Why PyBDSF?
PyBDSF was developed to serve the needs of LOFAR, the LOw Frequency ARray [1], functioning primarily in The Netherlands. LOFAR achieves orders of magnitude better sensitivity and resolution at low frequencies (15-80 MHz, 120-240 MHz) than previous radio interferometers. In addition, given the large primary beam, the field of view is large (2-16 degrees), and the spectral coverage is wider than usual (up to 48 MHz bandwidth and up to 62,464 channels). Lastly, up to 244 independent beams can be electronically generated on the sky. These capabilities make LOFAR an ideal survey instrument, but also create challenges in data processing. In particular, for surveys a good source extraction software is essential. Before PyBDSF was developed, a survey was made of the existing source extraction packages (SAD in AIPS, SFIND in MIRIAD and SExtractor). It was concluded that none of these packages were adequate to the task, and further, it would be difficult to modify any of these to suit the needs of LOFAR. Hence, PyBDSF was written (first called PyBDSM). However, in the recent years, new low-frequency telescope projects have started which are a similar to LOFAR in some parameters (e.g., MEERKAT, Mileura Array, ASKAP and other SKA pathfinders) and there has been considerable effort to develop source extraction software at some these project sites, e.g. DUCHAMP [2].
Traditionally, source extraction software, at least in radio astronomy, has defined the process as fitting (multiple) Gaussians to source pixels. This makes sense since all interferometric images are convolved with a Gaussian (fit to the main lobe of the dirty beam) after deconvolution. This process is adequate also because most radio images have primarily consisted of point (or slightly extended) sources. LOFAR images, however, are quite different. Note that the antenna diameter is 50 m, maximum baselines extend to 100 km or more, and in addition, LOFAR has almost no missing short spacing measurements (unless flagged due to RFI). Hence, the images will have a much wider range of scales of emission than usual - from point sources up to extended 3C sources. Decomposing such sources into Gaussians may not be very effective (as well as highly degenerate and hence not very useful). Hence alternative basis sets which can capture a variety of scales are essential. AIPS (Classic AIPS as well as CASA) has been experimenting with multi-resolution CLEAN methods for many years now and in the same spirit, we have included shapelet and wavelet decomposition as well. With the kind of image morphologies LOFAR routinely produces, complex ways of describing sources are needed, not just to catalog them but also to perform other filtering operations post-extraction for science purposes. Note that although PyBDSF is written for LOFAR, it will obviously work for images from any radio interferometric telescope.
Footnotes