Plugging CystallBall into WSclean

[AstroRipples]

Hi folks.

Maybe not necessarily the right place for it, but can't hurt to ask...

Francesca Loi clued me into CrystalBall as a way around a problem I'm having with WSclean. I'm trying to continue a previous WSclean run that I'm performing at high spectral resolution (i.e. using a large value of -channels-out) to better model the PSF over the bandwidth and mitigate the classic ring-like structures you see when you're not fully exploiting having a massive fractional bandwidth.

I've managed to get CrystalBall to successfully predict the source list into the MODEL_DATA column, so you'd think I'd be all set up to continue deconvolution, but no... WSclean is still trying to predict the model (which has some NaN/infs because of flagged channel ranges) which is exactly what I was trying to avoid in the first place... I can't be the only one who's come across this, so I thought I'd ask here if anyone can check my inputs. Maybe there's something I'm missing, but what I'm trying to run is:

wsclean -continue -verbose -no-update-model-required -log-time -j 48 -abs-mem 256 -temp-dir TempDir \
    -reuse-psf ${NAME} -reuse-dirty ${NAME} \
    -pol I -data-column CORRECTED_DATA_DI02 -niter 100000 -nmiter 2 -no-reorder -clean-border 1 \
    -multiscale -multiscale-scale-bias 0.65 -multiscale-gain 0.1 -multiscale-scales 0,5,17,25 \
    -channels-out 24 -join-channels -fit-spectral-pol 4 -save-source-list \
    -weight briggs 0.0 -size 16000 16000 -scale 0.6asec -padding 1.5 -no-negative \
    -weighting-rank-filter 3 -mgain 0.8 -auto-threshold 0.5 -local-rms -use-wgridder \
    -name ${NEW_NAME} \
    -fits-mask ${FITS_MASK} \
    ${MS_LIST}

I've found that I get the same issue regardless of whether I include the -reuse-psf and -reuse-dirty options so I'm a bit stumped 🤷

[IanHeywood]

Won't subtracting crystalball's MODEL_DATA prediction from the data column that you're imaging, and then not using -continue have the desired effect?

[SpheMakh]

@IanHeywood is correct, the option you should be using is -subtract-model not -continue see below documentation from WSClean

-subtract-model
   Subtract the model from the data column in the first iteration. This can be used to reimage
   an already cleaned image, e.g. at a different resolution.