Stochastic geolithological reconstruction coupled with artificial neural networks approach for hydrogeological modeling

When simulating fluid flow and solute transport a more accurate modeling of the lithologic, geological and structural characters of an aquifer is of extreme importance in order to improve the reliability of the numerical simulations. On the other hand the information available for the setting up of a hydrogeological model is subjected to ambiguities due to not univocal interpretations or to uncertainties linked to the methodologies of measurement of the variables of interest. Therefore, hydrogeological characterization of heterogeneous aquifers, if carried out up to a high degree of detail, should not identify a univocal model but a set of "equifinal" solutions. In the present paper the application of Artificial Neural Network approach coupled with a Nested Sequential Indicator simulation has allowed to obtain the distribution of hydrogeologic parameters that are not only conditioned by the in situ measured values but also by the soft information coming from geolithology. The results show a fairly good relationship between parameters such as Transmissivity and Storage coefficient and the geolithologic architecture of the examined aquifer.