Statistics Glossary → Variogram (semivariogram)
Variogram (semivariogram)concept
A function describing how dissimilarity between two points in spatial data changes with the distance between them; it is the basis of kriging interpolation.
The semivariogram is half the average squared difference between pairs of points separated by distance h: γ(h) = (1 / 2N(h)) Σ (z(xᵢ) - z(xᵢ + h))². Because nearby points usually take similar values, γ(h) is low at short distances and rises as distance grows. Beyond a certain distance the increase stops and the curve levels off: that level is the sill, and the distance at which it is reached is the range; points farther apart than the range are treated as spatially uncorrelated. A jump at zero distance (the nugget) reflects measurement error or variability at scales smaller than the sampling interval.
The empirical variogram consists of points computed for distance classes (lags); a valid theoretical model such as spherical, exponential, Gaussian or Matérn is fitted to them by weighted least squares. The Matérn smoothness parameter κ sets how smooth the surface is (κ = 0.5 is the exponential model). The number of distance classes and the maximum distance (cutoff) affect the result; because very distant pairs are few and unreliable, the cutoff is usually limited to about one third of the study area's extent. The classical (Matheron) estimator is sensitive to outliers, whereas the Cressie-Hawkins estimator is more robust. The variogram assumes stationarity (dependence on distance only); if dependence varies with direction (anisotropy), directional variograms should be examined.
An everyday example
You measure soil moisture in a field. Two points 1 metre apart have almost the same moisture, points 10 metres apart differ a little, and beyond 50 metres the difference no longer grows with distance: moisture at one end of the field is unrelated to moisture at the other. The variogram summarizes this picture as a curve and decides which neighbours, at which distances, matter most when estimating moisture at unmeasured points.
In MerQur
In Variogram + Kriging it is controlled by the 'Variogram model (Automatic, Spherical, Exponential, Gaussian, Matérn, Linear)' and 'Matérn κ', 'Number of distance classes (lags)' and 'Cutoff', 'Variogram estimator (Classical (Matheron), Robust (Cressie-Hawkins))' and 'Fitting weights' parameters.
Analyses where this concept appears
Variogram + Kriging (Spatial Interpolation)
Related concepts
Nugget, sill and rangeKriging (simple, ordinary, universal, block)Anisotropy