Event Reconstruction For Particle Detector Arrays Using Template-Based And Machine Learning Methods

The Universe hosts numerous, still largely unexplained, sources capable of producing particles with energies far beyond those achievable on Earth. Understanding these particles and their origins is crucial for shaping our view of the Universe and pushing the boundaries of high-energy physics. Gamma rays, as the most energetic photons, offer a unique perspective into this highest-energy regime, as their chargeless nature enables mostly unperturbed propagation through space, while their comparatively large interaction cross-section facilitates detection. At gamma-ray energies exceeding a few hundred GeV, ground-based instruments are essential for detection, with water Cherenkov detector (WCD) arrays especially effective in the TeV to PeV range. The Southern Wide-field Gamma-ray Observatory (SWGO) is a next-generation WCD experiment planned for construction in Chile, which aims to expand the view of WCDs to the Southern sky for the first time. As it is still in development, optimizing all aspects of the instrument, including the algorithms used for event reconstruction, is crucial. An accurate reconstruction of primary gamma-ray properties, such as energy, direction, and core position, is indispensable for ensuring excellent instrument performance and achieving SWGO's scientific goals of better understanding the high-energy Universe. In this work, two distinct approaches were developed to reconstruct the energy, core position, and arrival direction of gamma rays for the future SWGO. The first method employs a state-of-the-art template-based approach inspired by Joshi et al., 2019, which was adapted and further developed to serve as a standard approach for the energy and core reconstruction of SWGO. This enabled first-time performance comparisons between the SWGO candidate configurations. For the SWGO baseline configuration, the method achieves an energy resolution of up to 20%, exceeding current-generation instruments and the initial expectations of the experiment by up to 25% and 10%, respectively. A core resolution of roughly 3 m can be accomplished, enabling effective data quality cuts for the instrument. Furthermore, a novel template-based approach, which allows for a more realistic modeling of the detector response compared to the standard method, was proposed to reconstruct the arrival direction of gamma rays for WCDs. Under good Xmax reconstruction, it shows promise in improving the angular resolution above 100 TeV compared to the current standard algorithm. In addition to the template method, a novel machine learning approach was explored, utilizing Graph Neural Networks (GNNs) and Graphformer networks to improve the performance of the instrument further. The final GNN architecture was able to reconstruct the energy for the current SWGO array over the whole energy range, achieving either similar or better performance than the standard template-based approach, with resolutions close to 15%. The developed Graphformer networks were able to significantly improve the core and angular resolutions compared to their respective standard approaches. Improvements of up to 25% were observed across an energy range spanning roughly three decades, up to 100 TeV. At around 200 TeV, the method could achieve core resolutions below 3 m and angular resolutions of approximately 0.12 degree. While the Graphformer performance at hundreds of TeV is currently limited due to the lack of simulations, there is a strong potential for further performance improvements once more data is available. Both the template- and deep-learning-based algorithms promise an excellent event reconstruction capability for the future SWGO, indicating a strong point-source sensitivity and the ability to probe the highest-energy gamma rays produced across the Universe.