Conclusion: STATCONT works and we will use it for spectral baseline/continuum subtraction
[Nakanishi]: Still investigating difference between order of processing: STATCONT+PBCOR or PBCOR+STATCONT
[Nakanishi]: Doing PBCOR+STATCONT is much easier to do in processing
[Nakanishi]: I made some tests with auto-masking algorithm of tclean, in which three different ways of making mask are tested.
[A] frequency-independent mask; auto-masking is made on the bases on the frequency-integrated image
[B] normal (frequency-dependent) automask
[C] 5ch-binned mask; auto-masking is made on the 5ch-binned image to obtain higher S/N
The resultant images of (B) and (C) are almost same, so probably we do not have to use (C), as it takes longer time than normal automask.
The frequency-independent mask (A) is also good, but negative sidelobe patterns are less removed because the spatial shape of the negative lobes in the dirty image is also frequency-dependent.
The continuum image appears unaffected by the choice of masking algorithm.
I also quickly checked if the extended features that I reported at the previous telecon are real or not, by comparing CO 3-2 image with the CO 2-1 image.
The positive-latitude features are likely to be real, but the negative-latitude feature does not have obvious counterpart in the CO2-1 image. I am now trying to see if this feature becomes weak if selfcal is applied.
[Nanase]: Some low-level issues with ~0.5% of pixels