Effects of stochastic and natural seismic noise on the performance of waveform cross-correlation used to recover low-magnitude seismicity prior to the July 29, 2025, Kamchatka earthquake

Ivan Kitov

REVIEW 3 major objections 5 minor 3 references

Effects of stochastic and natural seismic noise on the performance of waveform cross-correlation used to recover low-magnitude seismicity prior to the July 29, 2025, Kamchatka earthquake

T0 review · 3 major / 5 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read Low-magnitude waveform-correlation detections made before the July 29, 2025, Kamchatka earthquake are statistically significant, not artifacts of random noise or remote seismic side-sensitivity.

desk verdict A useful but incomplete validation study: the random-noise control uses 20 master events while the catalog it certifies used 100, and the thresholds were tuned to suppress random events, so the central 'not biased' claim is not yet established. read the letter →

arxiv 2607.16226 v1 pith:UC4IRDEE submitted 2026-06-20 physics.geo-ph physics.soc-ph

Ivan Kitov This is my paper · ORCID

classification physics.geo-phphysics.soc-ph MSC 86A15 PACS 91.30.-f

keywords waveformcross-correlationlow-magnitudeseismicityKamchatkaearthquake2025seismicarraysrandomnoisenulltestmatchedfilterprecursorwhitening

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper defends the hundreds of very weak, waveform-cross-correlation detections made before the July 29, 2025, Kamchatka earthquake against two charges: that they are random noise coincidences, and that they are false signals from large remote earthquakes leaking into the array stations. By running the same detection pipeline on computer-generated random noise in place of real waveforms, the author builds a null bulletin that produces essentially no events; this is used to argue that the real pre-earthquake events are statistically significant. A second test with the 2011 Tohoku earthquake shows that remote signals can indeed create false local events from Kamchatka templates, but that adding Tohoku master events resolves the conflict and eliminates the false events. If true, the rising low-magnitude activity before the mainshock can be treated as a candidate precursory indicator rather than a processing artifact.

What carries the argument

The central machinery is the waveform cross-correlation (WCC) pipeline itself, run with a computer-generated pseudorandom noise waveform in place of the real seismic records. The random-noise run produces a null bulletin — the same local association and conflict-resolution steps, the same master events, the same event-definition criteria — against which the real cross-correlation event list (XSEL) can be compared. The StN scaling factor controls the amplitude of the added noise and is swept to find the value that maximizes valid detections; the side-sensitivity test uses the 2011 Tohoku signals as a stand-in for remote coherent noise.

What would settle it

Take one quiet hour of real array data with no reported events in the region, phase-scramble each channel to remove coherent arrivals while preserving the amplitude spectrum, and run the same WCC pipeline with the same master events; if the resulting bulletin contains more than a handful of events and approaches the real pre-earthquake XSEL rate, the white-noise null underestimates false detections and the statistical-significance claim is weakened.

Watch

Extended reading notes

Core claim

The central claim is that the waveform cross-correlation (WCC) event hypotheses used to build the precursory time series are not products of random noise or remote side-sensitivity. When random-noise waveforms are processed through the same detection, association, and conflict-resolution pipeline, the resulting null bulletin contains a negligible number of events, and the few that appear are confined to the loosest detection settings. The side-sensitivity test with the 2011 Tohoku earthquake shows that Kamchatka master events can generate false local events from high-amplitude remote signals, but adding Tohoku master events makes those false events disappear in conflict resolution. The autho

Load-bearing premise

The conclusion that the real XSEL events are statistically significant rests on treating a computer-generated white-noise waveform, scaled to the maximum amplitude of each hourly interval, as an adequate stand-in for real ambient seismic noise when counting false events; real noise is not purely stochastic, since it is a finite mixture of regular seismic phases.

Editorial extensions

If this is right

If the random-noise null is accepted, the low-magnitude events recovered before the 2025 Kamchatka earthquake are real, and their increase can be treated as a candidate precursory signal rather than an artifact.
The pattern of low-magnitude activity prior to Kamchatka 2025 matches the pattern prior to the 2013 Sea of Okhotsk earthquake, implying a repeating preparation signature for large subduction events.
Adding master events from adjacent seismic zones removes false events caused by remote high-amplitude signals, showing that the side-sensitivity problem is solvable by denser, geographically extended template coverage.
The optimal StN = 0.075 noise addition will be used to reprocess the Kamchatka and Okhotsk sequences and should improve XSEL bulletins by making the matched-filter detector closer to optimal.
The match statistics with independently reviewed events (up to 244 of 278 reviewed events matched in the first six post-seismic hours with 100 master events) show that most XSEL events correspond to real, independently reviewed earthquakes.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The null result depends on using white noise as the false-event generator; real ambient seismic noise contains coherent phases that could be partially similar to templates, so the true false-positive rate in a real low-seismicity day may be higher than the white-noise bulletin suggests. A more persuasive null would use real noise with all known events removed, or phase-scrambled noise that preserv
The side-sensitivity test uses a single far-away source (Tohoku); other geometries and azimuths could behave differently, so the paper's conclusion that side-sensitivity is a natural positional effect may not generalize to all remote sources or all array stations.
The paper defines the pre-quake increase as a candidate precursor indicator, not a prediction; whether the same signature appears before other mega-earthquakes can be tested by applying the identical pipeline to historical sequences before other subduction megathrusts.
The StN = 0.075 whitening result suggests that adding controlled noise is a tunable preprocessing step; one could apply adaptive scaling per station and frequency band to optimize detection without a single global parameter.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

This is a companion to Kitov 2026d, not a standalone result. What is genuinely new: a random-noise WCC bulletin as a null reference, a Tohoku side-sensitivity test, and the StN=0.075 stochastic-whitening effect. The Tohoku experiment is the strongest part — during six hours of Tohoku aftershocks, 20 Kamchatka MEs produce a couple hundred false Kamchatka XSEL events, and adding 12 Tohoku MEs makes those false events mostly disappear because the Tohoku MEs win conflict resolution. That is a concrete, interpretable result about the pipeline. The REB-match tables also give the XSELs a useful external anchor; high match rates on actual days, including pure REB events, say something real. The soft spot is exactly the stress-test concern. The random-noise control uses 20 MEs, while the catalog it is meant to certify used the 100 best MEs. Table 3 shows density matters: 100 MEs produce roughly two to five times more XSEL events on the same days. More templates mean more chances for a random detection chain to pass Local Association, and the near-zero random counts in Table 1 are not a bound for the 100-ME configuration. Add the paper's own admission that the EDC thresholds were designed to prohibit random events and tuned on quiet days, and the sentence saying the earlier results are 'not biased by random false detections' is overreach. The control is suggestive, not conclusive. Secondary issues: ran3 white noise is not ambient noise, and the Tohoku test covers one remote source, not the general case. There are no error bars or multiple realizations, and StN=0.075 is fitted on the same days rather than out-of-sample. The precursor framing relies on two retrospective earthquakes. None of these are fatal by themselves, but they do mean the paper is a feasibility study, not a confirmation. Who gets value: anyone working on matched-filter detection with arrays, especially on low-magnitude catalogs; the Tohoku side-sensitivity experiment is worth reading carefully. It deserves peer review — a serious referee could push for a 100-ME random control, multiple realizations, and a clearer separation between calibrated and tested claims. I would not cite it as evidence that the Kamchatka precursor is established until that control is run.

Referee Report

3 major / 5 minor

Summary. The paper assesses whether low-magnitude events detected by a waveform cross-correlation (WCC) pipeline before the July 29, 2025, Kamchatka earthquake—and used in a companion study to infer precursory activity—are statistically significant rather than artifacts of random noise or array side-sensitivity. The author replaces or augments actual IMS waveforms with computer-generated ran3 random noise scaled by a factor StN, runs the same WCC detection, Local Association, and Conflict Resolution pipeline, and compares the resulting XSEL bulletins with those from real data. Tables 1 and 4 show that for the 20-master-event configuration, high StN values produce zero or near-zero random false events, while a Tohoku earthquake test shows that remote real signals can generate many false Kamchatka events unless Tohoku master events are added. The paper concludes that the earlier WCC results are not biased by random false detections and that the XSEL statistical significance is confirmed.

Significance. If the central claim is correct, the paper provides an important null reference for WCC-based low-magnitude catalogs and supports the use of XSEL event counts as a precursory variable. The work has genuine strengths: it is transparent about the pipeline parameters (Appendix 2), it provides extensive quantitative comparisons to REB data (Tables 1–6), and it directly attempts to model both stochastic noise and one class of coherent false-event contamination. The Tohoku test is a useful demonstration that master-event coverage outside the target region can suppress side-sensitivity. However, the validation is not yet at the level needed to certify the 100-ME catalog used in the prior prediction study, because the random-noise control is run with only 20 master events and the event-definition thresholds were themselves designed to make random events rare.

major comments (3)

[WCC Local Association and Conflict Resolution with stochastic noise; Table 1 vs. Table 3] The random-noise false-event control is run only with the 20-ME feasibility set, whereas the catalog whose significance is being certified was built with the 100 best MEs. Table 3 shows that the 100-ME configuration produces roughly 2–5 times more XSEL events on the same days (e.g., 2025209, index 1: 457 vs. 239; index 12: 18 vs. 4). More templates mean more matched-filter opportunities and more chances for noise realizations to pass the Local Association thresholds; Conflict Resolution may suppress some, but the tail behavior is not shown to remain at zero. Therefore the near-zero random counts in Table 1 for StN=10 and 100 are not a bound on the random false-event rate for the 100-ME pipeline, and the sentence 'The results reported in [Kitov, 2026bd] are not biased by the random false detections' is unsupported. The authors should rerun the random-noise experiment with the full 100-ME
[Appendix 2 and Section 'WCC Local Association and Conflict Resolution with stochastic noise'] The validation is partly circular. Appendix 2 states that the EDC thresholds were estimated from quiet-day SNRcc distributions, and the main text notes 'By design, the EDC are intended to prohibit the creation of random events.' The near-zero random-noise bulletin is therefore, to a substantial degree, a consequence of the thresholds being calibrated to make random events rare, rather than an independent confirmation of statistical significance. The test verifies the implementation, but it cannot by itself certify the significance of the XSEL events. The paper should either use a pre-specified false-positive rate, show how the quiet-day SNRcc curves map to that rate, or demonstrate that the actual quiet-day new-XSEL counts (e.g., Table 3) are consistent with the claimed false-positive probability rather than with zero.
[Side-sensitivity of the WCC at array stations; Table 4] The random-noise null model is white ran3 noise, but the paper itself acknowledges that real ambient noise 'is not purely stochastic, as it is created by a finite mixture of regular seismic phases.' The Tohoku experiment (Table 4) shows that real coherent noise from a remote source is not negligible: the 20 Kamchatka MEs generate up to 235 false XSEL events during the six-hour Tohoku window. A single remote-source test is not a general bound on coherent-noise false positives from all possible external sources. Thus the conclusion that XSEL events are statistically significant because white-noise random counts are negligible underestimates the false-positive rate under realistic coherent noise. The authors should either apply the random-noise control to real noise records from quiet intervals at the same stations or otherwise quantify how often coherent non-target signals can produce LA-p

minor comments (5)

[Table 1 caption] Typo in day label: '20025210' should be '2025210'.
[Figure 2 caption] The caption lists 'PDAR' but the panel label and text refer to 'PDYAR'; please make the station names consistent.
[References and in-text citations] The citation '[Kitov, 2026bd]' is used inconsistently; the reference list contains only 'Kitov, I. O. (2026d)'. Please harmonize the citation key.
[General text] There are several typographical errors, e.g., 'cannon serve' for 'cannot serve' in the CMAR paragraph and 'PDAR' vs. 'PDYAR'. A careful proofreading pass is recommended.
[Section 'WCC Local Association and Conflict Resolution with stochastic noise'] The phrase 'there can be an extremely rare chance' is awkward; consider rephrasing for clarity.

Circularity Check

2 steps flagged · score 5.0 of 10

Random-noise significance test is partly self-fulfilling: the EDC thresholds that suppress random events were calibrated in same-author prior work and then used to certify the XSEL bulletin; REB matches provide external support but do not remove the circularity.

fitted input called prediction [Section 'WCC detection with stochastic noise' (after Fig. 4) and Section 'WCC Local Association and Conflict Resolution with stochastic noise']
"By design, the EDC are intended to prohibit the creation of random events. The SNRcc curves for quiet days were used to estimate the thresholds in the LA to guarantee very low probability of a random event being created. ... The results reported in [Kitov, 2026bd] are not biased by the random false detections."

The random-noise bulletin is generated through the same EDC thresholds that were fitted, from quiet-day SNRcc curves, specifically to make random events improbable. Finding near-zero random XSEL events is therefore a restatement of that threshold choice, not an independent measurement of the false-event rate. The claim that XSEL events are statistically significant is presented as the experiment's output, while the input already encodes the suppression of such events (mandatory top-station SNRcc 5.0/4.5, which the stochastic curves in Figs. 2-4 almost never reach).
self citation load bearing [Section 'WCC detection with stochastic noise', paragraph on EDC]
"The sets of EDC were estimated in [Kitov, 2026d] from the SNRcc frequency distributions at the IMS arrays using quiet days without REB events and very low numbers of XSEL events."

The threshold parameters that make the random-noise test come out near zero are imported from a same-author prior paper; they are not re-derived in this paper. The current paper's confirmation of statistical significance thus leans on a self-citation chain, with the cited work itself using the same quiet-day curves to design against the very random events the present paper claims to rule out.

full rationale

The central validation claim is not fully independent. The random-noise reference bulletin is generated through EDC thresholds that were deliberately estimated in [Kitov, 2026d] from quiet-day SNRcc curves to prohibit random events, so the near-zero random count is partly built into the test. Moreover, the control is run with only 20 MEs ('For this feasibility study, twenty MEs were selected'), while the catalog whose significance is certified was built with the 100 best MEs; Table 3 shows the 100-ME configuration produces 2-5 times more XSEL events (e.g., 457 vs 239 for index 1 on 2025209; 18 vs 4 for index 12), and the paper itself concedes 'The configuration of 20 MEs is too sparse to be used for these purposes.' Thus the random-noise null is not a direct bound on the 100-ME pipeline. However, the REB-matching statistics provide external grounding: on July 30 the 100-ME XSEL matched up to 244/278 REB events, and the paper's conclusion that most XSEL events correspond to analyst-reviewed seismicity does not reduce to the circular random-noise test. On balance the central hypothesis has independent content, but the headline significance claim contains a fitted-input-as-prediction step and a load-bearing self-citation; score 5.

Assumptions & free parameters 3 free parameters · 6 assumptions · 0 invented entities

The central statistical claims rest on several hand-set parameters: the EDC threshold set, which was tuned to suppress random events; the six origin-time tolerances; and the fitted StN=0.075 optimum. The random-noise null model and the representativeness of the Tohoku case are domain assumptions rather than proven facts. No new physical entities are introduced.

free parameters (3)

StN_optimal_noise_scaling = 0.075
Grid search over StN from 0.0 to 1.0 to maximize XSEL events and REB match rate; fitted to the July 2025 dataset and later described as calibrated for future reprocessing.
EDC_LA_threshold_set = weak: min station weight 0.80, min SNRcc 4.5, sums 14/17.5/21; strict: 0.855, 5.0, sums 15/18.5/22; grid radii 90/48 km
Thresholds and weights were 'guesstimated' from quiet-day SNRcc distributions and previous experience, and are designed to suppress random events; the random-noise false-event count is therefore partly controlled by these hand-set parameters.
origin_time_tolerances = 5.0, 3.0, 2.0, 1.0, 0.5, 0.25 s
Chosen by hand to define the twelve strict/weak version-case pairs; the rate of random associations depends directly on these tolerances.

assumptions (6)

ad hoc to paper Computer-generated white noise (ran3) is a valid surrogate for the false-positive component of real seismic noise at IMS array stations.
Used to build the random-noise XSEL bulletin in Section 'WCC detection with stochastic noise'; the paper acknowledges ambient noise is non-stochastic but treats ran3 noise as 'a better choice.'
domain assumption The LA/EDC thresholds calibrated on quiet days remain the correct false-positive thresholds during active and high-noise periods.
Invoked when counting XSEL events on July 28-30 and in the Tohoku experiments; if quiet-day thresholds fail under high activity, the random-noise and false-event counts are not comparable.
domain assumption REB bulletins from the IDC are a valid independent ground truth for matched and unmatched XSEL statistics.
Used throughout Tables 3 and 6 to assign real/false status to XSEL events; REB is human-reviewed but built from the same IMS station data and may share systematic biases.
domain assumption The 2011 Tohoku earthquake is a representative proxy for remote high-amplitude sources that can trigger side-sensitivity of Kamchatka templates.
Section 'Side-sensitivity...' uses one six-hour Tohoku window to model all remote side-sensitivity; other azimuths and source geometries are not tested.
ad hoc to paper A single random-noise realization with a given StN is sufficient to characterize the random false-positive rate.
Tables 1 and 4 report single counts without multiple realizations or confidence intervals; the conclusion 'zero random events for StN=100' is based on one realization.
standard math WCC acts as a matched-filter detector whose sensitivity improves when noise is whitened or stochastized.
Basis for the StN=0.075 'closer to optimal matched filter' claim; from Turin (1960) and cited prior work.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Effects of stochastic and natural seismic noise on the performance of waveform cross-correlation used to recover low-magnitude seismicity prior to the July 29, 2025, Kamchatka earthquake." pith.science (2026). https://pith.science/paper/UC4IRDEE

@misc{pith2026260716226,
  author       = {Pith},
  title        = {Pith review of: Effects of stochastic and natural seismic noise on the performance of waveform cross-correlation used to recover low-magnitude seismicity prior to the July 29, 2025, Kamchatka earthquake},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UC4IRDEE}},
  note         = {Machine review of arXiv:2607.16226}
}

read the original abstract

Waveform cross-correlation (WCC) applied to data from seismic arrays allows for the reduction of the detection threshold by approximately an order of magnitude. When applied to the data of the IMS, the smallest WCC detected events were by at least one magnitude unit lower than those reported by the IDC and revealed that the pattern of low-magnitude activity prior to the July 29, 2025, earthquake was similar to that prior to the May 24, 2013, Sea of Okhotsk earthquake. The consistent increase in the number of events with magnitudes approaching the corner value of the Kamchatka recurrence curve can likely be used as a precursory indicator of a mega-earthquake preparation. The WCC-based events are characterized by detections with a signal-to-noise ratio below the IDC detection threshold and are often not visible to analysts. This makes the statistical significance of the WCC event hypotheses uncertain due to the absence of a random noise reference and the potential side-sensitivity of the IMS arrays. Both problems are addressed. Random noise is used to calculate a WCC bulletin. This exercise shows a negligibly low number of WCC events and confirms the statistical significance of the WCC-based events. The side-sensitivity is modelled using the March 11, 2011, Tohoku earthquake. When master events (MEs) within the Kamchatka region are used in the WCC processing, a large number of false events are generated. When MEs from the Tohoku zone are added, all false events disappear since the Tohoku MEs win the conflict resolution process. When the random noise is scaled to a small fraction of the maximum amplitude within the processed interval, the WCC detection is enhanced with more detections and WCC-based events created. This is an effect similar to noise whitening, which makes the matched filter detector more sensitive and closer to its optimal performance under stochastic noise conditions.

Figures

Figures reproduced from arXiv: 2607.16226 by the authors.

**Figure 2.** Frequency distribution of SNRcc values at four stations (PETK, PDAR, MKAR, KURK, and WRA) on July 28, 2025. Twenty master events are used. Station may miss some of the MEs due to maintenance works, upgrade, or late start of operation. a) b) 1.E+00 1.E+01 1.E+02 1.E+03 1.E+04 1.E+05 1.E+06 0 1 2 3 4 5 6 # SNRcc 23689063 23329683 23316512 22605238 21227220 20280472 18683607 18517208 18425349 17328560 16438469 16313833… view at source ↗

**Figure 4.** Frequency distributions of SNRcc at station NOA. [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: Number of XSEL events as a function of the index of the version [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

3 extracted references · 1 linked inside Pith

[6]

The lowest possible value for the minimum SNRcc at one of the top stations from 3) is 5.0
[7]

The event hypotheses beyond a radius of 43.2 km are rejected

The grid radius is 48 km, defined by 12 steps of 4 km. The event hypotheses beyond a radius of 43.2 km are rejected. This radius can be less than half the distance between neighboring MEs, but it is important to increase statistical significance by reducing the flexibility in location. On the opposite side of the XSEL sensitivity is the weak LA version. T...
[1618]

https://doi.org Tromp, J., Tape, C., and Liu, Q. (2005). Seismic tomography, adjoint methods, time reversal and banana-doughnut kernels. Geophysical Journal International, 160(1), 195–216. https://doi.org/10.1111/j.1365-246X.2004.02453.x Turin, G. L. (1960). An introduction to matched filters. IRE Transactions on Information Theory, 6(3), 311–329. https:/...

arXiv 2005

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [6]

The lowest possible value for the minimum SNRcc at one of the top stations from 3) is 5.0

[2] [7]

The event hypotheses beyond a radius of 43.2 km are rejected

The grid radius is 48 km, defined by 12 steps of 4 km. The event hypotheses beyond a radius of 43.2 km are rejected. This radius can be less than half the distance between neighboring MEs, but it is important to increase statistical significance by reducing the flexibility in location. On the opposite side of the XSEL sensitivity is the weak LA version. T...

[3] [1618]

https://doi.org Tromp, J., Tape, C., and Liu, Q. (2005). Seismic tomography, adjoint methods, time reversal and banana-doughnut kernels. Geophysical Journal International, 160(1), 195–216. https://doi.org/10.1111/j.1365-246X.2004.02453.x Turin, G. L. (1960). An introduction to matched filters. IRE Transactions on Information Theory, 6(3), 311–329. https:/...

arXiv 2005