EXPLORATION RISK MODELING DRAFT

Exploration Risk Modeling

New software provides solution for Exploration Geologists and Engineers fighting overgrown decision trees

Torben Riis and Jerry S. Blinten Typically, the last consideration in an exploration
Caesar Petroleum Systems evaluation is the commercial chance of success. This
is usually expressed as an economic limit on reserve
Introduction size. A single reserve value is often used to represent
Traditionally, exploration project evaluations have a hurdle volume.
focused on estimating reservoir volumes and Figure 1 – Example Reserve Distribution
reserves. Typically, the objective is to answer the
following two key questions: 1.0
0.9
1. Does a subsurface trap exist that is capable of
accumulating and storing hydrocarbons? 0.8
Technical Chance of Success
0.7
2. Is there a sufficient amount of source rock Probability
0.6
available to fill the trap and is the source-trap Commercial Chance of Success
0.5
timing and migration pathways in-place to fill the
Economic Cut-off
0.4
trap?
0.3
The answers are expressed in a simple volumetric 0.2
equation containing area, thickness, porosity,
0.1
saturation, shrinkage, and a recovery factor as input
0.0
variables. In a probabilistic analysis, the uncertainties 0 60 120 180 240 300 360 420 480 540 600
in these inputs are estimated and the volumetric Reserves
equation is solved in a Monte Carlo simulation.
The consideration of the chance of failure (or
success) is equally important to the volumetric Decision tree analysis (Figure 2) is also a useful tool
estimation (Figure 1). The chance of failure reflects in an exploration evaluation. A common use of
the situation where proper conditions for source decision trees is to derive the risk-weighted net
generation, migration, timing, and seal integrity are present value of a prospect. A simple tree with a
not present. success and failure branch is often built. The success
branch represents a successful exploration effort and
A chance of failure (or success) may also be a cash flow forecast. Often, the decision tree contains
associated with reservoir volumetrics. It would only one “expectation” success branch. However,
represent the possibility that some input variables trees with “high”, “mid” and “low” reserve success
may fail to reach minimum threshold values. The total branches are also frequent. The failure branch
chance of failure is often expressed as the includes the net present value of an unsuccessful
corresponding technical chance of success (one exploration program. The tree can be calculated
minus the total chance of failure) and has the effect directly for a risk weighted net present value of the
of reducing the probability in reserves versus prospect and the combined cash flow can be used to
probability plots. determine a risk weighted rate of return.

By Torben Riis and Jerry S. Blinten, Caesar Petroleum Systems Page 1 of 5

EXPLORATION RISK MODELING DRAFT

Figure 2 – Example Decision Tree trees which encompass the possible development
scenarios with all the expected outcomes. However,
Result Success 50.0% 100.0 $ MM due to the work involved, this is rarely done.
25.0 $ MM
Explore Area? Yes For instance, a simple 3 prospect evaluation, where
25.0 $ MM Failure 50.0% -50.0 $ MM
only volumetric success or failure in considered,
results in 8 branches. If performance is considered, a
No 0.0 $ MM 27 branch tree is generated. Therefore, the growing
complexity of the decision tree, as more alternatives
Projects with multiple prospects or projects where the are included, inherently limits the use of decision
development plan depends on reservoir size or trees.
performance naturally lend themselves to decision The following chance of failure example serves to
tree analysis. In general, decision trees are an demonstrate that decision trees can be reflected in a
excellent way to visualize and communicate single Development Model. The approach is to
outcomes and alternatives in an exploration program. compare an uncertainty analysis of a simple
exploration example conducted by Monte Carlo
Traditional Exploration Analysis simulation with a traditional decision tree evaluation.
One problem with traditional probabilistic exploration
Finally, the example will be extended to include
analysis is that basic uncertainty is measured in
reserve risk and field sequence uncertainty to show
terms of volume rather than money.
that even very complex decision trees can be
There are numerous examples where development of modeled fast and accurately in a Monte Carlo
significant hydrocarbon accumulations has resulted in simulation.
commercial failure. On the other hand, many
prospects have been released only to be developed Chance of Failure Example
by competitors. The commercial decision support software package
The problem stems from the traditional definition of Petroleum Ventures and Risk (PetroVRTM ) has been
commercial chance of success. Imbedded in the used for the Development Modeling, Economic
volumetric equation for reserves is recovery factor. Modeling, Monte Carlo simulation and decision tree
Recovery factor is often very difficult to predict and is analysis.
ultimately dependent on the development plan. The As an example, a simple exploration and
commercial chance of success is dependent on development plan was built in PetroVR:
economic limit and the economic limit is in turn
dependent on the development plan. So, we have a The project includes exploration and development of
circular problem with the commercial chance of three prospects in the same license area. If
success dependent on reserves which is dependent successful, production from the fields will be routed to
on both recovery factor and the development plan. a shared Central Process Facility. An export pipeline
will be built for oil, and gas will be flared. Two
There are often large variances in economic limit exploration wells are drilled on each prospect, and
dependent on reserves and reservoir performance. drilling of these successful prospects will continue
For instance, it may be uneconomic to develop a immediately after exploration with two development
large accumulation with poor rates, but economic to drilling rigs available. Construction of the Central
develop limited reserves with high production rates. Process Facility would start subsequent to the
These problems can be solved with traditional success of exploration drilling. Figure 3 shows the
techniques via decision trees. It is possible to lay out Project Schedule and Process Flow Diagram.

By Torben Riis and Jerry S. Blinten, Caesar Petroleum Systems Page 2 of 5

EXPLORATION RISK MODELING DRAFT

Figure 3 - Project Schedule and Process Flow Diagram

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Drill 2 Alpha Expl. Wells
Drill 2 Bravo Expl. Wells
Drill 2 Charlie Expl. Wells
Build & Drill Alpha Facility
Build & Drill Alpha Facility
Build & Drill Alpha Facility
Alpha Facility
Build & Drill Bravo Facility
Build & Drill Bravo Facility
Bravo Facility
Build & Drill Bravo Facility
Build & Drill Charlie Facility
Build & Drill Charlie Facility
Build & Drill Charlie Facility Central Process Facility

Build Central Process Facility

4
Rig Schedule 2 Charlie Facility

In order to keep the decision tree at a manageable Figure 5 - Central Process Facility/Pipeline
size, it is initially assumed that the chance of failure Total CAPEX OPEX Capacity Construct
of the three prospects is the only uncertainty related Reserves (10 $)
6 6
(10 $/year) (b/d) (days)
to the project. This reduces the number of branches 6
(10 Stb)
at each chance node to two for each prospect (failure
0 0 0 0 0
or success) and the total number of branches in the
decision tree to eight. 500 230 15 150000 800

Two facility types with different capacity levels were >500 280 18 200000 900
considered for the Production and Central Process
Facilities. The type used depends on the individual
field and total reserves. In case a prospect fails, no Figure 6 - Reservoir Data
Prospect Reserves UR/Well Well Rate Chance of
production facilities will be built at that location and
6 6
(10 Stb) (10 Stb) (Stb/d) Failure
the size of the Central Process Facility will be based
on the total reserves of the remaining successful Alpha 150 5 2000 30%
prospects. Facility cost and capacity data and
Bravo 300 7 3000 60%
reservoir parameters are summarized in Figure 4 to
6. Charlie 400 7 4000 80%

Figure 4 - Satellite Production Facilities

Reserves CAPEX OPEX Capacity Construct The individual facility cost, capacity and construction
6
(10 Stb)
6
(10 $)
6
(10 $/year) (b/d) (days) time as a function of reserves are tabulated in
PetroVR for Alpha, Bravo, Charlie Facilities and for
0 0 0 0 0 the Central Process Facility. A table lookup function
300 100 10 50000 500 for the same variables which selects the appropriate
value dependent on the reserves is also entered.
>300 130 11 70000 600
The final step in generating the probabilistic
distribution of net present value is to run a Monte

By Torben Riis and Jerry S. Blinten, Caesar Petroleum Systems Page 3 of 5

EXPLORATION RISK MODELING DRAFT

Figure 7 – Decision Tree S(ABC)

Charlie Successful?
468.7 $MM Yes 20.0% NPV 10: 828.2 $MM
S(AB) F(C)
Bravo Successful? Yes 40.0% No 80.0% NPV 10: 378.8 $MM
169.7 $MM
S(AC) F(B)
Yes 70.0% Charlie Successful? Yes 20.0% NPV 10: 379.1 $MM
-29.7 $MM
S(A) F(BC)
Alpha Successful? No 60.0% No 80.0% NPV 10: -131.9 $MM
166.3 $MM
S(BC) F(A)
Yes Charlie Successful? Yes 20.0% NPV 10: 763.5 $MM
361.4 $MM
S(B) F(AC)
Explore License Area? Bravo Successful? Yes 40.0% No 80.0% NPV 10: 260.9 $MM
166.3 $MM 158.5 $MM
S(C) F(AB)
No 30.0% Charlie Successful? Yes 20.0% NPV 10: 249.6 $MM
23.2 $MM
F(ABC) S(A): Alpha Success
No 0.0 $MM No 60.0% No 80.0% NPV 10: -33.4 $MM F(A): Alpha Fail

Carlo simulation on the entire Development and in Figure 8 are consistent, demonstrating that the
Economic Model. A US Tax Economic Model is used decision tree can be substituted by a single project
to calculate the Net Present Value at 10% discount Development Model.
factor (NPV10).
Obviously, this is a very simplistic example. However,
Subsequently, the conventional decision tree is adding just one other uncertainty in the system will
created for comparison. Eight development grow the decision tree exponentially. Another branch
scenarios, one for each branch, are built in PetroVR. on the chance nodes will bring the total number of
branches to 27.
Based on the decision tree (Figure 7), the NPV10 of
the license area development is calculated at 166.3
Example Including Reserve Risk
$MM compared to the mean NPV10 value of 168.2
$MM from the Monte Carlo simulation after 500
and Field Sequence
iterations (Figure 8). The probability functions for the The example has served to demonstrate that a
Monte Carlo simulation and the decision tree shown decision tree can be modeled precisely with
PetroVR’s Monte Carlo simulation capabilities. In the
Figure 8 – Monte Carlo vs. Decision Tree following example three simple additions are made to
the Development Model to address reserve size and
Cumulative Probability field development sequence.
1
Uniform distributions were defined for recovery factor,
0.9
and triangular distributions were defined for net-pay
Cumulative Probability

0.8
thickness. Furthermore, the initial well rates for the
0.7
three prospects were assumed to follow a normal
0.6
distribution. The development start dates were
0.5
defined as a function of initial well rates, so the field
0.4
with the highest initial well rate is developed first
0.3
0.2
(Figure 9). While correlations can be entered, for this
Monte Carlo - 500 Iterations
example it was assumed that all the distributions
0.1 Decision Tree
0
were uncorrelated.
-200 0 200 400 600 800 1000
NPV ($MM)

By Torben Riis and Jerry S. Blinten, Caesar Petroleum Systems Page 4 of 5

EXPLORATION RISK MODELING DRAFT

Figure 9 – Risked Parameter Ranges Figure 10 shows the cumulative probability

Prospect Recovery F. Net Pay Well Rate
distribution of 500 iterations of the expanded
(Distribution) (Uniform) (Triangular) (Normal)
development model. Even though the distribution of
Value Range Range Mean/S.Dev
NPV(10) has been smoothed by the additional
probability distributions, it still contains a distinct
Alpha 20-28% 40-60ft 2000/200 bpd modality. Inspection of the frequency curve in Figure
Bravo 21-35% 50-100 ft 3000/150 bpd 10 reveals that points cluster in three groups. The
first cluster is zero or one successful field (the highest
Charlie 24-32% 80-120 ft 4000/300 bpd
peak is where all three prospects fail). The middle
cluster indicate two fields and the high cluster three
Assuming that each distribution is represented by 3 successful developments.
branches in a decision tree, it would require 262,144
branches to represent all possible combinations of Conclusions
reserves and development sequence. A traditional Two examples have been presented to demonstrate
approach aimed at simplifying the problem would the ability to model decision trees in a single
likely commence with a probabilistic analysis of total Development Model. For this to be successful, the
reserves including chance of failure. Subsequently, Development Model must have the capability to
field production profiles would be generated and change the development plan parameters “on the fly”
matched to recover the previously calculated in a Monte Carlo simulation.
reserves. The development sequence would probably
In the first example, the chance of failure was
be determined by well rates. This process is both
assumed to be the only risk variable. A traditional
time consuming and many simplifying assumptions
decision tree analysis was compared to a Monte
would be made.
Carlo simulation of the PetroVR Development Model.
Setting up a Monte Carlo simulation requires The results are identical.
discipline and the user must ensure that every
Next, net-pay thickness, recovery factor and initial
outcome in the Monte Carlo simulation makes sense.
well rate were introduced as risk variables. In
However, having all risked parameters contained in
addition, the development sequence was linked to
one Development Model helps understand the impact
well rates so the field with the highest initial rates is
of overall risk on the project, as opposed to an
developed first. This demonstrated that
approach of sequentially risking project parameters.
circumstances which would be complicated to model
in a decision tree can easily be incorporated in a
Figure 10 – Expanded Risk Analysis Results single Development Model.

Monte Carlo Simulation - 500 Iterations Finally, we have demonstrated that the capability to
100% 0.3 include all risk variables in a single Development
90% Cumulative Model is superior to performing a sequence of risk
Frequency 0.25 analyses on reserves, rates, timing etc.
Cumulative Probability

80%
70%
0.2 Working with a single probabilistic model will improve
Frequency

60%
understanding of the impact that uncertainty has on a
50% 0.15
project, as well as serving as a communication tool to
40%
0.1 management. Regardless of which modeling method
30%
is used, the responsibility for having reasonable
20%
0.05 assumptions on the model uncertainty will still lie with
10%
the geologist or engineer.
0% 0
-400 -200 0 200 400 600 800 1000
NPV ($MM)

By Torben Riis and Jerry S. Blinten, Caesar Petroleum Systems Page 5 of 5