EasyQuake: Putting Machine Learning to Work for Your Regional Seismic Network or Local Earthquake Study

Abstract

AbstractWe developed a Python package—easyQuake—that consists of a flexible set of tools for detecting and locating earthquakes from International Federation of Digital Seismograph Networks-collected or field-collected seismograms. The package leverages a machine-learning driven phase picker, coupled with an associator, to produce a Quake Markup Language (QuakeML) style catalog complete with magnitudes and P-wave polarity determinations. We describe how nightly computations on day-long seismograms identify lower-magnitude candidate events that were otherwise missed due to cultural noise and how those events are incorporated into the Oklahoma Geological Survey statewide network upon analyst manual review. We discuss applications for the package, including earthquake detection for regional networks and microseismicity studies in arbitrary user-defined regions. Because the fundamentals of the package are scale invariant, it has wide application to seismological earthquake analysis from regional to local arrays and has great potential for identifying early aftershocks that are otherwise missed. The package is fast and reliable; the computations are relatively efficient across a range of hardware, and we have encountered very few (∼1%) false positive event detections for the Oklahoma case study. The utility and novelty of the package is the turnkey earthquake analysis with QuakeML file output, which can be dropped directly into existing real-time earthquake analysis systems. We have designed the functions to be quite modular so that a user could replace the provided picker or associator with one of their choosing. The Python package is open source and development continues.