[Aavso-photometry] Re: Lightbox usable for Ic filter flats - Summary of recent SBIG Yahoo Group thread

[Wolfgang Renz]

Hello

Here a interesting summary of the recent SBIG Yahoo Group thread
"Flatfield techniques" 18.09.2004 - 20.09.2004:

Paul Warhurst:
I'm curious to know who is doing what with regards image calibration.  I've
seen a few posts down the line  suggesting for instance that dark subtracting
a flat is not needed (!?), and that mean stacks are not such a good idea
compared to median combines (what about charged particles and stars in
the twilight..).

Stan Moore:
--Dark-subtract the flats?
A well made flat exposure should have approx 30,000 ADU, which is 40k 
to 70k electrons, depending on camera gain.  The typical flat exp 
time is 5 seconds or less, for which the typical average dark current 
is maybe 0-3 electrons. The Poisson noise of 50k e- = 224 elections 
thus the dark current is a tiny fraction of the natural noise and 
subtracting that dark will have virtually no discernable S/N effect.  
If the dark is made from a single exp (e.g. auto-dark) then that dark 
subtraction actaully adds 10-15 e- of extra noise (but that is still 
buried by the flat's self-noise).
However, it may be useful to dark-subtract a bias, especially if your 
camera model has a strong (and bright) bias pattern.  Also some of 
the newer SBIG camera (e.g. ST-2000) have very high pedestals that 
are not expected by the calibration software (which usually assumes a 
100 ADU pedestal).  Subtracting a bias will remove the pesestal 
whatever it is (though some software adds a 100 ADU pedestal after 
the subtraction).
It is generally pointless to make special master-darks to match to 
your flats because a master-bias will do the job.
--Combine the flat-exps via mean, median or sigma-reject?
Statistically, the mean (or sum) will produce the very best S/N with 
the least dynamic distortion.  It is also the simplest procedure and 
has no hidden traps.  However, sky-flats with protruding stars or 
flats with the occasional cosmic ray may result in a blemish. (Flat 
exps are generally so short that there are very few frames with a 
cosmic ray).  Simply reviewing your frames and rejecting the 
occasional defective frame from the combine can usually deal with 
these blemishes.
Median or sigma-reject can be used to combine frames with blemishes 
but these procedures are much more complicated and have "hidden 
traps" that can easily result in significantly inferior S/N and/or 
dynamic distortion.  This is mainly a consequence of the usual need 
to normalize frames prior to a median/SR combine. The math of 
normalization and its S/N and field effects is a surprisingly deep 
topic that I am not going to get into here.  An exception to the 
requirement of normalization is intensity-stabilized flat devices 
that produce very uniform avg ADU across frames – these frames do not 
need to be normalized and can be directly combined via median or SR.  
But if you are median/SR combining sky flats then you must normalize 
them and if you do not really know what you are doing (or if you 
don't know what your software is really doing) then your flats may 
not be as optimal as you think.  There are ways to measure and verify 
the flat's S/N and dynamic distortion, but that's also a topic too 
deep for this email (though I am writing it up and will eventually 
post on my web site).

Paul Warhurst:
I think you won't find anybody trying to get robust flats for science
work using a mean flat, especially for twilight flats. Even near the
galactic poles there are always stars in the field, even if they're
unseen they add signal that can really ruin photometry. Making a nice
smooth picture is one thing, but I think many are interested in
scientific work too, and some of the methods I've heard about wouldn't
cut a lot of mustard most places engaged to that end.
I'm curious about these stable flat devices, so far all I've ever seen
that had truly stable output was beta lights used for linearity
testing detectors and they are quite green. What are they?

Chris Peterson:
For photometry, the first choice is usually median combined, normalized sky
flats. The second choice is some type of artificial flat (dome flat).
As you note, mean combined sky flats are not scientifically useful. I think
Stan is talking about what gives the best results for aesthetic imaging, which
is the focus of most people on this group.
I prefer sky flats, but get good results with a high temperature halogen
reflected from a gray card onto a white screen. The resulting light is very
stable and uniform, and has good spectral characteristics.

Ray Gralak:
Stan, do you have and actual examples of a sigma-clip combined
master flat producing inferior results? My experience has been
the opposite... a properly created sigma-clipped flat always 
produces a better result than a mean combine. 
Maybe you could post a set of *twilight* flats that you believe 
would produce a cleaner result with a mean combine? All of the 
flats should be in the linear range of the CCD and approximately 
near 1/3 to 1/2 full well (no fair using flats far below 1/2 full 
well! <g>).

Peter (erdmanpe):
Your mentioning the high temperature halogen bulb reminded me to ask
about it again <g>.  Do you have the make/model # available?  To me,
this looks like the best source illumination.

Chris Peterson:
Sorry, I still need to track down the receipt from when I ordered the bulb.
I can't read the number on the unit itself. I really need to do that- these
things don't last forever!
The one change I've made to my dome flat setup is that I used to use a piece
of white paper as my first reflector, and now I use a photographic gray
card, which shifts the light on the screen blueward. I now get more balanced
exposure times with I, R, V, and B filters.
It does seem like the 5000K rating on the light must be achieved by some
type of coating on the reflector (which looks dichroic, changing from blue
to pink with varying angle), since there is no filament material that could
be producing a blackbody output at that temperature.

Frank Barrett:
Stan, I always appreciate these analytical threads, as they usually result 
in some interesting debates.  However, sometimes I've observed that 
the truth lies distorted somewhere behind the assumptions:
>The typical flat exp time is 5 seconds or less, for which the typical
> average dark ... (Flat exps are generally so short that there are very
> few frames with a cosmic ray)...
It seems your assumption here is a very short flat exposure.  I use 
an artificial flat box which utilizes LEDs and I don't think I've 
ever taken a flat as short as 5 seconds.  So my question is, at what 
point is the exposure long enough that significant dark current 
and/or the "probability" of a cosmic ray hit become an issue such 
that dark frames and outlier rejection algorithms become more useful 
than harmful?
I do agree with your premise that shorter flat exposures are to be 
preferred owing to thier lower artificial and natural noise 
contributions and this has me thinking about a brighter light source 
for my flat box...hmmm...
BTW, I'd be very interested in reading about your analysis of the 
effects of normalization on S/N.

Peter (erdmanpe):
Chris, Thanks, that would save me some useless experimentation <g>.
How did you discover the blue shift from the photographic grey card,
just by test?  Presumably, "grey" just means less reflectivity, but
without a spectra bias.  Are all photographic grey cards similar, or
is yours something of a specialty item?  Sounds like a nice scheme to
increase the effective black body temperature.

Chris L Peterson:
Peter, I just noticed that an old gray card I had seemed to have a bluish cast. I
agree that in principle "gray" should mean "neutral", but I guess they don't
worry much about IR for normal photographic use. I think the main effect I'm
seeing is coming from IR attenuation. Visually, the final screen I image
doesn't look much different. But discriminating colors, especially
variations on white, is pretty impossible visually. The card I'm using is
just a standard 18% card like you can pick up for a few dollars at any photo
store.

Stan Moore:
Paul, Most "science work" is filtered and massive apertures that make 
starless sky-flats nearly impossible. 
Also, most professionals are aware of the mathematical nuances of 
pedestal, bias, and normalizations (or should be <g>).  My statement 
did not say that median/SR is wrong, only that there are some 
pitfalls that can result in poor master flats.

Chris L Peterson:
Stan, The vast majority of photometric work is done with meter or smaller
aperture scopes- hardly "massive apertures"!

Stan Moore:
> do you have and actual examples of a sigma-clip combined
> master flat producing inferior results? ...
Ray, Here is a good test for flat fields:
1) take a normal series of flat exposures
2) set one exposure aside (do not use in the combine)
3) combine the other exposures into your master flat
4) use that master to flat-field the set-aside exposure
5) examine the flatted exposure for:
5a) any ghostly "dark stars" mean that the master flat has stars. 
Bright stars are within the set-aside exp and do not impugn the flat.
5b) the flatted flat should be perfectly uniform across the field.  
If there is a gradient or ghost vignetting, donuts, etc. then there 
was a mistake in the normalization or handling of the pedestal/bias.
5c) compare actual S/N to Poisson S/N:
5c1) measure Standard Deviation (STD)
5c2) compute S/N = mean_ADU / STD
5c3) compute Poisson noise = sqrt(gain * mean_ADU) = Poisson S/N
5c4) compute ratio of actual S/N to Poisson = 5c2 / 5c3
This ratio should be between 0.9 and 1.1, note that measured S/N can 
exceed Poisson S/N due to the re-scaling of ADU from the normalized 
flat-division (i.e. after flat-division, not all ADU measurements 
actually correspond to electron counts).
6) Try a different method (mean vs median) and re-evaluate (steps 3-5).
Compare the S/N produced by the different methods.

Ray Gralak:
Stan, I have done steps 1-5b and have seen *visible* effects in 
mean-combined twilight flats that are not present in sigma-combined
flats. I'm sure there are spots in the mean-combined flat that 
have no "defects" and thus higher S/N than the same spot in the 
sigma-combined flat... but that does not make the mean-combined
flat the better choice because it can have many poorly corrected
spots. 

Stan Moore:
> ... at what point is the exposure long enough that significant 
> dark current and/or the "probability" of a cosmic ray hit 
> become an issue such that dark frames and outlier rejection
> algorithms become more useful than harmful?
Frank, Those are 2 different issues.  
Because flats are not "dithered", median/SR generally has no effect 
on dark current (a hot pixel is always hot).  So the first question 
might be: 
-- "At what point is the exposure long enough to benefit from a matched-
dark subtraction? (instead of bias subtraction or pedestal handling)" 
You can investigate this empirically by:
1) take a flat series and set-aside one of the frames
2) make a master flat via [bias subtract] from the remaining flats 
and apply it to the set-aside frame.
3) make another master flat via [matched dark subtract] from the 
remaining flats and apply it to the set-aside frame.
4) Compare the S/N (mean_ADU / STD) of the two differently flat-
fielded set-asides.  If the [matched dark subtract] S/N is 
significantly higher then go with that method, otherwise doing a 
[matched dark subtract] is unnecessary.  Note that it is entirely 
possible for the [bias subtract] flat to have superior S/N when the 
bias is a very high quality (because the [matched dark subtract] may 
contain more read noise, depending on its construction).
-- The other question is "at what point is the exposure long enough that 
the probability of a significant cosmic ray hit becomes an issue?"  
That depends on your location, elevation, the composition of the 
glass near your CCD, etc.  But I would say probably about 1 minute.
> BTW, I'd be very interested in reading about your analysis of the 
> effects of normalization on S/N.
Briefly, there are 3 considerations:
1) Proper handling of the bias/pedestal when normalizing (should be 
subtracted prior to normalization)
2) Correct type of normalization (multiplicative only – not offset)
3) Avoid large normalization factors, which can lead to sub-optimal 
S/N.  Large factors occur when the frames have significantly 
different avg ADU (e.g. twilight series) so this should not be an 
issue for flat-box techniques.


More to come.

Cear skies
  Wolfgang

-- 
Wolfgang Renz, Karlsruhe, Germany
Rz.BAV = WRe.vsnet = RWG.AAVSO