You are on page 1of 10

HOME

Faulted Geological Model Simulation of the Resolution Porphyry


Copper Deposit
G Verly1, K Brisebois2, W Hart3 and J Hammitt4
ABSTRACT
A significant portion of the risk in a mining operation is tied to the
geologic model that is being used to constrain the resource and reserve
estimates. Yet little has been published on geological model simulation in
the mining industry, perhaps because the original geologic setting has often
been disturbed by metasomatism and tectonism, making it often difficult to
interpret and simulate. One general method to realistically simulate the
geology consists of first establishing a priority ranking of the geological
features and then simulating them one at a time. This can be done by
starting with the interpreted outlines or shapes and incorporating the
uncertainties as specified by the geologist during the simulation process.
Two steps that could be part of this methodology are considered in this
paper: simulation of fault surfaces, and simulation of layered rock types
within the simulated faulted blocks. The algorithm consists of a series of
sequential Gaussian simulations that are merged together after various
rescaling to account for non-stationary uncertainties. A detailed description
of the procedure is given. Examples of results obtained on a porphyry-style
copper deposit are provided.

INTRODUCTION
A significant portion of the risk in a mining operation or study is
tied to the geological model that is being used to provide domains
for mineral resource and reserve estimation and/or conditional
simulation. Simulating the geology is a common procedure in the
petroleum industry but is rather the exception in the mining
industry. Perhaps one reason is that the original geological setting
of a mineral deposit has often been disturbed by metasomatism
and tectonism, making realistic simulations of the geology
difficult to produce. Some of the methods that can be used to
simulate different aspects of the geology are boolean modelling
(de Fouquet et al, 1989), indicator simulation (Alabert, 1987),
plurigaussian simulation (Armstrong et al, 2003), probability field
with local means (Srivastava, 2005) and potential field (Chiles et
al, 2007).
The case considered in this paper is the Resolution
porphyry-style Cu-Mo deposit, located in Arizona, USA. The
deposit is deep and sparsely drilled. The geological interpretation
is complex and contains faults, metamorphosed rocks of
sedimentary origin, intrusions, breccias and alteration assemblages. One general method to realistically simulate such complex
geology consists of establishing a priority ranking for each of the
geological features and simulating them one at a time, starting
with the interpreted outlines or shapes and incorporating the
uncertainties as specified by the geologist (Verly, Bridebois and
Hart, 2008). Details on how to simulate a fault block model
together with faulted rock types are presented in this paper along
with some results.
1.

AMEC, Suite 400, 111 Dunsmuir Street, Vancouver BC V6B 5W3,


Canada. Email: georges.verly@amec.com

2.

AMEC, 780 Vista Boulevard, Sparks NV 89434, USA.


Email: ken.brisebois@amec.com

3.

Resolution Copper Mining, LLC 102 Magma Heights, Superior AZ


85273, USA. Email: william.hart@riotinto.com

4.

MAusIMM, Geologist, Kennecott Exploration Company, Rio Tinto,


224 North 2200 West, Salt Lake City UT 84116, USA.
Email: jay.hammitt@riotinto.com

Advances in Orebody Modelling and Strategic Mine Planning I

GEOLOGY
Deposit geology
The Resolution Cu-Mo deposit is late-Cretaceous to earlyTertiary in age and is hosted within a buried, fault-bounded
sequence of Paleozoic and Precambrian sedimentary strata,
Precambrian diabase sills and Cretaceous-aged layered
volcaniclastic and siliciclastic rocks. Host strata are faulted and
have been intruded by porphyry bodies of late Cretaceous age.
Mineralised breccia bodies have also been identified, with some
being spatially related to faults and porphyry intrusions (Hammitt
and Ballantyne, 2007). The mineralised rocks are buried
unconformably beneath a 1000 m to 1500 m thick sequence of
barren sediments and volcanic rocks of Tertiary age. Figure 1
shows a plan view and two vertical sections of a portion of the
2007 geological interpretation.

Fault interpretation and uncertainty assessment


Figure 2 illustrates a common method used by geologists for the
identification of simple block bounding faults where folding is
assumed to be minimal, using a distinctive lithologic marker
intersected in multiple drill holes (eg a formation contact).

Stage 1
Distinctive lithologic marker intercepts in each drill hole are
visually identified in section view (red symbols) and grouped
based on their apparent relative degree of colinearity (or
coplanarity).

Stage 2
A 2D marker horizon is interpreted and constructed, defined by a
smooth line (red dashes) connecting all controlling data points for
the marker within each geometric grouping. Zones of abrupt and
significant changes in position (eg elevation) between different
segments are suspected locations for fault displacement. Fault
planes are then interpreted and constructed within these suspect
zones based on numerous geological principles including:

observed apparent sense of displacement of the marker


horizon;

actual fault intercepts in one or more drill holes (true snap


points);

typical fault geometries observed nearby on surface or in


underground workings; and

fault geometries predicted by well-established structural and

tectonic field studies within a similar geologic terrane, or


predicted by laboratory experiments on rock.

Stage 3
The spatial uncertainties of faults are assessed purely by distance
from controlling data points. Multiple fault snap points that are
spatially aligned may greatly reduce the uncertainty of a given
fault interpretation. Simple surfaces are then constructed that
define the limits of all permissible fault positions (blue dashed
lines).

Spectrum Series Volume 179

299

G VERLY et al

N=1006

E=206

N=1006
E=206

High Cu Shell

FIG 1 - Plan view and vertical sections through the geological interpretation. The black dots on the plan view indicate the drill hole intercepts.

FIG 2 - Diagram illustrating the methodology for the interpretation of simple faults (Stage 2, solid blue lines) in section and plan view, using
only lithologic markers intersected in irregularly-spaced or widely-spaced drill holes (Stages 1 and 2). Stage 3 illustrates the potential
degree of spatial uncertainty (orange polygons) inherent in the resulting fault model where all possible fault positions and orientations have
equal probability (dashed blue lines).

300

Spectrum Series Volume 17

Advances in Orebody Modelling and Strategic Mine Planning I

FAULTED GEOLOGICAL MODEL SIMULATION OF THE RESOLUTION PORPHYRY COPPER DEPOSIT

Faulted layered rock type interpretation and


uncertainty assessment
As described in the previous section on fault interpretation, the
boundaries for the Faulted Rock Types (layered rocks) are
interpreted and constructed on sections or in three dimensions by
first identifying the controlling drill hole data points for each
major stratigraphic horizon, and then constructing smooth lines,
polygons, or surfaces that pass through these controlling points
and enclose lithologically distinct domains. As with the faults,
every attempt is made to honor established geological principles of
stratigraphy. The resultant interpreted domains for the Faulted
Rock Types are then modelled as triangulations. Based on existing
information from both drilling and surface mapping, analysis of
the geological interpretation is performed far from drill hole
intercepts the limit of variation for the vertical position of
domain boundaries was assessed to be 30 m. For thickness, the
limit of variation was assessed as 20 per cent of the interpreted
thickness.

FAULT MODEL SIMULATION


Seven faults defining eleven fault blocks have to be simulated. The
general procedure consists in simulating one fault at a time as
a 2D surface and flagging a 3D grid accordingly. The seven
3D grids corresponding to the seven faults are then merged
together using the proper priorities. Each 2D surface simulation is
constrained within a variable and asymmetrical uncertainty bandwidth assessed by the geologist. The simulation methodology is
further described below and is illustrated in Figures 3 and 4.
For each of the 7 faults
Step 1: Pre-simulation processing
1a) Get original fault information:
- Fault surface (triangulation)
Drape the fault surface (dense grid)
- Uncertainty information (series of points with 2 max. distances from
interpretation)
- Snap points
1b) Get fault plane + offsets:
- Work out approximate fault plane
- Work out fault surface offsets from fault plane (draped grid)
1c) Rotate the fault information:
- Work out rotation to bring the fault plane to horizontal
- Rotate the offset information
- Rotate the uncertainty information
- Rotate the snap points
1d) Grid the rotated fault information:
- 2D horizontal grids
- Grid the offsets and the uncertainties

Step 2: 2D simulation
2a) 2D stationary simulation, conditional to snap points
2b) Rescale the simulation to account for variable uncertainty
2c) Add the offsets
100 rotated fault surfaces

Step 3: 3D simulation
3a) Rotate the 3D grid to be simulated
3b) Flag the rotated grid nodes
- Flag as above/below the simulated surfaces
100 fault simulations coded as 0/1 3D grids

Merge the simulated 3D grids


Merge grids with proper priorities
100 simulated fault grids, 7 faults / 11 fault blocks per realisation

FIG 3 - Faulted model simulation process.

Presimulation work
The available information for one fault consists of the fault
surface, the uncertainty information and three snap points for one
fault. The fault surface is modelled as a triangulation. The
uncertainty information is provided as a series of locations on the
fault surface with two maximum possible envelopes, one on each
side of the fault. The spacing between the locations is about 500 m
500 m. In a first step, the fault triangulation surface is sampled
on a dense grid using a draping technique. The approximate plane
of the fault is identified, and the normal offsets of the fault surface

Advances in Orebody Modelling and Strategic Mine Planning I

to the plane are calculated. These kinds of manipulations are


readily available in several resource modelling packages. Note that
the offset terminology is used for the distance from the interpreted
fault surface to the fault plane. It is not the displacement of the
fault as in normal geological parlance.
The fault plane is rotated to the horizontal. The coordinates of
the fault information (surface offsets from the plane, uncertainty
and snap points) are similarly rotated. The reason for the rotation
is that most 2D simulation algorithms do not handle 2D planes of
any orientation. Furthermore, the post-simulation grid manipulations are simplified in the rotated space. After rotation, the
fault surface offsets are vertical distances from the horizontal
plane. Because the angle of the fault surface with the fault plane
at any location is always small, the maximum possible offsets at
the recorded uncertainty locations are assumed to be vertical
after rotation. The fault surface is clipped by the other faults.
Different realisations will result in different clipping. To avoid
gaps between the simulated fault blocks, the simulated 2D grid
needs to be extended sufficiently beyond the extent of the rotated
interpreted fault surface. The rotated fault surface offsets and the
vertical uncertainties are estimated on a dense 2D grid (Step 1d,
Figure 3). For the offsets, the estimation method does not matter
much as the draped information is very dense. This
information, however, was available only on the clipped fault
surface and was estimated within the rotated outline of the
surface. The estimated values had to be extended to the limit of
the 2D horizontal grid. For the uncertainty, the available points
are much farther apart and ordinary kriging was used with a
spherical model with no nugget and very long ranges. Inverse
distance could have been used as well. No extensions were
necessary because the information was available for unclipped
surfaces. Figure 5 shows the estimated uncertainties on one side
of one fault.

Simulation
The rotated fault simulation is achieved in three steps:
1.

conditional simulation of standard normal scores,

2.

rescaling the simulated normal scores to account for the


non-stationary and asymmetric uncertainty bandwidth, and

3.

adding the offsets to the rescaled normal scores.

The conditional simulation is sequential Gaussian. A unit-sill


Gaussian variogram model with no nugget effect and a 1000 m
isotropic range has been used to ensure a very smooth surface. The
1000 m range was chosen after some experimentation. At this
stage, 100 realisations of standardised fault surface deviations are
available. Theoretically, the 100 simulated values at a given
location are standard normally distributed except within the zone
of influence of a snap point. The simulated values of one
realisation should also be standard normally distributed if the
surface extent is very large with respect to the variogram range.
Figure 6 shows the validation statistics for one fault surface. The
top left histogram (a) shows that the mean simulated value is very
close to zero as expected, whereas the standard deviation of a
given realisation is on average 0.66, which is lower than 1. This is
due to three snapping points that significantly reduce the possible
fluctuations and also due to the extent of the simulation being not
very large with respect to the 1000 m variogram range. The
bottom left graph (c) indicates that the variogram is reproduced by
the simulation even though significant fluctuations are observed
from one realisation to the next (d). Alfaro (2008) notes that wild
variogram fluctuations between realisations are to be expected
because the Gaussian variogram is not microergodic. The
probability plots (b) of the realisations are not straight lines
because the variances are less than one.
A local rescaling of the simulated normal scores is needed to
account for the non-stationary uncertainty. If the uncertainty was

Spectrum Series Volume 179

301

G VERLY et al

1) Rotate
F1

a) Fault

4) Flag rotated 3D grid

2) Create dense 2D grid

3) Simulate fault in 2D
a) simulate deviations
b) rescale deviations to account
for the uncertainty
c) add fault offsets

Fault
Plane

F1

Horizontal
b) 3D grid

5) Rotate back the grid


F1

F2

F1
6) Merge
grids
with
proper
priorities

F2

FIG 4 - Faulted model simulation main steps.

Rotated Fault -- Fault SE Side Uncertainty


400

600

800

1000

1200

1400

1600

1800

2000

2200

2400

1000

1000

800

40

14

40

49

202

74

800

14

17

30

206

77

200

46

400

600

Uncertainty (m)

800

1000

1200

40

1400
1600
Along Strike

3D Grid / Fault Intersept


0.

104

281

1800

2000

"Samples"

2200

Two sets of coordinates of the simulated 3D grid nodes are


available:
1.

original unrotated, and

2.

rotated as per the fault rotation.

-200

37

51

2400

2600

"Snap Points"

250.

FIG 5 - Gridded uncertainty on one side of the fault after rotation.

symmetrical on each side of the fault, the rescaling factor at a


given location would be a third of the maximum distance
estimated for the uncertainty at that location meaning that the
standard deviation of the rescaled simulated normal scores is a
third of the maximum distance. This ensures that approximately
99.75 per cent of the simulated fluctuations are within the
specified maximum ( 3 standard deviations). Simulated
fluctuations outside that interval are reset to the maximum
distance. The uncertainty at a given location, however, is
asymmetrical, and two rescaling factors are used, depending on
the sign of the simulated normal scores. The distribution of the
rescaled simulated normal scores consists of the halves of two
normal distributions with same zero mean but different standard
deviations. The two standard deviations are one third of the
maximum possible distances at that location. Figure 7 shows an
example of such a distribution with 10 m and 15 m standard
deviations, corresponding to maximum fluctuations of -30 m and
+45 m, respectively. Note that the mean of the resulting
distribution is not zero anymore.
One hundred realisations of fluctuations around the fault
interpretation are available at this stage. Figure 8 shows one
section with the trace of one rotated fault interpretation (red line
at deviation = 0), the maximum possible deviations (blue lines on
top and bottom) and 20 realisations of the simulated fluctuations

302

Post-simulation work

(grey lines). Note that the offsets from the horizontal have not yet
been added, which is why the interpreted fault trace lies at
deviation = 0. The final simulation step is adding the fault offsets
onto the rescaled fault fluctuations, resulting in 100 rotated
simulated fault surfaces on a dense 2D grid.

200

34

Down-Dip
400

Down-Dip
400

600

600

-200

2600

Spectrum Series Volume 17

The second set of coordinates is used to flag the rotated grid as


above or below the simulated rotated fault, which is equivalent to
flagging the original grid as one or the other side of the un- rotated
fault. The previous steps presimulation work, simulation, 3D
grid flagging are repeated for each fault. When this is
completed, seven simulated 0/1 fault indicator grids are available,
with 100 realisations each. The final simulated fault block model
is obtained by merging the seven individual fault simulated models
together with the proper priorities. Figure 9 shows eight realisations of the simulated fault model at mid-elevation in the model.

Results and discussion


Table 1 shows the statistics for the simulated fault block models.
To preserve confidentiality of the original data, each of the
interpreted fault block volumes has been reset to 100. The table
shows that the interpreted (modelled) fault block volumes are not
always reproduced. Differences from -14 per cent to +10 per cent
are observed. These differences are due to the asymmetric fault
uncertainties. Some simulated fault blocks have a greater chance
to become smaller than interpreted and vice versa. The table also
shows there can be very large differences between two realisations
for a given fault block. For example the average simulated volume
for Fault Block No 1 is 89, with minimum and maximum values
of 20 and 135. The coefficient of variation (CV) is 28.3 per cent.
Assuming a normal distribution, there is a 95 per cent probability
that the simulated Fault Block No 1 volumes are within 56.6 per
cent of the average.
The approach used to simulate the faults is considered
reasonable. Indeed, the results are statistically consistent with the

Advances in Orebody Modelling and Strategic Mine Planning I

FAULTED GEOLOGICAL MODEL SIMULATION OF THE RESOLUTION PORPHYRY COPPER DEPOSIT

Fault Simulation - Simulated NSCO Checks


0.300

99.99

A)

B)

99.9
99.8

Cumulative Probability

0.250

0.200

100 simulations
0.150

Nb. of data 5,109,700

y
c
n
u
q
re
F

0.100

mean
stdv

0.060
0.663

max
min

4.168
-3.137

99
98
95
90
80
70
60
50
40
30
20
10
5
2
1

0.050

100 Simul.
Mean Simul.

0.2
0.1
0.000
-4.00

-3.00

-2.00

-1.00

0.00

1.00

2.00

3.00

4.00

0.01
-3.0

5.00

-2.5

-2.0

-1.5

-1.0 -0.50

0.

0.50

1.0

1.5

2.0

2.5

3.0

NSCO

NSCO

C)

D)

1.20

1.20

0.80

0.80

io
V
m
ra
g

io
V
m
ra
g

0.40

Azm/Plg=0/0
Azm/Plg=0/0
Azm/Plg=90/0

0.40

Model
Mean simulation

Model
Mean Simul.
0.00

0.00
0.

200.

400.

600.

800.

1000.

0.

200.

Distance

400.

600.

800.

1000.

Distance

FIG 6 - Simulated normal score statistics. (A) Mean of realisation histograms; (B) realisation cumulative probability curves (grey) plus mean
of the curves (red thick line); (C) variogram model (green) plus mean of realisation variograms (red dashes); (D) same as in (C) plus ten
realisation variograms along two directions (grey full and dashed lines).

TABLE 1
Simulated and interpreted fault block volume statistics. The
interpreted (model) fault block volumes have been reset to 100.
Fault
block

100%

-30m -20m -10m

Simulation volume

Model
volume

Model to
simulation
% Chg

28.3%

100

-10.8%

16.8%

100

7.8%

86

22.2%

100

-13.9%

133

101

11.0%

100

1.4%

42

201

109

28.1%

100

8.6%

52

156

104

22.2%

100

3.6%

80

124

99

9.6%

100

-0.7%

63

171

103

20.6%

100

3.2%

Min

Max

Average

CV

20

57

135

89

147

108

33

114

79

5
6

95%

51

135

96

19.6%

100

-4.1%

10

501

176

110

29.9%

100

10.1%

67%

11

83

120

99

8.2%

100

-0.6%

0m

+15m

+30m

+45m

FIG 7 - Example of asymmetrical fault fluctuation distributions.


Two normal distributions with zero means but different standard
deviations are, in effect, glued together. In this example, the
standard deviations are ten and 15 m. Simulated values outside
the 3 standard deviation interval are reset to three standard
deviations.

Advances in Orebody Modelling and Strategic Mine Planning I

input data, the simulated faults are conditioned to the available


snap points and the fault location uncertainties as assessed by the
geologist have been reproduced. The simulated fault surfaces are
visually reasonable and have more chance to be close to the
interpretation than far away. On average, the interpreted fault
block volumes are generally well reproduced by the simulation.
Where there are discrepancies, they can be explained by the
asymmetrical input uncertainties. Although not necessary for the
Resolution deposit, different aspects of the simulation could be
modified and/or improved. With the current simulation

Spectrum Series Volume 179

303

G VERLY et al

-250 -200 -150 -100 -50

1000

1200

1400

1600

1800

2000

2200

2400

800

1000

1200

1400

1600

1800

2000

2200

2400

Along strike
Uncertainty bandwidths

Interpretation (0 Deviation)

Deviations
50
0 - 50 -100 -150 - 200 - 250

600

800

250 200 150 100

Deviations
0 50 100 150 200 250

Simulated fault deviations from interpretation


600

20 Realisations

FIG 8 - Simulated fluctuations from one fault interpretation.

11

21

31

41

51

61

71

81

FIG 9 - Eight realisations of the fault block model at mid elevation in the model. The red lines represent the original fault interpretations.
The grey lines represent the simulated fault traces.

methodology, the total number of faults before and after


simulation is static, as per the geological interpretation. The two
semi-normal distribution rule used to rescale the fault fluctuations
could be changed to generate wider or narrower fluctuations within the specified uncertainty bandwidths. The range of the Gaussian
variogram could be revisited. The simulation methodology could
be changed to generate surfaces that are more planar.

FAULTED ROCKTYPE SIMULATION


In the simulation described below, faulted rock types represent
rock types that are interpreted to have been offset by faulting.
These rock types consist of stacks of subhorizontal layers within a
background made of two rock types separated by an unconformity.
The faulted rock types are intruded by dykes and breccias that do
not appear to be offset by faulting. The general procedure for
simulating faulted rock types consists in processing one fault
block at a time. The preintrusive/prebreccia rock types are

304

Spectrum Series Volume 17

reconstructed. The 3D interpretation of each layer is converted to a


2D interpretation of elevations and thicknesses. 2D simulations of
the elevations and thicknesses are performed, conditional to drill
hole intercepts and accounting for the uncertainties specified by
the geologist. The 2D simulation results are converted back to 3D.
The faulted blocks are stitched back together at the very end. This
simulation methodology is further described below and is
illustrated in Figure 10.

Presimulation work
The available information consists of a 3D gridded geological
model, uncertainty information and drill hole sample locations.
The uncertainty information is the same for all layers a 95 per
cent confidence interval (two sigma) of 30 m from the
interpreted elevation and of 20 per cent from the interpreted
thickness. Note the deviations for thickness are in terms of a
relative percentage of the interpreted (modelled) thickness.

Advances in Orebody Modelling and Strategic Mine Planning I

FAULTED GEOLOGICAL MODEL SIMULATION OF THE RESOLUTION PORPHYRY COPPER DEPOSIT

For each fault block


Step 1 - Pre-simulation processing 3D
1a) Get the maximum simulated fault block footprint
1b) Get 3D Indicator Grid:
- Generate an indicator grid of the rock types
- Reconstruct pre-intrusive rock types
- Extend rock types from interpreted to maximum simulated fault block limit

Repeat Steps 2, 3, and 4 for each stack of layers in the fault block
Step 2 - Pre-simulation processing 2D
2a) Create 2D grids of the elevation and thickness of each layer
2b) Create 2D conditioning sample dataset from drill-hole intercepts

Step 3: Simulation 2D
3a) 2D stationary simulations of deviations, conditional to snap points.
3b) Rescale simulated deviations
3c) Incorporate rescaled deviations to interpretations.
- Adjust elevations for gaps or overlaps

Step 4: Simulation 3D
4a) Flag 3D grid from 2D simulations

Step 5: Post-simulation processing 3D


5a) Recombine the 3D grids corresponding to the stacks into one 3D grid.

Nine sets of 100 realizations (one set per fault block)


Stitch the faulted block rock types back together
For a given realization, pick the simulated rock type according to its location and
simulated fault block number at that location
100 simulated faulted rock type grids

FIG 10 - Faulted rock type simulation process.

As previously mentioned, the faulted rock types are processed


within one fault block at a time. In a first step, the
preintrusive/prebreccia faulted rock types are reconstructed
(Step 1b, Figure 10 and Figure 11). This can be done by either
using the preintrusive wireframe model, or by eroding away the
intrusives/breccias from the grid model. The reconstructed rock
types are then extended to the maximum possible fault block
limits. Again, this could be done using the original wireframe
model or numerically approximated on the grid model. The
maximum extent of the simulated fault block is obtained by
scanning the realisations of the previously simulated fault model.
Each reconstructed and extended stack of layers is processed in
turn. The 3D interpretation of each layer is converted to a 2D
grid of elevations and thicknesses. The few drill hole intercepts
one to five per fault block are converted to 2D snap points.

Simulation
A series of 2D simulations are performed for each stack:

same simulation of elevation fluctuations for all layers (will


tend to prevent crossovers), and

one simulation of thickness fluctuations per layer.


Standard normal scores are simulated first, followed by some
rescaling to account for the uncertainty. The simulations are
conditional sequential Gaussian using a unit-sill Gaussian
variogram model with no nugget effect and a 500 m isotropic
range to ensure smooth surfaces. The checks at this stage consist
in verifying that the averages of the simulated normal scores are
close to zero. The variances of the simulated normal scores are
not expected to equal one because of the conditioning and the
relatively small extent of the fault blocks compared to the
variogram range.
Rescaling the simulated normal scores is needed to account for
the uncertainty. The assumption is that the simulated fluctuations
far from snap points are normally distributed with 95 per cent of
the values within the specified uncertainty. In other words, the
specified uncertainty is equivalent to two standard deviations of
the normal distribution. The uncertainty for the elevation is 30 m
the simulated normal scores are multiplied by 15, and reset to 45
m if they exceed these values. The uncertainty for the thickness is

Advances in Orebody Modelling and Strategic Mine Planning I

20 per cent of the interpreted thickness the normal scores are


multiplied by ten per cent and then reset to 30 per cent if they
exceed these values. The final simulation step incorporates the
interpretation and simulation together. The simulated elevation at a
location is the interpreted elevation plus the rescaled simulated
deviation at that location. The simulated thickness is the
interpreted thickness plus the product of interpreted thickness and
the rescaled relative simulated deviation. Elevation corrections are
made to avoid gaps and overlaps between the simulated layers
within a same stack.
One hundred realisations of elevations and thicknesses are
available at this stage for each layer. Figure 12 shows one section
with the trace of one layer interpreted elevation and thickness
(red lines), the maximum possible deviations (blue lines) and 20
realisations of the elevation and thickness (grey lines). Three
snap points were available for the elevation but only two for the
thickness because of an incomplete hole intercept. The jagged
aspect of the thickness profiles is due to the exaggerated
resolution of the vertical grid.

Post-simulation work
One hundred realisations of one stack of layers within a fault
block are available at this stage as a series of 2D grids of layer
elevations and thicknesses. For each realisation, the layers top
and bottom elevations are calculated and used to flag one 3D
grid. The simulation procedure is repeated for the other stacks
within the fault block. The 3D grids are then merged together.
The procedure is repeated for the other fault blocks resulting in
11 individual simulated models. The final simulated faulted rock
type block model is obtained by merging the eleven models
together. For a given realisation and a given location, the
simulated fault block is first identified. The simulated rock type
is then picked from the rock type model that corresponds to the
fault block. Figure 13 shows eight realisations of a vertical
section through the faulted rock type model at about mid
northing in the model.

Results and discussion


Table 2 shows the statistics of the simulated faulted rock type
model. To preserve confidentiality, each of the interpreted rock

Spectrum Series Volume 179

305

G VERLY et al

FIG 11 - Faulted rock types presimulation steps. (A) Original interpretation; (B) maximum simulated fault block footprint identified
on left and right; (C) preintrusive/breccia reconstruction; (D) extension to fault block maximum simulated footprint; and
(E) 2D grid of elevations and thicknesses of the mid-layer (pzls).

1 000

1200

1400

250
200

150

800

1 000 Northing 1200

1400

75

75 150

Elevation

200
25

25

50

50

Thickness

Thickness

Thickness

Elevation

250

Elevation

300

300

Mid-Layer Simulated Profiles


800

800

1000

1200

1400

Nor thing

Interpreted elevation/thickness
Snap locations (within 100 m of section)

Elevation +/- 30 m/thickness +/- 20 per cent


20 Realisations

FIG 12 - Simulated elevation and thickness for one layer.

306

Spectrum Series Volume 17

Advances in Orebody Modelling and Strategic Mine Planning I

FAULTED GEOLOGICAL MODEL SIMULATION OF THE RESOLUTION PORPHYRY COPPER DEPOSIT

TABLE 2
Simulated and modelled faulted rock type volume statistics.
The model rock type volumes have been reset to 100.
Rock
type

Simulation volume
Min

Max

Average

CV

Model
volume

Model to
simulation
% Chg

additional uncertainty should probably be injected in the model. A


sensitivity study of the results to the range of the variogram would
be useful, as local variations could be important as the mine get
started.

CONCLUSIONS

kvs

98

102

100

0.8%

100

-0.1%

kqs

90

113

101

4.6%

100

1.2%

pzls

89

110

99

4.7%

100

-0.7%

mesc

89

114

101

4.8%

100

1.1%

qzite

91

104

99

2.5%

100

-1.5%

A general method to realistically simulate the geology is


suggested in this paper. The method consists of first establishing a
priority in the geological features and of simulating them one at a
time. The simulation starts with the interpreted outlines or shapes
and incorporates the uncertainties as specified by the geologist.
Two steps that could be part of such methodology were further
developed:

diab

97

103

100

1.4%

100

0.2%

1.

simulation of a faulted block model, and

2.

simulation of layered rock types within the faulted blocks.

type volumes has been reset to 100. The table shows that on
average the interpreted (modelled) rock type volumes are
relatively well reproduced by the simulation. Differences from
-1.5 per cent to +1.2 per cent are observed, which is considered
acceptable, especially since part of these differences is related to
the simulated fault blocks with their asymmetric uncertainties of
boundary positions. The table also shows the kind of differences
that can be observed between two realisations. For example the
average simulated volume for pzls (Paleozoic limestone) is 99
with minimum/maximum counts of 89 and 110 respectively. The
coefficient of variation (CV) is 4.7 per cent. Assuming a normal
distribution, there is a 95 per cent probability that the simulated
pzls volumes are within +/- 9.5 per cent of the average. These
differences are significantly less than those observed for the fault
block volumes (Table 1). This is partly due to the fact that the
same rock type is found in most faulted blocks. The variability of
the rock types within any given faulted block is more significant.
The approach used to simulate the faulted rock types is
considered reasonable. Indeed, the fault blocks are simulated. The
order relations between the simulated rock types are reproduced.
The simulated results are statistically consistent with the input
data. The simulated rock types are conditioned to the drill hole
intercepts. The fluctuations in elevation and thickness reflect the
uncertainty assessed by the geological staff. Still, several aspects
of the simulation could be improved. The method assumes that
there are enough holes to get the interpretation of rock sequence
100 per cent right at each node. This is optimistic and some

The method is relatively simple because the geological features


are simple once they are treated one-at-a-time. For example, the
faults are surfaces and the faulted rock types are layers. Both
features were simulated by simple 2D sequential Gaussian
simulations with various rescaling to account for the uncertainties
specified by the geologist. Last, the method is flexible in the sense
that it can cope with complex geological models, variable
uncertainties and a scarcity of data. The method is made powerful
by its incorporation of geological interpretation and uncertainty
assessments, although these can be subjective.

ACKNOWLEDGEMENTS
The authors wish to thank Resolution Copper Mining, LLC and
AMEC for permission to publish this paper. They also thank Harry
Parker and Geoff Ballantyne for their useful comments and
suggestions.

REFERENCES
Alabert, F, 1987. Stochastic imaging of spatial distributions using hard
and soft information, Masters thesis, Department of Applied Earth
Sciences, Stanford University, p 198.
Alfaro, M, 2008. Microergodicity and geostatistical simulation, in
Proceedings Eighth International Geostatistics Congress:
Geostatistics Santiago 08 (eds: J M Ortiz X and Emery), 1:409-417
(FCFM U: Chile).

11

21

31

51

61

71

41

81

C u Sh ell

FIG 13 - Simulated faulted rock type sections for eight realisations, in the northern part of the simulated area.

Advances in Orebody Modelling and Strategic Mine Planning I

Spectrum Series Volume 179

307

G VERLY et al

Armstrong, M, Galli, A, Le Loch, G, Geffroy, F and Eschard, R, 2003.


Plurigaussian Simulations in Geosciences, 160 p (Springer: Berlin).
Chiles, J P, Aug, C, Guillen, A and Lees, T, 2007. Modelling the
geometry of geological units and its uncertainty in 3D from structural
data: The potential-field method, in Orebody Modelling and Strategic
Mine Planning, second edition (ed: R Dimitrakopoulos), pp 313-320
(The Australasian Institute of Mining and Metallurgy: Melbourne).
De Fouquet, C, Beucher, H, Galli, A and Ravenne, C, 1989. Conditional
simulation of random sets, application to an argilaceous sandstone
reservoir, in Proceedings Third International Geostatistics Congress:
Geostatistics Avignon 88 (ed: M Armstrong), 2:571-530 (Kluver
Academic Publishers: Boston).
Hammitt, J and Ballantyne, G, 2007. Geologic interpretation and model
of the resolution deposit, unpublished proprietary data, Resolution
Copper Mining, LLC.

308

Spectrum Series Volume 17

Paul, A H, and Manske, S L, 2005. History of exploration at the Magma


Mine, Superior, Arizona, in Proceedings Geological Society of
Nevada Symposium 2005: Window to the World, pp 629-638 (eds:
H N Rhoden).
Srivastava, R M, 2005. Probabilistic modelling of ore lens geometry: An
alternative to deterministic wireframes, Mathematical Geology,
35(5):513-544.
Verly, G, Brisebois, K and Hart, W, 2008. Simulation of geological
uncertainty, resolution porphyry copper deposit, in Proceedings
Eighth International Geostatistics Congress: Geostatistics Santiago
08 (eds: J M Ortiz and X Emery), 1:31-40 (FCFM U: Chile).

Advances in Orebody Modelling and Strategic Mine Planning I

You might also like