A Nonparametric Approach to Automated S-Wave Picking

Abstract:

Although a number of very effective P-wave automatic pickers have been developed over the years, automatic picking of S waves has remained more challenging. Most automatic pickers take a parametric approach, whereby some characteristic function (CF), e.g. polarization or kurtosis, is determined from the data and the pick is estimated from the CF. We have adopted a nonparametric approach, estimating the pick directly from the waveforms. For a particular waveform to be auto-picked, the method uses a combination of similarity to a set of seismograms with known S-wave arrivals and dissimilarity to a set of seismograms that do not contain S-wave arrivals. Significant effort has been made towards dealing with the problem of S-to-P conversions.

We have evaluated the effectiveness of our method by testing it on multiple sets of microearthquake seismograms with well-determined S-wave arrivals for several areas around the world, including fault zones and volcanic regions. In general, we find that the results from our auto-picker are consistent with reviewed analyst picks 90% of the time at the 0.2 s level and 80% of the time at the 0.1 s level, or better. For most of the large datasets we have analyzed, our auto-picker also makes far more S-wave picks than were made previously by analysts. We are using these enlarged sets of high-quality S-wave picks to refine tomographic inversions for these areas, resulting in substantial improvement in the quality of the S-wave images. We will show examples from New Zealand, Hawaii, and California.