Enhanced Geologic Modeling with Data-Driven Training Images for Improved Resources and Recoverable Reserves

Deterministic geologic modeling methods accurately characterize large-scale continuous features of geological phenomena, but often fail in reproducing their inherent short-scale variability. The opposite is the case with stochastic methods that lack large-scale continuity yet contain reasonable short-scale variability. Both methods are limited in their ability to account for a balanced amount of geologic variability. Integrating both large and short scale geologic elements properly improves the prediction of mineral resources and reserves. This thesis develops a methodology that improves the characterization of the geologic variability in mineral deposits. The central idea is to combine deterministic and stochastic geologic interpretations and transfer the essential geological features into geostatistical models. The multiple point statistics (MPS) simulation method is suitable for this task. This technique utilizes training images for extracting and then reproducing complicated geomorphological features in the models. The method has been adapted to integrate information from different images. Generally, training images are designed based on conceptual models of the geologic phenomena; in this work, deterministic and stochastic geologic representations are used as data-driven