M33 combined 6cm+20cm catalog description (2018 March 04)

Source detection and fitting

The 6cm+20cm detection image was created by averaging the entire 6cm and 20cm bandpasses. The 20cm high and low frequency images were used, and 14 of the 16 125-MHz channels from the 6cm observations were used. (6cm channels 1 and 2 were contaminated by interference.) The 6cm and 20cm bands are equally weighted in the combined image. That's not optimal for a typical extragalactic source (which has a power law spectrum nu**alpha with alpha=-0.7) but is a reasonably "spectrum-neutral" weighting for the source detection. The resulting source lists do not appear to be a strong function of the weighting.

For this version of the catalog, the images at all frequencies were convolved to a fixed round beam with FWHM 5.9 arcsec. That reduces the resolution of the final image, but it improves the consistency of the multi-resolution filtering step. It also should produce much more accurate spectral indices when integrated over islands, where some fraction of the light might be spilling over the edges of the islands. So I think this should produce more accurate flux densities and (especially) spectral indices than the previous catalog.

A multi-resolution (median pyramid) image processing algorithm was used both for source detection and to subtract the background from the image to make the source detection threshold more uniform. The images without background subtraction can be compared to the original summed image to see the effect of the background subtraction.

The island map (segmentation map) shows where sources where detected. The numbers in the island map correspond to the "Island" field in the catalog. Islands were detected by thresholding the multi-resolution image. Islands detected at more than one resolution are merged. It is possible for several sources detected separately at high resolution to be embedded in one larger source at low resolution; in that case the morphology and sizes of the high-resolution islands are used to determine how to break the large island into pieces. The FITS file with the island numbers is available for download above.

This version does not include Gaussian fits to the flux distribution within the islands. It integrates the fluxes within islands. This can dramatically increase the noise compared with Gaussian fits in some cases. For example, when a point source (FWHM = 5.9 arcsec) is embedded within a broad halo, an unweighted summation over the island adds a lot of noise from the extended halo region even if it does not contribute much to the flux. A Gaussian fit, in contrast, weights the pixels with flux much more heavily and effectively ignores the broad component. The result is much lower noise on the integrated flux density from the Gaussian.

However, the total flux determined from the Gaussian fits is likely to be quite inaccurate (and biased to be underestimated) compared with the island integrated fluxes. That is why we have dropped the Gaussian parameters. But to recover some of the benefits of flux weighting, we have created a fully multiresolution detection algorithm that divides the image up into a stack of images with varying resolution. The islands for a source are determined in each resolution channel, and the flux for that channel is integrated using only the pixels from the same channel. The effect is to use the noisiest, high-resolution structures only over small regions, and to use smoothed, lower-noise channels for low resolution structures. This reduces the noise in the integrated flux densities significantly while retaining a summation that adapts to any flux distribution within the image. The island boundaries in each channel are determined using the FellWalker clump-finding algorithm.

The integrated flux densities are computed over matching images for 4 frequency bands (splitting the 20cm and 6cm bands in half). Then spectral indices are determined by fitting a power law to the 4 band fluxes. The moments in the island are used to determine the flux-weighted source position and major and minor axes (FWHM) of the equivalent elliptical Gaussian. Note that these sizes include the round 5.9 arcsec FWHM beam size. For an unresolved source the size will be approximately equal to 5.9 arcsec. Note that the size is not constrained to be larger than this value; sources with sizes smaller than the beam PSF will have integrated fluxes smaller than their peak fluxes.

The catalog also includes integrated and peak fluxes for the island computed from the detection image. The peak is just the brightest pixel in the island, and the integrated flux is the sum of all the flux within the island with a PSF-dependent correction factor (close to unity) for flux that spills outside the island. The same multi-resolution integration scheme is used for these detection image flux densities.

The images with islands overlaid show the positions of the sources. Red islands have source entries in the catalog; yellow islands either are too faint (less than 4-sigma) or failed in fitting, usually because the region was crowded.

When islands touch, the Gaussian fits to all the islands in the group are carried out simultaneously, so the effects of overlapping sources is properly included. That is an improvement over the previous version of the catalog. The resulting fits in crowded regions are much improved over the older catalog.

SNR, H II region, and X-ray catalogs

Catalog column descriptions

The 2018Mar04 M33 radio catalog from 6cm+20cm data includes 1886 sources, all of which are detected at 4-sigma or greater in island or in spectral index fit. It includes information about associated SNRs, H II regions and X-ray sources. The catalog was searched for radio sources that fall below the 4-sigma threshold but have associated matches with the other catalogs, but no such objects were found.

The catalog is available for download in text or FITS format. The text catalog has comments at the top that describe the contents. Here is a description of the columns in the text catalog:

The flag columns

The flag columns are intended to help make sense of cases with multiple matches (which are fairly common in this very crowded field). Multiple matches are listed in order of decreasing island overlap, so when the first match is considered most likely to be correct. The detection flags (Rflg, Sflg, Hflg, Xflg) are bit flags where the bits are described below.

These values are combined when a match satisfies more than one criterion. The most reliable matches will have flag bit 8 set. All of those will have bit 1 set as well, and most of them will also have bit 2 set. So the best matches have flag = 11, and most of the flag = 9 matches are also reliable. Flag values smaller than 9 indicate confusion about the best match, so the association should be treated with caution.

Here is the distribution of the various flag values in the radio catalog:

For many purposes we can choose samples that are not confused by excluding objects that have flag values other than 0, 9 or 11.

SNR, H II region, and X-ray catalog column descriptionss

The region catalogs (for SNRs, H II regions, and X-ray sources) have similar columns. Additional fields are added giving information about the radio catalog (along with the other external catalogs). The columns are slightly different for all 3 catalogs (because, e.g., the SNR catalog does not have "self-links" for SNRs, and the number of columns for associated objects can differ), but here is the description for the SNR region catalog: