SPE-181049-MS

Reservoir Uncertainty Analysis: The Trends from Probability to Algorithms

and Machine Learning

Maureen Ani, Gbenga Oluyemi, and Andrei Petrovski, Robert Gordon University; Sina Rezaei-Gomari, Teesside
University

Copyright 2016, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Intelligent Energy International Conference and Exhibition held in Aberdeen, United Kingdom, 6-8 September 2016.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents
of the paper have not been reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the Society of Petroleum Engineers, its officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written
consent of the Society of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300 words; illustrations may
not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
For over fifty years, reservoir development around the world has covered different reservoir types
and environments with vast technology, expertise and a growing variety of approaches. However, the
predominant challenge from which a myriad of other field development issues arise has been on how
to accurately characterise reservoir parameters because the obtained results are largely associated with
uncertainties due to subsurface geological complexities.
This paper focuses on the evolving advances and current practices in reservoir uncertainty modelling
and gives insight into the future trends. This work critically examines the foremost statistical reservoir
uncertainty analysis approaches, the current probabilistic and stochastic uncertainty modelling workflows
which are typically based on various numerical models, and the very recent development of embedding some
artificial intelligence algorithms (which include genetic algorithms, artificial neural networks, Bayesian
networks amongst others) in reservoir uncertainty modelling, which now points to a future of using
more sophisticated machine learning systems for achieving reservoir models and parameters with higher
confidence.
These evolving trends and approaches are discussed in more detail in this paper; with an in-depth analysis
of the associated workflows, fundamental principles, strengths, weaknesses, robustness and economics of
each approach. Also, reconciliation between the statistical, probabilistic, stochastic and artificial intelligence
methods present a deep insight into the prospects of using artificial intelligence for optimising the modelling
of reservoir uncertainties beyond the capabilities of conventional methods. Thus saving time and cost by
quantifying the uncertainties in reservoir properties as well as regenerating new best-fit reservoir attributes
using the robust uncertainty analysis networks and the pattern-recognition ability of machine learning
networks.
Hence, this paper presents a comprehensive review of the various uncertainty analysis methods, and also
analyses the confidence of artificial intelligence applications which are increasingly pushing the frontiers
to improved uncertainty modelling.
Keywords: Reservoir modelling Uncertainty modelling Artificial Intelligence
2 SPE-181049-MS

Introduction
When a technological process approaches maturity, it faces more complex demands for improvement in
order to sustain its value amidst economic and technological challenges. This is exactly the case in the
area of Reservoir Modelling. To work or make development decisions with the result of any reservoir
model, the uncertainty analysis of that reservoir model must be reliable. Thus over the years; in order to
meet the industry's demands as the scope of reservoir modelling expands to cover more complex systems
and unconventional resources, reservoir and subsurface engineers havetaken advantage of the consistent
improvement in the abilities of digital computers and computing processes to constantly push the frontiers
in the uncertainty analysis and validation of reservoir models.
There are quite a number of ways through which uncertainties arise in reservoir models. These
include dataset limitations, incorrect models, spatial variabilities in subsurface geological settings (Mallika,
Srividya and Venkatachalam, 2012), flawed geological interpretations, software misuse, amongst others.
Uncertainties in the output of reservoir models could mean volumetric error in the range of up to ±10%
to ±50% (Ref:Bentley, 2015) as well as other errors, and the consequences of overlooking such model
uncertainties could be very costly and this includes erroneous well placement planning, incorrect sizing of
surface facilities, flawed development plans, and wrong investment decisions. Hence; the importance of a
well-informed reservoir uncertainty analysis cannot be over emphasized.
Uncertainty with respect to reservoir characterisation can be defined as the reservoir modeller's best
estimate of how far a modelled reservoir quantity might deviate from the true value of that quantity. (Ref:
SLB Glossary) defines it as the degree to which a data deviates from predicted values.
Reservoir uncertainty analysis and quantification dates back to the early twentieth century (Lewis et
al., 1918), and several reservoir models have been developed ever since (Gorell and Bassett, 2001). While
uncertainty is often quantified and expressed in terms of standard deviation or variance, different methods
ranging from simple deterministic analysis to high level and intelligent computing algorithms as discussed
in this paper have been employed over the years in carrying out reservoir uncertainty analysis. For the
purpose of simplicity, the reference reservoir quantity that the uncertainty analysis in this paper will address
is the Stock Tank Oil Initially In-Place (STOIIP).

Mapping and Estimation Methods

The earliest attempts at analysing and managing reservoir uncertainties in reserves estimation were mainly
characterised by carefully contouring the geological structure of the reservoir and contouring the basal
water on a map (Washburne, 1915). Then, with the reservoir porosity experimentally obtained, the engineers
would make an estimation of the probable maximum yield based on manual experience which is used for
the prevention of excessive initial investment; although the estimate might of course far exceed the actual
results (Albertson, 1935 and Washburne, 1915). These were basically empirical methods based on analogy.
However, it had the same philosophy as the deterministic approach explained in the following section. The
only major difference being that the widely common deterministic methods that followed, which is still in
use today became more advanced with modern computing.
With the advent of reservoir behaviour simulation on digital computers by the 1950s (Sheldon et al.,
1960), the quantitative modelling of reservoir uncertainties gradually began to emerge as the associated
numerical programs were then supposed to be analysed by computers (Eckles, 1959).

Deterministic Approach
In this approach, the outcomes in the model are specifically determined through known relationships.
These outcomes are expressed in three levels – "Best Case, Most Likely, and Worst Case". Deterministic
uncertainty modelling approach works by entering a single value of each input parameter into an appropriate
SPE-181049-MS 3

mathematical model which yields solutions in the form of expected outcomes. This method has been in use
for as long as reservoir modelling itself, it is still in use and is a very common method.

Associated Workflows
When modelling reservoir uncertainty with this approach, a single earth model is built with a deterministic
construction of the reservoir attributes that best describes the geological setting within the best of the
reservoir modeller's knowledge of the reservoir. The reason of building just a singular model is often due
to computational (CPU) and time constraints, and this model is aimed at being as realistic as possible. In
some other cases, a few more models of the reservoir are built to describe some other possible scenarios
of the reservoir.

Strengths and Weaknesses

Deterministic models help in developing a good understanding of the reservoir system, however; it makes
no effort to assess the possibility of each outcome because it gives equal weight to each outcome. Thus, it
has a very low predictive capability which makes it impractical in cases of projecting cumulative production
over time. Moreover, deterministic models do not consider the effect of different inputs on the outcome.
Also, a single deterministic model considers only a few discrete outcomes and yields a singular prediction
value irrespective of the multiple uncertainties that will be inherent in the model, thus ignoring other sources
of uncertainty.

Probabilistic / Stochastic Approach

In this method, a likelihood theory is basically employed. Unlike the deterministic method, it involves
the random (stochastic) specification of specific parameters of the reservoir model through predefined
probabilistic algorithms. This method models the likelihood of many possible scenarios like presence
of sealing faults (no flow barriers) in the reservoir, non-sealing faults etc, in the estimation of a given
parameters for example, the STÒIIP.
An example of uncertainty quantification using this method is the Monte Carlo method. The results
from this method are usually reported as percentiles of the cumulative probability distribution. To carry
out uncertainty analysis on the STOIIP, a single value of each parameter is varied while the others are kept
constant.

Geostatistical Methods of Uncertainty Modelling

These methods statistically model the spatial distribution and behaviour of geological (Holden et.al, 1998;
Ringrose and Bentley, 2015, Caers, 2011) parameters across the reservoir (modelled on the reservoir grids).
These methods are known for their ability to build complete models using good geological knowledge of
the reservoir depositional system and sparse data.
As a more realistic knowledge of the geologic structure of the reservoir is limited at the decision-making
phase of a reservoir's development planning, geostatistical modelling offers the opportunity to model the
distribution of geologic properties across the reservoir.

Associated Workflows
It uses variograms (a function that describes the degree of spatial dependence between sample values as
separation between them increases), semivariograms (a function that indicates the spatial correlation in
observations measured at sample locations e.g. wells), and kriging (a Gaussian statistical technique used
with variograms, that interpolate between sample values as separation between them increases, to determine
the value of a point in a heterogeneous grid from known values in nearby grids) as tools.
4 SPE-181049-MS

The difference between variogram and semivariogram is that while semivariogram uses each pair of data
elements only once, the variogram uses all possible data pairs. More often, the semivariograms are used
instead of variograms (Ref: SLB Glossary).

Strengths and Weaknesses

Seeing how much is being achieved with geostatistical methods and how much is yet to be accomplished,
this apparently has been pushing the trend of uncertainty modelling towards embedding more intelligent
algorithms into reservoir uncertainty models over the last decade, and this has formally led into a new
paradigm in uncertainty modelling, discussed in the next section.

Artificial Intelligence (AI) Approach

The concept of AI can be described as the use of softwares that are ultimately designed to exhibit super
intelligence by recognizing patterns from a given set of data and information, and thus be able to draw
inferences that could be used in solving real world problems like reservoir uncertainty analysis and
management.
AI algorithms are classified as intelligent because they are supposed to be super-intelligent, and better
than human beings. They are supposed to recognise patterns in models and dataset(s), learn those patterns,
and proffer solutions to problems based on the patterns they have recognised.
Application of AI in solving Petroleum Engineering problems has increasingly become more popular
since it was introduced in the last two decades, because it has recorded substantial advances.
The application of AI is made possible by the availability of large volumes of data (big data), and the
accessibility of a huge amount of data points from seismics and multiple logs makes the application of AI
in reservoir uncertainty modelling more promising. Some of the AI methods that have been widely applied
to reservoir uncertainty modelling include artificial neural networks (ANN), support vector networks/
machines (SVM), Bayesian networks (BN) algorithm, fuzzy logic (FL) algorithms, amongst others.

Strengths and Weaknesses

With AI, one will be able better understand the nature of the reservoir geology as the software will have a
broader spectrum of data type at its disposal (including brown field data) to be used for analysis. Simulation
time is also reduced as parttern recognition allows similar deductions made on previous data to be made
on the currently processed data.

Future trend
From the discussions in the foregoing sections and following the evolution of approaches in the subject
of reservoir uncertainty modelling, it can be deduced that with the advancement of computing power, the
future points in the direction of more sophisticated machine data mining and learning systems in reservoir
uncertainty modeling.

Conclusion
There is no uniform methodology for reservoir uncertainty analysis in the oil and gas industry. Various
uncertainties exist and are included in any reservoir model. These include Hydrocarbon-in-Place (HCIP)
or reserves uncertainty, Gross Rock Volume (GRV) uncertainty, Hydrocarbon Contacts (OWC, GOC)
uncertainty, and all other reservoir attributes.
Reserve booking and Field Development Plans (FDP) are finalised with ranges of suitable resource
uncertainty categories (Sanyal and Sarmiento, 2005). For example, reserves are reported in three levels P90,
P50 and P10; representing the 90th percentile, the median, and the 10th percentile probability of accuracy in
corresponding to the estimated reserves volumes.
SPE-181049-MS 5

References
Abdollahzadeh, A., Reynolds, A., Christie, M., Corne, D., Davies, B. and Williams, G., 2012. Bayesian Optimization
Algorithm Applied to Uncertainty Quantification. SPE Journal, 17 (03), pp. 865–873
Abdus, S., Ghulam, M. and Buchwalter, J., 2007. Practical Enhanced Reservoir Engineering: Assisted with Simulation
Software. Tulsa, Oklahoma: Pennwell.
Adam, S. J., 2005. Quantifying Petrophysical Uncertainties. SPE Asia Pacific Oil and Gas Conference and Exhibition,
5-7 April. Jakarta, Indonesia: Society of Petroleum Engineers.
Ahmadi, M., Christie, M., and Gerritsen, M., 2013. Structural Uncertainty Quantification with Immersed Interface
Methods. SPE Reservoir Simulation Symposium. 18-20 February. Woodlands, Texas: Society of Petroleum Engineers.
Caers, J. (2011), Modeling Uncertainty in the Earth Sciences, West Sussex: Wiley-Blackwell.
Chen, H. W., Gavalas, G. R., Seinfeld, T. H. and Wasserman, M. L., 1973. A New Algorithm for Automatic History
Matching. 48th Annual Fall Meeting. Sept. 30-Oct. 3. Las Vegas: American Institute of Mining, Metallurgical, and
Petroleum Engineers, Inc.
Dani, K. C., Baskaran, V. K., Gupta, D. K., and Urkude, A. M., 2013. Hydrocarbon Risk and Uncertainty: Indian Scenario.
International Petroleum Technology Conference. 26–28 March. Beijing, China: IPTC.
Hantschel, T. and Kauerauf, A. I., 2009. Fundamentals of Basin and Petroleum Systems Modeling. Dordrecht: Springer.
Hayes, K., 2011. Uncertainty and uncertainty analysis methods. Hobart, Australia: CSIRO Mathematics, Informatics and
Statistics.
Jong-Se, L., 2005. Reservoir properties determination using fuzzy logic and neural networks from well data in offshore
Korea. Journal of Petroleum Science and Engineering, 49, pp. 182
Kamali, M. R., Omidvar, A. and Kazemzadeh, E., 2013. 3D Geostatistical Modeling and Uncertainty Analysis in a
Carbonate Reservoir, SW Iran. Journal of Geological Research, Article ID 687947
Kane, J. A., Rodi, W., Bube, K. P., Nemeth, T., Mikhailov, O., and Medwedeff, D., 2004. Structural Uncertainty
and Bayesian Inversion. SEG Int'l Exposition and 74th Annual Meeting. 10-15 Oct. Denver, Colorado: Society of
Exploration Geophysicists.
Oghena, A., 2007. Quantification of Uncertainties Associated with Reservoir Performance Simulation. [online] PhD
thesis, Texas Tech University. Available from: https://repositories.tdl.org/ttu-ir/handle/2346/17920 [Accessed 19 Dec
2015]
Ouenes, A. and Saad, N., 1993. A New, Fast Parallel Simulated Annealing Algorithm for Reservoir Characterization. 68th
Annual Technical Conference and Exhibition of the Society of Petroleum Engineers. 3-6 October. Houston: SPE.
Petroleum Experts, 2005. MBALReservoir Engineering Toolkit. 8th ed. Edinburgh: Petroleum Experts Ltd.
Ringrose, P. and Bentley, M., 2015. Reservoir Model Design. Dordrecht: Springer.
Schlumberger Oilfield Glossary, 2016.
Seldon, B., 2015. Buzzard Field Stage 2 Modelling Workflow: Closing the Loop on Static Uncertainties Using Workflows
and Olyx Trend Analysis. Recognising the Limits of Reservoir Modelling - and how to overcome them. 4-5 March.
London: The Geological Society of London.
Seyyed, G. A., 2009. Reservoir Modelling. [ID: 92] University of Calgary: Energy and Environmental Systems Group,
Institute for Sustainable Energy, Environment and Economy (ISEEE).
Steele, R. and Stephen, K., 2015. Geological Weaknesses in Geomodelling. Recognising the Limits of Reservoir Modelling
- and how to overcome them. 4-5 March. London: The Geological Society of London.
Tarek, A. and Mckinney, P., 2004. Advanced Reservoir Engineering. Saint Louis, MO, USA: Elsevier Science &
Technology.
Wilkinson, D. and Bailey, W. And Couët, B. October 2012. Method for Consistent Valuation of Assets with Multiple
Sources of Uncertainty. SPE Economics & Management, 4 (04), pp. 204
Winter, J., 2014. Principles of Igneous and Metamorphic Petrology. Essex: Pearson Education Limited.
Yang, C., Nghiem, L. X., Card, C., and Bremeier, M., 2007. Reservoir Model Uncertainty Quantification Through
Computer-Assisted History Matching. SPE Annual Technical Conference and Exhibition. 11–14 November. Anaheim,
California: Society of Petroleum Engineers.