🔭 astrophysics

Infrared-enhanced Photometric Redshifts for the Dark Energy Survey Y6 Gold catalogue

This study demonstrates that combining Dark Energy Survey Year 6 optical data with Wide-field Infrared Survey Explorer (WISE) infrared measurements significantly improves photometric redshift accuracy across all metrics, particularly at higher redshifts, while adding VISTA Hemisphere Survey data or low signal-to-noise infrared data provides marginal or no further benefit.

Original authors: M. M. Puebla, I. Sevilla-Noarbe, J. de Vicente, L. Toribio San Cipriano, J. Carretero, A. Drlica-Wagner, J. García-Bellido, D. Gruen, J. Gschwend, T. A. Manning, M. de la Osa, A. Porredon, N. Reynes

Published 2026-07-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: M. M. Puebla, I. Sevilla-Noarbe, J. de Vicente, L. Toribio San Cipriano, J. Carretero, A. Drlica-Wagner, J. García-Bellido, D. Gruen, J. Gschwend, T. A. Manning, M. de la Osa, A. Porredon, N. Reynes, E. Sánchez, P. Tallada, N. Weaverdyck

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

Imagine you are trying to guess how far away a distant galaxy is just by looking at a photograph of it. This is a bit like trying to guess a person's age just by looking at a blurry black-and-white snapshot; you can make a guess, but you might be way off. In the world of astronomy, this "guess" is called a photometric redshift. The Dark Energy Survey (DES) has been taking incredibly detailed "snapshots" of 5,000 square degrees of the southern sky using five different colors of light (filters named g, r, i, z, and Y). These snapshots are like a high-quality black-and-white photo: they are sharp and clear, but they only show a limited range of colors.

The problem is that for very distant galaxies, the most important clues about their distance (specifically a feature called the "4000 Å break") shift out of the visible light range and into the infrared, which our eyes—and the DES camera—can't see well. It's like trying to identify a fruit by its shape alone when the color, which would give it away, is hidden in the dark.

To solve this, the researchers in this paper decided to combine the sharp DES photos with "night-vision" data from two other surveys: WISE (which sees mid-infrared light) and VHS (which sees near-infrared light). Think of this as taking the original photo and layering it with a thermal camera image and a near-infrared scanner to get a full 3D understanding of the object.

The Main Discovery: The Magic of Mid-Infrared
The team tested this idea using a smart computer program called DNF (Directional Neighbourhood Fitting), which acts like a super-smart librarian. Instead of guessing, the librarian looks at the "neighbors" of a galaxy in a giant database of galaxies with known distances and says, "If this galaxy looks like these ones, it must be about this far away."

When they fed the computer just the DES optical data, the guesses were okay, but they got messy for distant galaxies. However, when they added the WISE data (specifically the W1 and W2 bands), the results improved dramatically.

The Result: Adding WISE data reduced the "scatter" (how much the guesses bounced around), the "bias" (how much the guesses were consistently wrong), and the "outlier fraction" (the number of completely wild guesses).
The Numbers: For galaxies between redshift 1.0 and 1.2, adding WISE data dropped the outlier rate from a messy 8.4% down to a much cleaner 5.1%. For the furthest galaxies (redshift 1.2 to 1.5), the improvement was even more dramatic, cutting the outlier rate from a chaotic 44% down to 25%. The bias also improved by more than 50% in these high-redshift ranges.

What Was Ruled Out: The "More is Better" Myth
Here is where the story gets interesting. You might think that adding more infrared data would always be better, like adding more spices to a stew. The researchers tested this by adding the VHS data (which provides J, H, and Ks bands) to the mix.

The Verdict: The paper explicitly argues that adding VHS data does not provide a significant improvement over just using DES and WISE, especially for galaxies closer than redshift 1.5.
The Analogy: It's like having a perfect recipe that calls for salt and pepper (DES and WISE). Adding a third spice, like paprika (VHS), didn't make the dish taste any better; it just added a tiny bit of extra work. The WISE data had already captured the key infrared features needed to solve the puzzle.
The Limit: The paper also notes that if the infrared data is too faint (with a signal-to-noise ratio of less than 10), it doesn't help at all. It's like trying to read a book in the dark with a flashlight that's about to die; the extra light is just too weak to be useful.

How Sure Are They?
The authors are very confident in these findings because they didn't just simulate this on a computer; they measured it using real data from the DES Year 6 Gold catalogue, which contains hundreds of millions of galaxies. They cross-matched this with real spectroscopic data (where the distance is measured directly with a spectrum, acting as the "truth") to test their guesses.

They found that the improvements from WISE are statistically significant.
They compared their results to previous studies (like Banerji et al. 2015) and found that while earlier work saw modest gains with near-infrared data, their deeper, more uniform dataset confirms that mid-infrared (WISE) is the real game-changer for distant galaxies.

The Final Takeaway
The team has released a new, upgraded version of the DES Y6 Gold catalogue. This new version includes these improved redshift estimates, created by combining the sharp optical eyes of DES with the infrared vision of WISE. They decided not to include the VHS data in this specific release because it didn't add enough value to justify the extra complexity.

In short, for the vast majority of galaxies in this survey, the combination of DES and WISE is the "golden ticket" for accurate distance measurements. It's a reminder that sometimes, you don't need more data; you just need the right kind of data to see the universe clearly. This work helps pave the way for future giant surveys, like the Vera C. Rubin Observatory, which will need these precise tools to unlock the secrets of dark energy.

Technical Summary: Infrared-enhanced Photometric Redshifts for the Dark Energy Survey Y6 Gold catalogue

Problem Statement
The Dark Energy Survey (DES) provides deep, homogeneous optical imaging across five broad bands ( $g, r, i, z, Y$ ) over 5000 square degrees. While the Year 6 (Y6) Gold catalogue represents the most refined DES dataset to date, optical-only surveys suffer from degeneracies in galaxy Spectral Energy Distributions (SEDs), particularly at intermediate and high redshifts where key spectral features (e.g., the 4000 Å break) shift beyond the optical regime. These limitations hinder the precision of photometric redshift ( $photo\text{-}z$ ) estimates, which are critical for cosmological probes like galaxy clustering and weak gravitational lensing. While previous studies suggested that adding near-infrared (NIR) data could mitigate these issues, a comprehensive assessment using the full depth and uniformity of the DES Y6 Gold catalogue, combined with modern infrared datasets, was lacking.

Methodology
The authors investigated the impact of combining DES optical data with infrared measurements from two sources: the VISTA Hemisphere Survey (VHS, DR5) and the Wide-field Infrared Survey Explorer (WISE), specifically the 2013 AllWISE Data Release and the deeper, unofficial unWISE catalogue.

Data Matching: A positional sky cross-match was performed between the DES Y6 Gold spectroscopic training sample (545,796 galaxies) and the infrared catalogues. A matching radius of 1" was selected based on the median Point Spread Function (FWHM) of the respective surveys (ranging from ~0.87" to 1.13" for DES/VHS and ~6.1"–6.4" for WISE).
Algorithm: Photometric redshifts were estimated using the Directional Neighbourhood Fitting (DNF) algorithm, the reference estimator for the DES Y6 Gold catalogue. The analysis utilized the Angular Neighbourhood Fitting (ANF) metric, which defines neighbors based on the similarity of color ratios rather than absolute magnitude proximity.
Comparative Samples: The study generated and analyzed four distinct matched catalogues to isolate the contribution of different infrared bands:
- DES + AllWISE (bands: $g, r, i, z, Y, W1, W2$ )
- DES + VHS (bands: $g, r, i, z, Y, J, K_s$ )
- DES + VHS + AllWISE (bands: $g, r, i, z, Y, J, K_s, W1, W2$ )
- DES + unWISE (using forced photometry on DECaLS DR10 sources for deeper matches).
Metrics: Performance was evaluated using three standard metrics: bias ( $\mu$ ), scatter ( $\sigma_{68}$ ), and outlier fraction ( $f_{out}$ ), calculated against spectroscopic redshifts ( $z_{spec}$ ) across different redshift and magnitude bins.

Key Results

Impact of WISE Data: The addition of mid-infrared WISE data (W1 and W2 bands) resulted in statistically significant improvements across all metrics. Compared to optical-only data, the DES+WISE sample showed a reduction in bias by more than 50% and scatter by approximately 15% at higher redshifts ( $z > 1.1$ ). The outlier fraction was also significantly reduced, dropping to manageable levels (e.g., ~25% at $z \approx 1.2\text{--}1.5$ compared to higher rates in optical-only data).
Impact of VHS Data: The addition of VHS NIR bands ( $J, K_s$ ) provided only marginal improvements over the optical data alone within the explored depth limits. The study found that at $z < 1.5$ , the VHS bands did not offer further significant gains beyond what was achieved with DES+WISE. The authors note that the H-band was excluded from the main analysis due to limited sky coverage, but checks indicated it offered no further enhancement.
Depth and Signal-to-Noise: Improvements from infrared data were most pronounced for sources with sufficient signal-to-noise. For very faint sources where the infrared signal-to-noise dropped below 10, the infrared data did not contribute to improvements beyond the optical dataset.
Comparison with Previous Work: The results generally align with and improve upon earlier findings by Banerji et al. (2015), who used earlier DES data. While Banerji et al. observed modest improvements with VHS, this study confirms that WISE data provides the dominant contribution to accuracy, particularly at high redshifts. The training-based DNF approach used here achieved lower bias and scatter values at low-to-intermediate redshifts compared to template-based methods used in previous comparisons.

Key Contributions

Updated Catalogue Release: The authors have produced and released an updated version of the DES Y6 Gold catalogue, now enhanced with infrared photometry (specifically WISE W1 and W2 bands via unWISE forced photometry). This value-added catalogue is accessible via the CosmoHub platform.
Quantitative Assessment: The paper provides a rigorous quantification of the relative contributions of different infrared datasets (VHS vs. WISE) to photo-z performance using the full DES Y6 Gold depth.
Methodological Validation: The study validates the use of the DNF algorithm with combined optical and infrared data, demonstrating its robustness for future large-scale surveys.

Significance and Claims
The paper concludes that the combined use of DES optical data and WISE W1/W2 data significantly improves photometric redshift metrics, particularly at higher redshifts where optical data alone struggles with SED degeneracies. The authors claim that under the constraints of the DES and VHS depths explored, the WISE data already captures the key infrared features necessary for accurate photo-z estimation, rendering the addition of VHS data marginal for $z < 1.5$ .

The significance of this work lies in its provision of a reference catalogue for current and future cosmological analyses. By demonstrating that mid-infrared data (WISE) is the primary driver of photo-z improvement in this context, the study informs data combination strategies for upcoming surveys like the Vera C. Rubin Observatory's LSST and Euclid, which will rely on precise photo-zs for next-generation cosmological discoveries. The authors modestly note that while the metrics derived from the spectroscopic training set indicate improvement, they are not directly extrapolable to the full catalogue without accounting for color coverage differences, but the overall trend of improvement is clear.

More like this