2.1. Rapid ASKAP Continuum Survey (RACS)

The Rapid ASKAP Continuum Survey (RACSFootnote ^c ; McConnell et al. Reference McConnell2020) is a collection of large-area ASKAP continuum surveys conducted in three frequency bands centred on 888, 1 296 and 1 656 MHz, corresponding to RACS-low, RACS-mid and RACS-high. Multiple epochs exist for each band, and the most relevant RACS versions for this work are RACS-low1 (McConnell et al. Reference McConnell2020; Hale et al. Reference Hale2021) and RACS-low2 (Duchesne et al. in preparation), observed at the same frequency and with the same PAF footprint as VAST. Improvements in the calibration and imaging strategy as described in Duchesne et al. (in preparation), lead to higher resolution, lower rms, and thus more detected sources in RACS-low2 compared to RACS-low1. Table 1 summarises all RACS surveys referenced in this paper, including their observing setup, sky coverage, and the specific roles they play in the VAST Extragalactic DR1 analysis. Note that throughout, when we refer to these RACS surveys, we refer to the curated all-sky catalogues as published and presented in the papers. Similar to the VAST survey (see Section 2.6), the raw data products for RACS are available on CSIRO’s ASKAP Science Data Archive (CASDA; Huynh et al. Reference Huynh, Dempsey, Whiting and Ophel2020) immediately after validation.

Table 1.

Summary of RACS survey versions referenced in this work. For each RACS epoch, the central observing frequency, PAF footprint (see Hotan et al. Reference Hotan2021) and sky coverage are listed, along with their key uses in this paper. [1]: McConnell et al. (Reference McConnell2020) [2]: Hale et al. (Reference Hale2021) [3]: Duchesne et al. in prep, [4]:Duchesne et al. (Reference Duchesne2023), [5]: Duchesne et al. (Reference Duchesne2024).

2.2. Survey footprint

Similar to the VAST pilot survey (Murphy et al. Reference Murphy2021), the VAST survey adopts the RACS-low tiling pattern (McConnell et al. Reference McConnell2020). The VAST Extragalactic Survey comprises 276 fields, with sky coverage shown in Figure 1; the VAST Galactic footprint is included in grey for comparison.

The extragalactic fields were selected to maximise overlap with existing multi-wavelength surveys. The equatorial regions (centred at $\mathrm{Dec}=0^\circ$ ) provide strong northern-hemisphere multi-wavelength coverage and can be observed by both northern and southern facilities, aiding follow-up of transient events. Fields between $-30^\circ \lt \mathrm{Dec} \lt -60^\circ$ and RA = 20–6 h were chosen for their overlap with the Dark Energy Survey (DES; D.E.S. Collaboration et al. Reference Collaboration2016). Additional southern fields extend the overall coverage and distribute the footprint across a wide range of right ascensions to facilitate scheduling. The equatorial fields overlap with FIRST, NVSS and VLASS, while fields below $\mathrm{Dec}=-30^\circ$ overlap with the Sydney University Molonglo Sky Survey (SUMSS; Bock, Large, & Sadler Reference Bock, Large and Sadler1999; Mauch et al. Reference Mauch2003).

The RACS tiling pattern introduces some field overlap; the right ascension spacing varies from $6.2^\circ$ at $\mathrm{Dec}=0^\circ$ to $10.6^\circ$ , corresponding to a projected separation of $\cos(56.3^\circ)\times10.6^\circ = 5.9^\circ$ at $\mathrm{Dec}=-56.3^\circ$ . As described in Section 3.1, VAST light curves use the central $6.67^\circ \times 6.67^\circ$ of each ASKAP field. Some sources are therefore observed as part of multiple fields and receive correspondingly more light curve points. Assuming a $6.67^\circ \times 6.67^\circ$ effective area for all 276 fields gives a total survey footprint of $\sim 12\,300\ \mathrm{deg}^2$ .

2.3. Survey strategy

The primary goal of the VAST Extragalactic Survey is to detect extragalactic synchrotron transients, which typically evolve on month-long timescales (see, e.g. Pietka et al. Reference Pietka, Fender and Keane2015). Such slow evolution means that relatively sparse repeat observations are sufficient to characterise them. The survey is also expected to detect radio stars, which are isotropically distributed at local distances (Driessen et al. Reference Driessen2024b).

The VAST Extragalactic Survey observes the full footprint (as described in Section 2.2) every two months. Each VAST observation has an integration time of 10–12 min, yielding a median image rms noise of $0.24$ mJy $\mathrm{beam}^{-1}$ . This implies that all VAST Extragalactic fields can be observed within 2–3 d of observing time (not considering overheads and the schedulability of other ASKAP surveys).

Figure 2.

Summary of VAST Extragalactic observations, giving the number of fields in each epoch. From 2024 the strict 2-month cadence was removed, meaning that a full reobservation of all 276 fields was often spread over multiple epochs. The greyed out epochs are not included in the VAST Extragalactic DR1 (see Appendix A for details).

Figure 2 summarises the number of fields observed over time and per epoch. Extragalactic observations began in epoch 36; earlier epochs contained only Galactic or pilot observations. Epochs vary in length–from 2 to 52 d–and their definition has evolved throughout the survey. Early in the full VAST survey, all 276 extragalactic fields were typically observed within a few days, producing short epochs. By late 2023, relaxed scheduling and the removal of the strict two-month cadence meant that full-field coverage was often distributed across multiple epochs. Several epochs (40, 43, 46, 72, 73, 74) captured nearly all extragalactic fields, with durations of 8, 16, 8, 20, 26, and 52 d, respectively. The increased length of epoch 74 can be attributed to a combination of scheduling difficulties (considering higher priority observations and avoiding Solar interference) and software changes that delayed processing. As of epoch 72, each extragalactic epoch includes one observation of every extragalactic field. Figure 2 also shows greyed out observations from December 2023 to April 2024 corresponding to epochs 48–55, which are excluded from VAST Extragalactic DR1 due to incorrect holography calibrations causing flux density inconsistencies (see Appendix A). After excluding these epochs, VAST Extragalactic DR1 includes 2945 observations from June 2023 to May 2025.

2.4. Main changes with respect to the VAST pilot surveys

Below we list the main changes in survey footprint and strategy between the full survey and the pilot survey (Murphy et al. Reference Murphy2021).

• • The full VAST Extragalactic Survey consists of 276 fields. This survey footprint has changed substantially compared to the pilot survey. In the pilot survey, only 113 Galactic and extragalactic fields were included.

• • The full VAST Extragalactic survey only observes in ASKAP low-band, with a central frequency of 888 and 288 MHz of bandwidth. In the pilot surveys, part of the VAST observations were made using the mid band (1 296 MHz). After the pilot surveys, we decided to remove the mid-frequency part of the VAST survey since this would allow for direct comparison of all VAST observations, without having to account for spectral effects. Furthermore, the radio frequency interference (RFI) conditions are worse in the mid-band (see, e.g. Section 2.1 in McConnell et al. Reference McConnell2020) and the low band has a larger field of view per observation.

• • In the full VAST survey we treat every observation and image independently. In the pilot surveys, images would be combined to a larger mosaic image. In the pilot, back-to-back observations allowed combining fields within an epoch for uniform sensitivity. In the full survey, scheduling constraints create gaps of up to weeks between adjacent fields, and therefore, mosaicking adjacent observations no longer makes physical sense.

We exclude the pilot survey observations (AS107 in CASDA) from this VAST Extragalactic data release for three main reasons:

• • There is a large temporal gap between the pilot observations and the start of the full survey, with no data taken between mid-2021 and mid-2023.

• • The full survey footprint is more than twice as large, so incorporating the pilot data – available for only a subset of sources – would lead to uneven light-curve sampling.

• • Significant improvements in ASKAP processing have introduced different systematic effects between the pilot and full surveys (see Section 2.5.1).

2.5. Data processing

The VAST data are processed using the same calibration and imaging strategy as used in the VAST pilot survey (Murphy et al. Reference Murphy2021) with two improvements introduced as a result of the experience gained with RACS-low1 (McConnell et al. Reference McConnell2020) and RACS-mid1 (Duchesne et al. Reference Duchesne2023; Duchesne et al. Reference Duchesne2024). The first is to convolve the point spread functions (PSF) of individual PAF beams to a common PSF to maintain integrated flux density integrity across the observed field. The second is the use of appropriate holography to correct for primary beam attenuation; the holography is updated each time a new set of PAF beams is formed, typically once every four months.

Figure 3.

A VAST image (VAST 2131-56) from epoch 74 with two cutouts. Cutout A: a 1 deg image centred on (J2000) (12:49:26.5, $-55$ :17:52.87) containing several bright sources, including the large radio galaxy 2MASX J21512991-5520124. Cutout B: a 0.3 deg image centred on (J2000) (21:36:18.9, $-58$ :00:12.68) containing a range of source morphologies. The central part of the full image, shown by the black dashed line, indicates the coverage included in the post-processed data products.

We use ASKAPSoft (Cornwell et al. Reference Cornwell2011) to create images and source catalogues for the VAST observations. When imaging, we removed short ( $\lt$ 100 m) baselines in order to (i) minimise solar interference; and (ii) as there is insufficient (u,v)-coverage to reliably recover extended structure in the short VAST observations. Figure 3 shows a typical VAST Extragalactic DR1 image. To allow for a comparison with the pilot survey data, we have chosen the same representative image as shown in Murphy et al. (Reference Murphy2021). The region indicated with the black dashed line indicates the coverage included in the post-processed data products (Section 3). In addition to the images, the Selavy source finding software (Whiting Reference Whiting2012), incorporated into the ASKAPSoft pipeline, produces source catalogues. We used the default parameter settings in ASKAPSoft, where sources are considered ‘detected’ if they exceed 5 times the local rms noise, which corresponds to $5\cdot 0.24\approx1.2$ $\mathrm{mJy\, beam}^{-1}$ in the VAST Extragalactic Survey (see Figure 8).

2.6. Data validation

ASKAP Survey Science Projects have no proprietary period, and are released on CASDA after validation. VAST observations are grouped in epochs, and all observations in a single epoch are released to CASDA simultaneously. Before the data are released on CASDA, a few quality control checks are performed on the astrometric accuracy and flux density scale relative to RACS-low1 (Hale et al. Reference Hale2021) and RACS-low2 (Duchesne et al. in prep). Systematic offsets in astrometry and or flux density are corrected in post-processing (see Section 3), but observations with large structures in astrometric offsets or large spreads in flux density offset are flagged as ‘uncertain’ or ‘rejected’. Additionally, each image is visually inspected to check the overall image quality. If the image quality is poor in a small part of the image, the observation is still included in VAST Extragalactic. However, if the overall image quality is severely reduced, for example, by artefacts across most of the image, it is rejected. This occurs when there is RFI or solar interference during the observation, or when a bad antenna is not flagged. Within the VAST Extragalactic DR1, only 4 observations were initially rejected and reobserved, compared to a total of 2945 observations.

2.7. Data flow

Figure 4 presents a high-level overview of the data flow of VAST observations. In Sections 2.2, 2.3, 2.5 and 2.6 we highlighted the steps that produce the raw images and source catalogues (available through CASDA). These steps are highlighted in purple in Figure 4. Section 3.1 below details the post-processing step before data is ingested into the VAST transient detection pipeline (Section 4.1), highlighted in purple and orange in Figure 4 respectively.

Figure 4.

A flow chart describing the high-level data flow of VAST observations. Each block refers to a section describing that step in more detail. The purple/blue steps result in the data products as available on CASDA quickly after an observation. The pink and orange steps are additional steps as introduced in this work.

3.1. VAST post-processing

The VAST post-processingFootnote ^d (Jiang et al. Reference Jiang, Dobie and O’Brien2024) applies corrections to the catalogues and images so we can extract reliable light curves. Below we discuss each of the steps in the post-processing pipeline.

3.1.2. Astrometric corrections

Per VAST field, we construct a RACS-calibrator sample. For all cross-matched sources, we calculate the right ascension and declination offsets between RACS and VAST. We define the astrometric offset as the median offset of all cross-matched sources in a single VAST field. We subsequently use this median offset as the astrometric correction to the VAST field, which is applied to all the VAST sources in the Selavy source catalogues and the FITS image.

Figure 5.

Astrometric offset between RACS-low2 and VAST before post-processing (in blue) and after post-processing (in orange). Each dot represents a single observation of a VAST source cross-matched to the RACS-calibrator sample (see text). The gradient in the orange datapoints represents the point density. The green dashed line shows the typical $2.5 \times 2.5 {''}$ pixel size in VAST images.

Figure 5 shows how the astrometric offset between RACS-low2 and VAST Extragalactic improves from the post-processing corrections. The blue dots represent the offset of RACS-calibrator sources in all VAST Extragalactic observations, as calculated using the source catalogues before post-processing. The orange dots represent the offsets of RACS-calibrator sources after we apply a single astrometric shift per observation. After post-processing, the offsets in both right ascension and declination are closer to zero, and the width of the offset distribution has decreased. The green dashed line shows the typical $2.5 \times 2.5 {''}$ pixel size in VAST images. The VAST post-processing reduces the average offset in right ascension from $-0.23 \pm 1.18 {''}$ to $-0.04 \pm 0.60 {''}$ , and in declination from $0.10 \pm 0.98 {''}$ to $0.00 \pm 0.44 {''}$ .

The astrometry correction in the post-processing effectively ties the astrometry of all VAST observations to the RACS-low2 astrometry. This is important for the VAST pipeline (Section 4.1), where we build light curves by associating new measurements to sources detected in previous epochs, based on their position. The post-processing reduces the position scatter of individual measurements of a single source, decreasing the number of false associations. The right side of Figure 6 shows the astrometry corrections that are applied to all 2 945 observations in the VAST Extragalactic Survey. Some observations have astrometry corrections of over 3^″, showing why these corrections are important to ensure proper associations of measurements in the VAST pipeline.

Figure 6.

Post-processing corrections for each of the 2 945 VAST Extragalactic DR1 observations. Left: flux density correction factor. Right: right ascension and declination corrections.

3.1.3. Flux density scale corrections

We perform a flux density correction in the post-processing to account for any flux density offsets that might be present in an individual VAST observation (for example due to observing/ionospheric conditions or calibration solutions). Outlier observations would introduce artificial variability to all sources in a field if unaccounted for. We use the RACS-calibrator sample to scale the VAST integrated flux density to the RACS integrated flux density.

We use the integrated flux density rather than the peak flux density to account for possible beam shape differences between the VAST and RACS observations. We calculate a single scaling factor per observation that is applied to the integrated and peak flux density of the sources detected in that observation. The left side of Figure 6 shows the flux density correction factors that are applied per observation in VAST Extragalactic DR1. The flux scaling factor is calculated by fitting the VAST integrated flux density as a function RACS-low2 integrated flux density for the RACS-calibrator sources. We use a Huber regressor to fit the data, as this statistic is more robust against different levels of variance and outliers, compared to the more commonly used orthogonal distance regressor (Huber Reference Huber1992). We opt against simply using the mean or median of the histogram of the ratio between the VAST and RACS flux density, as the errors on these quantities are ill-defined for skewed distributions.

We quantify how the flux density correction process impacts the light curve variability by calculating the root mean square deviation, $\sigma_{\mathrm{flux}}$ , in the light curve for all sources in the RACS-calibrator sample. In other words, how much do the individual measurements in the light curve deviate from the mean flux density? $\sigma_{\mathrm{flux}}$ is defined as

(1) \begin{equation} \sigma_{\mathrm{flux}} = \sqrt{\frac{\sum^{N_T}_{t=1} \left( S_t - \overline{S_t} \right)^2}{N_T}}\end{equation}

with $N_T$ the number of samples in the light curve, $S_t$ the integrated flux density at a given time t and $\overline{S_t}$ the mean integrated flux. The left panel of Figure 7 shows the median root mean square deviation for RACS-calibrator sources in a single observation over the duration of VAST Extragalactic before (blue) and after (orange) post-processing. The right panel of Figure 7 shows the ratio between the median $\sigma_{\mathrm{flux}}$ after and before post-processing. The flux density correction in post-processing reduces the median $\sigma_{\mathrm{flux}}$ to $0.82 \pm 0.08$ of its original value.

Figure 7.

Left: Median root mean square deviation ( $\sigma_{\mathrm{flux}}$ ) of RACS-calibrator sources in a single field over the duration of VAST Extragalactic. The blue distribution shows the median $\sigma_{\mathrm{flux}}$ before post-processing, while the orange distribution shows the median $\sigma_{\mathrm{flux}}$ after post-processing. Right: ratio between the $\sigma_{\mathrm{flux}}$ after and before post-processing. The red dashed line shows the median of this distribution at $0.82$ .

This single flux density correction factor removes artificial variability that is present for all sources in a single observation. The left side of Figure 6 shows that for most VAST Extragalactic DR1 observations, the flux density correction factors are close to 1.0. However, there are a few observations where the flux density correction factor corrects for flux density variations of 5–10%. If unaccounted for, these flux density variations could be interpreted as real astrophysical variability. Applying a flux density correction allows us to probe deeper in a search for astrophysical variability.

3.2. Image quality

A histogram of the rms noise in each Stokes I VAST Extragalactic image is shown in Figure 8. The median rms is 0.24 $\mathrm{mJy\, beam}^{-1}$ . This rms is influenced by source confusion, sidelobe confusion, and deconvolution artefacts. Figure 9 shows the median rms value over all observations for each field. An increased noise at higher declination is consistent with what was observed in RACS-mid (Duchesne et al. Reference Duchesne2023). There are three fields with significantly higher noise values, which can be attributed to particularly bright sources in the field. For example, VAST 1237+00, an equatorial field around RA=12:36:00, shows strong artefacts in the North-West portion of the field, around 3C273. This is a bright quasar with an integrated Stokes I flux density of around 64 Jy, which causes deconvolution artefacts over a large part of the image. The typical resolution of these images is 12–20^″.

Figure 8.

Distribution of median image rms values (computed over the central half of the image), for each observation in VAST Extragalactic DR1.

Figure 9.

Distribution of image rms values for each field after post-processing, showing the median of all observations for a single field. The yellow fields (with high rms) contain bright radio sources 3C161, Her A and 3C273, respectively.

3.3. Absolute flux density scale

We compare the overall flux density scale of VAST Extragalactic DR1 to SUMSS (Mauch et al. Reference Mauch2003), since the surveys are observed at roughly similar frequencies of 843 and 888 MHz, respectively, and the survey footprints overlap. The SUMSS fluxes have been calibrated against the Molonglo calibrator sample and have been found to be consistent with NVSS (Hunstead & Gaensler Reference Hunstead and Gaensler2000). Following the approach outlined by Hale et al. (Reference Hale2021), comparing VAST to SUMSS minimises the effect of a spectral index correction for flux density comparisons. Assuming an overall spectral index of $\alpha =-0.8\pm0.1$ , we expect flux density offsets of only $\pm 0.5\%$ when comparing VAST and SUMSS, whereas comparing VAST to a 1.4 GHz survey such as NVSS would impose a $\pm5\%$ flux density error due to the spectral index scaling.

Figure 10.

Flux density comparison between the average VAST peak flux density and the SUMSS peak flux density at a frequency of 888 MHz (assuming $\alpha=-0.8$ ). The black dashed line indicates the median of the flux density ratio distribution, while the grey dashed lines indicate the 16th and 84th percentiles.

We assess the absolute flux density scale of VAST sources. The filtering steps we undertake to ensure a reliable sample of sources and light curves are detailed in Section 4.3. We match these VAST sources to SUMSS, including only VAST sources that are highly compact (integrated flux density over peak flux density ratio below 1.2) and have no forced flux density extractions (i.e. all their flux density measurements are over $5\sigma$ ) and finally only SUMSS-VAST matches that are within 5^″. This yields a sample of 19381 SUMSS-VAST source pairs. Figure 10 shows the average VAST peak flux density as a function of the SUMSS peak flux density for each of these sources. The SUMSS peak flux density is scaled to VAST observing frequencies, assuming a spectral index of $\alpha=-0.8$ . The median flux density ratio is $0.98^{+0.14}_{-0.11}$ . The associated errors indicate the 16th and 84th percentile of the flux density ratio distribution. Lines with these slopes are overplotted on the distribution in Figure 10.

3.4. Absolute astrometry

The absolute astrometry in VAST depends on the absolute astrometric accuracy of RACS-low2 (Duchesne et al. in prep). There are two main considerations in the RACS astrometry: the astrometric precision and astrometric accuracy. The astrometric precision is determined by the ASKAP PAF beam-to-beam variations, which were found to be $\lesssim 2.5{''}$ . This was noted in McConnell et al. (Reference McConnell2020), and detailed in Figure 36 in Duchesne et al. (Reference Duchesne2023). The astrometric precision implies that the overall astrometry scale of RACS (and thereby VAST) cannot be much better than this $2.5{''}$ limit. This limitation comes from a lack of phase referencing to a suitable catalogue during processing. A new catalogue based on RACS-low3 data with astrometric calibration will be used for phase referencing in future VAST data releases (see Section 2.5.1).

The astrometric accuracy is determined by cross-matching to external surveys. When cross-matching RACS-low1 to the International Celestial Reference Frame 3 (ICRF3) catalogue (Charlot et al. Reference Charlot2020), it was found that there is a systematic offset between the two; $\Delta \mathrm{RA} \cdot \mathrm{cos(Dec)} = -0.6\pm0.6{''}$ and $\Delta \mathrm{Dec} = -0.4\pm0.7{''}$ (McConnell et al. Reference McConnell2020). Similar bulk offsets and scatter are found in RACS-mid (Duchesne et al. Reference Duchesne2023) and RACS-low2 (Duchesne et al. in preparation). The cause of this offset is currently not understood.

We evaluate the astrometric accuracy of the VAST sources presented in this data release. The filtering steps we undertake to ensure a reliable sample of sources and light curves are detailed in Section 4.3. We cross-match these VAST sources with the ICRF3. The ICRF3 consists of 4 588 radio sources with accurate positions measured using very long baseline interferometry at frequencies of 8.4 and 2.3 GHz. The median positional uncertainty is about 0.1 mas in right ascension and 0.2 mas in declination. We cross-match VAST to the ICRF3, yielding 76 matched pairs that are separated by less than 3”. The number of match pairs is insufficient to be robust against the ASKAP PAF beam-to-beam astrometry variations. To obtain a larger number of matches, we also cross-match the VAST sources against the VLASS epoch 1 Quick Look catalogue (Gordon et al. Reference Gordon2021). The VLASS catalogue consists of 3.4 million sources north of declination $-40^{\circ}$ observed at 3 GHz. The astrometric accuracy is around 1^″, increasing to $0.5{''}$ above declinations $-20^{\circ}$ (VLASS Quick Look catalogue user guide). We cross-match VLASS to VAST, yielding 183 793 matched pairs that are separated by less than 5^″. We note that this cross-matched sample of source only includes sources with declinations North of $-40^{\circ}$ (VLASS coverage).

Figure 11.

Astrometric offsets for sources in VAST Extragalactic DR1 (selection criteria detailed in Section 4.3) compared to VLASS (black) and ICRF3 (red). The blue dotted circles show radii of $1.2{''}$ and $2.5{''}$ .

Figure 11 shows the astrometric offset between the VAST sources that are cross-matched to ICRF3 (red) and VLASS (black). For the VAST to ICRF3 cross-matched sources we find $\Delta \mathrm{RA} \cdot \mathrm{cos(Dec)} = -0.4 \pm 0.4$ and $\Delta \mathrm{Dec} = -0.1 \pm 0.5$ . For the VLASS cross-matched sources we find $\Delta \mathrm{RA} \cdot \mathrm{cos(Dec)} = -0.5 \pm 0.6$ and $\Delta \mathrm{Dec} = -0.3 \pm 0.6$ . The median systematic offset for both samples is similar to the bulk offset noticed in the RACS surveys (McConnell et al. Reference McConnell2020; Duchesne et al. Reference Duchesne2023; Duchesne et al. in preparation). In the following, we consider the astrometric offsets of VAST compared to VLASS, as these numbers seem most consistent with the RACS offsets, and the ICRF3 cross-matched sample consists of only 76 sources.

We aim for a conservative and pragmatic astrometry scale to use when cross-matching VAST sources to multi-wavelength counterparts. To this end, we calculate the $3\sigma$ offset in the direction with the largest median offset and scatter. This is the right ascension offset, $-0.527-3\cdot 0.646 \approx 2.5{''}$ . Although the actual astrometric accuracy is not uniform, i.e. different systematic offset and variance in different directions, we conservatively use this $2.5{''}$ value as a cross-match radius. Additionally, we keep in mind that most beam-to-beam variations (astrometric precision) were contained within $2.5{''}$ . This $2.5{''}$ absolute astrometric uncertainty should be added in quadrature to the VAST averaged positional uncertainty ( $\sigma_{\mathrm{pos}}$ ). One will often find that the absolute astrometric uncertainty dominates.

We note that for individual sources of interest, it is possible to refine the astrometry by performing local corrections, for example with software like fits_warp (Hurley-Walker & Hancock Reference Hurley-Walker and Hancock2018), which calculates local astrometry corrections to radio images by cross-matching the radio source locations to optical or infra-red surveys.

4.1. Pipeline

The overall design of the VAST transient pipeline (Stewart et al. Reference Stewart and Serg2024) is described in detail in Murphy et al. (Reference Murphy2021) and Pintaldi et al. (Reference Pintaldi, Stewart, O’Brien, Kaplan and Murphy2021). For completeness, only a short description of the most important steps and concepts is provided here.

The main goal of the VAST pipeline is to associate sources from different observing epochs with each other via positional cross-matching. The pipeline takes as input a set of images, rms maps and mean background maps (all as FITS files) as well as the Selavy source finder component list. These data products are all default outputs from the ASKAPSoft continuum imaging process, which means the VAST pipeline can be run immediately after the ASKAPSoft data reduction and VAST post-processing. Once sources have been identified, forced flux density extractions are performed for previously detected sources that are not detected in every subsequent epoch. These forced flux density measurements are obtained by fitting a two-dimensional Gaussian at the average source position in any observation lacking a $\gt5\sigma$ detection. The fit solves only for the Gaussian peak amplitude, while the shape and orientation are fixed to match the PSF of the observation. We note that due to the nature of radio images, these fitted flux densities can be negative. This results in a light curve database for every source detected in one or more epochs, and a set of variability metrics.

4.2. Variability metrics

The VAST pipeline calculates two variability metrics ( $\eta$ and V) for each light curve. These metrics are commonly used to identify highly variable and transient sources (Swinbank et al. Reference Swinbank2015; Rowlinson et al. Reference Rowlinson2019). The flux density coefficient of variation over N measurements is defined as the ratio of the mean flux density to the sample standard deviation s,

(2) \begin{equation} V = \frac{s}{\overline{S_t}} = \frac{1}{\overline{S_t}} \sqrt{\frac{N}{N-1} \left( \overline{S_t^2} - \overline{S_t}^2\right)}\end{equation}

with $S_t$ an individual flux density measurement with uncertainty $\sigma_t$ . The significance of the flux density variability, $\eta$ , is defined as a reduced- $\chi^2$ expression in conjunction with the weighted mean flux density $\xi_{S_t} = \frac{\sum^n_{t=1} S_t/\sigma_t^2}{\sum^n_{t=1} 1/\sigma_t^2}$

(3) \begin{equation} \eta = \frac{1}{N-1} \sum_{t=1}^N \frac{\left(S_t - \xi_{S_t}\right)^2}{\sigma_t^2}\end{equation}

Sources that have both a high variability amplitude (V) and a high reduced $\chi^2$ statistic when compared to the weighted mean flux density ( $\eta$ ), are the most variable sources in the sample.

4.3. Selection criteria

The VAST pipeline run over the 2 945 VAST Extragalactic images in this DR1 results in 27.5 million individual measurements for 2.3 million sources. However, we disregard most of these sources ( $\sim$ 76%), as they are likely to be spurious or are found at low signal-to-noise ratio. We filter out

1. 1. sources that have fewer than five measurements;

2. 2. false ‘sources’ that are created based on a single flux density measurement that is due to part of an image having relatively high rms noise;

3. 3. sources caused by imaging artefacts around bright sources due to incomplete dirty beam subtraction;

4. 4. extended sources, which can be decomposed into multiple ‘sources’ by the source finder, making the flux density measurements unreliable;

5. 5. sources found at low signal-to-noise ratios.

Below we detail the consecutive filters and strategies we impose to reduce the spurious sources in our dataset. We suggest future improvements to this filtering approach in Section 6.2.

4.3.3. Filtering out imaging artefacts around bright sources

To limit contamination from imaging artefacts near bright sources, we require each source to have a nearest neighbour more than 30^″ away. This is roughly two times the median restoring beam size, which varies between 12^″ and 20^″ and peaks at 15^″. We assess this threshold using a sample of truly compact, isolated RACS-low2 sources (Duchesne et al. in preparation). This sample of sources has an ‘S’ source code in RACS (meaning they are fit by a single Gaussian), and a nearest neighbour source more than 2^′ away. We overlay truly compact sources from this sample across integrated flux densities of 2, 10, 50, and 200 mJy. The left panel of Figure 12 shows an overlay of a sample of 100 truly compact 2 mJy sources on top of each other. The right panel of Figure 12 shows how the peak of the noise distribution changes in radial annuli around the source. Only the stack of 2 mJy sources is shown here as an example. The largest change in noise distribution occurs between the 20–30^″ and 30–40^″ annuli for all flux density bins, supporting our choice of a 30^″ cutoff. A more elaborate filtering strategy could be designed by setting a larger radial filter around brighter sources, but that is beyond the scope of this paper.

Figure 12.

Left: $5{'} \times 5 {'}$ cutout of a summed stack of 100 2 mJy isolated compact sources. This figure is made for a stack of 10, 50 and 200 mJy sources as well. Right: peak of the noise distribution in each annulus drawn in the left figure.

We find that $\sim$ 1 million sources are filtered out by this distance to nearest neighbour criterion alone (i.e. they satisfy the extended sources and signal-to-noise ratio filter below), effectively reducing the survey size by about 220 $\mathrm{deg}^2$ .

4.4. Pipeline products and data availability

The main output of the transient pipeline is a light curve database that consists of a measurement, source and image table. The measurements table contains detailed information about each Selavy or forced flux density measurement per source and per timestamp. The source table contains the statistics per source averaged/calculated over all measurements, such as average location, maximum and average integrated/peak flux density, and the variability parameters. Finally, we include a table with detailed information on the images and observations that are part of VAST Extragalactic DR1. These tables represent the light curve database after applying the filtering steps detailed in the previous section, resulting in 0.5 million light curves. Appendix B describes the light curve database in detail. The light curve database tables are presented as CSV and parquet files (Vohra Reference Vohra2016) which can be read by data exploration libraries such as Pandas and dask.Footnote ^e To assess the quality and reliability of the light curves, image cutouts are important. For example, it can be the only way to easily identify spurious sources/measurements due to high local noise, or imaging artefacts around bright sources. A FITS cube is available for each source, containing layers of cutouts of the individual detections of the source. These FITS cubes are labelled with the unique source identifier from the light curve database (see Appendix B).

The light curve database and image cutouts for all sources in this data release are available on the CSIRO Data Access Portal (DAP) under the project ’VAST Extragalactic DR1: light curve database and cutouts’: https://doi.org/10.25919/nh9d-t846. This collection additionally provides an example notebook demonstrating how to access and work with the light-curve database and cutouts.

We encourage the wider community to make use of the light curves included in this data release. However, we urge the user to consider the following caveats:

1. 1. This data release will contain spurious sources and measurements, despite the filters detailed in the previous section. We encourage users to inspect the image cutouts for their sources, to ensure that the light curves only include reliable flux density measurements.

2. 2. This data release does not contain light curves for all sources in the VAST Extragalactic survey footprint. By filtering out all sources with a nearest neighbour within 30”, we exclude both bright sources that are decomposed into multiple components by the source finder, and any otherwise reliable sources that happen to have another source close by. Additionally, we exclude faint and extended sources in this data release.

We remind the reader that the following data products are available through CASDA under project code AS207:

• • the measurement sets and images per field (Stokes IQUV) http://hdl.handle.net/102.100.100/473995?index=1

• • the Selavy source catalogues (Stokes I) http://hdl.handle.net/102.100.100/473996?index=1

• • the images and Selavy source catalogues per field after post-processing (Stokes I) https://doi.org/10.25919/7597-df49

We do not release rms maps and background maps (see Murphy et al. Reference Murphy2021), as we expect limited interest in these data products. However, in case these are of use for your science goals, we encourage you to join the VAST collaboration.Footnote ^f

5.1. Known pulsars

All pulsars in the sample of variable sources are identified in an automated manner, by cross-matching the sample to the pulsar survey scraperFootnote ^g (Kaplan Reference Kaplan2022) with a cross-match radius of 10^″. 27 sources amongst our sample of most variable sources in VAST Extragalactic DR1 are known pulsars. Figure 15 shows the light curve and cutouts for ASKAP J102438.6-071920, cross-matched to millisecond pulsar PSR J1024-0719, as a representative example of a typical pulsar light curve. For two pulsars, PSR J1556+00 and PSR J0051+0423, corresponding to ASKAP J155540.5+004904 and ASKAP J005129.7+042259, respectively, a larger cross-match radius was required, as the archival pulsar positions have not been determined to sufficient accuracy with timing studies.

Thirteen of the 27 pulsars are confirmed binary millisecond pulsars (MSPs). Most notably, PSR J2222-0137 (Boyles et al. Reference Boyles2013) and PSR J2234+0611 (Deneva et al. Reference Deneva2013) are MSPs that have been used for tests of gravity (e.g. Guo et al. Reference Guo2021; Batrakov et al. Reference Batrakov2024; Stovall et al. Reference Stovall2019). PSR J2129-0429 (Hessels et al. Reference Hessels2011; Bellm et al. Reference Bellm2016) is classified as a redback MSP, while PSR J1227-4853 is a transitional redback (Roy et al. Reference Roy2015). PSR J2051-0827 (Stappers et al. Reference Stappers1996) and PSR J2039-5617 (Corongiu et al. Reference Corongiu2021) are eclipsing MSPs. In addition, two pulsars are thought to be recycled systems whose binaries were disrupted or transformed. PSR J1320-3512 (Belczynski et al. Reference Belczynski, Lorimer, Ridley and Curran2010) may have been separated during a second supernova explosion (Lorimer et al. Reference Lorimer2004), while PSR J2124-3358 (Reardon et al. Reference Reardon2016; Romani, Slane, & Green Reference Romani, Slane and Green2017) may have ablated its companion through spin-down power in an extreme ‘black widow’ phase (Ruderman, Shaham, & Tavani Reference Ruderman, Shaham and Tavani1989). The remaining MSPs are PSR J1024-0719 (Kaplan et al. Reference Kaplan2016; Bassa et al. Reference Bassa2016), PSR J2144-5237 (Bhattacharyya et al. Reference Bhattacharyya2016; Bhattacharyya et al. Reference Bhattacharyya2019), PSR J1337-4441 (Spiewak et al. Reference Spiewak2020), PSR J0034-0534 (Bell et al. Reference Bell, Kulkarni, Bailes, Leitch and Lyne1995), PSR J2145-0750 (Löhmer et al. Reference Löhmer2004; Bell et al. Reference Bell, Kulkarni, Bailes, Leitch and Lyne1995), and PSR B0820+02 (Kulkarni Reference Kulkarni1986).

5.2. Radio stars

We identify radio stars by cross-matching to within $2.5{''}$ of a star in the Sydney Radio Star Catalogue (SRSC; Driessen et al. Reference Driessen2024b). In addition, we cross-match our highly variable radio sources to the proper-motion-corrected Gaia positions of nearby stars. We classify matches with $2.5{''}$ as radio stars, with the exception of ASKAP J014700.3+075128 which has a $3.7{''}$ offset, see Section 5.2.1. We find that 40 of the most variable sources in the VAST Extragalactic DR1 are stars. Figure 15 shows the light curve and cutouts for ASKAP J030402.5+004550, a highly variable VAST source matched to LP591-256 (Gaia DR3 ID 3266760032372995072). This light curve is a typical light curve for a radio star. For the cutouts around radio stars (Figures 15 and 16), the VAST position of the source has been corrected according to the proper motion of the star, such that the radio detection is corrected to the time of the optical observation.

Of the 40 stars in this sample, 30 are known radio stars and 10 are newly detected radio stars. Of the known radio stars, Wolf 359 (aka CN Leo), is not yet in the SRSC but was previously identified as a radio star (Gudel et al. Reference Gudel, Schmitt, Bookbinder and Fleming1993). The same holds for HD 34736 (Semenko et al. Reference Semenko2024; Das et al. Reference Das2025). We have counted both of these, and the 28 stars in the SRSC as known radio stars. Some well-known radio stars that are (re)detected in VAST Extragalactic DR 1 are AT Mic and CU Vir. Some of the new radio stars have limited information, but of those with spectral type classifications, they range from M dwarfs to G-type stars. Many of the stars, both new and previously known radio stars, have previously been identified as active with activity indicators such as X-ray emission or Calcium H and K emission. We find that 8 out of the 10 new radio stars have high fractional circular polarisation (between 14% and 83%). We discuss the two new radio stars with low circular polarisation fractions (ASKAP J014700.3 $+$ 075128 and ASKAP J053954.1-051119) in more detail below.

Table 2.

Highly variable sources identified in VAST Extragalactic DR1. The coordinate of each source is given as the weighted average of all Selavy detections, where the weight is the inverse square of the positional error. $\sigma_{\mathrm{pos}}$ is the averaged positional uncertainty. $\eta$ and V are the variability parameters as described in Section 4.2. $N_T$ gives the number of epochs (observations) that cover the source location, which is equal to the number of points in the light curve. $N_f$ gives the number of forced flux extractions. $S_{\mathrm{max}}$ gives the maximum integrated flux density for this source. The last column shows the source the VAST detection has been matched to, or for the AGN, the survey that previously detected the AGN. The numbers in italics for the stars refer to the Gaia DR3 identifier. A reference to the light curve is included if a source is shown in the text, either as an example of the class (Figure 15) or for a more detailed discussion. A machine-readable version of this Table can be found in the Supplementary Materials.

Figure 15.

Example light curves per source class, each row represents a different source. Left: VAST Extragalactic DR1 light curve. Black points are integrated flux density measurements from Selavy. White points are the forced-fitted flux density for images where there was no Selavy detection. Middle: VAST cutout for the epoch with the maximum flux density, the ellipse in the lower right corner of the radio image shows the FWHM of the restoring beam. Right: RGB image (details in text) of optical data from the Legacy Surveys/D. Lang (Perimeter Institute) with the VAST Stokes I contours at 50%, 70% and 90% of the peak Stokes I flux density overlayed.

Figure 16.

Same as Figure 15 for the possible radio star ASKAP J014700.3+075128 and the radio star ASKAP J053954.1-051119.

5.3. AGN

We identify the AGN in our sample by cross-matching to FIRST, SUMSS, NVSS, and the VAST pilot survey. When a source is detected in VAST Extragalactic DR1 and in one of these archival surveys, and there is no stellar or pulsar counterpart, we interpret the variable to most likely be an AGN. This is because a detection in any of the aforementioned archival surveys implies that the source is radio bright over at least 4 yr, a variability timescale that is only found for AGN and slow synchrotron transients.

Figure 17.

Same as Figure 15 for the supernovae discussed in Sections 5.4.1 and 5.4.2 and the supernova candidate discussed in Section 5.4.3.

Using these relatively simple criteria, we classify 44 of our variable sources as AGN. Table 2 lists per AGN candidate in which survey it was detected previously. Note that we list only the first survey we find the source in; we check against archival surveys in the order we list them above. We cross-match sources against the transient name server, and visually inspect each light curve to check for a synchrotron-transient-type evolution. Our final set of AGN light curves all show variability consistent with strong QSO or blazar variability or scintillation.

Figure 15 shows two examples of sources classified as AGN. ASKAP J091741.5-031331 was detected in NVSS with a 1.4 GHz flux density of 23.6 mJy. Its optical counterpart is a known QSO (Gaia DR3 ID 3838208794064660352) at $z=1.428$ (Gaia QSOC; Delchambre et al. Reference Delchambre2023) and is highly variable in the optical. ASKAP J025503.3-561308 was detected in the VAST pilot survey. An optical/IR counterpart is visible in both the WISE W1 and W2 images (Cutri et al. Reference Cutri2013, Reference Cutri2021) and the Legacy Survey DR10 (Dey et al. Reference Dey2019), with a photometric redshift estimate of $z=0.835\pm0.051$ (Zhou et al. Reference Zhou2021). The optical and WISE colours suggest that this is a radio AGN hosted by a massive early-type galaxy. The radio flux density seems to be scintillating around the VAST detection threshold.

5.4. Other source classes

Six of the highly variable sources do not fall within one of the aforementioned categories. Two of these correspond to the radio counterparts of optically identified supernova, one is a supernova candidate, one is a brown dwarf, and there are two highly variable sources for which we have not been able to determine their origins. We discuss each of these sources briefly below, as most of them are part of (ongoing) follow-up campaigns, and will be described and modelled in detail in separate papers.

5.4.2. Supernova: ASKAP J044834.3+001448

ASKAP J044834.3+001448 is identified as the radio signature of optically identified supernova SN 2023mut (TNS: https://www.wis-tns.org/object/2023mut). Radio detections were reported previously from the uGMRT (Chandra et al. Reference Chandra, Chevalier, Ray, Nayana and Maeda2023) and AMI-LA (Sfaradi et al. Reference Sfaradi2023). Figure 17 shows the VAST Extragalactic DR1 light curve of SN 2023mut, where the duration of VAST Extragalactic DR1 seems to fully cover the radio outburst. SN 2023mut is a type IIb supernova at $z=0.0022$ . These results will be discussed in more detail as part of ongoing VAST projects, including multi-wavelength (radio) observations and detailed modelling.

Figure 18.

ASKAP J151721.1+052931 VAST Extragalactic DR1 light curve. Black points are integrated flux density measurements from Selavy. White points are the forced-fitted flux density for images where there was no Selavy detection. ASKAP J151721.1+052931 is cross-matched to T8 brown dwarf ULAS J151721.12+052929.0.

5.4.5. Unknown: ASKAP J160653.1-085406

Figure 20 shows the light curve and cutouts for ASKAP J160653.1-085406, a source showing a single burst. There are no archival ASKAP observations of ASKAP J160653.1-085406, nor are there any radio detections before the first VAST detection in archival radio surveys. There is no counterpart in Gaia, DES, WISE, 2MASS, SDSS, the Vista Hemisphere Survey (VHS) or the Skymapper Southern Survey. Figure 19 shows the dynamic spectrum for the single VAST detection. It is evident that the VAST image shows the averaged-out flux density of an extremely bright short-duration burst. The time resolution of the dynamic spectrum is 10 s, and the burst is contained well within one time bin, as there are no signs of leakage. The circularly and linearly polarised components are shown in the top panel in blue and red, respectively. The circular polarisation fraction is $|-30|/371\approx 8\%$ . ASKAP J160653.1–085406 was also independently detected as a single pulse with the Commensal Real-time ASKAP Fast Transients survey COherent upgrade backend (CRACO; Wang et al. Reference Wang2025) at the same time. The identification and follow-up observations of this source will be presented in Zhang, Wang, and Gupta et al. (in preparation).

Figure 19.

Stokes I dynamic spectrum of the single VAST detection of ASKAP J160653.1-085406 (see Figure 20). The light curves in the top panel show the total (I, black), circular (V, blue), and linear (L, red) intensity, respectively. The linear polarisation is defined as $L=\sqrt{Q^2+U^2}$ .

5.4.6. Unknown: ASKAP J230616.0-395740

Figure 20 shows the light curve and cutouts for ASKAP J230616.0-395740. This source is detected in two VAST Extragalactic DR1 epochs. There are no previous radio detections with ASKAP or any other public radio survey. There is no counterpart to ASKAP J230616.0-395740 in Gaia, DES, WISE, 2MASS, SDSS, the Vista Hemisphere Survey (VHS) or the Skymapper Southern Survey. The closest catalogued counterpart was found in DES, at $7.2{''}$ (Abbott et al. Reference Abbott2018). The star that is visible in Figure 20 is $15.8{''}$ offset from ASKAP J230616.0-395740. This difference cannot be accounted for by the proper motion of the star (Gaia DR3 ID 6543420093211391104) of 18.9 mas/yr. There are no known pulsars within 1 degree. We used the ESO archive to extract deep infrared upper limits from VHS (McMahon et al. Reference McMahon2013). J-band (1 166–1 339 nm) upper limits from VHS with a sensitivity of mag 21.3, and Ks-band (1 992–2 301 nm) to mag 20.3 (AB magnitudes), and these observations were taken on 2013-07-03.

Figure 20.

Same as Figure 15 for the sources with unknown progenitor discussed in Sections 5.4.5 and 5.4.6.

We evaluate whether the radio detections of ASKAP J230616.0-395740 could be explained by refractive scintillation of a background AGN that typically falls below the ASKAP VAST detection limit. To this end, we analyse archival ASKAP observations of ASKAP J230616.0-395740. There are 26 archival ASKAP observations that cover ASKAP J230616.0-395740 between April 2019 and June 2023 as part of VAST pilot surveys and RACS-low, -mid, and -high (McConnell et al. Reference McConnell2020; Duchesne et al. Reference Duchesne2023, Reference Duchesne2025). The source is not detected during any of these observations. Additionally, the amplitude of scintillation-induced flux density modulation would be extreme, going from non-detections ( $\sim 5\cdot 0.24 =$ ) $1.2$ mJy to 4.05 mJy, much more than the typical 20–30% modulations expected from refractive scintillation (Walker Reference Walker1998). Alternatively, the radio light curve of ASKAP J230616.0-395740 could be explained by extreme scintillation due to local Galactic screens (e.g. Bignall et al. Reference Bignall2007; de Bruyn & Macquart Reference de Bruyn and Macquart2015). These intra-day/intra-hour variables can have modulation indices up to 50%. The dynamic spectra for the two epochs in which the source is detected show no evidence of circular or linear polarisation, and the Stokes I flux density shows no bursty behaviour. At the moment it is unclear what type of source produces the radio emission in ASKAP J230616.0-395740. Further monitoring with VAST might shed light on its origin.

5.5. Solar system planets

Throughout VAST Extragalactic DR1, we detect Mercury four times, and Saturn five times (see Table 2). Solar System planets are detected routinely by ASKAP.Footnote ⁱ The planets in our Solar System show radio emission due to various processes. At frequencies below 100 MHz, electron cyclotron-maser emission is the dominant mechanism (Callingham et al. Reference Callingham2024). At higher observing frequencies the emission is a combination of thermal emission (Roberts Reference Roberts1963), from the heat absorption by the planet atmosphere, and non-thermal emission for the planets with a significant magnetic field. Saturn has a strong magnetic field, and non-thermal emission is generated as highly energetic electrons become trapped in the planetary magnetic field and spiral along its field lines (see e.g. Hao et al. Reference Hao2020). For Mercury, which does not have a strong magnetic field, the radio emission is dominated by thermal blackbody re-radiation from the solar insolation over a diurnal cycle (Burns et al. Reference Burns, Gisler, Borovsky, Baker and Zeilik1987; Ledlow et al. Reference Ledlow1992). As mentioned before, we do not consider the planet detections in Figure 14, as the main reason for variability comes from the fact that the Solar System planet positions are not fixed in an equatorial coordinate system.