Accuracy of the Kriging Error Estimate Through Cross Validation

Abstract

Cross-validation (CV) is used for error checking in kriging estimates. To quantify the error of a CV, 1875 eucalyptus trees were rigorously planted. A volume/tree map was constructed by kriging, generating 8736 points. Samples with different sizes (100, 150, 200, 250, 300 and 350) were taken from the 8736 points through systematic and random sampling and volume maps were constructed by kriging. A straight line was fitted to Estimated x Observed graph. The R2, the intercept and the slope were the errors parameters. The CV did not provide reliable results for the R2 in the systematic sampling with 100, 150, 200 and 250 samples, but it was reliable for the R2 in all randomized trials. The systematic sampling had good real results, R2 > 0.70 with 200 samples, while random sampling showed R2 > 0.70 only with 250 samples.