Deep learning for seismic phase detection and picking in the aftershock zone of 2008 M7.9 Wenchuan Earthquake

Abstract

The increasing volume of seismic data from long-term continuous monitoring motivates the development of algorithms based on convolutional neural network (CNN) for faster and more reliable phase detection and picking. However, many less studied regions lack a significant amount of labeled events needed for traditional CNN approaches. In this paper, we present a CNN-based Phase-Identification Classifier (CPIC) designed for phase detection and picking on small to medium sized training datasets. When trained on 30,146 labeled phases and applied to one-month of continuous recordings during the aftershock sequences of the 2008 M W 7.9 Wenchuan Earthquake in Sichuan, China, CPIC detects 97.5% of the manually picked phases in the standard catalog and predicts their arrival times with a five-times improvement over the ObsPy AR picker. In addition, unlike other CNN-based approaches that require millions of training samples, when the off-line training set size of CPIC is reduced to only a few thousand training samples the accuracy stays above 95%. The online implementation of CPIC takes less than 12 hours to pick arrivals in 31-day recordings on 14 stations. In addition to the catalog phases manually picked by analysts, CPIC finds more phases for existing events and new events missed in the catalog. Among those additional detections, some are confirmed by a matched filter method while others require further investigation. Finally, when tested on a small dataset from a different region (Oklahoma, US), CPIC achieves 97% accuracy after fine tuning only the fully connected layer of the model. This result suggests that the CPIC developed in this study can be used to identify and pick P/S arrivals in other regions with no or minimum labeled phases.

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

“…8 ). To compare the performances, we applied three deep-learning (DetNet 5 , Yews 4 , and CRED 7 ) detectors and one traditional (STA/LTA 11 ) detector to the same test set (Table 1 ). We should acknowledge that there is a level of tuning involved in each of these approaches (traditional and deep-learning detectors/pickers), and that the performance can vary based on this tuning.…”

Machine learning and fault rupture: a review

“…Wang et al [115] trained two separate convolutional networks to pick P and S arrival times, on a database of over 700000 picks in Japan. Zhu et al [113] also utlized on a CNN to study the aftershock sequences of the 2008 M W 7.9 Wenchuan Earthquake. For all studies, results in testing were found to outperfom existing methods.…”

“…Interestingly, models for phase picking appear to generalize better than models trained to detect earthquakes. Two recent studies report that their algorithms generalize well to regions outside of the training area, either without finetuning [115], or with minimal finetuning [113]. While this improved generalization could be attributed to larger training sets or model specificities, it might be the case that earthquake detection is intrinsically a more challenging problem in terms of generalization due to models indirectly learning specific Green's functions.…”

“…This procedure is applied to the identification of seismograms of good quality versus those of bad quality, with the goal of building a selection criteria of retaining only waveforms that have high signal-to-noise ratios, leading to more robust analyses. A recent study published by Zhu et al [113] similarly employs auto-encoders with the objective of separating seismic signals from noise. In this work, an auto-encoder trained on a large number of earthquake waveforms with high signal-to-noise ratios that are overlayed with noise waveforms was tasked with separating the signal from the noise.…”

“…A major effort in earthquake detection and location with ML algorithms has been focused on induced seismicity, either in geothermal fields [160], in mines [92], or in gas [103] and oil [91,100,101,113,123] extraction fields (in particular in Oklahoma). These studies aim at building automatic and more complete catalogs, and were described in Section 1.…”