Automatic seismic phase picking based on unsupervised machine learning classification and content information analysis

Abstract

Accurate identification and picking of P- and S-wave arrivals is important in earthquake and exploration seismology. Often, existing algorithms lack in automation, multi-phase classification and picking, as well as performance accuracy. A new fully-automated fourth-step workflow for efficient classification and picking of P- and S-wave arrival times on microseismic datasets is presented. First, time intervals with possible arrivals on the waveform recordings are identified using the fuzzy c-means clustering algorithm. Second, these signal intervals are classified as corresponding to P, S, or unidentified waves using the polarization attributes of the waveforms contained within. Third, the P-, S-, and unidentified-waves arrival times are picked using the Akaike information criterion picker on the corresponding intervals. Fourth, unidentified waves are classified as P or S based on the arrivals moveouts. The application of the workflow on synthetic and real microseismic datasets shows that it yields accurate arrival picks for both high and low signal-to-noise ratio waveforms.