Cosmology is undergoing a data revolution. Surveys such as the imminent Legacy Survey of Space and Time (LSST) to be conducted by the Vera C. Rubin Observatory will deliver huge galaxy catalogues that provide critical tools for understanding the nature of dark matter and dark energy. However, in order to obtain accurate cosmological constraints from these enormous datasets, we need reliable ways of estimating galaxy properties using only photometry. I will present pop-cosmos: a forward modelling framework for photometric galaxy survey data, where galaxies are modelled as draws from a population prior distribution over redshift, mass, dust properties, metallicity, and star formation history. These properties are mapped to photometry using an emulator for stellar population synthesis, followed by the application of a learned model for a survey's noise properties. Application of selection cuts enables the generation of mock galaxy catalogues. This enables us to use simulation-based inference to solve the inverse problem of calibrating the population-level prior on a deep multiwavelength catalogue, COSMOS2020. We use a diffusion model as a flexible population-level prior, and optimise its parameters by minimising the Wasserstein distance between forward-simulated photometry and the real COSMOS2020 survey data. The resulting model can then be used to derive accurate redshift distributions for upcoming photometric surveys, to facilitate weak lensing and clustering science. I will show applications of this framework, demonstrating how we are able to extract redshift distributions, and make inferences about galaxy evolution. I will also discuss the use of pop-cosmos as a prior for performing inference on individual galaxies in a highly scaleable manner, as well as ongoing work to analyse data from the Kilo-Degree Survey (KiDS) in preparation for LSST.
Niayesh Afshordi, Selim Hotinli