radon in groundwater - Mark- och vattenteknik - KTH

Radon in groundwater - Influencing factors and prediction methodology for a Swedish environmentRESULTS AND DISCUSSIONSThe 4439 private drilled wells analysed inthis study were found to be irregularly distributedin Stockholm County, with relativelyfew wells in the middle and southernparts of the county as shown in Fig. 1. Ananalysis of the data revealed that for themajority of the wells (73%), 222 Rn concentrationsvaried between 0 and 500 Bq/l and11% of all wells exceeded 1000 Bq/l, theSwedish regulatory limit for radon concentrationin drinking water. The maximumconcentration recorded was 63 560 Bq/lwhile the minimum was 4 Bq/l. Two otherwells also had very high radon concentrationsin the order of 14 000 Bq/l and 15 000Bq/l. The geometrical mean was found to be230 Bq/l and a standard deviation of 1227Bq/l was computed. The distribution ofradon values, viewed on a histogram, waspositively skewed and removal of outlierswas necessary (except for visual data mining)to avoid bias in statistical analyses. Outlierswere case-specifically defined as radon concentrationsexceeding 5000 Bq/l and removed.The traditional definition of outliersin the one-dimensional case that a point isan outlier if its distance from the mean isgreater than some factor times the standarddeviation (usually ± 2 standard deviations)was not considered, since its application didnot improve the skewedness of the dataset.Radon concentration values exceeding 5000Bq/l were observed in 21 wells, constitutingless than 0.5% of total available wells. Highradon concentrations can be encountered inwells drilled from granite rocks with anenriched content of uranium deposited onthe inner surfaces of fractures.Visual data miningThe visual analysis of 3D images was appropriatefor the problem since all of the variablesinvestigated could either be convertedto surface maps or visualised in GIS. It waspossible to detect visually whether a relationshipexisted between a thematic map andthe radon concentration in groundwater. Asmentioned in the methodology section, aradon surface needed to be built from pointdata prior to visualisation. In the project,radon measurements were not sufficient tobuild a representative surface over the wholearea. Interpolation of all point data to produceone surface was found to produce lotsof ‘U’ effects (sharp edges) around extremevalues and this was inevitable because visualdata mining involves analysis of all data,without the removal of outliers. Therefore,only a small sample of the study with anacceptable distribution of wells was subjectedto investigation.In the various visualisations, the followingwere observed:1) High radon values occurred predominantlyon low elevations and vice versa2) Areas overlain with till or clay had highradon concentrations in the bedrockgroundwater3) Peaks of high radon concentrations werenot always located in regions where highuranium content in bedrock had been recordedin airborne measurements.4) Radon concentrations from steep terrainwere generally not high but no clear-cutconclusions could be drawn from the thematicmap of relative altitude5) Granitic rocks were associated with highradon values6) The shorter the distance between a fracturezone and the bedrock well, the higherthe concentration of radon in groundwater.A multi-dimensional visualisation (Fig. 3)involving five different variables was aninteresting example of how the effects ofpeaks and depressions in a surface combinedwith colour effects allowed the observer toget insight into the data, draw conclusionsand directly interact with the data. In thatvisualisation, the two-dimensional geographicalextent and the radon concentrationsconstitute three spatial dimensions.The bedrock surface draped over the radonsurface is the fourth variable and finally thefracture lines constitute a fifth variable andare draped over the last surface produced.11