Re: Light pollution monitoring using all sky camera

[Jan Hollan <jhollan@amper....muni.cz>]

Dear friends,

even if this seems to be rather a topic for magnitude6 or some
google-accessible public forum, let's discuss it here for a while:

> I would suggest the following experiments to show the whole disaster of
> light pollution:
>
> Urban/sub urban places:
> Canon EOS 5D, Sigma 8mm, 1000 ASA, F 1:3.5, 30 seconds
>
> Darkest places:
> Canon EOS 5D, Sigma 8 mmm 1000 ASA, F 1:3.5, 5 minutes
> 	(guiding would show more stars in zenith and more light pollution)

To show it somehow, yes, to give any values, not exactly. 1000 ASA is nice
to see stars in camera-made jpegs, but of no advantage for photometry.
ISO 100 is better.

The reason is that the same (noisy enough, see below) signal is but
multiplied ten times more using ISO 1000, giving no more information --
actually, discarding information for bright stars, lamps etc. This is due
to diminishing the dynamic range of the image ten times. Maximum
recordable readings between 3 000 and 4 000 for ISO 100 become over 30 000
for ISO 1000 and do not fit into the 12-bit limit for Canon raw format any
more. Stars get overexposed.

Why "noisy enough": information would be lost if pixel readings of
darkframes (and of darkest parts of real images) would be mostly of the
same value, say 129 for Canon. They are not, 128, 129 and 130 are equally
common. 127 and 131 (or 128 and 132, for warmer parts of CMOS) are
percentils 10 and 90, even for shortest exposures (those used as offsets
when longer-exposure darkframes are scaled). This holds for D60 camera at
least, I suppose 5D is not much different.

You can still get nicely looking jpegs from raw images taken with ISO 100,
after subtracting (scaled, if needed) darkframes and multiplying the
result as needed before gamma correction (or even applying some clever
formula to get 256 levels out of the original 12-bit range).

So much for ISO. Another photometric topic is the sharpness. Small,
roughly circular star imags are an advantage, as this is what photometry
programmes search for when identifying the stars. Stars appearing as
coloured "comets" at 80 degrees from zenith are difficult to identify and
measure by any algorithm. However, having almost all light within one
pixel is not good. It may be tolerable for scientific cameras, but is bad
for those with RGB overlays. You need the same amount of light on all
three tiles of the colour matrix. The stars have to be slightly defocused
to achieve this. Otherwise no photometry of individual stars is possible,
just some statistics can be done. The statitics can give, apart from the
luminance calibration, an estimate of "zenith extinction" from many star
images of varying zenith distance.

But fish-eye photography offers more than a single number (extinction of
light in zenith) characterising the atmosphere. A map of "zenith
extinctions"  of different parts of sky could be produced, if there are
enough stars captured. Near cities or another sources of pollution, it
would be very interesting. And in fact, needed to interpret the sky
luminance values.

Unfortunately, I know of no software which could do automated photometry
of fish-eye sky views. I'd be much interested in such a thing, esp. if it
would run on linux (as a batch, processing many images without any
typing or clicking).

For sky luminance alone, no star images are needed when using raw images.
The stored exposure info should be reliable enough if manual setting had
been employed, and the recorded signal minus darkframe is proportional to
luminance. Once the constant of proportionality is known, then luminance
of any region consisting of at least one pixel of each colour can be
computed easily, for any image obtained by the camera. My programme
raw2lum using the dcraw input does the job. (I know that I should write a
paper about it and some user manual too, the existing directory
  http://amper.ped.muni.cz/light/luminance/
 is indeed not user-friendly). Occasionally, I could process your images,
if darkframes for them would be available as well. Then you could repeat
my procedure yourselves (the employed command lines are stored in the
raw2lum *.eps and text outputs).

I have no experience with focusing fish-eye lenses. As I see, not all DSLR
cameras offer full 180+ degree view, due to long fish-eye focal lengths
and small CCD or CMOS sizes. But I got excellent experience with an afocal
fish-eye converter FC-E8, at the old Nikon 990 (there is a Czech report
based on it, within http://amper.ped.muni.cz/noc/krnap/). This converter
or the larger type (FC-E9) might be perhaps adapted for another cameras,
preferably non-reflex ones with small optics and small sensors. But just
those ones which offer darkframes with almost no zero-value pixels
(compact nikons with *.nef output half of the pixels as zero, they are
unsuitable for faint light photometry). I'd gladly process all-sky images
obtained in this way.

cheers,
 jenik

PS.
 Let me remark the geometry for FC-E8, as hardwired into my
map_bsct sky map programme, is 2.4 sin(z/2.4):

K<string>[:<string>]  kind of geometry and surface:
  E or 0     Equidistant from projection pole (default), z,
  Conf or 1  Conformal (non-distorted angles), 2 tg(z/2) for a plane,
  G or 2     Gnomonic (central perspective),  tg(z)
  4          Equivalent (equal surface area), 2 sin(z/2)
  5          Nikon990 with fish eye convertor FC-E8,  2.4 sin(z/2.4)
  ...

Consequently, the space angle is slightly magnified away from zenith, the
formula being
       1.2 sin(z/1.2) / sin(z)
 giving 1.16 at z=90 degrees and 1.12 at 80 degrees.