Cross validation fitted models- krige.cv
On 03/27/2014 06:25 PM, Moshood Agba Bakare wrote:
Dear All, Common variogram models were fitted to empirical variogram with the following range of spatial dependency obtained. *model * *range* Spherical 14.2 Gaussian 7.5 Exponential 5.5 Linear 10.4 I intend to cross validate the models to determine the best that capture the spatial pattern.I thought of defining *maxdist option *to be the value of range so the the prediction can be done by the points within the range of spatial dependency. Am I doing the right thing? Using range as maxdist, will it give me a reliable RMSE and MSDR diagnostic statistics? #### Cross validation of the variogram models by ordinary kriging sph.kcv <- krige.cv(yield ~ 1, canmod.sp, model = sph.var,nmax = 100, nfold = 5, *maxdist = 14.2*) gau.kcv <- krige.cv(yield ~ 1, canmod.sp, model = gau.var,nmax = 100, nfold = 5,* maxdist = 7.5*)exp.kcv <- krige.cv(yield ~ 1, canmod.sp, model = exp.var,nmax = 100, nfold = 5, *maxdist = 5.5*) lin.kcv <- krige.cv(yield ~ 1, canmod.sp, model = lin.var,nmax = 100, nfold = 5, *maxdist=10.4*)
See also my previous reply earlier today -- if you want to have distances at which spatial correlation is below 5%, take range * 3 for Exponential and range * sqrt(3) for Gaussian. As you use ordinary kriging, also observations further away than the (effective) range play a role in the prediction, so that is not a reason why they should be ignored.
Edzer Pebesma Institute for Geoinformatics (ifgi), University of M?nster Heisenbergstra?e 2, 48149 M?nster, Germany. Phone: +49 251 83 33081 http://ifgi.uni-muenster.de GPG key ID 0xAC227795 -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 490 bytes Desc: OpenPGP digital signature URL: <https://stat.ethz.ch/pipermail/r-sig-geo/attachments/20140327/db1369fb/attachment.bin>