| Summary: | This thesis investigates the hydrodynamic behaviour of fracture and karst networks with a specific focus on a major carbonate aquifer north of Montpellier, southern France. The study is based on an extensive database including fracture measurements collected on outcrops at various scales and hydrodynamic data recorded in boreholes at the Terrieu experimental site. A novel method was developed to predict the type and preferential location of incipient karst in layered joint networks. Results indicate that the aperture ratio between the bedding plane and joints controls the magnitude of flow localisation in the networks. The aperture ratio range for various karst types was also quantified. A 3D local-scale conceptual model of the field site was developed by integrating a detailed structural characterisation of fracture networks at multi-scale outcrops with a dynamic examination of flow path networks using borehole logging and well test data. Numerical flow simulations devised by an experimental design technique were used to systematically assess the feasibility of the conceptual model and quantify the effects of geometrical and hydraulic properties on the hydrodynamic behaviour. Results indicate that the borehole connectivity to different flow features dominates the borehole transient responses. The hierarchical flow behaviour observed in the field site can to some extent be captured by the predictive flow model; however, a complete match between the real and simulated data may require an improved characterisation of the highly heterogeneous and anisotropic transmissivity field. This issue was further addressed using inverse modelling approaches. A quasi-Newton approach was firstly applied to jointly invert the steady-state cross-hole pumping test data. From the inversions, distributions of flow channels in the Terrieu site were obtained. Results show that the flow anisotropy is consistent with the fracture anisotropy mapped from outcrops close to the test site. The manner of using the multiple dynamic datasets controls emergent local transmissivity contrast. In addition, bounds for local and regional transmissivities in the studied system were determined. A novel stochastic quasi-Newton algorithm was then applied to infer the local-scale spatial transmissivity field and to quantify the uncertainty associated with the inverse analyses. Results indicate that preconditioning the transmissivity field using rock type and borehole connection data is the key to finding a rational solution for high-dimensional inverse problems.
|
|---|
