IOP Conference Series: Earth and

Environmental Science

PAPER • OPEN ACCESS You may also like

- SECOND AND THIRD SEASON QUaD
Weighted jackknife ordinary kriging - problem COSMIC MICROWAVE BACKGROUND
TEMPERATURE AND POLARIZATION
solution of the precision in mineral resources POWER SPECTRA
C. Pryke, P. Ade, J. Bock et al.

estimation - MEASUREMENT OF COSMIC

MICROWAVE BACKGROUND
POLARIZATION POWER SPECTRA
To cite this article: W S Bargawa 2018 IOP Conf. Ser.: Earth Environ. Sci. 212 012059 FROM TWO YEARS OF BICEP DATA
H. C. Chiang, P. A. R. Ade, D. Barkats et
al.

- THREE-POINT CORRELATION
FUNCTIONS OF SDSS GALAXIES:
View the article online for updates and enhancements. LUMINOSITY AND COLOR
DEPENDENCE IN REDSHIFT AND
PROJECTED SPACE
Cameron K. McBride, Andrew J. Connolly,
Jeffrey P. Gardner et al.

This content was downloaded from IP address 170.80.242.90 on 19/03/2024 at 21:00

ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

Weighted jackknife ordinary kriging - problem solution of the

precision in mineral resources estimation

W S Bargawa1*
1
Master of Mining Engineering UPNVY, Jalan SWK 104 Condongcatur, Yogyakarta,
Indonesia

*
corresponding author: [email protected], [email protected]

Abstract. The aim of introducing the weighted jackknife ordinary kriging (WJOK), as a robust
estimator, is to solve the outliers and the problems of uncertainty associated with grade estimates
of mineral resources. Jackknifing the ordinary kriging (OK) estimates and weighting the pseudo-
value yield more reliable estimates than ordinary kriging and jackknife ordinary kriging (JOK)
estimates. The new estimator is more conservative than the standard jackknife technique, which
is shown by a little standard deviation value. It obtains a simple way to calculate the confidence
interval associated with the estimation.

1. Introduction
One successful work of exploration is the accuracy of estimate grade of mineral resources which will be
used as reference data for the next phase of work. The estimates grade, reliable, is only possible if the
application of valuation technique is used according to the geological conditions of the mineral deposits
[1, 2]. Complex mineralized structure may cause the appearance of high grade outliers so as to generate
a highly skewed distribution of mineral grade [3-6]. This problem causes difficulty in determining the
grade estimation technique. Many approaches have been carried out to solve the problem of outliers but
did not yield the expected results. Moreover another problem is that the proposed approaches do not
produce a measure of uncertainty of grade estimation.
This paper uses a technique to estimate the grade of mineral resources namely WJOK technique. This
technique is a development of jackknife standard technique [7, 8]. Jackknife technique was previously
only used to reduce bias [9-11], determine the weight [12-13], build confidence interval [14-16], further
develop the application in the field of banking and financial aspects [17], and other applications [18-20].
This paper will combine classical statistics with independent data and spatial statistics with
corresponding data, so that it becomes an interesting new approach. The advantage of this technique is
simplicity in the mathematical concept but robust in the application on abnormal situations. This
technique is more easily understood and used by practitioners in the field.

2. Objective
The aim of this paper is to implement a robust and reliable estimator in mineral resources which have
high-grade outliers and have a long tail distribution and compare the new technique with other estimators
for measuring the uncertainty of estimation generated by the standard deviation.

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

3. Method and material

The method of estimation in this paper is a combination of weighted jackknife resampling techniques
[21] and ordinary kriging technique [22, 23] called jackknifing spatial data. The term of the spatial data
here is ordinary kriging grade estimates.

3.1. Jackknife technique in linear models

The basic concept of resampling techniques [7] can be studied through a linear model. The model
assumed reads as follows:
Y  A  e (1)
Assume
D0  AT A (2)
then the least squares estimator:
ˆ  D0 _1 AT Y (3)
and the residual vector is:

R  Y  Aˆ  I  AD0 AT Y
1
 (4)
Based on the above equation weights can be defined as follows:
1
wi  xi D0 xi , i  1, ...,n
T
(5)
The procedure of calculation standard jackknife technique initiated by defining pseudovalue:
Pi  nˆ  n  1ˆi i  1, ...,n (6)
Jackknife standard estimator obtained from equation (6), namely:
~
  n1  Pi (7)
Cressie [24] suggests this approach is often inaccurate when applied to the unbalanced models. This
weakness is reflected in the distance wi. Pseudo value Pi (equation 6) is defined symmetrically, while
the model is usually an unbalanced model. Based on the problems this paper proposes a weighted pseudo
value approach:
 
Qi  ˆ  n1  wi  ˆ  ˆi  ˆ  nD0 xi Ri
1
(8)
Jackknife weighted estimator is:
~
 w  n 1  Qi  ˆ (9)
and variance estimates are:
Vw  n n  p   Qi   w Qi   w 
1 ~ ~ T
(10)
Jackknife approach is very interesting, especially if the estimates result is not only an arithmetic
mean, but also a weighted average [21]. The weights are determined based on the spatial correlation
between samples.

3.2. Kriging technique

OK is a very popular technique because it is widely used for grade estimation in precious metals [25-
27]. OK estimator is defined:
n
Z *   wi Z i
i 1 (11)
Weight, wi, solved by OK equation system namely:
n

 w .
i 1
j ij     0i
(12)
with the constrained condition:
w i i 1 (13)

2
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

OK equation system above can be written in matrix form as follow:

D w  X  (14)
Weights solved using equation:
w  D1 X  (15)
Equation (15) is similar to the equation (5) so that the OK estimates can be identified with a regression
line in the linear model. Based on these arguments, OK can be understood as a linear model in the case
of spatial statistics. The combined concepts of spatial statistics namely OK techniques and jackknife
resampling technique is a good concept and very useful to solve the unbalance model.

3.3. Weighted jackknife ordinary kriging

Based on the above discussion an approach is obtained that is WJOK. In mathematical terms it can be
explained as follows: sample z1, z2, ….., zn as n representing the grade data, which is independently and
identically distributed (iid) at a location x 1, x2, ..., xn. Z* (x0) abbreviated Z* is an estimate of grades
(e.g. ordinary kriging estimate) at the location x0, using n sample, and the Z*-i is ordinary kriging
estimate using the n-1 sample with the i-th sample is not used, the jackknife estimator can be obtained
by calculating the weighted pseudo value (see equation 6 and 8):
Zp - i  Z * n(1  wi )(Z * Z * -i )i  1,2,..n (16)
Jackknife estimator is the average of weighted pseudo value [see equation (9)]:
n
ZJ  n 1  Zp - i (17)
1
i
Pseudo value formulation provides an important contribution that the precision of jackknife estimator
can be known through the variance or standard deviation of the WJOK, which are defined (see equation
10) as follows:
 i

 J  nn  11  Zpi  Z J 2 2
1
(18)
Based on Tukey [7] n pseudo value in equation (14), can be treated as an independent estimates. Based
on the above discussion, the statistics:
n 2 Z J  Z  /n  11   Zpi  Z J   2 i  1, 2,.,n
1 1
2 (19)
i

will have distribution t with the n-1 degrees of freedom. Where n is very large the distribution will be
close to normal, so it can be used in the estimation of the confidence interval [28]. Confidence interval
proposed by Tukey is as follows:
Z J  t , n 1 J (20)
with (1-2), 100% confidence interval for Z. In this case t, n-1 is the number of top percentile of
distribution-t with n-1 degrees of freedom.

3.4. Grade estimation and size precision

The main issue in grade estimation is the size precision of the estimates. The misunderstanding for
kriging variance is considered the size precision of the kriging estimate. Whereas OK variance or
standard deviation is not directly related to the data. The OK standard deviation is just a function of the
data configuration. OK standard deviation decreases with an increasing number of samples used to
estimate grade and distribution of samples around the block to be estimated. While the standard deviation
of JOK reflects the actual uncertainty from the process of grade estimation. If the samples around the
block to be estimated have a similar grade value, the uncertainty associated with the estimated block
grade will be low in accordance with the lower value of the standard deviation of jackknife ordinary
kriging. For blocks surrounded by grade sample with a mixture of high and low, the uncertainty
associated with the assessment grade.
Based on the discussion jackknife ordinary kriging standard deviation is an useful indicator to
quantify the uncertainty associated with the block grade estimates. Nevertheless the standard deviation
of every block does not offer many practical benefits. Therefore the standard deviation value is not

3
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

independent statistically. And it will be increased, which is reflected by the increasing value of the
standard deviation of JOK within these blocks, there is no easy way to use it. The urgent need
encountered in the estimation of mineral reserves is a measure of global estimate precision for a common
form that is much larger than the block.
JOK technique allows an easy way to solve the problem. The approach introduced in this paper is
block kriging short cut [29] and followed by jackknife technique. Grade estimation process is performed
locally by ordinary block kriging techniques. The weights received by each sample are collected and
totaled. The weight of each sample is divided by the total weight of the entire block to produce normal
weight for each sample. The weights reflect the effect of each sample against global grade estimate. The
term normal weight used for the sum of weights for all samples is one. Jackknifing global estimates with
each remove a sample of the data can produce jackknife estimates and standard deviation estimates. The
standard deviation can be used to calculate the confidence interval for jackknife estimate.
The case study conducted by reviewing nine blocks (2D illustrations for ease of explanation) which
continued to be a common shape (see figure 1 below).

1000E 1100E
1100N  0,017 1100N
0,022
0,019 
0,014

 
0,026
0,038 0,042

0,034 0,013
 
0,049 
 0,037
1000N 1000N
1000E 1100E

Figure 1. The general shape consisting of nine blocks and

eleven sample grades for estimation.

4
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

1000E
1100E
1100N 1100N

0,022 0,029 0,032

0,029 0,035 0,036 0,034

0,035 0,032

1000N 1000N
1000E
1100E

Figure 2. Estimates of block grades located in any shape

using OK.

Grade and standard deviation of the common shape will be estimated by block kriging shortcut (OK)
and jackknife. Based on the variography study of the data set used [8] in this case study can be obtained
semivariogram parameters as follows: C0 (nugget) = 0.0005; C1 = 0.0045 and range = 33 m. Search
radius of the block estimated sample was 16.5 m.
Practically the search of samples used for estimation was performed per block as in ordinary block
kriging. Figure 2 shows the estimation using OK techniques, while Figure 3 shows the estimates of JOK
for the same blocks. OK estimate average of nine blocks in Figure 2 is 0.032 oz/t, while the average of
the JOK is 0.033 oz/t (figure 3).
OK estimation of the same shape using block kriging shortcut equal to the average grade of the nine
blocks in figure 2 is 0.032 oz/t. Furthermore, OK estimation using block kriging shortcuts with the first
example is not used for the estimation. Furthermore, each sample i removed from the data (i = 1, 2, ...,
11). Based on the calculation, it obtained pseudo value (equation 3), JOK estimates (Equation 4) and
JOK standard deviation (equation 5). Global grade estimates for any form of case studies using block
kriging shortcut followed by jackknife technique is 0.033 oz/t. These results are similar to the average
grade of the nine blocks in figure 3. The calculation of standard deviation JOK to estimate the global
results is 0.0041 oz/t.

5
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

1000E
1100E
1100N 1100N

0,020 0,031 0,033

0,028 0,038 0,043 0,038

0,034 0,033

1000N 1000N
1000E
1100E
Figure 3. Block grade estimates located in any shape
using JOK.

The confidence interval for the global estimation can be calculated using equation (7). According to
Student-t table, the value of t, n-1,  = 0.025 and n-1 = 10 was 2.228. If fixed at 95% confidence interval
for jackknife ordinary kriging global estimation then obtained: (0.0332.228*0.0041) oz/ton. In other
words there is a 95% probability that the gold grade estimate of the volume (any shape) ranges between
the value of 0.042 oz/ton and 0.024 oz/ton.
The second case study conducted in porphyry copper deposits. The amount of data that is used by
20,720 assay obtained from 207 drill data includes copper grade (assay), specific gravity, and rock type.
Generally drill hole was carried out by the method of inclined drilling to the copper ore deposits.
Composite data based on geological considerations with interval length correspond at 15 m mining
bench height. Composite number is 6,400 data. 3D block model is made with block size of (20 x 20 x
15) m. Size of (20 x 20) m shows the north and east, while 15 m is a measure of mining bench height.
Each block in the model is marked with rock code which indicates the blocks that are inside or outside
the mineralized area. Interpolation is not performed on blocks that are outside the mineralization area.
Figure 4, 5, and 6 below is a scatter diagram between copper composites and block grade estimates
at the same location. The figure shows the result of the estimation of block grade using OK, JOK, and
WJOK methods. Linear regression statistics are shown in table 1. Statistics show a comparison between
the performance of the estimator used. Visual observation of the scatter plot (figure 4, 5, and 6) clearly
shows that the estimated grade using WJOK method tends to be closer to 450 regression line (bisector)
compared to the estimated grade using the OK and JOK method. Table 1 shows that the WJOK estimator
has the lower Y-intercept value, the higher the slope value compared with OK and JOK estimator.This
case study shows that the estimator WJOK is more accurate than the OK and JOK estimator.

6
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

Figure 4. Grade estimates using OK, grade estimates

were more spread.

Figure 5. The grade estimates using the JOK method

also tends to spread.

Figure 6. The grade estimates of the WJOK method

tend to follow the 450 regression line.

7
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

Table 1. Linear regression statistics based on the scatter plot between the copper composite
and grade estimate.

Parameter OK JOK WJOK

Y-Intercept 0.233 0.195 0.147
Slope 0.791 0.815 0.850
SEE 0.623 0.589 0.498
R 0.680 0.750 0.800

4. Discussion
Usually the addition of a number of exploration drillings in an area of investigation is conducted to
produce a more detailed geological information in order to obtain accurate data. Geologist and mining
exploration experts can estimate the effect of the presence or absence of drill point at a certain location.
If there is no drill point at a location it will produce different geological information if the location was
coupled with the drill point. The simulation of a present or an absent drilling point in one location will
display a spatial image of the geological model at that location.
Jackknife resampling technique is similar to the above scenario that is how the estimate results if one
particular sample of data is not used. Similarly the simulation for each sample will be removed from the
data and re-used in the next estimation. Combining estimation using all samples and estimation when
each sample is removed from the data can utilize all the information throughout the data. The idea is
similar to the addition of information from a drilling point at specific locations and how geological
information changes if the particular location does not have a drilling point.
The interesting concept is the basis for the use of WJOK techniques for ore grade estimation. The
case study in this paper is used to indicate the accuracy of the estimator and measurement of uncertainty.
In a case study of a bench composite with the relatively uniform grade distribution the third estimation
techniques (WJOK, JOK, and OK) produce similar estimates, but the standard deviation value of the
WJOK is smaller than the standard deviation of the standard jackknife OK. In this case the variance
(ordinary) kriging, can not be compared with the jackknife variance because it is not a measure of
uncertainty or estimates precision. During this time there is a fallacy that kriging variance is a measure
of precision whereas the kriging variance is merely an index of data configuration in the estimation.
The second case study shows that the WJOK is more robust than the JOK and OK techniques. OK
estimates are to be high due to the influence of outlier extrapolation. The estimate is less realistic as the
blocks are estimated to be surrounded by samples that have a relatively low value. While the block grade
estimates using WJOK technique indicates the most reliable results. The standard deviation value of the
WJOK is smaller than the standard deviation value of the JOK. The WJOK technique is more
conservative because it has a wider confidence interval than the JOK technique.

5. Conclusion
The discussion in the previous chapter can be summarized as follows:
1. WJOK estimator is a new technique that is more robust in the unbalanced model situation compared
with JOK and OK techniques.
2. The use of repetitive samples can maximize spatial correlation information between samples in order
to obtain the block grade estimates which is more accurate and can generate confidence interval
estimates.
3. Combining jackknife resampling approach and spatial statistics in the form of jackknifing ordinary
kriging estimates is a new approach that is highly prospective for mineral resource grade estimation.

8
ICEMINE IOP Publishing
IOP Conf. Series: Earth and Environmental Science 212 (2018) 012059 doi:10.1088/1755-1315/212/1/012059

References

[1] Gandhi S M, Sarkar B C 2016 Valuation of Mineral Properties Essentials of Mineral Exploration and
Evaluation (Elsevier) chapter 14 pp 321-338
[2] Njowa G, Musingwini C 2018 Resources Policy 56 3
[3] Guj P 2016 Ore Geology Reviews 78 336
[4] Aderhold J, Davydov V Yu, Fedler F, Klausing H, Mistele D, Rotter T, Semchinova O, Stemmer J and
Graul J 2001 J. Cryst. Growth 222 701
[5] Lazo P, Steinnes, Qarri F, Allajbeu E F S, Harmens H 2018 Chemosphere 190 337
[6] Lisitsin V 2015 Ore Geology Reviews 71 861
[7] Shao J, Tu D 1995 The Jackknife and Bootstrap (New York: Springer-Verlag) p 1
[8] Adisoma G S, Hester M G 1998 Mining Engineering 48 84
[9] Shyamal D, Peddada S D 1993 Handbook of Statistics 9 723
[10] Chambers M J 2013 Journal of Econometrics 172 142
[11] Chambers M J, Kyriacou M Statistics & Probability Letters 83 677
[12] Gong B 2018 Energy Policy, Volume 114 145
[13] Singh S, Stephen A. Sedory S A, Rueda M D M, Arcos A and Arnab R 2016 Tuning
of jackknife estimator, A New Concept for Tuning Design Weights in Survey Sampling 27
[14] Sang Y, Dang X and Zhao Y 2018 Journal of Statistical Planning and Inference in press
[15] Zhong P S, Chen S 2014 Journal of Multivariate Analysis 129 193
[16] Chen S, David Haziza 2018 Computational Statistics & Data Analysis 127 258
[17] Füss R, Miebs F and Trübenbach F 2014 Journal of Banking & Finance 43 14
[18] Lee H, Babu G J and Rao C R 2012 Journal of Statistical Planning and Inference 142 301
[19] B.E. Hansen, J.S. Racine 2012 Journal of Econometrics 167 38
[20] Kruse R, Kaufmann H and Wegener C 2018 Economic Modelling In press
[21] Bargawa W S 2016 Proc. Int. Conf. on the 1st International Conference on Science and Technology
(Yogyakarta), AIP Conf. Proc. 1755 120001-1
[22] Liang C P, Chen J S, Chien Y C and Chen C F 2018 Science of The Total Environment 627 1048
[23] Bargawa W S, Rauf A and Amri N A 2016 AIP Conf. Proc. 1705 020025
[24] Cressie N 1997 Technometrics, 39 1
[25] Espinar J A L, Igúzquiza E P, Olmedo J G and Olmedo C G Journal of Hydrology 561 348
[26] Donati M, Fouquet C D 2018 Science of The Total Environment 633 1480
[27] Bargawa W S, Amri N A 2016 AIP Conf. Proc. 1705 020001
[28] Miller R G 1974 Biometrika 61 1
[29] Isaaks E H, Srivastava R M 1989 Applied Geostatistics (New York: Oxford university Press) p 1