Speaker
Description
The large-scale structure (LSS) of the universe has become an important observational probe for cosmology. Its structure and growth over cosmic time can yield insight into dark matter, dark energy, and the behaviour of gravity on the largest scales. Using neutral hydrogen (HI) as a tracer of dark matter, the LSS can be mapped out as a function of redshift in a technique known as HI intensity mapping. Intensity mapping measures the integrated emission for a given volume and sky area as opposed to resolving each individual galaxy. This technique presents an efficient way to map out large volumes of the sky in a short amount of time but suffers greatly from foreground contamination. Left as is, foreground contamination leads to biased cosmological results. To this end, innovative foreground removal methods are needed, and the focus of this research is one such method known as Bayesian component separation. In this work, we formulate physically motivated models for the HI signal, foregrounds, and their covariances. We implement Gibbs sampling combined with Gaussian-constrained realisations to sample from the posterior distribution for L-band single-dish data from the MeerKAT telescope. This method is able to constrain millions of free parameters, allowing us to reconstruct the HI signal and foregrounds separately and recover the 1-dimensional HI power spectrum. This framework also enables the investigation of how observational systematics influence HI signal reconstruction and power-spectrum estimation. A thorough understanding of foreground contamination and observational systematics is essential for distinguishing potential signatures of new physics from spurious effects introduced by the data or its analysis.