Potential of Radon Deficit as a Monitoring Tool in Organic Soil Remediation: A Machine Learning-Based Predictive Approach

The characterization and monitoring of soils and groundwater affected by non-aqueous phase liquids (NAPLs) remains a challenge due to the difficulty and high costs associated with their spatial delineation through intrusive methods (e.g., core-recovery drilling). The radon deficit technique is a promising screening method that enables the identification of potentially impacted areas based on the ubiquity of this gas, its operational simplicity and capability for rapid in situ measurement, and its preferential partitioning into NAPLs. However, subsurface sampling does not allow discrimination between impacts occurring in the vadose zone and those in the saturated zone. This work proposes the application of machine learning algorithms (Random Forest) as a tool to analyze the spatial variability of radon activity data in contaminated sites, with the aim of quantitatively determining their dependence on information related to contamination processes in both the vadose and saturated zones, as well as evaluating the ability of these algorithms to assess  the potential of the radon deficit technique for monitoring remediation processes in NAPL-impacted sites.
This study uses information collected during sampling campaigns conducted at a NAPL-impacted site at various depths within the vadose and saturated zones. The collected data (radon activity, lithological characteristics, and organic contamination information) were integrated into a machine learning algorithm that enabled the spatial analysis of the joint behavior of the variables, resulting in a predictive model to assess the potential of the radon deficit technique for monitoring remediation processes.
The results suggest that the radon deficit is a useful screening and monitoring method for NAPL-impacted sites, and demonstrate the value of machine learning not only as a predictive tool but also as an analytical resource to interpret complex relationships and validate indirect environmental monitoring techniques.