← Back to Sessions

Machine Learning Earthquake Catalog Performance for Characteristic Alaska Settings

Description: 

Globally pre-trained machine learning (ML) earthquake phase-detection algorithms promise broadscale regional transferability in addition to real-time monitoring applications. These algorithms continue to gain research popularity and are routinely used to generate earthquake catalogs with thousands of additional events. We apply a globally pre-trained ML earthquake phase-detection algorithm to four different characteristic regions and sequences to generate a ML earthquake catalog. We compare the ML catalog with the Alaska Earthquake Center’s real time (RT) and analyst-reviewed (AR) catalogs. We establish a binary classification framework to compare the RT and ML catalogs to the AR catalog. We visually assess and label additional RT and ML events as earthquakes or other signals. Finally, we apply a minimum catalog inclusion criteria based on Alaska Earthquake Center analyst review standards to additional, visually-confirmed RT and ML catalog earthquakes, establishing a one-to-one performance comparison between catalogs. For each region, we find the ML catalog provides a consistently higher match with the AR catalog than the RT catalog. However, each region’s ML catalog introduces additional complications ranging from misidentification of non-earthquake signals to missed detection of large magnitude, felt earthquakes. These discrepancies warrant further training dataset scrutiny and suggest the establishment of a location based training dataset is necessary for consistent and reliable ML phase detection performance across the Alaska seismic network.

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: Sarah

Student Presenter: Yes

Invited Presentation: 

Authors