An environmental analysis of public UAP sightings and sky view potential

Brief

Medina, Brewer, and Kirkpatrick (2023) apply Bayesian small area estimation to 98,724 publicly reported UAP sightings aggregated to U.S. county level, regressing Poisson count data against five environmental predictors: light pollution, cloud cover, tree canopy, airport density, and military installation area, all z-score transformed. Air traffic density returns a coefficient of 1.099 (95% CI: 1.068–1.131) with 100% posterior credibility of a positive relationship; light pollution returns 0.923 (CI: 0.899–0.947) with 100% credibility of a negative relationship, corresponding to a 7.7% drop in reports per one standard deviation increase. Cloud cover produces no credible signal (48/52% credibility split). A Getis-Ord Gi* hotspot analysis places population-standardized sighting clusters in the western U.S. and northeast, with cold spots across the central plains and southeast.

Metadata

Category

Phenomenon

Venue

Scientific Reports

Type

Peer-reviewed

Year

2023

Authors

R. M. Medina, S. C. Brewer, S. M. Kirkpatrick

DOI

10.1038/s41598-023-49527-x

Access

Open access

Length

1.6 M

Programs

NUFORC, AARO, UAP Task Force, AOIMSG, Project Blue Book, Project Sign, Project Grudge

Instruments

MODIS (EarthEnv cloud cover), Landsat (USFS tree canopy), Getis-Ord Gi* hotspot analysis, INLA (Integrated Nested Laplacian Approximation)

Data sources

NUFORC, New World Atlas of Artificial Sky Brightness, EarthEnv Project (MODIS 2000–2014 cloud cover), USFS Multi-Resolution Land Characteristics Consortium tree canopy, ESRI ArcGIS airports dataset (19,850 entries), U.S. Census TIGER/Line military installation shapefiles

Tags

UAP-reports, geospatial-analysis, Bayesian-modeling, crowdsourced-data, airspace-density, light-pollution

Key points

• Dataset: 98,724 NUFORC reports (2001–2020) geocoded to U.S. county level, drawn from a full extract of 122,983 sightings spanning June 1930–June 2022; model estimated via INLA (Integrated Nested Laplacian Approximation) for computational efficiency.p.3
• Air traffic density is the strongest predictor: coefficient 1.099 (95% CI: 1.068–1.131), 100% posterior credibility of a positive relationship with sighting rates.p.5
• Light pollution coefficient: 0.923 (95% CI: 0.899–0.947), 100% credibility of negative relationship; one standard deviation increase in light pollution associated with a 7.7% decrease in sighting reports.p.5
• Tree canopy coefficient: 0.961 (95% CI: 0.915–1.010), 94% credibility of negative relationship, denser canopy correlates with fewer reports.p.5
• Military installation area coefficient: 1.013 (95% CI: 0.994–1.033), 92% credibility of positive relationship.p.5
• Cloud cover shows no credible relationship (coefficient 0.998; 48% positive / 52% negative credibility), contradicting the authors' initial hypothesis.p.5
• Hotspot analysis (Getis-Ord Gi*, k=8 nearest neighbors) identifies elevated per-capita sighting clusters in the western U.S. and northeast; cold spots concentrated in the central plains and southeast.p.5
• Co-authored by Sean M. Kirkpatrick (U.S. Department of Defense), who served as founding director of AARO, the first peer-reviewed UAP geospatial study to include a sitting DoD UAP office head as author.p.1

Verbatim

• “The model results find credible correlations between variables that suggest people see more "phenomena" when they have more opportunity to.”

  p.1

• “the coefficient for Mean Light Pollution is 0.923, indicating that a one standard deviation increase in light pollution will result in a 7.7% decrease in sighting reports.”

  p.5

• “it is impossible to discredit over 120,000 cases.”

  p.4

• “we posit that this dataset has value in understanding these sighting reports; that either this indicates people are seeing things they can't explain (or that they don't want to explain with more logical explanations), or this indicates where people are thinking more about UAPs.”

  p.6

Most interesting

• Cloud cover, the variable most intuitively expected to suppress sightings, showed zero credible relationship with report rates, likely because heavily clouded Pacific Northwest counties and clear-sky Mountain West desert counties both produce elevated counts, canceling the signal.
• NUFORC's full archive spans June 1930 to June 2022 and contains 122,983 U.S. sightings; the study restricts to post-2001 data partly because widespread internet access to file reports only emerged around 2000, creating a reporting-bias floor before that date.
• The 2021 DNI preliminary assessment left all but one of its 144 government-sourced UAP reports unexplained; the single resolved case was identified with high confidence as a deflating balloon.
• All five predictor Variance Inflation Factors fell below 2, well under the conventional multicollinearity threshold of 5, ruling out collinearity as a confound in the model.
• Satellites (including SpaceX Starlink) and drone activity were explicitly excluded as predictors despite being acknowledged as likely important factors, leaving a known gap in the model's coverage.
• The authors propose that elevated western U.S. sighting rates reflect a combination of physical geography (sparse canopy, open terrain), outdoor recreation culture, and cultural priming from Area 51 and Roswell, a rare instance of a peer-reviewed study treating paranormal ideation as a quantifiable spatial variable.