top of page
Image by SJ Objio
Title: Predictive Modelling of Nickel Potential Using Multisource Information and Random Forest
Organization

Vale - University of Granada

Period

June 2022 - June 2025

Status

Completed

Est. Budget

USD 2.2M

Est. Business Benefit

USD 25.0M

Indicative Value-to-cost

11.4x

Keywords

Nickel prospectivity, Random Forest, GIS, multisource data, mineral exploration, target ranking

Benefit basis

Estimated value from improved exploration targeting, reduced low-value drilling, better prospect ranking, and increased probability of identifying economically relevant nickel targets.

Business Challenges 
Mineral exploration requires robust models that can integrate complex geological, geochemical, geophysical, structural, spatial, and historical exploration information. Traditional prospectivity methods may be limited when relationships between variables are nonlinear or when data sources have different scales and statistical properties. The challenge was to map nickel potential more accurately and prioritise exploration targets in a data-rich but geologically complex environment.
Expanded Technical Solution
The project applied Random Forest modelling to nickel prospectivity mapping for Vale's Sudbury-related exploration context in Ontario, Canada. Multisource datasets were integrated into a spatial modelling environment and used to predict the probability or potential of nickel deposit occurrence. Random Forest was selected because it handles nonlinear relationships, categorical and continuous predictors, variable interactions, complex distributions, and variable-importance ranking. Logistic Regression was also used as a benchmark to validate the performance and interpretability of the Random Forest approach.
Technical Work Packages/Methods
  • Integrated geological maps, lithology, alteration, structural features, geophysical grids, geochemical samples, known mineral occurrences, drilling data, remote-sensing layers, and spatial derivatives.
  • Prepared GIS-based training datasets with positive occurrences and background/negative samples.
  • Developed Random Forest classification/regression models for nickel prospectivity and compared performance against Logistic Regression benchmarks.
  • Used cross-validation, spatial validation, ROC/AUC, precision-recall, calibration, and uncertainty mapping to assess model reliability.
  • Generated variable-importance rankings and interpreted them against known geological controls.
Key Deliverables
  • Integrated multisource nickel exploration dataset.
  • Random Forest prospectivity model and Logistic Regression benchmark.
  • Nickel prospectivity maps with confidence/uncertainty indicators.
  • Target-ranking methodology and geological interpretation of key predictive variables.
Results and Value Created
The project produced a data-driven method for prioritising nickel exploration targets and showed that Random Forest can integrate large multisource exploration datasets effectively. Key variables identified by the model aligned with geological expectations, improving confidence in the approach. The estimated benefit reflects more focused exploration spending, better target prioritisation, reduced drilling of low-probability areas, and increased probability of identifying high-value nickel mineralisation.
bottom of page