Extensive Air Showers Parameters Estimation Using Machine Learning Techniques with Simulations of the FAST Telescope

Abstract

We present the capabilities of the Fluorescence detector Array of Single-pixel Telescopes (FAST) observatory in the single telescope configuration as an important step towards the possible future large-field observatory for detecting ultra-high-energy cosmic rays. Reconstruction of main shower physics parameters are explored on noise-free simulated events using machine learning techniques in the challenging domain of a low-intensity transient signal. We find a very good correlation between the true and reconstructed energy of the shower even with the information from just the four photomultipliers of the single FAST telescope, and a reduced performance for the maximum of the shower development Xmax, using various architectures of artificial deep and convolutional neural networks, with a comparison to a benchmark gradient boost regression model. The resolution in the energy is found at the sub-percent level, while in Xmax it is~$5\%$. The relative difference between predicted and true values is under one percent for energy, while for Xmax it ranges from $-8\%$ to $+17\%$, which can be attributed to the limited information from the single FAST telescope configuration. The results constitute an important capabilities verification and a lesson learned with implications for established FAST prototypes as well as for more complex configurations of the FAST observatory under construction.