Robust Photometry for Roman High-Latitude Imaging Survey Cosmology Using Roman and Rubin Imaging

Technical Summary: Robust Photometry for Roman High-Latitude Imaging Survey Cosmology Using Roman and Rubin Imaging

Problem Statement
Accurate and precise photometry is a prerequisite for the Nancy Grace Roman Space Telescope's High-Latitude Imaging Survey (HLIS) cosmological analyses, influencing photometric-redshift ($z_{phot}$) performance, sample selection, and tomographic binning. However, achieving robust photometry in Roman data presents unique challenges:

1. Source Blending: Roman's depth leads to significant source blending, which is exacerbated in joint analyses with ground-based surveys like the Rubin Observatory (LSST) due to differing spatial resolutions and Point Spread Functions (PSFs).
2. Correlated Noise: Roman's Teledyne H4RG detectors and the image coaddition process introduce correlated noise across pixels. Ignoring this leads to underestimated flux uncertainties and overconfident color measurements.
3. Astronomical Shot Noise: Unlike many ground-based observations where sky background dominates, the shot noise from galaxies themselves is a non-negligible component of the noise budget for Roman weak-lensing samples.
4. Environment-Dependent Biases: Standard single-object fitting can introduce environment-dependent color biases (up to $\sim0.5$ mag in crowded regions), degrading $z_{phot}$ performance.

Methodology
The authors develop and validate slimfarmer, a model-fitting photometry pipeline designed specifically for Roman HLIS cosmology. The pipeline builds upon the profile-fitting package The Farmer (based on The Tractor) and incorporates specific treatments for Roman data:

• Detection and Grouping: Sources are detected on Roman coadded images (Y106, J129, H158, F184, K213) using SEP. Neighboring galaxies are grouped using dilated segmentation maps to enable simultaneous fitting.
• Multi-Object Fitting (MOF): The pipeline performs joint multi-band model fitting using The Tractor. Within each group, all sources are fit simultaneously to account for blended light. The model selection follows a $\chi^2$-based decision tree (Point Source, SimpleGalaxy, ExpGalaxy, DevGalaxy, or FixedCompositeGalaxy).
• Forced Photometry: To enable joint Roman–Rubin analysis, the pipeline uses the Roman-derived galaxy models (positions and morphology) to perform forced photometry on matched Rubin (u, g, r, i, z, y) image simulations. This ensures consistent color measurements despite differences in PSF and pixel scale.
• Uncertainty Quantification:
  • Correlated Noise: The pipeline propagates correlated noise from Roman coadds into photometric uncertainties using noise realizations processed through the same pipeline as science images. The variance is estimated using the correlation function of the normalized noise field.
  • Astronomical Shot Noise: A model-based treatment is applied to the inverse-variance weight maps, updating the shot noise contribution at each fitting iteration based on the current model prediction, rather than using the noisy science image directly.
• Validation: The pipeline is tested on DC25 Roman coadded image simulations (based on OpenUniverse2024 truth) and matched Rubin simulations. Validation includes comparing recovered fluxes and colors against truth catalogs and assessing the impact of different fitting strategies (MOF vs. Single-Object Fitting [SOF]) and noise treatments on $z_{phot}$ performance.

Key Results

1. Photometric Accuracy: slimfarmer recovers galaxy colors with a mode of residuals within 20 millimag for Roman-based colors and 70 millimag for Rubin-based colors. Magnitude residuals show small biases at the faint end, likely due to blending with undetected sources.
2. Uncertainty Calibration: Properly accounting for correlated noise is critical. Standard The Tractor uncertainties (ignoring correlations) underestimate empirical scatter by up to a factor of $\sim3$. slimfarmer's uncertainty estimates align closely with empirical scatter for both magnitudes and colors.
3. Blending Mitigation: Multi-object fitting is essential for controlling environment-dependent biases.
   • MOF (Roman) + SOF (Rubin): Induces a blue bias in crowded regions (up to $\sim0.5$ mag) because Rubin fluxes are overestimated due to unmodeled neighbors.
   • SOF (Roman) + MOF (Rubin): Induces a red bias (up to $\sim0.2$ mag) because Roman fluxes include blended neighbor light.
   • Consistent MOF: Applying MOF to both surveys eliminates these environment-dependent color biases, maintaining stable performance down to nearest-neighbor separations of $\sim0.45''$.
4. Impact on Photometric Redshifts:
   • Ensemble $n(z)$: Joint Roman–Rubin photometry yields the best-separated tomographic redshift bins. Ignoring Rubin bands or neglecting correlated noise significantly increases bin overlap.
   • Individual $z_{phot}$: Using FlexZBoost, the Fiducial configuration (MOF in both surveys + correlated noise correction) shows minimal dependence of outlier fractions and scatter on galaxy environment. Configurations using SOF in either survey exhibit strong environmental trends, with performance degrading at larger nearest-neighbor separations.

Significance and Claims
The paper establishes slimfarmer as a prototype photometric pipeline for Roman HLIS cosmology applications. The authors claim that robust Roman–Rubin color measurements and reliable $z_{phot}$ characterization require three specific ingredients:

1. Correlated-noise corrections to accurately quantify uncertainties.
2. Model-based treatment of astronomical shot noise to avoid biases in weight maps.
3. Consistent multi-object fitting across both Roman and Rubin imaging to mitigate blending-induced systematics.

The authors explicitly state that while these results validate the photometric framework for galaxy selection and redshift characterization, they do not constitute a conclusion on the optimal treatment of blending for weak-lensing shear measurements, which requires separate validation within a shear-calibration framework (e.g., METADETECTION). Future work will extend this validation to more realistic simulations, real early-observation data, and the propagation of these systematics into full cosmological constraints.