Combining geostatistics and numerical simulations to improve estimations of polluted zones

Léa Pannecoucke (1) , Mathieu Le Coz (2) , Xavier Freulon (1) , Chantal de Fouquet (1)

1 GEOSCIENCES - Centre de Géosciences
2 IRSN/PSE-ENV/SEDRE/LELI - Laboratoire de recherche sur le devenir des pollutions de sites radioactifs

Characterization of contaminated soil and groundwater around industrial plants is a major issue of site remediation. A classical approach consists in providing an estimation of the polluted zone extent thanks to observations (data of pollutant concentration) and geostatistical tools (e.g. kriging). However, this estimation might turn out to be of poor quality if only little data are available. Besides, flow and contaminant transport simulation is a widely used tool when it comes to assessing potential migration paths of pollutants through the subsurface. It is efficient even if information from sampling is not available, as long as the input parameters are consistent with the site under study. The approach developed in this work consists in combining classical geostatistical tools and simulations of flow and contaminant transport. This approach aims at improving the quality of the estimation of the polluted zone extent and reducing the associated uncertainties. Two methods are proposed, based on the building of an a-priori model of pollutant plume migration which is sampled to obtain unconditional simulations. For the first method, hundreds of those simulations are used to compute covariances. These covariances account for the spatial variability of the regionalized variable representing the pollution under study. Hence, we are able to compute non-stationary covariances that reproduce this variability better than a model based on observations only. For the second method, a few simulations well correlated to the observations are used as drift functions in a kriging with external drift, which enables to add information about the variable under study through auxiliary variables. In order to implement the two above-mentioned methods and assess their performances, a synthetic model of subsurface is built. This synthetic model is representative of an alluvial soil and includes a vadose zone of a few meters deep in which the flow and contaminant transport is simulated. The synthetic case is composed of a reference simulation, which represents a real scenario of pollution due to a tritium source, and for which only some observations are available in order to estimate the polluted zone extent. Then, hundreds of simulations are run with input parameters differing from the reference simulation in order to take into account the uncertainties about the input parameters. The extent of the reference polluted zone is then estimated using three methods: a classical geostatistical method considered here as a benchmark and the two above-mentioned methods combining geostatistical tools and simulations. The results show that the estimations are improved when using methods combining geostatistical tools and numerical simulations, even when few observations are available, which underlines the interest of the approach. Finally, the proposed approach could help to better estimate volumes of soils to be decontaminated in the context of remediation of industrial sites. It is not limited to radiological pollution and could be transposed to other types of pollution.

Domaines: Sciences de l'environnement

Léa Pannecoucke, Mathieu Le Coz, Xavier Freulon, Chantal de Fouquet. Combining geostatistics and numerical simulations to improve estimations of polluted zones. Sustainable Use and Management of Soil, Sediment and Water Resources, AquaConSoil, May 2019, ANVERS, Belgium. ⟨hal-02460182⟩