Beschreibung
Wide-area surveys that measure the clustering and weak lensing of galaxies are some of our most powerful cosmological probes, with Stage IV projects like the Vera C. Rubin Observatory, Euclid, and Dark Energy Spectroscopic Intrument (DESI) set to enable high-precision cosmological measurements in the coming years. Getting the most out of these surveys will rely on careful modelling of the galaxies they observe, both as individuals and as a population. I will present pop-cosmos – a stellar population synthesis- (SPS-)based generative model trained on 26-band UV-MIR photometry from the COSMOS2020 catalogue – our framework for tackling this challenge. I will show how we can use pop-cosmos to generate realistic mock catalogues for photometric weak lensing surveys – using the Kilo Degree Survey (KiDS) as an example – enabling us to forward model tomographic redshift distributions without needing a spectroscopic reference sample. I will also demonstrate how our generative model can be used as a prior when performing SPS parameter inference for individual galaxies, providing accurate redshift estimation from broadband photometry, and enabling the selection of tailored subsamples of galaxies based on their physical properties (stellar mass, star-formation rate, etc.). I will show for KiDS-1000 how we can use our SPS parameter posteriors to exclude massive quenched galaxies – thought to be more susceptible to intrinsic alignments – from a weak lensing source catalogue more cleanly than is possible with a simple colour cut, and for DESI how we can define star-forming and quiescent subsamples of galaxies for multi-tracer studies. I will finally highlight our latest results from performing Bayesian SPS parameter inference at Stage III survey-scale for ~41 million galaxies from KiDS-Legacy.