The integrated particle filter method: an approximate Bayesian earthquake early warning system
Stephen Wu, Institute of Statistical Mathematics
Monday, June 24, 2019 at 10:30 AM
- Building 3, Rambo Auditorium
- Sarah Minson
Motivated by the many false alarms during the 2011 Tohoku earthquake sequence in Japan, we developed a probabilistic approach to handle multiple concurrent earthquakes in earthquake early warning (EEW) system, called the integrated particle filter (IPF) method. We formulated a likelihood model that exploits information of both triggered and not yet triggered seismic stations, as well as the spatial distribution of seismic stations. A particle filter approach is used to perform real-time updating of the probabilistic estimations of source parameters, and the prediction uncertainty obtained in our model is used in an approximate Bayesian model class selection algorithm to estimate the number of concurrent events. Our algorithm results in over 90 per cent reduction in the number of incorrect warnings compared to the existing EEW system operating in Japan, and have shown good performance in places that do not have a dense seismic network.