Automatic P-Wave Arrival Detection and Picking with Multiscale Wavelet Analysis for Single-Component Recordings

TLDR

Wavelet analysis captures fine and coarse structures, with primary seismic features such as P‑wave arrivals persisting across multiple resolution scales while secondary arrivals fade at lower resolutions. The authors apply a sliding‑window discrete wavelet transform to single‑component seismograms, threshold the absolute coefficients at each scale, use an AIC autopicker to test consistency across scales, and then determine the arrival time with the AIC picker on the selected window. The algorithm achieves 81 % of picks within 0.2 s of analyst picks for the Dead Sea Rift dataset and 93 % within 0.1 s for the Parkfield dataset, with lower agreement on the Dead Sea data attributed to lower signal‑to‑noise ratio and possible analyst errors.

Abstract

We have developed an automatic P -wave arrival detection and picking algorithm based on the wavelet transform and Akaike information criteria (AIC) picker. Wavelet coefficients at high resolutions show the fine structure of the time series, and those at low resolutions characterize its coarse features. Primary features such as the P -wave arrival are retained over several resolution scales, whereas secondary features such as scattered arrivals decay quickly at lower resolutions. We apply the discrete wavelet transform to single-component seismograms through a series of sliding time windows. In each window the AIC autopicker is applied to the thresholded absolute wavelet coefficients at different scales, and we compare the consistency of those picks to determine whether a P -wave arrival has been detected in the given time window. The arrival time is then determined using the AIC picker on the time window chosen by the wavelet transform. We test our method on regional earthquake data from the Dead Sea Rift region and local earthquake data from the Parkfield, California region. We find that 81% of picks are within 0.2-sec of the corresponding analyst pick for the Dead Sea dataset, and 93% of picks are within 0.1 sec of the analyst pick for the Parkfield dataset. We attribute the lower percentage of agreement for the Dead Sea dataset to the substantially lower signal-to-noise ratio of those data, and the likelihood that some percentage of the analyst picks are in error.