This article describes the calibration and imaging of a single-pointing 6cm EVLA wideband continuum dataset on the galaxy NGC2967 (UGC5180) which was the location of the supernova candidate SN2010FZ. No supernova was detected in this observation, but the galactic continuum emission from this face-on spiral is adequately imaged. The data were taken in RSRO mode, with 1024 MHz of bandwidth in each of two widely spaced basebands (comprised each of 8 128 MHz spectral windows), spanning 4.5 to 7.5 GHz. We will use wideband imaging techniques in this tutorial.
This is a more advanced tutorial, and if you are a relative novice (and particularly for EVLA continuum calibration and imaging), it is strongly recommended that you start with the EVLA Continuum Tutorial 3C391 before tackling this dataset. We will not include basic information on CASA processing in this tutorial.
CASA Versions
This tutorial was written for the CASA Version 3.2.1 (release r15198 26 May 2011).
Obtaining the Data
The scheduling block (SB) processed appears in the EVLA archive under program AS1015 as
AS1015_sb1658169_1.55388.89474846065
and was run on 2010-07-11 from 21:28 to 22:28 UT (size 37.74GB).
For the purposes of this tutorial, we have provided the raw SDM data (as would be extracted from the archive) as well as measurement sets created by filling the data (with the importevla task) and upon time-averaging to 10s (after application of the online flags).
To start your tutorial, depending on which dataset you start with, proceed to:
To start with the raw SDM data: Start with the section below titled "#Importing your EVLA data from SDM". This is where you would start if you were reducing data from the archive.
The C-band data of interest is contained in scans 6-44 and spans spectral windows 2 to 17.
We use the importevla task to convert the SDM dataset from the archive to a CASA Measurement Set (MS).
# In CASAimportevla(asdm='/lustre/smyers/AS1015/AS1015_sb1658169_1.55388.89474846065', \
vis='SN2010FZ_filled.ms',online=True,flagzero=True, \
shadow=True,applyflags=False,tbuff=1.5,flagbackup=False)
Here we had the task create (but not apply) the online flagging commands, plus flags for zero-clipping and shadowing. The timeranges for the online flags were extended by 1.5sec (the integration time was 1sec) to account for some timing mismatches present in the EVLA data at this time. These online flags indicated times where the antennas were not on source (e.g. slewing) or had other detectable faults. The created flagging commands will be stored in the FLAG_CMD MS table and can be applied later. Note that if you set applyflags=True here then after filling the task will go ahead and apply the flags for you.
For the purposes of this exercise, in order to save time and disk space, we have turned off the automatic creation of flag column backups by setting flagbackup=False. If we make a mistake and need to recover flags then we will have to rerun all previous commands. We recommend that for real data processing that you leave the default value flagbackup=True in this and subsequent tasks.
You now have a MS called SN2010FZ_filled.ms in your working area. This should be 37GB like the SDM.
Application of Online Flags and Averaging your MS
If you are starting from the filled MS, you can find this at the AOC at:
/lustre/smyers/AS1015/SN2010FZ_filled.ms
Again, use the actual location of this file for your system.
NOTE: the following step will not work in Version 3.2.1 (you will get a blank plot) but should in later versions).
You can examine the commands stored in the FLAG_CMD table using flagcmd.
# In CASAflagcmd(vis='SN2010FZ_filled.ms',flagmode='table',optype='plot')
This will bring up a matplotlib plotter. You can have it plot to a PNG file instead:
# In CASAflagcmd(vis='SN2010FZ_filled.ms',flagmode='table',optype='plot',outfile='plotSN2010FZ_flagcmd.png')
# In CASAflagcmd(vis='SN2010FZ_filled.ms',flagmode='table',optype='apply',flagbackup=False)
This can take a while for our 37GB dataset. It took 20min on my workstation.
With the known bad data flagged, we can now split out the data we want and also average down in time to make a smaller MS.
For D-configuration (max baselines 1km) we can safely average to 3s or even 10s to reduce dataset size:
# In CASAsplit(vis='SN2010FZ_filled.ms',outputvis='SN2010FZ_filled10s.ms',datacolumn='data',timebin='10s')
This can also take a while for our 37GB dataset. It took 20min on my workstation.
You now have a MS called SN2010FZ_filled10s.ms in your working area. This should be 3.2GB in size.
Examining and Flagging your Averaged MS
If you are starting from the pre-flagged averaged split MS, you can find this at the AOC at:
Scan 6 is a dummy scan so we will use scans 7 to 44 when we process our data.
plotants plotter
To plot up the antenna positions in the array:
# In CASAplotants('SN2010FZ_filled10s.ms')
NOTE: if after this point or any other you get table locks, use clearstat to clear them:
# In CASAclearstat
Now we examine the MS looking for bad data to flag. We will use plotms to bring up an interactive GUI that will display 2-D Y vs.X style line plots. NOTE: We do not recommend using the editing/flagging features of plotms. It is very easy to mess up your data that way. Also, we will be restricting the scope of plotting so most box/flag operations would not get rid of all the bad data. We will instead use plotms to identify bad data and then use flagcmd to flag it. This will also allow full scripting of the flagging.
WARNING: The Flag button on the plotms GUI is close to other buttons you will be using, in particular the one that gets rid of boxes you have drawn. Be careful and don't hit the Flag button by mistake!
The useful spw are 2~17. To get an idea of the data layout, plot a single baseline/channel versus time:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='',spw='2~17:31~31', \
antenna='ea01&ea02',correlation='RR,LL',xaxis='time',yaxis='amp')
plotms ant2 vs ea01
Look for bad antennas by picking the last field and plotting baselines versus antenna ea01:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='2',spw='2~17:31~31', \
antenna='ea01',correlation='RR,LL',xaxis='antenna2',yaxis='amp')
You should be able to see that antenna 11 (= ea13) is bad (very low amplitude, it has no C-band receiver!) and that some of the spectral windows on 15 and 23 (ea17,ea25) are also on the low side. Boxing and using Locate will show that spw 10~17 are suspect for these antennas.
plotms ea02 vs frequency
Now look at the bandpass for ea02 - it is in the inner core and a prospective reference antenna. Exclude ea13 using negation in the selection:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='2',spw='2~17', \
antenna='ea02;!ea13',correlation='RR,LL',xaxis='frequency',yaxis='amp')
There is clearly less data for spw 11, and use of Locate shows spw 11 data only for ea02,ea03,04,08,09,11,12. We will later delete this incomplete spw. Note also the very strong RFI spike at 6614MHz (spw 10 ch 63) with clear ringing contaminating both spw 10 and 11. There is also a tremendous roll-off in spw 10. We will drop these spectral window when we process the data.
plotms ea02&ea20 iteration phase vs frequency
We can also step through the baselines to our antenna using iteraxis - use the ">" button to step through:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='2',spw='2~17',antenna='ea02;!ea13', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',iteraxis='baseline')
This will make it easier to isolate the bad antennas. Now plot the phases, iterating through baselines to ea02:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='2',spw='2~17',antenna='ea02;!ea13', \
correlation='RR,LL',xaxis='frequency',yaxis='phase',iteraxis='baseline')
You see the slopes due to residual delays. Mostly a turn or less over a 128MHz subband, but there are some outliers.
Step through to ea20. You see that there is a very large delay in RR (via locate) for
the first baseband (spw 0~7). We will delete this (will also delete LL so there are no orphan polarization products).
Note ea17 and ea25 baselines drop close to zero in the middle of upper baseband (e.g. plot 'ea17&ea25') so we will delete these.
To carry out flagging, we again use flagcmd in the mode where it takes a list of command strings:
# In CASAflaglist=['antenna="ea13"','antenna="ea17" spw="10~17"','antenna="ea25" spw="10~17"','antenna="ea20" spw="2~9"']flagcmd(vis='SN2010FZ_filled10s.ms',flagmode='cmd',command=flaglist,optype='apply',flagbackup=False)
These commands will be carried out as well as being added to the FLAG_CMD table (marked as applied).
Plot the data again now having flagged:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='2',spw='2~17',antenna='ea02', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',scan='7~44')
plotms field 0 ea02 amp vs frequency
Now our phase calibrator - it is weaker, and we now start to see the RFI:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='0',spw='2~17',antenna='ea02', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',scan='7~44')
Use the Zoom feature, Mark rectangles and use Locate to identify the frequency/channel of RFI. In particular, we note in our analysis:
6614MHz (spw 10 ch 63) super strong
6772-6778MHz (spw 12 ch 14-17)
7260-7264MHz (spw 16 ch 2-4)
7314-7340MHz (spw 16 ch 29-42)
7402-7418MHz (spw 17 ch 9-17)
7458-7466MHz (spw 17 ch 37-41)
7488MHz (spw 17 ch 52)
If you plot all antennas and avoid the band edges you see spw 16,17 are pretty wiped out:
# In CASAplotms(vis='SN2010FZ_filled10s.ms',field='0',spw='2~17:4~59',antenna='', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',scan='7~44')
For now we will not flag these channels but note them to mask out when creating continuum calibration tables and images.
Finally, split off the good scans and spw, this will allow us to work on the data without having to start completely over if we mess something up badly as well as letting us do simpler selection:
# In CASAsplit(vis='SN2010FZ_filled10s.ms',outputvis='SN2010FZ_flagged10s.ms',datacolumn='data',spw='2~9,12~17',scan='7~44')
You now have a MS called SN2010FZ_flagged10s.ms in your working area. This should be 2.8GB in size.
Prepare the MS for calibration by adding scratch columns. This will take a few minutes.
# In CASAclearcal('SN2010FZ_flagged10s.ms')
Setting the flux density scale
Before calibrating, we put a model for flux calibration source 3C286 into the MS (in the MODEL_DATA column we just created).
Have it set the flux on a per-channel basis.
NOTE: This uses the AOC Unix location for casapy models (yours may be different):
# In CASAsetjy(vis='SN2010FZ_flagged10s.ms',field='2',spw='',scalebychan=True, \
modimage='/usr/lib64/casapy/data/nrao/VLA/CalModels/3C286_C.im')
It reports to logger that its about 7.7Jy at lower end to 5.7Jy at upper frequency limit.
Calibrating delays and bandpass
First, we do a phase-only calibration solution on a narrow range of channels in each spw on the bandpass/flux calibrator 3c286 to flatten them before solving for the bandpass. Note where we saw RFI in the higher spw and avoid those channels. The range 23~28 should work. Pick a refant near center - ea02 is a reasonable bet:
# In CASAgaincal(vis='SN2010FZ_flagged10s.ms',caltable='calSN2010FZ.G0',field='2',spw='0~13:23~28', \
gaintype='G',refant='ea02',calmode='p',solint='int',minsnr=3)
plotcal G0 phase ant 0~15plotcal G0 phase ant 16~26
Plot the phase solutions (using full phase range instead of autorange)
# In CASAplotcal(caltable='calSN2010FZ.G0',xaxis='time',yaxis='phase',iteration='antenna', \
plotrange=[-1,-1,-180,180])
Step through the antenna-based solutions. They look good (and fairly flat over the scans).
NOTE: If you want to make single-page multipanel plots (like those shown to the right), particularly for
hardcopy (where it only shows the first page), you can do:
We can now solve for the residual antenna-based delays that we saw in phase vs. frequency.
This uses the gaintype='K' option in gaincal (this is not documented but is available in
Version 3.2.1). Note that this currently does not do a "global fringe-fitting" solution for delays,
but instead does a baseline-based delay solution to all baselines to the refant, treating these
as antenna-based delays. In most cases with high-enough S/N to get baseline-based delay solutions
this will suffice.
# In CASAgaincal(vis='SN2010FZ_flagged10s.ms',caltable='calSN2010FZ.K0',gaintable='calSN2010FZ.G0', \
field='2',spw='0~13:4~59',gaintype='K', \
refant='ea02',combine='scan',solint='inf',minsnr=3)
We pre-apply our initial phase table, and produce a new K-type caltable for input to bandpass calibration.
The delays found are sent to the terminal. They are up to a few nanoseconds.
plotcal B0 amp phase vs. freq for ea01
Now solve for the bandpass using the previous tables:
# In CASAbandpass(vis='SN2010FZ_flagged10s.ms',caltable='calSN2010FZ.B0', \
gaintable=['calSN2010FZ.G0','calSN2010FZ.K0'], \
field='2',refant='ea02',solnorm=True, \
bandtype='B',combine='scan',solint='inf')
Now plot this, in amplitude then phase:
# In CASAplotcal(caltable='calSN2010FZ.B0',xaxis='freq',yaxis='amp',iteration='antenna')plotcal(caltable='calSN2010FZ.B0',xaxis='freq',yaxis='phase',iteration='antenna', \
plotrange=[-1,-1,-180,180])
In the bandpass phase you no longer see the residual antenna delays but there are some band edge effects.
Note that some antennas have a little strange bandpasses at upper end of lower baseband in spw 5,6,7
(e.g. ea14,ea16,ea17,ea25).
To plot amp and phase for a single antenna versus frequency (see plots at right):
plotcal G1 phase ant 0~15plotcal G1 phase ant 16~26
Now calibrate phases for real with wider bandwidth. First the flux calibrator again, with a per-integration solution time:
# In CASAgaincal(vis='SN2010FZ_flagged10s.ms',caltable='calSN2010FZ.G1', \
gaintable=['calSN2010FZ.K0','calSN2010FZ.B0'], \
field='2',refant='ea02',solnorm=F,spw='0~13:4~59', \
solint='int',gaintype='G',calmode='p')
Next our phase calibrator, appending these solutions to previous table.
Exclude RFI channels here, and obtain one solution per scan:
# In CASAgaincal(vis='SN2010FZ_flagged10s.ms',caltable='calSN2010FZ.G1',gaintable=['calSN2010FZ.K0','calSN2010FZ.B0'], \
field='0',refant='ea02',solnorm=F, \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
solint='inf',gaintype='G',calmode='p',append=True)
The phases look reasonably connected:
# In CASAplotcal(caltable='calSN2010FZ.G1',xaxis='time',yaxis='phase',iteration='antenna', \
plotrange=[-1,-1,-180,180])
Now solve for amplitudes on a per scan interval. Do these separately using gainfield so phases don't get
transferred across fields. Uses linear interpolation of the previously determined phases by default.
Pre-apply the gaincurve also:
Since the flux on the gain calibrator is not scaled to its correct flux (but to 1.0Jy)
use fluxscale to transfer the amplitude gains from 3c286:
# In CASAfluxscale(vis='SN2010FZ_flagged10s.ms',caltable='calSN2010FZ.G2', \
fluxtable='calSN2010FZ.F2',reference='2',transfer='0')
As it so happens, the derived flux for J0925+0019 is about 1 Jy and flat spectrum (you can plot up the raw amplitudes for fields 0,2 and convince yourself this is indeed true and not a bug). Plot these:
# In CASAplotcal(caltable='calSN2010FZ.F2',xaxis='time',yaxis='amp',iteration='antenna')
The phase calibrator has consistent gains across the short run, though there is a slight offset to 3c286 (which is not nearby in any event).
Applying the Calibration and Final Editing
Next we actually apply all our accumulated calibration tables. We apply these to the
calibration fields individually using the appropriate gainfields and interpolation for each:
# In CASAapplycal(vis='SN2010FZ_flagged10s.ms',field='2', \
gaintable=['calSN2010FZ.K0','calSN2010FZ.B0','calSN2010FZ.G1','calSN2010FZ.F2'], \
gainfield=['','','2','2'],interp=['nearest','nearest','linear','nearest'], \
parang=False,calwt=F,gaincurve=T)
plotms cal applied fld2
For the nearby calibrator we did only scan-based phase solutions so we use nearest interpolation:
# In CASAapplycal(vis='SN2010FZ_flagged10s.ms',field='0', \
gaintable=['calSN2010FZ.K0','calSN2010FZ.B0','calSN2010FZ.G1','calSN2010FZ.F2'], \
gainfield=['','','0','0'],interp=['nearest','nearest','nearest','nearest'], \
parang=False,calwt=F,gaincurve=T)
Finally we apply calibration from field 0 to the target field 1. This takes a few minutes:
# In CASAapplycal(vis='SN2010FZ_flagged10s.ms',field='1', \
gaintable=['calSN2010FZ.K0','calSN2010FZ.B0','calSN2010FZ.G1','calSN2010FZ.F2'], \
gainfield=['','','0','0'],interp=['nearest','nearest','linear','linear'], \
parang=False,calwt=F,gaincurve=T)
We can examine the corrected data on 3c286 using our RFI mask from above and avoiding band edges
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='2', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',ydatacolumn='corrected')
We now see problems in spw 5 and 6 for baseline ea17&ea25, which gives a really strange response:
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='2', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
antenna='ea17&ea25', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',ydatacolumn='corrected')
You can exclude this through antenna negation:
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='2', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
antenna='!ea17&ea25', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',ydatacolumn='corrected')
Then use Locate for the other bad points, which seem to indicate spw 5,6,7 for ea14,ea16,ea17,ea25.
Exclude these and replot:
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='2', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
antenna='!ea14;!ea16;!ea17;!ea25', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',ydatacolumn='corrected')
plotms cal applied flagged fld2
This now looks clean except for the RFI in the upper subbands.
Do flagging based on these:
# In CASAflaglist=['antenna="ea14,ea16,ea17,ea25" spw="5~7"']flagcmd(vis='SN2010FZ_flagged10s.ms',flagmode='cmd',command=flaglist, \
optype='apply',flagbackup=False)
Now replot the corrected data (you may have to force reload if you plotted same thing right before this):
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='2', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',ydatacolumn='corrected')
Looks good. If we were more diligent we would go back and recalibrate, but this looks good enough for now.
Plot the phase:
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='2', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
correlation='RR,LL',xaxis='frequency',yaxis='phase',ydatacolumn='corrected')
plotms cal applied flagged fld2 phase
Note the characteristic "bowtie" pattern of the phases about the sub-band centers.
Here we can see the effect of the EVLA "delay clunking", where the delay steps through discrete values such that
the phase goes from -11deg to +11deg across the sub-band as the delay changes due to geometry. This is D-configuration so the delays change slowly, it will change faster in wider configurations. As of Q3 2011 we have not enabled the corrections for this
in the EVLA system so you will always have this remaining delay error in your data. In principle you could solve for delays on short timescales and take this out, in practice this in not possible for your weaker target source in any event.
Now lets plot the corrected data amplitude for the phase calibrator:
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='0', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',ydatacolumn='corrected')
plotms cal applied flagged fld0 ampplotms cal applied flagged fld0 phase
You can see the roll-off pumping up noise at the baseband edges (about 8-16 channels worth). Also, we can see some RFI we missed:
<6804MHz spw 8 below ch 30 lots of bad stuff (alot from ea18,ea22 but others too)
7168MHz spw 11 ch 20
pretty much all of spw 12,13
The ch 20 ones are all harmonics of the notorious 128MHz tone.
We will not flag these, but exclude them in imaging, so a good channel selection string for imaging might be:
We can also plot the corrected phase - looks fairly good
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='0', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
correlation='RR,LL',xaxis='frequency',yaxis='phase',ydatacolumn='corrected')
We can average over baseline and each scan:
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='0', \
spw='0:10~59,1~7:4~59,8:4~13;18~59,9~11:4~59,12:4~13;18~29;31~33;46~51;53~59,13:4~8;15~36;42~59', \
correlation='RR,LL',avgbaseline=True,avgtime='600s',xaxis='frequency',yaxis='phase',ydatacolumn='corrected')
plotms cal applied flagged fld0 phase averaged
In this case we can see the residual effect of the EVLA "delay clunking" described above. It is reduced due to the averaging
that we applied, but it is still there.
plotms cal applied flagged fld1 amp averaged
You can look at the target source field='1' but there are lots of data so you will need to do alot of averaging.
For example:
# In CASAplotms(vis='SN2010FZ_flagged10s.ms',field='1',avgtime='300s', \
spw='0:16~59,1~6:4~59,7:4~54,8:30~59,9~10:4~59,11:4~19;21~59', \
correlation='RR,LL',xaxis='frequency',yaxis='amp',ydatacolumn='corrected')
Alas, the upper baseband still has lots of low level RFI.
Now split off the data for calibrators and target, to avoid later issues that can corrupt the MSes.
These will take a bit of time...field 1 took 18min on my workstation.
# In CASAsplit(vis='SN2010FZ_flagged10s.ms',outputvis='SN2010FZ_split10s.ms',datacolumn='corrected',field='1')split(vis='SN2010FZ_flagged10s.ms',outputvis='SN2010FZ_3c28610s.ms',datacolumn='corrected',field='2')split(vis='SN2010FZ_flagged10s.ms',outputvis='SN2010FZ_J092510s.ms',datacolumn='corrected',field='0')
Imaging
This is D-config data at C-band, look at the Obs Status Summary:
[1]
Synthesized beam should be 12" at 6GHz with primary beam FOV 7.5arcmin (450").
Our data spans 4.5-7.5GHz : beam 9.6 at 7.5GHz and FOV 10' at 4.5GHz
A cellsize of 3" should work, with an imsize > 200 to cover 2xFWHM FOV
An imsize of 400+ will put the main beam inside the inner quarter safely
The Briggs robust (0.5) weighting is somewhere between uniform and natural
and will give reasonable resolution but still see some larger scale structure.
Due to the numerology of FFTW (which clean uses underneath for FFTs) optimal sizes,
imsize should be composite number with two and only two prime factors chosen from
2,3,5. Taking into account the x1.2 padding that clean uses internally to the imsize
you give it (and 1.2=2*3/5), we choose 640 or 1280 as our imsize (640=2^7*5). Other
reasonable sets would be 405,1215,etc. (405=3^4*5) or 432,648,1296 (these are 2^n*3^m*5).
In practice, if you give it non-optimal values for imsize, you may find that the transforms
take a bit longer, which is noticeable if you are doing interactive clean.
Lets interactively clean one of the lower baseband spw (5):
NOTE: this first time will take a few minutes at start to create scratch columns
in the MS in case we want to do self-calibration later.
Start carefully by boxing the bright source and setting iterations to 10 at first
Gradually add more boxes and increase the number of iterations
Since this is not much more than a snapshot you see the six-fold sidelobe pattern
of the extended emission in the center of the map. This decreases as you clean
out this emission.
Stop cleaning when the residuals look like noise (and you cannot clearly see sources).
To stop click the red X button.
The top figure to the right shows a zoom in on the end state of the clean, where
we have marked a number of boxes and cleaned them out.
Note that there are some strange sidelobe patterns in lower left, possibly
from a source outside the image area. We can make a bigger image starting from
our current model:
Sure enough, there is a bright source near the lower left (see middle panel at right).
Box it, clean it a bit, and look again. There is a second source in the mid-left (track
it down by its sidelobes). Box this one, clean it a bit, and when satisfied stop.
# In CASAviewer('imgSN2010FZ10s_spw5_clean1280.image')
The restored image is shown in bottom panel to the right. I have chosen the Grayscale1 instead of default color
map as I prefer grayscale to false color for assessing image quality.
Check the rms of the residuals using the imstat task:
# In CASAmystat=imstat('imgSN2010FZ10s_spw5_clean1280.residual')print'Residual standard deviation = '+str(mystat['sigma'][0])
I got 31.8uJy for mine.
Cleaning the lower baseband
clean spw0-7 restored image center
Now, image the entire lower baseband (spw 0-7).
Follow same iterative procedure as before, and get the best
residuals you can. Because of the bandwidth and frequency
synthesis, the sidelobe pattern is different than before and
it is much easier to see fainter emission.
Be careful cleaning sources that lie near or on sidelobe splotches.
Clean the central emission region way down first to reduce the sidelobe
level before adding components in the sidelobe areas.
# In CASAclean(vis='SN2010FZ_split10s.ms',spw='0:16~59,1~6:4~59,7:4~54', \
imagename='imgSN2010FZ10s_spw0to7_clean1280', \
mode='mfs',nterms=1,niter=10000,gain=0.1,threshold='0.0mJy', \
psfmode='clark',imsize=[1280,1280],cell=['3.0arcsec'],stokes='I', \
imagermode='csclean',cyclefactor=1.5, \
weighting='briggs',robust=0.5,calready=True,interactive=True)mystat=imstat('imgSN2010FZ10s_spw0to7_clean1280.residual')print'Residual standard deviation = '+str(mystat['sigma'][0])
I got 11.3uJy (and there is clearly structure left in the residual). To the right is a zoom-in on the center
of the restored image.
clean spw0-7 mfs nterms=2 in progressclean spw0-7 mfs nterms=2 load alpha with LELclean spw0-7 mfs nterms=2 tt0 and alpha
Lets try adding a spectral slope to the multi-frequency synthesis using nterms=2 on the lower baseband.
This will solve for images of the average and spectral slope simultaneously.
The dirty beam will have lower sidelobes so we turn up cyclefactor for csclean a bit:
# In CASAclean(vis='SN2010FZ_split10s.ms',spw='0:16~59,1~6:4~59,7:4~54', \
imagename='imgSN2010FZ10s_spw0to7_mfs2_clean1280', \
mode='mfs',nterms=2,niter=10000,gain=0.1,threshold='0.0mJy', \
psfmode='clark',imsize=[1280,1280],cell=['3.0arcsec'],stokes='I', \
imagermode='csclean',cyclefactor=4.5, \
weighting='briggs',robust=0.5,calready=True,interactive=True)mystat=imstat('imgSN2010FZ10s_spw0to7_mfs2_clean1280.residual.tt0')print'Residual standard deviation = '+str(mystat['sigma'][0])
I got 10.5uJy (somewhat better looking than the nterms=1).
The top screenshot to the right shows an intermediate but early stage of cleaning where we are looking at
the central emission and cleaning it out slowly.
Cleaning the lower baseband using two MFS Taylor terms
The mfs nterms=2 option creates two "Taylor Term" images - an average intensity image (with suffix .image.tt0)
and a spectral slope image (with suffix .image.tt1) which is intensity x alpha (where alpha is spectral index).
For convenience there is a spectral index image (with suffix .image.alpha). These Taylor expansions are with respect to the "Reference Frequency" of the image (by default the center frequency of the spw selected, but can be specified using the
reffreq parameter in clean). The convention for spectral index alpha is that
[math]\displaystyle{
S \propto \nu^\alpha
}[/math]
so negative spectral indexes indicate a "steep" spectrum (falling with frequency).
You can use the viewer to load the average intensity
# In CASAviewer('imgSN2010FZ10s_spw0to7_mfs2_clean1280.image.tt0')
and then use the Open Data panel to load imgSN2010FZ10s_spw0to7_mfs2_clean1280.image
which can then be blinked (optionally plotted side-by-side using the Panel Display Options
panel to set 2 panels in the x direction). Note there is lots of noise in the low-intensity
regions, and thus filtering the alpha image based on the values in the tt0 image is desirable.
You can use the immath task to make this filtered alpha image explicitly, using a
Lattice Expression Language (LEL) expression:
# In CASAimmath(imagename=['imgSN2010FZ10s_spw0to7_mfs2_clean1280.image.alpha','imgSN2010FZ10s_spw0to7_mfs2_clean1280.image.tt0'],mode='evalexpr',expr='IM0[IM1>5.0E-5]',outfile='imgSN2010FZ10s_spw0to7_mfs2_clean1280.image.alpha.filtered')
This will use 50uJy (or 5 x the sigma we found) as the cutoff.
You can then view or manipulate the filtered alpha image as normal.
We can also use LEL to filter the alpha image on the intensity on-the-fly when we load this raster in the Open Data panel
by specifying a LEL string in the LEL box instead of selecting the image from the directory
listing. The LEL string:
will replicate what we did above. The middle figure to the right shows the Open Data panel
with our LEL string in it. Just click the Raster button to load this.
The lower panel to the right shows the intensity and LEL-filtered alpha images side-by-side in the viewer, zoomed
in on the galaxy emission. Mousing over the alpha shows spectral indexes ranging from -1 to +1 in the center, with
the brightest emission with alpha -0.7 in the knots in the disk.
Cleaning using both basebands combined
For the ultimate image, use the "clean" part of the upper baseband in addition
to the lower (use spw 0-11). We will use mfs with nterms=2 (if you try nterms=1
on this wide bandwidth you will get much poorer residuals). Because of the added
work and extra data involved, this will take much longer than our other runs of
clean. Therefore, we will get a head start by doing a non-interactive clean using
the mask left from the previous clean (spw 0-7). We will insert a clean threshold
to limit runaway cleaning too far beneath the noise level. This took 40min on my
workstation.
I found a few more sources revealed in the outer parts of the image, and also more
emission around the galaxy disk in the center. So I drew new boxes, extended the box
in the center, and did about 1000 new iterations. At the end, what was left was dominated
by the error patterns from mis-calibration. Only self-calibration will get rid of
these. Stop cleaning for now. See the figure to the right for the interactive display panel
showing final residuals and mask (changing the colormap to Greyscale 1).
Check the residual levels:
# In CASAmystat=imstat('imgSN2010FZ10s_spw0to11_mfs2_clean1280.residual.tt0')sigma=mystat['sigma'][0]print'Residual standard deviation = '+str(mystat['sigma'][0])
Now I get 8.3uJy.
Analyzing the image and comparing with the Optical/Infrared
Lets see how close we got to expected noise and dynamic range.
# In CASAmystat=imstat('imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0')peak=mystat['max'][0]print'Image max flux = '+str(mystat['max'][0])mystat=imstat('imgSN2010FZ10s_spw0to11_mfs2_clean1280.model.tt0')total=mystat['sum'][0]print'Model total flux = '+str(mystat['sum'][0])snr=peak/sigmaprint'SN2010FZ peak S/N = '+str(snr)snr=total/sigmaprint'SN2010FZ total S/N = '+str(snr)
The output gives:
Residual standard deviation = 8.33040657806e-06
Image max flux = 0.00995613634586
Model total flux = 0.0329372803525
SN2010FZ peak S/N = 1195.15611304
SN2010FZ total S/N = 3953.86228077
What do we expect? If we do listobs on this MS we see the scans:
(listing columns truncated) and we estimate about 37min on target. We had
about 25 antennas on average, and our spw selection picked out
610 channels (2MHz each) for a total of 1220MHz bandwidth. If we plug this
into the
[ https://science.nrao.edu/facilities/evla/calibration-and-tools/exposure EVLA exposure calculator]
then we find that we expect a rms thermal noise level of 7.6uJy so we are close!
final image
Look at this in the viewer:
# In CASAviewer('imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0')
In the previous section we demonstrated how to process and display the spectral index image. You can do
the same for this final image. Here, we will do some rough analysis on the spectral index to determine
an intensity-weighted mean spectral index over the core region.
The .image.tt1 from our mfs is an intensity times alpha image. See the figure to the right.
Lets gate the Taylor-term images on intensity as before:
# In CASAimmath(imagename=['imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt1','imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0'],mode='evalexpr',expr='IM0[IM1>5.0E-5]',outfile='imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt1.filtered')immath(imagename=['imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0'],mode='evalexpr',expr='IM0[IM0>5.0E-5]',outfile='imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0.filtered')
We can identify a box containing the central emission (see figure of tt1 in viewer) and note the corners.
(We could also use the region tools from the viewer, but that is for another exercise.)
Let us compute the intensity-weighted spectral index over this box by averaging
these masked images using imstat and computing the ratio:
# In CASAmystat=imstat('imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt1.filtered',box='503,533,756,762')avgtt0alpha=mystat['mean'][0]mystat=imstat('imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0.filtered',box='503,533,756,762')avgtt0=mystat['mean'][0]avgalpha=avgtt0alpha/avgtt0print'SN2010FZ I-weighted Alpha = '+str(avgalpha)
We get
SN2010FZ I-weighted Alpha = -1.54687681754
The emission in this source is on the steep side. At this point we do not know how reliable this is or
what we expect (though our calibrators come out with correct spectral indexes if we image them the
same way). But this illustrates a way to extract spectral information from our wideband mfs images.
As a final comparison, here is our image side by side with a i-band image from the Sloan Digital Sky Survey (SDSS) registered
to our image:
final and sdss image
You can find this image and load it into your viewer, and blink against our 6cm image.
The figure below shows them side-by-side:
final and sdss image
We can also plot one as a raster and the other overlaid as contours. You can load the SDSS image
from the viewer Load Data panel and fiddle with contours. Once you know contour levels, you can
also use the imview task to load a raster and contour image:
# In CASAimview(raster={'file':'imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0'},contour={'file':'NGC_2967_UGC_5180_IRAS_09394+0033-i.fits','levels':[0.2,0.5,1,1.5,3],'base':0.0,'unit':1.0})
The figure below left shows the SDSS contours overlaid on our 6cm image (after fiddling with the
colormap shift/slope for the EVLA raster image). Likewise, we can plot the SDSS image as a raster
and overlay EVLA 6cm contours:
# In CASAimview(raster={'file':'NGC_2967_UGC_5180_IRAS_09394+0033-i.fits','scaling':-2.0,'range':[0,10]},contour={'file':'imgSN2010FZ10s_spw0to11_mfs2_clean1280.image.tt0','levels':[0.04,0.08,0.16,0.32,0.64,1.28,2.56],'base':0.0,'unit':0.001},zoom={'blc':[397,300],'trc':[1567,1231]})
This is shown in the figure below right. Is the compact 6cm emission in upper left associated with a
spiral arm?
6cm EVLA raster plus SDSS i-band contoursSDSS i-band raster plus EVLA 6cm contours
Credits: SDSS image courtesy David Hogg & Michael Blanton, private communication. Data comes from
SDSS DR7, see Abazajian et. al 2009.
What to do next: some exercises for the user
Here are a number of things you can try after completing this tutorial:
# Use self-calibration to improve the data and re-clean to make a better image.
# Use multi-scale clean by adding non-zero scales to the multiscale parameter.
# Image the calibrators. What sort of dynamic range can you get on them? Is self-calibration needed (and if so what dynamic range do you get when you use it)?
# Try the testautoflag task (in 3.3.0 and later) to automatically flag RFI from the upper sideband.
