How Yozora Atlas measures the dark
Every number on this site comes from public data and one fixed method. Nothing is written by hand. This page explains where the data comes from, how a satellite image becomes a sky reading, and how far you can trust the result.
What does a reading like 17.51 mean?
Each place gets one number: the brightness of the sky straight overhead on a clear night with no moon, in magnitudes per square arcsecond (mag/arcsec²). Astronomers count magnitudes backwards, so a bigger number means a darker sky. A sky with no artificial light at all reads about 22; this site uses 21.95. Glasgow reads 17.51. Each step of 1 is about 2.5 times brighter, so Glasgow's sky is about 60 times brighter than a natural one.
The Bortle class next to each reading turns the number into something you can picture, from 1 (the Milky Way casts shadows) to 9 (an inner-city sky).
Where does the light data come from?
From NASA's Black Marble product (VNP46A4): a yearly image of the Earth at night, taken by the VIIRS sensor on the Suomi NPP satellite and averaged over cloud-free, snow-free nights. Its pixels are about 500 meters across. The site uses the 2025 image for today's readings and the 2012, 2016 and 2020 images for the trend on each page.
VIIRS barely sees the blue part of white LED light. Towns that switched to LED streetlights can look dimmer from orbit than they are, so recent years in the trend probably read too dark.
How a satellite image becomes a sky reading
A satellite sees light going up. A stargazer cares about the light the air scatters back down. To get from one to the other, the light of every pixel within 250 km is added up, weighted so that nearby towns count far more than distant ones, with an extra fade over 100 km. The total is then scaled to the US National Park Service's sky glow model and turned into mag/arcsec², with 21.95 as the natural sky.
North of about 50 degrees the yearly NASA image has a faint background on every pixel, likely leftover aurora, snow and summer twilight. Added up over 250 km it would make the Arctic look like a suburb, so that background is removed in those northern map tiles only.
How accurate are the readings?
The model was checked against two references. Against the National Park Service's own sky glow model it ranks places in almost the same order (rank agreement 0.98, or 0.85 among dark places). Against 487 hand-held sky meter readings from the Globe at Night project in the US, 87 to 88% land in the right Bortle class or one class off.
In the UK the readings were compared with the meter readings that 12 dark-sky parks and communities published in their applications. On average the model reads about 0.1 too bright, which is within what a meter drifts. The darkest sites read too bright by more: Kielder by 0.45, Exmoor by 0.24.
A meter held under a streetlight can read 14 to 18 where the model says something darker. The model describes the sky above a place, not the glare on one street corner.
How are the darkest spots near each place chosen?
For each place the site looks for the darkest land at three distances: up to 50 km, 50 to 100 km and 100 to 150 km, in the same country. A dark hillside you can't get to is no use at night, so each of those points moves to the darkest car park, viewpoint, campsite, picnic site or trailhead from OpenStreetMap within 12 km (25 km in empty country), as long as its sky is at most 0.3 brighter.
Drive times come from OSRM route planning and start at the center of the place. A spot only counts as reachable if the route averages at least 30 km/h, is no more than 2.5 times the straight-line distance, and ends within 3 km of the spot. That rules out ferries, long detours around water and hikes into the backcountry. Before you go, check that the car park is open at night and that you're allowed to be there.
Parks work the other way round, because the park is where you want to be. Their pages list the darkest places you can drive to within about 60 km of the park's center, starting with the park's own coordinates. Drive times start at the nearest town you can drive to them from, out of the closest town of any size and the two closest with 5,000 people or more. Instead of the detour rule, park drives are capped at 5 hours. When no road reaches any of them, the page says so and links the nearest parks you can drive to.
Where do the clear-night odds come from?
From NASA POWER, which records cloud cover over every place on Earth each night. A night counts as clear when cloud covers less than 30% of the sky. The chart on each page is the average for each month from 2014 to 2023: what to expect in a typical year, not a forecast for a particular night.
How are the Milky Way dates calculated?
The positions of the sun, the moon and the center of the Milky Way are calculated for every night of 2027. A night counts when the bright core of the Milky Way stands more than 10 degrees above the horizon for at least an hour, while the sun is more than 18 degrees below the horizon and the moon has set. The best nights combine that with the clear-night odds for the month. Whether a particular night is clear is only known closer to the date.
How is the northern lights section worked out?
Every place where a strong storm (Kp 7) or less brings the aurora to the northern horizon gets one, from Alaska down to Toronto, Minneapolis and London. The aurora forms in a ring around the magnetic north pole, and the stronger a geomagnetic storm, the further south it reaches. Storm strength is measured by the Kp index, from 0 (quiet) to 9 (extreme). Each page gives the Kp level at which the aurora appears on the northern horizon there, using the place's magnetic latitude and the rule NOAA uses: the edge of the aurora sits at 66.5 degrees magnetic latitude at Kp 0 and about 2 degrees further south for each step. It can be off by one level, and a phone camera often shows a display the eye barely sees.
How often that level was reached comes from the Kp record of GFZ Potsdam, which goes back to 1932: the average number of nights a year with that level or more between 9 pm and 3 am local time, plus a quiet and an active year (the 10th and 90th percentile). The chart only counts nights when the sun gets more than 12 degrees below the horizon, which is why far northern places show no aurora nights in June. None of this is a forecast; for tonight, each page links to NOAA, Space Weather Canada or AuroraWatch UK.
Where do the statistics and the sky guides come from?
From the same numbers as the place pages, compared across every place. The statistics rank the towns by the quickest drive to a Bortle 4 sky, by how much brighter their sky got since the first satellite year, and every place by clear nights a year. The northern lights and Milky Way season guides gather the place sections by country and by state or province. Towns in these comparisons are the ones from GeoNames; the certified dark-sky communities count with the dark-sky places.
Which places get a page?
- 1013 towns and cities from GeoNames: 50,000 people or more in the US, Canada and the UK, 20,000 or more in Ireland, leaving out districts and suburbs within 8 km of a larger town
- 119 places certified by DarkSky International or, in Canada, the Royal Astronomical Society of Canada, from Wikidata
- 948 parks and protected landscapes: US national parks, monuments, forests, conservation areas and notable state parks from Wikidata, Canada's national parks and notable provincial parks from Wikidata (leaving out the Arctic parks no road leads to), the UK's national parks, National Landscapes and well-known dark landscapes, and Ireland's national parks and dark peninsulas and mountains
State and provincial parks and the added landscapes only get a page if the sky there is Bortle 4 or darker, and monuments, forests, conservation and recreation areas if it's Bortle 5 or darker.
The "darkest skies" rankings only compare places that have a page. Large parts of Alaska, Nevada, Utah and Montana are just as dark, often darker, but have no town or park to name them by. The same goes for the spots on each page: they are the darkest places you can drive to, and the roadless country around them can be darker still.
How often does the data change?
Once a year, when NASA publishes the next yearly night image. The Milky Way and moon dates move on to the next year at the same time.
Why is it called Yozora?
Yozora (夜空) is Japanese for "night sky".
Who makes Yozora Atlas?
Markus Green, in Nice, France. I build websites out of public datasets that are free to use but hard to read. A NASA night image is a grid of 5.7 million numbers per tile; what you want to know is whether the sky is dark an hour from your house. This site is the second step.
Sources and licenses
Night lights: NASA Black Marble VNP46A4 (NASA, no restrictions). Cloud cover: NASA POWER. Geomagnetic activity: Kp index, GFZ Helmholtz Centre for Geosciences (CC BY 4.0, Matzka et al. 2021). Places: GeoNames (CC BY 4.0), Wikidata (CC0). Parking areas, viewpoints, campsites and observatories: © OpenStreetMap contributors (ODbL). Drive times: OSRM. US state outlines: US Census Bureau via us-atlas. Sky glow scaling: US National Park Service. Checked against Globe at Night meter readings and published dark-sky applications.