Reducing Geological and Geomechanical Uncertainty in Volcanic Geothermal Systems Through Integrated Well Log and Drilling Data Interpretation: Insights from Northeastern Iceland

Characterizing subsurface rock properties in high-temperature geothermal systems is inherently challenging due to limited data availability. Extreme temperatures restrict the deployment of wireline logging tools, circulation losses often prevent the recovery of cuttings, and coring is very expensive . As a result, for large drilled intervals only minimal direct geological information is available, increasing uncertainty in subsurface characterization. The thesis addresses these challenges by developing data-driven frameworks that extract geological and mechanical insights from limited wireline logs and drilling parameters, specifically for volcanic geothermal environments in northeast Iceland. The thesis first establishes objective-driven data preprocessing as a foundation for reliable subsurface characterization. Rigorous workflows are developed for wireline and drilling data that emphasize parameter selection aligned with geological and mechanical objectives. These transparent and reproducible preprocessing workflows are a pre-requiste to assure sufficient data quality for interpretable machine learning results and drilling derived indicators of bit-rock interaction. Building on these foundations, an unsupervised machine learning in form of a clustering workflow is applied to a limited set of wireline log data to derive electrofacies in the absence of cuttings. Electrofacies represent physically consistent rock groupings that are interpreted in this thesis in terms of plausible lithological compositions and porosity states. The results show that volcanic rock properties can be inferred from limited well log data sets, and that data-driven clustering can mitigate uncertainties arising from incomplete or misleading cutting records. Furthermore, the proposed approach enables the reconstruction of subsurface heterogeneity across mafic to felsic volcanic successions and provides a consistent framework for interpretations across wells and different geothermal fields. The thesis further integrates electrofacies with preprocessed drilling parameters to investigate whether the drilling response can serve as an indirect indicator of rock mechanical properties. After minimizing operational effects, variations in rate of penetration and weight on bit are shown to reflect porosity-driven strength contrasts between volcanic units. Consistent drilling signatures at electrofacies boundaries reveal how lava flow stacking, brecciation, and intrusion emplacement influence drillability at the scale of the drill bit. This integration of elctrofacies and drilling parameters allows the development of a conceptual geological¿mechanical model that links subsurface volcanic architecture to the drilling response. Overall, this thesis demonstrates that limited wireline and drilling data do not constitute a fundamental barrier to subsurface characterization in geothermal systems. Through objective-driven preprocessing, unsupervised machine learning, and integrated drilling analysis, it is possible to extract robust geological and mechanical insight in data-constrained environments.