Faculté des sciences

Deterministic and probabilistic numerical modelling towards sustainable groundwater management : application to seawater intrusion in the Korba aquifer (Tunisia)

Kerrou, Jaouher ; Renard, Philippe (Dir.)

Thèse de doctorat : Université de Neuchâtel, 2008 ; Th. 2053.

This PhD endeavours numerical groundwater modelling considering heterogeneous and uncertain hydraulic parameters. It is made of three parts. First, we investigated the effects of dimensionality and heterogeneity of the hydraulic conductivity on dispersive seawater intrusion (SWI) processes. Multiple 2D and 3D unconditional simulations of hydraulic conductivity fields sharing the same statistics... Plus

Ajouter à la liste personnelle
    Summary
    This PhD endeavours numerical groundwater modelling considering heterogeneous and uncertain hydraulic parameters. It is made of three parts. First, we investigated the effects of dimensionality and heterogeneity of the hydraulic conductivity on dispersive seawater intrusion (SWI) processes. Multiple 2D and 3D unconditional simulations of hydraulic conductivity fields sharing the same statistics were generated then used to solve density-dependent flow and solute transport equations with a finite element code. Monte Carlo simulations were analysed in terms of dimensionless criteria including the penetration length and width of the saltwater wedge. Results showed that the 2D heterogeneity is affecting more strongly the SWI processes than the 3D heterogeneity. The saltwater wedge length in the 2D models is smaller than in the 3D ones while there is more mixing in 2D models. Most important, results showed that there is a critical ratio between advection and dispersion processes which is controlling the behaviour of SWI in heterogeneous porous medium. The second part of the thesis dealt with deterministic and probabilistic modelling and long term forecasts of SWI in the Korba aquifer (Tunisia). The study started by the development of a 3D density-dependent flow and solute transport model of the regional Korba aquifer. Then, two geostatistical models of the exploitation rates and of the hydraulic conductivities within the aquifer were built by combining incomplete direct data and secondary information including aquifer physical parameters. The effects of the uncertainty on the spatial distribution of the pumping rates and the uncertainty on the hydraulic conductivity field on the 3D density-dependent model were analysed separately and then jointly. To circumvent the large computing time required to run hundreds of 44-years transient models, the simulations were made in a parallel fashion on the EGEE Grid infrastructure as well as on a local Linux cluster. The deterministic numerical model allowed to estimate the current over-exploitation of the Korba aquifer to 135%. It also allowed to estimate the time lapse needed to turn back the initial head and slat distributions (before exploitation start) to about 150 years. The results of the stochastic simulations showed that both uncertainties led to a zone representing 12% of the aquifer area, where the groundwater heads and salt concentrations are not known with accuracy. Most important, results showed that reducing the pumping rates progressively by 50% until 2048 will not result in a recession of the saltwater wedge ; instead an additional 9.5% of the surface of the aquifer will be contaminated in 2048. In the third part of the thesis, the performances of kriging, stochastic simulations and sequential self-calibration inversion are assessed when characterizing a non-multi-Gaussian synthetic 2D braided channel aquifer. In a first step, the performance of the three methods was compared in terms of reproducing the original reference transmissivity or head fields. In a second step, the methods were compared in terms of accuracy of flow and transport (capture zone) forecasts. Results showed that the errors remain large even for a dense data network. In addition, some unexpected behaviours are observed when large transmissivity datasets are used. We also observed an increase of the bias with the number of transmissivity data and an increasing uncertainty with the number of head data. This was interpreted as a consequence of the use of an inadequate multi-Gaussian stochastic model.