[daip] IMAGR question

[Eric Greisen]

Eilat Glikman writes:
 > Hello,
 > 
 > I am conducting some tests on FIRST images and have run into a peculiar problem.  
 > My project involves stacking images of radio-quiet quasars from the FIRST survey 
 > to see if we can detect these objects and study their average properties.  This 
 > is an effective technique and we are able to detect a source when stacking 
 > thousands of cutouts.  
 > To calibrate these results we put 100 40uJy sources using UVMOD into FIRST UV 
 > files and rand IMAGR in the same way as the FIRST survey does to make its maps.  
 > As a control group we also input 100 4mJy sources with UVMOD and process the 
 > data in the smae way.
 > Our initial tests put the points in a 10X10 grid.  This resulted in a 60% bias 
 > in the stacked flux (inputting 40uJy sources resulted in 25uJy sources).  We 
 > repeated the tests on 80uJy, and 200uJy sources.  Since these sources were below 
 > the cleaning level (FLUX parameter in IMAGR is set to 5.0e-4 Jy) the stacked 
 > cutouts showed the dirty beam.  
 > We then hypthesized that perhaps the regular shape of the grid caused sidelobes 
 > to overlap and may have introduced this bias.  To correct this I made 100 
 > randomly positioned sources and repeated the test.  
 > The results of this test are even more bizarre: the 40uJy sources produced a 
 > similar bias of 60% reduction in fitted peak flux (using jmfit)
 > we also made a test with 4mJy sources and they too exhibited a bias of the same 
 > amount!  so cutting out the sources and stacking them yielded a 2.5mJy source.
 > 
 > We have conjectured several possibilities for this, but in an effort to save 
 > time I thought to ask an expert.
 > 
 > Does any of this make sense?  Could it be (with the 4mJy sources) that we are 
 > adding 400mJy to the map and that could be causing the bias?
 > Could this be a result of the snapshot nature of the FIRST UV images whose 
 > integration time is only 165sec?

I am unfamiliar with the methods used to reduce FIRST data.  Clean has
a number of problems - when allowed to run over the full area it sucks
flux out of real sources and puts it in the parts of the field that
should have been empty.  This is discussed in some detail in the NVSS
document and IMAGR has options to discard weak isolated Clean
Components at various stages.

I would begin by making an image using the IMAGR you just ran but with
NITER=0.  If the reduction in flux is still present, then the cause
lies somewhere else.  Run UVMOD in the mode where the original data
are discarded and reasonable noise added and make the uncleaned model
image.  Are the fluxes still biased?  Try an image with a strong
source (e.g. 2.0 Jy) at the coordinate origin and no Clean.  Is it
also reduced?  If so, what is the flux of your beam - not the peak
value which is forced to 1.0 but the integral over some area?

I am told that the FIRST survey manages to miss sources that it should
have seen fairly easily given the claimed noise limits.  In other
words that it is very incomplete at its lower fluxes.  Is the imaging
being done is a single facet or are you using multiple facets and
covering a wide range of angles?  Try 

tget imagr; 
cellsize=0; imsize=0
task 'setfc'
- set the other adverbs to cover a reasonable range of angle (30 arc
min radius or even 2 degrees) and include interfering sources (not
relevant to pure models of course) and see if the facets, cell size,
and imsize it recommends are similar to what you have been using.

If one does not take the 3D sky curvature into account, fluxes away
from the origin will be affected.

I am asking more questions than giving answers.  Snapshots are tricky
to image and calibrate - the trouble will lie in that area eventually.
ROBUST 0 rather than natural weight may matter too.

Eric Greisen