R
Rosie
Text
B-GSM: Bayesian Simultaneous Component Separation and Calibration of the Low-Frequency Sky
A data-driven Bayesian model of the diffuse radio sky below 400 MHz, built to remove the bright Galactic foregrounds that swamp the cosmological 21-cm signal. Rather than the principal-component fits of earlier global sky models, B-GSM uses nested sampling to infer full posterior distributions for the spectral behaviour and spatial amplitudes of the emission components, and Bayesian evidence to select how many components and which spectral parametrisation the data support. Crucially it conditions on both diffuse-emission maps and absolute-temperature datasets, so component separation and calibration are solved simultaneously with rigorous error quantification; the paper validates the framework on a synthetic sky with realistic partial coverage, thermal noise, and calibration uncertainty, recovering two curved-power-law components and sharply rejecting incorrect models.