I had frequently seen the definition for a "rigorous" spatial isotropic semivariogram being defined as:

$$ \gamma(h) = K(0) - K(h) $$

Where $K$ is a positive definite covariance matrix. If the nugget $\sigma$ is applied to the covariance function $K$, then it seems to me that it would always be that $\gamma(0) = 0$, but that $\gamma(0+\delta)>\sigma$ for all distances greater than zero.

However, often in practice a linear function is used to determine $\gamma$:

$$ \gamma(h) = \sigma + \frac{h}{\beta} $$

Should I really define the nugget in terms of the covariance matrix, or should I make it a separate term:

$$ \gamma(h) = \sigma + K(0) - K(h) $$

Am I making a mountain out of mole hill here? Are there any reasons to prefer one form or the other?

To visualize, is it more "correct" to pick one form over the other:

nonzero zero

zero zero

Based on the idea of nugget as allowing a non-zero offset at zero, I could see the former should be preferred: but this seems to imply that the nugget is no really part of the assumed covariance matrix.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.