Cosmic-ray energy reconstruction using machine learning techniques

TL;DR

This paper introduces machine learning-based methods, specifically neural networks, to improve the accuracy of cosmic-ray energy estimation in the HAWC observatory, enhancing previous likelihood-based techniques.

Contribution

The paper develops and evaluates neural network models for cosmic-ray energy reconstruction, updating existing estimators with modern machine learning approaches.

Findings

Neural network models outperform traditional estimators in accuracy.

Updated energy estimators improve cosmic ray spectrum measurements.

Models demonstrate robustness across different air shower observables.

Abstract

HAWC is a ground-based observatory consisting of 300 water Cherenkov detectors, which observes the extensive air showers induced by cosmic rays from some TeV to a few PeV and, in particular, gamma rays from 300 GeV to more than 100 TeV. One of the crucial features required for a detector of extensive air showers is the estimation of the primary energy of the events to study the spectra of cosmic and gamma rays. For HAWC there are currently two gamma-ray energy estimators: one relies on a ground density parameter, while the other utilizes an artificial neural network. For the cosmic ray energy estimation, there is only one estimator based on maximum likelihood procedures and measurements of the lateral charge distribution of the events. It is worthwhile to update the cosmic-ray energy estimator due to recent improvements of the extensive air shower offline-reconstruction techniques in HAWC. Therefore, we implemented an artificial neural network to reconstruct the primary energy of hadronic events trained with several observables that characterize the air showers. We trained several models and evaluated their performance against the existing cosmic ray energy estimator. In this work, we present the features and performance of these models.