Mask CRs in regions with 1-pass in some bands

[djschlegel]

Starting with DR8, CRs are identified only by the outlier-rejection comparing repeated observations in the same filter. We therefore no longer reject CRs where we only have a single observation in that filter. We should consider falling back to the CRs identified in the CP mask for those cases. An example is attached below near RA=229.602,Dec=-3.5768 .

[djschlegel]

In these cases with only 1 image in a filter at any pixels, we should use either the CP ood mask or the new cosmic masks being tested by Keming Zhang.

[moustakas]

For the record, here's this patch of sky in DR8-- http://legacysurvey.org/viewer?ra=229.6020&dec=-3.5769&layer=dr8&zoom=16

And below is the DR9-reprocessed image coadd with the dr9d container. Unfortunately there's still some junk. I need to dig and see if the ood image mask is being used correctly.

[moustakas]

The pipeline (currently only in the dr9d-dev branch) is doing the right thing as far as I can tell. We utilize CP-flagged cosmic rays in all reductions >4.8.0-- https://github.com/legacysurvey/legacypipe/blob/dr9-dev/py/legacypipe/decam.py#L67-L71

This image is on the edge of the imaging footprint, so I propose we close this issue. @djschlegel?

[dstndstn]

Closing unless folks are concerned enough about the edges of the footprint. If someone were interested in testing Keming's code in prod, that could be an option.

[djschlegel]

Let's check that the CP masks are in fact being used in the case of a single image. For example, in the dr9d reductions there's a long CR here that then shows up in the catalog as a bunch of red sources: http://legacysurvey.org/viewer-dev/?ra=266.8144&dec=17.6608&layer=dr9d-south&zoom=16&ccds8&bright The image is decam-553769-S9-z .

[djschlegel]

This may be somewhat worse in DR9 than DR8, since the CP re-reductions are less aggressively masking, especially for 90prime. The fraction of "cosmic ray" (CP mask=5) pixels has been reduced in 90prime & mosaic, and increased in decam: 0.20% -> 0.06% in mosaic 0.09% -> 0.01% in 90prime 0.03% -> 0.07% in decam (The above stats are from a random sampling of images.)

[moustakas]

Here's another example of poor CR/outlier masking-- http://legacysurvey.org/viewer/?ra=209.1709&dec=36.2504&layer=dr8&zoom=15

[moustakas]

Let's check that the CP masks are in fact being used in the case of a single image. For example, in the dr9d reductions there's a long CR here that then shows up in the catalog as a bunch of red sources: http://legacysurvey.org/viewer-dev/?ra=266.8144&dec=17.6608&layer=dr9d-south&zoom=16&ccds8&bright The image is decam-553769-S9-z .

@djschlegel here's where we are in this chunk of sky with the current code base. @dstndstn and I have verified that we're using the correct CP mask bits, and unfortunately we do end up with (here, 4 very red) spurious sources.

Unless you have a specific proposal, I propose we close this ticket (or push to DR10).