Inference of Fractured Rock Transport Properties by Joint Inversion of Push-Pull and Single-Hole Ground Penetrating Radar Data

Abstract:

Flow and transport characterization of fractured rock formations is very challenging and important for a multitude of applications that include groundwater extraction, nuclear waste storage and geothermal energy production. One popular hydrogeological method to study fractured rock is a push-pull test, in which injection and retrieval of a tracer is made at the same depth interval in a borehole. In theory, push-pull tests are not sensitive to changes in the heterogeneity of the tracer flow path since the retrieval at the injection location minimizes advective effects and makes the test more sensitive to time-dependent transport processes. This assumption is limiting in the presence of a natural hydraulic gradient or if non-neutrally buoyant tracers are used, but these limitations can be reduced by monitoring push-pull tests with ground penetrating radar (GPR). We present a methodology for combined modeling and inversion of a series of push-pull tests that we monitored with the single hole ground penetrating radar (GPR) method. For the GPR modeling we use a newly developed approach to simulate the GPR response in fractured rock. We coupled the GPR model to a flow-and-transport simulator that we use to define the electrical properties of the fracture filling.

The combined model can cope with heterogeneous fractures of any orientation, aperture and size and allows for the effect of density driven flow (that is strong during the saline tracer tests). We use the combined simulator to create synthetic datasets for both the time-series of the GPR traces at different locations and the tracer breakthrough curves. Since the combined problem is highly non-linear and the inverse solution is ill-posed, we use stochastic inversion techniques to obtain probabilistic estimates of the parameters of interest (fracture length, orientation and aperture distribution) and assess the use of different measures to compare the simulated and experimental data.