9:30 AM - 9:45 AM
[SSS08-03] Development of an earthquake swarm detection method using the ETAS model and AIC and its application to the offshore of the Boso Peninsula
Keywords:Earthquake swarm, ETAS model
An increase in the seismicity rate without a clear main shock is called an earthquake swarm (e.g., Mogi, 1963). Examples of observed earthquake swarms include seismicity associated with ascending crustal fluids (e.g., Tsuneishi & Nakamura, 1970) and slow slip events (SSEs) in subduction zones (e.g., Ozawa et al., 2003). Earthquake swarm detection is important for elucidating the relationship between aseismic phenomena and seismicity. In this study, we developed a new, simple method for detecting earthquake swarms based on the epidemic-type aftershock-sequence (ETAS) model (Ogata, 1988) and Akaike’s information criterion (AIC; Akaike, 1974). We then conducted an earthquake swarm detection along the Sagami Trough off the Boso Peninsula, where SSEs accompanied by earthquake swarms have been reported (e.g., Ozawa et al., 2003).
The ETAS model expresses the seismicity rate at time t (λ(t)) as the summation of the stationary background seismicity rate (μ) and aftershock rates (Σti<t Kexp(α(Mi-Mc))/(t-ti+c)p) following Omori–Utsu’s aftershock law (e.g., Utsu, 1957), where ti is the origin time of the i-th earthquake, Mi is the magnitude of the i-th earthquake, and α, c, K, and p are model parameters controlling aftershock rates.
In this study, based on the ETAS model improved by Okutani & Ide (2011), we developed a new model that takes into account an earthquake swarm by considering the increment of the background seismicity rate during an earthquake swarm period (μ1(t)). μ1(t) is expressed as the product of the background event increase during a swarm period (swc) and the normalized Gaussian function (1/(2π Tsws2 )1/2exp(-(t-Tswc)2/(2 Tsws2))), where Tswc is the peak time of the background seismicity rate during the swarm period, Tsws is the standard deviation of the normalized Gaussian function. Note that swc, Tswc, and Tsws are new model parameters.
In this study, the peak time of the background seismicity rate during an earthquake swarm period, Tswc, was grid-searched with 1-day intervals, and the other seven parameters (μ, α, c, K, p, swc, and Tsws) were estimated via the maximum likelihood method. We extracted dates when the difference of AIC (ΔAIC) between the model considering an earthquake swarm and the original ETAS model was less than -2 and considered those dates as peak dates of the background seismicity rate during an earthquake swarm period.
We then conducted an earthquake swarm detection along the Sagami Trough off the Boso Peninsula (34.8°N to 35.6°N, 139.9°E to 140.9°E) in the 0 to 50 km depth range. The analysis period was 11 years, from January 1, 2000 to December 31, 2010. This period includes earthquake swarms associated with large SSEs in October 2002 and August 2007. We used earthquakes of magnitude 2.0 or greater in the Japan Meteorological Agency catalog.
The results showed that there were 16 dates with a minimum ΔAIC of -2 or less, of which five dates had a substantial decrease in ΔAIC (-10 or less). Two of the five dates corresponded to the earthquake swarms associated with the 2002 and 2007 SSEs. The other three dates corresponded to earthquake swarms that occurred within small areas in Tokyo Bay and the southeastern part of the Boso Peninsula. All of these earthquake swarms occurred near the upper surface of the Philippine Sea Plate.
The successful detection of the significant ΔAIC decreases corresponding to the known earthquake swarms supports the usefulness of our method. The method also detected ΔAIC decreases corresponding to earthquake swarms that have not been reported in previous studies. Among them, at least two seismic sequences occurred simultaneously with smaller SSEs detected by Nishimura (2021). Some of the new earthquake swarms may correspond to unknown smaller SSEs, and further investigation using geodetic data, such as Global Navigation Satellite System data, is needed.
The ETAS model expresses the seismicity rate at time t (λ(t)) as the summation of the stationary background seismicity rate (μ) and aftershock rates (Σti<t Kexp(α(Mi-Mc))/(t-ti+c)p) following Omori–Utsu’s aftershock law (e.g., Utsu, 1957), where ti is the origin time of the i-th earthquake, Mi is the magnitude of the i-th earthquake, and α, c, K, and p are model parameters controlling aftershock rates.
In this study, based on the ETAS model improved by Okutani & Ide (2011), we developed a new model that takes into account an earthquake swarm by considering the increment of the background seismicity rate during an earthquake swarm period (μ1(t)). μ1(t) is expressed as the product of the background event increase during a swarm period (swc) and the normalized Gaussian function (1/(2π Tsws2 )1/2exp(-(t-Tswc)2/(2 Tsws2))), where Tswc is the peak time of the background seismicity rate during the swarm period, Tsws is the standard deviation of the normalized Gaussian function. Note that swc, Tswc, and Tsws are new model parameters.
In this study, the peak time of the background seismicity rate during an earthquake swarm period, Tswc, was grid-searched with 1-day intervals, and the other seven parameters (μ, α, c, K, p, swc, and Tsws) were estimated via the maximum likelihood method. We extracted dates when the difference of AIC (ΔAIC) between the model considering an earthquake swarm and the original ETAS model was less than -2 and considered those dates as peak dates of the background seismicity rate during an earthquake swarm period.
We then conducted an earthquake swarm detection along the Sagami Trough off the Boso Peninsula (34.8°N to 35.6°N, 139.9°E to 140.9°E) in the 0 to 50 km depth range. The analysis period was 11 years, from January 1, 2000 to December 31, 2010. This period includes earthquake swarms associated with large SSEs in October 2002 and August 2007. We used earthquakes of magnitude 2.0 or greater in the Japan Meteorological Agency catalog.
The results showed that there were 16 dates with a minimum ΔAIC of -2 or less, of which five dates had a substantial decrease in ΔAIC (-10 or less). Two of the five dates corresponded to the earthquake swarms associated with the 2002 and 2007 SSEs. The other three dates corresponded to earthquake swarms that occurred within small areas in Tokyo Bay and the southeastern part of the Boso Peninsula. All of these earthquake swarms occurred near the upper surface of the Philippine Sea Plate.
The successful detection of the significant ΔAIC decreases corresponding to the known earthquake swarms supports the usefulness of our method. The method also detected ΔAIC decreases corresponding to earthquake swarms that have not been reported in previous studies. Among them, at least two seismic sequences occurred simultaneously with smaller SSEs detected by Nishimura (2021). Some of the new earthquake swarms may correspond to unknown smaller SSEs, and further investigation using geodetic data, such as Global Navigation Satellite System data, is needed.