This thesis investigates fractured zones leading to preferential groundwater flow paths. In this context, we used the electrical resistivity tomography (ERT) and the self-potential (SP) methods to identify, characterize, monitor, and finally model preferential flow in hydraulically-active fractured zones at a scale representative of real world applications.
From an experimental point of view, we first identified the magnitude of self-potential signature, a -15 mV anomaly that could be associated with preferential groundwater flow in a shallow quartzite aquitard whereas the streaming potential theory was originally developed for porous media. This signature was confirmed experimentally in limestone aquifers at greater depths. The joint use of surface ERT and SP allowed the identification of water-bearing fractured areas which were electrically more conductive, presenting contrasts from 1 to 10 and which were hydraulically-active presenting negative SP anomalies ranging from -10 to -30 mV. We were also able to correlate hydraulic heads and SP gradients during a low and a high groundwater level period leading to interesting perspectives in understanding the dynamics of complex groundwater flow systems. Finally, a preferential flow and rapid transport path, over 10 m/h, was highlighted in a 20 m deep fractured and karstified limestone valley by monitoring a salt tracer test with only surface ERT. This methodology was being mostly used for relatively shallow and homogeneous aquifers up to now. Such information is crucial to set up new monitoring wells or to define the sampling rates of classic tracer test.
From a methodological point of view, we quantitatively assessed the efficiency of blocky and minimum-gradient-support regularizations in electrical imaging to recover sharp interfaces on numerical benchmarks and with field data. The usefulness of resolution indicators such as the cumulative sensitivity matrix and the resolution matrix were also assessed in this context. We demonstrated that noise characterization is crucial in time-lapse inversion and may supplant the choice of the time-lapse inversion scheme, calling for a systematic analysis of reciprocal measurements (or a subset of them). We also showed that, when using data differences in an inversion scheme, the data error, as estimated by time-lapse reciprocal measurements, depends on the mean measured resistance. These error characterization studies should always be performed if one wants to avoid wrong interpretations about the hydrodynamics. We further showed that focused inversion techniques (blocky inversion, minimum-gradient-support) may offer great perspectives when recovering model changes in time-lapse inversion.
Finally, ERT and SP were jointly used to conceptualize a physically-based and spatially distributed hydrogeological model, in particular to characterize the preferential flow paths. Predicted hydraulic heads and SP-derived hydraulic heads using the water table model showed a clear correlation, leading to perspectives in terms of hydrogeological model calibration. Further experiments are however needed to fully estimate the streaming potential apparent coupling coefficient, but the use of the full SP signals for hydrogeological model calibration is a clear perspective to this work.