11:30 〜 11:45
[S07-4-05] Evaluation of the P-wave detection method using higher order statistics
Detecting P-wave onsets in on-line processing is one of the important components for the real-time seismology. Conventionally, the STA/LTA method (the ratio between short-time average and long-time average) has been widely used for the P-wave detection (Allen, 1978). Since this method is reasonably robust and computationally less expensive, it has been used for long years, and the trigger time is used for the estimation of the location of earthquakes.
The P-wave detection method using higher order statics has been attracted researchers recently. The key of the method is to identify the change of distribution of amplitudes. In general, seismic noises tend to follow the Gaussian distribution. However, when a seismic signal arrives and the amplitude suddenly increases, the distribution of the amplitude becomes skewed. This change of distribution can be measured by the kurtosis, which is the forth moment of the data divided by the square of the variance. The kurtosis is zero for the Gaussian distribution, and it increases as the distribution is away from the Gaussian.
We applied this method to the strong motion dataset after the Tohoku earthquake in March 2011. The results are compared with the P-wave arrival time estimated by the STA/LTA method, which showed the arrival times estimated by the kurtosis approach was closer to the manually detected P-wave arrival times.
The P-wave detection method using higher order statics has been attracted researchers recently. The key of the method is to identify the change of distribution of amplitudes. In general, seismic noises tend to follow the Gaussian distribution. However, when a seismic signal arrives and the amplitude suddenly increases, the distribution of the amplitude becomes skewed. This change of distribution can be measured by the kurtosis, which is the forth moment of the data divided by the square of the variance. The kurtosis is zero for the Gaussian distribution, and it increases as the distribution is away from the Gaussian.
We applied this method to the strong motion dataset after the Tohoku earthquake in March 2011. The results are compared with the P-wave arrival time estimated by the STA/LTA method, which showed the arrival times estimated by the kurtosis approach was closer to the manually detected P-wave arrival times.