Reducing the VIRMOS CFH12K data: astrometry problems

Here's a quick report on what's been happening.
I was working on the central 4 pointings of the F14 VIRMOS images using the following procedure:
  • Assemble all the relevant images
  • Use astrometrix (TM) to correct WCS in the headers
  • Use photometric (TM) to set FSCALE in the headers
  • Use swarp (TM) to coadd the images.

I would work on all the CCD's of all the exposure of all the pointings simultaneously, trying to produce a mosaic of all 4 pointings. The first image (the F14I) was stacked by Henry McCracken. The astrometry in that catalog is based on the USNO catalog. I worked on the F14B and F14V data, using a catalog generated from the F14I as astrometric standards.

This procedure results in a F14B image that is perfectly registered to the F14I over most of the image. However there are certain areas where the registration is not so good as shown below. I believe that this poor registration is caused by the dither when the images were taken which has resulted in very small overlaps between the CCD's within a pointing, not to mention the small overlap between pointings.

The (hopefully animated) figure to the right one of these areas. One of the frames is the F14B; the other is the F14I (as labeled in the upper left hand corner). Clearly, although part of the images are properly registered the central part is not. It is also obvious that there is imperfect coverage in the F14I image, in the form a small L-shaped blank area.

(In Netscape, you can hit *ESC* to get rid of the animation)

The figure to the right shows the location of this offending area as a blue dot and circle. The boxes shows the location of the CCD's going into the mosaic. The different colour indicate different pointings. The mismatched part is at the join of 4 CCD's.

As above, but zoomed in to the mismatched area. The small overlap between the CCD's is more obvious.

The following plots show the the 4 central pointings of the F14 field in B. The black dots show sources used to compute relative photometry between the different exposures. The green dots show sources used to compute relative photometry between the different pointings. The pointings are labeled in red. The number of sources common to two different pointings is indicated by a number between the pointings.

Overlapping sources for the four central pointings of F14B field. There are clearly no overlapping sources between the P1 and P2 pointings. There is also no connection between P3 and P4. The left image shows that there only one source common to both the P1 and P3 pointings.

The situation is improved if one includes the sources in CCD00, as shown in the figure to the right.

CCD00 of the CFH12K imager is of course affected by the "brickwall" effect in the flat field, which means it is not appropriate to use it when calibrating photometry. CCD06 is never used since a typical image looks like the one at right.
The following table shows the number of usable overlapping sources from CCD to CCD between dithers within a pointing. (in this case pointing 3 of the 14h field, in the V band) The total number of overlapping sources is shown. There are a total of 7 exposures. As one can plainly see, there are a large number of overlapping sources for a given CCD on different exposures. In most cases however there are no sources common to two different CCD's.
   :     0     1     2     3     4     5     6     7     8     9    10    11
  0:  3316     0     0     0     0     0     0     0     0     0     0     0
  1:     0  3258     2     0     0     0     0     2     0     0     0     0
  2:     0     2  3696     1     0     0     0     0     0     0     0     0
  3:     0     0     1  4116     5     0     0     0     0     1     0     0
  4:     0     0     0     5  4006     0     0     0     0     0     0     0
  5:     0     0     0     0     0  5920     0     0     0     0     0     0
  6:     0     0     0     0     0     0     0     0     0     0     0     0
  7:     0     2     0     0     0     0     0  5706     3     0     0     0
  8:     0     0     0     0     0     0     0     3  3738     2     0     0
  9:     0     0     0     1     0     0     0     0     2  4496     2     0
 10:     0     0     0     0     0     0     0     0     0     2  6209     0
 11:     0     0     0     0     0     0     0     0     0     0     0  4026


Astrometry: I suspect (although further tests are necessary to show this) that including objects from adjacent data frames adds little to the astrometric solution and may even be adding noise.

Photometric calibration: Clearly if there are no overlapping sources, it is futile to try and do the photometric calibration between CCD's or between pointings. It is better then to select one exposure as being photometric, flag each CCD of that exposure as photometric and transfer the zero-point to the same CCD's in other exposures.

Source Extraction: In the gutters between the CCD's, there is considerably less coverage. This means there will enhanced noise, and consequently, these areas will be shallower than the central parts of the CCD's. This makes it undesirable to include these sections of the image in any analysis where some sort of strict magnitude limit is required.

In general, because of these small overlaps, it seems to me that it not optimal to combine all the CCD's from one pointing into one large image, or combine multiple pointings into an even larger image, but instead to reduce each CCD of each pointing separately. Ultimately, the final swarped images of each CCD's can swarped together to make a large image, if this makes it easier to run SExtractor, and/or improve the book-keeping.