UFO Reports Don't Cluster
Around US Earthquakes

The Folklore

Earthquake lights are an old folklore claim: visible flashes, glows, ball lightning, and columns of light reported in the hours before or during significant earthquakes. The claim is recorded in seismology literature since the 1965 Niigata earthquake in Japan, and continues to surface after every notable quake in popular media. A 2014 paper by Theis et al. proposed a piezoelectric mechanism, in which stress on certain rock types generates electrical charge that ionizes air. The mechanism remains contested.

If the claim holds at population scale, meaning that real people see real things during real earthquakes, then the simplest population-level signature is that eyewitness reports of weird sky phenomena should cluster around earthquake epicenters. The TerraPulse platform stores both layers needed to test that prediction.

The Test

We cross-matched two public catalogs:

• USGS (U.S. Geological Survey) earthquakes, magnitude (M) ≥ 4.5, located in US territory, 1910–2014. After deduplication: 9,379 events. Backfilled from the Advanced National Seismic System Comprehensive Catalog (ANSS ComCat) via the public International Federation of Digital Seismograph Networks (FDSN) web service.
• NUFORC (National UFO Reporting Center) sightings, US-resident only, with non-null coordinates, 1910–2014. 71,008 reports. Each carries an alleged sighting time, a latitude and longitude, and a state.

For each earthquake, we counted NUFORC reports within 100 km of the epicenter and within the 25-hour window from 24 hours before to 1 hour after the quake time. The asymmetric window reflects the folklore claim that earthquake lights appear before or during the quake. Post-event reports are a different phenomenon (people watching the news, looking up after the shaking).

Across 458,960 spatial pairs (NUFORC reports within 100 km of an epicenter), we got three coincidences.

Three vs Nine

The right comparison is not "is 3 a lot?" but "is 3 a lot compared to chance?" We answered this by shuffling: take the same 71,008 NUFORC reports, keep their locations fixed (so the population-density bias is preserved), and randomly re-assign their timestamps. Recompute the coincidence count under the shuffle. Do this 10,000 times.

Under random matching, the average coincidence count is 9.4 with a standard deviation of 3.5. The middle 95% of the random distribution lies between 3 and 17. Our observed three sits at the low edge of that range, at the 1.1th percentile.

The folklore predicts more coincidences than chance. We see fewer. The probability of observing at least as many coincidences as we did, under the random null, is 0.989. That's a clean null result on the folklore claim.

The Biggest Quakes Show Nothing Either

If earthquake lights are real, they should be most visible at the largest events. We broke the catalog into four magnitude bins:

• M 4.5–5.0 (5,479 quakes): 1 observed vs 6.7 expected
• M 5.0–5.5 (1,982 quakes): 0 observed vs 1.2 expected
• M 5.5–6.0 (1,162 quakes): 2 observed vs 1.5 expected
• M ≥ 6.0 (756 quakes): 0 observed vs 0.1 expected

The 756 events in the M ≥ 6 bin include some of the most consequential earthquakes in modern American history: the 1964 Anchorage M 9.2, the 1989 Loma Prieta M 6.9, the 2002 Denali M 7.9. None of them carries a coincident NUFORC report in the 100 km / 25-hour window. The folklore predicts that these are exactly the events that should produce the most visible lights. The data says they produced no eyewitness UFO reports detectable by the network.

Sensitivity

Single results on arbitrary parameter choices are weak. We re-ran the test under five variants of the window and magnitude floor. The null result is robust:

• Wider window (−48 h to +1 h): 9 observed vs 19.7 expected. Rate-ratio 0.46.
• Tighter window (−6 h to +1 h): 0 observed vs 2.9 expected. Rate-ratio 0.
• Lower magnitude floor (M ≥ 3.5, 32,429 quakes): 95 observed vs 124 expected. Rate-ratio 0.77.
• Higher magnitude floor (M ≥ 5.5, 1,918 quakes): 2 observed vs 1.6 expected. Rate-ratio 1.24, but the count of 2 is too small to be statistically significant (p = 0.48).
• 1995–2014 only (the era when NUFORC volume is heaviest, 3,050 quakes): 3 observed vs 5.9 expected. Rate-ratio 0.51.

The headline result survives every variant. The only point estimate above one is the M ≥ 5.5 bin, where the absolute count of 2 leaves the result well inside the noise floor.

What This Doesn't Say

A clean null is a finding, but it is a bounded finding. Three things this test does not rule out:

1. The Theis mechanism's geology-specific predictions. The leading mechanistic account for earthquake lights ties them to specific subbasalt geologies and to events with strong vertical stress changes (thrust and normal faults, not strike-slip). The historical reference cases concentrate in the European Rift, Mississippi Embayment, and Italian Apennines. Our test is unstratified by geology. A real signal confined to mechanism-favorable subsets would not be detected by a flat magnitude sweep.
2. Earthquake-induced reporting depression. The most parsimonious explanation for our depleted coincidence count is not a real "avoidance" phenomenon. Significant earthquakes cause local power outages, internet outages, and observer distraction. People in or near the disaster zone do not log on to file NUFORC reports in the hours during and after a quake. This depression of the post-event submission rate would mimic an avoidance pattern in our data regardless of whether earthquake lights are real. The test as designed cannot separate that confound from a true zero.
3. Sub-NUFORC-threshold phenomena. Some folklore accounts describe brief, faint, diffuse glows that a witness would notice but would not bother to report to a UFO database. Our test measures what the NUFORC submission population reports, not what every observer perceives.

In other words: the folklore's predicted footprint in this particular eyewitness-reporting network is not visible. The folklore's underlying physical mechanism is not ruled out by this test.

Power

The test is informative only against large folklore signals. At 80% statistical power, the smallest rate-ratio it can reliably detect is about 1.9. The folklore would have to produce roughly twice as many coincidences as chance to produce a result we could read in the data. Smaller effects are compatible with our numbers. The test is therefore well-suited to ruling out the strong form of the claim ("earthquakes cause lots of visible lights"), and not well-suited to ruling out a small effect.

Reproducibility

All scripts, extracted data files, the full 10,000-permutation null distribution, and the results JSON are in the TerraPulse repository at workspaces/earthquake-lights-nuforc-usgs/. ComCat data are public via the USGS FDSN web service. NUFORC data are public via the NUFORC archive. The random seed used for the permutation analysis (20260530) is recorded in the analyze.py script.

The full LaTeX paper, with the permutation null histogram, the magnitude-bin and sensitivity tables, and the references, is at paper/paper.pdf in the workspace.