Comparing the impact of deterministic and stochastic fracture networks on modelling hydraulic properties

Discrete Fracture Network (DFN) models have a widespread use when predicting hydraulic properties of fractured rock masses with different numerical and (semi-)analytical methods. However, recent advances in the way fracture network parameters are characterized in the field or in geophysical datasets are not completely reflected in input options of DFN simulators. For instance, to our knowledge no 3D DFN stochastic simulator is able to generate fracture networks with realistic topological relationships, and fracture spatial distributions different from a completely random Poisson distribution cannot be generated (so clustered or regular distributions cannot be modelled). This means that stochastic fracture networks cannot show realistic connectivity, with a strong impact on our possibility to model hydraulic properties.

Here we report on a comparative experiment where we have (i) reconstructed a 3D deterministic fracture network, based on rich outcrop data (Cretaceous platform limestones from Cava Pontrelli, Puglia, Italy), and stochastic DFNs with the same statistical parameters, and then (ii) we have modelled hydraulic properties with different semi-analytical (e.g. Oda method) and numerical methods (e.g. finite volumes implemented in DFNWorks).

Our preliminary results suggest that more advanced numerical methods are more sensitive to the quality of input data than simple semi-analytical methods. This is explained by the fact that for instance the Oda method simply ignores topology, connectivity, fracture height/length ratio and other important parameters.