I picked 6 of the February frames to check astrometric performance. The frames chosen were all February 12 (to avoid the clouds, clocking and rotation problems of February 2), and were of the 13h-14h RA fields to avoid crowding problems.
The images were all flatfielded using dark-subtracted median sky flats. The frames chosen are:
frame cam fil RA DEC g0493946 0 I 13:37 -01:45 g0493955 0 I 13:50 -01:45 g1493946 1 V 14:36 -01:04 g1493955 1 V 14:49 -01:04 g2493900 2 I 13:25 -00:55 g2493909 2 I 13:38 -00:45
Each image was inspected by hand using iraf and imexamine. A variety of stars across the image were examined. The full-width-half-maximum (fwhm) values as a function of x (Dec) and y (RA) were plotted. There was a correlation between fwhm and x value, with minimum fwhm at low x (high declination) and maximum fwhm at high x (low declination), with a basic linear trend inbetween. Rough values of fwhm are given below, where xbeg = x(1) and xend = x(768) im image space.
frame xbeg xend ave
g0493946 3.40 3.50 3.42
g0493955 3.32 3.60 3.45
g1493946 2.80 3.20 3.00
g1493955 2.80 3.20 2.90
g2493900 3.60 3.90 3.70
g2493909 3.60 3.90 3.80
The camera 1 images all showed definite elliptical shapes, elongated NE-SW. The camera 0 and camera 2 images were comatic.
I used daophot to locate about 200 bright stars in each frame. Daofind was used with a gaussian fwhm of 3.4 pixels; phot was then performed on each frame with various aperture sizes. These two operations resulted in a file with x,y positions for the 200 objects. Daofind positions are typically +-0.25 pixel as the centroiding algorithm is not optimized.
As a further improvement, I used the psf fitting routine PSF, along with ALLSTAR to generate a more accurate psf and therefore better coordinates.
The coordinate files from the PHOT and the ALLSTAR exercises were then compared against the GSC. The 6 plate constant solutions from SLALIB were generated. These gave the following results:
frame scale nstd phot psf
g0493946 13.732 153 2.88 1.75
g0493955 13.723 166 2.86 1.78
g1493946 13.734 141 3.19 1.93
g1493955 13.728 189 3.26 1.55
g2493900 13.765 126 2.51 1.36
g2493909 13.756 123 2.75 1.45
Note that the scales are similar for all cameras. This is primarily
because the clocking is identical for all 3 cameras, thereby making
the RA (y) scale identical.
I also plotted x,y vs delt(RA),delt(DEC) to check for any systematic trends. Examples of these would be pincushion/barrel distortion, which would show up as quadratic/cubic curves in x. No systematics were seen, except in y vs. delt(DEC), where there is a smooth parabola with minimum deviation at each end of the scan and maximum deviation in the middle. The only quick explanation I can give for this is a drift in the VCO.
The astrometric error is larger than I would expect. These images are showing 0.03*fwhm accuracy; we typically get a factor of 2-4 better than that with FASTT. One cause is obviously the psf of the images, as evidenced in the improvement shown between PHOT and PSF coordinates. Also, since the psf varies across the frame, further improvement is possible using a varying psf. No attempt was made to find the optimum centroiding algorithm for these images.
The size of the images was larger than expected. Typical camera lenses give 80-200 l/mm resolution, which is less than one pixel on the CCD. Seeing is typically 3-8arcsec at amateur sites, again less than one pixel. I would expect image sizes more like 1.2-1.5 pixels on a properly focussed TASS camera. Causes could include improper focus or lens aberrations. One test would be to stop down the lens to improve the optical quality, and see what the fwhm becomes.