Optimizing cut-off grade considering

grade estimation uncertainty – A

case study of Witwatersrand gold-
producing areas
by C.C. Birch
Affiliation:
S chool of Mining Engineering,
1

University of the Witwatersrand,

South Africa.

Synopsis
Correspondence to: Due to grade estimation uncertainty, two statistical errors can occur. The Type I error is where material
C.C. Birch
is classified as ore and mined, despite the true value being below the break-even grade. This material
is dilution. The Type II error is where the material is estimated to be below the cut-off grade and is
Email: classified as waste, although the true grade is actually above the break-even grade. This material is not
[email protected] mined and is lost. The uncertainty was assumed to follow a normal distribution in a previous study. For
this study, estimated block values are compared to those determined after mining (the best estimate
of the true grade). This actual data from four mines shows that the uncertainty follows a Laplace
Dates: distribution. There is no single solution regarding adjusting the cut-off grade away from the break-even
Received: 13 Oct. 2020 grade, considering estimation uncertainty, that could be applied to all gold mines. However, adjusting
Revised: 16 May 2022
the cut-off grade downwards (up to 22% for one mine) is noted when optimizing the profit considering
Accepted: 16 May 2022
grade uncertainties. This type of adjustment could open up significant mining areas and extend the life
Published: July 2022
of the mine.

How to cite: Keywords

Birch, C.C. 2022
uncertainty, Type I error, Type II error, cut-off grade, optimization, NPV; simulation, mixed-integer
Optimizing cut-off grade considering
grade estimation uncertainty – A
linear programming, @Risk, Excel Solver.
case study of Witwatersrand gold-
producing areas.
Journal of the Southern African
Institute of Mining and Metallurgy,
vol. 122, no. 7, pp. 337–346
DOI ID:
http://dx.doi.org/10.17159/2411- Introduction
9717/1403/2022
This study is a follow-up to a previous investigation (Birch, 2017) in which cut-off grade optimization was
undertaken for four South African narrow, tabular (Witwatersrand-type) gold mines while considering
various hypothetical degrees of uncertainty. It was found that where there are higher degrees of
uncertainty in the grade, reducing the cut-off grade compared to the break-even grade can improve the
financial returns. For the present study, the break-even grade is considered the grade where the income
from a mining block will cover all its mining and processing costs (as defined by the all-in sustainable
cost (AISC)). The management team uses the cut-off grade to determine the areas to be mined or left and
effectively split ore from waste. This study considers the actual grade uncertainty for four South African
gold mines. Each case study determines how the cut-off grades should be adjusted compared to the
break-even grade when taking grade uncertainties into account.

Background
Some mining companies apply the concept of break-even grade to individual ore blocks and assess if
each block could be exploited profitably. This break-even grade is then applied as the mining cut-off
grade (Minnitt, 2004) with some adjustments. With these approaches, the overall grade of the material
being mined becomes the average grade of the blocks above the break-even-derived cut-off grade. An
alternative approach is to apply the pay limit principle where the overall mining grade required to break
even financially is determined. Then the profit margin required is determined and added to the break-
even grade. The mining planners then check the mix of the various mining areas to ensure the required
mining grade is achieved.
However, these approaches fail to consider the interaction between the royalties for mineral
resources and the South African gold tax formula, which varies considering the mine’s overall
profitability. A revised approach to optimizing the value of gold mines was required as part of a research
project to understand the impact of the 2008 Mineral and Petroleum Resources Royalty Act on gold

The Journal of the Southern African Institute of Mining and Metallurgy VOLUME 122 JULY 2022 337
Optimizing cut-off grade considering grade estimation uncertainty
mines (Birch, 2016a). Considering the mine’s current planned the mining grade and expected metal content. South African gold
economic and production profile, this approach considers the mine block listings were run through a simple financial optimizer
entire cash flow and optimizes overall profit or net present value for the previous study. Various degrees of uncertainty for the block
(NPV) for the Measured and Indicated Resource. The financial grades were considered. This uncertainty was introduced using
optimizer uses mixed-integer linear programming (Excel Solver) @Risk software (a Monte Carlo simulation program developed
to maximize profit or net present value NPV by varying the cut-off by Palisade). The uncertainty distribution was assumed to be
grade. This, in turn, alters the overall tonnage and average mining symmetrical, and no bias was considered (100% block factor). The
grade (AMG) above the cut-off grade. The approach is simple, databases used for this research exercise had been corrected for
robust, and provides a valuable tool to quickly assess the impact conditional bias and smoothing (Tolmay, 2014). The uncertainty
of grade uncertainty without the need for extensive information dispersion was controlled using the standard deviation to give 5%,
from the mines themselves (Birch, 2016b). 6.7%, 10%, and 20% uncertainty.
The grade for each mining block is estimated by sampling The following points were noted.
the mineral deposit and projecting the values into the area to ➤ It was found that when optimizing for profit, three of
be evaluated. Various techniques are used, including nearest the four mines would benefit from a reduction in the
neighbour, inverse distance squared, and kriging. The estimated cut-off grade compared to the break-even grade.
value is used to determine if a mining block is classified as ore or ➤ This reduction ranged from 2–10%. The fourth mine
waste. There is, however, a degree of uncertainty regarding the would not benefit from any change in the cut-off grade.
estimated value due to sampling spacing, deposit heterogeneity, ➤ When optimizing for 9% NPV, three of the four mines
and the method of estimation used. would benefit from reducing the cut-off grade by
Accepting a hypothesis that should have been rejected is between 7 and 27%. The fourth mine would benefit by
a Type I error in statistics. The alternative error is rejecting a increasing the cut-off grade by 7%.
hypothesis that should have been accepted, and is called a Type ➤ When optimizing the NPV at 12%, the one mine would
II error (Underhill and Bradfield, 1994). Due to uncertainty in benefit from an 8% decrease in the cut-off grade. The
the estimation, the same two errors can occur. The Type I error NPV benefit is, however, only 1.3%. The other three
is where the material is classified as ore and mined, although the mines would not benefit from adjusting the cut-off
true value is below the cut-off grade and thus this is waste material grades and using the model results in NPVs lower than
and constitutes dilution. The Type II error is where the material is not adjusting them, as indicated in two cases.
estimated to be below the cut-off grade and is classified as waste,
The previous study assumed that the uncertainty follows
whereas the true grade is above the break-even grade and it could
a normal distribution, and there is no bias present in the
have contributed profits if it had been mined. This material is not
estimations. This research uses ‘actual’ uncertainty distributions
mined and is lost (Minnitt, 2017).
to determine if the same conclusions can be drawn.
Problem statement
Methodology
High uncertainties characterize mining projects. Such
The mining block listing is the starting point for the financial
uncertainties vary in magnitude and are very prevalent in the
model used for this study. This mining block listing contains all
geological data and subsequent grade estimation on which
potential mining areas, estimated grades, and volumes created
the project is based. The method of handling uncertainty in
from the geological block model. For South African gold mines,
mining projects may significantly affect the decisions made by
the mining blocks are the same size as the geological blocks
the management team (Kühn and Visser, 2014). There is no
(typically 30 m by 30 m for the Measured Resource blocks and
straightforward process for adjusting the cut-off grade away from
60 m by 60 m for the Indicated Resource blocks).
the break-even grade for narrow tabular orebodies, considering
The determination of the break-even grade is, in essence,
the uncertainties related to the grade estimation.
quite simple. It determines what grade a unit of ore requires to
Previous work return a profit. This is basically a break-even volume measurement
where the quantity is known (usually limited due to shaft capacity,
A previous investigation (Birch, 2017) aimed to determine what milling capacity, or other physical constraints). The unknown
adjustment should be made to the cut-off grade to reduce the is the commodity’s in-situ value. The other factors needed for
financial impact of dilution or lost ore in typical narrow, tabular the calculation of the break-even grade are the mine recovery
Witwatersrand gold mines. Three possible options were identified: factor (MRF), which is the mine call factor (MCF) multiplied
➤ Raise the cut-off grade to reduce the dilution – ‘the old by the plant recovery factor (PRF). The metal price is quoted
adage that a low-grade ton should never keep a high-grade in US dollars (for gold and silver, in troy ounces). These are all
ton out of the mill’ (Minnitt, 2017) projections and open to variation over the time over which the
➤ Keep cut-off grades the same, considering that the Type break-even grade is applied. Therefore, they contribute to the
I and Type II errors would balance each other if there is investors’ financial risk if they change significantly. This can be
no bias expressed as follows (School of Mining Engineering, 2021):
➤ Lower cut-off grades to ensure all value from the At break-even grade:
orebody is obtained (thus recovering a higher
TR = TFC +TVC
percentage of the lost ore).
but
The mines reviewed all have mining block listings. These
listings include the area (m2), channel width (cm), and the TR = UR* X
estimated in-situ grade in grams per ton (g/t). The tonnage and and
stoping width can be determined from these values, along with TVC = UVC * X

338 JULY 2022 VOLUME 122 The Journal of the Southern African Institute of Mining and Metallurgy
Optimizing cut-off grade considering grade estimation uncertainty
Therefore: to 14.5% to 17.6% at an annual inflation rate of 5% (typical for
UR * X = TFC + UVC * X South Africa).
The financial optimizer used for this study utilizes linear
Dividing both sides by X:
mixed-integer (Excel Solver) programming. Due to its availability
UR =(TFC÷X) + UVC to all Excel users, the Solver function built into Microsoft Excel
But was selected for this study (Meissner and Nguyen, 2014). The
UR = g/t * r * p variable for the Solver function is the cut-off grade, and the Solver
function is set to optimize the cut-off grade to maximize the
Therefore at break-even grade (g/t):
resulting profit or NPV. These cut-off grades are then compared
g/t = ((TFC÷X) + UVC)÷(r*p) to the base break-even grade to indicate how the cut-off grades
where g/t is the break-even grade in grams per ton; r is the mine should be adjusted considering the uncertainty in the estimated
recovery factor in %; p is the metal price in US$ per gram; TFC is in-situ grades.
the total fixed costs in US$; X is the milled tons; and UVC is the
unit variable costs in US$ per t. Sources of data
The cut-off grade is an extension of the concept of a break- Historical mining block listings were obtained from four gold
even grade. If only blocks above break-even grade are mined, this mines for this study. The mines are located in the West Rand and
becomes the cut-off grade. However, the cut-off grade undergoes the Free State gold mining regions of South Africa (Figure 1). They
variations over the mine’s life. Hall (2014) describes the basics all mine the narrow, tabular gold-bearing conglomerates of the
of cut-off grade theory. This book is an exhaustive study of the Witwatersrand Supergroup. For anonymity purposes, these mines
different techniques currently used in the mining industry. It will be called A, B, C, and D. It is important to note that these are
includes various value measures, including discounted cash flow not the mines that were used in the previous study. The data for
(DCF) and optimization of NPV. The South African Code for the the previous research was for four individual shafts belonging to
Reporting of Mineral Asset Valuation (The SAMVAL Code) income two different mines. The data used for that study was the block
approach to valuation requires a mine design with reasonable listing and financial information used for the mines’ Mineral
estimates of expected tonnages, grades, costs (fixed and variable), Resource Statements.
and recoveries (Border, 1991). Accompanying the historical block grades for the mined-
The cost of capital is usually calculated by the weighted out areas is the post-mining grade for the block. This grade is
average capital cost (WACC) and is expressed as a percentage. considered the ‘true’ grade. Although it is still an estimate it is
The WACC represents a company’s average cost of capital from the figure used by the mine surveyor to allocate the called-for
all sources. These include equities (common stock, preferred grade for the block. Discrepancies between the called-for gold
stock, and bonds) and other forms of debt. The cost of equity can estimates and the final recovered gold are dealt with in the break-
be determined using various methods, including the capital asset even formula where the MRF is considered. The distribution
pricing model (CAPM) and the Gordon growth model (GGM) of the uncertainties was based on these listings. They display
(School of Mining Engineering, 2021). The cost of capital of the a significantly different pattern from the normal distribution
mine is used as the discount rate for cash inflows and outflows. assumed for the previous study (Birch, 2017), and significant grade
The NPV is the difference between the present value of cash overestimation bias in three cases.
inflows and the present value of cash outflows over a period of The relationship between the volume (tons) available for
time (Investopedia, 2021). Real monetary mining discount rates mining and the cut-off grade, and the resulting average grade
of between 9% and 12% are suitable for South African mining above the cut-off grade, can be determined from the grade-
projects (Smith et al., 2007). That is equivalent in nominal terms tonnage curve.

Figure 1—The Witwatersrand gold-producing areas in South Africa (Minerals Council South Africa, 2020)

The Journal of the Southern African Institute of Mining and Metallurgy VOLUME 122 JULY 2022 339
Optimizing cut-off grade considering grade estimation uncertainty
A generic financial model has been utilized to compare a The distribution which best fits the data was selected using
typical deep-level gold mine producing 1.2 Mt of ore per annum. A the Akaike information criterion (AIC) value. AIC compares the
generic financial model has been used to highlight the effect of the relative ‘quality’ of a model (distribution) with the other models.
uncertainty and allow direct comparisons as to the effect of this @Risk ranks the various distribution model outputs according to
between the different mines. A gold price of US$ 1500 per ounce this value. The distribution with the smallest AIC value is usually
is assumed for this case study. For all the examples, annual fixed the preferred model (BPI Consulting, 2016). Figure 4 shows the
costs are assumed to be US$260 million, and the variable cost is Laplace and normal distribution curves for the four mines.
US$55 per ton. The split between fixed and variable costs is 75:25. It is clear from the figures that the Laplace distribution curve
A feature of South African gold mines is that labour costs are for the variation has a far higher kurtosis (the sharpness of the
typically 50% (Savant, 2012) of the total costs, resulting in a higher peak) than the normal distribution curve in all four examples.
fixed cost to variable cost ratio than more typical mining projects However, the normal curve reflects the mean variation of the
where a 50/50 split would be appropriate (Poxleitner, 2016). The input data better. It shows that the grades for mines A, B, and C
annual production rate and 75% MRF have been selected to give are over-estimated by 0.8, 0.4, and 1.0 g/t, respectively. Only mine
a break-even grade of 7.5 g/t. Break-even face grades of 7.5 g/t D has an overall variation of less than 0.1 g/t. Over- or under-
are typical for deep-level South African gold mines and fit into estimation is often reflected in the block factor value, with 100%
the break-even grades determined from the detailed financial indicating no bias. These block factors were found to be 94.3%,
figures used in the previous study (6.9–10 g/t). The NPVs were 93.2%, 92.4%, and 99.0%, respectively for the four mines.
determined using discount rates of 9 and 12%. This overestimation of the block grades is very significant.
Many blocks are being identified as ore while they are sub-
Analysis economic and should not be mined. Suppose the overestimation
is carried through to the post-mining true grade estimation for the
Estimated versus true block grades blocks. In that case, the MCF will also be lower than expected. The
The approximate value for one mining block and the true value MCF is used in the break-even grade calculation. This will raise
will be the same in a perfect world. There will be no blocks above the break-even grade.
the cut-off grade with a true value below the cut-off grade (Type I Financial modelling
error or dilution). There will also be no blocks calculated as below
cut-off grade that are actually above cut-off grade (Type II error or The tons above the cut-off grade and the AMG are determined
missing ore). This ideal relationship can be seen in Figure 2. from the block listing as inputs into the financial optimizer. The
The pre-mining estimated and post-mining true block grades block lists obtained for these mines only contained the gold
were obtained from four mines for this study. The number of grades in centimetre-grams/ton (cmg/t). Since the block listing did
not include block area, channel width, or stoping width, various
blocks included in the analysis ranged from 1757 to 2547 for the
assumptions have been made. These assumptions are assumed
respective mines. The estimated grades vs the true grades for the
not to impact the validity of this study because they do not alter
four mines are shown in Figure 3.
the comparative nature of the results. The same assumptions have
It can be observed that there is significant overvaluation in the
been made for all four financial models used in the current study.
estimated block grades for three of the mines. This is reflected in
the slope of the trend lines. For mine A, the slope is 0.85. For mine
B it is 0.87, and for mine C it is 0.86. This is significant because it
shows a clear bias in the estimation, which was not considered in
the previous study (Birch, 2017). For mine D, the bias is far less,
and the slope is 0.95. This overestimation of the block grades
has a marked impact on the suggested cut-off grade optimization
adjustments shown later in the analysis. The R2 values for the four
mines range from 0.54 to 0.69. This shows a significant spread of
the data-points and a poor correlation between the estimated and
true values. A perfect correlation would result in an R2 value of 1.0
(see Figure 2). This poor correlation results in numerous points
plotting in the graphs’ Type I (dilution) and Type II (lost ore)
areas.

Variation distribution
The difference between the estimated and true grades for each
block was determined. The distribution of these variations is
analysed using the Palisade @Risk Distribution Fit tool. For the
previous study, normal distributions with no bias were assumed.
When the variation distributions for the four mines in this study
were analysed, Laplace distributions were found to fit better than
normal distributions. In probability theory and statistics, the
Laplace distribution is a continuous distribution of probabilities, Figure 2—Ideal world situation where estimated and true values are the
same (zero uncertainty). All the data-points plot along a single line. No
called the double exponential distribution. It can be viewed as two dilution or lost ore occurs. The break-even cut-off grade of 7.5 g/t is shown,
exponential distributions spliced together back to back (Hilali, as are the Type I and II error areas. It is noted that no points plot into
2016). these two areas

340 JULY 2022 VOLUME 122 The Journal of the Southern African Institute of Mining and Metallurgy
Optimizing cut-off grade considering grade estimation uncertainty

Figure 3—Pre-mining estimated vs post-mining true block grades for mines A, B, C, and D

Some companies model accumulation value (cmg/t) rather compare the metal content with a very narrow 1 cm mineralized
than grade (g/t). Accumulation value is helpful for very narrow zone with 1000 g/t (for example, the carbon-rich facies of the
orebodies because it gives the metal content, rather than just Basal Reef) to one with a 100 cm mineralized width with a grade
the grade, over the mineralized width. The accumulation value is of 10 g/t. Both faces will have the same mining face grade of 10
calculated by multiplying the grade of the sample by the width the g/t if the minimum mining width is 100 cm. However, since this
sample represents. Multiple samples are taken and composited parameter is not widely used except on South African gold mines,
when sampling narrow tabular orebodies to get the average grade it has been converted to grams per ton for easy explanation. The
over the mineralized width. Each sample grade is multiplied by expected face grade for this study was determined for each block
the width (in centimetres), and then all the individual values are by dividing the cmg/t by a standard stoping width of 150 cm to get
summed. This combined value is then the value in cmg/t. This the face grade in grams per ton.
value can then be divided by the expected mining width, and the This block area needs to exceed the smallest mining unit
average mining face grade of the block determined. A 1000 cmg/t size determined by the mine design approach. Typical mining
over 1 m will result in a face grade of 10 g/t. It is possible then to panels in deep-level South African gold mines are 30 m long.

The Journal of the Southern African Institute of Mining and Metallurgy VOLUME 122 JULY 2022 341
Optimizing cut-off grade considering grade estimation uncertainty

Figure 4—Variation distributions for mines A, B, C, and D. The sharp- Laplace distribution curves in the four plots (in red) can be and compared to the flatter
normal distribution curves (in green)

342 JULY 2022 VOLUME 122 The Journal of the Southern African Institute of Mining and Metallurgy
Optimizing cut-off grade considering grade estimation uncertainty
The block sizes have been standardized at 30 × 30 m (also the ➤ LP Simplex (for problems that are linear)
Measured Mineral Resource block size). The specific gravity of ➤ Evolutionary (for problems that are non-smooth).
the quartzitic conglomerate ores on these mines is typically 2.72, It has been found for the financial optimizer model that
which has been used to determine the tonnage for each block. although the evolutionary approach is the slowest, it is the most
The other parameters used for financial modelling were described reliable. The GRG approach gives inconstant results, often not
previously. changing the initial value input for the cut-off grade. The LP
This study developed a model that can be optimized for Simplex approach has not been tested because the nature of the
either profit or NPV. Optimizing NPV can drive decision-making model is not considered linear.
to focus on short-term gains at the expense of the longer term. The financial optimizer determines the tonnage above the
Therefore, companies frequently use the total undiscounted cash cut-off grade and the AMG from the block listing sheet. The MRF
flow (Hall, 2014). All four mines' financial models were run using and ratio between stope tonnage and total milled tonnage are
zero (undiscounted profit), 9%, and 12% discount rates. This is obtained from the mines' ore flow sheets. These values form the
consistent with the previous study, which used the same discount primary inputs into the cash flow sheet, along with the production
rates. The NPV cut-off grade is higher than when optimized for rate, metal price, and mining costs. As Solver's cut-off grade varies,
profit. The NPV optimization model favours high-grading mining the tonnage above the cut-off grade and AMG inputs change to
because of the discounting. This leads to a shorter mine life and maximize the target selected (profit or NPV). This relationship is
lower mineral resource extraction (Birch, 2016b). shown in Figure 5.
The optimized cut-off grade obtained from the financial Mine D has been selected to demonstrate the grade-tonnage
optimizer excludes the estimated cost of the mineral resource curve (Figure 5). This is due to mine D demonstrating block
royalty ( South Africa, 2008) and income tax calculated using factors closest to 100% among the four mines included in this
the South African gold tax formula ( South Africa, 2019). The study.
purpose of excluding these financial liabilities from the model A grade-tonnage curve can be used to visually display the tons
is to consider profitability when determining the rate. Including above the cut-off grade and the average mining grade of the blocks
these in the model creates interesting interplays between the above the cut-off grade. A vertical line drawn at the cut-off grade
optimal cut-off grade and these variable costs. This is the subject will intersect both the tonnage and grade curves. A horizontal line
of further research by the author. This interplay adds a layer that drawn from the tonnage curve to the tonnage axis (the left side
is distracting from the purpose of this particular research project, in this case) will indicate the tons above the cut-off grade (3.3 Mt
which is studying the role of uncertainty in optimizing the cut-off in this case). These tons can be used as the tons available for the
grade. financial model. A horizontal line drawn from the intersection
The financial optimizer uses mixed-integer linear point with the grade curve to the grade axis (right side) will
programming (Excel Solver). The Solver feature is built into indicate the average grade of all the blocks above the cut-off grade
Microsoft Excel. It was selected for this study due to its (13.8 g/t in this case). This can then be used as the AMG for the
availability to all users of Excel (Meissner and Nguyen, 2014). financial optimization model.
Solver can be used to determine an optimal value (maximum or
minimum) for a formula in a target cell that is constrained or Results and discussion
limited by the values of other formula cells in a worksheet. Solver
adjusts the values in the decision variable cell (the cut-off grade) Overall bias not considered
to fulfil the constraint bounds of the cell and provide the desired The model was set to replace the pre-mining estimated block
result for the target cell (the profit or NPV) (Microsoft, 2022). value with the post-mining true value for each block. The Solver
Solver uses three algorithms (Microsoft, 2022): function was then run to optimize the cut-off grade. This exercise
➤ Generalized reduced gradient (GRG) nonlinear (for was run considering both optimization for profit (zero NPV) and
problems that are smooth nonlinear) NPV at 9% and 12%. Figure 6 displays how the cut-off grade for

Figure 5—An example of a grade-tonnage curve (mine D) with a cut-off of 7.5 g/t. The resultant tons above 7.5 g/t are 3.3 Mt, and the average mining grade is
13.8 g/t

The Journal of the Southern African Institute of Mining and Metallurgy VOLUME 122 JULY 2022 343
Optimizing cut-off grade considering grade estimation uncertainty

Figure 6—Profit optimization. The basic cut-off grade does not consider the grade estimation uncertainty, whilw the optimized cut-off grade considers the true
block values

Figure 7—NPV optimization. The basic cut-off grade does not consider the grade estimation uncertainty, while the optimized cut-off grade considers the true
block values

each of the four mines varies when considering the uncertainty of uncertainty, mine C shows a reduction in the cut-off
the block values when optimized for profit (zero NPV). grade, while the other three mines show an increase in
It can be observed that only mine A shows a decrease in the the cut-off grade is preferable.
cut-off grade when the grade uncertainty is considered. In the These trends indicate that the bias in the pre-mining block
previous study, the overall tendency was that the cut-off grade estimation grades is significant and alters the previous study's
when optimizing for profit is lower than the unoptimized break- findings.
even grade when uncertainty is considered. Figure 7 shows the
cut-off grade graphs optimized for 9% and 12% NPV. Overall bias considered
The following can be noted. Mines A, B, and C had block factors significantly below 100%
➤ Both the basic and grade estimate uncertainty-optimized at 94.3, 93.2, and 92.4%. Only mine D has a block factor close to
cut-off grades are higher than the break-even cut-off 100%, at 99.0%. The distribution model used in the previous study
grade of 7.5 g/t, which is expected when optimizing for to determine each model's true grades assumed a block factor
either 9% or 12% NPV. of 100%. Considering this study's initial optimization exercise
➤ The same trend as seen in the profit optimization is results, the impact of the overall bias in the estimated block
observed in the 9% NPV cut-off grade optimization. grades needed to be considered.
➤ Only mine A shows that a significantly reduced cut-off Whether to mine a block or not is based on the estimated
grade is beneficial. grade. Suppose the overall tendency is to overestimate the block
➤ For the 12% NPV optimization considering grade grades. In that case, more blocks will be considered ore than is

344 JULY 2022 VOLUME 122 The Journal of the Southern African Institute of Mining and Metallurgy
Optimizing cut-off grade considering grade estimation uncertainty

Figure 8—Profit optimization following correction to the bias to bring the block factor to 100%. The basic cut-off grade does not consider the grade estimation
uncertainty, while the optimized cut-off grade considers the true block values

Figure 9—NPV optimization following correction to the bias to bring the block factor to 100%. The basic cut-off grade does not consider the grade estimation
uncertainty, while the optimized cut-off grade considers the true block values

actually the case (Type I error), which will add to the dilution. For the exercise where the optimization maximizes the NPV,
Mines should constantly consider their block factor and adjust the overall tendency is to reduce the cut-off grade from the cut-off
their evaluation parameters to reduce this bias. If the same grade obtained without considering the uncertainty. This is shown
technique is used to evaluate the blocks post-mining, this will in Figure 9.
reflect an apparent loss when determining the mine call factor The following is noted.
(de Jager, 1997). ➤ The indicated cut-off grade reduction for 9% NPV is 23%
The planned block grade is adjusted to bring the block factor for mine A and 7% for mine C.
to 100% to correct this bias. This correction was applied to all four ➤ The reductions for mines B and D are minor at 1% and
mines. The model then selects the blocks based on this adjusted 3%, respectively.
planned grade rather than the originally planned grades. For the ➤ For the 12% NPV, the indicated cut-off grade reduction
profit optimization, the results indicate that two of the four mines for mines A and C is 4% and 14%, while for mine B there
(mines A and C) would benefit from a significant downward should be a 2% reduction.
adjustment in the cut-off grade from the break-even grade of ➤ No adjustment is suggested for mine D.
7.5 g/t (Figure 8).
The following is noted. Conclusion and recommendations
➤ The reduction in cut-off grade is 22% and 9%, The previous study (Birch, 2017) assumed a normal distribution
respectively, for mines A and C. with no bias for establishing the grade uncertainty. This study uses
➤ The other two mines (B and D) would require a 4–5% real data from four South African gold mines to establish whether
increase in cut-off grade. the assumptions and results from the previous research are valid.

The Journal of the Southern African Institute of Mining and Metallurgy VOLUME 122 JULY 2022 345
Optimizing cut-off grade considering grade estimation uncertainty
The true data indicates in all four cases that the distribution curve BPI Consulting. 2016. Deciding which distribution fits your data best. https://
of the uncertainty of the grade estimation follows a Laplace rather www.spcforexcel.com/knowledge/basic-statistics/deciding-which-distribution-
than a normal curve. This ‘double exponential’ curve displays far fits-your-data-best [accessed 8 April 2020].
greater kurtosis than a normal curve with tails extending further. De Jager, E. 1997. The analysis of the mine call factor in gold mining, with
In three of the four case studies, there is a significant bias. specific reference to Western Holding Mine. PhD thesis, University of the
This is displayed by block factors below 95%, indicating an Witwatersrand. http://sierra-eds.wits.ac.za/iii/encore/record/C__Rb1382121__
overestimation in the estimated grades. The initial financial SDe Jager, Ernst Jakobus.__Orightresult__X4?lang=eng&suite=def [accessed
optimization exercise indicated that reducing the cut-off grade 10 April 2020].
was preferable in a single case for the profit optimization and Hall, B. 2014. Cut-off Grades and Optimising the Strategic Mine Plan. Spectrum,
mixed results for the NPV optimization cases. This is contrary to S. (ed.). Australasian Institute of Mining and Metallurgy, Carlton. Victoria,
the previous study (where the optimization model indicated a cut- Australia.
off grade reduction was always preferable for profit optimization
Hilali, H. 2016. Laplace-distribution. https://www.coursehero.com/file/18659563/
and generally preferable for NPV optimization). In the previous Laplace-distribution/ [accessed 8 April 2020].
research the variances had no bias, and it was felt that the bias
Investopedia. 2021. Net present value (NPV) definition. https://www.investopedia.
was altering the trends. The bias was corrected to bring the block
com/terms/n/npv.asp [accessed 22 January 2022].
factors for all four to 100%, and the exercise was repeated.
When optimizing for profit, two of the four cut-off grade Kühn, C. and Visser, J.K. 2014. Managing uncertainty in typical mining project
optimization models suggest a reduced cut-off grade by up to 22% studies. South African Journal of Industrial Engineering, vol. 25, no. 2.
pp. 105–120. doi: 10.7166/25-2-800
from the break-even grade when considering grade uncertainty.
However, no reduction was indicated for the other two mines, and Meissner, J. and Nguyen, T. 2014. An introduction to spreadsheet optimization
a slight increase (4–5%) is suggested. With the 9% NPV model, the using Excel Solver. http://www.meiss.com/download/Spreadsheet-
Optimization-Solver.pdf [accessed 19 March 2020].
data from all the mines suggests that reducing the cut-off grade
is preferable. This reduction varied significantly from 1–23%. This Microsoft. 2022. Define and solve a problem by using Solver. https://support.
trend was also observed following the 12% NPV exercise, with microsoft.com/en-us/office/define-and-solve-a-problem-by-using-solver-
reductions between 2–14% in three mines. For the fourth mine, no 5d1a388f-079d-43ac-a7eb-f63e45925040 [accessed 22 January 2022].
reduction in the cut-off grade is suggested. Minerals Council South Africa. 2020. Gold. https://www.mineralscouncil.org.za/
It is recommended that all mines focus on their block factors sa-mining/gold [accessed 19 March 2020].
and other evaluation parameters to ensure no overall bias in the Minnitt, R.C.A. 2004. Cut-off grade determination for the maximum value of a
block estimations. This bias could also be carried through to small wits-type gold mining operation. Journal of the South African Institute of
estimate the mined grades (the true grade) and will result in gold Mining and Metallurgy, vol. 104, no. 5. pp. 277–283.
being called for that does not exist. This apparent loss will mean Minnitt, R.C.A. 2017. Poor sampling, grade distribution, and financial outcomes.
an increase in cut-off grades. Journal of the Southern African Institute of Mining and Metallurgy, vol. 117, no. 2.
Once the bias is corrected, running a financial optimization pp. 109–117. doi: 10.17159/2411-9717/2017/v117n2a2
exercise similar to this one, considering the grade uncertainty,
Poxleitner, G. 2016. Operating costs for mines. SRK Consulting
could unlock value by reducing the number of lost blocks. The (Canada). https://www.srk.com/sites/default/files/file/GPoxleitner_
value of the lost blocks appears to be higher than the mining OperatingCostEstimationForMiners_2016.pdf.
losses incurred by mining more un-pay blocks by reducing the cut-
Savant, R. 2012. Overview of mining costs. CPM Group. Brooklyn, NY. http://www.
off grades. However, every orebody has a unique grade-tonnage
goldconvention.in/iigc2012/presentation/CPM Group Overview of Mining
characteristic and financial circumstances. No single solution
Costs RS.pdf.
could be applied to all gold mines, and the optimization strategy
(profit or NPV) also affects the outcome. School of Mining Engineering. 2021. MINN7092A Course Notes. University of
the Witwatersrand.
Smith, G.L., Pearson-Taylor, J., Anderson, D.C., and Marsh, A.M. 2007. Project
References
valuations, capital investment and strategic alignment. Journal of the Southern
Birch, C. 2016a. Impact of the South African mineral resource royalty on cut-off African Institute of Mining and Metallurgy, vol. 107, no. 1. pp. 67–74. https://
grades for narrow, tabular Witwatersrand gold deposits. Journal of the Southern journals.co.za/content/saimm/107/1/AJA0038223X_3255 [accessed 4 November
African Institute of Mining and Metallurgy, vol. 116, no. 3. doi: 10.17159/2411- 2019].
9717/2016/v116n3a4
South Africa. 2008. Mineral and Petroleum Resources Royalty Act 28 of 2008.
Birch, C. 2016b. Impact of discount rates on cut-off grades for narrow tabular gold Government Gazette. 521:13.
deposits. Journal of the Southern African Institute of Mining and Metallurgy,
South Africa. 2019. Guide for Tax Rates/Duties/Levies (Issue 14). South African
vol. 116, no. 2. doi: 10.17159/2411-9717/2016/v116n2a2 Revenue Service, Pretoria. https://www.sars.gov.za [accessed 15 October 2019].
Birch, C. 2017. Optimization of cut-off grades considering grade uncertainty in Tolmay, L. 2014. On the reduction of algorithmic smoothing of kriged estimates.
narrow, tabular gold deposits. Journal of the Southern African Institute of Mining Journal of the Southern African Institute of Mining and Metallurgy, vol. 114, no. 8.
and Metallurgy, vol. 117, no. 2. pp. 149–156. doi: 10.17159/2411-9717/2017M17n2a6. pp. 667–672. https://www.saimm.co.za/Journal/v114n08p667.pdf [accessed 17
March 2020].
Border, S. 1991. Optimisation of cut-off grades during design of underground
mines. Proceedings of the Mining Industry Optimisation Conference, Sydney. Underhill, L. and Bradfield, D. 1994. IntroSTAT 5.0. Juta & Co., Kenwyn, Cape
Australasian Institute of Mining and Metallurgy, Melbourne. Town. u

346 JULY 2022 VOLUME 122 The Journal of the Southern African Institute of Mining and Metallurgy