- Dear group,
I am a master's student in forest ecology and am a "low intermediate"
geostatistics user. I have collected soil samples on 5 plots on an
irregular grid of points. Each 50x50m plot is separated by at least 75
meters. I hope to, among other things, characterize the spatial patterns
of several soil factors measured from each sample using variogram modeling.
I have been using the GS+ software and am quite pleased, but have a few
1. stationarity of the data: I have read a number of articles in the
ecology and agronomy literature discussing this concept, and have read some
conflicting things (one researcher refers to the concept of stationarity as
"troubling"). One article (by Jongman et al. I believe) says that
nonstationarity of the variance leads to a meaningless expression of the
autocorrelation function, but not of the semivariogram; with the variogram,
only nonstationarity of the mean is a concern. Several articles say that
both nonstationarity of the variance and mean need to be fixed before
variogram modeling. Which is it?
2. checking for stationarity: I just finished reading a paper by Hamlett,
Horton and Cressie which shows some exploratory techniques for variogram
analysis. They recommed doing local variance vs mean plots to check for
stationarity of the variance. However, Isaaks and Srivastava's book
mentions that this does not necessarily indicate nonstationarity, but a
"proportional effect", something which happens in non-normally distributed
data. In addition, the Hamlett et al paper seems to check for stationarity
of the mean by simply looking at the stem and leaf plot and checking its
symmetry, as well as looking at how the local mean changes across the
sampling area. Indeed, many of the papers and books I've read say "the
local mean should not change". Obviously, this is not literal, but nobody
seems to say just how much change is ok for your data to be stationary, and
likewise with the variance. How much change in local mean and variance is
3. trend removal: It's tempting just to say my data are non stationary and
do median polish and do my variograms with the residuals. Is this a valid,
albeit "black box", approach?
4. When I finally do arrive at my variogram models, I would like to use
them to block krig. I know that one approach is to do "universal kriging",
for which I unfortunately do not have the software nor the expertise.
Could I simply fit a first or second order trend surface to the data, model
the variogram with my residuals, krig using ordinary kriging, and then add
the trend back in at the end as is suggested in the "final thoughts" of
Isaaks and Srivastava's book and elswhere (unless I am misinterpreting what
I read)? That would make me very happy.
5. Speaking of this, I wonder about the merits of obtaining the residuals
for the variogram modeling from a 1st or 2nd order trend surface, vs.
obtaining them from median polish?
Thank you for your time; I will summarize any responses and post them to
*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!