A Multiscale Investigation of Fracture Networks Interpreted from Aerial and Drone Imagery

Abstract

This PhD thesis focuses on advancing the understanding and characterization of geological fracture networks using aerial and drone imagery, and automated fracture extraction methods. The research addresses three primary questions: the reliability and robustness of automated fracture extraction methods when applied to decimeter resolution aerial imagery, the effects of varying imaging resolutions on the interpretation of fault and fracture networks, and how spatial and resolution dependent variability in fracture networks impacts geological interpretations. The studies carried out throughout this research employ a multiscale approach, using datasets of varying resolutions, from aerial imagery at 0.2 m/px to field imagery at 0.002 m/px, to quantify the information loss across different resolutions. Automated extraction methods, specifically the Complex Shearlet Transform and Binary Thresholding Method, were tested for their reliability and reproducibility. Results reveal that currently, automated fracture extraction methods face challenges, specifically with the reliable extraction of fractures from low-resolution datasets. It was found that the accuracy and reliability of these methods are significantly impacted by the resolution of the input data and by exposure continuity. Reliability was quantified by measuring the Root Mean Square Error (RMSE) of fracture lengths between different automated approaches, which revealed significant discrepancies, indicating that these methods often produce inconsistent results when applied to the same datasets. The impact of image resolution on fracture network properties was investigated and quantified using manual approaches. High-resolution imagery (0.02 cm) captures fine-scale features but is limited by area-bound censoring. Medium resolution (2 cm) imagery offers a balanced view, identifying a broad range of fracture lengths and characteristics. Low-resolution imagery (20 cm) covers extensive areas ii but often misses finer fractures, leading to severe fine-scale fracture censoring. A multiscale approach and rigorous tracing methodology are essential for accurate geological analysis. Finally, a case study on the granitic rocks of west Hitra Island, Norway, using imagery at three different scales (regional, local, and outcrop scales) demonstrates that accurate geological interpretations are possible using imagery alone. This study also stresses the importance of applying clearly defined criteria when using manual methods with decimeter-resolution imagery to ensure reliable results.