Abstract: |
For several years now, the application of geostatistics to soil contamination has been studied in detail and specific approaches have been developed. Geostatistics are now recognized as a helpful tool to support decision making, which provides the ability to map site contamination and to assess the volumes of soils requiring remediation, together with an estimation of the uncertainty. But what about their practical implementation? From our experience, real case studies raised three major types of difficulties. First, the quality of the available data may require some approximations, secondly the clients may raise questions requiring specific theoretical developments, and last but not least communication about the results may require the use of appropriate vocabulary and illustrations. Making use of our experience, solutions to overcome these difficulties are presented on the case of a former gasworks located in France, with soils showing high levels of polycyclic aromatic hydrocarbons (PAH). In this framework, the aim of the geostatistical study was 1) to assess and locate the volume of contaminated soil, 2) to compare these estimates with the remediation results, and 3) to quantify and position the residual pollution after remediation. A highly heterogeneous support (from 0.1 to 2.3 m in height) and an irregular distribution of samples in the domain were the major drawbacks of the available data. After testing different methods, it appeared that the best compromises were obtained by fitting variogram models using all the data set for the range and only the data of small support for the sill, and by conditioning the PAH simulations using the whole data set. Confrontation of the geostatistical model to the remediation results was performed at different scales, and required a discussion on the cumulative effects of data support, heterogeneity of the pollutant phenomenon and sampling / analytical errors. Most of the residual pollution after remediation appeared to be associated to the areas where the probabilities of exceeding the regulatory levels were low, which questioned the relevance of the stationarity hypothesis and required understandable explanations to the decision makers. Finally the key to a successful communication seemed to be: providing the essential information but not the details, making understandable the different sources of uncertainty and producing illustrations directly usable by the decision makers. |
