CHALLENGES IN MODELLING

GEOMECHANICAL HETEROGENEITY
OF ROCK MASSES USING
GEOSTATISTICAL APPROACHES
H. Eivazy, K. Esmaieli, R. Jean

H. Eivazy1, *K. Esmaieli1, R. Jean2

1
Lassonde Institute of Mining, University of Toronto, Toronto, Canada
2
ArcelorMittal Mines Canada, Montreal, Canada

170 College St., Room 129, Toronto, ON, Canada

(*Corresponding author: [email protected])

October 18-21, 2016 • Rio de Janeiro /RJ • Brazil

 SURFACE MINING |
27

CHALLENGES IN MODELLING GEOMECHANICAL HETEROGENEITY OF ROCK MASSES

USING GEOSTATISTICAL APPROACHES

ABSTRACT

Modelling rock mass geomechanical heterogeneity plays an important role in optimization of

mine planning and design. Understanding the spatial variability of rock mass engineering properties allows
for a better prediction of the rock mass behaviour during excavation (i.e., rock mass stability, blast-
induced rock mass fragmentation), and during comminution (crushing and grinding behaviour of rock).
Geostatistical techniques are commonly used for resource modelling (i.e., modelling ore grade and
tonnage). Their application for geomechanical attributes also becomes routine for many mining operations;
however, the spatial modelling of the engineering geomechanical properties has its own challenges and
difficulties. Interpretation of the geospatial model of geomechanical properties without a good
understanding of these challenges, assumptions and limitations could lead to decision- making that does
not necessarily result in appropriate consequences. This paper addresses major challenges with spatial
modelling of geomechanical attributes using the information from a case study of an open pit mine in
Quebec, Canada, where more than 200,000 m. of geotechnical borehole data, drilled and logged in the pit
area since 1949, was used for 3D block modelling of various geomechanical attributes. The properties that
are modelled include RQD, RMR, and intact rock strength. Unlike ore-grade data, some of these
geomechanical variables are non-additive and direction dependent, which makes the process of 3D block
modelling more challenging. Geomechanical data are often scattered or widely dispersed across the
orebody and are generally sensitive to core drilling technology used for rock sampling. In addition, these
properties often show complex multivariate relationships. Using simple geostatistical estimation methods,
such as kriging, for geomechanical variables creates a unique model based on the available dataset.
However, the krige model is not reliable in representing the extreme rock mass quality. In other words, the
3D block model overestimates the poor quality rock masses and underestimates the good quality rock
masses. Using conditional simulation for spatial modelling of rock mass geomechanical properties can
generate multiple equi-probable models which allow for the reduction modelling biases related to the
estimation of non-additive variables.

KEYWORDS

Geomechanical properties, 3D block model, geostatistics, heterogeneity modelling

INTRODUCTION

Design of an optimal open pit mine is a challenging problem for mine engineers due to the spatial
variations of geomechanical properties of rock masses that form the pit area. A thorough understanding of
the geomechanical and geological behaviour of the entire rock masses across the mine is crucial for optimal
mine planning and design. Understanding the heterogeneity of rock masses’ geomechanical properties
enables the mine design engineers to identify the high -risk zones and the mine sectors that require more
exploration and data collection attempts. In open pit mine projects, modelling of the geomechanical
heterogeneity can improve the reliability of pit design and prediction of the rock mass behaviours during the
excavations (i.e., rock mass stability, blast-induced rock fragmentation) and prediction of rock mass
behaviour during comminution process (identification of geometallurgical domains). For example, such
models can be considered as inputs into the blast design process in order to determine the optimal powder
factor of an individual blast required for a target fragmentation (Bye, 2011). Consequently, this would
improve the mine-to-mill value chain by enhancing the loading rates, transportation and mill throughput.
28 | 24th WORLD MINING CONGRESS PROCEEDINGS

The common practice in open pit design is to create a deterministic boundary model of the
subsurface geomechanical properties (Hack et al., 2006). In this boundary model, which is generally used in
the earlier stages of a mine design with presence of limited data, the rock mass in the mine site is divided
into a number of geotechnical domains according to the collected data of subsurface condition and
engineering judgements (expert knowledge). Basically, some structural features, such as changes in
lithology, presence of major structures and characteristics of joint sets, are considered as the boundaries of
the geotechnical domains (Read and Stacey, 2009). Each geotechnical domain is assumed to be uniform with
in which the geomechanical properties are homogeneously distributed. An average value is generally
assigned to rock mass properties of each domain, and it is assumed that the average value represents the
property across the domain. These average values of rock mass geomechanical properties are then used in
numerical models for slope stability analysis to determine the safety factor. Finally, a sensitivity analysis is
generally performed to determine the effects of each geomechanical property on the pit slope stability in
order to explore the variation of the geomechanical properties of rock masses. This conventional
deterministic method ignores the local spatial variability of the rock mass properties within each geotechnical
domain because it assumes that the geotechnical domain is homogeneous. Research indicates that predictions
made based on these conventional deterministic boundary models have a lower probability of being accurate
and the resulting design decisions are generally questionable (Hammah et al. 2009).

An alternative approach for modelling the geomechanical heterogeneity in open pit mining is to
apply geostatistical methods. Geostatistics, or theory of regionalized variables (Matheron, 1963), is a branch
of applied statistics that models a natural phenomenon by quantifying the variability of one or more of the
so-called regionalized variables (Journel and Huijbregts, 1978). In the geostatistical techniques, both random
and structured characteristics of the regionalized variables are modelled in a mathematical context.
Geostatistical techniques were originally developed for ore reserve, where the main regionalized variable of
interest is the ore grade. The early efforts of using geostatistical methods in geomechanics were done by La
Pointe (1981), who applied geostatistics to investigate the spatial distribution of fracture properties in order
to improve the design of rock structures. Since then, many researchers have applied the geostatistical
interpolation and simulation methods for analysis of spatial distribution of geomechanical properties. Some
examples are the works by Vatcher et al. (2016), Doostmohammadi et al. (2015), Ferrari et al. (2014), Egaña
and Ortiz (2013), Marache et al. (2009), Stavropoulou et al. (2007), Marinoni (2003), Ayalew et al. (2002)
and Giles (1993). The principle reason behind the slower adoption of the geostatistical techniques in the
geomechanical mine design projects can be attributed to different challenges in the application. Modelling
the spatial heterogeneity of the geomechanical properties using the geostatistical tools requires special
attention to the specifications and limitations of the geomechanical variables. A better understanding of these
challenges in the spatial modelling of such variables leads to careful selection of the spatial modelling
approach in order to achieve the most reliable representation of the deposit. Each of the challenges and
restrictions could provide some uncertainties in the process of spatial modelling of the geomechanical
variable using these geostatistical techniques.

This paper focuses on the challenges in applications of the geostatistical methods for the
heterogeneity modelling of geomechanical variables. Of interest is the Rock Mass Rating (RMR)
geomechanical classification system developed by Bieniawski (1976) . RMR is commonly used for
geomechanical characterization of rock masses and estimation of their mechanical properties. RMR is
calculated as the sum of the ratings for its constituent parameters including intact rock strength; joint spacing
(the mean spacing between discontinuities and can be calculated as the reciprocal of the fracture frequency);
Rock Quality Designation (RQD) measured as the ratio of total length of all pieces of rock core larger than
10 cm (4 inch) to the total length of core run; joint condition (a measure of joint surface roughness and
alteration, and its filling material); and water condition (a measure of ground water condition in the project
area). An iron ore open pit mine in Quebec, Canada, was used as a case study to model spatial variation of
RMR geomechanical attribute using the geostatistical simulation method. Lessons learned from this case
study were then used to discuss the major challenges with the use of geostatistical methods for modelling
geomechanical heterogeneity of rock masses.
 SURFACE MINING |
29

Geostatistical Modelling of Geomechanical Properties in Mont-Wright Open Pit Mine

Mont-Wright (MW) is a major iron ore deposit that is owned and operated by ArcelorMittal Mines
Canada (AMMC) Corporation. The mining complex is located in northern Quebec in Canada, 16 km. west
of the mining town Fermont. It is also located approximately 1,000 km northeast of Montreal. The mine
site can be accessed from the closest city in the south, Baie-Comeau, via Highway 389. Mont-Wright has
operated since 1975 and in total 1,285 million tonnes of iron ore have been extracted up to 2012 (Savard
and Jean, 2012). The iron ore concentrates of the MW are shipped to the AMMC’s pelletizing plant and
shipping terminal located at Port Cartier on the north shore of the Gulf of St. Lawrence through a 416 km
private railway owned by AMMC. Figure 1 demonstrates the Quebec regional location map of the MW
mining complex. There are several active mining regions in Mont-Wright, including Paul’s Peak, Pit A, Pit
B and C-prim. The pit of study in this paper is the Paul’s Peak pit, which is the largest open pit within the
mining complex. Figure 2 depicts the Paul’s Peak open pit looking from the east wall. Currently, the Paul’s
Peak pit extends ~1800 m of length, ~600 m of width and ~320 m of depth. The pit will be expanded to a
depth of ~600 m by year 2045.

Figure 1. Map showing the location of MW mining

Figure 2. Paul’s Peak open pit (looking
from the east) complex in Quebec-Canada

Mont-Wright iron ore deposit is a part of the highly folded and metamorphosed south-western
branch of the Labrador Trough, where the most important rock type is the specular hematite iron formation
containing recrystallized quartz and hematite with minor magnetite and iron silicates (Savard and Jean 2012)
. The complex structural geology of the Mont Wright arises from the existence of sever folding into a series
of synclines and anticlines, as well as the presence of a secondary folding that has made wide zones of
specular hematite up to 300 m in width (Savard and Jean 2012). Figure 3 shows the spatial distribution of
rock types in a plan view of the Paul’s Peak pit. The view shows a folding system in the far east of Paul’s
Peak.

Since exploration of Paul’s Peak started in 1949, 871 drill holes with a total length of more than 200
km have been drilled and logged for geological and geotechnical information. Figure 4 shows a crosssection
of the pit with the RMR logging information of some drill holes have been superimposed. The geomechanical
properties measured/assessed along the boreholes include RQD, RMR, fracture frequency, joint condition
and intact rock strength. Table 1 provides a summary of the simple statistics for the geomechanical properties
in the drill-hole dataset. This table indicates that the rock mass in the Paul’s Peak formation has, on average,
a good quality of RMR and RQD with very high intact rock strength. In terms
30 | 24th WORLD MINING CONGRESS PROCEEDINGS

of fracturing, the rock masses have closely spaced joints, and their surfaces are characterized on average as
slightly rough with hard joint walls.

Figure 3. Lithological map of Paul's Peak open pit mine (year 2013)

Figure 4. A cross section of the Paul’s Peak pit showing the logged values of RMR along drill holes.

Table 1: Basic statistics for geomechanical properties in Paul's Peak

Property No. of samples Min. Max. Mean/Mode* Median STD CoV Description**
RQD (%) 65796 0 100 80 90 25 0.31 Good
Fracture Frequency 65794 0 65 5.3 4.0 6 1.29 Joints Closely
(m-1) Spaced
UCS (MPa) 65176 5 380 201 193 89 0.44 Very High
Slightly
Joint Condition 65176 0 25 17.2/20.0 20.0 5 0.31 rough/hard joint
wall
RMR 65176 12 100 78 81 16 0.21 Good

*Mode is recorded only for the Joint Condition

**Descriptions are based RMR 1976
 SURFACE MINING |
31

Using the geomechanical data of the drill-hole dataset, five 3D block models were developed using the
sequential Gaussian simulation (SGS) technique for each of the constituent geomechanical properties of
RMR including: RQD, joint condition, UCS and fracture frequency. The water condition was assumed as
dry, which is the common condition at the pit. The resulting 3D block models were then combined according
to the RMR76 procedure in order to develop the corresponding five RMR block models (summing of the
ratings for the RMR constituent parameters of each block). Table 2 compares the qualitative results of
simulated 3D block models and the geomechanical drill-hole data. According to this table, the built 3D block
models honour the quality of rock mass in terms of RQD and UCS. However, the simulated 3D block models
of fracture frequency provide rock masses of moderately spaced joints in comparison with the closely spaced
jointed rock mass logged in the geomechanical drill-hole data. This indicates that the simulated block models
slightly underestimate the fracture frequency (from mean 5.3 m. -1 in the geomechanical drill-hole data to
mean 3.75 m.-1 in simulated block models). Consequently, the resulting simulated block models of RMR
slightly overestimate the RMR values from good quality in the geomechanical drill-hole dataset to very good
quality in simulations; however, this overestimation is slight, as the average of simulated block models of
RMR is 82, which is a little more than mean 78 recorded in the geomechanical drill-hole data. Figure 5 shows
the Q-Q plot for the simulated RMR values versus the RMR values from the rock sample data. The simulation
honours the extreme rock mass qualities (low and high rock mass qualities). Figure 6 displays the 3D block
model of RMR which has been mapped on the Paul’s Peak walls. This figure shows the heterogeneity of
RMR in the Paul’s Peak pit. The majority of the rocks in the Paul’s Peak pit are very good quality in terms
of RMR.

Table 2: Comparison of simulation results and drill-hole data

Qualitative Description
Geomechanical Property
Drill-hole Data Simulation
RQD (%) Good Good
Fracture Frequency (m-1) Joints Closely Spaced Joints Moderately Spaced
UCS (MPa) Very High Very High
RMR Good Very Good

Figure 5. The Q-Q plot of the modelled RMR against the measured RMR values in the pit.
32 | 24th WORLD MINING CONGRESS PROCEEDINGS

Figure 6. One realization of the RMR model mapped on the Paul's Peak
pit walls (Looking towards East).

During the process of spatial modelling of geomechanical properties of the rock masses in the Paul’s
Peak open pit mine, several questions were raised regarding the reliability of the developed spatial models
and the nature of the input data used for the modelling. Application of geostatistical modelling for
geomechanical data has some challenges, which are common in most of the geomechanical modelling
projects and are discussed in the following section.

CHALLENGES IN GEOSTATISTICAL MODELLING OF GEOMECHANICAL PROPERTIES

Lack of geomechanical data

Despite the wide use of geostatistical techniques for 3D modelling in mineral and oil-and-gas
exploration projects, geotechnical projects are slowly adopting such methods for spatial modelling of
geomechanical properties. This is due to the fact that the geostatistical approaches require sufficient amount
of data for the statistical inference. Traditional open pit mine design and optimization have been limited by
the amount of detailed geomechanical data. Unlike ore grade data, which is generally closely and
systematically spaced across the orebody, geomechanical information is often scattered or widely dispersed
in the area of study, which can make it difficult to apply geostatistical approximation techniques properly.
Although many exploration drill holes are generally drilled to delineate an orebody, a limited budget is
allocated to geomechanical borehole drilling. Most of the core samples from exploration drillings that contain
ore are typically split for assay as soon as possible after drilling. This renders measurement of geomechanical
attributes very difficult and destroys opportunities for collecting valuable geomechanical data from mineral
exploration boreholes that generally have a higher density and better coverage of the area of investigation.
One of the reasons that mining and exploration companies hesitate to undertake a geotechnical core logging
prior to splitting of the core samples is that the conventional geotechnical core logging process is a very time-
intensive activity. Re-logging of the exploration cores after their handling and storage also causes significant
deviation in the recorded data (Macciotta et al. 2014) and can be technically difficult (Blenkinsop and Doyle
2010). In the Paul’s Peak open pit mine, the majority of the exploration drill holes were logged to collect
geomechanical properties prior to use the core samples for assay analysis. This additional effort provided
adequate amount of input data for spatial modelling of the geomechanical attributes of the open pit mine.
The simulation models demonstrate more uncertainty in the pit boundaries where the geomechanical data are
more widely dispersed.
 SURFACE MINING |
33

Subjectivity in geomechanical data collection

Unlike the ore grade data, in which the rock samples are carefully selected and assayed in the lab to
measure the constituent elements, there is a significant degree of subjectivity in the geotechnical data
collection during the site investigation (Egaña and Ortiz, 2013). This subjectivity comes partly from the
subjective definitions of the geomechanical properties, and inconsistency in reports (e.g., two core loggers
report two different values for a geomechanical variable of a core run). Logging the joint condition based on
joint surface roughness, shape of the joint surface and joint wall strength is a very subjective process. This
subjectivity in characterization also appears in logging other geomechanical properties such as RQD and
fracture frequency. RQD is measured as the ratio of total length of all pieces of sound, slightly or moderately
weathered rock cores longer than 10 cm (4 inch) to the total length of core run (Deere and Deere, 1988). Two
different core loggers could interpret the definition of the sound, slightly or moderately weathered piece of
rock differently.

Another source of subjectivity in geomechanical characterization is the estimation of a

geomechanical property using other measured variables. This is also true for the logging of ore grades in the
polymetallic deposits, where the values of some metal grades are estimated from other variables through
regression, deconvolution and construction of a metal balance sheet (Journel and Huijbregts, 1978). A similar
situation can be observed in the geotechnical logging, where, for example, the values of UCS for a rock is
reported using an experimental regression between the point load strength index and rock hardness index
(ISRM 1981) and the Uniaxial Compressive Strength. The values of RQD are sometimes estimated using the
relationship between RQD and the measured values of fracture frequency.

In the Paul’s Peak open pit mine case, the geomechanical database used for the geostatistical
modelling was logged by more than 83 different core loggers between 1949 and 2013. The experts who
logged the cores did not all have the same level of experience and geotechnical knowledge. However, it is
practically impossible to remove the uncertainties associated with data subjectivity from the database.

Bye (2011) states that it is necessary to recognize the uncertainties caused by the lack of data and
data subjectivity, and consequently identify the confidence level of the 3D spatial models using geostatistical
techniques. Both lack of data and subjective assessment of some geomechanical parameters could
significantly reduce the accuracy of the developed spatial models and consequently influence the decision-
making process for mine planning and design.

Homogeneity

The data used for geostatistical estimation and simulation should come from a consistent statistical
and geomechanical population (Egaña and Ortiz, 2013). One of the major sources of the uncertainty in the
geostatistical approaches is the inhomogeneity of the data used for modelling. Homogeneity of the data is
generally required for the statistical inference. Journel and Huijbregts (1978) suggest that homogeneity or
representativeness of data used for the geostatistical modelling must be constant through time and space.

In many mature mining projects like Paul’s Peak open pit operation, the homogeneity of the
geomechanical data through time and space is difficult to attain. The inhomogeneity of the geomechanical
data in time and space can be attributed to the advance of technologies and equipment used in different
periods of time. For example, newer core drilling technologies and the use of triple tube core barrel would
lead to better core recovery and generally better logged values for RQD and fracture frequency than older
technologies and equipment.

The temporal/spatial homogeneity of data can be acquired only if constant methodologies are
followed through the time and space for drilling, sampling and logging of the geomechanical variables. This
can be achieved by using advanced quantitative logging methods, such as high definition digital
34 | 24th WORLD MINING CONGRESS PROCEEDINGS

photography and image analysis of rock core samples; online non-destructive hardness testing of core
samples; and detailed geophysical characterization of drill cores (Dunham and Vann, 2007).

Additivity

The geostatistical estimation methods, such as ordinary kriging, estimate the values of a
regionalized variable (such as ore grade) using arithmetic averaging of the values in the sampled locations.
These methods assume that the regionalized variable being estimated is additive or linear. A regionalized
variable is additive when the arithmetic mean of its values is its average. The additivity assumption is valid
for regionalized variables, such as the ore grade where the support is constant (Journel and Huijbregts, 1978).
Unlike the ore grade, the geomechanical variables such as RQD and RMR are not additive because their
linear average would not necessarily be representative of the geomechanical behaviour of rock mass. For
example, the stability of a pit slope is more highly influenced by areas where the rock mass is of low quality
(low RMR or RQD); therefore, if a spatial model cannot accurately reproduce these poor rock mass quality
zones, the model will fail to represent the ground condition for the slope stability analysis. The non-linearity
of RMR also arises because RMR is calculated through a non-linear combination of its constituent variables
(Egaña and Ortiz, 2013). In calculating RMR, each geomechanical variable (RQD, UCS, joint spacing, joint
condition and water condition) is rated according to the class to which its value belongs.

Another factor that makes a regionalized variable non-additive is the support change effect. The
support change has impacts on the additivity of the RQD, fracture frequency, intact rock strength and
consequently the RMR. The support in the geomechanical logging is usually considered as the core run
length. Therefore, it is essential that the geomechanical variables are logged and reported in consistent and
constant core run lengths.

The estimates resulting from the linear averaging (e.g., using ordinary kriging) of the non-additive
geomechanical properties can only be of use for visualizing the trend and obtaining some comprehension of
the spatial features (Deutsch, 2013). The linear estimators provide the local average estimate where the
histogram of the original data is not honoured (Journel and Huijbregts, 1978; Deutsch, 2002; Eivazy et al.,
2015). Such local estimates are biased low or high depending on the type of geomechanical property
distribution and the type of linear estimation (Deutsch, 2013). Eivazy et al. (2015) demonstrated that using
geostatistical estimation methods for non-additive geomechanical properties like RMR can significantly
overestimate the poor rock mass classes (low RMR value) and underestimate the very good rock mass classes
(high RMR value); consequently, the estimated histogram becomes narrower than the sampling dataset.
Unlike the linear estimators, the geostatistical conditional simulation techniques could be a better choice for
spatial modelling of the geomechanical variables because they honour the histogram of the original data and
do not over/underestimate the extrema of such variables, or eliminate the biases associated with modelling
of non -additive variables. This was clearly demonstrated in the case study of the Paul’s Peak open pit mine.
In fact, conditional simulation methods are the only practical ways for modelling the heterogeneity of non-
additive geomechanical variables because they realistically represent the short-scale variability and avoid the
biases of the linear estimation methods. Such conditional simulation techniques also generate a collection of
point estimates inside a block.

Direction Dependency

The geomechanical variables such as joint spacing (or fracture frequency) and RQD measured along
a drill hole or a scanline survey are direction dependent. Depending on the structural heterogeneity of rock
mass, the values of RQD and fracture frequency logged in the process of core logging can significantly vary
when the orientation of the drill hole or the scanline traverse changes. Since RQD and joint spacing are the
two components in the calculation of RMR, this variable can also show directional behaviour. To address
the direction dependency in spatial modelling of geomechanical variables, it would
 SURFACE MINING |
35

be ideal if these geomechanical variables were logged using the drill holes/scanline traverses in various
orientations. This allows for better detection of rock mass geomechanical anisotropy. In practice, this may
not always be possible because the exploration drill holes are generally oriented with respect to the
orientation of orebody.

Upscaling

The upscaling from the point estimates to the 3D block models is a vital step in the modelling of
the regionalized variables. The upscaling of the estimated values of ore grade into 3D blocks is a routine
matter in the geostatistical modelling. This is due to the fact that ore grade is an additive regionalized variable
and the arithmetic average of the estimated values within a block can truly represent the upscaled value of
the 3D block. However, the upscaling of the geomechanical variables is a challenging and complex task. The
challenge in the upscaling of such geomechanical variables arises from the non-linearity and/or non-
additivity nature of these variables. The function or blend response model (Dunham and Vann, 2007) that
could effectively and realistically upscale a geomechanical variable is an important question that one should
think about it before starting the modelling process. The matter of upscaling of the geomechanical variables
is a key factor in selecting an appropriate geostatistical technique (Bye, 2011). In this regard, a thorough
understanding of each geomechanical variable and its impacts on the downstream decision makings (e.g. pit
slope design, geometallurgical modelling) is essential. This understanding can help in selection of the proper
geostatistical tool and the upscaling approach. Depending on the type of the geomechanical variable,
geostatistical linear, non-linear or simulation techniques might be appropriate. The upscaling problem of
geomechanical properties becomes even more challenging if one decide to use the 3D block models of these
properties as an input data for numerical modelling.

CONCLUSIONS

This paper has illustrated the challenges in applying geostatistical methods for modelling the
heterogeneity of geomechanical properties. These challenges include the lack of geomechanical data in many
mining projects; subjectivity of geomechanical variables in the data collection process; direction dependency
of the geomechanical variables; and non- additivity, non-linearity and upscaling issues associated with these
variables. These challenges are the major obstacles for the application of geostatistical techniques in the
geomechanical projects. Each of these challenges and limitations can influence the choice of the proper
geostatistical techniques that can be used for heterogeneity modelling. Without a good understanding of these
challenges and limitations, spatial modelling of the geomechanical variable might not lead to a reliable
interpretation of the modelling results and may adversely impact decision makings by mine design engineers.
Through a case study of an open pit mine elaborated on in this paper, it was shown that the best possible
choice of geostatistical methods for heterogeneity modelling of the geomechanical variables could be the
sequential Gaussian simulation (SGS). In this case study of an iron ore open pit mine, the geomechanical
variables were modelled in 3D block models using the sequential Gaussian simulation method. The
developed models effectively identify the high-risk zones of weak rock mass quality that are susceptible to
pit slope instability and require more monitoring efforts during exploitation.

ACKNOWLEDGEMENT
The authors would like to thank Natural Sciences and Engineering Research Council of Canada (NSERC)
for their financial support that made this research possible.

REFERENCES

AYALEW, L., REIK, G. & BUSCH, W. 2002. Characterizing weathered rock masses-a geostatistical
approach. International Journal of Rock Mechanics and Mining Sciences 39, 105-114.
36 | 24th WORLD MINING CONGRESS PROCEEDINGS

BIENIAWSKI, Z. T. 1976. Rock mass classification in rock engineering. Exploration for Rock
Engineering. Cape Town: Balkema.
Blenkinsop, T.G. and Doyle, M.G. 2010. A method for measuring the orientations of plannar structures in
cut core, Journal of Structural Geology; (32) 741-745.
BYE, A. R. 2011. Case Studies Demonstrating Value from Geometallurgy Initiatives. First AusIMM
International Geometallurgy Conference, 2011 Brisbane, Australia. The Australasian Institute
of Mining and Metallurgy, 9-30.
DEERE, D. U. & DEERE, D. W. 1988. The Rock Quality Designation (RQD) Index in Practice. Rock
Classification Systems for Engineering Purposes, ASTM, STP 984, Louise Kirkaldie, Ed.,
American Society for Testing and Materials, Philadelphia
DEUTSCH, C. V. 2002. Geostatistical Reservoir Modelling.
DEUTSCH, C. V. 2013. Geostatistical modelling of geometallurgical variables-Problems and solutions.
The second AUSIMM international Geometallurgy Conference. Brisbane, QLD, Australia.
DOOSTMOHAMMADI, M., JAFARI, A. & ASGHARI, O. 2015. Geostatistical modelling of
uniaxial compressive strength along the axis of the Behesht-Abad tunnel in Central Iran.
Bulletin of Engineering Geology and the Environment, 74, 789-802.
DUNHAM, S. & VANN, J. Geometallurgy, Geostatistics and Project Value — Does Your Block Model
Tell You What You Need to Know? Project Evaluation Conference 19-20 June 2007 2007
Melbourne, Victoria, Australia 189-196.
EGAÑA, M. & ORTIZ, J. M. 2013. ASSESSMENT OF RMR AND ITS UNCERTAINTY BY USING
GEOSTATISTICAL SIMULATION IN A MINING PROJECT. Journal of GeoEngineering 8,
83-90.
Eivazy, H., Esmaieli, K., Jean, R., & Albor, F. 2015. Application of 3D geotechnical block models in
design of open pit mines-a case study at Mont-Wright open pit mine. In Proc. of 37th Int. Symp.
on the Application of Computers and Operations Research in Mineral Industry (APCOM 2015),
23-27 May 2015, Fairbanks, USA.
FERRARI, F., APUANI, T. & GIANI, G. P. 2014. Rock mass rating spatial estimation by geostatistical
analysis. International Journal of Rock Mechanics & Mining Sciences 70, 162-176.
GILES, D. P. 1993. Geostatistical interpolation techniques for geotechnical data modelling and ground
condition risk and reliability assessment. In proceedings of Risk and Reliability in Ground
Engineering conference, (B.O. Skipp Ed.). Thomas Telford, London, UK.
HACK, R., ORLIC, B., OZMUTLU, S., ZHU, S. & RENGERS, N. 2006. Three and more dimensional
modelling in geo-engineering. Bull Eng Geol Env 65, 143-153.
HAMMAH, R. E., YACOUB, T. E. & CURRAN, J. H. 2009. NUMERICAL MODELLING OF SLOPE
UNCERTAINTY DUE TO ROCK MASS JOINTING. Int. Conf. on Rock Joints and Jointed Rock
Masses ‘09.
JOURNEL, A. G. & HUIJBREGTS, C. J. 1978. Mining Geostatistics, The Blackburn Press.
La pointe, P.R., 1980. Analysis of the spatial variation in rock mass properties through geostatistics. In Proc.
21st US Rock Mechanics Symposium, University of Missouri, May 28-30, 1980, 570-580.
Macciotta, R., et al. 2014. Empirical design of underground excavation support: assessing reliability
associated with rock mass classifications using core drilled for mineral exploration, 1 st Int.
Conf. on Applied Empirical Design Method in Mining, June 9-11, 2014, Lima, Peru.
MARCHESI, V. R., DA FONTOURA, S. A. B. & RUBIO, N. P. R. 2009. How 3D modelling can
improve quality & reliability of geotechnical projects. EUROCK-ISRM International Symposium.
MARINONI, O. 2003. Improving geological models using a combined ordinary–indicator kriging
approach. Engineering Geology, 69, 37-45.
MATHERON, G. 1963. Principles of geostatistics. Economic Geology 58, 1246-1266.
READ, J. & STACEY, P. 2009. Guidelines for open pit slope design. In: READ, J. & STACEY, P.
(eds.) Guidelines for open pit slope design.
SAVARD, P. & JEAN, R. 2012. Report of the Mineral Reserves and Resources. ArcelorMittal Mines
Canada.
STAVROPOULOU, M., EXADAKTYLOS, G. & SARATSIS, G. 2007. A Combined Three-
DimensionalGeological-Geostatistical-Numerical Modelof Underground Excavations in Rock.
Rock Mechanics and Rock Engineering, 40, 213-243.
VATCHER, J., MCKINNON, S. D. & SJÖBERG, J. 2016. Developing 3-D mine-scale geomechanical
models in complex geological environments, as applied to the Kiirunavaara Mine. Engineering
Geology, 203, 140-150.