The Importance of Resolution

A quick examination of the Tables will show that all of these quality indicators change significantly with the resolution. This is unfortunate but not unexpected. At a technical level, all but the DR depend sensitively on the difference image, which (given a complex source, and a CLEAN model which can change rapidly from pixel to pixel) can easily change considerably as the simulation and truth image are convolved. Both the peak and the rms noise will also change, and so naturally will the DR as well. A more sophisticated explanation is that the resolution effectively selects the range of Fourier spacings being used, and that there is no particular reason to suppose that a model which matches one range of spatial frequencies should match another set equally well. CHU proposed (and Holdaway (1994) implemented) the use of the visibility SNR curve to show the Fourier transform of the difference image, to show directly where in spatial frequency the imaging is going wrong. I have never found those plots very intuitive, and so have taken the approach of simply convolving the images to be compared to the same resolution before calculating and comparing their image qualities. From here on, unless otherwise noted, all comparisons refer to images convolved to the same resolution.