← Back to Gallery

A Comparison of Machine Learning Methods of Association

The association of phase picks to form events is one of the most fundamental components of seismology. Any error in association due to the inclusion of errant phase picks will introduce errors in the later characterization of the event, such as its location, magnitude, and focal mechanism. Large and dense sensor networks, such as nodal arrays (and DAS), offer unique challenges in association due to vast number of observations, high likelihood of errant picks or multiple events occurring within the array at the same time. Also the large number of stations can greatly increase the time it takes to perform the association. For this reason, machine learning might provide a more optimal method of association. In this work, we examine how well machine learning methods (e.g., GaMMA, Phaselink, and GENIE) can incorporate dense nodal arrays into regional networks and how well they handle the density of stations. We test their capabilities on two nodal deployments, one within Rock Valley Nevada (80 Nodes), and the LArge-n Seismic Survey in Oklahoma (LASSO) dense nodal array (>1800 nodes). These two nodal arrays will test two particular issues that the machine learning algorithms might face: 1) The Rock Valley Array is a small array (<10 kms) within a large regional network and it will allow tests of how the algorithms are able to merge the two during association. 2) The Oklahoma LASSO array spans 10’s of kms and contains multiple earthquake sequences and it will allow tests of each algorithm’s ability to associate events that might be happening simultaneously. We then compare their associated bulletins to those obtained using Rapid Earthquake Association and Location algorithm (REAL), a more traditional method of association. We find that there are often vast differences between findings of certain methods. Some methods finding roughly half the events of others while others cannot integrate the nodal array with the regional network. We even find that certain methods break apart larger events into multiple events. Prepared by LLNL under Contract DE-AC52-07NA27344.

Session: Network Seismology: Recent Developments, Challenges and Lessons Learned - III

Type: Oral

Room: Tubughnenq’ 5

Date: 5/2/2024

Presentation Time: 08:30 AM (local time)

Presenting Author: Colin Pennington

Student Presenter: No

Additional Authors