Royal Observatory of Belgium - Meridian Room
Learning ionosphere inversion priors via Gaussian processes
The spatially and temporally varying electron density of the ionosphere causes complex distortions to passing radio wavefronts, becoming dominant at frequencies $\leq 1$ GHz. Using a probabilistic description of the system we apply Bayesian inference to study and derive the phase distortions of radio astronomical data in multiple directions. The relative improvement to image quality is studied using this solution. The inferred correlation structures will, in a subsequent study, provide priors for a general tomographic inverse problem, in which we model the components responsible for the phase distortions from first principles. The Bayesian inferred phase screens are completely dependent on the calibration process of the measured phase distortions, and no systematic biases can hope to be overcome. The phase screens inferred from the general inverse problem, being from first principles, are more free from bias. In this talk we will give a brief overview of Bayesian inference, and then focus on the data and results.