Detecting and characterizing swarm-like seismicity

Tectonic earthquake swarms deviate significantly from the spatio-temporal evolution characteristic of mainshock-aftershock sequences. While earthquake sequences often begin with a dominant event called a mainshock, followed by a decaying rate of aftershocks governed by the Omori-Utsu law, earthquake swarms are defined by a gradual escalation of seismic activity lacking a singular, triggering large earthquake at the start of the cluster. In these sequences, peak magnitudes often emerge mid-sequence or later, frequently accompanied by distinct spatial migration and episodic bursts. This complex clustering evolution is driven by the interaction between steady tectonic loading and transient, short-term forcing mechanisms. Identifying these phenomena requires robust, unsupervised methodologies—a need that has spurred the development of various detection algorithms over recent decades.

This research provides a systematic evaluation of prevalent cluster-detection techniques and sequence characterization via the release of seismic moment over time. We apply four well-known (de-)clustering algorithms to identify clusters in space-time-magnitude space; subsequently, we evaluate each cluster using the statistical moment of the cluster source time function (i.e., the release of seismic moment over time). By utilizing thousands of synthetic catalogs generated through Epidemic-Type Aftershock Sequence (ETAS) modeling with time-varying background rates, we simulate realistic swarm behavior to test these tools. This synthetic framework allows us to define parametric boundaries that robustly differentiate swarms from mainshock-aftershock clusters. We then validate our findings against well-documented real-world datasets, including the 2010–2014 Pollino sequence and the Húsavík-Flatey transform fault in Northern Iceland. Additionally, show that the cluster classification to distinguish swarms from mainshock-aftershock sequences via the proposed statistics depends on the type of (de-)clustering algorithm used and, most importantly, on the cluster duration.  Accordingly, our results highlight that real-world application remains sensitive to algorithm choice and catalog completeness, suggesting that human oversight is still essential for precise swarm characterization and interpretation.