Assessment of interpretation uncertainty in spatial domains using data science and bayesian approximation

Research output: ThesisDoctoral Thesis

Abstract

Uncertainty can be Aleatory (data) or Epistemic (model), successful communication of quantitative uncertainties can lead to well informed decisions to reduce risk and efficiencies in data collection through understanding value of information.
In the context of the mining industry, reporting codes for public announcements have been developed setting minimum standards for Public Reporting of Exploration Results, Mineral Resources and Ore Reserves. These include an assessment of the quality and confidence in the data and work carried out, since Public Reporting aims to deliver in-formation, provided by Competent Persons, to investors that is both Material and Trans-parent. The JORC Code (2012) provides a mandatory system for the classification of minerals Exploration Results, Mineral Resources and Ore Reserves according to the levels of confidence in geological knowledge, technical and economic considerations in Public Reports. The JORC Code (2012) is also used in other jurisdictions. Recent in-sights of report compliance between December 2018 and November 2019 have high-lighted issues.
It has been found that the “if not, why not” nature of the Code was mostly ignored in Public Reporting of mining projects when considering the quality of geological spatial domains. To reduce ambiguity in Public Reporting, it is essential to either reduce the possible terms that can be used to communicate uncertainty or to quantify it.
There are four phases required to estimate a Mineral Resource (Preparation, Investigation, Model Creation and Validation), and estimation is highly dependent on the accuracy of the Preparation stage which is a result of the quality of the geological interpretation given for the mineralisation process and current spatial location. This interpretation seeks to spatially delineate geologically homogenous areas in the resource (spatial do-mains), corresponding to a single statistical population with a single orientation, where possible. In the estimation workflow, the creation of the spatial domain presents a challenge in terms of assessing the uncertainty in the geological interpretation often due to the manual and subjective interpretation incorporating expert opinion. It is industry practise to let expert knowledge guide the creation of spatial domains (explicit model-ling). Spatial domains which might include more than one economic element or with several mineralisation overprint events are also problematic as they create areas where multiple spatial domains can co-exist, making traditional methods of assessing the homogeneity of the spatial domain difficult.
In the research literature and industry practise, there has been limited work to quantify the interpretation uncertainty of spatial domains in the context of Mineral Resource estimation. Hence, this research aims to develop an empirically modelling framework to test the applicability of Bayesian approximation methods using multivariate quantitative elemental data combined with qualitative data, to predict and assess the interpretation uncertainty in the domain classification of drill hole intervals in early-stage Mineral Re-source projects.
This research has determined that pXRF, ICP and qualitative data can be used with a Bayesian approximation method to assess interpretation uncertainty in early-stage project geological models. It also found that geostatistical simulation of Gaussian and Categorical variables was not suitable for this task, however when combined allow a fuller assessment of uncertainty covering both interpretation uncertainty and spatial uncertainty.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • Charles Sturt University
Supervisors/Advisors
  • Horta, Ana, Principal Supervisor
  • Rahman, Azizur, Co-Supervisor
  • Coombes, Jacqueline, Co-Supervisor, External person
Place of PublicationAustralia
Publisher
Publication statusPublished - 2022

Fingerprint

Dive into the research topics of 'Assessment of interpretation uncertainty in spatial domains using data science and bayesian approximation'. Together they form a unique fingerprint.

Cite this