[R-sig-Geo] residuals by subtracting point values from grid values

Jon Olav Skoien j.skoien at geo.uu.nl
Tue Mar 24 18:48:44 CET 2009


Hi Els,

I think you want to do:

idw.pred.zn <- idw(zinc ~ 1, meuse.pred, meuse.val)

and for kriging cross-validation:

krige.pred.zn <- krige(zinc ~ 1, meuse.pred, meuse.val,model = vgm(140000,"Sph",600,0.1))
krige.pred.zn$res = krige.pred.zn$var1.pred - meuse.val$zinc
krige.pred.zn$zscore = krige.pred.zn$res/sqrt(krige.pred.zn$var1.var)
 

You cannot do crossvalidation with meuse.grid, as you dont have 
zinc-observations in the data set.
As Tom said, this is more or less the same as you can get a lot easier 
using krige.cv (with nfold=3), unless you are planning to recalculate 
the variogram for each fold.

Cheers,
Jon



Els Verfaillie wrote:
> Hi list,
>
>  
>
> I've splitted my dataset (e.g. the meuse dataset) into random 2 parts (2/3
> for prediction and 1/3 for validation). I've done an interpolation (e.g.
> IDW, but I also want to do it with different kriging techniques) based on
> the prediction dataset. Now I want to subtract the "zinc" values from the
> validation dataset (i.e. a SpatialPointsDataFrame with 52 points) from the
> "var1.pred" values from the interpolation (i.e. a SpatialPointsDataFrame
> with 3103 points) to obtain the residuals. This of course only on the
> location of the 52 points from the validation dataset. Then I want to
> calculate validation indices.
>
>  
>
> This is my code:
>
>  
>
> #Validation and prediction dataset
>
> permid <- sample(155)
>
> idportion <- 1:round(2*155/3)
>
> meuse.pred <- meuse[permid[idportion], ] 
>
> meuse.val <- meuse[permid[-idportion], ]
>
>  
>
> #idw based on meuse.pred
>
> idw.pred.zn <- idw(zinc ~ 1, meuse.pred, meuse.grid)
>
>  
>
> I know that I could use the "krige.cv" to do an n-fold cross-validation, but
> I would like to do it based on the above method.
>
>  
>
> Any suggestions? 
>
>  
>
> Thanks a lot!
>
>  
>
> Els 
>
>  
>
> ______________________________________________
>
> Dr. Els Verfaillie
>
> Carto-GIS cluster
>
> Ghent University (UGent) - Department of Geography
>
> Krijgslaan 281 - S8  B- 9000 Gent  Belgium
>
> ______________________________________________
>
>  
>
>  
>
>
> 	[[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
>



More information about the R-sig-Geo mailing list