The nugget effect is composed of at least two components, one is very short
range variability (shorter than the smallest inter-sample distance) and
secondly the noise-error componet of the data. Usually one does not have
data to separately estimate the latter but if data is available one can
use it in at least two ways, one is estimate/model the nugget effect (which
could be greater than the variance of the error component) and secondly one
can modify the kriging equations to allow smoothing as well as
interpolation. In the latter case the kriging estimator is no longer exact.
Given that you have a skewed distribution, you might want to consider a
data transformation such as logarithmic.
Also as an additional way of considering the size of the nugget term, you
might want to look at cross-validation results. This can also be helpful in
detecting unusual data values and locations. It might be useful to
temporarily delete these and re-model the variogram and again consider
cross validation results (both with the unusual values included and with
Donald E. Myers
Department of Mathematics
University of Arizona
Tucson, AZ 85721
*To post a message to the list, send it to ai-geostats@...
*As a general service to list users, please remember to post a summary
of any useful responses to your questions.
*To unsubscribe, send email to majordomo@...
with no subject and
"unsubscribe ai-geostats" in the message body.
DO NOT SEND Subscribe/Unsubscribe requests to the list!