GIS-based interpolation methods for spatial assessment of Geogenic Radon Potential

Modelling the spatial variability of Geogenic Radon Potential (GRP), based on spatially continuous geological, geographical and geochemical information as proxy data, is an important task to identify radon-prone areas and provide the local administration of a useful tool for land use planning and strategies aimed at radon health risk reduction. In this work, different interpolation techniques in a geographical information system (GIS) environment are applied and compared for estimating the spatial variation of GRP in the municipality of Celleno (Lazio, central Italy). The research activity has been conducted within the European LIFE-Respire project. Three regression models such as Ordinary Least Squares regression (OLS), Geographically Weighted Regression (GWR) and Empirical Bayesian Regression Kriging (EBRK) were applied to investigate the relationships between soil-gas radon concentrations and some proxy explanatory variables, and to generate spatial distribution of GRP. Ordinary Least Squares uses traditional regression techniques to model the relationships between a dependent variable and a set of independent variables (i.e. the explanatory variables). It provides a global model of the studied variable and creates a single regression equation to represent a process; this regression method assumes the relationship is constant over space, so the estimated coefficients of the regression are the same for all the study area. Geographically Weighted Regression is an extension of the traditional OLS regression, but it does assume that the relationships among the independent variables are not constant over space, then GWR calculates local regression coefficients and local r-squared values (R2) rather than global parameters. Finally, Empirical Bayesian Regression Kriging is a geostatistical interpolation that uses explanatory variables in raster format to improve predictions of the dependent variable. Regression models and semivariograms are estimated locally with simulations; and explanatory variables are transformed into principle components prior to modelling to solve multicollinearity problems. The regression models have been performed using the following proxy (i.e., explanatory) variables: the natural content of the radiogenic elements (Ra, U, Th, and K), the emanation coefficient of the outcropping rocks, the diffusive 222Rn flux from the soil, the soil-gas CO2 concentration, the Digital Terrain Model (DTM) and Topographic Position Index (TPI, a DTM-derived morphometric parameter), the permeability of the outcropping rocks (derived from the map of the hydrological complexes) and the gamma dose radiation of the shallow lithology. Soil-gas radon measurements were used as the response (i.e., dependent) variable of the applied regression models. Data has been organised in two subsets (training and test data) to be used in the validation process. Results from validation technique indicate that GWR provides a local model with a better performance (adjusted R2=0.882) than the global OLS model (adjusted R2=0.573). However, the application of the EBRK will result in the best model validation (R2=0.989) vs the validation of the GWR result (R2=0.863). Research was conducted and funded within two research projects: INAIL/CNR-IGAG (P19L06) and LIFE-Respire (LIFE16 ENV/IT/000553).