[R-sig-Geo] Advice on Cross validation_ Computational limits of spatial interpolation

[Paul Hiemstra]

Hi,

Hmmm, 3500 doesn't sound that huge to me...

I would say that more than, say, the 200 points closest to the
point-to-be-interpolated is more than enough. Gstat (and thus also
automap) supports only using the x nearest neighbors  for interpolation
through the 'nmax' argument. I'm not really sure if Tps has similar
settings, although I suspect that fitting the splines is done on the
entire dataset. Do take a look at the fastTps function in the fields
package, specifically the 'theta' argument.

good luck,
Paul

On 10/04/2011 03:56 PM, Sandeep Patil wrote:
> Hello
>
> The function Automap has been of great help to me and i have set it up 
> to perform cross validation for variables over continental USA.
>
> Essentially this is a very large dataset with about ~3500 points.
> There were many other dimensions , temporal etc that i wanted
> to explore but i could not get R packages setup  on univ supercomputer
> due to technical issues.
>
> This cross validation does not perform on my PC due to lack of
> computational units.
>
> Core 2 duo, 8 gb RAM. for Thin plate splines and kriging which are my 
> primary interest. It is obviously impractical to have so many points.
>
> I am dividing this dataset into smaller units and targeting ~300
> points per 
> polygon for analysis. Can anyone advise me on the influence of 
> number of points on the computational needs for TPS and kriging 
> based on their personal experience.
>
> I do know that Thin plate splines and Kriging are computationally
> demanding
> and my dataset is huge. Moreover it seems infeasible to do cross
> validation on it.
>  If anyone has any specific advice based on personal
> experience will they be willing to exchange a few emails?
>
>
> Thanks 
>
> -- 
> Sandeep R Patil
> Graduate Student
> M.S (Hydrological Sciences)
>


-- 
Paul Hiemstra, Ph.D.
Global Climate Division
Royal Netherlands Meteorological Institute (KNMI)
Wilhelminalaan 10 | 3732 GK | De Bilt | Kamer B 3.39
P.O. Box 201 | 3730 AE | De Bilt
tel: +31 30 2206 494

http://intamap.geo.uu.nl/~paul
http://nl.linkedin.com/pub/paul-hiemstra/20/30b/770