COMPARATIVE ANALYSIS OF 3D DOMAIN MODELLING ALTERNATIVES:

IMPLICATIONS FOR MINERAL RESOURCE ESTIMATES

by
Mustafa Tokoglu
 A thesis submitted to the Faculty and the Board of Trustees of the Colorado School of
Mines in partial fulfillment of the requirements for the degree of Master of Science (Mining and
Earth Systems Engineering).

Golden, Colorado

Date _________________

Signed: ______________________
______________________
Mustafa Tokoglu

Signed: ______________________
Dr. Kadri Dagdelen
Thesis Advisor

Golden, Colorado

Date _________________

Signed: ______________________
Dr. Priscilla Nelson
Professor and Head
Department of Mining Engineering

ii
 ABSTRACT

Domain modelling is a fundamental component of contemporary mineral resource

estimation. There exist four major modelling techniques that are distinct with respect to method
complexity, time and effort. A Comparative analysis is presented in this document constituting
estimation of resources using 3D estimation domains generated with four different modelling
approaches; (i) explicit modelling, (ii) implicit modelling, (iii) indicator kriging, and (iv)
conditional simulation. Furthermore, identical grade estimation method and parameters are
considered in order to demonstrate the discrepancies arising only from the choice of the domain
modelling approach and underlying assumptions.

Comparison of the outcomes indicated the significance of the domain modelling decision
on resource estimates of a polymetallic massive sulfide deposit located in western Turkey.
Economic implications are demonstrated in the form of range of outcomes for the extends of the
ultimate pit, ore tonnages (min:37.0 Mt, max: 46.7Mt), waste tonnages (min:201.2 Mt, max: 251.2
Mt), stripping ratios (min:5.18, max: 5.71), and total pit values (min: $1.05B, max: $1.45B).

The study showed that indicator kriging and simulation results are largely consistent with
each other due to similar estimation criteria considered. Moreover, these two techniques resulted
in notably larger volumes particularly for highly mineralized estimation domains compared to the
ones generated with explicit and implicit modelling techniques. Examination of solid models as
well as cross-sections revealed that major discrepancies are observed beyond outlying drillholes.
Therefore, it has been proposed that assumptions regarding extrapolation distance is the main
source of mentioned dissimilarities.

iii
 TABLE OF CONTENTS

ABSTRACT ................................................................................................................................... iii

LIST OF FIGURES ....................................................................................................................... iv

LIST OF TABLES .......................................................................................................................... x

ACKNOWLEDGEMENTS ........................................................................................................... xi

CHAPTER 1. INTRODUCTION ................................................................................................... 1

1.1. Problem Statement ........................................................................................................... 2

1.2. Thesis Goal and Objectives .............................................................................................. 3

1.3. Thesis Structure ................................................................................................................ 3

CHAPTER 2. BACKGROUND: GEOSTATISTICAL RESOURCE ESTIMATION .................. 5

2.1. Data Validation ................................................................................................................ 5

2.2. Geological Constraints on Mineralization........................................................................ 5

2.3. Exploratory Data Analysis ............................................................................................... 7

2.4. Variography.................................................................................................................... 11

2.5. Anisotropy ...................................................................................................................... 14

2.6. Model Fitting .................................................................................................................. 15

2.7. Estimation Criteria ......................................................................................................... 16

2.8. Search Neighborhood ..................................................................................................... 18

2.9. Grade Estimation ............................................................................................................ 20

2.10. Recoverable Resources .................................................................................................. 23

2.11. Model Validation............................................................................................................ 26

2.12. Resource Classification .................................................................................................. 28

iv
CHAPTER 3. GEOLOGY OF THE DEPOSIT ............................................................................ 31

3.1. Regional Geology........................................................................................................... 31

3.2. Local Geology ................................................................................................................ 32

3.3. Deposit Geology............................................................................................................. 34

3.4. Mineralization ................................................................................................................ 35

CHAPTER 4. STATISTICAL CHARACTERIZATION & ESTIMATION DOMAIN . . . . .

.. DEFINITION ......................................................................................................... 36
4.1. Quality Assurance and Quality Control ......................................................................... 37

4.2. Database Validation ....................................................................................................... 39

4.3. Statistical Data Analysis................................................................................................. 40

4.4. Delineation of Estimation Domains ............................................................................... 43

CHAPTER 5. ESTIMATION DOMAIN MODELLING ............................................................. 52

5.1. Explicit Modelling Approach ......................................................................................... 53

5.2. Implicit Modelling Approach ......................................................................................... 54

5.3. Indicator Kriging Approach ........................................................................................... 56

5.4. Conditional Simulation Approach .................................................................................. 58

CHAPTER 6. RESOURCE ESTIMATION & CLASSIFICATION ........................................... 60

6.1. Grade Estimation ............................................................................................................ 60

6.1.1. Block Model Setup ................................................................................................. 60

6.1.2. Geostatistical Data Analysis ................................................................................... 63

6.1.3. Estimation Strategy ................................................................................................. 64

6.1.4. Resource Classification ........................................................................................... 66

6.2. Model Validation............................................................................................................ 67

v
 6.3. Ultimate Pit Limit Analysis ........................................................................................... 71

CHAPTER 7. DISCUSSION & COMPARISON OF OUTCOMES ........................................... 75

CHAPTER 8. CONCLUSION...................................................................................................... 83

REFERENCES ............................................................................................................................. 86

APPENDIX A. CONTACT PLOTS ............................................................................................. 88

APPENDIX B. MODELLING METHODS ................................................................................. 91

APPENDIX C. 3D DOMAIN MODELS ..................................................................................... 89

APPENDIX D. MEASURED, INDICATED AND INFERRED RESOURCES OF…….

…………………...INDIVIDUAL ESTIMATION DOMAINS ................................................... 96
APPENDIX E. COMPARISON OF ESTIMATION DOMAINS……………...………………...97

vi
 LIST OF FIGURES

Figure 1.1 Alternative geological interpretations ........................................................................... 2

Figure 2.1 Sketch illustrating sharp and gradational contacts (Abzalov, 2016) ................................ 6
Figure 2.2 An illustration of higher uncertainty near contact (Rossi & Deutsch, 2014) .................. 7
Figure 2.3 Common representation of histogram (left) and probability plot (right) (Rossi &
…………..Deutsch, 2014) .............................................................................................................. 8
Figure 2.4 Cell declustering technique (left), plot of declustered mean vs. cell size (right)
…………..(Rossi & Deutsch, 2014) ........................................................................................................ 9
Figure 2.5 Polygonal declustering technique and declustering weights (Abzalov, 2016) .............. 10
Figure 2.6 Data pairing and h-scatterplots for three separation distances (Abzalov, 2016) .......... 11
Figure 2.7 Correlogram (left) versus covariance function (right) (Caers, 2011) ............................. 12
Figure 2.8 Relationship between correlogram, covariance function and variogram
…….…….(Caers, 2011) ............................................................................................................... 13
Figure 2.9 Principal components of a variogram (Rossi & Deutsch, 2014) ..................................... 13
Figure 2.10 Variogram tolerances (Deutsch, 2015) ............................................................................. 14
Figure 2.11 Variogram map (left) and contour map(right) showing anisotropy directions
……………(Abzalov, 2016) ........................................................................................................ 15
Figure 2.12 Common variogram models............................................................................................... 16
Figure 2.13 Hypothetical error distributions; underestimation (left), overestimation (middle)
…………… and no bias (right) (Isaaks & Srivastava, 1989) ............................................................ 17
Figure 2.14 Three different hypothetical error distributions; skewed distribution (left),
…………… symmetric distribution with small spread (middle), symmetric distribution with
,,,,,,,,,,,……,,large spread (Isaaks & Srivastava, 1989) ....................................................................... 17
Figure 2.15 Conditional bias. Globally unbiased but conditionally biased estimates (left),
……………both globally and conditionally unbiased estimates (Isaaks & Srivastava, 1989) .... 18

Figure 2.16 Cross-sectional estimation method (Abzalov, 2016)................................................. 20

Figure 2.17 Comparison of interpolation and simulation estimates (Rossi & Deutsch, 2014) .... 22
Figure 2.18 Volume-variance relationship (Rossi & Deutsch, 2014) ........................................... 24
Figure 2.19 Misclassification due to information effect (Vann et al., 2003) ................................ 27
Figure 2.20 Relationship between exploration results, mineral resources and ore reserves
……………(JORC Code, 2012)................................................................................................... 28
Figure 2.21 Illustration of three aspects considered in resource classification
……………(Rossi & Deutsch, 2014) ........................................................................................... 30
Figure 3.1 Tectonic Map of Turkey (van Hinsbergen, et al., 2016) ............................................. 32

vii
Figure 3.2 Geological Map of the Study Area .............................................................................. 33
Figure 3.3 Generalized Columnar Section .................................................................................... 33
Figure 3.4 SE-NW geological cross section ................................................................................. 34
Figure 4.1 Drillhole Locations ...................................................................................................... 36
Figure 4.2 Validation of drillholes collar positions ...................................................................... 40
Figure 4.3 Histograms of composite data ..................................................................................... 42
Figure 4.4 Estimation domains in a SE-NW cross-section ........................................................... 45
Figure 4.5 Box-plot diagrams for geostatistical estimation domains ........................................... 47
Figure 4.6 Illustration of top-cut determination from histogram .................................................. 49
Figure 4.7 Contact plots for oxide-sulfide boundary .................................................................... 50
Figure 5.1 An illustration of digitized contacts for HGS zone along a SE-NW cross-section ..... 53
Figure 5.2 Digitized contacts along orthogonal directions (left) and solid model of HGS zone
…………..generated by explicit modelling approach (right) ....................................................... 54
Figure 5.3 Solid model of HGS zone generated by explicit modelling (left) and implicit
…………..modelling (right) approaches .............................................................................................. 56
Figure 5.4 Indicator variogram constructed for HGS+LGS+ENR+GOS zones ........................... 57
Figure 5.5 Solid Models for HGS zone generated with explicit modelling, implicit modelling
…………..and indicator kriging approaches ........................................................................................ 57
Figure 5.6 Bar chart of the volumes for HGS solids generated with alternative approaches ....... 58
Figure 5.7 Map view of solid models for oxide zone generated by four alternative approaches . 59
Figure 6.1 Block model extends ................................................................................................... 61
Figure 6.2 Modelled Variograms for GOS (left) and HGS+LGS+ENR (right) ........................... 64
Figure 6.3 An illustration of resource classification ..................................................................... 67
Figure 6.4 Histogram of Cu grade estimates for original model vs. alternative models .............. 68
Figure 6.5 Grade-tonnage curves for original model vs. alternative models ................................ 68
Figure 6.6 Swath plot of Cu grades for original model vs. alternative models ............................ 69
Figure 6.7 Visual validation of resource model ............................................................................ 69
Figure 6.8 Histograms of block model grade vs. composite grade for six estimation domains ... 70
Figure 6.9 Grade-tonnage curves of block model vs. composites for six estimation domains..... 71
Figure 6.10 An illustration of the ultimate pit limits on a SE-NW cross-section ......................... 73
Figure 7.1 Comparison of HGS domain generated with alternative modelling approaches ........ 77
Figure 7.2 Swath plots for alternative scenarios ........................................................................... 78

viii
Figure 7.3 Misclassification of the central block .......................................................................... 79
Figure 7.4 Ultimate pit limits for alternative modelling approaches in section view ................... 80
Figure 7.5 Ultimate pit limits for alternative scenarios in map view............................................ 81
Figure 7.6 Bar charts for ore tonnage, waste tonnage and stripping ratio .................................... 81
Figure 7.7 Bar chart of pit values for alternative scenarios .......................................................... 82
Figure A.1 ClyGos-Gos contact plots ........................................................................................... 88
Figure A.2 Mpy-MpyMag contact plots ....................................................................................... 88
Figure A.3 HGS-LGS contact plots .............................................................................................. 89
Figure A.4 GOS-NGO contact plots ............................................................................................. 89
Figure A.5 HGS-BW contact plots ............................................................................................... 90
Figure A.6 LGS-BW contact plots................................................................................................ 90
Figure C.1 3D models of gossan zone (GOS)............................................................................... 89
Figure C.2 3D models of non-gossan oxide (NGO) ..................................................................... 90
Figure C 3 3D models of barren wall rock (BW) ......................................................................... 91
Figure C.4 3D models of high-grade sulfide (HGS) ..................................................................... 92
Figure C.5 3D models of low-grade sulfide (LGS) ...................................................................... 93
Figure C.6 3D models of enriched zone (ENR) ............................................................................ 94
Figure E.1 Comparison of LGS domain generated with alternative modelling approaches ........ 97
Figure E.2 Comparison of Gossan zone generated with alternative modelling approaches ......... 97

ix
 LIST OF TABLES

Table 4.1 Summary statistics for all the raw and composite data ................................................. 41
Table 4.2 Summary statistics for raw and composite data in length-weighted scenario .............. 41
Table 4.3 Summary statistics for primary protolith and ore types ................................................ 43
Table 4.4 Summary statistics for geostatistical estimation domains ............................................ 48
Table 4.5 High-grade cut-off values ............................................................................................. 49
Table 6.1 Block model fields ........................................................................................................ 61
Table 6.2 Specific gravity values of estimation domains ............................................................. 62
Table 6.3 Summary table of variogram models ............................................................................ 64
Table 6.4 Kriging search neighborhood parameters ..................................................................... 65
Table 6.5 Updated process recoveries .......................................................................................... 72
Table 6.6 Smelter Terms ............................................................................................................... 72
Table 6.7 Mineral resources of the deposit ................................................................................... 74
Table 7.1 Measured and indicated resources for alternative scenarios ......................................... 75
Table 7.2 Measured, indicated and inferred resources ................................................................. 77
Table C.1 Volumes for estimation domain solid models .............................................................. 95
Table D.1 Measured, indicated and inferred resources for oxide zone ........................................ 96
Table D.2 Measured, indicated and inferred resources for sulfide zone ...................................... 96

x
 ACKNOWLEDGEMENTS

Despite words being weak considering their support, I would like to express my sincere
gratitude to many people who have contributed to the research related to this thesis.

First of all, I would like to thank my advisor Dr. Kadri Dagdelen for his wisdom, clear
guidance, continual support, and encouragement throughout this study.

I wish to thank to members of the thesis committee who generously accepted to be a part
of my jury and supported me with their invaluable guidance. I certainly owe a great debt of
gratitude to Dr. Elizabeth Holley for her technical advice and suggestions especially during the
proposal stage of this study. Without her suggestions and critical approach, this work would never
have been completed on time. Dr. M. Stephen Enders was the one who made me aware of various
aspects of the industry during the time that I have spent at Mines. I would also like to thank Mr.
Dean Turner for his advice and fruitful discussions regarding technical aspects of contemporary
mineral resource estimation.

I would like to express my gratitude to my thesis advisor from MTA, Ms. Mutlu Yakici
Icli, who was friendly and helpful during my study.

I extend my thanks to authorities of Turkish Ministry of National Education (MEB) and

General Directorate of Mineral Research and Exploration (MTA) for their enormous generosity in
funding my studies at Colorado School of Mines.

At last but not the least, I would like to express my deepest gratitude to my family for their
endless support, patience and understanding.

xi
 CHAPTER 1

INTRODUCTION

Mineral resources are the primary assets of mining companies, and estimation of resources
is an important task forming the basis of capital intensive investment decisions. It is apparently a
challenging process calling for extensive expertise and input from several sources each of which
introducing its related uncertainties. In this perspective, quantification of uncertainties associated
with each component of estimation process is critical.

In the last decades, substantial role of quantitative analysis of geological risk and
uncertainty in evaluating projects’ upside potential and downside risks has been realized. Studies
indicated that mining companies have been severely impacted because of not meeting project
expectations. As noted by Vallee (2000) “… in the first year of operation after start-up, 60% of
mines surveyed had an average rate of production less than 70% of designed capacity”. Several
other studies on mining projects also revealed that uncertainties related with the resource model
are the one of the major contributors of mentioned shortfalls.

Identification of mineralization controls and subsequent three-dimensional geological

modelling is an essential component of resource estimation. It is a complex process involving
many unknowns and assumptions. Furthermore, geological modelling is important process having
pronounced economic implications for the resource estimates.

There exists a variety of 3-D geological modelling alternatives representing differences in

terms of complexity, time and effort involved. Traditional methods of geological modelling are
those involving time-consuming manual digitization of geological boundaries. On the other hand,
recent modelling techniques significantly reduce time and effort. The quality of the geological
model is clearly important because it influences downstream mining practices. Selection of the
modelling approach should involve consideration of not only complexity, time and effort but also
the potential influences of the method on subsequent steps in the mining process.

1
1.1.Problem Statement

A typical resource modeling routine involves construction of stationary estimation domains

which are representing statistically homogenous volumes ideally guided by knowledge about
major mineralization controls. If we assume following ellipses represent alternative geological
interpretations all honoring the drill hole intercepts, it is clear that uncertainty of geologic contacts,
that is a natural outcome of our limited information, has a great potential to have profound
implications on mineral resource estimates.

Figure 1.1 Alternative geological interpretations

The problem can simply be illustrated as in Figure 1.1. DH-1 and DH-2 are two vertical
drillholes for which mineralized intercepts are indicated with red. On the particular cross-section
in Figure 1.1a, continuous blue line and dashed brown lines represent the two alternative
geological interpretations both of which honor the mineralized intercepts observed in the drillholes
but having certain differences in term of the area (and volume in third dimension) shown as green
(A) and orange (B) colored portions in the figure above. Figure 1.1b illustrates an even more severe
case where the two mineralized intercepts are interpreted to be belonging to en échelon veins.

In this perspective, decision of the geological modeling approach has a high potential to
control the tonnage of the deposit. Moreover, slight variations in the mineralization geometry
might significantly impact unplanned dilution. In cases where there exist pronounced differences
in terms of metallurgical and geotechnical characteristics across geological domains, different
geological interpretations might result in significantly different outcomes for the downstream

2
mining practices. Therefore, geological modeling efforts that are not capturing the geological
variability of the deposit are likely to be suboptimal.

This study aims to address the question that how does domain modeling method utilized
may influence mineral resource estimates. It involves consideration of identical grade estimation
method and parameters with an objective to demonstrate the discrepancies arising only from the
choice of the domain modelling approach and underlying assumptions

1.2.Thesis Goal and Objectives

The primary objective of this thesis is to execute a comparative analysis of domain modelling
alternatives. It is further aimed to document implications of each modelling routine on mineral
resource estimation outputs of a polymetallic massive sulfide deposit located in western Turkey.
Following are the goals set to achieve the objectives of the thesis:

 Identification of the major geological controls on mineralization

 Execution of modelling alternatives to generate three-dimensional representations of the
deposit
 Estimation of mineral resources for each alternative scenario
 Comparison of resource estimates
 Address inconsistencies among outputs of different scenarios
 Quantify major risks arising from geological uncertainty
 Provision of conclusions

1.3.Thesis Structure

For ease of reference, the remainder of the thesis is divided into following major sections:

 Chapter 1 introduces the subject of the thesis and provides thesis goal and objectives as well
as its outline
 Chapter 2 presents a literature review of various resource estimation techniques together with
description of modelling alternatives
 Chapter 3 introduces the study area and describes the regional and local geology of the deposit
 Chapter 4 involves statistical and geostatistical characterization of the data and identification
of the major mineralization controls

3
 Chapter 5 covers the details of implementation of explicit modeling, implicit modeling,
indicator kriging and conditional simulation in geological modelling of the deposit
 Chapter 6 deals with the aspects regarding resource estimation and classification
 Chapter 7 involves comparison of the outcomes and discussions
 Chapter 8 addresses the conclusions

4
 CHAPTER 2

BACKGROUND: GEOSTATISTICAL RESOURCE ESTIMATION

Mineral resource estimation is a challenging task calling for extensive expertise and
contribution from several disciplines. It involves execution of a sequence of steps in construction
numerical models using limited amount of data available about the deposit. Although there exists
a variety of resource estimation techniques ranging in complexity, principle steps included in
construction of resource models are common. It involves collection and validation of data,
definition of geological constraints on mineralization, delineation of estimation domains, statistical
and geostatistical analysis of the data, grade assignment using an appropriate estimation technique,
and resource classification. Assessment of the uncertainty and validation of the results are also
fundamental components of a typical resource modelling.

There are numerous unknowns and assumptions involved in modelling practice. Therefore,
it is thought to be valuable to revisit and briefly summarize the primary steps and widely accepted
techniques in terms of their application and limitations.

2.1. Data Validation

Data validation is the first and most crucial step in resource modelling because quality of
the outcomes is directly related with that of the data. Therefore, validation of the data is needed to
ensure the data to be used for resource estimation is clean, useful and consistent. The checks should
ideally start in the field and generally cover sampling collection and preparation procedures,
topographic and down the hole surveying as well as quality control and quality assurance practices.
On the other hand, validation of resource database involves identification of numerical
inconsistencies with basic validation routines and constraints. Some of the automated validation
routines include from-to checks, stoichiometric checks and values within range checks.

2.2. Geological Constraints on Mineralization

Ore genesis is strongly controlled by complex geological processes. Thus, sound

understanding of geology is an essential component in mineral resource estimation. It should also
be emphasized that “a resource model with no geologic support is inadequate …” (Rossi &

5
Deutsch, 2014). Geological data gathered in different forms namely drillhole logs, products of
geophysical investigations, surface and underground maps can be utilized in construction of 3D
models representing geological characteristics of the deposit. This kind of models may also be
particularly valuable in delineation of distinct geotechnical and metallurgical zones. Traditional
method of constructing 3D geological models from input geological variables which relies on
interpretation and manual digitization of cross-sections is known as explicit modeling. On the other
hand, use of 3D interpolation functions in construction of geological surfaces forms the basis of
implicit modeling techniques. Implicit modeling has several advantages over explicit modeling.
For instance, three dimensional models of complex deposits can be generated in both time and
cost-effective manner. Furthermore, larger datasets can be incorporated and data from various
sources can be used in generation of a single model. However, none of these advancements
eliminate the vital role of geologist in modeling practice.

Stationary is a property of random functions (Isaaks & Srivastava, 1989), and also the key
condition for validity of the geostatistical models. Stationarity is also a requirement to make
inference. In its simplest sense, it represents statistically homogenous populations (Rossi &
Deutsch, 2014). Delineation of geostatistical stationary estimation domains is one of the
challenging tasks of a resource geologist that necessitates understanding of deposit geology and
statistical data characteristics. It is an iterative process calling for consideration of reasonable
combinations of geological attributes and assessment of defined domains to assure representation
of major mineralization controls.

An integral step of stationary domain definition is characterization of mineralization

contacts. Sharp and gradational contacts are the two primary contact types. Sharp contacts are
characterized by abrupt changes in mineralization grades across contact (Figure 2.1). On the other
hand, gradual change of grades represents gradational contacts (Figure 2.1).

Figure 2.1 Sketch illustrating sharp and gradational contacts (Abzalov, 2016)

6
 Contact analysis is performed with construction of diagrams representing grade
distribution profiles as a function of distance from contact. The ultimate objective of contact
analysis is to determine the estimation strategy. Hard boundary approach is usually applied to
deposits that are characterized by sharp contacts. On the other hand, soft boundary approach is
preferred for deposits with gradational contacts.

Estimation of dilution and mining losses calls for quantification of uncertainty associated
with the contacts. Conditional simulation techniques enable generation mathematically
equiprobable contact models which was not possible with deterministic models. Outputs of
conditional simulation capture the original variability of input data hence a better representation
of the reality.

Spatial data availability compounded with intrinsic variability usually results in higher
uncertainty along contacts as illustrated in Figure 2.2. It is a significant issue because reliability of
tonnage estimates directly depends on that of estimation domains. Therefore, quantitative analysis
of contacts through geologic interpretational and conditional simulation techniques is substantially
important.

Figure 2.2 An illustration of higher uncertainty near contact (Rossi & Deutsch, 2014)

2.3. Exploratory Data Analysis

Sound understanding of the data and mineral deposit is an essential foundation for resource
modelling. Tradition statistical techniques serve as effective data analysis tools that help enhance
understanding of the data and ensure data quality. The data analysis is also substantial for selecting
the most appropriate estimation method.
7
 Equal sample support is the primary requirement of the data analysis i.e., samples should
represent equal volume. Therefore, a special treatment known as compositing is performed to
regularize the sample support. There are various compositing methods. However, bench
compositing and downhole compositing are the most commonly applied ones. A key consideration
about compositing is selection of the compositing length. Abzalov (2016) suggests the chosen
composite size to be larger than average sample length and approximately equal to half of the
block size.

Frequency table and histogram are the most common presentations of univariate data i.e.,
data of a single variable (Figure 2.3). Probability plots are other univariate data analysis tools and
are utilized to identify the distribution and assess presence of multiple populations (Figure 2.3).
Summary statistics is useful as a descriptive tool that is capable to capture key features of the data
presented by histograms.

Figure 2.3 Common representation of histogram (left) and probability plot (right) (Rossi &
Deutsch, 2014)

Correlations and dependencies between variables sometimes help improve understanding

of the data. The methods that are utilized in the analysis of relationship between two variables are
in the scope of bivariate statistics. Scatter plots are used in comparison of different variables,
facilitate detection of clusters and identification of outliers. The relationship between variables are
quantified by correlation coefficient (ρ(h)). Q-Q plots and P-P plots are practical tools to compare

8
distributions of two variables. Together with scatterplots, these are also effective instruments for
error checking.

An important aspect of analysis of earth science data is clustering. Considering the

resources available for exploration, there is a natural tendency towards preferentially sample
higher-grade zones. It is an effective and legitimate strategy to allocate limited exploration
resource to delineate high potential areas faster. However, clustering introduces bias to both
statistical analysis and variography. Declustering is the method by which influence of data
clustering is overcome by assigned weights which are representative of data clustering.

Cell declustering technique divides the sample space into regularly sized cells (Figure 2.4).
The weights are assigned to data based on the number of data within the cell n, and number of
occupied cells L as follows:

= (2.1)
∗�

The weight calculated in this way are greater than zero and sum to one.

Figure 2.4 Cell declustering technique (left), plot of declustered mean vs. cell size (right) (Rossi
& Deutsch, 2014)

The estimate of weights obtained by cell declustering method is not unique and depends
on the cell size chosen. Plot of declustered mean vs cell size is an effective tool in selecting the
optimum cell size (Figure 2.4).

9
 As it is the most common case in mineral exploration, high grade areas are preferentially
sampled which result in overrepresentation of those areas. Our purpose is to prevent potential
overestimation by declustering. Therefore, the optimum cell size is selected to be the one for which
minimum declustered mean is obtained as shown in Figure 2.4. The opposite applies in a case
where low-grade area is preferentially sampled for a particular purpose.

Cell declustering perform well with the sample locations following a pseudo regular grid.
When it is not the case, polygonal declustering can be considered (Figure 2.5). It is a technique
by which declustering weights are assigned proportional to the area of polygons.

Figure 2.5 Polygonal declustering technique and declustering weights (Abzalov, 2016)

Another declustering technique is nearest-neighbor. It is applied to a regular grid of nodes

or blocks. With this technique closest datum is assigned to each grid block or grid node.

Analysis of local spatial behavior of the data is also concerned in exploratory data analysis.
Moving window is a useful tool to understand local mean and local variability across sampling
space. When the local mean and variance are found to be correlated, it is called as proportional
effect. Presence of proportional effect may indicate skewed data, lack of spatial homogeneity or
presence of spatial trends (Rossi & Deutsch, 2014). When it is not treated appropriately, it may
lead to distortions in variograms making their interpretation more difficult.

10
2.4. Variography

Spatial continuity or variability is an essential concept for geostatistical studies. Mineral

deposits are generated through successive geological processes. Most of the time, physical and
chemical factors that are controlling ore genesis result in distinctive patterns of spatial correlation.
Quantitative description of spatial continuity is fundamental in geostatistics.

The most obvious way to compare two values Z(u) and Z(u+h) whose locations are
separated by h (certain distance in a particular direction) is to take the difference, [Z(u) - Z(u+h)].
Then, plots can be generated by pairing all possible data values that are separated by a lag distance
“h” along a particular direction. These diagrams are known as h-scatterplots (Figure 2.6).

Figure 2.6 Data pairing and h-scatterplots for three separation distances (Abzalov, 2016)

In essence, samples located closer to each other are more likely to have similar values than
those located further apart. Therefore, the shape of cloud of points gets more diffuse as the
similarity decreases with increasing separation distance as it is observed in Figure 2.6. Correlation
coefficient (ρ(h)) makes quantification of information presented on a h-scatterplot. The relation
between correlation coefficient and distance along a particular direction can be analyzed with plots
that are known as correlation function or correlogram (Figure 2.7). Correlogram is a measure of
spatial continuity, and calculated as:

11
 �[ � − ∗ �+ℎ − ]
ρ(h) =
� �

(2.2)

m = E[Z(u+h)] = E[Z(u)] (2.3)

A similar measure of spatial continuity is the covariance function. The plots representing
the relationship between covariance and distance is known as covariance function (Figure 2.8).
Covariance is calculated as:

C(h) = [ − ∗ +ℎ − ] (2.4)

Figure 2.7 Correlogram (left) versus covariance function (right) (Caers, 2011)

Moment of inertia is another index to quantify the spread of point cloud around y=x line.
The plot representing the relationship between moment of inertia and separation distance is
traditionally called as semivariogram or simply variogram. Variogram is a measure of dissimilarity
of data values expressed as a function of increasing separation distance. It is calculated as:

γ(h) = E[ Z u – Z u + h 2
] (2.5)

For a stationary case, following relationships exists between variogram (γ(h)), covariance
(C(h)):

γ h =C 0 −C h or ℎ = 0 −γ h (2.6)

Constant and location independent mean and variance is the model decision forming the
basis of this relation (Rossi & Deutsch, 2014).

12
Figure 2.8 Relationship between correlogram, covariance function and variogram (Caers, 2011)

Nugget effect, sill and range are the principal features of the variogram (Figure 2.9).

Figure 2.9 Principal components of a variogram (Rossi & Deutsch, 2014)

At the origin where the separation distance (h) is zero the value of the semivariogram is
also zero. However, sampling errors or small-scale variability in the data values may result in a
discontinuity in the origin which is known as nugget effect. The semivariogram value is
continuously increasing with increasing distance. Then it reaches a plateau and flattens. The value
of the plateau is called as sill. The distance at which variogram flatten out and beyond which
samples are spatially uncorrelated is known as range.

Unfavorable field conditions make it difficult to find data pairs that are located separated
exactly by the same distance. Therefore, reasonable tolerances need to be specified for both
distance and direction. It is also common practice to introduce a limit for the bandwidth (Figure
2.10).

13
 Figure 2.10 Variogram tolerances (Deutsch, 2015)

Any sample falling within an area defined by lag distance, azimuth and associated
tolerances would be paired with the sample at the origin.

For any particular separation distance, all the pairs are considered for all possible directions
in a omnidirectional variogram for which angular tolerance is 90o. Due to large number of data
pairs, omnidirectional variograms are generally well-behaved and easily interpretable giving early
warning for erratic directional variograms (Isaaks & Srivastava, 1989).

As the name implies, downhole variograms are constructed along down the hole direction
hence adjacent samples are considered. Therefore, it provides a good estimate of nugget effect as
well as short scale continuity (Rossi & Deutsch, 2014).

2.5. Anisotropy

In many mineral deposits, the data values represent more continuity along certain direction
than others. This concept known as anisotropy. Geometric anisotropy is primarily concerned type
of anisotropy for which range of variograms are different for different directions. On the other
hand, zonal anisotropy represents the case where sill of the variogram changes with direction.

Directional variograms are utilized to identify anisotropy potentially existing in the data.
Conventional approach in finding major anisotropy axis rely on construction of directional
variograms in several directions and tracing one of the contours. These types of plots are known
14
as rose diagrams and represent elliptical shapes in the case of anisotropy. The diagrams
representing several contours of variogram values are known as variogram contour maps. An
alternative tool to quantify spatial anisotropy is the variogram map which is a 2D diagram
representing variogram values for different distances (Figure 2.11).

Figure 2.11 Variogram map (left) and contour map(right) showing anisotropy directions (Abzalov,
2016)

Longest continuity direction on either rose diagram, variogram contour map or variogram
map is the direction along which largest range value is observed. Similarly, minimum continuity
direction is the one along which range gets the smallest value.

2.6. Model Fitting

Spatial continuity is effectively described by experimental directional variograms.

However, solution of the kriging system might necessitate knowledge of variogram values for
some distance or direction for which we do not have a sample variogram value (Isaaks &
Srivastava, 1989). Therefore, special models that can be utilized to derive variogram values for
any possible vector is needed. Positive definiteness is the condition for a function to be a valid
variogram model and it guarantees the solution to exist and be unique.

Nugget effect, spherical, gaussian, exponential and linear are the basic models that are
characterized by different behaviors and functions (Figure 2.12).

15
 Figure 2.12 Common variogram models

It is not necessary to model the sample variogram using just one of the basic models. Owing
to the fact that any linear combination of positive definite models, called as nested structures, also
satisfies the condition of positive definiteness, two or more models can be linearly combined to
improve representativeness of sample variogram by the model. However, it is often recommended
to keep the model as simple as possible while capturing necessary detail in the sample variogram.

2.7. Estimation Criteria

Assessment of the performance of different estimation methods relies on comparison of

estimates ( ̂) with true grade values (v). Such a comparison is a straightforward process when the
performance is assessed for estimation at a single point. On the other hand, some estimation
criterion is needed for comparing estimates at several locations. A perfect estimation method is the
one that yields estimates matching the true values. However, all methods involve some error no
matter how sophisticated it is. Therefore, a method is adequate as long as it produces estimates
that are very close to the true values. Error can simply be defined as the difference between
estimate and true value at a given location. It is also often referred as residual. Univariate and
bivariate statistical tools can be utilized to assess the performance of estimation methods. Ideally,
estimation should result in residuals whose distribution centered at zero. This represents absence
of clear estimation error hence referred as unbiasedness condition.

16
Figure 2.13 Hypothetical error distributions; underestimation (left), overestimation (middle) and
no bias (right) (Isaaks & Srivastava, 1989)

Figure 2.13 represents three hypothetical error distributions. Mean of the residuals for the
histogram in the left is negative which is referred as underestimation. The histogram in the middle
illustrates the opposite case, thus it is overestimation. Mean of the residuals are centered at zero
for the histogram on the right so no bias is evident.

It is also preferable that the error distribution is symmetric and has a small spread. Median
and variance of error distribution are good checks on symmetry and spread, respectively.

Figure 2.14 Three different hypothetical error distributions; skewed distribution (left), symmetric
distribution with small spread (middle), symmetric distribution with large spread (Isaaks &
Srivastava, 1989)

Mean absolute error and mean squared error are two summary statistics incorporating both bias
and spread of error distribution. They are defined as:

� =� = ∑�= | | (2.7)

� � = �� = ∑�= (2.8)

It is also significant to assess the performance of an estimator for full spectrum of estimates
because global unbiasedness does not guarantee unbiasedness at different scales. It is known as

17
conditional unbiasedness and can be diagnosed from scatterplots of estimates versus the true values
(Figure 2.15).

Figure 2.15 Conditional bias. Globally unbiased but conditionally biased estimates (left), both
globally and conditionally unbiased estimates (Isaaks & Srivastava, 1989)

2.8. Search Neighborhood

Kriging strategy involves decisions about size and shape of the search neighborhood as
well as restrictions to prevent redundancy and ensure relevance of nearby samples to be used in
estimation. Shape of the search neighborhood is a decision that is guided by variogram analysis. It
is usually an ellipse whose major axes align with the anisotropy pattern. On the other hand, size of
the search neighborhood is a function of both variogram range and data spacing along with the
relative nugget i.e., the ratio of nugget variance to total variance. Quadrant search is a customary
practice that is utilized to deal with possible redundancy of samples. It typically involves definition
of maximum and minimum number of samples for any particular quadrant.

Kriging is often described as best, linear, unbiased estimator minimizing the error variance.
Vann et al. (2003) emphasized that kriging is the minimum variance estimator only when kriging
search neighborhood is properly defined. Determination of search strategy is a critical process
having substantial impacts on the quality of outcomes. It is not an arbitrary decision rather calling
for quantitative criteria to optimize search neighborhood.

Overly-restricted search neighborhoods result in conditional bias (Krige 1994, 1996, Vann et
al., 2003). Therefore, kriging neighborhood analysis involves quantitative measures to minimize
conditional bias through optimizing search neighborhood. Following are some of the most

18
commonly utilized evaluation criteria in quantitative kriging neighborhood analysis outlined by
Vann et al. (2003):

 Kriging variance
 Slope of regression for true values and associated estimates
 Proportion of negative kriging weights
 Weight of the mean value for simple kriging

Kriging variance is also referred to as estimation error and was proposed as a criterion to
optimize neighborhood. It depends on data configuration and variogram model but independent of
individual data values. It is an effective tool giving an idea about relative quality of estimation in
which search neighborhood resulting in lower kriging variance is more favorable.

Perfect estimation is the one for which the estimates exactly matching corresponding true
values. Therefore, it makes slope of regression to be equal to 1. Departure from unity for slope of
regression is a good check for degree of bias. Slope of regression (ρ |Zv ∗ is defined in term
of covariance (Cov [Zv, Zv*]) and variance (Var [Zv*]) for true (Zv) and estimated values (Zv*):

v [Zv,Zv∗]
(ρ |Zv ∗ ) =
Var [Zv∗]
(2.9)
It can also be expressed in terms of correlation coefficient (r) and standard deviations (StD)
as:

St Zv
(ρ |Zv ∗ ) = (2.10)
St Zv∗

Proportion of negative kriging weights is another quantitative criterion proposed to

optimize search neighborhood. Negative weights in kriging arise as a result of screening effect.
Vann et al. (2003) suggests that it is not problematic as long as negative weights represent a small
proportion (<5%). One of the objectives of quantitative kriging neighborhood is therefore to find
most optimum search neighborhood that minimizes percentage of negative weights.

Weight of the mean value in simple kriging is yet another measure of kriging quality. It is
a measure of weakness of screening effect and suggested to be close to zero for better quality
estimation.

19
2.9. Grade Estimation

Traditional or classical estimation methods are a family of non-geostatistical methods

including polygonal method, nearest-neighbor, triangulation, cross-sectional method, and inverse
distance weighting. They are useful tools for preliminary assessment of the resources providing
order-of-magnitude estimates.

Polygonal method is a 2-D estimation technique applicable for tabular deposits.

Construction of polygonal area of influence around drillholes and grade and thickness of the
drillhole is extrapolated to the surrounding polygon. Nearest-neighbor is a variant of polygonal
method. In this technique, the grade of the nearest data is allocated directly to blocks in the block
model. Both methods provide unsmoothed estimates. Consequently, artificial discontinuities are
produced. An alternative estimation method is triangulation that overcomes this limitation.
Triangulation is based on joining adjacent drillholes to form triangles. Then, the drillholes located
in the vertices are used to estimate resources of associated triangular area. In this method, results
of the estimation depend on the triangulation method.

Polygonal method and triangulation becomes ineffective when dealing with deposit with
complex shapes. In such cases, cross-sectional methods can be considered as an alternative
technique. Cross-sections are generated along traverses and either extrapolated half distance
between drillholes or linked to form wireframes (Figure 2.16).

Figure 2.16 Cross-sectional estimation method (Abzalov, 2016)

Inverse distance weighted is a weighted estimation method where weights are inversely
proportional to a power of distance between sample and target location, di. The weights, i, are
calculated as follows:

20
 /���
�� = (2.11)
∑� /���

Although choice of power is arbitrary, inverse distance square i.e., power of two, is most
commonly used one.

Kriging is a linear geostatistical interpolation technique where the weights are derived from
variogram model (Krige 1λ51; Matheron 1λ63, 1λ68). It is considered as “best linear unbiased
estimator” since estimates obtained by the solution of kriging system are weighted linear
combinations of samples within a specified neighborhood. It is unbiased because mean of the
residuals is tried to be equal to zero and kriging weights adding to one. The distinguishing feature
making it “best” among other methods is minimization of the error variance.

Some of the variable definitions used in geostatistics are as follows:

Z(ui) – the data value of each sample

Z*(u) – the estimate at target location

i– kriging weights of each sample

– the Lagrange parameter

C – the covariance matrix between sample points

W – the weight matrix

D – the covariance matrix between sample locations and the target point

The kriging system can be written in matrix notation as:

C * W = D (2.12)

(2.13)

21
 Then, the solution of the kriging system gives the kriging weights as follows:

W = C-1 * D (2.14)

Kriging is a geostatistical method based on theory of regionalized variables and stationarity

is the central concept. Three main classes of kriging are linear univariate, multivariate and non-
linear kriging. Ordinary and simple kriging are the two linear univariate kriging techniques that
are readily applicable for data characterized by a non-skewed distribution. Co-kriging and its
variants are the multivariate kriging techniques that make integration of data into a single coherent
model possible (Journel and Huijbregts 1978; David 1988; Goovaerts 1997). Disjunctive kriging,
Uniform Conditioning and various varieties of indicator kriging are the non-linear kriging
techniques all of which rely on non-linear transformation of data. Multiple indicator kriging is a
particularly effective variant of indicator kriging techniques for estimating resources of a deposit
characterized by multiple populations with different spatial trends (Journel & Huijbregts, 1978).

Risk and uncertainty are becoming increasingly concerned in mining industry. Traditional
geostatistical tools those including several types of kriging lack the ability to address these issues.
Additionally, unavoidable smoothing effect associated with kriging estimates results in
misrepresentation of original variability. Conditional simulation, on the other hand, is a more
sophisticated stochastic simulation technique that provides several equally probable realizations
of the model through utilization of kriging and Monte Carlo sampling approach (Figure 2.17).
Therefore, it provides ability to quantify uncertainty. Realizations of conditional simulation honor
conditioning hard data in term of both statistical (i.e., reproduce the histogram) and geostatistical
characteristics (i.e., reproduces the variogram).

Figure 2.17 Comparison of interpolation and simulation estimates (Rossi & Deutsch, 2014)

22
 Conditional simulation is applicable for both continuous variables (e.g. grade, thickness
and deleterious components) and categorical variables (e.g. geological structures and lithofacies).
Although there are variety of conditional simulation techniques available, Sequential Gaussian
Simulation (SGS) and Sequential Indicator Simulation (SIS) are the two most commonly utilized
methods. Sequential Gaussian Simulation is a Gaussian-based technique that requires
transformation of the data into standard Gaussian space. The main limitation of this technique is
about its applicability on highly skewed data. Sequential Indicator Simulation, on the other hand,
overcomes this limitation and it depends on transformation of data into indicators considering
prespecified cut-off values.

Similar to kriging, exploratory data analysis is the first and essential step in conditional
simulation. It is followed by definition of random path along which grid nodes will be simulated.
Based on which technique employed, data is transformed into either gaussian or indicator variable.
Later, an estimated is generated by kriging and conditional cumulative distribution function is
constructed at chosen location. Then a simulation value is randomly drawn from conditional
cumulative distribution function and assigned to the simulated node. Once estimation is completed
for all the nodes in the simulation grid, the values are back-transformed from gaussian to original
space. Lastly, reproduction of histograms and variograms are verified for validation purposes.

2.10. Recoverable Resources

Accurate estimation of tonnage and grade in a resource model necessitates consideration

of several types of dilution. Geologic contact dilution, operational mining dilution and internal
dilution are the main sources of dilution.

Geologic contact dilution is also referred to as external dilution. It is the type of dilution
that arises when waste outside of the orebody extracted with the mining block. The degree of
external dilution is governed by the orebody geometry, blasting practices and mining selectivity.
Sub-cell method and direct calculation of proportions of each unit within each block are the two
alternative methods by which external dilution can be incorporated into the model (Rossi &
Deutsch, 2014).

23
 Operational mining dilution is often a planned dilution source that occurs at the time of
production and can simply be estimated through geometric calculations.

Internal dilution is the type of dilution resulting from resource estimation at a support
(volume) which is different than that of original data. Block size is a crucial decision in resource
estimation. It primarily depends on data spacing. Journel and Huijbregts (1978) suggest 1/3 to 1/2
of data spacing to be an appropriate size to utilize available resolution and prevent artificial
smoothing. Height of the blocks is typically a function of operational parameters and mining
method. However, estimation of recoverable resources often necessitates consideration of a
smaller support representing mining selectivity. It is known as smallest mining unit (SMU) and
depends on many factors including mining method, available data and equipment size.

It is a well-known fact that variability at larger support is less than that in of smaller support
(Figure 2.18). This is a concept commonly known as volume-variance relationship. Change of
support methods are utilized to achieve representative distribution for a specific mining selectivity.
Following equation is known as Krige’s additivity relationship which forms the basis of change of
support techniques.

,� = ,� + �, � (2.15)

where D2(v,G), D2(v,V), and D2(V,G) represent dispersion variance for different scenarios and v,
V, G stands for increasingly larger support.

Figure 2.18 Volume-variance relationship (Rossi & Deutsch, 2014)

24
 Depending on the nature of correction, they are classified as global or local change of
support methods. Global change of support methods is those allowing to infer grade-tonnage
relationship at a target support from sample data. On the other hand, local change of support
methods allows estimation of recoverable resources within each mining panel (Abzalov 2006,
2016).

Affine correction, indirect lognormal and discrete gaussian change of support are the global
change of support techniques. They achieve volume-variance correction through utilizing the link
between support effect and spatial continuity. Therefore, variogram is an effective tool in
assessment of change of support.

Affine correction is a global change of support technique that is limited to non-skewed

data. Permanence of distribution shape is the primary assumption of this methods. It allows
calculation of quantile for target distribution (q’) in terms of quantile of the input distribution (q),
mean of both distributions (m) and variance correction factor (f) as follows:

′
=√ ∗ − + (2.16)

Variance correction factor in this equation is also referred to as variance adjustment factor,
and it represents the ratio of variance at target support to total variance.

Indirect lognormal correction is an alternative global change of support technique that

overcomes limitation of affine correction. Transformation of the quantiles for a positively skewed
distribution is in the form of an exponential equation as follows:

′
= (2.17)

where the coefficients a and b are defined as:

√�� +
= [ ] (2.18)
√�∗�� +

l �∗�� +
=√ l �� +
(2.19)

One of the primary features of change of support methods is to leave mean of the
distribution unchanged. Therefore, indirect lognormal correction technique necessitates rescaling
of the transformed values so that the mean of distributions match. This final adjustment is defined
mathematically as:

25
 " ′
= ′ ∗ (2.20)

where m’ represents the mean of transformed values while m is the mean of original distribution.

Discrete gaussian is the most sophisticated global change of support technique for which
Hermite polynomial expansion is used to express the distribution of SMU grades. It is also
applicable to skewed data and involves correction through transformation of distributions for
different supports to Gaussian units.

Uniform conditioning and localized uniform conditioning are the local change of support
methods that involve estimation of recoverable resources for each mining panel (Abzalov, 2016).
Uniform conditioning is a non-linear geostatistical technique by which tonnage and grade for SMU
size blocks constituting larger panels are calculated. However, actual locations of these small
blocks are not specified with this method. On the other hand, localized uniform conditioning
(LUC) was proposed by Abzalov (2006) in order to overcome this limitation of uniform
conditioning technique. LUC is a robust method allowing accurate production of grade-tonnage
functions. It involves partitioning of the panels into SMU size blocks and then ranking them in
increasing order of grade (Abzalov, 2006).

2.11. Model Validation

Validation of resource model is needed to ensure internal consistency as well as global and
conditional unbiasedness. It can be conducted using various statistical and graphical tools.
Comparison of histogram and summary statistics for estimates and input data of each domain
allows to verify global unbiasedness. Similarly, histogram of residual errors and scatterplot of
estimate versus true value are useful to check for validity of global estimates. On the other hand,
conditional unbiasedness condition is assessed through validation of local estimates for which
density of the data distribution is a major factor. Spider diagram is a useful tool to validate the
local mean. It enables comparison of average grades of both estimate and input data along different
directions. It is also particularly effective to check presence of systematic bias which will be
represented as consistent over or underestimation in the spider diagrams. Contour map of residuals
can also be utilized for similar purposes.

Graphical validation techniques rely on visual comparison of estimates with input data on
graphical plots of sections and plans. Comparison of kriged blocks with drillhole intersections is a

26
simple but powerful tool to assess the degree of smoothing. Excessive smoothing is a serious issue
which can also be diagnosed from grade-tonnage curves.

Cross validation is an alternative technique used in resource model validation as well as

comparison of different variogram models and search strategies. It depends on comparison of the
estimates obtained at successively removed sample locations with true values. Therefore, it allows
assessment of estimation errors only where information is available. Through utilization of
aforementioned statistical and graphical tools, cross-validation can be used as a both qualitative
and quantitative validation instrument. In most practical applications, misclassification might
result in much more profound consequences than inaccurate estimation. In those cases, cross
validation techniques can be utilized in examination of goals of the study as a goal-oriented tool.

Figure 2.19 Misclassification due to information effect (Vann et al., 2003)

There is always a risk of misclassification because ore/waste classification is done based

on estimates rather than true values. Therefore, minimization of misclassification might be a much
more relevant criterion to judge the goodness of estimates that various other statistical criteria
(Isaaks & Srivastava, 1989). The points in quadrant I and III in Figure 2.19 are misclassified. On

27
the other hand, those in the other two quadrants are correctly classified despite the estimates are
not the same as the true values.

Validation of grade estimates is the primary target with the previous validation procedures.
On the other hand, there are some other procedures which deals with accuracy of tonnage
estimates. Validity of geological interpretations is the main factor in accuracy of tonnage estimates.
Therefore, auditing and confirmation of validity of interpretations as wells as subsequent
extrapolation procedures are essential.

2.12. Resource Classification

Resource models are one of the primary assets of mining companies. In most mining
jurisdictions, publicly listed mining companies are required to comply with certain principles in
their public disclosures. Transparency, materiality and competence are the fundamental principles
that are common almost for all reporting codes that are related with disclosure of mineral resources.
Specific need of different mining jurisdictions resulted in development of a variety of reporting
systems (Rossi & Deutsch, 2014). Among those, JORC Code (Australia), NI 43-101 (Canada),
SAMREC (South Africa) and SEC Industry Guide 7 (United States) are the most widely used ones.

In all of the jurisdictions mentioned above, resources are classified as inferred, indicated
and measured in order of increasing level of geological knowledge and confidence (Figure 2.20).

Figure 2.20 Relationship between exploration results, mineral resources and ore reserves (JORC
Code, 2012)

28
 Reporting codes provide the definition of mineral resource and distinct resource categories.
They also attach non-prescriptive guidelines. Inferred category represents the lowest level of
confidence among three and is not allowed to be converted to mineral reserves. On the other hand,
indicated and measured resources have higher level of confidence sufficient to be converted into
reserves through application of “the Modifying Factors”. Modifying factors include a set of
technical, economic, legal and social factors to be considered for conversion of resources to
reserves.

There exists a variety of resource classification techniques that are based on different
criteria. Traditional geometric methods of resource classification use distances to drill holes,
number of samples used for estimation, multi-pass kriging estimation and reasonable combinations
of these (Rossi &Deutsch, 2014). An alternative resource classification technique is use of
methods based on kriging variance. However, limitations of all these methods motivated
development and application of conditional simulation in resource classification.

The technique which uses distance to drill holes has two varieties. The resource
classification is based on either distance from nearest sample or average distance of all samples
used in interpolation. It is also a common practice to consider the number of samples within
defined search neighborhood as a criterion. Multi-pass kriging technique relies estimation of
resources at several kriging iterations with different levels of restrictions generally defined in terms
of different search neighborhood and minimum number of samples (Rossi & Deutsch, 2014). Use
of kriging variance is a different approach to construct confidence intervals. In this method, error
arising from estimation is the criteria used in classification. However, its application is limited
with normally distributed data. Additionally, misleading results may result in the presence of
proportional effect.

Conditional simulation is becoming increasingly popular as a resource classification tool.

It quantifies the grade confidence considering data configuration and continuity (Glacken &
Snowden, 2001). Therefore, conditional simulation allows systematic and quantitative definition
of principle aspects considered in resource classification; volume, measure of uncertainty and
confidence represented as the probability to be within limits defined by uncertainty (Figure 2.21).

29
Figure 2.21 Illustration of three aspects considered in resource classification (Rossi & Deutsch,
2014)

Volume is the first aspect considered in resource classification and it represents a certain
production amount corresponding to a time increment such as annual or quarterly production rate.
Measure of “+/-“uncertainty and confidence interval, on the other hand, are used to express
variability of grade and probability to be within particular interval.

Following is a summary of the quantitative guidelines widely accepted in industry for resource
classification (Parker & Dohm, 2014):

Inferred: mineral resource category characterized by presence on insufficient geological

information to establish confidence levels

Indicated: mineral resource that is predicted to be ± 15% of the predicted grade 90% of the time
over an annual production increment

Measured: mineral resource that is predicted to be ± 15% of the predicted grade 90% of the time
over monthly or quarterly production increment

30
 CHAPTER 3

GEOLOGY OF THE DEPOSIT

The volcanogenic massive sulfide type polymetallic mineralization of the case study is
located in western Turkey. A Turkish mining company and an international mining company have
50/50 ownership of the property through their joint venture company. The deposit has been
discovered as a result of geochemical stream sampling program that has been completed in 2011.
In June 2011, the licenses have been transferred to the JV company which is the current operator
for the project. This chapter mostly relies on information provided in geology section of the
prefeasibility report published in 2016.

3.1. Regional Geology

Amalgamation of several continental fragments which were once separated by narrow

oceanic seaways of Tethys ocean resulted in a single landmass, Anatolia. Pontides, Anatolides-
Taurides and Arabian Platform are the three main tectonic units representing either Laurassian or
Gondwana affinities that comprise the geology of Turkey (Ketin, 1966).

The Pontides exhibit Laurassian affinity, and are located north of Izmir-Ankara-Erzincan
suture zone that formed as a result of complete closure of northern branch of Neo-Tethys ocean.
The Anatolides-Taurides show Gondwana affinities, and were detached from Gondwana by
southern branch of Neo-Tethys. The Arabian platform is the other major tectonic unit, and is in
contact with the Anatolides-Taurides along the Assyrian suture.

The study area is located within Afyon Zone. It is ascribed as a belt by Okay (1985) that is
located between Menderes Massif and the Tavşanlı Zone (Figure 3.1). On the other hand,
Goncuoglu (2007) describes the Afyon Zone as a part of Kutahya-Bolkardag Belt which represents
the north-facing passive margin of the Tauride-Anotolide Platform.

Relatively northern units that are located immediately south of the Neo-Tethys ocean
constitutes a distinct high-pressure/low-temperature metamorphic belt called as Tavşanlı Zone. On
the other hand, Afyon Zone comprises southerly located low-grade medium to high pressure
metamorphic units of Kutahya-Bolkardag Belt.

31
 Figure 3.1 Tectonic Map of Turkey (van Hinsbergen, et al., 2016)

3.2. Local Geology

Local geology of the study area comprises a variety of rock types that formed as a result of
a series of major tectonic events closely related with the evolution of the Tethys ocean. Figure 3.2
presents the geological map of the study area and its near vicinity together with the license
boundary indicated close to the center of the map.

Paleozoic metamorphics consisting of gneiss, schist, mica schist, amphibolite, marble,

phyllite and quartzite are tectono-stratigraphically the lower most unit in the study area (Figure
3.3). This unit is tectonically overlain by Triassic carbonates and Upper Cretaceous ophiolitic
mélange which is characterized by presence of olistostromal blocks and ophiolite sections.

Oligocene-Lower Miocene granitic intrusions are comprised of units that are products of
extensional tectonism and associated magmatic activities. It consists of granite porphyries and
aplitic dikes, and represents cross-cutting relationship with the units belonging to Paleozoic
metamorphics, Triassic carbonates and Upper Cretaceous ophiolitic mélange.

32
 Figure 3.2 Geological Map of the Study Area

Such a magmatic activity resulted in abundant skarn and hornfels formation along the
contact of Paleozoic metamorphic rocks and carbonaceous units belonging to Upper Cretaceous
ophiolitic mélange, respectively.

Figure 3.3 Generalized Columnar Section

All these units are stratigraphically overlain by Lower Miocene felsic to intermediate
volcanics represented by andesite and dacite complex intrusions, dykes, domes, lava flows and

33
volcanogenic sedimentary rocks. Among those pyroclastics are the most abundant volcanic rocks
particularly represented by ignimbrites. Pliocene terrestrial sediments and Quaternary alluvial
deposits sourced from the older lithologies unconformably overly all the units mentioned earlier.

3.3. Deposit Geology

The deposit is a polymetallic volcanogenic massive sulfide deposit (VMS) hosted by

Paleozoic metamorphic units. Among those quartz schist, chlorite-sericite schist and quartz-
feldspar schist are the most distinct lithologies outcropping around the region (Figure 3.4). Lower-
Middle Miocene volcanics are the second most abundant unit and particularly observable towards
southwest of the deposit.

Figure 3.4 SE-NW geological cross section

Quartz schist is beige-grey to beige-pale green colored unit which is characterized by

abundance of quartz porphyroblasts. This unit represents stratigraphically the lower most unit in
the study area.

Chlorite-sericite schist is the primary host lithology for ore and is consisting of minerals
including quartz, calcite, chlorite, muscovite and sericite. It is green to dark green colored and
presents a well-developed schistosity. Pyrite-chlorite-sericite schist is used by the company to
describe the parts of chlorite-sericite schist with a pyrite abundance greater than 15-20% by
volume.

34
 Quartz-feldspar schist is stratigraphically the uppermost unit. It is the unmineralized, beige-
pale green colored hangingwall lithology that is characterized by the presence of feldspar and
quartz porphyroblasts.

3.4. Mineralization

The polymetallic massive sulfide mineralization is hosted by metamorphic units exhibiting

evidence for greenschist facies metamorphic conditions. Among those units, chlorite-sericite schist
is the primary host lithology.

Exposed and near surface parts of the deposit is weathered and oxidized as a result of
interaction with meteoric water. In this oxide zone, meteoric water leached copper and zinc. The
intensely oxidized leached out parts of the sulfide body are named as gossan zone, associated with
elevated gold and silver grades especially along the sulfide-oxide contact.

The sulfide zone represents other the major zone of the deposit which is characterized by
two primary ore types namely massive pyrite and massive pyrite magnetite. These two names are
used to describe those parts of the deposit where mineralization completely overprints the host
lithology. Primary ore minerals observed in the sulfide zone includes sphalerite and chalcopyrite
while pyrite, tetrahedrite, tenantite, galena and magnetite constitute the gangue minerals.

Enriched zone and transition zone are the two further subdivisions of sulfide zone.
Enriched zone is the thin chalcocite enriched parts of the deposit which is particularly associated
with elevated copper grades. On the other hand, transition sulfide is a term used to describe parts
of the deposit indicating a transition from unmineralized country rock to high-grade primary
sulfide mineralization body.

The deposit is massive sulfide type polymetallic deposit exhibiting Cu + Au + Zn + Ag

mineralization. As suggested by the deposit type and field evidences, the deposit formed as a result
of volcanic-associated hydrothermal events. The attitude of the deposit is roughly NNE-SSW with
a dip of 20-35o NW. Subsequent deformational events are evident from abrupt breaks in the
mineralization especially along the strike direction. It is also inferred from the changes observed
in the attitude of the deposit. However, potential post mineralization features could not be well
studied and documented due to extensive cover.

35
 CHAPTER 4

STATISTICAL CHARACTERIZATION & ESTIMATION DOMAIN DEFINITION

The data to be used in this study are sourced from an international mining company. They
include collar, lithology, assay and survey files provided in .xslx format, a digitized surface
geological map, specific gravity readings, digitized topographical contours with 1 m contour
intervals, license boundaries, detailed documentation and results of geotechnical, metallurgical,
processing tests and geophysical surveys conducted for better understanding of the features of the
mineralization.

The drillhole database of the case study comprised 517 holes (333 diamond core and 184
reverse circulation) completed in four phases between 2013 and 2017, with depths ranging
between 20 and 376 m (average: 113.7 m). As presented in Figure 4.1, the holes were drilled in a
regular grid with an approximate grid resolution of 25 m. Considering the attitude of the deposit,
the drill grid is adjusted to NE-SW trend based on the attitude of the deposit.

Figure 4.1 Drillhole Locations

36
 The JV company utilized various exploration tools for better understanding and
characterization of the deposit in the last 7 years. Some of those are listed below:

 Geochemical studies:
 Stream sediment sampling (20 samples)
 Soil sampling @100m regular grid (786 samples)
 Rock chip sampling (353 samples)
 Geophysical studies:
 Ground based magnetic survey (32 N-S sections with a total 112.2km
line length)
 Induced polarization survey (22 NW-SE sections with a total 41.6 km
line length)
 Drilling:
 Phase 1 (11 core holes with a total length of 1,528.5m)
 Phase 2 (143 core holes + 81 RC holes with a total length of 17,114.8m
and 6,790.0m, respectively)
 Phase 3 (149 core holes +103 RC holes with a total length of 26,061.1m
and 6,042.0m, respectively
 Phase 4 (30 core holes with a total length of 1241.1m)

Two outlying drillholes located in the far north-east of the map do not geographically
belong to the estimation domain. Therefore, those peripheral drillholes are not considered in the
subsequent stages of this research. The vertical and sub-vertical drillholes (dip: -60 to -90 degrees)
comprising the drillhole database are located between 636,600-638,000E and 4,357,500-
4,359,000N.

4.1. Quality Assurance and Quality Control

Quality control and quality assurance (QA/QC) is an integral part of the exploration
program. Such information is regularly collected and analyzed to ensure the compliance of all the
stages of data collection, handling, analysis and documentation with industry standards. A
comprehensive chapter documenting the analysis of the QA/QC practices utilized by the operator
for the project that is included in the prefeasibility study is provided together with the database.
Therefore, this chapter is intended as a summary of that section.

37
 Standards, duplicates and blanks are the principal components of the quality assurance and
quality control system followed by the company.

There are three standards (certified reference material) used in the project with an intention
to assess the accuracy of assaying. G907-4 and G910-8 are the two standards for gold that are
sourced from Geostats Pty Ltd which provides confirmation at 3.84 g/t and 0.63 g/t, respectively.
On the other hand, GBM398-1 is the third standard with provided certified values of 0.183g/t Au,
1.482% Cu, 2.03% Zn, 2.667% Pb, and 5.1g/t Ag. This standard is also source from the same
company to be used for base metal intervals. There is a total of 1,571 gold standards and 244 base
metal standards inserted for 31,495 assays in the database covering the first three phases of drilling
which equated to 1 out of every 20 samples for gold and 1 out of every 130 samples for base
metals. Qualified person reported that only about 2% of the 1,572 standards samples differ more
that 10% from the original value which is an issue particularly associated with gold value of the
standard used for base metals. Therefore, qualified person has not suggested a potential for material
impact on the mineral resource estimates.

Insertion of blanks samples as the last sample to the end of every single drillhole is another
common QA/QC procedure. It done with an intention to control contamination. QP reported that
there is a total of 1,134 blank samples inserted into the whole sample stream which makes an
average blank insertion rate of roughly 1 out of 28 samples. Assay results of blank samples are
compared against pre-determined cut-off representing trace level and ore value. Out of 1,134
blanks inserted just one sample reported to be over economic gold grade and small percentage of
blanks are out of the tolerance (trace level).

Duplicates are the third form of QA/QC instruments, and are utilized to assess the precision
of the assay procedure. Split core and reverse circulation coarse rejects represent the two primary
form of duplicates used for this project. Based on the QA/QC report, there identified a total of
1,219 duplicates within the database with a duplicate inserted as every 25th sample in the sample
stream. Comparison of duplicate samples with the original samples on scatterplots confirmed the
precision of the procedure.

Round robin test is another form of check included as a part of QA/QC system for the
project. It involved analysis of 269 check samples for gold by a third party laboratory using the

38
same method of analysis. This practice is conducted as an interlaboratory test in order to assess
the reproducibility of the process.

Yet another analysis included in the QA/QC chapter of the prefeasibility study is the
comparison of diamond drilling and reverse circulation drill results. Nearest neighbor based
comparison of results indicated that the data gathered from RC samples is generally comparable
with that of diamond drilling to ensure statistical reliability despite being slightly conservative for
the oxide zone.

All in all, QA/QC analysis conducted by the QP indicated that the database for the project
is in acceptable quality in terms of data gathering, handling, analysis and documentation. Thus, it
is decided to be reliably used in the modelling and subsequent analysis.

4.2. Database Validation

Acknowledging the significance of quality of the data on reliability of the estimates, a

number of actions are taken to ensure validity and integrity of the database. Project drillhole
database is first checked for interval errors (missing and overlapping intervals). It is followed by
checks and corrections for negative and non-numeric grade values. There aren’t any negative grade
values detected during validation. Non-numeric grade values, on the other hand, are commonly of
two types: (i) typo errors resulted during data entry, and (ii) grade values reported to be outside of
the range defined by lower detection limit (LDL) and upper detection limit (UDL). There aren’t
any non-numeric value errors detected in the form of typo errors and grade values larger than UDL.
There are grade values reported to be lower than LDL, which necessitated a degree of data
manipulation. For those intervals, grade value is assigned as the half of the lower detection limit
which is a common industry practice.

Comparison of collar positions with respect to topography was also an essential component
of the database validation to ensure collars honor the surface topography. Validation of the collar
positions included both visual inspection of the drillhole collars with respect to topography as well
as quantitatively. The latter approach involved projection of drillhole collar onto surface
topography in order to generate a new variable that will store projected collar coordinates. Then,
that variable is used in the calculation of difference in the elevation of drillhole collars from the
topography as presented in Figure 4.2.

39
 Figure 4.2 Validation of drillholes collar positions

Both visual inspection and the graph above indicated some of the collar positions does not
honor the topographic surface. There are nine drillholes labeled above whose collar positions are
off by more than 2.5m from the topography. DRD-295 is the drillhole whose collar position
presented the most severe discrepancy which is - 9.5m. The database was then checked one more
time for a potential data entry and transfer error. The collars still did not honor the topographic
surface; thus, they are excluded from the estimation procedure.

4.3. Statistical Data Analysis

Preparation and statistical characterization of the data is the following step. Data
preparation involves regularization of raw data through compositing. The drill cores and RC chips
recovered from the deposit are sampled with an average length of 1.58 m and vast majority of
sampling conducted at 1 m, 1.5 m and 2 m intervals. Based on 25 m drill grid resolution, 10 m is
thought to be an appropriate horizontal block dimension to fully utilize the available resolution
and prevent artificial smoothing.

The deposit is a near surface polymetallic mineralization for which open pit mining method
at a bench height of 5 m is proposed and incorporated into the preliminary feasibility study.
Considering average sample length and proposed bench height, 5m decided to be a reasonable
composite size. Vertical and sub-vertical dip of the drillholes in the database makes fixed-length
downhole compositing preferable over optimum length compositing approach. The composite data

40
is compared with raw data in terms of summary statistics to assure the compositing approach does
not introduce bias (Table 4.1).

Table 4.1 Summary statistics for all the raw and composite data
Count Min/Max Mean Standard Deviation Coefficient of Variation
Variable Raw Composite Raw Composite Raw Composites Raw Composites Raw Composites
Au 0/150 0/61.44 0.245 0.169 1.447 0.962 5.9 5.69
32,561 10,305
Ag 0/5621 0/1039 9.09 6.4 54.66 28.17 6 4.4
Cu 0/16.4 0/12.25 0.201 0.138 0.613 0.452 3.06 3.28
32,723 10,459
Zn 0/32.2 0/15.88 0.343 0.238 1.203 0.805 3.5 3.38

As presented in above table mean of the composite data is not identical to raw data.
However, this does not necessarily mean compositing introduced a bias. It is rather an outcome of
unweighted calculation of average grade which particularly has a pronounced impact on the results
for the raw data due to non-regular sample lengths ranging from 0.4 m to 9.0 m.

Table 4.2 Summary statistics for raw and composite data in length-weighted scenario
Count Min/Max Mean Standard Deviation Coefficient of Variation
Variable Raw Composite Raw Composite Raw Composites Raw Composites Raw Composites
Au, g/t 0/150 0/61.44 0.17 0.17 1.35 0.97 7.88 5.66
32,561 10,305
Ag, g/t 0/5621 0/1039 6.5 6.5 41.6 27.6 6.4 4.3
Cu, % 0/16.4 0/12.25 0.14 0.14 0.51 0.45 3.61 3.22
32,723 10,459
Zn, % 0/32.2 0/15.88 0.24 0.24 0.99 0.81 4.08 3.36

Comparison of length weighted mean grades of raw and composited data for all the
variables resulted identical or almost identical values as shown in Table 4.2. Furthermore, standard
deviation and coefficient of variation values are reduced for all variables significantly which is a
natural outcome of reduced spread of the data.

Another objective of data analysis is assessment of data distribution. Based on the

histograms presented in Figure 4.3, grade values span couple orders of magnitude and strongly
positively skewed exhibited by a pronounced asymmetry.

41
 Figure 4.3 Histograms of composite data

Cumulative probability plots are also utilized in the assessment of the data distribution.
Similarly, non-linear shape of such plots indicates that the composite data for gold, silver, copper
and zinc grades represent neither normal nor log-normal distribution. On the other hand, multi-
modal shape of the histograms for log-transformed data together with presence of inflection points
on cumulative probability plots suggest presence of multiple data populations which we are already
aware based on geological information provided in chapter 3.

Summary statistics of the primary protolith rock types and generic names for the ore times
identified earlier are presented in Table 4.3. It clearly shows that ClyGos and Gos are the two units
that are characterized by elevated gold and silver grades. It is primarily a natural outcome of the
leaching and oxidation process took place in the near surface portions of the deposit resulting in
relative enrichment of comparatively immobile components (gold and silver) and depletion and
downward migration of mobile components (copper and zinc) within the gossan zone.

Massive pyrite (MPy) and Massive pyrite magnetite (MPM) are the two units which are
associated with the largest mean base metal grades in the deposit. As mentioned earlier in chapter
3, these are the units indicating the ore type encountered with the chlorite-sericite schist (ClSerSch)
where the host rock is completely overprinted by the mineralization. When the intercept belonging
to these two units examined in 3D, they look like the continuation of gossan zone below oxide-
sulfide boundary along the plane defining the attitude of the mineralization.

42
 Table 4.3 Summary statistics for primary protolith and ore types

ne

ch
ag
Zo

rS
ch

ll
lp
Sc
s

Se
M
rS

a
o

u
ch

er
Cl
py

Py
yG

Cl
rc
Se

-S
s

t
Go

Ov

QS

Ov
QF

Qz
En

Py
M

Tr
Cl

Cl
Count 4642 177 44 226 486 358 21 1160 950 1911 106 167 10305
Min 0.01 0.01 0.04 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01
Max 61.44 24.61 4.06 22.02 7.74 3.27 1.44 3.07 2.37 1.11 0.93 5.74 61.44
Au
Mean 0.07 1.75 1.12 1.42 0.84 0.52 0.10 0.06 0.02 0.01 0.03 0.29 0.17
Std Dev 0.94 3.56 0.79 2.63 0.78 0.65 0.32 0.20 0.11 0.03 0.10 0.59 0.98
CV 14.61 2.03 0.71 1.85 0.93 1.25 3.17 3.11 5.75 3.01 3.15 2.03 5.70
Count 4642 177 44 226 486 358 21 1160 950 1911 106 167 10305
Min 0.2 1.0 4.2 1.0 1.0 1.0 1.0 0.4 1.0 1.0 1.0 1.0 0.2
Max 804.0 1039.8 115.8 660.2 692.3 215.7 12.7 136.8 64.2 25.6 325.2 84.9 1039.8
Ag
Mean 2.9 57.9 44.6 40.4 31.6 20.6 2.2 3.1 1.4 1.1 6.1 11.6 6.5
Std Dev 17.8 129.3 30.4 73.4 41.1 27.6 3.1 10.0 3.2 0.7 31.9 17.5 28.7
CV 6.13 2.23 0.68 1.82 1.30 1.34 1.40 2.88 2.28 0.65 5.26 1.52 4.35
Count 4643 177 44 226 486 358 21 1160 951 1911 106 167 10459
Min 0 0 0.79 0 0.05 0.05 0 0 0 0 0 0 0
Max 3.55 1.20 11.77 1.95 12.25 1.96 0.23 1.89 1.64 0.74 2.43 2.69 12.25
Cu
Mean 0.04 0.10 3.40 0.15 1.01 0.73 0.02 0.16 0.01 0.01 0.12 0.37 0.14
Std Dev 0.13 0.18 2.28 0.23 1.15 0.36 0.05 0.20 0.08 0.03 0.31 0.34 0.46
CV 2.96 1.82 0.67 1.60 1.14 0.49 2.77 1.24 10.62 4.87 2.50 0.91 3.30
Count 4643 177 44 226 486 358 21 1160 951 1911 106 167 10459
Min 0 0 0.02 0 0.01 0.01 0 0 0 0 0 0 0
Max 4.99 1.83 7.94 1.39 15.88 11.94 0.92 6.09 1.96 1.09 2.69 7.22 15.88
Zn
Mean 0.11 0.10 2.60 0.11 1.90 1.39 0.06 0.18 0.02 0.02 0.37 0.67 0.24
Std Dev 0.31 0.23 2.03 0.16 2.04 1.90 0.20 0.50 0.09 0.06 0.54 1.10 0.81
CV 2.99 2.19 0.78 1.43 1.07 1.37 3.29 2.76 5.16 4.06 1.47 1.64 3.41

Enriched zone is a subdivision of sulfide zone which is described as the thin chalcocite
enriched parts of the deposit. As presented above, it is the unit with the highest average grades for
all the variables included in the summary statistics table.

It is clear based on all these observations and analysis that the database is composed of a
number of data populations which are needed to be identified to proceed with subsequent phases
of the study.

4.4. Delineation of Estimation Domains

Estimation domains represent geostatistical stationary zones characterized by statistically

homogeneous populations. Traditional method of estimation domain delineation is an iterative
process involving definition of based on simple statistics and geological knowledge. On the other
hand, alternative approach involves utilization of multivariate statistical techniques. In this study,
a two-step approach is followed in delineation of estimation domains for the Cu-Au-Zn-Ag
deposit.

Traditional method is the first step in the domaining practice. It mainly involved utilization
of available knowledge about the genesis of the deposit. Initially there are around a dozen geologic
attributes (protolith rock types and ore types) involved in the raw data base some of which are
filtered out as a result of 5m downhole compositing or found to be only present in the two drillholes

43
located far north-east (DRD-186 and DRD-188) that are decided to be excluded from the study.
Then, the genetically related attributes such as ClyGos-Gos and MPy-MPM are grouped in order
to identify primary estimation domains in oxide and sulfide zones, respectively. As an assessment
of the domaining practice, contact plots and statistical comparisons are conducted. Contact plots
were particularly helpful to characterize the behavior of grade across the boundary between two
units. In a case where two units are going to be combined into a single population, we expect not
to have a sharp change. On the other hand, comparison of coefficient of variation for each single
unit considered to be combined into a population and that of the population was the major statistical
consideration. Assessment of the performance of the domaining practice was based on a check
whether there is any improvement in the coefficient of variation value. The domaining approach
based on these three criteria (genetical association, boundary analysis and statistical comparison)
indicated that ClyGos-Gos and MPy-MPM can each be combined into a population which are later
decided to be representing the dominant components of the two domains within oxide and sulfide
zones, respectively.

Second step of estimation domain delineation involved utilization of geostatistical hierarchical

clustering algorithm that is a machine learning based process used to group samples into
statistically homogenous domains according to their spatial dependency and degree of similarly in
terms of grade and geological features like lithology, mineralization or alteration (Fouedjio, 2016).
With this method two statistically distinct sub units are identified for each of the chlorite-sericite
schist, pyrite-chlorite schist and transition sulfide units. These units later grouped into previously
identified major domains for oxide and sulfide zones. As a result of this two-step domaining
approach following six estimation domains (Figure 4.4) are identified for the deposit:

 Domain 1 is named as gossan zone (GOS). Gossan zone is composed of clay gossan (ClyGos)
and gossan (Gos) units that are exhibiting elevated gold and silver grades. It is also the unit
where mobile elements are leached out and transported towards the sulphide zone. Therefore,
gossan zone is associated with low copper and zinc grades.
 Domain 2 is named as non-gossan oxide (NGO. Non-gossan oxide consists of the intercepts
that are between the oxide-sulfide boundary and the topographic surface but not considered
within the gossan zone.

44
 Domain 3 is named as barren wall rock (BW). Barren wall rock represents the poorly
mineralized or non-mineralized parts of the deposit in the sulfide zone. It is mainly made up
of chlorite-sericite schist 1, quart-feldspar schist (hangingwall rock) and quartz-schist
(footwall rock).
 Domain 4 is named as high-grade sulfide (HGS). High-grade sulfide represents the primary
sulfide mineralization of the deposit, and it is composed of massive-pyrite, massive-pyrite
magnetite, pyrite-chlorite-sericite schist 2 and transition sulfide 2.
 Domain 5 is named as low-grade sulfide (LGS). Chlorite-sericite schist 2, pyrite-chlorite-
sericite schist 1 and transition sulfide 1 comprise the low-grade sulfide which forms a low-
grade mineralization envelope around high-grade sulfide zone.
 Domain 6 is named as enriched zone (ENR). Enriched zone is characterized by presence of
elevated grades for all gold, copper, zinc and silver which is initially considered as a part of
high-grade sulfide. However, this zone is later modelled as a separate domain due to distinct
response of the material during metallurgical tests. Enriched zone is the domain that consists
only of material represented by chalcocite enriched thin intercepts which are described as
enriched zone during geological logging.

Figure 4.4 Estimation domains in a SE-NW cross-section

45
 Box-plot diagrams presented in Figure 4.5 clearly indicates statistical dissimilarities
between estimation domains. In addition to relative composition of different elements, it also
useful to make comparisons in terms of relative variability and distribution of the grades.

As stated earlier, the gossan zone (GOS) and the enriched zone (ENR) are the two
estimation domains that are associated with the largest mean precious metal concentrations.
However, these two units are distinct from each other with respect to the variability of the grades
in that the box representing the range between 25th and 75th percentiles of the data considered
within the gossan zone is longer, thus the gossan zone is more variable compared to enriched zone.
One another observation is regarding data distribution. Median and mean grades (indicated by blue
and red lines, respectively) are close to each other for the enriched zone. On the other hand, mean
grade is much larger that the median value for gossan zone which suggests a skewed distribution.
It is also evident from the figure that high-grade sulfide is also an important domain in terms of
these components.

Similar comparisons can also be done for copper and zinc grades from the lower two
diagrams of Figure 4.5. In terms of these two variables enriched zone and high-grade sulfide are
the two most significant estimation domains. Low-grade sulfide is exhibiting a strongly
asymmetric distribution and has the third largest mean copper and zinc grade.

The box-plot diagrams which are utilized here to make comparisons between geostatistical
estimation domains in terms of statistical properties are provided in log-transformed format with
an intention to enhance the presentation and facilitate relative comparison. Therefore, statistical
summary table (Table 4.4) would be more convenient for absolute numeric comparisons.

Statistical analysis of estimation domains delineated for the deposit indicated that the two-
step domaining approach yielded satisfactory results suggested by significantly improved
coefficient of variation values particularly for gossan zone, high-grade sulfide, enriched zone and
low-grade sulfide. Besides, there is still some room for improvement through introducing high-
grade cut-offs to deal with outlier values suspected both from box-plots and statistical summary
table.

46
Figure 4.5 Box-plot diagrams for geostatistical estimation domains

47
 Table 4.4 Summary statistics for geostatistical estimation domains

High-grade cut-off is also known as top-cut is essential in order to prevent risk of smearing of
the high-grades which might result in overestimation. Following are some of the widely utilized
alternative approaches in determination of high-grade cut-off values:

 Sum of the data mean and twice the standard deviation

 Four times the data mean value
 The point where the ragged tail starts on the histogram

Among these three alternatives, last approach found to be the most appropriate option for this
particular case that will eliminate the long tail of high-grade value representing the outlier as
illustrated below (Figure 4.6).

48
 Figure 4.6 Illustration of top-cut determination from histogram

High-grade cut-off values identified in this manner for all variables in each domain are
presented in Table 4.5. It also includes a column where the number of samples cut for each scenario
is indicated. Based on that the capped assay values only represent a small percentage (0.64%) of
all composites.

Table 4.5 High-grade cut-off values

Domain Variable Mean (µ) Standard Deviation (σ) µ+2σ 4µ From histogram Top-cut # Samples Cut
Cu 0.1 0.17 0.44 0.40 0.9 0.9 4
Zn 0.09 0.15 0.39 0.36 0.9 0.9 1
1
Au 1.29 2.69 6.67 5.16 11.6 11.6 5
Ag 44.8 102.7 250.20
0 179.20 0 350 350 6
Cu 0.07 0.19 0.45 0.28 1.6 1.6 2
Zn 0.1 0.35 0.80 0.40 2.65 2.65 4
2
Au 0.31 1.43 3.17 1.24 10.75 10.75 3
Ag 8.7 32.5 73.70
0.00 34.80
0.00 236 236 3
Cu 0.03 0.12 0.27 0.12 1.7 1.7 4
Zn 0.08 0.39 0.86 0.32 6.3 6.3 2
3
Au 0.04 0.24 0.52 0.16 4.7 4.7 3
Ag 1.5 12.9 27.30
0.00 6.00
0.00 135 135 2
Cu 0.88 0.83 2.54 3.52 6.5 6.5 2
Zn 1.65 2.03 5.71 6.60 11.75 11.75 1
4
Au 0.67 0.69 2.05 2.68 3.7 3.7 4
Ag 26 37 100.00
0.00 104.00
0.00 181 181 1
Cu 0.18 0.39 0.96 0.72 4.6 4.6 1
Zn 0.3 0.7 1.70 1.20 4.3 4.3 11
5
Au 0.14 1.31 2.76 0.56 9.15 9.15 1
Ag 4.4 12.3 29.00 17.60 110 110 3
Cu 3.26 2.17 7.60 13.04 7.65 7.65 1
Zn 2.33 1.84 6.01 9.32 6.05 6.05 1
6
Au 1.02 0.83 2.68 4.08 3.35 3.35 1
Ag 50.4 94.9 240.20 201.60 150 150 1

49
 Contact characterization or boundary analysis is another consideration which is aimed to
describe the grade trends and behavior near domain boundaries. Contact plots were particularly
effective in this analysis. The boundary between oxide and sulfide zones is one of the major
contacts that is examined within this part of the study. Figure 4.7 below shows the behavior of the
grades for gold, silver, copper and zinc near the oxide-sulfide boundary.

Figure 4.7 Contact plots for oxide-sulfide boundary

There are a number of inferences made from these plots: (i) grades for both precious and
base metals represent distinct behavior in oxide and sulfide zones, (ii) the change of the grades
across the boundary is sharp, (iii) near the boundary precious metal grades are noticeably higher
in the oxide zone which is already explained by relative enrichment of gossan zone as a result of
leaching and oxidation process, (iv) opposite case is valid for the base metal grades i.e., copper
and zinc grades are considerably higher in the sulfide zone especially near the oxide-sulfide
boundary, and (v) gold and silver grades are typically increasing within the oxide zone towards

50
the contact. As a result of all these observations, the oxide-sulfide boundary is will be treated as a
hard boundary during grade estimation process.

The boundaries between all other populations were analyzed in this manner will be treated
as hard boundaries in order to prevent smearing of grades across estimation domains (Appendix
A).

51
 CHAPTER 5

ESTIMATION DOMAIN MODELLING

Domain modeling is a fundamental component of contemporary mineral resource

estimation. A geological model is a simplified three-dimensional representation of the physical
features. For a mineral deposit those features might consist of lithology, alteration, mineralization,
oxidation state and structural elements. Geological models are constructed with an intention to
delineate the primary controls on the mineralization. On the other hand, a domain model is a more
comprehensive construction which calls for both geological and statistical input. Furthermore,
domain models use the knowledge about the mineralization controls, and are utilized as a key
guiding tool during estimation of a property.

Stationarity is the key characteristic of robust estimation domains. In the simplest terms, it
represents statistically homogenous data populations. Delineation of stationary estimation domains
is an iterative process that involves consideration of many aspects. There are six estimation
domains identified for the deposit through a two-step domaining approach whose details are
presented earlier in chapter 4. Following stage of the study involves construction of 3D estimation
domain model.

The tools and approaches employed for mineral resource estimation have been evolved
noticeably in the last couple decades. There exist a variety of 3-D domain modelling alternatives
exhibiting differences in terms of complexity, time and effort requirement. Considering its
implications for downstream mining practices, quality of the model is crucial. In this perspective
choice of the modelling approach is a significant consideration.

This part of the study involves construction of alternative domain models through
execution of four alternative modelling approach which are (i) explicit modelling approach, (ii)
implicit modelling approach, (iii) indicator kriging approach, and (v) conditional simulation
approach. It compasses the details for the assumptions involved, the main steps followed as well
as results obtained for all these alternatives.

52
5.1. Explicit Modelling Approach

Explicit modelling is the modelling approach utilized in construction of solid models

representing statistically homogeneous estimation domains. This method represents the traditional
technique for construction of 3D models which involves manual digitization of interpreted
geologic features on cross-sections. Despite being a time-consuming approach, explicit modelling
provides complete user control and facilitate better understanding of the mineralization controls as
well as major features of the deposit.

There are 517 drillholes constituting the drillhole database that are drilled at a pseudo-
regular pattern which is oriented along NE-SW where grid resolution is roughly 25 meters.
Analysis of mineralized intercepts also revealed that this trend is consistent with the strike of the
mineralization. Therefore, orientation of cross-sections is chosen as NW-SE, roughly parallel to
the dip direction. Figure 5.1 illustrates digitized contacts for high-grade sulfide (HGS) zone for
one of the cross-sections along the chosen orientation.

Figure 5.1 An illustration of digitized contacts for HGS zone along a SE-NW cross-section

Digitized contacts along NW-SE cross-sections belonging to the estimation domains

largely defined the geometry. A further improvement for the explicit model has been achieved
through considering the orthogonal direction as the orientation for the long-sections (NE-SW). The
mineralization geometry is delineated from 53 cross-sections and 26 long-sections as shown in
Figure 5.2A.

53
 In the next step of estimation domain modelling, manually digitized cross-sections are
linked to form three-dimensional triangulations, also known as wireframes, with the help of tie
lines. Figure 5.2B is an illustration of the solid model belonging to high-grade sulfide zone
constructed with this way.

Figure 5.2 Digitized contacts along orthogonal directions (left) and solid model of HGS zone
generated by explicit modelling approach (right)

One of the major assumptions made during generation of 3D models by explicit modelling
approach is extrapolation distance. Common practice within the industry is to use nearly half of
the distance between adjacent drillholes. For this particular study, the same strategy has been
followed. Another substantial consideration is the final adjustment of the wireframes based on
cross-cutting relationships defining the chronology. In this respect, construction of the resource
domain model for the deposit through explicit modelling approach necessitated exhaustive review
of all available geological information and facilitated better comprehension of the deposit in terms
of attitude and continuity of the mineralization.

5.2. Implicit Modelling Approach

Implicit modelling technique is a more recent approach compared to explicit modelling. It

relies software driven algorithms (e.g. radial basis function) for generation of surfaces

54
corresponding to boundaries between different data populations. In this study, Leapfrog Geo is
used to generate the implicit model of the estimation domains comprising the deposit.

The estimation domain codes assigned as a new attribute for the drillholes considered for
this research is one of main inputs. Resolution of the model is a critical aspect which determines
the minimum length of input data that will be considered within implicit modelling practice.
Therefore, it is a crucial decision calling for understanding end use of the product. The deposit is
planned to be exploited with open pit mining using 5m benches. Considering the bench height and
average sample length of the raw data, the compositing length is decided to be 5 m. Thus,
resolution of the model is set as 5 m. Other key information required for generation of preliminary
model with the implicit modelling approach is age relationships because it directly controls 3D
extends of each individual unit. The oxide-sulfide boundary modelled using points representing
oxide bottom in each drillhole is used as the limit of two primary zones (i.e. oxide zone and sulfide
zone) of the deposit. The topographic surface and oxide-surface boundary defines the limits of
oxide zone where gossan zone is modelled as the younger unit. On the other hand, sulfide zone
occupies the space between oxide-sulfide boundary and 1,015 m elevation which stands for bottom
limit of the model in vertical direction. Enriched zone, high-grade sulfide, low-grade sulfide and
barren wall rock are the four units comprising sulfide zone for which units are given in younger to
older. It is clear that the relationship defined for some of these units does not necessarily indicative
of the timing of formation. For instance, high-grade sulfide and low-grade sulfide are the two units
that are probably formed concurrently, thus, suggested relationship is used merely for geometric
purposes.

The results obtained for the preliminary model uses the defaults of the implicit modelling
where use of isotropic search strategy generated solids that does not reflect the spatial
characteristics of the deposit. In this respect, the knowledge gathered regarding attitude of the
mineralization in terms of dip, dip azimuth and pitch was a valuable contribution guiding implicit
modelling. Final editing of the solids following introduction of trend is the last and an important
step to ensure the model is clean of unrealistic isolated bodies or other artifacts.

55
Figure 5.3 Solid model of HGS zone generated by explicit modelling (left) and implicit
modelling (right) approaches

The solid model generated for high-grade sulfide zone in this manner is presented in Figure
5.3B while the product of explicit modelling approach is shown in Figure 5.3A. Visual comparison
of outcomes suggests that both methods resulted in globally consistent solid models exhibiting
differences in terms of details.

5.3. Indicator Kriging Approach

Application of indicator kriging provides an alternative approach for modelling categorical

variables. The codes assigned as a new variable for each estimation domain are used in generation
of a new set of codes indicating whether or not an intercept belongs to a particular estimation
domain. Therefore, the codes generated in this way are in binary format and are called as indicator
values or variables. They are essential components of indicator kriging technique where these
binary values are used in quantification of spatial continuity using indicator variograms. Following
is the indicator variogram constructed to assess the spatial continuity of high-grade sulfide, low-
grade sulfide, gossan and enriched zones (Figure 5.4).

The results of the variography study for the indicator variables belonging to these zones
are also presented along with the indicator variograms in the form of model parameters. It is

56
noteworthy that this approach was effective to capture the major attitude of the mineralization.
Therefore, indicator variogram results are considered as satisfactory for subsequent analysis.

Figure 5.4 Indicator variogram constructed for HGS+LGS+ENR+GOS zones

Next step involved in this approach for modelling estimation domains delineated for the
deposit is execution of indicator kriging routine using information derived from indicator
variograms as a guide for estimation. As a result of this process, 10m x 10m x 5m blocks of the
block model are assigned to represent an element of an estimation domain. Finally, shells
corresponding to each estimation domain are generated as separate 3D wireframes.

Figure 5.5 Solid Models for HGS zone generated with explicit modelling, implicit modelling and
indicator kriging approaches

Figure 5.5 illustrates the products of explicit modelling, implicit modelling and indicator
kriging approaches to generate solid models of high-grade sulfide zone. Side by side comparison

57
of the outcomes is an easy way to reveal discrepancies between solid models. It is clear that
indicator kriging approach was also successful to capture the major features of high-grade sulfide
zone as suggested by previous two models but there exist significant dissimilarities particularly in
terms of spatial continuity of the domain.

5.4. Conditional Simulation Approach

Conditional simulation technique constitutes the fourth approach employed in estimation

domain modelling for the deposit. Sequential indicator simulation is the specific conditional
simulation technique which enables generation of mathematically equiprobable outcomes,
commonly referred as realization, for categorical variables. Therefore, this technique is utilized in
generation of a set of realizations for the deposit in which the indicator variogram model obtained
for modelling with indicator kriging approach is used. There are ten realizations with this
approach. Figure 5.7 presents the solid models of oxide zone modelled through four alternative
approaches. It provides the outcomes only for gossan and non-gossan oxide in plan view in order
to facilitate visualization and comparison.

Similar to other modelling approach, sequential indicator simulation method was also
effective to capture the main trend of the mineralization. Furthermore, it resulted in comparable
volumes for estimation domains with that of indicator kriging approach which are considerably
larger than volumes generated by both explicit and implicit modelling approaches. Figure 5.6
below presents the volumes for HGS domain for alternative modelling approaches.

Figure 5.6 Bar chart of the volumes for HGS solids generated with alternative approaches

58
Figure 5.7 Map view of solid models for oxide zone generated by four alternative approaches

59
 CHAPTER 6

RESOURCE ESTIMATION & CLASSIFICATION

The mineral resource estimates for the deposit is developed from a 3D block model
generated using a commercial mine planning package (MineSight3D). The estimates are based on
part of the database which is found to be reliable to be used in this study. As stated earlier, primary
objective of this research is to execute a comparative analysis of 3D domain modeling alternatives.
It further aims to document implications of each modelling routine on mineral resource estimates.
Therefore, this part of the study involves estimation of resources based on identical estimation
parameters, method and assumptions. This is why, this chapter is dedicated to present the primary
aspects regarding grade estimation, model validation and resource classification applied for
estimation domains generated through explicit modelling approach.

6.1. Grade Estimation

The polymetallic massive sulfide deposit is associated with anomalous concentrations of

copper, gold, zinc and silver which are considered in the estimation of resources. On the other
hand, metallurgical test work results do not indicate presence of any deleterious components that
may materially affect the concentrate quality. Moreover, economic contribution of lead grades is
negligible (mean grade of non-composited data:0.08%), thus excluded in the subsequent analysis.
That being the case, grade estimation is focused on procurement of estimates for aforementioned
precious and base metals concentrations.

6.1.1. Block Model Setup

The block model setup is a term used to describe the extends and orientation of the model
as well as block size and model fields. Considering the strike of the deposit and orientation of the
drillholes, the block model is rotated 45o clockwise as shown in Figure 6.1 where model limits are
indicated as callouts at four corner points and coordinates are provided for Universal Transverse
Mercator projection system European Zone 35.

60
 Figure 6.1 Block model extends

Block size is a crucial decision in resource estimation which is a function of data spacing.
Journel and Huijbregts (1978) suggest 1/3 to 1/2 of data spacing to be an appropriate size for
optimum utilization of the available resolution. Considering 25m average drillhole grid spacing,
10m is decided as the horizontal block dimensions. Moreover, vertical block dimension is chosen
based on proposed 5m bench height. Therefore, the block model comprised of 10m x 10m x 5m
sized regular blocks.

Model fields are the last component of the model setup. They represent the variables used
to store information for each individual block within the model. Table 6.1 is the complete list of
model fields and their descriptions.

Table 6.1 Block model fields

Model Field Description
East, North, Elevation Block center coordinates
DOM Estimation domain
Au, Ag, Cu, Zn Block grade estimates
NSR Net smelter return ($/t)
TP Percent of the block below topographic surface
SG Specific gravity
FLAG Variable used for resource classification purposes

East, north and elevation are three default variables used for block center coordinates.
DOM variable for each block indicates which of the six domains they are a part of. In cases where
61
a block is a part of more than one domain, then the DOM code is set as the domain comprising the
majority of the block. Au, Ag, Cu and Zn are variables used for kriging or inverse distance
weighted block estimates. NSR stands for net smelter return which is a variable generated for
ultimate pit limit analysis. Similarly, TP is generated for ultimate pit limit analysis and it indicates
the percentage of individual blocks below topographic surface. Flag is used to store codes that will
be employed later for resource classification purposes.

Finally, SG is used for specific gravity. It is one of the most crucial parameters in resource
estimation considering its direct control on tonnage, metal content, and stripping ratio. Therefore,
quality of resource estimates and subsequent technical and financial studies clearly reliant on
accuracy and representativeness of specific gravity measurements as well. Specific gravity values
assigned to blocks belonging to one of the six estimation domains determined as a result of
statistical analysis carried out using over 5000 specific gravity measurements. SG measurements
are performed as a part of sample preparation program. Water immersion technique utilized in
determination of specific gravity requires measurement of mass of sample for air-dried (A), wax-
coated (B) and water submerged (C) cases. Then, the specific gravity is calculated as follows:

� � � � � =
−
− − 0.86

SG values for each estimation domain determined in this manner are shown in Table 6.2.

Table 6.2 Specific gravity values of estimation domains

v
De
n
ea
ax
in

d
M

St

Gossan 1.62 4.61 2.56 0.42

Non-gossan oxide 1.84 4.71 2.73 0.56
Barren wallrock 1.83 4.88 2.80 0.40
High-grade sulfide 2.47 4.89 4.30 0.47
Low-grade sulfide 1.94 5.91 3.28 0.63
Enriched Zone 2.14 4.80 3.47 0.85
Table 6.2 clearly indicates the distinctions between estimation domains in terms of specific
gravity. Barren and poorly mineralized parts of the country rock (BW) and domains comprising
oxide zone (GOS and NGO) are associated with lower specific gravities. On the other hand, three

62
units that are associated with sulfide mineralization (HGS, LGS and ENR) are the ones with
relatively larger specific gravity values. Following determination of summary statistics presented
above, average specific gravity value is assigned to individual blocks based on which estimation
domain they belong to.

6.1.2. Geostatistical Data Analysis

Quantification of spatial continuity of the data through variography is one of the most
fundamental aspects of geostatistical resource estimation. Therefore, variography was a crucial
tool in geostatistical analysis of estimation domains. Initially, variography has been carried out for
each single domain which resulted in erratic experimental variograms. Therefore, it has been
realized that it is quite difficult to characterize the spatial continuity with this approach. Later,
variography study is decided to be performed for one variogram in each of the oxide and sulfide
zone. Composite data belonging to gossan zone is considered for the oxide zone while the data for
enriched zone, high-grade sulfide and low-grade sulfide are used together for variography.

As explained earlier in chapter 2, variography study aims to quantify spatial continuity or

variability of the data which is expressed in terms of principal features of the variogram namely
nugget effect, sill and range. Nugget effect represents the discontinuity at the origin which means
the semivariogram value for data pairs that are separated very small distance apart. Therefore, most
practical tool to quantify the nugget effect is downhole variogram. On the other hand, omni-
directional variograms (where angular tolerance is 90o) are utilized in quantification of the total
sill. Then, directional experimental variograms are constructed at a lag spacing roughly capturing
the drillhole grid resolution for log-transformed composite data. Logarithmic transformation of the
data considerably improved the experimental variograms and facilitated subsequent analysis.
Later, resultant experimental variograms are modelled with spherical variogram structures and

63
previously determined nugget and total sill values. Model fitted variograms for GOS and
HGS+LGS+ENR are presented in Figure 6.2.

Figure 6.2 Modelled Variograms for GOS (left) and HGS+LGS+ENR (right)

Final step in the geostatistical analysis of the data is back-transformation of the outcomes.
Principal features of the variogram identified in this manner are compiled in Table 6.3. The
anisotropy suggested by nonidentical ranges along the major, semi-major and minor axis
orientations is consistent with the main attitude of the deposit.

Table 6.3 Summary table of variogram models

Model Orientation Range, m
Domain Structure Nugget Total Sill Major Semi Major Minor Major Semi Major Minor
Gossan zone Spherical 1.00 3.70 31 152 -8 110 71 62
Non-gossan oxide - - - - - - - - -
Barren wallrock - - - - - - - - -
High-grade sulfide
Low-grade sulfide Spherical 0.07 0.8 152 31 -8 139 64 42
Enriched zone

6.1.3. Estimation Strategy

Estimation strategy here is used to describe considerations regarding grade estimation

mainly including estimation method, search and high-grade restriction strategies employed.

Ordinary kriging is used in estimation of precious and base metal components for 10m x
10m x 5m block of the resource model. There exist many reasons behind the choice of this
estimation method. First of all, it is the most widely used linear univariate geostatistical estimation
technique because of relative ease of use. More importantly, deposit geology and statistical
characteristics of the data population after domaining does not necessitate use of more
sophisticated techniques in grade estimation.

64
 Grade estimation is made using ordinary kriging on whole block basis and sub-blocking
and block partials are not considered.

Search strategy is the other key component of grade estimation. It involves considerations
for (i) minimum and maximum number of samples, (ii) size and geometry of the search
neighborhood, and (iii) measures to deal with data redundancy.

Minimum and maximum number of samples are two important estimation parameters
determining the estimation accuracy. Ideally, quantitative kriging neighborhood analysis might be
considered for optimum number of samples. However, the numbers used in this study are as
follows:

 Minimum number of samples : 1

 Maximum number of samples : 10
 Maximum number of samples per hole : 3

Size and geometry of the search neighborhood is another aspect of search strategy. Similarly,
quantitative kriging neighborhood analysis is an effective approach to obtain optimum parameters
for search neighborhood. On the other hand, size and geometry of the search neighborhood is
guided by the variography study. In estimation of grades for block which are a part of gossan zone,
high-grade sulfide, low-grade sulfide and enriched zone, the kriging neighborhood is defined as
ellipses to reflect the anisotropy of spatial continuity captured by directional variograms. On the
other hand, grade estimation for non-gossan oxide and barren wall rock blocks are guided by
isotropic search neighborhoods. Parameters defining the size and geometry of search
neighborhood for each estimation domain are presented in Table 6.4.

Table 6.4 Kriging search neighborhood parameters

Range, m Orientation, degrees
Major Semi Major Minor Major Semi Major Minor
Gossan zone 110 71 62 152 31 8
Non-gossan oxide
100 100 100 - - -
Barren wall rock
High-grade sulfide
Low-grade sulfide 139 64 42 31 152 8
Enriched Zone

Yet another aspect of search strategy involves measures against data clustering. It is a
common practice to deal with data clustering through a quadrant or octant search when there exists
a risk related with preferential sampling hence data clustering. However, a regular pattern has been

65
followed for drilling, thus no actions were taken to deal with data clustering through dividing the
ellipsoid into sectors.

High-grade restriction is the final consideration regarding search strategy. It describes

actions to prevent smearing of extremely high-grade value hence over-estimation. Extremely high-
grade values are also named as outliers. They are identified and reported for each estimation
domain earlier in chapter 4. Grade capping and high-grade restraining are two of the commonly
utilized treatment methods. Former involves truncation of the histogram before the outlier value
in order the eliminate influence of extreme value. On the other hand, latter necessitates use of
restricted search neighborhood. For estimation of Au, Ag, Cu and Zn grades of the block model,
high-grade restraining method is utilized where the size of the search neighborhood is set as 10%
of the base case whose parameters are provided in Table 6.4. With this technique, the outliers
within the restricted kriging neighborhood is considered for grade estimation. Beyond this region
extremely high-grade value is capped to the top-cut value. Such an approach allows both
representation of high-grade value by the blocks in the very near vicinity and elimination of the
risk of excessive smearing.

6.1.4. Resource Classification

There exists a number of reporting guidelines that are widely accepted around the world.
Those codes ask for reporting of confidence of estimation as a part of the disclosure. Resource
classification is a way to meet that requirement. It involves delineation of resource classes as
measured, indicated and inferred with decreasing level of geological knowledge and confidence.
Therefore, resource classification is a way to express confidence of the estimates.

Multiple-pass estimation approach is utilized in classification of resources at this study. The

classification approach involves progressively decreasing the size of the search neighborhood.
FLAG is the model variable generated to store codes for resource classification purposes. Primary
steps involved in identification of resource classes are as follows:

 initially FLAG variable for all the blocks are set as 0

 first pass of estimation is carried out for a search neighborhood whose sizes are 1.5 times of
the base case summarized in Table 6.4, then FLAG variable for the blocks estimated during
this pass are modified to 3

66
 second pass of estimation is done for a search neighborhood with the sizes shown in the base
case, then FLAG variable for the block estimated during this pass are modified to 2
 Third pass of estimation involves a search neighborhood whose sizes are 0.5 times of the base
case, and the FLAG variable of the blocks estimated during this pass are modified to 1

As a result of this procedure, FLAG variable for each block are assigned as one of the four
integer values. Figure 6.3 is an illustration for the outcomes of mentioned resource classification
procedure in a cross-section.

Figure 6.3 An illustration of resource classification

As presented in the cross-section, such a procedure resulted in resources classified as

measured (FLAG =1) where the highest density of data exists. With decreasing data density,
resources are classified as indicated (FLAG =2) and inferred (FLAG =3), respectively.

6.2. Model Validation

Validation of the resource model involves a number of checks to ensure internal

consistency, global and conditional unbiasedness of the product. Validation of the case study
resource model involves comparison of the model against input data as well as alternative
estimation techniques.

First part of the procedure involves assessment of the performance of the estimation
technique against two alternative estimation methods namely nearest neighbor and inverse distance
weighted. In order to assess relative performance of the estimation method, two alternative models
are generated for identical composite data and guided by same search strategy. Histograms are the

67
first tool used for comparison of the outcome. The histogram presented as Figure 6.4 below clearly
indicates consistent results achieved by original resource model and two alternative models.

Figure 6.4 Histogram of Cu grade estimates for original model vs. alternative models

Grade-tonnage curve is another graphical validation technique employed to assess relative

performance of the estimation method with respect to alternatives.

Figure 6.5 Grade-tonnage curves for original model vs. alternative models

Grade-tonnage curves presented above suggests comparable performance for these three
techniques in estimating the copper grades using the same input data (Figure 6.5).

One another tool used in this part of the comparison is swath plot which allows spatial
comparison of the estimates. The swath plot in Figure 6.6 is a graphical display of average copper
grade for a series of 10m slices along NW-SE direction which also suggest consistency between
results of three estimation techniques.

68
 Figure 6.6 Swath plot of Cu grades for original model vs. alternative models

Second part of model validation involves assessment of the quality of estimates by visual
inspection of cross-sections. It further incorporates some of the graphical validation tools presented
earlier to make comparisons of resource model estimates with composite data.

Visual validation may be the most important validation approach utilized for this study. It
relies on inspection of block grade estimates against composite data as shown in Figure 6.7. The
figure suggests that block grade estimates largely capture the grades and the spatial relationships
in the input data.

Figure 6.7 Visual validation of resource model

Comparison of the histograms of block grades with composite grades on domain basis
forms the other step in model validation. The histograms presented in Figure 6.8 indicates the

69
global unbiasedness of copper grade estimates. Furthermore, relatively smaller variance of the
block model grades compared to composite grades is a natural outcome of smoothing effect of
kriging particularly evident in the histogram for high-grade sulfide.

Figure 6.8 Histograms of block model grade vs. composite grade for six estimation domains

Comparison of grade-tonnage curves for block model and composites represents the last
step of model validation. Grade-tonnage curves for block model and composites are provided in
Figure 6.9 on domain basis. The curves generated for estimation domains which are associated
with relatively enriched base metal mineralization (HGS, LGS and ENR) suggests satisfactory
results. On the other hand, other three domains reported under-estimation of the copper grades
over all cut-off grades. Analysis of grade-tonnage curves for precious metal components especially
in two estimation domains (GOS and NGO) forming the oxide zone do not represent the same
problem. Besides, base metal grades within the oxide zone do not have economic significance
because there isn’t any process proposed to extract copper and zinc from the oxide zone. Therefore,
the estimates are decided to be reliable enough to be used in subsequent analysis.

70
 Figure 6.9 Grade-tonnage curves of block model vs. composites for six estimation domains

6.3. Ultimate Pit Limit Analysis

CIM best practices allow classification of part of the mineral endowment as resource that
demonstrates “reasonable prospect for eventual economic extraction” (Abzalov, 2016). The
deposit is a near surface polymetallic deposit which is planned to be extracted with open-pit
mining. Therefore, determination of component of reasonable prospects for economic extraction
achieved by ultimate pit limit analysis using Lerchs-Grossman algorithm in MineSight3D.

Estimated precious and base metal grades, metal prices, pit slope angle, smelter terms,
process cost and recoveries are the main input parameters of this analysis.

CIM best practices are not prescriptive about the metal prices. They rather leave the
responsibility to the qualified person to decide appropriate prices which should be reasonable over
the project life. On the other hand, Securities Exchange Commission (SEC) requires 3-year trailing
average metal prices in estimation of reserves. The metal prices used for resource estimation are:
$1200/oz Au, $18.00/oz Ag, $3.00/lb Cu and $1.20/lb Zn. Lately, those prices are slightly adjusted

71
($1250/oz Au, $18.25/oz Ag, $3.00/lb Cu and $1.10/lb Zn) for mentioned mineral resource
estimation study. Therefore, recent values are also considered in this research.

Precious metal components are initially planned to be extracted by heap leaching of oxide
ore at an average 3,000tpd rate roughly over three years. While sulfide material is planned to be
processed at a flotation circuit to produce copper and zinc concentrates at a 6,500tpd rate.
Subsequent metallurgical testing indicated that tank leaching would be a better alternative to
extract high-grade gold and silver from the oxide zone with significantly higher recoveries.
Furthermore, material classified as enriched zone, thin chalcocite enriched parts of the deposit
associated precious and base metal grades, is found to be refractory in nature, thus currently it is
not planned to be extracted. Table 6.5 presents updated recoveries used for resource estimation
purposes.

Table 6.5 Updated process recoveries

Oxide Zone Au: 88.0% Ag: 63.0%
HGS LGS & BW ENR
Copper Concentrate Zinc Concentrate Copper Concentrate Zinc Concentrate Copper Concentrate Zinc Concentrate
Sulfide Zone Cu: 70.00% Zn: 81.50% Cu: 73.80% Zn: 74.17% Cu: 0.00% Zn: 0.00%
Au: 17.20% Au: 16.00% Au: 40.80% Au: 13.00% Au: 0.00% Au: 0.00%
Ag: 12.30% Ag: 21.50% Ag: 40.10% Ag: 18.96% Ag: 0.00% Ag: 0.00%

Smelter terms comprised of payable percent of the metal content, treatment charge, refining
charges are presented in Table 6.6 below. It is a CIF based agreement, thus smelter terms also
include associated insurance, transportation and port charges.

Table 6.6 Smelter Terms

Au payability : 99.0% Ag payability : 98.0%
Oxide Zone
Au transportation: $5.133/oz Ag transportation: $1.602/oz
Copper Concentrate Zinc Concentrate
Payable Metal Payable Metal
Cu: pay lesser of 96.5% or Cu content less 1% Zn: pay lesser of 85% or Zn content less 8%
Au: pay lesser of 90.0% or Au contect less 1g/t Au: pay 70%after (1g/t deduct from Au content)
Ag:pay lesser of 90% or Ag content less 30g/t Ag: Pay 70% after (93.31 g/t deduct from Ag content)

Treatment Charge : $90.0/ dry t Treatment Charge : $142.51/ dry t

Refining Charges Escalator of $0.10/zinc price above $1850/t
Sulfide Zone Copper : $0.09/lb (if Zn>$1850/t, treatment=$259.80+(Zn pr-1850)*.1)
Gold : $5.00/oz de-escalator of $0.14/dollar zinc price below $1850/t
Silver : $0.50/oz (if Zn<$1850/t, treatment=$259.80-(1850-Zn pr)*.14)

Moisture (assumption) : 12% Moisture (assumption) : 12%

Ocean Freight : $30.00/wet t Ocean Freight : $30.00/wet t
Port Charge : $15.35/wet t Port Charge : $15.35/wet t
Land Freight : $12.07/wet t Land Freight : $12.07/wet t
Insurance : 0.15% of CIF Insurance : 0.15% of CIF

72
 As mentioned earlier in this chapter, a separate block model variable (NSR) is generated
to store economic block values. Net smelter return value is then calculated for each individual
block using grade estimates and provided smelter terms.

In ultimate pit limit analysis, pit slope angles suggested as a result of the slope stability
analysis performed by Fugro-Sial is used. Mentioned study reports the maximum safe slope angles
for weathered and intact rock as 42o and 48o, respectively. Weathered unit is reported to be present
near surface parts of southeast side of the deposit. For the sake of simplicity, 48o is used as the pit
slope angle in the analysis.

Mining and processing costs are the other parameters needed for ultimate pit limit analysis.
Prefeasibility study provided the cost items for mining, heap leaching and flotation. However, the
heap leaching is later planned to be replaced with tank leaching. Therefore, the processing cost
used for tank leaching reflects the assumption of the author of this research. Briefly, mining,
flotation and tank leaching costs used for determination of ultimate pit are $1.5/t, $18.5/t and
$14.5/t, respectively. (processing costs includes general and administrative costs).

An illustration of the ultimate pit determined based on all these technical and economic
parameters using the Lerchs-Grossmann algorithm is presented in Figure 6.10.

Figure 6.10 An illustration of the ultimate pit limits on a SE-NW cross-section

73
 Ultimate pit limits define the limits of optimum economic extraction. Therefore, all the
waste block within the ultimate pit are needed to be removed to expose ore blocks. That’s exactly
why, mining cost is considered as a sunk cost for the blocks within the pit limits. Moreover,
ore/waste classification of those blocks are based on whether net smelter return value is large
enough to justify extraction of economic components by either flotation or leaching. Therefore,
NSR value of a block need to be larger than $18.5/t and $14.5/t to be classified as an ore block in
oxide and sulfide zones, respectively.

The resources reported as inferred, indicated and measured categories for oxide and sulfide
zones in terms of tonnage and average grade based on explicit modelling approach.

The resources represent the part of the deposit that demonstrated reasonable prospects for
eventual economic extraction i.e. within the ultimate pit and above given NSR cutoff for oxide
($18.5/t) and sulfide ($14.5/t) material. As a result, measured and indicated (M&I) resources are
calculated as 4.33 and 30.49 million tons for oxide and sulfide zones, respectively (Table 6.7).

Table 6.7 Mineral resources of the deposit

Oxide (NSR cutoff: $18.5/t) Sulfide (NSR cutoff: $14.5/t)
Resource Class Tonnage, Mt Au, g/t Ag, g/t Cu, % Zn, % Tonnage, Mt Au, g/t Ag, g/t Cu, % Zn, %
Inferred - - - - - 0.34 0.58 24.4 0.53 1.57
Indicated 0.03 0.93 50.3 0.07 0.07 1.90 0.62 18.3 0.43 0.94
Measured 4.30 1.78 52.3 0.14 0.18 28.59 0.55 20.0 0.60 1.20
Measured & Indicated 4.33 1.77 52.3 0.14 0.18 30.49 0.56 19.9 0.59 1.18

74
 CHAPTER 7

DISCUSSION & COMPARISON OF OUTCOMES

The comparative analysis of 3D domain modeling alternatives constitutes the primary

objective of this research. It is aimed to assess the impact of the choice of domain modelling
approach, thus the study involves execution of resource estimation routine using identical
estimation parameters, method and assumptions. Earlier sections of this document are particularly
focused on presentation of main steps and aspects such as estimation method and assumptions.
This chapter, on the other hand, is dedicated for global and local comparison of outcomes as well
as discussion of sources and implications of major discrepancies documented among different
scenarios.

Classification and reporting resources in a format which is compliant with one of the
widely recognized reporting systems constitutes an essential task of resource geologists. Resource
classification is a way to communicate relative confidence of the estimates in that measured,
indicated and inferred resources are distinct with respect to the confidence hence risks associated
with estimates. Furthermore, best practices allow conversion of economically mineable part of
measured and indicated resources into ore reserves as a result of a study involving consideration
of a number of modifying factors. That being the case, it is a common practice to report measured
and indicated resources separately from resources belonging to inferred category. In this respect,
comparison of tonnage and average grade for measured and indicated resources is substantial for
global comparison of outcomes.

Table 7.1 Measured and indicated resources for alternative scenarios

Oxide Sulfide
Tonnage, Mt Au, g/t Ag, g/t Cu, % Zn, % Tonnage, Mt Au, g/t Ag, g/t Cu, % Zn, %
Explicit Model 4.33 1.77 52.3 0.14 0.18 30.49 0.56 19.9 0.59 1.18
Implicit Model 4.20 1.81 53.3 0.16 0.22 32.47 0.55 20.0 0.61 1.20
Indicator Kriging 6.10 1.82 55.0 0.15 0.19 37.56 0.56 21.3 0.60 1.27
Simulation - 1 6.11 1.77 54.4 0.15 0.17 38.12 0.58 21.2 0.61 1.27
Simulation - 2 6.04 1.85 55.9 0.15 0.19 37.45 0.57 21.2 0.59 1.29
Simulation - 3 6.03 1.83 55.1 0.15 0.19 37.83 0.57 21.3 0.60 1.28
Simulation - 4 6.05 1.80 54.7 0.15 0.19 38.44 0.58 21.3 0.60 1.29
Simulation - 5 6.04 1.85 54.1 0.15 0.20 37.44 0.57 21.1 0.60 1.27
Simulation - 6 6.15 1.83 55.6 0.15 0.19 36.31 0.57 21.1 0.60 1.28
Simulation - 7 6.01 1.82 55.9 0.15 0.18 36.52 0.56 21.0 0.61 1.27
Simulation - 8 6.14 1.82 54.1 0.15 0.20 35.27 0.58 21.4 0.61 1.29
Simulation - 9 6.11 1.84 55.0 0.15 0.19 38.46 0.57 21.2 0.61 1.29
Simulation - 10 6.10 1.82 54.1 0.15 0.19 37.35 0.59 21.5 0.60 1.30

75
 Table 7.1 above presents the measured and indicated resources for oxide and sulfide zones
of alternative scenarios in terms of tonnage and average grade of precious and base metals within
associated pit limits at NSR cutoff of $18.5/t and $14.5/t for oxide and sulfide material,
respectively. It is clear that there exists significant discrepancies between outcomes especially with
respect to tonnage of estimates. Moreover, there are differences for average grade of zinc and
silver. It is also evident that the estimates for sequential indicator simulation approach is consistent
internally and with that of indicator kriging approach all of which are significantly larger than the
resource estimates of explicit and implicit modelling approach. Despite being global in scale, such
a comparison suggests either (i) mineralized estimation domains guiding resource estimation are
too conservation for explicit and implicit modelling approaches or (ii) indicator kriging and
sequential indicator simulation results are too optimistic.

Another component of the global comparison of outcomes is visual comparison of 3D

estimation domains particularly the ones associated with elevated precious and base metal grades.
A preliminary observation made during visual inspection is that highly mineralized domains are
notably larger for indicator kriging and simulation compared the ones for explicit and implicit
modelling approaches. Moreover, major discrepancies are observed especially beyond the outmost
drillholes where the extend of estimation domain in other words degree of extrapolation is
controlled by modelling assumptions i.e., range of indicator variogram defines the extend of an
estimation domain beyond the outmost drillhole for indicator kriging and simulation approach
while it is limited to the half of the distance between adjacent drillholes in explicit modelling
approach. Those parts of the deposit are associated with lower confidence resource estimates.
Thus, it is a fair expectation to observe notable differences especially for inferred and indicated
resources. Table 7.2 presents inferred, indicated and measured resources for alternative scenarios.

It is clear that there exists pronounced differences between inferred and indicated resource
estimates for alternative scenarios. However, this does not account for major part of the
discrepancy between the outcomes.

76
 Table 7.2 Measured, indicated and inferred resources
Oxide, ktons Sulfide, ktons Oxide & Sulfide, ktons
Inferred Indicated Measured Inferred Indicated Measured Inferred Indicated Measured
Exp - 24.9 4,304.5 341.7 1,896.5 28,591.3 341.7 1,921.4 32,895.8
Imp - 2.2 4,201.6 53.1 2,115.8 30,355.8 53.1 2,118.0 34,557.4
Ik 0.8 91.9 6,004.6 349.0 4,222.8 33,339.0 349.8 4,314.6 39,343.6
Sim1 8.1 97.5 6,010.3 368.1 4,002.6 34,115.4 376.2 4,100.0 40,125.6
Sim2 7.3 128.3 5,911.3 373.7 3,789.9 33,661.0 381.0 3,918.2 39,572.4
Sim3 8.7 104.9 5,925.1 326.6 4,014.2 33,810.9 335.3 4,119.1 39,736.0
Sim4 - 114.5 5,932.2 363.8 4,607.4 33,831.5 363.8 4,721.9 39,763.7
Sim5 - 67.4 5,972.5 394.1 3,927.1 33,510.4 394.1 3,994.5 39,483.0
Sim6 3.1 144.7 6,002.0 337.3 3,457.5 32,855.5 340.4 3,602.3 38,857.5
Sim7 0.8 77.9 5,935.0 45.0 3,110.3 33,410.1 45.9 3,188.2 39,345.1
Sim8 0.3 87.4 6,050.4 82.7 2,638.1 32,630.7 83.0 2,725.5 38,681.1
Sim9 - 83.0 6,028.1 367.4 4,032.8 34,427.5 367.4 4,115.8 40,455.5
Sim10 - 80.3 6,017.4 401.2 3,717.2 33,632.9 401.2 3,797.5 39,650.3

Global comparison of outcomes by visual comparisons and utilizing resource tables

suggested some preliminary sources for discrepancies. However, a further analysis is needed to
both to illustrate the implications and to identify the causes of the discrepancies. Thus, local
comparisons of outcomes are made to address those issues.

Visual inspection and comparison of estimation domains along cross-sections comprises

the first and most important part of the local comparison.

Figure 7.1 Comparison of HGS domain generated with alternative modelling approaches

Figure 7.1 presents SE-NW cross-sections for high-grade sulfide estimation domain
generated with four alternative domain modelling approaches. Products of explicit and implicit
modelling approach are three-dimensional volumes while those for indicator kriging and

77
simulation are coded blocks. Therefore, the comparison is made on the basis of domain coded
blocks for the sake of consistency.

Based on the cross-sections provided, all the modelling approaches are effective to capture
attitude of the estimation domain. However, there exists significant distinctions in terms of
continuity and extends of the domain modelled. As stated earlier, extend of the estimation domain
is directly controlled by estimation parameters and assumptions. Figure 7.1 indicates that indicator
kriging and simulation results are relatively less continuous compared to those for explicit and
implicit modelling. This is an outcome of the fact that indicator kriging and simulation results are
purely mathematical solutions for the modelling practice. While explicit and implicit modelling
approaches involve considerably more user interference, thus they allow generation of
geologically more meaningful volumes.

Swath plots generated roughly along the dip direction of the deposit are the other tools
utilized for local comparison purposes (Figure 7.2). The plots presented below facilitates local
comparison of average mean grade estimates for precious and base metals for alternative modelling
approaches followed.

Figure 7.2 Swath plots for alternative scenarios

They suggest the discrepancy between the average grades of 10 m NW-SE slices is more
distinct for gold and silver than those for copper and zinc. Furthermore, the results for indicator
kriging and simulation are largely consistent. Gold and silver have the largest grade disparities
comparing explicit/implicit to indicator kriging/conditional simulation models. The swath plot

78
delineates maximum divergence along the middle of the SE-NW traverse which is potentially a
result arising from local domain modelling differences in higher grade Au-Ag in gossan.

Global and local comparisons of the outcomes explicitly revealed the discrepancies for
alternative resource estimates in terms of tonnage and average grade which are two important
technical considerations. However, other technical and economic aspects are also needed for a
more comprehensive assessment and illustration of implications of the choice of estimation
domain modelling approach.

Misclassification of estimation domain is one of the most significant problem. Considering

distinctions between physical properties (e.g. density) and technical factors such as recovery and
treatment charges among different material types, misclassification has pronounced economic
implications.

Figure 7.3 Misclassification of the central block

Figure 7.3 presents a hypothetical case to illustrate the significance of the problem. Cherry
brown and orange blocks indicates high-grade sulfide and low-grade sulfide domains, respectively.
While white colored NE-SW polyline represents the boundary between two estimation domains.
In the case presented above, central block is classified into different material types due to a slight
difference between the boundary polylines. As a result, pronounced differences observed in terms
of net smelter return (NSR) and net economic block values of the same 10mx10mx5m centrally
located block. NSR of the block for the given precious and base metal grades is calculated as
$61.65/t and $68.49/t when it is classified as HGS and LGS, respectively. On the other hand, net
economic block values are calculated as $98,142.7 (HGS) and $86,081.3 (LGS). The results can
even be more serious when an ore grade block is misclassified as a waste block, and hence
excluded from the resources.

79
 Ultimate pit limit represents the extend of economic extraction. Therefore, the size of the
ultimate pit is an indication of the scale of potential future operation as well as size of part of the
deposit meeting reasonable prospects for eventual economic extraction criteria. That being the
case, it is used to illustrate the implications of the modelling approach.

Figure 7.4 shows the ultimate pit limits for alternative modelling approach in a SE-NW
cross-section. Similar to the previous observations, indicator kriging and simulation results are
significantly distinct from those for explicit and implicit modelling approaches. Additionally, the
degree of dissimilarity between the outcomes is clearly a function of the data density.

Figure 7.4 Ultimate pit limits for alternative modelling approaches in section view

Figure 7.5 presents the extends of the ultimate pits belonging to alternative scenarios in
map view. Such a comparison is thought to be particularly an important aspect for surface
infrastructure placement. The figure below suggests that none of the ultimate pits are consistently
larger or smaller than the others.

80
 Figure 7.5 Ultimate pit limits for alternative scenarios in map view

The comparative analysis whose details has been presented so far indicates that estimation
domain modelling decision also has implications in terms of ore, waste tonnages and stripping
ratio.

Figure 7.6 Bar charts for ore tonnage, waste tonnage and stripping ratio

81
 Histograms of the ore tonnage, waste tonnage and stripping ratio belonging to alternative
scenarios are presented in Figure 7.6. Based on the histograms, explicit and implicit modelling
approaches resulted in smaller ore and waste tonnages compared to the ones obtained by indicator
kriging and simulation. The histogram of the stripping ratios, on the other hand, indicates that
explicit modelling approach yields the largest value among alternative results.

Similarly, histogram of total value of the pits for alternative scenarios is utilized to illustrate
the implications of the modelling choice in term of economic considerations.

Figure 7.7 Bar chart of pit values for alternative scenarios

Figure 7.7 above clearly shows the significant discrepancies between the pit values of
different scenarios. Once again, it has been aimed to illustrate implications of estimation domain
modelling alternatives on resource estimates not to decide on the optimum approach. Therefore,
one of the most valuable findings of the study is the range of outcomes defined by the extremes of
the histogram. It is crucial to emphasize that the range of outcomes for technical and economic
considerations are directly related with underlying method and assumptions.

Finally, comparison of outcomes revealed that explicit-implicit pair and indicator kriging-
conditional simulation pair are significantly different while the models yield similar results within
the pairs. Mentioned similarity of indicator kriging and simulation results is due to identical
estimation criteria considered. On the other hand, explicit and implicit modelling results are
comparable because implicit model is guided by the geological knowledge obtained during explicit
modelling.

82
 CHAPTER 8

CONCLUSION

Accurate estimation of resources is a significant task in mineral project evaluation. Therefore,

the study presented in this document focused on an essential component of such a substantial
process. It constitutes estimation of resources for alternative scenarios for which 3D estimation
domains are generated with four different modelling approaches namely explicit modelling,
implicit modelling, indicator kriging and sequential indicator simulation. In order to demonstrate
the discrepancies resulted only from the choice of the domain modelling approach and underlying
assumptions, identical grade estimation method and parameters are considered. Comparison of the
outcomes indicated the significance of domain modelling decision on resource estimates of a
polymetallic massive sulfide deposit located in western Turkey. As a result of the study following
conclusions are made:

 The study demonstrated economic implications of four different modelling decision that are
distinct from each other with respect to method complexity, time and effort. Although it
involves time consuming manual digitization of interpreted geological features, explicit
modelling is the suggested modelling technique because of complete user control. It has been
seen that implicit modelling results when guided by geological knowledge particularly
regarding attitude of the mineralization can be preferable when modelling need to be completed
in relatively shorter time. Application of indicator kriging provides an alternative method to
model categorical variables. Even though it can still be consulted for procurement of order of
magnitude estimates, indicator kriging is not found particularly preferable for domain
modelling purposes. Finally, conditional simulation technique can be effectively utilized for
quantification of uncertainties as long as modelling parameters are guided by sound geological
knowledge.
 It is a challenging task to model volcanogenic massive sulfide type deposits due to complex
deposit geometries and post mineralization deformations. However, this particular deposit was
an ideal case study considering fairly simple deposit geometry and huge amount of data
availability.

83
 There exist distinct domain discontinuities along the strike direction suggesting presence of
potential NE-SW structures. However, structural features are poorly documented due to
extensive cover.
 Fairly tabular mineralization geometry suggests that metamorphism didn’t result in significant
deformation of the original deposit geometry.
 The deposit is associated with economic concentrations of gold, copper, zinc and silver. On
the other hand, lead (non-composited mean grade: 0.08%) and trace elements are in negligible
concentrations. Therefore, grade estimation is focused on procurement of block grade
estimates for gold, copper, zinc and silver.
 Polymetallic nature of studied volcanogenic massive sulfide deposit necessitated calculation
of net smelter return value for each individual block in the model based on estimated metal
grades, price, cost and concentrate recoveries, etc.
 Measured plus indicated resources for the case where the domain models are generated with
explicit modelling are estimated to be 4.33 million tons (average grade of Au:1.77 g/t, Ag:52.3
g/t, Cu:0.14 % and Zn:0.18 %) and 30.49 million tons (average grade of Au:0.56 g/t, Ag:19.9
g/t, Cu:0.59 % and Zn:1.18 %), respectively.
 Measured plus indicated resource estimates for oxide zone range from 4.20 million tons to 6.15
million tons. While those for sulfide zone range from 30.49 million tons to 38.46 million tons.
 The cases involving modeling estimation domains using indicator kriging and sequential
indicator simulations resulted in resource estimates and pit values whose variability is less than
5%. It is primarily due to identical variogram model parameters and indicator codes used in
both methods.
 Analysis showed that the explicit modelling (37.0 million tons) and implicit modelling (38.0
million tons) resulted in comparable ore tonnages which are significantly smaller that the
values obtained for indicator kriging (45.6 million tons) and conditional simulation (min: 43.1
million tons and max: 46.7 million tons).
 Explicit modelling (211.2 million tons) and implicit modelling (201.2 million tons) results are
also somehow comparable in term of waste tonnages. Furthermore, waste tonnages for
indicator kriging (248.3 million tons) and conditional simulation (min: 225.5 million tons and
max: 251.2) are similarly larger than the results for other two techniques.

84
 Explicit modelling (5.71) and implicit modeling (5.30) results are distinct with respect to
stripping ratio. The results obtained by the case where domain modeling is performed by
explicit modeling technique represents the most unfavorable case with the largest stripping
ratio.
 Different ultimate pit limit extents identified by this analysis might be considered as a guide
for decisions regarding surface infrastructure placement.
 The range of outcomes defined by the minimum ($1.04 billion for explicit modelling) and
maximum ($1.45 billion for simulation #9) pit values is proposed as a way to express downside
risks and upside potential of the project for given modelling assumptions.
 The analysis of both cross-sections and 3D models of estimation domains generated using four
different methods indicated that major discrepancies observed beyond the limit defined by the
outermost drillholes. Therefore, underlying assumption of modeling techniques regarding
extrapolation distance is the main source of mentioned dissimilarities.
 Aforementioned differences in domain models especially for domains that are associated with
elevated metal grades (i.e. gossan zone and high-grade sulfide) may represent potential drill
targets

The discrepancies identified between the outcomes of the alternative scenarios explicitly
demonstrated the substantial role of the geological understanding in mining project evaluation.
Considering its potential influences as well as economic implications on downstream mining
practices which are discussed earlier, quality of the estimation domain model, which is utilized as
a key guiding tool during estimation of a property, is clearly important. Therefore, decision of the
modelling approach should involve consideration of not only complexity, time and effort
requirement but also its potential influences on subsequent applications.

All in all, the estimation of resources requires detailed consideration of various aspects involving
contribution of experts from various disciplines. Like a chain, they are linked in that the quality of
the overall resource estimate will be equal to the quality of the weakest link.

85
 REFERENCES

Abzalov, M. (2006). Localized Uniform Conditioning (LUC): A new approach for direct block
modelling of small blocks. Mathematical Geology, 393-411.
Abzalov, M. (2016). Applied Mining Geology. Switzerland: Springer.
Caers, J. (2011). Modeling Uncertainty in the Earth Sciences. Wiley-Blackwell.
David , M. (1988). Handbook of applied advanced geostatistical ore reserve estimation.
Amsterdam: Elsevier.
Deutsch, J. L. (2015). Experimental Variogram Tolerance Parameters. Retrieved from
http://geostatisticslessons.com/lessons/variogramparameters
Fouedjio, F. (2016). A hierarchical clustering method for multivariate geostatistical data. Spatial
Statistics, 333-351.
Glacken, I. M., & Snowden, D. V. (2001). Mineral Resource Estimation. In Mineral Resource
and Ore Reserve Estimation – The AusIMM Guide to Good Practice (pp. 189-198).
AUSSIM.
Goncuoglu, M. C. (2007). Geology of the Kutahya-Bolkardag Belt. Menderes Massif
Colloquium Proceedings, (pp. 39-43).
Goovaerts, P. (1997). Geostatistics for natural resources evaluation. New York: Oxford
University Press.
Isaaks, E., & Srivastava, R. (1989). An Introduction to Applied Geostatistics . Newyork: Oxford
University Press.
JORC Code. (2012). Australaisian code for reporting of exploration results, mineral resources
and ore reserves. Melbourne: AUSIMM.
Journel, A. G., & Huijbrets, C. J. (1978). Mining Geostatistics. New York: Academic.
Ketin, I. (1966). Anadolu'nun tektonik birlikleri. Maden Tetkik ve Arama Enstitusu Dergisi, 20-
34.
Krige, D. (1951). A statistical approach to some basic mine valuation problems on the
Withwaterstand. SAIMM, 119-139.
Krige, D. (1994). An analysis of some essential basic tenets of geostatistics not always practiced
in ore valutations. Aplication of Computers and Operations Research in the Mineral
Industries (pp. 15-28). Slovenia: APCOM.
Krige, D. (1996). A Practical Analysis of the Effects of Spatial Structure and of the Data
Available and Accessed, on Conditional Biases in Ordinary Kriging. Geostatistics, (pp.
799-810). Wollongong.

86
Matheron, G. (1963). Principles of Geotstatistics. Economic Geology, 1246-1266.
Matheron, G. (1968). Basics of Applied Geostatistics. Mir, Russia.
Okay, A. (1985). Metamorphic belts in northwest Turkey. Ketin Symposium Proceedings (pp.
83-92). Geological Society of Turkey.
Parker, H. M., & Dohm, C. E. (2014). Mineral Resource-Crirsco. Retrieved from Evolution of
Mineral Resource Classification from 1980 to 2014 and Current Best Practice:
http://www.crirsco.com/docs/H_Parker_Finex.pdf
Rossi, M., & Deutsch, C. (2014). Mineral Resource Estimation. Delray Beach, Florida: Springer.
Vallee, M. (2000). Mineral Resource + Engineering, Economic and Legal Feasibility=Ore
Reserve. CIM Bulletin, 53-61.
van Hinsbergen, D. J., Maffione, M., Plunder, A., Kaymakci, N., Ganerod, M., Hendriks, B. W.,
Vissers, R. L. (2016). Tectonic evolution and paleogeography of the Kirsehir Block and
the Central Anatolian Ophiolites, Turkey. Tectonics, 983-1014.
Vann, J., Jackson, S., & Bertoli, O. (2003). Quantitative Kriging Neighborhood Analysis for the
Mining Geologist - A Description of the Method with Worked Case Examples. 5th
International Mining Geology Conference (pp. 215-223). Bendigo, Victoria: AUSSIM.

87
 APPENDIX A

CONTACT PLOTS

Figure A.1 ClyGos-Gos contact plots

Figure A.2 Mpy-MpyMag contact plots

88
Figure A.3 HGS-LGS contact plots

Figure A.4 GOS-NGO contact plots

89
Figure A.5 HGS-BW contact plots

Figure A.6 LGS-BW contact plots

90
 APPENDIX B

MODELLING METHODS

This section of the report is prepared to provide further details regarding major steps and
assumptions made during generation of three-dimensional estimation domain models using four
modelling techniques: (i) explicit modelling, (ii) implicit modelling, (iii) indicator kriging, and
(iv)conditional simulation. A digital appendix is also included in the form of a flash drive which
can be referred for review purposes. Original dataset containing whole drillhole database as well
as other complementary data are included in the folder names as Original Data. On the other hand,
part of the database that has been used in this study can be found under folder named as Model
Data.

Explicit Modeling
It represents traditional technique for construction of three-dimensional models. Explicit
modelling heavily relies on manual digitization of interpreted geological features on cross-
sections.
Main steps of explicit modelling:
 Step 1- Delineation of stationary estimation domains
 Step 2 - Interpretation of continuity of estimation domains and construction of cross-
sections and long-sections
 Step 3- Digitization of interpretations on cross-sections
Contacts digitized along cross-sections and long-sections are provided under X-
Sections folder (MineSight3D) in the digital appendix

 Step 4- Building wireframes (also known as triangulations) through linking digitized

boundaries representing the contacts between adjacent estimation domains
Domain wireframes are provided under Explicit approach subfolder of Estimation
domains folder in the digital appendix

 Step 5- Final adjustment of wireframes based on cross-cutting relationships

 Step 6- Validation of the model with input data, sections and maps
91
 Assumption(s): Additional remarks:

Major assumption made during generation of The best practices suggest interpretation of
3D models by explicit modelling technique geology on all three orthogonal planes
is extrapolation distance. Similar to the including cross-sections, long-sections and
common industry practice, nearly half of the plan views. This study involves definition of
distance between adjacent drillholes is the mineralization geometry along NE-SW
considered for extrapolation purposes. and NW-SE sections.

Implicit Modelling
It is a more recent technique compared to explicit modelling. Implicit modelling relies on software
driven algorithms (i.e. radial basis function) for generation of surfaces corresponding to boundaries
between data populations.

Main steps of implicit modelling:

 Step 1- Delineation of stationary estimation domains

 Step 2- Modelling surfaces representing boundaries between adjacent estimation
domains using RBF

 Contact surfaces are modelled in Leapfrog Geo. However, they are not provided with
the digital appendix

 Step 3- Definition of contact surface chronology and identification of the background

lithology

Contact surface chronology is defined from younger to older as gossan zone (GOS)
and non-gossan oxide (NGO) for the oxide zone, enriched zone (ENR), high-grade sulfide
(HGS), low-grade sulfide (LGS) and barren wall rock (BW) in the sulfide zone

92
Main steps of implicit modelling (cont.)

 Step 4- Generation of 3D solid models based on modelled contacts and defined surface
chronology

Domain wireframes are generated in a commercial modelling software, Leapfrog

Geo. They have later been imported into MineSight3D. Domain wireframes are provided
under Implicit approach subfolder of Estimation domains folder in the digital appendix

 Step 5- Validation of implicit model by comparing with input data, cross-sections and
surface geological maps

Assumption(s): Additional remarks:

Main assumption made in generating 3D  Preliminary implicit model which is

domain models using implicit modelling generated based on isotropic search
technique is model resolution which strategy was not successful to capture
determines the minimum length of input the spatial characteristics of estimation
data that will be considered for modelling domains. Thus, the model is improved
purposes. Considering composite length by introducing a trend reflecting the
(5m), resolution of the model is set as attitude of the deposit (strike: N15E and
5m. dip: 25o) which has been identified
during explicit modelling
 A further improvement is achieved
through use of 3D polylines both to limit
extents of each volume and to get rid of
modelling artifacts observed in the form
of isolated bodies

93
Indicator Kriging
Application of indicator kriging, a non-linear kriging technique, provides an alternative approach
for three-dimensional modelling of categorical variables.

Main steps of indicator kriging:

 Step 1- Delineation of stationary estimation domains
 Step 2- Non-linear transformation of data into six set of binary indicator codes (0:
interval not belonging to the estimation domain, 1: interval belonging to the estimation
domain)

Step 3- Construction of directional indicator variograms

Directional indicator variograms are provided under Variograms subfolder of MSDA

folder in the digital appendix. V1 and V456 prefixes are used for the indicator variograms
for oxide and sulfide zones, respectively

 Step 4- Characterization of spatial continuity of individual estimation domains by

fitting variogram models for experimental directional variograms

Fitted variogram model parameters file is provided as a text file within Variogram
subfolder of MSDA folder in the digital directory

 Step 5- Execution of indicator kriging with generated indicator codes and variogram
model parameters to generate 3D solid models of estimation domains

Estimation domain codes used to store domain information for the block model
named as BM-IK provided under Block models folder of the digital directory are as follows:
GOS: 1, NGO: 2, BW: 3, HGS: 4, LGS: 5 and ENR:6

 Step 6- Validation of solid models by comparing with input data, cross-sections and
surface geological maps

94
 Assumptions: Additional Remarks:

 Ellipsoidal search whose dimensions are the Even though, 3D domain

same as the ranges of indicator variogram along modeling is performed on 10m x
major continuity directions 10m x 5m sized regular block
 For variography intercepts belonging to gossan basis, it has been thought that
zone are considered for oxide zone while those contacts between adjacent
for high-grade sulfide, low-grade sulfide and estimation domains might be
enriched zone employed for sulfide zone better defined for smaller sized
 Indicator threshold = 0.5 (i.e. if the indicator blocks.
kriging estimate for a block is greater than 0.5,
then the block belongs to associated estimation
domain)

Conditional Simulation
Conditional simulation is the fourth technique by which 3D solid models can be generated. The
method is utilized to generate mathematically equiprobable realizations in other words outcomes.

Main steps of conditional simulation:

 Step 1- Delineation of stationary estimation domains
 Step 2- Non-linear transformation of data into six set of binary indicator codes (0:
interval not belonging to the estimation domain, 1: interval belonging to the estimation
domain)
 Step 3- Construction of directional indicator variograms

Directional indicator variograms are provided under Variograms subfolder of MSDA

folder in the digital appendix. V1 and V456 prefixes are used for the indicator variograms
for oxide and sulfide zones, respectively

 Step 4- Characterization of spatial continuity of individual estimation domains by

95


Main steps of conditional simulation (cont.)
 Step 4- Characterization of spatial continuity of individual estimation domains by
fitting variogram models for experimental directional variograms

Fitted variogram model parameters file is provided as a text file within Variogram
subfolder of MSDA folder in the digital directory

 Step 5- Execution of sequential indicator simulation with generated indicator codes and
variogram model parameters to generate 10 realizations for 3D solid models of
estimation domains

Estimation domain codes used to store domain information for the block models
named as BM-SIM1 to BM-SIM10 provided under Block models folder of the digital
directory are as follows: GOS: 1, NGO: 2, BW: 3, HGS: 4, LGS: 5 and ENR:6

 Step 6- Validation of solid models by comparing with input data, sections and maps

Additional remarks:
Assumptions:
Similar to indicator kriging,
 Ellipsoidal search whose dimensions are the
consideration of smaller blocks
same as the ranges of indicator variogram along
could be effective to better define
major continuity directions
the contacts between adjacent
 In variography, intercepts belonging to gossan
estimation domains.
zone are considered for oxide zone while those
for high-grade sulfide, low-grade sulfide and
enriched zone employed for sulfide zone
 Indicator threshold = 0.5 (i.e. if the estimate for
a block is greater than 0.5, then the block
belongs to associated estimation domain)

96
 APPENDIX C

3D DOMAIN MODELS

Figure C.1 3D models of gossan zone (GOS)

89
Figure C.2 3D models of non-gossan oxide (NGO)

90
Figure C.3 3D models of barren wall rock (BW)

91
Figure C.4 3D models of high-grade sulfide (HGS)

92
Figure C.5 3D models of low-grade sulfide (LGS)

93
Figure C.6 3D models of enriched zone (ENR)

94
 Table C.1 Volumes for estimation domain solid models

Solid Model Volumes, million m3

GOS NGO BW HGS LGS ENR
Explicit Modelling 2.12 8.13 765.64 3.30 14.98 0.34
Implicit Modelling 1.90 8.32 764.92 3.68 15.39 0.30
Indicator Kriging 3.14 10.11 760.17 4.72 16.13 0.24
Simulation 1 3.06 10.19 760.26 4.99 15.80 0.21
Simulation 2 3.09 10.17 760.25 4.68 16.08 0.24
Simulation 3 3.24 10.01 760.23 4.76 16.07 0.21
Simulation 4 2.98 10.28 760.06 4.86 16.08 0.26
Simulation 5 3.00 10.25 760.23 4.76 16.01 0.25
Simulation 6 3.02 10.23 760.69 4.76 15.58 0.24
Simulation 7 3.35 9.91 760.27 4.79 15.92 0.28
Simulation 8 3.06 10.20 760.53 4.72 15.76 0.24
Simulation 9 3.11 10.15 760.25 4.82 16.03 0.16
Simulation 10 3.05 10.20 760.26 4.75 15.97 0.28

95
 APPENDIX D

MEASURED, INDICATED AND INFERRED RESOURCES OF INDIVIDUAL ESTIMATION DOMAINS

Table D.1 Measured, indicated and inferred resources for oxide zone
Gossan Zone, ktons Non-gossan Oxide, ktons Oxide Zone, ktons
Inferred Indicated Measured Inferred Indicated Measured Inferred Indicated Measured
Exp - 18.1 1,972.8 - 6.8 2,331.6 - 24.9 4,304.5
Imp - - 1,780.4 - 2.2 2,421.2 - 2.2 4,201.6
Ik 0.8 89.2 3,027.8 - 2.7 2,976.8 0.8 91.9 6,004.6
Sim1 8.1 93.3 2,947.5 - 4.1 3,062.7 8.1 97.5 6,010.3
Sim2 7.3 123.0 2,908.7 - 5.3 3,002.6 7.3 128.3 5,911.3
Sim3 8.7 104.9 3,029.4 - - 2,895.7 8.7 104.9 5,925.1
Sim4 - 104.0 2,857.4 - 10.5 3,074.9 - 114.5 5,932.2
Sim5 - 60.7 2,845.7 - 6.7 3,126.8 - 67.4 5,972.5
Sim6 3.1 137.6 2,871.9 - 7.1 3,130.1 3.1 144.7 6,002.0
Sim7 0.8 68.9 3,056.2 - 9.0 2,878.9 0.8 77.9 5,935.0
Sim8 0.3 83.0 2,905.1 - 4.4 3,145.3 0.3 87.4 6,050.4
Sim9 - 79.9 2,949.4 - 3.1 3,078.6 - 83.0 6,028.1
Sim10 - 72.7 2,975.3 - 7.6 3,042.1 - 80.3 6,017.4

Table D.2 Measured, indicated and inferred resources for sulfide zone
Barren wallrock, ktons High-grade Sulfide, ktons Low-grade Sulfide, ktons Sulfide Zone, ktons
Inferred Indicated Measured Inferred Indicated Measured Inferred Indicated Measured Inferred Indicated Measured
Exp 6.9 524.9 3,802.1 6.0 374.2 13,225.3 328.8 997.4 11,563.9 341.7 1,896.5 28,591.3
Imp 6.9 493.4 4,163.6 - 238.5 14,966.1 46.2 1,383.9 11,226.1 53.1 2,115.8 30,355.8
Ik 254.7 1,330.7 3,734.2 66.7 1,637.1 17,118.4 27.5 1,254.9 12,486.5 349.0 4,222.8 33,339.0
Sim1 285.5 1,243.7 3,749.4 81.2 1,901.9 18,092.9 1.4 857.1 12,273.1 368.1 4,002.6 34,115.4
Sim2 290.3 1,366.4 3,641.2 77.9 1,475.5 17,254.9 5.5 948.0 12,764.9 373.7 3,789.9 33,661.0
Sim3 254.7 1,303.8 3,832.3 68.0 1,700.4 17,265.6 4.0 1,010.0 12,712.9 326.6 4,014.2 33,810.9
Sim4 291.9 1,357.3 3,609.1 67.7 1,945.0 17,439.1 4.3 1,305.1 12,783.2 363.8 4,607.4 33,831.5
Sim5 289.6 1,276.6 3,454.7 71.0 1,718.3 17,247.5 33.5 932.3 12,808.3 394.1 3,927.1 33,510.4
Sim6 289.5 1,268.0 3,444.5 43.1 1,658.2 17,243.0 4.7 531.3 12,168.0 337.3 3,457.5 32,855.5
Sim7 5.6 547.9 3,691.0 36.8 1,604.5 17,268.7 2.7 957.9 12,450.4 45.0 3,110.3 33,410.1
Sim8 5.6 521.2 3,486.3 48.7 1,395.2 17,034.0 28.4 721.6 12,110.4 82.7 2,638.1 32,630.7
Sim9 254.7 1,334.2 3,846.7 110.6 1,472.8 17,718.4 2.1 1,225.8 12,862.3 367.4 4,032.8 34,427.5
Sim10 291.9 1,251.1 3,572.6 94.6 1,673.5 17,164.4 14.7 792.6 12,895.8 401.2 3,717.2 33,632.9

96
 APPENDIX E

COMPARISON OF ESTIMATION DOMAINS

Figure E.1 Comparison of LGS domain generated with alternative modelling approaches

Figure E.2 Comparison of Gossan zone generated with alternative modelling approaches

97