R.F. Minchin, PASA, 16 (1), in press.
Next Section: Summary Title/Abstract Page: Finding the Bivariate Brightness Previous Section: Determining the BBD | Contents Page: Volume 16, Number 1 |
PICASSO - an automated galaxy finder
We are developing an automated galaxy finder for use on HIPASSdata cubes. The results of the tests so far are encouraging. It is currently useable for finding candidate sources in real data cubes, although a by-eye inspection is still necessary to ensure reliability.
The finder, PICASSO, contains a number of sub-processes which are written in FORTRAN 77 and PERL. The most important of these are finder, which identifies sources and associates a quality level with them, and fitter, which provides accurate 3D positions and line widths for the galaxies. Other sub-processes remove degenerate sources and parameterize the galaxy using routines from Miriad (Sault, Teuben & Wright, 1995)
The quality level output by finder is defined as being the signal measured in the pixel being examined, summed over a number of velocity channels, divided by the noise measured in a surrounding annulus, taken over the same number of channels. The quality of a source is therefore its signal to noise as seen by the finder.
The quality data from finder can be used to determine a maximum distance to which the galaxy would be detected, as the only input into the quality that should change with distance is the flux of the galaxy (the noise on the cubes remains almost constant in the velocity range being examined). The distance to which the galaxy can be seen, dmax is therefore simply related to the minimum quality that is accepted, Qmin as:
Where Q and d are the detected quality and distance. Therefore V/Vmax (Schmidt, 1968) can be determined from the quality and used to measure the completeness of the sample.
Testing on simulated sources injected onto an HIPASSdata cube with real noise indicated that sources could be found reliably above quality 7 and in the range
(where k is the number of channels, each channel being 13.6 km wide), the range over which the finder searches (see figure 1). Quality 7 was associated, using these simulations, with a signal to noise ratio of around 10. It has since been found that pre-processing which was being applied to the cube in order to reduce processing time and memory load was degrading the signal to noise. This pre-processing has now been abandoned and quality should now correspond with signal to noise. As the pre-processing was degrading the cube before it was searched by the finder, this does not affect the reliability results given here.
The finder was then tested on real data. The data used was from the Deep project (Disney et al., 1999) and consisted of 3 independant cubes at HIPASScoverage (the `shallow' fields) and the full Deep cube at
HIPASScoverage (the `deep' field). Initially no velocity limit was applied to the source region, and the reliability of the shallow fields was judged by whether a source was also found in the deep field. A countour map of this reliability is given in figure 2, showing that problems were encountered.
It was found that the density of unreliable sources on the real data was 3 times higher than on the simulated cube, which also had real noise. It was also found that a large proportion of the `reliable' sources were actually the peaks of baseline ripple due to strong continuum sources. The large number of unreliable sources was mainly due to noise spikes superimposed on lower continuum peaks. In order to avoid this baseline structure, the region used was limited in velocity to v < 8000 km , at the same time a lower limit to the velocity of v > 200 km was introduced to reject contamination from HIin our Galaxy. This lower limits is raised to 1000 km for the BBD sample in order to reject nearby galaxies with poorly-defined distances.
The sources in this region were then checked by eye on the deep field cube and a judgement made as to their reliability. In addition to this, the source list from the deep field was compared to an independant by-eye survey of the deep field that had been carried out in Cardiff (Disney et al., 1999). It was determined that, at the Q > 7 level, PICASSO had a reliability of just over 70% and found about same number of true galaxies as the by-eye survey, with an overlap between the lists of 75%. At the Q > 10 level, PICASSO was found to be around 95% reliable. This region contains just over 60% of the sources found. The reliability for the shallow and deep fields, and an average of the two, is given in figure 3.
As part of the post-processing of the PICASSO data, moment maps are constructed using a 44
window over a range in velocity of twice the fitted velocity width, centred on the fitted position. The spatial size is chosen to be three times the FWHM of the Parkes beam. There is only a small chance of confusion as galaxies would need to be close in velocity as well as spatially. A gaussian fit and a base-line offset are fitted simultaneously to the zeroth-order moment map using the imfit routine in Miriad. As long as the galaxy is a point source, the peak value of the gaussian is the integrated flux of the galaxy, which can be used to determine the mass of the galaxy. The imfit routine also gives an error on the peak, which gives an estimate of the error on the mass. It is hoped, although it has not yet been tested, that this will give another discriminator against spurious detections and therefore further reduce the subjectivity of the final sample by cutting out false candidates before the by-eye inspection.
A slight caution should be attached to the reliabilities determined for PICASSO on these cubes. Although real data cubes were used, the observations used to construct them were all made at night time, so these cubes are vitually free of the solar ripple that can affect standard HIPASS cubes. However, the region is fairly close both to the galactic plane and to the strong radio source in NGC 5128 and contains a relatively high number of strong continuum sources. The polarisation subtraction used to construct the real-noise cubes for the injection of the simulated sources means that these cubes are similarly free of solar ripple. The reliability of PICASSO when run on a standard cube has been found to vary widely from cube to cube, although this is not a problem for this project due to the by-eye inspection of the spectra before selection of the final sample.
Next Section: Summary Title/Abstract Page: Finding the Bivariate Brightness Previous Section: Determining the BBD | Contents Page: Volume 16, Number 1 |
© Copyright Astronomical Society of Australia 1997