Sign in Subscribe
Mining industry

Evaluating the Uncertainty and Predictive Performance of Probabilistic Models in Mining Geology

Strategic Insights

— 2 min read

Researchers from the University of Sydney have made a significant contribution to the field of mining geology by developing a holistic approach to assess the uncertainty and predictive performance of probabilistic ore grade models. The research, supported by the Australian Centre For Robotics and Rio Tinto, aims to bridge the gap between model development and deployment by providing industry guidelines for evaluating the uncertainty and predictive performance of probabilistic ore grade models. The proposed model assessment targets three objectives, ensuring that predictions are reasonably calibrated with probabilities, facilitating large-scale simultaneous comparisons for multiple models across space and time, and objectively measuring the spatial fidelity of models.

Key Takeaways:

  • The research aims to develop a holistic approach to assess the uncertainty and predictive performance of probabilistic ore grade models in mining geology.
  • The proposed model assessment targets three objectives: ensuring calibrated predictions, facilitating large-scale comparisons, and objectively measuring spatial fidelity.
  • The research uses extensive data from a real copper mine in a grade estimation task and demonstrates the proposed methods using ordinary kriging and Gaussian process models.
  • The assessments are underpinned by statistics that evaluate the model's predictive distributions relative to the ground truth.
  • The proposed methods enable competing models to be evaluated consistently and the robustness and validity of probabilistic predictions to be tested.
  • The research highlights the increased difficulty of future-bench prediction (extrapolation) relative to in situ regression (interpolation).

The experiments were designed to emphasize data diversity and convey insights into the challenges of predicting future-bench data. The research makes cross-study comparison possible, irrespective of site conditions, and provides a standardized and interpretable way to evaluate the uncertainty and predictive performance of probabilistic models.

Statistics:

  • The research uses extensive data from a real copper mine in a grade estimation task.
  • The proposed methods are demonstrated using ordinary kriging and Gaussian process models.
  • The assessments use statistics that evaluate the model's predictive distributions relative to the ground truth.
  • The proposed methods enable competing models to be evaluated consistently.
  • The research highlights the increased difficulty of future-bench prediction (extrapolation) relative to in situ regression (interpolation).

Sources:

  • Evaluating the Uncertainty and Predictive Performance of Probabilistic Models Devised for Grade Estimation in a Porphyry Copper Deposit. Modelling, 2025,6(2):50 (Available at https://doi-org.sdpl.idm.oclc.org/10.3390/modelling6020050).
  • NewsRx. University of Sydney Researchers Update Understanding of Modelling (Evaluating the Uncertainty and Predictive Performance of Probabilistic Models Devised for Grade Estimation in a Porphyry Copper Deposit). Journal of Engineering. July 7, 2025; p 5997.

Read more