Open Access Repository

Machine-assisted modelling of lithology and metasomatism

Downloads

Downloads per month over past year

Hood, SB ORCID: 0000-0002-5680-7597 2019 , 'Machine-assisted modelling of lithology and metasomatism', PhD thesis, University of Tasmania.

[img]
Preview
PDF (Whole thesis)
Hood_whole_thes...pdf | Download (28MB)

| Preview
[img] Archive (Zipped appendices)
Hood_Appendices...zip | Download (63MB)

Abstract

Machine learning (ML), a subfield of artificial intelligence (AI), includes computational methods to identify patterns in data efficiently, objectively, and repeatably. These methods have increasingly been adopted in economic geology, which incorporates the science of mineral exploration. Computational techniques such as clustering, classification, and automation algorithms are several decades old yet are only recently moving from use by Earth informatics specialists towards common application by mineral exploration geologists. This thesis explores and progresses the application of these methods in a mineral exploration context, using multivariate data to interpret bedrock lithology and metasomatism (the alteration of minerals in rocks by fluid). The goal of this research is to demonstrate how machine learning and automated analysis, guided by the geological experience of a domain expert, may be more routinely used.
Three core thesis chapters each detail a case study where ML is applied to create maps or spatial models of lithology or metasomatism. In the first study, remote sensing data for a regional mineral exploration project in western Eritrea are used to produce maps of predicted bedrock, where substantial transported cover had previously been mapped. This study illustrates the utility of selecting appropriate input features for ML by a domain expert. By separating input features and producing iterative maps, practical bedrock maps are generated. The spatial representation of class membership probability and entropy are also shown to add value for domain expert interpretation of bedrock predictions.
The second case study uses whole-rock geochemical data from the Minto Cu-Au mining area in northwest Canada to produce cross-sections of granitoid dykes and sills. Machine learning is applied to interpret protolith rock geometry where overprinting ductile deformation and metasomatism obscure diagnostic textures of primary minerals. Data are first normalised and then clustered into natural groupings that represent protolith lithologies or rock-type subunits. These clusters then inform supervised classification to assign protolith equivalent labels to samples of altered rocks. Results highlight the need to account for the mathematical constraints on compositional geochemical data before applying ML, because cross-validation metrics yield similar values even when classification results are geologically spurious. The results allow reconstruction of protolith geometry and an understanding of how rock type may have influenced later partitioning of hydrothermal fluids and ductile deformation. A workflow is provided, to guide domain experts who wish to apply this method.
In the final case study, a statistically-robust approach is used to quantify and model metasomatism within a shear-zone at the Hamlet orogenic Au mine inWestern Australia. Whole-rock geochemical data from drill core were first processed to transform them from compositional space (i.e., summing to a constant such as ppm or weight percent) to Euclidean mathematical space (i.e., appropriate for statistical analyses). These transformed data were then used to construct synthetic data matrices, produced in a way that better represents the variable composition of protolith and metasomatised samples. Matrices are then used to compute mass balance estimates for geochemical elements using a Monte Carlo method, an approach which facilitates creating confidence intervals about the averaged results. The element enrichment and depletion values were plotted in three-dimensions to investigate potential metasomatic trends. Results indicate broad spatially cohesive regions of Bi, Na and K enrichment interpreted to correspond with fluid-focusing structures within the Au mineralised shear plane. The method presented in this case study improves the understanding of element mobility in a shear-associated, greenstone-hosted Au deposit. However, the approach also provides a way to pre-process large geochemical datasets before use with ML.
The research in this thesis demonstrates the practical value that can be gained by merging domain expertise with ML. The techniques described herein are generally well-understood and well-studied, yet their application by mineral exploration geologists has only recently become commonplace. The findings of this thesis will contribute significantly to the uptake of such computational methods for the objective and repeatable construction of reliable maps and models of lithology and metasomatism. Importantly, this research provides real-world examples of robust and auditable workflows, which produce results that can be integrated by non-technical domain experts: economic geologists exploring for mineral deposits.

Item Type: Thesis - PhD
Authors/Creators:Hood, SB
Keywords: Machine Learning; Modelling; Exploration; Economic Geology; Domain Expert
DOI / ID Number: 10.25959/100.00032368
Copyright Information:

Copyright 2019 the author

Additional Information:

Chapter 3 appears to be the equivalent of a post-print version of an article published as: Hood, S. B., Cracknell, M. J., Gazley, M. F., Reading, A. M., 2019. Improved supervised classification of bedrock in areas of transported overburden: applying domain expertise at Kerkasha, Eritrea, Applied computing and geosciences, 3-4. 2019 © The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

Chapter 4 appears to be the equivalent of a post-print version of an article published as: Hood, S. B., Cracknell, M. J., Gazley, M. F., 2018. Linking protolith rocks to altered equivalents by combining unsupervised and supervised machine learning, Journal of geochemical exploration, 186, 270-280

Chapter 5 appears to be the equivalent of a post-print version of an article published as: Hood, S. B., Cracknell, M. J., Gazley, M. F., Reading, A. M., 2019. Element mobility and spatial zonation associated with the Archean Hamlet orogenic Au deposit, Western Australia: Implications for fluid pathways in shear zones, Chemical geology, 514, 10-26. © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/).

Related URLs:
Item Statistics: View statistics for this item

Actions (login required)

Item Control Page Item Control Page
TOP