IMPROVING FRACTURE NETWORK CHARACTERIZATION AND DISCRETE FRACTURE NETWORK FLOW SIMULATIONS USING UNMANNED AERIAL VEHICLES

Stochastic methods are commonly used to generate discrete fracture networks (DFNs) models for simulation of fluid flow and solute transport in fractured media. This is due to limitation of field methods in fully characterizing natural fracture networks which makes it difficult to develop models based solely on deterministic fractures. To improve the realism of stochastic DFNs and their predictive capabilities, field observations regarding fracture orientation, length, spatial organization, and density need to be incorporated into the DFN generation process. Traditional fracture mapping techniques that rely on airborne and satellite imageries are often resource intensive and have a coarse resolution that limits identification of smaller fractures. The recent advance in the use of unmanned aerial vehicles (UAVs) and image processing techniques, such as structure-for-motion algorithms, allow for high-resolution mapping which leads to better characterization and statistical analysis of the geometrical attributes and spatial organization of fracture networks. This study presents a workflow that integrates high-resolution fracture mapping with detailed statistical analyses of fracture properties to compute network-equivalent permeability. An UAV survey was conducted over a 95 m by 95 m migmatite outcrop in northern Togo. Fracture traces from the UAV map were then used to infer statistical properties of fracture orientation and length. The spatial organization and degree of fracture clustering were quantified using a two-point correlation method. A total of 4 orientation sets were delineated from the 1184 fractures mapped. Fracture lengths were found to span over two orders of magnitude (0.07 - 21.21 m) and follow a power-law distribution with an exponent of 2.2. The spatial correlation dimension was 1.47 which suggests a strong clustering of fractures. Equivalent permeability of the network was computed using an ensemble average of 50 stochastically generated networks and were found to be in good agreement with values derived from pumping tests in the region. These results suggest that high-resolution UAV surveys coupled with detailed statistical analyses can significantly improve the development of high fidelity DFN models for prediction of fluid flow.