Continuum imaging and self-calibration

[relevant workers: transform, flag, selfcal, mask]

Split, average and flag target visibilities

Following cross-calibration, CARACal creates a new .MS file which contains the cross-calibrated target visibilities only. This is done by the transform worker. In case the cross-calibration tables have not been applied to the target by the crosscal worker, transform can do so on the fly while splitting using the CASA task MSTRANSFORM.

Optionally, the transform worker can average in time and/or frequency while splitting. Depending on the science goals, it might be useful to run this worker more than once. E.g., the first time to create a frequency-averaged dataset for continuum imaging and self-calibration, and the second time to create a narrow-band dataset for spectral-line work. The possibility of running this worker multiple times within a single CARACal run allows users to design the best CARACal workflow for their project.

Before self-calibrating it might also be good to flag the target’s visibilities. (Typically the target is not flagged before applying the cross-calibration.) This can be done with the flag worker (which was probably already run on the calibrators’ visibilities before cross-calibration) setting flag: field to target.

Image the continuum and self-calibrate

Having cross-calibrated, split, optionally averaged and flagged the target, it is now possible to iteratively image the radio continuum emission and self-calibrate the visibilities. The resulting gain tables and continuum model can also be transferred to another .MS file (particularly useful for spectral line work). All this can be done with the selfcal worker.

Several parameters allow users to set up both the imaging and self-calibration according to their needs. Imaging is done with WSclean, and the parameters of this imaging software are available in the selfcal worker. Calibration is done with either Cubical or MeqTrees, and also in this case the selfcal worker includes the parameters available in those packages.

Additional parameters allow users to decide how many calibration iterations to perform through the parameter selfcal: cal_niter. For a value N, the code will create N+1 images following the sequence image1, selfcal1, image2, selfcal2, … imageN, selfcalN, imageN+1.

Optionally, users can enable selfcal: aimfast, which at each new iteration compares the new continuum image with the previous one and decides whether the image has improved significantly. In case it has not, no further iterations are performed. In this case therefore selfcal: cal_niter is the maximum number of iterations.

While imaging it is usually convenient to identify where to clean. Within CARACal this can be done in several different ways through the parameter selfcal: image: clean_mask_method:

• with WSclean automated masking method, which cleans blindly down to a masking threshold, defines the clean mask as the ensamble of all cleaned pixels, and then re-cleans them down to a deeper clean cutoff;
• with SoFiA, which makes a clean mask for the Nth imaging run from the (N-1)th image;
• with a clean mask made by the mask worker or supplied by the user.

Several parameters allow users to control the calibration step in the selfcal worker. Users can set the time and frequency solution intervals. Gain phase and amplitude can both be solved for, each with its own time and frequency solution interval (more standard phase-only self-calibration is also possible). We refer to the selfcal page for a full description of all available modes and parameters.

Gain and model transfer

If the self-cal loop was executed on a frequency-averaged .MS file, it might be necessary to transfer the resulting gains and continuum model back to the original, full-frequency-resolution .MS file. This is done with selfcal: transfer_apply_gains (using Cubical) and selfcal: transfer_model (using Crystalball), respectively. The latter allows users to limit the model transfer to the N brightest sources, to sources in a region, or to point sources only. Be aware that the model transfer step can be very time consuming for large .MS files.