Checking 16,074 Night-Sky Brightness Measurements
Against the Light-Pollution Map

On one side, night-sky background brightness measured in the field by amateur astronomers worldwide with SQM instruments; on the other, the "theoretical value" returned by a light-pollution lookup service. Only by understanding where each comes from can you compare them correctly.

The theoretical map describes an ideal night sky, so a fair comparison is needed: keep only the — readings taken zenith-up, under clear skies, with the moon below the horizon and a sensible value. Each point below is one measurement — theoretical on the x-axis, measured on the y-axis; the closer to the diagonal, the better the agreement.

The earliest data is from 2005, while the theoretical map is the 2024 edition. If a location brightened over these years, its older readings should read darker than the 2024 model (positive residual). Looking at residuals year by year — is the trend really there?

The real trap is not the passage of time, but that the sites sampled in different years are simply not the same set. Look at the per-year "mean theoretical brightness" and sample count — the last three years have been flooded by a handful of bright city auto-stations.

2005–2023 was mostly amateur observation at dark sites (theoretical ≈ 20.5–21); from 2024, hourly-reporting SQM-L units like the one in suburban Cincinnati pull the sample toward the city (theoretical ≈ 18.7). Without controlling for site, this "change in sampling" gets misread as "the sky changing with the year".

The cleanest approach is to stay fixed on a single point. For the — sites that span at least 4 different years, each is fitted for the slope of "measured SQM ~ year": positive = darkening, negative = brightening.

Sites tracked for the most years

Plotting the — sites on a world map, coloured by theoretical brightness — bright spots (cities) glow red, dark ones (wilderness) turn blue-black, faintly tracing the map of human lights.

The sections above proved the zenith theoretical map is accurate — but it only answers "how dark is the zenith". At the darkest places (theoretical SQM ≳ 21.7, Bortle 1–3), zenith brightness saturates: the zenith readings of classes 1, 2 and 3 are almost identical, and the zenith alone cannot tell them apart. What truly sets the observing class is space — how large the surrounding dark area is, whether light domes sit on the horizon, and whether mountains block those domes below the line of sight.

The zenith saturates, but space can separate them

Ground light domes × mountain shielding

The "artificial glow" at the dark end comes from ground light domes near the horizon: light from the surrounding towns is summed by the Walker/Garstang propagation law Σ A(s)/d^2.5 (radius ~120 km), capturing the whole horizon rather than just the nearest city. A further terrain-gating layer is added: elevation is sampled along each azimuth to compute the horizon angle, and the first source beyond >2° is blocked below the horizon by the mountains and excluded from the glow. For the same zenith darkness, a valley backed by mountains can be far darker than open wilderness.

The more negative the glow index, the darker. Tivoli and Wiruna are open on all sides, so terrain does not change their glow; but at Crestone the 5–16° Sangre de Cristo ridge to the east blocks the Front Range conurbation 100 km away → trimming 56%, actually overtaking the open Wiruna; Kendrick Park is shielded 52% by the 3850 m San Francisco Peaks → rescued from a "3-class dark-pocket drop" back to class 2. This is exactly the layer the zenith SQM cannot see at all, yet which decides the observing class.

Further reading · The full recalibration table and the glow and terrain-gating formulas are in "Dynamic Light-Pollution Rating · Design & Technical Document"; the field analysis of why the dark-end zenith reading fails is in "The Dark-End Divergence".