Gaussian source models

[nicholebarry]

Gaussian source models are off by about 10%. If a standard scaling factor in both RA/DEC widths is applied at a constant 5-10%, the residuals are better (for both near-zenith and off-zenith Fornax models). Since DEC is well-behaved in a x-y coordinate system, this indicates that there is some sort of proportionality constant/conversion that is missing (as opposed to a problem in just RA, which does not map well to a x-y coordinate system). See pull request #143 for plots.

[miguelfmorales]

I am wondering if this is an array filling factor issue. Essentially when the Gaussian model was made from the dirty image, it implicitly assumes the dirty image was made from a filled uv plane.

If this was the case the visibilities would be right based on the model, but the round trip of dirty-image, Gaussian components, visibilities, dirty image would have an amplitude error. This error would be implicit in the 1-2 step. One test is it would get worse as you extended the maximum baseline (keeping everything else equal), as the filling factor of the MWA would fall.

Within FHD point sources have an associated normalization factor (which might even be being applied here), but the appropriate factor will change with the size of the Gaussian. Essentially the spatial pattern of the Gaussian talks with the spatial pattern of the filling factor.

Just throwing mud at the wall.

[rlbyrne]

@nicholebarry is this resolved with #291?

[nicholebarry]

I don't remember... I'll queue up a job and let you know