[R-sig-Geo] Best approach to implement Spatial Interpolations

[Edzer J. Pebesma]

Cesar Martinez Izquierdo wrote:
> Hello list,
> we are trying to implement spatial interpolations (IDW and several types 
> of  Kriging) in our open source GIS application, which is written in Java.
>
> We plan to use JRI to connect with R from Java, and then use the 
> features of "gstat" R-package to calculate the interpolations.
>
> However, we don't want R to draw the results of interpolations, we just 
> want to get the raw result which we will draw in our GIS client.
>
> Now, the questions:
> - Do you think gstat is the best R package for this task?
>   
As the author, I will refrain from commenting on this. It has a large 
user base, a code that has been open source and stable for a long time, 
and it has several options not found in other packages, notably block 
kriging, cokriging, and kriging in a local neighbourhood with an 
efficient search algorithm (quadtrees) for huge/massive data sets.
> I've read that it doesn't allow not-projected data. Is there a package 
> which provides similar algorithms and works with not-projected data?
>   
Recently, this functionality has been added, only the quadtree algorithm 
does not work in this case; I'm still in doubt whether quadtrees are 
useful on a sphere. Consider this code somewhat experimental, but I'm 
willing to help debug it if needed. It does, for example not check 
whether covariance functions provided are positive definite on a sphere, 
and has little (if any) functions specifically dedicated for this 
purpose. It does compute great circle distances though.

Another package that seems to deal with unprojected coordinates that you 
may want to look into is package fields.
> - Do you have some documentation regarding this topics (more exactly: 
> documentation about how to get the results of a kriging, etc with JRI)?
>
>   
Perhaps the JGR user interface is an example?

I'm very interested in your developments and experiences; please keep 
me/us updated on this mailing list.
--
Edzer
> Any comments and suggestions are welcome.
>
> Regards,
>
>