Deep Learning for Picking Seismic Arrival Times

Abstract

Arrival times of seismic phases contribute substantially to the study of the inner working of the Earth. Despite great advances in seismic data collection, the usage of seismic arrival times is still insufficient because of the overload manual picking tasks for human experts. In this work we employ a deep-learning method (PickNet) to automatically pick much more P and S wave arrival times of local earthquakes with a picking accuracy close to that by human experts, which can be used directly to determine seismic tomography. A large number of high-quality seismic arrival times obtained with the deep-learning model may contribute greatly to improve our understanding of the Earth's interior structure.Plain Language Summary Deep learning is currently attracting immense research interest in seismology due to its powerful ability to deal with huge seismic data collections. In this study we developed a deep-learning model (PickNet) that can rapidly pick a great number of first P and S wave arrival times precisely from local earthquake seismograms. The picking accuracy of the arrival times provided by our PickNet model is close to that by human experts. The data are good enough to be used directly to determine high-resolution 3-D velocity models of the Earth. Our PickNet model can deal with seismic waveforms provided by data centers of different earthquake networks. Furthermore, our PickNet model is also a potential tool for automatically picking later seismic phases accurately. A large number of high-quality seismic arrival times can be used to illuminate the Earth structure clearly. Hence, this study may greatly contribute to improve our knowledge of the Earth's interior.

Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking

“… Model μ σ Pr Re F1 MAE MAPE Training data Training size Ref . EQTransformer 0.00 0.11 0.99 0.96 0.98 0.01 0.00 Global 1.2M This Study PhaseNet −0.02 0.11 0.96 0.93 0.94 0.09 0.01 North California 780K 8 GPD 0.03 0.14 0.81 0.83 0.82 0.10 0.01 South California 4.5M 10 PickNet 0.08 0.17 0.75 0.75 0.75 0.10 0.03 Japan 740K 2 PpkNet 0.02 0.15 1.00 0.91 0.95 0.10 1.85 Japan 30K 5 Yews −0.02 0.13 0.83 0.55 0.66 0.11 0.01 Taiwan 1.4M 4 Kurtosis −0.10 0.13 0.89 0.39 0.55 0.11 0.01 — — 17 FilterPicker ...…”

“…We now compare the picking performance with five deep-learning (PhaseNet 8 , GPD 10 , PpkNet 5 , Yews 4 , PickNet 2 ) (Supplementary Fig. 9 ) and three traditional (Kurtosis 17 , FilterPicker 18 , and AIC 19 ) (Supplementary Fig.…”

“…The error distributions for some of the deep-learning pickers are not uniform and cluster at sporadic times perhaps due to their moving-window scheme. All of these models (GPD 10 , PpkNet 5 , and Yews 4 ) use wider labels compared to the other models (PhaseNet 8 , PickNet 2 , and EQTransformer). However, it is difficult to narrow down the exact reason behind their non-normal error distributions.…”

“…Earthquake monitoring has a growing need for more efficient and robust tools for processing of increasingly large data volumes, is conceptually straightforward, and has a large quantity of available labeled data, which make earthquake detection and phase picking attractive targets for the new wave of machine learning applications in seismology. To date, earthquake signal detection and phase-picking form the largest portion of this relatively young sub-field 2 – 10 . Despite the differences in approaches and results, most of these studies find important advantages to deep-learning-based methods compared with traditional approaches 11 , 12 .…”

Rapid Characterization of the July 2019 Ridgecrest, California, Earthquake Sequence From Raw Seismic Data Using Machine‐Learning Phase Picker

“…In recent studies, machine-learning phase pickers have shown promise in both picking efficiency and picking accuracy (e.g., Ross et al, 2018;Wang et al, 2019;Zhu & Beroza, 2019). However, the precision and resolution of earthquake locations determined from those machine-learning picks have not been well studied.…”