You are on page 1of 12

The Application of Multiple-Point Geostatistics in

the Modeling of Dike; a case study of Sungun


Porphyry Copper, Iran
Hassan REZAEE1, Omid ASGHARI2, Mohammad KONESHLOO3

University of Tehran, Iran,


h.rezaee@ut.ac.ir

University of Tehran, Iran,


o.asghari@ut.ac.ir

Shahrood University of Technology, Iran,


koneshloo@shahroodut.ac.ir
Peer-reviewed IAMG 2011 publication
doi:10.5242/iamg.2011.0302

Abstract
The aim of this study is to assess the application of the new-born technique of Multiple-Point
geostatistical method to simulate complicated structures in a mining context. The postmineralization intrusions in porphyry systems may cause such structures hence the dike system
of an Iranian Porphyry Copper deposit has been tried for this study. The TI was extracted from
the upper benches of the mines block model and using the local dike proportion and other controlling parameters in SNESIM an E-type grid was produced that reproduced the dike-like structures on the whole mining domain.

Hassan REZAEE, Omid ASGHARI, Mohammad KONESHLOO, Julin M. ORTIZ

Introduction

Geostatistical techniques have opened a new era in the field of earth phenomenon characterization. It had a profound role in oil and gas reservoir and mineral deposit simulation and modeling.
Variogram has been known as a strong tool to identify spatial variations of phenomenon under
study. It takes into account two points at a while only and in the case of non-linear continuity it
suffers from a main inefficiency. Due to this reason it cannot distinguish vastly different patterns
of heterogeneity (Journel, 2005). Consequently the variogram is also an incomplete measure of
uncertainty in complex and non-linear structure; in a reservoir modelling context, curvilinear
geometries, such as sinuous channels in a fluvial reservoir or incised valleys over topography,
cannot be modeled using only traditional two point statistics such as a variogram (Strebelle,
2002).
In every mineral modeling research the post mineralization processes put several problems on
experts way of modeling. As the physical and chemical properties of ore, extracted from different zones differ from one another, so their behavior will be different in processing mill. Having
the 3D model of present facies in a mineral deposit has a beneficial role in production planning.
Since, dikes are of one of the most prevalent phenomena in mining projects and their structure is
complicated enough to not be modeled by traditional variogram-based methods in this study it
has been tried to acquire a taste for the application of MP statistics in dike modeling.

Multiple-Point Geostatistics

To go beyond the variogram, a more powerful tool is needed by which not only the variograms
capabilities are considered but also it fails are met. Training Image has been introduced as a replacement for variogram. Journel (2005) defined the training image as a conceptional model of
the random process. It is also a quantitative model. The training image can be produced using
different methods but it completely depends on the data source we have available in modelling
phases. Tuanfeng, (2008) refers to the various ways through which TI can be come along e.g.,
Object-based algorithms, Process-based models (Boisvert et al, 2008), Sequence stratigraphy etc.
Multiple-Point (MP) geostatistics has emerged (Guardiano and Srivastava, 1992, Journel, 2001,
Strebelle, 2002) as a new powerful field for obtaining realistic geostatistical models that one of
its great capabilities is to integrate consistently a large variety of different sources of information, each of which has been taken on different scales. The process of an MPS simulation can
simply be defined as follows:
At the first step and to go beyond the traditional variogram, the variogram should be replaced
with a more powerful tool; which in multiple point geostatistics has been known as Training Image (TI) (Journel, 2001). In MPS we use TI as a tool to identify the dominant variation patterns.
Albeit there has not been a long time from the advent of MPS; there are several ways of TI production. The production of TI is the most important stage of the simulation process that all the
outputs are dependent on it. By simulating TI, MPS simulation process can be begun which is
consisted of three main stages (Caers, 2001):
Pattern extraction
Pattern identification
Pattern reproduction
At first, one should extract the existing pattern in phenomenon under study from simulation
space, containing the nodes on which the target value is unknown,. Figure 1 gives an example of
such MP pattern simulation in very simple words. Finding such pattern in the simulation grid, the
2

The Application of Multiple-Point Geostatistics in Dike Modeling

available TI will be searched for the same pattern, this will be followed by computing the probability of each facies and then it will be the turn for the next node in the random process. De
pending on the variable whether categorical (such as facies), or continuous variable (for
example, grade, or any other Geophysical property of the deposit) relevant algorithm
should be used to simulate the phenomena. There is one renowned algorithm known as
SNESIM which is provided in SGeMS software, for categorical variables simulation.

Figure 1: Illustration of sequential multiple-point simulation (After Tuanfeng, 2008)

Since Strebelle (2000) introduced snesim in his PhD thesis, within a short period of time, many
other MP simulation algorithms have been developed from, some already in a state of beta testing such as Arpat and Caers, 2001 and Zhang et al, 2004. After that several researches have been
done to make a use of MP in the geosciences. (Dovera, 2006, Feyen, 2005, Levy, 2008).
One of the outstanding factors of MP is the ability to integrate data from different sources and
taken on different scales. Hard data like drill hole data and soft one such geophysical data (in this
study it is local proportion data) can easily be considered in simulation process. In this study we
plan to use the local proportion data as an auxiliary variable to constrain the realization the procedure of doing so will be described:

2.1 Soft data integration


Soft (secondary) data may be available to constrain the simulated realizations. They are
mainly are acquired by remote sensing techniques such as seismic data but different sort of
soft data can be used. They may be acquired on the realization grid on a very corse resolu
tion but their effect can be easily tasted (Remy, 2009). The rst step is to derive soft data
conditional probability
from the soft data . Then at each unknown node a sequen
will be combined with the sofat data probabil
tial simulation will be done then the
ity
to get the posterior probability
. Finally, a facies indicator value is
drawn using this nal probability
, which is conditioned to both hard and soft
data. Journel (2002) proposed a permanence of ratios algorithm to combaine such prob
abilities.
The basic assumption of this algorithm is that the relative contribution of data event is
the same before and after knowing :

Eq. 1
where, the distances to the event occuring are defined as
3

Hassan REZAEE, Omid ASGHARI, Mohammad KONESHLOO, Julin M. ORTIZ


Also, Journel proposed the introduce a power into Eq. 1 to have the ability to control the
relative role of secondary data:

Setting


increases the impact of soft data, coversly, Setting

does the vice versa.

Study area

The Sungun porphyry copper deposit is located in northwestern Iran (Azarbaijan province) and is
associated with diorite-granodiorite to quartz monzonite of Miocene age which intruded Eocene
volcano-sedimentary and Cretaceous carbonate rocks. Field observations and petrographic studies demonstrate that emplacement of the Sungun stock took place in several intrusive pulses,
each with associated hydrothermal activity (Hezarkhani, 1998). (See Figure 2)

Figure 2. Below: Geological map of Iran showing Sahand-Bazman belt (modified from: Stocklin,
1976; Shahabpour, 2007); Above: Geological map of Sungun deposit area showing various types of
intrusive rocks of dominantly Miosene age and the outline of Cu-Mo porphyry type mineralization.
(Modified from Mehrparton, 1993 and Hezarkhani, 2006).
4

The Application of Multiple-Point Geostatistics in Dike Modeling

3. 1 Dike system
The Hypogene zone often contains considerable thickness of un-mineralized porphyry. The dikes
(DK1a orange one in Figure 2, DK1b the green dike in Figure 3) strike NNW-SSE, dip steeply to
the west and have thickness from a few centimeters to several tens of meters (Figure 3). They
appear to have acted as a barrier to hydrothermal and Supergenes processes and consequently
sometimes mark the boundary between Leached and Supergene material. They also frequently
act as a focus for high-grade copper-molybdenum mineralization in the adjacent monzonite porphyry host rock (Hezarkhani, 1998). DK1 is the most prevalent type of dike that exists in the
study area which mainly is composed of Quartz Diorite to Quartz Monzonite. Based on isotope
studies or other evidences like chronological, alteration and mineralogical one, three main types
of dike can be derived from DK1 known as DK1-a, Dk1-b and Dk1-c, each of which has its own
features. Some of distinguishing features are as follow:
a. DK1-a
This the most prevalent type of dike DK1 and also all dikes in the area. Samples belonged to
DK1-a have been exposed to weak and medium to high degree of phyllic alteration. Based on
macroscopic factors three subgroups of DK1-a can be derived from: a. Mineralized dikes that
have some formed pyrites, b. Non-mineralized dikes and c. Mineralized dikes containing Subhedral and Euhedral Orthoclase.
b. DK1-b&c
On the other hand, samples from DK1-b have been intruded in before DK1-a and has very weak
phyllic and propylitic alterations. They dont bear any mineralization and the amount of pyrite is
very low. Besides, DK1-c samples are in very weak propylitic alteration and sometimes they
entombed in the DK1-b dikes.
Therefore in an industrial point of view these dikes can cause severe problems as their physical
and chemical features are strongly different rather than host rock. There should be an exhaustive
model including these variations.

Figure 3: Dike series, and the selected area for this study
5

Hassan REZAEE, Omid ASGHARI, Mohammad KONESHLOO, Julin M. ORTIZ

Result and Discussion

The most important factor of every MP research is to produce the TI that represents for the
whole area under study. One of the methods through which TI can be produced is to use the geological data that have been gathered during the past exploartion or current mining activities
phases. In this study the sam way was adopted to get TI. The TI was extracted from the block
model (Figure 4) from operating mine as the first 200 meters of the mentioned block model was
considered to do so. What is shown in Figure 4 is the block model from which the later TI was
extracted. This is the area in which the model has the most validity that can garantee the results
of being used in mine.


Figure 4: Block model used to extract TI (colors indicates to elevation to well illustrate the block
model used in this study)

Primarily, as there huge number of nodes can be found in this model that are irrelevant to what
we are looking for therefore we cookie-cut it to a smaller volume but more useful in the computational efficiency point of view. For example, the area that contained one very thick dike was
removed from the TI as it does not need any simulation and based on simple geological rules it
can be extended to the deeper elevations. The existence of such thick dike in the domain under
study can cause several problems as it disturbs the stationarity conditions as because the average
thickness of dike depends on location. (See Figure 5)

Figure 5: The smaller TI that is considered for simulation, some surplus surrounding and the southern thick dike are removed from the final TI
6

The Application of Multiple-Point Geostatistics in Dike Modeling

Since, the dense geological surveying has been done on this first 200 meters area of mine, more
confidence can be established upon this layer. As it was mentioned before there are several kinds
of dike, each of which has their own features, but in this study which can be cosidered as the first
study on such phenomenon, only a state of being dike or not case is considered. Therefore, regional variable has been considered as an indicator state between dike and non-dike cases:

Following the way, the final TI that was considered for simulation purposes is shown in
Figure 6right.


Figure 6: Left: TI including codes for each of the dikes; Right: The final TI that was inputted to
SNESIM algorithm

Although, the surplus parts of the TI was removed to make it as representative as possible but
there may arise another problem too that is: the dike occurence probability differs from part to
part and is the factor through which the next simulation process can be affected with. As it is
shown in Figure 7 there two distinct parts with different dike proportions can be found so one
should account for a secondary variable that represents the local dike proportion. This variation
can be extended to depth so it demands another variable to account it.


Figure 7: Different dike proportions in the study area

Hassan REZAEE, Omid ASGHARI, Mohammad KONESHLOO, Julin M. ORTIZ

Several methods can come up with to calculate such local proportions like Nearest Neighbor
(NN) and proceed the way by using a Moving Average on the NN results. But for this study as
huge number of nodes are going to be simulated, this method suffers from the lack of computational efficiency. Since full Indicator Kriging (IK) bears this onus too it was considered to do so,
as after a variography stage IK method was applied on the whole area (covering the whole operating mine) and as the IK results are to some how un-stable the process was followed by applying a Moving Average on the IK estimates (Probabilities) to make it smoother and coarser to be
used. Figure 8 refers to the output of applying Moving Average on IK results.


Figure 8: Moving Average on the IK estimates

As can be seen, the higher value (Red color) for dike existence coincides with the local proportion that we expected. The same grid was produced for the Sungun Porphyry (Non-Dike domain). On the next stage, the SNESIM of SGeMS open source software was applied on the grid
using the following input parameters:

The Application of Multiple-Point Geostatistics in Dike Modeling

Figure 9: The input parameters for SNESIM algorithm in SGeMS

The conditioning data used for this study were drill hole data including the code for each sample
point that represented each of the dike or non-dike states. There were summing to 41520 data
points that cover the whole area under study. Servosystem correction factor is important if TI
proportions are different from the conditioning data proportions. It should be used moderately to
impose the desired proportions, but allowing some fluctuations. The Servosystem correction of
0.7 is due to the big difference between TI and Drill hole proportions as the 0.17 and 0.30 are the
values for dike proportion for TI and conditioning data respectively.
Eventually, the simulation algorithm was applied on the grid that was produced before (Containing 930,000 nodes). One of the realizations is illustrated in Figure 10 and as can be seen dikelike structures have been produced in the area that shows the ability of this method to model this
kind of phenomena. The conspicious effect of local proportion is appeared in Figure 9 where the
parts with lower probability of dike occurence is simulated in this figure.

Hassan REZAEE, Omid ASGHARI, Mohammad KONESHLOO, Julin M. ORTIZ


Figure 10: One of the SNESIM realizations, red nodes refers to the Dike occurrence

Moreover, the effect of conditioning data can be better understand when one comes with an Etype grid. To get a representative simulation, 100 realizations were produced and the next step of
project was to calculate the E-type 3D grid. Finally, the E-type map was produced using the
mean value of 100 values for each node and is illustrated in Figure 10. Again the capability of
SNESIM is appeared in the form of dike-like structures in the simulated area. As can be seen in
Figure 10 there are some parts that the dike structure has not been simulated that are mainly on
the corners and the bottom. The lack of conditioning data may cause such problems.


Figure 11: E-type map of the 100 realizations of the study area
10

The Application of Multiple-Point Geostatistics in Dike Modeling

Conclusion and future works

In geosciences studies there are many circumstances in which the modeling of phenomena raises
several conflicting issues. Among different geostatistical techniques Multiple-Point method has
shown its great capability to solve such problems. In this study one of them was investigated and
the final E-type grid point was an evidence of this ability. The art of Multiple-Point geostatistics
appears when one comes with a comparison between the other tools that are being used in industrial projects such as the geological priory geological data. Since MPS has the ability to simulate
the dike model for the deeper areas of mine under exploitation. It extremely decreases the error
of dike modeling in deeper parts of mine as it was done in this study.
Using the geological data containing useful information on dike structures the TI was produced
that is much simpler than carrying out variography step to come with an estimate for each block
in the study area. The simplest state of Multiple-Point technique was used here and it is suggested to apply SNESIM with more input data. For example, coding the TI by different kinds of
dike in the same area can bring useful information up on dike models as the chronology of these
dikes differs. A segmentation procedure can be adopted too as it was seen that dike proportion
varies with coordinate. Furthermore, the affinity and rotation factors can play an important role
in making the realizations as fit as possible to reality.

11

Hassan REZAEE, Omid ASGHARI, Mohammad KONESHLOO, Julin M. ORTIZ

References
ARPAT, B., G., J., CAERS, A., HAAS.; (2001): Characterization of West-Africa submarine channel reservoirs: A neural network based approach to integration of seismic data. In
2001 SPE Annual
BOISVERT, B., O., LEUANGTHONG, J., M., ORTIZ, C., V., DEUTSCH; (2008): A methodology to construct training images for vein-type deposits, J, Center for Computational
Geostatistics, Department of Civil and Environmental Engineering, University of Alberta,
Computers & Geosciences 34 491 502
CAERS, J.; (2001): Geostatistical reservoir modelling using statistical pattern recognition,
. Journal of Petroleum Science and Engineering 29 2001 177188
GUARDIANO, F., SRIVASTAVA, M.; (1992): Multivariate geostatistics: . beyond bivariate moments. In: Soares, A. Ed., Geostatistics Troia. Kluwer, Dordrecht, Holland, pp.
133144.
FEYEN, L.; CAERS, J.; (2005): Geostatistics for environmental applications. Proceedings
of the Fifth European Conference on Geostatistics for Environmental Applications 2005
pp. 197-207
HEZARKHANI, A.; (2006): Petrology of Intrusive rocks within the Sungun Porphyry
Copper Deposit, Azarbaijan, Iran. Journal of Asian Earth Sciences. 73, 326-340.
HEZARKHANI, A., WILLIAMS-JONES, A., E.; (1998): Controls of alteration and mineralization in the Sungun porphyry copper deposit, Iran: Evidence from fluid inclusions and
stable isotopes. Economic Geology, 93, 651-670.
JOURNEL, A., G., (2002): Combining knowledge from diverse sources: An alternative to
traditional data independence hypotheses. Mathematical Geology, 34(5):573596.
JOURNEL, A. G.; (2005): Beyond covariance: the advent of multiple-point geostatistics,
tics O. Leuangthong and C. V. Deutsch (eds.), Geostatistics, Banff, 2004, 225-233.
LEVY, M., HARRIS, P., H., STREBELLE, S.; (2008): Multiple-Point Statistics (MPS) /
Facies Distribution Modeling (FDM) Carbonate Reservoir Modeling Using Search and
Discovery Article #40293
MEHRPARTOU, M.; (1993): Contributions to the geology, geochemistry, Ore genesis
and fluid inclusion investigations on Sungun Cu-Mo porphyry de- posit, northwest of Iran.
Unpublished PhD Thesis. University of Hamburg, Germany, 245 p.
REMY, N., BOUCHER, A., AND WU, J.; (2009) : Applied Geostatistics with SGeMS,
user guide, Cambridge University Press.
STREBELLE, S.; (2000): Sequential simulation drawing structures from training images,
unpublished PhD Thesis, Stanford University.
STREBELLE, S.; (2002): Conditional Simulation of Complex Geological Structures Using Multiple-Point Statistics, Mathematical Geology, Vol. 34, No. 1.
TUANFENG, ZH., SWITZER, P. JOURNEL, A. G.; (2004): Sequential conditional simulation by identification of training patterns, in ibid.
TUANFENG, ZH.; (2008): Incorporating Geological Conceptual Models and Interpretations into Reservoir Modeling Using Multiple-Point Geostatistics, Earth Science Frontiers,
2008, 15(1): 2635.
12

You might also like