AUTOMATED GEOGRAPHIC INFORMATION SYSTEM FOR ASSESSING RADON HAZARD IN AN URBANIZED AREA

he paper presents the development of a geoinformation system for assessing the potential radon hazard of residential buildings, based on a hybrid approach that integrates gradient boosting (for data detrending, taking into account anthropogenic factors such as the year of construction, foundation type, and material porosity) and ordinary kriging (for spatial estimation of residuals). After the pre-processing stage, which included outlier removal and data transformation, the application of gradient boosting allowed for the elimination non-stationarity and the explanation of 51.52 % of the variance in the average annual equivalent equilibrium volumetric activity of radon and thoron daughter products in the air of residential buildings, taking into account the uncertainty of the estimate. The spatial structure of the residuals was successfully approximated by a spherical variogram model, and their interpolation using ordinary kriging demonstrated acceptable accuracy (MAE = 0.12 normalized units, RMSE = 0.16 normalized units). The additive combination of the results from both models provided.