2014 年 67 巻 2 号 p. 41-55
The extremely active seismicity after the 2011 off the Pacific Coast of Tohoku Earthquake (Tohoku Earthquake) caused serious issues for the Earthquake Early Warning (EEW) system in Japan. Multiple concurrent earthquakes generated many false alarms due to the failure of correctly determining the source location. In this study, we propose a method to distinguish multiple concurrent events for EEW. We used the outputs of a B-delta method and principal component analysis, together with P-wave arrival times and amplitudes of triggered stations. The information of non-triggered stations is also used to constrain the locations. We constructed a likelihood function using the source parameters and solved this optimization problem using the particle filter. The appropriate choice of a likelihood function improves the accuracy of the location determination, and the probabilistic framework evaluates the uncertainty of the hypocenters. We applied this method to 72 events for which warnings were issued after the Tohoku Earthquake. The performance of EEW was greatly improved, reducing the false alarms from 22 to 0. The percentage of inaccurate warnings was also reduced from 66% to 6%.