The Night Sky Is Getting 9.6 Percent Brighter Every Year, and 51,351 Naked-Eye Observations Say So

At 19,262 places around the world where people walked outside and counted stars, the satellite that maps Earth at night recorded the ground below getting slightly darker. Across North America, upward radiance measured at those exact sites by the VIIRS Day/Night Band fell 0.80 percent a year. Over those same locations, artificial sky brightness was rising 10.4 percent a year.

Both numbers come out of the same study. The gap between them runs to about 11 percentage points a year, and it is an artifact of what the two methods can and cannot see. In January 2023, Christopher Kyba and three coauthors published in Science the result of running 51,351 naked-eye observations through a single model of how skyglow has changed. The global figure was an increase of 9.6 percent per year in artificial sky brightness from 2011 to 2022. Satellite studies over overlapping periods had been reporting something closer to 2 percent globally, and at the observers' own locations they were reporting a small decline. The eyes were registering something the radiometers structurally could not.

What 51,351 Star Counts Actually Measured

The data came from Globe at Night, a citizen-science program run by NSF's NOIRLab and developed by NOIRLab astronomer Connie Walker. The mechanics are almost aggressively simple. During a campaign window, you go outside more than an hour after sunset with the Moon down, let your eyes adapt for ten minutes, find the target constellation, and pick which of eight star charts matches what you can actually see. That choice is an estimate of naked-eye limiting magnitude, the faintest star you can detect. As the background glows brighter, the faint stars go first, so the number encodes sky brightness whether the observer thinks about it that way or not.

Kyba's team screened the archive hard before fitting anything, throwing out entries contaminated by twilight, moonlight, or snow on the ground, which cut 83,989 cloud-free reports to 51,351 usable ones. Binning those by month and by the atlas's 30-arcsecond grid left roughly 29,000 data points driving the trend fit. The model carries five free parameters, including the annual rate of change in artificial skyglow and the rate at which participants simply make mistakes. That error term landed between 1.7 and 2.5 percent depending on the region, which is a reassuringly small number for a dataset built by strangers on phones in the dark.

The regional results split sharply. Europe came in at 6.5 percent per year, North America at 10.4 percent, the remaining continents pooled at 7.7 percent, and the global average at 9.6 percent with a stated uncertainty of 0.4. That global average is weighted by where the participants are, and only Europe and North America had enough spread across the land area and across the decade to stand on their own; the authors say the other continents should not be read as representative. The team then checked itself with a second, cruder method: fit a straight line to the monthly gap between the limiting magnitude people reported and the magnitude a static sky would predict. Under the assumption that skyglow held constant, that gap drifted downward at 0.044 magnitude per year. Correct the model for a 9.6 percent annual increase, and the drift vanishes, sitting at 0.001 magnitude per year. In the NOIRLab release announcing the paper, Kyba put the consequence this way: a child born where 250 stars are visible would see about 100 by their eighteenth birthday.

Why the Satellites Saw the Opposite

Only one kind of instrument monitors the whole Earth at night, and it is blind to the part of the spectrum that matters most here. The VIIRS Day/Night Band cannot detect light below about 500 nanometers. White LEDs, the technology that took over new outdoor lighting during the 2010s, put their emission peak between 400 and 500 nanometers. The satellite is looking straight past the blue spike.

That swap happened fast. Kyba's paper cites global LED market share for new general lighting rising from under 1 percent in 2011 to 47 percent in 2019, with LEDs taking 66 percent of new outdoor lighting in the United States by 2020. When a city converts its streetlights to LED and keeps the same perceived brightness, the radiance the satellite records can actually go down, because the energy moved into a band the sensor does not count. That is one reason the Day/Night Band saw North American observing sites dimming by 0.80 percent a year while the people standing at those sites lost stars.

Orientation compounds the problem. Satellites measure light headed up. Skyglow is dominated by light headed sideways, out of windows, off signs, across parking lots and building faces, because a photon launched toward the horizon crosses roughly an order of magnitude more atmosphere before it escapes and has far more chance to scatter back down. In the early evening, when most people are actually outside looking up, a large fraction of the light escaping a city is not streetlighting at all. "This shows that existing satellites aren't sufficient to study how Earth's night is changing," Kyba said in the NOIRLab announcement.

Blue Light Scatters Almost Three Times as Well

Rayleigh scattering by air molecules varies as the inverse fourth power of wavelength, which is the same relationship that makes a clear daytime sky blue. Run the numbers on the specific swap that happened over the last fifteen years, and the result is stark. A low-pressure sodium lamp emits essentially at the sodium doublet, 589.0 and 589.6 nanometers, and the high-pressure sodium lamps that lit most streets before the changeover were broader but still piled into the amber. A white LED's blue pump sits near 450. By the inverse-fourth-power rule, that blue photon is about 2.9 times more likely to be scattered by the atmosphere than the amber one. Real lamps are broadband and lumens are weighted by daytime eye response, so that single ratio is not the whole calculation, but it points the same direction the full spectral models do: a city that keeps its lumen output flat and only changes the color of its lamps still pushes more light back down onto itself.

Then there is the eye. Human night vision runs on rods, whose spectral sensitivity peaks near 507 nanometers rather than the 555 nanometers of daylight vision. The shift moves your peak sensitivity toward exactly the wavelengths that scatter best and that the satellite cannot see. "Since human eyes are more sensitive to these shorter wavelengths at nighttime, LED lights have a strong effect on our perception of sky brightness," Kyba said.

Weather sets the gain on all of it. Aerosol scattering is close to wavelength-neutral and strongly forward-peaked, and it cuts both ways. More haze scatters more light back down near its source, so the dome over the city itself burns brighter. The same particles also take light out of the beam as it travels, so the glow carries less far. The counterintuitive result is that a crisp dry night delivers a distant city's dome to a rural site more effectively than a humid one does, even though the humid night looks worse from inside town. Cloud is the extreme case. Kyba measured this directly in Berlin more than a decade ago, finding overcast amplified sky luminance by a factor of 10.1 inside the city and 2.8 at a site 32 kilometers out. Inside the city, an overcast night ran 4.1 times brighter than a clear moonlit night in the countryside. If you have ever driven toward a city under a low deck and watched the base of the clouds turn amber, you have watched the amplifier work. Over genuinely dark ground, the same cloud does the opposite and makes the night darker, because there is nothing below it to reflect.

The Bortle Arithmetic, and What It Isn't

The popular version of this paper, repeated in a lot of stargazing coverage since 2023, is that your dark site has about a decade before it drops a Bortle class. The arithmetic behind that claim is real, and it is easy to reproduce. John Bortle's ladder, laid out in his 2001 Sky & Telescope article, steps in half-magnitude increments through the lower classes: Class 3, rural sky, spans a limiting magnitude of 6.6 to 7.0, and Class 4, rural-suburban transition, spans 6.1 to 6.5. Take the 0.044 magnitude per year that Globe at Night participants were actually losing, divide half a magnitude by it, and you get about eleven years, or roughly ten to fourteen once you carry the paper's error bar. One decade, one class.

The paper does not say that, and the projection is mine rather than a finding. It also does not survive contact with a genuinely dark site. That 0.044 magnitude per year is an average across a participant pool that clusters in residential areas, where artificial light already dominates the sky background. The 9.6 percent figure describes growth in the artificial component only, not in the total. Natural airglow, zodiacal light, and unresolved starlight do not grow at all.

Compound it yourself, and the difference is enormous. Hold the natural background fixed, grow only the artificial part at 9.6 percent a year, and ask how long the total takes to climb half a magnitude. Where artificial glow already equals the natural background, roughly a suburban-edge sky, the answer is about eight and a half years. Where it adds only a tenth on top of natural, which is barely past the 8 percent excess that Falchi's atlas uses as the line between a light-polluted sky and a clean one, the answer is about 22 years. And even that overstates the visible damage, because limiting magnitude and sky brightness do not move one for one; the paper's own fitted slope puts about 1.43 magnitudes of limiting magnitude on a full factor-of-ten change in sky radiance. Both of those calculations are mine, done from the published rate and stated assumptions, not results the authors reported.

Bortle had something to say about the metric in the first place. He warned in that same 2001 piece that naked-eye limiting magnitude "is a poor criterion. It depends too much on a person's visual acuity (sharpness of eyesight), as well as on the time and effort expended to see the faintest possible stars." He was right about any single observer, which is precisely why Kyba's method does not use one. It aggregates tens of thousands of estimates, fits an explicit error rate, and corrects for the fact that the pool of participants changes from year to year. Bortle's objection is a limit on your personal site rating, not on a continental trend.

The Map and the Clock

Kyba's study did not measure sky brightness from scratch. It measured change against a fixed spatial baseline, and that baseline is the 2016 world atlas of artificial night sky brightness by Fabio Falchi and colleagues, evaluated for 2014. The atlas modeled the propagation of artificial light through the atmosphere using high-resolution satellite data and ground photometry: more than 80 percent of the world's population and more than 99 percent of the United States and European populations live under light-polluted skies, and the Milky Way is invisible to more than a third of humanity, including 60 percent of Europeans and close to 80 percent of North Americans.

The two papers do different jobs, and they are most useful together. Falchi's atlas is the map, a snapshot of where the glow sits. Kyba's is the clock, telling you how fast that snapshot goes stale. A map built on 2014 conditions and a rate of 9.6 percent per year are not two independent facts about the sky; the second one is the expiration date on the first. Falchi, for his part, coauthored a commentary that ran in the same issue of Science, arguing that the results demonstrate the need for a nighttime satellite with real multiband capability in the visible.

What Filters and Longer Exposures Buy Back

Sodium lighting was, from a filter designer's point of view, a gift. Low-pressure sodium is nearly monochromatic, and high-pressure sodium, while genuinely broader, still concentrates most of its output into pressure-broadened lines in the amber, so a notch filter could cut the offending region at modest cost to everything else. That is why those filters worked so well in the 1990s and 2000s that they felt like cheating. White LED offers no such target. A blue pump feeding a broadband phosphor produces a continuum across the visible band, and a filter that rejects that continuum rejects your subject along with it.

What still works is narrowband, with a hard limit on where. An Optolong L-eXtreme or a similar dual-band filter passes hydrogen-alpha and doubly ionized oxygen and blocks nearly everything else, which is why people image emission nebulae from suburban driveways and get away with it. It does nothing for a galaxy, a globular cluster, a star field, or a Milky Way landscape, because those targets emit the same broadband light the filter is there to kill. Broadband options are a partial hedge rather than a fix: the widely used Optolong L-Pro was designed around the old emission-line offenders, while the IDAS LPS-D3 was cut with LED street lighting specifically in mind and is widely reported to do better against it. Neither one restores a dark sky, and both cost you throughput.

Exposure is the honest lever, and it is expensive. Once sky background is your dominant noise source, which it is for essentially any wide-field night exposure on a modern sensor, signal-to-noise on a faint target scales as the square root of total integration divided by sky brightness. Double the sky, and you need double the total integration to end up where you started. At 9.6 percent per year, sky brightness doubles in a little over seven and a half years. That is my arithmetic on the published rate, and it assumes the 2011 to 2022 trend simply continues, which the paper does not claim. If it does hold, then where artificial light already dominates the background, your integration budget for a given result doubles on the same schedule. Two hours becomes four, and a project with a four-hour ceiling on a clear night becomes one that needs two clear nights. The other lever is unglamorous and more effective than any filter: drive further, and shoot the target when it is highest, because sky brightness climbs steeply toward the horizon. If you want a structured route into the technique, Elia Locardi's cityscape and astrophotography tutorial covers the planning and stacking workflow that this math makes mandatory.

The measurement itself needs no gear at all. Globe at Night has logged 329,384 observations since 2006, and as of late August 2026, the current year stands at 9,045 against a goal of 20,000. The next northern-hemisphere campaign window runs September 2 through 11, 2026, and the constellation is Cygnus, nearly overhead after dark. Ten minutes of dark adaptation, one honest look at the Swan, one chart chosen from eight, and the form submitted before you go back inside. That is the same act, repeated 51,351 times, that caught what the satellites could not see.

Lead image by ESO/P. Horálek, M. Wallner, CC BY 4.0. Source.