Digital Camera as Sky Meter: the Full Scoop

As I reported in a previous note, I recently purchased a Canon A60
digital camera to experiment with it as a sky-brightness meter.
To make a long story short, I have been quite pleased with the
results. I find it very convenient to use, and it is amply
sensitive to measure sky brightness in any urban or suburban
setting. However, I have some doubts about its value under
truly dark skies.

I have used the camera to measure sky brightness at my four primary
observing sites, as follows:

zenith zenith distance and bearing to
location reading NELM major source of light
-------- ------- ---- ---------------------
New Lebanon, NY 11 6.7 Albany, 25m NW, 900K people
Westford, MA 23 5.7 Boston, 30m ESE 3.5M people
Lincoln, MA 31 5.2 Boston, 15m ESE 3.5M people
Cambridge, MA 100 4.7 Boston, 5m ESE 3.5M people

These figures should all be taken with a grain of salt. I have
reason to believe that the camera overstates the sky brightness
in New Lebanon, and all other reasonably dark sites. The zenithal
NELMs were acquired on different nights from the camera readings,
and are not very reliable anyway; I find limiting magnitude to be
exceedingly time consuming, highly variable from one observer to
another, and also highly variable for myself from one time to
another. The distances to the nearest cities are estimates to
the center of population and/or industry, and are funky in all
cases. Downtown Albany is actually only 20 miles from my site
in New Lebanon, but the vast majority of its population and
employment lies on the far side of downtown from me. Boston
is asymmetric the other way; the far side of downtown is the
Atlantic Ocean. All three of the MA sites are well within
the area used by the Census Bureau to derive that 3.5M figure,
although Westford is not too far from the edge.

Even allowing for all the unknowns, though, it is striking that
the 3-fold increase in brightness from Lincoln to Cambridge
corresponds only to an 0.5m decrease in NELM, the same as the
1.5-fold increase in brightness from Westford to Cambridge.

Equally interesting to me is the way that light pollution is
distributed at any given site according to altitude and azimuth.
The tables below show the zenithal reading for each site plus
two columns giving readings at 30 degrees and 60 degrees above
the horizon at 8 points of the compass. The E and SE figures
are missing for Lincoln due to trees.

NewLeb Westford Lincoln Cambridge
30 60 30 60 30 60 30 60
Zenith 11 23 32 100
N 16 10 36 23 60 37 178 114
NE 14 11 43 24 76 39 180 122
E 15 10 47 25 -- 45 234 132
SE 13 9 48 24 -- 42 234 134
S 13 10 42 25 64 39 196 124
SW 14 10 39 26 50 33 168 112
W 13 10 33 22 51 33 172 110
NW 18 11 29 21 46 33 172 112

I also measured the sky brightness closer to the horizon than
the 30-degree figure given above. In all cases, the brightness
was inversely proportional to the altitude within the error of
my measurement, e.g. the brightness at 15 degrees was 2X the
brightness at 30 degrees, the brightness at 10 degrees was 3X
the 30-degree brightness, and so on.

It is interesting to note that the zenithal brightness in
Lincoln, which is fairly dark as suburbs go, is only slightly
lower than in an otherwise dark location at Full Moon.
It is also interesting to note that although the light
dome of Albany seems immensely obtrusive in New Lebanon,
that even directly above the trees in the worst part of the
sky at New Lebanon is darker than the zenith in Westford,
which in turn is much better than most suburbs I have been in.

---------------

Full details of how I arrived at these figures, and the
experiments that I did to verify their validity, would be
a major subject, and are still undergoing investigation.
I will give just the essentials here.

Basically, I mounted the camera on a tripod in the "portrait"
orientation and took two shots for each compass reading,
with the altitude determined by a graduated level and
confirmed by matching star patterns. All shots were at
the widest zoom setting, where the FOV is about 40x55
degrees, allowing two shots to cover the sky from horizon
to zenith with ample overlap.

All the shots above were 15-second exposures done at the ASA 400
setting; in the future, I plan to use ASA 200, which is equally
sensitive at low light levels while running into fewer problems
with non-linearity at high light levels. All the shots above
were done with the Black-and-White "effect", but in the future
I plan to use normal (Color) mode. It turns out to make no
difference at all whether the RGB colors are averaged by the
camera before downloading to the computer or are averaged by
the photo processing after download, so why throw away potentially
interesting information? I shot at the lowest possible resolution,
640x480, and would happily have gone lower still if I could.

After downloading, I opened each photograph in Corel Photo-Paint
and resampled it to anywhere from 3% to 6% of the original linear
dimensions, depending how much detail I wanted to preserve.
Resampling gets rid of very substantial noise at the level of
individual pixels, although it definitely also discards potentially
useful information. Then I measured the brightness of resampled
pixels in the areas of interest on an 8-bit grayscale.

I determined by numerous experiments that the pixel reading is
nearly proportional to the brightness up to a reading of about 60,
and then can be corrected as shown below. All the readings given
above are *after* correction. The "actual brightness", of course,
is on a purely arbitrary scale determined by the camera's
response at ASA 400, 15 seconds, for readings under 60.

camera actual correction
reading brightnes formula
60 65 1.5x - 25
70 80 2x - 60
80 100
90 120
100 140 2.5x - 110
110 165
120 180 3x - 170
130 220
140 250
150 280 4x - 320
160 320 6x - 640
170 380

There is also a failure of linearity for low light levels. A
totally black shot yields a camera reading of 3 or 4 after
resampling at ASA 400, consistently 1 at ASA 200, and 0 at
ASA 100 and ASA 50. The camera reading stops being linear
as the light level approaches zero, but instead approaches
that black-shot level asymtotically. That is why ASA 200
yields as much information as ASA 400 at low light levels.

Somewhat more baffling, I detected a "floor" of 9 - 11 for
readings in New Lebanon, regardless of sky brightness. For
instance, the readings seemed unchanged after the Moon had
risen. Granted, the Moon was a fairly thin crescent (4 days
before new) and only 8 degrees above the horizon, but the
subjective effect on the Milky Way was dramatic; it was still
easily visible, but very much washed out as compared to the
view before moonrise.

I believe that the floor of 9-11 is due to some nonlinear
effect from faint stars. One reason that I believe this is
that the floor seems to rise when I shoot at full resolution
rather than the 640x480 which I normally use. Another reason
is that the 11 reading always lies in or near the Milky Way,
while the 9 reading lies far from the Milky Way. It is also
extremely intriguing that the Milky Way is no more visible
in the shots from New Lebanon than in the shots from Westford,
or even from Lincoln, although it is certainly immemsely
much more prominent to the naked eye. This hints to me at
some kind of non-linearity, possibly due to the fact that
stars are point sources.

- Tony Flanders

Interesting. Look out for temperature-dependent effects, especially when
measuring the darkest skies.

In message , Tony
Flanders writes

Somewhat more baffling, I detected a "floor" of 9 - 11 for
readings in New Lebanon, regardless of sky brightness.

What was the value for the inside of your lens cap taken at the same
time?

The CCD dark current is both temperature (and so time dependent) and
position sensitive on the array. If you take a long enough exposure on
most consumer digicams you will be able to see which is the "warm"
corner nearest to the readout electronics and stray IR photons.

I believe that the floor of 9-11 is due to some nonlinear
effect from faint stars. One reason that I believe this is
that the floor seems to rise when I shoot at full resolution
rather than the 640x480 which I normally use. Another reason
is that the 11 reading always lies in or near the Milky Way,
while the 9 reading lies far from the Milky Way. It is also
extremely intriguing that the Milky Way is no more visible
in the shots from New Lebanon than in the shots from Westford,
or even from Lincoln, although it is certainly immemsely
much more prominent to the naked eye. This hints to me at
some kind of non-linearity, possibly due to the fact that
stars are point sources.

Sometimes it is the eye's own non-linearity that means the camera shows
you more accurately what is really there, but the eye fails to see
enough contrast to latch on.

Try a few shots with M33 in the frame - that ought to show if it is
diffuse faint starlight. I'd also be inclined to shoot at full
resolution - you have no idea what dirty tricks the camera firmware may
use to downsample from the CCD to 640x480.

Very interesting results and with simple easily available equipment too!
- I hope more people try measuring their skies this way.

Regards,
--
Martin Brown

Martin Brown wrote in message ...

Somewhat more baffling, I detected a "floor" of 9 - 11 for
readings in New Lebanon, regardless of sky brightness.

What was the value for the inside of your lens cap taken at the same
time?

Yes, I thought of that. Readings from trees in the same frames, or
in frames near by, are consistently 4, the standard value for a dark
frame. Note that the camera does have some kind of correction for
thermal effects at speeds below 1 second. I suspect that it internally
takes a shot with the shutter closed and subtracts it from the actual
shot. In any case, a 15 sec exposure starts 15 sec after you press
the shutter release.

- Tony Flanders

Believe that the camera is auto shooting a dark frame equal to the
normal long exposure shot. The dark frame captures the image of the
electronic noise and subtracts it from the light shot.

Bob Berta

(Tony Flanders) wrote in message m...
Martin Brown wrote in message ...

Somewhat more baffling, I detected a "floor" of 9 - 11 for
readings in New Lebanon, regardless of sky brightness.

What was the value for the inside of your lens cap taken at the same
time?

Yes, I thought of that. Readings from trees in the same frames, or
in frames near by, are consistently 4, the standard value for a dark
frame. Note that the camera does have some kind of correction for
thermal effects at speeds below 1 second. I suspect that it internally
takes a shot with the shutter closed and subtracts it from the actual
shot. In any case, a 15 sec exposure starts 15 sec after you press
the shutter release.

- Tony Flanders

In message , Tony
Flanders writes
Martin Brown wrote in message
...

Somewhat more baffling, I detected a "floor" of 9 - 11 for
readings in New Lebanon, regardless of sky brightness.

What was the value for the inside of your lens cap taken at the same
time?

Yes, I thought of that. Readings from trees in the same frames, or
in frames near by, are consistently 4, the standard value for a dark
frame. Note that the camera does have some kind of correction for
thermal effects at speeds below 1 second. I suspect that it internally
takes a shot with the shutter closed and subtracts it from the actual
shot. In any case, a 15 sec exposure starts 15 sec after you press
the shutter release.

Seems highly likely then.

Another thing to consider is that the in camera JPEG compression will
generate some positive bias in the lowest values near stars due to the
effect of quantising the coefficients. Unresolved point sources on a
flat black background are just about worst case pathological for JPEG.

If the camera has a TIFF mode at full it would be interesting to see if
using that made any noticeable difference. It might or might not
depending on the actual JPEG compression settings used in the camera.

Regards,
--
Martin Brown