Method and sources
Every number on this site is computed from public data at build time. This page explains how, and how far to trust it.
Sky darkness
We start from NASA's Black Marble annual night-lights product (VNP46A4, snow-free composite), which records how much artificial light leaves each half-kilometer of the ground. Light at a point is not the same as sky brightness there: the glow of a city spreads across the sky for tens of miles. So for every location we add up the light from every source within 200 km (124 miles), weighted by distance with a Walker-type law, brightness ∝ distance−β. We then convert that to an estimated sky quality meter reading (SQM, magnitudes per square arcsecond) with SQM = m0 − 2.5·log10(1 + k·Bγ), and to approximate Bortle classes.
The free parameters (the distance exponent β, a near-field softening length, the natural-sky brightness m0, the scale k and a curve exponent γ) are fitted to Globe at Night citizen-science SQM readings from 2021–2025, keeping only readings taken under a clear sky with the Sun more than 18° and the Moon more than 5° below the horizon. We measure the error with 5-fold spatial cross-validation (each reading is predicted by a model that never saw readings within its 2° block).
| Calibration readings | 1853 readings at 624 locations |
| Chosen distance exponent β | 2 |
| Cross-validated mean absolute error | 0.63 mag/arcsec² |
| Cross-validated RMS error | 1.09 mag/arcsec² |
| 80% of locations within | ±0.93 mag/arcsec² |
| Bortle class right, or within one class | 56% / 84% |
| Range shown on pages | estimate ± 0.95 |
Error by type of sky
| Sky at the reading | Locations | Mean error | Bias (model − meter) |
|---|---|---|---|
| <18.5 (city) | 148 | 1.25 mag | +1.14 mag |
| 18.5-20.3 (suburb) | 238 | 0.41 mag | -0.12 mag |
| 20.3-21.3 (rural) | 190 | 0.37 mag | -0.23 mag |
| >=21.3 (dark) | 48 | 0.84 mag | -0.83 mag |
Two biases, stated plainly. In cities the model reads too dark, by about 1.1 mag: many meters sit under a streetlight that a satellite looking down cannot see. At dark sites the model reads too bright, by about 0.8 mag: real skies there are darker than our estimate. Across all 1853 readings at 624 locations, 80% of locations are within ±0.93 mag, and every figure is published as the estimate ±0.95.
How we show it
- Stargazing spots (confirmed and unconfirmed): "≈ Bortle N (range A–B)", where N is the Bortle class of the central estimate and A–B the classes the published SQM range covers, shown next to that SQM range. Spots are rural or dark, where the model is most accurate (rural mean error 0.37 mag) and where its bias is conservative: it reads them too bright, so the real sky is usually as dark as shown or darker.
- County centers: the SQM range only, no Bortle class, because centers are often towns, where the error is largest. Where a center estimates brighter than SQM 19.5 we add that satellite estimates miss streetlights, so skies in town are usually brighter than the estimate.
Calibration data: Globe at Night, NOIRLab/NSF, licensed CC BY 4.0.
Limits: satellites see light going up, not light shining sideways; the sensor is not sensitive to the blue part of LED light; hills, trees, haze and snow are not modelled. The value for a county is for its center point.
Night access
A dark sky is no use if the gate is locked. We only recommend a place ("night access confirmed") when the managing agency's own page, which we fetched and keep a quote from, shows that the public can be there at night: a campground (overnight for registered campers), a certified dark-sky place, or a park with published night or 24-hour hours. A script re-fetches every page and checks the quote is still there; any that fail drop out. This build has 210 confirmed places across the five states.
Each confirmed place says what kind of night access its page shows, taken from that page, never guessed: "Overnight: registered campers only (campground)", "Open 24 hours", "Night use allowed (dark-sky park)", "Open into the night (published hours)" or "After-hours stargazing by permit". A campground page, or a park's name on a state campground reservation list, is treated as access for registered campers only, even where the page also says the campground is "open 24 hours". The quote from the page is shown next to each place.
Places on land that allows hunting (national forests, state forests and fish & wildlife areas in PAD-US, or where the official page says so) carry a hunting note linking to that state's official season dates.
Everything else, open-access federal and state land of 300 acres or more in PAD-US (national forests and parks, Army Corps lakes, state parks, forests and wildlife areas), is listed only as "public land nearby: check hours and hunting seasons before you go", always below the confirmed places. We leave out military land, wildlife refuges, nature preserves and local parks, most of which close at night.
When to go
New moons, moonrise and moonset, astronomical twilight (Sun 18° below the horizon), the Milky Way core (Sagittarius A*, at least 10° up during full darkness) and meteor-shower peak times are computed for each county center with astronomy-engine. Meteor peaks use the solar longitudes in the International Meteor Organization's calendars; 2028 dates are provisional.
Clear nights
Monthly mean night-time cloud amount from NASA POWER (CERES SYN1deg, 2001–2025). The data are on a 1°×1° grid, so every county in the same cell, typically about seven, shows the same numbers. Average cloud amount is not the same as the chance of a perfectly clear night, but the months rank the same way.
Distances
All distances are straight-line miles from the county's center. Driving distance is longer, sometimes much longer in hilly country.
Sources and credits
- NASA Black Marble VNP46A4 (annual night lights, Collection 2). NASA open data. Courtesy of NASA; no endorsement implied. Used for: Sky brightness model.
- Globe at Night observations, 2021–2025. Globe at Night, NOIRLab/NSF, licensed CC BY 4.0. Used for: Calibration of the sky brightness model.
- NASA POWER climatology API, CLOUD_AMT_NIGHT (CERES SYN1deg, 2001–2025). Data obtained from the NASA Langley Research Center (LaRC) POWER Project funded through the NASA Earth Science/Applied Science Program. Used for: Clear nights by month.
- USGS Protected Areas Database of the United States (PAD-US) 4.1. U.S. Geological Survey (USGS) Gap Analysis Project (GAP), 2025, PAD-US 4.1. Public domain. Used for: Public land and site locations.
- USGS GNIS Domestic Names. Public domain. Used for: Locations of a few lakes and features.
- U.S. Census Bureau 2024 Gazetteer (counties) and Geocoder. Public domain. Used for: County centers; two park addresses.
- DarkSky International, International Dark Sky Places. Facts (name, designation, year) and links only. Not affiliated with or endorsed by DarkSky International. Used for: Certified places.
- International Meteor Organization, Meteor Shower Calendar 2027 (and 2026). Dates restated as facts; peak times computed from IMO solar longitudes. Used for: Meteor shower peaks.
- astronomy-engine 2.1.19 by Don Cross. MIT License. Used for: Moon, twilight, Milky Way, shower peak times.
- timezone-boundary-builder (via geo-tz). Data © OpenStreetMap contributors, ODbL. Used for: Each county’s time zone.
- OpenStreetMap (Nominatim) and recreation.gov facility records. OpenStreetMap data © OpenStreetMap contributors, ODbL. Used for: Exact locations of 13 federal campgrounds whose official pages give no coordinates.
NASA POWER accessed 2026-11-16. DarkSky International list checked 2026-11-16. Pages built 2026-11-16.