Title: PETROLEUM ENGINEERING 689 Special Topics in Unconventional Resource Reserves Lecture 11 Applied Prob
1PETROLEUM ENGINEERING 689Special Topics
inUnconventional Resource ReservesLecture 11
Applied Probabilistic Reserves Texas AM
University - Spring 2007
2Outline
- Summing Reserves
- Bootstrap Method
- Choosing A Distribution Type
- Scoping Analysis
3Learning Objectives
- How to sum reserves classes at the well, field
and company level - Be able to generate reserves classes from DCA
- Know when to choose a log-normal distribution or
something else - How to set up a probabilistic scoping analysis.
4Summing Reserves
- One cannot add proved reserves for zones to get
well proved reserves. Nor can one add well
reserves to get field reserves. Mathematically
incorrect. Leads to the wrong answer. - - E.C. Capen, SPE paper 73828
5Summing Reserves
- In practice, the vast majority of reserves
analysts will add reserves per well to get field
reserves regardless of the reserves class - This works only if the mean reserves per well are
being added to get the mean reserves for the
field (or Zone ? Well or Field ? Company) - The reserve class thresholds of Proved (P10),
Probable (P50) and Possible (P90) are not equal
to the Mean and have to be summed by a different
method - Read SPE Paper 73828
6Summing Reserves
- Very few reserves analysts actually apply a true
probabilistic approach to generate P10, P50 P90
reserves, further complicating the issue - Without a proper approach to compute P10, P50 and
P90 reserves, there is no proper technique for
summing - The proper technique is not hard as long as the
threshold probabilities were computed using
actual statistical methods.
7Summing Reserves
- In Practice, Many Reserves Analysts Use Something
Close To The Mean When Assigning Proved Reserves - Producing Wells Mean is probably assigned as
Proved more often than not - Proved Undeveloped Locations / Behind Pipe
Probably still some Mean assignments
8Summing Reserves
- Lets Start With Definitions
- Proved Definitions vary, but words such as
reasonable certainty or confident are often
used. In practice, industry claims it is moving
to assigning reserves to the proved class if
there is a 90 probability that the reserves
level will be met or exceeded. - Applying the 90 probability that reserves will
exceed a certain threshold should correspond
statistically to P10 on a probability plot. - Proved P10 90 chance that reality will be
above the proved value.
9Summing Reserves
- Lets Start With Definitions
- Probable Now generally defined as meeting or
exceeding a P50 threshold - Possible New generally defined as meeting or
exceeding a P90 threshold (P90 means only a 10
chance that reality will be at or above the
threshold ? 90th percentile on a cumulative
probability plot (ascending)).
10Summing Reserves
11Summing Reserves
- Why Cant You Add Two P10 Proved Reserves Cases
to Get a Combined P10 Proved Case? - If you sample the distribution only one time, you
will get the P10 value or less 10 of the time,
but if you sample the distribution more than one
time the chances of getting all the samples at
the P10 value or less become exceedingly small - Adding P10 proved reserves will underestimate the
combined proved reserves
12Rolling the Dice Example
- With One Die
- Assume Proved Reserves a one in six chance
- Hurdle corresponds to a One on the die (P16.7)
- With Two Dice
- Chance of rolling a one on each die is 1/6 1/6
1/36 ? much smaller chance of happening than
when rolling only one die. - If Proved meant it had a 1/6 chance of
happening, then using the One on a die as the
test for Proved only works if only one die is
rolled (i.e. only one well is drilled).
13Outcomes of rolling two dice
- If Proved lowest 1/6 of outcomes, then any
two-dice combination equaling 4 or less would be
Proved.
14Outcomes of rolling six dice
- For Six Dice, Proved Could Be a Large Number of
Dice Combinations, With Some Individual Dice With
Numbers Well Above One.
The chance that a roll of six dice produces six
ones is 1 in 46,000!
15Outcomes of rolling six dice
- Possible Totals From 6 to 36
- Proved With Six Dice
- Lowest 1/6 of all possible outcomes
- Any combination of dice totaling 17 or less
- Average value per die of 2.8, or 280 the
Proved value when rolling only one die - Some individual dice could have the highest
possible value of 6
16Summing Reserves
- How Do We Add Reserves?
- Reserves are added by maintaining the uncertainty
in the combined population of outcomes. - Statistically, this can be accomplished by adding
the Means and the Variances. - See SPE 73828 for formulas
17Summing Reserves
- Another Way To Add Reserves is to Sample The
Overall Distribution From The Perspective Of An
Entire Drilling Project. - Example Reserves For Undeveloped Drilling
Locations - Assume the reserves distribution on the following
slide - Assume a certain number of locations are
available for drilling
18Summing Reserves
19Summing Reserves
- If We have One Drilling Location
- P10 Proved Reserves 0.046 Bcf
- P50 Probable Reserves 0.222 Bcf
- P90 Possible Reserves 1.321 Bcf
- What if We Have Two Drilling Locations?
- We have to determine P10, P50 P90 outcomes for
two-well pairs
20Summing Reserves
- We Can Perform A Monte Carlo Simulation of
Two-Well Outcomes - Randomly sample the distribution twice
- Compute the average reserves from this two-well
sample - Treat the two-well average reserves as a new
random variable - Monte Carlo the two-well average variable
21Summing Reserves
This Distribution is Defined As RiskLognorm(0.5,
1, RiskTruncate(0.01, 5))
22Summing Reserves
23Summing Reserves
- Now The Proved Reserves Per Well Have Increased!
- 0.091 vs. 0.046 Bcf
- But The P90 (Possible) Reserves Have Decreased
- 0.914 vs. 1.32
24Summing Reserves
25Summing Reserves
26Summing Reserves
27Summing Reserves
28Summing Reserves
- As The Size Of The Drilling Program Increases
- The variance decreases (spread between the P10
P90 reserves - The Mean expectation remains unchanged
- The per-well P10 P50 reserves increase, with
the P50 reserves nearing the Mean reserves - The P90 reserves decrease
- The average per-well reserves distribution
approaches a normal distribution for large
drilling programs
29Summing Reserves - Implications
- If Two Companies Have Similar Acreage In The Same
Play - The company with more drilling locations, can
book higher per-well proved reserves (P10) - Both companies would have the same expected value
per-well since the Mean is used to determine the
value of the wells
30Summing Reserves - Implications
- Current Industry Practice of Using P10 as a
Criteria for Proved Reserves - Creates a large gap between Proved reserves (P10)
and Asset Value (Mean), especially in
Unconventional Resources - Makes Proved reserves of little business value
other than as a floor to corporate value - Does not reflect the Expected Value of the
reserves - Shows the need for conveying the entire reserves
distribution rather than just this one small piece
31Summing Reserves
- To Sum Reserves, Do One Of Two Things
- Model the combined process and/or distributions
using Monte Carlo techniques - Add the Means and Variances using statistical
equations - Either requires you to generate probability
distributions
32Summing Reserves
- The Prior Examples Are For Assigning Reserves To
Wells Before They Are Drilled - To Apply This To Producing Wells, The Concept Is
The Same, But You Must Have A Method To Determine
P10, P50 P90 Production Forecasts
33Bootstrap Method
- One Method For Determining Probabilistic Reserves
For Production Data Is The Bootstrap Method - Outlined in Lecture 9
34Bootstrap Method For DCA Analysis
- Single-Well Forecast Confidence Levels
- P10, P50 P90 for the remaining life of a single
existing well - Deals Effectively With Noise in Production Data
- More noise greater range between P10 P90
- Does Not Require Any a priori Knowledge about DCA
Parameters (Qi, Di, b) But it Generates Them
for You!!! - Repeatable No Matter Who Does The Analysis
35Bootstrap Method For DCA Analysis
- See SPE Papers 36633 95974
- Randomly Samples Historic Production Data To
Create a Synthetic Production Curve - Uses Sampling With Replacement
- If you have 50 historic data points, perform 50
random samples with replacement - Some historic data points will be duplicated,
some omitted - Regression-Fit the synthetic curve to find DCA
parameters (Qi, Di, b) - Repeat until a large number of synthetic
production curves are generated Now you have a
Qi, Di, b distribution set to work from.
36Bootstrap Method For DCA Analysis
- Much Improved Forecasts May Be Possible If You
Use Subsets of Your Production Data (i.e. recent
data) And If You Use the Modified Bootstrap
Method.
37DCA Forecast Using All Historical Data
38DCA Forecast Using Last 4 Years of Data
39DCA Forecast Using Last 2 Years of Data
40Bootstrap Example
41Bootstrap Example
- 100 Iterations Of Generating Copies of The
Production Data Using The Bootstrap Method - Used Excel Solver to regress on each copy to
generate Qi, Di b - Ranked each iteration by EUR
- Picked P10, P50 P90 curves based on ranking
42Bootstrap Example
43Bootstrap Example
44Bootstrap Example
- The Distribution of The Bootstrap Outcomes is a
Normal Distribution - Therefore P50 Mean
- This may not be the case for all Bootstrap
outcomes on other data sets
45Bootstrap Example
46Bootstrap Example
- A Least Squares Regression Of The Raw Data Gives
Us Essentially The Same Curve As The P50 Case - The Least Squares Fit Is The Usual Way Reserves
Auditors Make Decline Curve Projections For
Proved Reserves - Proved Least Squares P50
- If normally distributed, then Least Squares
Mean Proved reserves can be added. - These Proved reserves ARE NOT P10 reserves
47Bootstrap Example
- What Does This Mean?
- In practice, Proved Developed Producing Reserves
may in fact be similar to a P50 or Mean case, and
if so then can be added - Industry is not prepared to apply a more
sophisticated probabilistic approach to generate
a true P10, P50, P90 distribution of production
forecasts
48Bootstrap Example
- This Example Only Addresses Uncertainty In The
Production Data - Deals with how to weight individual data points
- Could be tied to a reservoir simulator for
forecasting rather than regressing on a decline
curve equation - Does not address uncertainty in reservoir
description of reservoir model
49Distribution Types
- How Do You Choose A Distribution Type For Various
Data? - Reservoir pay, perm, porosity, OGIP, etc.
- Reserves
- Economic Inputs product prices, operating
costs, etc. - Some Data Tend To Follows A Certain Distribution
Type
50Distribution Types
- The Central Limit Theorem (CLT)
- Adding or subtracting distributions will tend
towards a normal distribution - Multiplying or dividing distributions will tend
towards a log-normal distribution
51Distribution Types
- Triangular Distributions From SPE 73828
- Distributions Added/Multiplied etc. As Shown Below
52Distribution Types
53Distribution Types
- Recommend Uniform, Triangular, Normal
Log-Normal Distributions Unless You Have Good
Reason To Use Something Different - Per-Well Reserves, Permeability, Drainage Area
- Log-Normal
- Porosity, Drilling Package Reserves
- Normal to Log-Normal
- Decide For Yourself If You Have Sufficient Data
54Distribution Types
- Correct Distribution Will Be A Straight Line On A
Probability Scale
55Scoping Analysis
- Scoping Analysis Example
- Use spreadsheet Scoping.xls
- We will build a scoping analysis in class,
complete a matrix of outcomes and show how to
make business decisions from this matrix of
outcomes.
56Learning Objectives/Accomplishments
- Now You Should Be Able To
- Sum reserves classes at the well, field and
company level - Generate reserves classes from Decline Curve
Analysis - Know what distribution type to choose
- Set up a probabilistic scoping analysis
57End Lecture 11 Applied Probabilistic Reserves