b'AEGC 2021Short abstractsSubsequent geophysical methods include detailed groundThe sensitivity of a project to changes in geological gravity, airborne magnetics, down-hole (DHEM) and airborneinterpretation has been neglected in the past, due to a lack electromagnetic (AEM) surveys. DHEM has identified multipleof time to generate and develop different geological models. conductors associated with known, and potentially mineralisedA resource model should portray the best understanding of zones. Gravity and magnetic survey data and inversions havegeological processes and observations. To report a resource improved the understanding of the geology and structure of thefrom a geological model requires three components - volume, intrusion. An AEM survey flown in late 2020 highlighted knowndensity and grade or quality, each carrying a degree of mineralisation at Gonneville as well as identifying multiple newunderlying uncertainty.anomalies to the north within the broader Julimar Complex. A volumetric interpretation of geological observations is only Geophysical techniques will continue to provide a key roleas good as the knowledge, experience, bias and patience in exploring the Gonneville deposit, targeting extensions ofof the geoscientist building the model. In reality, several known mineralisation as well as delineating new areas forpossible interpretations could be generated by multiple continued exploration within the Julimar Complex. geologists. Geological uncertainty is just as important as grade uncertainty, yet often gets overlooked, primarily because 124: Geographic quantile regression forest: a newunlike grade uncertainty, there is no easy way of capturing or method for spatial modelling of mineral commodities communicating it.Mr Kane Maxwell 1,2, Dr Mojtaba Rajabi2 and Prof Joan Esterle3 Advances in machine learning have opened up new possibilities, and this presentation outlines a new method for 1 Matrix Geoscience recognising domain uncertainty. Using a case history with 2 University of Queensland data from the Lisheen base metal mine in Ireland, the author 3 School of Earth and Environmental Sciences, The University ofwill show how several possible interpretations for geological Queensland domain boundaries were generated from the same drilling data. All solutions honour the data, highlighting the underlying Spatial interpolation (modelling) is required for resourceuncertainty that exists in most geological settings.estimation in all mineral commodities. For spatial modelling of most commodities, geostatistical methods such as kriging andRecognising that uncertainty exists is the first step towards a more hybrid kriging are most popular because they are generallyrealistic resource statement. The ability to measure the variation in more accurate than deterministic methods, can quantifyinterpretation of the resource models provides mine planners and uncertainty and use auxiliary information to improve predictivepotential investors with a quantitative assessment of risk.accuracy. However, geostatistical methods have the primary disadvantages that they have onerous pre-processing steps126: Integration of high-resolution HyLogger spectral such as variogram modelling and the incorporation of additionalscanner and TESCAN Integrated Mineral Analyser for auxiliary information which has non-linear relationship with the target variable is difficult. To address this, a machine learningmineralogical characterisation of shalemethod based on quantile regression forest algorithm is proposedMr Muhammad Iqbal 1, Prof Reza Rezaee2, Prof Gregory Smith2 as an alternative approach for spatial modelling. This newlyand Mr Hasnain Ali Bangash3,4proposed method (termed geographic quantile regression forest), does not require variogram modelling, and can also quantify1 Western Australia School of Mines, Curtin University, Western uncertainty and incorporate auxiliary information. To evaluate theAustraliaperformance of the new method, the accuracy of predictions of2 Curtin Universityspecific coal properties is compared to inverse distance weighting3 University of Western Australia, Perth, WA(popular in the coal industry), and two geostatistical methods.4 Rio Tinto Exploration, Perth, WAData from an active mine site in the Bowen Basin, Queensland Australia is used for the comparison. In addition, the accuracy ofThe mineralogy of shales is a fundamental parameter because the predictions in two geological domains of the mine site, whichit has a direct influence on petrophysical and geomechanical have different spatial variation due to the impacts of intrusion,properties. However, thick shales comprise a heterogeneous is also compared. Using evaluation metrics from leave-one-outsuccession of very fine-grained strata in which only some thin cross-validation, this paper demonstrates that geographic quantilebeds are optimum for production of hydrocarbons. Hence, regression forest method has the highest accuracy, lowest biasmore continuous high-resolution mineralogical information is and highest precision of all methods across all coal propertiescrucial to obtain a better understanding of the heterogeneity and geological domains. Disadvantages of the new methodand fill the gaps between samples. This study aims to solve this compared to deterministic and geostatistical methods are that theproblem for the Goldwyer Formation shale in the Canning Basin, method is more computationally demanding, less intuitive andWestern Australia. A continuous mineralogical evaluation over is not available in existing geological packages. However its highthe core interval was carried out using the Hylogger spectral accuracy and advantage over geostatistical methods makes it ascanner. The spectra are validated with detailed core logs and candidate for future inclusion in geological model packages. TESCAN integrated mineral analyser (TIMA) analysis. The total organic carbon content (TOC) was determined by Rock-Eval 125: Recognising the impact of uncertainty in resourcepyrolysis. The results indicate that the Goldwyer Formation shale models is heterogeneous in terms of mineralogy and organic richness. Four main rock types are identified in the Goldwyer Formation Mr Steven Sullivan (RT1-4), each with distinct Hylogger spectra, TIMA based mineral distribution maps and TOC values. The RT-1 is an argillaceous Measuring uncertainty in resource models provides mine plannersshale with TOC ~2.5 wt% dominated by illitic clay minerals and potential investors with a quantitative assessment of risk. (50%). The RT-2 is an organic rich black shale with TOC 4 wt% 83 PREVIEW AUGUST 2021'