Fitting a model to the empirical semivariogram

Semivariogram/Covariance modeling is a key step between spatial description and spatial prediction. The main application of geostatistics is the prediction of attribute values at unsampled locations (kriging).

The empirical semivariogram and covariance provide information on the spatial autocorrelation of datasets. However, they do not provide information for all possible directions and distances. For this reason and to ensure that kriging predictions have positive kriging variances, it is necessary to fit a model (in other words, a continuous function or curve) to the empirical semivariogram/covariance.

Learn about the relationship between semivariograms and covariance

Different views of the empirical semivariogram/covariance values

The Geostatistical wizard provides three different views of the empirical semivariogram values. You can use one, two or all three of them to aid you in fitting a model to the data. The default view shows binned and averaged empirical semivariogram/covariance values.

Default view

Binned values are shown as red dots, and are generated by grouping (binning) empirical semivariogram/covariance points together using square cells that are one lag wide. Average points are shown as blue crosses, and are generated by binning empirical semivariogram/covariance points that fall within angular sectors. Binned points show local variation in the semivariogram/covariance values, whereas average values show smooth semivariogram/covariance value variation. In many cases it is easier to fit a model to the averaged values, as they offer a less cluttered view of the spatial autocorrelation in the data and show smoother changes in the semivariogram values than the binned points.

The Show points control can be set to Binned and Averaged (as shown in the figure above), Binned, or Averaged (as shown in the figures below).

Binned values only

Averaged values only

Additionally, lines can be added to the plot. The lines are local polynomials fitted to the binned empirical semivariogram/covariance values. If the Show search direction option is set to True, then only the local polynomial fitted to the empirical semivariogram/covariance surface in the Show search direction tool’s central axis transect is displayed, as shown in the following figure:

Semivariogram plot with lines

The semivariogram/covariance model you fit to the empirical data should:

Keep in mind that your knowledge of the phenomenon may dictate the shape of the model as well as its nugget, range and partial sill and anisotropy values, even though the model does not fit the empirical data too well (recall that the empirical data is just a sample of the real phenomenon you want to model, and may not be fully representative of all of its spatial and statistical aspects).

Different types of semivariogram/covariance models

Geostatistical Analyst provides the following functions to model the empirical semivariogram:

The selected model influences the prediction of the unknown values, particularly when the shape of the curve near the origin differs significantly. The steeper the curve near the origin, the more influence the closest neighbors will have on the prediction.

As a result, the output surface will be less smooth. Each model is designed to fit different types of phenomena more accurately.

The diagrams below show two common models (Exponential and Gaussian) and identify how the functions differ:

Exponential model

Gaussian model

9/11/2013