Combining data from different arrays

At ERIS the data sets for this tutorial will be in /home/user/datastore/COMB

Contents

Combining data from different arrays
1. MERLIN + VLA
2. EVN + MERLIN

MERLIN + VLA

EVN + MERLIN

Markarian 273 was observed by Bondi et al., 2005 MNRAS 361 748 using MERLIN and the EVN at 5 GHz. We thank Marco Bondi et al. for kindly providing the data. The data were correlated separately and the individual data sets were calibrated using the same calibration sources including the phase referencesource J1337+550, and the target (which is too faint to self-calibrate) was split out. The data sets are fully calibrated and edited; we only have to combine them.

The naturally-weighted MERLIN-only map has a beam size of ~50 mas; the EVN-only map has a beam size of ~10 mas and is insensitive to emission on scales >60 mas. The beam size ratio of 1:25 (EVN : MERLIN) means that a lowest contour at the same flux density in Jy/beam corresponds to a brightness temperature 25x higher in the EVN image. In fact, the rms is slightly lower in the EVN image (as the collecting area is larger) but still only the brightest hot-spots are detected.

Load and assess the data

Load the files

/home/user/datastore/COMB/MRK273-EVN-6CM-SPLIT.FITS  
/home/user/datastore/COMB/MRK273-MERLIN-6CM-SPLIT.FITS

into AIPS (remember douvcomp -1).

Each file contains just the Mrk 273 data. The headers and history files show that the data from each array have already been averaged into a single channel using AVSPC. You can check that Mrk 273 was observed at about the same frequency and time by checking the headers. Note the position, frequency, bandwidth, date etc. Normally you should also check that the phase-reference source was observed at the same position but in this experiment we confirm that was the case; see the MERLIN+VLA example for what to do with misaligned data.¹

You can use UVPRT to check that the data were observed very closely or simultaneously in time with respect to the reference day in the header.²

Knappen et al. (1997) published a MERLIN 5-GHz image taken without the Lovell telescope, in 1992. This shows that the detectable source lies within ~1 arcsec. 4.5.4.1 Bandwidth- and time-smearing in the MERLIN User Guide or lectures at this School give expressions and a table for bandwidth smearing which you can use to check whether the data here will give any problems.

Correcting the amplitude scale

Even though the data sets were taken contemporaneously the amplitude scales were set separately; for MERLIN using 3C286 and for the EVN using Tsys measurements. Both these methods are prone to uncertainty at 5 GHz so the first step is to compare the flux density on the same uv scales. EVN baselines are shown in yellow and MERLIN in green and by zooming in we see that the ~2.5-3 Mlambda baseline overlaps. In these data that is of course the same (Ca-Lo) baseline which can be selected by antenna number, but in e.g. MERLIN+VLA (A-array) data a similar overlap can be selected by uv- and HA-range. Plotting visibility amplitude against time for Ca-Lo shows that there is a mis-scaling by a factor of 0.7423 (EVN apparently brighter than MERLIN) - you can get a single value by averaging to a single bin in UVPLT instead of the 5 bins shown here.

Use PRTHI to examine the history file for the MERLIN data, PRTASK 'SETJY'. This shows that a flux density of 2.514 Jy was used for OQ208. Some recorded MERLIN flux calibration source values show that this is a typical or slightly high value, so probably the EVN data have the flux scale set too high.

We multiply the EVN data to be consistent with the MERLIN flux scale (leaving all the model parameters as zero):

AIPS 1:task 'UVMOD'
AIPS 1:INNAME     'MKN273-6CM '
AIPS 1:INCL       'SUM IF'
AIPS 1:FACTOR      0.7423

Weighting, combining and imaging

The maximum EVN baseline of ~40 Mlambda implies a full resolution of ~5 mas and so a pixel scale of 1.5 mas is needed, which in turn implies an image size of 2048 pixels to ensure all possibly detectable emission is well within the field. Make test images (natural weighting) with no cleaning for both data sets (MKN273MER.UVMOD.1; MKN273-6CM.SUM IF.1) to get a report of the gridding weights. These are in a ratio ~500:1 EVN : MERLIN (check for yourself) so, to produce a combined data set with equal contributions, the inverse of this is used. The position but not the frequency agrees exactly and the arrays are different.

AIPS 1: task 'DBCON'
AIPS 1: INNAME     'MKN273-6CM'      
AIPS 1: INCLASS    'UVMOD'            
AIPS 1: IN2NAME    'MKN273MER   '      
AIPS 1: IN2CLASS   'SUM IF'            
AIPS 1: REWEIGHT      1    0.002  
AIPS 1: OUTNAME    'MERLEVN_.002  '      
AIPS 1: OUTCLASS   '      '            
AIPS 1: OUTSEQ        0                
AIPS 1: DOPOS         1, 0 
AIPS 1: DOARRAY      -2

Experiment with IMAGR. This highest-resolution image was made from the 1:0.002-weighted combined image with

AIPS 1: task 'IMAGR'
AIPS 1: INN 'MERLEVN_.002'
AIPS 1: INCL 'DBCON'
AIPS 1: INSE 1
AIPS 1: CELLSIZE    0.0015 0.0015  
AIPS 1: IMSIZE     2048 2048    
AIPS 1: ROBUST     0
AIPS 1: UVBOX      5; UVBXFN     4 
AIPS 1: UVWTFN ' '
AIPS 1: FACTOR=-1
AIPS 1: GAIN  0.05
AIPS 1: BMAJ  0.010; BMIN 0.010
AIPS 1: MINPATCH 512
AIPS 1: NBOX 1;CLBOX  157 585 1421 1509
AIPS 1: NITER 5000

The single box was set the same for various test images to ensure comparability but you may get better results with interactive boxing.

In order improve the sensitivity to slightly extended emission e.g. to try and detect the SW component, we repeated DBCON with more MERLIN in the mix, e.g. REWEIGHT 1 0.01, and used natural weighting in IMAGR:

AIPS 1: tget IMAGR 
AIPS 1: INN 'MERLEVN_.01'
AIPS 1: INSE 2
AIPS 1: ROBUST     5
AIPS 1: UVWTFN 'N'
AIPS 1: FACTOR=-0.3
AIPS 1: BMAJ  0.020; BMIN 0.020

which gave a medium-resolution image

Note that each combination and weighting gives reliable images on a limited range of scales; for example, the highest resolution (e.g. Bondi et al. Fig 2 (left) shows that at 5-mas resolution even the brightest peak fragments. This has significant implications for the search for the location of the AGN in Mrk 273. On the other hand, you would not use either this or the 10-mas resolution image (right) for investigating the structure of the more extended emission since even where this is detected, it is prone to artificial fragmentation due to the small beam size. These images do show that adding in shorter spacings from MERLIN does improve the quality of the combined image even at the same resolution as the EVN-only map.

1 If data are observed at slightly different frequencies you might have to compensate for the spectral index.
2 Sometimes it is possible to combine data for variable sources observed at different times if the only components which have varied are compact and can be subtracted out.