Moustakas and I encountered some somewhat incomprehensible results on his LDSS3 images: We find that the twilight flat (the estimate of the pixel-to-pixel sensitivity of the CCD based on images taken of the bright sky at the beginning or end of the night) does not properly flatten the night sky in the science frames.
This incident reminded me of the great pain we suffered on SDSS when we realized that it is not just dangerous but also wrong in principle to base your (large-scale) flat estimate on twilight or sky levels; sky level variations can have both multiplicative and additive components! Unfortunately, I don't think this is the problem with Moustakas's images, and unfortunately the observing plan on the LDSS3 run (not made by Moustakas and not executed by him) did not involve enough calibration data to get at the root of the problem.
My suggestion: mask out the troublesome parts of the images! Yes, it is throwing away data. However, if there is one lesson I have learned abundantly when taking imaging data it is that you never, ever, ever take enough calibration data. Calibration data take time, so they reduce the amount of science data you get, so you tend to skimp; but calibration data make it possible to use all of the science data you do take, and to use them optimally.
Me commenting on old posts!
ReplyDeleteThe calibration issue is really an important thing, but the problem also gets to TAC's philosophy! If you put into a telescope time proposal that you need around 20-30% of the telescope time to take calibration frames (e.g. in my case, long exposure "blank" sky frames, taken in different times of the night) they will probably chop this fraction of your project's awarded time ...