Multivariate Transformations Overview
Use the Multivariate Transformations option to perform conventional Principal Component Analysis (PCA), decorrelate-only Principal Component Analysis (sometimes called sphering, or spectral decomposition), or Minimum Maximum Auto-correlation Factors (MAF). All of these transformation methods take some input multivariate data and populate a set of new variables which are decorrelated and, in the case of conventional PCA, ranked in order of importance.
The Multivariate Transformations option will also back transform an estimated, or simulated, block model back into original units.
PCA and MAF are both efficient techniques for modelling many interrelated variables. In almost all deposits there are multiple variables of interest (gold, silver, sulphur, etc., or nickel, silica, magnesium, iron, etc.). These variables are often related to one another due to their common geologic origins. It would be incorrect to model them separately and ignore the relationships. It is also often difficult, or impractical, to use full cokriging as it requires fitting an elaborate linear model of coregionalization. PCA and MAF tackle the problem in a slightly different way; instead of trying to model the relationships directly they transform the input variables into a set of linearly uncorrelated variables called principal components. These components may then be modelled independently and back transformed into the original variables.
The information entered through the Multivariate Transformations option will be saved in a multivariate transformation specification file ( mvt.spec ). The transformation particulars will be stored in a transformation file (pca.tfn). The transformation file should be retained to perform the backward transformation.
In order to use Multivariate Transformations, you must have a database that contains fields for all variables of interest, and fields for all resulting components. These fields need to be in the same table (record). Additionally for minimum maximum auto-correlation factors you will require the point coordinates.
