Analysis of flat-field images

[awhoward]

The propose of this Issue is to enumerate a set of analysis methods for the DRP that can assess the quality of the flat-fielding procedures. These methods should apply to individual flats, a set of flats (say, 100 flats taken back-to-back), and DRP-produced master flats.

We'll need to consider if these plots should be for individual slices (e.g., SCI2) or all of them overplotted in different colors.

Individual flats and DRP-produced master flats (L1) files:

• Report the measured SNR of each order as (Sam's procedure): 1. Divide per-order L1 spectrum by a smoothed version of itself (say, 100-pixel moving average). 2. SNR = 1 / stddev(middle 1000 pixels of order). These will be a set of numbers, not plots.
• Report the per-order "fractional SNR" = "measured SNR" / "theoretical SNR". These will be a set of numbers, not plots.
• A plot version of the above, with spectral order on the horizontal axis.
• A grid of subplots per CCD like the plot below.
  

A set of flats:

The set of flats for a given day can be determined using the 'time series database' and selecting all files matching OBJECT=autocal-flat-all.

• A plot like the one below from Sam, which is a variation on the measurements above. The difference is that in the plot below, SNR is computed for each subset of the N_total exposures.
• A plot like the one below, which shows a zoomed-in region a set of L1 spectra. This might be done for a 6x6 grid, as we do for the extraction window QLP plots.

Master flats:

• Make the above plot with the blue trace. This assumes that all of the files that go into the master files can be determined.

[awhoward]

Ideas for attacking this problem from the DRP meeting on March 26:

• @bjfultn suggested looking at this in the 2D frames. Make our own rectangular extraction boxes and see if the fixed pattern noise persists.
• Look at the smooth lamp pattern to see if the same structure is there.
• Assess the overscan region to see if the fixed pattern noise is present. @RussLaher said that this is unlikely to be the source because the entire overscan is used to computed clipped mean.

[awhoward]

To produce the plots that depend on knowing which files contributed to a master flat, one needs to look at FITS keywords like these in one of the master flat extensions (partial list shown below):

INFL0   = 'KP.20240322.00007.72_2D.fits'                                        
INFL1   = 'KP.20240322.00086.26_2D.fits'                                        
INFL2   = 'KP.20240322.00165.11_2D.fits'                                        
INFL3   = 'KP.20240322.00243.89_2D.fits'