[R-sig-Geo] Relative Mean Error for comparing interpolation methods

[Seyed Jalil Alavi]

 Dear All

I am comparing kriging and IDWmethods in mapping Forest Site Productivity 
using10-fold cross validation.

Bothkriging and IDW methods produced negative mean error. Now I want to use 
relativemean error for comparing these methods other than RMSE and mean absolute error.

here is the results of kriging;

mean ofresponse  variable: 34.76982

meanerror:  -0.03613827

meanabsolute error: 1.598008

RMSE:2.053376


how can Icalculate relative mean error?

I readsomewhere we can use this function for calculating relative mean 
errorhighlighted with red color:

OK_CV<- krige.cv(Site_Form ~1, ~X+Y, Data, model = model1.out, nfold=10)

# meanerror, ideally 0:

ME_OK<- mean(OK_CV$observed - OK_CV$ var1.pred)
ME_OK

### MeanAbsolutely Error
MAE_OK<- mean(abs(OK_CV$residual))
MAE_OK

###Relative Mean Error
MEr_OK<- (ME_OK/mean(Data$Site_Form))*100
MEr_OK

### RMSE
RMSE_OK<- sqrt(mean(OK_CV$residual^2))
RMSE_OK

###Relative RMSE

RMSEr_OK<- (RMSE_OK/mean(Data$Site_Form))*100
RMSEr_OK


if I use theabove function for relative mean error, the result will be negative!!!

I wouldbe very grateful if anyone can help me to calculate 
relativemean error in R.

Regards

SJA




	[[alternative HTML version deleted]]