Description
Weak Lensing Cosmology Beyond Two Points: Propagating Systematics through Forward Modelling and Simulation-Based Inference
Weak gravitational lensing is a powerful probe of cosmic shear as it traces the large-scale structure of the Universe. The cosmic shear data from galaxy surveys is typically reduced to two-point correlation functions. These only contain a fraction of the constraining power on the cosmological model of the Universe present in the whole cosmic shear data measured by galaxy surveys. These surveys are limited by an assortment of instrumental, observational and astrophysical systematic effects which can be non-trivial to include in the modelling of the data and its uncertainty, especially, for higher-order statistics of the shear field. Such effects can be incorporated within accurate cosmological simulations to propagate them fully simulation-based inference (SBI).
In this talk, I will present my work on forward modelling systematics in weak lensing data for both ground- and space-based weak lensing surveys, such as the Kilo-Degree Survey (KiDS) and Euclid, respectively. I will show how the spatially varying selection of galaxies, their redshifts and their shapes can be forward-modelled at the level of the catalogue. Since the systematics are simulated at the same stage as they would enter the real data, their impact is naturally incorporated in the higher-order statistics of the cosmic shear field. At the same time, this approach also accounts for the correlations between the selection function and other contaminant signals such as variations in the point-spread function or the clustering of sources. The inclusion of such anisotropic and correlated systematics, while subdominant in the two-point correlation functions, becomes more significant in the higher-order correlations of the shear field. This biases the signal of n-point correlation functions and the shear field as a whole, while also severely impacting the accuracy of the uncertainty estimates for two-point statistics. I will demonstrate how biases in the cosmological inference can be avoided through sequential neural likelihood estimation SBI from forward simulations of the KiDS-1000 and KiDS-Legacy data. I will also highlight my ongoing work on forward modelling systematic effects in Euclid, the unique challenges of a space mission (such as radiation damage of the detector) and how simulation-based inference can address the challenges when conducting cosmological parameter inference with higher-order statistics of cosmic shear observables as measured by Euclid.
