Improving Automatic Post-Processing at the Southern California Seismic Network With Machine Learning Algorithms
Description:
During routine seismic network operations, data is automatically analyzed in real-time to identify seismic events and provide initial locations and magnitudes. Additionally, subnet triggers are used to collect unassociated phase picks from the real-time system that may correspond to an undetected event. Automatic post-processing may be applied to small events (M<~3) to add and refine picks and improve the hypocenter and magnitude before analyst review. The Southern California Seismic Network currently employs the Earthworm/AQMS real-time system, which uses a short-term average/long-term average (STA/LTA) phase picker and includes a post-processing module (hypomag).
We aim to improve automatic event picks and hypocenters and to reduce analyst workload by incorporating machine learning algorithms into our existing systems and workflow. Recently, we integrated the deep-learning picker PhaseNet into our event post-processing. This new hypoPN module has produced ~2-3 times more automatic picks of slightly better average quality than the original hypomag, leading to more accurate hypocenters. However, more time and analysis are needed to determine whether this change has reduced analyst workload overall. Following the success of hypoPN, we’ve begun work on automating subnet trigger processing. Weekly trigger counts can reach a few hundred, with around 10% converted to seismic events. Parsing through these triggers takes time from other analyst work for only a small increase in cataloged events. With modern algorithms, automatic trigger processing should be viable. We plan to apply PhaseNet and the machine-learning event-associator GaMMA to find seismic events in the triggers, with the ultimate goal of eliminating the need for analysts to review triggers.
Session: Network Seismology: Recent Developments, Challenges and Lessons Learned [Poster Session]
Type: Poster
Date: 5/1/2024
Presentation Time: 08:00 AM (local time)
Presenting Author: Ellen
Student Presenter: No
Invited Presentation:
Authors
Gabrielle Tepp gtepp@caltech.edu California Institute of Technology |
Ellen Yu Presenting Author Corresponding Author eyu@caltech.edu California Institute of Technology |
Weiqiang Zhu zhuwq@berkeley.edu University of California, Berkeley |
Erika Jaski ejaski@caltech.edu California Institute of Technology |
Zackary Newman znewman@caltech.edu California Institute of Technology |
Nick Scheckel nick@caltech.edu California Institute of Technology |
|
|
|
Improving Automatic Post-Processing at the Southern California Seismic Network With Machine Learning Algorithms
Session
Network Seismology: Recent Developments, Challenges and Lessons Learned