Pith. sign in

REVIEW 3 major objections 6 minor 43 references

VORA turns earthquake phase association into unsupervised clustering of origin times on a Voronoi station graph, and the same code runs from local arrays to a global pick set at record speed.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-14 11:36 UTC pith:N3BVASNL

load-bearing objection Solid, practical associator that is genuinely fast and scale-agnostic; the dual-phase premise is a real limit but the paper owns it and the rest of the evidence holds. the 3 major comments →

arxiv 2607.10450 v1 pith:N3BVASNL submitted 2026-07-11 physics.geo-ph

VORA: Rapid Association of Earthquake Phases from Local to Global

Junhao Song , Weiqiang Zhu , Haoyu Wang , Utpal Kumar , Taka'aki Taira , Richard M Allen This is my paper
classification physics.geo-ph PACS 91.30.Ab91.30.Dk93.85.Bc
keywords earthquake phase associationVoronoi tessellationorigin-time clusteringunsupervised clusteringseismic network geometryRidgecrest sequenceglobal-scale seismicityDBSCAN
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Modern seismic networks and deep-learning pickers flood analysts with millions of P and S arrivals that must be grouped into common earthquake origins. VORA estimates a candidate origin time from each P–S pair (or from a waveform model) using only a narrow range of velocity ratios, then asks which of those times come from neighboring stations. Neighbors are defined by the Voronoi diagram of the operating stations, so the spatial graph automatically adapts when stations are added or removed and works unchanged from a dense local array to a global network. A modified DBSCAN clusters the origin-time intervals across those neighbors; an optional EM step splits merged clusters when two events fall inside the same cells. On synthetic catalogs with average inter-event times as short as 5 s, and on real Ridgecrest picks at both local and global scales, VORA is the fastest of the three methods tested and keeps high recall and precision with a single hyper-parameter set. The result is a training-free associator that can keep pace with the growing volume of automated picks.

Core claim

The paper claims that earthquake phase association reduces to unsupervised spatio-temporal clustering of hypothetical origin times once station neighborhoods are defined by Voronoi tessellation, and that this formulation is both faster and more scale-invariant than existing grid-search or trained methods.

What carries the argument

Voronoi station graph plus origin-time intervals: each station owns a Voronoi cell; stations that share an edge are neighbors; candidate origin times (Eq. 1 or waveform prediction) are clustered only across those neighbors, turning association into a single unsupervised pass.

Load-bearing premise

Enough stations must report both a usable P and an S pick (or a direct origin-time estimate) so that neighboring origin-time intervals can form a cluster; missing S phases cause real events, including the Ridgecrest mainshock, to be lost.

What would settle it

Run the identical VORA configuration on a dense aftershock sequence whose S-phase detection rate is deliberately degraded below the level of the Ridgecrest mainshock window; if recall collapses while competing methods that use P-only or multi-phase logic remain stable, the central claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The manuscript introduces VORA, an unsupervised earthquake phase associator that estimates hypothetical origin times from P–S pairs (Eq. 1, or direct waveform prediction) and clusters them over Voronoi-defined station neighborhoods, with an optional EM sub-clustering step (Eq. 2) for events that merge in space–time. The central claims are that VORA is training-free, governed by few physically interpretable hyperparameters, achieves the fastest runtime among the three methods tested, maintains high recall and precision under dense seismicity, and uses the same framework from a local 121-station network to a 4461-station global pick set covering the 2019 Ridgecrest sequence. Support comes from eight synthetic sequences (5000 events each, average inter-event times from 64 s to 5 s), a local Ridgecrest PhaseNet pick set compared with GaMMA and PyOcto, and a two-week global pick subset with USGS/SCSN catalog matching.

Significance. If the performance and scalability claims hold under realistic pick incompleteness, VORA would be a practically important contribution: a fast, training-free associator whose station graph adapts to non-uniform and evolving networks without region-specific retraining or absolute velocity models. The Voronoi neighborhood construction is a clear methodological strength relative to fixed-K or pure distance graphs, and the demonstrated local-to-global reuse of the same hyperparameter set is rare among published associators. The runtime advantage on a single CPU (Fig. 5) is valuable for large automated pick streams. These strengths are real even if the intense-seismicity numbers need more careful qualification.

major comments (3)
  1. [§3.1 Synthetic Stress Test; Fig. 5] §3.1 and Fig. 5: The synthetic stress test that underpins the strongest performance claim (recall >0.8 and precision >0.9 at Δt̄ = 5 s, plus the associated runtime ranking) explicitly assumes that every station always records both P and S. The manuscript states this assumption directly and does not inject controlled missing-S or false-pick rates. Because Eq. 1 and the dual-phase seed construction require usable P–S pairs, those headline numbers have not been demonstrated under the incompleteness that the Discussion and Fig. 9 show is decisive on real data (including the missed Ridgecrest mainshock). A controlled missing-S / missing-P sweep on the same synthetic suite, or a clear separation of claims between dual-phase-complete and realistic pick streams, is needed before the intense-seismicity claim can stand as written.
  2. [Abstract; §3.2; §3.3; Fig. 8–9] §3.2–3.3 and Abstract: Real-data recall (0.86 local Ridgecrest; daily global recalls in Fig. 8) is measured by matching associated events to SCSN/USGS catalogs, not by pick-level association accuracy under known ground-truth associations with realistic dual-phase incompleteness. Catalog matching is standard but does not close the gap left by the idealized synthetics: events that never form dual-phase seeds cannot be recovered by the optional EM step. The Abstract’s phrasing that VORA “maintains robust performance (high recall and precision) even under intense seismicity” currently blends the dual-phase-complete synthetic result with real-data catalog recall and should be tightened accordingly.
  3. [§2.4 Optional EM-Based Clustering; Eq. (2); Fig. 3–4] §2.4, Eq. (2): The optional EM sub-clustering relies on a constant pseudo-slowness K per event and seed stations identified as local minima of S–P intervals. The paper notes that quality “depends heavily on how precisely the seed stations are identified” and that the approach works better for denser networks, but provides no quantitative success rate for the sub-clustering step across the synthetic suite (e.g., fraction of merged clusters correctly split, sensitivity to seed errors). Because this step is the only remedy offered for the hardest overlapping cases that drive the Δt̄ = 5 s regime, its failure modes should be reported with the same recall/precision protocol used for the main clustering pass.
minor comments (6)
  1. [§2.1–2.3] Clarify in §2.1–2.3 whether origin-time intervals from a VP/VS range are treated as closed intervals for overlap, and how multi-ratio candidates from the same station are deduplicated before DBSCAN-style growth.
  2. [Fig. 5] Fig. 5d: state explicitly whether runtime includes only association or also preliminary location, and whether GaMMA/PyOcto were run with their recommended parallel settings disabled solely for fairness.
  3. [§3.3] §3.3: the 20 s maximum P–S separation (local-only detection) is important; state it earlier when claiming “global-scale” association so readers do not infer teleseismic association capability.
  4. [Fig. 6; §3.2] Fig. 6: VORA’s systematically larger epicentral residuals are attributed to requiring both P and S; a short quantitative note on average station count per event versus GaMMA/PyOcto would make that trade-off clearer.
  5. [Introduction; References] Minor wording: “archiving comparable” → “achieving comparable” (Introduction); ensure consistent spelling of Munchmeyer/Münchmeyer across text and references.
  6. [Open Research Section] Open Research: code is promised after peer review; a frozen commit or archive at revision would strengthen reproducibility claims for a methods paper.

Circularity Check

0 steps flagged

No circularity: association is unsupervised clustering of origin-time intervals from measured picks; performance is scored against independent synthetic ground truth and external catalogs/methods.

full rationale

VORA’s derivation chain is self-contained and non-circular. Hypothetical origin times come from measured P/S arrivals and a prescribed VP/VS range via the standard Wadati-style formula (Eq. 1), or from external multi-task pickers; those intervals are then clustered with a Voronoi neighbor graph and optional EM sub-clustering. The clusters are the associations by construction of unsupervised clustering, not a claimed first-principles prediction that reduces to a fitted free parameter. Recall, precision, and runtime are measured against synthetic ground-truth catalogs (Hungarian matching) and independent reference catalogs (SCSN, USGS), and are compared to two external methods (GaMMA, PyOcto). Self-citations (e.g., Zhu et al. 2022 for GaMMA/picks, Zhu et al. 2025 for origin-time prediction) supply comparison baselines or related tools; none is a load-bearing uniqueness theorem or ansatz that forces the reported metrics. The dual-phase completeness assumption in the synthetic suite is a validity/generalization concern, not a circular reduction of outputs to inputs. No equation or claim reduces the central performance results to a fit or self-definition by construction.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central performance claims rest on a small set of physically motivated but free hyper-parameters, standard seismological assumptions about phase arrivals and velocity ratios, and the geometric definition of Voronoi neighbors. No new physical entities are postulated.

free parameters (4)
  • minimum number of stations per event = 3
    User-chosen threshold (set to 3 in all experiments) that directly controls which clusters are retained; changes recall/precision trade-off.
  • VP/VS ratio range = [1.63, 1.83]
    Prescribed interval used to turn each P–S pair into an origin-time interval (Eq. 1); chosen by hand as [1.63, 1.83] and kept fixed across all scales.
  • highest neighbor order = 2
    Controls spatial extent of clustering (1st- and 2nd-order used); selected by the authors rather than derived.
  • maximum P–S separation for pairing = 20 s
    20 s cutoff used in the global experiment; limits detection range and is not derived from first principles.
axioms (4)
  • domain assumption Both P and S phases (or a direct origin-time prediction) can be obtained at stations surrounding an event with sufficient reliability for Eq. 1 to produce coherent origin-time intervals.
    Stated in §2.1 and acknowledged as a limitation when S picks are missing (Discussion, Fig. 9).
  • domain assumption The VP/VS ratio varies less regionally and with depth than absolute velocities, so a narrow fixed range yields usable origin-time estimates.
    Invoked to justify Eq. 1 and the claim of velocity-model independence (§2.1).
  • domain assumption Stations whose Voronoi cells share an edge or vertex are the spatially relevant neighbors for a common source.
    Core geometric premise of §2.2; standard Voronoi property applied to seismic stations.
  • ad hoc to paper A constant or slowly varying pseudo-slowness K is adequate for the optional EM sub-clustering step (Eq. 2).
    Introduced in §2.4; authors note it is ‘not perfect but proven effective’.

pith-pipeline@v1.1.0-grok45 · 20077 in / 2725 out tokens · 41205 ms · 2026-07-14T11:36:45.715340+00:00 · methodology

0 comments
read the original abstract

Earthquake phase association, which groups seismic phase arrivals into common origins, is a key step towards more complete and reliable seismicity catalogs. It has become a challenging task because of the massive phase datasets produced from dense seismic networks and advanced phase picking methods. Here we present VORA (Voronoi tessellation- and Origin-time-based Rapid Associator), an efficient and scalable earthquake phase associator that treats association as an unsupervised spatio-temporal clustering problem. Specifically, VORA depends on two primary constraints: the estimated earthquake origin time (temporal) and seismic station adjacency (spatial). Recent advances in deep learning models have enabled detection of S- and P-phases with similar effectiveness, making it straightforward to estimate corresponding earthquake origin times from candidate phase pairs given a prescribed range of velocity ratios or directly from raw waveforms. We then leverage the Voronoi diagram to define each station's neighbors, cluster the origin times across neighboring stations, and apply an optional sub-clustering step to separate overlapping events. Benchmarks on synthetic and real datasets show that VORA achieves the fastest runtime and maintains robust performance (high recall and precision) even under intense seismicity. Applying it to a two-week global-scale pick dataset covering the 2019 Ridgecrest earthquake sequence further demonstrates that the same framework scales from local networks to a global station set. VORA requires no training and generalizes across local-to-global regions, varying velocity models, and evolving network geometries, helping to meet the growing demands of expanding seismic networks and the increasing volume of automated phase picks.

Figures

Figures reproduced from arXiv: 2607.10450 by Haoyu Wang, Junhao Song, Richard M Allen, Taka'aki Taira, Utpal Kumar, Weiqiang Zhu.

Figure 1
Figure 1. Figure 1: Schematic view of spatio-temporal clustering of earthquake hypothetical origin times. The left plot shows the Voronoi diagram built using the synthetic stations (black trian￾gles). The origins (time and location) of three synthetic earthquakes are denoted by crosses. Phase arrivals and the estimated earthquake origin times at the nearby stations are shown in the right plot. The hypothetical origin times fo… view at source ↗
Figure 2
Figure 2. Figure 2: Multi-scale views of an example Voronoi diagram generated from the locations of globally distributed seismic stations. (a) Global scale, showing a hemisphere centered on North America. Triangles mark the locations of seismic stations, and green lines denote the Voronoi edges that compose Voronoi cell boundaries. (b) A zoomed-in regional view of California. (c) A further zoomed-in local view focusing on The… view at source ↗
Figure 3
Figure 3. Figure 3: An example illustrating the optional sub-clustering step for two prescribed earth￾quakes that happen at the same time, and also close in space based on the EM algorithm. (a) Map showing the locations of two earthquakes (stars) and the same seismic stations (triangles) shown in [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Synthetic test input and output with a 5-second average time separation between earthquakes (∆¯t). (a) Map showing the distribution of the 5000 synthetic earthquakes (colored dots) and the same stations shown in [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Stress testing VORA and two other existing methods (GaMMA & PyOcto) on the 8 synthetic datasets. (a) The changes of recall rate with ∆¯t (average time separation between earthquakes). (b) The changes of precision rate with ∆¯t. (c) The changes of total run time with ∆¯t used by each method with single CPU to complete the association. (d) Association speed comparison across three methods (all with single CP… view at source ↗
Figure 6
Figure 6. Figure 6: Association results of VORA and two other existing methods (GaMMA & PyOcto) on the ML-based phase picks dataset from Zhu et al. (2022) for the 2019 Ridgecrest earthquake sequence. (a) The spatial distribution of associated earthquakes by VORA, and the histograms of origin time difference and epicentral distance compared with the common events from the SCSN catalog. (b) Similar to (a), GaMMA results based o… view at source ↗
Figure 7
Figure 7. Figure 7: Association results of VORA on the two-week global-scale ML-based phase picks dataset (Ni et al., 2025) (a) Global distribution of seismic stations, color-coded by the number of picks. Stations without any picks generally lack available waveforms (Ni et al., 2025) and are therefore excluded from the association process. (b) Global distribution of associated and pre￾liminarily located earthquakes, color-cod… view at source ↗
Figure 8
Figure 8. Figure 8: Daily variations in VORA recall rates across different regions and pick datasets. Red and blue dashed lines represent recall rates using PhaseNet picks from Ni et al. (2025) for the Ridgecrest and US West Coast (excluding Ridgecrest) regions, respectively. The solid red line denotes the recall rate obtained using PhaseNet picks from Zhu et al. (2022) for the Ridgecrest region. –20– [PITH_FULL_IMAGE:figure… view at source ↗
Figure 9
Figure 9. Figure 9: Example input and output of VORA surrounding the origin of the 2019 Ridgecrest mainshock. (a) All available PhaseNet phase picks (Ni et al., 2025) from the stations within the geographic bounds [-130°W, -110°W, 32°N, 43°N], spanning a 1-hour window centered on the mainshock time. (b) VORA association results, represented by clustered hypothetical origin times similar to [PITH_FULL_IMAGE:figures/full_fig_p… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

43 extracted references · 2 linked inside Pith

  1. [1]

    Bulletin of the Seismological Society of America , volume=

    Waveform relocated earthquake catalog for southern California (1981 to June 2011) , author=. Bulletin of the Seismological Society of America , volume=. 2012 , publisher=

    1981

  2. [2]

    Science , volume=

    Hierarchical interlocked orthogonal faulting in the 2019 Ridgecrest earthquake sequence , author=. Science , volume=. 2019 , publisher=

    2019

  3. [3]

    The Seismic Record , volume=

    Machine-learning-based high-resolution earthquake catalog reveals how complex fault structures were activated during the 2016--2017 central Italy sequence , author=. The Seismic Record , volume=. 2021 , publisher=

    2016

  4. [4]

    Geophysical Research Letters , volume=

    Fracture-mesh faulting in the swarm-like 2020 Maacama sequence revealed by high-precision earthquake detection, location, and focal mechanisms , author=. Geophysical Research Letters , volume=. 2023 , publisher=

    2020

  5. [5]

    Reviews of Geophysics , volume=

    Achievements and prospects of global broadband seismographic networks after 30 years of continuous geophysical observations , author=. Reviews of Geophysics , volume=. 2022 , publisher=

    2022

  6. [6]

    Seismological Research Letters , volume=

    A plan for a long-term, automated, broadband seismic monitoring network on the global seafloor , author=. Seismological Research Letters , volume=. 2020 , publisher=

    2020

  7. [7]

    Bulletin of the seismological society of America , volume=

    A double-difference earthquake location algorithm: Method and application to the northern Hayward fault, California , author=. Bulletin of the seismological society of America , volume=. 2000 , publisher=

    2000

  8. [8]

    Journal of Geophysical Research: Solid Earth , volume=

    Large-scale relocation of two decades of Northern California seismicity using cross-correlation and double-difference methods , author=. Journal of Geophysical Research: Solid Earth , volume=. 2008 , publisher=

    2008

  9. [9]

    Seismological Research Letters , volume=

    Distributed acoustic sensing turns fiber-optic cables into sensitive seismic antennas , author=. Seismological Research Letters , volume=. 2020 , publisher=

    2020

  10. [10]

    Bulletin of the seismological society of America , volume=

    Automatic earthquake recognition and timing from single traces , author=. Bulletin of the seismological society of America , volume=. 1978 , publisher=

    1978

  11. [11]

    Geophysical Journal International , volume=

    PhaseNet: a deep-neural-network-based seismic arrival-time picking method , author=. Geophysical Journal International , volume=. 2019 , publisher=

    2019

  12. [12]

    Nature communications , volume=

    Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking , author=. Nature communications , volume=. 2020 , publisher=

    2020

  13. [13]

    Bulletin of the Seismological Society of America , volume=

    Generalized seismic phase detection with deep learning , author=. Bulletin of the Seismological Society of America , volume=. 2018 , publisher=

    2018

  14. [14]

    Seismological Research Letters , volume=

    Rapid earthquake association and location , author=. Seismological Research Letters , volume=. 2019 , publisher=

    2019

  15. [15]

    arXiv preprint arXiv:2310.11157 , year=

    PyOcto: A high-throughput seismic phase associator , author=. arXiv preprint arXiv:2310.11157 , year=

    Pith/arXiv arXiv

  16. [16]

    Journal of Geophysical Research: Solid Earth , volume=

    PhaseLink: A deep learning approach to seismic phase association , author=. Journal of Geophysical Research: Solid Earth , volume=. 2019 , publisher=

    2019

  17. [17]

    Bulletin of the Seismological Society of America , volume=

    Earthquake phase association with graph neural networks , author=. Bulletin of the Seismological Society of America , volume=. 2023 , publisher=

    2023

  18. [18]

    Journal of Geophysical Research: Solid Earth , volume=

    Earthquake phase association using a Bayesian Gaussian mixture model , author=. Journal of Geophysical Research: Solid Earth , volume=. 2022 , publisher=

    2022

  19. [19]

    Journal of the Meteorological Society of Japan

    On the travel time of earthquake waves.(Part II) , author=. Journal of the Meteorological Society of Japan. Ser. II , volume=. 1933 , publisher=

    1933

  20. [20]

    Geophysical Research Letters , volume=

    Foreshock migration preceding the 2025 Mw 8.8 Kamchatka earthquake: Insights from single-station observations , author=. Geophysical Research Letters , volume=. 2026 , publisher=

    2025

  21. [21]

    arXiv preprint arXiv:2506.06939 , year=

    Towards end-to-end earthquake monitoring using a multitask deep learning model , author=. arXiv preprint arXiv:2506.06939 , year=

    Pith/arXiv arXiv

  22. [22]

    Pure and Applied Geophysics , volume=

    HYPOSAT--An enhanced routine to locate seismic events , author=. Pure and Applied Geophysics , volume=. 2001 , publisher=

    2001

  23. [23]

    Seismological Research Letters , volume=

    PhasePApy: A robust pure Python package for automatic identification of seismic phases , author=. Seismological Research Letters , volume=. 2016 , publisher=

    2016

  24. [24]

    Seismological Society of America , volume=

    An earthquake detection and location architecture for continuous seismograms: Phase picking, association, location, and matched filter (PALM) , author=. Seismological Society of America , volume=

  25. [25]

    Nouvelles applications des param

    Voronoi, Georges , journal=. Nouvelles applications des param. 1908 , publisher=

    1908

  26. [26]

    Seismological Research Letters , volume=

    Parsimonious seismic tomography with Poisson Voronoi projections: methodology and validation , author=. Seismological Research Letters , volume=. 2020 , publisher=

    2020

  27. [27]

    Journal of Geophysical Research: Solid Earth , volume=

    Adaptive coda-wave imaging with voronoi tessellation , author=. Journal of Geophysical Research: Solid Earth , volume=. 2023 , publisher=

    2023

  28. [28]

    Seismica , volume=

    A Global-scale Database of Seismic Phases from Cloud-based Picking at Petabyte Scale , author=. Seismica , volume=

  29. [29]

    Seismological Research Letters , volume=

    A high-resolution seismic catalog for the initial 2019 Ridgecrest earthquake sequence: Foreshocks, aftershocks, and faulting complexity , author=. Seismological Research Letters , volume=. 2020 , publisher=

    2019

  30. [30]

    Geophysical Research Letters , volume=

    Rapid characterization of the July 2019 Ridgecrest, California, earthquake sequence from raw seismic data using machine-learning phase picker , author=. Geophysical Research Letters , volume=. 2020 , publisher=

    2019

  31. [31]

    Geophysical Journal International , volume=

    Deep learning for deep earthquakes: insights from OBS observations of the Tonga subduction zone , author=. Geophysical Journal International , volume=. 2024 , publisher=

    2024

  32. [32]

    Seismological Society of America , volume=

    Hex: Hyperbolic event extractor, a seismic phase associator for highly active seismic regions , author=. Seismological Society of America , volume=

  33. [33]

    IEEE Transactions on Geoscience and Remote Sensing , volume=

    Better together: Ensemble learning for earthquake detection and phase picking , author=. IEEE Transactions on Geoscience and Remote Sensing , volume=. 2023 , publisher=

    2023

  34. [34]

    Seismica , volume=

    Picking regional seismic phase arrival times with deep learning , author=. Seismica , volume=

  35. [35]

    ACM transactions on database systems (tods) , volume=

    DBSCAN revisited, revisited: why and how you should (still) use DBSCAN , author=. ACM transactions on database systems (tods) , volume=. 2017 , publisher=

    2017

  36. [36]

    IEEE Transactions on Aerospace and Electronic Systems , volume=

    On implementing 2D rectangular assignment algorithms , author=. IEEE Transactions on Aerospace and Electronic Systems , volume=. 2016 , publisher=

    2016

  37. [37]

    Journal of Geophysical Research: Solid Earth , volume=

    Which picker fits my data? A quantitative evaluation of deep learning based seismic pickers , author=. Journal of Geophysical Research: Solid Earth , volume=. 2022 , publisher=

    2022

  38. [38]

    Bulletin of the Seismological Society of America , volume=

    Double-difference tomography: The method and its application to the Hayward fault, California , author=. Bulletin of the Seismological Society of America , volume=. 2003 , publisher=

    2003

  39. [39]

    Earth and Planetary Science Letters , volume=

    Measuring earthquake source parameters in the Mendocino triple junction region using a dense OBS array: Implications for fault strength variations , author=. Earth and Planetary Science Letters , volume=. 2016 , publisher=

    2016

  40. [40]

    Journal of Geophysical Research: Solid Earth , volume=

    AI-PAL: Self-supervised AI phase picking via rule-based algorithm for generalized earthquake detection , author=. Journal of Geophysical Research: Solid Earth , volume=. 2025 , publisher=

    2025

  41. [41]

    Earth, Planets and Space , year=

    Predictable and unpredictable aspects of earthquakes from P-wave onsets of acceleration seismograms , author=. Earth, Planets and Space , year=

  42. [42]

    Geophysical Research Letters , volume=

    Real-time earthquake location based on the Kalman filter formulation , author=. Geophysical Research Letters , volume=. 2020 , publisher=

    2020

  43. [43]

    Bulletin of the Seismological Society of America , volume=

    Real-time evolutionary earthquake location for seismic early warning , author=. Bulletin of the Seismological Society of America , volume=. 2008 , publisher=

    2008