Recoverable Resource Estimation by nonlinear geostatistics – Module 4: Advanced multivariate modeling

Where this module sits

The “Recoverable Resource Estimation” course is delivered in four modules that you can take separately or in sequence.

Module 1 — Uniform Conditioning. Unbiased grade-tonnage curves from sparse drilling, localized to the SMU level.
Module 2 — Multiple Indicator Kriging and Conditional Expectation. Two nonlinear estimators and how to choose between them.
Module 3 — Simulations. Multiple realizations with Turning Bands, SGS, and direct block simulation, post-processed into grade-tonnage curves.
Module 4 — Advanced multivariate modeling. Trend, decorrelation by PCA, MAF and PPMT, imputation and ratio transforms.

Content

• Trend modeling. Two distinct operations that are easy to confuse: modeling the trend itself and estimating with a trend already in place. When to separate them, and what a misplaced trend does to your estimates.

• Three multivariate decorrelation techniques – Principal Component Analysis (PCA), Minimum/Maximum Autocorrelation Factors (MAF), and Projection Pursuit Multivariate Transform (PPMT). All three address the same problem: cokriging several correlated variables at once quickly becomes unmanageable, while simulating them separately breaks their relationships. These methods turn the variables into independent factors you can handle one at a time, then transform them back. Where they differ is how far the decorrelation holds — at zero distance only, at two distances, or across the full joint distribution.

• Multivariate imputation. Filling the gaps in an incomplete dataset, using the correlations between variables to estimate the values you don’t have — so you can run the transforms above without discarding every sample that has a missing assay.

• Ratio transforms. For variables bound by a sum or a ratio, such as metal-to-oxide relationships or grades that must total a fixed amount, so the constraint survives estimation and simulation rather than being broken.