How dark is the sky in Surbiton?
Looking at light pollution maps Surbiton is on the fringe of urban/suburban London, quite near to Heathrow Airport, M25 and M4 so a provisional brightness figure of 19 is reasonable. However, we can measure the actual brightness by identifying stars, looking up their measured magnitudes (brightnesses) and using that to calibrate our measurements. We also need to know the background darkness. We found that the brightness was closer to 18 than 19, 18.1 being a reasonable figure.
The night
Astrocam ran 598 frames of 59.9 s at gain 1. Capture opened when the sun reached 10 degrees below the horizon at 19:01 UTC and closed at 04:48.
| hour (UTC) | frames | frame mean (ADU) | stops above black (64) |
|---|---|---|---|
| 19 | 56 | 102.813 | 5.3 |
| 20 | 61 | 95.100 | 5.0 |
| 21 | 60 | 90.187 | 4.7 |
| 22 | 60 | 87.570 | 4.6 |
| 23 | 60 | 85.152 | 4.4 |
| 00 | 60 | 82.725 | 4.2 |
| 01 | 60 | 82.153 | 4.2 |
| 02 | 60 | 81.845 | 4.2 |
| 03 | 60 | 82.336 | 4.2 |
| 04 | 60 | 110.972 | 5.6 |
The darkest single frame of the night landed at 02:26:23 UTC, 03:26 BST, at a frame mean of 81.711 ADU, which is 4.1 stops above black. The six darkest frames all fall between 02:23 and 02:28.
The moon
The moon was a waxing gibbous, 79.2 percent illuminated, 10.96 days old. It does not spoil the result because it had already set at 00:19 UTC, two hours and seven minutes before the darkest frame. At 02:26 it sat 17.2 degrees below the horizon; the sun was 29.1 degrees down.
That matters for the ranking below. This night reached its trough under a gibbous moon that had merely got out of the way. The two September nights that beat it were genuine new moon nights, 1.2 and 4.8 percent illuminated. On moon terms this was the least favourable of the three by a wide margin.
| night | darkest frame (ADU) | stops above black (64) | moon illum at trough |
|---|---|---|---|
| 2026-09-10 | 78.812 | 3.9 | 1.2 % |
| 2026-09-13 | 79.671 | 4.0 | 4.8 % |
| 2026-09-21 | 81.711 | 4.1 | 79.2 %, set 00:19 |
| 2026-08-19 | 83.830 | 4.3 | 49.7 % |
(The 69.888 ADU frame from 2026-06-09 is excluded: that is the imx219 era, POSINDEX 1, a different sensor with a different black level and pedestal. It is not comparable.)
The plate solve
We found a plate scale and a photometric zeropoint like this using the green channel of one frame, extracted from the Bayer mosaic (the two greens sit on the anti-diagonal, so the in-capture 180 degree rotation leaves them where they are). Five 600 px crops were solved against the Tycho-2 indexes on muppet, and all five solved.
| crop | native arcsec/px | zeropoint (Gaia G) | sky above black (ADU) | \(\mu_G\) |
|---|---|---|---|---|
| centre | 60.88 | 11.526 | 33.5 | 18.14 |
| left | 56.84 | 11.392 | 30.0 | 17.98 |
| top | 58.92 | 11.539 | 27.5 | 18.30 |
| bottom | 60.74 | 11.431 | 35.0 | 17.99 |
| right | 58.54 | 11.609 | 21.5 | (excluded) |
The right crop is excluded: 23 matched stars against 36 to 70 elsewhere, and a median astrometric residual of 129 arcsec against 37 to 43 arcsec for the rest.
Exposure time cancels, because the star and the sky are measured in the same frame:
where:
- \(\mu_G\) is the sky brightness in Gaia G magnitudes per square arcsecond
- \(\mathrm{ZP}\) is the zeropoint, the magnitude of a star that gives 1 ADU, found by matching stars in the frame to their Gaia magnitudes
- \(p\) is the native plate scale in arcsec per pixel; the green channel has one pixel per 2 by 2 Bayer block, so each covers \(4p^2\) square arcseconds
- \(S_\mathrm{sky}\) is the sky level above black in that pixel, in ADU
Results
Plate scale: 58.9 arcsec/px native, 0.016366 deg/px. The stale value of 0.0190 deg/px was 16 percent too coarse. The 56.8 to 60.9 spread across the field is real radial distortion of about 7 percent, so a single scalar will always be an approximation here. Implied horizontal field of view is 75.4 degrees, not the 66 degrees asserted in camera.json's focus notes.
Black level 64. The sensor values are offset by a pedestal to prevent noise causing values going negative, which would be clipped. It's harmless, as long as it's constant, big enough and known, but it does reduce the effective maximum value to 1023 minus the pedestal. Not usually a concern. We measured it. Part of the frame is permanently blocked by trees and roofline, which gives a dark reference in every frame. On the darkest nights the darkest 32 by 32 patches read 64.2 to 64.4, but that small excess is scattered light, not the sensor: across seven frames from dark to overcast the tree level rises with the sky, by up to 13 ADU in green, and the green excess over red and blue grows in step, which is the signature of reflected light. Extrapolating to no light gives red about 64.05 and blue about 64.10, the sensor's nominal 64 within the error. Values below 64 are recorded, so nothing is clipped, and the read noise is about 0.3 ADU, around 2 electrons. (An earlier version said the first percentile was "exactly 64.00", which was integer pixel values piling up at 64 and 65, and a later one said 64.3; the trees are lit, so 64 it is. Checked by astro-science, 2026-09-25.)
Sky surface brightness: \(\mu_G\) = 18.1 mag/arcsec², with a spread of 18.0 to 18.3 across the four good crops. The result is barely sensitive to the black level: moving it by 2 ADU either way shifts mu by 0.07.
Summary
Three effects push the true zenith figure darker than 18.1:
- Astrocam points at the celestial pole, so it is looking at 51.4 degrees altitude through airmass 1.28, not at the zenith. Worth perhaps 0.2 to 0.4.
- Gaia G is a very broad band and the sky's LED and sodium spectrum is not a star's spectrum, so there is an unmodelled colour term of a few tenths.
- The 6 px photometric aperture loses some of the PSF wing, which biases the zeropoint low and mu low, perhaps 0.1.
18.1 is Bortle 8 (city). We might get darker nights, so maybe 18.5 might happen at the zenith, on a really clear night, which would be Bortle 7. Bortle 7 has not yet been observed but might be. Any wisps of cloud brighten the sky and a deep frost might reduce this, however we want a representative figure and 18.1 is that.