Kriging is cheating, the best estimate is a myth, and it's time resource modellers embraced uncertainty to do it right—and ethically

In a rousing keynote at AusIMM’s 2025 Mineral Resource Estimation Conference, Dr Clayton Deutsch—director and professor at the School of Mining Engineering, University of Alberta—challenged the audience to confront a fundamental flaw in how mineral resource estimators approach their craft. His message was clear: the longstanding industry convention of pursuing a “best estimate” through traditional Kriging workflows is not only technically deficient, it is—when misapplied—ethically questionable.

“Best estimates are a myth,” he told a packed room in Perth. “When you apply Kriging with restricted search parameters to smooth out your model and meet economic expectations, you’re cheating. And most of us are doing it.”

The provocation landed as intended.

Kriging Is Not the Enemy—But It’s Not the Answer

Deutsch was careful to clarify that Kriging itself isn’t the problem. Ordinary Kriging, he said, is an excellent tool for short-range grade control applications. It provides conditionally unbiased estimates and maximises Kriging efficiency. For production settings—where dense drilling and real-time reconciliation are available—it remains a mainstay.

“But for long-range resource modelling, the same method falls short,” he argued. “You’re smoothing excessively. You don’t get the right tonnes, you don’t get the right grade, and you’re underestimating uncertainty.”

The common workaround—restricting the search radius, applying multiple passes, or manually tuning search strategies—may look better on a report, but it distorts the underlying reality of geological variability.

“We cheat to make it look like we’re doing better,” he said. “And the irony is, we do this while pretending it’s still an objective estimate. It’s not.”

Embracing Uncertainty as the Core Task

At the heart of Deutsch’s argument lies a philosophical shift: uncertainty isn’t a nuisance to be minimised—it’s a core attribute of the subsurface environment.

“Mineral deposits are fixed in time and space. They are not changing, but they are inherently unknown. And because they’re variable at all scales and our drilling is sparse, uncertainty is intrinsic and permanent,” he said.

It’s not simply a matter of drilling more holes to reduce uncertainty—uncertainty can never be entirely eliminated. Instead, resource professionals must quantify it explicitly and rigorously, using simulation, not deterministic estimation.

In a diagram he referred to as his “uncertainty cartoon,” Deutsch laid out a modern modelling framework that includes:

• Prior parameter uncertainty (proportions, histograms, variograms),

• Measurement error and missing data handling,

• Uncertain geological boundaries and domains,

• Ensemble simulations of grade within those domains.

This approach generates a full distribution of possible outcomes for each block, allowing practitioners to not only calculate expected values, but to meaningfully assess risk and variability.

From Passive Estimators to Active Risk Managers

For Deutsch, this shift has professional and ethical implications. By failing to communicate uncertainty, practitioners are not just missing technical nuance—they’re violating their duty to stakeholders.

“If you know there’s geological uncertainty—and we all do—but you don’t disclose or incorporate it, that’s not just bad practice. That’s unethical,” he said. “We have a responsibility to do high-quality work, and that means quantifying uncertainty.”

Resource professionals, he argued, must transition from being passive estimators of ore to active managers of risk. That means using uncertainty to drive classification decisions (e.g., measured, indicated, inferred), scenario testing, project planning, and investment decision-making.

“This is not about abandoning realism in favour of academic theory,” Deutsch stressed. “It’s about building models that are useful—technically rigorous and transparent in their limitations.”

Simulation Is Ready for Prime Time

A central thrust of the keynote was Deutsch’s assertion that the simulation tools and methodologies required to model uncertainty are already mature, validated, and widely available.

“Forty years ago, simulation was theoretical,” he said. “Today, we know how to handle parameter uncertainty, spatial bootstrapping, non-stationarity, domain uncertainty, and reconciliation. We’ve run hundreds of these models—including for world-class deposits. The tools exist.”

He cited a recent example where simulation—not Kriging—was used to support a public resource estimate for what is reportedly the largest copper deposit discovered in the last 30 years.

“All resources and reserves were based on the ensemble of realisations,” he said. “No inverse distance. No Kriging. And it passed disclosure.”

Deutsch emphasised that all methods discussed in his talk—simulation, uncertainty quantification, trend modelling, variogram validation—are published, peer-reviewed, and publicly accessible. “There is no excuse to keep doing it the old way,” he said.

Machine Learning: A Supporting Role

While some industry practitioners pin their hopes on machine learning to revolutionise modelling, Deutsch was quick to caution that ML, while powerful, is not a panacea.

“Machine learning is everywhere now—it’s like computers in the 80s,” he said with a wry smile. “Yes, it’s useful. Yes, it’s embedded in our workflows. But on its own, it doesn’t address the information effect or the inherent limitations of sparse sampling.”

Instead, he advocated for a hybrid approach where machine learning supports—but doesn’t replace—geostatistical simulation and geological reasoning.

A Call to Action—and Accountability

Deutsch’s final message was a challenge to his peers: do better, not because the regulators demand it, but because professional ethics require it.

“We can no longer claim ignorance,” he said. “We know how to do it right. If you’re still applying Kriging with arbitrary search restrictions to control smoothing, you’re cheating. And if you’re not communicating uncertainty, you’re misleading stakeholders.”

He called on practitioners to stop hiding behind precedent, software limitations, or management inertia. “Change requires force and energy,” he said. “But it’s time.”

The Road Ahead

As the session closed, Deutsch left the audience with one final provocation: “The next time you submit a model built on restricted-search Kriging, ask yourself—am I really modelling the deposit? Or am I modelling what I wish it looked like?”

For the mineral resource estimation community, the challenge has been laid down: the path to more honest, defensible, and useful models lies not in chasing mythical best estimates, but in embracing and quantifying uncertainty.

And there is no better time to start.