Performance Comparison Of Various Kriging Techniques Using Exhaustive Data Set

Using an exhaustive data set of Walker Lake, four different kriging techniques plus the inverse distance squared weighting were compared with regards to each one's performance in accurately estimating the block grades. The North-West quadrant of the Walker Lake data area was used because of its extreme variability of values and its distinct contour of the values. There were 1900 exhaustive data points in the selected area from which a total of 121 samples were drawn following the sampling strategy given in Isaaks and Srivastava (1989). The average of 5 x 5 exhaustive data points was treated as the true, known block grade. A total of 780 blocks, each block corresponding to the average of 5x5 exhaustive data points, were then estimated using the 121 sample points. The resulting estimates were then compared with the true known values to determine each technique's prediction performance. For this particular portion of the Walker Lake and for the given sampling strategies used, the obtained results for the global estimation are mixed. For the variable U which is much more variable, the rankings are; 1) Jackknife Kriging (JK), 2) Indicator Kriging (IK), 3) Inverse Distance Square weighting (IDS), 4) Ordinary Kriging (OK), and 5) Log-Jackknife Kriging (LJK). For the variable V, the results are; 1) OK, 2) IK, 3) JK, 4) LJK, and 5) IDS. When we apply several cutoff grades to the estimates to determine the conditional mean grades and tonnage, the above rankings generally get shifted around. Two exceptions to these cases are; a) UK for U and b) IDS for V. These two consistently ranked at the bottom.