Data Driven Decisions of Stationarity for Improved Numerical Modeling in Geological Environments

• Author / Creator

  Martin, Ryan

• Generating representative models of geological domains is critical for decision making and process optimization in natural resource exploitation. Partitioning geological datasets is an important step undertaken early in geostatistical analysis to ensure that subsequent modeling stages consider only related subsets of the domain. The partitioning is hierarchical; the large scale focuses on the geological properties of the domain to ensure the resulting models are geologically representative, and the smaller scale considers the statistical properties of the modeling variables to ensure the assumptions made in geostatistical algorithms are reasonable.Implicit modeling tools are commonly used to generate geological boundaries, partitioning the domain at the large scale. Commercial implicit modeling tools allow explicit control over all aspects of generated models, abstracting away the actual construction of geological surfaces. However, a manual interpretation of local structural features is required in many geological settings to ensure the boundary models are reasonable. Furthermore, the interpretation must be manually implemented by digitizing polylines or specifying local orientations throughout the domain. An implicit modeling framework is developed to address the inference and interpretation of local features from the dataset. The framework utilizes a well-established strategy for interpolating large datasets, modified for geological modeling by introducing unique local orientations of continuity. An additional methodology is developed to automatically generate local orientations from previous boundary models. The combination of automatic inference of local orientations with the developed boundary modeling algorithm results in a data-guided workflow for capturing curvilinear features in geological domains. The resulting models improve the geological feature reproduction of geological models.Following the large-scale geological partitioning, explicit decisions about statistically related samples must be made. This decision of stationarity is an important assumption made by geostatistical algorithms used to estimate value at the unsampled locations. Deviations from stationarity imply errors and biases will result in the generated models. There is a consensus in geostatistical literature that improved decisions of stationarity result in improved domains for geostatistical modeling. Increasingly the stationary partitioning of geological datasets is achieved through clustering, spatial clustering and other machine learning algorithms. Yet, the justification of a set of stationary domains is lacking. Several improvements to stationary domaining are proposed in this work, to aid in the generation and justification of groups of samples used for statistical modeling. Metrics for comparing two different possible stationary configurations are developed that allow the practitioner to choose better domains for geostatistical modeling. Furthermore, two new clustering algorithms are introduced that ease the burden of parameterization, and permit assessment of the uncertainty associated with this component of geostatistical analysis.This thesis makes several contributions to two related areas of geostatistics that have large consequences on the generated models. Geological boundary models are required to ensure the generated models have the correct geological context. Similarly, decisions of stationarity are required to ensure the dataset is amenable to statistical modeling. The combined contributions of this thesis improve partitioning of geological datasets at all scales for geostatistical modeling.

• Subjects / Keywords

  • Decision of Stationarity
  • Geostatistics
  • Spatial Clustering
  • Implicit Modeling

• Graduation date

  Spring 2019

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-w1db-c150

• License

  Permission is hereby granted to the University of Alberta Libraries to reproduce single copies of this thesis and to lend or sell such copies for private, scholarly or scientific research purposes only. Where the thesis is converted to, or otherwise made available in digital form, the University of Alberta will advise potential users of the thesis of these terms. The author reserves all other publication and other rights in association with the copyright in the thesis and, except as herein before provided, neither the thesis nor any substantial portion thereof may be printed or otherwise reproduced in any material form whatsoever without the author's prior written permission.