Speaker
Description
HI Intensity Mapping (IM) surveys with the telescopes (and precursors) of the SKA Observatory are a promising, independent, and insightful way of studying the large-scale of the Universe. However, strong astrophysical foregrounds contaminate the signal, and their coupling with instrumental systematics further increases the complexity of the data cleaning. As an effort of the HI IM Focus Group of the SKAO Cosmology Science Working Group, we have performed the first Foreground Subtraction Blind Challenge for HI single dish IM.
We asked: if we were given some actual data today, what could we achieve using the available pipelines? Nine different cleaning pipelines joined the Challenge: based on statistical learning techniques (PCA, ICA, GMCA), on a new hybrid algorithm (mixGMCA), on a blind polynomial fitting method, and a more astrophysical-informed parametric fit to foregrounds. We performed the cleaning on realistic data cubes for both a MeerKAT and a SKAO1-MID set-up, with no knowledge of the ground truth or the beam's full shape. Here, I discuss the pipelines that joined the Challenge and compare their cleaned maps against the input maps. We found a large scatter in the results among methods, observational set-ups, and pre-processing steps. We particularly stress how mixGMCA, devised ad-hoc for IM, outperforms the methods from which it stems. Blind Challenges are an excellent tool for testing methods and best practices, and learning as a community.
| Reasearch area | Cosmology |
|---|
