Anticipating geochemical compatibility to reuse excavated soils at urban scale: Are usual statistical tools effective?

Abstract Reusing low to moderately excavated soils within land redevelopment projects necessitates among others to verify the geochemical compatibility of these soils with the receiving site. Optimizing excavated soil reuse at redevelopment project or urban scale is compelled by comparing their concentrations with threshold values based on quarter or urban geochemical background. Because urban soil geochemistry varies both at horizontal and vertical scale, background might vary spatially. It is thus necessary to consider also the vertical dimension of the geochemical background. However, available in depth data on urban soils consist mainly in pollution diagnoses. The representability of these data is then called into question, along with the method employed to determine geochemical background. To answer this question, we compare herein three standard statistical computation algorithms and examine their sensitivity to various parameters: probability distribution, number of data, proportion of values below the limit of quantification, and sampling schema heterogeneity. After performing a set of theoretical computations, simulations are run and application tests are conducted on actual datasets. Results reveal a variability in the calculated thresholds depending not only on the applied statistical methods used, but also on the distribution law, number of samples and sampling heterogeneity. It appears to be impossible therefore to apply these statistical methods on datasets that are not dedicated to the geochemical background without conducting a preliminary data study, ultimately followed by a data sort. It is in fact necessary to verify dataset consistency with the notion of an anthropogenic geochemical background. In addition, according to the objectives associated to the valorisation of excavated soils (economic benefit of excavated soils and/or environmental/health protection being prioritized), the thresholds (based on background) targeted may be more or less conservative by virtue of adapting both the computational method and algorithm used.