Abstract

Accurate redshift calibration and physically motivated source sample selection are central challenges for precision weak lensing cosmology. In this talk, I will present our recent work on forward modelling redshift distributions and inferring physical properties of weak lensing galaxies in the Kilo-Degree Survey (KiDS-1000) using a generative model for the galaxy population. The framework enabling this work is pop-cosmos, a calibrated galaxy population model that allows principled Bayesian inference of individual galaxy redshifts and physical properties for millions of KiDS-1000 sources. Validation against spectroscopic samples demonstrates low bias and scatter in the inferred photometric redshifts, while physical property inference enables the construction of weak lensing catalogues that mitigate intrinsic alignment systematics. We also develop a forward-modelling framework to infer the redshift distributions of the KiDS-1000 galaxies. By applying a KiDS data and survey selection model to synthetic photometric data of mock pop-cosmos galaxies, we directly characterize the redshift distributions in each of the five tomographic bins of KiDS-1000, thereby bypassing the need for spectroscopic reweighting used in conventional redshift calibration methods. Overall, our results demonstrate how the pop-cosmos galaxy population model can deliver accurate redshift distributions as well as galaxy properties, crucial for connecting galaxy evolution physics and weak lensing cosmology with Rubin LSST and Euclid.