b'ASEG newsCommitteesunanticipated anomalies. Results forThe approach was first tested withinKewell Dome target, which lies 40 km each profile suggested that metabasalticthe North Stawell Minerals tenement,northwest of Magdala, outside the extent clasts may be distributed throughout thetaking advantage of the quality andof the training dataset. It also highlights Moornambool Metamorphic Complexcoverage of the AGG survey data. Aftersome other areas of interest, including more widely, more frequently, and acrosstraining the neural network on thetargets with a similar character to Kewell a greater range of scales than expected.northern or southern half of the datafurther beneath the Murray Basin to the An example from the Magdala forwardthen testing it on the withheld half,east of Lake Hindmarsh, and highlights model is included in Figure 2. the gravity and magnetic neural netssignals beneath the Newer Volcanics at were each able to reliably identify, withMortlake. The composited model results The quality and coverage of the AGGspatial coherence, the anomalies thatand some highlighted areas are included data supplied by North Stawell Mineralscorrelated to the dome structures. Thein Figure 3.allowed for an additional opportunitymodel was then generalised to the to test a machine learning model thatbroader Stawell Corridor using the 2019The model represents a method would carry the learnings from forwardNational Compiled Gravity Grid 1VDof quantifying the potential field modelling into a regional potential field- instead of the AGG data. Neural nets forinterpretation process. The results are based predictive targeting model. Thegravity and magnetics were trained onhighly dependent on the quality of the forward models were used to map thea labelled dataset in the North Stawelltraining dataset, both in terms of the constrained dome extents, which wereMinerals study area, which was extendedresolution of the input data and the then used to label segments from gravityto include Magdala and the Stawellaccuracy with which the input data was and magnetic lines as dome targetGranite, then tasked to classify the newinterpreted. As such its very important signals for a training dataset. The gravityunlabelled data along the rest of theto have a thorough understanding of the and magnetic signals were each fed tocorridor. Outputs from the gravity andtraining area, as was the goal with the neural networks to train it on the signal,magnetic neural nets were compositedcharacterisation of the dome targets in then the trained networks were appliedto form a combined predictive potentialthe forward modelling. However, with to a set of potential field lines spatiallyfield targeting model. The modelthe machine learning-driven approach, separated from the training set. successfully predicts the location of thea thorough exploration model of a small Figure 1. Stawell Corridor study locality and geological map. Figure 2. Magdala ground gravity survey profile model.12 PREVIEWFEBRUARY 2024'