Physics-informed AI for mineral prospectivity

Last updated: August 2026

About

Machine-learning methods can identify patterns in large exploration datasets, but a purely data-driven model can reproduce sampling bias or correlations that have little relationship to mineral formation. A useful prospectivity model needs to explain why a region is favourable and to represent uncertainty in both data and physical interpretation.

This project will combine geological observations with plate reconstructions, mantle-convection outputs, and physically informed indicators of tectonics, melting, and landscape evolution. It will develop interpretable methods that identify the role of different controls rather than only assigning a prospectivity score.

Research questions

• How can geodynamic knowledge improve statistical models of mineral prospectivity?
• Which indicators contain causal information, and which reflect sampling bias?
• How can uncertainty in reconstructions propagate into prospectivity estimates?

Methods and data

Projects can use geological maps, geophysics, geochemistry, mineral occurrences, plate reconstructions, geodynamic model outputs, interpretable machine learning, and uncertainty quantification.

Possible projects

Possible directions include building a physics-informed prospectivity model for a selected commodity, testing transferability between provinces, or separating geodynamic controls from reporting and sampling bias.

Essential background

Useful preparation includes geology, geophysics, data science, computer science, mathematics, statistics, or programming.