Re: flatfields

[Michael Gutzwiller <deepsky@fuse.net>]

Michael Gutzwiller wrote:
> 
> Now I'm getting totally confused.  It would sure be nice to have a TASS
> triplet so I can see some of these effects myself!
>   Mike G. wrote:
> >I then compared the magnitudes of the Tycho stars and the instrumental
> >magnitudes and found they differed by about 0.140 mag sigma, much worse
> >than I expected.  I then graphed the mag difference as a function of RA,
> >result - a random scatter plot.  I then graphed the difference as a
> >function of Dec, voila - a straight line!  The error was a very strong
> >function of declination!  A straight line fit showed a slope of 0.160
> >mag per degree.  That's right, a difference of about 0.450 mags from one
> >edge to the other!  This same slope was repeated in subsequent images on
> >the same night.
>   We've combined data from several TASS sites, with cameras pointed at
> quite different declinations, and the data agree with the 0.05mag sigma.
> This would be impossible if the slope you propose is true.  On the other
> hand, Michael R. had trouble on his one good night with the Vermont triplet
> with a slope of comparable size.  These hints at huge gradients, at least
> on some of the frames, are very disturbing.

It is true that a slope as large as I found in the images I studied last
light would have made for worse photometry than the 0.050 we have been
getting.  I believe we were luckier before for two reasons.  First the
"optimized" PSF based aperture magnitudes I am using now would be more
sensitive to large differences in the PSF from one side to the other. 
Nearly all of the measurements used for the AAS meeting were based on
larger apertures which are less sensitive to the slope. Second, as the
focus is "improved" the difference in ratio between the FWHM on one side
of the chip to the other is exaggerated.  The focus for the I filters
have always been fuzzier and two thirds of our measurements are I
filtered.

> 
> >To date both Tom and I have been using sky backgrounds to generate flat
> >vectors for use in the star program.  These flat vectors are typically
> >slowly undulating and look just like what you would expect if a two
> >dimension flat with dust doughnuts and some vignetting were squashed to
> >one dimension.  What you may not know is that these flats typically show
> >a 5 percent total range, peak to trough.
>   All that you have shown is that the sky brightness, as seen after passing
> through all of the optics, in your frames does not vary by more than 5 percent.
> I could conceive of a situation in which you have vignetting in your lens
> that is stronger (vignettes more) at lower declinations, and at the same
> time your sky is getting brighter at lower declinations.  The two effects
> would then compensate.

While feasible I find this scenario unlikely.  Also if the slope were
due to flat field problems it would have affected all size apertures
equally.

> That is why I keep harping on some external, evenly
> illuminated target so that you know exactly what illumination pattern the
> camera is looking at.  It isn't that difficult to do some tests, and then
> we know what is going on!  I don't like the idea of using sky-generated
> flatfield vectors from cloudy nights or nights where the Milky Way passes
> through the strip, even without considering the gradients from city lights,
> moon, etc.  The problem just gets worse with the bigger fields expected
> from the mark IV cameras.  We need to solve the problems now, not later.
>

I agree and I will attempt to generate some flats by another method for
comparison.  I think I will also see if I can generate PSF's for
different declination regions for comparison purposes.
 
> Glenn G. wrote about photometric limitations.  It is certainly true that
> the 'all-sky' measurements require photometric skies, and that transparency
> variations due to clouds, dust, smoke, etc. will limit the accuracy of the
> results.  Mike G. correctly pointed out that, if the sky variation is a
> slowly varying function, it can be subtracted out and does not
> effect the extraction process itself (the zero point of course may change
> during a single frame or from frame-to-frame on non-photometric nights).
> Clouds also affect the signal/noise since they dim the stars, and you
> won't be able to go as deep on cloudy or moonlit nights.
>   However, I observed in Indiana for many years before moving out
> west, and claim that even Midwestern sites will see 10 percent or more
> photometric nights.  That is why I separated the variability project into
> two segments:  use the photometric nights to generate a master star list
> with good mean magnitudes.  Then do differential photometry for each star on
> all of the nights, using nearby companions to remove nightly transparency
> variations, and setting the zero point for the ensemble with the master
> star list.  The trick is in creating a good master star list, and that means
> getting lots of data so that you can discard the discrepant measures.
> 
>   I still think TASS can do 1 percent photometry.  Hopefully, there are just
> a couple of 5 percent effects now that can be discovered and solved so that
> TASS can reach this level.  The closer you get to one percent, the harder
> the problems will be to solve, but the more useful the data become.

I agree wholeheartedly.  These should be solvable problems and I do feel
we are closing in on their solution.

Mike G.