Earthquake Early Warning in Japan: Performance after the Tohoku earthquake and the improvement

Current earthquake early warning (EEW) systems lack the ability to appropriately handle multiple concurrent earthquakes, which led to many false alarms during the 2011 Tohoku earthquake sequence in Japan. This research uses a Bayesian probabilistic approach to handle multiple concurrent events for EEW. We integrates P-wave picking time and maximum displacement amplitude from both JMA and Hi-net seismic stations into a single algorithm. Importance Sampling method with a sequential proposal probability density function is used to estimate the earthquake parameters, i.e., hypocenterlocation, origin time, magnitude and local seismic intensity. A real data examplebased on two months data (March 9 to April 30, 2011) around the time of the 2011 M9 Tohoku earthquake is studied to verify the proposed algorithm. Our algorithm results in over 90% reduction in the number of incorrect warnings compared to theexisting EEW system operating in Japan.