The following table of images shows an M31 frame with standard science processing (top left), with the background restored using the camera stage adjusted background models (top right), background restored again but with the PATTERN.CONTINUITIY correction disabled (bottom left), and the background unsubtracted reduction Bill did previously (bottom right). The two background restored images did not have the mask data applied, but have effectively the same masks as those seen on the automated JPEGs.

This profile plot is the same region as those done previously, showing Bill's reduction and the background restored without CONTINUITY (top), and the profile of the background restored/CONTINUITY=T (bottom).

The important points:

  1. Background restoration appears to generate largely the same solution as not subtracting the background at all. This is good, as we expect that we should be able to subtract a model and then readd that same model to return to the initial case.
  2. OTA-to-OTA offsets still exist in the background restored images. Again, this is likely an error in the dark model, and as noted above, we add back what we've previously subtracted. The camera stage background adjustment seems to have not completely fixed this, most likely because this dark error occurs on one edge of the OTA, and the adjustment solution is a function of the offsets on all edges.
  3. PATTERN.CONTINUITY introduces a constant offset on all OTAs, as there is an open degree of freedom. The added adjustments to each cell allows for the OTA-to-OTA offsets to be minimized as well. However, sharp gradients in the image are broken by this correction.

The processed data for all of these reductions (20 exposures) is available for further study.


The following images show the vertical median profile (along pixel columns) of OTA45 and OTA46 after a series of detrend steps have been applied. Three exposures were chosen for this, one from M31 (the middle pairs), and two from 3PI (the top and bottom pairs). The first image shows the raw overscan subtracted image along with the dark corrected image. The bottom image continues from there, adding the flat and pattern corrections. The important points to notice:

  • The raw data shows a cell-to-cell "scoop" shape that is caused by the dark current.
  • Upon dark correction, the "scoops" are largely removed from each cell, although OTA46 (x > 5000) now has a downturn (anti-scoop?) shape for the M31 exposure. I suspect that this is due to the presence of a video cell on OTA46 in this exposure, resulting in a different dark current relative to the dark model. The bottom exposure also has a video cell in OTA46, and I think there's some evidence of a similar overcorrection in that OTA.
  • After applying the flat field correction, the cells of the individual OTAs match each other well. Further adding the pattern.continuity correction seems to do little but shift the OTA background to be slightly more flat.
  • The bumps in the M31/OTA45/x=4700 and bottom/OTA46/x=5900 are from bright features on those exposures that are not fully removed by the median filtering, and should not be confused with issues with detrending.
  • Manually extrapolating the M31 dark corrected edge value to something not "anti-scooped", and then applying the expected flat field correction seems to provide a value that would be consistent with no offset between OTAs (measured OTA46 edge value ~178, flat field corrected value ~151, "correct OTA46 value" ~185 => expected correct flat field value ~157. measured OTA45 edge value ~160).

Since this seems to be an issue with the darks, and more specifically, with the video dark response, I'm not sure that this is going to be resolved until after the next round of detrend creation and verification. As some of this M31 data is older, we may not ever be able to generate a valid video dark for that old data, as we only recently started taking video dark exposures.