I’m running the assembleCoadd task on images from DECam using version w.2022.06 of the pipelines. The images are of very dense fields in the galactic plane. I am finding that for some of the visits and detectors, processing fails due to the variance re-scaling factor (from warp to coadd) becoming too large (>10, the configurable limit). For some detectors, the processing succeeds, but the scaling factor found seems large (from processing logs):
lsst.assembleCoadd.scaleWarpVariance: Renormalizing variance by 6.668616
lsst.assembleCoadd.scaleWarpVariance: Renormalizing variance by 5.655051
and for some the processing fails:
lsst.assembleCoadd.scaleWarpVariance: Pixel-based variance rescaling factor (529.884715) exceeds configured limit (10.000000); trying image-based method
lsst.assembleCoadd: Unable to rescale variance of warp (Variance rescaling factor (527.710998) exceeds configured limit (10.000000)); leaving it as-is
...
lsst.assembleCoadd.scaleWarpVariance: Pixel-based variance rescaling factor (841.338161) exceeds configured limit (10.000000); trying image-based method
lsst.assembleCoadd: Unable to rescale variance of warp (Variance rescaling factor (842.515557) exceeds configured limit (10.000000)); leaving it as-is
...
lsst.assembleCoadd.scaleWarpVariance: Pixel-based variance rescaling factor (732.610963) exceeds configured limit (10.000000); trying image-based method
lsst.assembleCoadd: Unable to rescale variance of warp (Variance rescaling factor (734.059905) exceeds configured limit (10.000000)); leaving it as-is
In comparison for a set of extragalactic (sparser) fields I am also processing, the rescaling factor is <1:
lsst.assembleCoadd.scaleWarpVariance: Renormalizing variance by 0.759862
...
lsst.assembleCoadd.scaleWarpVariance: Renormalizing variance by 0.770071
...
lsst.assembleCoadd.scaleWarpVariance: Renormalizing variance by 0.683306
...
lsst.assembleCoadd.scaleWarpVariance: Renormalizing variance by 0.679214
...
lsst.assembleCoadd.scaleWarpVariance: Renormalizing variance by 0.709469
I am not sure what is causing the problem, but it seems to be quite a normal occurance as ~50% of the detectors face this issue. (The 50% number might be double counting some, I’m not sure.) I’ll be proceeding with doScaleWarpVariance: false to avoid this issue, as it might not be critical for our science case. I saw this post about variance re-scaling which might be good reading for me, but posting this as an “FYI” and in-case anyone has thoughts about this.