A new automatic S-onset detection technique: Application in local earthquake data

Abstract

Algorithms that deal with the automatic S-onset time identification problem are a topic of ongoing research. Modern dense seismic networks used for earthquake location, seismic tomography investigations, source studies, early warning, etc., demand accurate automatic S-wave picking. Most of the techniques that have been proposed up to now are mainly based on the polarization features of the seismic waves. We propose a new time domain method for the automatic determination of the S-phase arrival onsets, and present its implementation on local earthquake data. Eigenvalue analysis takes place over small time intervals, and the maximum eigenvalue which is obtained on each step is retained for further processing. In this way, a time series of maximum eigenvalues is formed, which serves as a characteristic function. We obtain a first S-phase arrival time estimation by applying the kurtosis criterion on the derived characteristic function. Furthermore, a multiwindow approach combined with an energy-based weighting scheme is also applied, to reduce the algorithm's dependence on the moving window's length and provide a weighted S-phase onset. Automatic picks were compared against manual reference picks, resulting in mean residual time of 0.051 s. Moreover, the proposed technique was subjected to a noise robustness test and sustained a good performance. The mean residual time remained lower than 0.1 s, for noise levels between −1 up to 8 dB. The proposed method is easy to implement, because it is almost parameter free and demands low computational resources.

Development of an adaptive multi‐method algorithm for automatic picking of first arrival times: application to near surface seismic data

“…The onset time of the considered phase can be determined using the maximum CF k derivative (Lois et al . ; Langet et al . ).…”

“…Final picking : In analogy with the multi‐window and weighting scheme proposed by Lois et al . (), the final pick time ( tP final ) is achieved by performing the weighted average of the set of picks obtained by the previous pickers ( tP i ; i = 1, 2, 3) according to: where Q i is the quality factor of the picked phase calculated within a predefined time windows and is given in dB (Küperkoch et al . ): where A n is the root mean square (RMS) amplitude of the noise calculated within a time window of length ( nois_wind = 3T d ) before determining tP i .…”

“…The very beginning of this increase matches well the human-operator picking (Baillard et al 2014). The onset time of the considered phase can be determined using the maximum CF k derivative (Lois et al 2013;Langet et al 2014). To address this issue, we applied the successive transformations proposed by Baillard et al (2014) to CF k followed by smoothing; in a similar fashion as in MNW method, we can also apply low pass filtering to CF k by using the same function and the smoothing parameters.…”

“…All potential first-arrival onsets and their associated errors picked by the sub-algorithms are shown in Final picking: In analogy with the multi-window and weighting scheme proposed by Lois et al (2013), the final pick time (tP final ) is achieved by performing the weighted average of the set of picks obtained by the previous pickers (tP i ; i = 1, 2, 3) according to:…”

“…10(a) shows that about 35% of the differences are found at 10 milliseconds, and 75% are within [0, 20] milliseconds. Figure 10(b) shows a statistical analysis per quality class, which is determined in analogy with the empirical uncertainty categories (A-D) of Lois et al (2013). Clearly, all distributions of the four classes are found on the positive side of the difference between automatic and actual picks.…”