EarthByte Group develops machine learning recipe to find copper-gold deposits along the Andes

In a paper just published in the journal Tectonics, EarthByter and Natural Sciences, University of Sydney alumnus Nathaniel Butterworth and colleagues from the School of Geosciences, University of Sydney and Data61/CSIRO have developed a spatio-temporal machine learning recipe to identify subduction zone tectonic environments in which porphyry copper-gold deposits tend to form. The new approach could pave the way for future mineral discoveries.

The Andes are the longest mountain chain in the world, and host some of the largest copper (± gold) deposits in the world. These deposits are generally associated with intrusive-extrusive subduction zone magmatism. Subduction has been active along South America for hundreds of millions of years, but known copper-gold deposits have formed only during certain time periods, concentrated along particular stretches of the margin. This raises the question: what are the particular tectonic environments leading to ore deposit formation?

In a collaborative project, geophysicists from the EarthByte Group and machine learning experts from Data61 developed a novel data mining approach to isolate the tectono-magmatic parameters leading to the formation of metalliferous copper-gold deposits during subduction. The group used a plate tectonic model to compute convergence rates and directions, as well as the age of the downgoing plate through time.

They tested two alternative supervised machine learning methods to analyse over a dozen tectonic parameters in terms of their association with copper-gold ore deposits versus regions and geological periods barren of major deposits. The two methods were the random forest (RF) ensemble and support vector machines (SVM). They found that SVMs are superior for this particular application and determined that a combination of four key parameters is linked to ore deposit formation:

1) Rapid convergence rates of ~100 km/Myr,
2) Subduction obliquity of ~15°,
3) a subducting plate age between ~25–70 Myr, and
4) a location far from the lateral subducting trench edge (>2000 km) – meaning the likelihood of finding these ore deposits increases towards the central Andes as opposed to their southern or northern extremities.

These parameters are significant because they are linked to the thickness of oceanic sediments during subduction, the nature of small-scale convection around the subduction zone, and the availability of partial melt in the mantle wedge. When coupled, these parameters influence the genesis and exhumation of porphyry copper deposits. The data and python scripts used for the analysis are available as supplementary material with the paper, in the spirit of enabling Plate tectonics 2.0

The paper can be downloaded here.