next previous
Up: The ATESP radio survey


Subsections

  
2 Source detection and parameters

The ATESP survey consists of 16 radio mosaics with spatial resolution $\sim 8\hbox{$^{\prime\prime}$ }\times 14\hbox{$^{\prime\prime}$ }$. The survey was designed so as to provide uniform sensitivity over the whole region ($\sim 26$ sq. degrees) of the ESP redshift survey. To achieve this goal a larger area was observed, but we have excluded from the analysis the external regions (where the noise is not uniform and increases radially). In the region with uniform sensitivity the noise level varies from 69 $\mu$Jy to 88 $\mu$Jy, depending on the radio mosaic, with an average of 79 $\mu$Jy (see $\sigma_{\rm fit}$ values reported in Table 3 of Paper I and repeated also in Table B1 of Appendix B, at the end of this paper). For consistency with Paper I, in the following such sensitivity average values are denoted by the symbol $\sigma_{\rm fit}$. This means that sensitivity values have been defined as the full width at half maximum (FWHM) of the Gaussian that fits the pixel flux density distribution in each mosaic (see Paper I for more details).

A number of source detection and parameterization algorithms are available, which were developed for deriving catalogues of components from radio surveys. We decided to use the algorithm Image Search and Destroy (IMSAD) available as part of the Multichannel Image Reconstruction, Image Analysis and Display (MIRIAD) package (Sault & Killeen 1995), as it is particularly suited to images obtained with the ATCA.

IMSAD selects all the connected regions of pixels (islands) above a given flux density threshold. The islands are the sources (or the source components) present in the image above a certain flux limit. Then IMSAD performs a two-dimensional Gaussian fit of the island flux distribution and provides the following parameters: position of the centroid (right ascension, $\alpha$, and declination, $\delta$), peak flux density ( $S_{\rm peak}$), integrated flux density ( $S_{\rm total}$), fitted angular size (major, $\theta_{\rm maj}$, and minor, $\theta_{\rm min}$, FWHM axes, not deconvolved for the beam) and position angle (PA).

IMSAD proved to have an average success rate of $\sim 90\%$ down to very faint flux levels (see below). Since IMSAD attempts to fit a single Gaussian to each island, it obviously tends to fail (or to provide very poor source parameters) when fitting complex (i.e. non-Gaussian) shapes.

2.1 Source extraction

We used IMSAD to extract and parameterize all the sources and/or components in the uniform sensitivity region of each mosaic[*]. As a first step, a preliminary list containing all detections with $S_{\rm peak} \geq 4.5\sigma $ (where $\sigma$ is the average mosaic rms flux density) was extracted. Detection thresholds vary from 0.3 mJy to 0.4 mJy, depending on the radio mosaic.

  
2.2 Inspection

We visually inspected all islands ($\sim 5000$) detected, in order to check for obvious failures and/or possibly poor parameterization, that need further analysis. Problematic cases were classified as follows:

The goodness of Gaussian fit parameters was checked by comparing them with reference values, defined as follows. Positions and peak flux densities were compared to the corresponding values derived by a second-degree interpolation of the island. Such interpolation usually provides very accurate positions and peak fluxes. Gaussian integrated fluxes were compared to the ones derived directly by summing pixel per pixel the flux density in the source area, defined as the region enclosed by the $\geq 3\sigma$ flux density contour. Flux densities were considered good whenever the difference between the Gaussian and the reference value $S-S_{\rm ref}$ was $\leq 0.2 S_{\rm ref}$. Positions were considered good whenever they did fall within the $0.9 S_{\rm peak}$ flux contour.


  \begin{figure}
\resizebox{8.8cm}{!}{\includegraphics{H2076f1.ps}}
\end{figure} Figure 1: Peak flux density distribution for all the ATESP radio sources (or source components) with $S_{\rm peak} \geq 4.5\sigma $

  
2.3 Re-fitting

For the first three groups listed above ad hoc procedures were attempted aimed at improving the fit.

Single-component fits were considered satisfactory whenever positions and flux densities satisfy the tolerance criteria defined above.

In a few cases Gaussian fits were able to provide good values for positions and peak flux densities, but did fail in determining the integrated flux densities. This happens typically at faint fluxes ($<10\sigma$). Gaussian sources with a poor $S_{\rm total}$ value are flagged in the catalogue (see Sect. 3.4).

The islands successfully split in two or three components are 67 in total (64 with two components and 3 with three components).

For non-Gaussian sources we adopted as parameters the reference positions and flux densities defined above. The source position angle was determined by the direction in which the source is most extended and the source axes were defined as largest angular sizes (las), i.e. the maximum distance between two opposite points belonging to the $3\sigma $ flux density contour along (major axis) and perpendicular to (minor axis) the same direction. All the non-Gaussian sources are flagged in the catalogue (see Sect. 3.4).


  \begin{figure}
\par\resizebox{8.8cm}{!}{\includegraphics{H2076f2.ps}}\par
\end{figure} Figure 2: Local to average noise ratio distribution for all the sources with $S_{\rm peak} \geq 6\sigma _{\rm fit}$. Local noise is defined as the average noise in a $\sim 8\hbox {$^\prime $ }$ sided box around each source. The distribution is well fitted by a Gaussian with FWHM = 0.14 and peak value equal to 1.01 (solid line). The excess at large $\sigma _{\rm local}/\sigma _{\rm fit}$ values is due to the presence of systematic noise effects (see text)


  \begin{figure}
\resizebox{8.8cm}{!}{\includegraphics{H2076f3.ps}}\par
\end{figure} Figure 3: $S_{\rm peak}/\sigma _{\rm local}$ vs. $S_{\rm peak}/\sigma _{\rm fit}$. As expected for random noise distributions, the two signal-to-noise ratios show a tight correlation. Dashed lines indicate the $S_{\rm peak}=6\sigma _{\rm fit}$ (vertical line) and $S_{\rm peak}=6\sigma _{\rm local}$ (horizontal line) cut-off respectively. Also shown are the lines (dotted) between which, under the assumption of a normal noise distribution, one would find 99.9% of the sources


next previous
Up: The ATESP radio survey

Copyright The European Southern Observatory (ESO)