Re: Flats

[<aah@nofs.navy.mil>]

There are two things you are attempting to do with flatfield frames.
First, any modification to the incoming light before it reaches the
CCD will make a uniform illumination become non-uniform.  Such modifications
include dust motes, vignetting, etc.  This external calibration is the
hard part.
  Second, pixel-to-pixel variations reflect the slight inherent QE
differences between pixels, such as physical pixel size variation.
These effects are at the ~2percent level and become very important when
trying to do accurate photometry.  However, these are the easiest to
calibrate and remove.
  For the internal calibration, you just need any bright light source.
You don't care if that light source has some gradient, as long as the
gradient is over a wide area compared to the size of the pixel.  This
is the approach we use for our astrometric measures, since it is just the
pixels near the centroid that affect the localization.  In fact, the
astrometric flatfield screen at the 1.55m is just a bedsheet illuminated
off-axis by a Kodak slide projector.  We can centroid to ~1/100 pixel
in this manner.  Similarly, if you do a poor job of overall flatfielding,
but a good job locally, and reposition the field 'pretty good' between
exposures, the differential photometry small subsections can be quite good.
  On the other hand, external calibration is really hard if you want to
do sub-percent absolute photometry from one edge of the field to the other.
This is where we get into the questions of twilight flats vs dome flats vs
master sky flats, for example; which one gives the flattest response over
the entire field?  If you just want to measure to millimag precision variables
in a field, with respect to comparison stars located near those
variables, this flatness is a second order effect.  I worry about it because
I am generating an all-sky catalog, and so I need to do a two-step flatfielding, getting
high signal/noise for the pixel-pixel variation and then fitting a smooth function
to a more uniform, but lower, signal/noise flat to remove the gradients in the
high signal/noise flat.  It can be done, but is computationally more involved.
  Differential millimag work in small subsets of the CCD field is difficult enough.
Don't try to get a millimag-uniform flat!
Arne