From: 	Frank Masci <fmasci@ipac.caltech.edu>
	Subject: 	new light on image-derived position biases
	Date: 	February 21, 2012 5:59:30 PM PST
	To: 	Ken Marsh, Timothy Conrow, Roc Cutri, Tom Jarrett, 
                John Fowler, Howard McCallon

Hi all,

I have revisited the image-derived source position astrometric bias problem and analyzed the outputs from different image co-adders (a task initiated last December). Putting all the pieces together, I have isolated the origin of the problem. This doesn't mean I have an explanation, but finding its origin is a start. If you recall, our tweak to the PSF reference positions used to define the astrometric tie-point(s) "on a point source profile" certainly helped eliminate the W1 image-derived position biases (and to a lesser extent W2), but not for W3 and W4. Maybe a little in W3 and W4, but not enough.

This may not come as a surprise, but the presence of non-negligible systematic errors in the W3 and W4 distortion and/or band-to-band calibration appears to have an adverse effect on *all-coadders*, independent of the pixel-interpolation method used. The non-linear transformations in the distortion/reprojection routines are amplifying these systematics in a way that I don't understand. Again, the fact that coadd-derived source positions in W1 (and somewhat W2) agree well with the astrometric catalog positions tells me (i) the WCS _and_ distortion calibration in these bands is pretty good, and (ii) all co-adders are capable of carrying out the transformations if all is "perfect" going in.

I've summarized my findings here:
http://wise2.ipac.caltech.edu/staff/fmasci/coadd_comp.html

Regards, Frank