Variogram Modeling Rules of Thumb
The following rules of thumb are general tips for variogram modeling:
-
Know your data! Before calculating the experimental variogram, calculate regular non-spatial statistics. Use the Variogram Report to display the data minimum, maximum, median, mean, standard deviation, variance, and skewness. Create a post map or classed post map in Surfer to display scatter plots. Use Grapher to create histograms and cumulative frequency plots.
-
Do not over model. The simplest model that reproduces the important features of the experimental variogram is the best model.
-
When in doubt, use the default variogram model for gridding. A simple linear variogram model usually generates an acceptable grid; this is especially true for initial data analysis. Remember, however, that the kriging standard deviation grid generated using the default variogram is meaningless.
-
Unless there is a clear, unambiguous, physical justification, do not use an anisotropy ratio of greater than 3 to 1. If the experimental variogram appears to support an anisotropy of greater than 3 to 1, and there is no unambiguous, physical justification for such a severe anisotropy, there may be a trend in the data. Consider detrending the data before carrying out your variogram analysis.
-
Try the model and see how the resulting grid looks in a contour map. If you have competing candidate variogram models, generate a grid and contour map from each. If there are no significant differences, choose the simplest variogram model.
-
The range of the variogram is often close to the average size of physical anomalies in the spatial fluctuation of the Z values. In the absence of a reliable experiment variogram, this rule of thumb may be used to postulate a variogram range.
-
An experimental variogram that fluctuates around a constant value is not an ill-behaved variogram. It is an indication that the Z values are uncorrelated at the scale of the typical sample spacing. In such a situation a contour map, regardless of the gridding method used, is an unreliable representation of the data, and more data at closer sample spacing are needed for detailed local characterization.
-
If the following three conditions are met, then the sample variance is a reasonable approximation for the variogram sill:
-
The data are evenly distributed across the area of interest, as displayed in a post map.
-
There is no significant trend in the data across the area of interest.
-
The dimension of the area of interest is more than three times the effective variogram range.
Back to Modeling the Variogram Anisotropy
Next to Variogram Frequently Asked Questions