Flat-field structure

Record here investigations into the structure of flat fields, particularly the unexpectedly large radial gradients

Tape Bump Temperature Dependence

Thanks to help from Liz, I retrieved data for the 4 focal plane temperature sensors from 1 Nov 2012 to 9 Feb 2013 (these data are stored every 10 minutes in the telemetry database). I searched these data to identify all of the dates with temperature changes of >2 degrees and identified three temperature change events:

  • 2012-12-26
  • 2012-12-30 through 2013-01-03 (camera thermal cycle)
  • 2013-01-09

The tape bumps appear to be stable in the intervals between these events. For example, ratios of flats on these four pairs of dates do not show evidence of changes: 1 Nov 2012 to 22 Dec 2012, 28 December 2012 to 29 December 2012, 4 January 2013 to 8 January 2013, and 13 January 2013 to 3 February 2013.

The bottom line is that the tape bump features are stable, provided the focal plane temperature changes by less the 2 degrees. If the temperature does change, the tape bump features will stabilize once the temperature does. However, the new equilibrium will be different from the previous equilibrium.


  • Track the date/time of each significant (>2 degree) change of the focal plane
  • Only apply flats to data if no significant temperature change has occurred between their (respective) acquisition times
  • Only construct superflats from flats that do not bracket a significant temperature change
  • Flag pixels in the tape bump features, as objects in these regions may have less reliable photometry and shapes

-- Paul Martini, 11 Feb 2013

Tape Bump Evolution

Here is how the tape bumps evolved through January:

Evolution of tape bumps through Jan 2013

This figure shows the ratio the same tape bump feature shown below from 4 Jan through 3 February. The main point is that the tape bumps stabilized to their current (early Feb) values between 8 and 13 January. This corresponds to about 2 weeks after the thermal cycle. The features from Jan 4-8 are also consistent with one another, so the features actually appear to have changed abruptly between 2 equilibrium values between Jan 8 and Jan 13, rather than steadily changed over time.

For comparison, here are the same ratios (and same tape bump) in the Y-band flats:

These images are about 160x160 pixels and the grayscale extends from 0.985->1.005 (the mean in the ratio of the flats is 1). The largest deviations in the tape bumps are 3-4%.

I divided several sets of g-band ratios by the corresponding Y-band ratios (e.g. the ratio of g-band flats from Jan 4/Feb 3 divided by the ratio of Y-band flats from Jan 4/Feb 3) and the 'sharp' features almost completely vanish, but there is still some residual difference. This is also apparent by close inspection of the g and Y band flat ratios.

-- Paul Martini, 5 Feb 2013

Update on Tape Bump Features

This is an update to the Change in the Tape Bump Features listing below.

I calculated the ratio of the flatcor output for g-band, chip 4 on a range of dates. The figure below shows a zoom in on one of these features (centered at approximately 90, 2096) for the ratio of dates indicated in each panel. Note that the first two rows are approximately a week apart, while on the third and fourth rows the date separation progressively increases to about a month and then two months.

Grid of flat field ratios centered on one tape bump

These images are about 160x160 pixels and the grayscale extends from 0.985->1.005 (the mean in the ratio of the flats is 1). The largest deviations in the tape bumps are 3-4%.

These figures indicate that there was a change between Dec 7 and 28, as well as the well-known one between Dec 28 and Jan 4 (the warm-up). These features appear to have stabilized by Jan 22 and have remained unchanged since then. While I have not looked at all CCDs on all of these dates, some spot checks indicate that the behavior seems consistent across the chips.

Another note: the average value in these features is 1, which implies flux is conserved and suggests an electric field effect that is analogous to, but more localized than, the glowing edges. Given their small-scale nature, it does not seem practical to correct them with the WCS solution and these pixels should be flagged.

-- Paul Martini, 4 Feb 2013

Change in the Tape Bump Features

The SN scanners reported a change in the 'tape bump' flat field features with time. This report includes an example that John Marriner reported on at the 9 Jan 2013 DESDM meeting.

I compared flat fields obtained before and after the thermal cycling of the camera at the beginning of January and these features are quite different on many (but not all) chips. Here is an example that shows the ratio of flats (flatcor output) from 20130108212414_20130107 to 20121124213907_20121124 (left) to the ratio of 20121119091456_20121118 to 20121124213907_20121124 (right). The first ratio compares chip 4 in g-band before and after the cool down cycle (Jan 7 to Nov 24) and the second ratio compares the same chip/filter combination with data obtained about a week apart in November. Note that the tape bump features are not apparent in the ratio of the two November flats. The full range on both grayscales is +/- 1%.

Ratios of two pairs of g-band dome flats for chip 4

I spot checked some other flats (again the flatcor output) obtained closer in time to the thermal cycling of the camera (ratio of late December to early January) and the tape bumps are similarly apparent in the before/after flats. The tape bumps are also apparent in the ratio of flats obtained in November and December. The tape bumps are not apparent in the ratio of flat obtained on 4, 5, and 7 Jan, which indicates the camera equilibrated relatively quickly in this respect.

My preliminary conclusion is that these features are stable over timescales of about a week (provided the camera is not thermally cycled!), but change on progressively longer timescales.

-- Paul Martini, 13 Jan 2013

9 Jan Vinu Vikraman

The star (illumination) flat for griz filters using the standard stars. The method is described in the document

13 Dec Vinu Vikraman

Figures show the ratio of star (illumination) flat to dome flat for griz filters.

27 Oct 2012 Gruendl

There is a utility present within the CP software stack (xchipfltnorm) that builds normalized flat fields built by mkflatcor. When mkflatcor runs it inserts into the header a value, SCALMEAN, which is the mean of scale values applied to the individual images that were combined by mkflatcor. xchipfltnorm takes a set of flats and uses these values to scale individual flat images relative to one another.
The code should be present within the postgres branch of ImageProc (it has not been merged in the DESDM stack). If there is interest I will see about merging it into the trunk (probably in imdetrend) and will beef up its documentation.

12 Oct 2012 GaryB

Here is an image of the flat fields of all filters (ugrizY and open, from left->right, top->bottom). Each amplifier has had a multiplicative factor applied which minimizes the RMS difference between this amp and its neighbors along strips that are 40-80 pixels from the edges. This edge-matching will do a crude job of equalizing the gain * QE variations between chips & amps. It will not be quantitatively exact since (a) it assumes QE is constant within an amp, and (b) it assumes that the matched strips have the same illumination, which they don't, because they are substantially seperated. In any case we get some view of the illumination pattern from the domes. The section along the bottom shows the ~12% center-to-edge gradient in illumination that has not been anticipated by any optics modeling.

7 Oct 2012 Jim Annis

See report on radial variations in flat fields and stellar response at docdb-6684