5:15 PM - 7:15 PM
[SCG45-P33] On the recurrence of slow slip events using the Brownian Passage Time distribution: Four case studies in Japan
Keywords:Slow Slip Event, Statistical forecasts
Slow slip events (SSEs) have been observed in many subduction zones where megathrust earthquakes occur, detected using geodetic instruments (Beroza & Ide, 2011; Obara & Kato, 2016; Schwartz & Rokosky, 2007). These events typically occur at downdip extensions of strongly locked seismogenic zones on plate interfaces, releasing accumulated interplate stress over periods ranging from days to years. The resulting stress perturbations may influence the occurrence of megathrust earthquakes (cf. Graham et al., 2014; Ito et al., 2013; Kato et al., 2012; Ozawa et al., 2012; Radiguet et al., 2016; Ruiz et al., 2014; Socquet et al., 2017; Voss et al., 2018; Yokota & Koketsu, 2015). Therefore, understanding SSEs provides valuable insights into the mechanisms of megathrust earthquakes.
We analyze a statistical property of SSEs, the recurrence intervals of SSEs. In the literature of the recurrence analysis of earthquake events, for the Kamaishi-Oki earthquake, which was considered to behave as a characteristic earthquake, predictions based on the mean interval and standard deviation of past occurrences did not match unless the confidence interval was extended to 99%. This highlights the importance of efficient statistical evaluation of the recurrence for practical predictions (c.f., Matsuzawa et al., 2002). For the recurrence interval analysis of slow earthquakes, several statistical modeling of low frequency earthquakes has been proposed (Langline et al., 2017; Tan and Marsan, 2020; Ide and Nomura, 2022; Nishikawa, 2024).
In this presentation, we analyze their recurrence intervals in four different regions, the Kii Peninsula (Araki et al., 2017; Ariyoshi et al., 2021), Yaeyama (Heki and Kataoka, 2008; Tu and Heki, 2017, Kano et al., 2018), Boso Peninsula (Fukuda, 2017; Ozawa et al., 2019), and the Bungo Channel (c.f, Ozawa et al., 2024), using the Brownian Passage Time (BPT) distribution. The BPT distribution models the timing of events when a perturbed linear load state process reaches a threshold and has been widely applied to analyze recurrent earthquake sequences (c.f., Nomura et al., 2009). The coefficient of variation (CV), that is, the variance of the recurrence interval divided by the mean recurrence interval, is a key parameter in recurrence analysis: if 0 < CV < 1/sqrt(2), the quasi-stationary failure rate exceeds the mean failure rate (Matthews et al., 2002) and the failure rate may not be completely random or completely deterministic.
We performed maximum likelihood estimation (MLE) and assess uncertainties using the Fisher information matrix and bootstrap techniques. While sequential parameter estimation exhibits temporal variability in CV across the four regions, (i) CV is not around zero, declining the complete randomness of the occurrence; (ii) the values are generally higher than 0.24 (The Headquaters for Earthquake Research Promotion, 2002) and 0.35 (Nomura et al., 2011), those used in the evaluation of seismic activities in Japan; (iii) the values consistently remain below 1/sqrt(2) throughout the observation period, even considering uncertainty estimates. Further details will be reported.
We analyze a statistical property of SSEs, the recurrence intervals of SSEs. In the literature of the recurrence analysis of earthquake events, for the Kamaishi-Oki earthquake, which was considered to behave as a characteristic earthquake, predictions based on the mean interval and standard deviation of past occurrences did not match unless the confidence interval was extended to 99%. This highlights the importance of efficient statistical evaluation of the recurrence for practical predictions (c.f., Matsuzawa et al., 2002). For the recurrence interval analysis of slow earthquakes, several statistical modeling of low frequency earthquakes has been proposed (Langline et al., 2017; Tan and Marsan, 2020; Ide and Nomura, 2022; Nishikawa, 2024).
In this presentation, we analyze their recurrence intervals in four different regions, the Kii Peninsula (Araki et al., 2017; Ariyoshi et al., 2021), Yaeyama (Heki and Kataoka, 2008; Tu and Heki, 2017, Kano et al., 2018), Boso Peninsula (Fukuda, 2017; Ozawa et al., 2019), and the Bungo Channel (c.f, Ozawa et al., 2024), using the Brownian Passage Time (BPT) distribution. The BPT distribution models the timing of events when a perturbed linear load state process reaches a threshold and has been widely applied to analyze recurrent earthquake sequences (c.f., Nomura et al., 2009). The coefficient of variation (CV), that is, the variance of the recurrence interval divided by the mean recurrence interval, is a key parameter in recurrence analysis: if 0 < CV < 1/sqrt(2), the quasi-stationary failure rate exceeds the mean failure rate (Matthews et al., 2002) and the failure rate may not be completely random or completely deterministic.
We performed maximum likelihood estimation (MLE) and assess uncertainties using the Fisher information matrix and bootstrap techniques. While sequential parameter estimation exhibits temporal variability in CV across the four regions, (i) CV is not around zero, declining the complete randomness of the occurrence; (ii) the values are generally higher than 0.24 (The Headquaters for Earthquake Research Promotion, 2002) and 0.35 (Nomura et al., 2011), those used in the evaluation of seismic activities in Japan; (iii) the values consistently remain below 1/sqrt(2) throughout the observation period, even considering uncertainty estimates. Further details will be reported.