Deep Learning Enhanced Earthquake Catalog for Northern California
Description:
Improving the completeness and accuracy of earthquake catalogs can reveal new insights into spatio-temporal trends in seismicity, identify fault structures in improved detail, and provide rich datasets for tomography. We present work on a comprehensive earthquake catalog for northern California between 2000 - 2023, spanning from Parkfield to the Mendocino Triple Junction, and from the west coast into the Sierras and western Nevada. To develop this catalog, we use PhaseNet-based picks (Zhu and Beroza, 2018) followed by GENIE-based association (McBrearty and Beroza, 2023). The original GENIE is a graph neural network associator that relies on all pairs of source and station nodes. We implement a more memory efficient version that uses only a subset of all source-station pairs. We also experiment with using amplitudes and phase types in addition to arrival times. These modifications improve the performance of the model and enable accurate associations across the full spatial domain, despite the heterogeneity in station coverage, source distribution, and time-varying seismic networks.
The combination of these two deep learning processing steps, followed by double-difference relocation (Waldhauser and Ellsworth, 2000), reveals dense seismicity throughout all of the expected seismogenic zones of northern California, while increasing the rates of detection compared to the USGS catalog by 4 – 8x depending on the detection thresholds used. We observe significant earthquake activity at Geysers geothermal field, along the Calaveras and Hayward faults, and at Long Valley Caldera and the Mendocino Triple Junction. The highest event activities occur during the 2003 San Simeon, 2004 Parkfield, and the 2014 Napa Valley earthquake sequences. We assess the reliability of the results based on known seismicity, expected earthquakes statistics, and association assignments across the seismic network with respect to source locations and event magnitudes.
Session: Network Seismology: Recent Developments, Challenges and Lessons Learned - III
Type: Oral
Date: 5/2/2024
Presentation Time: 09:00 AM (local time)
Presenting Author: Ian
Student Presenter: Yes
Invited Presentation:
Authors
Ian McBrearty Presenting Author Corresponding Author imcbrear@stanford.edu Stanford University |
Gregory Beroza beroza@stanford.edu Stanford University |
|
|
|
|
|
|
|
Deep Learning Enhanced Earthquake Catalog for Northern California
Category
Network Seismology: Recent Developments, Challenges and Lessons Learned