Responsivity Map Comparisons

Responsivity Map Comparisons

Summary

o Ground flats from MIC2 were compared to a set of flats made from flight
  data taken Dec 29 (early set) and Jan 4 (late set). The early and late
  flats were also compared against each other to explore stability.

o Input data were from "early" scans: 00499x, 00500x, 00501x, 00502x, 00503x,
  and "late" scans: 00626x, 00627x, 00628x, 00629x, 00630x.
  
o W1,W2 flats were made using the classic stacking (or standard) method, and
  W3,W4 flats used the gradient (change in zodi) method. This method measures
  the amount by which each pixel responds to the change in overall (median)
  background level.
  
o Overall, there are significant differences in the low-frequency structure
  of the MIC2 flats compared to flight, warranting a replacement of the
  default flats in WSDS processing, at least for W1,W2.
  
o There are some new hot/low-response pixels in W1,W2 compared to the MIC2
  products. These will be dealt with in a new bad-pixel mask.
  
o Furthermore, there appear to be a family of noisy pixels which fluctuate
  from high-to-low response between the early and late flight flats
  [see blink animations in Sect 1]. It's reassuring that these pixels are
  associated with the largest variances in the uncertainty maps.
  
o The W1,W2 flats made from flight data ~ 1 week apart are remarkably stable.
  [see difference images in Sect 2]. These will be repeatedly made to check
  for stability.
  
o Excluding effects of bright sources and latents in W3,W4, these bands also
  appear to be remarkably stable in their relative pixel-to-pixel response.
    
o Bright sources, streaks, and latents therefrom in W3,W4 are a menace and
  make it tricky to construct and match flats to science data in a moving
  window. It's not known whether we should make these on a per-orbit
  basis (with probably a special set for data close to anneals) or on some
  longer timescale. The goal is to minimize damage to data that do not
  contain the same latents contaminating a flat. For example,
  suppose one constructs a flat from N frames, K of which contain long-term
  latents and whose signature survive in the flat. When the flat is
  applied to the N frames, the long-term latents in the K frames will
  (or should) divide out, but we'll be left with depressions in the
  remaining N - K frames. Bright sources + latents come and go, so some
  damage is inevitable if this approach is taken.
  
o Testing the "late" flight flats on a single frameset reveals that there are
  a new class of hot/low-response pixels in W1,W2 cf. to MIC2, and that the 
  flight flats make a significant difference in flat-fielding W3,W4. 

o The remaining issue is: from which W3, W4 data shall we construct a default
  set of "latent-free" flats? Is it worth making flats on a per-orbit basis?
  If the relative responsivity at high spatial frequencies is stable, as
  results currently show, we could aim towards a pristine superflat and let
  the latents do as they please. They'll be tagged by both the transient-pixel
  detector, and the model-based latent module in source space.