The semi-variogram function (if it reaches a sil at
some point) is directly analogous to the covariance
between samples given the distance (and possibly
direction) between them. If you take:
total sill (final height of semi-variogram)
semi-variogram value at a given distance
you get the covariance between the values at two
locations separated by that distance. If you divided
through by the total sill you (theoretically) get the
correlation between them.
So it isn't really a probability function but a
You can derive probabilities for unsampled locations
by theory or simulation through the kriging process.