Professional Documents
Culture Documents
IADC/SPE 151130 Uncertainty Quantification of Real-Time Pore Pressure Models Caused by Normal Compaction Trend Line Setting
IADC/SPE 151130 Uncertainty Quantification of Real-Time Pore Pressure Models Caused by Normal Compaction Trend Line Setting
This paper was prepared for presentation at the 2012 IADC/SPE Drilling Conference and Exhibition held in San Diego, California, USA, 6–8 March 2012.
This paper was selected for presentation by an IADC/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 International Association of Drilling Contractors or the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect
any position of the International Association of Drilling Contractors or 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 International Association of Drilling Contractors or 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 IADC/SPE copyright.
Abstract
Pore pressure related wellbore instability is a recognized drilling challenge requiring attention as early as possible to mitigate
and remediate severe drilling hazards such as kicks. Methods exist to quantify the effects of certain input parameters on pore
pressure models, but the effect of the different, partly manual stages in pore pressure modeling on the uncertainty has rarely
been addressed in the past.
This paper highlights uncertainty sources associated with the basic stages in pore pressure modeling: shale discrimination
using the gamma ray log and definition of the normal compaction trend line, either manually or using a regression analysis.
Furthermore, filtering of the data prior to trend line definition may also add to some amount of uncertainty. These uncertainties
in pore pressure prediction introduced by the modeling stages is demonstrated on an example data set from an offshore well. In
particular, we show that applying different shale discrimination approaches can introduce pore pressure variations.
Furthermore, a variation in depth intervals over which the trend line is automatically determined by linear regression can add
more uncertainty to the pore pressure model.
An approach is presented to quantify pore pressure uncertainty by automatically analyzing the variation in normal compaction
trend lines. The application of the proposed method aims to constrain and quantify the error associated in modeling pore
pressure in real-time, based on surface and subsurface (MWD and LWD) data. Modeling both pore pressure and its uncertainty
creates additional value for safer drilling, because rig and remote monitoring personnel are made aware of a dynamic updated
pore pressure range in addition to the possible pressure regimes interpreted during pre-drill modeling.
Introduction
Wellbore stability problems related to overpressure are recognized as an important drilling challenge that may increase drilling
costs considerably. Therefore, early attention is required to mitigate and remediate severe drilling hazards.
In many cases and particularly in young sedimentary basins, the generation of overpressure is related to the sedimentation and
compaction process. Usually, under normal burial conditions, sediment porosity decreases with depth due to compaction as the
fluid entrapped in the sedimentary sequences escapes. However, in case of high sedimentation rate or in very impermeable
sediments, the release of pore fluid might be restricted, and fluid pressure (pore pressure) increases above hydrostatic, i.e. the
sediment has higher porosity than under normal conditions (undercompaction or compaction disequilibrium). Methods for
compaction based pore pressure models are thus applicable to porosity-indicating logs from impermeable formations only,
which makes using data from pressure tests from permeable formations challenging in its application to shale-based pore
pressure.
All compaction based modeling approaches make use of various input data and involves several workflow steps, each of which
can contribute some amount of uncertainty to the predictive model. In total, Aadnoy (2011) estimated the uncertainty in pore
pressure modeling from downhole logs to be around ± 30%. A frequently used approach to account for uncertainty in pore
pressure evaluation is to simply introduce a safety factor, ranging for example between 0.5 and 1 ppg (e.g. Falcao, 2002).
However, using such safety factors imposes limits to the well design and is particularly critical in deep water operation where
the pressure window for drilling is very narrow (Falcao, 2002). Therefore, a better way of constraining uncertainty is always
desired.
2 IADC/SPE 151130
A variety of approaches have been tested to estimate the uncertainties associated in pore pressure modeling. For example,
Falcao (2002) used a regional data base containing pore pressure data from formation pressure tests to reduce the safety factor
based on the statistics of the formation pressure tests. Dodds et al. (2001) and Dodds and Fletcher (2004) evaluated the risk of
encountering high pore pressures using a decision-making process for the entire well planning process and Interval Probability
Theory to obtain uncertainty. Several authors evaluate uncertainty caused by input data (e.g. downhole logs, seismic) based on
Monte-Carlo methods (e.g. Liang, 2002; Doyen et al., 2003; Malinvero et al., 2004; Moos et al., 2004).
The usual approach of pore pressure modeling involves different steps of human interaction. Although Matthews (2004) stated
that human interpretation is responsible for the greatest part of uncertainty in pore pressure modeling, the influence of human
decisions is often not taken into account. Some studies also investigated the uncertainty caused by the selection of different
normal compaction trend lines and Eaton exponents in case of the Eaton method (e.g. Malinvero et al., 2004; Sayers et al.,
2006). However, the uncertainty in pore pressure associated with the steps of human interaction; in particular the setting of
1. Shale discrimination. This is required to ensure that only clean shale intervals are included into the compaction
model. Typically the gamma ray log is used to distinguish between shale and non-shale formations. The analyst draws
one or more shale discrimination lines on the gamma ray log and all data (e.g. resistivity or acoustic data) with the
corresponding gamma ray values higher than the gamma ray value at the cut-off line are analyzed for pore pressure.
2. Filtering of shale points. Often the shale resistivity or shale acoustic data will be filtered to further reduce scattering
associated with bad hole condition, recorded noises, etc. However, this step is optional.
3. Calculation of overburden gradient. Overburden gradient is required to calculate pore pressure following equation (1)
or (2). If available, the density log will be integrated to calculate the overburden. Otherwise, acoustic or seismic data
can be used by applying a velocity – density transform.
4. Definition of a normal compaction trend line. Usually the shallow section of the well, which is considered to be
normally (hydrostatically) pressured is used to analyze the normal compaction profile, i.e. the decrease in porosity
with depth. In most cases it is manually interpreted.
5. Calculation of pore pressure. Pore pressure is calculated using equation (1) or (2).
6. Calibration and determination of a definitive pore pressure curve. If additional information on pore pressure is
available (e.g. formation tests in permeable formations, geological or structural information), the analyst usually
adjusts the compaction trend line accordingly - and sometimes even the Eaton exponents - in a way that the
interpreted shale pore pressure profile shows a proper fit to these data. Modeling a number of pore pressure curves
from different input data (resistivity, acoustic), the analyst has better options to constrain the definitive pore pressure
profile.
IADC/SPE 151130 3
Uncertainty in the pore pressure model can result from uncertainties associated with the input data (resistivity, acoustic logs)
or could be related to the different modeling steps. In this paper, we will not deal with uncertainty associated with the input
data. Instead, we will concentrate on uncertainty introduced in modeling step 1 (shale discrimination), step 2 (filtering) and
step 4 (definition of normal compaction trend line). As we focus on automation of the pore pressure modeling workflow, the
manual definition of the normal compaction trend line in step 4 is replaced by an automated algorithm as described in
Wessling et al. (2011).
• In a very detailed approach, a higher number of shale lines (almost 30 lines for the 11000 ft interval) were picked by
an interpreter. This style will be called “detailed” in the following discussion.
• In another approach, only one shale line was picked per casing interval. Such an approach is often recommended (e.g.
Matthews, 2004)
• Finally, for a third approach the gamma ray log was subdivided into four major units and one shale line was picked
for each of these units. Shale lines were placed at slightly lower API values compared to the first two approaches.
These different styles result in a different number of shale points generated for the analysis (Table 1). While the first two
approaches do not show much difference in the number of shale points, the third approach included a significantly higher
number of shale points as the interpreter placed the shale lines at a lower API value or reduced discrimination compared to the
other two approaches.
Table 1:Statistics (coefficient of determination R² and slope for the trend lines) for the different shale discrimination approaches
Number Shale points not filtered Shale points filtered
Style of shale points R² slope R² slope
Detailed 1223 0.27 0.000063 0.41 0.000053
One shale line per casing interval 1044 0.10 0.000033 0.34 0.000025
Major units 1688 0.20 0.000058 0.42 0.000053
4 IADC/SPE 151130
Normal compaction trend lines in Fig. 1 were calculated using a linear regression over the interval 4000 – 8803 ft. Two normal
compaction trend lines plot almost over one another, only the trend line based on the shale discrimination approach with only
one shale line per casing interval shows some variation. Table 1 shows that there is some variation in slope which is here
referred to as the gradient of the normal compaction trend. The quality of the fits varies significantly and is generally higher
for the regressions based on the filtered shale points than for the data sets without filtering, as indicated by the higher values of
R². The pore pressure curves in Fig. 1 do not show much difference in the normally pressured part where the gamma ray log is
homogeneous and also the resistivity shale point sets are similar.
Different filters (pore pressure modeling step 2) were applied to the shale point data set from the detailed approach (Fig. 2): a
moving average filter over 99 data points, a 99 point corrected mean filter (cmf), and a depth based median filter with a
maximum interval length of 100 ft. Although the trend lines calculated for the three shale resistivity data sets are very similar
and show identical profiles in Fig. 2, there are some differences in the calculated pore pressure curves. In particular, using the
Fig. 2: Pore pressure analysis based on the same shale point data set (“detailed” in Fig. 1) but with different filtering methodology
applied to the shale points. Left track: Gamma ray log (green). Second track from left: Resistivity shale points (not filtered) with trend
line. Third track from left: Resistivity shale points smoothed with three different filters and corresponding trend lines. Right track:
pore pressure gradient calculated from the filtered resistivity shale points.
Table 2: Statistics (R² and slope for normal compaction trend lines) for applying different filter methodology
Methodology R² Slope
No filter 0.27 0.000063
99 pt filter 0.41 0.000053
cmf 0.41 0.000053
Depth based 0.29 0.000053
play a minor role in pore pressure uncertainty magnitudes. In case of strong scattering in the porosity-indicating log and
frequent fluctuations, appropriate choice of depth interval for calculating the linear regression is crucial to obtain a normal
compaction trend line which is representative. While the analyst will generally visually inspect the data set for an appropriate
depth interval when manually placing the trend line, an automatic algorithm will require specific criteria for this choice. Here
we investigate and quantify the uncertainty in the pore pressure model associated with the selection of the depth interval which
is used to calculate the normal compaction trend line.
For the algorithm, two depth intervals need to be defined to calculate a series of trend lines (TL): a start interval containing n
data points and an end interval containing m data points (step 1 in Fig. 3). In step one, a linear regression analysis is performed
over the interval beginning at the first data point from the start interval (i=1) and ending at the first data point from the end
interval (j=1), yielding the normal compaction trend line TL1,1. Step n of the analysis defines the interval for linear regression
from data point i=n to j=1, giving TLn,1 (center track in Fig. 3). The final analysis is performed for i=n, j=m to obtain trend line
……
Start Start Start
interval interval interval
n n n
Interval f or
linear
Interval f or
regression
linear
regression Interval f or
Trendline Trendline linear Trendline
TL1,1 TLn,1 regression TLn,m
Depth (ft)
Depth (ft)
Depth (ft)
Case Study
This paper extends the work previously done by Freitag et al. (2004) by applying the uncertainty quantification approach
described above. Resistivity and acoustic data from a deepwater well in the Gulf of Mexico were used to model pore pressure
applying Eaton’s method. Formation pressure and mud weight data were available to calibrate the pore pressure model so that
drilling was successfully conducted even through the narrow pressure window.
To compensate for a fault at about 12000 ft depth Freitag et al. (2004) bisected the normal compaction trend lines of the
porosity indicating logs (acoustic and resistivity) at that depth. The acoustic trend line above the fault was shifted towards
lower slowness values, yielding a higher pore pressure in the upper section to match the mud weight. Below the fault the
resistivity trend line was shifted towards lower resistivity values, yielding in lower calculated pore pressure (see Freitag et al.,
6 IADC/SPE 151130
Figure 4: Examples of average pore pressure curves from using a constant starting interval and varied end interval positions for the
calculation of the normal compaction trend line. Definition of start and end interval is illustrated in Fig. 3. Left track: Average pore
pressure curves (solid lines) ± 1 standard deviation (dashed lines) determined from the resistivity log; right track: Average pore
pressure curves (solid lines) ± 1 standard deviation (dashed lines) calculated from the acoustic log.
Table 3: Overview of uncertainty expresses as standard deviation associated with different end intervals as in Fig. 4.
Standard deviation
Resistivity Acoustic
Lower depth interval (ft) 6500 – 8000 7500 – 9000 8803 – 10115 10300 – 10800 10600 - 11100 11000 – 11500
Normal compaction zone 0.215 0.145 0.105 0.105 0.075 0.060
Under compaction zone 0.855 0.440 0.285 0.475 0.260 0.180
Overall 0.610 0.305 0.185 0.375 0.200 0.140
The proposed algorithm has been applied to resistivity and acoustic data. Only shale intervals were included into the modeling
procedure (see step 1 of the pore pressure modeling sequence above) and shale points were filtered using a moving average
filter over 99 data points (step 2 in the pore pressure modeling procedure). The results show that the uncertainty associated
with variations in the normal compaction trend line remains reasonable. Furthermore, it also allows for a successful
reconstruction of pressure conditions in the narrow pressure window at the bottom of the well.
After defining upper and lower intervals for the porosity indicating logs (resistivity; upper interval: 3068-4573 ft; lower
interval: 8803-10115 ft / acoustic data; upper interval: 9000-9500 ft; lower interval: 11000-11500 ft) the algorithm was run
with the same parameters as used by Freitag et al. (2004), i.e. the typically applied Eaton exponent of 1.2 for resistivity data
and 3.0 for acoustic data were applied. Normal formation pressure was assumed to be 8.75 ppg.
While Freitag et al. (2004) bisected their compaction trend lines, the normal compaction trend line algorithm works with one
single trend line only. However, a shift in normal compaction trend line was required to adjust the modeled pore pressure
IADC/SPE 151130 7
profiles to pressure conditions below the fault zone at 12000 ft (see Fig. 4 for position of the fault zone) as deduced from
formation pressure tests and mud weight. It was decided to perform a manual parallel shift of the pore pressure curves at
12000 ft to compensate for the bisection. Note that the parallel shift does not change trend line slope nor pore pressure
uncertainty. The average pore pressure curve derived from the resistivity log was parallel shifted by -2 ppg below 12000 ft to
compensate for the overlying fault and second trend line, respectively. The average pore pressure curve from the acoustic log
was parallel shifted by +1 ppg above 12000 ft to take account for the upper normal trend line in the reference data. Both pore
pressure curves were plotted with a ± 1standard deviation corridor in Fig. 5. Uncertainty corridors for pore pressure curves
calculated with resistivity and acoustic data agree particularly well below 14500 ft and show a reasonably small variation of ±
0.26-0.30 ppg below the mud weight. Also note that the selected mud weight was sufficiently high to exceed not only the
average pore pressure but also the pore pressure plus one standard deviation.
Discussion
Different shale discrimination approaches lead to variations in the pore pressure curves because each of the shale
discrimination approaches results in a different number of shale points, and hence the data density with respect to these points
varies locally (Fig. 1). Thus a different sub set of shale points is included for each regression analysis to determine the normal
compaction trend line. Although the difference in normal compaction trend line seems to be minor in our example, in deeper
sections even small differences between the trend lines may have a larger effect when the trend lines show increased
separation.
Moreover, the shale point sub sets were filtered applying identical filters with equal parameters. As number and density of data
points is different for each sub set, the filtered curves show differences and consequently, the pore pressure curves differ. This
is particularly evident below 9000 ft in this example (Fig. 1), where the gamma ray log shows a slight decrease with depth. In
that depth interval, the one line per casing interval approach gives a lower data density than the other shale picking approaches
and the associated pore pressure curves shows less scattering.
The different filter methods also lead to divergent pore pressure curves. Several filters were applied to the same shale point sub
set. This has a minor effect on the calculation of the normal compaction trend line, because low pass filtering does not affect
the overall trend in the data. Nevertheless, differences in predicted pore pressure profiles occur as the different filters have
varying capabilities to reduce scatter in the data. In this example, the moving average and the corrected mean filter do not
show much difference because both were applied over a long window (99 data points). In cases where the window size is
reduced, the two filters can show varying results. The depth based filter was less able to reduce scattering in the data in this
case. This is because the depth based filter does not average over large gaps in the shale data (i.e. the presence of thick sand
intervals) so that some intervals consist of only a few points.
It was investigated in more detail and in a quantitative manner how far the choice of the depth interval, which is used to
calculate a normal compaction trend line, can contribute to uncertainty in the pore pressure curve. By varying the end interval
for calculating the normal compaction trend line with depth, the authors simulated the real-time case with data continuously
streaming in. By extending the depth interval for the computation of the normal compaction trend line to a greater depth, more
data is included into the analysis. This gives a better definition of the normal compaction trend line when drilling deeper as
demonstrated by the decreasing uncertainty of the pore pressure curve (Fig. 4).
8 IADC/SPE 151130
References
Aadnoy, B.S. 2011. Quality Assurance of Wellbore Stability Analyses. Paper SPE / IADC 140205 presented at the SPE/IADC Drilling
Conference and Exhibition in Amsterdam, The Netherlands, 1 – 3 March 2011.
Bartetzko, A., Wessling, S., Pei, J., and Dahl, T. 2011. Challenges for wellbore stability workflow automation. Oil and Gas European
Magazine 2, 65-71.
Dodds, K.J., Fletcher, A., Hennig, A., Bekele, E., Johnson, M., Abriel, W., Higgs, W., and Strudley, A. 2001. An Overpressure Case History
Using a Novel Risk-Analysis Process. Paper OTC 13102 presented at the 2001 Offshore Technology Conference, Houston, Texas, 30
April - 3 May, 2001.
Dodds, K., and Fletcher, A. 2004. Interval probability process mapping as a tool for drilling decisions analysis – the R&D perspective. The
Leading Edge, June 2004, 558 – 564.
Doyen, P.M., Malinverno, A., Prioul, R., Hooyman, P., Noeth, S., den Boer, L., Psaila, D., Sayers, C.M., Smit, T.J.H., van Eden, C., and
Wervelman, R. 2003. Seismic pore pressure prediction with uncertainty using a probabilistic mechanical earth model. SEG Expanded
Abstracts 22, 1366, doi:10.1190/1.1817542
Eaton, B.A. 1975. The equation for geopressure prediction from well logs. Paper SPE 5544-MS presented at 50th Annual Fall Meeting of the
Society of Petroleum Engineers of AIME, Dallas, Texas, 28 September – 1 October, 1975.
Falcao, J.L. 2002. Uncertainties in the Pore Pressure Evaluation in Deepwater: a Statistical Approach. Paper SPE 78247-MS presented at the
SPE/ISRM Rock Mechanics conference in Irving, Texas, 20-23 October, 2002.
Liang, Q.J., 2002. Application of Quantitative Risk Analysis to Pore Pressure and Fracture Gradient Prediction. Paper SPE 77354-MS
presented at SPE Annual Technical Conference and Exhibition, San Antonio, Texas, 29 September, - 2 October, 2002.
Malinverno, A., Sayers, C.M., Woodward, M.J., and Bartman, R.C. 2004. Integrating diverse measurements to predict pore pressure with
uncertainties while drilling. Paper SPE 90001-MS presented at the SPE Annual Technical Conference and Exhibition, Houston, Texas,
26 – 29 September, 2004.
Matthews, M.D. 2004. Uncertainty – Shale Pore Pressure from Borehole Resistivity. Paper ARMA/NARMS 04-551presented at Gulf Rocks
2004, the 6th North America rock Mechanics Symposium (NARMS) in Houston, Texas, June 5 – 9, 2004.
Moos, D., Peska, P., Ward, C., and Brehm, A. 2004. Quantitative Risk Assessment Applied to Pre-drill Pore Pressure, Sealing Potential, and
Mud Window Predictions from Seismic Data. Paper ARMA/NARMS 04-499presented at Gulf Rocks 2004, the 6th North America rock
Mechanics Symposium (NARMS) in Houston, Texas, June 5 – 9, 2004.
Morita, N. 1995. Uncertainty analysis of borehole stability problems. Paper SPE 30502presented at the SPE Annual Technical Conference
and Exhibition, Dallas, October 22-25, 1995.
Sayers, C., den Boer, L., Nagy, Z., and Hooyman, P. 2006. Well-constrained seismic estimation of pore pressure with uncertainty. The
Leading Edge, December 2006, 1524-1526.
Wessling, S., Bartetzko, A., Wang, X., Pei, J., and Dahl, T. 2012. Automated Wellbore Stability Systems- Experience with Pore Pressure
Modeling. Paper IPTC 14935 presented at the International Petroleum Technology Conference, Bangkok, Thailand, 07–09 February
2011.