11:00 AM - 1:00 PM
[SSS11-P01] 5-year performance check for the real-time automatic system for aftershock forecasting
Keywords:Automatic aftershock forecasting system, NIED Hi-net , Omori-Utsu law, Gutenberg-Richter law, Bayesian inference
To generate aftershock forecasts from the early stage, NIED started to operate the real-time forecasting system based on the Hi-net automatic hypocenter catalog since April 2017 (Omi et al., 2019). The system simultaneously evaluates the b-value of the Gutenberg-Richter law and the detection rate of earthquakes that change in elapsed time after the mainshock and then estimates aftershock probabilities based on the Omori-Utsu formula (Omi et al., 2013). After three hours from an M>5 earthquake detected by the Hi-net system, the system starts to calculate the aftershock probabilities for three different forecast duration of one day, three days, and seven days. This calculation runs every hour until four days after the first event.
In this presentation, we introduce 11 automatic results for the recent M>6 earthquakes, including retrospective two cases of the 2016 Kumamoto earthquakes (the largest foreshock and the mainshock).
First, we have checked the stability of the parameters of the Omori-Utsu and the Gutenberg-Richter laws for imperfectly detected aftershocks in the early stage after the M>6 event. These parameters are unstable immediately after the occurrence of a large earthquake unless incorporating the prior standard estimates, say, throughout Japan. After the 6-9 hours or later from the M>6 event, however, we confirm that the system is able to estimate stable results for most aftershock sequences. The exception is the case of the earthquake sequence of Off-Fukushima in Feb. 2021, which needs 12-15 hours until the parameter became stable. This was the only intraplate earthquake among the eleven events selected in this study.
Next, we extracted the aftershock probabilities that at least one M>5 earthquake would occur during the forecast periods and confirmed the presence of an earthquake of M>5 in the Hi-net catalog for the same period. Until 48 hours after the first event of M>6, M>5 earthquakes occurred after the Kumamoto Earthquakes (both the foreshock and the mainshock), the Hokkaido Eastern Iburi Earthquake, the Off-Fukushima Earthquake, and Hyuganada Earthquake in 2019. For three main shocks that occurred in the overriding plate, the system successfully forecasted within the first three-hour span that the aftershock activity would be more active than the average aftershock activity in Japan. On the other hand, the seismic activity after the Off-Fukushima Earthquake was estimated as comparable to the average. Since the magnitude of this earthquake was greater than 7, we can well expect that the M>5 aftershock probability should be high. All other earthquakes, including the Hyuganada Earthquake, were predicted to be less active than the average. Although the aftershock probability was less than 20% after the Hyuganada Earthquake, an M5.0 earthquake occurred 12 hours after the mainshock.
The system uses the Hi-net automatic hypocenter catalog, not including the S-net and DONET data yet. To enhance the system performance for offshore events, it is essential to improve the accuracy of hypocenter location by adding data from these offshore networks.
Acknowledgment: This study is supported by the MEXT Project for Seismology toward Research Innovation with Data of Earthquake (STAR-E) Grant Number JPJ010217.
In this presentation, we introduce 11 automatic results for the recent M>6 earthquakes, including retrospective two cases of the 2016 Kumamoto earthquakes (the largest foreshock and the mainshock).
First, we have checked the stability of the parameters of the Omori-Utsu and the Gutenberg-Richter laws for imperfectly detected aftershocks in the early stage after the M>6 event. These parameters are unstable immediately after the occurrence of a large earthquake unless incorporating the prior standard estimates, say, throughout Japan. After the 6-9 hours or later from the M>6 event, however, we confirm that the system is able to estimate stable results for most aftershock sequences. The exception is the case of the earthquake sequence of Off-Fukushima in Feb. 2021, which needs 12-15 hours until the parameter became stable. This was the only intraplate earthquake among the eleven events selected in this study.
Next, we extracted the aftershock probabilities that at least one M>5 earthquake would occur during the forecast periods and confirmed the presence of an earthquake of M>5 in the Hi-net catalog for the same period. Until 48 hours after the first event of M>6, M>5 earthquakes occurred after the Kumamoto Earthquakes (both the foreshock and the mainshock), the Hokkaido Eastern Iburi Earthquake, the Off-Fukushima Earthquake, and Hyuganada Earthquake in 2019. For three main shocks that occurred in the overriding plate, the system successfully forecasted within the first three-hour span that the aftershock activity would be more active than the average aftershock activity in Japan. On the other hand, the seismic activity after the Off-Fukushima Earthquake was estimated as comparable to the average. Since the magnitude of this earthquake was greater than 7, we can well expect that the M>5 aftershock probability should be high. All other earthquakes, including the Hyuganada Earthquake, were predicted to be less active than the average. Although the aftershock probability was less than 20% after the Hyuganada Earthquake, an M5.0 earthquake occurred 12 hours after the mainshock.
The system uses the Hi-net automatic hypocenter catalog, not including the S-net and DONET data yet. To enhance the system performance for offshore events, it is essential to improve the accuracy of hypocenter location by adding data from these offshore networks.
Acknowledgment: This study is supported by the MEXT Project for Seismology toward Research Innovation with Data of Earthquake (STAR-E) Grant Number JPJ010217.