A DFN-based High Performance Computing Approach to the Simulation of Radionuclide Transport in Mineralogically Heterogeneous Fractured Rocks

Geological repositories for nuclear waste are based multi-barrier concepts using engineered and natural barriers. In fractured crystalline rocks, the efficiency of the host rock as transport barrier is related to the processes: advection along fractures, diffusion into the rock matrix and retention onto the available sorption sites. Anomalous matrix penetration profiles were observed in experiments (i.e. REPRO carried out by Posiva at the ONKALO underground facility in Finland and the Long Term Sorption Diffusion Experiment, LTDE-SD, carried out by SKB at the Äspö Hard Rock Laboratory in Sweden). The textural and mineralogical heterogeneity of the rock matrix was offered as plausible explanation for these anomalous penetration profiles. The heterogeneous structure of the rock matrix was characterised at the grain-scale using a micron-scale Discrete Fracture Network (DFN), which is then represented onto a micron-scale structured grid. Matrix fracture free volumes are identified as reactive biotite-bearing grains whereas the rest of the matrix domain constitutes the inter-granular regions. The reactive transport problem mimics the ingress of cesium along a single transmissive fracture. Part of the injected mass diffuses into the matrix where it might eventually sorb onto the surface of reactive grains. The reactive transport calculations are carried out using iDP (interface between DarcyTools and PFLOTRAN). The generation of the DFN is done by DarcyTools, which also takes care of solving the groundwater flow problem. Computed Darcy velocities are extracted and used as input for PFLOTRAN. All the simulation runs are carried out on the supercomputer JUQUEEN at the Jülich Supercomputing Centre. The results are compared with those derived with an alternative model, where biotite abundance is averaged over the whole matrix volume. The analysis of the cesium breakthrough computed at the fracture outlet shows that the averaged model provides later first-arrival time estimates compared to the grain-scale model. The breakthrough computed by the two models converge at late times. These results suggest that spatially averaged parameters of mineral distribution are adequate to predict the late time behavior of breakthrough curves but could lead to severe overestimation of the radionuclide first-arrival.