PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking Method

Abstract

As the number of seismic sensors grows, it is becoming increasingly difficult for analysts to pick seismic phases manually and comprehensively, yet such efforts are fundamental to earthquake monitoring. Despite years of improvements in automatic phase picking, it is difficult to match the performance of experienced analysts. A more subtle issue is that different seismic analysts may pick phases differently, which can introduce bias into earthquake locations. We present a deep-neural-networkbased arrival-time picking method called "PhaseNet" that picks the arrival times of both P and S waves. Deep neural networks have recently made rapid progress in feature learning, and with sufficient training, have achieved super-human performance in many applications. PhaseNet uses three-component seismic waveforms as input and generates probability distributions of P arrivals, S arrivals, and noise as output.We engineer PhaseNet such that peaks in probability provide accurate arrival times for both P and S waves, and have the potential to increase the number of S-wave observations dramatically over what is currently available. This will enable both improved locations and improved shear wave velocity models. PhaseNet is trained on the prodigious available data set provided by analyst-labeled P and S arrival times from the Northern California Earthquake Data Center. The dataset we use con-arXiv:1803.03211v1 [physics.geo-ph] 8 Mar 2018 2 tains more than seven million waveform samples extracted from over thirty years of earthquake recordings. We demonstrate that PhaseNet achieves much higher picking accuracy and recall rate than existing methods.

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

“…From the training process of discriminating seismic events from noise on large datasets, the weights of the local filters collectively capture the intrinsic features that most effectively represent seismograms for the given task of phase picking. In the next sections, we show that CPIC, trained on a much smaller labeled dataset, achieves comparable classification accuracy as reported in Ross et al (2018a) and Zhu & Beroza (2018). CPIC is further tested on a one-month continuous aftershock dataset for phase detection.…”

“…Unlike recent CNN studies that rely on an exceptionally rich training dataset of labeled samples (Zhu & Beroza, 2018;Ross et al, 2018a) to achieve good accuracy and robustness against noise, we design CPIC and study its performance on a relatively small training set prior to applying it on a large volume of unlabeled data. This is a typical scenario when analyzing the aftershock dataset of a major earthquake: strong aftershocks at a later time can be easily picked by existing algorithms or analysts; however, the real targets are the numerous number of aftershocks right after the mainshock that are missed by traditional methods (Kagan, 2004;Peng et al, 2006).…”

“…Unlike waveform similarity methods, CNNs trained on labeled datasets do not need a growing library of templates and seems to generalize well to waveforms not seen during training. These recent developments have led to CNNs being applied to diverse seismic data sets (Kong et al, 2018), including volcanic events (Luzn et al, 2017), induced seismicity (Perol et al, 2018), aftershocks , as well as regular tectonic earthquakes recorded by regional seismic networks (Ross et al, 2018b,a;Zhu & Beroza, 2018). However, most of these works rely on a large volume of labeled training data which is only available in well-studied regions, such as California,…”

A Machine‐Learning Approach for Earthquake Magnitude Estimation

“…• Table S1 to develop a full deep-learning pipeline ; ; Zhu and Beroza (2018); ) for earthquake signal processing and monitoring.…”

“…Our goal in this study was to develop a method for a fast and reliable estimation of earthquake magnitude directly from raw seismograms recorded on a single station. This is part of a larger project aiming to develop a full deep‐learning pipeline (Zhu et al (); Mousavi et al (); Zhu and Beroza (); Mousavi et al, , Mousavi et al, ]) for earthquake signal processing and monitoring.…”