Automatic S-phase arrival determination of seismic signals using nonlinear filtering and higher-order statistics

Automatic seismic P- and S-phases arrival identification has been a challenging scientific goal for the last two decades for seismologists, as it provides them with important seismological information. Based on recent bibliography, various approaches result in efficient P-phase identification, while S- phase identification is still an open problem, since there is an overlap of the S- phase arrival and the P- phase coda. They usually employ simple energy ratio criteria, the linear seismic wave polarity assumption, neural networks or heuristic methods based on seismologists' experience, but they seem to have moderate performance, especially for noisy cases. The proposed method uses a Wavelet Transform-based filtering technique that separates the STationary from the NonSTationary parts of a mixed signal, namely the WTST-NST filter, combined with Higher-Order Statistics (HOS). Initially, the WTST-NST filter is used to remove the undesired P-phase coda (given the P-phase arrival) and the background noise from the seismic data. Then, the HOS are applied on the de-noised signal to detect the S-phase arrival at the location of the maximum value of HOS. The method has been applied on real seismic data recorded in Greece, characterized by human experts as highly noisy cases. Experimental results indicate that the proposed method identifies the S-phase arrival efficiently, since evaluation analysis proves high identification accuracy, when compared to the analysts' findings and due to its low computational complexity, real-time implementation is feasible.