Data: About 30 year after the first radon actions in Belgium, several datasets are available which allow the study of the correlations between indoor radon data and various related quantities. Indoor radon: the database covers the Southern part of Belgium (Walloon region) with an average density of about 1 data/km2, whereas the Northern part (Flemish region), where no affected area is known, only has about 0.1 data/km2. Soil radon and permeability: Ardenne, a radon-affected area of ~4000 km2, was densely explored (> 1 data/km2, including thoron), but only small campaigns went in the whole country (~30000 km2, ~0.003 data/km2), one being planned in summer 2017. Airborne survey of K, Th, U: uncalibrated data are available on a 100m x 100m grid. Soil K, Th, U/Ra: 219 data for U, somewhat less for K and Th, used for the calibration of the airborne data. Terrestrial gamma doserate: 379 measurements; calculated values are also derived from the airborne data. Qualitative information: for each measurement, geological, lithological and pedological information is available. Analyzing the relations between all these factors could allow developing a model that would predict areas affected by 222Rn, even without any measurement in homes. Because of the strong difference between Ardenne and the whole country in the information available on soil Rn and permeability, Ardenne deserves a specific approach. It is unfortunately not possible to include anthropogenic factors influencing indoor radon, because the available information was never encoded in the computerized databases. Ardenne: Ardenne is globally a strongly radon-affected area. Despite its homogeneity for geology, lithology and soil type, an important variability is observed in the indoor radon risk (local GM from ~45 to ~450 Bq/m3) as well as in soil Rn (1 to 400 kBq/m3), soil permeability (<10-15 to 6.10-11 m2), and airborne U (0.1 to 5.7 ppm). These variables were transformed in order to obtain roughly normal distributions. As the datasets were not collected at the same sampling points, a step of interpolation / smoothing was necessary for some of them before the analysis. Three methods were tried: (a) data mapped on a kilometric grid, by moving average (indoor Rn, soil Rn, soil permeability) or interpolation (airborne U); (b) data mapped at the soil sampling points by moving average (indoor Rn) or interpolation (airborne U) without processing soil Rn and permeability; (c) mean values determined for the squares of a 5x5 km2 grid (3x3 and 4x4 were also tried). Pearson's correlation coefficients of the data obtained with these methods show the absence of correlation between indoor Rn, soil Rn , soil permeability or airborne soil U in this Rn-affected area. Without surprise, further study by principal component analysis (PCA) does not give interesting results. This leads us to consider grouping by geological and/or lithological unit rather than mapping on a grid. Considering the weighted correlation coefficients between mean values calculated for 20 possible classes defined as lithology-geology pairs reveals better though still rather weak correlations between indoor Rn and soil Rn or permeability, but also a surprising negative correlation with airborne U. The construction of preliminary multivariate regression models by using classical OLS and spatial GWR was performed with Ardenne dataset in order to individuate those geological parameters that better define the geogenic radon potential of the region. Regression models were constructed by using dataset for 5x5 km2, a 4x4 km2 and a 3x3 km2 grids. Results indicate a general good performance of spatial regression models compared with the classical OLS global regression models. The geographical variability test computed for all the different grids shows a spatial variability in terms of model selection criteria, thus strongly supporting the use of spatial models. The 5x5 km2 dataset provides the best model. Whole Belgium: the extension to the whole country allows to take into account in the analysis not only strongly radon-affected areas like Ardenne, but also regions where the indoor radon risk is low, such as the Flemish region. Methods (a) and (c) cannot be applied to the too scarce soil Rn and soil permeability data. Method (c) was applied considering all other quantities available at country level and correlations were determined. Method (b) can be used to evaluate the indoor Rn GM and airborne U at the sampling points of soil Rn / permeability. The correlations observed are still weak, but with the expected pattern: positive correlation between indoor Rn, soil Rn and airborne U, positive also between indoor Rn and soil permeability, but negative correlation between soil permeability and soil Rn or airborne U. Organising data in groups according to geology, lithology or soil class is only possible for simple classifications with few groups, but much better correlations can then be observed. This is a good incentive for deeper studies like PCA and spatial regression models.
First steps in the multivariate analysis of Belgian radon data
Ciotoli G;
2017
Abstract
Data: About 30 year after the first radon actions in Belgium, several datasets are available which allow the study of the correlations between indoor radon data and various related quantities. Indoor radon: the database covers the Southern part of Belgium (Walloon region) with an average density of about 1 data/km2, whereas the Northern part (Flemish region), where no affected area is known, only has about 0.1 data/km2. Soil radon and permeability: Ardenne, a radon-affected area of ~4000 km2, was densely explored (> 1 data/km2, including thoron), but only small campaigns went in the whole country (~30000 km2, ~0.003 data/km2), one being planned in summer 2017. Airborne survey of K, Th, U: uncalibrated data are available on a 100m x 100m grid. Soil K, Th, U/Ra: 219 data for U, somewhat less for K and Th, used for the calibration of the airborne data. Terrestrial gamma doserate: 379 measurements; calculated values are also derived from the airborne data. Qualitative information: for each measurement, geological, lithological and pedological information is available. Analyzing the relations between all these factors could allow developing a model that would predict areas affected by 222Rn, even without any measurement in homes. Because of the strong difference between Ardenne and the whole country in the information available on soil Rn and permeability, Ardenne deserves a specific approach. It is unfortunately not possible to include anthropogenic factors influencing indoor radon, because the available information was never encoded in the computerized databases. Ardenne: Ardenne is globally a strongly radon-affected area. Despite its homogeneity for geology, lithology and soil type, an important variability is observed in the indoor radon risk (local GM from ~45 to ~450 Bq/m3) as well as in soil Rn (1 to 400 kBq/m3), soil permeability (<10-15 to 6.10-11 m2), and airborne U (0.1 to 5.7 ppm). These variables were transformed in order to obtain roughly normal distributions. As the datasets were not collected at the same sampling points, a step of interpolation / smoothing was necessary for some of them before the analysis. Three methods were tried: (a) data mapped on a kilometric grid, by moving average (indoor Rn, soil Rn, soil permeability) or interpolation (airborne U); (b) data mapped at the soil sampling points by moving average (indoor Rn) or interpolation (airborne U) without processing soil Rn and permeability; (c) mean values determined for the squares of a 5x5 km2 grid (3x3 and 4x4 were also tried). Pearson's correlation coefficients of the data obtained with these methods show the absence of correlation between indoor Rn, soil Rn , soil permeability or airborne soil U in this Rn-affected area. Without surprise, further study by principal component analysis (PCA) does not give interesting results. This leads us to consider grouping by geological and/or lithological unit rather than mapping on a grid. Considering the weighted correlation coefficients between mean values calculated for 20 possible classes defined as lithology-geology pairs reveals better though still rather weak correlations between indoor Rn and soil Rn or permeability, but also a surprising negative correlation with airborne U. The construction of preliminary multivariate regression models by using classical OLS and spatial GWR was performed with Ardenne dataset in order to individuate those geological parameters that better define the geogenic radon potential of the region. Regression models were constructed by using dataset for 5x5 km2, a 4x4 km2 and a 3x3 km2 grids. Results indicate a general good performance of spatial regression models compared with the classical OLS global regression models. The geographical variability test computed for all the different grids shows a spatial variability in terms of model selection criteria, thus strongly supporting the use of spatial models. The 5x5 km2 dataset provides the best model. Whole Belgium: the extension to the whole country allows to take into account in the analysis not only strongly radon-affected areas like Ardenne, but also regions where the indoor radon risk is low, such as the Flemish region. Methods (a) and (c) cannot be applied to the too scarce soil Rn and soil permeability data. Method (c) was applied considering all other quantities available at country level and correlations were determined. Method (b) can be used to evaluate the indoor Rn GM and airborne U at the sampling points of soil Rn / permeability. The correlations observed are still weak, but with the expected pattern: positive correlation between indoor Rn, soil Rn and airborne U, positive also between indoor Rn and soil permeability, but negative correlation between soil permeability and soil Rn or airborne U. Organising data in groups according to geology, lithology or soil class is only possible for simple classifications with few groups, but much better correlations can then be observed. This is a good incentive for deeper studies like PCA and spatial regression models.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
