Re: [TASS] Flat fielding

[Andrew Bennett <andrew.bennett@NS.SYMPATICO.CA>]

On Tue, 14 Sep 1999 10:46:45 -0700, "Arne A. Henden"
<aah@NOFS.NAVY.MIL> wrote:
> ...
>  There are lots of these sources of error.  You should
>understand the process before you start calling on
>the statistician gods.

I was struck speechless. The effect is now wearing off.

The whole reduction process is an exercise in statistics -
the conversion to I magnitudes a particularly nasty one
involving extrapolation and all kinds of possibilities
for selection effects. It is just not possible to
understand the process without becoming at least a bit
of a statistician.

But, to return to my original point (and the subject of
the thread) and rephrase it as offensively as I can as a
series of assertions:

1) Flatfielding by dividing by a vector derived from a
single image is a waste of time and introduces errors.
The large scale errors (gradients) can be partially
removed by FlatComp. The small scale errors arising from
incomplete removal of stars cannot be so removed and
unnecessarily increase the final photometry errors.

2) One should instead use a flat vector derived from
a large number of images. Because the drift scan method
averages over a complete column, the pixel to pixel
scatter is very small and should be ignored. The flat
vector should be a low order polynomial fit to the
large scale (vignetting) response.

None of this applies to the MK IV which is a whole
new can of worms.

Disclaimer: These conclusions are based on the TN's, etc.
plus a purely mental examination of the raw data to
which I do not have access. Any connection with fact
is accidental.

Andrew Bennett, Avondale Vineyard, Nova Scotia, Canada.