[R-sig-Geo] Spatial interpolation possibilities

[Paul Hiemstra]

On 12/03/2010 03:38 PM, Arnaud Mosnier wrote:
> Dear GeoUseRs,
>
> > From the large number of methods that can be used to spatially interpolate
> data, we tried several of them (Inverse Distance Weighted, Natural
> Neighbour, Spline, Kriging, Co-Kriging) on our dataset. We used ArcGIS to
> apply the first three and R for kriging and test possibilities of
> co-kriging.
>
>   We do not have external independent data to test results of interpolations.
> The solution is probably to do a kind of cross-validation but doing it into
> ArcGIS seems difficult to implement.
> I know that I can do Inverse distance weighted interpolation in R but ...
>
> - Is it possible to do Natural Neighbour and Spline interpolation in R ? And
> if yes how ?
>    
Hi Arnaud,

Spline is possible, look at the Tps function from the fields package. 
I'm not sure on Natural Neighbor, but voronoi tessalations are available 
in R.
> ArcGIS tools propose to consider "barriers" when using Inverse Distance
> Weighted and Spline tools. Thus, for example, interpolation occurring North
> from an island do not consider information located South of this island ...
> a quite important thing in our case.
>
> - I wondering if it is possible to do the same with gstat or another package
> ?
>    
You could split up the dataset in two regions and interpolate them 
seperately.

Finally, if you want to cross-validate, look at the compare.cv function 
from the automap package. It provides statistics and some spatial plots 
of the cross-validation residuals which I found very useful (I wrote 
them myself ;)). When cross-validating non-gstat interpolations you need 
to make sure that the output of the cross-validation is compatible (i.e. 
a SpatialPointsDataFrame with a few specific attributes). I have an 
example of how to do this with Tps, which can serve as inspiration. If 
you are interested I can send the code off list.

Paul
> Thanks for your help !
>
> Arnaud
>
> 	[[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-Geo mailing list
> R-sig-Geo at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>    


-- 
Paul Hiemstra, MSc
Department of Physical Geography
Faculty of Geosciences
University of Utrecht
Heidelberglaan 2
P.O. Box 80.115
3508 TC Utrecht
Phone:  +3130 253 5773
http://intamap.geo.uu.nl/~paul
http://nl.linkedin.com/pub/paul-hiemstra/20/30b/770

currently @ KNMI
paul.hiemstra_AT_knmi.nl