Machine-learning method for reconstructing near-Earth low-energy cosmic ray fluxes from ground-based SOPO neutron monitor data

Nowadays, there are a large number of ground-based neutron monitors. These detectors measure neutron fluxes produced by the interaction of primary cosmic rays (CR) with Earth’s atmosphere. Since their invention in the 1950s, they have become very popular for studying variations of CR intensity on different time scales. These detection systems are stable and therefore can conduct continuous measurements for decades. However, they exhibit low sensitivity to cosmic ray fluxes with energies ≤1 GeV, whereas heliospheric variations of these fluxes are most pronounced at lower energies. The rapid growth of the space industry has made it possible to perform direct measurements of CR fluxes by a variety of scientific instruments installed on board spacecraft. Satellite-borne detectors measure cosmic ray fluxes across a wide energy range, from tens of keV up to tens of TeV. Nonetheless, experimental data collected by spacecraft is not always immediately accessible and is often published with a significant delay. Moreover, the radiation in outer space is gradually reducing the efficiency and stability of satellite systems, so their operating time in orbit is limited. Ground-based detectors are less susceptible to radiation damage, and their data is always available for scientific research. Therefore, developing methods to reconstruct top-of-atmosphere CR fluxes from ground-based detector data is increasingly urgent. This paper presents the method for calibrating the high-latitude SOPO neutron monitor by machine-learning algorithms, using data on cosmic proton fluxes with energy >500 MeV from the P11 channel of the proton sensor installed aboard the GOES-16 satellite. The machine model obtained as the result of training was employed to reconstruct the low-energy cosmic ray flux from SOPO data. Then it was compared with actual data from the GOES-16 proton sensor.