Progress toward modernizing the Southern California Seismic Network (SCSN) earthquake catalog with cloud-native machine learning approaches

Clara Yoon, Ryan Tam, Gabrielle Tepp, Rayomand Bhadha, Zachary E. Ross, Ellen Yu, Weiqiang Zhu, Allen L. Husker, & Michael Black

Submitted September 11, 2022, SCEC Contribution #12183, 2022 SCEC Annual Meeting Poster #117

Earthquake monitoring systems used at regional seismic networks run software on local servers that process continuous seismic data in real-time and output an earthquake catalog. The Southern California Seismic Network (SCSN) has used the Earthworm-based AQMS software since 2001 for operational earthquake monitoring. AQMS uses a short-term-average/long-term-average (STA/LTA) algorithm for event detection and phase identification, and hypoinverse with a 1D velocity model for event location. Analysts review every detected event and may either accept the automatic solution (~10% of events) or adjust the picks as needed. Over the past several years, the SCSN catalog has been complete down to M1.8, though it includes many smaller events as low as ~M0.

The SCSN is developing new data processing systems that take advantage of modern technologies to improve earthquake product quality and timeliness. We are making progress on Quakes2AWS - a cloud-based, modular software system for real-time earthquake monitoring, designed to allow easier testing and integration of new scientific algorithms, including machine learning approaches. Quakes2AWS leverages Amazon Web Services (AWS) serverless technology for scalable data processing and infrastructure-as-code tools to easily set up the system. As the first step of Quakes2AWS, we replaced the AQMS automatic picker with two different deep-learning based picker models, GPD (Ross et al., 2018) and PhaseNet (Zhu and Beroza, 2019), while using AQMS for all other processing. Testing shows that these models automatically pick phases with a similar accuracy to analysts and perform well at detecting lower magnitude events, especially during active sequences, which would enable rapid availability of more complete catalogs. We recently progressed further toward an entirely machine learning approach for earthquake catalog generation by incorporating and testing the unsupervised Gaussian Mixture Model Association (GaMMA) algorithm (Zhu et al., 2022) for event association, event location, and magnitude estimation, which we hope will further improve performance during periods of high earthquake activity.

Key Words
earthquake monitoring, earthquake catalog, cloud computing, machine learning

Yoon, C., Tam, R., Tepp, G., Bhadha, R., Ross, Z. E., Yu, E., Zhu, W., Husker, A. L., & Black, M. (2022, 09). Progress toward modernizing the Southern California Seismic Network (SCSN) earthquake catalog with cloud-native machine learning approaches. Poster Presentation at 2022 SCEC Annual Meeting.

Related Projects & Working Groups
Computational Science (CS)