Re: puzzled about OK predictions

["Edzer J. Pebesma" <e.pebesma@geog.uu.nl>]

Ashton, I'm glad to hear you're working with gstat in R.

Kriging has not constraints as to what the range of the
predicted values should become -- they may indeed get outside the
range of the data. The reason this occurs is because kriging weights
can (and will) become negative. This phenomenon occurs more strongly
when you use a Gaussian variogram without a nugget. I would suggest
to try a Matern model with a suitable smoothness parameter (it has
the Exponential and Gaussian as two special cases). 

Another "solution" that may work is add a (small) nugget component.
Please share your experiences on this.

Best regards,
--
Edzer