Geostatistical estimation of the hydraulic conductivity field under the Parma Dam

The city of Parma (Italy) located on the large alluvial fan of the Parma River, is preserved upstream by a flood control dam. The structure has the purpose of damping the Parma River floods storing a portion of the flood volume and releasing it, downstream, at a controlled rate. A concrete stilling basin is located downstream the dam in order to dissipate the energy of the discharged flow. The underneath deposits are surrounded by a grout wall that reaches an impervious clayey layer with the aim of realizing a confined "box". Groundwater levels inside the box are controlled by a 110 m long drainage trench located upstream the stilling basin (under the dam) and 3 m below its floor. Unfortunately, according to field data (boreholes and head observations) the box is not completely sealed both upstream and downstream. The aquifer beneath and surrounding the structure has been investigated by means of stratigraphical, lithological and hydrogeological characterization and in situ permeability tests. The groundwater monitoring network consists of 14 piezometers with dataloggers. The aquifer below the structure is multilayered, with prevailing silty gravels and relatively thin silty and clayey strata, and it can be simplified in three layers: a phreatic aquifer (0 to 20 m depth), a thin clayey layer (20 to 25 m) and a regional confined aquifer (beneath 25 m). In this work, the hydraulic conductivity field and a single value of the storage coefficient of the phreatic aquifer beneath the dam and the stilling basin has been investigated by means of a Bayesian Geostatistical Approach (BGA). The BGA method allows to estimate the hydraulic conductivity field (unknowns) constrained by prior information on its structure and, in the same time, to counteract uncertainties in the predictions and errors in the observations. The prior soft knowledge about the structure of the unknowns is related with the degree of smoothness and/or continuity of the parameters and serves also the role of regularization. The geostatistical inversion was performed by means of a computer code, bgaPEST, developed according to the free PEST software concept. The methodology needs a numerical model of the study area able to reproduce the boundary conditions, the geometry of the aquifer and the hydraulic heads in the observation points. In this work the groundwater flow process and the sensitivities of observations to parameters (required by the inverse procedure) has been simulated by means of an adjoint state formulation of MODFLOW_2005. The numerical model consists of a finite difference grid of 22 rows, 51 columns and 13 layers. Each cell is 4 m x 4 m x 2 m except the deeper ones that are truncated to the clayey layer. The upstream boundary conditions are the reservoir water level and a barrier simulating the grout wall that delimits the box below the dam. The downstream boundary condition is no flow except where the grout wall does not reach the clayey layer. This region has been described by means of a general head boundary condition; a drain condition was instead considered below the dam. In 2008, during a test period (5 months) of the hydraulic structure, the reservoir was filled capturing the tails of the spring flood events. The resulting head levels in the aquifer monitoring points and the reservoir water level were recorded. The hydraulic conductivity field and the storage coefficient of the numerical model were estimated considering an observation period of 15 days (one observation per day for each monitoring point). The entire model was then validated considering all the observations collected during the 5 months test period. The BGA methodology was able to estimate the hydraulic parameters and the storage coefficient of the aquifer, identifying the local heterogeneities. The calibrated numerical model allows to understand the interactions between the reservoir and the aquifer in different scenarios and to forecast head levels due to strong flood events.