3:30 PM - 5:00 PM
[STT41-P05] Ghost events by the hypocenter determination method based on the back-propagating wavefront from receivers
Keywords:hypocenter determination method
In the JpGU 2022 spring meeting, we reported new hypocenter determination method based on the reverse propagating wavefront from receivers. The advantage of this method is that discriminating seismic phase and grouping the arrival times into each event are not needed. We showed that the method works fine with the synthetic arrival time data in simple event and receiver distribution.
In this study, we apply our method to real observed dataset, 2022-01-22 Hyuga-nada earthquake, 2016 Kumamoto earthquakes, and 2015-01-1 ordinary (no large events) dataset in Japan. Over 90% events are retrieved. Next, we apply to 4 hours dataset just after 2011-03-11 Tohoku earthquake, and the extracted event rate is not so high (77%), although we can grasp a rough event distribution, but ghost (false) offshore events are extracted (Figure 1).
The reason of the appearance of the ghost event will be that 2 waves of 2 events near the coast of their origin time difference of about 1 minute is interpreted as P and S waves of one offshore event. To exclude such events, we set the lower limit of arrival time number according to the shortest distance from seismic receivers. Only the event of the arrival time number over the limit is adopted. We can suppress the appearance of the ghost event, although the event extraction rate is a little lower (75%) (Figure 2).
In our calculations, space grid interval divided by seismic velocity is longer than time grid interval. In order to trace the wavefront continuously across the grids, we assign the score to 1 (P phase) or 2 (S phase) grid points before and after exact arrival time point along time axis. The score weight is triangle shape. If we give the score only to the grid point exactly at arrival time, it happens that the wavefront disappears at some grids in our algorithm. If the event origin is there, it is difficult to pick up the event. To avoid it, we smooth the score distribution over the grid by assigning the score to a few grids after and before the exact wavefront.
We try other weight shapes. When we change the triangle weight to flat one, the results are almost same, except Tohoku earthquake. In the case of Tohoku earthquake, the event extraction rate is drastically lower (38%). When we restrict the point to exact arrival time (delta function), the event detecting sometimes fails and stops in the middle of the whole process in the other cases not mentioned here. As for Tohoku earthquake, the event extraction rate is lower (64%) but higher that of flat weight. The triangle weight seems be best for a while.
In this study, we apply our method to real observed dataset, 2022-01-22 Hyuga-nada earthquake, 2016 Kumamoto earthquakes, and 2015-01-1 ordinary (no large events) dataset in Japan. Over 90% events are retrieved. Next, we apply to 4 hours dataset just after 2011-03-11 Tohoku earthquake, and the extracted event rate is not so high (77%), although we can grasp a rough event distribution, but ghost (false) offshore events are extracted (Figure 1).
The reason of the appearance of the ghost event will be that 2 waves of 2 events near the coast of their origin time difference of about 1 minute is interpreted as P and S waves of one offshore event. To exclude such events, we set the lower limit of arrival time number according to the shortest distance from seismic receivers. Only the event of the arrival time number over the limit is adopted. We can suppress the appearance of the ghost event, although the event extraction rate is a little lower (75%) (Figure 2).
In our calculations, space grid interval divided by seismic velocity is longer than time grid interval. In order to trace the wavefront continuously across the grids, we assign the score to 1 (P phase) or 2 (S phase) grid points before and after exact arrival time point along time axis. The score weight is triangle shape. If we give the score only to the grid point exactly at arrival time, it happens that the wavefront disappears at some grids in our algorithm. If the event origin is there, it is difficult to pick up the event. To avoid it, we smooth the score distribution over the grid by assigning the score to a few grids after and before the exact wavefront.
We try other weight shapes. When we change the triangle weight to flat one, the results are almost same, except Tohoku earthquake. In the case of Tohoku earthquake, the event extraction rate is drastically lower (38%). When we restrict the point to exact arrival time (delta function), the event detecting sometimes fails and stops in the middle of the whole process in the other cases not mentioned here. As for Tohoku earthquake, the event extraction rate is lower (64%) but higher that of flat weight. The triangle weight seems be best for a while.