Automatic Phase Picking Model for Ocean Bottom Seismic Data: Phasenet Model Trained Using Japanese S-net Data
Description:
With the recent enhancement of seismic networks, a vast amount of seismic waveform data has been obtained. Advanced information technologies, such as deep learning, are being utilized for processing this data, with seismic phase picking being a prime example. This has significantly simplified the process of deriving earthquake catalogs from seismic waveform data. Numerous automatic picking models, including PhaseNet (Zhu and Beroza, 2019) and EQTransformer (Mousavi et al., 2020), have been proposed. The performance of machine learning models depends not only on the model architecture but also on the training data. Naoi et al. (2024) successfully improved the performance of the PhaseNet model by training it with data from Japan's permanent seismic network. In this study, I focus on the application to ocean bottom seismic data and conducted transfer learning using PhaseNet from Naoi et al. (2024) with data from Japan's ocean bottom cable seismic observation network, S-net (NIED, 2019). During this process, I corrected the axis orientation of the data based on components related to gravity recorded in S-net's acceleration data, ensuring one of the three components was aligned vertically. We also compared the performance with OBSTransformer (Niksejel and Zhang, 2024) and PickBlue (Bornstein et al., 2024), both trained with ocean bottom seismic data. Currently, our model shows higher accuracy measured based on arrival time differences between the manual and automatic pick, while OBSTransformer and PickBlue show higher detection rates.
Acknowledgements: I used seismic data from S-net maintained by NIED (NIED, 2019), and SeisBench, a platform for machine learning in seismology (Woollam et al., 2022). This study was supported by MEXT Project for Seismology toward Research Innovation with Data of Earthquake (STAR-E) Grant Number JPJ010217.
Session: Building and Decoding High-resolution Earthquake Catalogs With Statistical and Machine-learning Tools [Poster]
Type: Poster
Date: 4/15/2025
Presentation Time: 08:00 AM (local time)
Presenting Author: Takahiko
Student Presenter: No
Invited Presentation:
Poster Number: 40
Authors
Takahiko Uchide Presenting Author Corresponding Author t.uchide@aist.go.jp National Institute of Advanced Industrial Science and Technology |
|
|
|
|
|
|
|
|
Automatic Phase Picking Model for Ocean Bottom Seismic Data: Phasenet Model Trained Using Japanese S-net Data
Category
Building and Decoding High-resolution Earthquake Catalogs With Statistical and Machine-learning Tools