This project continues to progress methods for the automated classification of lithology and alteration zonation from geological, geophysical and geochemical data and flexible approaches to visualising research results. Our focus continues to optimise the value of machine learning in practical workflows, given a resource project’s development stage (e.g., area selection, target prediction, resource evaluation, and resource development) and has progressed with the PhD research of Stephen Kuhn. In 2019, a paper was published detailing geochemical reconnaissance-stage research using machine learning to incorporate geochemical sampling and preliminary ground-based mapping. Currently in its final peer review stages, a new paper, again based on Steve’s research, details the application of machine learning prediction uncertainty to highlight the most likely areas for the occurrence of key lithologies. Steve is in the last stages of completing his PhD thesis, which has been generously supported by First Quantum Minerals, and is expected to submit early in 2020.

Visualisation development activities also took place with the development of a suite of interactive software for geoscientific visualisation. The software, developed by Peter Morse, incorporates understanding of human colour perception and perceptually uniform colour-space, enabling well-posed visualisation to enhance the confidence levels in visualised data and to characterise uncertainty. In 2019 a paper was published detailing application to 2D seismic contour maps, generating new insights for AuSREM seismic data. In 2020, a demonstration study is described in a new paper currently in peer review, extending this work to 3D global tomography in solid Earth geophysics. Peter is in the final stages of his PhD and will submit in early 2020. The software suite is available opensource on Github.

Professor Anya Reading, gave an invited talk and sat on a panel at the Stanford (Earth) Women in Data Sciences Workshop in November. This event highlighted the use of data science in various geoscience research areas.

This project continues to progress methods for the automated classification of lithology and alteration zonation from geological, geophysical and geochemical data and flexible approaches to visualising research results. Our focus continues to optimise the value of machine learning in practical workflows, given a resource project’s development stage (e.g., area selection, target prediction, resource evaluation, and resource development) and has progressed with the PhD research of Stephen Kuhn. In 2018, a paper was published on geophysical reconnaissance-stage research with demonstration studies incorporating geochemical sampling and preliminary ground-based mapping following through the year with two further papers submitted/ in preparation. Significant progress was made for a case study in North America whereby the interim products of a machine learning exercise highlight the most likely areas for the occurrence of key lithologies. This represents very valuable progress in the application of machine learning methods to areas with lithological units that are difficult to distinguish.

Visualisation development activities also took place with the development of interactive software which enables insights to be drawn from contour maps allowing for a better understanding of human colour perception and the confidence levels in the data set. A demonstration study is described in a paper in preparation and the extension of the work to 3D through the balance of luminosity and transparency is ongoing through the work of Peter Morse.

The project leader, Professor Anya Reading, gave an invited presentation at the first meeting on Machine Learning in Solid Earth Geoscience in Santa Fe, New Mexico (February 2018), on ‘Optimising Insights from machine learning in geoscience’, and contributed as co-Director to the institute-wide UTAS Research Theme on Data, Knowledge, Decisions.

This project aims to test and refine supervised and unsupervised learning methods for the automated classification of lithology and alteration zonation from geological, geophysical and geochemical data. By exploring unique characteristics of individual ore deposit styles (e.g., orogenic gold, sedimentary copper, etc.), this project will identify appropriate scales of investigation, and optimal input data, given a resource project’s development stage (e.g. area selection, target prediction, resource evaluation, and resource development). Resulting models will be used to independently validate existing geological maps, while also identifying mineralisation targets, especially in areas concealed by cover.

Development activities focussed on geological feature discovery included the publication of a computer application by PhD student Peter Morse that enables a first pass reconnaissance of large datasets using an animated desktop computer user interface. The application is flexible for use on datasets held locally, but also compatible with cloud computing architecture which enables an interface with large, national databases. The project leader, Professor Anya Reading, gave a keynote presentation ‘Machine Learning using high-D spatial data’ at the Colorado School of Mines Symposium on Machine Learning in Geophysics, and contributed as co-Director to the institute-wide UTAS Research Theme on Data, Knowledge, Decisions.

Research achievements in the project area centred on the PhD research of Stephen Kuhn have progressed very significantly in 2017. Highlights include the completion of a project that uses only airborne geophysics and remote sensing data to refine a geological map in the reconnaissance stage of exploration, and substantial progress on a project which used a case study approach to optimise strategies for Machine Learning at early to late stages of project maturity. Notable achievements also include the geologically referenced use of metrics to quantify the uncertainty associated with prediction of lithology. Based on the ongoing success of Earth Informatics being undertaken through the TMVC Hub, Stephen Kuhn and co-supervisor Dr Matthew Cracknell were invited to contribute to a training workshop at the Australian Exploration Geoscience Conference to be held in Sydney in early 2018.

This project aims to test and refine supervised and unsupervised learning methods for the automated classification of lithology and alteration zonation from geological, geophysical and geochemical data. By exploring unique characteristics of individual ore deposit styles (e.g., orogenic gold, sedimentary copper, etc.), this project will identify appropriate scales of investigation, and optimal input data, given a resource project’s development stage (e.g. area selection, target prediction, resource evaluation, and resource development). Resulting models will be used to independently validate existing geological maps, while also identifying mineralisation targets, especially in areas concealed by cover.

During the year, robust methods for the selection of relevant variables, and predictive model uncertainty quantification, have been applied and tested on existing geological models from the Eastern Goldfields region of Western Australia and the Zambian Copperbelt in Central Africa. Results indicate that automatically defined relevant geochemical features align closely with features used by domain experts to interpret geological domains. Furthermore, uncertainty maps provide a quantitative means of streamlining data acquisition for increasing confidence in output geological models.