Optimal Valuation And Kriging - 9.1 Definitions: Kriging Estimator And Kriging Error

[Consider a square block W in an ore body S2, to be valued from one centre sample w, and one corner sample w2 (Fig. 9.1). The sample set is ws = Iw,;w21. We can estimate the block average µw by the average of the sample values x, and x2: µw = 'z (x, + xz). (9.1) The error made is measured by the estimation variance: o' (ws to W) = 2 - (ws; W) - - (ws;ws) - - (W; W). (9.2) However, we know that the centre sample w, gives more information about the block value than the corner sample w2. In the example studied in § 8.4, where w, and w2 were considered to be points, we found that of (w, to W) was nearly one third smaller than o2E (w2 to W), therefore when estimating W one should give more weight to w, than to w2. In general, when evaluating a block W from a set of n samples ws, where WS = l w;w2; ... wn1, a better estimator of µw can be obtained if we give different weights to the sample values, depending on the size and the positions of the samples. Some desirable properties of such an estimator µw are: ? It should be a linear function of the sample values xi: n µw = bi xi (9.3) =t where bi = weight given to sample x . ? It should be unbiased, i.e. its expected value should be equal to the true block value: E [µw - µw] = 0. (9.4) ? The mean squared error of estimation of µw should be a minimum E [(µw - µw)2] = a minimum. (9.5) The estimator which satisfies these properties is a minimum variance, linear and unbiased estimator, and is known as the kriging estimator. The corresponding error of estimation is the kriging error. This estimator is also known as the BLUE (for Best Linear Unbiased Estimator) estimator. This was named by Matheron in acknowledgement of the early work of Krige's in the optimization of the evaluation of the South African gold mines. The following notations will be used: µK kriging estimator. o2k = o2K (w,, to W) kriging error.]