Indicator Kriging for Cutoff Grade Decisions: When and How

Ordinary kriging (OK) is the workhorse of resource estimation. It’s what most of us reach for, it’s what most software defaults to, and for most deposits it’s the right answer. But there’s a class of problems where OK quietly fails — and where indicator kriging (IK) is the tool that actually answers the question you’re asking.

The question is usually some version of “what’s the probability that this block is above cutoff?” OK answers a different question: “what’s the expected grade of this block?” Those sound similar. On a well-behaved deposit with smooth grade distributions, they give the same answer. On a nuggety, skewed, or multi-population deposit — which describes half the gold projects I work on in Indonesia — they don’t.

This post covers when IK beats OK, how to build the binary transforms and probability maps, how to select thresholds, and a worked example on an Indonesian gold deposit.

What indicator kriging actually does

IK transforms the grade data into a series of binary indicators. For each threshold (cutoff) you choose, every composite becomes either 1 (above cutoff) or 0 (below cutoff). You then krig the indicators — not the grades — to get, for each block, a value between 0 and 1 that represents the probability the block is above that cutoff.

That’s the key word: probability. OK gives you an expected value. IK gives you a probability distribution. From the IK output, you can:

• Map the probability that any block is above cutoff — directly useful for mine planning
• Reconstruct the local grade-tonnage curve, accounting for the local data distribution
• Quantify the uncertainty, not just the central estimate

The trade-off: IK is more work. You run a separate variogram and a separate kriging pass for each threshold. Five thresholds means five variograms and five kriging runs. The variograms are also noisier (binary data has less information than continuous data), and the results can be sensitive to threshold selection.

When IK beats OK

Reach for indicator kriging when one or more of these apply:

1. High nugget effect

When the nugget is more than 40–50% of the sill, OK smooths aggressively and the block estimates cluster around the mean. High-grade blocks disappear. On a high-nugget deposit, the question “is this block above 2 g/t?” is more answerable than “what is the grade of this block?” — because the grade estimate has too much variance to be useful, but the probability estimate is still informative.

If you’re seeing high nugget values on your variogram (covered in the variography visual guide), that’s a signal to consider IK.

2. Strongly skewed or multi-modal distributions

OK assumes the local distribution is roughly symmetric. On a lognormal gold distribution with a long tail, the kriging estimate is pulled toward the mean and the high-grade blocks are underestimated. IK doesn’t assume symmetry — it works on the indicator transforms, which are binomial and well-behaved regardless of the underlying grade distribution.

On multi-modal distributions (which usually mean you have a domaining problem, but sometimes the populations genuinely overlap in space), IK lets you estimate the probability of each population separately.

3. The decision is binary

If the mining decision is “above cutoff → ore, below cutoff → waste,” then the probability of being above cutoff is the quantity that drives the decision. The expected grade is a step removed. IK gives you the probability directly.

This is especially relevant for selective mining — when the SMU is small and the mine is making per-block ore/waste calls, the probability map is more useful than the grade estimate.

4. You need local grade-tonnage curves

OK gives you one global grade-tonnage curve. IK gives you a local grade-tonnage curve for every block — the distribution of possible grades at that location, conditional on the surrounding data. This matters for mine planning: the risk of a block falling below cutoff at mining is different in a high-probability area vs a marginal area, even if the OK grade estimate is the same.

When OK is still the right answer

IK is not always better. Use OK when:

• The grade distribution is roughly symmetric (most porphyry Cu deposits, most laterite Ni)
• The nugget is low (<30% of the sill)
• The deposit is well-drilled and the local data density is high (OK’s smoothing is less of a problem when you have lots of data)
• You need a single grade estimate per block for downstream processing (most resource statements)

For most Indonesian porphyry and laterite projects, OK is the right tool. For most Indonesian epithermal and orogenic gold projects, IK deserves serious consideration. The decision is geology-driven, not software-driven.

Building the indicator transforms

The mechanics:

1. Choose your thresholds. These should correspond to meaningful cutoffs — the current economic cutoff, plus 2–4 thresholds bracketing it. For a gold deposit with a 1.0 g/t cutoff, I’d typically use 0.5, 1.0, 2.0, 5.0, and 10.0 g/t.
2. Transform the composites. For each threshold, create a new column where each composite is 1 if grade ≥ threshold, 0 otherwise.
3. Run variography on each indicator. The indicator variogram is different from the grade variogram — typically shorter range, higher nugget. Model each one.
4. Krig each indicator. For each block, you get a probability estimate (0 to 1) for each threshold.
5. Reconstruct the local distribution. The ordered probabilities at the thresholds define a discrete distribution. From this, you can calculate the expected grade (the IK mean), the probability above any cutoff, and the local grade-tonnage curve.

A note on thresholds

Threshold selection matters. Too few thresholds and the local distribution is too coarse to be useful. Too many and the indicator variograms become noisy (each indicator has less data). Five to seven thresholds is the practical range.

Choose thresholds that:

• Bracket the economic cutoff (so the decision-relevant probability is well-resolved)
• Span the grade distribution (don’t cluster them all at one end)
• Have enough data above each threshold (a threshold above the P99 has ~3 ones and the rest zeros — the variogram is meaningless)

Worked example — Sulawesi epithermal Au

A project I worked on last year: low-sulfidation epithermal Au in Central Sulawesi. 65 holes, 2m composites, single vein domain. Grade statistics:

• N = 380 composites
• Mean = 3.2 g/t Au
• Median = 0.8 g/t Au
• P95 = 14.5 g/t Au
• P99 = 48 g/t Au
• Max = 185 g/t Au
• Nugget = 45% of sill (high)

Classic skewed, high-nugget epithermal. OK was producing a model where the high-grade core of the vein was smeared into a blob at ~2 g/t and the actual bonanza intercepts were invisible. The mine planner couldn’t tell which blocks were genuinely high-grade vs which were averaged-up waste.

I ran both OK and IK and compared. Thresholds: 0.5, 1.0, 2.0, 5.0, 10.0, 25.0 g/t.

Note what the indicator variograms tell you: the high-grade indicators have shorter ranges and higher nuggets. The high-grade mineralization is less spatially continuous than the low-grade halo — geologically sensible for a vein system where the bonanza shoots are local and the low-grade halo is broad.

The OK model and the IK model at the 1.0 g/t cutoff told different stories:

• OK: a smooth 2–3 g/t blob covering the vein footprint. Every block in the vein was “above cutoff” with a grade estimate of 1.5–3.5 g/t. The mine planner saw uniform ore.
• IK: a probability map where the core of the vein had >90% probability of being above 1.0 g/t, the margins dropped to 50–70%, and the edges were 20–40%. The mine planner saw a clear high-confidence core and a marginal halo — and could plan the mining sequence and the dilution accordingly.

The IK model also gave a local grade-tonnage curve per block, which let the mine planner quantify the risk on the marginal blocks: “this block has a 60% probability of being above cutoff, so 40% chance it’s waste — plan the haul road accordingly.”

Top-cut interaction

IK and top-cutting interact. The indicator transform is robust to outliers — a 185 g/t composite and a 25 g/t composite both become 1 at the 10 g/t threshold. This is one of IK’s advantages: it’s less sensitive to the top-cut decision than OK.

But you still need to be careful. If you’re reconstructing the local grade distribution from the IK probabilities, the high-grade tail of that reconstruction depends on the highest threshold. If your top threshold is 25 g/t and you have a few 100+ g/t composites, the reconstructed distribution above 25 g/t is unconstrained. The IK mean (expected grade) can still be inflated by these values.

The practical approach: apply the top-cut to the grade data before running IK (for the mean reconstruction), but recognize that the probability maps at the lower thresholds are top-cut-insensitive and can be used directly for mine planning.

Post-processing — what to do with the IK output

The IK output is a set of probability maps (one per threshold). The useful derivatives:

• Probability above economic cutoff: the primary mine planning map. Blocks with >70% probability → confident ore. 50–70% → marginal, plan for selectivity. <50% → probable waste.
• IK mean grade: the expected grade, reconstructed from the indicator probabilities. Comparable to the OK estimate but often less smoothed on high-nugget deposits.
• Local grade-tonnage curve: per block, the distribution of possible grades. Useful for risk-adjusted mine planning.
• Classification input: the probability spread (how peaked vs flat the local distribution is) is a measure of uncertainty that can feed into resource classification — high probability = high confidence = Indicated/Measured; spread-out probability = lower confidence = Inferred.

How Orebit GeoSuite helps

The GeoSuite Resource Estimation module supports both OK and IK, with the IK workflow fully integrated:

• Threshold selection assistant — suggests thresholds based on the grade distribution (percentiles, economic cutoff, natural breaks)
• Indicator variography — per-threshold experimental variograms with model fitting, side-by-side with the grade variogram for comparison
• Parallel kriging — runs all indicator kriging passes in parallel (5 thresholds in roughly the time of 2 sequential OK passes)
• Probability map output — per-threshold probability maps, with the economic cutoff highlighted
• Local grade-tonnage reconstruction — per-block conditional distributions, with the IK mean and the probability above cutoff
• OK vs IK comparison — side-by-side grade maps and grade-tonnage curves, so you can see whether IK is actually adding value on your deposit
• Validation suite — the same seven checks from the block model validation checklist, applied to the IK model

The OK vs IK comparison is the feature I use most. On three of the last five gold projects I’ve worked on, IK clearly added value. On the other two, OK was fine and IK was unnecessary overhead. The comparison tells you which is which — you don’t have to guess.

Try Core free → · Get the bundle — Rp 99K → (single modules Rp 49K · one-time · yours forever)

Bottom line

Indicator kriging answers a different question than ordinary kriging — probability of being above cutoff, not expected grade. On high-nugget, skewed, or binary-decision deposits (most Indonesian gold), that’s the question you actually need answered. On well-behaved deposits (most Indonesian porphyry and laterite), OK is fine and IK is overhead.

The decision is geology-driven. Run the comparison. If IK gives you a materially different and more useful picture — probability maps, local grade-tonnage, uncertainty quantification — use it. If the OK and IK models agree, save yourself the extra variograms and ship the OK model.

Either way, the question you’re answering is “what’s the probability this block is ore?” — not “what’s the grade?” The grade is a means to that end, not the end itself.

Working on a high-nugget gold deposit where OK is smoothing away your high grades? Email hello@orebit.id with the variogram and the grade distribution — I’ll tell you whether IK is worth the extra work on your project.