further investigations of photometry vs. position in field

[Stupendous Man <richmond@a188-l009.rit.edu>]

  I've spent another couple of days working on the 7x7 grid
measurements from DS24 (images with names like hvra2604797 to hvra2604863).

  The good news is that I finally have a pretty good set of 
tools for finding and removing low-order polynomial corrections
from a set of grid images.

  The bad news is that it doesn't work all that well with the
this particular set of data.  I found that a combination of 
linear terms like this:

                correction =  a*(x-x0) +  b*(y-y0)  +  c*r

where "r" is the radial distance from center of frame (x0, y0),
does a pretty good job of removing systematic errors in the 
I-band.  Higher-order terms don't help much.  Using the 
correction factor reduces the scatter in measurements of the
same star at different positions by about a factor of 2, as
I described in TN 89.

  I did notice that there were 4-7 frames in I-band which
always left outliers in the residuals.  It appears that those
images may have been taken through thin clouds.

  When I looked at the V-band data, the problems with those
same frames became much worse.  That would make sense if clouds
were responsible, since the extinction in larger in V-band than I-band.
I couldn't find a good subset of the frames; by that, I mean that
couldn't find a subset which caused the residuals to show a
simple "linear + radial" pattern.

  I do think that this approach will work, eventually, but it 
needs a better set of data.  There is at least one other set
of grid frames on DS24 (and maybe two others), so I'll check 
them out soon.

  _IF_ we can define the radial pattern clearly in each camera,
then we can probably apply the correction to all existing photometric
measurements pretty easily.  This won't get rid of the error due
to the linear North/South gradient in the sky, but it will be better
than doing nothing.  Removing the North/South gradient will be
harder, I think, because it probably varies quite a bit from
night to night (and depending on the RA, too, to some extent).

  I anticipate that we can reduce the systematic errors in all
_new_ data to the level of <= 0.020 mag in the near future.
At the moment, this takes some massaging by hand after the regular
processing, but it may be possible to incorporate these corrections
into the pipeline.  

  Of course, since classes resume tomorrow morning, it will probably
be weeks before I can manage to do anything more along these lines :-(

                                            Michael