b'AEGC 2023Short abstractsdeployed with the AEM system and because its affordabilitymeasurements. To assess the predictability of the final model and simple operation. The mining exploration communityobtained, we perform and compare different cross-validations uses magnetics as a proxy for geological mapping. We proposestrategies: 1) random cross validation, 2) spatial kfolds cross a method to use that structural texture from appropriatelyvalidation, 3) spatial block validation and 4) a leave-one-pre-processed mag grids as a structural guide in inversions ofout cross validation. These results show that the spatial AEM data sets. The method is based on our previous work thatdependencies existing in the predictors, the drill holes, strongly utilised the cross-gradient approach to introduce structuralinfluence the estimation of the performance of our model, information in inversion workflows. The algorithm requiresand that a classic random cross-validation overestimates the some modifications in order to be usable with the two- confidence in the results obtained. We therefore recommend dimensional gradients derived from a magnetic data grids.using kfolds or block validation strategies when using drill holes We compare results of this approach to those obtained byas a predictor for the location of a certain type of geological employing a full 3D joint electromagnetic and magnetic vectorfeature.inversion (MVI) of the same data.Monitoring subsurface changes using amplitudes Structurally self-constrained magnetotelluric inversion. of active and passive seismic signals recorded by 1 1 2 downhole distributed acoustic sensors: CO 2 CRC Otway Hoel SeilleGerhard Visserand Janelle Simpson Stage 3 field study.1 CSIRO2 GSQ Pavel Shashkin 1, Stanislav Glubokovskikh2, Roman Isaenkov1, Konstantin Tertyshnikov1, Sinem Yavuz1, Boris Gurevich, Julia Many Australian metallogenic provinces are buried underCorrea2, Todd Wood2, Barry Freifeld3 and Roman Pevzner1thick post-mineralisation cover. The detection of the depth to prospective basement or discrete mineral deposits is1 Curtin University challenging in this type of geological environment, and2 Lawrence Berkeley National Laboratory common geophysical data modelling approaches requires3 Class VI Solutionsconstraints to enhance the reliability of the resultingTracking changes of formation properties along the wellbore can geophysical images. In this work, we propose a probabilisticbe important for monitoring production and injection of fluids. If magnetotelluric (MT) data driven workflow for deriving thosea well is instrumented with an optical fibre, this fibre can be used structural constraints. These are then used to constrain aas a distributed seismic receiver array (DAS) to detect temporal deterministic 2D MT inversion. The workflow is assessed usingchanges in amplitudes of seismic waves generated by active synthetic data MT calculated from a realistic 2D Earth and thenor passive seismic sources. Stage 3 Otway Project presented tested on real data acquired in Queensland. We show that aan opportunity to test this technique using continuous active geologically realistic image with sharp resistivity boundaries(using permanent seismic sources) and passive (using regional can be recovered, making the precise spatial detection ofearthquakes) downhole DAS recording conducted over a period prospective basement or discrete mineral deposits possible,of 22 months before, during, and after a small CO 2injection at the even in presence of a thick cover. Otway International Test Centre. Analysis of P-wave amplitudes extracted from downhole DAS gathers shows a consistent Depth to basement mapping using a Bayesianamplitude anomaly at the CO 2injection level, visible in all active and passive observations after the start of injection. This indicates workflow: prediction and performance assessment. that the anomaly is caused by changes of elastic properties in Hoel Seille and Gerhard Visser the reservoir interval caused by CO 2saturation. The amplitude changes from active seismic sources are remarkably crisp and CSIRO consistent, whereas amplitudes of earthquake waves show significant scatter between earthquakes even without subsurface Mapping the basement morphology undercover is crucial tochanges. Thus, the use of the technique in passive mode requires identify structures significant for mineral exploration. Usingmultiple events to distinguish the time-lapse anomaly from geophysics, this mapping is done indirectly and the solutionstime-lapse noise. Ubiquity of these events even in a tectonically obtained present important uncertainty. Uncertainty can bequiet region (such as Australia) makes this technique a viable and estimated using Bayesian methods and can be reduced whencost-effective option for downhole monitoring.introducing constraints such as direct drill hole observations. When assessing the reliability of a probabilistic model, it is common to use cross-validation techniques, to evaluateControls on shelf-margin architecture and sediment how accurately it can predict unseen observations, in thatpartitioning in the Hammerhead shelf-margin (Bight case drill hole depth to basement observations. However, aBasin, southern Australia): Quantitative 3D seismic bias exists in the spatial distribution of the drill holes usedstratigraphy (QSS).as predictors. Certain exploration strategies, often based of geophysical observations, cause a dependency betweenJohn Shepherd, Simon Lang, Victorien Paumard, Annette adjacent observations that is incompatible with the statisticalGeorge and Daniel Peyrotindependence required by the cross-validation techniques, and can cause an overestimation of the predictability of a model. The University of Western AustraliaIn this paper, we assess the predictability of a depth toUnderstanding the stratigraphic architecture of shelf-basement probabilistic model derived from the Bayesianmargin clinoforms is key to determining how sediments are inversion of magnetotelluric data constrained by drill holestransported to deep-water settings and how the interplay of FEBRUARY 2023 PREVIEW 138'