Enhancing the Sensitivity of RNO-G Using a Machine-learning Based Trigger

The Radio Neutrino Observatory in Greenland (RNO-G) is an array of radio detector stations which has been designed to study ultra-high energy (E ≳ 10¹⁸ eV) neutrinos. The experiment, when completed, will have the best sensitivity in this energy range and will yield a major advancement in our understanding of the sources and propagation of the highest energy cosmic rays. While RNO-G will be sensitive to primarily E ≳ 100 PeV neutrinos, the optical-based detectors only have a large enough exposure to study up to ∼ 1–10 PeV, leaving a gap in the energy range between the two detection methods. For RNO-G, the energy threshold is set by our ability to distinguish the Askaryan pulses, created from neutrino interactions, from the irreducible background of thermal noise. Using modern machine learning techniques, an online trigger can be implemented to identify small-amplitude pulses from in-ice cascades and thereby decrease the energy threshold of RNO-G. Such an advancement will increase the expected amount of observed neutrinos as well as close the gap between radio- and optical-based observatories. We present a convolutional neural network for classification of neutrino events that can be run as a second-stage trigger.