iNSIGHTS 1.0

SensOre iNSIGHTS on Geological Prospectivity Modelling

Transforming Embedded Geological Information into Model-Ready Data

SensOre recognizes the potential of non-numeric data to be transformed to inform machine learned prospectivity models. We’ve developed advanced programmable text-mining solutions to convert geological data into numbers. Machine learning algorithms are then used to predict spatial adjacency for commodities and deposit types of interest for each point object. Knowledge extraction is driven by contiguity associations, determining the spatial relationships of specific words and phrases from SensOre’s proprietary and numerically coded geological dictionary with Au, Cu, Li, or Ni deposits. This reflects a cause-and-effect relationship between mapped geological objects and mineralisation and helps to fully utilise our geological knowledge and understanding of mineral systems of interest.

With statistical output visualisation, SensOre has constructed new geological prospectivity maps over WA for a range of commodities and deposit types including Cu-VMS, LCT Pegmatite, Ni Intrusion-Related, and Phosphate. SensOre’s geological prospectivity models are on track to be extended to other states and continental Australia as the SensOre data cube expansion evolves towards completion over the coming months.

SensOre’s insights into geological prospectivity can be used in exploration for terrane and region selection, and for project generation and ranking. Strategically, these maps can be used to probabilistically assess the geological prospectivity of any tenement for the commodities modelled, making them an invaluable tool for business development decision-making.

iNSIGHTS 2.0

SensOre iNSIGHTS on Growing Australia Data Hypercube

Significant State-and Continent-Scale Expansion of Data Capture

SensOre Ltd is boldly changing the way exploration targeting is carried out by harnessing and fusing the vast quantities of underutilised geoscience and environmental data from the public domain into its Australian Data hypercube. SensOre’s most valuable (non-human) asset is its Data Hypercube. The data in the database resides in a common, globally scalable gridded geospatial architecture consisting of layers of geoscience data including geology, geochemistry, and geophysics. It is central to all SensOre models generated on open data and underpins our targeting services allowing SensOre to generate high-quality, data-driven exploration targets.

State-scale geoscience data is now ingested for WA, SA, and NSW

SensOre achieved a significant milestone of compiling, cleaning, merging, and ingesting the vast quantities of all openly available state-scale geoscience data across SA and NSW, adding to the completed data hypercube for WA. This was accomplished with the support of industry clients in SA and competitive government co-funding in NSW from the Critical Minerals and high-Tech Metals Activation fund. The Data Hypercube currently contains approximately 2,000 layers of geoscience data with more than 63 billion data points occupying the cells within the Data Hypercube.

The Data Hypercube includes SensOre’s Discoveries Database, an extensive metallogenic database of compiled and cleaned deposit and mineral occurrence data from many sources. This database is updated quarterly and is expanding to include more commodities. This database provides accurate point location, resource footprints, and mineral endowments and grades calculated from (mines, resource, reserves, and historical production), deposit depth including tested deposit depth and depth to top of deposit, mineral deposit type. The Discoveries Database is one of our most valuable data assets as it functions as one of our primary training datasets containing 61,660 occurrences for gold, nickel, copper, and lithium.

Combined, the Data Hypercube and the Discovery Database are deployed in SensOre’s machine learning (ML) workflows to target critical minerals, including Au, Ni, Cu, Co, PGE’s, Li and REE’s.

Continental-scale geoscience data expansion

A second significant milestone was achieved this year with respect to our continental-scale geoscience data compilations. We have made considerable advancements in acquiring, processing, compiling, and merging continental-scale geophysical and geochemical data. SensOre has ingested CSIRO’s ASTER, and hydrogeochemistry data.

Intrepid Geophysics has delivered their advanced processed continental merged compilations for magnetics, gravity and radiometrics data and approximately 54 derivative layers each for magnetics and gravity and 27 derivative layers for radiometrics. These data sets were added to our 430 m resolution Data Hypercube over the continent and are now suitable for application of machine learning algorithms as well as to our 80m resolution grids which aggregate to some 165 billion data cells over the continent.

Why are SensOre’s Data Hypercube expansions worth talking about?

• SensOre’s Data Hypercube expansion is addressing a protracted, industry-wide dilemma namely, the underutilisation, under-appreciation, and poor management of both new and legacy geoscience and environmental data that continues to obscure the potential utility of geoscience data.

• We’ve done the hard yards of compiling, cleaning, feature engineering and fusing massive amounts of geologically and environmentally significant data into a highly curated, proprietary database so you don’t have to.

• The construction of Data Hypercube represents several millions of dollars of investment and years of human resources dedicated to curating geoscience data conservatively estimated to have cost industry more than $20B to collect across three states.

iNSIGHTS 3.0

SensOre iNSIGHTS on igROCK and Simclust ML Tools

Supervised and Unsupervised ML Tools Generate Geological Insights from Geochemistry

SensOre Ltd has productionised a range of innovative ML tools and workflows to derive geological knowledge from geochemistry data. The power of these tools are realised by leveraging data from our substantial geoscience Data Hypercube and subject matter expertise in both data science and geosciences. SensOre’s ML solutions address protracted, everyday problems and challenges faced by geologists across the mining value chain and provide information and insights that allow decisions to be made on how projects should progress. Two of our most sought-after ML tools are igROCK and SimClust for Geochemistry. When used in tandem, these supervised and unsupervised approaches produce powerful and reinforcing results that increase understanding and confidence in the data collected.

IGROCK: A supervised ML tool that leverages SensOre’s Data Cube

Correctly identifying protolith types is critical when defining the spatial distribution of geological sequences. Especially where hydrothermal alteration and/or weathering and/or metamorphism is pervasive (and let’s be honest, if the rocks are interesting rocks they will often be altered!). To make things even more difficult a massive amount of the samples geologists have to log come from RC or aircore drilling programs. To help geologists make sense of their geochemical samples SensOre has developed igROCK – a supervised igneous protolith classification system. igROCK has been trained using igneous geochemical databases comprising over 400,000 samples. It uses the extended multi-element geochemistry to automate the classification of igneous rocks into 42 different rock types. Key to igROCK’s performance is the appreciable time SensOre’s geoscientific staff members have spent cleaning (and relabelling) these training databases prior to training the igROCK model (you would be surprised how many samples originally labelled as fresh ultramafics are running less than 1 % MgO!). Protolith identification is typically subjective and experience based. igROCK provides an automated solution to help geologists spend more time on the fun stuff (like interpreting where fluids have come from and planning where to put the next hole).

SimCLUST™️: An unsupervised ML tool that leverages SensOre’s Data Cube

SimClust Geochemistry is an unsupervised ML tool used to identify and classify natural geochemical populations of surface and drillhole point data using multiple, multivariate statistical models. The results can be used for rock characterisation, exploration vectoring, identification of altered equivalents, and differentiating regolith from saprock to name a few applications. The image below shows the Kundana Drill hole assay data in SimClust space.

Supervised and Unsupervised ML: A match made in heaven

When the results from igROCK are used to inform the unsupervised outputs from SimClust, magical things happen. When combined, igROCK and SimClust provide the geologist with valuable insights regarding different protoliths and altered equivalents and can help reinforce or challenge previous interpretations.

The coloured image below shows the Kundana Drill hole assay data SimClust clusters trained by igROCK.

The geologist can now answer questions such as which samples within these clusters are mafic? Are there multiple mafic groups or just one? Are there multiple fractionation trends representing multiple magmatic pulses in the AOI? Does a cluster represent multiple subclusters? In our example using the Kundana Drillhole assay data, igRock is able to inform SimClust there is ultramafic, mafic and felsic units in the first cluster, there is a felsic and intermediate subcluster in the second cluster, and that the remaining clusters are all felsic.

The map below shows the igROCK results plotted spatially, revealing coherent geological trends for the ultramafic, mafic and felsic units.

A Joint Supervised and Unsupervised ML: Case Study

The case study presented here is from bottom of hole chips collected from aircore surveys from an area near Mulga Bill. As if logging Aircore chips isn’t hard enough, the weathering at Mulga Bill is deep (150 m!), and the alteration is pervasive. The image below represents the geologists original interpretation of the samples. 

The next image shows the results from igRock. As you can see there are some significant differences between the two classifications. The original logging considers the samples in the east to be predominately mafics, while igROCK has classified the samples in east as being predominately intermediates with some lenses of ultramafics. That’s great, but how do we know igRock is providing reasonable predictions? When combined with the outputs from SimClust we can see that igROCK’s interpretation is far more consistent. igROCK is also able to inform SimClust as to the location of cluster boundaries.

Why are SensOre’s SimClust and igROCK ML tools worth talking about?

SensOre’s SimClust and igROCK geochemistry workflows are powerful proxy mapping tools in low-data density regions and in areas under cover.

The igROCK protolith classification system provides information critical in defining the spatial distribution of geological sequences where the primary mineral assemblages are obscured or have been destroyed by hydrothermal alteration and/or weathering. igROCK also provides a data-driven approach to classify difficult to log samples from RC and air-core programs.

SimClust provides powerful unsupervised validation for igROCK as well as a number of SensOre’s other supervised ML tools including AGLADS and iGOSSAN, which can be found in our brochure.

How do you get access to these resources?