Stochastic Inversion of Three‐Dimensional Discrete Fracture Network Structure With Hydraulic Tomography

Abstract

Groundwater flow through rocks with a low-permeability matrix is usually dominated by the presence of fractures, associated with pronounced local permeability contrasts. Multiple connected fractures yield preferential flow paths along a fracture network permeating the rock mass. Implemented in a model, the network is mostly represented either by a single or multiple continuum method that translates the hydraulic properties of the fractures into an upscaled effective permeability tensor or explicitly as a discrete fracture network (DFN). Combinations of both methods are also possible, such as realized by the discrete fracture matrix model (Berre et al., 2019). Dense fracture networks with many interconnections are more appropriate for the representation in a continuum model. In contrast, if a few fractures dominate the hydraulic conditions, resolving the fractures explicitly in flow models allows for a more detailed insight into preferential flow and transport paths, specific processes such as flow focusing, spatial fracture connectivity, and quantification of the individual influence of single fracture pa-

Hybrid Discrete Fracture Network Inversion of Hydraulic Tomography Data From a Fractured‐Porous Field Site

“…This was also shown by Fischer et al (2020) for a karstic and fractured aquifer, where the inferred transmissivities of fracture and matrix differ by several orders of magnitude. However, for an almost impermeable rock matrix (e.g., in most crystalline rocks), where hydraulic connections are exclusively maintained by the fractures, the classical DFN inversion (without matrix flow) can be sufficient (Ringel et al, 2021(Ringel et al, , 2022. If, on the other hand, the permeability contrast between matrix and fractures is very small, for example, due to only partially open fractures or high fracture surface roughness, the heterogeneities can be assumed continuous, so that the travel time inversion is advantageous due to its simplicity and computational efficiency.…”

“…For that purpose, we develop a novel inversion strategy using a hybrid DFN model. It is based on a DFN inversion algorithm that has been applied for fractured crystalline rock masses assuming an impermeable rock matrix (Ringel et al., 2021, 2022; Somogyvári et al., 2017). However, at fractured‐porous sites the characteristics of purely fractured rocks and a porous rock matrix with considerable permeability are combined and therefore it might not be sufficient to rely on either a continuum model or a DFN model alone.…”

High‐Resolution Characterization of Excavation‐Induced Fracture Network Using Continuous and Discrete Inversion Schemes

“…The Markov Chain Monte Carlo (MCMC) method is one of the most common techniques of Bayesian inference in geosciences. MCMC has been applied to solve inverse problems using various sources of information such as hydraulic tomography (Ringel et al, 2019(Ringel et al, , 2021, tracer tomography (Jiménez et al, 2016;Somogyvári et al, 2017), fracture network intersections with boreholes (Mardia et al, 2007), and in-situ stress state field data (Afshari Moein et al, 2018). The transdimensional reversible jump Markov Chain Monte Carlo (rjMCMC) is an extended variant of MCMC, in which the number of parameters can vary among subsequent iterations during the inversion process.…”

Three‐Dimensional Thermoporoelastic Modeling of Hydrofracturing and Fluid Circulation in Hot Dry Rock

“…In general, any single monitoring method can hardly eliminate uncertainties, rooted in large‐scale fractured rock mass, to precisely capture every detail of the fracture. Consequently, a need exists to integrate two or more monitoring techniques (e.g., the combination of borehole imaging for local scale and microseismic mapping for global scale) with inverse approaches (e.g., machine learning method) to minimize uncertainty, even using quite sparse data (Chakravarty & Misra, 2022; Ringel et al., 2021; Yao et al., 2018).…”