A comparative analysis of five commonly implemented declustering algorithms

Declustering of earthquake catalogs, that is determining dependent and independent events in an earthquake sequence, is a common feature of many seismological studies. While many different declustering algorithms exist, each has different performance and sensitivity characteristics. Here, we conduct a comparative analysis of the five most commonly used declustering algorithms: Garnder and Knopoff (1974), Uhrhammer (1986), Reasenberg (J Geophys Res: Solid Earth 90(B7):5479–5495, 1985), Zhuang et al. (J Am Stat Assoc 97(458):369–380, 2002), and Zaliapin et al. (Phys Rev Lett 101(1):4–7, 2008) in four different tectonic settings. Overall, we find that the Zaliapin et al. (Phys Rev Lett 101(1):4–7, 2008) algorithm effectively removes aftershock sequences, while simultaneously retaining the most information (i.e. the most events) in the output catalog and only slightly modifying statistical characteristics (i.e. the Gutenberg Richter b-value). Both Gardner and Knopoff (1974) and Zhuang et al. (J Am Stat Assoc 97(458):369–380, 2002) also effectively remove aftershock sequences, though they remove significantly more events than the other algorithms. Uhrhammer (1986) also effectively removes aftershock sequences and removes fewer events than Gardner and Knopoff (1974) or Zhuang et al. (J Am Stat Assoc 97(458):369–380, 2002), except when large magnitude events are present. By contrast, Reasenberg (J Geophys Res: Solid Earth 90(B7):5479–5495, 1985) only effectively removed aftershocks in one of the test regions.