January 3rd 2022

Using mathematics to find an important rock type


Håkonhals pegmatite mine in the Tysfjord area in Hamarøy Municipality. Photo credit: Axel Müller, Universitetet i Oslo (UiO)
Machine learning can be used to find a type of rock called pegmatite, which contains minerals and elements that are important to the green transition.

In a new study published by the research project GREENPEG, researchers conclude that using mathematical algorithms on satellite data can be a good technique for locating areas with pegmatites. That can make it quicker to start analysing potential new areas for prospecting. 

A green energy revolution will require large quantities of a variety of minerals, such as high-purity quartz, lithium, rare-earth metals, beryllium, tantalum and caesium. All of these can be found in pegmatites. 

Initial “screening”

“Yes, it turns out that mathematical algorithms can help to identify areas with possible pegmatites without you actually having to physically investigate them”, says Marco Brönner, who leads the Geophysics Section at the Geological Survey of Norway (NGU). 

“The method can be used for an initial ‘screening’ of an area in conjunction with prospecting, prior to mining companies surveying the pegmatite deposits more thoroughly for possible exploitation”, believes Brönner.

From space

Tysfjord in Nordland was chosen as the Norwegian testing area, since it has known pegmatite deposits, which make it possible to test whether the algorithms work. Researchers used images from the Sentinel-2 satellites to create maps of the various elements they identified – without knowing what those elements represented. 

By using geological maps, aerial photographs and geophysical data from airborne surveys, it was possible to determine what they actually were. The led to them being classified into four types: pegmatite, granite, water and vegetation. 

Not just pegmatites

In the article Proceedings of SPIE – Identification of pegmatite bodies, at a province scale, using machine learning algorithms: preliminary results, the researchers describe how you can build a mathematical model that automatically recognises the four classes, and how it is possible to test the model. 

“We must continue working to improve the procedure to reduce the level of misclassification. In the future, we would also like to be able to identify weathered pegmatite”, says Brönner, who points out that satellite data has enormous potential, and not just for pegmatites:

“In principle we can already distinguish between different types of rock by using multispectral and hyperspectral imagery from, for instance, an Earth observation satellite operated by the American space agency NASA.” 

Financed by the EU

Meanwhile, in 2022 the European Space Agency (ESA) will start the Copernicus Hyperspectral Imaging Mission (CHIME), with an even wider spectral range capable of mapping more mineral types. NGU already uses hyperspectral imaging on core samples, so it is in a position to exploit this new opportunity. 

“I expect the use of hyperspectral data from satellites and planes, together with machine learning, to enable far more efficient geological mapping in the future. That will be useful for NGU, but also for the mineral industry and land use planning”, says Section Manager Marco Brönner.

The study was financed through the EU’s research programme Horizon 2020 as part of the project New Exploration Tools for European Pegmatite Green-Tech Resources (GREENPEG).

References: Ana Claudia M. Teodoro, Douglas Santos, Joana Cardoso-Fernandes, Alexandre Lima, Marco Brönner, “Identification of pegmatite bodies, at a province scale, using machine learning algorithms: preliminary results,” Proc. SPIE 11863, Earth Resources and Environmental Remote Sensing/GIS Applications XII, 1186308 (12 September 2021); doi: 10.1117/12.2599600 Link: Identification of pegmatite bodies, at a province scale, using machine learning algorithms: preliminary results (spiedigitallibrary.org)