School of Mining Engineering

A REVIEW AND ASSESSMENT OF THE

APPROPRIATENESS OF THE MINERAL RESOURCE
ESTIMATION AND CLASSIFICATION PRACTICES, AS
CURRENTLY APPLIED TO THE MG1 CHROME SEAM ON A
BUSHVELD COMPLEX CHROME MINE

Andre Hanekom

Student number:886823

Supervisor: Prof Christina Dohm

A research report submitted to the Faculty of Engineering and the Built

Environment, University of the Witwatersrand, Johannesburg, in partial fulfilment

of the requirements for the degree of Master of Science in Mining Engineering.

Johannesburg, 2020
DECLARATION

I declare that this report is my own, unaided work. I have read the University Policy

on Plagiarism and hereby confirm that no plagiarism exists in this report. I also

confirm that there is no copying nor is there any copyright infringement. I willingly

submit to any investigation in this regard by the School of Mining Engineering and

I undertake to abide by the decision of any such investigation.

Furthermore, I declare that I have obtained authorisation with regards to the

research study by the company the research is conducted on.

30th of October 2020

_________________________ __________________

André Hanekom Date

i
ABSTRACT

The applicability of both the Mineral Resource estimation and classification

practices of the MG1 chrome seam on a Bushveld Complex mine are addressed
in this research. The Growth technique (a 2D gridding algorithm) available in
Geovia MinexTM software, currently used for resource estimation. Mineral Resource
classification is based on the application of fixed radii (150m, 300m and 600m for
Measured, Indicated and Inferred Resources respectively) to verified MG1
chromitite borehole intersections. The Competent Person can at his/her discretion,
manually adjust classified Mineral Resource boundaries resulting from the radii for
resource declaration. These practises stem from experience gained over many
years of mining the MG1 chromitite resource. Establishing formal methodologies
supported by research and comparative studies for MG1 chrome resource
estimation and classification are the focus areas for this research investigation.
The benefit of having technical support for the resource estimation and
classification practices is that this will assist the Competent Person (CP) in his/her
justification for estimation and classification methodologies in the resource
declaration of the mine, a requirement of the Mineral Resource reporting codes.
Having formal backing for these practices is considered value adding to the mine.

Growth technique estimation is compared to alternative estimation methodologies

namely Inverse Distance Weighting (IDW) and Ordinary Kriging (OK). The fixed
radii resource classification procedure is weighed up against published Mineral
Resources classification approaches found in the resources industry. Within the
current sampling configuration IDW and Ordinary Kriging estimation results were
globally similar to those of the Growth technique. Locally however, IDW to the
power two and OK produced more representative estimates.

Mineral Resource classification methods based on geostatistical parameters and

measures confirmed that the fixed borehole confidence radii used for resource
classification are applicable. Spatially the geostatistical classification approach
resulted in a more sensible representation of the Mineral Resource categories. In
conclusion the influence and impact of geological factors on resource classification
are stressed. The development and implementation of a Mineral Resource
classification scorecard approach is recommended, as such would result in a
transparent, defendable, and robust classification methodology meeting mineral
Resource Reporting Code requirements.

ii
ACKNOWLEDGEMENTS

First and foremost, I want to thank God for blessing me with this opportunity and
providing me with the knowledge to complete this research project.

Furthermore, I want to thank my family, friends and colleagues for their support
and believing in me throughout the process of this research project. I also want to
thank the mining company which granted me permission to conduct this study,
making use of the relevant software, and providing me with the data for this
research.

I want to acknowledge the following individuals for their contributions:

• Professor Christina Dohm (University of the Witwatersrand), for her

guidance, assistance, support and patience, and challenges which kept me
determined to complete this research project.
• My wife, Leani Hanekom, for all her love, support, patience and for
motivating me.
• The Superintendent of Geology which worked for the mining company for
preparing the data for this research project and documentation that
explains the Resource estimation and classification procedure of the mine,
including technical inputs.
• Justin Glanville, for technical advice and guidance, including inputs towards
data conversions and macro’s which assisted me in the analysis of the
technical part of the research.
• Antonio Umpire, for providing technical reviews and software guidance.
• Isobel Clark, for reviewing and providing guidance on variography.
• Felix Walraven (Seequent Pty Ltd) for providing software licenses for
Leapfrog® and Datamine® for the use of this research project.

iii
TABLE OF CONTENTS

DECLARATION ...........................................................................................i

ABSTRACT ................................................................................................. ii

ACKNOWLEDGEMENTS .......................................................................... iii

TABLE OF CONTENTS ............................................................................. iv

LIST OF FIGURES ................................................................................... vii

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

APPENDIX ................................................................................................. xi

LIST OF UNIT SYMBOLS ......................................................................... xii

LIST OF ABBREVIATIONS ..................................................................... xiii

1 Introduction.......................................................................................... 1

1.1 Project Background ............................................................................... 1

1.2 Estimation and Classification Methodologies in Place at MCo ............... 2

1.3 Research Project Objective ................................................................... 3

1.4 Problem Statement ................................................................................ 4

1.5 Research Hypothesis and Assumptions................................................. 4

1.6 Research Motivation .............................................................................. 4

1.7 Research Questions and Methodology .................................................. 4

1.8 Research Project Outline ....................................................................... 6

2 Literature review .................................................................................. 8

2.1 Resource Estimation, Classification and Reporting ................................ 8

2.2 Factors Impacting Mineral Resource Estimation .................................. 10

2.3 Resource Estimation Methodologies .................................................... 11

2.3.1 Growth technique ......................................................................... 12

2.3.2 Alternative Estimation Techniques ................................................ 13

2.4 Resource Classification Review ........................................................... 15

2.4.1 Classification Practices ................................................................. 15

iv
 2.5 Reality of Mineral Resource Estimation in the Mining Industry ............. 17

3 Geology and data collection .............................................................. 18

3.1 Regional Geology ................................................................................ 18

3.2 Project Geology ................................................................................... 21

3.1.2 Stratigraphy .................................................................................. 22

3.3 Data collection.......................................................................................... 23

3.3.1 Drilling Techniques ....................................................................... 25

3.3.2 Collar Positions............................................................................. 25

3.3.3 Down Hole Surveys ...................................................................... 25

3.3.4 Geological Logging ....................................................................... 25

3.3.5 Sampling ...................................................................................... 26

3.3.6 Bulk Density ................................................................................. 27

3.3.7 Database Integrity, Quality and Validation .................................... 27

3.4 Geological Modelling ........................................................................... 30

3.4.1 Geological model ............................................................................... 32

3.4.2 Geological structures .................................................................... 34

4. Exploratory data analysis (EDA) ....................................................... 37

4.1 Data Selection and Representativeness .............................................. 37

4.2 Compositing and Descriptive Statistics ................................................ 38

4.3 Graphical representation of the data ................................................... 41

4.3.1 Histograms ................................................................................... 41

4.3.2 Normal probability plots ................................................................ 43

4.4 Domaining ........................................................................................... 46

4.4.1 Stationarity ................................................................................... 46

5. Variography (spatial analysis) ........................................................... 51

5.1 The Semi- variogram and variogram model ......................................... 51

5.2 Variography of variables under study ................................................... 54

5.2.1 Variogram Model Range of Influence and Resource Classification59

v
6. Block modelling and estimation ......................................................... 62

6.1 Block Model Size and Search Neighbourhood ..................................... 62

6.1.1 Quantified Kriging Neighbourhood Analysis (QKNA) results ......... 64

6.2 Estimation Comparisons and Analysis ................................................. 64

6.2.1 Estimation results, comparisons and validation .................................. 68

6.2.2 Comparison of Estimation Techniques within the MCo Resource

Classification .............................................................................................. 80

7. Factors that can influence the MG1 Resource classification ............. 84

7.1 Non-numeric Factors Relevant to Mineral Resource Classification ...... 84

7.2 Borehole Radii Approach in Resource Classification ........................... 87

8. APPROPRIATENESS of MCo’s MINERAL RESOURCE

CLASSIFICATION ................................................................................... 93

8.1 Classification Based on the “two thirds” of the Range Method ............. 93

8.2 Classification Based on Kriging Input and Output Parameters ............. 94

8.2.1 Kriging Variance, Kriging Efficiency and Slope of regression ........ 95

8.2.2 Resource Classification Comparison Results ............................. 102

8.3 Principles of a Mineral Resource classification Scorecard-based

Approach...................................................................................................... 104

9. Conclusions ..................................................................................... 107

10. Recommendations .......................................................................... 110

References............................................................................................. 112

APPENDIX ............................................................................................. 123

vi
LIST OF FIGURES

Figure 1: Satellite image: general area of the research project indicated by the
red circle (Google Maps , 2020). ............................................................ 2
Figure 2: Linkages between Exploration results, Mineral Resources and Mineral
Reserves (SAMREC, 2016). .................................................................. 9
Figure 3: Simplified map of the Bushveld Complex showing the location of the
various limbs (Kinnaird et al, 2004). ..................................................... 19
Figure 4: Typical Stratigraphic column of the Rustenburg Layered Suite with
general zone thicknesses; major chromitite layers of the Critical zone,
with Cr2O3 and PGE compositional data (column on the right) based on
Scoon and Teigler (1994). (Kinnaird et al, 2002). ................................. 20
Figure 5: Floor folds in the Magaliesberg Quartzite Formation in the south-
western Bushveld Complex, west of the property (modified after Davey,
1992 cited in Dube, 2010). ................................................................... 22
Figure 6: General Stratigraphy underlying the project area (MCo, n.d.). ............ 23
Figure 7: Borehole collars from available surface drilling data provided by MCo.24
Figure 8: Sample blank analysis for Cr2O3% with QA/QC thresholds. ................ 29
Figure 9: SARM146 sample analysis for Cr2O3% with QA/QC thresholds. ......... 30
Figure 10: Orthogonal view of the interpreted MG1 top contact of the 3D
geological model. ................................................................................. 32
Figure 11: 3D Geological model overlain by MCo’s major geological structural
interpretation (plan view). ..................................................................... 33
Figure 12: Un-composited Cr2O3% sample data. .............................................. 39
Figure 13: Composited Cr2O3% sample data. ................................................... 39
Figure 14: Histograms and fitted normal distributions for Cr2O3%, FeO%, SiO2%,
CR:FE, SG and Channel/Seam Width from left to right and top to bottom
respectively. ......................................................................................... 42
Figure 15: Normal cumulative probability plots for Cr2O3%, FeO%, SiO2%,
CR:FE, SG and Channel/Seam Width. ................................................. 44
Figure 16: Swaths plots for Cr2O3% sample composites................................... 48
Figure 17: Swaths plots for CW sample composites. ......................................... 49
Figure 18: Swaths plots for SG sample composites ........................................... 50
Figure 19: Experimental semivariogram with a spherical model (Snowden, 2001).
............................................................................................................. 52
Figure 20 : Components of the spherical model (Sinclair & Blackwell, 2002). .... 54

vii
Figure 21: Isosurfaces for Cr2O3%.......................................................................55
Figure 22: Isosurfaces for CW. .......................................................................... 55
Figure 23: Isosurfaces for SG. ........................................................................... 56
Figure 24: Spherical Semivariogram model for Cr2O3%. .................................... 57
Figure 25: Spherical Semivariogram model for true channel width (m). ............. 57
Figure 26: Semivariogram with spherical model for SG...................................... 58
Figure 27: Cr2O3% block model estimates (Growth technique and alternative
methods) .............................................................................................. 69
Figure 28: Volume percentage based on Cr2O3% grade distribution of the
estimation scenarios ............................................................................ 70
Figure 29: Percentage difference of Cr2O3% block estimate means and the
declustered composite sample mean. .................................................. 72
Figure 30: Histograms of the Cr2O3% block estimate scenario’s. ....................... 75
Figure 31: Histogram of original sample input data (Cr2O3% composites). ......... 76
Figure 32: Q-Q Plots of the Growth technique estimate versus the alternative
estimation methods. ............................................................................. 77
Figure 33: Scatter Plots of the Growth technique estimate versus the alternative
estimation methods. ............................................................................. 78
Figure 34: MCo’s resource classification for the current sample data
configuration. ....................................................................................... 83
Figure 35: Possible pothole influence zones in the project area. ........................ 86
Figure 36: Example of a “spotted dog” – what not to do (Guidelines Review
Committee (JORC), 2014).................................................................... 88
Figure 37: Minimum borehole spacing for each Coal Resource classification
category for the various South-African coal deposit types (SANS
10320:2004 cited in Hancox & Pinheiro, 2017)..................................... 90
Figure 38: Example of a Coal Resource classification based upon SANS
10320:2004 (Exxaro, 2015). ................................................................. 91
Figure 39: Example of the borehole distance gridding functionality in Geovia
MinexTM (GEOVIA, 2014). .................................................................... 92
Figure 40: Visual comparison between the “two thirds” (left) and MCo’s (right)
Resource classification approaches. .................................................... 94
Figure 41: Resource classification criteria based upon the cumulative log
probability of the KV values of OK_1. ................................................... 96

viii
Figure 42: Resource classification criteria based on KV(left) compared to that of
MCo (right) resource classification perimeters. ..................................... 97
Figure 43: Resource classification criteria based on KV and variogram range of
influence compared to the MCo resource classification perimeters. ..... 98
Figure 44: Correlation between efficiencies of block valuations and slopes of
regression (Krige, 1996). ...................................................................... 98
Figure 45 : Resource classification criteria specified in Table 18 compared to
MCo's resource classification perimeters. .......................................... 101
Figure 46 : Correlation between Kriging Efficiency and slope of regression, with
calibrated resource classification criteria based on Table 18 (modified
after Krige (1996)). ............................................................................. 102
Figure 47: Comparative Mineral resource classification methods, compared to
MCo’s resource classification perimeters based on volume percentage.
........................................................................................................... 103

ix
LIST OF TABLES

Table 1: Pros and Cons of the IDW technique (Gentile, et al., 2012). ................ 14
Table 2: Number of boreholes used for statistical analysis ................................. 38
Table 3: Descriptive statistics of the un-composted and composited borehole
data. ..................................................................................................... 40
Table 4: Rules for analysing COV (modified after Noble , 1992). ....................... 41
Table 5: Composite data statistics (geological influenced intersections removed).
............................................................................................................. 45
Table 6: Parameters of modelled spherical variograms for Cr2O3%, CW and SG.
............................................................................................................. 58
Table 7: Comparison between two-thirds variogram range technique and the
mining company’s standard for Mineral resource classification. ........... 61
Table 8: Kriging neighbourhood parameters based on QKNA. ........................... 64
Table 9: Tested estimation scenario’s, with parameters used for comparison
towards the Growth technique. ............................................................. 66
Table 10: Variogram model parameters, used in estimation’s comparisons
towards Growth technique.................................................................... 67
Table 11: Statistics of the estimation methods and sample composites. ............ 73
Table 12: Estimated Cr2O3% grades and tonnages of all estimation methods
within MCo’s defined resource classification perimeters. ...................... 81
Table 13: Grade (Cr2O3%) percentage differences of the Growth estimate
compared to the alterative estimates as per MCo’s resource
classification perimeters. ...................................................................... 81
Table 14: Alternative estimated (volume weighted) Cr2O3% percentage
differences compared to the Growth estimate as per MCo’s classification
perimeters. ........................................................................................... 82
Table 15: OK_1 Kriging variance classification criteria based on Figure 41. ...... 96
Table 16: Mineral Resource classification categories based on Kriging efficiency
ratios (Mwasinga,2001) . .................................................................... 100
Table 17: Resource classification criteria commonly used for coal resources
(Mwasinga,2001) ............................................................................... 100
Table 18: Resource classification using KE ratio and SOR that closely resemble
MCo’s resource classification perimeters. .......................................... 101

x
APPENDIX

Appendix 1: Swath plot comparison between the alternative estimation methods

(IDW2;IDW4;OK_1;OK_2;OK_3) and the Growth technique’s estimation for
Cr2O3%...............................................................................................................123

xi
LIST OF UNIT SYMBOLS

UNIT SYMBOL
centimetre cm
Chromium Cr
Chromium(III) oxide Cr2O3
Coefficient of determination R2
grams per cubic centimetre g/cm3
Iron Fe
Iron(II) oxide FeO
meter m
Million tonnes Mt
Million years Ma
More than >
percentage %
silicon dioxide SiO2
TM
Trademark
Registered trademark ®

xii
LIST OF ABBREVIATIONS

ABBREVIATION DEFENITION
2D Two dimensional
3D Three dimensional
ASX Australian Securities Exchange
B.L.U.E Best Linear Unbiased Estimation
BC Bushveld Complex
BV Block Variance
CAD Computer-aided design and drafting
CDF Cumulative Distribution Function
COV Coefficient of variation
CP Competent Person
CR:FE Chromium to Iron ratio
CVV cross-validation variance
CW Channel Width
DHS Drill Hole Spacing
DIFF Difference
EDA Exploratory Data Analysis
GPS Global Positioning System
ICP-OES Inductively coupled plasma - optical emission spectrometry
IDW Inverse Distance Weighting
IDW2 Inverse Distance Weighting to the power 2
IDW4 Inverse Distance Weighting to the power 4
IND Indicated
INF Inferred
IRUPs Iron-rich ultramafic pegmatites
ISO International Organization for Standardization
JORC Joint Ore Reserves Committee
JSE Johannesburg Stock Exchange
KE Kriging Efficiency
KV Kriging Variance
LG1 Lower Group one
LG2 Lower Group two
LG6 Lower Group six
LG7 Lower Group seven

xiii
ABBREVIATION DEFENITION
MG1 Middle Group one
MG2 Middle Group two
MG3 Middle Group three
MG4 Middle Group four
MINTEK Council for Mineral Technology
MRE Mineral Resource estimation
N North
N/A Not applicable
NNW-SSE North North West - South South East
NR Neighbourhood restrictions
NZX New Zealand's Exchange
MAX Maximum
MIN Minimum
MEAS Measured
OK Ordinary Kriging
PGE Platinum Group Elements
PGM Platinum Group Metals
QA/QC Quality assurance and Quality control
QQ Quantile-Quantile
RBF Radial Basis Function
RC Reverse Circulation
RCI Resource Classification Index
RLS Rustenburg Layered Suite
SABS South African Bureau of Standards
The South African Code for the Reporting of Exploration
SAMREC Results, Mineral Resources and Mineral Reserves
SANAS South African National Accreditation System
SARM South-African Reference Materials
SG Specific Gravity
SOR Slope of Regression
STANDEV Standard Deviation
UG1 Upper Group one
UG2 Upper Group two

xiv
1 INTRODUCTION

The data used in this research study is considered confidential and reporting of
results is drafted in such a manner that the details and origin of the information
remains confidential. The company’s confidentially is respected by not disclosing
the specific location, or specific project description throughout this research report
the abbreviation “MCo” refers to “the Mining Company”. Furthermore, any internal
documentation, reporting or research information provided by MCo and used in
this research study is referenced in such a manner to not breach the confidentiality
agreement. The intellectual property on which the research study was conducted
is retained by MCo.

MCo agreed to make geological and borehole data available in order to compare
the mine’s estimation and classification practices with other methods using the
same data set, on condition that the confidentially agreement is respected. This
permission implied that apart from data access, information on practices had to be
accessed through interaction with relevant stakeholders at the mine

1.1 Project Background

The research project was conducted on one of MCo’s underground chrome mines
that extracts chromitite ore, for producing beneficiated chrome, including ferro-
chrome for the local market and international export markets. The mining right for
mineral extraction is owned by MCo. Again respecting the confidentiality a Google
map image, indicating the general area in which the MCo project is located in the
North-West province, east of Rustenburg and west of Pretoria appears in Figure 1.

The project is situated in the western Bushveld Complex (BC) and is underlain by
the Rustenburg Layered Suite (RLS), which hosts the Middle Group chromitite
seams that are of interest for chromitite ore extraction. The two main chromitite
seams mined underground for economic value are the Middle Group one (MG1)
and Middle Group two (MG2) seams. Underground mining of the MG1 and MG2
chromitite seams are predominantly by means of hybrid-conventional mining for
the MG1 and bord and pillar mining for the MG2. The chromitite seams are
generally homogeneous in terms of mineralisation and grade, with the MG1 seam,
generally higher in grade.

1
Figure 1: Satellite image: general area of the research project indicated by the red
circle (Google Maps , 2020).

1.2 Estimation and Classification Methodologies in Place at MCo

MCo’s Mineral Resource estimation and classification practices at the time of this
research study stems from years of practical experience and empirical knowledge
of the chromitite resources in the Bushveld Complex(BC) gathered and defined by
geology and mining professionals from exploration to resource estimation through
to life of mine planning.

Estimation
The Growth technique is a proprietary algorithm within Geovia Minex™ software;
currently the software package of choice for Mineral Resource estimation and
classification at MCo. This estimation method is based on a two dimensional (2D)
gridding algorithm, that calculates best-fitting surfaces for stratiform deposits,
considering regional trends while honouring drillhole data. The technique is
applied to the Cr2O3%, SiO2%, FeO%, Cr:Fe ratio, Specific Gravity (g/cm3, and
MG1 Seam Thickness(m) variables for MG1 resource estimation.

Classification
The current Mineral Resource classification methodology, based on prescribed
empirical borehole radii for the delineation of the chromitite resource categories
relates to the borehole grid spacing. As the drilling grid expands, the radius from
seam intersections of verified drilled borehole locations increases and

2
consequently the resource confidence decreases. For a Measured Resource, a
150m radius is typical, a 300m radius for an Indicated Resource and a 600m radius
for an Inferred Resource. The Competent Person can manually adjust the resource
confidence categories for closer borehole spacing based on his/her own discretion.

Comparative Studies
The methodologies for resource estimation and classification have evolved within
the mining industry with advancement of the technological age, advanced software
applications allowed for the application of spatial estimation techniques as
opposed to classical approaches. Although MCo recognised this development and
have adopted modern estimation and classification methods, technical
comparative studies confirming the current resource estimation and classification
methodology have not been carried out.

1.3 Research Project Objective

The intension of MCo comply with the requirements of the SAMREC(2016) and
JORC(2012) reporting codes for Exploration Results, Mineral Resources and
Mineral Reserves, specifically that the Competent Person (CP) should be able to
defend their choice of estimation and classification methodologies employed in the
Mineral Resource declaration of a company.

The objective of this project research is to address the MCo’s aim of adherence to
the reporting codes and to provide the CP with defendable researched Mineral
Resource estimation and classification guidelines and recommendations based on
acceptable industry practices. Furthermore, the study will assess the
appropriateness of MCo’s Mineral Resource estimation and classification
approach, in comparison to alternative estimation methods (IDW & OK), and
classification methods based on geostatistical studies, by using the same data set.
The “same data set” refers to the data that was used in the most recent Mineral
Resource and classification, 2017 end year reporting of MCo.

The goal is to either confirm the appropriateness of current resource estimation

and classification approaches or to propose different methodologies based on
researched practices of Mineral Resource estimation an classification. The results
of the study could also assist other chrome mines within the BC in approaches
towards Mineral Resource estimation and classification.

3
1.4 Problem Statement

Firstly, the MCo currently uses the Growth technique for Mineral Resource
estimation, a commonly applied method in the coal industry. The appropriateness
of this estimation technique has not been established through comparative studies
of other estimation methods (e.g. IDW or OK), in the project area being considered.

Secondly, the borehole radii mineral resource classification approach (similar to

the coal industry), has been entrenched as a standard for several years, and have
also not been compared to other alternative methods (e.g. geostatistical methods),
used in the mining industry.

1.5 Research Hypothesis and Assumptions

Applying statistical and geostatistical techniques and analyses, as opposed to the

current methodology of a growth algorithm for estimation and the application of
fixed radii for resource classification categories, can improve the resource
estimation confidence. This is expected to have a positive impact on the MG1
resource classification. This does not necessarily mean that there will be more
resources in the higher confidence categories, but rather that the confidence in
estimates within all the resource categories will be improved in terms of confidence
measurement or determination.

1.6 Research Motivation

A general lack of published research regarding the estimation and classification of

chromitite resources, established by the detailed literature search and review of
estimation and classification methodologies in the mining industry was a significant
driving force in the motivation for the investigations carried out by this research.

The value of the research lies in a benefit to the CP, compiling or reviewing the
MG1 chromitite resource statement for the mine, who will be able to muster support
for their choices of resource estimation and classification practices based on the
outcomes of the research.

1.7 Research Questions and Methodology

The methodology followed throughout this research study is based on a

quantitative approach, in conjunction with industry research, including internal
documentation supplied by MCo. Due to the confidentiality agreement with the

4
company, references of any internal literature, reports or information used, is not
provided in this research study. The research report is structured to incorporate a
review of the Mineral Resource estimation and classification of the MG1 chromitite
seam, by using comparisons of known industry-standard estimation and
classification methods, with the current methods employed by MCo. The
comparative results were not statistically measured for significant differences and
was rather compared with logical conclusions.

The key questions this research aims to answer are as follows:

• Irrespective of the use of geostatistical parameters for resource estimation,

are other factors such as geological parameters influencing the confidence in
the resource estimation?

• Is the Growth estimation methodology as used MCo applicable for the MG1
orebody and can it provide measurable or quantifiable estimation confidence?

• Is there merit in applying alternative estimation methods such as Ordinary

Kriging (OK) and IDW, to the relevant variables (grade, seam width, and
Relative Density), i.e. do these methods provide estimates that are closer to
reality? Is there an underlying spatial variability in terms of grade, seam width
and SG in the MG1 seam?

• Are the predetermined radii sizes for resource estimation applicable for the
declaring the current resource confidence associated with the orebody?

• How do statistical, geostatistical, geological and (or) scorecard approaches

compare to the current resource classification practise in place at MCo?

The methodology followed to answer the research questions and address

the objectives of the study is listed below:

• Conduct a literature review and assessment of the current Mineral Resource

estimation techniques, and classification practices applied to the MG1
chromitite seam at the mine.

• Assess the geological setting (regional & local) and the main geological factors
that influence the resource estimation confidence conducted at the MCo. This
was achieved by a literature review of the BC geology and supported by a
numerical indicator modelling conducted in Leapfrog® Geo software.

5
• Investigate via an Exploratory Data Analysis(EDA) the underlying statistical
distribution of the relevant variables used for estimating the MG1 resource,
which is a crucial factor of determining an applicable Mineral Resource
estimation technique.

• Analyse spatial variability and continuity of the variables used for Mineral
Resource estimation based on a literature review of variography including the
calculation and modelling of variograms, also apply Leapfrog® Geo numeric
modelling (Isoshells) to enhance this analysis.

• Carry out comparative statistical and geostatistical analyses of the Growth

estimation technique, including the classification methodology used by MCo.
In principle, comparisons was also conducted by using a Mineral Resource
classification score card approach.

1.8 Research Project Outline

This research project consist of 10 chapters, followed by the Reference list and an
Appendix.

Chapter 1 offers an introductory background to the research project. Commencing

with the limiting confidentiality aspect of the project and how it is addressed.
Followed by a summary of current estimation and classification practices of MCo.
Proceeding with the research objective namely to assess MCo’s Mineral Resource
estimation and classification practices and to either confirm or provide
recommendations based on acceptable industry practice whilst adhering to
reporting codes. The two fold problem is stated, the research hypothesis and
assumptions are presented. The research is motivated by the value add of the
outcome and benefit expected for the CP generating the Mineral Resource
declaration for MCo. Finally, the research questions and methodology are
discussed.

Chapter 2 provides an holistic overview of the Mineral Resource estimation and

classification practices within the mining industry. The importance of reviewing the
appropriateness of these practices is highlighted as they can have an major impact
on the efficiency, productivity and financial wellbeing of a mine.

In Chapter 3 the Geological setting and the MCo’s data regime is reviewed,
background of the data used within this study is also discussed. The MCo’s

6
Geological modelling procedures are reviewed, in accordance to Mineral Resource
estimation and classification practices. The 3D Geological modelling methodology
conducted in Leapfrog® Geo by the researcher, is presented in comparison with
MCo’s current geological structural model.

Chapter 4 elaborates on the Exploratory Data Analysis conducted to gain

understanding of the underlying grade distribution and continuity of the relevant
variables used in the Mineral Resource estimation and classification at MCo.
Analysis and investigation on domaining, stationary and influences of geological
complexities on the data conducted are argued.

Chapter 5 provides a high level literature overview of variography, the semi-

variogram model, and the spatial analysis conducted for this research, with
concluding results. The “two thirds” variogram range of influence classification
method, is also compared with MCo’s current Mineral resource borehole radii
criteria.

Chapter 6 describes the block modelling and estimation process followed, and the
determination of a search neighbourhood using a quantified Kriging neighbourhood
analysis. The results of applying alternative estimation techniques (IDW & OK) in
comparison to the Growth technique are also discussed.

Chapter 7 describes the main influencing factors affecting the MG1 Resource
classification practices currently employed by MCo.

Chapter 8 summarises the Mineral Resource classification comparison results,

conducted by using geostatistical parameters.

Chapter 9 concludes the research findings based of comparisons conducted on

MCo’s current Mineral Resource estimation and classification practices.
Furthermore, the advances and disadvantages of the Growth technique are
discussed, and the findings of the research questions of this research study are
also addressed.

Chapter 10 summarises the recommendations based upon the conclusions and

outcomes of this research project. Furthermore, guidelines to improve the current
Mineral Resource estimation and classification practices of MCo are also
discussed.

7
2 LITERATURE REVIEW

The literature review of this study is focused on providing an holistic overview of

the Mineral resource estimation and classification practises within the mining
industry. Furthermore, an overview of the estimation techniques that were
compared for this study is also described.

2.1 Resource Estimation, Classification and Reporting

Resource estimation is critical within a mining company and used to determine

performance, describe risk and allow informed decisions to be made (Snowden,
2001). The generation of Ore Reserves is underlain by the estimation of Mineral
Resources. It is therefore critical for resource estimations to be reliable to provide
a level of confidence in day-to-day operations and in feasibility studies. Geologists
and Engineers that conduct these estimates should be mindful of the errors and
uncertainty associated with estimations (Dominy et al, 2002).

The most critical goal of resource estimation is to predict future grade and tonnage
of resources that may be mined, in other words, the final estimation for the purpose
of distinguishing between ore and waste. Mining companies must live up to
production expectations and deliver adequate returns to investors who provide
funding. Therefore, the emphasis is on local precision and exactitudes of
estimates, including the realistic expected value that could be recovered (Deutsch
& Rossi, 2014).

Noppé (2014), explains that mining is inherently a risky type of business from
technical, environmental, social, including uncertainties economically with regards
to an exploration prospect to a feasible project advancing to a developed mine,
that is associated with operating, market and safety risks. The risk in a project can
better inform internal and external stakeholders with improved transparency,
consistency and balanced interpretations of technical confidence. Therefore,
minimum standards, recommendations, and guidelines for public reporting of
Exploration Results, Mineral Resources and Mineral Reserves are specified by
International reporting codes.

Although the Reporting codes provide guidelines on more specific matters in terms
of estimation and classification, the guidelines remain non-prescriptive and rely on
the judgement of the Competent Person (CP) that reports the estimates (Njowa,
2008).

8
The SAMREC and JORC codes have been included in the listing rules of securities
exchanges (ASX, JSE and NZX), that are associated with specific requirements
for exploration and mining companies that report to the relevant exchanges
(Noppé, 2014).

Furthermore the reporting codes (SAMREC and JORC), make recommendations,

provide guidelines and set minimum standards and criteria for Mineral Resource
estimation and classification.

The researcher, therefore, believes that it is critical to consider SAMREC 2016,

specifically Clause 24 covering the Reporting of Mineral Resources and Table 1
Section 4, which emphasises the geological model interpretation, the nature and
appropriateness of the resource estimation method and resource classification
technique.

MCo aims to comply with the SAMREC reporting Code. Figure 2 below is the
SAMREC (2016) framework for the classification of tonnage and grade estimates,
reflecting the geoscientific confidence and associated factors of technical and
economic evaluation.

Figure 2: Linkages between Exploration results, Mineral Resources and Mineral

Reserves (SAMREC, 2016).

9
Mineral Resource reporting Codes, such as the SAMREC code, do not describe
how Mineral Resources should be classified, and only requires the CP to describe
the criteria and methods that were used in order to classify Mineral Resources into
different categories of confidence. However, the criteria and methods used, should
adhere to the guidelines per definition of the different classification categories
(Measured, Indicated and Inferred), on the basis of data quality, geological and
grade continuity.

This still leads to subjective decisions of classifying a Mineral Resource and would
depend solely on the criteria compiled by the Competent Person and (or) mining
company, hence the reason for proper documentation and justification required
(Emery, et al., 2004). The underlying difficulty of quantifying the extent of
confidence associated with a Mineral Resource estimate has resulted in various
and different methods, terminology, and definitions, with mining companies drafting
their own independent classification standards (de Souza, et al., 2010).

De-Vitry (2003) points out a valid reason for this in stating that:
“ … no prescribed criteria and rules will work for all situations or even
between different ore types within the same deposit”.

2.2 Factors Impacting Mineral Resource Estimation

There are many variables and disciplines influencing the precision of estimation all
of which are important to achieving accurate resource estimation results. This
includes, but is not limited to mining, metallurgy, geostatistics and geology
mineralisation prior to resource estimation and realistic geological interpretation,
including block modelling (Snowden, 2001).

Dominy et al (2002) state that reliable information and high-quality data is critical
to sound investment and operational decisions, they also highlight the importance
of the transparency of the estimator with regards to the inherent risks of the
estimation.

Micon International (2016) highlights four main factors that have an influence on
resource and reserve estimation and are causes of mine failure. These are
identified as Geology in the estimation; Orebody characterisation; Top cutting of
outlier assays and grade interpolation. Factors also affecting estimation, are poor
data quality used to define the Mineral Resource, difficulties with regulating the

10
amount and density of exploration data that is used to estimate the Mineral
Resources and define the Mineral Resource confidence categories.

The inability of mine production to reconcile with Mineral Reserve estimates in

terms of tonnage and grade are often associated with mine failure. Errors with
regards to these measures are unavoidable. However, by applying appropriate
estimation procedures, the expectancy of estimates produced by professionals
could be within 10 percent of reality (Sinclair & Blackwell, 2002).

In summary, the literature review of the above three authors brings focus to the
importance of understanding the reliability of and confidence in all elements playing
a role in Mineral Resource assessment. Lacking this insight and appreciation could
lead to compromised Mineral Resource estimation and classification, including
consequential inaccurate mine planning, estimation of operational costs and
capital expenditure, leading to compromised Mineral Reserve estimates.

2.3 Resource Estimation Methodologies

One of the fundamental decisions during the process of resource estimation is the
selection of the appropriate estimation method. Inappropriate estimation methods
could result in errors of ±50% associated with the estimate. The estimation method
of choice should be based on the geology, grade distribution complexity and
degree of high-grade outlying values (Dominy et al, 2002). Isaaks & Srivastava
(1989) emphasise- that no single estimation method is appropriate for all Mineral
Resources.

Furthermore, it is important to understand the capabilities associated with the

method that is used. As an example, polygonal estimation methods contain one
fixed biased answer and are more suitable for producing volume-weighted global
mean grade estimates. The IDW technique is an unbiased estimator; it is a purely
geometric technique and it is impossible to determine an estimation error variance
as estimates only depend on the distance between samples and point to be
estimated.

A geostatistical technique, namely kriging, provides the best linear estimate

possible with obtainable data (Dohm,2015), and minimises the error variance of
the estimation (Matheron,1963).

The appropriateness of an estimation technique is related to estimation goal

Deutsch & Rossi (2014) mention for example if the goal is simplicity and

11
reproducibility, then IDW estimation may be suitable. Should the goal be to reveal
large scale geologic trends then both block kriging and IDW could be appropriate.

Considering that the goal determines the resource estimation technique, a review
of the Growth method, Ordinary Kriging and Inverse Distance Weighting (IDW)
techniques, which are in general applicable to chromitite resource estimation,
follows.

2.3.1 Growth technique

The Growth technique is considered as a general-purpose gridding method

available in Geovia MinexTM software, which was designed for the purpose of
geology and mine planning solutions for coal, including other stratified orebodies
(Dassault Systèmes, n.d.).

Dassault Systèmes GEOVIA Inc. (2016) describes the Growth technique as

“… a proprietary 2D gridding algorithm that calculates the best-fitting
surfaces for strataform deposits, taking into account the regional trends
while honouring the drillhole data, given the appropriate gridding
parameters. The growth technique is suited to modelling borehole data,
contour strings or spot heights and can handle features such as faults,
trend lines (ridges) or discontinuity lines (creeks, gullies).”

The Growth modelling technique is predominantly used in conjunction with gridded

seam models. Irregular spaced data configurations are used to estimate attributes
at grid values that are regularly spaced. The gridded model represents a deposit
as a succession of layers and is appropriate for layered ore bodies, for example
coal; bauxite; laterites and phosphate deposits (Surpac Minex Group, 2005).

Surpac Minex Group (2005) cites Batcha & Reese (1964); Crain (1970); Jones et
al (1986), and states that the Growth technique occurs as a two-staged process.
The first stage consists of surrounding the known data by four mesh points (nodes)
on a grid. The second stage grows away from the initial four mesh points and
estimate the remaining grid nodes by growing the mesh outwards. As the process
continues the original points are ignored and only estimated points are used to fill
the mesh points. A few passes are required until all the grid points have estimated
values.

12
The Growth technique is predominately used for modelling surfaces of coal seams,
including estimation of coal quality, in the mining industry.

2.3.2 Alternative Estimation Techniques

Sinclair and Blackwell (2002) call attention to the fact that in any given situation
where alternative estimation methods deliver different numeric grade results, then
these results can generally be used for comparative purposes.

A brief review of the alternative estimation techniques commonly used within the
Chrome industry namely Ordinary Kriging (OK) and Inverse Distance Weighting to
the power 2 (IDW2) are therefore described in the sub-sections that follow. These
estimation techniques were also used to compare MCo’s current estimation
method.

2.3.2.1 Ordinary Kriging

Deutsch et al (2014) state the following:

“… early computerised OK techniques were developed by Krige, Sichel
and others in South Africa. Matheron formalised the methods for
mining resource and reserve estimation and named it kriging after the
South African pioneer Danie Krige”.

OK is a geostatistical interpolation method that delivers local estimations. It is

known as a “best linear unbiased estimator” (B.L.U.E). The estimates are “linear”,
as OK aims to estimate weighted linear combinations; attempts to equalise the
mean error to zero, being “unbiased”; and referred to as “best”, as OK tries to
minimise the variance, associated with the estimation errors (Isaaks & Srivastava,
1989).

OK makes use of a weighted average approach with regards to neighbouring

samples to interpolate an “unknown” value at a specific location. The weights are
improved by the use of the semi-variogram model, describing the inter-relationship
of known and unknown values, including the location of sample values.
Furthermore, the OK technique provides a “standard error”, and has the option to
measure levels of confidence in the estimate, (Geostokos (Ecosse) Limited, n.d.).

13
2.3.2.2 Inverse Distance Weighting

Hengl (2007) cites Shepard (1968) stating that the IDW interpolation method is
possibly one of the oldest techniques in spatial prediction.

Inverse distance methods are closely related to weighted average methods. The
method is grounded on the calculation of weights for samples related to the
distance from the samples to a specific point or block to be estimated. Weights are
calculated according to the inverse of the distance between a known sample value
and estimation point. To ensure that the estimate is globally unbiased, the sum of
the weights are standardized to one. The weights decline based on the inverse a
specific power (exponent) of distance. The Inverse Distance squared (1/d2)
technique is typically employed with attributes associated with uniform variation,
for example, coal beds, in-situ bulk density values and strata-bound deposits
(Deutsch & Rossi, 2014).

The IDW technique (1/d, 1/d2, 1/d2.7, 1/d3, etc.) is one of the traditional and most
widely used Mineral Resource estimation methods, that considers point samples
on which properties such as grade is estimated on a regular grid. This method is
often critiqued for being subjective, however it remains popular as it produces
results relatively close to geostatistical estimates, which has a rational theoretical
basis for mine valuation (Sinclair & Blackwell, 2002).

One of the major shortfalls of the IDW technique, is that it cannot provide a
measure of accuracy or precision (Myers, 1994), as opposed to geostatistical
estimation methods. Gentile, et al. (2012) highlight pros and cons of the IDW
technique which are presented in Table 1.

Table 1: Pros and Cons of the IDW technique (Gentile, et al., 2012).

Pros Simple and intuitive

Computationally fast
Cons The choice of the interpolation parameters are empirical
The interpolation is always exact – no smoothing
Sensitivity to outliers and sampling configuration (clustered and isolated
points)
No estimation of interpolation errors (deterministic)

The most commonly used IDW power functions for Resource estimation to the
second or third power, the choice is somewhat subjective (Glacken & Snowden,
2001).

14
Babak & Deutsch (2008) and Weber & Englund (1992) discuss several reasons
why the IDW technique may be preferred over Kriging techniques and found that
in several comparative studies the IDW estimates performed better. Although the
IDW method is considered as an unbiased method, it does not account for spatial
correlation (variography) and does not attempt to minimise the estimation variance,
that Kriging does.

Another issue in using IDW is the decision of search neighbourhood parameters,

however if used in conjunction with variography and a QKNA process, it can be
resolved, as used in this study.

2.4 Resource Classification Review

The final task after resource estimation is the resource classification as it is

required to report the level of confidence associated with the Mineral Resource
(Glacken, & Snowden, 2001). The classification describes whether the Mineral
Resources have enough confidence to be considered assets (Deutsch & Rossi,
2014).

The JORC code aims to guide Mineral Resource classification through guidelines
such as the requirement for a Competent Person to base the classification results
on as many objective factors as possible (Glacken & Snowden, 2001). Resource
classification needs to be transparent and disclose all material issues for the
benefit of all stakeholders (Snowden, 2001).

These requirements ultimately aim to address the subjective debate with regards
to classification resulting in misclassification. Subjectivity is also influenced by the
lack of an easy and established method to measure the confidence of varying
classification approaches (Arik, 2002).

Furthermore, there seem to be unreasonable expectations of the resource model,

due to an industry trend of using statistical descriptions of uncertainty, to
supplement traditional resource classification criteria (Deutsch & Rossi, 2014).

2.4.1 Classification Practices

Some common Mineral Resource classification practices include utilising drill holes
and samples near each block, the kriging variance (KV) method which provides an
index of the data configuration and the use of radii to estimate blocks (Deutsch &
Rossi, 2014).

15
The KV is known as a useful criterion and an objective measure for resource
classification, geostatistical confidence with respect to the data configuration and
it is a good indicator of overall sample spacing due to dependency on the
arrangement and continuity of samples around the estimated block (Glacken &
Snowden, 2001).

The KV classification method makes use of the cumulative probability (CDF) or the
histogram of the kriging variance to assign resource confidence categories to the
resource (Arik, 2002).

Silva & Boisvert (2014) proposed a new classification technique, calculating the
cross-validation variance (CVV) during block kriging, and eliminating drillholes (one
or more) that contains the highest weights and classify the block by using the
resultant kriging variance.

Snowden (2001) specifies a comprehensive list of traditional and geostatistical

Mineral Resource classification assessments or measures with related advantages
and disadvantages. The list below only summarises some of the assessments of
confidence using different aspects:

• The semi-variogram: measuring the anisotropy of the mineralisation, and

assessment of confidence can be conducted with regards to drill hole
spacing in relation to the range of influence.
• Number of samples per block: classification based on the number of samples
within a search ellipse of a specified block.
• Distance: is useful for deposits that are isotropic (same variability in different
directions), by using the average distance of samples from an estimated
block.
• Kriging efficiency (KE%) and Regression slope: these measures can be used
to investigate the appropriate block size during estimation, as well as to
regulate the confidence of block estimates. In the case of a perfect
estimation, the KE will be 100%, with the Regression slope being equal to 1.

Arik (2002) proposed the Resource Classification Index (RCI) that could be
effective for resource classification, and utilises a combined variance, to combine
various desirable classification measures into one.

Silva & Boisvert (2014) also highlight that many authors namely Wawruch and
Betzhold (2005), Dohm (2005), Dominy et al (2002), and Snowden (2001), suggest

16
using conditional simulation for resource classification, that presents an improved
approach for accessing uncertainty compared to the KV and other methods.
Deutsch et al (2006) however, recommend that conditional simulation should only
be used as a supporting tool and that the classification criteria should remain
geometric for final results.

Silva (2015) states that the most common Resource classification techniques used
are geometric techniques, and although there are various types of geometric
measures used the most popular are Drill hole spacing ( DHS) and Neighbourhood
restrictions (NR).The general principle of the DHS technique is the use of spacing
between drill holes in order to classify blocks close to a block under consideration.
The NR technique is based upon the distance of samples close to the classifying
blocks and constraints associated with the number, including the configuration of
samples within a search radius.

Deutsch, et al. (2006) suggests that geometric techniques (i.e. DHS and NR),
should be based on thresholds such as standard industry practises of a country;
geological setting; experience from comparable deposits: a calibration with
uncertainty measured by geostatistical calculations and skilled judgement of the
CP.

2.5 Reality of Mineral Resource Estimation in the Mining Industry

Glacken, & Snowden (2001) emphasise that future resource estimation will be
influenced by technology and more sophisticated estimation algorithms. Deutsch
& Rossi (2014) highlight that automation of many steps in the resource estimation
process means less trained professionals might be required, which could be an
advantage seeing there are already relatively few highly trained professionals for
resource estimation. Furthermore, some of the other benefits include repeatability
and transparency as well as real-time diverse data type collection, incorporation
and processing of resource modelling, (Deutsch & Rossi, 2014; Glacken &
Snowden, 2001).

Some disadvantages of the automation of resource estimation include unrealistic

geological models, possible mistakes and a false sense of confidence being
portrayed. These authors believe that professionals and software providers should
become better equipped to address these disadvantages as well as the uncertainty
associated with resource estimation, (Deutsch & Rossi, 2014).

17
3 GEOLOGY AND DATA COLLECTION

3.1 Regional Geology

There are three main mineral commodities that are mined from the layered mafic
rocks of the RLS, of which the most important is the Platinum Group Elements
(PGE), followed by chromitite and then vanadiferous magnetite. The RLS is
estimated to contain 80% of the worlds PGE reserves (Morrissey, 1988 cited in
Maier and Barnes,1999).

The mining area in this study is underlain by the mafic and ultramafic rocks of the
Bushveld Complex (BC). The saucer-shaped BC intruded the sedimentary rocks
of the Transvaal Supergroup, 2055–2060Ma, largely along an unconformity
between the Magaliesberg quartzite of the Pretoria Group and the overlying
Rooiberg felsites (Cheney and Twist (1991) cited in Hunt, 2006).

The BC hosts more than 80% of the world’s Chromium and Platinum Group
Elements (PGE) deposits (Crowson, 2001 cited in Cawthorn, 2010), and is
recognised as the largest layered intrusion on earth (Von Gruenewaldt,1977) cited
in Manyeruke (2006).

The BC is comprised of three suites of plutonic rocks, namely the mafic and
ultramafic Rustenburg Layered Suite(RLS), the Rashoop Granophyre suite and the
Lebowa Granite Suite (Von Gruenewaldt et al, 1985).

The RLS consists of an approximately 9km thick succession of layered mafic and
ultramafic rocks, that are exposed in 5 major limbs; the Eastern, Western, the far-
western limbs, the Northern (Potgietersrus) limb and the Bethal (Southern) limb
(Kinnaird et al, 2004). Figure 3 shows the locations of the limbs.

18
Figure 3: Simplified map of the Bushveld Complex showing the location of the
various limbs (Kinnaird et al, 2004).

The RLS can be subdivided into five major zones known as the Marginal, Lower,
Critical, Main and Upper Zones (Figure 3, after Kinnaird et al, 2002).

Layers of chromitite were deposited throughout the Critical zone, generally at the
base of the crystallisation cycles. Chromitite seams are divided into Lower, Middle
and Upper Groups. The Lower and Upper Group seams occur in the Lower and
Upper Critical zones respectively, with the Middle Group chromitite seams that are
in between the boundaries of the Lower and Upper Group chromitite of the Critical
Zone. Due to the chromitite layers occurring in a succession, the seams get named
according to their location, with numbers starting from the bottom up. The
Lowermost Group gets named LG1 followed by LG2 to LG7 in the Lower Group
(consisting of 7 layers), progressing to MG1, MG2, MG3 and MG4 (consisting 4
layers) in the Middle Group, and ends on two layers in the Upper Group UG1 and
UG2, followed by the Merensky reef (SRK consulting, 2008).

19
Figure 4: Typical Stratigraphic column of the Rustenburg Layered Suite with
general zone thicknesses; major chromitite layers of the Critical zone, with Cr2O3
and PGE compositional data (column on the right) based on Scoon and Teigler
(1994). (Kinnaird et al, 2002).

The chromitite layers predominantly mined for the extraction of chromite content is
LG6 and MG1 to MG4. Generally, the LG6 and MG1 are associated with the
highest chromite content being mined in the BC. The bottom layer of the sequence,
the Marginal zone, is associated with noritic rocks, overlain by the Lower Zone
(lithological units of dunites, harzburgites and pyroxenites), the Critical Zone (cyclic
units of chromitite, pyroxenites, norites, and anorthosite), the Main Zone (cyclic
units of norites and gabbronorites), and the Upper Zone (layers of magnetite,
anorthosite, troctolite, ferrogabro to diorites). (Kinnaird et al, 2002).

20
The Critical Zone only occurs in the Western and Eastern Limbs and its enrichment
in the platinum group elements (PGE), relative to the marginal rocks, culminates
in the formation of the world-famous Merensky Reef and UG2 reefs, producing
close to 60% of the World’s PGM supply. The Merensky Reef is a feldspathic
pyroxenite with a thin basal chromitite stringer which is associated with the highest
PGM contents. The UG2 comprises of a main chromitite layer, carrying most of the
mineralisation, followed by a poorly mineralised pegmatoidal pyroxenite footwall.
The Critical Zone is thus of utmost economic importance and contains chromitite
layers carrying between 50 and 85% chromite. The best Cr: Fe (Chrome to Iron
ratio) values are found in the lowest layers, which currently produce about 40% of
the World’s chromium (Saager, 2005).

3.2 Project Geology

The mine’s lease area is located within the Western Limb of the BC. The rocks
associated with the Lower and Upper Critical Zone of the Complex underlies the
project area, with chromitite seams interlayered with norite, anorthositic norite,
mottled anorthosite and pyroxenite. The Marginal zone is also present within the
project area and is associated with quartzite xenoliths. This represents the
transgressive relationship between the Rustenburg Layered Suite and its floor.
Distinct undulations such as the Spruitfontein and Kookfontein upfolds within the
quartzite sedimentary floor, subdivide the South-Western BC into the Brits,
Marikana, Rustenburg and Boshoek sectors which could be responsible for the
lateral variations in the stratigraphy, chromitite thicknesses including chemical
characteristics (CAE, 2013).

MCo chromitite deposits are situated in the Marikana section of the BC (Figure 5).

21
 Project

Figure 5: Floor folds in the Magaliesberg Quartzite Formation in the south-western

Bushveld Complex, west of the property (modified after Davey, 1992
cited in Dube, 2010).

3.1.2 Stratigraphy

On the project area, there are sub-outcrops of chromitite seams belonging to the
Middle Group chromitite layers within the RLS and are overlain by 0.5m to a few
meters of black turf. Figure 6 represents the general stratigraphy underlying the
project area.

22
Figure 6: General Stratigraphy underlying the project area (MCo, n.d.).

3.3 Data collection

The borehole database, that incorporates collar, survey, stratigraphy, lithology and
assay data was provided by MCo in the form of csv files, exported from Geobank
Micromine® data management software. The borehole database dates back to
1999. Prior to exporting and handing over the borehole data to the researcher,
MCo conducted both a data validation in Geobank and a seam validation in Minex

23
software. The resource estimation for chromitite grade is based on surface
exploration borehole data only.

The total number of boreholes within the database received from MCo is 196
surface boreholes (Figure 7) consisting of wireline diamond and percussion drilling,
including deflections. Note to comply with the confidentiality on the location of the
Project area coordinates are not shown in Figure 7.

North

Figure 7: Borehole collars from available surface drilling data provided by MCo.

The main chromitite variables under consideration are Cr2O3%, SiO2%, FeO%, Cr
to Fe ratio, Specific Gravity(g/cm3) and chrome seam width. Statistics were
calculated for all the variables. However, the objective of the research is to improve
confidence in the Mineral Resource estimate and classification, therefore, the
focus has been on the economic driver (Cr2O3%), Specific Gravity(g/cm3) and
seam width(m) of the MG1 chromitite being the relevant variables for the resource
grade and tonnage estimates.

Additional data such as, validated underground survey pegs; top contact MG1 off-
sets by means of survey equipment, and interpreted geological structural plans,
were also provided by MCo.

24
3.3.1 Drilling Techniques

The Borehole database generally comprises of wireline diamond drilling of BQ size

(36mm diameter) mother holes and single deflections of TWB size (45mm
diameter) core. The drilling of deflections was initiated from 2008 and comprise of
approximately 24% of the diamond drilling that was conducted. Deflections were
drilled for additional assay confirmations, including geotechnical and metallurgical
test work. Reverse circulation (RC) drilling was also conducted in 2000, with 52
boreholes drilled in the opencast area of the property, however the drilling
information was only used for orebody and structural delineation, and not deemed
representative for Resource estimation by MCo.

Contracts with drilling contractors specifically stipulate more than 95% core
recovery and not less than 99% within the economic mineralised zones.

3.3.2 Collar Positions

Final borehole collar positions were conducted with calibrated survey

instrumentation, with a total station and differential GPS. The measurements were
configured within the geomatic software of the instrumentation to automatically
convert all data from WGS84 to LO27 coordinate system.

3.3.3 Down Hole Surveys

Down-the-hole surveys were conducted for approximately 31% of the Boreholes,

which excludes RC drilling. The average deviation of the surveys is less than two
degrees and a maximum of less than four degrees from vertical, that is considered,
acceptable.

3.3.4 Geological Logging

MCo has standard logging codes in place for capturing data into a Geoscientific
Information Management system, Geobank Micromine® software. However, a
formal logging procedure is still within a final drafting phase.

Detailed geological information is captured and the chromitite mineralisation of

economic importance is captured as “Economic Zones” (EC Zone) that is used for
Resource estimation. The level of logging that is conducted includes lithologies,
mineralisation, alterations, core bedding angles and geotechnical data if available.
According to records within the database, about 50% of the borehole core from

25
1999 to 2006 drilling is in core storage. Drilled, and hardcopy logging documents
for the drillhole database is safely stored for cross-reference if needed. Borehole
core from 2006 to the most recent exploration program, is all (100%) in storage,
with supporting hard copy documentation available.

3.3.5 Sampling

According to internal company reports of 2003, the sampling protocol that was
followed on historical boreholes(mother holes) of the MG1 chromitite seam was to
split the seam into two individual samples on the internal chromitiferous
disseminated pyroxenite layer, and further split into half-core for assaying. A
diamond rotary saw is used to cut and split the core. The MG1 chromitite layer is
separated from adjacent waste rock by cutting it at the top and bottom contacts.
Consistently one half of the core was sampled for assay analysis, and the other
half retained in the core tray and allocated the same sample number as on the
sample tickets, included in the plastic bags used for the assay sample. The protocol
specifies that internal waste or partings more than 4-5cm in thickness would have
to be sampled individually for assay analysis.

Deflections are sampled with a similar procedure, however if there were no

deflection drilled, the mother hole chromitite intersection would have been sampled
as a quarter core. Deflections would typically be used for friability analysis; two half
core samples or if requested full core samples will be analysed. The same
sampling protocol has been followed up to the latest exploration drilling projects.

The bagged samples are placed in a larger sample bag which contains all samples
for one borehole. Each sample bag is adequately tagged with a sample
identification number, as well as the larger sample bag containing all the samples.
Subsequently, a laboratory request form for analysis is completed, upon delivery
at the laboratory and cross-reference checks are conducted together with the
laboratory Superintendent, prior to signing acceptance.

3.3.5.1 Sample Preparation and Analysis

MCo make use of an on-site laboratory that prepare and analyse all sampling that
is obtained from exploration projects or underground. The Laboratory adheres to
all the requirements of the South African Bureau of Standards (SABS) Standard
ISO 17025: 1999 specifications in order to maintain international standard
operating procedure and to retain accreditation. Furthermore, the company’s

26
laboratory is also accredited by The South African National Accreditation System
(SANAS).

Inductively Coupled Plasma Optical Emission Spectroscopy (ICP-OES) is the

analysis method that is used for determining percentages of Cr2O3, SiO2 and FeO,
of which an additional wet chemical analysis is conducted for Cr2O3.The Cr:Fe
(ratio) is determined from the Cr2O3% and FeO% analysis.

The sample preparation and analysis is comprehensively documented in the June

2004 Mineral resource and Reserve report from the company, which complies to
the SAMREC code.

3.3.6 Bulk Density

Lipton and Horton (2014) mention three fundamental inputs to a Mineral Resource
estimate, namely grade, volume and bulk density. The importance of density stems
from the fact that tonnage is estimated as a product of volume and density.
Furthermore, grade estimates usually get weighted by tonnage, for reporting,
which has a direct impact on the validity of grade.

Makhuvha et al (2014) also highlight that density have an impact regards to

operational factors, such as mine planning; mine design; equipment selection and
operational performance. The authors also point out essential procedures for
mining and exploration projects such as selecting the appropriate method for
density determination; comparing two or more techniques and quality control
measures, to reduce risk and improve operational performance.

MCo uses Specific Gravity (SG) as a measure for density that is used for Mineral
resource estimation, to report tonnage. The SG measurements are conducted at
the company’s laboratory and are reported together with the chemical analysis
conducted for grade elements on the certificate of analysis. Records of the
procedure for the specific method used for SG determination could not be
obtained, however internal documentation from 2004, stated that an Archimedes
method was used.

3.3.7 Database Integrity, Quality and Validation

Several authors (Assibey-Bonsu, 2016), (Chanderman, 2015), (Duggan, et al.,

2017) and (Snowden, 2001) emphasise the critical importance of database
integrity, quality and validation throughout the initial stages of Mineral resource

27
estimation and classification. The integrity of a Mineral Resource estimation
database is directly proportional to the reliability and confidence of the estimation
results. Errors are considered to accumulate, due to the fact that geological
interpretations and Mineral Resource estimations are grounded on the initial
sampling database (Minnitt, 2014 cited in (Chanderman, 2015)).

The database integrity of MCo is upheld by ensuring that external and internal
audits are frequently conducted on the drill hole database, procedures, and
random checks on geological logging compared to physical observations of the
borehole core. The company employs competent trained staff to ensure the quality
of the data obtained is maintained, and by having transparent guidelines instated
within the geological department.

Although the data received from MCo was validated using the Geobank®
(database management software) and Geovia MinexTM, the author of this research
study re-checked and validated the database prior to Mineral Resource estimation
and throughout the geological modelling process.

The researcher carried out additional validation by carrying out cross-reference

checks and due diligence was conducted on the following:

• Zero values, missing, duplicated or erroneous collar coordinates

• Reconciliation of collars and the surface topography

• Overlaps, gaps, duplicates, negative values of borehole intervals, including

sampling, especially the mineralised zone (MG1)

• Outlier, missing, zero values or anomalous assay results within the

mineralisation zone (MG1) based preliminary histograms, and normal
probability plots

• Visual inspections with regards to the borehole data in relation to geological

structures that could have an influence on the mineralised intersections and
assay results

• Visual inspections of elevation difference with regards to MG1 intersections

The collars coordinates of the boreholes were deemed acceptable, however during
reconciliation with the surface topography two borehole collars were found to have
a more than 2m difference, that is outside the acceptable margin. Subsequent to

28
investigation, it was found that the topography available had survey data of a waste
rock dump that influenced the reconciliation, and the collars were accepted as is.

3.3.7.1 Quality Assurance and Quality Control (QA /QC)

The SAMREC code (2016) guidelines for Quality Assurance and Quality Control
QA/QC states that the Competent Person should be able to:
“… demonstrate that adequate field sampling process verification
techniques (QA/QC) have been applied, e.g. the level of duplicates,
blanks, reference material standards, process audits, analysis, etc.”

The company’s laboratory is accredited to ISO 9000, and conforms to a Quality

Assurance system. The company has well documented QA/QC procedures, as it
is standard practise at the laboratory.

However, no documentation on a QA/ QC protocol with regards to the field

sampling process, prior to sending samples for assay at the laboratory for historical
borehole core samples were supplied by MCo. More recent records do however
show a drive towards implementing control samples, such as blanks and standards
into the sample stream. Records were obtained for 12 borehole samples assayed
for Cr2O3% in 2017, where 18 blank and 12 standard samples were included in
the sample batches (Figure 8 and Figure 9). Blank samples comprise of pure
quartzite, and standards are certified reference material, SARM 146, prepared and
distributed by MINTEK.

Figure 8: Sample blank analysis for Cr2O3% with QA/QC thresholds.

29
Figure 9: SARM146 sample analysis for Cr2O3% with QA/QC thresholds.

Blank and standard sample analysis is used to determine contamination, and

accuracy respectively with regards to the analytical process of a laboratory.
Standard practice in the chrome industry allows for a one percent threshold in
terms of blank samples for Cr2O3% analysis. Standard samples or certified
reference material analysis usually represents a good reflection of the accuracy of
overall sample batch or batches, if results are within at least two standard
deviations from the certified mean. Standard samples are also used to test the
biasness of the analysis and good practice is for the results of the laboratory to be
within 5% difference of the certified mean value.

Based on the limited records found, it is evident that the laboratory reflects
acceptable accuracy and biasness towards the 12 boreholes that were assayed.
However, in terms of the contamination analysis, less than 80% of the blanks
passed the 1% Cr2O3 threshold. This indicates that the company should further
investigate to the origin and preparation of the quartzite material used as the
blanks, as this material is not certified. Further investigation with regards to the
laboratory procedures internally should also be investigated.

3.4 Geological Modelling

Chanderman, et al. (2017) considers a geological model of an orebody prior to

Mineral Resource Estimation (MRE) and Classification as a fundamental process,

30
as it acts as the limits in which estimates are defined. Thus, MRE and classification
fundamentally depend on the confidence of a model and a good understanding of
the geology of an orebody, as it impacts the interpretation and the analysis process
of conducting a Mineral Resource evaluation.

MCo generally uses a two-dimensional platform (CAD and contouring software),

and to some extent three-dimensional interpretations when updating an ongoing
2D Geological structural model, based on historical drilling information;
geophysical surveys; underground mappings; underground survey pegs and three-
dimensional MG1 seam contours generated from the data. The MG1 seam
contouring is created in Surfer Golden software® and (or) Geovia Minex™ to assist
with geological interpretations, and dip regime of the orebody. The 2D geological
structural model and MG1 contours is updated as new information is obtained
through underground drilling, annual exploration, including updated underground
mappings from mining advance.

The models that are used for Mineral Resource estimation (MRE) and classification
consists of a combination of the 2D geological structural model, to incorporate an
estimate of the geological loss, and a 3D gridded MG1 seam model that is
generated in Geovia Minex™, to report volume. Furthermore, the models are also
used for interpretation during resource classification, of structures that could have
an influence on the confidence of the classification. The economic zone, which
consists of MG1 borehole intersections, and relevant underground data of the MG1
seam, forms the basis of the 3D MG1 seam model.

The 3D gridded MG1 seam model consist of meshes, that are interpolated from
borehole and other relevant data, by using the Growth technique. A borehole seam
interpolation is used to insert seam intervals to complete the stratigraphic
sequence for each borehole before generating model seam roof and seam floor
grids (meshes). This ensures that there are no missing seams in the boreholes and
prevent the crossing of gridded meshes. The MG1 seam model, does however,
not take into account the geological structures when conducting interpolations.
This could have an effect on the dip regime interpretations, and possibly reporting
of the MG1 volume.

31
3.4.1 Geological model

For this research a 3D geological model of the MG1 roof (top contact) was
constructed in Leapfrog® Geo (Figure 10) based on current and historical data of
various sources listed below:
• Surveyed topography
• Existing 2D geological structural model of MCo
• 3D underground surveyed pegs and MG1 top contacts
• Validated MG1 borehole intersections
Leapfrog® uses a Fast-Radial Basis Function (RBF)TM algorithm (Spragg, 2013),
to implicitly create interpolated meshes of triangles from the input data, in 3D
space.

Figure 10: Orthogonal view of the interpreted MG1 top contact of the 3D geological
model.

The 3D geological model was constructed to assist with the following:

• Confirmation and review of the current structural interpretation of MCo
• Orebody delineation and dip regime
• Influence of geological complexities that can affect grade continuity
• Geological loss and data influence associated with geological structures
• Conversion for MG1 drillhole seam intersection data, from apparent to true
thicknesses using the MG1 top wireframe

32
The mineralisation of the MG1 orebody was modelled within definite hard
boundaries of the exterior country rock, pyroxenite, that represent the footwall and
hanging wall units.

In carrying out the geological modelling phase it became clear that the strike of
the MG1 chromitite is generally east-west, dipping 150 to the north-northeast and
north-northwest. Whilst the seams are assumed to be planar localised undulations
do exist. Steep dipping depressions, known as pothole structures, occur in the
project area, these tend to decrease or excessively increase the thickness, and
locally increase the dip of the seam. The pothole occurrences are observed in both
the borehole core and underground mapping data incorporated into the 3D
geological model. Major known geological structures intersected and interpreted
within the project area are faults, dykes and pothole structures. MCo’s
interpretation of the geological structures generally correlate well with the 3D
geological model that was constructed (Figure 11).

Figure 11: 3D Geological model overlain by MCo’s major geological structural

interpretation (plan view).

33
3.4.2 Geological structures

The major geological structures present within the project area, that could have a
potential influence on geological and grade continuity are Faults, Dykes, Potholes
and Iron-rich ultramafic pegmatites (IRUPs) and are detailed in the sub-sections to
follow. Add “IRUPs” to the abbreviation list the you do not have to do it in the IRUP
section below.

Major Faults:

Major faults or fault zones within the project area can be described as fault
structures or zones that displace the chromitite seam(s) more than or equal to two
meters in elevation. Generally, the faults that have been intersected and
interpreted from geological modelling is normal (dip-slip) faults, with the occasional
reverse fault intersection. Associated with these faults are a number of smaller dip-
slip faults (fault zones) with throws less than 2m. The major faults within the project
area range from 2-6m displacements, with one ±22m downthrow fault to the east
of the property. The major strike direction of the faults is North West-South East
dipping North-East or South-West respectively.

The faults are commonly associated with shear zones, dense sympathetic jointing,
slickensides, alteration (clay minerals – serpentine, chlorite, talc) and water
presents (seepage/ discharge). These conditions and fault displacements
contribute to zones of geological loss within the MG1 chromitite seam.

Dykes:

The dykes interpreted from the geological modelling and intersected in

underground workings are later stage igneous intrusions that replace the chromitite
seams with a generally pyroxenitic (dolerite) rock mass, causing a geological loss
in terms of mining. Pilanesberg age dykes trending NNW-SSE was determined by
previous geophysical modelling and opencast mining on the project area. The
dykes generally trend in a similar direction as the faults and conjugate. Minor sills
associated with these dykes that intruded through weakness planes of the bottom
contact of the MG1 seam are evident from underground mapping.

The dykes “pinch” and “swell” along its length (dip and strike) and sometimes
bifurcates into thin branches, with thicknesses of ±1-20m. The dykes that were
intersected in underground workings are associated with minor to moderate
alteration, more severe alteration with the presence of water. Steep seam “drags”

34
resulting in reduced seam thicknesses of minimum (±0.50m) or maximum
increased thickness of (±2.0m) have been observed in underground workings that
are associated with the influences of dykes.

Potholes:

Other structural features that are exposed within the project area in the
underground workings are potholes, which are large “dish-or-pear“ shape
structures with diameters estimated between 30 to 80m and a depths varying
between 1m and 16m. Many Geologists generally describe potholes as
synmagmatic circular depressions or slumps (Hoffman, 2010). Furthermore
Hoffman (2010) highlights that
“… in Mineral Resource modelling, potholes may be characterized
either as known losses, where the data acquisition activity has allowed
reasonable definition or as unknown losses, where there is an
expectation that potholes will be present with some statistical
proportion”.

The main characteristics of potholes within the project area, observed in

underground workings and geological modelling, that impact economic extraction
of the ore are listed below:

• Associated poor rock mass conditions due to high density of jointing and
alteration of the pyroxenite host rock, that leads to uneconomical dilution
when mined with mechanized machinery. Furthermore, also associated
with safety risks that result in unfavourable mining conditions.
• Steep apparent and true dips of the MG1 seam, measured up to ±300.This
leads to off-seam mining, uneconomical ore:waste ratios and decreased
ore extraction.
• Steepening of the MG1 seam within areas close to pothole structures,
termed “seam rolls”, which can prevent feasible extraction of ore.
• Reduced seam widths (±0.30m), “splitting”, and “pinching” of the MG1
seam, resulting in large areas of total geological loss.

Internal company reports from 2004, mention that no regional trend has been
established regarding the occurrence of potholes. However, in 2017 the MCo’s
Geology team, attempted to improve the predictions by a applying the methodology
suggested by Hoffmann (2010), by including an analysis of strike/dip regime
contours.

35
Hoffmann (2010) and a Geology department report from 2017 concluded that the
size and local occurrence of potholes are highly unpredictable, and require
sufficient data from underground workings, exploration/underground drilling or
geophysical seismic surveys to accurately estimate/predict the influence on the
Mineral Resource.

Even using all the sources of relevant data in the geological model, there will be
large areas of the unmined Mineral Resource that will be associated with
uncertainty. Hence, conducting a thorough “pothole influence” analysis in
conjunctions with other methods, is crucial for estimating geological loss and
influence. It is also important to consider a level of uncertainty when declaring the
Mineral Resource in classification categories.

IRUPs :

Cawthorn, et al.(2018) consider IRUPs, as rocks primarily consisting of

clinopyroxene olivine and magnetite, that intrude/cut layered successions
irregularly, and are usually associated with sharp contacts. Due to their magnetic
characteristics Aeromagnetic surveys are useful to identify the areas that could
potentially influence economical chromitite seams. IRUPs can cause total
geological loss, including deformation of chromitite seams. According to an internal
company report in 2003, minor occurrences of IRUPs have been observed within
the project mining area, however, could be exposed throughout the life of mine.
Underground mapping and borehole data used for geological modelling also
confirmed that currently there is no presence of major IRUPs, that cause geological
loss or uncertainty with regards to the MG1 Mineral Resource.

36
4. EXPLORATORY DATA ANALYSIS (EDA)

The initial process, prior to resource estimation is an exploratory statistical

approach, in order to understand the underlying characteristics of the related data
and investigating stationarity of the geology associated with spatial continuity of
grade, lithology, mineralogy, including alteration (Ortiz & Emery, 2006). Stationary
refers to geological domains that are generally isotropic and homogeneous. The
nature of the tabular MG1 orebody is vertically bound by prominent chromitite
mineralisation within lithology that does not contain significant mineralisation and
is also generally homogeneous within the mineralisation zone itself. However,
continuity can vary along dip and strike of the orebody, that can influence the
statistical analysis of the data, and if variant across specific areas, should be
treated as separated domains. MCo considers the MG1 orebody to be estimated
as a single domain, and the exploratory data analysis was conducted based upon
that principle.

The main aim of the statistical analysis was to gain an understanding of the general
distribution of the Cr2O3%; SiO2%; FeO%; CR:FE (ratio), SG and true thickness
(seam width) relevant to the estimation of the MG1 Chromitite seam .Descriptive
statistical analyses were carried out to identify outliers or irregularities in the
borehole database. This was done to ensure that the geological modelling,
including the block model estimation, is not influenced by local variations due to
geological structures or unrealistic assay values. Furthermore, the results of the
statistical analysis were used to investigate the appropriate resource estimation
technique to be utilised for a comparative review of the current practice at the mine.

4.1 Data Selection and Representativeness

The final drill hole database used for the exploratory data analysis, resource
estimation and classification was based on the representative data selection
criteria of MCo to ensure that the investigation conducted in this research study
is carried out on the same data and that results would be directly comparable.

MCo considers borehole data nonrepresentative for resource estimation and

classification based on the following criteria:

• Diamond drilling of MG1 intersections that are less than 40m from surface
are deemed to be in the weathered zone of the orebody

37
 • Percussion drilling within the opencast area of the mining perimeter
• MG1 intersections that are influenced by geological structures (potholes;
dykes; faults and IRUPs)
• variables that are indicated to be erroneous or outlying (based on statistical
analysis)

The number of acceptable boreholes based on the MCo selection criteria and used
for statistical analysis of the different variables of interest are shown in Table 2.

Table 2: Number of boreholes used for statistical analysis

Seam Thickness
Cr2O3% SIO2% FeO% CR:FE SG
(CW)
Number of
91 91 91 91 72 73
boreholes

4.2 Compositing and Descriptive Statistics

Composting of originally borehole sampled grade values is usually conducted prior

to resource estimation, however, it is not a mandatory requirement. Generally,
resource estimation software requires the user to ensure that input data is of
constant support, meaning that sample lengths are distributed equally according
to grade. Original sampled variables consist of lengths that are usually not
homogenous and can consist of internal waste; incomplete intervals; small scale
analysis and high variability. Compositing homogenises and decreases the
variability of samples to an appropriate data scale, which also relates to more
robust geostatistical analysis, including variography (Deutsch & Rossi, 2014).

The MG1 sample data was composited based on the total length of each seam
intersection, prior to statistical analysis. This method is known within the chrome
industry and also part of the company’s procedure. The final validated composite
values are then used throughout the resource estimation process. The implication
of this practice is that there is no real grade vertical variability within the seam.

The histogram in Figure 12 presents a summary of the un-composited data, i.e.

all the samples that were cut according to the protocols discussed in Section 3.3.5.

38
Figure 12: Un-composited Cr2O3% sample data.

The results subsequent to composting is presented in Figure 13, i.e. variable length
composite that would include internal waste for the main economic variable
Cr2O3%.

Figure 13: Composited Cr2O3% sample data.

The descriptive statistics for the composited and un-composted of the relevant
variables are presented in Table 3.The true channel width (CW) of each selected
MG1 borehole intersection, were calculated during the compositing process using
the geological modelled surfaces (wireframes), to obtain estimated true dips, that

39
were used for conversion calculations form the apparent drilled intersection
lengths.

Table 3: Descriptive statistics of the un-composted and composited borehole data.

UN-COMPOSITED DATA STATISTICS

MISSING VALUES

SKEWNESS
STANDDEV
VARIANCE

KURTOSIS
RECORDS

SAMPLES

RANGE

COV%
MEAN
MAX
MIN
VARIABLE

LENGTH(m) 179 179 0 0.08 1.50 1.42 0.65 0.03 0.17 0.50 8.55 26

CR2O3% 179 178 1 33.99 44.33 10.34 40.84 1.52 1.23 -0.96 4.81 3

FEO% 179 178 1 21.00 25.80 4.80 24.10 0.52 0.72 -1.08 2.57 3

SIO2% 179 178 1 3.08 14.38 11.30 6.84 1.54 1.24 2.20 12.08 18

CR:FE 179 178 1 1.36 1.67 0.31 1.49 0.00 0.05 0.29 1.18 3

SG 179 140 39 3.61 4.41 0.80 4.12 0.02 0.14 -0.67 0.85 3

COMPOSITED DATA STATISTICS

MISSING VALUES

SKEWNESS
STANDDEV
VARIANCE

KURTOSIS
RECORDS

SAMPLES

RANGE

COV %
MEAN
MAX
MIN

VARIABLE

LENGTH(m) 91 91 0 0.22 1.60 1.38 1.27 0.04 0.21 -2.33 6.78 17

CR2O3% 91 91 0 38.90 43.25 4.35 40.87 0.90 0.95 0.18 -0.45 2

FEO% 91 91 0 21.62 25.42 3.80 24.09 0.38 0.62 -1.27 2.67 3

SIO2% 91 91 0 4.71 9.27 4.56 6.82 0.67 0.82 0.35 0.48 12

CR:FE 91 91 0 1.40 1.67 0.27 1.49 0.00 0.04 0.89 3.61 3

LENGTH_SG(m) 72 72 0 0.22 1.60 1.38 1.26 0.05 0.22 -2.21 6.03 18

SG 72 72 0 3.77 4.36 0.59 4.12 0.02 0.12 -0.38 -0.44 3

True CW (m) 73 73 0 0.66 1.56 0.90 1.26 0.03 0.18 -1.88 3.42 14

Observing the descriptive statistics, it is evident that there is a general reduction in

the variability, range, and coefficient of variation (COV) across all the variables
from un-composited samples to composited samples and the number of records
also reduced. The shape of the distribution of the main variable Cr2O3% also
transformed from a negatively skewed to a more normal shaped distribution, as
expected (Figure 13). The homogeneity of the orebody can be noted in the fact

40
that the mean and COV value of un-composited data does not differ significantly
from the composited data. On average, it can be observed that there are about 2
individual samples per MG1 borehole intersection, and the mean length roughly
double subsequent to compositing. Furthermore, the means of the un-composited
and composited Cr2O3% values remained similar, with a reduction in variance,
which is to be expected from the process of compositing the samples.

From the COV values of all variables it can be deduced that the distributions are
relatively symmetrical, and several estimation methods will be applicable for
Resource Estimation, this interpretation is based on Table 4 compiled by Noble
(1992).

Table 4: Rules for analysing COV (modified after Noble , 1992).

COV (%) Interpretation

Simple symmetrical grade distribution. Resource estimation is

0 - 25
easy, many methods will work.

Skewed distributions with moderate difficulty in resource

25 -100
estimation. Distributions are typically lognormal.

Highly skewed distributions with a large grade range. Difficulty

100 - 200
in estimating local resources is indicated.

Highly erratic, skewed data or multiple populations. Local

Above 200
grades are difficult or impossible to estimate.

4.3 Graphical representation of the data

4.3.1 Histograms

One of the most basic statistical tools to analyse data graphically is by use of a
histogram (Deutsch & Rossi, 2014). The histograms of the composited data for the
relevant variable are given in Figure 14 . Furthermore, the histograms were also
used to identify outliers, and local changes in the distribution that could assist in
the investigation of geological and (or) grade domains, for Resource estimation, if
required.

41
Figure 14: Histograms and fitted normal distributions for Cr2O3%, FeO%, SiO2%,
CR:FE, SG and Channel/Seam Width from left to right and top to bottom
respectively.
Based on the histogram plots and descriptive statistics the following conclusions
were drawn:
• Cr2O3 %: Normally distributed
• FeO %: Negatively distributed
• SiO2 % and CR:FE: Slightly positively distributed
• SG: Slightly negatively distributed
• Channel width: negatively distributed

42
The histograms for CR:FE and CW, Figure 14 indicate noticeable outliers, of high
and low values respectively. This could be due to geologically related influences
or random outliers inherent within the orebody and will be investigated in the
sections to follow.

4.3.2 Normal probability plots

Probability plots can be used to easily identify deviations from normality and also
assist in identifying inconsistencies within a dataset (Revuelta, 2018).

Figure 15 presents the normal probability plots for all the relevant variables.
Deviations from a straight line in these plots identify gaps within the distribution
which can be preliminary indications of possible population differences for
domaining and (or) grade capping. Grade capping is a method to constrain values
above the pullulation threshold, resetting values to the threshold value (Deutsch &
Rossi, 2014).

43
Figure 15: Normal cumulative probability plots for Cr2O3%, FeO%, SiO2%, CR:FE,
SG and Channel/Seam Width.

Based on the normal probability plots, and visual inspections with regards to the
outliers and deviations from a Gaussian distribution in relation to geological
structures (mainly potholes, dykes, and faults) the following were concluded:

• Cr2O3%; FeO%; SiO2%; SG - Lower and upper bound values are largely
influenced by borehole intersections within or in close proximity of faults,
dyke or pothole structures.

• CR:FE - A result of Cr2O3% and FeO%, outliers were detected in boreholes

located >700m outside the project area.

44
 • CW – A distinct group of eight low values, less than 1m were identified, of
which five values could be associated with a geological structural influence.
The remaining three values could be associated with underlying structures
which have not been delineated due to a lack of supporting borehole and
underground data.

• Based on the current data plots, the deviations and differences in populations
are fairly random in terms of location, and do not constitute in forming distinct
domains, and rather governed by geological influences.

• Due to the relatively few composited samples and the low variability in the
variable distributions, and low COV values observed in Table 3, grade
capping is not appropriate in this instance.

The geological influenced data was temporarily removed from the selected
borehole database in order to investigate the impact on the statistics (Table 5).

Table 5: Composite data statistics (geological influenced intersections removed).

SKEWNESS
STANDDEV
VARIANCE

KURTOSIS
RECORDS

RANGE

COV %
MEAN
MAX
MIN

VARIABLE

CR2O3% 77 39.30 42.97 3.67 40.87 0.77 0.88 0.21 -0.59 2

FEO% 77 23.10 25.13 2.03 24.18 0.18 0.42 -0.22 -0.12 2

SIO2% 77 5.21 8.45 3.25 6.79 0.47 0.68 0.13 0.26 10

CRFE 77 1.40 1.56 0.16 1.49 0.00 0.03 -0.25 0.21 2

SG 60 3.89 4.36 0.47 4.12 0.01 0.12 -0.18 -0.74 3

True CW (m) 65 1.13 1.56 0.43 1.31 0.01 0.08 0.22 1.10 6

It is evident that subsequent to the removal of the geologically influenced

intersections the data distributions of all the relevant variables are more
symmetrical and closer to normality, as expected. This elementary analysis
confirms the notion that Faults, Dykes, Potholes and IRUPs do have an influence
on the mineralisation.

45
4.4 Domaining

According to (Glacken & Snowden, 2001) and (Chanderman, et al., 2017) the
process of domaining should occur subsequent to creating a sound geological
model, where domains have been identified. In the simplest term, a domain can be
referred to as a volume or area which contain similar mineralisation characteristics
with regards to outside the specific area, in this case, the domain.

Domains can be classified as areas that also consist of similar continuity in relation
to geology and (or) value, which is usually grade or thickness. Geological and
grade domains are not always similar, although there could be a relationship.
Sinclair and Vallee (1994) cited in Sinclair & Blackwell (2002) refer to two
categories of continuity in Mineral Resource estimation, which are geological
continuity and value continuity. Domain boundaries are commonly defined by
assay and geological information (Sinclair & Blackwell, 2002).

Glacken and Snowden, (2001) highlight that frequently there are situations where
the mineralisation domains are embedded within a geological unit, meaning that
the resource grade model is constrained completely by the geological modelling.

This is true for this specific research project with regards to the MG1 chromitite
unit. The chromite mineralisation is constrained by the Middle Group 1 geological
unit, representing the domain for Mineral Resource estimation. However, economic
and product beneficiation recovery-related characteristics, including geological
structural influence could result in different domains being created based on strike
or dip direction of the orebody. This research project did not cover all the details of
the related attributes, as the current Mineral Resource estimation of MCo classifies
the MG1 orebody as a single estimation domain. Furthermore, the Mineral
Resource classification of MCo is mainly based on the borehole intersections of
the main economic variable Cr2O3% and is associated with various radii sizes to
represent resource confidence.

4.4.1 Stationarity

Chanderman (2015) indicates that geostatistics depend on the assumption that an

area that is being estimated should be stationary. Stationary can be referred to as
a domain that is homogeneous; without a dominant trend; constant continuity;
without extreme outliers of grade and represents a single population (Coombes,
2008).

46
Swath plots were used to investigate stationarity of the three variables mostly
influencing the Mineral Resource estimation, namely Cr2O3%, Channel width (CW)
and Specific gravity (SG). Swath plots, also known as sectional plots, are specified
width slices which can be used to plot composite averages of each slice on a graph.
The swaths are usually created through various directions of an orebody, indicating
the spatial distribution of a specific variable and can assist in verifying stationary
domaining. Swath plots are commonly used as a validation method for comparing
estimated block and sample grade distributions, (Webster, n.d.).

Slices of 120m and 180m were used for Cr2O3%, CW and SG composites in X
(North), and Y (east) directions respectively (Figures 16-18).

The data support for the swaths plots could be considered limited, however, the
results indicate that the value distribution within the single domain used is fairly
consistent. Although there are identified anomalies, it is concluded that these are
either as a result of the few observations in a specific swath (more likely) or are
associated with possible geological influences based on the position of the
borehole intersections with regards to major geological structures (faults and
dykes).

47
Figure 16: Swaths plots for Cr2O3% sample composites.

48
Figure 17: Swaths plots for CW sample composites.

49
Figure 18: Swaths plots for SG sample composites

50
5. VARIOGRAPHY (SPATIAL ANALYSIS)

The research project utilises Ordinary Kriging (OK); a geostatistical method for
spatial grade estimation; for comparison of the reliability and appropriateness of
the current Mineral Resource estimation and classification practices of MCo. The
OK estimation process not only produces a block estimate (ZK*) but also the
minimum estimation variance (K2), which reflects the confidence in the estimate,
at any location in the focus area (Dohm, 2015).

The spatial relationship of relevant variables needs to be estimated and modelled

as a requirement for any geostatistical estimation method, by means of a semi-
variogram (Morgan, 2011). Weights that are assigned to samples in the Kriging
estimation process which is based on the use of the semi-variogram model,
associated with the continuity in two or three dimensions of the variables under
consideration (Mpanza, 2015).

Apart from the requirement of the semi-variogram model in geostatistical

estimation algorithms, it is also a tool to quantify and investigate the underlying
spatial variability of the variables that are considered for estimation (Gringarten &
Deutsch, 2001). In the simplest description for a semi-variogram, is defined by
differences in grade, or a numeric variable, that is plotted in relation to the distance
of a sample population (Coombes, 2008).

It should be emphasised that prior to variography thorough Exploratory Data

Analysis must be conducted, that includes the establishment of domains, if present
[Coombes (1997) and Snowdengroup(2017a)].

5.1 The Semi- variogram and variogram model

According to Sinclair & Blackwell (2002) the fundamental application of the

semivariogram, is that it i) acts as an autocorrelation function that quantifies the
average continuity of grade in three-dimensional space and ii) the isotropic or
anisotropic nature of the grade distribution can be defined which is a crucial
concept regardless of the estimation technique that is utilised. The anisotropy can
be measured mathematically by the use of a semivariogram (Snowden, 2001), that
can then be applied in geostatistical estimations.

Geostatistical techniques all stem from the basis of the theory of regionalised
variables, that was developed by a French mathematician, Georges François Paul

51
Marie Matheron, and his theory of regionalised variables can be sited in Matheron
(1963). Geostatistical algorithms aim to use the spatial relationship between
variables, calculated by the semi-variogram, in order to apply a form of weighting
towards the estimation of a point or block with an unknown value (Glacken &
Snowden, 2001).

The function γ(h) represents the semi-variogram and is half the squared difference
in value of sample pairs which is separated by a lag distance (h). The values of the
experimental variogram (γ*(h)) can be calculated based on different lag distances
(h) by the formula:
𝑛ℎ
∗ (ℎ)
1 2
𝛾 = ∑(𝑍(𝑥𝑖 ) − 𝑍(𝑥𝑖 + ℎ))
𝑛ℎ
𝑖=1

Z(xi) is the regionalised variable value at a sample point xi

Z(xi + h) is the value of another point at a lag distance (h) from point xi
nh is the number of sample pairs for a specific distance h

The semi-variogram is calculated in different directions to measure variability,

which increases as the spacing between samples increases (Deutsch & Rossi,
2014).

The experimental variogram values (𝛾 ∗ (ℎ)) are plotted against h on a graph and a
model is fitted to represent an approximation of a smoothed mathematical function
(Figure 19).

Figure 19: Experimental semivariogram with a spherical model (Snowden, 2001).

52
The three main components of the fitted semivariogram model are described by
(Bohling , 2005) and (Snowden, 2001):

• Sill: The value of the modelled semivariogram, where the graph levels and
is typically equal to the population variance.

• Range: the range at which the lag distance of the semivariogram

approaches the sill value. The values that are beyond the range can be
considered to be spatially uncorrelated.

• Nugget: Theoretically the value at the origin of the semivariogram should

be equal to zero. The nugget or also referred to as the nugget effect is the
intrinsic variability, including the sample variability at a zero-lag distance.

There are various variogram models types, represented by different mathematical

functions and used in geostatistical applications, the most frequently used for
Mineral Resource estimation is however the spherical model (Sinclair & Blackwell,
2002).

A single structured spherical variogram model (Figure 20) was also used in this
research and is represented by this formula defined in the equation: `

0 𝑖𝑓 ℎ = 0
3
3 ℎ 1 ℎ
𝛾(ℎ) = 𝐶0 + 𝐶1 [ ( ) − ( ) ] 𝑖𝑓 0 < ℎ < 𝑎
2 𝑎 2 𝑎
{ 𝐶0 + 𝐶1 𝑖𝑓 ℎ ≥ 𝑎 }

C0 = the nugget effect, that represents local, short-range geological variability,

including subsampling; analytical and sampling variability. According to Bohling
(2005) , “the nugget represents variability at distances smaller than the typical
sample spacing, including the measurement error.”

C1 = the structural component of the model, and also referred to as the sill
component.

C = (C0 + C1) = the total sill, which theoretically is the total variance of the sample
data.

a = the range (or range of influence), which represents the increase of the distance
of the average variability from zero to C1, or from C0 to the total sill.

h = the lag distance or also referred to as the sample spacing.

53
Figure 20 : Components of the spherical model (Sinclair & Blackwell, 2002).

5.2 Variography of variables under study

According to Gringarten & Deutsch (2001) an understanding of the variogram in

three-dimensional space is necessary to completely identify geological continuity.
Furthermore, different directions of continuity should be tested to be consistent with
geological deposition, including accurate quantification of the spatial variability for
algorithms associated with geostatistical estimations. Variogram values can be
plotted in various directions and then subsequently contoured for an easier method
of viewing and interpreting variograms (Coombes, 2008).

Variogram contours can assist in identifying the maximum direction of continuity,

which is usually associated with the lowest variability in the longest range. Recent
developments in Resource estimation software have made this process easy to
plot, for example, Supervisor® and Leapfrog Edge®, that was also used for this
study with regards to variography.

The numeric modelling functionality in Leapfrog , a Radial Basis Function (RBF)

interpolation method, was used throughout the variography analysis to assess the
spatial continuity with regards to the distribution of the variables. Grade shells in
the form of isosurfaces (RBF contour surfaces) were created for the relevant
variables Cr2O3%, CW and SG, and are presented in Figures 21-23, respectively.

54
Figure 21: Isosurfaces for Cr2O3%. Figure 22: Isosurfaces for CW.

55
Figure 23: Isosurfaces for SG.

Based on the results of the isosurfaces of the relevant variables, it is evident that
there is no prominent continuity direction. Furthermore, based on the current
sample support, it was also observed that the spatial continuity is not related to
geological structures. However, geological structures do have an influence on the
distribution of the variables with regards to a post-depositional sense. It is possible
to argue that based on the initial EDA, swath plots and isosurfaces that domaining
should be further investigated, especially for Cr2O3% and CW, as domains will
impact the variogram analysis. There can, however, also be strong geological
structural influences that could be associated with the grade distribution
characteristics.

The decision of treating the sample data of the three primary variables (Cr 2O3%,
CW and SG) as a single domain was established in Chapter 4, and the
experimental semi-variograms from the data were calculated for the domain.

Standardised experimental semivariograms were calculated in different directions,

both grade capping as well as cutting were applied to the Cr2O3% variable. Due to
the few data available it was however not possible to identify any preferred
direction of continuity for any of the three variables and therefore, omni-directional

56
semivariograms spherical variograms models were fitted to the experimental
semivariograms for the three variables (Figures 24-26).

Omni-directional variograms can also be referred to variograms that are isotopic,

the spatial variability of the variable is the same in all directions. The variogram is
considered to be isotropic, where the calculated semivariogram is not dependent
on the direction of the lag distance (Krige, 1981).

Figure 24: Spherical Semivariogram model for Cr2O3%.

Figure 25: Spherical Semivariogram model for true channel width (m).

57
Figure 26: Semivariogram with spherical model for SG.

Double structured spherical semivariogram models were fitted to the Cr2O3%, and
SG(g/cm3) and a single structured semivariogram model was appropriate for the
CW(cm), the model parameters are presented in Table 6. The percentages in
brackets represent the ratios of each variability component to the total sill (C) of
the semivariogram model.

Table 6: Parameters of modelled spherical variograms for Cr2O3%, CW and SG.

Nugget 1st Structure 2nd Structure st

1 Structure 2nd Structure
Variable Effect C0 Sill C1 Sill C2
Range a1 Range a2
(C0/C)% (C1/C)% (C2/C)%
Cr2O3%. 0.20 (18%) 0.55 (50%) 0.35 (32%) 470 530
CW 0.17 (17%) 0.83 (83%) - 500 -
SG 0.20 (20%) 0.50 (50%) 0.30 (30%) 200 430

The general rule of thumb is to model the semi variogram to the sample variance,
which if the variogram is normalised, to a sill value of 1. However, in the event that
the total sill is more than one, it means that experimental sample variance is more
than the sample variance of the data. According to Morgan (2011) in the case
where the experimental semivariogram contains a minor difference greater than
the sample variance, the semi-variogram should be fitted to the sill of the
experimental semivariogram.

58
Morgan (2011) concludes on this statement, by stating that:

“… the sample variance is calculated assuming random, independent

samples (which is generally not the case – the samples are spatially
correlated.)”.

The standardised experimental sill for the Cr2O3% (Figure 24) is greater than one.
Therefore a Sill greater than 1 was modelled for this variable.

The variations of experimental variogram values that were observed in all (Cr2O3%,
CW and SG) the semivariograms, is most probably associated with a lack of
sample data and large sample/ borehole spacing.

Morgan (2005) emphasises that

“… the most important aspect of the fitted semi-variogram model is the

nugget effect and the slope near the origin.”

and also refers to Armstrong (1998) which states that

“… the behaviour of the semi-variogram at and near the origin has a

significant influence on kriging results, as well as their stability”.

The researcher also reached this conclusion from a practical exercise executed
during a geostatistical course at the University of the Witwatersrand (Dohm,2015).

Due to the lack of closely spaced borehole data, with an average spacing of
±130m, and limited deflections, a limited amount of sample pairs close to the origin
of the semivariograms was available. This reality had a subsequent influence on
the estimation of the modelled nugget effects for the relevant variables (Cr2O3%,
CW and SG).

It could be argued that the modelled nugget effects are higher than expected in a
Bushveld Complex chromitite environment. Although the nugget effects are
subjective, especially for Cr2O3% and SG, the variograms provide a good indication
of the range of influence associated with the spatial variance (or correlation) of the
variables considered.

5.2.1 Variogram Model Range of Influence and Resource Classification

Dohm (2004) mentions a method based on the range of influence of the variogram
model as proposed by Snowden (1996) and that was at the time often used for
Mineral Resource classification, that is:

59
 “Resources are classified as Inferred when drill holes are further apart
than the range of influence of the variogram. The drill spacing at which
a distinction between Measured and Indicated is made is based on a
rule of thumb and is taken as the distance equivalent to two-thirds of
the total variability i.e. two-thirds of the sill of the variogram model”.

However, Dohm (2004) clearly states that ranges of the variogram are not
considered to be adequate for all resource classification situations, especially in
the Witwatersrand Gold environments where high nugget effects, and short
variogram ranges are commonly encountered.

This two thirds the range of influence classification methodology has been
observed to be applied for Chromitite and Platinum Mineral Resource
classifications and was sited in African Rainbow Minerals (ARM) (2008), including
Canadian Aviation Electronics (CAE) (2013) annual Mineral Resource and
Reserve reports.

There is thus some merit in using this technique, as a foundation, although one
should also consider other factors, such as geological complexities and the
Competent Person’s judgement and experience. Furthermore, it appears that the
technique would only be useful in highly homogeneous orebody environments, with
long range variogram models, zero nugget and not associated with a high
frequency of geological complexity.

Isaaks & Srivastava (1989) mentions “in fitting this model (spherical model) to a
sample variogram it is often helpful to remember that the tangent at the origin
reaches the sill at about two-thirds of the range”. The researcher is of the opinion
that the technique is based upon the fact that most of the spatial correlation is
generally captured within two-thirds of the theoretical spherical variogram and
could serve as a means of confidence, depending on the characteristics of the
orebody, and should consider geological complexities.

The “two-thirds” classification technique was compared to MCo’s current

classification standard (Table 7). It is evident that MCo’s classification radii are
more conservative, likely for good practical reasons. The risk of unknown or
unidentified geologically complex or pothole influenced areas are too high to allow
for declaration margins such as ±350m and 530m borehole radii for Measured,
including Indicated Mineral Resources respectively.

60
Table 7: Comparison between two-thirds variogram range technique and the
mining company’s standard for Mineral resource classification.

Confidence Confidence radii (m) - Confidence radii (m) -

%Diff
classification two thirds method mining company
Measured 353 150 136%
Indicated 530 300 77%
Inferred >530 600 -
*Variogram range for Cr2O3% = 530m

61
6. BLOCK MODELLING AND ESTIMATION

MCo utilises a 2D gridding algorithm, known as the Growth technique, described

in section 2.4 of this research report, for estimation. The technique requires a
gridded mesh, of which the mesh size is determined by the estimator, and as a
starting point Dassault Systèmes GEOVIA Inc. (2016) recommends using one-
quarter to one-fifth of the sample data spacing. MCo uses a mesh size of 25m x
25m for estimating the three variables (Cr2O3%, CW and SG) within the Geovia
MinexTM software.

In this research report two other estimation techniques (OK & IDW) are used for
comparison to the Growth technique, based on estimated block models created in
Leapfrog® Edge.

6.1 Block Model Size and Search Neighbourhood

Snowdengroup (2017b) considers the parent block size used in a Mineral

Resource model to be one of the most important parameters, as it influences the
quality of grade estimates, especially when using OK. Furthermore, Chanderman
(2015) highlights the fact that Kriging aims to minimize the estimation variance,
however this can only be achieved when the search neighbourhood is properly
defined.

The search neighbourhood for OK comprise of the parent block size; minimum and
maximum number of samples considered; search range and point discretisation of
the estimated parent block. The search neighbourhood can also be considered as
the “input or estimation parameters” for a kriging estimate, which also includes a
variogram model, discussed in detail by Coombes (2008) and Amwaama (2018).

Yamamoto (2005) indicates that linear estimation methods, such as OK and IDW,
inherently have a smoothing effect that leads to conditional bias. Conditional bias
is well documented in Deutsch (2007), and in simple terms, it refers to the
occurrence that the expected true values of the blocks not being equal to estimated
grades. The influence of the smoothing effect is dictated by the continuity of
mineralisation (variogram model); number of samples used in the OK estimate and
the search criteria established for the kriging neighbourhood (Deutsch, 2007).

Vann, et al. (2003) describes a well-known method for quantitatively assessing a

kriging neighbourhood, which assists in minimising conditional bias and optimising

62
the kriging estimation parameters, prior to the estimation. The method known as
Quantified Kriging Neighbourhood Analysis (QKNA), has the objective to
determine a search neighbourhood and block size combination, to achieve
conditional unbiasedness by assessing measures of the following criteria: Slope of
Regression (SOR); Kriging Efficiency (KE); weights of the means for a simple
Kriging; Kriging weights (which includes proportion of negative weights) and
Kriging variance.

Kriging efficiency (KE) and Slope of regression (SOR) are derived from the
following formulas (De-Vitry, 2003) :

KE = (BV – KV)/BV

SOR = (BV – KV + µ)/ (BV – KV + 2µ)

Where:

BV is the Block variance (theoretical variance of blocks within the estimation

domain).

KV is the Kriging variance

µ the LaGrange multiplier (used in solving the OK matrix system)

The use of the SOR and KE as a criteria for measuring the conditional bias is well
documented in literature by Deutsch (2007) ; Sinclair & Blackwell (2002) ;
Coombes (2008) ; Vann, et al. (2003); Deutsch, et al. (2014) and Nowak &
Leuangthong (2016), whom all describe the detailed Geostatistical background in
a mathematical sense. Nowak & Leuangthong (2016) reviews the use of the SOR
and KE as measures in defining the optimal kriging neighbourhood and highlights
that both these measured should be handled with caution, which is supported by
the case study on an epithermal gold deposit in British Columbia.

The QKNA process proposed by Vann, et al. (2003) was used in this research
project to inform the selection of appropriate block size, minimum and maximum
number of samples, including the dimensions of the search range combinations
that contain the least conditional biasness.

Vann, et al. (2003) recommends testing a range of block sizes with a range of data
configurations that are well informed, less well informed, and poorly informed. The
margins that best define conditional unbiasness are:

• A SOR and KE that are equal to one or expressed as a percentage of 100%

63
 • Negative weight proportion of less than 5% of the total kriging weights
• Analysis of the Kriging Variance, or also known as the estimation variance
and attempt to obtain a minimum variance with regards to the other
measurement criteria

The initial block size that was considered in this report prior to QKNA for the OK
and IDW estimation was based on the composite data configuration of the relevant
variables (C2O3%, CW, and SG). The sample composite spacing is on average
130m, and 60m x 60m block size was considered as a baseline for QKNA, based
upon the industry standard for blocks not to be smaller than ½ the drill hole spacing
(Coombes, 2008).

6.1.1 Quantified Kriging Neighbourhood Analysis (QKNA) results

The QKNA was conducted using a trial version of Supervisor® software on the
sample composites for Cr2O3%, as it is considered as the main variable of this
study. The variogram is a critical input for a QKNA (Snowdengroup, 2017b), and
similar variogram model parameters were used in Supervisor®, as modelled in
Leapfrog Edge® for Cr2O3%. The initial selected block model size (60mx60m) were
tested together with a range of different block sizes, however the 60m x 60m block
size was deemed representative and used throughout the QKNA process.

The kriging neighbourhood parameters that were established during the QKNA
process based on the measurement criteria is shown in Table 8. The search radius
has been determined by the modelled Cr2O3% semivariogram presented in Table
6 (Section 5.2) for all the data configurations.

Table 8: Kriging neighbourhood parameters based on QKNA.

Kriging neighbourhood parameters - based on QKNA

Min Max
Data configuration Search Range Discretisation
samples samples
Well informed 530m 4 12 4x4x2
Less well informed 530m 6 15 4x4x2
Poorly informed 530m 4 12 4x4x2
* 60m x 60 m x 2m Block sized used based on QKNA

6.2 Estimation Comparisons and Analysis

In order to analyse the appropriateness of MCo’s Mineral Resource estimation

method, namely the Growth technique, comparisons were conducted with known

64
industry estimation techniques. The Mineral Resource estimation techniques used
for comparisons were OK and IDW. This is not to say that estimation techniques
used for comparison are more appropriate, however, it is merely used as a
measure of the algorithm commonly used to estimate chromitite deposits in the
Bushveld Complex.

Only the Cr2O3% variable was used for the comparison as assumption has been
made that if alternative methods are used for comparisons for the other variables
(CW and SG), the same conclusions can be established based on different tested
algorithms.

Various scenarios were considered to achieve similar results as obtained from the
Growth technique, and comparisons with the original search neighbourhoods
established from QKNA results. Table 9 reflects different variogram parameters
and Table 10 the different search neighbourhoods considered in the comparison
process.

Estimation comparisons were conducted on a 60m x 60m x1m block size, in 2D

space. The estimated gridded meshes conducted in Geovia MinexTM, were
converted into 25m x 25m block models, and subsequently regularised into 60mx
60m block models for comparable visualisation and statistical analysis in
Leapfrog® Geo/Edge software. The conversions were conducted in the Datamine
Studio RM®.

65
Table 9: Tested estimation scenarios, with parameters used for comparison towards the Growth technique.

Scenario code

(Range-minor)

(Range-minor)
(Range-Major)

(Range-Major)
(Range-Semi-

(Range-Semi-
Structure 1

Structure 2

Structure 1

Structure 1

Structure 1

Structure 2

Structure 2

Structure 2
Model type

(Sill) (Sill)
Scenario

Nugget

major)

major)
(Sill)
1 IDW2 N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A

2 IDW4 N/A N/A N/A N/A N/A N/A N/A N/A N/A N/A

3 OK_1 Spherical 0 0.75 0.35 470 470 50 530 530 50

4 OK_2 Spherical 0.2 0.55 0.35 470 470 50 530 530 50

5 OK_3 Spherical 0.2 0.55 0.35 470 470 50 530 530 50

*Variogram models are normalised

66
Table 10: Variogram model parameters, used in estimations comparisons towards Growth technique.

Estimation search neighbourhoods for Cr2O3%

S1 S1 S2 S2 S3 S3
Search Search Search
Scenario Estimation Minimum Maximum Minimum Maximum Minimum Maximum
Scenario ranges ranges ranges
code type no of no of no of no of no of no of
(Search 1) (Search 2) (Search 3)
samples samples samples samples samples samples
Inverse
Max:530 Max:1060 Max:2650
distance
1 IDW2 Int:530 4 12 Int:1060 4 10 Int:2650 3 8
weighting to
Min:N/A Min:N/A Min:N/A
the power 2
Inverse
Max:2500
distance
2 IDW4 Int:2500 1 5 N/A N/A N/A N/A N/A N/A
weighting to
Min:N/A
the power 4
Max:530 Max:1060 Max:2650
Ordinary
3 OK_1 Int:530 4 12 Int:1060 4 10 Int:2650 3 8
Kriging
Min:N/A Min:N/A Min:N/A
Max:530 Max:1060 Max:2650
Ordinary
4 OK_2 Int:530 4 12 Int:1060 4 10 Int:2650 3 8
Kriging
Min:N/A Min:N/A Min:N/A
Max:530 Max:1060 Max:2650
Ordinary
5 OK_3 Int:530 6 15 Int:1060 4 12 Int:2650 3 10
Kriging
Min:N/A Min:N/A Min:N/A

* Block size used : 60m x 60m x1m (2D estimation) , expansion factors used for Search 2 = 2 and Search 3= 5, bottom / top capping was not applied.

67
The estimation statistical analysis was conducted on a volume-weighted approach
with regards to the block model estimate comparison, with equal dimension in the
z-direction (height) of the block size, which is one meter.

However, tonnage-weighted statistics were tested for some of the estimation tests,
and insignificant differences were found, compared to a volume-weighted
approach.

6.2.1 Estimation results, comparisons and validation

Throughout the comparison of the estimation results the differences were

calculated by subtracting the Growth techniques results from an alternative
estimation method, then dividing by the Growth technique and converted to a
percentage by multiplying by a 100 to obtain a percentage difference.

Various authors such as Webster (n.d.), Glacken & Trueman (2014) and WSP
(2015) recommend common methods of validation for Mineral Resource estimates
of which the following were used to validate and compare the different estimation
scenarios with the Growth technique:

• Visual comparison of block model grades and input data

• Comparing estimate and input data declustered means
• Using alternative Resource estimate algorithms and comparing descriptive
statistics and grade distribution graphically by using histograms, QQ Plots,
including scatter plots.
• Swath plots (sectional plots) comparing averages of input composite data
and block model estimates

The spatial results of the Cr2O3% block model estimation scenarios considered
together with declustered composites are shown in Figure 27.

The Cr2O3% grade distribution based on volume percentage of the block model
estimates are summarised with bar charts in Figure 28.

68
Figure 27: Cr2O3% block model estimates (Growth technique and alternative
methods)

69
Figure 28: Volume percentage based on Cr2O3% grade distribution of the estimation scenarios

70
Based on the general visual inspection of the Cr2O3% grade estimates and
declustered composites (Figure 27), all the estimations methods honour the input
data to an acceptable level. However, the Kriging estimates that were conducted
with variograms with 18% nugget effects (OK_2 & OK_3), over smooth the grade
estimates relevant to the input data. The over smoothing, is associated with the
effect that the variogram model has on the weight distribution of samples used to
estimate a block. In this case the over smoothing effect is observed when a higher
nugget effect is used for the variogram model, as appose to OK_1 results with a
zero nugget effect. Furthermore, increasing the minimum and maximum number
of samples for the Kriging neighbourhood of OK_3, did not have a significant
impact on the estimate.

Visually the grade distribution of the Growth technique and the inverse distance
weighting to the power of 4 are similar and this was confirmed by the volume grade
distribution plot (Figure 28). Both these methods are sensitive to high or low-grade
samples that are sparely spaced and creates a “smearing” effect of low or high-
grade zones.

The observation confirms that increasing the power of the IDW algorithm, the
estimated value starts receiving the value of the closest sample composite,
resulting in an estimate that is similar to a Polygonal method (Diadato & Ceccarelli
(2005) cited in Babak & Deutsch (2008) or Nearest neighbour method (Glacken &
Snowden, 2001). This means that in the event that there are low or high-grade
composite samples sparsely spaced from surrounding data, the estimated values
around the sample will receive a high or low grade until the surrounding data is
closer to a block that requires estimation.

Glacken & Snowden (2001) highlights that most resource estimation techniques,
are associated with a grade smoothing interpolation and can be categorised into
non-geostatistical (IDW) and geostatistical (Kriging) estimation methods.

Based on the visual observation for the IDW2, compared to IDW4 and Growth
technique, the “smearing” effect is mitigated by IDW2, and grades are smoothed
more conservatively.

Three Ordinary Kriging estimation scenarios were conducted (Figure 27), in order
to test the effect of different search neighbourhood parameters and using a
variogram with a zero nugget effect for OK_1.

71
Due to the limited amount of close-spaced data the nugget effect initially modelled
at 18%, can be considered as subjective, as Chromitite deposits in the BC are
known for low nugget effects (<5%). The grade distribution of the OK_1 estimates,
using a variogram model with zero nugget effect, visually appears to represent
grade trends of the input data better than with the other estimation comparison
scenarios, including the Growth technique.

Glacken & Trueman (2014) states that for understanding the global performance
of the Resource estimation during validation, the declustered sample mean, should
be similar to the estimation mean. The block estimates variance, should also show
reduction as appose to the sample data variance.

The declustered mean for the Cr2O3% composites was determined with a
declustering tool available in Leapfrog® Edge (Leapfrog3d-Mining and Minerals
team, 2019), and an ellipsoid approach based on the general spacing of the data
was used. The declustered mean for the composite data was estimated to be
41.11% Cr2O3, and the differences in percentage compared to the estimate means
are shown in Figure 29.

% Difference from Cr2O3% declustered mean

GROWTH

OK_3

OK_2

OK_1

IDW4

IDW2

0.00% 0.05% 0.10% 0.15% 0.20% 0.25% 0.30%

Percentage

Figure 29: Percentage difference of Cr2O3% block estimate means and the
declustered composite sample mean.

Although the Growth technique estimated mean difference is the highest, the
estimate means are within 0.5% difference of the declustered sample mean, which
is deemed acceptable.

72
The descriptive statistics of the block model estimates and the sample composite
data are presented in Table 11.

Table 11: Statistics of the estimation methods and sample composites.

Descriptive statistics of block model estimates - Cr2O3%

Standard
Volume

quartile

quartile
Median
Variance

Lower

Upper
Mean

COV

Max
dev

Min
Method

IDW2 1918800 41.15 0.735 0.018 0.541 39.18 40.65 41.10 41.61 43.22

IDW4 1918800 41.17 0.944 0.023 0.891 38.91 40.45 41.11 41.75 43.25

OK_1 1918800 41.16 0.809 0.020 0.655 39.28 40.55 41.13 41.73 43.02

OK_2 1918800 41.14 0.672 0.016 0.451 39.60 40.64 41.18 41.63 42.69

OK_3 1918800 41.14 0.670 0.016 0.448 39.60 40.63 41.18 41.63 42.69

GROWTH 1918800 41.22 0.909 0.022 0.826 39.14 40.56 41.12 41.82 43.23

Descriptive statistics of Cr2O3% sample composites

Cr2O3%
N/A 40.87 0.953 0.023 0.908 38.90 40.11 40.88 41.43 43.25
composites

Generally, the statistics compare well between the block estimates, the alternative
estimation technique means are within 0.5% of the Growth technique mean.
Reduction in the variance from the original composites is also noted, however
minimal (<10%) with regards to the Growth technique and IDW4, as appose to the
Ordinary Kriging technique, including IDW2 scenarios that reduced more than
25%. Base on the statistics, a global bias is observed for all the block estimates,
of over estimating, if compared to the composite mean. This could be the effect of
the sparse sample data that was used for the estimation process, however as
mentioned, the global performance are better measured by using the declustered
mean of the composites. Slightly higher coefficient of variation (COV) was noted
from the Growth and IDW4 estimates, compared to IWD2 and Ordinary Kriging
results. This can be attributed , as the global means of all the block estimates that
are almost similar, however the standard deviations are higher for the Growth and
IDW4, which will result in higher COV values.

73
The reduction in the variance is explained by Isaaks & Srivastava (1989):

“This reduced variability of estimated values is often referred to as

smoothing and is a consequence of combining several sample values
to form an estimate”.

Isaaks & Srivastava (1989) also highlights in an exercise conducted, that the
polygonal method that only uses a single sample, had estimates that are
unsmoothed.

Glacken & Trueman (2014) states that when a lower power is used in inverse
distance ( i.e. IDW2), the greater the smoothing, and the higher the power (i.e
IDW4), the estimate approaches a nearest neighbour method, as observed in
Table 11.

Rezaee, et al. (2011) also mentions the smoothing effect associated with OK
estimates:

“OK estimates do not reproduce the sample histogram because of

reduced variance as a consequence of the smoothing effect. In the OK
estimation process low values are overestimated and high values
underestimated making the estimated histogram narrower than the
sample histogram.”

Morgan (2005), conducted an exploratory comparison by using the polygonal,

IDW2, and OK methods, and also confirmed the smoothing effect. However,
Morgan (2005), clearly states the advantages of Kriging of being the best linear
unbiased estimator, and provides a KV, as a measure of quality of the estimation.
Furthermore, Morgan (2005) cites Journel & Huijbregts (1978), which states that
“reproducing spatial variability is not the goal of kriging estimates, but rather the
goal of geostatistical simulations”.

The histograms of the Cr2O3% estimation results for the different estimates are
shown in Figure 30 and the histogram of the Cr2O3% input data on which the
estimations are based on appear in Figure 31 for further comparison of results.

74
Figure 30: Histograms of the Cr2O3% block estimate scenarios.

75
Figure 31: Histogram of original sample input data (Cr2O3% composites).

The kriging results with a higher nugget (OK_2 & OK_3) reduced the variance of
the estimates more than the other estimates, including the input data, however it
overly smoothed the grade distribution of the estimates.

This highlights the importance of accurately fitting the nugget-effect and slope near
the origin of the variogram model, due to the significant influence towards kriging
results (Morgan, 2005).

The Growth and IDW4 estimates underlying grade distribution is similar and
compares the best with the input data distribution, with minimal reduction of
variability, based on the histograms and statistics. The IDW2 and OK_1 grade
distributions have more similarities, with the smoothing effect confirmed (Table 11),
with reduction in the variability.

The grade distribution of the alternative estimation techniques was also compared
with the Growth technique grade distribution by the use of Q-Q plots (Figure 32)
and scatter plots (Figure 33) to investigate correlation.

76
Figure 32: Q-Q Plots of the Growth technique estimate versus the alternative
estimation methods.

77
Figure 33: Scatter Plots of the Growth technique estimate versus the alternative
estimation methods.

78
The results confirm the grade distribution similarities of the Growth and IWD4 grade
estimates, including a good positive correlation (R2 >90%) between the estimated
values, with a narrow scatter envelope between the values. The scatter correlation
between the growth and IDW2 estimates also presents a positive correlation of
>90%, however the scatter envelope starts to increase towards the high grades,
and effect is more prominent with regards to the Ordinary kriging estimates,
although still presents a positive correlation of >80%.

The effects of smoothing and variance reduction from the IWD2 and Kriging
estimates can be seen in the differences of grade distributions compared to the
Growth estimate distribution. Higher grade distribution is observed in the upper-
quartiles for the Growth estimates and vice-versa on the lower-quartiles for low-
grade distribution, compared with IDW2, including Ordinary Kriging distributions.

Swath plots were generated in 120m slices through the estimated block model
grades, which included the sample composites in a North-South and East-West
direction (Appendix A:1).

Generally, it is recommended to use swaths/slices that are equal to the block model
size, however due to a lack of composite data, the slices were multiplied by a factor
of two, to capture more data within a swath. Swath plots are used to compare two
datasets (Webster, n.d.), and could also indicate “smearing” or smoothing of high
grades (Leuangthong & Nowak, 2015). A swath plot represents the averages of
the grade estimates and composite data at various swath/slice intervals that are
then plotted on the same graph. Chanderman (2015), also states that the purpose
of the swath plot is to investigate gross over or under underestimation associated
with the block model estimates.

Due to the limited amount of sample composites that are captured in a swath
interval, the swath plot interpretations can be somewhat subjective, nonetheless
the swath plots were also used to compare the grade trends between the growth
and alternative block estimates. Generally, the average differences based on the
swath intervals for all the estimates, compared well to the composite values and
the means are within 1%, which is considered acceptable. Consistent
overestimation was noted in the swaths plots for all the estimated grades, probably
due to the effect of a lack of sample numbers available, as well as sparse spacing
of data along swath intervals.

79
However, the swaths plots do provide a grade trend comparison between the
growth estimates and alternative estimation methods. As previously observed, it is
evident that Growth and IDW4 generally possess the same grade trends, with
minor differences, with regards to the swath plots.

The Ordinary Kriging estimates do appear to smooth the grade trend at places;
however, minor differences were noted with grades slightly higher and lower than
the Growth estimates. The IDW2 grade estimates are consistently lower than the
Growth estimates and globally appears to follow the grade trends of the composite
data the best of all the estimated grades.

Based on the estimate validation and comparisons, the following is surmised;

globally the means of the estimates appear to be similar leading to the conclusion
that the estimation methodologies considered do not differ significantly on a global
scale. However, local variation is observed between the estimates compared,
which is revealed by the differences in the variance of the estimated grade
distributions of the blocks.

6.2.2 Comparison of Estimation Techniques within the MCo Resource

Classification

In order to compare local variation of the grade estimates, and the impact of the
estimates when applying MCo’s resource classification method were to be used,
the estimates are reported according to MCo’s resource classification perimeters
(Table 12). The estimate via the growth method, was obtained directly as reported
by MCo using Geovia MinexTM. In order to obtain tonnage for the alternative
estimates, CW and SG were estimated using the IDW2 technique, to discount the
influence of variogram modelling.

80
Table 12: Estimated Cr2O3% grades and tonnages of all estimation methods within
MCo’s defined resource classification perimeters.

RESOURCE CLASSIFICATION
MEASURED INDICATED INFFERED TOTAL
IDW2 41.18 41.70 41.67 41.27

Estimated Cr2O3%
IDW4 41.18 41.81 42.02 41.29

weighted
-Tonnage
OK_1 41.21 41.79 41.22 41.29
OK_2 41.19 41.70 41.22 41.26
OK_3 41.19 41.68 41.22 41.26
GROWTH 41.22 41.91 42.36 41.35
*ALTERNATIVE (Mt) 6.57 1.13 0.22 7.92

Tonnes
MCo (Mt) 6.68 1.14 0.24 8.06
%DIFF TONNES -2% -1% -8% -2%
* The tonnage for all alternative estimates (IDW2,IDW4,OK_1,OK_2,OK_3) are equal

The percentage grade difference between the Growth estimate for each Resource
class that is based upon the results of table 12, is presented in Table 13.

Table 13: Grade (Cr2O3%) percentage differences of the Growth estimate

compared to the alterative estimates as per MCo’s resource classification
perimeters.

Cr2O3% difference compared to GROWTH –

Tonnage weighted
RESOURCE
IDW2 IDW4 OK_1 OK_2 OK_3
CLASSIFICATION

MEAS % DIFF -0.1% -0.1% 0.0% -0.1% -0.1%

IND % DIFF -0.5% -0.2% -0.3% -0.5% -0.6%

INF % DIFF -1.6% -0.8% -2.7% -2.7% -2.7%

TOTAL % DIFF -0.2% -0.1% -0.1% -0.2% -0.2%

The greatest % tonnage differences between the alternative estimation methods

and the Growth technique are observed in the Inferred category, these
percentages are considered to be within a margin of acceptance.

It might be argued that the choice of applying a different estimation technique

(IDW2), to estimate the variables that account for tonnage (SG and CW) and this
could affect the tonnage estimation. However, MCo’s resource classification
perimeters were used to report the Cr2O3% differences between the Growth
estimates based on volume as well, with a fixed block model height and the volume

81
results are given in Table 14. The conclusion reached based on Table 13 remain
unchanged and the biggest difference is with the Inferred Resource category.

Table 14: Alternative estimated (volume weighted) Cr2O3% percentage differences

compared to the Growth estimate as per MCo’s classification perimeters.

Cr2O3% difference compared to GROWTH –

Volume weighted
RESOURCE IDW2 IDW4 OK_1 OK_2 OK_3
CLASSIFICATION
MEAS % DIFF -0.1% -0.1% 0.0% -0.1% -0.1%

IND % DIFF -0.6% -0.3% -0.4% -0.6% -0.7%

INF % DIFF -1.5% -0.7% -2.6% -2.6% -2.6%

TOTAL % DIFF -0.2% -0.1% -0.2% -0.2% -0.2%

The differences in grade per resource class, compared to the Growth estimates,
can also be logically considered as insignificant, with regards to the Measured and
Indicated Resource classes. However, the most difference is observed in the
Inferred Resource class, where data is sparse, and the IWD2 and the OK results
have less of a “smearing” effect on the grade estimate. The close resemblance of
the IDW4 and Growth method is observed in the Inferred class and is due to the
fact that the IDW4 generally mimics the Growth technique. The comparison did not
include the total geological losses for reporting, and only included the geological
losses associated with major dykes, as applied by MCo, from the classification
perimeters.

The Resource classification perimeters of MCo (Figure 34) are determined by a

combination of pre-determined MG1 Chromitite intersected borehole radii (150m
for Measured; 300m for Indicated; 600m for Inferred) and the resource estimator’s
judgement where confidence can be overruled by the general standard, based
upon complex geological influences, unexpected grade trends, or dip regimes,
where confidence is uncertain. This judgement call can be observed in the top part
of the Mineral Resource, East and West with regards to an Inferred Resource
classification.

82
Figure 34: MCo’s resource classification for the current sample data configuration.

83
7. FACTORS THAT CAN INFLUENCE THE MG1

RESOURCE CLASSIFICATION

This chapter describes the main influencing factors that can affect the MG1
resource classification practices that are currently employed by MCo.

7.1 Non-numeric Factors Relevant to Mineral Resource

Classification

Mineral Resource classification at MCo is a function of the prescribed search radii

associated with the resource categories and the adeptness of the CP to adjust the
resource class boundaries based on their personal knowledge and experience. In
the research process four additional factors that can affect the resource
classification boundaries have been identified. Three of these are inherent to the
MG1 Chromitite deposition, all three require additional drilling information to
confirm interpretation confidence and the fourth is related to the value of historic
information. The factors are:

• Areas of geological complexity: usually associated with Potholes Faults,

Dykes and IRUPs, modelling of the size and shape thereof can have an
influence on the estimated resource and reserve tonnages.

• Grade continuity and apparent outlier values: the spatial positions of outlier
intersections can impose, uncertainty particularly in the areas of their
occurrence within the resource.

• Sudden changes in the dip regime or seam thickness of the orebody: often
associated with geological structures such as Potholes confirm can influence
confidence in the modelling and the orebody delineation and consequently
the resource classification .

• Historical mining knowledge of the Mineral Resource: gathering available

information, of known underground or opencast mining, and applying this
knowledge to the resource classification can lead to the increase in
confidence or reduction of uncertainty in the three topics discussed above.

84
Potholes are the most unpredictable geological structures due to the fact that
these vary in size, dimension, including direction and can be located anywhere
within the chromitite borehole intersection grid. The best method for confirmation
of pothole delineation is a combination of underground workings, underground and
exploration drilling and/or seismic information (if available).

The bulk of the resource area often only has exploration drilling available from
which predictions of potholes can be projected. Pothole structures are usually
thinning or increasing seam thickness, other indicators are the reduction of internal
layering with regards to the overlying Chromitite layer (MG2), outlier grade trends,
increased dip regimes and increasing internal waste layering within the MG1
chromitite layer. MCo generally use a ±30m radius around a borehole intersection
suspected to have intersected a pothole structure when only exploration data is
available, and as additional drilling is carried out and information becomes
available the structure delineation becomes more detailed.

The information made available from MCo’s borehole database, included

previously estimated potholes plotted on structural maps, underground mappings
and dip regime wireframes, that were used to construct a possible pothole
influence zone model. The current pothole map was scrutinised and for the
purposes of this research, a pothole influence model was created (Figure 35). The
model was derived from Leapfrog Geo® numeric RBF indicator interpolation
functionality (Leapfrog, n.d.).Construction of the model is based on borehole
intersections that have been declared as potholed intersections, and further refined
based upon underground mappings, the MG1 seam dip regime, decreased
interseam thickness, and thinning/increasing seam thickness.

85
Figure 35: Possible pothole influence zones in the project area.

MCo does account for potholes, faults, dykes and IRUPs (when and where
encountered), with a the application of a geological loss factor, to mitigate the effect
these structures on the estimated tonnage of the declared Mineral Resource
statement.

However, about ±80% of the possible pothole influence zones are within the
Measured category of the resource. A question that can be raised is whether there
is adequate data and geological knowledge/confidence available to support and
justify the declaration of all of these areas as Measured. Furthermore, bearing in
mind that high confidence does not mean high grade. The potholes will have an
influence and effect on the reserve classification and consequently on the mine
planning. Another point of observation that is debatable, is the allocated ±30m
radius applied around pothole intersected boreholes in sparsely drilled areas, as
in some areas this pothole prediction may not be applicable and the size could be
increased.

86
7.2 Borehole Radii Approach in Resource Classification

The guidance that the SAMREC code (2016) provides for declaring a Measured
Mineral Resource are the following :

“A Measured Mineral Resource requires that the nature, quality,

amount and distribution of data are such as to leave the Competent
Person with no reasonable doubt that the tonnage and grade of the
Mineralisation can be estimated to within close limits and any variation
within these limits would not materially affect the economics of
extraction. This category requires a high level of confidence in, and
understanding of, the geology and the controls on mineralisation”.

“A Measured Mineral Resource estimate should be of sufficient quality

to support detailed technical and economic studies leading to Mineral
Reserves which can serve as the basis for major development
decisions”.

Geological influences, such as potholes do have an effect on the economics of ore

extraction, as it will potentially amount to a higher geological loss or overly diluted
grade of the resource, which results in areas that are economically unmineable.
However, the more prominent point to be observed is the effect that geological
continuity can have on resource classification , if the classification method used,
does not adequately cater for geological or grade continuity between data points.

Technically it is more difficult to account for geological continuity than for chromitite
grade continuity and more reliance would be on the judgement of the CP based
his/her experience and interpretation of the data available. Although MCo’s
standard practice allow room for the CP to exert their own judgement when making
informed decisions in areas that do not conform to the general standard resource
classification, there is still a risk of only accepting the pre-determined classification
radii, if not thoroughly interpreted or reviewed. The risk is that the Resource
classification can result in a “spotted dog” effect (Figure 36).

As early as 2006, Stephenson, et al. (2006) stressed the importance recognising

the geological and grade continuity in Mineral Resource classification:

“ Spotted-dog resource classifications should be unacceptable to both

geologists and mining engineers. A final smoothing step to produce
resource classifications commensurate with a realistic level of

87
 geological and grade estimation confidence should be seen as a
necessary and integral part of the resource estimation process”.

Stephenson, et al.(2006) further argue that “spotted dog” classification may be

contradictory to reporting codes and mentions the SAMREC code, and further
emphasised that:
“all of these standards discuss continuity of geology and grade in terms
of drillholes plural, implying correlation between drillholes, not around
individual drillholes”.

The SAMREC code (2016), states that for a Measured Resource,

“geological evidence is derived from detailed and reliable exploration,

sampling and testing and is sufficient to confirm geological and grade
or quality continuity between points of observation”.

Figure 36: Example of a “spotted dog” – what not to do (Guidelines Review

Committee (JORC), 2014).

The JORC Guidelines Review Committee (Guidelines Review Commitee (JORC),

2014), in the Australian Guidelines For The Estimation and Classification of Coal
Resources considers the “spotted dog”, as a poor practice:

“… a Resource estimate classification which displays the poor practice

of estimating Measured, Indicated and Inferred Resources over
disconnected circles of influence around individual Points of
Observation or along a line of Points of Observation”.

88
This concern has been raised by other authors such as Nengovhela (2018), Lewis
(2018) and Stephenson, et al. (2006).

Nengovhela (2018) emphasises the risk of practitioners applying default software

suggestions for resource confidence and this has led to less focus being put on the
importance of geology in Mineral Resource classification:

“The advent of computers for geological modelling and the increased

application of geostatistics in estimating Mineral Resources has seen
most practitioners limiting the assignment of confidence to the
suggestions made by software package algorithms. This has further
seen a reduced focus on the impact geology has on the confidence of
an estimated block, which is something that industry needs to
address”.

Although MCo does not make use of geostatistics to determine confidence, the
underlying issue is associated with a set methodology or recipe, that could deter
focus on interpretation and impact of geology on the confidence of a Mineral
Resource, if not properly addressed by the CP.

Lewis (2018): Highlights consequences of ignoring geological continuity in using

radii around borehole intersections and questions the credibility of those who claim
competence in doing so. Lewis (2018) also cautions against this practices as it
could have detrimental effects on Mineral Reserves estimates:

“A ‘Spotted Dog’ output is potentially misleading since fundamental

requirements, such as continuity of geology and mineralization
between drill holes, and the imprecise nature resource estimation,
have been ignored. As such, it undermines the credibility of the
responsible QP’s (Qualified Person) claim to competence and relevant
experience in the commodity or type of deposit”.

Lewis (2018), further argues that:

“… mis-classified resource estimates can subsequently cause

substantial problems for engineers undertaking a mine design and
applying modifying factors to the resource estimates to produce
Mineral Reserve estimates”.

MCo’s resource classification approach does not lead to a “spotted dog” effect
where close drillhole spacing, allows for the overlapping of classification radii,

89
however areas where spacing increases, the effects of the “spotted dog” is present.
Furthermore, the radii approach for classification, cannot and does not cater for
the impact of geological continuity (i.e. pothole influence zones or geological
complexities), and for this it would have to rely on the experiential judgement and
interpretation of the CP.

Resource classification based upon a minimum borehole spacing based on coal

quality data, is a common practice in Coal Resource reporting based upon the
SANS 10320:2004 (Figure 37), in conjunction with the SAMREC Code.

Figure 37: Minimum borehole spacing for each Coal Resource classification
category for the various South-African coal deposit types (SANS 10320:2004 cited
in Hancox & Pinheiro, 2017)

90
Stephenson et al.(2006) highlights that

“Competent Persons for Mineral Resources must keep in mind the

purpose of their work and should use their experience and judgment to
avoid or smooth out “spotted dog” classifications, providing a result
commensurate with the level of geological and resource”.

An example of an Exarro Mineral Resource statement of classifying Coal resource

based upon the SANS 10320:2004 classification guidelines, where a Competent
Person’s judgement was used and mitigated the ‘spotted’ dog effect can be seen
in Figure 38. Exarro also uses Geovia MinexTM for Resource Estimation and
Classification, with a distance-gridding functionality, that is adapted with hand-
drawn polygons to adhere to the CP’s understanding of the geological continuity
for classification (Exxaro, 2015).

Figure 38: Example of a Coal Resource classification based upon SANS

10320:2004 (Exxaro, 2015).

The functionality of the borehole distance gridding in Geovia MinexTM can be best
described by GEOVIA (2014) and the use thereof for Mineral Resource
classification by Mitra (2017).

The borehole distance gridding functionality creates distance contours around

individual boreholes, based upon specified distances. An example is shown in

91
Figure 39 of three increasing distances (red; green and blue), around individual
boreholes. In the event that such criteria based on radii are accepted for a Mineral
Resource classification, the classification can easily start representing a “spotted
dog” effect.

Figure 39: Example of the borehole distance gridding functionality in Geovia

MinexTM (GEOVIA, 2014).

92
8. APPROPRIATENESS OF MCO’S MINERAL RESOURCE

CLASSIFICATION

There are various Resource classification techniques that are globally used by
Resource estimation practitioners, some of the more commonly techniques used
were discussed in section 2.5.1 of this research report.

The use of point data density based upon a borehole intersection spacing or radii
criteria, in conjunction with geological continuity (considering complex geological
areas) for classifying Mineral Resources, is not uncommon in the Chrome industry.
A recent Mineral Resources and Mineral Reserves statement of using point data
spacing criteria in Western Bushveld Complex for MG1 was observed in Merafe
Resources (2018). Sinclair & Blackwell (2002) considers the “distance from a
sample site” method as inadequate, if not used in conjunction with other
parameters and leads to cylindrical shapes that are difficult or impractical to use in
practice.

In other chromitite Mineral Resources and Mineral Reserves statements, such as

African Rainbow Minerals (ARM) (2016), Resource classification was conducted
by considering geological and geostatistical parameters. Mining companies, such
as Anglo American have entrenched a scorecard based approach to classify
Mineral Resources, incorporating a range of factors that can influence uncertainty
and confidence (Anglo American, 2017).

8.1 Classification Based on the “two thirds” of the Range Method

The “two-thirds” range of the variogram classification technique mentioned in

section 5.2.1, is visually compared to MCo’s resource classification approach in
Figure 40. The “two-third” technique was applied solely on the variogram ranges
and did not account for geological complexities, neither were manual (hand-made)
adjustments made.

93
Figure 40: Visual comparison between the “two thirds” (left) and MCo’s (right)
Resource classification approaches.

Although the “two thirds” technique would require some manual adjustments, it can
be concluded that MCo’s resource classification approach is more conservative,
however it is considered to be appropriate.

Based on the critique by Stephenson et al.(2006) and Sinclair & Blackwell (2002),
it would be however more appropriate and practical to manually override the radii
perimeters to produce a classification that is more similar to the Exarro (2015)
approach shown in the example Figure 38, where a more realistic interpretations
of the geological and grade continuity have been incorporated in the classification
boundaries.

8.2 Classification Based on Kriging Input and Output Parameters

Using the input parameters and output measures of a Geostatistical estimation

process have become common in classifying Mineral Resources, as these assist
with providing an indication of uncertainty associated with an estimate, which can
be difficult to determine solely by the CP’s interpretation of geological and grade
continuity.

Using the kriging input and output measures also creates a sense of
standardisation in mining company’s resource classification practices, as the
interpretations can be documented and aim to adhere to the transparency of
Mineral Resource reporting. However geostatistical input parameters, including

94
output measures are still just estimates and cannot replace the final judgement,
interpretation and experience of the CP or estimator. Hence all these
factors/measures should be combined when classifying Mineral Resources.

The basis of geological influence is the most important and generally takes primacy
over Mineral Resource confidence that is based on mathematical indicators
(Snowden, 2001). Glacken & Snowden (2001) highlight that resource classification
should in most cases be a combination of criteria, numerical and geological
measures, with a manual interpretation and an override by the CP on the final
classification.

8.2.1 Kriging Variance, Kriging Efficiency and Slope of regression

The advantage that a Kriging estimate has over non-geostatistical estimations is

that it provides a measure of quality of the estimation, providing a minimum error
variance, that is known as the Kriging Variance (KV). Snowden (1996) explains
that the “kriging variance represents the expected value of the squared error
between the actual grade and the estimated grade”.

The KV can be useful as criteria for resource classification, as it is dependent on

the sample configuration and continuity (the semivariogram) associated with an
estimated block, however independent of the actual grade estimates (Glacken &
Snowden, 2001).

There are numerous ways of using the Kriging variance for resource classification
that has been proposed by various authors such as Arik (2002) , Emery, et al.
(2004) and Silva & Boisvert (2014). Dohm (2004) explains that the KV can be
calculated without producing an estimate, as the KV calculation only depends on
the semivariogram model, the sampling grid and block configuration of the block to
be estimated. For this reason the author did not find the use of the KV in resource
classification surprising and provides a reasoning for the popularity of this practice:

“Measured Mineral Resources arise from interpolated blocks, which

have lower kriging variances and therefore higher confidence
associated with them. Indicated Mineral Resources occur when blocks
are extrapolated, this means that these blocks have higher kriging
variances and thus a reduced confidence is associated with them. Any
blocks extrapolated beyond the range of influence of the variogram are
classified as Inferred blocks”.

95
Arik (2002) and Mwasinga (2001) suggests a basic method to establish the
resource classification classes by using the cumulative histogram or cumulative
probability (cdf) of the KV, to derive classes based on visual inspection or
predefined criteria. Major changes in the shape of the histogram and cdf can be
used a guide to establish the different classes of confidence. Arik (2002) though
suggest calibrating the categories based on production data (if available) or
previous experience with regards to the orebody.

The cdf of the KV values for the OK_1 estimate was used to establish resource
classification thresholds, using the above methodology. The cumulative log
probability graph of the KV is show in Figure 41 and the thresholds identified in
Figure 41 are presented in Table 15.

Figure 41: Resource classification criteria based upon the cumulative log
probability of the KV values of OK_1.

Table 15: OK_1 Kriging variance classification criteria based on Figure 41.

Resource classification Kriging variance

MEASURED < 0.6
INDICATED 0.6 - 0.9
INFFERED >0.9

96
The resource classification criteria based the OK_1 estimation KV (Table 15) is
compared to MCo’s resource classification perimeters in Figure 42.

Mining company’s Resource

class perimeters

Figure 42: Resource classification criteria based on KV(left) compared to that of

MCo (right) resource classification perimeters.

Based on visual observation (Figure42) the resource classification using the KV,
measured category spatially overlaps relatively well with the measured category
(white perimeter outline) of MCo. However, due to the nature of KV, that relies the
sample configuration artefacts are created on the edges of the mining perimeter
for the inferred category, requiring some manual adjustments, should this
methodology be applied.

An additional criterion was introduced to the criteria specified in Table 15, added
incorporated the range of influence of the variogram. If the block estimates were
within the range of influence of the variogram it would be considered to be
classified as Indicated, any estimate beyond the variogram range and a KV value
more than 0.9, was classified as inferred. This resource classification comparison
with the resource classification perimeters is presented in Figure 43.

97
 Mining company’s Resource
class perimeters

Figure 43: Resource classification criteria based on KV and variogram range of

influence compared to the MCo resource classification perimeters.

Generally, the incorporation of the variogram range of influence, together with the
KV produced a spatially more realistic resource classification result, and closely
resembles the underlying MCo resource classification.

Based on some 70 cases, of various ranges of spatial and data patterns (Figure
44), Krige (1996) concluded that in this instance a correlation (87.5%) between the
Kriging Efficiency and Slope of Regression was observed. Furthermore that poor
efficiencies were associated with poor spatial structures, including limited number
of samples are available for estimation.

Figure 44: Correlation between efficiencies of block valuations and slopes of

regression (Krige, 1996).

98
Glacken & Snowden (2001) mention that the Slope of regression and Kriging
Efficiency can be used for Resource classification as a numerical approach as
proposed by Krige (1996).

The Kriging Efficiency (KE) is measure that represent the robustness and
effectiveness of the kriging estimates relating to the true block values. KE is directly
related to the KV, and a perfect estimate will be associated with values of 1 (or
100%) and 0 for a poor estimate (De-Vitry, 2003). The slope of regression (SOR)
is a measure of conditional bias with regards true values and estimated kriged
values (Deutsch, et al., 2014). The slope of regression is an indication of how good
a kriged estimation is, and a perfect estimate should have a value close to 1
(Mwasinga, 2001).

Deutsch, et al. (2014) explains the relationship between KE and KV:

“A high efficiency means that the kriging variance is low, and the
variance of the block estimates is approximately equal to the variance
of the true block values. A low efficiency implies a high kriging
variance relative to the block variance. The kriging variance varies
from block to block, so the kriging efficiency will vary as well”.

KV, KE and SOR can be used as measures asses the quality of an kriging estimate
(Deutsch & Deutsch, 2012 and Amwaama, 2018), although slope of regression
maps may be more useful (Vann, et al., 2003).

Clark (2015) highlights that Kriging efficiency can be negative, and refers to Krige
(1996), and that in such cases the Mineral Resource should be classified as
Inferred. Clark (2015), also states that

“… as a general rule, any block with negative kriging efficiency should

never be included in a Measured Resource category”.

99
Mwasinga (2001) recommends using Kriging efficiency margins as ratios for
Mineral resources classification (Table 16) and also mentions that based on
personal communication with Krige, that the Indicated and Inferred categories can
be relaxed, depending on observed geological continuity.

Table 16: Mineral Resource classification categories based on Kriging efficiency

ratios (Mwasinga,2001) .

Mwasinga (2001) also presents other common criteria that are used for
classification of coal resources in Table 17, these incorporate drillhole spacing;
variogram range; KE ratio and SOR.

Table 17: Resource classification criteria commonly used for coal resources
(Mwasinga,2001)

Kriging
Drill Hole Slope of
Tool Variogram range efficiency
Spacing Regression
Ratio

Blocks within sampled area

and using range of
Measured 250m x 250m 0.5 0.95
influence (66% of the
variogram range)

Blocks within sampled area

but beyond range of
Indicated 350m x 350m 0.3-0.5 0.80-0.95
influence (34% of the
variogram range)

Blocks within deposit but

Inferred 500m x 500m remote from data <0.3 <0.80
(extrapolated blocks)

The Kriging efficiencies and slope of regressions obtained from the OK_1 estimate
were calibrated to resemble MCo’s resource classification perimeters (Figure 45)

100
based upon the criteria presented in Table 18. No manual adjustments were made
to the classification, and geological complexities were also not considered.

Table 18: Resource classification using KE ratio and SOR that closely resemble
MCo’s resource classification perimeters.

Resource classification Kriging efficiency ratio Slope of Regression

MEASURED > 0.4 > 0.8
INDICATED 0 - 0.4 0.5 - 0.8
INFFERED <0 < 0.5

Mining company’s
North Resource perimeters

Figure 45 : Resource classification criteria specified in Table 18 compared to

MCo's resource classification perimeters.

The visual observation of the comparison of using the criteria specified in Table
18, and MCo’s approach, for resource classification are similar. However, due to
MCo's radii approach, it is not surprising to observe slight differences. The
geostatistical parameters are not bound by a fixed borehole intersection radii and
represent the continuity between borehole intersections more realistically. The
MCO classification reflect the “spotted dog” effect, especially where there is a
decrease in sample data, however, should manual over-rides be conducted the
MCo classification criteria could be considered appropriate when compared to

101
criteria used in Table 18. It is also observed that MCo classification criteria
(specified radii), is not as effective in classifying the Inferred category, as the KE
and SOR criteria.

The classification criteria used in Table 18, is plotted on Krige’s (1996), Kriging
efficiency and slope of regression correlation graph (Figure 46), and although
debatable, it does present an acceptable representation of confidence, based on
the underlying spatial correlation of the sample configuration at hand.

Figure 46 : Correlation between Kriging Efficiency and slope of regression, with

calibrated resource classification criteria based on Table 18 (modified after Krige
(1996)).

8.2.2 Resource Classification Comparison Results

The results of each comparative Mineral Resource classification method is

presented in Figure 47, based upon a volume percentage of each classification
category (Measured, Indicated & Inferred). The Inferred and Measured
classification volumes were adjusted for the comparative classification methods to
incorporate the manual interpretations of MCo.

102
Figure 47: Comparative Mineral resource classification methods, compared to
MCo’s resource classification perimeters based on volume percentage.

It has been established that the “two thirds” classification method would not be
applicable for the sampling configuration as the method does have a tendency to
overstate the Measured category, when only relying on 66% of the spatial
correlation of the Cr2O3% values. However, the two thirds method presents a more
realistic approach towards the Inferred category, when criteria on the estimates
being beyond the range of variogram are considered.

Although the KV and KV_2 classification methods do not differ much in terms of
percentage volume of classification with regards to the “two thirds” method, the
allocation of resource categories is spatially more realistic, especially when adding
the component of the range of influence of the variogram. The KE_SOR method of
Krige classification is closely related to MCo’s percentage volume towards
resource classification, however a slight increase is present for the Measured and
Inferred categories, due to the fact that the criteria do not rely on a radii of borehole
intersections. In fact, none of the alternative methods rely on specific dimension of
borehole intersection radii, and the “gaps” that are created by the “spotted dog”
effect is not present, which therefore will result in increases of the Measured
category and decrease in the Indicated category. Resulting in more tonnage being
allocated to the Measured category, and less tonnage towards the Indicated
category, representing a more realistic approach recognising geological and grade
continuity.

103
8.3 Principles of a Mineral Resource classification Scorecard-based

Approach

MCo’s current resource classification method is dependent on drill hole spacing

and is associated with distance radii around validated chromitite borehole
intersections with assayed Cr2O3% grades Measured (150m); Indicated(300m) and
Inferred (600m) . The predetermined radii were designed to cater for geological
and grade continuity with regards to confidence, in addition the CP or estimator
can practice his/her own judgement for manual adjustments or reducing the drill
spacing margin used for classification at own discretion.

However, it appears that MCo’s classification method lacks in quantifying the

confidence of geological influence in a formal procedure. This can be highlighted
with regards to the example of the possible “pothole influence zones” indicated in
Section 7.1, that could possibly have an effect on the Mineral Resource
classification categories. Furthermore, criteria measuring the estimation
confidence associated with the Growth technique estimates, are cannot bet
considered in MCo’s Mineral Resource classification procedure. Due to the fact
that the Growth technique is devoid of estimation confidence statistics, analogous
to those available in geostatistical methods, and discussed in Section 8.2.
However, there could be other factors besides geostatistical measures of
confidence, and geological complexities, that should be incorporated to define the
Mineral Resource classification categories.

Mining companies have implemented scorecard approaches with regards to

classifying Mineral Resources, as such an approach is transparent, defendable
and robust methodology, it incorporates qualitative and quantitative attributes of
various elements used to measure resource confidence. Examples of large mining
companies that have adopted a Mineral Resource classification scorecard
approach are Anglo American and De Beers, that are discussed in Mohanlal &
Stevenson (2010) and Duggan , et al. (2017).

Mohanlal & Stevenson (2010) highlights that

“the scorecard approach to the classification of mineral resources
ensures compliance and provides assurance that the CP has taken
due consideration of all pertinent factors when assigning a resource
classification”.

104
Parker & Dohm (2014) recommend key factors, that can be considered for the use
of a classification scorecard approach, although not all factors could be applicable,
depending on the deposit. The key factors are listed below:
• Geometry of the orebody : Geological confidence (structural interpretation,
geological loss, structural complexities, etc.); drilling method; confidence in
survey data (collar and down-hole); confidence in geological logging; drillhole
spacing.

• Data integrity: sampling and analytical data ; QA/QC; data security.

• Spatial correlation: considering variography and quality of variogram

models.

• Estimation methodology: estimation confidence (KE, SOR and KV);

validation techniques.

• Bulk density: Method used for determination; estimation method applied.

• Other factors: mineralogy; geometallurgical.

The process of using a scorecard approach, based on the key factors above is
explained consistent with Parker & Dohm (2014):

Factors are assigned non-linear scores, as specific weightings, according to

importance that relates to the orebody. The scores for example can be low (1),
medium (3) or high (5), and each factor is then multiplied by a rating of confidence
no confidence (zero) to high confidence (five). The total weighted score for all
factors, that were incorporated and the importance weighting factor, is then applied
onto a block model. Perimeter strings can be used to constrain specific areas of
confidence of each factor used. The weighted score can then be visualised and
margins can be assigned to classify the Mineral Resource as Measured, Indicated
or Inferred. Each classification category can be smoothed if required to produce
the final Mineral Resource classification allocations.

MCo take cognisance and ensure validation of several factors listed by Parker &
Dohm (2014), in accordance with the guidance of the SAMREC(2016) code.
However the factors are not formally part of the classification procedure or
quantified based on confidence. The main and only factor that is quantified to some
extent is the drillhole spacing, based on MCo’s radii approach. Other factors
relating to the confidence of the geometry of the orebody; data integrity ; estimation

105
methodology and bulk density is assumed to be accounted for through validation
processes, standard compliant procedures, audits, inputs from geology
department and the expert judgement of the CP. The Growth technique that is used
by MCo for estimation, does not require variogram analysis, which contributes to a
spatial correlation factor that can be used in a scorecard approach. The same
accounts for Ordinary Kriging output perimeters such as KV, KE and SOR, which
are not available for the Growth technique, that can form part of an estimation
methodology confidence factor.

Other factors, such as geometallurgical factors, could possibly also have an

influence on the resource classification, with regards to the confidence of Friability
index of the MG1 chromitite resource that is fed to the plant. As mentioned, factors
that can influence the confidence of the Mineral Resource, should be applicable to
the orebody and adopted to the Mineral Resource practises currently in place.

It is evident, based upon the analysis and research in this report, that MCo can
improve and optimize their current resource classification methodology. The
researcher believes that a customized scorecard based approach would lead to a
more structured, defendable, and transparent approach. Furthermore, a scorecard
approach should normally contain a few non-negotiable factors, that would require
inputs from geoscience experts (geologists), that should be formally drafted and
confidence relating to various factors plotted for visible assessment.

106
9. CONCLUSIONS

It appears that the Growth technique performs similar (within comparable margin)
to the alternative estimation techniques where there is an adequate amount of
sample data. This can be attributed to the fact that the grade distribution of Cr2O3%
is generally homogenous and close to a gaussian (normal) distribution and
associated with a low nugget effect. However, in areas where the sample data
spacing increase, the Growth technique tends to “smear” and inflate or exaggerate
grade values, and alternative methods such as IDW2 and OK_1 performed better.

The Growth technique and IDW4 method estimates showed similar results , this
can to some extent be attributed to the fact that an IDW4 method is similar to a
nearest neighbourhood method. It is therefore be concluded that other variables
(SG; Channel Width; SiO2% and FeO%) estimated using the Growth technique will
reveal similar tendencies, if compared to the alternative estimation methods.
According to MCo’s standard, alternative estimation methods can be motivated for,
however the nearest neighbour method cannot be used for Mineral Resource
estimation.

Regarding MG1 resource estimation, the findings in this research report regarding
the Growth technique do not conclude that the estimation method is inappropriate
for the MG1 orebody, however, the CP and resource practitioners should be aware
of the disadvantages of the technique when applied to specific sampling
configurations. The major disadvantages of the Growth technique are listed below:

• If not mitigated, the Growth technique tends to “smear” or exaggerate grade

values in areas that sample spacing increases. Subsequently, where the
“smearing” effect is present, it can affect the reported grades of the Mineral
Resource categories can be affected ad for example observed in the Inferred
category of the MG1 orebody.

This characteristic have also been observed when using the IDW4 method,
that are almost similar to a nearest neighbour method and tends to mimic the
Growth technique. The IDW2 and OK_1 estimation scenarios were able to
mitigate this effect, by gradually smoothing grades estimates.

• Similar to IDW2 and IDW4 estimation method the Growth technique also fails
to provide a measure for grade estimation errors or any quantification of the
quality of estimation that OK provides. Kriging output parameters such as

107
 KV, KE, and SOR can be used in this regard by providing measures of for
Mineral Resource estimation confidence and thus resource classification.

• The Growth technique does not take into account spatial variability or
correlation of variables that are estimated, compared to OK estimates that
are based on a variogram model of this spatial variability. It is however
accepted that the variograms produced in the research project might be
considered as being subjective, due to a lack of closely spaced sampling
data. Nonetheless the variograms were able to portray a good representation
of the range of influence of spatial correlation. The same disadvantage is true
for the IDW2 and IDW4 estimation methods.

The value in applying alternative estimation methods such as IDW and OK, is that
they can serve as global and local validation tools for the Growth technique. This
could result in more realistic estimates by adjustment of perimeters that are used
for the Growth technique and mitigate the exaggeration of grade estimates as
identified by IDW and OK estimates. Ordinary Kriging also has the ability to
determine an appropriate search neighbourhood, and as mentioned can assist with
assessment of the quality or error of estimates, that the Growth technique lacks.

Kriging, is however a more complex estimation method, and realistic or accurate

estimates rely significantly on meaningful variogram models and an appropriate
search neighbourhood. Although the Growth technique has drawbacks, similar to
IDW methods, these are easier techniques to apply, and there is less room for error
with regards to the estimation procedure.

It is difficult to reach a conclusion regarding whether the estimation results using

IDW and OK methods are more realistic than the Growth technique, as the
estimation methods performed fairly similar on a global scale. However, estimates
of the Growth technique can be argued to be subjective, on a local scale, where
sample spacing increases, or associated with outlying lower or higher values.
Therefore, it is concluded that on a local scale and under certain sampling
configuration conditions, the alternative estimation methods do appear to be a
more realistic estimation approaches.

Despite the limited amount of closely spaced data for Cr2O3%, Channel width and
SG variography, the semivariogram models, including deflection results, did reveal
spatial variability that can be considered realistic in this depositional environment.
The variogram models did also assist in providing ranges of influence for the spatial

108
correlation of these variables. More data is however required, and the data base
can be increased through the use of underground sampling programs or using
additional data from MCo’s surrounding adjacent mining sites; continuity is not
affected by manmade lease boundaries; to further refine the variography and
improve the estimation of a nugget effect.

By using geostatistical Mineral Resource classification methods, it was concluded

that the underlying radii distances used for MCo’s resource classification
categories could be considered to be applicable.

However, the circular approach of borehole radii, results in a “spotted dog” effect
where sample spacing increases, and the KV_2 approach, represented a more
realistic and practical approach with regards to the classification categories.

Analysis of possible “pothole influence zones”, revealed that a more detailed and
rigorous geological approach be applied when allocating mineral resource
categories, as the standard practice of MCo could result in overruling or ignoring
the importance of the influence that geological factors can have on classification
categories. The research report has shown that there is room for improvement and
optimisation with regards to MCo’s Mineral Resource classification procedure.

By comparing in principle, MCo’s classification methodology to a scorecard

approach, it is evident that an oversight of relevant factors can easily happen in
allocating Mineral Resource classification categories. Therefore, the researcher is
of the option that the development of scorecard approach would add significant
benefits, that would result in a more realistic, defendable and transparent Mineral
Resource classification methodology than the currently implemented method.

The only approach to come close to having an idea of how accurate or realistic
MCo’s or the alternative estimation methods are, would be to reconcile the Mineral
Reserves, to the Run-of-mine tonnages and grade over at least a six month period
of production. This process should also incorporate local grade control data and
block models, and then derived back to the in-situ Mineral Resource estimation
grades and tonnage. Currently MCo does not have an established grade control
system in place, that contains validated sample assay data, and due to
confidentially, run-of-mine data, was not made available for study in this research
project.

109
10. RECOMMENDATIONS

• Implement a field QA/QC procedure for samples that are requested from

analysis, additional to the current internal laboratory procedures. This will

ensure more confidence and control of the quality of the sample data that is

used in Mineral Resource estimation.

• Use parameters (if possible), that could mitigate the “smearing’’ effect of the

Growth technique, especially where sample spacing increases or associated

with high or low outlying values. The impact of using capping high outlying

values could also be considered, to conform values to the general population

of value distribution.

• Further analysis is required with regards to variography, to improve the

understanding of the spatial variability associated with the relevant variables

that are used for Mineral Resource estimation. Sampling information from

underground and adjacent mine sites of MCo can assist with the expansion

of the data base, and this is expected to improve variogram models.

• Mitigate the “spotted dog” effect where present by manual overrides to have

a more realistic representation of the geological and grade continuity in terms

of resource classification. Furthermore, this will also result in a more practical

block model or gridded mesh, with regards to classification of Mineral

Reserves and application in the mine value chain.

• Consider geostatistical criteria such as KV, KE, SOR and range of influence

of the variogram, as guidelines to further refine the resource categories,

when using a borehole radii approach as it is currently implemented. This will

ensure that an statistical element of estimation confidence and spatial

variability is accounted for with regards to Mineral Resource classification.

110
 KV is also a useful tool to use to indicate requirements of infill drilling to

increase confidence, as it incorporates both the drillhole sample spacing and

the range of influence of the variogram.

• Geological influence zones should be analysed in more detail and more

rigorously incorporated into the resource classification process.

• Developing and incorporating a Mineral Resource classification scorecard

approach, that incorporates data integrity, geological, spatial correlation,

statistical/geostatistical and other orebody specific confidence factors, will

improve and optimise the current Mineral Resource classification

methodology.

111
REFERENCES

African Rainbow Minerals (ARM), 2008. Mineral Reserves and Resources.

[Online] Available at:
https://www.arm.co.za/im/files/annual/2008/files/ARM_RR2008.pdf [Accessed
27 September 2019].

African Rainbow Minerals (ARM), 2016. Mineral Resources and Mineral

Reserves. [Online] Available at: https://www.arm.co.za/im/files/annual/2016/ARM-
RR16.pdf [Accessed 19 11 2019].

Anglo American, 2017. Ore reserves and Mineral resources 2017. [Online]
Available at: https://www.angloamerican.com/~/media/Files/A/Anglo-American-
PLC-V2/documents/annual-updates-2018/aa-ore-reserves-and-mineral-
resources-2017.pdf [Accessed 30 12 2019].

Arik, A., 2002. Comparison of resource classification methodologies with a new

approach. Phoenix,Arizona, In APCOM 2002: 30th International Symposium on
the Application of Computers and Operations Research in the Mineral Industry,
pp. 57-64.

Assibey-Bonsu, W., 2016. The basic tenets of evaluating the Mineral Resource
assets of mining companies, as observed through Professor Danie Krige’s
pioneering work over half a century. The Journal of the Southern Africa Institute
of Mining and Metalurgy, Volume 116, pp. 635-643.

Awaama, M. M., 2018. Comparison of Kriging Neighbourhood Analysis (KNA)

results from two different software packages – a case study, Johannesburg:
University of the Witwatersrand.

Babak, O. & Deutsch, C., 2008. Statistical approach to inverse distance

interpolation. Stochastic Environmental Research and Risk Assessment, 23(5),
pp. 543-553.

Batcha, J. B. & Reese, J. R., 1964. Surface determination and automatic

contouring for mineral exploration extraction and processing. Colorado School of
Minex Quarterly, Issue 59, pp. 1-14.

112
Bohling , G., 2005. Introduction to geostatistics and variogram analysis. [Online]
Available at: https://www.coursehero.com/file/12867665/Variograms/ [Accessed
24 August 2019].

Canadian Aviation Electronics(CAE), 2013. Resource Classfication Report for the

Lesedi Project , Johannesburg : CAE.

Cawthorn, G., Latypov, R. & Vuthuza, A., 2018. Origin of discordant ultramafic
pegmatites in the Bushveld Complex from externally-derived magmas. South
African Journal of Geology, 15 February, 121(3), pp. 287-310.

Cawthorn, R. G., 2010. The Platinum Group Element Deposits of the Bushveld
Complex in South Africa. Platinum Metals Rev, 54(4), pp. 205-215.

Chanderman , L., Dohm, C. & Minnitt, R., 2017. 3D geological modelling and
resource estimation for a gold deposit in Mali. The Journal of the Southern
African Institute of Mining and Metallurgy, Volume 117, pp. 189-197.

Chanderman, L., 2015. 3D geological modelling and mineral resource estimate

for the fe2 gold deposit, Sadiola mine – Mali, Johannesburg: University of the
Witwatersrand.

Cheney, E. S. & Twist, D., 1991. The conformable emplacement of the Bushveld
mafic rocks along a regional unconformity in the Transvaal succession of South-
Africa. Precambrian Research, Volume 52, pp. 155-132.

Chilès, J.-P. & Delfiner, P., 2012. Geostatistics Modeling Spatial Uncertainty. 2nd
ed. New Jersey: John Wiley & Sons,Inc..

Clark, I., 2001. Practical Geostatistics. Alloa: Geostokos Limited.

Clark, I., 2015. Regression revisited (again). The Journal of The Southern African
Institute of Mining and Metallurgy, Volume 115, pp. 45-50.

Coombes, J., 1997. Handy Hints For Variography.. In: V. N. Misra & J. S. Dunlop,
eds. Melbourne: The Australasian Institute of Mining and Metallurgy:, pp. 127-
130.

Coombes, J., 2008. The Art and Science of Resource Estimation. First ed. Perth :
Coombes Capability.

CRAIN, I. K., 1970. Computer interpolation and contouring of two dimensional

data a review. Geoexploration, Issue 8, pp. 71-86.

113
Cressie, N. A., 1993. Statistics for Spatial Data. Revised Edition ed. Newyork:
John Wiley & Sons, Inc.

Crowson, P., 2001. Minerals Handbook 2000-01: Statistics and Analyses of the
World's Minerals Industry. Edenbridge,Kent: UK: Mining Journal Books Ltd.

Dassault Systèmes GEOVIA Inc., 2016. Minex 650 usersguide. s.l.:Dassault

Systèmes GEOVIA Inc..

Dassault Systèmes, n.d. Geovia Minex™. [Online] Available at:

https://www.3ds.com/products-services/geovia/products/minex/ [Accessed 29
Septemeber 2019].

Davey, S. R., 1992. Lateral variations within the Upper Critical Zone of the
Bushveld Complex on the farm Rooikoppies 297 JQ, Marikana, South Africa.
South African Journal of Geology, 95(3-4), pp. 141-149.

de Souza, L. E., Costa, J. F. & Koppe, J. C., 2010. Comparative analysis of the
resource classification techniques: case study of the Conceic¸a˜o Mine, Brazil.
Applied Earth Science IMM Transactions section B, 119(3), pp. 166-175.

Deutsch, C. V., Leuangthong, O. & Ortiz C, J., 2006. A Case for Geometric
Criteria in Resources and Reserves Classification, Paper 301: Centre for
Computational Geostatistics ( CCG) - University of Alberta, annual report 08.

Deutsch, C. V., 2007. The Slope of Regression for Kriging Estimators, University
of Alberta : Centre for Computational Geostatistics.

Deutsch, C. V. & Rossi, M. E., 2014. Mineral Resource Estimation. Dordrecht:

Springer.

Deutsch, J. L. & Deutsch, C. V., 2012. Kriging, Stationarity and Optimal

Estimation: Measures and Suggestions, CCG Annual Report 14: Paper 306.

Deutsch, J. L., Szymanski, J. & Deutsch, C. V., 2014. Checks and measures of
performance for kriging estimates. The Journal of The Southern African Institute
of Mining and Metallurgy, Volume 114, pp. 223-230.

De-Vitry, C., 2003. Resource classification – a case study from the Joffre-hosted
iron ore of BHP Billiton’s Mount Whaleback operations. Mining Technology
(Trans. Inst. Min. Metall. A), Volume 112, pp. 185-196.

114
Dohm, C., 2004. Quantifiable Mineral resource classification: A logical approach.
In: O. Leuangthong & C. V. Deutsch, eds. Geostatistics Banff. Dordrecht, The
Netherlands: Springer, pp. 333-342.

Dohm, C. E., 2015. Geostatical methods in Mineral Resource evaluation.

Johannesburg: University of the Witwatersrand, course notes,MINN7006.

Dominy, S. C., Noppé, M. A. & Annels, A. E., 2002. Errors and uncertainty in
mineral resource and ore reserve estimation: The importance of getting it right.
Exploration and Mining Geology , 11(1-4), pp. 77-98.

Dube, Z. A., 2010. The Merensky Reef and UG2 layers at De Wildt, western limb
of the Bushveld Complex, Johannesburg: University of the Witwatersrand.

Duggan , S., Grills , A., Stiefenhofer, J. & Thurson, M., 2017. Development of a
best-practice mineral resource classification system for the De Beers group of
companies. Journal of the Southern African Institute of Mining and Metallurgy,
117(12), pp. 1127-1132.

Duggan, S., Grills, A., Stiefenhofer, J. & Thurston, M., 2017. Development of a
best-practice mineral resource classification system for the De Beers group of
companies. The Journal of the Southern Africa Institute of Mining and Metalurgy,
Volume 117, pp. 1127-1132.

Emery, X., Ortiz, J. M. & Rodriguez, J. J., 2004. Quantifying Uncertainty in

Mineral Resources with Classification Schemes and Conditional Simulations, s.l.:
Department of Mining Engineering : University of Chile.

Exxaro, 2015. Consolidated Mineral resources and ore reserves report (CMRR).
[Online] Available at:
https://www.exxaro.com/assets/files/Consolidated_Mineral_Resources_Ore_Res
erves_Report.pdf [Accessed 01 12 2019].

Gentile, M., Courbin, F. & Meylan, G., 2012. Interpolating point spread function
anisotropy. Astronomy & Astrophysics, pp. 1-20.

Geostokos (Ecosse) Limited, n.d. What is Kriging anyway?. [Online] Available at:
http://www.kriging.com/ [Accessed 05 November 2017].

GEOVIA, 2014. The GEOVIA Blog. [Online] Available at:

https://blogs.3ds.com/geovia/using-borehole-distance-gridding-minex/ [Accessed
18 11 2019].

115
Glacken , I. M. & Snowden, D. V., 2001. Mineral Resource Estimation. In: A. C.
Edwards, ed. The AusIMM Guide to Good Practice (Monograph 23). Melbourne:
The Australasian Institute of Mining and Metallurgy, pp. 189-198.

Glacken, I. & Trueman, A., 2014. Overview – Mineral Resource Estimation. The
AusIMM Guide to Good Practice, Monograph 30(2nd), pp. 263-275.

Google Maps , 2020. Google Maps. [Online] Available at:

https://www.google.com/maps/@-
25.6753781,27.4774792,55155m/data=!3m1!1e3 [Accessed 12 02 2020].

Gringarten , E. & Deutsch, C. V., 2001. Teacher's Aide Variogram Interpretation

and Modeling. International Association for Mathematical Geology, 33(4), pp.
507-534.

Guidelines Review Commitee (JORC), 2014. Australian guidelines for the

estimation and classification of coal resources. [Online] Available at:
http://www.jorc.org/docs/Coal_Guidelines_2014_-_Final_Ratified_Document.pdf
Accessed 18 11 2019].

Hancox, J. & Pinheiro, H., 2017. The new SANS 10320:2016 versus the 2014
Australian guidelines for the estimation and classification of coal resources—what
are the implications for southern African coal resource estimators?. The Journal
of the Southern African Institute of Mining and Metallurgy, Volume 117, pp. 1-8.

Hengl, T., 2007. A Practicle Guide to Geostatistical Mapping of Enviromental

Variables. Luxembourg: European Communities.

Hoffman, D., 2010. Statistical size analysis of potholes: an attempt to estimate

geological losses ahead of mining at Lonmin’s Marikana mining district. The
Southern African Institute of Mining and Metallurgy, Volume The 4th International
Platinum Conference, Platinum in transition ‘Boom or Bust’, pp. 97-104.

Hunt, J. P., 2006. Geological Characteristics of Iron Oxide-Copper-Gold (IOCG)

Type Mineralisation in the Western Bushveld Complex, Johannesburg: University
of the Witwatersrand.

Isaaks, E. H. & Srivastava, R. M., 1989. Applied Geostatistics. New York: Oxford
University Press.

116
Jones, T. A., Hamilton, D. E. & Johnson, C. R., 1986. Contouring Geologic
Surfaces With The Computer (Van Nostrand Reinhold Catalysis Series). 1987th
ed. New York: Springer.

JORC code, 2012. Australasian Code for Reporting Exploration Results, Mineral
Resources and Ore Reserves, prepared by: The Joint Ore Reserves Committee
of The Australasian Institute of Mining and Metallurgy, Australian Institute of
Geoscientists and Minerals Council of Australia : JORC.

Journel, A. G. & Huijbregts, C. J., 1978. Mining Geostatistics. London: Academic

Press.

Kinnaird, J. A., Kruger, F. J. & Cawthron, R. G., 2004. Rb-Sr and Sm-Nd isotopes
in fluorite related to the granites of the Bushveld Complex. South African Journal
of Geology, Volume 107, pp. 413-430.

Kinnaird, J. A., Kruger, F. J., Nex, P. A. & Cawthorn, R. G., 2002. Chromitites of
the Bushveld Complex - process of formation and PGE enrichment.
Johannesburg: Economic Geology Research Institute (incorporating the Hugh
Allsopp Laboratory).

Krige , D. G., 1996. A practical analysis of the effects of spatial structure and of
data available and accessed, on conditional biases in ordinary Kriging.
Wollongong-Australia, International Geostatistics Congress.

Krige, D. G., 1981. Lognormal-de Wijsian Geostatistics for Ore Evaluation. In: G.
S. Baker & D. Phil, eds. South African Institute of Mining and Metallurgy
Monograph series. 2nd ed. Johannesburg: The South African Institute of Mining
and Metallurgy, pp. 1-50.

Leapfrog3d-Mining and Minerals team, 2019. Understanding declustering in

Leapfrog Edge. [Online] Available at:
https://blog.leapfrog3d.com/2019/03/07/understanding-declustering-in-leapfrog-
edge/ [Accessed 06 11 2019].

Leapfrog, n.d. Leapfrog Geo 4.5 Help. [Online] Available at:

http://help.leapfrog3d.com/Geo/4.5/en-GB/Content/interpolants/indicator-
interpolant.htm [Accessed 17 11 2019].

117
Leuangthong, O. & Nowak, M., 2015. Dealing with high-grade data in resource
estimation. The Journal of The South African Institute of Mining and Metallurgy,
115(1), pp. 27-36.

Lewis, W., 2018. micon international limited - mineral industry consultants.

[Online]
Available at: https://www.micon-international.com/spotted-dog-still-issue-mineral-
resource-classification/ [Accessed 18 11 2019].

Lipton, I. T. & Horton, J. A., 2014. Measurement of Bulk Density for Resource
Estimation – Methods,Guidelines and Quality Control. The AusIMM Guide to
Good Practice, Monograph 30(2nd), pp. 97-108.

Lonmin Platinum, 2013. Regional Locality Plan Mining Rights. North-West

province South-Africa : Unpublished.

Maier, W. D. & Barnes, S. J., 1999. Platinum-Group Elements in Silicate Rocks of

the Lower, Critical and Main Zones at Union Section, Western Bushveld
Complex. Journal of Petrology, 40(11), pp. 1647-1671.

Makhuvha, M., Arellano, R. M. & Harney, D. M., 2014. Determination of bulk

density, methods and impacts, with a case study from Los Bronces Mine, Chile.
Applied Earth Science , 123(3), pp. 196-205.

Manyeruke, T. D., 2006. The petrography and geochemistry of the Platreef on the
farm Townlands, near Potgietersrus, northern Bushveld Complex, Pretoria:
University of Pretoria.

Matheron , G., 1963. Principles of Geostatistics. Reprinted from Economic

Geology , Volume 58, pp. 1246-1266.

Merafe Resources, 2018. Mineral resources and Mineral reserves report. [Online]
Available at: http://www.meraferesources.co.za/reports/ir_2018/pdf/appendix-
1.pdf [Accessed 19 11 2019].

Micon International , 2016. Why Mines Fail. [Online] Available at:

https://www.micon-international.com/minex/ [Accessed 6 Novemeber 2017].

Mitra, N., 2017. The GEOVIA Blog - Resource Classification in GEOVIA Minex.
[Online] Available at: https://blogs.3ds.com/geovia/resource-classification-geovia-
minex/ [Accessed 18 11 2019].

118
Mohanlal, K. & Stevenson, P., 2010. Anglo American Platinum’s approach to
resource classification case study—Boschkoppie/Styldrift minewide UG2 project.
The 4th International Platinum Conference, Platinum in transition ‘Boom or Bust’,
The Southern African Institute of Mining and Metallurgy.

Morgan, C. J., 2005. Analysing spatial data via geostatistical methods,

Johannesburg: University of the Witwatersrand.

Morgan, C. J., 2011. Theoretical and practical aspects of variography: in

particular, estimation and modelling of semi-variograms over areas of limited and
clustered or widely spaced data in a two-dimensional South African gold mining
context, Johannesburg: University of the Witwatersrand.

Morrissey, C. J., 1988. Exploration for platinum. In: H. Prichard, ed. Geo-Platinum
87. London: Elsevier Applied Science, pp. 1-12.

Mpanza, M., 2015. A comparison of ordinary and simple kriging on a PGE

resource in the Eastern limb of the Bushveld Complex, Johannesburg: The
University of the Witwatersrand.

Mwasinga, P. P., 2001. Approaching resource classification: General practices

and the integration of geostatistics. In: Xie, Wang & Jiang, eds. Computer
Applications in the Mineral Industries. Tokyo: A.A.BALKEMA PUBLISHERS, pp.
97-104.

Myers, D. E., 1994. Spatial interpolation: an overview. Geoderma, Volume 62, pp.
17-28.

Nengovhela, A. C. N., 2018. The application of geostatistics in coal estimation

and classification, Johannesburg: University of the Witwatersrand.

Njowa, G., 2008. Salient issues in the reporting of mineral resources and mineral
reserves in the vein and reef mining. The Southern African Institute of Mining and
Metallurgy, Volume Narrow Vein and Reef, pp. 1-15.

Noble , A. C., 1992. 5.6 Ore Reserve/Resource Estimation. In: H. L. Hartman, ed.
SME MINING ENGINEERING HANDBOOK. Littleton: Society for Mining,
Metallurgy, and Exploration, Inc., pp. 344-359.

Noppé, M., 2014. Communicating confidence in Mineral Resources and Mineral

Reserves. The Journal of The Southern African Institute of Mining and
Metallurgy, Volume 114, pp. 213-222.

119
Nowak, M. & Leuangthong, O., 2016. Conditional Bias in Kriging - Let's Keep It ,
Vancouver & Toronto : SRK Consulting (Canada) Inc..

Oritz, J. M. & Emery, X., 2006. Geostatistical estimation of mineral resources with
soft geological boundaries: a comparative study. The Journal of The South
African Institute of Mining and Metallurgy, Volume 106, pp. 577-584.

Parker, H. M. & Dohm, C. E., 2014. Evolution of Mineral Resource Classification

from 1980 to 2014 and Current Best Practice. [Online] Available at:
http://www.crirsco.com/docs/H_Parker_Finex.pdf [Accessed 30 12 2019].

Revuelta , M. B., 2018. Mineral Resources:From Exploration to Sustainability

Assessment. 1st ed. Madrid: Springer .

Rezaee, H., Asghari, O. & Yamamoto, J. K., 2011. On the reduction of the
ordinary kriging smoothing effect. Journal of Mining & Environment, 2(2), pp. 25-
40.

Saager, R., 2005. The Bushveld Complex in the context of the Geology of South
Africa's Kaapvaal Craton. [Online] Available at:
www.bacfund.ch/download/6thPD/Prof%20Saager%20Part.pdf [Accessed 12 10
2018].

SAMREC code, 2016. The South African Code for the reporting of exploration
results, Mineral resources and Mineral reserves, Marshalltown: Prepared by: The
South African Mineral Resource Committee (SAMREC) Working Group.

Scoon, R. N. & Teigler, B., 1994. Platinum-group element mineralization in the

critical zone of the western Bushveld Complex; 1. Sulfide poor-chromitites below
the UG-2. Economic Geology, 89(5), pp. 1094-1121.

Shepard, D., 1968. A two-dimensional interpolation function for irregularly-spaced

data. New York, ACM Press,proceedings of the 1968 ACM National Conference,
pp. 517-524.

Silva, D. S., 2015. Mineral Resource Classification and Drill Hole Optimization
Using Novel Geostatistical Algorithms with a Comparison to Traditional
Techniques, s.l.: Department of Civil and Environmental Engineering: University
of Alberta.

120
Silva, D. S. & Boisvert, J. B., 2014. Mineral resource classification: a comparison
of new and existing techniques. The Journal of The Southern African Institute of
Mining and Metallurgy, Volume 114, pp. 265-273.

Sinclair, A. J. & Blackwell, G. H., 2002. Applied Mineral Inventory Estimation.

Cambridge: Cambridge University Press.

Snowdengroup, 2017a. 4 key aspects to Variography. [Online] Available at:

https://snowdengroup.com/news/4-key-aspects-variography/ [Accessed 24
August 2019].

Snowdengroup, 2017b. Supervisor 8.7 – Multi-block kriging neighbourhood

analysis. [Online] Available at: https://snowdengroup.com/news/supervisor-8-7-
multi-block-kriging-neighbourhood-analysis/ [Accessed 28 September 2019].

Snowden, V., 1996. Practical interpretation of resource classification guidelines.

West Perth, AusIMM 1996 Annual Conference.

Snowden, V., 2001. Practical interpretation of mineral resource and ore reserve
classification guidelines. West Perth, The AusIMM Guide to Good Practice
(Monograph 23), pp. 643-652.

Spragg, K., 2013. Leapfrog3d. [Online] Available at:

https://blog.leapfrog3d.com/2013/05/08/leapfrog-interpolation-basics/ [Accessed
09 02 2020].

SRK Consulting , 2008. An Independent Mine Feasibility Study for the Sky
Chrome Project , North-West Province South- Africa: Report number 355663.

Stephenson, P. R. et al., 2006. Mineral Resource classification– it’s time to shoot

the ‘spotted dog’!. Australasian Institute of Mining and Metallurgy, pp. 91-95.

Surpac Minex Group, 2005. Klipstone. [Online] Available at:

http://www.klipstone.com.au/wp-
content/uploads/MINEX_GROWTH_METHOD_GRIDDING.pdf [Accessed 02 10
2019].

Vann, J., Jackson, S. & Bertoli, O., 2003. Quantitative Kriging Neighbourhood
Analysis for the Mining Geologist — A Description of the Method With Worked
Case Examples. Bendigo, 5th International Mining Geology Conference.

Von Gruenewaldt, G., 1977. The mineral resources of the Bushveld Complex.
Minerals Science and Engineering, 9(2), pp. 83-95.

121
Von Gruenewaldt, G., Sharpe, M. R. & Hatton, C. J., 1985. The Bushveld
Complex; introduction and review. Economic Geology, Volume 80, pp. 803-812.

Wawruch, T. M. & Betzhold, J. F., 2005. Mineral resource classification through

conditional simulation . In: Geostatistics Banff 2004. The Netherlands: Springer,
pp. 479-489.

Weber, D. & Englund, E., 1992. Evaluation and Comparison of Spatial

Interpolators. Mathematical Geology, Issue 24, pp. 381-391.

Webster, R., n.d. AMC Consultants - Mineral Resource Validation. [Online]

Available at: https://amcconsultants.com/experience/mineral-resource-validation/
[Accessed 04 11 2019].

Webster, R., nd. Mineral Resource Validation. [Online] Available at:

https://amcconsultants.com/experience/mineral-resource-validation/ [Accessed
6 July 2019].

WSP, 2015. Newfoundland and Labrador. [Online] Available at:

https://www.nr.gov.nl.ca/NR/mines/MRR2015-Lookback/WSP-
Tricks_Block_Model_NL_CIM_2015_R1.pdf [Accessed 04 11 2019].

Yamamoto, J. K., 2005. Correcting the Smoothing Effect of Ordinary Kriging

Estimates. Journal of the International Association for Mathematical
Geosciences, 37(1), pp. 69-94.

122
APPENDIX

Appendix 1: North-South and West-East swath plot comparisons between the

alternative estimation methods (IDW2;IDW4;OK_1;OK_2;OK_3) and MCo’s
Growth technique estimation for Cr2O3%.

Swath plots - Growth technique compared to IWD2

123
Swath plots - Growth technique compared to IWD4

124
Swath plots - Growth technique compared to OK_1

125
Swath plots - Growth technique compared to OK_2

126
Swath plots - Growth technique compared to OK_3

127