Taming Additive Systematics via Redshift-Bin-Optimized Star-Galaxy Separation

Abstract

Contamination from stars in the galaxy samples of large-scale structure surveys can bias cosmological constraints if not tightly controlled. This is especially true for lens samples used for galaxy clustering and galaxy-galaxy lensing probes, where contamination is a primary source of additive systematics. We propose an improved approach to star-galaxy separation and an optimal weighting scheme to jointly mitigate additive and multiplicative contamination of the density field at the map level. Our star-galaxy separation approach exploits the fact that photometric galaxy samples used for cosmological inference populate different regions of color-space than the full photometric dataset on which star-galaxy cuts are typically applied, and therefore optimizes star-galaxy separation for the galaxy samples in each redshift bin. This serves as a complementary approach to morphological star-galaxy separators, which can have complicated dependencies on PSF and blending systematics. We demonstrate the method using the Dark Energy Survey Y3 MagLim lens sample, for which we obtain forced NIR unWISE photometry via cross-matching with DECaLS DR9 to define redshift-bin-optimized color cuts. We identify and remove residual stellar contamination in the DES Y3 lens sample, which varies strongly across redshift bins ($1.3-5.5\%$) and across the footprint.