The American Meteor Society Says
Fireballs Have Surged.
What do the Sensors Actually Say?

The Claim

In March 2026 the American Meteor Society — the volunteer organization that aggregates eyewitness fireball reports — published a quarterly summary stating that early-2026 bright-meteor witness counts were running roughly 3.9 standard deviations above the historical norm. News outlets ran with it: ZME Science published "Inside 2026's Massive Fireball Surge"; EarthSky described "a flurry of fireballs"; aggregators quoted "nobody can explain what changed."

The catch is that AMS counts are eyewitness reports. They depend on how many cameras are pointed at the sky, how widely unusual sky events spread on social media, and how many people know AMS exists. All three have risen steeply on multi-year timescales — smartphone penetration, dashcam adoption, security-camera saturation, TikTok and Instagram amplification. Any of those can inflate witness counts without changing the rate at which rocks actually hit the atmosphere.

The Independent Check

NASA's Jet Propulsion Laboratory maintains a separate catalog: the Center for Near-Earth Object Studies (CNEOS) Fireballs and Bolides Database. It doesn't ask people; it gets its data from government sensors — mainly Department of Defense satellites that watch for the heat signature of explosions in the upper atmosphere, plus infrasound arrays that hear the boom from far away. CNEOS only catches a fraction of all impacts (estimates run 10–20% of kiloton-scale events), but the fraction it catches is selected by physics, not by whoever happens to be holding a phone.

If the AMS surge reflected an actual change in the impactor population, the same anomaly should show up in CNEOS. If it's a reporting-infrastructure artifact, CNEOS should look normal. That's the test.

Step One — Cleaning the Catalog

The CNEOS bolide catalog in our database holds 1,064 records. After deduplicating on timestamp and rounded location we get 1,062 unique events spanning April 1988 through May 2026 — this feed is already essentially clean (only two duplicates), unlike the thinner, poll-duplicated snapshot an earlier version of this analysis ran on.

Step Two — How Many Fireballs in 2026?

January through May of 2026, CNEOS logged 13 events. Is that a lot? It depends entirely on what baseline you compare against:

On the complete catalog there is no surge under any baseline. Even the full 1998–2025 comparison comes in at 0.91× — below one — and every modern-sensor baseline (2013, 2014, 2015 onward) lands at 0.85–0.88×. Thirteen events is statistically unremarkable; if anything it sits slightly below expectation, not above it. The 2.14× ratio that drove the headlines came from a thin, incomplete catalog that undercounted the baseline years — on the complete CNEOS feed it disappears entirely.

One more thing worth noting: the 13 events break down as Jan = 2, Feb = 1, Mar = 3, Apr = 4, May = 3 — spread across the window with no single month standing out as a spike. A real surge isn't normally that flat.

Step Three — Can We Detect the AMS Claim at All?

A reasonable question at this point: maybe our sample is just too small to find anything? Eleven events isn't much. So we asked a sharper question: if the AMS 3.9σ effect were physically real and reflected in the impactor population, what would CNEOS look like?

Against the post-2014 baseline of 15.2 expected events, a 3.9σ effect would imply about 30.4 events in our window. We observed 13. Thirteen is 3.16 standard deviations below what the AMS claim implies. The one-sided Poisson probability of observing 13 or fewer events under that AMS-implied rate is p ≈ 0.0003.

That's the substantive answer to the public's question. We are not weakly failing to detect a possible surge. We are statistically rejecting the AMS-equivalent claim in the sensor data, at the standard 5% threshold, across every baseline we tried.

Step Four — The Energy Distribution Tells the Same Story Backwards

Beyond just counting events, we can ask: are the 2026 events bigger than usual? A real flux increase from a swarm of debris or a comet shedding might mean more kiloton-scale events specifically. We compared the energy distributions:

The 2026 distribution is statistically different from the prior — but it's smaller, not larger. The Cohen's d effect size is moderate and negative, with a bootstrap confidence interval that comfortably excludes zero. None of the 13 events exceed the prior 90th percentile. If a swarm of bigger rocks were hitting Earth, the distribution should shift right; it shifted left.

Why? CNEOS's detection floor has been improving for two decades. The per-year 10th percentile of energy fell from around 50 (×10¹⁰ J) in 2003–2010 to a stable 2–3 since 2015. The 2026 P10 is 2.2, on the modern plateau. We're catching smaller events now than we used to, which pulls the median down — even as the rate of big events stays flat.

Step Five — The Famous Events from the News

Three 2026 fireballs got particular press coverage: a March 8 European event that AMS logged 3,229 witness reports for; a March 17 Ohio fireball that recovered an actual meteorite; and a March 21 Houston airburst estimated at 26 kilotons. We checked whether each appears in CNEOS:

Two of the three most-reported 2026 fireballs are absent from the sensor catalog. With a sample size of three, this is illustration rather than statistical evidence — but it shows in plain view what the witness-versus-sensor coverage gap actually looks like. AMS records many bright meteors that CNEOS sensors miss, either because the event fell below the sensor energy threshold or because no sensor was looking the right way.

Figure: Twenty-Eight Years of Sensor Counts

What This Result Does and Does Not Say

It says: the rate of kiloton-scale impacts measured by independent government sensors is not elevated in 2026. The AMS-equivalent 3.9σ effect is statistically rejected in the sensor data at p ≈ 0.0003 across every baseline we tried. The energy distribution shifts the wrong direction for a "more big rocks" hypothesis, with a confidence interval that excludes zero. Per-year detection-floor trends fully explain the lower 2026 median as a sensor-improvement artifact rather than a real population change.

It does not say: that fireballs aren't happening, or that AMS is wrong about people seeing more fireballs. People are almost certainly reporting more bright meteors than they used to — what we're saying is that the underlying flux of rocks hitting the atmosphere hasn't changed to match. The simplest explanation is the obvious one: more cameras, more sharing, more reports. The sky has not changed; the audience for it has.

Reproducibility

All extraction scripts, analysis code, deduplicated catalog, and the full results.json live in the cneos-fireball-2026-surge workspace. Every number on this page comes from that JSON. Pipeline: PostgreSQL extraction of the CNEOS fireball catalog (now the canonical cneos_bolide spine; this analysis originally ran on the since-consolidated neo_fireball metric), deduplication on (timestamp, ±0.01° lat/lon), Poisson rate test across five baseline windows with bootstrap overdispersion CIs, AMS-equivalent power statement, two-sample Kolmogorov–Smirnov and Mann–Whitney on log-energy, Cohen's d with bootstrap CI on 10⁴ resamples, and per-year P10 detection-floor measurement.