Inhomogeneous Dust Biases Photometric Redshifts and Stellar Masses for LSST

Spectral energy distribution (SED) modeling is one of the main methods to estimate galaxy properties, such as photometric redshifts, z, and stellar masses, M_*, for extragalactic imaging surveys. SEDs are currently modeled as light from a composite stellar population attenuated by a geometrically homogeneous foreground dust screen. This is despite evidence from simulations and observations that find large spatial variations in dust attenuation due to the detailed geometry of stars and gas within galaxies. In this work, we examine the impact of this simplistic dust assumption on inferred z and M_* for Rubin LSST. We first construct synthetic LSST-like observations (ugrizy magnitudes) from the Numerical Investigation of Hundred Astrophysical Objects (NIHAO)-SKIRT catalog, which provides SEDs from high-resolution hydrodynamic simulations using 3D Monte Carlo radiative transfer. We then infer z and M_* from the synthetic observations using the PROVABGS Bayesian SED modeling framework. Overall, the homogeneous dust screen assumption biases both z and M_* in galaxies, consistently and significantly for galaxies with dust attenuation A_V ≳ 0.5, and likely below. The biases depend on the orientation in which the galaxies are observed. At z = 0.4, z is overestimated by ∼0.02 for face-on galaxies and M_* is underestimated by ∼0.4 dex for edge-on galaxies. The bias in photo-z is equivalent to the desired redshift precision level of the LSST “gold sample” and will be larger at higher redshifts. Our results underscore the need for SED models with additional flexibility in the dust parameterization to mitigate significant systematic biases in cosmological analyses with LSST.