Every Bortle or SQM estimate you see in a planning app answers the same question — "how dark is this site?" — with the same basic method: take a satellite radiance measurement, propagate it through a model of atmospheric scattering, and turn the result into a sky brightness with one constant. PlanAstro's self-hosted estimate works exactly that way, computed from the portal's own copy of satellite data with no outbound request at request time. This article is about what happened when we checked that one constant against real measurements, in a country whose sky it clearly did not describe well.

The data

Since 2018, Japan's Ministry of the Environment (環境省) has run a twice-yearly national star-watching campaign, 「デジタルカメラによる夜空の明るさ調査」: a digital camera pointed at the zenith (35–50 mm, 30 s, f/5.6, ISO 800), reduced against standard stars in the same frame, reported in mag/arcsec². The campaign has produced 3,531 measurements at 1,392 sites between 2018 and 2025 — calibrated photometry, not a citizen-science estimate, and published under a public data license (PDL1.0) compatible with CC BY 4.0. It is exactly the kind of source PlanAstro's calibration script already knew how to fit against, alongside the U.S. National Park Service's own all-sky measurements and Globe at Night's worldwide citizen readings, which is what the shipped global constant was originally fitted on.

Source: 環境省 星空観察.

The result

Filtered to the 215 raster cells with enough independent readings to trust, the model — using its existing global constant of 0.2990 — read the Japanese sky about 0.75 magnitudes too dark: it landed on the exact Bortle class about 36% of the time, and within one class 87% of the time, which is a lot worse than the model does elsewhere. The bias was uniform enough across brightness levels to point at one missing factor in the constant, not at a geometry problem in the propagation model itself.

A second, independent instrument backs that reading. Globe at Night's own SQM measurements inside Japan — different camera, different observers, already part of the worldwide dataset — refit to a similar regional value on their own. And the same by-region refit applied to Globe at Night everywhere else lines up with climate, not geography: the dry western United States refits to 0.25, Iberia to 0.29, the humid south-eastern United States to 0.36, north-western Europe to 0.34, and humid Japan to somewhere around 0.44–0.49 depending on exactly how the region is drawn. Japan sits at the extreme end of a dry-to-humid gradient the model already shows elsewhere — it is not a separate phenomenon.

What shipped

Refitting the constant directly on the Japanese campaign's own measurements — the same V-band-like camera method the bias was found in — gives 0.832 across those 215 cells, with the residual bias falling to +0.05 mag and the RMS error from 0.75 mag down to 0.37 mag. That fitted number, together with its provenance, is what PlanAstro ships as a regional calibration entry for Japan: no individual measurement and no site position is stored or redistributed, only the fitted constant, because the copy of the campaign's results that carries exact coordinates sits under separate terms that reserve reproduction. The openly licensed result tables — prefecture, municipality and a free-text place name only — are what authorizes shipping the constant at all.

The region itself is drawn as two simple polygons over open water (Hokkaidō down to the Yaeyama archipelago, plus a small ring around the Ogasawara Islands), deliberately with no smoothing at the edge: two sites on either side of the Korea Strait at the same measured radiance will read a different SQM, by design, rather than pay for an invisible few-kilometer improvement with a much more complex boundary.

On the real raster, before and after this calibration:

The limits, stated plainly

This is a solid sample for one country — 215 cells, 2,799 filtered measurements — but it is one instrument, reduced by one campaign, and the caveat that matters most is unresolved: the campaign's camera reads close to the V band, while the global constant this regional entry corrects for is fitted mostly on Globe at Night's SQM band, which typically reads a few tenths of a magnitude darker under a light-polluted sky. We do not yet know the exact offset between the two, so we cannot yet say whether the "true" Japan-specific constant, expressed in the same band as the rest of the world, is closer to 0.832 or to the roughly 0.49 that Globe at Night's own (much smaller) Japanese sample gives on its own. Both numbers are shipped in the documentation, and neither is hidden behind the other.

Other caveats worth naming: a 2′ raster cell cannot separate a city center from its inner suburbs; terrain that screens a site from a distant city is not modeled; and the underlying satellite composite is a single year (2025) of VIIRS data. None of this changes the global constant used everywhere else — fitting one constant across the world plus Japan would have made the rest of the world's estimate worse to only partly fix Japan's, so Japan gets its own regional entry instead, which changes nothing for any other position on the map.

How you can help

If you have access to a calibrated SQM reading or a Globe-at-Night-style measurement at a known place, especially in a region PlanAstro has not calibrated yet, it is worth more to a model like this than almost anything else we could add in code. The calibration script and the regional-calibration table are in the open-source repository, documented in full in docs/SKY_QUALITY_RASTER.md — adding another region follows the exact same steps this one did.

Sources and attribution. This product was made utilizing VIIRS Nighttime Lights (VNL) annual composite data produced by the Earth Observation Group, Payne Institute for Public Policy, Colorado School of Mines, licensed CC BY 4.0 — see Elvidge, C.D., Zhizhin, M., Ghosh, T., Hsu, F.-C., Taneja, J. (2021), "Annual time series of global VIIRS nighttime lights derived from monthly averages: 2012 to 2019," Remote Sensing 13(5), 922, doi:10.3390/rs13050922. Japanese measurements: 出典：環境省 星空観察. U.S. measurements: National Park Service Night Skies Program (public domain). Worldwide citizen measurements: Globe at Night, NSF NOIRLab, CC BY 4.0. The Japan-specific constant is calibrated on the Ministry of the Environment's published results; this does not imply their endorsement or validation of PlanAstro.