Online Seismic Event Picking via Sequential Change-Point Detection

Seismic event picking plays a key role in seismology studies. The goal of seismic event picking is to detect the onset of a seismic event, which typically causes an increase in the amplitude of the recorded signal. In this paper, we present a sequential change-point detection method for online seismic event detection, based on the generalized maximum likelihood statistic. We assume that the signals prior to the event are i.i.d. Gaussian random variables with zero mean and known variance, and after the event are i.i.d. Gaussian with zero mean and an increased unknown variance. We form a generalized likelihood ratio (GLR) based statistic by replacing the unknown variance with its maximum likelihood estimate, which takes a simple form and has a recursive implementation. An event is detected whenever the statistic exceeds a prescribed threshold. We compare the performance of our GLR procedure relative to the commonly used short-term-average/long-term-average (STA/LTA) algorithm, which is the state-of-the-art for seismic event detection, using large-scale seismic dataset and demonstrate the benefits of our GLR statistic. We also present a joint detection method to utilize the capability of seismic sensors to record signals through three independent channels, to achieve much better detection performance. We also present a method to combine GLR procedure with P-wave and S-wave filtering.