b'AEGC 2023Short abstractssulphides from surficial clay IP responses. The method was theclosure consisting of channel reservoir with thickness of ~15-20 subject of dual-frequency in an MSc thesis by Bob White inm that has been mapped from the 3D seismic amplitudes and the 1980s. In a 2022 presentation to the ASEG Sydney branch,confirmed by wells. Interpretation of the thin and lower quality Steve Collins suggested that the heterodyne method couldoil reservoirs in the form of secondary channel and floodplain potentially be useful as an airborne exploration. method,sandstone deposits from the seismic has not been successful. due to its fundamental frequency independence. SomeThe inability to discriminate and delineate the geological subsequent theoretical modelling showed that heterodyneand/or fluid facies is the main challenge to further explore effects could be detected using unconventional time-domainand develop the field. The challenge is worsened by the waveforms. uncertainty in the well logs and the poor quality nature of land seismic data. An advanced pre-stack geostatistical inversion Using a modified airborne EM system, we will report on the testsstudy has been carried out aiming to solve the observed undertaken over two or three sulphide deposits to establishkey issues: i) discrimination of different reservoir facies from if the proposed methodology has sufficient sensitivity to be aelastic properties derived from 3D seismic amplitudes; ii) novel ASSET in the geophysical toolbox. enhancement of the quality of the seismic to resolve the inherent uncertainty associated with the AVO responses; Rock physics driven machine learning for quick andiii) mitigation of the ambiguity of false AVO anomaly due to carbonaceous shale that had led to unsuccessfully drilled well. improved reservoir characterisation The applied geostatistical inversion study workflow includes Jyoti Malik iterative seismic petrophysics and rock physics modelling to produce a good quality and consistent set of well logs; Machine learning has been used in petroleum industry fromrobust seismic data conditioning for removal of coherent a long time, but its usage was limited due to hardware andand incoherent noises, and alignment of seismic events, with data constraints. With advancement in hardware capabilities,the resultant seismic AVO response calibrated with well data; machine learning usage has expanded in various domains.deterministic inversion of conditioned multiple angle stacks Still, in many real situations, inadequacy of well data requiredand litho-facies estimation using Bayesian inference to provide in seismic reservoir characterisation poses a challenge to useunderstanding on the intricacies of the aforesaid challenges recently developed deep machine learning methods, e.g.,before application of geostatistical inversion. Joint facies and convolutional neural networks (CNN). Theory guided machineelastic properties inversion facilitated by Bayesian-based learning (TGML) generates large amount of 1D synthetic datageostatistical inversion using Multigrid Markov Chain Monte to capture variability in the conditions of reservoir using aCarlo algorithm has resulted in highly detailed subsurface rock physics model(RPM), conforming to regional geology andfacies models that show excellent match at most of the 14 depositional setup. Corresponding amplitude variation withblind wells not used in the study.offset (AVO) responses are used for training and validating a CNN. Concept of transfer learning is used to validate the CNN training on real well properties before applying to 3D seismicData integration to quantify structural risk and GRV data for predicting several elastic and reservoir propertiesuncertainties over the western flank of the Cooper simultaneously. BasinHere, we present a case study on WestTryal dataset fromAlessandro Mannini, Jon Cocker, Diogo Soares Cunha and Northern Carnarvon Basin, Australia with limited well control inLaurent Souchethe survey area. A RPM is established and geological knowledge about the area is used to simulate various scenarios of reservoirBeach Energy variation to predict elastic properties in 1D. Each set of reservoir and elastic properties is regarded as a synthetic well. A real- Beach Energy has successfully discovered and produced world wavelet is used to compute AVO responses for eachhydrocarbons from the Western Flank of the Cooper Basin synthetic well. With this, there is a lot of well and seismic to besince 2002. The entire sector operated by Beach Energy is used in the deep neural network for machine learning. Trainedcovered by good quality Pre Stack Time Migrated Seismic and validated CNN is then transferred & applied on 3D seismicData (PreSTM). Multiple drilling campaigns, executed over the data to predict multiple elastic and reservoir properties, acousticyears, have confirmed that structure and migration are the impedance, Vp/Vs, porosity, volume of clay and water saturationgreatest risk to finding additional hydrocarbons in the area. simultaneously. A comparison is made between porosity andSince a proper assessment of the pre-drill gross rock volume volume of clay predicted from conventional workflow and the(GRV) is also crucial to inform future exploration campaigns, one precited from TGML, showing improvement in quality ofextensive efforts were made to fully understand the GRV prediction and value addition by removing workflow repetition. distribution using a stochastic approach, as well as the risks associated with the presence of four way closures. Once a depth conversion project, exploiting the value of the latest Maximising value from available data via advancedstochastic technology along with the benefits of a Pre Stack geostatistical inversion in the Growler Field. Depth Migration (PreSDM) reprocessing, was completed, risks (probability of a structure being present), uncertainties (GRV Alessandro Mannini 1 Diogo Soares Cunha1 and Jimmy Ting2 distribution) and reservoir depth estimation were validated by 1 Beach Energythe post drilling results of 13 exploration and appraisal wells. 2 GeoSoftware Post drilling results confirmed that the chosen approach is more precise and accurate compared to previous attempts to The Growler field produces oil from the middle Birkheadquantify risks and uncertainties carried out in the past using formation. The main production area is a low relief four-way dipthe same input data.117 PREVIEW FEBRUARY 2023'