I am comparing two ordinary krigged models (one on log-transformed data, one on original data) in the cross validation window of Geostatistical Analyst, and I see conflicting results between the prediction errors and the scatterplot of the predicted versus observed values. The prediction errors (root-mean-square, mean standardized, root-mean-square standardized, and average standard error) appear better for the model on the original data, but for the model on the log-transformed data, the relationship between the observed and predicted values (as indicated in the scatterplot) is better than for the original data. Can anyone explain why this would happen? I have included the cross validation plot below. Thank you!