A leading global mining company needed more reliable density estimates for a major undeveloped high-grade iron ore project to support resource modelling and mine planning.
Datarock developed machine learning models integrating drilling, assay, and geophysical data to deliver high-resolution density predictions across the resource.
Challenge
- Averaging densities across large domains masked local variability.
- Misalignment between samples, assays, and wireline geophysics introduced noise.
- High porosity and friable material in RC holes produced unreliable readings and sampling bias.
Solution
- Two complementary models developed to predict density and porosity across the resource.
- Primary model integrates assay data, normative mineralogy, material types, geo-domains, and wireline geophysics.
- Secondary model uses the same inputs minus wireline geophysics, for areas with sparse or no coverage.
- All sources merged into consistently spaced composites and validated via spatial cross-validation.
Result
- Density predictions generated across 1,000+ drill holes and hundreds of thousands of metres, spanning multiple deposits.
- Both models generalised effectively to new drilling data.
- 3D visualisations confirmed capture of known structural trends.
- Spatial error maps highlight zones of lower confidence and flag where additional sampling adds value.
What this means for resource modelling
The models deliver density and porosity estimates at 2 m intervals, replacing broad domain averages with high-resolution predictions. The result is consistent, defensible inputs for resource estimation, mine planning, and economic modelling , with built-in guidance on where additional drilling would improve confidence.
