Gamma-ray astronomy: GammaLearn’s deep learning put to the test using real data from the LST-1 telescope

On the first large Cherenkov telescope at the future CTAO observatory on La Palma, a neural network analyses the complete image of atmospheric showers, whereas standard analysis relies on just a few parameters.

The Cherenkov Telescope Array Observatory (CTAO) is expected to offer up to ten times the sensitivity of current Cherenkov telescopes. Its first prototype, the Large-Sized Telescope (LST-1), is already operating at the Roque de los Muchachos Observatory on the island of La Palma.

These instruments observe the brief glow produced when a gamma ray hits the atmosphere. Classical analysis summarizes each image by a few shape parameters; GammaLearn, a deep learning framework designed for this data, instead exploits all the recorded time and charge information to estimate the energy, direction of arrival and nature of the incident particle. The expected gain is mainly at low energies, around 20 GeV, which LST-1 is able to detect.

The study goes beyond the simulation stage: the authors apply GammaLearn to real observations of LST-1 and compare its sensitivity curves to those of standard analysis. This is an important challenge for the study of distant extragalactic sources such as the active nuclei of galaxies, used to test fundamental physics and cosmology.