RE: Star's magnitude adjustments

["Gutzwiller, Michael" <mgutzwiller@lanvision.com>]

Arne wrote:
	
	  The correction vector to the flatfield vector should be either
included
	in the flatfield vector itself and used to reprocess the data,
or included
	in the 'instrumental magnitude'.  A single zero point adjustment
per frame
	should be all that is reported to the database.  The correction
vector 
	(or flat compensation in Mike's terminology) is
	just another calibration step to get good instrumental
magnitudes and is
	no more likely to be 'backed out' than is the flatfield itself.

Sounds good to me.  The current program that Glenn and I have been
testing will reprocess a star list to either determine the correction
vector or use one to reprocess a star list.

	  I would like some more information about the flat compensation
program
	when it is available, namely how you are determining this
correction for
	non-photometric nights, what kind of fit you are using, the
number of
	stars used to determine the fit, a handle on the fit error, etc.
There is
	always the potential of making a systematic error in applying
such a
	correction, so we should check it *very* carefully before using
it in
	a production mode.
	
I'll put the following in a technical note when the program is complete
but here's how it works now:

1.	Each star list for a camera for one night is read in one at a
time.
2.	The star list is checked to see if it has a low background
variation to guard against non photometric images.
3.	The stars are matched against a catalog of "good" Tycho stars.
"Good in this case means:
*	the star is not marked as variable
*	Bt and Vt mag errors < 0.060
*	-5.0 < Dec < 5.0
*	No other Tycho star within 0.014 degrees in Dec and 0.023
degrees in RA (the larger RA cut is to avoid bright stars streaking into
this star)
*	V, R, and I are calculated by using Arne's transformation
*	-0.2 < (Bt - Vt) < 2.0 (valid range for Arne's transformation)
*	V, R, and I > 7.5 (avoids saturation)
	With these cuts there are a total of 16,859 Tycho stars in the
catalog which works out to an average of 42 per image.
4.	For each matched star the instrumental magnitude is compared to
the catalog magnitude and the difference and the x coordinate are saved.
5.	After all the good star lists are processed the differences are
separated into several equal regions in x. (Typically 8 but
configurable.)  Each region overlaps it neighbors by 50%.  For example
if there were 4 regions the x values in the regions would be:
		a 0 - 287
		b 96 - 479
		c 288 - 671
		d 480 - 767
6.	A median line fit is made to the differences in each region as a
function of x.  The median is robust to avoid bias due to outliers. The
overlap between regions allows smoother fits by avoiding mismatches at
end points.
7.	An array of (number of regions + 1)  adjustments are calculated.
Each adjustment point is at the normal boundary of each region.  For
example the points for the 4 regions above would be:
a 0
b 192
c 384
d 576
e 767
8	The adjustment is calculated by averaging the values of the
lines fitted for each region containing the point.  This again smoothes
out the adjustment.

Glenn and I are working to determine the best number of regions and
other configuration values.

The flat compensation program looks for lists generated from photometric
images based on the background variation.  A photometric night will have
many lists pass the cut while a non photometric night will have few.  If
a non photometric night does not generate a reasonable number of
photometric images the flat compensation vector from a nearby night can
be used.

Mike G.