Machine learning for mineral exploration

Can the application of innovative technologies accelerate the discovery of critical resources? Unearthed's Holly Bridgwater explores the possibilities...

By Holly Bridgwater, Industry Lead – Crowdsourcing, Unearthed

Economic mineral deposits have become increasingly difficult to find. During exploration, the iterative process of collecting different datasets, followed by geological interpretation, can take an extremely long time.

Vast amounts of data are collected and processed, very often without any significant mineral discovery.

Unearthed's Holly Bridgwater prepares to don 3D glasses
Holly Bridgwater is an exploration geologist, crowd sourcing lead at Unearthed, and an advocate for industry adoption of open data initiatives. Holly is working with the South Australian Government Department for Energy and Mining to deliver ExploreSA: The Gawler Challenge. Image: Roy Vandervegt

Therefore, explorers are seeking new approaches and innovative processes that can drive up discovery rates and speed up the exploration life cycle. This will result in more efficient global discovery of mineral deposits that contribute to production of the critical metals we need to grow future industries.

So, how can industry leverage data science to its full potential at a time when significant new mineral discoveries are becoming more and more rare?

Data science: can we believe the hype?

Data science technologies, including machine learning approaches, are gaining exposure and interest in the resources sector. But, how does machine learning play a part in discovery? What’s the hype and how does it really work?

Machine learning and artificial intelligence (AI) are some of the most overused buzzwords right now. This means that there is a lot of hype around them, along with a general lack of understanding and regular misuse of the terms and their applications.

Seen as a subset of AI, machine learning is the scientific study of algorithms and statistical models that computer systems use to perform a specific task without using explicit instructions, relying on patterns and inference instead.

Typically, machine-learning algorithms build a mathematical model based on sample data, known as training data, in order to make predictions or decisions without being explicitly programmed to perform the task.

Applications of machine learning to exploration are often thought of as ‘black box’ approaches. A considerable amount of the work involved is getting relevant and clean data sets, which is one of the biggest challenges and barriers to applying machine learning to exploration targeting.

Historic data is full of errors, is messy, and can be redundant because it isn’t relevant to the question that we are asking. And often, we may not have access to that many layers of useful data.

A typical scenario is a project where only magnetics and gravity and some limited chemistry is available, and at a low resolution, which does not give many variables to train a model on. Where we do have lots of data (and geologists love collecting a lot of data), we often don’t consider why we are collecting it, so we may not collect it in a useful or consistent way.

Consistency is a huge challenge, particularly with chemical assay data. The high level of variability across the data in methodology, elements and detection limits makes it very difficult to use in machine learning approaches.

Practitioners do not claim that machine learning is a silver bullet for finding the next economic deposit, rather that it is an approach that allows you to combine many layers of data and identify key relationships between them, in ways that we haven’t previously been able to.

Machine learning combined with domain knowledge can be a powerful new tool for geologists in assessing their projects.

Can data science techniques accelerate discovery?

Through developing and running crowd-sourced, open data exploration competitions, such as the Explorer Challenge and the upcoming ExploreSA: The Gawler Challenge, our team at Unearthed has found that this approach to exploration targeting can generate hundreds of independent machine-learning models and targets from data scientists and geoscientists around the world in just a few months.

Holly Bridgwater at the South Australia Drill Core Reference Library
Holly Bridgwater at the South Australia Drill Core Reference Library. Image: Roy Vandervegt

The submissions received on the Explorer Challenge displayed an amazing range of approaches. From cutting-edge machine learning to advanced physical modelling, the submissions represented thousands of hours of work, developing and applying robust techniques applicable to the problem of target generation.

Bringing data scientists and geologists together for the challenge resulted in novel ways to apply modern data science techniques to geological problems in a meaningful and explainable way.

At the conclusion of the competition, the best of the submissions were combined through our proprietary method into one aggregated target map. This approach reduces uncertainty, dramatically shortens the life cycle and may significantly increase mineral discovery rates.

Our next competition, in collaboration with The Government of South Australia Department for Energy and Mining, ExploreSA: The Gawler Challenge, goes one step further in being open with data.

All targets generated, including those from the winners of the A$250,000 prize pool, will be publicly shared to increase innovation and understanding in the resources sector by enabling access to data science approaches and modern geoscience thinking.

We are excited about the global community of innovators that this competition and world-class open data set will attract; some of whom won’t have worked in the mining industry before.

We can’t wait to see how they will apply their diverse skills, fresh ideas and novel approaches to mineral exploration to accelerate discovery in South Australia.

2 comments on “Machine learning for mineral exploration

  1. Thanks Carly. Hopefully you can pursue this effort.
    Excellent write-up.

    Edmundo TULCANAZA


    Simple,brief, excellent and constructive article on the use of machine learning in mineral exploration.
    Thank you…..

Leave a Reply

%d bloggers like this: