ANALYSIS OF THE INFLUENCE OF NATURAL AND ANTHROPOGENIC FACTORS ON RADON FLUX DENSITY IN MOSCOW USING MACHINE LEARNING METHODS

During routine measurements of radon flux density for construction purposes in Moscow, the areas were found with abnormally high radon flux densities exceeding 400 mBq m-2 s-1. These values far exceed those average values around 24-40 mBq m-2 s-1 for local sandy and clay soils, which is compounded with the fact that the area under study doesn’t contain any active faults or natural soils rich in uranium. Therefore, the question arises, whether these high values are of technogenic or natural origin. This paper uses machine learning algorithms to find the answer to these questions. Machine learning algorithms including random forest trees and artificial neural networks were used to try and predict radon flux density anomalies on a city scale. Predictors used included maps of geodynamically active areas, lineaments, distances to heavy rail infrastructure such as metro tunnels and surface-level rail. Additionally, normal predictors of radon such as 226Ra concentration in soil, quaternary soil type and elevation were used for the predictions. Predictions were made for both anomaly-free and anomaly included datasets. Training data included radon flux data for Moscow with both anomalous and background values which included 931 data points, of which 112 was classified as anomalous (more than 400 mBq m-2 s-1). Based on the predictions obtained, factors which influence radon flux density and those that may produce anomalous values were underlined.