Characterizing Pick Error Models for Local Seismic Phases

Accurate manual picking of seismic arrivals is crucial for earthquake location, seismic tomography, and training deep learning phase-picking models. An error model serves as an effective tool for quantitatively assessing pick quality. However, establishing models for local phases (i.e., Pg and Sg) is challenging due to the lack of a common dataset of seismograms picked by diverse experts. In this study, we construct a large dataset by collecting waveforms and bulletins from the China Earthquake Networks Center and determining the pick differences between analysts across different provincial earthquake agencies for common phases, which consists of 49,983 Pg and 48,217 Sg phases. Results indicate that the pick quality of Pg phase is superior to that of Sg phase, and the pick quality of stations in Northwestern China is better than that in Northern China. The magnitude and distance dependence on error suggest the main controlling factor of pick quality may be the signal-to-noise ratio (SNR). To address the observed decreasing trend of pick error for SNR ≤ 100 and a general constant on seismograms with high SNR (> 100), we propose piecewise error models as a function of SNR, which can provide pick errors for most applications in seismology quantitatively.