b'AEGC 2021Short abstractsFirst, a deep convolutional neural network instancethe technique to the Cloncurry region of the Mount Isa Inlier. segmentation model is trained to identify cohesive zones withinWe evaluate the effectiveness of using the compact magnetic core photos that are associated with the desired artefacts to berock property estimates to define targets that might otherwise masked. Following this, a convolutional neural network that ishave been missed in the early phase of exploration. Historical capable of inferring geologically reasonable textures in place ofairborne data acquired prior to mining of most deposits the irregularly-shaped masked artefacts is trained, using partialhelps us to design appropriate strategies for targeting similar convolutions that are conditioned on valid pixels. prospect styles.The improvement that these image-preprocessing algorithmsKey words: magnetisation, depth, IOCG, skarn, AI.have on image-based geological modelling will be presented.14: Almost automatic geological mapping from AEM21: Experimental investigation of the impact of cross-surveys flow on immiscible CO 2versus WACO 2displacement efficiency in permeability heterogeneity sandstone Dr David Annetts 1 and Dr Juerg Hauser2 porous media 1 CSIRO Mineral Resources Dr Duraid Al-bayati 1, Dr Quan Xie1 and A/Prof Ali Saeedi12 CSIRO MineralsQualitative interpretation of airborne electromagnetic surveys is1 Curtin Universitygenerally focused toward determining the geology representedReservoir rocks are rarely homogeneous, however, instead, by the inversion result for each station. However, it is a timevariations in permeability and porosity occur on a variety of consuming and often subjective process. It is well known thatlength scales. Geologic heterogeneity exerts a major influence machine learning algorithms excel at the automatic classificationon multiphase flow in the reservoir over many length scales, of features in images after training. Here we test a supervisedfrom the micron scale up to the kilometre scale. This is of learning approach for airborne electromagnetic data collectedgreat importance in many industrial and environmental in the La Grange groundwater area, Western Australia. We usecontexts, such as enhance hydrocarbon recovery and geologic machine learning to identify the most likely geological settingCO 2storage. The most common and characteristic structure at each station and use this to derive the probable extent of thein porous sedimentary rocks is layering (i.e. the continuous seawater interface. We employ standard techniques, such as crossparallel layers of different lithologic and physical properties) validation, to benchmark machine learning algorithms such asis considered as. Layers are observed at many different length nearest-neighbour, naive Bayes and support vector networks.scales, including; lamination (millimetre-thick layers), and The good agreement between a qualitative interpretation andbedding (centimetre- to metre-thick layers). The presence of the best-performing machine learning algorithm, here a randomheterogeneities inevitably influences the formation of viscous forest algorithm, for the seawater interface extents suggestsfingers and play a major role for generating of channelling that automatic classification has the potential to speed up thethrough preferential path. This manuscript presents the interpretation of large airborne electromagnetic surveys. Ourresults of an experimental investigation into the effect of results also suggest that careful use of machine learning algorithmscrossflow on displacement efficiency during immiscible trained on high-quality interpretations can lead to more objectivecontinuous CO 2and water alternating CO 2(WACO 2 ) in layered geological maps particularly when airborne electromagnetic datasandstone porous media. A manufactured core sample are collected in order to map regional geology. constructed by attaching two half-cylindrical homogeneous natural sandstone plugs of different permeability. Core 20: Rock property and depth mapping from magneticflooding tests were conducted at a constant temperature of data applied to greenfields exploration targeting in the343 K and pressure of 9.6 MPa. The results from this paper Cloncurry District are very important to overcome the current challenges in capturing the importance of crossflow influence as well as Dr David Pratt 1, Dr James Austin2 and Dr Clive Foss2 mitigating the effect of geological uncertainties on current and future CO 2storage projects.1 Tensor Research2 CSIRO Mineral Resources22: Groundwater and gas sampling informing A new method for building a model of the rock propertyhydrogeological conceptualisation of the Precipice and distribution on an unconformity surface (Pratt etal., 2019)Hutton Sandstone aquifers of the southern Surat Basinpresents new opportunities for greenfield exploration in complex geological environments. The method uses theDr Julie Pearce 1, Dr Harald Hofmann1, Mr Iain Rodger1, A/Prof inherent 3D information present in the magnetic tensor toPhil Hayes1, Prof Sue Golding2 and Ms Kim Baublys1create a model segment on the unconformity surface for every magnetic anomaly on every line in the aeromagnetic1 University of Queenslandsurvey. Because the tensor is a 3D spatial derivative of the2 School of Earth and Environmental Sciences, University of magnetic field, it automatically reduces the influence of regionalQueenslandmagnetic field changes to focus the inversion process on theThe southern Surat Basin in Queensland, Australia, may be a unconformity surface. An expert AI system builds a coherent 3Dprospective region for geological storage of CO 2within the geological model from the individual model segments, which itPrecipice Sandstone with the Evergreen Formation acting as then uses to constrain the joint inversion of the tensor data. a sealing cap-rock of the reservoir. The Hutton Sandstone, We explore the ways in which the maps and rock propertiesa major regional aquifer, overlies the Evergreen Formation. can be used to enhance greenfields exploration by applyingUnderstanding the baseline current hydrogeological and AUGUST 2021 PREVIEW 64'