Professional Documents
Culture Documents
Assessing risk in grade-tonnage curves in a complex copper deposit, northern Brazil, based on an efficient
joint simulation of multiple correlated variables. Application of Computers and Operations Research in the Minerals Industries, South African Institute of
Mining and Metallurgy, 2003.
Risk quantification in grade-tonnage curves is critical for capital investment in mining projects
and can be obtained through geostatistical simulations of orebodies. A practical difficulty may
arise in multi-element deposits, as the joint modelling of the related attributes using the traditional
co-simulation approaches is computationally intensive and may be impractical for use in the
industrial environment. This paper presents the construction of risk-integrating grade-tonnage
curves for a complex copper deposit in northern Brazil, by jointly simulating its key geochemical
attributes of interest: Cu, Fe and K. The joint conditional simulation of these elements is based on
Minimum/Maximum Autocorrelation Factors (MAF). MAF is an approach, based on principal
components, that spatially decorrelates the variables involved to non-correlated factors. MAF’s
spatial decorrelation at any lag distance is the main and critical difference of this approach from
the principal component approach attempted in the past. In the MAF approach, the independent
factors are individually simulated and back-transformed to the conditional simulations of the
correlated deposit attributes that reproduce the cross-correlations of the original variables.
Keywords: Joint simulation; mini/max autocorrelation factors; grade-tonnage curves.
2 [ T
]
Q2 M ( ∆ )Q2T = Q2 [ΓY ( ∆ )] + [ΓY ( ∆ )] Q2T
1
[4]
generate the three min/max autocorrelation factors in
Sectors 11 and 12 is shown in Table II. MAF are calculated
where matrix ΓY(∆) is an asymmetric matrix variogram at by multiplying the vector of elements Cu, Fe and K by a
lag distance ∆ for the regular PCA factors Y(x)=Z(x) A, vector of loadings from the rows of the transformation
where A = QΛ-1/2. In practice, several ∆ lag distances may matrix. It should be noted that the MAF loadings are quite
be used for values lower than the range and the resulting different from the ones derived by PCA8. The lag ∆ in
eigenvectors averaged. Equation [4] used in this example is 20 metres and was
Given the MAF transformation above, the joint derived experimentally by testing several lag distances to
simulation of multiple correlated variables using the MAF assure a suitable decorrelation and stable MAF
approach proceeds as follows: decomposition. Figure 3 shows examples of cross-
• Normalize the variables to be simulated variograms between MAF from the present study that
• Use MAF to generate the MAF non-correlated factors demonstrate variable decorrelation. Experimental
• Produce variograms for each MAF variograms and cross-variograms for Cu, Fe and K are
• Conditionally simulate each MAF using any Gaussian shown and discussed in more detail in a subsequent section.
simulation method
• Validate the simulation of factors Variography of MAF
• Back-transform simulated MAF to variables and Variography on each MAF is performed. Figure 4 shows
denormalize the experimental and model variograms fitted to the three
• Validate the final results MAF in Sector 11 and Sector 12. Note that all variogram
• Generate additional simulations, as needed. models are spherical. MAF variograms are subsequently
In most cases, the reblocking of the generated realizations used in the simulation of each factor and the validation of
to a block support model is required and may be seen as an the MAF simulation results. MAF variograms show clear
additional step in the above algorithm. spatial patterns, as expected. It should be noted that MAF
variograms are linear combinations of the variograms of the
original (normal score) variables.
The deposit and data available
The copper oxide deposit considered in this study is located Conditional simulation of MAF
in northern Brazil (Figure 1). It is approximately 2 km long Conditional simulation is performed independently on the
and 500 m wide and appears as a prominent NW–SE three MAF using a sequential algorithm 9 based on the
aligned hill. The copper mineralization is hosted by copper generalized sequential Gaussian simulation method10. The
oxide minerals, mainly malachite, in a weathered simulations are performed on a grid of 320075 nodes within
hydrothermal breccia and in a large weathering cap where the geological limits of Sector 11 and 239375 nodes within
copper is disseminated and enriched in sheet-like mineral- Sector 12. Twenty simulations are generated in this study
rich saprolitic units, as shown schematically in Figure 1 for each of the two sectors and are validated in detail for
(bottom). Metallurgical studies indicate that iron and reproduction of data, histograms and variograms. The
potassium are key elements for predicting copper recovery, validation of the MAF simulations is not presented here as a
indicating the need to evaluate all these three elements subsequent section presents the validation of realizations in
throughout the deposit. The orebody mineralization model the data space.
Back transformations of MAF Table II. Subsequently, the normal score Cu, Fe and K
The realizations of MAF were transformed back to realizations are back transformed to the data space, and
simulated normal score variables by multiplying a column point support realizations are reblocked to 10 x 10 x 8 metre
vector of simulated MAF in each grid node with the blocks. Figure 5 shows three realizations of Cu through
corresponding inverse matrix of the MAF loadings in horizontal sections and a vertical section of the deposit.
.200 .100
.100 .050
.000 .000
.0 2.0 4.0 6.0 8.0 .02 1.02 2.02 3.02 4.02 5.02 6.02 7.02 8.02
CU CU
.100
.100
.000 .000
123. 10123. 20123. 30123. 40123. 50123. 60123. 42. 20042. 40042. 60042. 80042. 100042.
K K
Frequency
.0200
.020
.0100
.0000 .000
2.7 7.7 12.7 17.7 22.7 0 10.0 20.0 30.0
FE FE
Table II
Transformation matrix AMAF for Sectors 11 and 12
1.20 1.20
Gamme (h)
Gamme (h)
0.90 0.90
0.60 0.60
0.30 0.30
0.00
70 140 210 280 350 0.00
80 160 240 320 400
Lag Distance (h)
Lag Distance (h)
1.20 1.20
Gamme (h)
Gamme (h)
0.90 0.90
0.60
0.60
0.30
0.30
0.00
120 240 360 480 600 0.00
90 180 270 360 450
Lag Distance (h) Lag Distance (h)
1.20 1.20
Gamme (h)
Gamme (h)
0.90 0.90
0.60 0.60
0.30 0.30
0.00 0.00
160 320 480 640 800 60 120 180 240 300
Lag Distance (h) Lag Distance (h)
Figure 4. Experimental and model variograms of the three MAF factors in Sector 11 (left column) and Sector 12 (right column)
Simulation 5
Cu%-Colour Profile
Simulation 10
Simulation 10
Simulation 15
Simulation 15
Figure 5. Selected conditional simulations of Cu for Sectors 11 and 12 for level #325 (left column) and cross-section 250NW (right column)
Validation of the joint Cu-Fe-K simulation results simulation based on MAF, which used the variograms of
Several validation checks are performed to assess the the MAF independent factors.
results of the joint simulations of Cu, Fe and K using the
MAF transformations in both Sectors 11 and 12. Validation Risk assessment results and discussion
involves calculation of histograms, experimental The mining project considered in this study represents a
variograms and cross-variograms of the simulated US$120 million capital investment and aims to operate an
realizations to ensure reproduction of original data SX-EW copper recovery plant using traditional sulphuric
characteristics. heap leach technology. One of the most important factors in
Figure 6 shows histograms from simulated realizations of the economics of this project is the variability of
each variable in both sectors, and these can be compared to recoverable copper and the associated risk. Preliminary
the histograms of original data in Figure 2. Figure 7 shows metallurgical tests, summarized in Table III, indicate that
plots of variograms and cross-variograms for data and recoverable copper is controlled by the content of iron and
simulations in Sector 11. All results suggest that the potassium, in addition to the original copper grade. This
reproduction of the original data spatial characteristics by highlights the importance of integrating the geological
simulation is excellent. The reproduction of data model of the deposit with jointly simulated orebody
variograms and cross-variograms for Sector 12 is also scenarios and metallurgical results so as to better identify
excellent. Recall that the variograms and cross-variograms the possible variability of copper ore output (Cu tonnes)
of original variables are not directly used in the joint from the SX-EW recovery plant.
median .2400
.150
Frequency
lower quartile .1214 median .4460
minimum .0130 lower quartile .2028
.200 minimum .0180
.100
.100
.050
.000
.000
.0 2.0 4.0 6.0 8.0
.02 1.02 2.02 3.02 4.02 5.02 6.02 7.02 8.02
Back Transform
Back Transform
Frequency
lower quartile 742.0432 lower quartile 1316.7340
minimum 123.0000 minimum 42.0000
.100 .100
.000 .000
123. 10123. 20123. 30123. 40123. 50123. 60123. 42. 20042. 40042. 60042. 80042. 100042.
Back Transform Back Transform
.0200 .0200
.0100
.0100
.0000
.0000 .0 10.0 20.0 30.0
2.7 7.7 12.7 17.7 22.7
Back Transform
Back Transform
Figure 6. Histograms for joint simulation of Cu, K, Fe using MAF. Sector 11 is shown on the left column and Sector 12 in the right
Given the above reasoning, the assessment of Cu average (e-type) from the simulations. It is apparent from
variability and risk in grade-tonnage curves for the deposit all grade-tonnage graphs that recoverable copper tonnage
is based on recoverable Cu. Recoverable Cu is calculated as variability is more clearly related to the in situ copper and
a function of the jointly simulated Cu, Fe and K contents potassium content than to iron content. Although there is no
for each block of the orebody models and the relationships restriction to a specific cutoff grade for Fe and K in the
derived from metallurgical tests, shown in Table IV. The metallurgical process, it is clear that for Fe-rich ores no
joint simulations described in the previous section provide wide variability in copper output is apparent in the grade-
the information needed to generate realizations of tonnage curves. This is supported by the geological model,
recoverable Cu at the support scale of 10 x 10 x 8 metre where Fe-rich grade zones are normally rich in malachite, a
blocks, and to subsequently generate grade-tonnage curves mineral with better kinetics in the metallurgical process,
for the deposit based on a range of Cu, Fe and K grades. compared with iron for acid consumption, and therefore not
The resulting grade-tonnage curves are then suitable for reducing copper recovery. Variability in recoverable copper
risk analysis as well as for ‘average type’ assessments for grade is, however, mainly dependent on the in situ copper
project economics. content. These results reflect the related metallurgical tests
Figures 8, 9 and 10 present the grade-tonnage curves for and the recovery functions developed, where the in situ
recoverable Cu as a function of Cu, Fe and K cutoffs. All copper content is the predominant control on recoverable
Figures plot the results from individual simulations and the copper.
Table III
Preliminary metallurgical Cu recovery tests for the deposit under study
Table IV
Recoverable Cu equations from metallurgical tests and for different rock types
1.40
300000.00
Avg Cu (Recovered)
1.20
Tonnage Cu (t)-
1.00
Scaled
Grade %
200000.00 0.80
0.60
100000.00 0.40
0.20
0.00 0.00
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1
Cutoff Cu%
Figure 8. Grade-tonnage curves for recoverable Cu as a function of Cu cutoffs. Thin lines represent individual simulations and thick dashed
lines represent the average (e-type) from the simulations
400000.00 0.70
0.60
300000.00
Avg Cu (Recovered)
0.50
Tonnage Cu (t)-
0.40
Scaled
Grade %
200000.00
0.30
100000.00 0.20
0.10
0.00 0.00
0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00
Cutoff Fe%
Figure 9. Recoverable copper tonnage as a function of iron content; results from individual simulations and average (e-type) from the
simulations
The grade-tonnage curves in Figures 8, 9 and 10, show tonnes, the results suggest that the potential exists for the
that the assessment of variability of recovered copper can project life to be substantially expanded.
be investigated by integrating all three elements with the At present, no restrictions to any specific cutoff grades
geology of the deposit. This assessment of variability for Fe and K are considered in the evaluation of the deposit.
provides a clear description of the inherent uncertainty of However, it is clear from Figure 9 that for the Fe-rich ores,
potential copper outputs at the plant when analysing cash which are mainly composed of hematite and goethite
flows and other financial aspects in the evaluation of the oxidized breccia zones, there does not appear to be wide
present project. For example, consider a preliminary variability in recoverable copper outputs in the grade-
operational Cu cutoff grade for the project of 0.6% Cu. tonnage curves. This may result from the regular
Figure 8 suggests that the output from the SX-EW recovery occurrence of malachite with Fe-rich ores, which has better
process could range from 145000 to 210000 tonnes of kinetics in the metallurgical process, compared with iron,
copper. With a planned annual Cu production of 45000 for acid consumption.
Tonnage Cu (t)-
0.30
200000.00
0.20
100000.00
0.10
0.00 0.00
0 400 800 1200 1600 2000 2400 2800 3600 4000
Cutoff K (ppm)
Figure 10. Recoverable copper tonnage as a function of potassium content; results from individual simulations and average (e-type) from
the simulations