Unraveling the complex interplay between soil characteristics and radon surface exhalation rates through machine learning models and multivariate analysis

Abstract

This research seeks to elucidate the intricate interplay between soil characteristics and the rates of radon surface exhalation rate. To achieve this aim, Light Gradient Boosting Machine (LightGBM) and eXtreme Gradient Boosting (XGBoost) machine learning (ML) algorithms are employed, supported by Multivariate Analysis (MA). An analysis was performed on a collection of soil samples, examining radon surface exhalation rates and other pertinent properties such as moisture content, particle size distributions, and the concentrations of Ra-226, Th-232, and K-40. The analysis revealed several key factors influencing radon exhalation rates, namely Ra-226 concentration, moisture content, and larger soil particles. To visualize the intricate relationships between these variables, contour plots of experimental and ML-generated data were created. These visual representations demonstrated that elevated soil moisture levels decrease radon exhalation rates. In contrast, higher concentrations of Ra-226 and a greater proportion of large soil particles led to an increase in exhalation rates. This endeavor presents these complex relationships in an accessible manner, furthering our understanding of the factors in radon surface exhalation.MA techniques, including Hierarchical Cluster Analysis (HCA) and Principal Component Analysis (PCA), were initially employed to investigate the complex interactions of soil attributes on radon exhalation. HCA identified three distinct clusters but faced limitations in detecting strong negative impacts. PCA successfully captured these inverse effects, indicating that the first two principal components accounted for approximately 80% of the total variance, primarily attributed to Ra-226 concentration, moisture content, and the percentage of large soil particles. However, neither technique could quantify the effects of soil attributes on radon exhalation rates. LightGBM outperformed XGBoost, but both successfully quantified the impacts of the studied soil characteristics on radon exhalation. Sensitivity analysis confirmed the robustness and accuracy of both models.This study highlights that XGBoost and LightGBM algorithms can effectively quantify radon exhalation rates based on soil characteristics, providing valuable insights for environmental policies, land use planning, and radon mitigation strategies.