Reconstruction of muon bundles in KM3NeT detectors using machine learning methods

TL;DR

This paper investigates reconstructing muon bundle properties in KM3NeT ARCA/ORCA using machine learning on data and simulations. It employs two muon-simulation pipelines, MUPAGE and CORSIKA, to generate training data, and trains LightGBM regressors to estimate muon-bundle energy, primary cosmic-ray energy, and muon multiplicity. Results show reliable energy reconstruction under favorable detector configurations, while primary-energy and high-multiplicity predictions are more challenging due to limited statistics and modeling gaps; comparisons with data expose data–MC discrepancies that motivate further improvements in shower modeling. The work demonstrates the feasibility of ML-based muon-bundle reconstruction in undersea neutrino telescopes and points to directions such as lower-level feature exploitation and graph-based architectures for full-detector track reconstruction.

Abstract

The KM3NeT Collaboration is installing the ARCA and ORCA neutrino detectors at the bottom of the Mediterranean Sea. The focus of ARCA is neutrino astronomy, while ORCA is optimised for neutrino oscillation studies. Both detectors are already operational in their intermediate states and collect valuable data, including the measurements of the muons produced by cosmic ray interactions in the atmosphere. This work explores the potential of machine learning models for the reconstruction of muon bundles, which are multi-muon events. For this, data collected with intermediate detector configurations of ARCA and ORCA was used in addition to simulated data from the envisaged final configurations of those detectors. Prediction of the total number of muons in a bundle as well as their total energy and even the energy of the primary cosmic ray is presented.

Figures (12)

• Figure 1: Energy spectra of various CR primary particles measured by several experiments. The "LHC" arrow indicates the kinetic energy achievable with the Large Hadron Collider (as of 2023). The figure was adapted from The_CR_spectrum.
• Figure 2: Pictorial summary of the design of KM3NeT neutrino telescopes. From left to right: the components of an optical module, the module integrated into a detection unit, an artist's impression of a detection unit and a building block.
• Figure 3: The processing chain for muon simulations in KM3NeT. The colour coding distinguishes three possibilities: CORSIKA simulation marked in purple, MUPAGE simulation in orange and experimental data indicated by the grey colour. The order of execution is from top to bottom, along the vertical arrow. At the trigger stage, both simulations and experimental data can be compared against each other.
• Figure 4: Different machine learning models in terms of achieved weighted $R^2$-score and weighted Pearson correlation coefficient $c$. The models were trained on a fraction (50k) of the ARCA115 training dataset because of the infeasibly long runtime and RAM memory required by some of the models.
• Figure 5: Feature importance computed over 10 random trials of removing a given feature. The colour coding groups features into clusters based on their inter-correlation.