Tin-Tungsten Prospecting
with Machine Learning
A regional study applying machine learning to integrate geological and geophysical datasets and identify areas with similar signatures to known Sn–W mineralisation.
- Clustered mineralisation: Most occurrences are concentrated in a few districts, limiting detection beyond known areas.
- Subtle and inconsistent signals: Some deposits have weak or variable geophysical expression.
- Noisy and uncertain data: Correlated datasets, vegetation effects and inconsistent occurrence locations reduce data reliability.
- Machine learning prospectivity workflow: Built using 222 in-situ occurrences and 30 m resolution geological, geophysical and radiometric data.
- Feature reduction and data integration: PCA reduced inputs from 21 to 16 variables, improving signal quality.
- Spatially robust modelling: 222 CatBoost models trained using a 2 km holdout approach and combined into a single prospectivity map.
- Measured predictive performance: Models correctly identified 45% of known occurrences.
- Strong detection in clustered systems: Performance was highest in major mining districts.
- Key geological drivers identified: Gravity, elevation, magnetics, distance to granite and radiometric signals were most influential.
Read our blog article related to our Prospectivity Modelling
Trusted by exploration teams worldwide.
Uncover subtle, complex exploration signals across geochemistry, geophysics, and remote sensing datasets that traditional methods miss.
Hyperspectral
and computer vision
expertise
Specialised modelling and texture analysis unlock high-value mineral signatures in airborne and satellite datasets for sharper targeting.
Transparent models
with reduced risk
Our explainable analytics clearly
quantify uncertainty, validate models,
and improve target confidence.
”Datarock regional prospectivity models have been integrated into our global generative prospectivity workflows
Principal GeoscientistMajor mining company
Get your prospectivity consultation
Understand where to focus next, with clear, data-driven target ranking.



