Multi-element geological data clustering for domaining
Multi-element clustering groups samples by how similar they are across all measured elements and position at once, instead of one element at a time, to help define geological domains.
In GEOREKA you choose a reference sample, and the tool scores every other sample from 0 to 100 for similarity to it across all numeric attributes plus the X, Y, Z position. A score of 100 means perfect similarity and normally applies only to the reference point itself. Plotting those scores in 3D can reveal patterns that single-element kriging misses, such as a gold halo around a copper core. The tool is still a work in progress.
Background: single-element interpolation
In many deposits the orebody outline is made by interpolating the element of interest with kriging, which is a weighted average of nearby points, using distance and a variogram. Kriging is generally applied to one element at a time, and although multi-element methods exist, they are complex and rarely used across many elements.
The approach: similarity in a higher-dimensional space
Instead of clustering in 3D space alone, the method treats every numeric attribute as a dimension: element concentrations such as As, Au, Ag, Mo, Pb and Zn, plus X, Y and Z. In this higher-dimensional space, points are grouped by similarity. You select a reference point, and the tool estimates a similarity score for every other point, scaled from 0 to 100.
Example: the Parker Challenge dataset
With two reference points, one in a high-grade copper area and another about 390 m away at a high-grade gold location, the copper clustered as expected. The gold did not: its cluster appeared as a halo around the high-copper area. Gold, arsenic and molybdenum were correlated, which may mean that gold domains need their own geostatistical treatment.
Example: geochemistry with geophysics
In a second example, geological data was combined with geophysical measurements, conductivity and gamma radiation. The tool separated ore from waste and found a significant third group, which suggests that lithological units could be predicted from geophysics using a single reference drillhole.
Limitations
- The choice of reference point strongly influences the result, and its sensitivity needs more study.
- Categorical data is not used, because it is subjective and not continuous.
- Interpreting the patterns depends on data quality and geological understanding.
A full walk-through is available as a video.
Apply it
Cluster your own multi-element data
Interactive multi-element analysis in GEOREKA links 2D and 3D plots, so a pattern in the chemistry can be located in the model. Geological modelling in GEOREKA →
Common questions
Multi-element clustering FAQ
What is domaining?
Domaining is deciding which samples belong to which geological domain, so that each domain can be modelled and estimated as a separate population.
How is multi-element clustering different from kriging?
Kriging usually interpolates one element at a time. Multi-element clustering compares all numeric attributes and the position together, scoring each sample against a chosen reference.
Can geophysical data be included?
Yes. In the example, conductivity and gamma radiation were combined with the geochemistry, and the tool separated ore, waste and a third group.