Dear all, I using the function krige.cv (in gstat library) to compare different kriging methods (ordinary, external drift, universal, lognormal,...). To do that I'm using among other the mean square normalized error (mean(out$zscore^2)). I'm wondering if I can use this statistic to compare results on a logarithmic scale (lognormal kriging) and normal scale. To compare the mean error (mean(out$residual)) I'm using backtransformed values:
out$var1.pred.backtransform=exp(out$var1.pred+0.5*out$var1.var)-1 out$observed.backtransform=exp(out$observed)-1 out$residual.backtransform=out$observed.backtransform-out$var1.pred.backtransform ME=mean(out$residual.backtransform) To compare the mean square normalized error do I have to backtransform the zscore value: out$residual.backtransform=out$observed.backtransform-out$var1.pred.backtransform out$var1.var.backtransform=?????????? out$zscore.backtransform=out$residual.backtransform/(sqrt(out$var1.var.backtransform)) if yes, is there a way to backtransform the prediction variance to the normal scale? Many thanks Sam [[alternative HTML version deleted]] _______________________________________________ R-sig-Geo mailing list R-sig-Geo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-geo