Application of Computers and Operations Research in the Mineral Industry –

Dessureault, Ganguli, Kecojevic & Dwyer (eds)

© 2005 Taylor & Francis Group, London, ISBN 04 1537 449 9

The application of two stage indicator kriging in gold vein modeling

W. Sulistyana & S. Saptono

Department of Mining Engineering, Jalan, Indonesia

ABSTRACT: Accurate prediction of recoverable reserves is vital to the planning of a mining operation at all
stages of its development. Method such as ordinary kriging estimate the mean grade of a block that is fairly large.
The usual outcome is that large blocks rarely turn out to be all ore or all waste, thus making reserve estimates
an incorrect estimate of what will be mined. Two stage indicator kriging offers a solution to this problem by
estimating the distribution of grade values within a large block, rather than just estimating the mean grade of the
block. Knowing the distribution of grade value within the block, it is then easy to calculate the proportion of the
block that is above cutoff grade and the grade of the ore above cutoff grade. This paper shows that the two stage
indicator kriging model is quite applicable and reliable in modeling the gold vein deposit.

1 INTRODUCTION mineralization stage, which influenced the geometry

of the deposit. High grade Au–Ag mineralization is
The estimation problem, termed recoverable reserve generally found in silicic alteration associated with
estimation, is especially difficult for deposit which shear zones and the mentioned clay minerals. The
present a highly skewed grade distribution and/or con- strike and dip of Cikidang vein respectively are N21◦ E
tain mineralization which is interspersed with waste and 75◦ to west.
material such as in gold vein deposit. To handle this
difficult problem various recoverable reserve estima-
tors have been introduced. These estimators have not, 2 AIM
however, enjoyed the widespread acceptability and use
afforded the ordinary kriging (OK) estimator as they The aims of this paper are: (a) to implement an applica-
are difficult to comprehend and apply. If recoverable tion of geostatistical method through case study with
reserve estimators are to attain the level of accept- gold vein modeling, (b) to evaluate performances of
ability of ordinary kriging, methods of estimating ordinary kriging and two stage indicator kriging with
recoverable reserves must be developed which are sim- block model which describes 3D depiction of geom-
ple, robust, and easy to apply. The application of two etry of mineralization and grade distribution in gold
stage indicator kriging (TSIK) is presented here. The vein deposit.
TSIK method has precedents in uranium reserve esti-
mation, refer to Chávez and Parker (1979), Froidevaux
and Journel (1982). It was also used by M. Springett for 3 METHOD
early reserve estimates for the Mesquite Mine (1984).
Parker (1991) suggested a variant, simplified indica- 3.1 Block model construction
tor kriging (SIK), which was renamed IMOK because Using block model the 3D depiction of geometry
of the unappealing connotation of the former acronym of mineralization, ore grade distribution, and the
for the Rossi and Parker paper in 1993. resources in gold vein deposit can be obtained. The
This paper presents a case study illustrating the first step in geologic modeling is to plot cross sections
application of two such estimators, TSIK and OK on with geologic data of each drill hole on sections. The
the Cikidang gold vein deposit. The Cikidang gold geology is then interpreted on the cross section and a
epithermal deposit is located in Lebak District West numerical code is assigned to each rock type to rep-
Java Indonesia and is operated by PT Aneka Tam- resent the geology in digital form for computer input.
bang. Cikidang deposit is postulated as a tension gash Geology from the section is then plotted onto hori-
fracture infilled by gold bearing hydrothermal fluid, zontal plan maps. It shows where each cross section
forming quartz vein and altered the wall rocks. The intersects the plan map and the rock types at the inter-
Cikidang oblique fault with its minor faults (nor- section. After the geologic modeling is complete, plan
mal and reserve faults) was formed during the post maps of the model can be plotted and checked.

259

Copyright © 2005 Taylor & Francis Group plc, London, UK

 Table 1. Statistical summary of Au assays Table 2. The result of Au grade composite statistics.
(57 core drill holes and 10 mining stopes).
Statistics Gold composite
Statistics Gold assay
Number of data 495
Number of samples 681 Maximum (g/t) 91.10
Maximum (g/t) 104.37 Minimum (g/t) 0.02
Minimum (g/t) 0.02 Mean (g/t) 12.80
Mean (g/t) 9.71 Median (g/t) 8.10
Median (g/t) 3.91 Standard deviation (g/t) 15.30
Standard deviation (g/t) 14.28 Variance (g/t)2 233.16
Variance (g/t)2 203.99 Skewness 2.26
Skewness 2.68 Kurtosis 9.25
Kurtosis 12.04 Coefficient of variation 1.19
Coefficient of variation 1.50

The limits of the ore reserve model are plotted

on a topographic map of the area. Topographic con-
tour are digitized to obtain several thousand data
point with the northing, easting, and elevation of
each point. Gold resource modeling in this study
have coordinate boundary as follow: North direction:
0 N–1,300 N; East direction: 150 E–600 E; Elevation:
1,075 m–1,400 m. The block size is (10 × 2 × 5) m.

3.2 Data analysis

Exploration database construction consists of making
assay and composite database, descriptive statistics,
and variography study. There are altogether 57 core
drill holes at an average spacing of about 75 m and
20 mining stopes. The length of original assay is
2.0–3.5 m. A statistical summary (length weighted) of
those is given in Table 1. As can be seen, the assays
are positively skewed with a coefficient of variation
of 1.50.
Statistical analysis and geological interpretation
indicate that the data and blocks in the model could be
divided into two populations, i.e. vein and non-vein.
Decile analysis of the data show that high-grade cap
is not required. The top decile (90–100%) has 30.4%
Figure 1. Along strike Cikidang gold composite variogram.
of the total metal content (Waterman, 1999). If the
top decile has more than 40% of the metal, cutting
the high assay may be warranted (Parrish, 1997). To 3.3 Variogram modeling
obtain the same geometrical support, assay data are Variography study is done by searching data to along
averaged (weighted average) into 2 m composites. The strike of vein (N21◦ E) and down dip of vein (75◦ to
weighted average is found by first tabulating the indi- west). Along strike and down dip gold variograms
vidual length li and their corresponding
grades gi . The are calculated using all data. Figures 1 and 2 (see
products li gi are formed and summed
li gi . This sum Appendix) show the obtained variograms. These vari-
is then divided by the sum of the length li to yield the ograms are calculated through covariances. It is gen-
desired grade. Composite statistics give mean grade of erally more difficult to obtain traditional variograms
12.80 g/t Au, as shown in Table 2 (all composites have for gold deposits, and the Cikidang vein deposit is no
2 m long). exception. The figures illustrate that all variograms are
Tables 1 and 2 show that composite grade mean is quite well behaved and they can be fitted with the same
higher than assay grade mean, because approximately spherical models.
60% data are taken from high grade vein area with 2 m Along strike and down dip indicator variograms are
interval sampling. calculated at one discriminator grade, i.e. 3 g/t Au.

260

Copyright © 2005 Taylor & Francis Group plc, London, UK

 Figure 2. Along down dip Cikidang gold composite Figure 3. Along strike indicator variogram @ 3 g/t gold
variogram. discriminator.

This first indicator variogram is used to define blocks,

which exist within mineralization area (up to 3 g/t Au).
Figures 3 and 4 (see Appendix) show the obtained
indicator variograms. The variogram is also quite well
behaved and it can be fitted with spherical model as
well.
Gold grade estimation parameters, which are
obtained from ordinary and indicator variograms and
data distribution, are shown in Table 3.

3.4 Two stage indicator kriging

Estimation of the reserves for the gold vein deposit
boils down to two steps. First, the amount of ore within
a volume must be determined and second, the grade of
the ore must be determined. The process is called two
stage indicator kriging.
As the name implies, two stage indicator kriging
involves two steps. First, a study of the gold vein
deposit is done to determine a cutoff value, i.e. 3 g/t
Au that clearly separates mineralized material from
non-mineralized ground. Next, a variogram is calcu-
lated for the indicator values. The TSIK model used
only a 3 g/t Au discriminator to define the mineralized
vein. Using the indicator variogram, ordinary kriging
is then used to estimate the propotion of each block that
is waste. It should be obvious that if we know the per-
centage of each block that is waste, we also know the Figure 4. Along down dip indicator variogram @ 3 g/t gold
amout of each block that is ore. The second step is to discriminator.

261

Copyright © 2005 Taylor & Francis Group plc, London, UK

 Table 3. TSIK and OK estimation parameters.

TSIK

No Parameter The 1st stage The 2nd stage OK

1. Search distance: 50 m 45 m 45 m
along strike of vein
2. Search distance: 35 m 40 m 40 m
down dip of vein
3. Search distance perpendicular 1.50 m 5m 5m
to vein
4. Minimum number of sample 2 2 2
5. Maximum number of sample 10 10 10
6. Variogram/Indicator variogram:
• nugget 0.050 100 (g/t)2 100 (g/t)2
• sill 0.155 200 (g/t)2 200 (g/t)2
• range 75 m 45 m 45 m
7. Geological control: Grade
estimations (TSIK and OK)
are only done in vein area

determine the grade of the ore within each block. The

395 N
200 E

grade of the ore in the block should be correlated to the

grade of ore observed in nearby samples. Thus, the sec- 1250 1250
ond step is to estimate the grade of the ore using only

40.4
nearby samples that are ore grade. Ordinary kriging
is used to make these estimates. The variogram used
for the kriging correspond to the variogram calculated 1200 1200
from data that are above the cutoff grade.
Gold resource modeling in this study use TSIK
technique with OK technique for comparison. The
OK estimation parameters for block estimation are 1150 1150
roughly similar to those of the TSIK model (Table 3).
For validating the results, the scatter plot between
data value and its estimates is used. Data value used
in this study is composite data, because the pro-
1100 1100
duction data is not available at the same estimate
395 N

395 N
200 E

402 E
locations. 0 20 40 60 80
METERS

EAST-WEST SECTION 395

4 RESULT >20 g/t

10 - 20 g/t
5- 10 g/t
<5 g/t
The result of the gold vein modeling using TSIK and SCALE
OK techniques can be seen at Figures 5 and 6 (see 1:1000
INDEPENDENT
Appendix). Figure 6 shows that the section of OK MINING CONSULTANTS, INC. TSIK MODEL
block model is controled by ore vein that is obtained TUCSON, ARIZONA

from geological digitized, whereas the vein shape of Figure 5. East-West section (N 395) of TSIK model.
TSIK model (Figure 5) reflects the actual condition.
The TSIK model could distinguish high grade area and
low grade area. 5 DISCUSSION
The comparison of tonnage, grade, and gold metal
content of TSIK and OK models using three compos- The comparison in this study, the distribution of
ites minimum can be seen at Table 4. Both of TSIK samples used in grade estimation is plotted against
and OK models show high grade estimates because the distribution of block estimates at sample locations.
they use drilling data (57 core drill holes) and stope The cumulative frequency plot of the TSIK and OK
sampling (20 mining stopes) in the vein area. estimates, which are plotted against the data value,

262

Copyright © 2005 Taylor & Francis Group plc, London, UK

 395 N
200 E

14.9 15.1 16.6 17.8

1250 1250

14.6 14.9 19.2 14.8 19.3

33.3
46.146.9 50.3 58.4
14.7
36.2
43.6
10.8

1200 1200

46.8 44.1
26.1 40.8
35.5 42.7
43.340.3
2.1
16.4

24.1
34.937.0

2.1
31.9 40.5
18.1

3.8
21.134.2
19.7

2.1
6.6 6.19.8
8.2

17.5
29.4
1150 1150
11.2
4.9
11.3
5.2
7.4
6.6 5.0
6.4
4.7
4.7
3.9
4.7 3.5
6.0
6.6
6.9
3.4

1100 1100
3.5
395 N
200 E

395 N
402 E
0 20 40 60 80
METERS

EAST-WEST SECTION 395

>20 g/t
10 - 20 g/t
5- 10 g/t
<5 g/t Figure 7. Scatter plot of the composite grade against the
SCALE TSIK estimate.
1:1000
INDEPENDENT
MINING CONSULTANTS, INC. OK MODEL
TUCSON, ARIZONA

common feature found in kriged estimates (referred

Figure 6. East-West section (N 395) of OK model. to as smoothing) especially where one is estimating
grade in a block volume which is much larger than the
Table 4. Comparison of tonnage, grade, and gold metal sample.
content of TSIK and OK models. However, the two distributions are quite close to
Cog Tonnage of TSIK Tonnage of OK model
each other.
(g/t) model This study does not use jackknifing for validating
the results.The ideal scatter plot between the data value
0 404,169 821,517 and its estimate is the first bisector, i.e. the 45◦ line
4 400,569 623,949 with zero intercept. This means every single estimate
6 376,274 529,883 matches the data value perfectly at all locations. In the
10 295,729 376,286 real world this is, of course, unattainable. The scatter
plot of the TSIK and OK estimates, which are plot-
Cog Grade (g/t) of TSIK Grade (g/t) of OK ted against the data value (composites), is shown in
(g/t) model model Figures 7 and 8 (see Appendix).
Table 5 shows the linear regression statistics for
0 18.62 12.32
each scatter plot. Comparison of both TSIK and OK
4 18.76 15.55
6 19.59 17.40 models indicate that the TSIK technique is quite
10 22.66 21.32 reliable in gold vein modeling.
Gold vein deposit, in this study, is characterized
Cog Au metal content of Au metal content of by • a low average concentration, high variance, and
(g/t) TSIK model (kg) OK model (kg) a skewed distribution of assay values which tends to
produce outliers, • poor continuity of mineralization,
0 7,525.63 10,121.09 • complex intermingling of ore and waste, and • dif-
4 7,514.67 9,702.41 ficulties in obtaining reliable samples. Each of these
6 7,371,21 9,219.96 factors adversely affects the reliability of ore reserve
10 6,701.22 8,022.42 estimation. TSIK could help to solve several of the
above problems.
Outliers make interpretation of variogram difficult,
show that in the low-grade range, the estimate is plot- thus lending suspicions to the validity of OK results.
ted above the composite grade, which indicates over With TSIK, variogram are calculated using indica-
estimation. Similarly, in the high grade range the block tor value (0 or 1). The variogram calculation is thus
estimate is lower than the composite grade. This is a unaffected by outliers because there are none.

263

Copyright © 2005 Taylor & Francis Group plc, London, UK

 6 CONCLUSION

Due to using borehole data and stopes sampling

obtained from high grade ore vein, the two models
TSIK and OK show that the result of grade estimating
are high. TSIK model have relatively high grade mean
than OK model, because low grade have screened with
determination a 3 g/t Au discriminator. However, Au
metal content of TSIK model is relatively conservative
than OK model.

ACKNOWLEDGEMENTS

The author would like to thank: Mario E. Rossi, great

advised for this paper; Gatut S. Adisoma (PT. Freeport
Indonesia), John M.Marek and Michael G. Hester
for applying IMC software; Nafiri and Lidana for
supplying a huge assistance so that this paper could
have done.
Figure 8. Scatter plot of the composite grade against the
OK estimate.
REFERENCES

Table 5. Linear regression statistics: scatter plot of Froidevaux, R., and Journel, A.G. Prediction of Local Ore
composite grade against estimated grade. Recoveries and Ore-Waste Ratios at the Silver Bell Ura-
nium Mine, Mathematical Geology, vol. 14, 1982. pp.
TSIK OK 645–660.
Journel, A.G. Non Parametric Estimation of Spatial Distribu-
Number of samples 255 235 tion, Mathematical Geology, vol. 15, 1983, pp. 445–468.
Intercept 3.02 4.22 Kim, Y.C., Medhi, P.K., and Roditis, I.S. Performance Eval-
Slope 0.65 0.73 uation of Indicator Kriging in A Gold Deposit, Mining
Standard error of estimate (SEE) 2.36 5.90 Engineering, 1987, pp. 947–952.
Coefficient of correlation (R) 0.70 0.60 Parker, H.M., and Chávez, W.X. The Use of Geostatistics
in Delineation of Underground Sandstone-type Uranium
Deposits, 108th Annual Meeting AIME, New Orleans,
1979.
Parker, H.M. Statistical Treatment of Outlier Data in Epither-
The poor continuity and often complex intermin- mal Gold Deposit Reserve Estimation, Mathematical
gling of ore and waste mean that precious metal Geology, vol. 23, 1991, pp. 125–199.
deposits usually have to be mined in a highly selective Parrish, I.S. Geologist’s Gordian Knot: To Cut or Not To Cut,
manner. This causes problems in trying to contruct an Mining Engineering, April, 1997, pp. 45–49.
ore reserve model from wide spaced exploration data Rossi, M.E., and Parker, H.M. Estimating Recoverable
that will correctly estimate the mineable reserves when Reserves, Is It Hopeless?, Geostatistics for the Next Cen-
the deposit is actually mined in a selective manner. OK tury (Dimitrakopoulos, R., ed.), Kluwer Academic Publ.,
Boston, 1993, pp. 259–276
method estimate the mean grade of a block that is fairly Rossi, M.E., Douglas, I.H., and Parker, H.M. Introducing
large. However, when the deposit is mined, selection Economics in Grade Control: The Expected Revenue
is based on a smaller volume, called the selective min- Method, SME Annual Meeting, February 14–17, 1994,
ing unit (SMU). The usual outcome is that large blocks Albuquerque, New Mexico.
rarely turn out to be all ore or all waste, thus making Waterman, S. The Application of Indicator Kriging in Ciki-
our reserve estimate an incorrect estimate of what will dang Gold Modeling, M.Sc. Thesis. Bandung Institute of
be mined. TSIK obtains a solution by estimating the Technology, 1999.
mean grade of the block.

264

Copyright © 2005 Taylor & Francis Group plc, London, UK