Machine-learning correction of the local effects on neutron monitor and muon detector count rates at Syowa Station in the Antarctic

Abstract

Solar modulation of galactic cosmic rays around the solar minimum in 2019-2020 looks different in the secondary neutrons and muons observed at the ground. To compare the solar modulation of primary cosmic rays in detail, we must remove the possible seasonal variations caused by the atmosphere and surrounding environment. As such surrounding environment effects, we evaluate the snow cover effect on neutron count rate and the atmospheric temperature effect on muon count rate, both simultaneously observed at Syowa Station in the Antarctic (69.01 S, 39.59 E). A machine learning technique, Echo State Network (ESN), is applied to estimate both effects hidden in the observed time series of the count rate. We show that the ESN with the input of ERA5 reanalysis data (temperature time series at 1000, 700, 500, 300, 200, 100, 70, 50, 30, 20, and 10 hPa) at the closet position can be useful for both the temperature correction for muons and snow cover correction for neutrons. The corrected muon count rate starts decreasing in late 2019, earlier than the corrected neutron count rate, which starts decreasing in early 2020, possibly indicating the rigidity-dependent solar modulation in the heliosphere.