This is an update 25 July 2013 from Gary on comparing star flats derived by me (penn), Nicolas Regnault (paris), and Brian Yanny from Anne Bauer's code (fnal). I won't go into details here on making these flats. But here are some specifics about these particular versions:
- From the Dec 2012 star flat exposure series.
- Using catalogs reprocessed at run 20130612140044_20121222
- The penn plots below are forcing the stars' MAG_AUTO to agree across exposures.
- I'm not sure which magnitude the fnal star flats use.
- The paris star flats are made with Nicolas's software and may be based on different dome flats.
- The penn flats allow a 4th-order polynomial per CCD.
- paris and fnal define star flats in superpixels of 1024 (512) pixels square, respectively.
- All the plots below show the focal plane in a gnomonic projection about the telescope axis on the sky. N is up, E is to the right.
- Everything I show will be in magnitude units.
Here are the three star flats, each set to zero mean, with the same color scale of +-0.05 mag. grizY are from left to right. They look very similar, which is good!
Penn vs FNAL star flats
Below is the difference between the penn and fnal star flats. I have removed the mean since the absolute level is irrelevant. I have also removed any linear gradient that may be present across the FOV, since there is a degeneracy between this and the zeropoints of the starflat exposures which can be difficult to remove. The left-hand column gives the histogram of differences, which are plotted on the right side. There is ~3 mmag RMS difference except in i band. The i-band differential resembles the color term of the star flat, so it is possible this difference is because penn and fnal star flats use different reference colors. But it would need to be 1 mag different, which seems unlikely.
The half-chip that is discrepant at the right-hand side is #31, which has time-variable offsets on one amp.
Penn vs Paris star flats
Here is the difference between the penn and paris star flats. The RMS is a little larger. The g-band RMS can be reduced to a bit, to 3.8 mmag RMS, by applying ~0.25 mag shift of the reference color of the penn flat. Some fraction of the RMS here is from the "scalloping" visible where the flat has gradients across the larger superpixels. It might be worth recalculating with smaller superpixels, then take another look.
This plot shows the difference between the star flat derived from Feb Omega Cen data and the December data. The g-band differences are large, and Y is larger than is comfortable. I next need to check if the dome flats used in Feb differ in a related way from those used on the December data.