Nowcasting for Improving Radon-222 Forecasting at Canfranc Underground Laboratory

Abstract

Physics’ rare event investigation, like the dark matter direct detection or the neutrinoless double beta decay research, is typically carried-out in low background facilities like the underground laboratories. Radon-222 ($^{222}Rn$) is a radionuclide that can be emitted by the uranium decay in the rock, thus the monitoring and the prediction of Rn contamination in the air of the laboratory is a key aspect to minimize the impact of this source of background. In the past, deep learning algorithms have been used to forecast the radon level, however, due to the noisy behavior of the $^{222}Rn$ data, it is very difficult to generate high-quality predictions of this time series. In this work, the meteorological information concurrent to the radon time series from four distant places has been considered—nowcasting technique—in order to improve the forecasting of $^{222}Rn$ in the Canfranc Underground Laboratory (Spain). With this work, we demonstrated and quantified the improvement in the prediction capability of a deep learning algorithm using nowcasting techniques.