Variogram modeling

Introduction:


Characteristics of semivariogram:

Sill: The semi variance value at which the variogram levels off. Range: the maximum distance beyond which there is no correlation between pair of samples.

Nugget: The nugget represents variability at distances smaller than the typical sample spacing, including measurement error.

Need of variogram model:



For the sake of kriging we need to replace the empirical semivariogram with an acceptable semivariogram model. The reason is that semivariogram models used in the kriging process need to obey certain numerical properties in order for the kriging equations to be solvable.

It should be noted that the variogram models introduced here are those that we consider to be the basic models. They are simple, isotropic models, independent of direction. The basic variogram models can be conveniently divided into two types; those that reach a plateau and those that do not.

Variogram models of the first type are often referred to as transition models. The plateau they reach is called the sill and the distance at which they reach this plateau is called the range. Variogram models of the second type do not reach a plateau, but continue increasing as the magnitude of h increases.

Using h to represent lag distance, a to represent (practical) range, and c to represent sill, the most frequently used models are: Spherical Model: Perhaps the most commonly used variogram model is the spherical model.

It has a linear behavior at small separation distances near the origin but flattens out at larger distances, and reaches the sill at a. In this type we have a large range. i-e the sill is reached at high range.

The Exponential Model: Another commonly used transition model is the exponential model. This model reaches its sill asymptotically, with the practical range a.

Like the spherical model, the exponential model is linear at very short distances near the origin, however it rises more steeply and then flattens out more gradually. In fitting this model to a sample variogram its helpful to remember that the tangent at the origin reaches the sill at about one fifth of the range.

The Gaussian Model: The Gaussian model is a transition model that is often used to model extremely continuous phenomena. Like the exponential model, the Gaussian model reaches its sill asymptotically.

The distinguishing feature of the Gaussian model is its parabolic behavior near the origin. The Linear Model: The linear model is not a transition model since it does not reach a sill, but increases linearly with h. All the four models are given in the following figure.

Fitting model to variogram:



The actual process of fitting a model to an empirical semivariogram is much more of an art than a science, with different authorities suggesting different methods and protocols.

Since empirical semivariograms are often quite noisy, quite a bit of subjective judgment goes into selecting a good model.

Example:


Here is the semivariogram for porosity data, with three fitted models. In each case the sill value was fixed at the overall variance of 0.78 and the range was estimated using weighted nonlinear regression (weighting by number of data pairs for each lag):

The fitted ranges for the three models are 4141 m for the spherical, 5823 m for the exponential, and 2884 m for the Gaussian. The Gaussian model gives the best fit, but the spherical is a close second.