Scholarlink

<p>Earthquake Early Warning (EEW) systems often have difficulty estimating the source location in the first seconds after detection, especially for events in islands and peninsulas where station coverage is sparse and azimuthal geometry is biased. To address this limitation, we develop a prior-informed integrated particle filter (πIPF) method by enhancing the extended integrated particle filter (IPFx) source estimation framework with earthquake occurrence probabilities derived from the Hierarchical Spatiotemporal Epidemic-Type Aftershock Sequence (HIST-ETAS) model. The method incorporates HIST-ETAS information in two ways: (1) ETAS sampling, which uses ETAS-derived spatial weights to guide the initial particle distribution (prior), and (2) an ETAS-based likelihood term with an adaptive weight that decreases as more stations trigger and azimuthal coverage improves. We evaluate performance on the June 3, 2024 Noto Peninsula aftershock (M 6.0) as an example of unstable initial estimation. ETAS sampling narrows the initial spread and modestly improves early location estimates, but its benefit is confined to the first few seconds as sequential reweighting and resampling quickly align particles with the data-driven likelihood. In contrast, the ETAS-based likelihood maintains average horizontal location error below 10&#xa0;km throughout the time window and keeps final errors below 2&#xa0;km. The ETAS term in the likelihood provides stabilizing constraints when picks are few and coverage is poor; then it is gradually decreased to zero once sufficient observations accumulate, leaving the solution determined solely by waveform data. To test robustness, we apply the approach to 129 earthquakes with maximum seismic intensity ≥ 4 that occurred in 2020. Using HIST-ETAS information computed up to the day before each event, both ETAS sampling and ETAS likelihood reduce the 95th-percentile location error by &gt; 30% and the 95th-percentile magnitude error by ~ 10% relative to the original IPFx method. These results demonstrate that incorporating information on past seismicity into real-time inference improves the stability and accuracy of EEW source estimation, with particular benefits in poorly instrumented or azimuthally unbalanced settings common to islands and peninsulas, while avoiding lasting bias once data quality increases.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A prior-informed integrated particle filter (πIPF) method for earthquake early warning using ETAS model

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月21日

查看全文
  • Masumi Yamada,
  • Stephen Wu,
  • Keisuke Yano,
  • Koji Tamaribuchi,
  • Keishi Noguchi,
  • Naoki Hayashimoto,
  • Hiroshi Tsuruoka

摘要

Earthquake Early Warning (EEW) systems often have difficulty estimating the source location in the first seconds after detection, especially for events in islands and peninsulas where station coverage is sparse and azimuthal geometry is biased. To address this limitation, we develop a prior-informed integrated particle filter (πIPF) method by enhancing the extended integrated particle filter (IPFx) source estimation framework with earthquake occurrence probabilities derived from the Hierarchical Spatiotemporal Epidemic-Type Aftershock Sequence (HIST-ETAS) model. The method incorporates HIST-ETAS information in two ways: (1) ETAS sampling, which uses ETAS-derived spatial weights to guide the initial particle distribution (prior), and (2) an ETAS-based likelihood term with an adaptive weight that decreases as more stations trigger and azimuthal coverage improves. We evaluate performance on the June 3, 2024 Noto Peninsula aftershock (M 6.0) as an example of unstable initial estimation. ETAS sampling narrows the initial spread and modestly improves early location estimates, but its benefit is confined to the first few seconds as sequential reweighting and resampling quickly align particles with the data-driven likelihood. In contrast, the ETAS-based likelihood maintains average horizontal location error below 10 km throughout the time window and keeps final errors below 2 km. The ETAS term in the likelihood provides stabilizing constraints when picks are few and coverage is poor; then it is gradually decreased to zero once sufficient observations accumulate, leaving the solution determined solely by waveform data. To test robustness, we apply the approach to 129 earthquakes with maximum seismic intensity ≥ 4 that occurred in 2020. Using HIST-ETAS information computed up to the day before each event, both ETAS sampling and ETAS likelihood reduce the 95th-percentile location error by > 30% and the 95th-percentile magnitude error by ~ 10% relative to the original IPFx method. These results demonstrate that incorporating information on past seismicity into real-time inference improves the stability and accuracy of EEW source estimation, with particular benefits in poorly instrumented or azimuthally unbalanced settings common to islands and peninsulas, while avoiding lasting bias once data quality increases.

Graphical Abstract