The Normal-Score Transform

Gaussian Methods, Skewed Data

Sequential Gaussian simulation and the cleanest forms of kriging assume the variable is normally distributed. Real reservoir properties rarely are: permeability is strongly right-skewed, spanning orders of magnitude. Applying a Gaussian method directly to raw skewed data gives distorted, unreliable results.

Transform by Rank

The normal-score transform solves this without inventing data. It sorts the samples and replaces each value with the standard-normal quantile of its rank: the smallest value becomes a low normal score, the median becomes zero, the largest becomes a high score. Any input histogram, however skewed, becomes a perfect bell. Because the mapping is purely by rank it is monotonic, so it never reorders the data, it only rescales it.

Model, Then Back-Transform

The workflow is to transform to normal scores, compute the variogram and run kriging or simulation in that Gaussian space where the methods are valid, then back-transform the results through the same mapping to recover the original units and the original histogram. The back-transform is essential: skipping it would leave the model in abstract normal-score units instead of porosity or permeability.