The Italian radon risk map

Radon (²²Rn) is a naturally occurring radioactive gas that occurs in rocks and soils, and its migration pathways are influenced by geological faults. These processes can significantly increase radon leakage into buildings, posing a significant health risk. Classified as a carcinogen by the World Health Organisation, exposure to radon has required the establishment of national reference levels across Europe under Directive 2013/59/EURATOM and the identification of Radon Priority Areas (RPAs) to guide remediation initiatives. This legislation emphasises the need for both collective and individual risk management, using advanced radon risk assessment tools.

In this study, we present an innovative approach to construct a geogenic radon hazard index (GRHI) map for Italy using a robust bottom-up methodology. Our approach integrates several geological proxies related to radon source (e.g. geology, radionuclide content) and migration pathways (e.g. faults) using supervised auto-machine learning (Autogluon). A dataset of approximately 30,000 soil radon measurements was divided into training and test datasets. A conceptual model with ten predictors was developed to estimate soil radon concentrations at unsampled locations on a 1x1 km grid. The LightGBMLarge algorithm resulted in the best model (R²_test = 0.524) which was validated by a combination of statistical metrics. The SHAP analysis highlighted the relative importance of the predictors in the model.

The GRHI map was further combined with census section data (ISTAT database) and population density to produce a risk map from Collective Risk Areas (CRA) to Individual Risk Areas (IRA). This final map serves as a valuable tool for national and regional administrations to identify IRAs in accordance with Directive 2013/59/EURATOM (Article 103).

This research addresses the lack of a standardised European methodology for radon risk assessment. It provides a comprehensive framework to bridge the gap between collective and individual risk. Through the integration of geological knowledge with machine learning and demographic data, this work provides useful information for the improvement of radiation protection and public health strategies.