ANN-based approach for the estimation of aquifer pollutant source behaviour Maria Laura Foddis, Philippe Ackerer, Augusto Montisci and Gabriele Uras ABSTRACT The problem of identifying an unknown pollution source in polluted aquifers, based on known contaminant concentration measurements, is part of the broader group of issues called inverse problems. This paper investigates the feasibility of solving the groundwater pollution inverse problem by using artificial neural networks (ANNs). The approach consists first in training an ANN to solve the direct problem, in which the pollutant concentration in a set of monitoring wells is calculated for a known pollutant source. Successively, the trained ANN is frozen and is used to solve the inverse problem, where the pollutant source is calculated which corresponds to a set of concentrations in the monitoring wells. The approach has been applied for a real case which deals with the contamination of the Rhine aquifer by carbon tetrachloride (CCl 4 ) due to a tanker accident. The obtained results are compared with the solution obtained with a different approach retrieved from literature. The results show the suitability of ANN-based methods for solving inverse non-linear problems. Maria Laura Foddis (corresponding author) Gabriele Uras Department of Civil, Environmental Engineering and Architecture – Sector of Applied Geology and Applied Geophysics, University of Cagliari, via Marengo 3, 09123 Cagliari, Italy E-mail: ing.foddis@gmail.com Philippe Ackerer Laboratory of Hydrology and Geochemistry of Strasbourg (LHyGeS), University of Strasbourg, 1 rue Blessig, 67084 Strasbourg Cedex, France Augusto Montisci Department of Electrical and Electronic Engineering (DIEE), University of Cagliari, via Marengo 3, 09123 Cagliari, Italy Key words | ANN inversion, groundwater pollution source identification, inverse problems INTRODUCTION Groundwater is an important source for the production of drinking water. Consequently, the protection of ground- water resource quality appears of extreme importance for life support systems. Nevertheless, groundwater is exposed to man-made pollution that might prevent its use for drinking as well as for other domestic, industrial and agricultural purposes. When groundwater is polluted, the restoration of quality and removal of pollutants is a very slow, hence long, and sometimes practically impossible task. In the field of groundwater resource contamination, it should be highlighted that in some cases pollution might result from contaminations whose origins differ in time and place from the point where the contaminations were wit- nessed. To tackle such situations, it is necessary to develop specific techniques for identifying the behaviour of unknown contaminant sources from both spatial and temporal points of view. Getting to know the initial conditions of pollution is consistent with the implementation of the European Union Directive 2004/35/EC. This Directive, based on the ‘polluter- pays’ principle, concerns the environmental liability in relation to the prevention and compensation of environmental damages. The application of Directive 2004/35/EC requires the development of novel methodologies, such as that pro- posed in this work, for the identification of unknown pollution sources in contaminated aquifers. The problem of identifying an unknown pollution source in contaminated aquifers, based on known contaminant concentration measurement, is part of the wide group of issues called inverse problems. During the last decade, several studies have been dedicated to the development of different methods for solving inverse problems, however, works using artificial neural networks (ANNs) are less popular. 1285 © IWA Publishing 2015 Water Science & Technology: Water Supply | 15.6 | 2015 doi: 10.2166/ws.2015.087 Downloaded from https://iwaponline.com/ws/article-pdf/15/6/1285/413061/ws015061285.pdf by guest on 18 June 2020