2.1 Calibration and flagging

Figure 1 shows example primary beam attenuation patterns (with main, grating, and side lobes) for a zenith scan [pointed towards declination $-26.7^{\circ}$ , Figure 1(a)] and a low-elevation scan [pointed towards declination $+18.6^{\circ}$ , Figure 1(b)] with the pointing directions indicated by green stars. The main lobe is closest to the pointing direction, but at this frequency and particularly for low elevations, the grating lobes approach the sensitivity of the main lobe. Cook et al. (Reference Cook, Seymour and Sokolowski2021) used observations of bright ‘calibrator’ sources to ensure bandpass and direction-independent complex gains can be solved when the primary beam grating lobes contribute significant amounts of power. We found that following this approach presented two issues: (1) we were unable to achieve good calibration solutions for some dedicated calibrator scans for unknown reasons, and (2) the gains drift over the course of the night and good calibration solutions become less effective as the night progresses. The alternate approach being used in other MWA processing strategies is to perform in-field calibration for every snapshot observation (e.g. Franzen et al. Reference Franzen, Hurley-Walker, White, Hancock and Seymour2021, but see also Duchesne et al. Reference Duchesne, Johnston-Hollitt, Zhu, Wayth and Line2020; Lynch et al. Reference Lynch2021; Hurley-Walker et al. Reference Hurley-Walker2022a). For these 300-MHz observations, the in-field approach suffers a higher failure rate than the dedicated calibrator scans, reducing to only 26% of observations arriving at good bandpass and gain solutions. In the context of these 300 MHz observations, ‘in-field’ refers to the main lobe as well as grating lobes above an attenuation of 20%. A local sky model is constructed for each observation from a subset of the GLEAM global sky model (Hurley-Walker et al. Reference Hurley-Walker2022a).Footnote ⁱ We take the 1 000 sources with the highest attenuated brightness using their frequency-dependent models. We note that while the sky model is largely based on GLEAM, the sky above $\delta_{\text{J2000}} \gtrsim +30^{\circ}$ is filled in with sources from the NRAOFootnote ^j VLA Sky Survey catalogue (NVSS; Condon et al. Reference Condon, Cotton, Greisen, Yin, Perley, Taylor and Broderick1998) at 1 400 MHz. These northern sources have a two-point spectral index that is derived after cross-matching to the VLA Low-frequency Sky Survey redux catalogue (VLSSr; Lane et al. Reference Lane, Cotton, van Velzen, Clarke, Kassim, Helmboldt, Lazio and Cohen2014)Footnote ^k at 74 MHz.

Figure 1. Example primary beam response for a zenith pointing [declination $-26.7^{\circ}$ , (a)] and a low-elevation pointing [declination $+18.6^{\circ}$ , (b)]. The attenuation is displayed with square-root stretch, and the white contours trace $[0.2, 0.5, 0.9]$ . The green stars indicate the main lobe, in the pointing direction, and significant grating lobes are those above the 0.2 contour.

Figure 2. Flow diagram of the calibration procedure. The diagram shows the processing steps to assign good calibration solutions to each ObsID after initial pre-processing, and prior to peeling/outlier source subtraction, and imaging.

Our calibration process makes use of a mixture of in-field solutions and dedicated calibrator solutions, selecting solutions that reduce the image rms noise. This process is outlined in a flow diagram in Figure 2 and summarised as follows. We initially obtain calibration solutions from each nightly calibrator scan, using the same in-field calibration approach on the calibrator observation. We then apply the solutions to the survey ObsIDs, perform additional flagging, then create a shallow quick-look image.

Then, on uncalibrated data we perform in-field calibration with all ObsIDs. We then consider in-field solutions ‘good’ if they satisfy

1. 1. $\sigma_{\text{amplitude}} \lt 40$ Jy,

2. 2. $\sigma_{\text{phase}} \lt 2^{\circ}$ ,

3. 3. $\lt35$ % solutions flagged from data that are not initially flagged,

where $\sigma_{\text{amplitude}}$ and $\sigma_{\text{phase}}$ are the standard deviations of the amplitude and phase of the complex gains, checking both the XX and YY polarisations separately. We ignore XY and YX instrumental polarisations as we do not perform polarisation calibration and assume all calibrators are unpolarised. Solutions identified as good are applied directly to their host snapshots, and for ObsIDs with ‘bad’ solutions the nearest-in-time good solution is applied with no interpolation. Another set of shallow quick-look images is then made after further flagging.

The two quick-look images are compared for each ObsID, and we record the set of solutions that produced the lowest rms noise in the image and use those solutions to calibrate the data. In Table 1 we note the fraction of ObsIDs for each night that use either dedicated calibrator solutions (‘C’), the in-field solutions derived from the ObsID itself (‘S’), or the nearest-in-time in-field solutions (‘N’).

At multiple stages in the calibration process, prior to imaging, we perform a significant amount of RFI flagging. Alongside AOFlagger, we also follow Cook et al. (Reference Cook, Seymour and Sokolowski2021) and make heavy use of the automated flagging tasks rflag and tfcrop within the Common Astronomy Software Applications (CASA; CASA Team et al. Reference Team2022)Footnote ^l package. This is done before and after calibration. We also use the gain solutions to identify poorly-performing MWA tiles that are typically flagged for a given observing night. While the total percentage of flagged data varies between observations, we see approximately 60% of data flagged, which is a reasonable increase from the 20–30% normally seen at lower frequencies from other GLEAM observations processed with the same pipeline.

2.2 Imaging

2.3 Patching gaps with grating lobes

The lowest declination strip is at $-72^{\circ}$ and the mainlobe full-width at half-maximum (FWHM) is only $\approx 15^{\circ}$ so does not cover the South Celestial Pole (SCP). While there are no pointings towards the SCP, most of the observations in the declination $-26.7^{\circ}$ and $-19.9^{\circ}$ strips have significant primary beam grating lobes that provide coverage below $\delta_{\text{J2000}} \lt -72^{\circ}$ , including the SCP, with maximum sensitivity at $\delta_{\text{J2000}} \approx -86^{\circ}$ and $\approx -76^{\circ}$ , respectively. Figure 0(a) shows an example pointing with a grating lobe with a peak in sensitivity at $\delta_{\text{J2000}} \approx -86^{\circ}$ , covering the SCP. To ensure complete coverage of the Southern Sky we image these lobes for all the relevant observations, using the same calibration solutions for the given snapshot observation. We also repeat the peeling process, assuming the grating lobe is the pointing centre (as opposed to the main lobe) and continue with imaging and post-processing in an otherwise identical fashion.

In addition to the grating lobes covering the SCP region, we also use the equivalent grating lobes at $\delta_{\text{J2000}} \approx +32^{\circ}$ and $\approx +23^{\circ}$ from the declination $-26.7^{\circ}$ and $-33.5^{\circ}$ strips to provide survey coverage up to $\approx +40^{\circ}$ . We refer to these additional data as separate declination strips in Table 1.

2.4 Post-imaging corrections

2.4.2 The point-spread function

After imaging and source-finding with aegean, we inspect the ratio of integrated and peak flux densities ( $S_{\text{int}}/S_{\text{peak}}$ ), comparing the median for each snapshot and each declination strip to check that the point-spread function (PSF) is appropriately estimated for each snapshot. To avoid extended sources, we construct a reference point source catalogue by combining the Sydney University Molonglo Sky Survey (SUMSS; Bock, Large, * Sadler Reference Bock, Large and Sadler1999; Mauch et al. Reference Mauch, Murphy, Buttery, Curran, Hunstead, Piestrzynski, Robertson and Sadler2003, at 843 MHz, $\approx 45\times 45$ arcsec $^{2}$ angular resolution), Molonglo Galactic Plane Survey (MGPS-2; Murphy et al. Reference Murphy, Mauch, Green, Hunstead, Piestrzynska, Kels and Sztajer2007), and the Rapid ASKAPFootnote ^s Continuum Survey (RACS; McConnell et al. Reference McConnell2020) 888-MHz catalogue (RACS-low; Hale et al. Reference Hale2021, at $25 \times 25$ arcsec $^{2}$ angular resolution). We use SUMSS/MGPS-2 for the sky south of $\delta_{\text{J2000}}\lt-80^{\circ}$ and RACS-low elsewhere. For RACS-low sources, we use the unresolved definition from Hale et al. (Reference Hale2021, see their section 5.2.1), and for SUMSS/MGPS-2 we consider a source unresolved if $S_{\text{int}}/S_{\text{peak}} \lt 1.2$ . We also remove sources in the SUMSS/MGPS-2 and RACS-low catalogues that have neighbours within 4 and 3 arcmin, respectively. We cross-match this unresolved and isolated RACS-SUMSS/MGPS-2 catalogue to the GLEAM-300 snapshot source list to identify point sources, but we do not enforce any similar selection criteria on the GLEAM-300 data.

Figure 3(a) shows the declination $-55.0^{\circ}$ signal-to-noise (SNR) weighted mean $S_{\text{int}}/S_{\text{peak}}$ ratios for cross-matched unresolved sources as a function of ObsID. In general, this ratio should tend towards $\gt1$ ( $=1$ for an ideal point source), where residual extended sources and ionospheric blurring push the ratio above 1. With median ratios $\lt1$ there are likely calibration, deconvolution, and/or a mismatched PSF issues. We find that for most snapshots this ratio is less than 1. The declination $-55.0^{\circ}$ strip has a median across all sources/snapshots of $0.91_{-0.06}^{+0.05}$ , which reduces to $0.88_{-0.05}^{+0.03}$ for sources with peak flux densities above $100\sigma_{\text{rms}}$ . The median across all declination strips is $0.96_{-0.06}^{+0.07}$ (reducing to $0.90_{-0.05}^{+0.08}$ for $100\sigma_{\text{rms}}$ sources). For comparison, the equivalent GLEAM cross-match to the unresolved RACS-SUMSS/MGPS-2 catalogue has a median of $0.99_{-0.03}^{+0.02}$ at 200 MHz (with $0.96_{-0.06}^{+0.11}$ above $100\sigma_{\text{rms}}$ ). The median ratios $\lt1$ are likely related to calibration errors in this case due to mismatch between sky model and true sky (i.e. point sources being treated as extended or vice versa).

Figure 3. Basic image properties per snapshot for declination strip $-55.0^{\circ}$ . (i)–(ii) SNR-weighted mean $S_{\text{int}}/S_{\text{peak}}$ per snapshot after applying $f_{\text{PSF}}$ . The dashed horizontal line is drawn at 1, and the dotted horizontal lines are drawn at a ratio difference of 20%. (iii)–(iv) SNR-weighted mean flux density ratios, $S_{\text{300 MHz}} / S_{\text{model}}$ , comparing to the GGSM measurement, before and after applying $f_{\text{PSF}}$ and $f_{\text{scale}}$ . The horizontal lines are as in (i). (v) median rms noise ( $\sigma_{\text{rms}}$ ) per snapshot after applying the brightness scale correction factor. Dashed horizontal lines indicate 20, 50, and 100 mJy beam $^{-1}$ . In (i)-(iv) the grey shaded region is drawn between $\pm1\sigma$ and in (v) between the 16-th and 84-th percentiles. Snapshots are coloured by their calibration solutions. Black circles: solutions derived from a dedicated calibrator scan; red squares: nearest-in-time best solutions from other observations; blue diamonds: solutions derived from the observation itself. Note for this declination strip, only a small number of snapshots used the dedicated calibrator solutions in the first night of observing. Note the y-axis in all panels is logarithmically scaled.

We attempt to correct for this by scaling the PSF (as the image restoring beam) which will leave the surface brightness in the images untouched, but allow for more accurate integrated flux density measurements. To avoid over-correcting individual snapshots, particularly those with a small number of sources, we opt to fit generic polynomial models to contiguous groups of ObsIDs for each night and for each declination strip separately. The groups are defined by ObsIDs that are not separated by more than 60 mins, and where the logarithmic ratios do not change by more than 0.05 to avoid jumps that require high-order polynomial models. This effectively creates a piece-wise polynomial model over the night. For fitting, we also use a sliding window filter across 10 ObsIDs and remove those with ratios greater than three times the standard deviation of the window. We also remove snapshots from fitting if they had less than 100 sources in the original source lists prior to cross-matching. For each group with a number of snapshots $N_{\text{obs}}\geq10$ , we trial polynomial models up to degree $\lceil{N_{\text{obs}} / 3}\rceil$ , using the Bayesian Information Criterion (Schwarz Reference Schwarz1978) to select the model from the range of fitted models. Snapshots that do not lie within a fitted group will take the median value for the night, or a value of 1 if no groups were fit for that particular night, and those within the group range will take a fitted value from the selected model. These PSF correction factors, $f_{\text{PSF}}$ , are applied to the FWHM of both the PSF major and minor axes as $\theta_{\text{corrected}} = \theta \sqrt{f_{\text{PSF}}}$ . Figure 4 shows two example nights of $S_{\text{int}}/S_{\text{peak}}$ highlighting the results of the piece-wise polynomial fitting.

Due to the lack of coverage of the RACS-SUMSS/MGPS-2 catalogue, we also cross-match the $\delta_{\text{J2000}} = +32.0^{\circ}$ grating lobe strip to the NVSS and check the results against the $\delta_{\text{J2000}} = -26.7^{\circ}$ mainlobe strip. We find similar ratios of $S_{\text{int}}/S_{\text{peak}}$ within the standard deviation for the ObsIDs, but with more significant scatter. For this reason, we assume the $\delta_{\text{J2000}} = +32.0^{\circ}$ grating lobe declination strip mirrors the main lobe in terms of the resultant PSF correction and use the corrections derived from the mainlobe images. After application, the median ratio for unresolved sources across all snapshots is $1.02_{-0.03}^{+0.04}$ ( $0.98_{-0.02}^{+0.02}$ for $100\sigma_{\text{rms}}$ sources), bringing the 300 MHz data closer to the expected ratio of 1.

2.4.3 Snapshot brightness scale

After application of the primary beam model, we also compare the measured flux densities of sources in the 300 MHz images compared to the sky model to check the overall brightness scale of the snapshot images. To compare, we cross-match sources detected above $10\sigma_{\text{rms}}$ and with $S_{\text{int}}/S_{\text{peak}} \lt 1.2$ in the 300-MHz snapshots to the sky model and scale the sky model flux densities to 300 MHz using the reported spectral indices. This is done after application of the PSF correction factors. We noticed residual variation to the brightness scale as a function of ObsID. Figure 3(c) shows the SNR-weighted mean measured to model flux density ratio ( $S_{\text{300 MHz}} / S_{\text{model}}$ ) per ObsID as a function of time for the declination $-55.0^{\circ}$ strip. We also see variation as a function of time/ObsID for all declination strips, with some link to specific calibration solutions (i.e. the ratio between observations that share a set of solutions shows less variation generally). While not shown in the SNR-weighted mean values, we also see no significant variation in brightness scale as function of position across the images, but the residual variation as a function of ObsID is likely related to variation in the local sky model for each ObsID and how nearest-in-time ‘good’ solutions are applied to observations.

Figure 4. SNR-weighted mean $S_{\text{int}}/S_{\text{peak}}$ of sources cross-matched to the ‘unresolved’ RACS-SUMSS/MGPS-2 catalogue as a function of time for two example nights (top panels) and the residual ratio (bottom panels). The black points indicate SNR-weighted mean values for a given ObsID, and the red solid lines indicate fitted polynomial models to each group as described in the text and the blue dashed lines with blue markers indicate a fixed value was assigned (only relevant for the bottom panel in this case). The grey shaded regions show $\pm 1\sigma$ for each snapshot. The residuals after applying the model values are shown in the bottom panels.

To correct this residual brightness scale variation, we perform a similar polynomial fitting procedure for the ratio $S_{\text{300 MHz}} / S_{\text{model}}$ , using the SNR-weighted mean ratio per snapshot, with model-fitting weights determined by the sum of weights from flux density measurement uncertainties at 300 MHz. We restrict the ratios to 0.1–10, with a slightly larger logarithmic jump of 0.1 to determine where fitting group boundaries are defined. After obtaining the models, we apply the brightness scale correction factor, $f_{\text{scale}}$ , to snapshots by evaluating the models for snapshots within a group that was fit. For snapshots that were not part of a fitted group, we take the measured SNR-weighted mean ratio as $f_{\text{scale}}$ if over 25 sources were in the cross-match, otherwise snapshots not in a group and with few cross-matches are discarded. Figure 5 shows the SNR-weighted mean flux density ratios along with the fitted models for two example nights from the declination $-72.0^{\circ}$ strip (top panel) the declination $-19.9^{\circ}$ strip (bottom panel), highlighting the general shape and variance of the polynomial models.

Figure 5. Models of the per-ObsID brightness scale correction factors for two example nights (top panels) and the residual ratio (bottom panels). The black points indicate weighted mean values for a given ObsID, and the red solid lines indicate fitted polynomial models to each group as described in the text. The shaded, grey regions show the weighted standard deviation. The blue dashed lines with blue markers indicate a fixed value was assigned based on the weighted mean of the individual ObsID.

The GLEAM flux density scale uncertainty is reported to be 80% for $\delta_{\text{J2000}} \lt -83.5^{\circ}$ and our sky model makes use of NVSS-VLSSr sources above $\delta_{\text{J2000}} \gtrsim +30^{\circ}$ so it is important to check the validity of the brightness scaling process for the grating lobe images in those regions. We compare the flux density ratios calculated for the $\delta_{\text{J2000}} -86.0^{\circ}$ and $+32.0^{\circ}$ images, which are derived from the same observations and calibration solutions.

Figure 6 shows the flux density ratios of the ObsIDs with the sky model in the far grating lobe images (top panel $\delta_{\text{J2000}} = +32.0^{\circ}$ and bottom panel $\delta_{\text{J2000}} = -86.0^{\circ}$ ) along with the difference in the ratios for each ObsID (middle panel). The general form of the flux density ratios is similar between the two declination strips except around complex regions (e.g. ObsIDs where the main lobe covers the Galactic Centre), but with some residual small-scale features. The standard deviation of the difference between the ObsIDs is $\approx 16\%$ , which we take as an intrinsic brightness scale error for these regions (see Section 4.6.2 for details on the final brightness scale uncertainty). This reduces to $\approx 6$ % after application of $f_{\text{PSF}}$ and $f_{\text{scale}}$ . We note that comparisons of the far grating lobe data with the main lobe data for those ObsIDs returns almost identical standard deviations, though with an additional time-independent offset likely related to the primary beam models which is implicitly corrected for in this post-image scaling.

Figure 6. SNR-weighted mean flux density ratios, $S_{\text{300 MHz}} / S_{\text{model}}$ , comparing to the extrapolated sky model measurement for grating lobe images that make up the $\delta_{\text{J2000}} = +32.0^{\circ}$ (top panel) and $-86.0^{\circ}$ (bottom panel) declination strips. The middle panel shows the difference in the SNR-weighted mean ratios between the two strips for each ObsID. The shaded region in the middle panel is drawn at $\pm 1\sigma$ (16%), and the dashed and dotted lines in each panel are drawn at a ratio of 1 and $\pm 20\%$ , respectively. The points are coloured by their calibration solutions type, as in Figure 3. The y-axis scale is logarithmic, though has a reduced range compared to the flux density ratios on Figure 3.

3.1 Mosaics

After post-imaging corrections, snapshots are combined via linear mosaicking to form full-sensitivity images at 300 MHz. Because of the smaller FoV at 300 MHz, observations typically have less overlap compared to the GLEAM/GLEAM-X observations in the lower part of the MWA band even with the additional declination strips. A smaller FoV allows us to mosaic smaller regions than the full declination strip mosaics used for GLEAM, though there is still some redundancy in the mosaic processing. Therefore, we follow a similar mosaicking process used for the RACS-low (Hale et al. Reference Hale2021) to create full-sensitivity images in regions covering

(1) \begin{equation} f_{\text{pad}}[15^{\circ} \times {9}^{\circ}] \, .\end{equation}

The regions are placed with equal spacing between $-72^{\circ} \leq \delta_{\text{J2000}} \leq +18^{\circ}$ , with dimensions in $\alpha_{\text{J2000}}$ scaled by $\cos \delta_{\text{J2000}}$ . These regions are shown in blue on Figure 7. We include a padding factor, $f_{\text{pad}} = {1.2}$ , in the region size to ensure overlap between the regions so that extended sources are less likely to be cut off during source-finding.

Figure 7. Mosaic and source-finding regions in orthographic projection. The grating lobe regions are shown in red (SCP cap region and high-declination regions). Only regions on the front side of the sphere are shown.

Figure 8. Example mosaic covering $21.0^{\circ} \times 11.7^{\circ}$ containing the radio galaxy Fornax A. The image represents the region of sky close to lowest rms noise ( $5.6_{-0.6}^{+0.8}$ mJy beam $^{-1}$ ) in the survey, partly due to the strong and well-modelled in-field calibrator.

We include additional regions covering the SCP and high-declination grating lobe images (highlighted red in Figure 7). The SCP mosaic is $30 \times 30$ deg $^2$ and the large high-declination mosaic regions are centred on $\delta_{\text{J2000}} = +32.0^{\circ}$ , with dimensions of $26.3 \times 19.2$ deg $^2$ , following a similar padding and spacing as the main lobe regions.

There are a total of 217 mosaic regions, covering $-90^{\circ} \leq \delta_{\text{J2000}} \lesssim +40^{\circ}$ . These regions are formed from the linear mosaic of all snapshots with image centres within 5 degrees of region boundary or with centres within the region boundary. Due to the generally lower quality of the grating lobe images, we only use the grating lobe images in the specific grating lobe regions. This results in 87–461 snapshots per non-SCP region and 2 046 for the SCP region (which draws also from the $\delta_{\text{J2000}} = -72.0^{\circ}$ snapshots). We use a position-dependent rms map as weights to stack the snapshots via a weighted average. The rms map is generated by BANE, where we increase the grid and box size (100 and 500 pixels) used for calculating the rms to ensure a smoothly varying rms estimate that largely follows the primary beam sensitivity.

Prior to forming the individual mosaics, we also convolve all snapshots within a mosaic group to the lowest common angular resolution of that particular group. We make use of the convolution tools from RACS-tools Footnote ^t which makes use of the common_beam function from the radio-beam Python packageFootnote ^u to find the smallest PSF that the group can be deconvolved by, ensuring all snapshots can be convolved to a matching resolution. To ensure that the resolution does not become too degraded, we remove snapshots that have a PSF major axis ( $\theta_{\text{major}}$ ) that is $\gt \overline{\theta_{\text{major}}} + 2\sigma_{\theta_{\text{major}}}$ where $\overline{\theta_{\text{major}}}$ is the mean PSF major axis of the snapshot group and $\sigma_{\theta_{\text{major}}}$ is the standard deviation. For the SCP, we restrict this further to a $0.5\sigma_{\theta_{\text{major}}}$ variation to avoid an overly-large PSF in some of the SCP images and because there is a significant amount of overlap in images for this region. In addition to removal of snapshots with large PSFs, we also filter snapshots based on the median root-mean-square noise and we remove 99 snapshots where the Moon was present within the image boundaries.

Figure 8 shows the example mosaic containing the bright radio galaxy Fornax A. That particular mosaic presents a ‘best-case’ for imaging quality as Fornax A provides a strong model for in-field calibration and most snapshot images in this region arrive at good solutions. The median rms noise for the mosaic in Figure 8 is $5.6_{-0.6}^{+0.8}$ mJy beam $^{-1}$ . Figure 9 highlights the imaging quality in the Galactic Plane, showing a mosaic covering the plane and Galactic longitudes of $285^{\circ} \lesssim l \lesssim 308^{\circ}$ . The median noise in this region is higher than the Fornax A example, at $14.1_{-3.1}^{+10.9}$ mJy beam $^{-1}$ , owing to the number of complex and bright extended sources that are not as well-modelled or well-calibrated as Fornax A.

Figure 9. Example mosaic covering $24.5^{\circ} \times 11.7^{\circ}$ covering Galactic longitudes $285^{\circ} \lesssim l \lesssim 308^{\circ}$ , showing the background ripples due to the extended sources. The median rms noise of the mosaic is $14.1_{-3.1}^{+10.9}$ mJy beam $^{-1}$ .

A region with low rms noise at high-declination is shown in Figure 10 – a median rms noise of $16.5_{-2.7}^{+3.7}$ mJy beam $^{-1}$ is a best-case scenario for the most northern part of the survey. High-declination mosaics are constructed from low-elevation pointings and grating lobes which have lower sensitivity. Finally, Figure 11 shows the full mosaic covering the SCP. Due to the significant overlap in images here the median rms noise is $12.2_{-1.9}^{+4.0}$ mJy beam $^{-1}$ . This region features noticeable ripples radiating from the SCP, likely residual RFI in the declination $-26.7^{\circ}$ scans. The same ripples can be seen in the mainlobe images, though at a much lower level. In addition, all of the grating lobe regions (SCP and high-declination) tend to feature more significant large-scale background undulations than elsewhere.

Figure 10. Example mosaic at high declination covering $26.3^{\circ} \times 19.2^{\circ}$ , representing the lowest noise ( $16.5_{-2.7}^{+3.7}$ mJy beam $^{-1}$ ) for the northern part of the survey.

Figure 11. The full SCP mosaic, with median noise $12.2_{-1.9}^{+4.0}$ mJy beam $^{-1}$ . The colourscale stretch is chosen to highlight the large-scale background features.

3.2 Source finding and source lists

Once mosaics are made, we use aegean to generate lists of 2-D Gaussian components. We use BANE again to create a position-dependent rms noise and background map for each mosaic, used by aegean for source-finding thresholds and measurements. For the mosaics, we use a $4\sigma_{\text{rms}}$ threshold for source detection (the ‘seedclip’ parameter in aegean), and $3\sigma_{\text{rms}}$ for growing source detections (‘floodclip’). This process is consistent with source-finding from other GLEAM data releases, focusing on compact sources. While our initial source-finding threshold is $4\sigma_{\text{rms}}$ we remove any components with a peak flux density $\lt5\sigma_{\text{rms}}$ from each source list. In addition, we remove sources with reported integrated flux density uncertainties greater than 100% of the measured value, with $\approx 1$ –20 sources removed from a median $\approx 2\,500$ sources per mosaic. For example, the mosaic region with Fornax A in Figure 8 contains 3 861 components above $5\sigma_{\text{rms}}$ .

3.3 Completeness

Due to difference in image properties, we investigate how complete the source lists for each mosaic region are expected to be prior to combining them into a single catalogue. We follow the process used for other GLEAM data releases (Hurley-Walker et al. Reference Hurley-Walker2017, but see also Franzen et al. Reference Franzen, Hurley-Walker, White, Hancock and Seymour2021; Hurley-Walker et al. Reference Hurley-Walker2022a; Ross et al. Reference Ross2024). We inject point sources onto our images using AeRes (Hancock et al. Reference Hancock, Murphy, Gaensler, Hopkins and Curran2012, Reference Hancock, Trott and Hurley-Walker2018) with pseudo-random positions (requiring no neighbours within 4 arcmin) and for multiple realisations at a range of logarithmically-spaced flux densities from 0.012–1 Jy. The number of sources varies per mosaic region, corresponding to $\approx 10$ sources per square degree matching the overall source density in the final catalogue (Section 4). The completeness is defined by the number of injected sources that are recovered at each injected flux density level after repeating the same source-finding process used for the normal images. We note that this method largely assesses the source-finding approach and does not take into account how other instrumental effects such as (u,v) coverage would affect recovery of sources.

The median flux density limit at 50% completeness is $45_{-16}^{+12}$ mJy (ranging from 21–141 mJy) and at 95% is $89_{-33}^{+89}$ (ranging from 45–891 mJy) across the mosaic regions. We show the 50% and 95% complete flux density limits as a function of position in Figure 12. The data are shown as a average for each tile, with nearest-neighbour interpolation to generate a contiguous map. Generally, completeness decreases in the Galactic Plane where large, extended Galactic radio emission is present, and around areas with significant artefacts following positional variation of the rms noise across the survey (see Section 4.3). Completeness metrics are included in the FITS headers for each mosaic image.

Figure 12. 50% (top) and 95% (bottom) flux density completeness limits across the survey. The limits are represented by an average over each mosaic region, with nearest-neighbour interpolation. Note each panel uses a different colourscale.

4.1 Merging source lists

The mosaic regions have a significant overlap by construction so any combined source catalogue needs to have duplicate measurements from sources detected across adjacent mosaics removed. We follow the method of Duchesne et al. (Reference Duchesne2024) and construct the full catalogue by concatenating the source lists one-by-one. As each source list is added, we cross-match the incoming source list to the partially constructed catalogue with a separation of $0.5 \times \langle\Theta_{\text{major}}\rangle$ , where $\Theta_{\text{major}}$ is the fitted major axis FWHM of the Gaussian component. Any sources cross matched within half their averaged size are considered duplicate measurements of the same source. For duplicated measurements, we opt to keep the measurement from the mosaic lowest rms noise at the location of the source. The resultant merged catalogue contains 338 080 components, with a median density of $\approx 10$ deg $^{-2}$ . Figure 13 shows the component density across the whole survey using HEALPixFootnote ^v binning, generally highlighting regions of lower/higher sensitivity (see Section 4.3).

4.2 Catalogue columns

The information included in the catalogue is similar to a single frequency from GLEAM-X, with component coordinates, measured sizes and flux densities, and additional metadata useful for further measurements or comparisons to other catalogues. The columns are summarised as:

Figure 14. HEALPix representation of the local rms noise at the position of sources in the constructed catalogue.

1. 1. name: Component name following IAU convention: GLEAM-300 JHHMMSS $\pm$ DDMMSS.

2. 2. RAJ2000: J2000 right ascension of the component in decimal degrees.

3. 3. err_RAJ2000: Uncertainty on right ascension from fitting the component position in decimal degrees.

4. 4. DEJ2000: J2000 declination of the component in decimal degrees.

5. 5. err_DEJ2000: Uncertainty in declination from fitting the component position in decimal degrees.

6. 6. local_rms: Estimate of the local root-mean-square noise in Jy beam $^{-1}$ .

7. 7. background: Estimate of the local background in Jy beam $^{-1}$ .

8. 8. peak_flux: Peak flux density of the component in Jy beam $^{-1}$ .

9. 9. err_peak_flux: Uncertainty in the peak flux density in Jy beam $^{-1}$ .

10. 10. int_flux: Integrated flux density of the component in Jy.

11. 11. err_int_flux: Uncertainty in the integrated flux density of the component in Jy.

12. 12. a: FWHM of the major axis of the component in arcsec.

13. 13. err_a: Uncertainty in the FWHM of the major axis in arcsec.

14. 14. b: FWHM of the minor axis of the component in arcsec.

15. 15. err_b: Uncertainty in the FWHM of the minor axis in arcsec.

16. 16. pa: Position angle of the component in degrees.

17. 17. err_pa: Uncertainty in the position angle in degrees.

18. 18. bmaj: Local major axis of the beam in arcsec.

19. 19. bmin: Local minor axis of the beam in arcsec.

20. 20. bpa: Local position angle of the beam in degrees.

21. 21. flags: Aegean fitting flags.

22. 22. flux_scale_err: Declination-dependent brightness scale fractional uncertainty (Section 4.6.2).

4.3 Noise

The rms noise in the final mosaic images varies across the survey, with a median value of ${9.1_{-2.8}^{+5.5}}$ mJy beam $^{-1}$ . Figure 14 shows the HEALPix-binned map of the rms noise as reported in the catalogue. The noise is raised around some bright sources and in regions of the Galactic Plane, with some additional increase in noise around $\alpha_{\text{J2000}} \approx 270^{\circ}$ and $\delta_{\text{J2000}} \approx -13^{\circ}$ to $\approx +1.6^{\circ}$ due to calibration challenges in this region. Some of the high-declination mosaics feature higher noise due to complex bright sources within the images (Cygnus A) and within the main lobe (the Galactic Centre) – these are peeled/removed when outside of the FoV, but significant artefacts still remain in some cases. Figure 9 also shows typical large-scale background features present near the extended sources in the Galactic Plane.

4.4 Image artefacts

In common with most all-sky surveys, the 300-MHz images feature noticeable artefacts at the ${\approx} 1$ % level, from both position-dependent and position-independent errors around bright sources. These artefacts can be mitigated with self-calibration and are reduced when using in-field calibration, although with our processing strategy self-calibration is not done due to a high failure rate and in-field calibration is only done for a subset of the snapshots. Common artefacts following the general shape of the PSF around bright sources are shown in Figure 15, highlighting the form of the errors for $\delta_{\text{J2000}} \gtrsim 0^{\circ}$ (Virgo A, left) and for $\delta_{\text{J2000}} \ll 0^{\circ}$ (PKS 1932 $-$ 46, right). Despite the lack of self-calibration, we are able to achieve a dynamic range of at least $\approx {90}$ with typical errors ${\approx} 1$ %.

4.5 Catalogue reliability

A common method of determining the number of false positive detections around bright sources (hence the reliability of the source finding) is to repeat source-finding on the same images multiplied by $-1$ (i.e. the ‘negative’ image, or ‘inverted’ image; e.g. Intema et al. Reference Intema, Jagannathan, Mooley and Frail2017; Hale et al. Reference Hale2021; Hurley-Walker et al. Reference Hurley-Walker2022a; Ross et al. Reference Ross2024). Assuming the noise and artefacts are Gaussian and symmetric, a measure of source-finding reliability, r, is then defined as

(2) \begin{equation}r = 100 \times \left(1 - N_{\text{negative}} / N_{\text{positive}}\right) \, \% ,\end{equation}

for $N_{\text{negative}}$ sources found in the negative image and $N_{\text{positive}}$ sources found in the original image.

To help reduce the number of artefacts that are included in the catalogue, we filter catalogued Gaussian components around bright ( $S_{\text{bright}} = S_{\text{300}} \gt 1$ Jy) sources. This process is modified from the artefact filtering done for GLEAM-X DR1/DR2 (Hurley-Walker et al. Reference Hurley-Walker2022a) with the inclusion of a simplified flux-dependent radial filter (see e.g. Knowles et al. Reference Knowles2022, for a similar method). We do this prior to merging source lists. We define a beam- and flux-dependent filter around sources that scales linearly between 5– ${20}\times\theta_{\text{major}}$ for bright sources between 1–5 Jy. Sources with flux densities $\gt5$ Jy use a fixed ${20}\times\theta_{\text{major}}$ radius. All catalogued components within the defined radius of a bright source are considered artefacts if they are sufficiently fainter than the bright source by $S_{\text{faint}}\times 100 \lt S_{\text{bright}}$ . In addition, if a ‘negative’ source lies within the filter radius, we reduce the filter radius to match the maximum angular separation between the bright source and any negative sources. We then consider any positive source within the same absolute flux limit as the negative source to also be an artefact. Negative sources satisfying these conditions are also removed from their respective source lists. An example 2.4-Jy source is shown in Figure 16, with sources within the ${12.1} \times \theta_{\text{major}}$ radius shown, with markers indicating whether they are artefacts or not. The positive artefacts either side of the source are common source sidelobe artefacts seen in that region.

Figure 15. Common artefacts around bright sources for declination $\gtrsim 0^{\circ}$ (Virgo A, left) and declination $\ll 0^{\circ}$ (PKS 1932 $-$ 46, right). The dynamic range (DR) is shown, estimated based on peak values of the sources and the most significant nearby artefacts.

Figure 16. Example of the artefact filtering process. The panel is centred on a 2.4 Jy component. Positive and negative components within the initial filter radius are shown, with markers indicating whether they are considered artefacts or not (see legend). The single dashed-grey contour is drawn at $-3\sigma_{\text{rms}}$ .

We perform the assessment of the reliability after removing artefacts on the final catalogue and negative catalogue. The overall reliability in the catalogue is 99.58%, and Figure 17 shows the reliability of all mosaics combined as a function of SNR ( $S_{\text{peak}}/\sigma_{\text{rms}}$ ). Reliability generally increases with increasing SNR, with $\approx 99$ % reliability achieved at $\approx 6.0\sigma_{\text{rms}}$ , and 100% reliability above $\approx 10\sigma_{\text{rms}}$ . We note Franzen et al. (Reference Franzen, Hurley-Walker, White, Hancock and Seymour2021) reports an overall higher reliability for the GLEAM SGP catalogue, likely a result of self-calibration reducing artefacts around bright sources. We note that the SCP region has comparatively higher overall reliability at 99.94%, while the high-declination regions more closely match the main survey at 99.71%.

Figure 17. Reliability as a function of SNR. The dashed horizontal line shows 100% reliability, and the bars on each point show the SNR bin width. Note the minimum SNR in source lists is confined to $5\sigma_{\text{rms}}$ .

4.6 Photometry

4.6.1 The point spread function

As the individual mosaics have different angular resolutions, the PSF for sources in the catalogue also varies across the sky. Figure 18 shows the position-dependent major and minor axes of the image PSF as recorded in the catalogue using HEALPix binning. The median major and minor axes are ${128^{\prime\prime}8} \times {112^{\prime\prime}5}$ , and range from 108^″8–275^″7 and 89^″4–222^″8, respectively. The PSF major and minor axes largely vary as a function of declination, with some additional variation around specific regions, relating to areas with either fewer snapshots, more RFI flagging, or poorer calibration. In particular, the high-declination and SCP regions feature the largest PSFs, corresponding to the lower elevation of the grating lobe directions.

Figure 18. HEALPix representation of the PSF major (top) and minor (bottom) axes across the sky as recorded at source positions in the catalogue.

Figure 19. The ratio of $S_{\text{int}}/S_{\text{peak}}$ as a function of SNR for all sources in the GLEAM-300 catalogue, represented in hexagonal bins. The horizontal line indicates a ratio of 1.

Figure 19 shows the ratio of $S_{\text{int}}/S_{\text{peak}}$ as a function of source SNR for all sources in the catalogue. The overall median ratio is ${0.99^{+0.19}_{-0.10}}$ , reducing slightly to ${0.98^{+0.06}_{-0.02}}$ for sources with $\text{SNR}\gt100$ , consistent with results from the individual snapshots prior to mosaicking (Section 2.4.2).

4.6.2 The brightness scale

The brightness scale is set by the input sky model we use for calibration and post-processing so we expect the final mosaics to be consistent with GLEAM and GLEAM-X by extension. To assess any residual brightness scale errors we compare the GLEAM-300 catalogue measurements to scaled flux densities from GLEAM, GLEAM-X DR2, and GLEAM SGP (scaled from 200 MHz measurements), the VCSS1 (scaled from 340 MHz), WISH (scaled from 352 MHz), and the TGSS ADR1 (scaled from 147.5 MHz). These surveys provide data products that are the reasonably close in sensitivity, resolution, and/or frequency to GLEAM-300. We note that GLEAM-X DR2 and GLEAM SGP inherits a flux density scale from GLEAM, and GLEAM overall is set to the Baars et al. (Reference Baars, Genzel, Pauliny-Toth and Witzel1977, hereinafter B77) flux density scale. WISH is flux-corrected to the NVSS, which is largely set to the B77 scale as well. For TGSS ADR1, the brightness scale is set to the low-frequency models from Scaife * Heald (Reference Scaife and Heald2012), tied to the (RCB; Roger et al. Reference Roger, Costain and Bridle1973) brightness scale. There is an expected $\approx 4$ % difference between B77 and RCB at these frequencies. VCSS1 is flux-calibrated using Perley * Butler (Reference Perley and Butler2017) models for 3C 286 and 3C 138, which again differ by a few percent from the B77 and RCB scales.

For consistency, we take the median $\alpha$ from all cross-matched GLEAM sources ( $-0.8$ ) to scale all surveys to 300 MHz. During cross-matching we only consider cross-matches within 30 arcsec and only take reasonably compact sources ( $S_{\text{int}}/S_{\text{peak}} \lt 1.2$ in both catalogues, where appropriate) and with no neighbours within 240 arcsec. An Eddington bias (Eddington Reference Eddington1913) correction is also applied to each catalogue, following equation 4 from Hogg * Turner (Reference Hogg and Turner1998). Figure 20 shows the scaled flux density ratio $(S_{\text{300}} / S_{\text{survey}}$ ) between the catalogues as a function of the signal-to-noise ratio in the GLEAM-300 catalogue. For comparisons with GLEAM, GLEAM-X DR2, GLEAM SGP, and TGSS ADR1 we see GLEAM-300 flux density measurements are largely in agreement, with average offsets within $\approx 5\%$ for sources above $100\sigma_{\text{rms}}$ . VCSS1 and WISH flux density comparisons show GLEAM-300 flux densities are $\approx 10\%$ lower, with some additional structure in their distributions.

Figure 20. Flux density ratios of sources cross-matched between GLEAM-300 and GLEAM (a), GLEAM-X DR2 (b), GLEAM SGP (c), VCSS1 (d), WISH (e), and TXS (f), after scaling flux densities to 300 MHz and correcting for Eddington bias, as a function of SNR in the GLEAM-300 catalogue. Solid horizontal line is drawn at 1, and the dashed lines are drawn at $\pm 20\%$ . Note the y-axis is scaled logarithmically.

Table 2. Brightness scale uncertainty as a function of declination.

To estimate the uncertainty in the brightness scale we also cross-match the catalogue to the sky model that was used for calibration and post-imaging brightness scale corrections. From the 84-th percentile of the $\gt100\sigma_{\text{rms}}$ cross-matched sources, we suggest an initial brightness scale uncertainty of 9% for declinations covered by the main lobe images ( $-78^{\circ} \leq \delta_{\text{J2000}} \leq +24^{\circ}$ ). For the grating lobe regions ( $\delta_{\text{J2000}} \lt -78^{\circ}$ and $\delta_{\text{J0000}}\gt+24^{\circ}$ ) we get $\approx 13$ % for the 84-th percentile, though we noted in Section 2.4.3 that the two regions had differences represented by a standard deviation of $\approx 16$ %. To be conservative, we suggest the 16% represents the uncertainty in the brightness scale calibration of the individual snapshots, so consider this the initial brightness scale uncertainty for these regions. These initial uncertainties should be added in quadrature to the GLEAM external brightness scale uncertainty for the relevant declination (see table 4 from Hurley-Walker et al. Reference Hurley-Walker2017), except we do not consider the 80% brightness scale uncertainty reported for $\delta_{\text{J2000}} \lt -83.5^{\circ}$ to be necessary here. This yields the full declination-dependent brightness scale uncertainty reported in Table 2 and is included in the catalogue for each source.

4.6.3 Spectral energy distributions

One of the goals of this work is to provide an additional ‘high’-frequency data point to GLEAM and/or GLEAM-X measurements to constrain the shapes of source spectral energy distributions (SEDs), particularly in the presence of spectral curvature. As a point of comparison we measure the integrated flux density of Fornax A within a polygon region containing the radio galaxy. We also make the equivalent measurements from the 16 GLEAM SGP mosaics between 107–227 MHz. We compare these GLEAM-300 and GLEAM SGP measurements to the VLA measurements and spectral model from Perley * Butler (Reference Perley and Butler2017). The spectral data are shown in Figure 21(a) with the Perley * Butler (Reference Perley and Butler2017) logarithmic polynomial model overlaid. The GLEAM-300 data agrees with the VLA measurements and model, though we see some discrepancy between the VLA data and GLEAM SGP data. The SGP data appear to show a steeper overall integrated spectrum for Fornax A. This is likely due to use of robust image weighting combined with a 30- $\lambda$ (u,v) cut and 15- $\lambda$ Tukey taper (corresponding to sensitivity up to $\approx 1.5$ deg), so artificial steepening of extended sources is expected, and care should be taken when interpreting spectra in cases of extended sources. Overall, the agreement with the Perley * Butler (Reference Perley and Butler2017) suggests the flux density scale (and associated uncertainty) is sensible for the GLEAM-300 data. We do not show a comparison to the original GLEAM as the Fornax A images at some frequencies contain artefacts from the linear mosaic process.

Figure 21. Example SEDs including a 300-MHz measurement. (a) shows the SED of Fornax A including the GLEAM SGP measurements and VLA measurements from Perley * Butler (Reference Perley and Butler2017). The fitted logarithmic polynomial model from Perley * Butler (Reference Perley and Butler2017) is also shown. (b)–(c) show unresolved point sources with standard power law spectra, (d)–(f) show curved power law spectra, and (g)–(i) comprises variable/flat spectrum sources, all selected after cross-matching the GLEAM-300 catalogue with GLEAM-X DR2 and the RACS catalogues. Power law (PL) and curved power law (CPL) models are fit for illustrative purposes. Note both the x- and y-axes are scaled logarithmically.

As an example of where the GLEAM-300 datapoint sits within the ecosystem of widefield Southern Sky surveys, we also cross-match GLEAM-300 to GLEAM-X DR2 and the three RACS bands: RACS-low at 887.5 MHz (McConnell et al. Reference McConnell2020; Hale et al. Reference Hale2021), RACS-mid at 1 367.5 MHz (Duchesne et al. Reference Duchesne2023, Reference Duchesne2024), and RACS-high at 1 655.5 MHz (Duchesne et al. Reference Duchesne2025). We repeat a similar strict multi-matching process employed by Duchesne et al. (Reference Duchesne2025), which focuses on isolated and compact point sources detected above $10\sigma_{\text{rms}}$ in all five catalogues under consideration. This choice is to avoid the aforementioned concern with extended source spectra and to avoid issues arising from differing angular resolution and sensitivities between the surveys. This limits the resultant cross-matched catalogue to 9 880 sources. We then fit generic power law and curved power law models as in for example, (Ross et al. Reference Ross2024). For the GLEAM-X DR2 data, we opt to use 30-MHz wideband measurements as opposed to 7.68-MHz narrowband measurements for consistency with the GLEAM-300 30-MHz measurement.

Figure 21(b)–(i) show a set of example sources with fitted SEDs that include the GLEAM-X, GLEAM-300, and RACS measurements. The sources are selected from the cross-matched sources to showcase a range of spectral morphologies. Figures 21(b)–(c) show a selection of sources with standard power law spectra, and Figures 21(d)–(f) show example sources with well-modelled curved power law spectra. In addition to the power law and curved spectra, we also show a few examples of more exotic spectra arising from source variability. Figure 21(g) shows the millisecond pulsar PSR J0437 $-$ 4715 (Johnston et al. Reference Johnston1993) which was also detected in the MWA circular polarisation survey (Lenc et al. Reference Lenc2017). Figure 21(h) shows a ‘flat-spectrum’ blazar source, 4C $+$ 06.69 (e.g. Healey et al. Reference Healey, Romani, Taylor, Sadler, Ricci, Murphy, Ulvestad and Winn2007), and Figure 21(i) shows NVSS J051042 $-$ 094813 (Condon et al. Reference Condon, Cotton, Greisen, Yin, Perley, Taylor and Broderick1998), which also has a non-power law spectrum, though we do not explore these sources further here. With the addition of the GLEAM-300 data point, the gap in spectral coverage between GLEAM and RACS (231–888 MHz) is now more accurately filled-in across the whole Southern Sky.

4.7 Astrometry

For assessment of the astrometry, we also cross match the GLEAM-300 catalogue to the SUMSS and the NVSS. We also compare to the other GLEAM catalogues, TGSS ADR1, and VCSS1 again. We consider isolated and compact sources (in each catalogue) and calculate the astrometric offsets as GLEAM-300 $-$ external survey for sources above $50\sigma_{\text{rms}}$ in the GLEAM-300 catalogue. Figure 22 shows the positional offsets in $\alpha_{\text{J2000}}$ and $\delta_{\text{J2000}}$ for the survey comparisons.

Figure 22. Right ascension ( $\alpha$ ) and declination ( $\delta$ ) offsets between GLEAM-300 and GLEAM (a), GLEAM-X DR2 (b), GLEAM SGP (c), TGSS ADR1 (d), VCSS1 (e), SUMSS (f), and the NVSS (g). Only sources with an SNR $\gt50$ in the GLEAM-300 catalogue are shown. The solid black lines are drawn at 0 offset, the dashed black lines are drawn at the mean offset value, and the dotted black lines are drawn at $\pm 1\sigma$ about the mean offset.

As GLEAM is largely used for phase calibration and post-imaging astrometric corrections, we see bulk $\alpha_{\text{J2000}},\delta_{\text{J2000}}$ offsets of $-0^{\prime\prime}06 \pm 1^{\prime\prime}40$ and $+0^{\prime\prime}52 \pm 1^{\prime\prime}52$ , noting that the smallest pixel size in the mosaic images is $17^{\prime\prime}9$ (with a maximum pixel size of 446). Other surveys tend towards a bulk offset in declination, up to a few arcsec, with median bulk offsets in $\delta_{\text{J2000}}$ of $+1^{\prime\prime}55 \pm 3^{\prime\prime}29$ for GLEAM-X DR2, $+0^{\prime\prime}60 \pm 3^{\prime\prime}52$ for VCSS1, $+1^{\prime\prime}99 \pm 3^{\prime\prime}30$ for the TGSS ADR1, $+2^{\prime\prime}19 \pm 3^{\prime\prime}07$ for SUMSS, and $+1^{\prime\prime}30 \pm 3^{\prime\prime}11$ for the NVSS. The GLEAM SGP on the other hand features the largest offset in $\alpha_{\text{J2000}}$ , with $\Delta\alpha\cos\delta = -1^{\prime\prime}39 \pm 2^{\prime\prime}32$ along with a marginal offset in $\delta_{\text{J2000}}$ . GLEAM-X DR2 uses both SUMSS and NVSS for its astrometric corrections so we expect the offsets against SUMSS and NVSS seen by Ross et al. (Reference Ross2024) should be present here too. Of particular note is elongation of the offset distributions towards the lower right of each plot, except in the SUMSS comparison. Figure 23 shows the median-binned declination offsets as a function of declination for the GLEAM, GLEAM-X DR2, and TGSS ADR1 cross-matches. The GLEAM cross-matches deviate in the SCP region and towards the Northern Hemisphere, whereas both the GLEAM-X DR2 and TGSS offsets show similar structure as a function of $\delta_{\text{J2000}}$ . The GLEAM-300 catalogue has largely inherited an inherent astrometric uncertainty from GLEAM.

5.1 Diffuse and extended radio emission

The GLEAM-300 catalogue provides crucial frequency coverage for understanding the spectral evolution of diffuse non-thermal phenomena in several environments. Both the so-called ‘remnants’ or ‘fossils’ associated with switched-off AGN (e.g. Murgia et al. Reference Murgia2011; Quici et al. Reference Quici, Turner, Seymour and Hurley-Walker2025) and the steep-spectrum radio sources associated with cosmic structures – groups and clusters of galaxies (see e.g. Eckert et al. Reference Eckert, Gaspari, Gastaldello, Le Brun and O’Sullivan2021; van Weeren et al. Reference van Weeren, de Gasperin, Akamatsu, Brüggen, Feretti, Kang, Stroe and Zandanel2019, respectively) – require broadband spectral information to understand the physics of the particle acceleration mechanisms at work.

Oftentimes these spectra show significant curvature, such as the remnant radio galaxy NGC 1534 (Duchesne * Johnston-Hollitt Reference Duchesne and Johnston-Hollitt2019), the ‘ultra-steep spectrum jellyfish’ associated with Abell 2877 (Hodgson et al. Reference Hodgson, Bartalucci, Johnston-Hollitt, McKinley, Vazza and Wittor2021), or the ultra-steep spectrum fossil in Abell 3266 (Duchesne et al. Reference Duchesne, Johnston-Hollitt, Riseley, Bartalucci and Keel2022; Riseley et al. Reference Riseley2022). The latter object in particular shows a clear break in the spectrum between 216 MHz (measured by the MWA) and 943 MHz (measured by ASKAP) and constraining the break frequency, as well as the type of break – whether the spectrum is truly curved or shows a sharp break in the injection index, for example – would provide insights into the ageing and evolution of these sources.

Similarly, in cosmic structures such as galaxy groups, bulk motions and interactions in the environment can re-distribute and re-accelerate fossil plasma onto different scales (e.g. Brienza et al. Reference Brienza2022; Candini et al. Reference Candini2023) but the mechanisms by which this occurs are still poorly understood. The GLEAM-300 measurements occupy a critical frequency regime that has been historically served only down to $\delta_{\text{J2000}} \approx -53^{\circ}$ (the southern limit of the GMRT); the catalogue published herein significantly broadens the scope of such studies.

Furthermore, the frequency coverage provided by the GLEAM-300 catalogue is essential in supernova remnants (see Dubner * Giacani Reference Dubner and Giacani2015, for a recent review) studies which exhibit a variety of spectral behaviours connected to their particle acceleration mechanisms, interactions with the surroundings and energy losses. The GLEAM-300 frequency band is beneficial for identifying turnovers in the spectra at low radio frequencies, which can signify absorption mechanisms such as free-free absorption in ionised gas (e.g. Castelletti et al. Reference Castelletti, Supan, Peters and Kassim2021), and in particular the 300-MHz data can provide an anchor for the unabsorbed power-law portion of the spectrum at a frequency where interstellar medium (ISM) absorption has not yet taken effect.

The GLEAM-300 data provide a unique opportunity to probe these effects by extending the spectral coverage into an under-explored range. This dataset effectively bridges the very low frequency of GLEAM (72–231 MHz) and the higher frequencies provided by surveys like the Evolutionary Map of the Universe (EMU; Norris et al. Reference Norris2021; Hopkins et al. Reference Hopkins2025) at 944 MHz while awaiting the new SKA infrastructure. Expanding studies of supernova remnants with this catalogue will help constrain the physical processes governing particle acceleration and evolution in these sources.

5.2 Transients and variability

Figure 23. Median-binned declination offsets as a function of declination for cross-matches to GLEAM, GLEAM-X DR2, and TGSS ADR1. The error bars are drawn at $\pm 1\sigma$ for each bin. Bins are offset by $2^{\circ}$ for each survey for clarity.

Figure 24. Jupiter detections in the mosaic images. The positions of Jupiter in the individual snapshots are indicated by green crosses, and sources detected at those locations are the weighted-average detection in the mosaic. In the left panel, both radio sources at the marked positions are Jupiter detections. In the right panel, only one source is a Jupiter detection.

Over the last few years, widefield imaging surveys have become an increasingly popular tool for finding sources that vary on minute- to hour-long time scales, a hitherto relatively unexplored parameter space (e.g. Hyman et al. Reference Hyman, Roy, Pal, Lazio, Ray, Kassim and Bhatnagar2007; Murphy et al. Reference Murphy2013, Reference Murphy2021; Hurley-Walker et al. Reference Hurley-Walker2022b). Such surveys can also potentially detect shorter duration (sub-second) events that are sufficient bright (e.g. Wang et al. Reference Wang2021; Sett et al. Reference Sett, Bhat, Sokolowski and Lenc2023; Mcsweeney et al. Reference Mcsweeney, Moseley, Hurley-Walker, Grover, Horváth, Galvin, Meyers and Tan2025). The sensitivity of a given survey towards different classes of transient sources is a non-trivial function of frequency. At lower radio frequencies ( $\lesssim 200$ MHz), detection can be inhibited by higher sky temperatures, intrinsic low-frequency turnovers, as well as the temporal smearing effects of interstellar dispersion and scattering. On the other hand, transient sources’ typically steep spectra ( $-4 \lesssim \alpha \lesssim -1$ ) make detection difficult at higher frequencies ( $\gtrsim 1$ GHz). This survey, at 300 MHz, may therefore be sensitive to transients that, for one reason or another, are difficult to detect at other frequencies. The individual snapshots have a median noise of $\sigma \approx {68}$ mJy beam $^{-1}$ ; we expect that they will be sensitive to transient events whose brightness (integrated over the two-minute snapshot) is $S_{\text{300}} \gt 5\sigma \approx {340}$ mJy beam $^{-1}$ . In particular, this survey is likely to detect new long period transients whose single pulses can last anywhere from a few tens of seconds (de Ruiter et al. Reference de Ruiter2025; Hurley-Walker et al. Reference Hurley-Walker2024) to several minutes (Hurley-Walker et al. Reference Hurley-Walker2023; Lee et al. Reference Lee2025). To enable searches of such transients, the individual source lists for each two-minute snapshot are being made available.

We note that Jupiter is detected in snapshots taken on 20–21 February 2016 and 25–26 April 2016 and appears in the mosaic images. The movement across sky over the two nights in February is seen as two separate sources, whereas only a single radio source is seen over the two nights in April. We show the mosaic detections in Figure 24, with green markers indicating the direction of Jupiter in the individual snapshots. Note that Jupiter is not included in the CLEAN mask so has noticeable PSF sidelobes around it. Three sources in the final catalogue are co-located with Jupiter’s positions: GLEAM-300 J112611 $+$ 051306, GLEAM-300 J112546 $+$ 051601, and GLEAM-300 J110033 $+$ 075214.

5.3 General data improvements and other additions

One of the limitations of our current GLEAM-300 processing strategy is the lack of consistent in-field calibration. Franzen et al. (Reference Franzen, Hurley-Walker, White, Hancock and Seymour2021) highlighted that self-calibration (after initial calibration from other methods) can be used in place of in-field calibration, providing similar results. We find through MWA processing experience that the level of improvement will also depend on the direction being observed and what is in the sky model. Either more consistent in-field calibration or a successful self-calibration procedure would improve the overall sensitivity of the GLEAM-300 survey and reliability at low SNR. At present it is not clear why these processes do not work consistently, though it is likely the combination of inaccurate source models, certain primary beam pointings being more difficult due to attenuated bright sources at the edges of the primary beam lobes, as well as other time/observation-dependent effects such as ionospheric activity.

An interesting avenue for further work in this area will be to develop a linear polarisation counterpart to GLEAM-300, akin to the POGS effort at lower MWA frequencies (Riseley et al. Reference Riseley2018, Reference Riseley2020). A detailed understanding of the magnetoionic properties and environments of radio sources requires broadband linear polarisation measurements, and the primary figure of merit is complete sampling of a broad range of wavelength-squared ( $\lambda^2$ ) coverage. GLEAM-300 now provides well-understood visibility data from which the required data products – full Stokes continuum cubes – can be produced and analysed. This is however a non-trivial exercise because additional calibration effects require attention, such as XY phase differences, ionospheric Faraday rotation measure corrections, and characterisation and correction of polarisation leakage. But the gains can be significant, as a polarisation counterpart of GLEAM-300 would help to fill a key gap in $\lambda^2$ coverage between POGS and the ASKAP SPICE-RACS catalogue (Thomson et al. Reference Thomson2023) at considerably higher frequency. These intermediate- $\lambda^2$ measurements are particularly valuable to understand the details of depolarisation effects (see, e.g. Farnsworth, Rudnick, * Brown Reference Farnsworth, Rudnick and Brown2011) that dramatically reduce the density of linearly polarised sources seen at MWA frequencies compared to GHz-regime observations like those with ASKAP (e.g. O’Sullivan et al. Reference O’Sullivan2023; Piras et al. Reference Piras2024).

Whilst the focus of this work is on producing a compact source catalogue at 300 MHz, the short baselines of the MWA provide excellent sensitivity to extended diffuse emission from the Milky Way and nearby galaxies. The MWA in the Phase I compact configuration was sensitive to scales of up to $\sim7.4$ deg based on the minimum baseline of 7.7 m at 300 MHz. As described in the introduction, GMIMS-LBS (Wolleben et al. Reference Wolleben2019) covers the entire GLEAM-300 band and almost the entire survey area (up to $\delta_{\text{J2000}}=+20^{\circ}$ ). Having been observed with Murriyang, a 64 m single dish telescope, GMIMS-LBS can provide the missing ‘zero spacings’ in GLEAM-300 and provide sensitivity in spatial scales across the entire sky.

By way of example we have combined the GLEAM-300 mosaic of the region surround the LMC and SMC (field ‘J0326-72’) with GMIMS-LBS using the ‘feathering’ algorithm (Weiß et al. Reference Weiß, Neininger, Hüttemeister and Klein2001).Footnote ^w We show the original and ‘feathered’ image in Figure 25, highlighting the additional recovered signal from the large-scale features in the LMC in particular. We intend to combine the entirety of GLEAM-300 with GMIMS-LBS as an enhanced data product in future work.

Figure 25. Example mosaic region containing both the LMC and SMC, comparing the GLEAM-300 image by itself (a) and the GLEAM-300 image feathered with an equivalent image from GMIMS-LBS (Wolleben et al. Reference Wolleben2019) (b). The brightness scaling is the same in both panels, and the feathered image highlights additional extended emission within/around both the LMC and SMC as well as Galactic emission in this region.