CYBERSKA

 A Cyberinfrastructure platform to meet the needs of data intensive radio astronomy on route to the SKA

GALFACTS Technical Team

GALFACTS Technical Team

GALFACTS project technical development team
Parent Group: GALFACTS Consortium

Navigation

No pages created yet

  • Public

Automated Source Finding - Duchamp

Last updated 3770 days ago by Samuel George Comments (13)

Here are some initial results of automated source finding with the GALFACTS dataset.

The Duchamp source finding software has been run on the GALFACTS C-field 13 Jul 2010 Release. I have completed some preliminary source finding in the C-field 13 Jul 2010 Release with Duchamp. This is for a median averaged Stokes I image over 700 channels (1481.374048-1510.774048 MHz) for all pixels (the FITS file used can be found here).

Some 284 sources were found in the I image, though some are obviously false detections (e.g. one of the sources is the whole milky way, and another is the "flotilla" of the 7 beams), it seems to find the bright sources, well. The full list of sources can be found here and a composite view of all the sources found, and some brief information can be found here. Also a KVIS annotation file can be found here.

Summary of statistics determined by Duchamp:
Detection threshold = 2.159 K
Noise level = 0.0329605, Noise spread = 0.708679
Full stats:
Mean   = 0.237711    Std.Dev. = 0.976606
Median = 0.0329605    MADFM    = 0.477996 (= 0.708679 as std.dev.)

Polarization Properties of the sources.

From this source list the peak Stokes Q and U are extracted from the data cube and plotted for all of the sources is Q/I vs U/I for each channel (note this is without background subtraction). This gives us an indication of the polarization of the source and to see how the leakage corrections are working. A PDF for all 284 sources (clearly some are not sources and the Q/I vs U/I nicely illustrates this) can be found here (and for a larger axes range (1,-1 in x/y) see: here). We also average the Q/U data, averaging every 10 channels: the median, mean plots. There are a few interesting sources in the field and so to investigate further we have also looked at the behaviour of Q/I and U/I with frequency (red curve is Stokes U/I and the blue curve is Stokes Q/I). Here is a few interesting sources:

Source 52: a possible example of baseline problems in the dataset (red curve is Stokes U/I and the blue curve is Stokes Q/I).

image

image

Source 107: a source with rotation.

image

image

Source 114:

image

image

If we channel collapse a larger number (e.g. 50 channels) then the these plots become clearer and we are able to easily connect the points to give an idea of what is happening with frequency. The plot can be found here. The data points are joined together by a red line. The first data point is marked by a blue point (i.e. the first channel) and the last a green one. Source 24 (a bright source with I = 9.58 K) shows an interesting shape with frequency and could be due to rotation over the band. Source 27 (I =7.9K) also has curved shape with frequency.  Source 99 (one of the brightest sources at 11.88K) also shows a possible arc.

 

Stokes I Comparison with NVSS

image

Note preliminary plot. PDF version. Within the beam of the GALFACTS survey (I assumed 3 arcmins) I have matched the list of sources produced from Duchamp with that of the NVSS. I have not converted the units of the GALFACTS dataset, but there doesn't appear to be a "clean" correlation, however the spurious points are likely false detection and the closest NVSS source is merely coincidence. TODO: look into which sources are obviously false detection. There does however, appear to be some form of correlation, though interesting it we were to fit a linear model to this then it would not go through 0,0 - thus is our zero offset correct?

Comparision of positional offsets between the NVSS and GALFACTS:

 

image

(PDF version)

Stokes I Comparison with GB6.

 

image

image

PDF version of flux vs flux. Within the beam of the GALFACTS survey (I assumed 3 arcmins) I have matched the list of sources produced from Duchamp with that of the GB6. I have not converted the units of the GALFACTS dataset.

PDF version of the position offsets.

Comments

  • Samuel George 3783 days ago

    This page (and the corresponding plots / data) has now been made public to all members of the portal. I've also uploaded a plot with the smaller axes range (-0.3, 0.3 in Q/I and U/I), however I've not included the autoscaled plots at the moment as there are spurious points with much larger data and this is currently messing up the range plotted.

  • Samuel George 3782 days ago

    The plot of Q/I versus U/I over the smaller range has now been updated to have smaller points and include the I flux in the subplot title - I've just over wrote the file and can be found here.

  • Samuel George 3780 days ago

    I it worth noting that the Q/I vs U/I plots do not have background emission subtracted, this will be added asap.

  • Samuel George 3780 days ago

    If we channel collapse a larger number (e.g. 50 channels) then the these plots become clearer and we are able to easily connect the points to give an idea of what is happening with frequency. The plot can be found here. The data points are joined together by a red line. The first data point is marked by a blue point (i.e. the first channel) and the last a green one. Source 24 (a bright source with I = 9.58 K) shows an interesting shape with frequency and could be due to rotation over the band. Source 27 (I =7.9K) also has curved shape with frequency.  Source 99 (one of the brightest sources at 11.88K) also shows a possible arc.

  • Jeroen Stil 3778 days ago

    Hi Sam,

    Source 107 may be polarized, but I don't think you see Faraday rotation. It looks more like a baseline ripple to me.

  • Russ Taylor 3777 days ago

    HI Sam.  Could you make delta alpha vs delta delta plots for the NVSS versus GALFACTS positions?  It would be good to check the GALFACTS source positions.

  • Russ Taylor 3777 days ago

    Your plot of GALFACTS vs NVSS flux density looks as expected.  Looks like there is a cut off of about 2 K on the GALFACTS source list. Down to that level there is a linear relationship with NVSS flux with the exception of what I take to be spurious sources.  Also GALFACTS peak flux can be much higher than the NVSS peak flux if the source is resolved by the NVSS, so the cluster of points above the relationship is probably NVSS resolved sources.  Can you redo the plot using only sources that are unresolved with the NVSS, or alternately using the integrated NVSS flux density (or do both)?

  • Russ Taylor 3777 days ago

    When Duchamp reports a peak flux density for a source, is this with a local background removed?

  • Samuel George 3777 days ago

    When Duchamp reports a peak flux density for a source, is this with a local background removed?

    It appears to return the peak flux without the local background removed, or more in the mode I am currently running this in (I'm not sure if it can, will investigate) it does not remove the background (the value is the same as the non background removed pixel value in the FITS file).

    I've implemented the background removal in annulus around the detected sources for the Q/I vs V/I, Q/I / U/I with frequency and the averaged (over 50 channels) Q/I vs U/I plots. I've not updated the above plots.

  • Samuel George 3777 days ago

    delta alpha vs delta delta plots for the NVSS versus GALFACTS positions

    Here is the plot of the position offsets between the NVSS and the GALFACTS positions (all sources are matched with the GALFACTS beam) and I've updated the above page (PDF version):

     

    image

  • Samuel George 3776 days ago

    I've also produced a plot of the position offset with the GB6 and this shows a similar offset to the NVSS:

     

    image

  • Jeroen Stil 3774 days ago

    Two comments: we just checked that kvis fitted positions are significantly different from Duchamp positions. Kvis agres with the NVSS it seems. The issue may be in the way Duchamp was run. Last week I found with Ben that the position of bright sources depends strongly on input (try raising detection threshold).

    It is also curious that the outliers tend to be in one quadrant relative to the cluster of points.

  • Samuel George 3769 days ago

    I've now completed the same analysis as on here but for SAD to check what was found with Duchamp (and the issue might be relanted to the mode of operation).