In nominal ops campaign 2 we executed the flatfield and dark sequence four separate times, at intervals of roughly 3 days. I have examined the resulting skyflats, primarily to ask the following question: is there any evidence for variation in the flats over time, and if not, do we need to take this many?

The data was all reduced in the standard fashion. The flats were processed via a median combine with outlier rejection. A matching set of skydark data, taken within a few hours (usually about 6 hours earlier) were subtracted. The results were normalized to 1. I took the ratio between each flat, and the next in time. These results are shown below. Anything appearing white was higher in flat N vs flat N+1. In order to ignore the latents, I have looked at statistics in rectangular subregions that appear to be latent free.

Unsurprisingly, all of the ratio images have means of 1+/- 0.001. Given that they were normalized to 1, this was the expected result. There are no noticeable gradients indicating substantial differences between the flats taken at different times, with the exception of the first channel 3 flat. The following table gives the measured pixel-pixel dispersion in each ratio pair:

For the ratio a/b, if a and b are near 1 and the dispersion in a and b are similar, then sigma(a/b) is roughly root(2)sigma(a). We then get that the intrinsic dispersion in the flats is about 2% in channel 1, 1% in channels 2 and 3, and about 0.3% in channel 4.

Note that it is a little hard to inperpret the latents, since they can arise either in the flat data, or in negative from latents present in the skydarks that were applied to the flat data. This gets even more confusing when looking at the ratios, which adds another sign flip. However, an exmaination of the actual flatfield data frames indicates that the latents seen here are generally a result of actual latents in those frames, not in the darks.

Channel 1 - this channel has significant latent contamination in every instance except the third. The first flat was badly contaminated by observations of Polaris.

Channel 2 - this channel shows no evidence of time-variability. All of the flats are latent free. This channel by far is the best cal

Channel 3 - the first flat is noticeably different from the following three, all of which strongly resemble each other. In all likelihood this slope is due to changes in the dark offset.

Channel 4 - the flats have significant latent image issues. The annealing of channel 4 erased the latents every 12 hours, but new ones built up during the PAO that the flats were in.

In the end, the decision was to activate only the third flat for channel 1, all the flats for channels 2 and 3, and the last two flats for channel 4.

Several changes are being made to the ordering of the calibrations starting in NOMOPS4. The most important of these regards the timing of the flat-fields. In IOC the worst latents we saw (particularly in channel 4) were triggered by the flat-field sequence. Hence, in operations we placed the flats up against a downlink, thus annealing any flat-generated latents. In reality, we are building up enough latents during a given PAO that the odds of getting a decent flatfield, a highly critical calibration, were low. Starting in campaign 4, we will be doing:

downlink
anneal
darks
flats
science 
calibrators
downlink

in that order. This should ensure that we execute the dark and flat sequences with relatively latent-free detectors. After this goes into effect we can re-evaluate the state of the calibrations. Right now we need to take as many as we are just to ensure a non-zero probability that at least one set of channel 1 flats will come out usable.