Robust Workflows for Seismic Reservoir

LUNCHEON
Characterisation
Patrick Connolly, SEG Distinguished Lecture
BP, London

In the talk I describe a workflow for seismic reservoir charac-
terisation. The specific problem I address is the estimation of
net-to-gross from a clastic reservoir, however the workflow is
adaptable for a variety of other applications.
The three main components of the workflow are:
Vs and density, with associated reflectivity equal to the
gradient term in the 2-term AVO equation. Extended elastic
impedance (EEI) (Whitcombe et al, 2002) is a coordinate rota-
tion in AIGI space (using log-log axes). An EEI curve is para-
meterised by the rotation angle, chi.

1. Extended elastic impedance analysis followed by 2-term

AVO coordinate rotation of the seismic intercept and
gradient datasets.

2. Coloured inversion.

3. Attribute extraction followed by calibration and

detuning.

This workflow has evolved within BP over the past decade

or so. The component processes are all fundamentally very
simple; they all operate within the seismic bandwidth
(other than the initial well data analysis) and the number of
parameters required to implement them is small. This leads
to a very robust workflow; the subjectivity of the parameter
selection is minimised so the results tend to be repeatable.

Extended Elastic Impedance

Extended elastic impedance is based on the gradient imped-
ance/acoustic impedance crossplot domain (Whitcombe &
Fletcher, 2001). Gradient impedance (GI) is a function of Vp, Figure 1. Elastic parameter projections in AIGI space.

Patrick Connolly graduated from Birmingham University with a BSc in physics in 1977 and joined
Seismograph Service Ltd as a data processor working in Oman, New Zealand, Pakistan and London. He
moved to Britoil in Glasgow 1982 and worked as a seismic programmer until the company was bought
by BP in 1989.

He has remained with BP working as a seismic analyst and occasional interpreter in Aberdeen, Houston
and London. He has been a member of a number of exploration and appraisal teams working in the West
of Shetland, Gulf of Mexico and offshore Angola. Since 2001 he has worked in BP’s E&P Technology divi-
sion where his current role is Manager, Seismic Reservoir Characterisation and Surveillance R&D.

Patrick was awarded the SEG Virgil Kauffman Gold Medal in 2001 for his development of elastic impedance technology.
He was an EAGE distinguished lecturer in 2007.

★ ★ ★ ★ ★ ★

APRIL LUNCHEON MAY LUNCHEONS

DATE: April 26, 2010
May 10, 2010
TIME: 11:30 A.M. Lunch
GeoCanada Luncheon: Ray Zahab
LOCATION: Telus Convention Centre, Calgary
TICKETS: Contact CSEG office May 12, 2010
TELEPHONE: 403-262-0015 or Fax: 403-262-7383 GeoCanada Luncheon: Brent Butt

Continued on Page 8

April 2010 CSEG RECORDER 7

Luncheon Cont’d
Robust Workflows…
Continued from Page 7
The chi angle can be selected to optimise the correlation of the fractal curves have power law amplitude spectra (Turcotte, 1997)
EEI curve with a target reservoir parameter such as VShale or and hence, because we’ve seen that impedance logs can be made
porosity or with an elastic parameter such as bulk or shear to correlate with gamma ray curves, they too will have power
modulus, Poisson’s ratio or shear impedance. law spectra (Stefani & De, 2001).

The equivalent process can then be performed on seismic data by

combining intercept and gradient datasets using the same coor-
dinate rotation weightings (Hendrickson, 1999). In practice for a
number of reasons the optimal seismic angle will probably be
different from the optimum angle calculated from the log data so
an additional optimisation process is required. The result is a
seismic dataset with an optimum correlation with a desired
parameter which can then form the basis for a reservoir charac-
terisation workflow.

It has been recognised for some time that combinations of AVO

parameters can provide useful approximations to a number of
elastic parameters (Dong, 1996, Goodway et al, 1997). Any
simple combination of intercept and gradient is equivalent to a
coordinate rotation (figure 1). By performing this rotation in the
impedance domain in intercept-gradient space the correlation
coefficient between the EEI log and elastic parameter provides a
quantitative assessment of the quality of the approximation.
These approximations are often very good.
Figure 3. Bed thickness distributions showing a power law relationship over much
This approach can also be used to demonstrate the equivalence of the range.
of different AVO domains. For example Poisson’s Ratio and
shear impedance typically have high correlations at specific rota- Coloured inversion (Lancaster & Whitcombe, 2000, Lancaster &
tion angles within the AIGI domain implying that AIPR, AISI Connolly, 2007) is the process to transform zero phase seismic
and AIGI domains are just stretched and squeezed versions of data of arbitrary (but stationary) amplitude spectrum to band-
each other. limited impedance having a power-law spectrum limited by the
bandwidth of the seismic. This process will result in optimal
resolution within a given bandwidth. It is also a very simple
Coloured inversion process requiring only an estimate of the power-law exponential
As much as we’d like to from calibration well data and optimisation of the bandwidth.
be able to invert our
seismic data to absolute Seismic net-to-gross
impedance in many
circumstances the addi- We can think of net-to-gross as being the ratio of the value of
tional low frequency data an average measurement, such as gamma ray, across the reser-
required to do this is voir to the value of that measurement for the 100% net case
unreliable. A safer option
is to base seismic charac-
terisation on band-limited
impedance. To transform
seismic to band-limited
impedance (BLI) we need
to understand the charac-
teristics of impedance
amplitude spectra.

Geology has long been

recognised as being, at
not at least partially, scale
invariant; photographs of
outcrops require some-
thing to provide scale
(figure 2). Mathematically Figure 2. Outcrop (without scale).
this means that bed thick-
ness distributions (figure 3) tend to follow a power law relation-
ship (Talling, 2001) with the implication that say gamma ray
curves are fractal. The importance of this for geophysicists is that Figure 4. Cross-plot of average band-limited impedance against gross thickness for
a turbidite reservoir.

Continued on Page 10

8 CSEG RECORDER April 2010

Luncheon Cont’d
Robust Workflows…
Continued from Page 8
(all measurements relative to the non-net baseline). If impedance
correlates with GR the process will also work using an
impedance log.
That concept can be extended to band-limited impedance. The
100% net value will now depend on gross thickness because of
tuning effects but this can be modelled if we know the wavelet.
The scaling of the tuning curve (figure 4) can be deduced either
from geological insight (for example that the maximum net-to-
gross of a turbidite reservoir is likely to be 100%) or by cali-
brating to well data (Connolly, 2007).
Seismic data that has been optimised to correlate with a gamma
ray curve by using EEI analysis and then transformed to band-
limited impedance by coloured inversion can therefore provide a
basis for net-to-gross estimation by a fairly simple detuning and
calibration process. Provided apparent thickness is always used as
the basis for the net-to-gross estimate the process remains consis-
tent and works for sub-tuning as well as thicker reservoirs the
maximum thickness being limited by the lack of low frequencies.
Robustness and uncertainty
The three processes in the workflow are all fundamentally
simple. They each require a small number of parameters, which
have fairly well defined methods of optimisation. This helps to
make the workflow robust and repeatable. Minimising the
subjectivity puts us in a position to consider making quantitative
estimates of the uncertainty of our characterisation results; to put
error bars on our reservoir predictions.
To do this we must consider the sources of uncertainty. I propose
grouping these into three categories; geological uncertainty, cali-
bration uncertainty and seismic uncertainty.
Geological uncertainty is caused by the sub-seismic scale vari-
ability in the rock properties and of the fine-scale layering. We
can only estimate the impact of these effects statistically. One
approach is to use pseudo-wells; a large number of geologically
realistic synthetic wells constrained by calibration data. These
can provide a statistical relationship between the seismic
attribute and the required reservoir parameter (Connolly &
Kemper, 2007).
Calibration uncertainty arises from the process of detuning and
calibrating the seismic attributes. The level of uncertainty can be
directly estimated and is dependent on the quality of the cali-
bration data and the reliability and consistency of the wavelet
estimation.
Seismic uncertainty is the inherent limited fidelity of the seismic
amplitudes caused by the inability of the processing to fully
GeoCanada 2010 Conference:
(CSEG's Annual Joint Convention)
“Working with the Earth”
May 10-14, Calgary Alberta
www.geocanada2010.ca <http://www.geocanada2010.ca>
10 CSEG RECORDER April 2010
correct for all the transmission effects, to remove all the noise and
to perfectly image the data. Estimating the amplitude uncer-
tainty is difficult but an indication of the level can be obtained by
comparing independent surveys at the same location.
In principle each of these sources of uncertainty can be inde-
pendently estimated and summed to provide an uncertainty
map; in the example discussed this will be the standard devia-
tion of the seismic net-to-gross. A capability to routinely do this
would provide numerous benefits;
1. A better understanding of the relative size of different
sources of uncertainty potentially leading to improvements
in characterisation processes.
2. An ability to put meaningful errors bars on reservoir predic-
tions.
3. The use of uncertainty maps to constrain reservoir model-
ling to ensure that models are consistent, but not driven by,
the seismic data.
More fundamentally an ability to estimate uncertainty will allow
us to better integrate different types of data and hence develop
more accurate models of the subsurface. R
References
Connolly, P., 2007 (October), A simple, robust algorithm for seismic net pay estimation,
The Leading Edge.
Connolly, P. and Kemper M., 2007 (October), Statistical Estimation of Reservoir
Characterisation Uncertainty, The Leading Edge.
Dong, W, 1996: A sensitive combination of AVO slope and intercept for hydrocarbon indi-
cation: EAGE, Amsterdam.
Goodway, B., Chen, T., and Downton, J., 1997, Improved AVO fluid detection and
lithology discrimination using Lamé petrophysical parameters; “lr”, ”mr”, ”lm fluid
stack”, from P and S inversions. SEG Extended Abstracts.
Hendrickson, J.S., 1999, Stacked; Geophysical Prospecting, 47, 663-705.
Lancaster, S. and Connolly, P., 2007, Fractal layering as a model for Coloured Inversion
and Blueing, EAGE Extended Abstracts.
Lancaster, S. and Whitcombe, D.N., 2000, Fast-track ‘coloured’ inversion, SEG
Extended Abstracts.
Stefani, J. P. and De, G. S., 2001, Power-law behaviour of well log spectra, EAGE/SEG
Research Workshop on Reservoir Rocks, Extended Abstracts.
Talling, P.J., 2001. On the frequency distribution of turbidite thickness. Sedimentology,
48.
Turcotte, D. 1997, Fractals and Chaos in Geology and Geophysics, Cambridge
University Press.
Whitcombe, D.N. and Fletcher, J.G. 2001, The AIGI crossplot as an aid to AVO analysis
and calibration, SEG Extended Abstracts.
Whitcombe, D.N., Connolly, P.A., Reagan, R.L. and Redshaw, T.C., 2002, Extended
elastic impedance for fluid and lithology prediction: Geophysics, Vol 67.
IMPORTANT GEOCANADA DATES:
April 29, 2010 Online Registration ends
Short Course Registration ends
May 7, 2010 Exhibitor Registration closes