Machine learning as a tool for analysing the impact of environmental parameters on the radon exhalation rate from soil

Highlights

• Demonstration of the effect of rain with time-lag on radon exhalation rate values.

• Use of machine learning methods to determine the most important parameters for the radon exhalation rate.

• Confirmation of the strong relationship between radon and CO₂.

Abstract

Interest in radon (Rn) is not limited only to its impact on health and its dose to the public, but due to its properties, the techniques to analyse its behavior can be used in many fields such as radiotherapy, atmospheric physics, geophysics, geohazards, mineral exploration, and even planetary science.

Nowadays machine learning methods provide extremely important tools for intelligent environmental data analysis, processing and visualization.

We describe application of machine learning to environmental sciences with an emphasis on the radon exhalation rate in order to express responses from multivariable time-series data collected at a measuring site near the Sakurajima volcano (Kagoshima, Japan).

Introduction

The earth's crust contains small amounts of the primordial radionuclides uranium (²³⁸$\text{U}$) and thorium (²³²$\mathrm{T}\mathrm{h}$). These nuclides undergo a series of decays until a stable isotope of lead is reached. Decay products are usually isotopes of solid elements; however, in these series, two decay products are noble gases: radon (²²²$\mathrm{R}\mathrm{n}$) in the uranium series and thoron (²²⁰$\mathrm{R}\mathrm{n}$) in the thorium series. Radon concentrations in soil pores are dependent upon the radium content of the soil, emanating power for radium, and soil moisture content. Radon thus formed is transported through the porous medium (soil) caused by diffusion and advection processes and finally, it can be exhaled into the air.

Three main physical processes are involved in the generation and transfer of Rn from mineral grains to soil gas and in the movements of Rn-bearing soil air: 1) emanation related to produced Rn in soil, i.e. the Rn atoms escape from the grains in the interstitial space between them; 2) Rn transport, caused by diffusion and convection; and 3) exhalation, i.e. the Rn atoms transported to the soil surface are exhaled into the atmosphere.

The gases measured in soil and those exhaled from it can, however, be strongly altered by environmental conditions, such as atmospheric pressure, soil temperature, or moisture. Accurate knowledge of the influence of environmental conditions is required to decipher information from deeper phenomena in the earth.

Generally, Rn produced in subsoil contributes to surface radioactivity if upward Rn transport by groundwater and gas diffusion occurs in the unsaturated zone and by fluid convective movement due to the geothermal gradient, also in the same zone. A strong Rn signal can be observed only if a convection or an advection process occurs. Rn activities increase with the increase in the flow rate of the soil-gas, as the increased flow rate increases gas velocity, which gives ²²²$\mathrm{Rn}$ less time to decay (${\mathrm{T}}_{1/2}$=3.825 d) and allows more extraction from the fissure walls but for higher flows, dilution of Rn by carrier fluids and gases may occur. CO₂ is a well-defined carrier gas for noble gases such as Rn and helium, which are unable to reach the surface due to their low mobilities and very small quantities, respectively (Walia et al. (2009)). Rn that reached the surface is exhaled (defined as exhalation rate and refereed to in this article as $RnE$) and its speed is related to specific activity of radium in soil, the emanation fraction, the dry soil bulk density and temperature.

Relatively little has been published on continuous $RnE$ monitoring and multivariation analysis because of measurement difficulties due to changes of environmental conditions. A study conducted by Kojima (1998) demonstrated that $RnE$ does not follow changes of atmospheric pressure. However, his work showed correlations of $RnE$ with wind velocity and with the pressure difference between the surface and at 100 cm depth (decreasing the pressure difference raises $RnE$). Kojima concluded that the seasonal variation of $RnE$ is determined by a balance between the pressure difference corresponding to the force for vertical flow of soil-gas and the soil moisture content affecting the air permeability in soil.

More complex investigations using automatic measurement stations with multiparameters analysis were presented by Mazur and Kozak (2014) and Yang et al. (2019). Results from both investigations confirmed the complexity of the radon exhalation process and its dependence on many factors, both environmental and meteorological. Mazur and Kozak showed that the highest value of $RnE$ can be observed for the high soil temperature with no significant dependency on the precipitation. In the multivariate model presented by Yang et al. the temperatures of soil and air are dominant, however opposite tendencies were found. Moreover, they concluded that dependencies of $RnE$ are more complicated than those described by the theory of diffusive radon flux from soil.

In our study, the primary goals were to create a model and test the potential of machine learning methods, particularly the gradient boosting machine (GBM) algorithm, to analyse the influence of several measured variables and to overcome some of the difficulties in measurement of radon exhalation rate from the soil surface.