Semivariogram Modeling: Visualize Spatial Correlation

simulator intermediate ~12 min
Loading simulation...
Range = 150 m — spherical model with nugget ratio 5.9%

A spherical variogram with nugget C₀=5, sill C=80, and range a=150 m shows strong spatial dependence (nugget ratio 5.9%). Samples within 150 m are significantly correlated.

Formula

γ(h) = C₀ + C[1.5(h/a) − 0.5(h/a)³] for h ≤ a (spherical)
γ(h) = C₀ + C[1 − exp(−h/a)] (exponential)
γ(h) = C₀ + C[1 − exp(−h²/a²)] (Gaussian)

The Foundation of Geostatistics

The semivariogram is the cornerstone of geostatistical analysis. Introduced by Georges Matheron in the 1960s, it quantifies a deceptively simple observation: measurements taken close together tend to be more similar than measurements far apart. By plotting the average squared difference between paired observations against their separation distance, the variogram reveals the spatial structure hidden in scattered data — the range over which correlation exists, the total variance, and the amount of random noise.

Anatomy of the Variogram

Three parameters define a variogram model. The nugget C₀ is the y-intercept — variance that exists even at zero separation, caused by measurement error or micro-scale variation. The sill (C₀ + C) is the plateau the variogram reaches — equal to the total variance of the data. The range a is the distance at which the sill is reached — beyond this distance, measurements are essentially uncorrelated. The ratio of nugget to sill indicates how much of the total variation is spatially structured versus random.

Choosing the Right Model

This simulation lets you compare three standard authorized models. The spherical model is linear near the origin and reaches its sill at exactly distance a — it is the workhorse of mining geostatistics. The exponential model approaches the sill asymptotically, reaching 95% at distance 3a — suitable for phenomena that transition gradually. The Gaussian model is parabolic near the origin, indicating extreme short-range smoothness — useful for topographic surfaces and other very continuous fields, though it can cause numerical instability in kriging.

From Variogram to Prediction

Once a valid variogram model is fitted, it becomes the engine for kriging interpolation. The variogram determines exactly how much weight each nearby sample receives when estimating an unsampled location. Strong spatial dependence (low nugget ratio) means kriging can produce precise, smooth maps. Weak spatial dependence (high nugget ratio) means each sample influences only its immediate neighborhood. Variogram modeling is both science and art — automated fitting helps, but geological knowledge of the deposit is essential for choosing appropriate models and parameters.

FAQ

What is a semivariogram?

A semivariogram γ(h) measures how dissimilar measurements become as the separation distance h increases. It is defined as half the average squared difference between paired observations at distance h. The shape reveals nugget (microscale variation), sill (total variance), and range (correlation distance).

What is the nugget effect?

The nugget C₀ represents variance at zero distance — caused by measurement error or spatial variation at scales smaller than the sampling interval. Named after gold nuggets causing erratic assay values in adjacent mining samples.

How do you choose between spherical, exponential, and Gaussian models?

The spherical model reaches its sill at exactly the range — good for phenomena with a clear correlation cutoff. The exponential model approaches the sill asymptotically (95% at 3a) — suitable for gradual transitions. The Gaussian model is parabolic near the origin — appropriate for very smooth, continuous phenomena.

Why is the variogram important for kriging?

The variogram model is the engine of kriging. It defines the spatial weights used to interpolate — nearby samples get more weight, with the exact weighting determined by the variogram's shape. A poor variogram model leads to unreliable kriging estimates and incorrect uncertainty quantification.

Sources

Embed

<iframe src="https://homo-deus.com/lab/geostatistics/variogram/embed" width="100%" height="400" frameborder="0"></iframe>
View source on GitHub