Identification of the unknown pollution source in the Alsatian aquifer (France) through groundwater modelling and Artificial Neural Networks applications

Philippe Ackerer (1) , Maria Laura Foddis (2) , Augusto Montisci (3) , Gabriele Uras (2)

1 LHyGeS - Laboratoire d'Hydrologie et de Géochimie de Strasbourg
2 DIT - Department of Land Engineering
3 DIEE - Department of Electrical and Electronic Engineering [University of Cagliari]

Groundwater is the main source of drinking water and it has a vital importance in developed and developing countries. Computational models of groundwater are required to ensure its suitable management. Such models are useful to design remediation strategies in polluted aquifers. This work aims at studying the spreading of a dangerous chemical - carbon tetrachloride (CCl4) - that contaminated a part of the Alsatian aquifer (France) because of a tanker accident in 1970. The exact amount of the chemical infiltrated is unknown and this constitutes the main issue for its individuation and remediation. The purpose of this study is to define the behaviour of the unknown pollution source in terms of temporal variations, injection rates and duration of the activity. An Artificial Neural Network (ANN) has been developed to identify the characteristics of the source and to solve the groundwater inverse problem: on the basis of known contamination concentrations data in pumping wells, the pollution source temporal evolution is reconstructed. The ANN learns to solve a problem by developing a memory, associating a large number of input patterns examples, with resulting set of outputs or effects. ANN are characterized by a flexible structure capable of approximating almost al input-output relationships. To develop an ANN, it is necessary to generate a data set of patterns for training, validation and test procedure. In the case studied, the simulation models of solute transport in saturated groundwater flow are generated considering different scenarios of the source behaviour. These models examples are created with the software TRACES (Transport or RadioActiver Elements in the Subsurface) developed at the IMFS (Fluid and Solid Mechanics Institute) of Strasbourg “HOTEIT and ACKERER (2003)”. Conceptual model and model design of carbon tetrachloride pollution in the Alsatian aquifer has been the subject of various studies developed by the Institute de Mécaniques des Fluides et des Solides of Strasbourg. The numerical simulation model used to generate the necessary patterns for the ANN is based on data measured between 1970 and 2004.

Mots clés: Groundwater pollution; pollution source identification; groundwater modelling; Artificial Neural Networks

Domaines: Environnement et Société

Philippe Ackerer, Maria Laura Foddis, Augusto Montisci, Gabriele Uras. Identification of the unknown pollution source in the Alsatian aquifer (France) through groundwater modelling and Artificial Neural Networks applications. World Wide Workshop for Young Environmental Scientists: 2010, May 2010, Arcueil, France. ⟨hal-00521423⟩