Improvements to monoscopic analysis for imaging atmospheric Cherenkov telescopes: Application to H.E.S.S

TL;DR

This paper introduces advanced analysis techniques, including machine learning and new image parameters, to enhance monoscopic event reconstruction in IACTs, significantly improving angular resolution, lowering energy thresholds, and boosting sensitivity for gamma-ray detection.

Contribution

The paper presents novel improvements in monoscopic analysis for IACTs, notably a machine-learning orientation algorithm, intensity-dependent cuts, and new image parameters, enhancing performance at low energies.

Findings

57% improvement in angular resolution at 100 GeV

50% reduction in low-energy threshold for source analysis

41% increase in sensitivity at low-energy threshold

Abstract

Imaging atmospheric Cherenkov telescopes (IACTs) detect gamma rays by measuring the Cherenkov light emitted by secondary particles in the air shower when the gamma rays hit the atmosphere. At low energies, the limited amount of Cherenkov light produced typically implies that the event is registered by one IACT only. Such events are called monoscopic events, and their analysis is particularly difficult. Challenges include the reconstruction of the event's arrival direction, energy, and the rejection of background events. Here, we present a set of improvements, including a machine-learning algorithm to determine the correct orientation of the image, an intensity-dependent selection cut that ensures optimal performance, and a collection of new image parameters. To quantify these improvements, we use the central telescope of the H.E.S.S. IACT array. Knowing the correct image orientation, which corresponds to the arrival direction of the photon in the camera frame, is especially important for the angular reconstruction, which could be improved in resolution by 57% at 100 GeV. The event selection cut, which now depends on the total measured intensity of the events, leads to a reduction of the low-energy threshold for source analyses by ~50%. The new image parameters characterize the intensity and time distribution within the recorded images and complement the traditionally used Hillas parameters in the machine learning algorithms. We evaluate their importance to the algorithms in a systematic approach and carefully evaluate associated systematic uncertainties. We find that including subsets of the new variables in machine-learning algorithms improves the reconstruction and background rejection, resulting in a sensitivity improved by 41% at the low-energy threshold.