You are on page 1of 5

An overview to geostatistical modeling methods of the

categorical variables for mineral resources and hydrocarbon


reservoirs

Enayatollah Ranjineh Khojasteh*


Faculty of Mining Engineering, Sahand University of Technology, Sahand New Town, Tabriz,
Iran

* Corresponding Author’s E-mail: khojasteh@sut.ac.ir, enayatkhojasteh@gmail.com

Abstract
Considering the key role of the mineral resources and hydrocarbon reservoirs and consequently, the
inevitability of having their precise models, geostatistical tools are regarded as efficient and functional
solutions for these problems. They not only provide accurate predictions, but they also quantify the
corresponding uncertainties. To obtain a precise geostatistical models, an exact estimation of their
categorical variables such as lithofacies and alteration zones are necessary. In this article, four
geostatistical methods of modeling categorical variables were reviewed and their advantages and
disadvantages were mentioned. The discussed methods include Sequential Indicator Simulation (SIS),
Truncated Gaussian simulation (TGS) and Truncated Pluri-Gaussian Simulation (PGS),
methods based on using Markov chain and transition probabilities.
Keywords: Sequential Indicator Simulation (SIS), Truncated Gaussian and Pluri-Gaussian
Simulation (TGS and PGS), Markov Chains, Transition Probabilities.

1. INTRODUCTION

Ore deposits play an important role in economic and strategic development and prosperity of nations and the
whole world [1]. This role is even more considerable for the developing countries [2]. Moreover, the raise of
the world's population, their increasing commodity demands, and the depletion of the existing mineral
resources obligated the prospect and exploration of new ore deposits. In addition, the trend of the industries
towards the consumption of the strategic metals and elements, made these new prospecting campaigns more
inevitable. Yet, a critical issue here is an accurate construction of the mineral resources models with
spending reasonable exploration expenditures. To make a realistic and accurate mineral resource model, one
should first construct and predict the important categorical variables, especially the geological domains such
as lithological and alteration zones. This point can be considered within the context of the estimation
domains related to the geostatistical stationarity zones in predicting the continuous variables such as the
concentration of the important and major elements like Copper in a porphyry deposit [1].
To predict the domains of the categorical variables, several methods including deterministic and
stochastic frameworks exist. To account for the uncertainty of the models geostatistical methods offer a wide
variety of tools and techniques which not only make accurate predictions but also they provide the models
with the uncertainty measures which make these methods very efficient and flexible [3].

Considering their efficiency and power, geostatistical models, we focus on reviewing different just
geostatistical models for predicting categorical domains.

Several geostatistical methods for modeling categorical variables have been developed. For example,
Sequential Indicator Simulation (SIS), Truncated Gaussian simulation (TGS) and Truncated Pluri-Gaussian

1
Simulation (PGS), methods based on using Markov chain and transition probabilities, and Multiple-Point
Statistics (MPS).
Each of the above-mentioned methods have their own advantages and disadvantages and can be considered
for specific problems. Moreover, the methods addressed have also applications in other cases such as remote-
sensing, image processing, agriculture, etc.
Following, more details about the first three mentioned methods have briefly been discussed.

2. SEQUENTIAL INDICATOR SIMULATION (SIS)

SIS is a common and rather conventional geostatistical simulation method for categorical variables such as
lithology and alteration. This technique is based on calculating the probably of the occurrence of each
category (i. e. Conditional Cumulative Density Function: CCDF) using Multiple Indicator Kriging (MIP) at
each simulation grid-node.
For MIK, an Indicator (Ik) variable is defined for every category (k). The Ik would be 1 when the category k
exists at that point and 0 otherwise. Then the Indicator variograms are calculated for every category, as
following:

2
1 N (h)
 I * ( h)   I ( x )  I ( x  h)
2 N (h)  1 (1)

With separation/lag distance of h and N(h) the number of pairs of points, and their locations of x [3, 4].
Subsequently, appropriate models are fitted to each indicator variogram and the inferred models are
used for kriging with the Indicator values (Ik) as the input/conditioning data as explained above.
The MIK outputs would be the probabilities of occurrence for each category at that location. In other words,
the local CCDF would be calculated.
In the SIS algorithm, first, a simulation grid is constructed and a random simulation path which
meets all the simulation grid-nodes is chosen. Then, sequentially and conditioned to the neighboring data
(including the previously simulated grid-node data), MIK is performed at every simulation grid-node and
hence its CCDF is calculated. Next, a random value (i.e. the simulated category type) is drawn from this
CCDF and assigned to the simulation point. This procedure is continued till all the grid-nodes are simulated.
These steps are repeated n times, to obtain n realizations [5].
Though simple and straightforward, SIS still contains some drawbacks. For instance, the generated
patterns through this method is usually very scattered and disconnected. This feature could make it only
appropriate when the geological environment tends to a high degree of diagenesis. Yet, for more continuous/
connected geological settings, they can be regarded unrealistic. Moreover, due to ignoring the inter-class
correlations, SIS may not produce geologically sound outcomes for the complex geological environments
with more number of categories (like several lithofacies) [6]. Additionally, the indicator variograms could be
unclear to fit appropriate models to them in the sparse datasets. There are also a number of other problems in
the bases of SIS such as the existence of order-relation violations in its underlying MIK framework.
Therefore, a number of other methods for geostatistical modeling of categorical variables (e.g.
lithofacies and alteration zones) could be considered to overcome or mitigate these problems [7].

3. TRUNCATED GAUSSIAN (TGS) AND TRUNCATED PLURI-GAUSSIAN SIMULATIONS (PGS)

One approach in geostatistical modeling of a categorical variable (like rock types) could be to generate these
categories through the simulation of a continuous variable, and truncating it in a number of thresholds. This
can be conducted (TGS) by making a simulation of a Gaussian variable and truncating it in a number of
thresholds which reflect the proportions of corresponding categories. Figure 1 illustrates a schematic example
of TGS method along a 1D profile.

2
Figure 1. (a) Truncating a continuous Gaussian variable at thresholds to create a realization of categorical
variable. (b) a Gaussian distribution with two thresholds (yt1, yt2) and three continuous variables (y1, y2, y3) [5].

Although the method is simple and functional and has the potential to take the subjective and
interpretive information into account (e.g. by considering the logical facies sequences), it is unable to model
the non-sequential transitions [8]. For instance, as it is clear in Figure 1, the categories/facies 1 and 3 cannot
be
in their immediate adjacencies and should pass through the category 1. This can be a problem when non-
sequential transitions are expected, according to the observations and geological interpretations.
To overcome this problem, two or more Gaussian continuous variables (e.g. G1, G2,...) are used
instead of only one Gaussian variable. This method is called Truncated Pluri-Gaussian Simulation (PGS).
PGS is more flexible with different geological scenarios. Nevertheless, it is more complicated and requires
several decisions by the user, for example about the transition rules and the G1, G2,… correlations. This can
make the method vulnerable to the user’s choices ad decisions [9]. Still, PGS can be regarded a powerful tool
for geostatistical modeling of facies, alteration zones, and other categorical variables.

4. METHODS BASED ON USING MARKOV CHAINS (MC) AND TRANSITION PROBABILITIES


(TP)

Most of the common geostatistical tools for modeling categorical variables make use of variograms or
covariogram for modeling spatial structure of the existing variables. However, these methods may face
shortcomings in modeling the vague structures, taking the interclass correlations into account, and
integrating subjective and interpretive information into account. A solution to overcome these problems is to
use the Transition Probabilities (TPs) instead of variograms and Markov Chains (MCs) to models the
corresponding TPs. TPs can be defined as the probabilities of transitioning from a category to another one, in
terms of separation distance. Unlike the variograms, TPs are not symmetric and can consider more facts
about the underlying systems. Figure 2 shows an example of TPs ad their corresponding MC models.
Using TPs and MCs facilitates the calculation of the interclass correlations among the various
categories/facies and increase the capacity of integrating subjective and interpretive information in the model
[10]. The generated patterns are usually more realistic and consistent with the geological facts in the
sedimentary environments. For instance, a clearer layering is evident in the generated models using these
methods.

3
Figure 2. An example of TPs ad their corresponding MC models [3].

Several methods use TPs and MC models for modeling categorical variables. For example, the
Transition Probability- Markov Chain (TP/MC) method is among these techniques. TP/MC is based on re-
formulation of cokriging equations using TPs and making predictions/simulations and then optimizing the
generated simulations by simulated annealing (quenching) according to the MC models. To fit suitable MC
models to TPs, TP/MC provide a range of tools for different conditions which enables the user to integrate
the subjective and interpretive information in the model. Discrete-lag MC, Embedded MC, transition rates,
and Maximum Entropy models are the most important of these methods [10]. Lately, Carle and Fogg [11],
also expanded the capabilities of this method by considering soft data within this framework. Nevertheless,
TP/MC indirectly uses the MC models and has a rather slow run-time. Therefore, a number of other methods
like Coupled Markov Chain (CMC) [12], Markov Chain Random fields (MCRF) [13], and Generalized
Coupled Markov Chain model (GCMC) [14] were developed which use direct MC and TPs to predict the
categorical variables and hence has a much quicker run-speed than even that of SIS (almost three times as
fast).
Lately, Mahmudi et al. [15], combined TP/MC and GCMC to obtain a fast, yet integrative method
(Combined Generalized Coupled Markov Chain (CGCMC) method), to keep the potentials of both methods
and they made modifications to obtain even more realistic modeling framework.

5. SUMMARY AND CONCLUDING REMARKS

Geostatistics is a powerful tool in modeling mineral and hydrocarbon reservoirs, but it also can be
implemented in several other applications such as remote-sensing, image processing, agriculture, etc.
Moreover, the model uncertainties are quantified in this framework. To obtain accurate Geostatistical models,
 categorical predictions (e.g. lithofacies, alteration zones, etc.) are necessary.
precise
In the current article, four geostatistical techniques for modeling categorical variables were briefly
reviewed and their key advantages and disadvantages were mentioned. Transition Probabilities (TPs) with
Markov Chain (MC) models show promising results, and in some of them like GCMC and CGCMC, quicker
run-speed, in these problems. They also represent a higher capacity of integrating subjective and interpretive
information in the model. TGS and PGS methods also entail several advantages esp. with considering the
geological conditions of the investigated system.

4
This paper provides a brief guide and overview for selecting the most appropriate technique among them and
introduction for the readers.

REFERENCES

1. Rossi, M.E. and Deutsch, C.V., (2013), “Mineral resource estimation,” Springer Science & Business
Media.
2. Akhtar, A., (2005), “Mineral resources and their economic significance in national development:
Bangladesh perspective,” Geological Society, London, Special Publications, 250(1), pp.127-134.
3. Ranjineh Khojasteh, E., (2013), Geostatistical three-dimensional modeling of the subsurface
unconsolidated materials in the Göttingen area, PhD dissertation, University of Göttingen, Germany.
4. Isaaks, Edward H., and R. Mohan. Srivastava, (1989), “An Introduction to Applied Geostatistics,” New
York: Oxford University Press. Print.
5. Pyrcz, M.J. and Deutsch, C.V., (2014), “Geostatistical reservoir modeling,” Oxford university press.
6. Li, W. and Zhang, C., (2008), “A single-chain-based multidimensional Markov chain model for
subsurface characterization,” Environmental and Ecological Statistics, 15(2), pp.157-174.
7. Carle, S.F. and Fogg, G.E., (1997), “Modeling spatial variability with one and multidimensional
continuous-lag Markov chains,” Mathematical Geology, 29(7), pp.891-918.
8. Ghojeh Beyglou M.H., Falahat, R., Ranjineh Khoasteh, E., (2022), “A geostatistical analysis approach for
facies and porosity modelling of a heterogeneous sandstone reservoir to compare four practical stochastic
methods,” Zeitschrift der Deutschen Geselschaft Fur Geowissenschaften, Schweizerbart Verlag, 173: 299–310,
Stuttgart.
9. Armstrong, M., Galli, A., Beucher, H., Loc'h, G., Renard, D., Doligez, B., Eschard, R. and Geffroy, F.,
(2011), “Plurigaussian simulations in geosciences,” Springer Science & Business Media.
10. Carle, S.F., (2000), “Use of a transition probability/Markov approach to improve geostatistical
simulation of facies architecture,” (No. UCRL-JC-141551). Lawrence Livermore National Lab.(LLNL),
Livermore, CA (United States).
11. Carle, S.F. and Fogg, G.E., (2020), “Integration of soft data into geostatistical simulation of categorical
variables,” Frontiers in Earth Science, 8, p.565707.
12. Elfeki, A.M., (2006), “Reducing concentration uncertainty using the coupled Markov chain approach,”
Journal of hydrology, 317(1-2), pp.1-16.
13. Zhao, C., Gong, W., Li, T., Juang, C.H., Tang, H. and Wang, H., (2021), “Probabilistic characterization
of subsurface stratigraphic configuration with modified random field approach,” Engineering Geology,
288, p.106138.
14. Park, E., (2010), “A multidimensional, generalized coupled Markov chain model for surface and
subsurface characterization,” Water Resources Research, 46(11).
15. Mahmoudi, M., Ranjineh Khojasteh, E. and Sharghi, Y., (2022), “Facies modelling for hydrocarbon
reservoirs using a combined Markov chain method,” Geocarto International, pp.1-22.

You might also like