The Earthquake-Lights Signal That Isn't: Two Big Quakes, a Double-Counting Trap, and 104 Years of UFO Reports

Dek. A naive coincidence count of 104 years of US earthquakes against NUFORC UFO reports comes out sharply positive — 2.9× chance, p = 0.0002, with the magnitude dependence the folklore predicts. It's a mirage: the excess is two California mega-quakes (Loma Prieta and Northridge) and the same handful of reports counted once for every aftershock. Count each report once, or set those two quakes aside, and it collapses to chance (rate-ratio ≈ 1.0). Across 104 years, US earthquakes produce no population-scale sky-anomaly signal.

Workspace slug: earthquake-lights-nuforc-usgs Status: Complete. Date: 2026-05-30.

---

Abstract

The folklore claim known as "earthquake lights" — visible luminous phenomena (flashes, glows, ball lightning, columns of light) appearing in the hours before or during significant earthquakes — has been recorded in seismology literature since at least the 1965 Niigata event and continues to be cited in popular media. If the claim holds at population scale and at the unstratified level of an open eyewitness-reporting network, NUFORC reports of sky anomalies should cluster in space and time around earthquake epicenters at a rate above chance.

We cross-matched the US M ≥ 4.5 earthquake catalog (ANSS ComCat, 1910–2014, n = 9,379 after deduplication) against the US-resident NUFORC UFO sighting database (n = 70,822 in joint window) using a 100 km spatial radius and a −24 h / +1 h temporal window around each epicenter, comparing observed coincidences to a permutation null that shuffles UFO timestamps while preserving their spatial distribution.

The naive coincidence count is strikingly positive: 38 observed against a null mean of 13.0, a rate-ratio of 2.93 (95% CI [1.58, 7.60]), p = 0.0002 — and it rises monotonically with magnitude, from 2.6× at M 4.5–5.0 to 7.3× at M ≥ 6, exactly the pattern the folklore mechanism predicts. Taken at face value this would be a population-scale confirmation of earthquake lights.

It is an artifact. The excess is produced by two mechanisms, both unrelated to any luminous phenomenon. First, aftershock multiplicity: the coincidence statistic counts (quake, report) pairs, so a single UFO report near a mainshock is counted again against each of that mainshock's aftershocks. The 38 pairs resolve to just 14 distinct reports, and counting each report once erases significance (rate-ratio 1.26, p = 0.22). Second, two mega-quakes: the 1989 Loma Prieta (M 6.9) and 1994 Northridge (M 6.7) sequences, both in dense California metros, dominate the count; removing those two sequences drops the pair-count rate-ratio to 1.00 (p = 0.52) and the distinct-report rate-ratio to 1.01 (p = 0.53).

We conclude that the public record does not support the earthquake lights folklore in the specific form "US M ≥ 4.5 earthquakes produce sky anomalies that the NUFORC submission population would notice and submit at a population-level rate above chance." The apparent signal is a textbook confound trap: real quake–report co-location (California has many of both), amplified by aftershock double-counting and two famous events, with no excess in the confound-controlled statistic. The test does not address the geology-specific predictions of the leading proposed mechanism (Theis et al. 2014), sub-NUFORC-threshold luminous phenomena, or reporting behavior specific to individual disasters.

Introduction

Earthquake lights are an old folklore claim with a small but persistent scientific literature. Theis et al. (2014) catalogued 65 historical accounts back to 1600, and proposed a piezoelectric mechanism linking subbasalt stress to atmospheric ionization. The mechanism remains contested and the empirical base — historical eyewitness accounts at named events — has not, to our knowledge, been tested against a population-level cross-match.

The TerraPulse platform stores both layers needed for such a test:

• NUFORC (National UFO Reporting Center): a self-reported sighting archive, 1906–2014, of which 70,854 are US-located with coordinates and a US state assignment.
• USGS Earthquake Catalog: the US M ≥ 4.5 event list from the early 20th century onward, held in the TerraPulse eq dex.

If the folklore signal is real at the magnitude that would produce population-level reports, NUFORC submissions should be enriched within a small spatial radius of and a few hours before significant earthquakes, relative to a null that preserves the spatial and temporal distribution of UFO reports under random matching. As we show, testing that prediction naively produces a positive result — and diagnosing why is the substance of this paper.

Data

NUFORC

National UFO Reporting Center reports, read from the TerraPulse nuforc dex through the platform provenance registry (source nuforc, metric ufo_sighting). Each record carries an alleged-sighting timestamp_utc, latitude, longitude, a US state code, and a free-text description. The submission timestamp is stored separately in the record metadata and is not used in this analysis.

US reports are identified by NUFORC's own two-letter state assignment (the 50 states plus DC and territories), guarded against foreign reports whose state code collides with a US code (an explicit non-US country is excluded) and against geocoding errors (the coordinate must fall inside the US bounding box, latitude 18°–72° N, longitude −180° to −65° E). This yields 70,854 US reports, 1910–2014.

NUFORC volume is heavily concentrated post-1995, reflecting the growth of the online submission system and the population-level uptake of UFO reporting. Pre-1990 volume is sparse; pre-1970 is anecdotal.

USGS earthquake catalog

ANSS ComCat (Advanced National Seismic System Comprehensive Catalog), events between 1906-11-10 and 2014-05-08, magnitude ≥ 4.5, within a bounding box covering the United States (latitude 18°–72° N, longitude −180° to −65° E), post-filtered to US territory via the catalog's place field. The seismic layer is read from the TerraPulse eq dex through the provenance registry (source usgs_earthquake, metric earthquake_magnitude). We apply the canonical TerraPulse earthquake deduplication: round timestamp_utc to the nearest minute, group by (time, latitude, longitude, magnitude), keep first.

Final catalog: n = 9,379 US M ≥ 4.5 events, 1906-11-15 to 2014-05-05.

Joint window

The intersection of both layers is 1910-01-01 to 2014-05-05, a span of 104.3 years. After filtering, n_quakes = 9,379, n_ufo = 70,822.

Method

Spatial-temporal cross-match

For each earthquake k, let N_k be the set of NUFORC reports within 100 km of the epicenter (great-circle distance). For each report u ∈ N_k, the pair (k, u) is a coincidence if

t_q − 24 h ≤ ts(u) ≤ t_q + 1 h

where t_q is the earthquake time and ts(u) is the alleged-sighting time. The window is asymmetric (24 h before, 1 h after) because the folklore predicts pre-event luminosity; we cap the after-window at 1 h to capture syn-event reports while excluding rubbernecking and news-cycle bias. Spatial neighbours are computed with a BallTree on the haversine metric; the 808,528 spatial pairs (report within 100 km of an epicenter) are enumerated as a flat table.

Permutation null

The null preserves the spatial distribution of UFO reports — population density, urban concentration, reporting-network bias — by holding each report's coordinates fixed and shuffling the UFO timestamp vector globally, recomputing the coincidence count. We repeat this 10,000 times (numpy RNG seeded at 20260530) and report the rate-ratio (observed / null mean, 95% CI from the 2.5th–97.5th null percentiles) and a permutation p-value (fraction of permutations ≥ observed, plus one; Phipson and Smyth 2010).

Two statistics: pairs vs distinct reports

The pair-count statistic above has a known vulnerability: aftershock multiplicity. A single UFO report near a mainshock is re-counted against every aftershock in that mainshock's sequence, inflating the count without adding independent evidence. We therefore also compute a distinct-report count — the number of distinct UFO reports that fall within 100 km and −24 h / +1 h of at least one qualifying quake, so a report near a mainshock plus six aftershocks is counted once, not seven times. The permutation null recomputes the same distinct-report count under shuffled timestamps, keeping the comparison valid.

Mega-quake exclusion

To test whether the count is dominated by individual events, we re-run both statistics after removing the two largest, most-reported sequences: all quakes within 7 days and 150 km of the Loma Prieta (1989-10-18, M 6.9) and Northridge (1994-01-17, M 6.7) mainshocks (30 events).

Results

The apparent signal

The naive pair count is strongly positive:

• Observed coincidences across the 104 y joint window: 38.
• Permutation null mean: 12.98 ± 4.79 (95% CI [5, 24]).
• Rate-ratio: 2.93 (95% CI [1.58, 7.60]).
• p = 0.0002.

And it strengthens with magnitude — the folklore-predicted signature:

Magnitude bin	n_quakes	Observed	Null mean	Rate-ratio	p
4.5 – 5.0	5,479	24	9.25	2.59	0.0013
5.0 – 5.5	1,982	7	2.51	2.79	0.0180
5.5 – 6.0	1,162	4	0.80	4.98	0.0128
6.0 +	756	3	0.41	7.33	0.0081

At face value this is a population-scale confirmation of earthquake lights: a significant excess of sky-anomaly reports near epicenters, growing from 2.6× at moderate magnitude to 7.3× for the great earthquakes the folklore most often invokes. The result is also robust to the window and magnitude-floor choices (see Sensitivity). It is, nonetheless, an artifact — and the rest of the Results section dismantles it.

The signal is a double-counting artifact

The pair count treats each (quake, report) pair as independent evidence. It is not. The 38 coincidence pairs resolve to just 14 distinct UFO reports across 14 distinct quake-days. Counting each report once:

• Distinct-report observed: 14.
• Null mean: 11.15 ± 3.33 (95% CI [5, 18]).
• Rate-ratio: 1.26; p = 0.22.

The moment aftershock multiplicity is removed, significance disappears. The pair-count inflation is concentrated on two days: 1989-10-18 contributes 14 pairs across 7 quakes (the Loma Prieta mainshock and its aftershocks) but only a few distinct San Jose reports; 1994-01-17 contributes 11 pairs across 11 quakes (Northridge) from a similarly small set of reports.

The signal is two earthquakes

Geographic and temporal concentration confirm it. Of the 38 coincidence pairs, 33 are in California; by year, 1989 (Loma Prieta, 14) and 1994 (Northridge, 12) account for 26 of 38. Removing those two sequences (30 quakes):

Statistic	Observed	Null mean	Rate-ratio	p
Pair count (full catalog)	38	12.98	2.93	0.0002
Pair count (excl. mega-quakes)	12	11.95	1.00	0.52
Distinct reports (full catalog)	14	11.15	1.26	0.22
Distinct reports (excl. mega-quakes)	11	10.87	1.01	0.53

Excluding Loma Prieta and Northridge, the pair-count rate-ratio falls from 2.93 to 1.00 — the entire apparent signal is those two events. The distinct-report rate-ratio, already null, stays flat at 1.01. Neither confound control leaves anything behind.

Sensitivity

We repeated the pair-count test across five window/floor variants to confirm the apparent signal is not a tuning artifact — it is not; the co-location is real, which is precisely why the confound controls above are the load-bearing analysis rather than the parameter sweep.

Variant	n_quakes	Observed	Null mean	Rate-ratio	p
Baseline (−24 h / +1 h, M ≥ 4.5)	9,379	38	12.98	2.93	0.0002
Wider window (−48 h / +1 h)	9,379	63	25.55	2.47	0.0001
Tighter window (−6 h / +1 h)	9,379	19	3.61	5.26	0.0004
Lower floor (M ≥ 3.5)	32,428	310	136.83	2.27	0.0003
Higher floor (M ≥ 5.5)	1,918	7	1.21	5.76	0.0016
1995–2014 only	3,050	8	9.11	0.88	0.68

The naive positive persists across every window and floor choice — because California genuinely hosts many earthquakes and many UFO reports. There is one exception, and it is the tell: the 1995–2014 sub-window is null (rate-ratio 0.88, p = 0.68). That sub-window is the only variant that excludes both Loma Prieta (1989) and Northridge (1994), and it is the only variant with no signal — the same result the explicit mega-quake exclusion gives, arrived at by cutting on time. The apparent signal is entirely a pre-1995, two-event phenomenon; the modern era that carries the bulk of NUFORC's reporting volume shows nothing.

Discussion

What this rules out

The confound-controlled analysis rules out a folklore signal large enough to survive as an excess of distinct reports near US M ≥ 4.5 earthquakes. Counting each report once, observed (14) is statistically indistinguishable from the permutation null (11.15; p = 0.22), and with the two mega-quake sequences removed the record is a textbook null (rate-ratio 1.00–1.01). Within the limits below, the public record does not support the claim that visible luminous phenomena precede or accompany US earthquakes at a rate detectable by the NUFORC reporting network.

Why the naive count misleads

The pair-count positive is a clean worked example of how spatiotemporal cross-matching manufactures spurious "confirmations."

1. Aftershock multiplicity. Large earthquakes come with dense aftershock sequences co-located with the mainshock. A single report near the mainshock is re-counted against every aftershock, multiplying one piece of evidence into many. The permutation null shares the pair structure, so the p-value is internally valid — but the effective sample size is the number of distinct reports (14), not the number of pairs (38). Reporting only the pair count overstates the evidence by nearly 3×.
2. Event concentration. Two California mega-quakes in dense metros, each among the most-reported earthquakes in US history, supply the bulk of the count. A test that is powered only by two events is not a population-scale test; it is two anecdotes with a p-value.
3. Reporting behavior. Loma Prieta struck during a live World Series broadcast; Northridge levelled parts of Los Angeles. Both drove intense media coverage, power outages, transformer arcs, and heightened public attention to the sky. A UFO report in San Jose the evening of Loma Prieta is as consistent with disaster-driven reporting and reinterpretation as with a luminous geophysical phenomenon. The permutation null preserves spatial structure but does not preserve event-conditional reporting surges.

What this does not rule out

1. The Theis-mechanism prediction is not directly tested. The leading mechanistic account (Theis et al. 2014) ties earthquake lights to specific geological substrate (p-hole-rich subbasalt, rift environments) and fault geometry. Our test is unstratified by geology or fault type; a real signal confined to mechanism-favorable subsets would not be detectable here.
2. Sub-NUFORC-threshold phenomena. The folklore literature includes brief, faint, diffuse glows that would not provoke a NUFORC submission. Our test addresses what NUFORC's submission population reports, not what every observer perceives.
3. Non-NUFORC observer populations. NUFORC reports are filtered through self-identification as a UFO observer, and through a US-language reporting culture.
4. Individual events. This paper argues the population-scale claim fails. It does not claim that no luminous phenomenon occurred at any specific earthquake; Loma Prieta in particular has a genuine eyewitness-lights literature. It argues only that such events are too few and too confounded to establish a population-level rate above chance.

Power

Under the distinct-report null (mean 11.15, std 3.33), the minimum detectable rate-ratio at 80% power (one-sided, α = 0.05) is approximately (11.15 + 2.49 × 3.33) / 11.15 ≈ 1.74. The test is well-powered against a large folklore signal and poorly powered below a ~50% enrichment. The observed distinct-report rate-ratio (1.26) sits comfortably inside the null.

Caveats

• NUFORC volume non-uniformity. Submission volume is concentrated in 1995–2014; most spatial-pair opportunities come from this sub-window.
• NUFORC timestamp precision. Sighting times are minute-precise for some submissions and rounded for others; the 25 h window is robust to this, a tighter follow-up would not be.
• US-only. The result is specific to NUFORC's US reporting culture.

Comparison to prior TerraPulse work

This is the second TerraPulse cross-match to find that anomalous-witness reports do not corroborate an underlying geophysical claim. The Skinwalker Ranch cross-match (workspaces/skinwalker-ranch-baseline/cross-match) found 7 public sensor layers stay quiet across 10 filming-day pins. Both cut against the family of claims that anomalous-witness reports track underlying geophysical events — and both illustrate that the honest test is the confound-controlled one, not the first number that clears p < 0.05.

Conclusion

We tested one population-level prediction of the earthquake lights folklore: that significant US earthquakes should produce eyewitness sky-anomaly reports clustered around their epicenters at a rate above chance, as reflected in the NUFORC submission record. A naive coincidence count says yes — 2.9× chance, p = 0.0002, with a magnitude dependence that mimics the mechanism. The confound-controlled analysis says no: the excess is 14 distinct reports inflated to 38 pairs by aftershock multiplicity and dominated by two California mega-quakes; counting reports once gives a rate-ratio of 1.26 (p = 0.22), and excluding the two events gives 1.00 (p = 0.52).

This negative result is bounded. It rules out a large-effect-size folklore signal in the NUFORC submission population, at the population-aggregated scale, without geological stratification. It does not test the Theis et al. (2014) mechanism's geology-specific predictions; it does not address sub-NUFORC-threshold luminous phenomena; and it does not claim that no luminous phenomenon occurred at any individual earthquake. What it demonstrates cleanly is that the population-scale earthquake-lights claim survives only as long as the analysis does not control for the two most elementary confounds in event cross-matching.

Reproducibility

• Scripts: scripts/extract_nuforc.py (reads the nuforc dex), scripts/extract_quakes.py (reads the eq dex via read_quakes_from_dex.py), scripts/analyze.py (pair-count headline), scripts/reanalyze_confound.py (distinct-report + mega-quake exclusion), scripts/sensitivity.py.
• Data outputs: data/ufo_us.parquet, data/quakes_us_m45.parquet, data/coincidences_observed.parquet, data/permutation_null.parquet, data/results.json, data/reanalysis_confound.json, data/sensitivity.json.
• Both layers originate from the TerraPulse dexes (nuforc, eq) through the provenance registry.
• RNG seed: 20260530.

References

1. Theis, F., et al. (2014). Prevalence of earthquake lights associated with rift environments. Seismological Research Letters 85 (1): 159–178.
2. Phipson, B., and Smyth, G. K. (2010). Permutation P-values should never be zero. Statistical Applications in Genetics and Molecular Biology 9 (1).
3. ANSS Comprehensive Catalog. U.S. Geological Survey. https://earthquake.usgs.gov/data/comcat/
4. National UFO Reporting Center. https://www.nuforc.org/