Inferring radio beams from sparse sampling patterns: a case study with the Murchison Widefield Array

Accurate primary-beam models are essential for high-dynamic-range radio experiments, particularly searches for the redshifted 21-cm signal from the Epoch of Reionisation, where beam errors can couple bright foreground emission into the faint cosmological measurement. Using reprocessed satellite observations we infer the 137-MHz, zenith-pointed power beams of Murchison Widefield Array receiving elements. We develop two complementary Bayesian approaches for quantifying departures from nominal electromagnetic simulations: a physically motivated model of complex variations among individual dipoles and a hierarchical spline model for flexible direction-dependent structure. The measurements reveal shallower and displaced nulls together with coherent sidelobe deformations. The dipole model captures much of this broad structure, whereas the spline recovers additional localised features while remaining largely robust to isolated pass-shaped artefacts. Despite their sparse and distinct sampling patterns, four independent 20-day subsets, each containing approximately 20% of the full dataset, recover consistent null and sidelobe morphology across the visible hemisphere. These results demonstrate that opportunistic satellite transmissions can support practical, repeatable beam reconstruction, while motivating future chromatic, phase-referenced measurements of the full Jones response.