Professional Documents
Culture Documents
BRIAN RUSSELL and DAN HAMPSON, Hampson-Russell Software Ltd., Calgary, Alberta
JIM SCHUELKE and JOHN QUIREIN, Mobil E&P Technical Center, Dallas, Texas
broad term that encompasses all geo- the seismic trace itself, which Data Data
statistical methods that utilize more indicate changes in re f l e c t i o n
than one attribute to predict some coefficient size. For example, if
physical property of the earth. Bivari- the acoustic impedance shows a Log Attribute
ate geostatistics is obviously the sim- dramatic change, as in a gas sand, Properties
plest subset of the multivariate tech- the seismic trace will display the
niques and thus the standard classic “bright spot”.
technique of cokriging could be called 2) Extraction of acoustic impedance Statistical
multivariate geostatistics. However, from the seismic data by trace Relationships
in this paper we will use the name to inversion. If we simply integrate
refer to geostatistical methods that the trace, the result will lack a
use more than two variables. Even low frequency component due to Calibration
then, there are many different meth- the effect of the seismic wavelet.
ods that fall under this heading. There This is usually corrected by
are three major subcategories: adding the low frequency com-
ponent from a velocity model. Residual
Correction
1) The extension of cokriging to (Schultz et al. object to calling the
include more than one secondary result a true seismic attribute,
attribute to predict the primary since it has already been influ-
parameter. enced by the well log data). Interpretation
2) Methods that use the covariance 3) The use of prestack data to extract
matrix to predict the parameter information about intercept and
Figure 1. The data-driven statistical
from a linearly weighted sum of gradient, and hence Poisson’s
interpretation (after Schultz et al.).
the input attributes. ratio or shear wave reflectivity,
3) Methods that use artificial neural using the formula:
networks (ANNs) or nonlinear dφ(t)/dt = the instantaneous fre-
optimization techniques to com- R(θ) = R P+G sin2θ, quency.
bine attributes into an estimate These are the three primary
of the desired parameter. where: R P = intercept; G = gradi- attributes, but many more can be
ent; R S ≈(RP - G)/2; and ∆σ ≈ (RP derived from the basic three.
This paper will consider only + G)*4/9. 5) Attributes which are derived spa-
techniques described under the sec- In the above formulation, sever- tially from multiple traces, such
ond point. Point (3) is discussed in a al approximations have been as coherency.
series of papers in TLE Schultz et al, made. We have assumed that
(1994). For a discussion of methods Vp/Vs ≈ 1/2 and have dropped Although relationships have been
in point (1), refer to Multivariate Geo - the third term in the A k i - inferred between these attributes and
statistics by Wackernagel (Springer- Richards formula. reservoir parameters, the physical
Verlag, 1995). 4) The use of instantaneous attrib- basis is not always clear, and we may
utes derived from seismic data. want to derive statistical, rather than
Using seismic attributes to estimate These attributes are based on the deterministic, relationships. This
log properties. The idea of using definition of the complex trace, approach, which Schultz et al. call a
multiple seismic attributes to predict which, in polar form, give us the data-driven methodology, is summa-
log properties was first proposed by three classical seismic attributes: rized in Figure 1.
Schultz et al. in their series of TLE the amplitude envelope, instan- In the most general case, we look
articles. They point out that the tra- taneous phase, and instantaneous for a function that will convert m
ditional approach to deriving reser- frequency. Mathematically: different attributes into the desired
voir parameters from seismic data property. This may be written:
has been to look for a physical rela- C(t) = s(t) + jh(t), P(x,y,z) = F[A1(x,y,z,),....,
tionship between the parameter to Am(x,y,z)] (1)
be mapped and some attribute of the where C(t) = complex trace; s(t) =
seismic data, and then use that single A(t)cosφ(t) = the seismic trace; where P = the property as a function
attribute over a 2-D line or 3-D h(t) = A(t)sinφ(t) = the Hilbert of coordinates x, y, z; F = the func-
volume to predict the reservoir para- transform trace; A(t) =(s 2 ( t ) + tional relationship; and Ai, i= 1, ... , m
meter. This could be termed deter- h2(t))1/2 = the amplitude envelope; = the m attributes.
ministic reservoir parameter predic- φ(t) = tan-1(h(t)/s(t)) = the instan- The simplest possible case would
tion. Examples include: taneous phase; and ω(t) = be a linearly weighted sum:
=
1 1 1 1 1 1 1
amplitude envelope, and instanta- w2 ∑ E1 ∑ E I ∑ E ∑ E F ∑ E φ
2
neous frequency. In symbols: 1 1 1 1 1 1 1
φ(x,y) = w 0 + w1I(x,y) +
w3 ∑ F1 ∑ F I ∑ F E ∑ F ∑ Fφ
1 1 1 1 1
2
1 1
(5)
φ1 = w0 + w1I1 + w2 E1 + w3 F1
φ2 = w0 + w1I2 + w2 E2 + w3 FN
M M M M M
φN =w0 + w1IN + w2 EN + w3 FN
σ Aσ P
where σAP = covariance between A
and P; σA = standard deviation of A;
σP = standard deviation of P; A is the
attribute; and P is the parameter.
For the N attributes, the correla-
tion coefficients can be ordered from
best to worst. However, it should
be noted that when we determine
the best combination of attributes
to use in the final sum, the order of
the attributes used may not be the
same as the order of their correla-
tion coefficients. This may seem
strange at first, but consider a sim-
ple example. One of the attributes
could just be a linearly scaled ver-
sion of another attribute, and it
would obviously have just as good
a correlation coefficient. But adding
Figure 4. Crossplot of seismic trace against sonic log. this attribute to the sum would not
improve the fit.
The quality of the fit can be deter-
mined by finding the RMS error
between the known parameter and
the estimated parameter. By using
this criterion, we can find the opti-
mal combination of attributes to use.
number that should not be exceeded) diction that has been in use for so
and other techniques (such as ANNs, many years.
the prediction can get even closer.
Obviously, there are many other References. P. S. Schultz, S. Ronen,
applications of the technique that we M. Hattori, and C. Corbett, 1994,
have just shown. We are preparing a “Seismic guided estimation of log
sequel to this paper in which we will properties, Parts 1, 2, and 3”: (TLE,
show a 3-D case study. It is our feel- 1994). Hans Wackernagel, 1995, Mul -
ing that the introduction of geosta- tivariate Geostatistics: Springer-Ver-
tistical techniques to our industry lag, (1995). LE
gives us a powerful tool which will
challenge the deterministic model of Corresponding author: Brian Russell, 403-
seismic processing and lithology pre- 266-3225; brian@hampson-russell.com