Assessment of hydraulic parameters in the phreatic aquifer of Settolo (Italy): a stochastic approach

In this work we present and test against real field data an inversion approach for the identification of hydraulic parameters at the aquifer scale. Our test field is the alluvial phreatic aquifer of Settolo, located along the left bank of the Piave River in a piedmont area in Northeastern Italy, with an extension of approximately 6 km2 and exhibiting heterogeneities of the geological structures both at the local and intermediate scales. The area is characterized by the alluvial sediments (mainly gravel in a sandy matrix) deposited by the Piave River during the Last Glacial Maximum, being the subsurface, with an average aquifer thickness of 50 m, crossed by paleo-riverbeds that probably represent the main hydrogeological unit from which water is withdrawn. The interactions between watercourses and the aquifer, the recharge linked to the precipitation, as well as the dynamics of partially penetrating extraction wells must be properly reproduced for an effective protection and a sustainable exploitation of the water resources. In order to do so, in cooperation with Alto Trevigiano Servizi S.r.l., the local water resources management company, a careful site characterization is in progress since 2009, with a number of different measurements and scales involved. Besides surface ERT, water quality surveys, and a tracer test, we highlight here the role of 18 continuously monitored observation wells, available in the study area for the measurement of the water table dynamics and the calibration/validation of groundwater models. A preliminary comparison with the results of a three-dimensional Richards model demonstrated that the site can be properly described by means of a two-dimensional finite element solver of the nonlinear Dupuit-Boussinesq equation, saving CPU time and computer storage. Starting from an ensemble of randomly generated and spatially correlated hydraulic conductivity (K) fields, the fit between water table observations and model predictions is measured through the Nash-Sutcliffe (NS) coefficient. On the basis of a NS cutoff value, the best performing realizations are used to sample the locations in which K has the lowest uncertainty (coefficient of variation, CV, less than a pre-assigned threshold). These parameters are then used for a conditional generation of a new ensemble of K fields and the procedure continues until convergence. With reference to the recorded piezometric time series, we report on the capability of the inversion procedure and its sensitivity to the choice of the key parameters, namely, the NS and CV thresholds, as well as the prior geostatistical parameters, in particular the correlation anisotropy and characteristic length, of the initial ensemble of hydraulic conductivity fields.