Evaluation of Machine Learning Assisted Earthquake Phase Detection Performance in Different Tectonic Regions and Environmental Noise on the Alaska Seismic Network
Many available machine learning (ML) phase-detection algorithms promise real-time monitoring applications and to be broadly transferable across different seismic regions when trained on a global dataset. One such algorithm, Earthquake Transformer (EQT), developed by Mousavi et al. 2020, is repeatedly used by the research community to generate localized earthquake catalogs in various tectonic regions. We systematically apply the global pre-trained EQT package to disparate seismogenic regions in Alaska as a first assessment of ML-assisted, phase-picking competency on the Alaska Seismic Network.
We select three regions with varying station azimuthal coverage, environmental noise, and seismicity characteristics to evaluate overall ML performance. We define the minimum performance goal as reproducing the Alaska Earthquake Center’s (AEC) existing real-time catalog. The three chosen regions are (1) Purcell Mountain in northwestern Alaska, (2) Andreanof and Fox Islands along the Aleutian Island chain, and (3) Yakutat and Icy Bay in southeastern Alaska. The Purcell Mountain region represents near-ideal ML conditions with moderate magnitude, shallow-crustal events recorded with good azimuthal station coverage and low environmental noise. The Aleutian and Yakutat/Icy Bay regions incorporate poor azimuthal station coverage and varying earthquake source characteristics in addition to volcanic and glacial signals, respectively. Application of EQT in each region produces an alternative earthquake catalog that is directly comparable to both the AEC’s augmented STA/LTA monitoring catalog and analyst-reviewed catalog. Leveraging all three catalogs provides insights into the ability of ML to accurately detect seismic arrivals in the wide range of conditions present in Alaska.
Session: Network Seismology: Recent Developments, Challenges and Lessons Learned
Type: Oral
Room: 202A/B
Date: 4/20/2023
Presentation Time: 10:00 AM (local time)
Presenting Author: Sarah Noel
Student Presenter: Yes
Additional Authors
Sarah Noel Presenting Author Corresponding Author sknoel@alaska.edu University of Alaska Fairbanks |
Michael West mewest@alaska.edu University of Alaska Fairbanks |
|
|
|
|
|
|
|
Evaluation of Machine Learning Assisted Earthquake Phase Detection Performance in Different Tectonic Regions and Environmental Noise on the Alaska Seismic Network
Category
Network Seismology: Recent Developments, Challenges and Lessons Learned
Description