Kriging

Kriging is a geostatistical method of interpolation for estimating unknown values from some original sample data set. It differs from other techniques such as inverse distance in that it uses the concept of spatial continuity between data. The aim is to minimise the variances of the estimates and arrive at the best data value for each unknown point.

Part of the input is a semi variogram model that fits the input data. The validity of the Kriging result depends on how well the semi variogram models the data.

This means that you should not use Kriging until you are satisfied that you have created the best possible semi variogram model for the data, as its closeness of fit will determine the confidence you have in the Kriged output. Kriging functions operate on points, blocks, or polygons defined in an outline file.

Before Kriging a data set, you need to analyse the data and create a semi variogram that will be used in the modelling process. The semi variogram model you produce must be saved in a form set, which will then be used as an input to the Kriging function.

Analysis should also include:

• Investigating the distribution of sample values using histograms, probability plots, and scattergrams for correlation between variables and possible geological patterns.
• Transformation of values; logarithms, indicators and rank uniform transforms.
• Calculation and interpretation of experimental variograms; identification of any trend and possible anisotropies.
• If necessary, studying of trend components, removal of trend and study of residuals including distribution of residuals and identification of outlines.
• Modelling the semi variogram graph. Choosing a model, specifying parameters and assessment of fit.
• Cross validation of the semi variogram model. Calculation of error statistics. Remodelling if required.

The following Kriging functions are available for selection on the Block Model tab, in the Estimation group:

Kriging

When the Kriging method is set to SIMPLE, the global mean is assumed to be constant throughout the study area. If the Mean response is blank, the global mean is automatically calculated from the input data.

When the Kriging method set to ORDINARY, the local mean is recalculated each time the search neighbourhood is positioned on a new block centroid, but is kept constant within that neighbourhood.

Other Kriging methods are also available on the Block Model tab, in the Estimation group.

For more information, refer to the Kriging topic.

Rank Kriging

Use the Rank Kriging option for non-parameteric or distribution free statistics. Your data will be transformed so that it has a uniform distribution (where the probability of each value occurring is equal).

Samples are ranked from lowest to highest. Each observation is assigned a rank. These ranks are then transformed into a percentage value against the number of samples + 1. This means a value of 75 is three-quarters of the way along the data list.

For more information, refer to the Rank Kriging topic.

Multiple Indicator Kriging

Indicator Kriging allows you to preferentially exclude data at a defined cutoff. It estimates the probability of values being above or below the cutoff. Probability is associated with the mean of a bin to produce a weighted-average (e-type) estimate.

Use Multiple Indicator Kriging to define a series of cutoff values and apply the same indicator method again for each cutoff.

For more information, refer to the Multiple Indicator Kriging topic.