🔭 astrophysics

EMU: Cross-correlating EMU Pilot Survey 1 with Dark Energy Survey to validate the radio galaxy bias and redshift distribution

This paper validates a statistical method for determining the redshift distribution of radio galaxies by cross-correlating EMU Pilot Survey 1 data with Dark Energy Survey optical data, demonstrating that the recovered distribution aligns with state-of-the-art simulations and supports future cosmological analyses with large-scale radio surveys.

Original authors: Chandra Shekhar Saraf, David Parkinson, Jacobo Asorey, Catherine L. Hale, Benedict Bahr-Kalus, Maciej Bilicki, Stefano Camera, Andrew M. Hopkins, Konstantinos Tanidis

Published 2026-06-26

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Chandra Shekhar Saraf, David Parkinson, Jacobo Asorey, Catherine L. Hale, Benedict Bahr-Kalus, Maciej Bilicki, Stefano Camera, Andrew M. Hopkins, Konstantinos Tanidis

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Mapping the Invisible Universe

Imagine trying to map a city at night, but you can only see the streetlights from a distance. You can see where the lights are clustered (the neighborhoods), but you have no idea how far away each light actually is. Some are right next to you; others are miles away.

This is the problem astronomers face with radio galaxies. Radio telescopes (like the ASKAP telescope in Australia) are excellent at finding thousands of these "streetlights" across the sky. However, unlike optical telescopes that can sometimes tell you exactly how far away a star is, radio waves are very hard to measure for distance. Without knowing the distance (redshift), astronomers can't build a true 3D map of the universe; they only have a flat, 2D picture.

The Solution: The "Shadow" Trick

This paper describes a clever trick to figure out how far away these radio galaxies are without needing to measure each one individually.

Think of it like this: Imagine you are in a dark room with a wall covered in sticky notes (the radio galaxies). You can't see how far away the notes are. But, you have a friend standing next to the wall holding a flashlight with a known pattern of light (the Dark Energy Survey, or DES, which is an optical survey where distances are known).

If you shine the flashlight, the sticky notes that are close to the wall will cast shadows that align perfectly with the light pattern. The notes that are far away won't line up as well. By seeing how the "sticky notes" (radio galaxies) line up with the "flashlight pattern" (optical galaxies) at different depths, you can statistically figure out where the sticky notes are located in 3D space.

What They Did

The researchers took data from two major surveys:

EMU Pilot Survey 1: A radio map of the southern sky containing about 184,000 radio galaxies.
Dark Energy Survey (DES): An optical map of the same area containing millions of galaxies with known distances, split into six different "depth slices" (like layers of a cake).

They didn't try to match individual radio galaxies to individual optical galaxies. Instead, they looked at the statistical patterns. They asked: "Do the radio galaxies cluster together in the same way that the optical galaxies in 'Layer 1' cluster? What about 'Layer 2'?"

The Results

By running these patterns through a computer model, they successfully reconstructed the distance distribution of the radio galaxies.

The Shape: The resulting map of distances looked very similar to what the most advanced computer simulations (called TRECS) predicted.
The Peak: They found that the radio galaxies are most densely packed at a specific distance (redshift of about 0.85). Interestingly, this peak is slightly further away than some older simulations suggested, but it matches well with other deep-space surveys like MIGHTEE and COSMOS.
Validation: They checked their work by comparing it to a third map (the Planck CMB lensing map, which is like a map of the universe's gravity). The patterns matched up, confirming their method works.

Why This Matters

This paper proves that you don't need to measure the distance of every single radio galaxy to understand the universe's structure. You can use a "shadow" method—cross-referencing with a known map—to statistically infer the distances.

This is a crucial step for the future. As the Evolutionary Map of the Universe (EMU) survey expands to cover the whole sky, and as the massive Square Kilometre Array (SKA) comes online, this method will allow astronomers to turn flat radio maps into detailed 3D models of the cosmos, helping them understand how the universe grows and changes over time.

In a Nutshell

The authors showed that by comparing a radio map of "unknown distance" galaxies with an optical map of "known distance" galaxies, they could statistically deduce the 3D layout of the radio galaxies. Their method worked, producing a distance map that matches our best computer simulations and validates the tools we will use for the next generation of giant radio telescopes.

Technical Summary: EMU Cross-correlation with DES to Validate Radio Galaxy Bias and Redshift Distribution

Problem Statement
Radio continuum surveys offer a powerful method for mapping the large-scale structure (LSS) of the Universe due to their ability to cover large sky areas and their insensitivity to dust extinction. However, a critical limitation for cosmological analyses using these surveys is the lack of direct redshift information for radio sources. The dominant emission mechanism, synchrotron radiation, produces a featureless power-law spectrum, making redshift estimation extremely challenging. Without an accurate redshift distribution ($dN/dz$) and a robust model for galaxy bias, systematic errors can propagate into cosmological parameter constraints, such as $\sigma_8$ . While previous studies have relied on cross-matching with optical surveys (which yields photometric redshifts for only a fraction of sources) or simulations (e.g., TRECS, SKADS), there is a need for a method that utilizes the entire survey area to statistically infer the redshift distribution without relying on individual source identification.

Methodology
This study employs a cross-correlation technique between the two-dimensional radio continuum map from the Evolutionary Map of the Universe Pilot Survey 1 (EMU-PS1) and the photometric optical galaxy sample from the Dark Energy Survey (DES). The analysis proceeds as follows:

Data Preparation:
- EMU-PS1: Utilizes a catalogue generated by the PyBDSF source finder, containing $\sim$ 224,100 sources. The analysis applies flux density cuts at 180 $\mu$ Jy and 400 $\mu$ Jy to create conservative samples. Weight maps are constructed to account for observational noise and source count variations.
- DES: Uses the magnitude-limited (MagLim) galaxy sample divided into six tomographic redshift bins ( $0.20 \le z < 1.05$ ). Photometric redshifts are derived using the Directional Neighbourhood Fitting (DNF) algorithm.
- Planck: Includes the Planck CMB lensing convergence map ( $\kappa$ ) to provide an independent validation of the radio bias and redshift distribution.
Power Spectrum Estimation:
- Angular power spectra ( $C_\ell$ ) are computed using a pseudo- $C_\ell$ estimator (NaMaster) to handle incomplete sky coverage.
- The analysis measures the auto-power spectrum of EMU-PS1 ( $C^{EMU}_\ell$ ), the auto-power spectra of DES bins ( $C^{DES}_\ell$ ), the cross-power spectra between EMU-PS1 and DES bins ( $C^{EMU \times DES}_\ell$ ), and the cross-power spectrum between EMU-PS1 and Planck lensing ( $C^{EMU \times \kappa}_\ell$ ).
- A covariance matrix is estimated using 1,000 mock realisations generated via GLASS, incorporating the survey footprints and Poisson noise.
Modeling and Inference:
- Redshift Distribution ( $n(z)$ ): Two models are tested. First, the redshift distribution from the Tiered Radio Extragalactic Continuum Simulation (TRECS), which includes contributions from Active Galactic Nuclei (AGNs) and Star-Forming Galaxies (SFGs). Second, a parametric model (adapted from Nakoneczny et al. 2024) with three free parameters ( $z_0$ , $r$ , $\gamma$ ) designed to physically motivate the peak and tail of the distribution.
- Galaxy Bias: Linear, scale-independent bias models are adopted. For the parametric $n(z)$ , a constant amplitude bias model is used for SFGs and polynomial evolution for AGNs (or a combined effective bias).
- Parameter Estimation: A maximum likelihood framework (using PyMC) is employed to constrain the bias parameters and the redshift distribution parameters by fitting the joint data vector of the cross-power spectra.
Validation:
- The analysis pipeline is validated using mock catalogues where the true $n(z)$ and bias are known. This confirms the method's ability to recover the peak location of the redshift distribution.
- The impact of residual incompleteness in the radio catalogue is assessed, showing that while auto-power spectra are sensitive to incompleteness, the cross-power spectra used for inference remain robust.

Key Results

Detection Significance: The cross-correlation between EMU-PS1 and the six DES tomographic bins is detected with significances ranging from $3.5\sigma$ to $5.5\sigma$ in the multipole range $50 < \ell < 250$ .
DES Bias Validation: The galaxy bias estimated for DES bins using only clustering measurements is consistent with previous DES results (e.g., Abbott et al. 2022) within $1\sigma$ , validating the analysis pipeline.
Redshift Distribution Recovery:
- TRECS Model: The TRECS simulated $n(z)$ , when combined with a mixture model for galaxy bias, fits the cross-power spectra, the EMU-PS1 auto-power spectrum, and the cross-correlation with CMB lensing well. The total $\chi^2$ values are 7.78 (180 $\mu$ Jy) and 3.15 (400 $\mu$ Jy) for 20 degrees of freedom.
- Parametric Model: The parametric model (Eq. 11) also provides a good fit to the cross-power spectra without requiring prior knowledge of the AGN/SFG mixture. The best-fit distribution peaks at $z \sim 0.85$ . This peak is higher than that predicted by TRECS and LOFAR deep fields but is consistent with the COSMOS 3GHz and MIGHTEE surveys.
Bias Evolution: The recovered galaxy bias evolution for EMU-PS1 is consistent with both the TRECS predictions and independent estimates from the literature, even when using the parametric model that does not explicitly separate source populations.
Systematics: The study notes a discrepancy where the parametric model fails to fit the EMU-PS1 auto-power spectrum at the 180 $\mu$ Jy cut, suggesting residual systematics in the radio data (potentially incompleteness) that do not affect the cross-correlation results.

Significance and Claims
The authors claim that this work demonstrates the feasibility of estimating redshift distributions and galaxy bias for large-area radio continuum surveys directly from cross-correlations with optical surveys, without relying on individual source redshifts or simulations alone.

Validation of Method: The cross-correlation technique successfully recovers the shape of the redshift distribution and validates the bias model, providing a complementary approach to direct optical matching.
Cosmological Utility: The results validate the use of the parametric $n(z)$ model and the TRECS simulation for future cosmological analyses. This is significant for upcoming large-scale surveys like the full EMU survey and the Square Kilometre Array Observatory (SKAO), where direct redshift measurements for all sources will be impossible.
Robustness: The methodology is shown to be robust against internal systematics and residual incompleteness in the radio catalogue when using cross-power spectra, though the authors caution that the small survey area of the pilot ( $\sim$ 270 deg $^2$ ) limits the precision of the constraints compared to what will be achievable with the main survey.

The paper concludes that cross-correlation with optical surveys is a viable and necessary tool for the precision cosmology era of radio astronomy, enabling the mitigation of redshift distribution uncertainties that currently limit the scientific return of radio continuum data.