UltraFine+® Next Gen
Greenfields mineral exploration in varied cover is often hindered by a failure to detect, understand, and evaluate near surface geochemical anomalies. The UltraFine+® Next Gen Analytics research project combines the UltraFine+® soil analysis method with high quality spatial data, creating machine learning-derived, first-pass data products to make a step change for the exploration industry.
Developed by CSIRO, LabWest and 30 industry sponsors as part of the MRIWA M462a Project 2020-2023.
The UltraFine+® Next Gen Analytics research project (CSIRO/MRIWA Project M462a) was conducted from 2020 to 2023 by CSIRO in collaboration with over 30 industry sponsors and state geological surveys. The project goal was to facilitate a paradigm shift for precious, base and critical metals exploration in Australia by combining UltraFine+® soil analyses methods with intelligent data integration tools, adding value to routine soil sampling in frontline exploration and shaping mineral exploration approaches for decades.
The UltraFine+® Next Gen Analytics project’s site locations (as of August 2022; Click image to view the live version).
It has been common practice to use soil geochemistry in mineral exploration with little regard for soil parameters or landform settings and how these relate to buried mineralisation. The UltraFine+® Next Gen Analytics research addressed this challenge by delivering an analytical refinement of the UltraFine+® soil analysis method and by adding relevant mineral proxies via spectral mineralogy and soil properties pH, EC and particle size distribution to the soil analytical workflow and interpretation.
The CSIRO/MRIWA Project M462 refined and improved the UltraFine+® soil analytical method and is commercially available through LabWest.
The second component of the research – CSIRO/MRIWA Project M462a Next Gen Analytics – identified machine learning approaches to integrate soil geochemistry with spatial data. The resulting regolith landscape models can be used to identify statistical outliers by landscape type in geochemical soil surveys. This improves our ability to identify targets and false positives as well as understand the spatial variance and influence of regolith types.
Examples of machine learning derived landscape maps (top row) compared to satellite imagery (middle row) and existing regolith maps (bottom row). From left to right, machine-learned maps of (left) salt lakes across the Western Australia and Northern Territory state-territory boundary, (middle) alluvial channel systems in Queensland, and (right) sand dune in the Nothern Territory.
With the development of a robust soil analytical technique and new data products to fully assess underappreciated soil properties, the UltraFine+® Next Gen Analytics research provides the next generation analytical tools for mineral explorers to make qualified decisions on when and where to explore further. The overall aim with of this research project was to simplify and accelerate the ability of explorers to review their data and go from assessing geochemical surveys to discovery or moving on with confidence that they haven’t missed an opportunity.
Soil geochemistry in landscape context. Left: Boxplots for all Li data (white box) and for Li by landscape type (coloured boxes). The dashed line indicates the upper 25 % boundary for the whole sample population. Easily observed soil anomalies are samples above the dashed horizontal line. Those shown below the dashed line would not be easily observed without the landscape context. Right: (A) Spatial distribution of Li outliers (white triangles) calculated for the whole sample population over a map of machine-learning derived proxy regolith types (B) Spatial distribution of Li outliers (coloured triangles) calculated by landscape type over a map of machine-learning derived proxy regolith types.
For further details head to the Publications tab.
Since the project completion in November 2023, the CSIRO has conducted further research and ground tested the machine-learned landscape modelling in Australia (find the peer-reviewed publication here) and the landscape modelling and statistical outlier definition by landscape type functionality of the Next Gen Analytics tools will be commercially available through CSIRO’s Exploration Toolkit as LandScape+ from April 2025.
(A) Simplified sketch of the major components of the Next Gen Analytics machine learning workflow developed during the M0462a project. The workflow derives first-pass interpretation of the UltraFine+® soil analytical results in landscape context for a given project site. Figure reproduced from Noble et al. 2024. (B) Simplified sketch of the major components of the LandScape+ machine learning workflow which will be commercially available from April 2025. This application will provide landscape models with or without soil samples.