Rapid Earthquake Discrimination for Earthquake Early Warning: A Bayesian Probabilistic Approach Using Three-Component Single-Station Waveforms and Seismicity Forecast

摘要

The utility of Earthquake Early Warning (EEW) relies on the robust and rapid classification of near-site earthquake source signals from noise and teleseismic arrivals. To achieve this goal, we propose using the three-component acceleration and velocity waveform data and epidemic-type aftershock sequence (ETAS) seismicity forecast information in parallel, which will produce a posterior prediction by combining the predictions from the heterogeneous sources using a Bayesian probabilistic approach. We collected 2481 three-component strong-motion records for training and testing. The rapid prediction is available as quickly as 0.5 s after the trigger at a single station and updates every 0.5 s up to 3.0 s, achieving a precision rate of 94.7% at the first prediction with the classification accuracy increasing with time. The leave-one-out cross-validation method also demonstrates confidence of robust performance for future earthquake signal detections. We compared the method with the tau(c) - P-d EEW classification criterion and find that our prediction is 83% faster. Because the method evaluates two independent sources of information simultaneously under an ensemble model, the new strategy has shown fast predictions with promising results and the implementation of this methodology could provide significantly faster and more reliable EEW warnings to regions near the earthquake's epicenter, where the strongest shaking is observed.