Open tony-johnson opened 7 years ago
We have run into a problem with this. When the IQ is too good we end up with artifacts in the matched images. This results in unusable coadds. I am not sure if this is because of numerical instability as the PSF gets close to critically sampled or some other bug.
I think the thing to do in the short term is to use a window in QI to make the coadds.
@SimonKrughoff Yes - now issued at #435
Hi @tony-johnson @SimonKrughoff
At the DESC meeting Twinkles session we decided to use the Year 1 visits to make the good seeing template image, and then difference all subsequent (Year 2 - Year 10) images against these 6 templates. It seems as though Simon's 2.6 - 3.0 pixel IQ window works pretty well, and we'll just have to swallow the loss of all visits with IQ better than 3.0 pixels (0.6") for the Run 3 analysis. What else, if anything, do you two need in order to run the 2-branch DM workflow that we saw last week? How did your reconnection conversation go?
Cheers
Phil
I think things went really well. I think the workflow is in place. The remaining work is to figure out the data that go into each step.
The DM pipeline should be adapted to run the workflow from the newly update Image Differencing cookbook, using 3.00 (pixels) as the value for modelPsf.defaultFwhm and select.maxPsfFwhm.
This should then be run on the entire Run3 output at NERSC.
First step is for @SimonKrughoff to make sure that the cookbook is complete, then Tony will run the workflow.
In more detail, here are the issues in this epic: