VLASS > VLASS SE continuum users guide

VLASS SE continuum users guide

The VLASS Single Epoch (SE) continuum images are intended to be the reference continuum images for each epoch of VLASS. The SE continuum images are of significantly higher quality than the Quick Look images in terms of both dynamic range near bright sources and peak flux densities due to the application of self calibration and better cleaning. Spectral index information is also supplied, though due to the limitations of the imaging algorithms employed this is only trustworthy for compact sources. Users requiring accurate spectral index maps should use the cube products to obtain them.

Introduction

The SE continuum images are higher quality versions of the Quick Look (QL) images that have previously been released. They are made to a finer pixel scale (0.6-arcsec per pixel) than QL, cleaned more deeply, and are self-calibrated. Validation of these images is described in detail in upcoming VLASS Memo 17. The SE images have improved dynamic range close to bright objects by a factor ~2 relative to QL, as well as fewer artifacts. Spectral index information, valid for compact sources (point sources, or sources well-fit by a single gaussian component) is available using the supplied spectral index maps or first Taylor term (tt1) images. A summary of the SE Continuum image characteristics may be found in Table 1.

SE_image_char.png

Table 1: SE Continuum image characteristics. In the formulae for positional accuracy and spectral index accuracy, rho is the signal-to-noise ratio of a compact source (click the table for a higher resolution version).

Release plan

Processing of the VLASS SE continuum imaging begins with Epoch 2, specifically VLASS2.1 (click here for the approximate VLASS processing schedule). For the initial release, we focus on areas of the sky that can be well-imaged using the mosaic gridder in CASA; implementation of the more advanced AW-project gridder in a computationally-efficient form is still a few months away at the time of writing). Within these areas, we further prioritized VLASS tiles where images had been used as part of the Single Epoch continuum validation process, as described in upcoming VLASS Memo 17.
 

Calibration

Calibration has so far been performed using the VLASS recipe in the 2021.1.1.32 pipeline running with CASA 6.1.3.3. We expect no significant changes to the underlying SE pipeline in future releases. The VLASS pipeline recipes may be found in recent joint releases of CASA and the pipeline. The calibration recipes run are hifvcalvlass_compression (or hifvcalvlass if no gain compression fix is required). There were small differences between the running of the pipeline for the Quick Look VLASS 2.1 calibrations and the Single Epoch.

For VLASS 2.1 QL:

  • gain compression recipe (hifvcalvlass_compression) only run if gain compression seen at >5-10%.
  • - RFI flagging extended to include adjacent channels (increases noise without significantly improving image quality, dropped after VLASS2.1)

For VLASS 2.1 SE:

  • gain compression recipe run if gain compression seen at >1-3%.
  • RFI flagging not extended.

From VLASS2.2 onwards the calibration pipeline will only be run once, with the same recipe for QL and SE. The flux densities are calibrated in the pipeline using the Perley-Butler 2017 scale using either 3C286 or 3C138.

The imaging pipeline

Full details of the SE imaging algorithm are given in VLASS Memo 15, but are summarized here and in Figure 1. First, an initial set of images is made with uniform weighting (which minimizes the sidelobes of the synthesized beam). The initial continuum image is first cleaned to 10-sigma using a mask made from a QL component list generated by using PyBDSF. The model from this is then used to perform a single iteration of phase-only self calibration. The self-calibrated visibilities are then reimaged, cleaned with the QL mask to 5-sigma and smoothed to a 5-arcsec beam. This smoothed image is then used to generate a new source list using pyBDSF and a corresponding mask. The original QL mask and the mask from the smoothed image are combined.

The final production image is made using three successive cleaning stages: (1) a clean to 3-sigma using the QL mask from the first step described above (2) a further clean down to 3-sigma using the combined mask from the previous step and (3) a final additional clean to 4.5-sigma with no mask (effectively cleaning anywhere within the primary beam). This iterative cleaning scheme ensures that clean divergences are minimized, while at the same time including diffuse flux in the clean down to 4.5-sigma.

flowchart_web.png

Figure 1: the VLASS SE continuum imaging pipeline. (Click on the image for a higher resolution version.)

All the SE images in the initial release are made with the mosaic gridder (without conjugate beams, which make the spectral indices more robust, but increase the image noise significantly when used with the mosaic gridder). The AW-project gridder has a number of improvements over the mosaic gridder, but images made with AW-project currently take 1-2 orders of magnitude more computational resources than those made with the mosaic gridder. Improvements to the aw-project gridder are being made rapidly, however, so we are restricting the current release to observations where the mosaic gridder produces image quality within the survey requirements (i.e. tiles with median image zenith distances <45 deg and maximum zenith distance of an image <50 deg - typically in the northern sky). Once the improved AW-project gridder is available, we will complete the sky for Epoch 2.

The neglect of w-terms in the gridding algorithms used in cleaning results in position errors on the scale of ~0.1-1 arcsec depending on the zenith distance of the observation (see VLASS Memo 14). These errors occur when using the mosaic gridder---that does not incorporate w-correction---as well as when using the AW-projection gridder with a single w-plane. We mitigate these errors with a single image-plane correction that is applied to the SE images as the last stage of the production workflow. Using the mosaic gridder also affects the measurement of spectral indices, as detailed later in this document and in VLASS Memo 14.

Known issues

Characterization of the validation images revealed several issues, detailed further in VLASS Memo 17. Many of these issues are common to all VLA snapshot observations:

Image fidelity - extended sources

Comparison of VLASS data to  pointed snapshot observations confirms that the typical flux density measurement errors for compact sources are typical for VLA observations, and are within the expected flux calibration and system noise uncertainties (~3%). For extended sources, there is an additional ~10% error in the surface brightness and total flux density measurements, most likely arising from residual phase and amplitude calibration errors as is typical for images where uv-plane coverage is limited. The error increases with source size due to limited sampling of short baselines in the B-configuration, reaching to an error of ~+/-20% for sources about 1-arcmin in size.

Spectral index fidelity

As part of the imaging process, in-band spectral index images are made, and are used to derive spectral indices for compact sources in the survey. The spectral index maps for extended sources produced by this method are not of good quality, suffering from artifacts related to the limited overlap in the uv-plane between the highest and lowest frequency data and small errors in the bandpass calibration. We therefore recommend that users needing accurate spectral index information, particularly for extended sources, derive them from the SE Cube images rather than use the SE Continuum images. There are also problems analogous to chromatic aberration in data taken at large zenith distances (typically in the far south of the survey) when imaged with the mosaic gridder. Tests show that the spectral index variation across a beam FWHM is +0.3 on one side of the beam and -0.3 on the other for observations taken at zenith distances ~45 deg. We therefore will reserve the tiles taken at a median zenith distance >45 deg. (about 50% of the survey) for the AW project gridder. Nevertheless, for compact sources the mosaic gridder images can be used to derive component spectral indices (averaged over 5 pixels centered on the component peak) accurate to +/-0.2 (in the absence of noise), though there remain residual correlations with zenith distance within these errors.

In addition to the systematic effects listed above, the uncertainty on the spectral index measurements due to noise  is dominated by the tt1 image. The uncertainty in VLASS spectral indices Delta(alpha) ~5/rho, where rho is the signal-to-noise ratio of the component. Thus, to obtain a Delta(alpha)< 0.2, source components with rho > 25 are needed (i.e. flux densities >~4 mJy). At low signal-to-noises (<25), there is also a bias towards lower absolute spectral index values as the noise in the tt1 image is normally distributed with a mean of zero. This results in the spectral index estimate given by the tt1/tt0 tending towards zero as the signal-to-noise ratio drops. The magnitude of this effect for VLASS is about 0.2 in the spectral index between the high signal-to-noise regime and the catalog limit (~5 sigma).

Artifacts around bright sources

Many bright (>~ 100 mJy) sources show strong spikes radiating from them that are not removed by cleaning in the SE images. These are typically restricted to within ~30 arcmin of the source, but in cases of very bright or poorly calibrated sources can extend across most of a 1 deg2 image. Also, the originating source can often be outside of the 1 deg2 subimage. Such sources external to the subimage can lead to artifacts above 5-sigma in the mosaic image. These artifacts are probably explicable via a combination of amplitude, pointing and baseline-based calibration errors (the phase errors are well-corrected with self-calibration). We estimate the rate of >5-sigma artifacts on VLASS SE images from these spikes to be ~2/deg2 on average, with about half of these capable of being successfully flagged in the component catalog using a peak-to-ring metric (e.g. Gordon et al. 2021).

Component Catalogs

A component catalog for each image is produced using pyBDSF. These are combined into tile-level catalogs by merging the component lists. The concatenated list contains duplicates due to the small overlap between images; these were removed by selecting whichever source was closest to the phase center of one of the images and rejecting the other(s). Component names of the form ``VLASS2SEDR1 J...'' are added at this step. Also, at this point a modified version of the peak-to-ring metric for flagging possible sidelobes from bright sources (Gordon et al. 2021) was applied. This flag is triggered if (1) there are no other source components within 20'' and (2) the peak in a ring 5''-10'' from the component is >3 sigma and >50% of the component peak and (3) there is another peak >25% of the component peak diametrically opposite. Ultimately, once SE imaging for an Epoch is complete, the tile-level catalogs will be combined and duplicates removed to create a release level catalog.