NUFORC Cross-Correlation: Filtering UFO Sightings Against Known Phenomena

Author: Claude (TerraPulse Lab)

Status: Complete

Created: 2026-04-03

GitHub Issue: #83

Hypothesis

A significant fraction of NUFORC UFO sightings can be attributed to known phenomena (self-classified fireballs, CNEOS bolides, and geomagnetic storms/aurora). The unexplained residual will show systematic patterns revealing observer bias.

Data Sources

Source	Records	Span
NUFORC sightings	80,331 (with coordinates)	1906–2014
CNEOS fireballs	391 (with coordinates)	1998–2026
Dst index (CDAWeb)	12,651 daily values	1991–2026

Findings

Filter cascade: 24.1% attributed, 75.9% unexplained

All percentages use total sightings (N = 80,331) as the denominator. Counts are exclusive (overlap removed).

Filter (exclusive)	Attributed	% of total
Self-classified "fireball"	4,946	6.2%
CNEOS fireball match (±1hr, ±500km)	0	0.0%
Geomagnetic storm (Dst < -50, ±1 day)	13,134	16.3%
Overlap (fireball intersect storm)	1,262	1.6%
Unexplained	60,989	75.9%

Gross counts before overlap removal: 6,208 fireballs (7.7%), 14,396 storm matches (19.5% of 73,675 with Dst coverage).

Key finding: CNEOS is too sparse to explain individual sightings

Zero matches at standard tolerance (±1hr, ±500km). Only two matches at loose tolerance (±2hr, ±1000km). CNEOS detects ~20 bolides/year, far too few to correlate with ~6,000 annual NUFORC reports. The sightings and the bolides measure different populations.

Key finding: 19.5% of Dst-covered sightings coincide with geomagnetic storms

14,396 sightings (19.5% of 73,675 with Dst coverage; 16.3% of all 80,331 after overlap removal) occur within ±1 day of Dst < -50 nT. However, the expected random match rate is 31.2% (11.7% of Dst-record days have Dst < -50, and the ±1-day window triples the effective window). The observed rate is below this baseline, meaning the match reflects storm-day frequency rather than a causal excess. Sensitivity: Dst < -30 captures 32,163 (43.6%), Dst < -100 captures 3,220 (4.4%).

Shape analysis: "fireball" is perfectly depleted from unexplained set

The self-classification filter works: "fireball" shape goes from 7.7% of all sightings to 0.0% of unexplained; every fireball-shaped sighting was removed by definition. All other shapes (light, triangle, circle, sphere, disk) show enrichment factors of 1.07-1.12, a mechanical consequence of fireball removal. A chi-squared test on the non-fireball shape distributions confirms no significant difference: chi2(12) = 6.7, p = 0.88 (N = 54,984). Sightings with no reported shape (2.4%) show enrichment of 1.06, consistent with other non-fireball shapes. (Note: an earlier version of the analysis incorrectly reported null-shape sightings as fully depleted due to a Polars null-handling bug; this has been fixed.)

Null test: strong summer seasonal pattern

chi2(11) = 3,466 (days-per-month null), p < 10^-15, N = 60,989. July peak (7,287) vs February trough (3,587) , a 2:1 ratio. The days-per-month correction (vs uniform: chi2 = 3,646) has no effect on significance. This is consistent with observer availability bias (more people outdoors in summer evenings) rather than a physical phenomenon.

Visualizations

Filter cascade waterfall - attribution breakdown
Shape comparison - all sightings vs unexplained

References

NUFORC - National UFO Reporting Center, https://nuforc.org/
NASA CNEOS Fireball Database, https://cneos.jpl.nasa.gov/fireballs/
CDAWeb OMNI Dst index, NASA GSFC, https://cdaweb.gsfc.nasa.gov/