Flats for Pipeline 3 Reprocessing

The flat histories are being reviewed for all nights to better determine how to approach the need for canonical flats during Pipeline 3 reprocessing. This is a status report for the results so far.

At present it seems reasonably possible that single canonical flats averaged from all good flats during a given hardware period will give good, consistent results for extended periods of up to several months. Currently all southern J flats have been examined by choosing a reference canonical flat from within a hardware period and subtracting it from all of the individual nightly flats in that period. Conclusions from this preliminary check follow:

Typical night-night variations are less than +/- 1%
Systematic variations seen across a given hardware period are also typically less than +/- 1%
The single largest discrepancies are seen between evening and morning flats

Some date ranges seem worse than others
Variations can be of order +/- 2-3% and are typically seen as gradients across the corners
It is unclear whether either the evening or morning flats can be considered more "accurate" than the others, though evening flats (the most common variety) are taken while the detectors drop down from saturation.

It appears that for at least J band, using grand canonicals averaged from the entire period will give a good representation of the flatfield for that period. By averaging in both morning and evening flats, it may be possible to "split the difference". Another possiblility would be to attempt to average in equal numbers of morning and evening flats, but for some hardware periods, very few morning flats were taken, so it would be hard to adopt this policy consistently.

Below I illustrate a typical evening/morning discrepancy. The first date in the image is the night of the flat, the second is the night of the reference canonical that has been subtracted.

The image on the left is a morning flat while the one on the left is evening. The reference canonical used evening flats.

Below are four flats referenced to the same canonical spanning just over 4 months. There is little evidence for strong long-term evolution of the responsivities of the whole array, though a scattering of pixels do show a bit of evolution. Much larger uncorrelated night-to-night variations can be seen on various dates:

I've included the entire sequence of southern J flats, encoded in 5 Quicktime movies to make the disk requirements manageable. They may be accessed below:

J Flats 980319s-990116s
J Flats 990118s-990503s
J Flats 990510s-990927s
J Flats 991002s-000330s
J Flats 000418s-001031s

R. Hurt, last updated 11/14/00