Automated Generation of Site‑Specific Discrete Fracture Networks: Mixture‑Model Orientation Clustering and Fracture‑Count Calibration

Worldwide, three host rocks are considered for hosting high‑level radioactive waste repositories: salt, clay, and crystalline formations. In crystalline rocks, a reliable, site‑specific representation of the fracture network is critical for safety assessment because fracture characteristics control both mechanical integrity and hydraulic response. In this study we use fracture data from three 10‑m drillholes at the Bedretto Underground Laboratory, Switzerland. The wells are spaced 2m-apart giving access to the local fracture network.

We present an automated workflow to generate discrete fracture networks (DFNs) from field observations. The workflow has two main steps. Step 1: orientation clustering — we fit a mixture model to fracture orientation data to derive orientation sets, then validate the fitted parameters against the field observations using statistical tests. Step 2: abundance calibration — we perform Monte‑Carlo simulations using the best‑fit orientation sets and filter realizations by comparing simulated per‑meter fracture counts to in‑situ observations.

Our results show that mixture models reliably recover orientation parameters, but stochastic simulations are sensitive to the random seed; we therefore recommend ensemble simulations and sensitivity analysis. Combined with the proposed fracture‑count calibration, our approach produces robust, site‑specific DFN realizations suitable for numerical safety assessments and hydraulic or mechanical modelling of fractured rock mass.