Variogram Modeling: 5 Mistakes That Make Your Kriging Wrong

Variograms are the part of resource estimation where geologists get uncomfortable. The math feels opaque, the software defaults are seductive, and the connection between the variogram model and the final estimate isn’t always obvious. So people click buttons, accept the auto-fit, and move on.

I understand the temptation. But your variogram IS your kriging estimate. The variogram tells the algorithm how similar samples are as a function of distance and direction. Get the variogram wrong and every single block estimate is biased — not randomly, but systematically, in a way that only shows up when the mine starts digging and the model doesn’t match reality.

After 12 years of reviewing variogram models for Indonesian gold and copper deposits, here are the five mistakes I see most often.

Mistake 1: Using the default search direction

Most software defaults to calculating the variogram along the drillhole axis — usually vertical. For most Indonesian deposits, the mineralization is NOT vertical. It’s along a shear zone dipping 60° to the northeast, or a breccia pipe plunging 30° to the south.

If you calculate the variogram along the wrong direction, you’re mixing mineralized and unmineralized samples. The variogram looks terrible — high nugget, no structure, short range. So the geologist increases the lag distance, the variogram smooths out, and nobody realizes the model is estimating along the wrong axis.

Fix: Always calculate the experimental variogram along the mineralization direction. Use a directional variogram with:

• Strike: parallel to the mineralized trend
• Dip: down the mineralized plane
• Across: perpendicular to the mineralization

If you don’t know the direction, calculate a variogram map (a 2D surface showing variogram values in all directions) and identify the anisotropy axes visually.

Mistake 2: Ignoring the nugget effect

The nugget effect is the variogram value at zero distance. It represents variability at scales smaller than your sample spacing — micro-scale geological noise, sampling error, assay error.

Indonesian gold deposits, especially epithermal systems, have high nugget effects — often 40-60% of the total sill. This is normal. It tells you that gold distribution is erratic at short distances, which is exactly what you’d expect in a vein system.

The mistake is trying to reduce the nugget by:

• Increasing the lag distance — this hides the nugget but doesn’t eliminate it. Your model still has short-scale variability; you’re just not modeling it.
• Compositing to longer lengths — this reduces the nugget by averaging out short-scale variability, but it also reduces the resolution of your model.
• Using a single composite length for all domains — a 1m composite in a narrow vein has a different nugget than a 2m composite in a disseminated zone.

Fix: Accept the nugget. A high nugget is geological information, not an error. If your nugget is 50% of the sill, that means half the variability in your deposit is at scales smaller than your drill spacing. Your kriging will be conservative — it will smooth estimates toward the mean — and that’s correct. Don’t fight it.

Mistake 3: Fitting the same model to all domains

I reviewed a project in Kalimantan where the geologist had used the same variogram model — spherical, nugget 0.3, sill 1.0, range 120m — for six different domains. The domains included a high-grade vein, a low-grade halo, a breccia zone, and three separate stockwork zones.

Each domain has its own geological continuity. A high-grade vein has short-range continuity along the vein direction and almost no continuity across it. A disseminated halo has longer ranges and less anisotropy. Using the same variogram for all domains means you’re assuming all mineralization behaves the same way — which is geologically impossible.

Fix: Calculate a separate variogram for each domain. If a domain doesn’t have enough data for a robust variogram (fewer than 30-50 composites), use a variogram from a geologically similar domain and document the assumption.

Mistake 4: Not validating the variogram with kriging

A variogram model can look beautiful on screen — smooth curve, good fit to the experimental points, sensible ranges — and still produce terrible kriging estimates. The variogram is not an end in itself; it’s input to kriging. You need to validate it by checking kriging performance.

Validation methods:

• Cross-validation (jackknife): Remove each sample one at a time, estimate its value using the surrounding samples, and compare the estimate to the actual value. If the variogram is correct, the errors should be unbiased (mean near zero) and the variance should match the kriging variance.
• Kriging neighborhood analysis: Test different search ellipses and sample counts to find the configuration that gives the most accurate estimates.
• Swath plots: Compare kriged estimates to the raw data along swaths (e.g., elevation bands, easting bands). The kriged model should follow the data trends.

I’ve seen variograms that looked perfect but produced kriging estimates that systematically overestimated high-grade blocks by 15-20%. The variogram range was too long, causing the estimator to borrow too much from distant high-grade samples. Cross-validation caught it. The “beautiful” variogram was wrong.

Mistake 5: Using a variogram from a different deposit

“This is an epithermal gold deposit, and I have a variogram from another epithermal deposit, so I’ll use that.”

No. Two epithermal deposits can have wildly different variograms. One has narrow veins with 30m ranges and 60% nugget. Another has disseminated mineralization with 200m ranges and 20% nugget. Both are “epithermal.” The variogram depends on the specific geological processes that formed and modified the mineralization at YOUR deposit.

Fix: Always calculate a deposit-specific variogram. If data is insufficient (early exploration stage), use a range of variogram scenarios and report the sensitivity of the resource to each. This is more honest than importing a variogram from another deposit.

The Indonesian-specific challenge

Indonesian deposits often have a complication that makes variography harder: the weathering profile. The oxide zone has different grade continuity than the sulfide zone. The transition zone — where oxidation is partial and irregular — can have a variogram that looks nothing like either the oxide or sulfide.

Practical approach:

1. Separate oxide and sulfide domains before variography
2. Calculate separate variograms for each weathering state
3. Treat the transition zone as its own domain if it has enough data
4. Document the difference — if oxide has a 150m range and sulfide has an 80m range, that’s geologically meaningful and should be in the report

The bottom line

Your variogram is not a software output. It’s a geological interpretation expressed in mathematical form. Every parameter — nugget, sill, range, anisotropy — has a geological meaning. If you can’t explain what each parameter means in terms of your deposit’s geology, you don’t understand your variogram well enough to use it for a resource estimate.

Take the time. Calculate directional variograms. Model each domain separately. Validate with cross-validation. Document your reasoning. The result is a kriging estimate that you can defend in front of a Competent Person review, a due diligence team, and eventually, the mine itself.