technical large sept2007

8/16/2019 Technical Large Sept2007

http://slidepdf.com/reader/full/technical-large-sept2007 1/13

© 2007 EAGE 39

technical articlefirst break volume 25, September 2007

IntroductionIt is vital to the planning of drilling wells to have an esti-

mate of the expected pressure regime to be encountered in

the subsurface. Direct concerns are the safety of the personnel

and equipment, in particular minimising the associated risks.

Furthermore, it facilitates more effective planning and order-

ing of the required material. With respect to the reservoir, the

right drilling mud weight is important. If it is too low, a blow

out might occur and conversely, if it is too high, the formation

might be damaged by invasion of the drilling fluid.

In recent years the prediction of pore pressure has under-gone various developments. Geoscientists, such as reservoir

and processing geophysicists, are more involved in the appli-

cation of processes to do predictions based on seismic in 3D.

Historically, it was only done on site in 1D. The advancement

of seismic processing allows a more and more accurate esti-

mation of the velocities. The combination of the knowledge

of the engineers and geoscientists is extremely valuable in try-

ing to predict the pressure cubes. Besides extracting the fore-

cast pressure profiles at future drilling locations from these

volumes, they have an interpretation value. They allow scan-

ning to provide alternative well locations, see how the pressure

regime is distributed with respect to the structure and geolo-gy, and help with the identification of sealing or leaking faults.

This last item is even better facilitated when used in combina-

tion with other now standard seismic attribute volumes that

help to identify fluids and sands, such as those from AVO and

elastic inversion.

It is the process of deriving as accurate as possible velocities

from seismic processing which is one of the key factors in pre-

dicting reliable pressures. Obviously, this is where the process-

ing geophysicists come in. On the other hand, this needs to

be combined with an in-depth analysis of the well data, pres-

sure data, and drilling data by an experienced person such as

a petrophysicist and a pressure engineer. The role of the res-

ervoir geophysicist is to integrate the subsurface informationwith the specially conditioned seismic data to achieve the most

reliable results.

In this article we describe the extensive pore pressure pre-

diction (PPP) processing of 3D marine seismic data acquired

over Blocks 01 and AB in the Caspian Sea, Turkmenistan. It

directly follows a pre-stack depth migration (PSDM) under-

taken at the first stage of the complete processing project. This

makes it a unique case study, but it is proven to be impor-

tant (Dutta et al., 2002). In cooperation with the client (Petro-

nas Carigali), a team of processing and reservoir geophysicists

(CGGVeritas) and drilling experts (Knowledge Systems) pro-

vides the necessary input of disciplines. We would like to stress

that it is not the intention of the authors to explain the con-

cepts of pore pressure - extensive literature exists, for exam-

ple, Bell (1998), Bowers (1999, 2001, 2002), Bruce and Bow-

ers (2002), Sayers (2006), and Terzaghi (1943).The data consists of two vintage surveys. The Block 01 sur-

vey is located over the Livanova field approximately 210 km

southeast of Baku covering an area of 1089 km2. The field’s

coverage is 80 fold. The Block AB data is acquired as two sep-

arate surveys: 1A and 1B with the same acquisition parameters

but different shooting direction. The block AB data is located

to the southeast of the Block 01 survey. The Block 1A survey

covers a full fold area of 388 km2. The Block 1B survey is 272

km2 with 54 fold coverage. After the merging of the two sur-

veys, the area covers approximately 1500 km2.

Initial well data pressure analysisAn initial pore pressure prediction is undertaken. This is purely

based on well and pressure data, and will give an understand-

ing of the behaviour and various responses: do the curves react

to varying pore pressure at all? Firstly, the Gamma ray logs

are used to construct filtered petrophysical measurements in

shales (predictions are made in shales and measurements are in

sands). Subsequently, various stress–velocity relationships pre-

dict pore pressure, which are refined to calibration data such

as mud weight, pressure measurements and drilling events. A

second prediction based on resistivity logs is carried out.

First of all, the vertical stress of the overburden needs to be

computed, i.e. the overburden gradient (OBG), by integrating

the composite density function. Most of this composite densityfunction comes from well A, which should be reliable accord-

ing to the caliper-bit size overlay. The reason for constructing a

composite density function is due to missing logging sections in

various wells, as well as measurement accuracy errors related

to hole enlargement. As Gardner (Gardner et al., 1974) over-

Large-scale pore pressure prediction afterpre-stack depth migration in the Caspian Sea

Norbert van de Coevering,1* Hazim Hameed Al-Dabagh,2 Liau Min Hoe,3 and Tony Jolly4

1 CGG AP (a CGGVeritas company), Kuala Lumpur, Malaysia.2 Formerly Petronas Carigali, now Lukoil Overseas, London, UK.3 Petronas Carigali, Kuala Lumpur, Malaysia.4 Knowledge Systems, Houston, USA.*[email protected]



© 2007 EAGE40

technical article first break volume 25, September 2007

Figure 2 Well D with high resistivity values at ~2700 m.

Figure 1 Well C with low resistivity values at ~1200 m.



© 2007 EAGE 41


estimates the densities in the shallow zone (first ~500 m), Mill-

er’s relation which is based purely on measurements by Oster-

meier et al. (2001) is used. In this section there are no meas-

ured densities, which is very common.

Basic pore pressure gradients (PP) are predicted for the ini-

tial input wells using the Miller (2002) and Bowers (1994) cal-

ibrated pressure models. Both these methods use the sonic log

and represent nothing more than a velocity–effective stress

relation. It can be seen that for all these wells, some crossing of

the PP with the MW (mud weight) curves takes place. Also, the

match with the MDT measurements is sometimes not good. It

should be noted, however, that pressures are predicted in shales

and measured in sands. Some of the mud weight curves were

not reliable. In well B, the sonic is estimated, which makes

direct quality control comparisons less reliable. The pressure

model parameters need to ensure predictions above the hydro-

static pressure (~ 9 ppg). It seems from the analysis that the

sonic logs do detect over pressure, hence it makes sense to use

velocities for pore pressure prediction.Eaton’s method (1972, 1975) of predicting pressure uses

resistivity logs. With this type of log, a reliable prediction

is not always possible. Some suspiciously low and high val-

ues, compared to the normal compaction trend (NCT) were

observed. For well C in Figure 1, for example, the resistivity

values around 1200 m are very low, giving anomalously high

pore pressure. In well D in Figure 2 we have very high values,

giving extremely low pressure predictions. For well A in Figure

3, the sharp pressure increase at TD appears to be real, like-

ly due to high pressure water. There is a step increase in MW.

Note that the large discrepancies between the MDT measured

pressures and the sonic predicted pressures in previous Figures1-3 will be addressed later through the application of a vary-

ing compaction trend. The discussion also explains why errors

can remain. For comparisons, in Figure 4 the smoothed pre-

dicted individual well pressures using sonic (Miller and Bow-

ers) and resistivity are displayed. From this, it is decided that

the emphasis will be placed on sonic pressure predictions and

that the resistivity based pressure predictions are not reliable

enough.

For the fracture gradient (FG) prediction, normally the

matrix stress ratio is determined from LOT (leak off test)

measurements. The only data available to use were the LOT or

FIT (formation integrity test) values, not the pressure vs. time

graphs. This makes calibration more difficult (Postler, 1997).

As some of the wells had shear sonic available, this could be

used in the determination of an effective stress ratio K0=PR /

(1-PR)

by computing Poisson’s Ratio (PR) directly from the sonic and

shear sonic. With an average value from the actual PR logs,

K0 is derived as being 0.795 (close to a common best experi-

ence value of 0.8) and used to predict the FG with the Mat-

thews and Kelly method (Matthews & Kelly, 1967). The FGcan also be computed directly using the PR logs (Eaton, 1968,

1997). Subsequently, comparisons with the selected LOT and

FIT data points can be made. For a comparison of the individ-

ual well FG predictions, see Figure 5 (Matthews and Kelly in

blue, using PR log in yellow, a ‘definitive’ pick in greenish col-

our interpreted by the geomechanical expert).

From the figures presented previously, it seems that one

simple NCT most likely does not suffice. Both temporally and

laterally there seems to be considerable variation in the com-

paction trend. This can have various causes, e.g. compartmen-

talization or tectonic uplift concentrated in an area. At this

stage, the depth at which the biggest change occurs in the com-paction trend for each of the wells is analyzed. These depths

Figure 3 Well A showing a sharp pressure increase at TD which appears real .



© 2007 EAGE42


Figure 5 Individual FG comparisons.

Figure 4 Initial PP results filtered using Bowers, Miller (sonic) and resistivity.



© 2007 EAGE 43


are largely consistent with the depth of the RS4 marker. SeeFigure 6 for a description of the two Miller models, one above

(‘Miller 1’) and one below (‘Miller 2’) the RS4 horizon. In Fig-

ure 7 it can be seen that the PP prediction based on this multi-

ple NCT gives better results in relation to the MW curves and

MDT measurements. There is a danger here in adjusting the

NCT to predict desired pressures, but thorough analysis has

shown the pressure variation exists.

Seismic processing and velocitiesAs this PPP is preceded by a PSDM, we have as input CRP

gathers and their associated final velocities. These final RMO

velocities have been manually picked on a coarse grid to apply

the final RMO correction. The velocity model coming out ofthe last iteration from the PSDM is used as input to this RMO

analysis. See Table 1 for a summary of the PSDM processing

sequence.

The high-density simultaneous velocity analysis technique

picks a high density VRMS

and (effective eta – this parame-

ter accounts for the anisotropy of the layered rock, i.e. velocity

variations with direction) field (Siliqi et al., 2003a). This ena-

bles the events to be flattened to higher incidence angle than

would be the case with a 2nd order stacking velocity correction.

Such an automated velocity analysis was performed using a

grid of 8 inlines x 8 crosslines (100 m x 100 m) and an angle

mute of 50

0

on the gathers, using the RMO velocity analysisas an input guide.

Automatic picking velocity processes are generally noisi-

er and some geo-statistical filtering is required for depth con-

version and pore pressure prediction. To remove the noise in

the velocity cube, the velocity analysis was performed with

an internal filtering step. Advanced geo-statistical algorithms,

built into the software, allowed for this filtering by factorial

1 Merge of block 01 and block AB surveys

2 Despiking

3 Initial model building

4 Iteration 1 – global tomographic inversion5 Iteration 2 – global tomographic inversion

6 Iteration 3 – global tomographic inversion

7 Kirchhoff pre-stack depth migration

Offsets = 55 (block AB) and 81 (block 01)

Aperture = 5 X 5 km, dip limit = 70°

8 Final velocity analysis

Manual picking on a 500 X 500 m grid

Using VRMS field (converted from migration interval

velocity) as a guide function

Table 1 Summary of PSDM processing sequence.

Figure 6 Initial pore pressure model parameters using a multiple NCT with depth.



© 2007 EAGE44


kriging (Siliqi et al., 2003b). First the velocity cube is split into

a trend cube, holding the structural component of the veloc-

ity and a residual cube holding the fine scale variations and

the noise. Such decomposition has the advantage of the resid-

uals being stationary, which is the condition to perform fac-

torial kriging, and the structural component of the velocity is

preserved.

Factorial kriging was then performed on the residuals to

remove the noise. The final filtered velocity cube is the sum

of the filtered residuals and the trend cube. The filtering was

first applied on the RMS velocity cube with the following

parameters:

Smoothing window to compute trend:

11 traces x 11 traces x 300 ms

Smoothing window on high frequency part:

6 traces x 6 traces x 20 ms

To enable use of the seismic velocity field for pore pressure pre-

diction, it has to be converted from RMS to interval velocity.

Normally, the regularly sampled velocity field (in time) is con-

verted using the Dix approximation (Dix, 1955), but this can

lead to some instabilities. Furthermore, we want to preserve as

much information as possible, hence a small sample rate has to

be used. To achieve this goal of a high-frequency, stable veloc-

ity conversion, a novel approach has been used as follows:

The gathers, after removing the RMO correction, are high-

er order NMO corrected using the new dense velocity and

eta field. These gathers are stacked to form the highest qual-

ity stack as input. Based on a user-defined amplitude thresh-

old, the stack is used to form a ‘skeleton’ of values of 1 (else-

where the values are 0) along coherent events or where there

is sufficient signal.The RMS velocity field, at seismic sam-

pling, is multiplied with this skeleton; and in order to get the

final interval velocity attribute, the RMS to interval velocity

conversion is done using the irregularly sampled RMS veloci-

ty field. Because this is made consistent with the geology and

along coherent events, the vertical resolution is similar to the

seismic resolution. The accuracy of the interval velocity at this

resolution depends on the seismic noise level, which generally

increases with depth. It is difficult to estimate directly, but it

should be higher than with a standard Dix conversion.

For an example of the skeleton, see Figure 8. To be con-

sistent with the depth conversion decided upon during the

PSDM post-processing phase, this new final interval velocityattribute volume in time was converted to depth, using exact-

ly the same average velocity field for conversion of the PSDM

seismic imaged volume.

Further calibration and compaction trendA cross-plot of the new seismic interval velocities along the well

tracks against the well sonic velocities on the left in Figure 9

shows that in parts the scatter in values is still present. When

Figure 7 Composite PP results using multiple NCT.



© 2007 EAGE 45


we leave out wells E and D on the right, we can see that this

situation readily improves. Even though the correlation is

quite good, it was decided to test calibrating the seismic inter-

val velocity field to the well sonic velocities with the hope of

possibly improving further on the input to the pore pressure

prediction. For this process, the residuals between the seismic

and well velocities will be computed and kriged in 3D. These

residuals will then be added to the initial seismic velocity field.

This method was tested using a kriging radius of 5 km and 10

km. Well A was initially not used in the kriging to serve as a

‘blind well’ test and to provide a validity check for the kriging

parameters. The result at this well can be seen in Figure 10,

based on which it was decided to apply the kriging step with

a radius of 5 km.

Earlier, a calibrated Gardner relation was derived based on

the well data to estimate the density from the P-wave velocity

(Gardner et al., 1974). This relation is applied to the new cali-

brated interval velocity cube. Along the well tracks, this seismi-

cally derived density can be cross-plotted against the well den-sity logs (Figure 11), where the colour indicates the well. The

scattering was attributed to the well data quality, as the cali-

brated Gardner relation purely based on well logs gave a very

high correlation between density and velocity. Many observed

large differences between the bit-size and borehole-size logs

support this argument. This will influence both log measure-

ments. If there is a perfect prediction, all points would have

fallen along the red line. A first order regression inside the grey

polygon (excluding outliers) results in the best fit represent-

ed by the black line. The resulting equation, given in the same

figure, can be applied to the ‘seismic’ density cube to perform

a residual calibration. Subsequently the OBG can then be cal-

culated.

Figure 12 shows the correlation of the OBG: the blue

lines represent the OBG from calibrated seismic density at

the wells; the purple lines are the OBG curves from seismic

densities at the wells before this calibration step. It can be

seen that this final residual calibration is improving the fit

to the OBG from the original well densities (in red). Well D

was compared against the OBG from the composite density

function due to hole damage. This OBG volume, computed

using the density volume derived from the calibrated interval

velocity volume (through kriging) with the additional residu-

al density calibration step, was used in the final pressure vol-

ume calculations.

It was concluded in the initial well data analysis that a var-

ying NCT would be used, a shallow zone from the mud-lineto RS4 and a second zone below the RS4. In order to see the

lateral variations, the interval velocity attribute volume was

scanned to establish the spatial variation of the NCT curves

on a defined grid of 1 km x 1 km. To generate the NCT vol-

ume computations, these grid points will be interpolated to

cover the final seismic grid. This ‘scanning’ consists of extract-

ing the velocity trace at the selected locations and defining

the best lambda parameter for the Miller equation, so that a

Figure 8 Example of a derived skeleton for RMS to interval velocity conversion .



© 2007 EAGE46


Figure 10 Velocity kriging – test results at blind well A.

Figure 9 Seismic versus well velocity cross-plot QC.



© 2007 EAGE 47


Figure 12 OBG comparisons: well versus seismic versus calibrated seismic based.

Figure 11 Well density versus seismic density (calibrated Gardner) QC.



© 2007 EAGE48


Figure 14 Well vs. seismic based pressure predictions for older wells.

Figure 13 Derived NCT parameter maps at water bottom and RS4.



© 2007 EAGE 49


NCT curve is obtained that fits the respective velocity trace.

This is done for the velocity traces between the water bottom

and the RS4 horizon and downwards from the RS4 horizon.

The resulting lambda NCT parameter maps for the water bot-

tom and RS4 horizon are given in Figure 13. For both of these

zones, there is a distinct difference in the structurally higher

eastern area of the seismic data. For the RS4, one can clearly

see the crestal area along the centre.

Analysis and conclusionNow that the OBG and NCT are defined, the PP gradient and

the FG can easily be derived with the Miller pressure model. In

Figure 14 and 15 the results of the seismic and well pressure

prediction are displayed for a few selected old and new wells

(these wells were not used in the petrophysical calibration,

they became available at the end of the project). Good quality

well ties are obtained for the newest wells F and G between

the seismic pore pressure prediction and the sonic pore pres-

sure prediction. For the other wells, in general, the ties are fairor reasonable - at least the same trend is present in the sonic

PP and the seismic PP. This is proof that the velocity attribute

volume is accurate for the pressure estimation, which has

achieved good results in the two separate parts of Block 01

and Block AB.

An OBG, PP, and FG volume have been produced for the

survey, as well as a PP standard deviation cube to have an idea

of the uncertainty. For an example of depth slice see Figure

16 and for a profile near well location H, see Figure 17. Here

one can see structural influences on the pressure, as well as

detailed local variations. The sharp variation from high (red)

to lower (blue-green) pressures in the depth slice are related to

the change in compaction trend at the depth of the RS4 hori-

zon and the anticlinal-faulted structure. The proposed well H,

for example, goes just through a higher pressure zone. The

sharp variations here may well be due to compartments as well

as the structure.

The manual RMO analysis of the PSDM phase of the

project does not have sufficiently high frequency content to

be able to use it successfully in pore pressure prediction. This

RMO however provided a good input velocity model for the

dense automated velocity and eta analysis. Furthermore, on

the petrophysical model one can conclude that a spatially and

vertically varying NCT is required in this area to predict porepressures. Both resistivity and sonic data responded to pore

pressure in which case seismic velocities should show pressure

changes in shales. There are non-pressure related effects on the

resistivity logs. Normal pressure portions of the well, events

and MW logs are mostly used for pressure calibration. For use

Figure 15 Well vs. seismic based pressure predictions for new wells.



© 2007 EAGE50


with seismic data, the Gardner sonic to density transform and

the Miller velocity to effective stress transform are calibrated

using well data.

DiscussionIn the calibration of the pore pressure model for this area, it

is important to note that petrophysical models work in shales

and pressures are measured in sands. Shale and sand pressures

do not appear to be in equilibrium in the study area (concluded

through comprehensive investigation of all available well data

and geomechanical reports), which is required for calibration

of the model (in shales) to measured pressures (in sands). Also,

the measured pressures themselves are highly variable: struc-

ture, buoyancy, and depletion can influence the sand pressures.

The temperature differences in the Caspian Sea are expected to

have minimum impact on the pressures. As a result, it seems

that events and MW logs probably provide the most reliable

calibration data. The fact that the pressures are not in equi-

librium and there is a high variability in the pressures may

be linked to the compartmentalization. The des-equilibrium

means that the uncertainty in the predicted pressures increases.

The observation that the two newest wells, F and G, give good

ties is noteworthy. As these wells have not been used in the

calibration of the petrophysical model (available only towards

the end of the project), they serve as so-called ‘blind’ test wells.

These wells are the first wells drilled using synthetic oil-based

mud, and there is an uncertainty in the quality and origin of

the older well data. With respect to the seismic velocity model,

Figure 16 An example depth slice (4500 m) through the estimated PP volume.

Figure 17 An example PP section through proposed well location H.




it can be further stated that there is a high uncertainty in the

deeper section (~ > 5 km) in the velocities due to the offset

limitations. In the shallow zone (up to ~ 1 s, i.e. slightly more

than 1 km), due to the limited offsets, no reliable velocity pick-

ing is possible.

Despite all the difficulties, it seems that a successful attempt

has been made in predicting the pressures in this difficult area.

The main challenges are: (1) the pressure compartmentalisa-

tion, (2) non pressure related influences on resistivity logs, (3)

variable pressures, (4) sands and shales often not in pressure

equilibrium, and (5) an extremely difficult velocity and imag-

ing environment. It is still required to undertake real time pre-

diction of pore pressures while drilling. The current volumes

can be used as a reference. Due to the challenges listed and the

seismic limitations, there will always be an associated uncer-

tainty and error with respect to the pressure predictions. But

the best attempt has been made, and PSDM has increased the

accuracy and reliability of the velocity model. The pressure

model used can be updated when new well data becomes avail-able. Of vital importance and value is the combination of well

mechanical and operational knowledge and seismic process-

ing expertise.

AcknowledgementsWe would like to thank Petronas and Petronas Carigali for

their cooperation and allowing us to present these results.

Special gratitude to Nashrol Ariff Hussain and Peter Majid

in Petronas Carigali, Turkmenistan for their cooperation. In

addition, thanks are due to CGGVeritas and KSI for their sup-

port. Personal thanks from Norbert to Chris Manuel for the

assistance during the project.

References and suggested readingBell, D. [1998] Velocity Estimation for Pore Pressure Prediction,

Pressure Regimes. In Huffman, A. and Bowers, G. (Eds).

Sedimentary Basins and Their Prediction. AAPG Memoir 76,

Bowers, G.L. [1999] State of the Art in Pore Pressure

Estimation. DEA-119.

Bowers, G.L. [1994] Pore Pressure Estimation From Velocity

Data: Accounting for Overpressure Mechanisms Besides

Undercompaction. IADC/SPE 27488.

Bowers, G.L. [2001] Determining an Appropriate Pore-

Pressure Estimation Strategy. OTC 13042.

Bowers, G.L. [2002] Detecting high overpressure. The Leading

Edge, 21, 174-177.

Bruce, B., Bowers, G.L. [2002] Pore pressure terminology. The

Leading Edge, 21, 170-173.

Dix, C.H. [1955] Seismic Velocities from Surface Measurements.

Geophysics, 20, 68-66.

Dutta, N., Mukerji, T., Prasad, M., and Dvorkin, J. [2002]

Seismic Detection and Estimation of Overpressures Part II:

Field Applications. CSEG Recorder, September, 58-73.

Eaton, B.A. [1968] Fracture Gradient Prediction Techniques

and Their Application in Drilling, Stimulation and Secondary

Recovery Operations. SPE 2163.

Eaton, B.A. [1972] Graphical Method Predicts Geopressures

Worldwide, World Oil, May, 51-56.

Eaton, B.A. [1975] The equation for geopressure prediction

from well logs. SPE 5544.

Eaton, B.A., Eaton, T.L. [1997] Fracture Gradient Prediction

for the New Generation, World Oil, October, 93-100.

Gardner, G.H.F., Gardner, L.W., and Gregory, A.R. [1974]

Formation Velocity and Density – the Diagnostic Basis forStratigraphic Traps. Geophysics, 39, 6, 2085-2095.

Matthews, W.R. and Kelly, J. [1967] How to Predict Formation

Pressure and Fracture Gradient. Oil and Gas Journal , 20

February, 92-106.

Miller, T.W. [2002] A Velocity/Effective Stress Methodology

that Predicts Finite Sediment Velocities at Infinite Effective

Stresses. EAGE/SEG Summer Research Workshop, Galveston,

Texas.

Postler, D.P. [1997] Pressure Integrity Test Interpretation. SPE/

IADC 37589.

Ostermeier, R.M., Pelletier, J.H., Winkler, C.D., and Nicholson,

J.W. [2001] Trends in shallow sediment pore pressures -Deepwater Gulf of Mexico. SPE/IADC 67772.

Sayers, C.M. [2006] An introduction to velocity-based pore-

pressure estimation. The Leading Edge, 25, 1496-1500.

Siliqi, R., Le Meur, D., Gamar, F., Smith, L., Toue, J-P.,

Herrmann, P. [2003a] High density moveout parameter fields

V and . Part one: Simultaneous automatic picking, 73rd Ann.

Internat. Mtg., SEG Expanded Abstracts.

Siliqi, R., Le Meur, D., Gamar, F., Smith, L., Toue, J-P.,

Herrmann, P. [2003b] High-density moveout parameter fields

V and . Part Two: Simultaneous geostatistical filtering, 73rd

Ann. Internat. Mtg., SEG Expanded Abstracts.

Terzaghi, K. [1943] Theoretical Soil Mechanics, John Wiley &

Sons, Inc., New York, 1943.

Visit www.eage.org/jobsLooking for a new Challenge?

Surf to our website and have a

look at the latest Job openings!

technical large sept2007

Documents