@niandra82 Yes, some points are really close, but still not more than, say, 300-400 meters over a 284 km^2 area. What would you do to prevent this "ill" conditioning?

if I let optim() find automatically the best values, both the space and time nuggets are set to 0, which causes the original error and perfect correlation like you said...

@niandra82 as far as choosing a "nugget", why would I not be allowed to chose a value at wish? If I do so, the error may be bigger than using an automatic routine optim() to find best values that minimize errors but I still choose a variogram model with the parameters I specify.

@niandra82 station "s" is not in the data used to make prediction. That point is removed during CV and it is specified as newdata as prediction location. What I meant is that the variogram model was fitted to the empirical variogram. When calculating the empirical variogram and finding a theoretical variogram model, all available observations can be used.

@niandra82 I think the last error is because I forgot to use as( ,"STSDF") to my data in the krigeST() function. However, let me know if I answered your question...thanks for your help again!

Error in cbind(v0, X) : number of rows of matrices must match (see arg 2)

Also, If I manually add a nugget variance, so that it is not equal to 0 anymore, I now get the following error in the krigeST() function:

When I do cross-validation on station "s", I use the space-time variogram model fitted to the empirical variogram on all my observations (using their recorded values). As "newdata" in my ST kriging estimation, I specify station "s" (thus as if it was unknown) as the only one point on which to estimate the temperature. Is that what you were asking for?

I get that error exactly when doing cross-validation to predict one station at a time given all others. I use a neighborhood to speed up the computing time but I get the same error when I use the entire observed process. I will try by manually changing the nugget from 0 to 1 and see. The only drawback is that by doing that, my space-time variogram fit gets worse (~3.2 RMSE) after adding a nugget manually.

Ok, thanks gung and Glen_b for your ideas. I guess since I only have 10 observations that I'd need to use a non parametric Poisson regression if at all, since the sample size is too small...the 10 observations correspond to the 10 cities...within which I have a measure of this # of nodes (count data, Y variable), and total # number of days (X1) taken to create this "nodes" (or edits) on a map.

Which generalized model approach would you use?...remember I only have 10 observations so it's not just the 'normality' of data the problem...