Astrometric solution variation

2 April 2014, Gary Bernstein


Vinu Vikram has run SCAMP on all of the star flat exposure series taken during SV and Y1. These fits are done on catalogs produced by Margaret Gelman's runs of a truncated Final Cut pipeline applied on the raw images. These runs have id's 201403******.

Unlike normal DESDM processing, SCAMP was run with STABILITY_TYPE=INSTRUMENT which allows a fresh set of PV terms to be derived for each filter. Also all ~20 exposures in a given filter are matched to each other as well as to a reference catalog. In normal DESDM runs the PV's are frozen, and each exposure is matched only to a reference catalog. In previous processing the PV's were (I think) always held to a single set of values derived quickly during commissioning in September 2012 by me and Mischa Schirmer. This was expected to be a quick-and-dirty solution.

The open questions are:

  1. Is the commissioning solution an adequate representation of the data (e.g. RMS < 100 mas) for purposes of creating DESDM stacks?
  2. Are there differences between the filters' solutions?
  3. For a given filter, does the solution change with time over SV and Y1?


After Vinu derived the solutions with SCAMP, I took the .head files that were produced and mapped a grid of (x,y) pixel positions on each CCD back to positions on the sky using the derived WCS. Then I plotted the difference between pairs of solutions as a vector diagram.

Note that before plotting these differences, I removed the best-fit translation and linear function of sky coordinates from the difference vectors. In other words shifts, scale changes, rotations, and shears are taken out of these plots. This is because (a) we expect these shifts etc to occur from small pointing errors, differential refraction, stellar aberration, polar misalignment, etc., and (b) in the DESDM pipeline, the shift/scaling of each exposure is re-fit by comparison to the reference stars. So these affine transformations are irrelevant to the issue of getting a better nonlinear WCS.

These plots show:

  • Red arrows, which show the mean shift of each of the 60 ccds (CCDs 2 and 61, now dead, are not plotted0.
  • Black arrows show the shift of points on each CCD relative to the CCD centroid. So if you add the red and black vectors you would get the total differential for a given grid point.
  • Statistics in the corner give the RMS and the maximum shift across the array in the x and y coordinates. These are x and y sky coordinates, by the way, not pixel coordinates. So the serial registers are vertical in these plots.

The attached PDF files give these difference diagrams between the different filters and for a fixed filter between all adjacent epoch pairs. I highlight some of them here.

The first plot shows the difference between the commissioning solution and an i-band solution from 20121222. This shows that the answer to the first question is that there are significant errors in the commissioning solution. There are errors of 300 mas in the commissioning solution. The pdf is commission_minus20121222i.pdf

The second plot that I highlight is the difference between the g and i band solutions for 20121222 (see 20121222colors.pdf for the r-i, z-i, and Y-i differentials). There are differences of >300 mas between the bands, directed radially as we would expect from our axially symmetric optical system. The redder bands have less difference than g but it is still worth using a different WCS for each filter.

The third question is about changes with time. The pdf files below give the change in each filter between successive star flat dates. I show here two typical frames, the change in the r band flat between 20131115 and 20141118, and in i band between 20130830 and 20131115. Note that the camera had a warmup in this period. The changes between epochs are small enough (<60 mas) that they can be ignored in DESDM. We can see that some of the chips have rotated, by the ~micron that the mounting design allows (according to Tom Diehl).


I would conclude that it would be wise for us to use different instrument WCS solutions for each filter, but we can use the same ones through Y1. Attached here are 5 ".head" files giving the 20131115 solutions for each filter. These could be used. There are variations of the WCS within a season that the most careful analyses will need (weak lensing shapes, stellar proper motions), but the DESDM pipeline does not need to incorporate them.