Analogue outcrops can be a source of fracture information beyond the resolution of seismic and well profiles, providing geometric, topological, and geomechanical characteristics of the discrete fracture network (DFN). Fracture characterization is a task of interpreting and extracting these attributes in situ or in the digital models, requiring a specialized understanding of the fracture network and its relations regarding families of fractures and the estimation of attributes. Besides that, the outcrops may have limited exposure to certain characteristics important for generating more accurate models. Our approach to fracture characterization and modeling incorporates artificial intelligence techniques to mitigate bias in the DFN analysis and in parameter estimation when analyzing digital outcrop models. The proposed methodology includes the clustering of 3D data (strike/dip) using spherical clustering with a modified K-Means relying on spherical statistics. Considering the fractal propagation of the fracture network, our methodology also includes a power-law estimation using a Stochastic Gradient Descent algorithm. Using an outcrop within the Potiguar basin that exposes Jandaíra formation carbonate rocks as a case study, our methodology aggregated fracture planes and delineaments to create deterministic and stochastic volumetric models with fracture persistence. We also found that the 2D and 3D data had a stronger correlation in terms of connectivity, fracture intensity, and length distribution in both deterministic and stochastic models. With P₃₂ sampling, achieving a Pearson correlation of 99% against the P₂₁ fracture intensity and of 92% against the 2D fracture connectivity.
Racolte et al. (Sat,) studied this question.