One of the most powerful features in GeoInsights 3D is the geochemical clustering tool. While detailed geological logging is valuable, it often faces challenges:
- Inconsistency between geologists
- Variable logging quality
- Subtle mineralogical changes not visible to the naked eye
- Time-consuming manual review of large datasets
Geochemical clustering provides a rapid, objective first pass at identifying major rock types and alteration domains. GeoInsights 3D offers two approaches: straightforward K-means clustering of your raw data, or a more advanced option that first transforms the data using Principal Component Analysis (PCA) before clustering. Both methods have their merits — direct clustering is simpler and more transparent, while PCA transformation can help reduce noise and highlight subtle patterns.
To demonstrate the core workflow, we’ll use the simpler K-means approach on a dataset where the geology consists of a weathered profile of transported cover, saprolite and saprock.
Remember, while clustering is powerful, it should always be integrated with geological understanding and validated against known geology, structure and controls on mineralisation.
Feature Selection: Before clustering, it’s important to select appropriate elements. Using biplot and correlation matrix visualisations, we can identify key elements while removing those that are redundant or noisy. As geochemical data typically shows skewed distributions, we apply a natural log transform to normalise the data.
Pro Tips:
- Start with major elements (Al, Fe, Ca, Mg, Na, K) for broad lithological domains
- Add pathfinder elements to refine alteration and mineralisation patterns
Clustering: Clustering: Applying k-means clustering to the transformed geochemical data revealed three distinct domains. While the weathering profile contains additional complexity, this three-cluster solution provides a simple test case, capturing the main lithological units of interest. The choice is supported by both the scree plot’s elbow at k=3 and the coherent spatial groupings that form in the 3D viewer
Statistical Validation: Summary statistics and box plots reveal distinct geochemical fingerprints for each cluster:
- Cluster 0 (Saprock): Elevated base metals
- Cluster 1(Lower saprolite): Lower concentrations of mobile elements (Na, Mg, and Ca) compared to cluster 0
- Cluster 2 (Cover): Depleted in base metals and mobile elements
Comparison with Logged Data: The lithology vs cluster heatmap provides a quick correlate clustering results against traditional logging. For example, the heatmap allows us to cross check our clusters against logged geology: Cluster 0 correlates strongly with the saprock (rssr) as well as the minor andesite (vanv) and monzonite (icmo) units, Cluster 1 correlates with the upper (rssu) and lower (rssl) saprolite units, and Cluster 2 correlates most strongly with the logged gravel (ssgr) units.
Spatial Validation: The critical final step is checking that clusters form coherent geological shapes in 3D space — random or scattered patterns would indicate poor clustering results. In our case, these distinct geochemical signatures generally align well with the logged weathering boundaries. The clusters display strong horizontal continuity, reflecting the expected geometry of weathering horizons.
When viewed in 3D alongside the lithological logging, the geochemical domains show better hole-to-hole correlation than the original logging, particularly in the southeastern portion of the project. Here, two holes show significant logging inconsistencies compared to the broader dataset — a common challenge when multiple geologists are involved in logging weathered material.
While visual logging can effectively identify these weathering horizons, determining the exact boundaries between lower saprolite, upper saprolite, and saprock units often remains subjective. The objective nature of geochemical clustering helps standardise these boundary determinations. If we wanted to delve into further detail, applying k-means with k=4 would reveal distinct geochemical differences between the northern and southern saprolite units, reflecting variations in weathering intensity across the deposit.