Bill,
for what I know about GSLIB and gstat, they have different
definitions for the range parameter of the exponential model.
Gstat uses the `former' GSLIB convention, whereas GSLIB2 puts
in the 3 somewhere. See
http://www.ai-geostats.org/FAQ_software_conventions.htm
for details. So, your examples should result in different
outcomes. Another thing that worried me is the sill value
of 32000000 in your variogram: this may lead to a numerically
unstable solution in gstat. GSLIB, IIRC, standardizes
covariances to correlations before solving the system. Try
to divide your concentration data by 1000, and retry. If you
are only concerned with predictions, you can also set the sill
to 1 (and multiply kriging variance by 32000000 later on), as
multiplying a variogram with a constant does not affect kriging
predictions.
Best regards,
--
Edzer
