Geological and hydrological data collected at the Terrieu experimental site north of Montpellier, in a confined carbonate aquifer indicates that both fracture clusters and a major bedding plane form the main flow paths of this highly heterogeneous karst aquifer. However, characterising the geometry and spatial location of the main flow channels and estimating their flow properties remain difficult. These challenges can be addressed by solving an inverse problem using the available hydraulic head data recorded during a set of interference pumping tests.

We first constructed a 2D equivalent porous medium model to represent the test site domain and then employed regular zoning parameterisation, on which the inverse modelling was performed. Because we aim to resolve the fine-scale characteristics of the transmissivity field, the problem undertaken is essentially a large-scale inverse model, i.e. the dimension of the unknown parameters is high. In order to deal with the high computational demands in such a large-scale inverse problem, a gradient-based, non-linear algorithm (SNOPT) was used to estimate the transmissivity field on the experimental site scale through the inversion of steady-state, hydraulic head measurements recorded at 22 boreholes during 8 sequential cross-hole pumping tests. We used the data from outcrops, borehole fracture measurements and interpretations of inter-well connectivities from interference test responses as initial models to trigger the inversion. Constraints for hydraulic conductivities, based on analytical interpretations of pumping tests, were also added to the inversion models. In addition, the efficiency of the adopted inverse algorithm enables us to increase dramatically the number of unknown parameters to investigate the influence of elementary discretisation on the reconstruction of the transmissivity fields in both synthetic and field studies.

By following the above approach, transmissivity fields that produce similar hydrodynamic behaviours to the real head measurements were obtained. The inverted transmissivity fields show complex, spatial heterogeneities with highly conductive channels embedded in a low transmissivity matrix region. The spatial trend of the main flow channels is in a good agreement with that of the main fracture sets mapped on outcrops in the vicinity of the Terrieu site suggesting that the hydraulic anisotropy is consistent with the structural anisotropy. These results from the inverse modelling enable the main flow paths to be located and their hydrodynamic properties to be estimated.