variogram with nominal data

Hi Julian, and list,

the variogram calculates the average squared difference between z-values separated by particular lags. Calculating differences with nominal data, and certainly IDs, is wrong. You will get results, but garbage in, garbage out applies here.

That said, you can calculate indicator variograms on nominal data. See a recent paper by Ola Ahqvist and myself, or other papers, mostly in soil science. (What fun to self-cite!): 
Spatial and semantic dimensions of landscape heterogeneity, Landscape Ecology (2010).

I don't think the gstat variogram functionality for nominal data extends to more than two binary classes, but is not hard to implement more than that. I have R script on my machine at work if you would be interested.

Hope this helps,

Ashton Shortridge

Quoting Julien Beguin <julien.beguin.1 at ulaval.ca>:

Dear list,

I would like to evaluate spatial dependence for about 10000 spatial 
data points. However, the variable associated with those points is 
not continuous as in the data "meuse" but is a land cover type (e.g. 
WATER). What should I consider as variable Z in the variogram() 
function with gstat? is it right to take the ID of each point as in 
the code below?
-----------------------------------------------------------------------------
Z <- data.frame[ which(data.frame$landcover=='WATER'),]$ID
X <- frame$X
Y <- frame$Y
mydata1 <- data.frame(Z, X, Y);
coordinates(mydata1) <- c("X", "Y");
vario01 <- variogram(Z~1, mydata1, cutoff=50000, alpha=c(0,45,90,135))
------------------------------------------------------------------------------

Any advice would be very appreciated. Thank you for your time,

Julien Beguin
------------------
Ph.D. student
Laval University
Qu?bec, Canada

	[[alternative HTML version deleted]]