AI-PAL: Self-Supervised AI Phase Picking via Rule-Based Algorithm for Generalized Earthquake Detection

Copyright and License

© 2025 The Author(s).

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.

Acknowledgement

We thank the JGR editor Rachel Abercrombie, the associate editor, and two anonymous reviewers for their constructive comments on this work. We thank Prof. Ziyadin Cakir for sharing the rupture model and afterslip model for the 2020 Elaziğ (Turkey) earthquake. Additionally, Yijian Zhou would like to thank Dr. Yun Wang from Carnegie Mellon University (now at Facebook) for inspiring the application of RNN sequence labeling in the phase picking task, which should have been acknowledged in Zhou et al. (2019). This work is jointly supported by the National Key R&D Program (Grant 2022YFF0800601) and the University of California, Riverside Senate CoR grant.

Data Availability

Figures in this paper are plotted with GMT and Matplotlib. The continuous seismic data for Ridgecrest-Coso (2008–2019), the waveform-relocated SCSN catalog, the QTM catalog, and the refined focal mechanism solutions by Cheng et al. (2023) are downloaded from the SCSN (SCEDC, 2013). The continuous data for the East Anatolian Fault Zone (EAFZ, 2020–2023/04) is collected through multiple sources: (a) TU network, from the Disaster and Emergency Management Presidency (AFAD, Last accessed 2024/11), downloaded manually, (b) KO network, from Kandilli Observatory And Earthquake Research Institute (KOERI), downloaded with Obspy, and (c) GE, CQ, and IM network, all available through Obspy. The relocated AFAD catalog by Lomax (2023) is available at Zenodo. The focal mechanism solutions for Mw > 4 events (2020–2023) are also available at AFAD. The active faults data of the Ridgecrest-Coso region is available at United States Geological Survey (USGS, Last accessed 2024/11); that for East Anatolian Fault Zone comes from the GEM Global Active Faults Database (Styron & Pagani, 2020), and we adopt a more detailed fault data for the EPF area from the Active Faults of Eurasia Database (AFEAD, Zelenin et al., 2022). The surface rupture data of the 2019 Ridgecrest earthquake comes from Ponti et al. (2020); the fault traces of the 2023 Turkey sequence is available by USGS (2024). The coseismic slip model of the 2019 Ridgecrest earthquake comes from Yue et al. (2021); that for the 2023 Turkey earthquake comes from Ren et al. (2024); the rupture model and afterslip model for the 2020 Elaziğ (Turkey) earthquake comes from Cakir et al. (2023). The referred velocity models for the RC-Coso region include: CVM-S4 velocity model (Kohler et al., 2003), Hutton et al. (2010) for the whole Southern California, and Shelly (2020) for the Ridgecrest source region. The referred velocity models for the EAFZ include that from Güvercin et al. (2022), Acarel et al. (2019), and Ding et al. (2023). The AI-PAL workflow is open sourced at Github (Zhou, 2024).

Supplemental Material

Supporting Information S1

Data Set S1

Data Set S2

Data Set S3

Data Set S4

Errata

The originally published version of this article contained typographical errors. Throughout the article, the section headings were misnumbered. The errors have been corrected, and this may be considered the authoritative version of record.