Resource classification methods: drill spacing, simulation precision and Nyquist sampling adequacy

What classification needs to express

Under JORC and NI 43-101 (CIM definitions), classification expresses confidence in geological and grade continuity. The codes do not prescribe a formula, so practitioners use a mix of drill spacing, number of informing holes, estimation quality and judgement. A tool can propose classes consistently across millions of blocks, but the choice of thresholds and the final report remain with the Competent or Qualified Person.

Lens 1: distance to drilling

The classic approach assigns a class from the distance to the nearest samples, and often the number of holes within that distance. GEOREKA measures the distance in three ways: a simple isotropic distance, a distance under one global anisotropy ellipsoid, and a distance under the local anisotropy field. Under LVA, a block that is close to a hole across strike but far along strike is treated as far if continuity is longer along strike, which is what the geology says.

Lens 2: simulation precision (DHSA)

Drillhole spacing analysis by conditional simulation runs many equally likely realisations of the grade and measures the variation in each block, as a coefficient of variation or as an interval such as P10 to P90 relative to the median. Blocks whose realisations agree closely are more precisely known. Thresholds are set as percentages. The method answers how precise the estimate is at the current spacing, but it needs a simulation run first (see conditional simulation).

Lens 3: Nyquist sampling adequacy

The Nyquist–Shannon sampling theorem says a signal can only be recovered if it is sampled often enough compared with its fastest variation. Lu and Wang (2026) applied this to geospatial prediction. If drill spacing is too coarse relative to the scale of variation, the fine structure is aliased and no estimator can recover it. This sets a floor on accuracy that belongs to the data configuration, not to the model.

GEOREKA turns this into a number for each block, ε: the fraction of the grade variance that the local spacing cannot resolve. It depends on only two things, the covariance kernel and the ratio of local drill spacing to the range of continuity. The kernel matters: a rough exponential kernel has much more variance at fine scales than a smooth one, and needs much denser drilling for the same ε.

• Kernel-matched. ε is calculated for the kernel your interpolation used, with the range read from the interpolation run for each domain, so classification and estimation speak of the same continuity.
• Locally anisotropic. Local spacing is measured under the LVA field, so under-drilling across strike raises ε automatically.
• Thresholds. Blocks are labelled Measured, Indicated or Inferred from ε limits. Because the absolute value of ε depends on the kernel and the deposit, the thresholds can also be set as percentiles of each domain’s own ε distribution, which gives a relative classification that adapts to the deposit.

Each block also records ε and an error floor, so the result can be inspected and not just coloured.

How the three compare

	Distance and count	Simulation precision	Nyquist adequacy
Question	How close is the drilling?	How precisely is the grade known?	Can the drilling resolve the structure?
Needs a simulation	No	Yes	No
Depends on the estimator	No	Yes	No, it is a floor
Uses LVA	Optional	Through the simulation	Yes, for local spacing
Cost	Low	High	Low

Reading them together

The lenses are complementary. If precision is poor and Nyquist adequacy is also poor, the problem is the sampling and only more drilling will help. If Nyquist adequacy is good but precision is poor, the estimator or the domain is more likely the limit. Where all three agree, the classification is easier to defend.

What to be careful about

• Nyquist adequacy depends on the covariance range. A range fitted to sparse data tends to be too long, which understates ε, so use a cross-validated range.
• The checks measure sampling and estimation quality. They say nothing about geological or QA/QC risk, which a class also reflects.
• Classification output should be smoothed and reviewed. Isolated blocks of a higher class inside a lower one rarely survive a reasonable Competent Person review.

Keep reading

Classification relies on a range and kernel from a validated interpolation, see validating a resource estimate, and on the local anisotropy field in LVA explained.