On the Bias Jumps

-Eli Rykoff

Something strange is going on with the bias jumps at 2048...

I'm taking as an example 180985, CCD 41, somewhat arbitrarily.

I use swarp to subtract the sky and rebin to a 2" scale to highlight the sky levels. This is the finalcut image which shows the problem:

On the Flat Fields

Below if you squint you can just see the bias problem in the flat. But there has to be a better way to visualize this. And there is! I took the offending flat:

Note this is problem on all the chips; chip 41 is chosen just as a random example of the problem.

And calculated the ratio between that and another (hopefully normal) flat:

And the bias shift pops out plain as day.

So...where did this come from?

I looked at the header of the t0227 offending flat and downloaded all 91 of the input images. Visual inspection shows a slew of them (00182422 - 00182431 inclusive) are not healthy images.

The gradient on this "flat" goes from 16000 - 1600, a factor of 10!

But... are 10% of the images being bad in a superflat enough to cause this sort of problem? And why does this show up as a bias shift in the middle?

Ahhh...Aaron Roodman had a great idea. What's happening is that the median in the superflat is removing the bad image pixels ... EXCEPT in the middle, where the gradient crosses over (I'm assuming each image is normalized before taking a median?). And then it slightly contaminates each side, making it look like a bias shift.

Original Images: Ignore the horizontal stripe which is an artifact of the sky subtraction

But if we go back to the raw_obj image (same scaling) which has had DECam_crosstalk run on it then it looks fine! (Well, it could use a good flat-fielding).

Similarly, the raw-raw image looks fine (though it's hard to get the scaling right because of the dynamic range).

So this is being inserted into the image somewhere along the line...Alex suspects the flat field...

Looking at the flat field applied to this image:

And rebinning we see:

A-ha! A bias jump made its way into the flat field and has now corrupted the downstream images.