Kriging concept question

Hello,

I have recently read an interesting article about integrating Limited Area Models (LAMs) into kriging with external drift for temperature (Libert? et al. link below).

www.wmo.int/pages/prog/www/IMOP/.../IOM.../P2(05)_Perini_Italy.doc

This URL is incomplete. Please check.

As I understand, it seems that the authors generated variograms with the differences between the closest LAM grid point and the ground point. Then, they used kriging with external drift where the ground observation stations were the primary data and the secondary data was the LAM.

Conceptually, is it possible to generate a variogram with differences and then krig the data rather than the differences?

I am not 100% sure what do you mean by "differences". If you mean 
residuals, then the answer is positive (the variogram HAS to be of the 
residuals and not of the target variable). KED (or what I prefer to call 
regression-kriging) in a moving window is in fact probably more general 
approach than global analysis. For example you can be very relaxed about 
stationarity assumptions. The problem is that, for each prediction 
point, you will need to fit both the trend model (LAM?) and estimate a 
local variogram (using global variogram and local regression models 
means 'cheating'), which is probably very time consuming! In fact, I am 
not aware of any applications of KED with moving window (maybe I missed 
some?).

For more info see also section 2.2 in my book:
http://spatial-analyst.net/book/Regressionkriging

HTH

T. Hengl
GEOSTAT 2010 - THE Spatial Data Analysis event of the year!
http://geostat2010.info

I would appreciate anyone's explanations or corrections to my understanding.

Thank you,

Cara

	[[alternative HTML version deleted]]



------------------------------------------------------------------------

_______________________________________________
R-sig-Geo mailing list
R-sig-Geo at stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-geo