Mine data mining
How to find the metals for a sustainable economy
The world economy needs a massive injection of base metals to enable it to convert from an oil and gas economy to a sustainable economy (see below).
Demand side metal projections are high. BHP estimates copper demand for battery-powered electrical vehicles (EVs) will need 80 kg of copper, four times as much as an internal combustion engine. BHP forecasts by 2035 there could be 140 million EVs on the road (8% of the global fleet), versus 1 million today. BHP projects 8.5 million tonnes more copper is needed, a third more than today’s total global copper demand. Nickel demand for EV battery manufacturing projections show growth of 29% per annum until 2030. To enable this mining boom, investors and explorationists need a present-day exploration effort that leverages existing data to increase odds of exploration success and massively shorten the time required to open new mines.
Present situation in mining exploration
Present mining exploration consists of drilling geophysical anomalies identified by various geochemistry and geophysical techniques. This current process is slow, expensive, and risky and is not finding and producing enough mines. The Prospectors & Developers Association of Canada (PDAC) calculates the odds of finding a mine from exploration to mine(s) commercialization as 0.01%.
The very low odds of successful exploration mean mines are harder to find and thus take years to find and develop with typical quoted time lags for exploration to opening a mine of 15 years. The odds continue to get worse, and the supply of new mines is plunging. Between 1983 until 2020, there has been a 67% drop in mines the United States, and between 2011 to 2020, there has been a 32 % drop in new gold mines, and a 57 % drop in new base metal mines worldwide.
Existing exploration companies take a balkanized exploration approach. Some explore only next to old mines in mining districts. They anticipate luck will rub off from those old mining district deposits. Greenfield explorationists have PDAC’s odds. We have nothing against both approaches – we wish them well. But this hasn’t stopped a decrease in new mines.
Geological data and exploration companies
Industry and governments approach this exploration time lag issue like a gambler at a horse race who gathers data about each horse and jockey. The “horses” are various types of geological data being gathered, and the “jockeys,” are companies with skill levels to ride a resource race. This horse and jockey data gathering approach is done to find out what works to find deposits, and who best finds economic mines. Horses alone carry the jockeys, but it is horses – the geological data – that win the races.
Thousands of terabytes of data to “help” with a low odds exploration success process is piling up. Until recently, no one took a look at all this public data, from a data mining perspective, to find what works and does not work, using a supercomputer capable of looking and finding economic deposits across vast areas.
The Eureka Maps approach: look for ore deposits both around old deposits, and in entirely new places, using data mining algorithms that scale to a planetary level. This approach presents opportunities for novel discoveries of orebodies at a different scale, the planet.
Using sophisticated data mining techniques and supercomputers, at slightly better accuracy levels than IBM gets on data mining in other industries, we are able to predict geological anomalies holding resources with better than 95% accuracy. This massively reduces the risks, costs and time associated with the current 15-to-20-year cycle of exploration to mine opening.
The techniques include decision tree algorithms, solving both regression and classification problems to high accuracy levels. Decision trees create a training model that predicts the class or value of the target variable by learning simple decision rules inferred from prior data (training data). Decision trees, for predicting a class label for a record start at a root of the tree. Algorithms compare values of the root attribute with the record’s attribute. On the basis of comparison, branches corresponding to a value are then added and then jump to the next node while improving the overall algorithm fit to the dataset. Test confusion matrices are then run on new known data predicting the accuracy of decision trees. Actual ground truthing by assay, using various methods, on predicted deposits, then follows, providing a data miner’s version of geological ground truthing.
Decision trees come in two broad flavors
Categorical and/or binary decision trees. This type trains on a categorical target variable or binary targets, in our case, economic levels of minable deposits across wide areas of exploration as these algorithms can handle both various statistical distributions and ground faults.
Regression algorithm style decision trees. This type works on continuous grade variables, similar to kriging, i.e. ore grade changes. Regression algorithms are applicable on specific orebody statistical distributions within fault boundaries.
Eureka’s approach to wide area exploration focuses most on the first type of decision tree. It is an approach developed over many years. In 2003, we published a British Columbia province-wide exploration decision tree data mining prediction result for high production oil and gas wells in The Oil and Gas Journal. This was the first time a data mining technique applied to exploration in geology was published. In 2004, geological data mining work on Exxon Mobil’s $12.5-billion program revealed three deposits within a 50-sq. km study area. These deposits today are working heavy oil sand mine at Kearl Lake, Alberta. Between 2020, until present, this approach was scaled up using a supercomputer to planetary level searches for gold, nickel and copper.
What does the resulting data mined metal classification look like? The first example is a 10 g/t gold classification in southern Nova Scotia, as compared to a simplified geological picture (See Fig. 2).
Next is a greenfield gold resource, next to a 500 kV high tension power line in British Columbia. This data was mined using BC litho-geochemistry data trained to see what gravity anomaly map portions contain potential gold resources (See Fig. 3). This approach also works to expand on resource discovery around brownfield sites.
An interesting point: the data mining algorithm shows a folding rock pattern of orebodies typically found in quartz in metasedimentary rock of Meguma Group geology which is associated with high grade visible gold in Nova Scotia (Fig. 6). The data mining algorithm has independently discovered a deposit shape at Rawdon mine that geologists in Nova Scotia are familiar as an observed deposit fold shape that holds visible gold in quartz.
What to expect with better exploration forecasting?
There will be a much shorter lag time from exploration to commercial mines with the improved forecasting of deposits. A more accurate method to forecast mines means much shorter lag times for mine development.
Exploration programs using traditional methods for base metals typically have 15-to-20-year lag periods before beginning production. The lag time is expected to be less than five years from prospected resource to mine development using a method that gets drilling programs onto the resource quicker, and funding, when combined with the NI 43-101 process which requires a company to file a technical report at certain times, prepared in a prescribed format.
To meet the upcoming demand for electrical vehicles and
the decarbonization of our mining industry, the world needs far more metals. Supercomputer-level data mining across wide areas is one way to meet upcoming metal base metal demand.
James Cormier-Chisholm, PGeo., BSc geology, environmental diploma, and MBA, owns Eureka Maps, a geological data mining company. He became interested in data mining after working on the Voisey’s Bay nickel project as a consultant. Contact him via email@example.com