Professional Documents
Culture Documents
3UDFWLFDOVHLVPLFSHWURSK\VLFV7KHHIIHFWLYHXVHRIORJGDWD
IRUVHLVPLFDQDO\VLV
TAD M. SMITH, Apache Corporation
Downloaded 03/01/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
A quick scan of the SEG web site shows that the phrase
“seismic petrophysics” has been used explicitly in the
title of a paper or abstract six times, the earliest of which was
with the justification that many of these edits are below seismic
resolution, and are therefore not important. In many instances
this may well indeed be true, but given that we are in search of
by Williams et al. in 1996 (“The Hugoton cross-well survey; ever more subtle seismic responses, it is my opinion that these
A direct look at stratigraphy, seismic petrophysics and shale edits should be performed with care anytime we are generating
anisotropy”). However, the first attempt at a definition and an log-based seismic models or using log data to interpret seismic
expanded description of “seismic petrophysics” was published responses. Failure to properly edit log data can lead to erroneous
by Wayne Pennington in The Leading Edge in 1997 (“Seismic assumptions and expectations from seismic amplitudes.
petrophysics: An applied science for reservoir geophysics”).
He defines seismic petrophysics as follows: “…the purposeful Identification of spurious log data
application of rock physics theory, as calibrated by laboratory Well log data can be faulty for any number of reasons, the most
and well measurements, to the interpretation of seismic data…” common of which are poor wellbore conditions (e.g., washouts)
The discussion of “seismic petrophysics” in Pennington’s pa- and problems with the measurements (e.g., cycle skips on sonic
per, as well as most subsequent uses of the phrase in the pub- data). Examination of caliper log data is the most common tech-
lished literature, focused largely on what today is simply referred nique for identifying areas in the wellbore that potentially con-
to as “rock physics.” The 2004 paper by Kittridge et al. (“Seis- tain bad data. This is especially relevant for density data, as the
mic petrophysics for clean sandstones: Integrated interrogation density tool is a pad type of device and needs to be in contact
of lab- and well-based data for improved rock physics model- with the wellbore wall (Figure 1).
ing”) specifically mentions the importance of fully integrating It is important to recognize that not all wellbore washout will
geological and petrophysical data into the rock physics workflow, result in bad data. Conversely, lack of washout in the wellbore
although this is also implied in the paper by Williams et al. This
is an important aspect of seismic petrophysics that I subscribe
to; that is, the need to understand the underlying geology and
petrophysical properties of the reservoir (and the bounding lay-
ers) prior to developing any rock physics or AVO models. An
important corollary is that rock physics modeling should not be
considered a substitute for careful analysis of log or core measure-
ments. Application of the wrong rock physics model may allow
in-situ velocities to be predicted, but it will not yield any insights
into the underlying controls on the elastic properties of the sub-
surface rocks and, worse, may lead to incorrect forward models.
This could lead to potentially misleading and expensive mistakes,
as I will attempt to demonstrate throughout this document.
Because of the breadth and multidisciplinary nature of seis-
mic petrophysics, it is not possible in an overview paper to cover
in any detail all the various topics that could be discussed. In-
stead, I have decided to focus on four aspects of seismic petro-
physics that I consider to be the “workhorses” of the discipline:
(1) log editing and conditioning, (2) shear velocity prediction,
(3) fluid replacement modeling, and (4) porosity modeling.
While others may disagree with my assessment of what consti-
tutes the “workhorses” of seismic petrophysics, it is my opinion Figure 1. Washout effects on density. Caliper data are often the
that flawed application of these four fundamental processes will best indicator of potentially bad data in well logs. In this example,
compromise all subsequent rock physics and AVO modeling. the density excursion to low values across the washout zone is not
supported by evaluation of other curves (e.g., deep resistivity). Also
Data editing and conditioning notice the deviation of the density response from the shale background
trend at a depth of approximately 7200 ft. It is important to recognize
Careful conditioning of well-log data prior to the application that not all wellbore washout will result in low-quality density data.
of a rock physics or geophysical model is an often overlooked, Notice that zone A has some washout as indicated by the caliper
or over-simplified, aspect of the rock physics workflow. Editing log (approximately 1–1.5 inches of washout), yet the density data
frequently requires depth shifting, estimation of pseudodata to are consistent with the density trend in this well and for the area.
Conversely, analysis of the caliper log across zone B shows over six
replace bad log data, and invasion corrections (if necessary). Un- inches of washout. The low densities across zone B will affect the
fortunately, geophysicists often oversimplify the editing process, modeled AVO response, and should be replaced with pseudodata.
does not guarantee that the log data are useable. It is therefore
also useful to examine multiwell depth-trend plots and crossplots
for identification of anomalous data (Figure 2a and 2b). In most
subregions within a basin, multiwell trend-plot analysis will
show relatively consistent sonic and density behavior with depth.
Large deviations from this expected trend can therefore be useful
Downloaded 03/01/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
Generation of pseudodata
Numerous empirical transforms are available in the published
literature to assist with the generation of pseudo density or ve-
locity data (shear-wave prediction will be discussed in a later
section). For density, commonly used models include those of
Gardner et al. (1974), Castagna et al. (1991), and Wang (2000).
For sonic data, some examples of empirical models include those
by Wyllie et al. (1956), Raymer et al. (1980), Han (1986), Eb-
erhart-Phillips (1989), and many others (also see Avseth et al.,
2007). It is important to recognize that all published coefficients
are lithology specific, and do not apply on a global scale. Figure 2. Effective use of trend plots for identifying bad density data.
Although these empirical transforms provide useful guides (a) Evaluation of multiwell trend plots is an effective technique for
for understanding velocity and density behavior, they generally quickly identifying spurious data (sonic and density). In this example,
do a poor job of predicting velocity and density logs for the pur- note the anomalously high density values from approximately 5000–
5100 ft. These high values could be due to an unexpected lithology
pose of seismic analysis (Figure 3). The most rapid and robust (e.g., carbonates); yet comparison to other logs in the same well (e.g.,
approach for log prediction, in my experience, is via multilinear resistivity) and examination of mudlog data show no unusual rock
regression (MLR; Figure 4). During the MLR training process it types. (b) Examination of the caliper and DRHO data also indicate
is important that the user not incorporate bad data, or data that that there was an unexplained problem with the density measurement
have been affected by near wellbore damage. This applies to both across this zone (mudballs?). In this particular example, the sonic data
may also be affected by poor borehole conditions (track 7). Failure to
the curve being predicted, as well as to the curves being used for examine the density data within the framework of a multiwell analysis
the prediction. could lead the user to miss this problem area, potentially leading to
The logs that have the highest correlation with velocity typi- problems with seismic well ties.
cally are the following:
4) Gamma Ray (or some indication of lithology; e.g., Vclay)
1) Neutron porosity
2) Resistivity The order in which the above logs will best correlate with
3) Density velocity or density will vary, in part, on the degree of consolida-
4) Gamma ray (or some indication of lithology; e.g., Vclay) tion in the subsurface and the composition of the rock. It is also
important to recognize that MLR is best applied over zones of
The logs that have the highest correlation with density typi- limited depth range in order to avoid compaction effects.
cally are the following:
Shear-velocity prediction
1) Neutron porosity (if available) It is common practice in rock physics to estimate a shear log for
2) Velocity the purposes of AVO modeling. The need to estimate a shear
3) Resistivity velocity is often necessary because no measured shear velocity
Figure 3. Application of empirical trends. It is common practice to apply various empirical trends to estimate density and sonic logs. In this plot,
density is calculated using Gardner et al. (1974) and Castagna et al. (1991). Neither model adequately predicts density, highlighting the risk
of uncalibrated application of these models. It is possible, however, to calibrate the “Gardner coefficients” on a local basis to better estimate density.
is available in the well, or because the quality of the measured than the shear-wave velocity predicted using the coefficients of
shear-wave data is too low for modeling purposes. It is also use- Greenberg and Castagna (1992); most of this large deviation oc-
ful to estimate a shear velocity for the purposes of evaluating the curs in the shale, whereas the sandstone shear velocities closely
quality of a measured shear log, as well as guiding the editing conform to the predicted values (Figure 5b). Although these
process of a measured shear log. large deviations are not direct evidence that the data are in error,
such large deviations typically do not occur with high quality
Recognizing bad data shear-velocity data. As a general approximation, if the measured
There are some simple techniques for assessing the quality of a shear velocity deviates from the estimated shear log by more
measured shear log, the most effective of which is an examina- than 15%, the data quality should be carefully assessed.
tion of the measured data for anomalously low or high VP/VS The quality of an S-wave log is probably best assessed by ex-
ratios (Figures 5 and 6); negative Poisson’s ratios (VP/VS < 1.414) amination of VDL displays of stacked semblance plots, if avail-
are clearly in error and need to be corrected (Figure 6). In many able (Figure 7). This provides the user with perhaps the most
instances, comparison of the measured shear data to an esti- robust tool for recognizing spurious shear data. If raw waveform
mated shear log can provide useful guidance on shear-log qual- data are available, it is possible (desirable) to reprocess the data in
ity. For example, examination of the shear-wave data in Figure an attempt to improve the results.
5 shows that the measured shear log is upwards of 25% slower
Figure 8. VP/VS crossplots. For both (a) and (b), the blue and green
lines are computed for sands and shales using the Greenberg and
Castagna coefficients. (a) Three data sets are posted on this crossplot
to illustrate some pitfalls associated with application of Greenberg
and Castagna coefficients. As long as the rocks consist primarily of
clean quartz sands (e.g., the Han samples; these are the 10 samples
in this data set with no clay) and clay-rich shales, the application of
Greenberg and Castagna coefficients is often an effective predictor
of shear-wave velocity and VP/VS ratios. If, however, the rocks are
Figure 7. Anomalously high Poisson’s ratios are often observed from composed of minerals other than quartz, Greenberg and Castagna
P- and S-wave sonic data (also see Figure 5). (a) Note the high PR will typically predict VP/VSratios that are too low. The glass beads of
values at approximately 13,100 ft. Occasionally, this can be due to Berge et al. (blue circles) are included on this plot simply to illustrate
unusual lithologies (e.g., ash beds and some coals). The high PR values the important effect of composition on the resultant VP/VS ratios. For
in this example are not due to an unusual lithology, but due to a weak velocities less than 3.5 km/s, Greenberg and Castagna would predict
flexural wave across this interval and the automatic tracking of a later VP/VS ratios that are close to the measured values. For velocities greater
event. (b) In the absence of the raw waveforms, it would be difficult to than approximately 4 km/s, Greenberg and Castagna would predict
determine the reason for the PR values observed across this zone. VP/VS ratios that are too low. Although glass beads are not a naturally
occurring material, they are useful for illustrating the important effect
of composition on VP/VS ratios. Similarly, application of sandstone
coefficients to a limestone interval will generate VP/VSs ratios that are
potentially have an large impact on the computed AVO gradient far too low. Failure to understand the composition of the rock can
response. lead to the prediction of the wrong VP/VS ratios, and consequently
Equation 1 becomes a powerful tool for generating a pseudo misleading AVO models. Finally, for poorly consolidated sediments
Kdry during the fluid substitution process. If a reliable shear mod- (VP < ~2.5 km/s), Greenberg and Castagna predict will predict VP/
VS ratios that are too low. The unconsolidated sands included on this
ulus can be computed from measured data, or if one can be rea- crossplot are from synthetic sand mixtures, with velocities measured at
sonably estimated (from a pseudo shear velocity), then a pseudo various effective stress states (data courtesy of Manika Prasad, Colorado
Kdry can be computed by simple rearrangement of Equation 1: School of Mines). Vernik et al. have proposed a modified model to
account for these low-velocity rocks (a). (b) Various heuristic models
can also be used to better understand the fundamental controls on VP/
VS ratios (see text for discussion). In this plot, quartz sand models
are generated by combining Gassmann’s equation with Krief et al.
This equation can be combined with Gassmann’s equation to (1991) and Nur et al. (1995). For pure quartz, both models predict
solve some of the difficult problems frequently encountered dur- VP/VS ratios that are lower than those predicted using Greenberg
ing the fluid substitution process (shaly sands, invasion correc- and Castagna. The higher VP/VS ratios predicted by Greenberg and
tions, and negative PR values). In order to do this, the user must Castagna may be due to the presence of brine-saturated microcracks in
the rock matrix.
Figure 10. AVO modeling results from the well shown in Figure
10. (a) AI versus PR crossplot. Application of the estimated shear
velocity results in a larger contrast in PR than that calculated from
the measured data. (b) Half-space model using data from (a). Note
that the AVO gradient response is enhanced when using the estimated
shear log. (c) Offset synthetic response. Note the following: (1) the AVO
gradient response is enhanced when using the estimated shear-velocity
log, and (2) at the seismic bandwidth, the AVO response is complex,
and not representative of the underlying rock properties. Rock property
plots (a and b) would predict a class I to IIP response, whereas an
apparent class III response is observed on the offset synthetic models (c).
fold: (1) most sands typically contain minerals other than quartz,
and (2) most framework grains contain microstructural defects
(i.e., cracks) that will lower the effective bulk modulus and shear
modulus of the mineral matrix.
A key element of the approach outlined above is the neces- Figure 12. Simple mixing model (Voigt-Reuss-Hill) for dry frame
Poisson’s ratio, assuming that the PRdry = PRmatrix. Note that the smaller
sity for understanding the underlying composition of the rock. the amount of quartz in the system, the higher the dry frame PR. Some
Applying the properties of quartz to a nonquartz reservoir may minerals, such as plagioclase feldspar, have petrophysical properties
result in fluid substitutions (or estimated shear logs) that over- very similar to quartz, and are therefore difficult to recognize from
estimate the predicted AVO effects (Figure 9). conventional logging suites. Note, however, that PRdry for a sandstone
with abundant plagioclase feldspar will be considerably higher than
Porosity modeling that for a pure quartz sandstone. It should not be expected that all
reservoir rocks have PRdry values of 0.12 and lower.
Porosity modeling in sedimentary rocks is usually done via ap-
plication of empirical or heuristic models (Han, 1986; Eberhart-
Phillips, 1989; also see Avseth et al., 2007 and 2010 for more a between porosity and velocity in sedimentary rocks, due in
detail discussions). Although these are useful approaches, they large part to variable pore geometries as well as the presence of
may not accurately describe the underlying velocity-porosity highly variable concentrations of microcracks in the rock ma-
systematics for many reservoirs, especially over large ranges in trix (Smith et al., 2010). In addition, for many reservoir rocks,
porosity. This is because there is often no unique relationship changes in porosity often correspond to changes in clay content
Figure 13. Frame property model and predicted P-wave velocity using Equation 2 (see text). In this approach Kdry is calculated using Equation
2, which is then used in Gassmann’s equation to compute a P-wave log for the in-situ conditions (red curve in track 4). Note the close match
between the measured and predicted log. Application of Equation 2 is an effective technique for correcting difficult problems during the fluid
substitution workflow.
and/or the amount and types of diagenetic cements and diage- will extrapolate back to some value representative of wet shale or
netic alteration. Consequently, application of empirical or heu- clay at 0% porosity, and not a P-wave velocity of pure quartz.
ristic porosity models may impose on the reservoir a velocity- On the other side of the spectrum, velocity behavior in low-
porosity relationship that does not exist in the subsurface. Prior porosity rocks (porosity < 10–12%) is often dominated by the
to application of any porosity model, the controls on porosity presence of variable concentrations of microcracks in the rock
behavior should be assessed and utilized in the development of matrix (Smith et al., 2010; Ruiz and Cheng, 2010). As a result,
the porosity model. In addition, examination of local well data, velocity-porosity relationships can be poorly defined in these sys-
as well as geological considerations, should also provide insights tems, with the porosity having only a negligible effect on the P-
into the range over which porosity is modeled. wave velocity (Smith et al., 2010a and 2010b). In these systems,
For many poorly consolidated sand and shale sequences, velocities often extrapolate to values lower than those predicted
the P-wave velocity for shale is similar to the P-wave velocity for by the rock matrix at 0% porosity.
brine-saturated sandstones (e.g., Smith and Sondergeld, 2001). Regardless of the reservoir being modeled, it is important
In these scenarios, porosity degradation often occurs simply by to build an understanding of the local controls on velocity, and
the replacement of sandstone layers with shale. As a result, there the causes for porosity degradation or enhancement. This is pos-
are frequently only small changes in P-wave velocity with changes sible only when core data are available and can be analyzed. For
in porosity (Figure 15a). This is analogous to the “sorting” trend the more common scenario when only log data are available, I
of Avseth et al. (2010). When modeling velocity and porosity in find that the most effective means for evaluating the underlying
such systems, it is important to recognize that P-wave velocities controls on velocities is to examine the manner in which Kdry
Figure 15. Velocity systematics. (a) P-wave velocity versus porosity. Note that there is only a small change in P-wave velocity with changing
porosity. Constant Vclay lines are from Eberhart-Phillips (1989). Examination of these data implies that porosity changes are due to changes in silt
and clay content (e.g., the “sorting trend” of Avseth et al, 2010). Also note that extrapolating to 0% porosity does not result in the matrix velocity
for quartz. Failure to account for the important role of clay (more specifically, composition) in the porosity-velocity systematics would lead to
application of the wrong porosity-velocity model, and potentially misleading AVO models. (b) Dry frame bulk modulus and shear modulus versus
porosity. Note that the Kdry/μ ratio increases with decreasing porosity. This is consistent with increasing clay content as porosity decreases. Direct
observation of variations in the dry frame properties with porosity is an effective means for better understanding velocity-porosity systematics,
and for constructing porosity models that are representative of the reservoir in the subsurface. Perhaps most importantly, the ability to use a rock
physics model to match the measured velocities does not necessarily imply that it is representative of the underlying controls on velocities.
A constant theme that exists throughout this “opinion piece” pretation: Cambridge University Press.
is the need for the geophysical community (and, for that matter, Avseth, P., T. Mukerji, G. Mavko, and J. Dvorkin, 2011, Rock-physics di-
the petrophysical community) to rigorously incorporate knowl- agnostics of depositional texture, diagenetic alterations, and reservoir
edge of the underlying controls on the elastic response of the res- heterogeneity in high-porosity siliciclastic sediments and rocks—A
review of selected models and suggested work flows: Geophysics, 75,
ervoir and its bounding layers. As we search for ever more subtle
no. 5, 7531–7547.
AVO responses, it is important that we take the time to fully un- Berge, P. A., B. P. Bonner, and J. G. Berryman, 1995, Ultrasonic velocity-
derstand the underlying geological and petrophysical properties porosity relationships for sandstone analogs made from fused glass
of the reservoir, and not assume that our sands are dominated beads: Geophysics, 60, no. 1, 108–119, doi:10.1190/1.1443738.
by quartz, or that they conform to some preferred rock physics Castagna, J. P., M. L. Batzle, and T. K. Kan, 1993, Rock physics-the link
model. The application of rock physics models in the absence of between rock properties and AVO response, in J. Castagna, and M.
this important calibration step can be misleading, and can lead M. Backus, eds., Offset-dependent reflectivity—theory and practice
to potentially expensive mistakes. of AVO analysis: SEG., 131–171.
Dvorkin, J., D. Moos, J. Packwood, and A. Nur, 1999, Identifying
References patchy saturation from well logs: Geophysics, 64, no. 1, 1–5, doi:
Avseth, P., T. Mukerji, and G. Mavko, 2005, Quantitative seismic inter- 10.1190/1.1444681.
Eberhart-Phillips, D., D.-H. Han, and M. D. Zoback, 1989, Empiri- Smith, T. M., C. M. Sayers, and C. H. Sondergeld, 2010, Rock properties
cal relationships among seismic velocity, effective pressure, poros- in low-porosity/low permeability sandstones: The Leading Edge, 28,
ity, and clay content in sandstone: Geophysics, 54, no. 1, 82–89, no. 1, 48–59, doi:10.1190/1.3064146.
doi:10.1190/1.1442580. Smith, T., C. Sondergeld, and T. Ali Ousseini, 2010, Microstructural
Gardner, G. H. F., L. W. Gardner, and A. R. Gregory, 1974, Formation controls on electric and acoustic properties in tight gas sandstones;
velocity and density—the diagnostic basics for stratigraphic traps: some empirical data and observations: The Leading Edge, 29, no. 12,
Downloaded 03/01/13 to 141.117.79.62. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/