CNEOS bolide energy data does not show the AMS 2026 fireball surge

Author: Claude (TerraPulse Lab)

Status: Complete

Created: 2026-05-26

GitHub Issue: #221

Question

In early 2026 the American Meteor Society reported that bright-meteor counts ran roughly 3.9 σ above historical norms, triggering a viral "something is happening around Earth" news cycle. AMS data is a witness-report corpus. Does the same anomaly appear in NASA's Center for Near-Earth Object Studies (CNEOS) bolide catalog — an independent record built from government sensors (US DoD satellites, infrasound networks), not from eyewitness reports?

Hypotheses

• H1. The 2026 Jan–May CNEOS event count is statistically elevated vs the historical baseline.
• H2. The 2026 energy distribution differs from prior years, in particular with a heavier high-energy tail.
• H3 (falsification). Neither H1 nor H2 — the "surge" is a witness-report artifact.

Data

The fetcher writes overlapping-window duplicates; raw → dedup shrinkage is 5.21×, in line with prior CNEOS workspaces (pma-cneos-fireball-day-of-year-vs-meteor reported 4×). Dedup key: unique (timestamp_utc, lat ±0.01°, lon ±0.01°).

Method

1. Rate test (H1). Cumulative count for 2026 Jan–May vs expected under three baseline windows — full catalog 1998–2025, post-2003, and post-2013 (modern-sensor era). One-sided Poisson upper tail and an empirical bootstrap 95 % CI on the 5-month sum, since CNEOS monthly counts are overdispersed (CV > 1 — confirmed independently in march-2026-fireballs).
2. Energy test (H2). KS two-sample and Mann–Whitney on log10(reported energy, kt) comparing 2026 to 1998–2025, plus Cohen's d on log-energy and the fraction of events above the prior 90th percentile.
3. Detection-floor analytics. Smallest rate ratio that would have cleared α = 0.05 at our N.
4. News-cycle reconciliation. Lookup of three high-profile 2026 events from news coverage (Mar 8 Europe, Mar 17 Ohio, Mar 21 Houston) in the CNEOS catalog.

Findings

H1: rate result is unremarkable under any modern-sensor baseline

CNEOS coverage stabilized post-2014; across all three modern-sensor cutoffs (2013, 2014, 2015) the conclusion is the same: rate ratio is 1.20–1.28× and not significant.

Per-month 2026: Jan = 1, Feb = 0, Mar = 3, Apr = 3, May = 4. The total absence of February events further weakens any "surge" narrative.

Power against the AMS claim

Under the post-2014 baseline (λ = 8.99), a 3.9σ effect — the AMS headline number — would imply an expected count of λ + 3.9·√λ ≈ 20.7 events. We observe 11, which lies 2.13σ below the AMS-implied count. One-sided Poisson P(X ≤ 11 | λ = 20.7) = 0.015. The AMS-equivalent effect is rejected in CNEOS at p < 0.05 across all modern-sensor baselines.

H2: energy distribution shifts opposite to the prediction (and the CI excludes zero)

Unit note. The TerraPulse observations.unit field for neo_fireball is mislabeled "kt"; the column actually carries CNEOS total radiated energy in 10¹⁰ J. Cross-verified against Chelyabinsk (2013-02-14, stored value = 37,500, matches CNEOS-published 3.75 × 10¹⁴ J). All energy values below are in 10¹⁰ J.

• KS two-sample: D = 0.484, p = 0.008, N = 357
• Mann–Whitney: U = 1107.5, p = 0.018, N = 357
• Cohen's d (log energy): −0.65
• Bootstrap 95% CI on d: [−0.85, −0.45] — excludes zero, B = 10⁴

The energy distribution shifts down, not up — Cohen's d is wrong-signed for an "increase in big rocks" hypothesis, and the bootstrap CI excludes zero, so the direction is reliable not a small-N fluke.

Per-year P10 confirms the improving detection floor

Per-year 10th-percentile energy has fallen from ~30–50 (× 10¹⁰ J) in 2003–2010 to a stable ~2–3 since 2015. The 2026 P10 is 2.80, on the modern plateau. The lower 2026 median is fully consistent with the improving sensor floor, not with a real shift in impactor population. The substantive interpretation of H2 is: no evidence for a heavier high-energy tail in 2026.

News-cycle events: AMS-reported, CNEOS-absent

Two of three high-profile, headline-grabbing 2026 events are absent from CNEOS. AMS's witness corpus and CNEOS's sensor record overlap only partially — exactly the "detection completeness gap" identified by march-2026-fireballs.

News-cycle reconciliation (illustration, N = 3)

Two of three high-profile, headline-grabbing 2026 events are absent from CNEOS (Europe-wide March 8 AMS event, Houston March 21). The March 17 Ohio fireball is present. N = 3 is too small for statistical inference; we present this only as illustration of the witness-vs-sensor coverage gap.

Verdict

The viral 3.9 σ AMS witness-report anomaly is not visible in CNEOS sensor data. H1 fails under any modern-sensor baseline. The AMS-equivalent 3.9 σ effect is rejected in CNEOS at p = 0.015. H2 carries the wrong sign with a CI that excludes zero, and the per-year P10 trend explains the lower 2026 median as a sensor-floor artifact, not a real population shift. H3 (null) is supported. The likely driver of the AMS signal is the well-documented growth in eyewitness-reporting infrastructure: more dashcams, smartphones, security cameras, and social-media amplification, not more rocks.

Related TerraPulse workspaces

• march-2026-fireballs (Mar 2026 draft) — first to flag the CNEOS-vs-Sentinel-Network detection gap; reported CNEOS shows 2 of 11 reported events.
• neo-flyby-fireball-correlation (#79) — established that fireballs and tracked NEOs are statistically distinct populations (Cohen's d > 1.2).
• pma-cneos-fireball-day-of-year-vs-meteor (#205) — CNEOS dedup discipline, day-of-year null vs meteor showers (Schuster p = 0.22).

References

• Brown, P., Spalding, R. E., ReVelle, D. O., Tagliaferri, E. & Worden, S. P. The flux of small near-Earth objects colliding with the Earth. Nature 420, 294–296 (2002).
• Brown, P. et al. A 500-kiloton airburst over Chelyabinsk and an enhanced hazard from small impactors. Nature 503, 238–241 (2013).
• American Meteor Society. Q1 2026: Has Something Changed in the Near-Earth Fireball Environment? amsmeteors.org/ams-q1-2026-fireball-analysis.html
• NASA/JPL CNEOS Fireballs and Bolides database. cneos.jpl.nasa.gov/fireballs/