Case study IGO

The collaboration between Aurizn's data science team and IGO's minerals exploration experts aimed to improve the mineral exploration process by applying deep learning techniques to a large survey area of over 30,000km². The geospatial datasets used included ground gravity, airborne magnetic, and geochemistry data from over 9,000 drill holes with elemental assays.

The data analytics process involved interpolating geophysics to a 10m grid and converting geochemistry to prospectivity binary labels using provided elemental thresholds and ratios.

The machine learning components included unsupervised clustering of similar geophysics and supervised deep learning models using expert-derived binary prospectivity labels.

The model was trained using geophysics data as inputs and prospectivity as outputs on 80% of drill holes. The model's performance was tested and verified on 20% of drill holes, with qualitative assessment from known prospects and geochemist feedback. Multiple algorithms and visualisations were employed to ensure the model's robustness.

Regular feedback from the exploration team was crucial for geological interpretation, culminating in the delivery of GeoTIFF heatmaps indicating high prospectivity regions for IGO's drilling efforts. The heatmap provided a colour scale covering full confidence range (0-1) and a narrowed confidence range (e.g., 0.5-1) to highlight only the more likely prospective areas.

This machine learning pipeline has been established for the next survey area, streamlining the process for other greenfields and brownfields exploration.