Bayesian approach to signal estimation in gamma-ray astronomy with Gammapy

A fundamental challenge for observations with Imaging Atmospheric Cherenkov Telescopes is the treatment of the dominant background of cosmic-ray initiated air showers. Traditional methods for signal estimation rely on gamma-hadron separation cuts to remove a large fraction of background events, thereby reducing the efficiency of gamma-ray detection. In this work we adopt and extend a method known as Bayesian Analysis including Single-event Likelihoods (BASiL), in which these separation cuts are avoided by including into the likelihood function the probabilities of observed events originating from gamma or cosmic rays. With this approach, we retain 10 to 20% more signal throughout the analysis, particularly improving low signal-to-noise regimes and lowering the energy threshold. From the posterior probability of the source signal, credible intervals can be used to derive statistical uncertainties on flux points, while the computation of the Bayes factor allows the assessment of the probability of source detection. By adapting the open-source package Gammapy, we combine BASiL and the forward folding method to fit models and extract source spectra even from highly background-dominated data. This has potential applications to weak-source analyses or transient phenomena. We present results from simulated sources in the context of the H.E.S.S. telescopes, both in the 1-dimensional (data binned in energy) and a novel 3-dimensional analysis extension (data binned in energy and spatial coordinates), investigating how the method performs under distinct observation conditions and source characteristics.