Earthquake transformer\u2014an attentive deep-learning model for simultaneous earthquake detection and phase picking

S Mostafa Mousavi,William L Ellsworth,Lindsay Y Chuang,Weiqiang Zhu,Gregory C Beroza

doi:10.1038/s41467-020-17591-w

S Mostafa Mousavi, William L Ellsworth + Show 3 more

Open Access

https://doi.org/10.1038/s41467-020-17591-w

Copy DOI

Abstract

Earthquake signal detection and seismic phase picking are challenging tasks in the processing of noisy data and the monitoring of microearthquakes. Here we present a global deep-learning model for simultaneous earthquake detection and phase picking. Performing these two related tasks in tandem improves model performance in each individual task by combining information in phases and in the full waveform of earthquake signals by using a hierarchical attention mechanism. We show that our model outperforms previous deep-learning and traditional phase-picking and detection algorithms. Applying our model to 5 weeks of continuous data recorded during 2000 Tottori earthquakes in Japan, we were able to detect and locate two times more earthquakes using only a portion (less than 1/3) of seismic stations. Our model picks P and S phases with precision close to manual picks by human analysts; however, its high efficiency and higher sensitivity can result in detecting and characterizing more and smaller events.

Highlights

Earthquake signal detection and seismic phase picking are challenging tasks in the processing of noisy data and the monitoring of microearthquakes
Phase picking is the measurement of arrival times of distinct seismic phases (P-wave and S-wave phases) within an earthquake signal that are used to estimate the location of an earthquake
Both detection and picking can be viewed as identifying distinct variations in time-series data, phase picking is a local problem compared to detection, which uses a more global view of the full waveform and consists of information from multiple seismic phases including scattered waves

Summary

Introduction

Earthquake signal detection and seismic phase picking are challenging tasks in the processing of noisy data and the monitoring of microearthquakes. This is due to the extreme sensitivity of earthquake location estimates to earthquake arrival time measurements - 0.01 second of error in determining P-wave arrivals can translate to tens of meters of error in location Both detection and picking can be viewed as identifying distinct variations in time-series data, phase picking is a local problem compared to detection, which uses a more global view of the full waveform and consists of information from multiple seismic phases including scattered waves. Humans focus on a certain region of an image with high resolution while perceiving the surrounding image at low resolution and adjusting the focal point over time Our model emulates this through two levels of attention mechanism, one at the global level for identifying an earthquake signal in the input time series, and one at the local level for identifying different seismic phases within that earthquake signal. The events we detect are located to demonstrate the generalization of the model to other regions and its ability to improve earthquake source characterization

Objectives

Methods

Results

Conclusion