Unbounded variogram models like the power-law model can't be expressed by a covariance model, since they don't have a finite sill.
We need another class next to CovModel (maybe UnboundedModel), that provides only a variogram method. Only parameter in common with CovModel is then nugget (and maybe rescale).
Possible models are:
Random field generation is currently depending on the spectral density of a model derived from its covariance function. Thus, we would need to implement new random field generators like turning bands or sequential gaussian for these models.
For Kriging, these models could be used immediately, but we need #191 for it to work.
Unbounded variogram models like the power-law model can't be expressed by a covariance model, since they don't have a finite sill.
We need another class next to
CovModel(maybeUnboundedModel), that provides only avariogrammethod. Only parameter in common withCovModelis thennugget(and mayberescale).Possible models are:
UnboundedLinearPowerLaw(Webster 2007)Schlathermodel: https://onlinelibrary.wiley.com/doi/full/10.1002/sta4.134Random field generation is currently depending on the spectral density of a model derived from its covariance function. Thus, we would need to implement new random field generators like turning bands or sequential gaussian for these models.
For Kriging, these models could be used immediately, but we need #191 for it to work.