5.1. Noise analysis

5.1.1. Measured noise

To determine the effectiveness of the mosaicking process in reducing the RMS, we performed a jack-knifing test for each GAMA field. We generated mosaics with varying numbers of included observations in a range from four to the maximum available, doubling the number of observations included each time. For each increment of the number of observations included, we created 100 mosaics with the included snapshot images being selected randomly from the total available. All observation snapshots used in the mosaics were resampled to the same grid.

The noise for each mosaic was estimated with BANE, measuring the average RMS in each GAMA region. The results are shown in Figure 3, which shows the effect of a limiting noise on the mosaics. For both fields, the noise decreases as expected theoretically (i.e. as the square-root of the number of observations), but then deviates with the co-addition of more than 16 snapshots for GAMA23 and more than 64 for GAMA9. Hence, the thermal noise dominates down to these levels, but after mosaicking more than 64–128 snapshot images, the decrease in RMS reduction tapers off considerably suggesting that the fields are reaching a confusion limit, which is likely a combination of sidelobe and classical confusion. The lower decrease in RMS observed in the GAMA9 field was likely caused by a higher sidelobe confusion from the large number of bright sources in the field.

Figure 3.

Jackknife plot for the mean RMS measured at the centre of the mosaic (y-axis) vs the number of observations mosaicked (x-axis). Each increment of the observation count was mosaicked 100 times with a randomised observation selection. The dashed grey lines represent the theoretical optimum noise reduction by the square-root of the number of observations. The dashed sandy brown line represents the median RMS achieved in GLEAM-X DRII (1.5 mJy/beam). The RMS reductions follows the theoretical optimum noise reduction for a mosaic count up to 64 snapshots for GAMA23 and 128 snapshots for GAMA9. After this point, as further snapshots are mosaicked, the RMS reduction tapers off indicating it has reached the sidelobe and/or classical confusion limit.

Estimation of the final noise in the GAMA23 deep field mosaic is difficult because the image approaches the classical confusion limit. BANE does not naturally function in this regime as it expects sparse input images dominated by Gaussian thermal noise. For this reason, we used BANE to characterise the structure of the noise while implementing a novel approach to estimate the magnitude for the MIDAS mosaics.

We masked known sources in the $5^{\circ}\times 12^{\circ}$ GAMA23 region of our deep image using a higher frequency, deeper image produced from the EMU survey by:

• • Convolving the EMU deep field image of the GAMA23 region at 887.5 MHz produced by Gürkan et al. (Reference Gürkan2022) to the same beam size as the GAMA23 MIDAS mosaic.

• • Re-projecting the convolved and the original EMU mosaics to the same pixel grid as the GAMA23 MIDAS mosaic.

• • Masking pixels in the MIDAS GAMA23 mosaic where there was a corresponding pixel above the reported $5 \sigma$ threshold ( $5 \times 0.038$ mJy/beam) in either the original EMU mosaic, or the convolved EMU mosaic. This is equivalent to $5 \times 0.12$ mJy/beam at the 215.68 MHz observed in MIDAS assuming a spectral index of $\alpha=-0.8$ .

• • Dilating all masked pixels with a fully connected $3 \times 3$ structure function for two iterations using the scipy ndimage Python package.

This resulted in 49% of the GAMA23 region being masked with the remaining unmasked pixels in the region consisting mainly of noise. An example cutout of the implemented filter is presented in Appendix A1 exhibiting its effectiveness in removing sources.

A histogram of the remaining pixels was plotted with a bin size of 5 $\unicode{x03BC}$ Jy/beam and then a Gaussian fitted with scipy curve_fit as seen in Figure 4 with the resultant values described in Table 6. A Gaussian with a zero point $x_0=-96\pm 1$ $\unicode{x03BC}$ Jy/beam and $\sigma = 331 \pm 1$ $\unicode{x03BC}$ Jy/beam provided a very good fit with a small amount of additional positive flux observed due to incomplete masking. The masking of the bright areas leaves the underconvolved negative sidelobes of each bright source resulting in a 96 $\unicode{x03BC}$ Jy/beam offset to the zero point. This offset value is arbitrary and does not offer any meaningful insight. Therefore, we concluded the RMS within the GAMA23 region is $0.331 \pm 0.001$ mJy/beam.

Table 6.

The resultant values calculated in Section 5.1 for a Gaussian fitted to the remaining pixel’s values in the GAMA9 and GAMA23 mosaics after filtering out sources, used to estimate the RMS of the mosaics. The first column is the region mosaicked, the second column is the estimated classical confusion noise, the third is the theoretically predicted RMS calculated by Equation (4), the fourth is the $x_0$ value of the fitted Gaussian and the fifth is the reported $\sigma$ .

Table 6. Long description

A table with three rows and five columns comparing predicted and measured values for GAMA9 and GAMA23 mosaics. The columns are labeled as Mosaic, Predicted (mJy/beam) with sub-columns σ_conf and σ_p, and Measured (mJy/beam) with sub-columns x_0 and σ. Row 1: Mosaic, GAMA9; σ_conf, 0.358; σ_p, 0.833; x_0, -0.171 ± 0.001; σ, 0.923 ± 0.003. Row 2: Mosaic, GAMA23; σ_conf, 0.231; σ_p, 0.371; x_0, -0.096 ± 0.001; σ, 0.331 ± 0.001.

Figure 4.

Distribution of the pixel intensity in the GAMA9 region (left) and GAMA23 region (right) as a function of flux density for the deep images after masking all pixels above $5 \sigma$ in the EMU image. In the GAMA9 histogram, the black dashed line is a Gaussian with $x_0 = -0.171$ mJy/beam and $\sigma = 0.923 $ mJy/beam fit to the histogram. For the GAMA23 image, the black dashed line represents a Gaussian with $x_0 = -0.096$ mJy/beam and $\sigma = 0.331$ mJy/beam.

To map the structure of the noise in the GAMA23 mosaic, BANE was run on the deep image with the default arguments. We then applied the same mask used to measure the RMS in the deep image to the RMS map output by BANE. The median value for the unmasked pixels in the GAMA23 region was then used to determine the scaling factor $f=\sigma/\tilde{i}$ where $\sigma$ is the RMS determined from the Gaussian fit and $\tilde{i}$ is the median of the GAMA23 region in the masked RMS map. The entire RMS map was then multiplied by the scaling factor, calculated as $f=0.81$ , to determine the RMS distribution in the GAMA23 deep field image.

Due to the lower elevation of the GAMA9 field, the resulting RMS is higher with a larger combined natural and sidelobe confusion limit. In addition, the sidelobe noise is spatially correlated to the noise in the mosaic. The same methods that were used to calculate the RMS for the GAMA23 mosaic were used for the GAMA9 deep image.

A higher frequency deep ASKAP image mosaic for this field has not been published; however, the project ‘Survey with ASKAP of GAMA-09+X-ray’ (SWAG-X; Moss et al. Reference Moss, Hotan, Whiting and Huynh2020) has produced several tiles of this region. To create the base image required for the mask, we obtained seven tiles around GAMA9 from the SWAG-X survey that were conducted with ASKAP at 887.5 MHz. We then co-added the tiles with SWARP and estimated the RMS with BANE, using the mean of the RMS within the GAMA9 deep field ( $\sigma=0.054$ mJy/beam) to set our threshold of $5\sigma$ for the pixel masking. The resultant mask covered 65% of the GAMA9 deep field.

Figure 5.

RMS maps for the GAMA9 (left) and GAMA23 (right) MIDAS mosaics where the rectangles represent the regions targeted with GAMA9 in green and GAMA23 in yellow. The circles represent the area used for the completeness estimation and source counts for GAMA9 (orange) and GAMA23 (blue).

A Gaussian with $x_0 = -0.171 \pm 0.001$ mJy/beam and $\sigma = 0.923 \pm 0.001$ mJy/beam was fit to the histogram of the masked region. As was done with the GAMA23 deep image, BANE was run on the GAMA9 image and the RMS map was scaled accordingly with a calculated scale factor $f=0.89$ . The resulting RMS is significantly higher than the GAMA23 image due to the observing limitations of the field. The off-zenith observations result in a higher RMS, then the addition of the bright sources in the sidelobes increases the RMS even further while additionally introducing a noise structure to the mosaic. This constraint on the RMS is corroborated when examining the ratios between the GAMA fields for the average RMS of the individual snapshots (outlined in Table 1) and the ratio of the RMS for the final mosaics (outlined in Table 6), where $\sigma_\mathrm{G9}/\sigma_\mathrm{G23} = 2.8$ for both.

Due to the attenuation of the primary beam, the RMS increases significantly at the edges of each mosaic as seen in Figure 5. In the GAMA23 deep field image, the RMS reaches close to 6 mJy/beam at the edge while in the GAMA9 mosaic the RMS increases to slightly over 10 mJy/beam at the edge. However, due to the large field-of-view of the MWA, the RMS in the central $60\,\deg^2$ GAMA fields remain relatively uniform.

5.1.2. Theoretical noise

We estimated the theoretical classical confusion noise ( $\sigma_\mathrm{conf}$ ) for each MIDAS mosaic using the methods derived by Condon (Reference Condon1974) where

(3) \begin{equation}\sigma_\mathrm{conf} = \left(\frac{q^{3-\gamma}}{3-\gamma}\right)^{\frac{1}{\gamma-1}} \left( k \frac{ \pi\,B_\mathrm{maj} B_\mathrm{min} }{4 (\gamma-1) \ln(2)}\right)^{\frac{1}{\gamma-1}}\end{equation}

in which $B_\mathrm{maj}$ and $B_\mathrm{min}$ are the major and minor axes for the average restoring beam obtained by WSClean from the final fitted PSF listed in Table 8 and q is the SNR that we take to be $q=1$ at the confusion limit. Using the LOFAR power-law source counts by Williams et al. (Reference Williams2016), we estimate the normalisation $k=4 \times 10^{-6}$ with the source count slope $\gamma=1.5$ at flux densities around 1 mJy. Note this slope has a conservatively steep value as the two fields only just reach the up-turn in the source counts at this flux density.

Combining the theoretical classical confusion noise with the theoretically optimal noise for n number of observations mosaicked using the mean RMS from the individual snapshots ( $\overline{\sigma_1}$ ) where

(4) \begin{equation}\sigma_p = \frac{ ( \overline{\sigma_1} - \sigma_\mathrm{conf} ) }{ \sqrt{n} } + \sigma_\mathrm{conf}\end{equation}

resulted in the predicted RMS ( $\sigma_p$ ) for the MIDAS mosaics listed in Table 6. However, this estimation does not account for sidelobe confusion that dominates GAMA9. Additionally, the confusion noise is highly sensitive to the source count slope which is rapidly changing at the faintest fluxes. We observe agreement with our predictions for GAMA23, while GAMA9 is slightly under-predicted as expected due to the large amount of sidelobe noise contributed by Hydra-A.

After scaling by a common spectral index ( $\alpha=-0.8$ for $S_\nu\propto\nu^\alpha$ ) and comparing to wide-field ( $\gt $ $60\text{ deg}^2$ ) radio surveys in the Southern hemisphere, Figure 6 shows that the GAMA9 mosaic reaches a depth deeper than GLEAM (Hurley-Walker et al. Reference Hurley-Walker2017, Reference Hurley-Walker2019), GLEAM-X (Hurley-Walker et al. Reference Hurley-Walker2022; Ross et al. Reference Ross2024) and GLEAM-300 (Duchesne et al. Reference Duchesne2025a). It achieves an RMS comparable to surveys such as the Rapid ASKAP Continuum Survey (RACS-low; McConnell et al. Reference McConnell2020, RACS-mid; Duchesne et al. Reference Duchesne2023, RACS-high; Duchesne et al. Reference Duchesne2025b) and the Very Large Array Sky Survey (VLASS; Lacy et al. Reference Lacy2020), probing a lower-frequency complementary range. However, the GAMA23 mosaic probes significantly deeper, offering a depth comparable to the expectant GAMA Legacy ATCA Sky Survey (GLASS; Huynh et al. in prep) and close to the 887.5 MHz early EMU release (Norris et al. Reference Norris2021). However, the MIDAS mosaics are much shallower than the LoTSS survey (Shimwell et al. Reference Shimwell2017, Reference Shimwell2019, Reference Shimwell2022) of the northern sky and the expected RMS for the Giant Metrewave Radio Telescope (GMRT; Seymour et al. Reference Seymour2020) survey.

The MIDAS 215.68 MHz deep field data along with the plethora of higher frequency data available for the GAMA23 field offers an opportunity to probe this area of the sky further with the potential for new insight into galaxy clusters, halos, relics, AGN and more.

Figure 6.

Comparison of the 215.68 MHz MIDAS GAMA9 and GAMA23 deep images to wide-field surveys ( $\gt $ $60\text{ deg}^2$ ) performed at radio wavelengths assuming a constant spectral index of $\alpha=-0.8$ . Here the x-axis represents the wavelength of the survey measured in MHz and the y-axis reflects the RMS of the deep field image measure in mJy/beam. The solid circles are completed published surveys while the open circles are estimations for unpublished surveys currently in progress.

5.2. Source catalogues

The source catalogues for the MIDAS GAMA9 and GAMA23 surveys were created with AEGEAN using a blind source finding method. We used AEGEAN’s default parameters except for the seedclip, PSF, RMS, and background values. The seedclip was set to $4 \sigma$ to detect pixels above this flux density for initial positions used in the component fitting. For the additional parameters, we utilised the PSF generated in Section 4.4 along with the rescaled RMS map and the constant background value calculated in Section 5.1. This resulted in a total of 67 985 sources found in the GAMA9 mosaic and 106 221 sources found in the GAMA23 mosaic, where we define a source as an individual fitted Gaussian component. The catalogues were reduced to sources with a peak flux density above the $5 \sigma$ detection level where the RMS is determined from the rescaled RMS maps produced in Section 5.1. Next, the sources within 1^′ from the edge of the mosaic were removed from the GAMA catalogues to ensure that there were no sources with flux density overlapping the edge of the mosaic.

Table 7.

The number of sources detected within each survey, and the rows represent each cut that was applied. Each column lists the numbers of sources remaining after each cut where the full image columns are the number of sources within the full survey mosaic and the region columns are the number of sources within the GAMA9 (RA $+129^{\circ}$ to $+141^{\circ}$ , Dec $-2^{\circ}$ to $+3^{\circ}$ ) and GAMA23 (RA $+339^{\circ}$ to $+351^{\circ}$ , Dec $-35^{\circ}$ to $-30^{\circ}$ ) regions, respectively.

Table 7. Long description

A table with two rows and seven columns comparing beam dimensions and source counts in GAMA9 and GAMA23 fields. The columns are labeled B maj (arcsec), B min (arcsec), B pa (degrees), Beam area (deg^2), and Beams per source. The rows are labeled GAMA9 and GAMA23. Row 1: B maj, 64.3; B min, 48.7; B pa, -66.1; Beam area, 2.73 times 10^-4; Beams per source, 5.0. Row 2: B maj, 57.3; B min, 43.9; B pa, -27.4; Beam area, 2.20 times 10^-4; Beams per source, 32.4.

Table 8.

The beam dimensions calculated from the source dimensions of unresolved sources in the MIDAS catalogues along with the resultant beams per source calculated for each GAMA field’s mosaic.

Table 8. Long description

The table presents beam dimensions and resultant beams per source for GAMA9 and GAMA23 fields. It has four rows and six columns. The columns are labeled as Cut, GAMA9 Full image, GAMA9 Region, GAMA23 Full image, and GAMA23 Region. The rows are labeled as Initial, S peak < 5 sigma, On edge, and S int / S peak > 16. Row 1: Initial, 67985, 5043, 106221, 10591. Row 2: S peak < 5 sigma, 53515, 4227, 80523, 8423. Row 3: On edge, 53431, 4227, 80428, 8423. Row 4: S int / S peak > 16, 53419, 4225, 80407, 8417.

We visually inspected sources with high integrated to peak flux density ratios in both MIDAS mosaics. It was determined all sources with a ratio $S_\mathrm{int}/S_\mathrm{peak} \gt 16$ were spurious and therefore removed from the catalogues. Due to the higher noise around bright sources caused by sidelobes, sources nearby are difficult to validate as they could be artefacts from the sidelobes. As a precaution, we flagged sources that were within 4^′ of a 1.5 Jy sources as lower reliability. As a result, 429 sources in the GAMA9 catalogue and 852 sources in the GAMA23 catalogue were flagged.

As detailed in Table 7, this resulted in a total of 53 419 sources in the final GAMA9 catalogue and 80 407 sources in the GAMA23 catalogue. The 24 column names in the MIDAS catalogues are outlined in Appendix C. For GAMA9, 3.4% of islands were multi-component while 5.0% of the islands in GAMA23 had multiple components.

The average beam dimensions were calculated from the source dimensions of compact sources within the MIDAS catalogues selected with the RMS envelope described in Section 5.3. The average $B_\mathrm{maj}$ and $B_\mathrm{min}$ for sources within the respective GAMA regions were calculated and are outlined in Table 8. Within the GAMA9 region, 4 225 sources were detected with a beam size of $2.73 \times 10^{-4} \text{ deg}^2$ for the mosaic, resulting in a source density of 70.42 $\text{sources} / \text{deg}^2$ (52.0 beams per sourceFootnote ^o ). However, a much larger number of sources were detected within the GAMA23 region with a total of 8 417 sources. With a beam size of $2.20 \times 10^{-4} \text{ deg}^2$ for the mosaic, we approach the classical confusion limit with a source density of 140.28 $\text{sources} / \text{deg}^2$ (32.4 beams per source).

Figure 7.

Ratio of integrated flux density to peak flux density as a function of SNR for the MIDAS GAMA9 (left) and GAMA23 (right) source catalogues. The green line is the upper envelope, Equation (5), set to define the limit for unresolved sources that we approximate to be 95% of the total sources. The small bump observed in GAMA23 around $S_\mathrm{peak}/\sigma \approx 100$ and the less noticeable bump observed in GAMA9 around $S_\mathrm{peak}/\sigma \approx 60$ is suspected to be a result of sources on the detection limit in the individual observations; these sources would be deconvolved in some observations, but not in others.

5.3. Flux density scale evaluation

To evaluate the flux density scales of our catalogues, we compared the MIDAS catalogues to the wideband integrated flux densities from the GLEAM catalogue (Hurley-Walker et al. Reference Hurley-Walker2017) and to the GLEAM-X DRII catalogue (Ross et al. Reference Ross2024) for GAMA23. The GLEAM(-X) flux densities were scaled to 215.68 MHz using an assumed constant spectral index of $\alpha = -0.8$ . At the time of writing, there are no public data releases from GLEAM-X that cover the GAMA9 field.

The MIDAS catalogues were matched to GLEAM(-X) with a maximum separation distance of 1^′. The MIDAS catalogue was reduced to only include compact source, additionally any MIDAS source with a neighbour within 2^′ was removed to ensure that sources within the GLEAM(-X) catalogues were not resolved into multiple sources within the MIDAS catalogues. Compact sources were selected from the catalogue based on their integrated flux density to peak flux density ratio. We plotted the ratios as a function of SNR as seen in Figure 7 where we estimated an upper envelope of

(5) \begin{equation}\frac{S_\mathrm{peak}}{S_\mathrm{int}} = 0.98 - \frac{1.6}{(S_\mathrm{peak}/\sigma)^{0.6}}\end{equation}

to capture 95% of the sources, where $S_\mathrm{peak}$ is the measured peak flux, $S_\mathrm{int}$ is the measured integrated flux and $\sigma$ is the measured local RMS. A small bump is observed in the ratios around $S_\mathrm{peak}/\sigma \approx 60$ –100 that is likely caused by unCLEANed sources. The masking stage discussed in Section 3.2.2 results in only sources $\gt 3\sigma_i$ being CLEANed (where $\sigma_i$ is the local RMS in an individual 2-min snapshot). Therefore, sources with a flux density near this detection limit will be CLEANed in some snapshots and not others, resulting in more unpredictable source morphologies after mosaicking. Sources below this detection limit will not be CLEANed at all.

A correctable bias in the flux density measurements can be introduced from measurement errors known as Eddington bias (Eddington Reference Eddington1913). Flux density measurements for sources will scatter above and below their true values. As there are a greater number of the fainter sources (due to the steep source counts) that will be measured brighter than there are bright sources that will be measured fainter, the flux density measurements for the faint sources are much more likely to be overestimated. To evaluate the Eddington bias in our flux density scale comparisons, we use the maximum-likelihood value ( $S_\mathrm{ml}$ ; Hogg & Turner Reference Hogg and Turner1998; Hurley-Walker et al. Reference Hurley-Walker2017),

(6) \begin{equation}S_\mathrm{ml} = \frac{S}{2} + \frac{S}{2} \sqrt{1 - \frac{4q+4}{(S / \sigma)^2}}\end{equation}

where S is the measured integrated flux density, $\sigma$ is the local RMS measured for the source, and $q=1.54$ is the logarithmic source count slope at the GLEAM flux densities (Franzen et al. Reference Franzen2016; Hurley-Walker et al. Reference Hurley-Walker2022; Ross et al. Reference Ross2024). This correction made a negligible difference in our results so we chose not to apply it.

The flux density scale comparisons with GLEAM(-X) in Figure 8 exhibits good agreement with GLEAM. For GLEAM-X DRII, we observe agreeance at an SNR $\gt {\sim} 100$ ; however, at lower SNR values, we observe a discrepancy where GLEAM-X reports lower source flux densities. This is likely due to the CLEANing methods used in GLEAM-X where a $3\sigma$ auto-mask was used. As discussed in Section 3.2.2, we found this level of auto-mask would CLEAN noise peaks removing flux from real sources and introduce a small offset into the flux density measurements. This results in an under-prediction of flux density, which is more evident for fainter sources and aligns with the observed discrepancy.

Figure 8.

The integrated flux density ratios between GLEAM and GAMA9 MIDAS at the top, GLEAM and GAMA23 MIDAS in the middle and GLEAM-X and GAMA23 MIDAS at the bottom. The x-axes represents the MIDAS SNR and the y-axes is the flux density ratio between the surveys. The solid black line represent a ratio of 1. Here we observe good agreement with GLEAM through the entire flux density range. We observe agreement with GLEAM-X at an SNR $\gt {\sim} 100$ , however at lower SNR values GLEAM-X reports lower flux density scales. We attribute this to the CLEANing methods used in GLEAM-X resulting in an over-estimate of the observed vs true flux density values.

Figure 8. Long description

Three scatter plots compare flux density ratios between different surveys. Panel A: The top scatter plot shows the ratio of flux densities between GLEAM and GAMA9 MIDAS. The x-axis represents the MIDAS signal-to-noise ratio (SNR) and the y-axis represents the flux density ratio. The solid black line indicates a ratio of 1. The color gradient from red to yellow indicates the number of sources, with red representing fewer sources and yellow representing more sources. Panel B: The middle scatter plot shows the ratio of flux densities between GLEAM and GAMA23 MIDAS. The x-axis represents the MIDAS SNR and the y-axis represents the flux density ratio. The solid black line indicates a ratio of 1. The color gradient from green to yellow indicates the number of sources, with green representing fewer sources and yellow representing more sources. Panel C: The bottom scatter plot shows the ratio of flux densities between GLEAM-X and GAMA23 MIDAS. The x-axis represents the MIDAS SNR and the y-axis represents the flux density ratio. The solid black line indicates a ratio of 1. The color gradient from pink to yellow indicates the number of sources, with pink representing fewer sources and yellow representing more sources.

The flux density scaling for MIDAS was performed using GLEAM as a reference during post-imaging as described in Section 3.3. GLEAM reports an uncertainty of 8% for its flux density scale accuracy, therefore we prescribe an 8% uncertainty for the MIDAS flux density scaling when compared to other surveys.

5.4. Astrometry

The precision of the astrometry for the final MIDAS GAMA9 and GAMA23 source catalogues was evaluated by comparing the source positions with the joint NVSS and SUMSS positional-based catalogue of unresolved sources described in Section 3.3, which we will refer to as the reference catalogue. The GAMA deep field catalogues were reduced to sources $\gt 50\sigma$ and matched to the reference catalogue with a matching distance within 1^′, where the source positions in the reference catalogue were assumed to be the correct positions. We then calculated the offsets for the deep field sources relative to the reference sources and calculate the mean offset and standard deviation for the offset per catalogue.

For the GAMA9 catalogue, there were a total of 8 749 sources above $50\sigma$ matched to the reference catalogue. The resulting mean RA offset was $+0.007'' \pm 0.706''$ and the mean Dec offset was $+0.002'' \pm 1.440''$ as seen in Figure 9.

Figure 9.

The astrometric offsets for the GAMA9 source catalogue (left) and GAMA23 source catalogue (right) compared to a reference catalogue containing unresolved sources from NVSS and SUMSS. The x-axes and y-axes are the offsets for the RA and Dec respectively measured in arcseconds. A total off 8 749 sources was compared resulting in an RA offset of $+0.007'' \pm 0.706''$ and a Dec offset of $+0.002'' \pm 1.440''$ for GAMA9 and a total 13 355 sources for GAMA23 resulting in an average RA offset of $-0.009'' \pm 1.054''$ and a Dec offset of $+0.010'' \pm 1.149''$ . The red dotted lines represents the mean RA and Dec offsets.

The GAMA23 catalogue matched 13 355 sources above $50\sigma$ to the reference catalogue. As seen in Figure 9, an average RA offset of $-0.009'' \pm 1.054''$ and a Dec offset of $+0.010'' \pm 1.149''$ was calculated.

We additionally confirm astrometry matching to RACS-high which offers a higher angular resolution of $11.8'' \times 8.1''$ but experiences astrometry limitations with positional uncertainties of $0.6''$ in right-ascension and $0.7''$ in declination (Duchesne et al. Reference Duchesne2025b). An average RA offset of $+0.040''\pm3.916''$ and Dec offset of $-0.095''\pm3.666''$ was calculated between the GAMA9 catalogue and the RACS-high catalogue. For the GAMA23 catalogue, an average RA offset of $+0.24''\pm3.97''$ and a Dec offset of $-0.091''\pm3.904''$ was calculated. Due to the limitations of RACS-high, the results have large uncertainties; however, they are in good agreement.

5.5. Completeness

To estimate the percentage of true sources recovered at a given flux density, which we define as the completeness, we injected simulated sources into random positions in each deep field image at selected flux density levels. To accomplish this, we used AeRes from the AEGEAN package that enables adding or removing sources from an image with specified positions, flux densities, and shapes. We ran 26 simulations for flux density levels ranging from 1 to 316 mJy, evenly spaced in log space, injecting ${\sim}10$ point-like sources per square-degree at a constant flux density for each simulation.Footnote ^p To ensure that we were not introducing an unnecessary confusion measurement, a minimum separation distance between simulated sources was set at 5^′. No minimum separation distance was established between the simulated and real sources to ensure that the classical confusion limit was accurately taken into account.

Blind source finding with AEGEAN was run on each simulation and the percentage of injected sources recovered was calculated. To estimate the uncertainty for the completeness calculations, we ran the full simulation process 100 times for each region. The mean for each flux density level was then used for our completeness values and the standard deviation for the uncertainty.

The completeness curves for the GAMA regions are shown in Figure 10, although part of the shape of the curves will be due to the increase in the RMS towards the edges of the fields of view that is observed in Figure 5. To attempt to minimise this effect, we measure the completeness for a circular area in the centre of each region. For the GAMA9 field, a circular area with a radius of $15^{\circ}$ was used resulting in an estimated completeness of 50% at ${\sim}7.0$ mJy and 90% at ${\sim}15.6$ mJy. A $12^{\circ}$ radius was used for the circle at the centre of the GAMA23 region resulting in a completeness of 50% at ${\sim}2.4$ mJy and 90% at ${\sim}5.0$ mJy.

Figure 10.

The completeness for a circular area with a radius of $15^{\circ}$ in the GAMA9 (green) mosaic and a circle with a radius of $12^{\circ}$ in the GAMA23 (blue) mosaic as a function of flux density. The errors were calculated as the standard deviation from 100 simulations. The dashed black line represents 100% completeness. A 90% completeness was estimated at ${\sim}10.06$ mJy for GAMA9 and ${\sim}4.97$ mJy for GAMA23.

5.6. Source counts

The source counts for each deep field catalogue were calculated for a circular area at the centre of each GAMA region. For the GAMA9 mosaic, we used a circle with a radius of $15^{\circ}$ centred at RA $+135^{\circ}$ , Dec $+0.5^{\circ}$ ; for the GAMA23 mosaic, we used a circle with a radius of $12^{\circ}$ centred at RA $+345^{\circ}$ , Dec $-32.5^{\circ}$ (i.e. the same regions over which we determined the completeness). The source Hydra-A was excluded from the GAMA9 MIDAS source counts as it is a very bright outlier. The sources detected in each circular region were binned by flux density level in 15 bins; the faintest 12 bins are equal size in log space, while the three brightest bins were double the size to account for the lower number of sources. The value for the start of the first bin was calculated as five times the average RMS of all selected sources, while the end of the final bin was set as the value of the brightest source.

The GAMA9 source catalogue was binned from $7.44 \times 10^{-3}$ to 23.3 Jy with a bin width of $\Delta \log_{10}(S) = 0.194$ Jy for the first 12 bins and a width of $\Delta \log_{10}(S) = 0.388$ Jy for the brightest 3 bins. For GAMA23, we used a bin width of $\Delta \log_{10}(S) = 0.217$ in log-space for the first 24 bins and $\Delta \log_{10}(S) = 0.434$ Jy for the 3 brightest bins, covering a flux density range from $2.58 \times 10^{-3}$ Jy to 21.0 Jy.

Figure 11.

Source counts normalised to Euclidean values for GAMA9 (blue) and GAMA23 (gold) as a function of flux density in comparison to low-frequency counts from literature scaled to 215.68 MHz. Here we observe good agreement with the 200 MHz counts by Franzen et al. (Reference Franzen, Vernstrom, Jackson, Hurley-Walker, Ekers, Heald, Seymour and White2019), but a minor deviation from Venville et al. (Reference Venville, Parkinson, Hurley-Walker, Galvin and Ross2024) at a flux density below ${\sim} 40$ mJy. We attribute this deviation to the lack of a completeness correction applied to their source counts. We observe good agreement with the 144 MHz source counts by Shimwell et al. (Reference Shimwell2026) with a slight deviation at the bright end with GAMA9 and a small deviation at the faint end with GAMA23 as it approaches the GAMA9 observing limit. Comparing to the ${\sim}150$  MHz counts by Mandal et al. (Reference Mandal2021) and Williams et al. (Reference Williams, Intema and Röttgering2013), we observe agreeance down to ${\sim} 25$ mJy. Below this we observe a deviation in the counts that is likely due to the differing resolutions and flux density limits of the surveys along with the different source finding methods employed. The black and grey dashed lines represent a fitted third-degree polynomial to the GAMA23 and GAMA9 source counts respectively.

Figure 11. Long description

A line graph compares source counts normalized to Euclidean values for GAMA9 (blue) and GAMA23 (gold) as a function of flux density. The x-axis represents the 215.68 megahertz flux density in millijanskys, ranging from 0.1 to 100000. The y-axis represents the source counts normalized to Euclidean values, ranging from 10 to 10000. The graph includes multiple data series: GAMA9 (blue squares), GAMA23 (gold diamonds), Shimwell 2026 - 144 megahertz (red triangles), Mandal 2021 - 150 megahertz (green triangles), Williams 2013 - 153 megahertz (red circles), Franzen 2019 - 200 megahertz (purple crosses), and Venville 2024 - 200 megahertz (brown stars). The black and grey dashed lines represent a fitted third-degree polynomial to the GAMA23 and GAMA9 source counts respectively. The data series show varying degrees of agreement and deviation across different flux densities, with specific trends and patterns highlighted in the caption.

Here in Figure 11, we present the source counts for our deep field catalogues normalised to Euclidean values. These source counts have been corrected for the incompleteness calculated in Section 5.5 and are listed in Tables 9 and 10. The corrected normalised source counts ( $C_N$ ) were calculated by

(7) \begin{equation}C_N = S^{2.5} \frac{dN}{a\,dS}\end{equation}

where S is the centre of the flux density bins in Jy, dN is the number of sources in the bin ( $N_s$ ) multiplied by the correction factor for completeness calculated from the linear fit between the two closest flux density bins determined in Section 5.5, a is the area sampled in steradian, and dS is the width of the bin in Jy.

Table 9.

The source counts calculated with the GAMA9 mosaic. The first column is the bin ranges, the second is the bin centres, and the third is the number of sources detected in the bin. The fourth column is the correction factor applied to the number of sources, calculated from the completeness presented in Section 5.5. The fifth column is the source counts normalised to Euclidean values along with their uncertainty estimated from Poisson statistics, the uncertainty due to cosmic variance and completeness errors added in quadrature.

Table 9. Long description

A table titled ‘The source counts calculated with the GAMA9 mosaic.’ The table has five columns: Bin range, Bin centre, Number of sources, Correction factor, and S^2.5 dN/dS. The table contains 15 rows of data. The Bin range column lists the flux density ranges in milliJanskys (mJy). The Bin centre column lists the centre of the flux density bins in millijanskys (mJy). The Correction factor column lists the correction factors applied to the number of sources. The S^2.5 dN/dS column lists the source counts normalized to Euclidean values along with their uncertainty. The table provides detailed data on the source counts and their corrections for incompleteness.

The uncertainty for the source counts was estimated using Poisson statistics for a confidence interval equivalent to a $1 \sigma$ Gaussian described by Gehrels (Reference Gehrels1986) where for the upper ( $\lambda_u$ ) and lower ( $\lambda_l$ ) limits,

(8) \begin{align}\lambda_u &\approx \sqrt{dN + 1} + 1 \nonumber\\[3pt]\lambda_l &\approx \sqrt{dN - 0.25}.\end{align}

For large values of dN these uncertainties tend towards $\lambda=\sqrt{dN}$ ; however, these uncertainties are formally correct for $dN\lesssim 20$ .

Additionally, we add cosmic variance to the uncertainty, which we calculate for each bin using the LOFAR Two-metre Sky Survey (LoTSS) Data Release 2 catalogue (Shimwell et al. Reference Shimwell2022). The data release specifies a potential flux density variation dependant on pointing which may limit this analysis, but we continue with the assumption the effects are small. The flux density of these sources was scaled from 144 to 215.68 MHz under the assumption of a constant spectral index of $\alpha = -0.8$ . The source counts for each bin in each region were measured for 100 randomly positioned circular areas within the area surveyed by LoTSS DR2. The size of the circles was equal to those that were used to measure the deep field GAMA source counts, $15^{\circ}$ for GAMA9 and $12^{\circ}$ for GAMA23. The centre location of the circles were determined by selecting random sources in the LoTSS catalogue. To ensure that the region was not on the edge of the measured field and had complete coverage, we searched an area of 2^′ at the upper and lower extremes for the RA and Dec coordinates of the circle’s edges to ensure there were sources present. The cosmic variance for the flux density bins below 2 Jy was calculated to be ${\sim} 5$ –10%, but for the brighter bins the cosmic variance increases significantly to ${\sim} 20$ –35%. The error on the corrected source counts was taken as the sum in quadrature of the Poisson error, completeness correction error, and uncertainty due to cosmic variance.

The resulting normalised source counts seen in Figure 11 are compared to the counts generated from other low-frequency surveys scaled by $\alpha=-0.8$ . We can not compare source counts from surveys at very different frequencies due to selection biases resulting in changes to spectral index distribution between the frequency ranges. This means a simple spectral index conversion could not be used for comparison with surveys completed at frequencies significantly different from the 215.68 MHz we observed at (Kellermann Reference Kellermann1964; Franzen et al. Reference Franzen2016). Unfortunately, no source counts have been previously calculated at 216 MHz, so we made comparisons with surveys conducted at 144, 150, 153, and 200 MHz.

For our comparisons, we used counts by Williams et al. (Reference Williams, Intema and Röttgering2013) who utilised observations of the National Optical Astronomy Observatory (NOAO) Boötes field with the Giant Metrewave Radio Telescope (GMRT) at 153 MHz producing an image reaching an RMS of 2 mJy/beam covering an area of $30 \text{ deg}^2$ . Additionally, we included source counts at 200 MHz from Franzen et al. (Reference Franzen, Vernstrom, Jackson, Hurley-Walker, Ekers, Heald, Seymour and White2019) using the GLEAM survey that reaches an RMS of 10 mJy/beam and covers an area of $25\,000\text{ deg}^2$ , which reduces potential influence of cosmic variance and improves accuracy at the bright end. We also include the 200 MHz source counts produced by Venville et al. (Reference Venville, Parkinson, Hurley-Walker, Galvin and Ross2024) that utilises the central $5\,600 \text{ deg}^2$ area of the GLEAM-X DR2 survey which contains the lowest and least variable RMS properties but do not correct for completeness. Finally, we included source counts produced from the LoTSS survey which offer better insight at the faint end. We include the 150 MHz counts by Mandal et al. (Reference Mandal2021) that utilises the deep field observations from the survey reaching an RMS of 0.1 mJy/beam for a $25\text{ deg}^2$ area along with the counts published in LoTSS-DR3 by Shimwell et al. (Reference Shimwell2026) covering 88% of the northern sky, achieving an average RMS of 0.092 mJy/beam.

Table 10.

The source counts calculated with the GAMA23 mosaic. As with Table 9, The first column is the bin ranges, the second is the bin centres, and the third is the number of sources detected in the bin. The fourth column is the correction factor applied to the number of sources, calculated from the completeness calculated in Section 5.5. The fifth column is the source count normalised to Euclidean values along with their uncertainty estimated from Poisson statistics, the uncertainty due to cosmic variance and completeness errors added in quadrature.

Table 10. Long description

A table titled ‘The source counts calculated with the GAMA23 mosaic.’ The table has five columns: Bin range, Bin centre, Number of sources, Correction factor, and S^2.5 dN/dS. The table contains 15 rows of data. The Bin range column lists the flux density ranges in milliJanskys (mJy). The Bin centre column lists the centre of the flux density bins in millijanskys (mJy). The Correction factor column lists the correction factors applied to the number of sources. The S^2.5 dN/dS column lists the source counts normalized to Euclidean values along with their uncertainty. The table provides detailed data on the source counts and their corrections for incompleteness.

Comparing to the 200 MHz counts, we see good agreement with Franzen et al. (Reference Franzen2016) for the entire flux density range, reaffirming the alignment with GLEAM as observed in Section 5.3. The comparison with Venville et al. (Reference Venville, Parkinson, Hurley-Walker, Galvin and Ross2024) shows good agreement down to ${\sim} 40$ mJy, below this level their source count slope becomes steeper. This is consistent with the lack of a completeness correction applied to their source counts, where the GLEAM-X completeness estimations exhibit a decrease in this range (Ross et al. Reference Ross2024). Additionally, the under-predicated flux densities in GLEAM-X for fainter sources as discussed in Sections 3.2.2 and 5.3 will have additional effects on the reported source counts at lower flux densities.

For the comparisons with the 144 MHz source counts by Shimwell et al. (Reference Shimwell2026), we see good agreeance for GAMA9 with a slight deviation at the bright end and good agreement with GAMA23 with a small deviation at flux densities $\lt $ 10 mJy/beam where we approach the observing limit of GAMA9. The comparisons with the ${\sim}$ 150 MHz source counts, exhibit agreement with Mandal et al. (Reference Mandal2021) down to ${\sim}$ 25 mJy. Williams et al. (Reference Williams, Intema and Röttgering2013) shows increasing scatter around these flux density levels, but we observe agreeance with their higher calculations. Below this point we see a deviation in the counts where several factors are likely at influence. Cosmic variance due to the smaller survey area may be a contributing factor. Additionally, the source counts are based on surveys with a higher resolution that likely resolves some sources into multiple fainter sources instead of a single brighter source observed in MIDAS. Finally, the MIDAS source catalogues were produced with AEGEAN where we classified each 2D Gaussian component as a radio source, while Williams et al. (Reference Williams, Intema and Röttgering2013) and Mandal et al. (Reference Mandal2021) utilise PYthon Blob Detector and Source FinderFootnote ^q (PyBDSF) that groups 2D Gaussians connected via continuous emission into a single source. Mandal et al. (Reference Mandal2021) then performed additional steps to deblend physically distinct radio sources that were classified as a single source and consolidate sources with multiple separate radio components (e.g. radio lobes) into a single source. These variations in source classification will result in a discrepancy in source counts between surveys.

The resulting source counts may be adequately described by third-degree polynomials. Fitting the polynomial

(9) \begin{equation}\begin{split}\log_{10} \left( S^{2.5} \frac{dN}{dS} \right) = a \log_{10}{(S)}^3 + b \log_{10}{(S)}^2 + c \log_{10}{(S)} + d\end{split}\end{equation}

to the GAMA9 and GAMA23 counts, where S is the 215.68 MHz flux density measured in Jy for the respective fields, results with the constants outlined in Table 11.

Table 11.

The resulting constants for a third-degree polynomial fitted to the GAMA9 and GAMA23 source counts.

Table 11. Long description

A table with two rows and five columns comparing the constants for a third-degree polynomial fitted to GAMA9 and GAMA23 source counts. The columns are labeled Field, a, b, c, and d. The row labels are GAMA9 and GAMA23. Row 1: Field, GAMA9; a, -0.0288 ± 0.0160; b, -0.256 ± 0.040; c, 0.267 ± 0.032; d, 3.33 ± 0.02. Row 2: Field, GAMA23; a, -0.0537 ± 0.0119; b, -0.283 ± 0.030; c, 0.300 ± 0.024; d, 3.36 ± 0.02.

5.7. Spectral properties of the sample

Sources within the GAMA23 MIDAS catalogue were matched with the 887.5 MHz deep field source catalogue produced from the ASKAP EMU Early Science project by Gürkan et al. (Reference Gürkan2022) and the GAMA spectroscopic redshift catalogue (Driver et al. Reference Driver2022). A maximum separation distance of 1^′ was used for the source matching, resulting in a total of 12 640 sources cross-matched to the EMU catalogue and 5 861 spectroscopic redshifts identified. We detected 846 sources in the GAMA23 MIDAS catalogue within the area imaged by EMU that did not have a corresponding source within 1^′ in the EMU catalogue. Most of these non-detected sources were due to the different source finding methods. Gürkan et al. (Reference Gürkan2022) used PyBSDF for their source finding and then visually inspected and consolidated radio galaxies with clearly distinct jets and cores into a single source. Our use of AEGEAN attempts to fit 2D Gaussians to the sources, resulting in the core and jets or multiple sections of a complex source being defined as separate sources.

Furthermore, many of the sources not detected in EMU are on the edge of the EMU mosaic in high RMS areas, including some sources on the image border, causing them to be cropped. However, there were 60 sources located in the lower RMS areas near the centre of the EMU mosaic that were not detected suggesting spectra with steep negative spectral indices, typical of older emission from remnant radio galaxies or galaxy cluster relics.

From the sources detected in both catalogues, we calculated their spectral indices and identified 35 sources with a spectral index $\alpha^{216}_{888} \lt -2$ and 5 sources with a spectral index $\alpha^{216}_{888} \gt 0.7$ . These are atypical of standard synchrotron emission from radio galaxies and could point towards interesting physics or objects. In Figure 12 we present the spectral distribution for the point sources calculated between the flux densities in the GAMA23 MIDAS survey at 215.68 MHz and the EMU survey at 887.5 MHz along with several of the non-detected sources. For the non-detected sources, the upper limit for the spectral index was calculated from EMU’s $5 \sigma$ source detection limit 0.19 mJy, aligning well with the slope observed for the detected sources in the negative spectral index region. The matched sources were filtered based on the EMU catalogue to those reported as a point source, while non-detected sources were selected from the GAMA23 MIDAS catalogue for a region $8^{\circ}$ by $4^{\circ}$ at the centre of the GAMA23 field to ensure the corresponding area within the EMU image had a low RMS. To lower the likelihood of false detections, we filtered out detected sources within 3^′ of sources with an integrated flux density $\gt 0.8$ Jy where noise can exceed $5\sigma$ . To filter out sources that may be classified differently between the catalogues, the pixel value at each source location was measured, along with the mean of the pixel values for a 20 pixel by 20 pixel region around it. We then selected the sources where these values were below the $5 \sigma$ EMU detection threshold. A brief visual inspection of the 60 sources was performed to verify they were likely genuine.

Figure 12.

The integrated flux density at 215.68 MHz as a function of the spectral indices $\alpha$ between 215.68 and 887.5 MHz for sources in the GAMA23 MIDAS catalogue matched to a source within 1^′ of the GAMA23 EMU catalogue (Gürkan et al. Reference Gürkan2022). The sources were filtered to unresolved sources along with a selection of sources not detected in EMU as described in the text. The cyan markers are the upper-limits on the spectral index for the selection of sources not detected within EMU. The crosses represent 3 examples of steep spectrum sources, selected to highlight the potential of the catalogue. Here we select one negative spectrum source that is detected in both MIDAS and EMU, one negative spectrum source not detected in EMU and one positive spectrum source.

Figure 12. Long description

Panel: A scatter plot with integrated flux density on the vertical axis and spectral indices on the horizontal axis. The vertical axis ranges from 10^-3 to 10^3 Jy, and the horizontal axis ranges from -3 to 2. The plot includes hundreds of data points, with a concentration of points around the center forming a dense cluster. Several data points are highlighted with different colored crosses, representing specific sources. Cyan markers indicate upper limits on the spectral index for sources not detected in EMU. The plot also includes a legend identifying the highlighted sources. The data points show a general trend where most points are concentrated in the middle, with fewer points scattered towards the edges. The plot aims to illustrate the relationship between integrated flux density and spectral indices for sources in the GAMA23 MIDAS catalogue.

Here, we provide a closer inspection for 3 example sources with extreme negative and positive spectral indices detailed in Table 12. From the 35 sources with a spectral index $\alpha^{216}_{888} \lt -2$ , we randomly selected a source that was detected in both surveys (MIDAS J225807-321424) along with one of the brightest sources in MIDAS that was not detected in EMU (MIDAS J225817-320052). From the 5 sources with a spectral index $\alpha^{216}_{888} \gt 0.7$ , we selected the source with the highest spectral index, MIDAS J224728-341906. These sources are detailed in Figure 13 using optical data overlaid with the observed radio contours. The optical data is Z-band imaging from the Visible and Infrared Survey Telescope for Astronomy (VISTA) Kilo-degree Infrared Galaxy (VIKING) survey (Findlay et al. Reference Findlay, Sutherland, Venemans, Reylé, Robin, Bonfield, Bruce and Jarvis2012) and the Dark Energy Spectroscopic Instrument Legacy Imaging Surveys (Dey et al. Reference Dey2019), while the radio contours are from the MIDAS mosaics, GLEAM-X, EMU Early Science and VLASS.

Table 12.

Details for the three example sources selected from the MIDAS catalogue with extreme spectral indices where 35 sources were identified with a spectral index $\alpha^{216}_{888} \lt -2$ and 5 sources with a spectral index $\alpha^{216}_{888} \gt 0.7$ . We selected a random source from the negative spectrum population that was detected in both MIDAS and EMU along with a source with one of the steepest negative spectrum that was not detected in EMU. From the positive spectrum population, we selected the source with the highest positive spectral index.

Table 12. Long description

A table comparing three example sources with extreme spectral indices from the MIDAS catalogue. The table has 12 rows and 10 columns. Column headers are: RA (deg), Dec (deg), alpha sub 888, Int flux density (mJy), Peak flux density (mJy/beam), Local RMS (mJy/beam), GLEAM-X source, EMU source, GAMA source, Redshift (Host). Row 1: MIDAS J225817-320052, 344.57462, -32.01452, < -2.7, 8.74, 3.53, 0.338, GLEAM-X J225817.9-320049, EMUJ225808.0-321353 (Mismatch), GAMAJ225806.74-321411.1, 0.317 (Host). Row 2: MIDAS J225807-321424, 344.52914, -32.24020, < -2.5, 6.94, 6.14, 0.344, GLEAM-X J225807.1-321424, EMUJ225808.0-321353 (Mismatch), GAMAJ225806.74-321411.1, 0.198. Row 3: MIDAS J224728-341906, 341.86905, -34.31851, 0.85 ± 0.07, 3.34, 3.19, 0.348, EMUJ224728.4-341905, GAMAJ224728.42-341904.8, 0.120.

Figure 13.

Postage stamp radio overlay images of three rare, interesting sources from the GAMA23 field along with their SEDs. The optical background data is Z-band imaging from VISTA. The contours are GAMA23 MIDAS 215.68 MHz (blue), EMU Early Science 887.5 MHz (sky blue), EMU Early Science 887.5 MHz convolved to the MIDAS resolution (orange) as described in ${\rm{\S}}$ 5.1, GLEAM-X wideband (170-231 MHz, red) and VLASS 3 GHz (purple). The contours start at the reported $5\sigma$ threshold with 50 even increments in log-space to $40\sigma$ for MIDAS J225817-320052 and MIDAS J225807-321424 and three even increments for MIDAS J224728-341906. The yellow crosshairs mark the locations of the sources detected in MIDAS. The SEDs contain the measured integrated flux densities for the sources in each respective survey with the exception of the EMUES values for MIDAS J225818-32005 and MIDAS J225806-32142; as these sources were not detected, we have used their $5\sigma$ detection threshold for the upper limit.

Figure 13. Long description

Panel A: A radio overlay image of MIDAS J225817-320052 with contours representing different radio surveys. The image shows right ascension on the x-axis and declination on the y-axis. The spectral energy distribution (SED) plot for MIDAS J225817-320052 displays flux density (mJy) on the y-axis and frequency (MHz) on the x-axis. The plot includes data points from EMU ES, GLEAM-X, and MIDAS surveys. Panel B: A radio overlay image of MIDAS J225807-321424 with similar contour representations. The SED plot for MIDAS J225807-321424 also shows flux density (mJy) on the y-axis and frequency (MHz) on the x-axis, with data points from EMU ES, GLEAM-X, and MIDAS surveys. Panel C: A radio overlay image of MIDAS J224728-341906 with contours. The SED plot for MIDAS J224728-341906 displays flux density (mJy) on the y-axis and frequency (MHz) on the x-axis, with data points from EMU ES, VLASS, and MIDAS surveys.

MIDAS J225817-320052 with an integrated flux density of $8.74 \pm 0.34$ mJy measured in MIDAS was not detected within the EMU mosaic, but is observed in GLEAM-X (GLEAM-X J225817.9-320049) with an integrated flux density of $12.08 \pm 1.74$  mJy in the wideband mosaic centred at 200 MHz. Although no source was detected at this source location within the EMU mosaic, diffuse radio emission is observed that is below the EMU detection threshold. The source in MIDAS is likely a radio lobe where a potential host (EMUJ225819.9-320156, GAMAJ225819.84-320154.0) is observed in the EMU mosaic south-southeast to the radio emission with a spectroscopic redshift of $z=0.317$ . Additionally, a second brighter source, which is likely the opposing lobe, is observed in MIDAS and GLEAM-X on the opposite side. When convolving EMU to the same beam size as MIDAS, only a large single source, which encompasses the area of the host and lobes, is observed with a peak between the potential host and the southern lobe; this is likely due to the uneven contribution of the radio lobes. The EMU source detected south of the lower lobe appears to be an unrelated background galaxy. Asymmetric double radio sources are unusual. Jet powers are likely similar from either pole of a black hole so the differing morphology is most likely due to environmental affects. For example, a denser medium on the southern side could keep a jet collimated longer meaning that we still see a hotspot. Or the southern lobe could be angled away from us so we are seeing it at a younger age than the northern lobe (which would be angled towards us).

MIDAS J225807-321424, randomly selected from the $\alpha^{216}_{888} \lt -2$ population with an integrated flux density of $6.94 \pm 0.34$ mJy in MIDAS matched to EMUJ225808.0-321353 with an integrated flux density of $0.205 \pm 0.034$ mJy resulting in a spectral index of $\alpha^{216}_{888}=-2.5 \pm 0.1$ . The MIDAS source position corresponds to the galaxy cluster GAMAJ225806.74-321411.1 (Robotham et al. Reference Robotham2011) and has a reported spectroscopic redshift of $z=0.198$ (Driver et al. Reference Driver2022). This source is also detected in the GLEAM-X wideband mosaic centred at 200 MHz with an integrated flux density of $8.37 \pm 0.97$ mJy. The detection within the 887.5 MHz EMU image occurs north-northeast of the cluster, which appears to be a separate source to the MIDAS detection and therefore is a mismatch resulting in an incorrect spectral index. Within the 215.68 MHz GAMA23 MIDAS mosaic, the detection occurs near the centre of the cluster, at this location in the convolved EMU mosaic, the radio emission is below the $5\sigma$ detection threshold of 0.190 mJy. Setting an upper limit for the source within EMU as $5\sigma$ , we estimate a spectral index of $\lt $ $-2.5$

From the sources with positive spectral indices $\alpha^{216}_{888} \gt 0.7$ , we selected MIDAS J224728-341906 which had the highest spectral index in the sample with $\alpha^{216}_{888} = 0.85 \pm 0.07$ . It was observed with an integrated flux density of $3.34 \pm 0.35$ mJy at 215.68 MHz within the GAMA23 MIDAS mosaic which is too faint to be detected in GLEAM-X. Within EMU at 887.5 MHz, it was detected with an integrated flux density of $11.2 \pm 0.1$ mJy and was detected in VLASS at 3 GHz with an integrated flux density of $19.24 \pm 0.13$ mJy (Gordon et al. Reference Gordon2021). The source position corresponds to a known galaxy with a spectroscopic redshift of $z = 0.120$ (Driver et al. Reference Driver2022), previously observed in radio (Wright et al. Reference Wright, Griffith, Hunt, Troup, Burke and Ekers1996; Mauch et al. Reference Mauch, Murphy, Buttery, Curran, Hunstead, Piestrzynski, Robertson and Sadler2013; Gordon et al. Reference Gordon2021), infrared (Cutri et al. Reference Cutri2013), optical (Maddox et al. Reference Maddox, Sutherland, Efstathiou and Loveday1990) and ultraviolet (Bianchi et al. Reference Bianchi, Herald, Efremova, Girardi, Zabot, Marigo, Conti and Shiao2011). The low-frequency MIDAS measurement may be used to help constrain the spectral peak and fit for this source. This source has no reported evidence of variability, so is likely a very young high-frequency peaker. Young radio galaxies peak due to synchrotron self-absorption and that peak moves to lower frequencies as the radio source expands. Future work could model the radio spectra to help determine the mechanism causing the the turn-over.