Study On The Automatic Identification Of Local Seismic P-Phases

Detection and identification of seismic phase plays an important role in many seismological fields, such as earthquake location, earthquake early warning and the earth’s deep interior. Automatic recognition of seismic phases can greatly improve the speed of earthquake rapid report and earthquake early warning, saving precious time for post-seismic emergency and succor.We have summarized three kinds of methods for automatic seismic phases identification used in practice:single-feature methods (including energy analysis, polarization analysis, higher order statistics, fractal dimension, Akaike information criterion (AIC), spectrum analysis), multi-feature methods (such as full wave phase analysis, correlation method, artificial neural network), and comprehensive analysis methods. After analyzing and comparing the benefits and drawbacks of these methods, we point out the developing direction of automatic recognition of seismic phases.On the basis of previous studies, we present a new method named wavelet packet Kurtosis-AIC (WP-KAIC) for automatic identifying P-wave first arrivals. Our method consists of three parts. Firstly, the weighted STA/LTA (short term average/long term average) method is used to detect effective seismic event automatically and to roughly pick up its first P-wave arrival time tp. Secondly, three scales of discrete wavelet packet transform is applied to decomposing and restructuring the original signal within the time window [tp-3s, tp+3s]. Finally, the Kurtosis-AIC values of the three-scale reconstruction signals are calculated respectively and then stacked together. The minimum value of the stacked AIC curve is taken as the first P-wave arrival time.In order to test the new method, we have applied the WP-KAIC method to synthetic theory seismograms of a virtual event, which is designed by a real seismic in Yunnan region. Adding white Gaussian noise and real seismic noise to synthetic seismograms with different SNR (signal-to-noise ratio), we used the theoretical arrival time calculated with a ray tracing technique as the reference standard, and compared our method with the weighted STA/LTA algorithm and Kurtosis-AIC method for automatic first P-wave detection. Results show that our method has greater noise immunity and higher P-wave recognition accuracy.Taking 722 near earthquake vertical records of Yunnan region as an example, we have studied the effect of the filtering method, SNR, and the clarity of first breaks on the P-phase identification accuracy. Results show that FIR (finite impulse response) digital filter with optimal frequency band can significantly improve the SNR and P-wave recognition accuracy. Compared with the factor of SNR, the clarity of first breaks has more effect on the accuracy of P-wave identification. Using the first P-phase arrival times picked by experienced seismologists as the reference standard, we find our method is better than the weighted STA/LTA algorithm and Kurtosis-AIC method. Comparison of the P-wave travel time curve of manual identification with that of automatic identification further verify our method is reliable.