Application of machine learning methods to improve the radon deficit technique

The Radon deficit technique is a promising screening method for identifying and mapping potential subsurface organic pollution hotspots and thus, for the optimization of intrusive characterization campaigns. Radon (²²²Rn) a naturally procuded radionucleid and particularly suitable for use as a natural tracer due to its preferential partitioning with non aqueos phase liquids (NAPLs) and and ease of in situ analytical detection (Kram et al., 2001). The ability of the ²²²Rn technique to locate organic pollution hotspots and provide a semiquantitative analysis has been widely assessed in sites affected by NAPLs (De Miguel et al., 2018, De Miguel et al. 2020). However, the Radon measurement is affected by several confounding factors, such as variations in soil water saturation and ground-level temperature. Machine learning can be used to study and model these confounding factors and improve the interpretation of in situ radon analytical information.

Machine learning is a class of statistical techniques that have proven to be a powerful tool for modelling the behaviour of complex systems in which response quantities depend on assumed controls or predictors in a complicated way (Janik, 2018). The first purpose of this work is the application of machine learning to analyse sampled data of time series outdoor ²²²Rn. The algorithms "learn" from complete sections of multivariate series (containing measurements of soil water content, soil temperature and meteorological information), derive a dependence model. The model trained in this work can be used to improve the accuracy and reliability of the radon deficit technique, making it a more valuable tool for identifying and mapping subsurface contamination.

 

De Miguel, E., Barrio-Parra, F., Elío, J., Izquierdo-Díaz, M., Jerónimo, García-González, E., Mazadiego, L.F., Medina, R., 2018. Applicability of radon emanometry in lithologically discontinuous sites contaminated by organic chemicals. Environ. Sci. Pollut. Res. 25, 20255–20263. https://doi.org/10.1007/s11356-018-2372-9

De Miguel, E., Barrio-Parra, F., Izquierdo-díaz, M., Fernández, J., García-gonzález, J.E., 2020. Applicability and limitations of the radon-deficit technique for the preliminary assessment of sites contaminated with complex mixtures of organic chemicals: a blind field-test. Environ. Int. 138, 105591. https://doi.org/10.1016/j.envint.2020. 105591.

Janik, P. Bossew, O. Kurihara, 2018,Machine learning methods as a tool to analyse incomplete or irregularly sampled radon time series data, Scie. Tot. Environ.,630, 1155-1167, https://doi.org/10.1016/j.scitotenv.2018.02.233.

Schubert, M., 2015. Using radon as environmental tracer for the assessment of subsurface non-aqueous phase liquid (NAPL) contamination – a review. Eur. Phys. J. Spec. Top. 224, 717–730. https://doi.org/10.1140/epjst/e2015-02402-3.