Some comments on Tom Charnock's queries:
A) Generally, the rule of thumb is to limit the search neighborhood,
to avoid stretching the stationarity assumption. Of course, this is
only fine and dandy if you have sufficient data. This could range
from three to six wells in an offshore development to hundreds of
thousands of points in a remote sensing problem.
B) Consider that cross-validation can be used to fine tune
search neighborhood parameters such as radii, points per
quadrant, anisotropy ratios, and so forth.
C) Limiting the number of CPU cycles can be a moot point with
today's hardware capabilities. Also, techniques are available
to increase computation speed, e.g. sparse matrix solvers in
special kriging cases.
D) Realistically speaking, 100 points is about the maximum for
a problem involving a unique neighborhood (using all points
instead of using a local neighborhood). Covariance matrix is
inverted only once, and doesn't change from location to location.
*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!