Selection of downgoing neutrinos and background characterization in the KM3NeT/ORCA telescope

The KM3NeT/ORCA neutrino telescope, located in the Mediterranean Sea, is optimized to study atmospheric neutrino oscillations and determine the neutrino mass ordering. Currently, the atmospheric neutrino flux is known with a systematic uncertainty of approximately 10‐20%. Precise measurement of the downgoing atmospheric neutrino flux is crucial to constrain this uncertainty, as downgoing neutrinos do not undergo matter‐induced oscillations before reaching the detector. However, selecting downgoing neutrinos is highly challenging due to an overwhelming background of downgoing atmospheric muons, resulting in an initial background‐to‐signal ratio of 10^5. This thesis develops a methodology to successfully isolate the downgoing atmospheric neutrino signal from the atmospheric muon background using a containment strategy. A virtual detector boundary, or mesh, is generated based on the physical positions of the outermost functional Digital Optical Modules (DOMs). To process the data, a highly automated and scalable Snakemake workflow is developed, processing raw offline hit‐level files to compute novel geometrical and light‐collection variables. The geometric variables include the particle track length inside the detector and the distance between the reconstructed starting point and the detector to evaluate the containment of the events. The probabilistic light‐collection variable is introduced to validate the reconstructed starting points of the events. The event selection is optimized in sequential stages based on data‐to‐Monte Carlo agreement. Initial pre‐cuts restrict the sample to well‐reconstructed events within an energy range of 3 to 100 GeV, requiring high track reconstruction quality and limiting the allowed number of triggered hits. By requiring a measurable track length within the instrumented volume and applying geometric containment criteria to ensure the reconstructed starting vertex lies strictly inside the detector boundary, the misreconstructed atmospheric muon background is significantly suppressed. The application of the novel probabilistic cut, which evaluates the absence of expected upstream light, successfully eliminates 99.9% of the challenging background category (muons mistakenly reconstructed as starting inside the detector) while preserving the vast majority — 82.8% — of the neutrino signal. This selection reduces the atmospheric muon event rate by several orders of magnitude, effectively overcoming the initial 105 background‐to‐signal ratio and achieving parity between the expected daily rates of background muons and signal neutrinos. Finally, an extrapolation to future, larger KM3NeT/ORCA geometries demonstrates that geometric containment efficiency will substantially improve as the detector expands, providing a promising foundation for future precision measurements.