[R] Compare predict and experimental values

[Roger Bivand]

On Fri, 30 Sep 2005, Toni Viúdez wrote:

> Hi everybody:
> I just generate different maps of interpolation with different methods but 
> with the same package -(gstat, i used the krige command). And now for 
> evaluate the best method i want to compare the predict values versus 
> experimental, but i don't know how do it.
> Could anybody tell me how should do it?.

?krige.cv or ?gstat.cv provide kriging cross validation, including IDW. 
The leave-one-out cross validation RMSE agree with those in Geostatistical 
Analyst in ArcGIS. That is one possibility that you may choose if it fits 
your needs, but "best method" is only valid up to the data you have and 
the relative placing of the point locations. Surprisingly often, external 
drift does matter, and once other variables are taken into account, the 
residual spatial autocorrelation does not contribute so much.

> Thanks in advance.
> 

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no