← Back to Gallery

Testing Machine Learning Phase Pickers to Develop a High-Resolution Earthquake Catalog With a 398-Instrument Nodal Array on Kodiak Island, Alaska

Portable seismic node instruments enable spatially dense seismic data acquisition for high-resolution imaging and event detection in diverse field areas. The Kodiak node array, deployed during the Alaska Amphibious Community Seismic Experiment (AACSE), consisted of 398 nodal seismometers deployed along roads at ∼200–300 m spacing on northeastern Kodiak Island for four weeks in May-June 2019. The network overlies the southern asperity of the 1964 Mw9.2 Great Alaska earthquake and offers a high-resolution snapshot of thrust zone microseismicity several decades after a large megathrust earthquake. Compared with collocated AACSE broadband stations, the node array is far more numerous and closely spaced, providing an opportunity to detect small events which are critical for understanding fault zone structure and earthquake interactions.

Here we create an earthquake catalog using a machine-learning (ML) workflow. To test existing ML picking models on the node data, we first create a node waveform dataset by manually picking 2,467 P and 2,392 S from 23 small and local events using event times from the AACSE catalog. We then use the dataset to test performance on node waveform of 29 pretrained ML picking models available from Seisbench. The picking models have variable performance on the node waveform dataset. Overall, PhaseNet pretrained with INSTANCE, EQTransformer pretrained with INSTANCE, and EQTransformer retrained with local AACSE data resulted in the highest recalls, smallest median residuals, and reasonable number of picks. We are using these three models as starting models for transfer learning, and will apply the re-trained models to continuous node data to detect more regional and low-magnitude earthquakes compared with the AACSE catalog.

Session: Structure and Behavior of the Alaska-Aleutian Subduction Zone [Poster Session]

Type: Poster

Room: Exhibit Hall

Date: 5/3/2024

Presentation Time: 08:00 AM (local time)

Presenting Author: Hanqi Zhu

Student Presenter: Yes

Additional Authors