical clustering (multiple scales of aftershock activity) is identified as a priority. Moreover, the authors suggest that the η‑distance and its 2‑D decomposition could be applied to other geophysical phenomena (e.g., volcanic tremor clustering, induced seismicity) where space, time, and magnitude (or analogous marks) are relevant.

In summary, the study provides a rigorous probabilistic foundation for the η‑distance, demonstrates that η follows a Weibull law in a homogeneous Poisson setting, and introduces a practical 2‑D η‑analysis that cleanly separates clustered aftershocks from background seismicity. The approach outperforms traditional declustering schemes in robustness to parameter choice and in its natural integration of magnitude information, making it a valuable addition to the toolbox of seismologists and hazard modelers.