You are on page 1of 12

Iron ore resource modeling and estimation

using geostatistics
Cite as: AIP Conference Proceedings 2245, 070016 (2020); https://doi.org/10.1063/5.0006928
Published Online: 08 July 2020

Waterman Sulistyana Bargawa, and Recky Fernando Tobing

ARTICLES YOU MAY BE INTERESTED IN

Mapping of infiltration rate using Horton method in Kedungwaru Village, Karangsambung,


Kebumen, Central Java
AIP Conference Proceedings 2245, 020007 (2020); https://doi.org/10.1063/5.0007717

Optimization of Sump drying with alternatives concept at coal open mine PT Bukit Asam
South Sumatra
AIP Conference Proceedings 2245, 030008 (2020); https://doi.org/10.1063/5.0007077

Identification technique of alteration zones on site Kutacane, South-East Aceh, verified by


petrography and XRD analyses
AIP Conference Proceedings 2245, 030001 (2020); https://doi.org/10.1063/5.0007231

AIP Conference Proceedings 2245, 070016 (2020); https://doi.org/10.1063/5.0006928 2245, 070016

© 2020 Author(s).
Iron Ore Resource Modeling and Estimation Using
Geostatistics
Waterman Sulistyana Bargawaa), Recky Fernando Tobing

Department of Mining Engineering, UPN Veteran Yogyakarta, Jl. SWK 104 Yogyakarta, Indonesia

a)
Corresponding author: waterman.sb@upnyk.ac.id

Abstract. Modeling and estimation of ore grade are very essential in geostatistical ore resource estimation. Resource
modeling is generally carried out on gold, copper, nickel and bauxite ores. This study applies the geostatistical method for
modeling and estimation of iron ore grade. The objective of the study is to apply estimation techniques (OK, ordinary
kriging; IDW, inverse distance weighting, and NNP, nearest neighbor polygon) and evaluate the accuracy of these
techniques in iron ore resources. This study uses detailed exploration, which are 68 drill holes with 170 iron ore grade
composite data. In the iron ore resource estimation, the block modeling method is applied. The results showed RMSE (root
mean square error) values of various estimation techniques. Based on statistical analysis, visualization of comparisons
between borehole data and models, and probability plots, the accuracy of each iron ore resource estimation technique in
the study area can be determined. All estimation techniques have the same accuracy on low CV (coefficient of variance)
values. The relative kriging standard deviation values determine the classification of measured iron ore resources.

Keywords: Iron ore, Resource, Modeling, Geostatistics, Kriging

INTRODUCTION

Ore resource estimation has an important role in determining the number of resources [1], ore grade distribution,
3D model estimates, mine boundary, and mine life [2]. The accuracy of ore resource estimates determines production
targets, ore mining procedures [3], and investment decisions [4]. The accuracy of the estimation method is very crucial
for evaluating grade-tonnage of resources and reserves [5-6].
Various methods have been developed in estimating grade and quantity of ore resources [7-8]. Conventional ore
resource estimation methods such as cross-section, triangulation, polygon methods have rarely been used while the
nearest neighbor polygon, inverse distance weighting, kriging method is widely used in estimating ore grade of mineral
resources [9-11].
The objective of this study is to compare three interpolation techniques, namely ordinary kriging (OK), inverse
distance weighting (IDW) and nearest neighbor polygon (NNP). The technique uses omnidirectional, and anisotropy
properties in the search for samples used to estimate iron ore resources. The accuracy is based on statistical analysis,
comparison of visualization and probability plots. This research was conducted on iron ore resources. The research
location is in Merangin Regency, Jambi Province, Indonesia. The results of the detailed exploration of iron ore
produced 68 drill holes with 170 composite grades. The composite interval is 6 m, which is adjusted to high bench iron
ore mining. The dimensions of the block model are (6x6x6) m.

2nd International Conference on Earth Science, Mineral, and Energy


AIP Conf. Proc. 2245, 070016-1–070016-11; https://doi.org/10.1063/5.0006928
Published by AIP Publishing. 978-0-7354-2004-5/$30.00

070016-1
RESEARCH METHODOLOGY

Estimation methods: Estimation techniques use OK, IDW, and NNP. Parameters of distance and power usage affect
the results of the IDW estimation method [12]. The formula for this method [13] is:
∑ ×

= (1)

OK estimation method considers [14-15] the following:


Block model estimation use the equation:

= ∑ (2)
Solve weight, wi, using the equation:
∑ . , + = , dan ∑ = (3)
Kriging variance can be expressed in the equation as follows:
=∑ . , − , + (4)
The Measure of Accuracy: The difference between the data and the interpolation value is the error value at that
point [16]. The error is defined as follows:

= ∑ ∗
− (5)

Cross-validation considers the value of root mean square error (RMSE). RMSE is defined as follows [17]:

= ∑ ∗ − (6)

RESULTS AND DISCUSSION


Statistics Analysis: Statistical analysis was performed on assay and composites of iron ore grade. This analysis
aims to determine the characteristics of the data for selecting estimation techniques. TABLE 1 shows the results of the
statistics analysis.
TABLE 1. Descriptive statistics of assay and composite
Parameter Assay Composite
Minimum 59 59
Maximum 64.2 64.1
Mean 62.22 61.97
Variance 2.19 1.69
Std. Dev 1.48 1.30
CV 0.023 0.020
Skewness -0.56 -0.35
Kurtosis 2.44 2.28
Median 62 62.07
N 837 170
Based on the statistical analysis (TABLE 1), assays and composites have a low standard deviation, and have
low CV value (<0.5). FIGURE 1 shows the distribution map of drill holes in the study area.

070016-2
FIGURE 1. Distribution of Drill Holes at Research Area

Variography: This study uses 16 variograms and changes every 22.5⁰. Simulation is carried out to compare between
omnidirectional variograms (FIGURE 2) and in specific directions (FIGURE 3-4).

FIGURE 2. Omnidirectional Variogram

FIGURE 3. Variogram of N120.5⁰E Direction

070016-3
FIGURE 4. Variogram of N30.5⁰E Direction
FIGURE 2 above shows the variogram fitting of omnidirectional models. This model has the same range for
different directions. While FIGURE 3-4 shows the variogram that has a different range according to the direction
(anisotropy model). TABLE 2 shows variogram parameters from omnidirectional and variogram from various
directions.
TABLE 2. Variogram Parameters
Parameter Omnidirectional variogram 30.5⁰ direction 120.5⁰direction
Nugget 0 0 0
Sill 1.019 1.29 0.909
Range 18.008 21.263 10.388
Major /Semi-major 1 2.046 -
Major /Minor 1 1.232 -
IDW (Omnidirectional): Using statistical analysis, the IDW model is obtained a histogram (FIGURE 5) with a
skewness value of -0.34 and kurtosis was 2.21.

FIGURE 5. Histogram of Estimate Grade using IDW Model (Omnidirectional)

After the block modeling process, estimation of block grades is carried out to estimate the grade-tonnage [18] of
iron ore resources (see TABLE 3).
TABLE 3. Grade-tonnage Estimation Using the IDW Model (Omnidirectional)
GradeInterval Average Grade
Ore Tonnage
%Fe) (% Fe)
59 - 60 20,813 59.41
60 - 61 160,071 60.30
61 - 62 126,075 61.41
62 - 63 281,443 62.37
63 - 64 152,314 63.29
64 - 65 21,775 64.02
Total 762,491 61.93

Based on iron ore average grade 61.93% Fe, ore density 5 t/m3, resource tonnage was 762,491 (see TABLE 3).

070016-4
IDW (Anisotropy): As the calculation before, based on statistics (IDW model), the skewness value is -0.95 and
kurtosis is 4.35 (FIGURE 6).

FIGURE 6. Histogram of IDW Model (Anisotropy)

Block model and estimation of iron ore grades were carried out to estimate the grade-tonnage (see TABLE 4).
TABLE 4. Grade-tonnage Estimation Using the IDW Model (Anisotropy)
Grade Ore Average
Interval Tonnage Grade (% Fe)
59 – 60 11,281 59.74
60 – 61 49,355 60.50
61 - 62 245,017 61.62
62 - 63 350,583 62.33
63 - 64 30,639 63.20
Total 686,874 61.94

The quantity of resources is 686,874 tons with an average grade of 61.94% Fe, ore density of 5 t/m3. The highest
tonnage tendency is in the range of 62-63% Fe (TABLE 4). The difference in direction of a sample search causes
different grade-tonnages (TABLE 3-4). In this study, the direction of anisotropy corresponds to the spatial correlation
between data (TABLE 4).
OK: The estimation parameter consists of the maximum number of data is 15 and a minimum of 3 samples, while
the value of the sill parameter, major, minor, semi-major ranges, and directions use the results of the previous
variogram fittings. FIGURE 7 shows the histogram of the OK model, while TABLE 5 shows the grade-tonnage of
the OK model.

FIGURE 7. Histogram of Grade Estimate of OK Model

TABLE 5. Grade-Tonnage Estimation Using the OK Model


Grade Interval Volume (m3) Average Grade (% Fe) Ore Tonnage
59 - 60 1,066 59.83 5,331
60 - 61 7,162 60.66 35,811
61 - 62 64,643 61.61 323,213
62 - 63 68,92 62.36 344,760
63 - 64 5,447 63.19 27,237
Total 147,270 61.96 736,352

070016-5
TABLE 5 shows OK model with the tonnage of resources of 736,352; an average grade of 61.96% Fe. The tonnage
is almost similar to the IDW model grade estimation. The highest tonnage tendency is in the range of 62-63%.
NNP: FIGURE 8 shows the histogram of grade estimate using the NNP Model.

FIGURE 8. The histogram of Grade Estimate of the NNP Model


TABLE 6 shows the average grade of 61.37% Fe. If ore density is 5t/m3, the resource tonnage is 775,784 (TABLE
6). The highest tonnage is in the range of 62-63% Fe.
TABLE 6. Grade-Tonnage Estimation Using the NNP Model
Ore Grade
Grade Interval
Tonnage Average (% Fe)
59 – 60 6,355 59.01
60 – 61 36,237 60.72
61 – 62 310,560 61.65
62 – 63 400,913 62.29
63 - 64 21,718 63.18
Total 775,784 61.37

Statistical analysis: TABLE 7 shows the coefficient of variance of all estimation techniques show similarity value
(<0.5). The next analysis is cross-validation between block model estimates and composite data.
TABLE 7. Recapitulation of Results of Statistical Analysis
Parameter IDW Omnidirectional IDW Anisotropic OK NNP
N 3842 3352 3674 3975
Power 2 2 - -
Minimum (%Fe) 59.1 59.6 59.4 59
Maximum (%Fe) 64.05 63.5 64.6 64.1
Mean 61.91 61.93 61.92 61.97
Variance 1.46 0.43 0.52 1.69
Std. Dev 1.20 0.65 0.72 1.30
CV 0.019 0.010 0.011 0.021
Median 62.07 62.01 62.12 62.03

Cross-Validation: FIGURE 9-10 show cross-validation between composite grade and estimated grades for each
estimation technique. A low CV value results in similarity to the estimated average grade.

070016-6
y = 0,494x + 31,354 y = 0,929x + 4,388
R2 = 0,816 R2 = 0,924

(a) (b)

FIGURE 9. Cross-validation between composite vs omnidirectional IDW model (a); composite vs anisotropic IDW model (b)

y = 0,902x + 6,033
R2 = 0,921

FIGURE 10. Cross-validation between composite and OK model

TABLE 8 shows the results of the linear regression of three estimation techniques. RMSE of IDW and OK
techniques show relatively similar values. R2 values of both IDW and OK techniques are close to one. TABLE 8
shows that IDW and OK techniques have Y-intercept values close to zero and slope values close to one. Both of these
methods show good accuracy to be applied in the research area.
TABLE 8. Results of Linear Regression of Each Method
Method
Statistics
IDW IDW OK
Parameter
Omnidirectional Anisotropy
RMSE 0.638 0.179 0.187
R2 0.816 0.924 0.921
Intercept 31.354 4.388 6.033
Slope 0.494 0.929 0.902

Probability Curves: FIGURE 11 shows the IDW model curve (omnidirectional) has an overestimate at low grades
and underestimates at high grades. The IDW model curve (anisotropy) and OK model almost coincide with the
composite curve.

070016-7
FIGURE 11. Probability Curve of Composite, IDW, OK and NNP Models

Visualization Analysis: Visualization analysis uses cross-sections, and uses seven color trends, blue indicates a
range of 0-59% Fe; light blue 59-60% Fe; cyan 60-61% Fe; yellow 61-62% Fe; green 62-63% Fe; orange 63-64% Fe;
and red >64% Fe.

(a) (b)

FIGURE 12. Composite on 183431,388 E Section (a), Composite vs. Omnidirectional IDW Model (b)

FIGURE 12 shows the composite grades and the omnidirectional IDW model. The red color shows the distribution
of grades up to 64% Fe, while the green color shows the grade distribution up to 62% Fe. FIGURE 12 (b) shows
yellow blocks (61% Fe) in the vicinity of the composite. It indicates the underestimation of the omnidirectional IDW
model.

070016-8
(a) (b)

FIGURE 13. Composite vs. Anisotropy IDW Model on 183431,388 E Section (a), Composite vs. OK Model on 183431.3 E
Section (b)

FIGURE 13 (a) shows the composite grades and IDW anisotropy grades estimate. The red color shows the
distribution of grades up to 64% Fe, while the green color is the distribution of grades of up to 62% Fe. FIGURE 13
(a) shows blocks, which are close to the composite tend to be the same as composite grades.
FIGURE 13 (b) shows the composite grades and OK model estimates. FIGURE 14 shows a little underestimation
on the block surrounding the composite (yellow), but the OK model overall follows the color of the composite.

FIGURE 14. Composite vs. NNP Model on 183431.3 E Section

FIGURE 14 shows the composite grade and NNP model. NNP extrapolates grades up to half of the distance
between drill holes. This is the principle of polygons which is to divide the distance between the drill as an area of
influence. Drastic color changes will occur at half the distance between the drill holes. This shows the weakness of the
NNP model. The choice of estimation method becomes flexible in the low CV value (CV <0.5).
Resources Classification: Resource classification is implemented on the IDW method, and OK (TABLE 9). The
resource classification of the IDW model is based on the density of drill hole points [19]. The level of confidence in
the classification of resources will be higher at close drilling distances. The classification of resources in the OK
method is based on the relative kriging standard deviation (RKSD) [20], where:
Measured 0.3 ≤ Indicated ≤ 0.5 Inferred
Based on the distance between drill holes and RKSD (FIGURE 15) values (<0.3), iron ore resources are classified
as measured.

070016-9
0.025

Relative Kriging Standard


0.020
0.015

Deviation
0.010
0.005
0.000
2 3 4 5 6 7 8 9 10 11 12 13 14 15
Number of Composite

FIGURE 15. The Number of Composite vs. Relative Kriging Standard Deviation

TABLE 9 shows the results of NNP, IDW, and OK in modeling and estimating iron ore grades. Based on the table,
the IDW omnidirectional model show the highest tonnage of iron ore resources which indicates the number of blocks
estimated in the area of influence of the omnidirectional IDW. This is incompatible with spatial correlation as indicated
by anisotropic IDW. Anisotropic IDW produces a conservative tonnage of resources compared to other techniques.
Whereas the average grade estimation results show similar results.
TABLE 9. Resources Classification
Resources
No. Method Measured Indicated Inferred
Vol (m3) Tonnage %Fe Vol. Tonnage Fe Vol. Tonnage Fe
1. IDW
152,498 762,491 61.93 - - - - - -
Omnidirectional
2. IDW Anisotropy 137,375 686,874 61.94 - - - - - -
3. OK 147,270 736,352 61.96 - - - - - -

CONCLUSION
The conclusions obtained from this study are:
1. If the CV data is <0.5 various estimation methods such as IDW, OK, NNP shows the same accuracy results.
The choice of the estimation method is easier with the same accuracy.
2. Based on probability analysis, IDW, OK, NNP estimation methods show similar accuracy. This is indicated
by the coincidence of the four probability lines.
3. Based on the relative kriging standard deviation value, the classification of iron ore resources in the study area
is measured resources.

REFERENCES

[1] D. Agam, “Adapting pattern recognition approach for uncertainty assessment in the geologic resource
estimation for Indian iron ore mines,” In Proc. International Conference on Signal Processing,
Communication, Power, and Embedded System (SCOPES), 2016, pp. 1816–1821.
[2] I. O. Ferreira, “In bathymetric surfaces”: IDW or kriging, Bulletin of Geodetic, pp. 1982–2170, 2017.
[3] A. Krzemien, “Beyond the pan-European standard for reporting of exploration results, mineral resources,
and reserves,” Resources Policy, pp. 81–91, 2016.
[4] S. K. Haldar, Chapter 8: “Mineral Resource and Ore Reserve Estimation.” Mineral Exploration (Second
Edition), pp. 145-165, 2018.
[5] S. Sun and S. Anwar, “R&D activities and FDI in China’s iron ore mining industry.” Economic Analysis
and Policy, pp. 47-56, 2019.
[6] T. Morita, K. Higashida, Y. Takarada, S. Managi, “Does acquisition of mineral resources by firms in
resource-importing countries reduce resource prices?” Resources Policy, 58, pp. 97-110, 2018.

070016-10
[7] W. S. Bargawa, “Weighted jackknife ordinary kriging - problem solution of the precision in mineral
resources estimation,” IOP Conf. Series: Earth and Environmental Science 212 (012059), pp.1-9, 2018.
[8] W. S. Bargawa, “Mineral resources estimation using weighted jackknife kriging.” Advances of Science
and Technology for Society. AIP Conf. Proc. 1755, pp. 120001-120006, 2016.
[9] I. Zulkarnain and W. S. Bargawa, “Classification of coal resources using drill hole spacing analysis
(DHSA),” Journal of Geological Resource and Engineering, 6, pp. 151-159, 2018.
[10] W. S. Bargawa, “Mineral resources estimation based on block modeling.” Progress in Applied
Mathematics in Science and Engineering Proceedings. AIP Conf. Proc. 1705, pp. 020001-1-020001-8,
2016.
[11] W. S. Bargawa, A. Rauf, and N. A. Amri, “Gold resource modeling using pod indicator kriging.” Progress
in Applied Mathematics in Science and Engineering Proceedings. AIP Conf. Proc. 1705, pp. 020025-1-
120025-8, 2016.
[12] V. Senapathi, and C. R. Paramasivam, “An introduction to various spatial analysis techniques,” in GIS and
Geostatistical Techniques for Groundwater Science, 2019, pp. 23-30.
[13] B. I. Harman, H. Koseoglu, and C. O. Yigit, “Performance evaluation of IDW, Kriging and multiquadric
interpolation methods in producing noise mapping”: A case study at the city of Isparta, Turkey. Applied
Acoustics, 112, pp. 147-157, 2016.
[14] N. Paraskevis, C. Roumpos, N. Stathopoulos, and A. Adam, 2019 International Journal of Mining Science
and Technology, In press, corrected proof.
[15] K. Kang, C. Qin, B. Lee, and I. Lee, “Modified screening-based kriging method with cross-validation and
application to engineering design.” Applied Mathematical Modelling, 70, pp. 626-642, 2019.
[16] F. Aràndiga, “Adaptive rational interpolation for point values.” Journal of Computational and Applied
Mathematics, 349, pp. 212-224, 2019.
[17] L. Mentaschi, G. Besio, F. Cassola, and A. Mazzino, “Problems in RMSE-based wave model validations.”
Ocean Modelling, 72, pp. 53-58, 2013.
[18] S. A. Hosseini, O. Asghari, X. Emery, and M. Maleki, “Forecasting the grade-tonnage curves and their
uncertainty at the Mehdiabad deposit-Yazd, central Iran,” Bollettino di Geofisica Teorica ed Applicata,
vol. 58, n. 3, September, pp. 217-232, 2017.
[19] M. E. Rossi and C. V. Deutsch, Mineral Resource Estimation, Springer, 2014.
[20] A. J. Sinclair, and G.H. Blackwell, Applied Mineral Inventory Estimation, Cambridge University Press,
2005.

070016-11

You might also like