Anda di halaman 1dari 4

Density estimation from the sonic log: A case study

James P. DiSiena* and Fred J. Hilterman, Geophysical Development Corporation


Summary
In this case study, we estimate the bulk densities which would
be measured by the density log in a well. We base this
estimate on the sonic log, a derived lithology log and
methods based on
density trend curves. Two
Gardner et
(1974) relationship and an alternate approach
that utilizes an
trend analysis are evaluated. In comparison with the observed density, the Gardners relationship
underpredicts the shale density and overpredicts the sand
density. A modification of Gardners equation (Castagna et
al.,
which utilizes different coefficients for each
lithology, produces a better estimate. However, the results
vary from well to well. A local data base within our study area
provides an empirical calibration to improve upon the
Gardner-type relationships for this area.
Approximately 1000 square miles with 50 wells make up our
study area in offshore Louisiana, centered on South Marsh
Island Block 106. These logs constitute a local data base for
determining trends in velocity and density for a two component lithology of sand and shale. We identify a linear relationship between the density and logarithm of velocity for
both sand and shale. Mixing the sand and shale relationships based on their volume lithologic fractions, we arrive at
In comparison to the modified
our density estimate.
Gardners method, a comparable to better estimate of the
densities is obtained. Furthermore, the linear relationship allows for easy fine-tuning of the local density prediction. If a
portion of the well has a density log, we can calibrate the
relationships for the remainder of the well. These results
show a remarkable fit to the density curve, with errors of less
than two percent. When discrepancies are evident, the
predicted curve can be used to edit other logs or to indicate
the presence of gas.
Introduction
Density can play a very important role in seismic modeling.
This statement especially holds true when trying to match
amplitudes and seismic character as in the case of seismic
thology analysis. In practice, amplitude versus offset (AVO)
analysis in the Gulf of Mexico requires calibration by modeling
sequences that contain hydrocarbons. An extensive
data base of digital well logs across the offshore Gulf of
Mexico (over 3000 wells) provides the input to our modeling
and analysis of rock-property trends. When analyzing local
study areas within this data base, one finds depth intervals on
the order of a thousand feet where the contrast in velocity
between sands and shales becomes very small. For these
areas, the density contrast between sands and shales contributes a major portion of the seismic response.
Unfortunately in the Gulf of Mexico, density log data remain
less available than sonic logs. Even when available, the density log is often run over just a small portion of the well.
Furthermore, drilling conditions can have an adverse effect on
the density log, especially in porous gas sands. To proceed

SL2.3

with modeling for these cases, we need methods to estimate


the density from other logs. The effectiveness of the
estimations will be based on analyzing the similarity with the
observed density log, along with comparing AVO model
results.
Approximately 1000 square miles with 50 wells make up our
study area in offshore Louisiana, centered on South Marsh
Island Block 106. Logs from these wells constitute a local
data base for determining trends in velocity and density, for
sands and shale. Our study incorporates individual analysis
on many of these wells and on several wells neighboring this
area. To describe our methodology, we will focus in on the
Transco A2 well in Block 107, identified by sequence # 6054.
Basically, other logs in the well are used to estimate the
density log. First, published relationships between velocity
and density provide a quick means to estimate the density.
Then, we evaluate the adequacy of these relationships within
our study area. In follow-up to this evaluation, we developed
an alternate method for density estimation based on our local
trend analysis.
Gardners Density
Gardner et
(1974) relationship can transform sonic log
velocity into density:
density (z),
V velocity (z),
The inadequacy of this relationship for our area becomes
evident when comparing a synthetic seismogram generated
with the observed density log (Figure 1) to one with Gardners
estimate (Figure 2). A ray-trace approach without transmission loss was used to generate the
gather
(Hilterman, 1990). Shown in the left-most panel are the
moved-out traces from near offset (left) to the far offset
(right). The four panels of repeated traces display:
0.23 V

Stack
D Sigma
Sigma-

Stack of the gather,


Normal incidence term from AVO inversion,
term related to Poissons ratio contrast and
AVO indicator, NI multiplied by the
term.
Density Estimate From Gardners

6054

Density estimation from the sonic log

By using a mixing relationship based on volume fraction of


the lithology, the composite density becomes:
(4)
is the sandstone volume fraction of the rock (1 .O to 0).

Comparing the models, a major difference can be seer


around 2.1 to 2.3 seconds (7000 to 8000 ft). In this regior
the observed density produces a series of reflections not
evident on the other model. Upon further analysis and
comparisons to other density estimation methods, the cause
of this difference becomes evident. Gardners relationship is
based on an average rock type and does not account for
lithology variations. Figure 3 shows velocity and density
trends, sorted by sand and shale, averaged over 20 ms intervals, and plotted in two-way time. In this region, the velocity
shows little to no contrast. between sand and shale. However,
the observed density shows a substantial variation between
sand and shale of almost 0.2 g/cc.

Figure 4 shows the logs and compares the various approaches for density estimation approaches with the
observed density log from well sequence #6054. Initially, we
will focus on Gardners density estimate. From left to right,
the log tracks show the sand volume lithology curve, the sonic
log in velocity units, and the observed density log in the well.
The next track overlays the observed density curve (light line)
with the classic Gardner density estimate (bold lines).
Similarly, the next track shows Castagnas modification of
Gardners density relationship. As noted by others, the
classic Gardners relationship overestimates the density for
sandstone and underestimates the density for shale. The
modified Gardners relationship, based on lithology, tracks the
variations in the observed density. The modification reduces
the density of the sands, which gives a better estimate at the
sandstone intervals. As for the shale values, they are only
slightly improved (increased). As others suggest (Castagna
et al., 1993), both relationships apply over a wide range of
rocks and conditions, therefore local calibration could improve
these estimates.

Castagna et al. (1993) suggested the use of different coefficients for each lithology to better fit Gardners data and to
account for differences between lithologies in the velocitydensity relations. Their equations, in a form similar to the
classic Gardners relationship, with velocity V (ft/s) and
density (g/cc) are:
(2)
(3)

Trend Analysis
We can assemble trend curves for velocity and density from
the subset of wells in the data base which contain a density
and/or sonic log along with a predicted sand percentage
curve.

1097

Density estimation from the sonic log

From this trend analysis, we can evaluate velocity-density


relationships. For example, to predict the exponential form of
the equivalent Gardners equation for this area, we crossplot
the logarithm of the density trend against the logarithm of the
sonic trend. For this area, the relationship did not appear
linear. After evaluating other functional relationships in
crossplots, one relationship stood out. Namely, a strong
linear relationship appears when density is plotted against the
logarithm of the velocity. Figure 6 plots Gardners equation in
the classic linear velocity-density form. On the same graph,
our semi-log relationship for sand and shale is overplotted on
the velocity and density trend analysis data points.
The ideal situation would be to compute this linear relationship for rock of 100% sand and for 100% shale. In this
fashion, we can combine the end-member relationships to
estimate a composite rock density that is a fraction of sand
with shale. So, to refine these relationships for predicting the
density from the sonic, a cut off of >75% sand and ~25%
sand is used for the input trend curves. This provided
sufficient data points for the statistical analysis, while providing better discrimination between the sand and shale
trends. Both of the trends show excellent fits with correlation
coefficients of greater than 0.99.
From the cross-plots, we obtain a relationship for the shale
and sand densities as a function of velocity:
+ 1.61
shale

Figure 5 shows four trends developed from this data base,


with values averaged over 500 ft depth intervals and plotted
as a function of depth for sand and shale, namely the velocity
(Figure
and the density (Figure 5b). For these trends,
values with a sand percentage
contribute to the sand
trends, while those values with a sand percentage of
Note, the trend curves show
make-up the shale trends.
scatter at depths approaching 9000 ft as some of the wells
begin to respond to geopressure.

sand

0.8 1 + 0.759

In a similar fashion to the application of the modified


Gardners equation for mixed lithologies, we use a simple
linear mixing relationship for adding the two component
system of sand and shale:
sand

While the velocity trends for sands and shale appear quite
similar, the density trend shows a systematic increase in
going from sands-to-shale. As the individual well showed,
this implies that lithology has a strong influence on density in
this area.

1098

shale.

Density estimation from the sonic log

The actual velocity (ft/s) of the sonic log at a given depth is


substituted into equations (5) and (6).
We applied this method of density prediction to well 6054. In
comparison to the classic Gardners or the modified
Gardners, we obtain a much better estimate of the densities.
Overall there appears to be a similar systematic bias of
underestimating the shales and overestimating the sand
density (Figure 4, second track from the right). We can
attribute this to several factors, one of which is the 75%-25%
sand/shale cut-off in the trend analysis. Unfortunately, raising
the cut-off would limit our ability to predict reliable trends as
the number of data points in the logs especially for the clean
sands are dropped. By adjusting the relationships to compensate for the 75% sand/shale cut-off, a revised estimate of
the density can be obtained. In this well, the calibration
amounted to subtracting 0.05 from the sand density equation
(5) and adding it to the shale density equation (6). This
amount was consistent in size with the extrapolation of the
variations between the 50% and 75% trend curves to an
effective 100% sand cut-off. This calibration appears to be
constant over the entire portion of the well analyzed. However, this calibration changes slightly from well to well in the
study.

Observations and Conclusion


To further demonstrate and evaluate these density estimation
schemes, we applied them to other wells in this study area
and to neighboring wells outside the area. Consistently, the
Gardners relationship produced the poorest results. As
designed, the modified Gardners equation did decrease the
sand densities and increase the shale densities, often with
good results. Overall, the trend analysis method proved to be
at least as good as the modified Gardners relationship. With
calibration, the trend analysis method typically produced the
best match to the observed density in the well. Moreover, the
best application for this method arises when a portion of the
well did have a density log. This portion, even if small, can
be enough to calibrate the trend analysis density estimate
over the remainder of the well.
A key source of systematic bias for both methods that rely on
lithology is the estimate of the sand percentage curve from
the SP and/or gamma ray curve for the log of interest. As for
impacting the trend analysis, any bias in the sand percentage
curve should average out and not be significant given the
large population of wells used to develop the trends. Other
sources of potential difficulties occur when applying these
methods across geopressured zones or varying age rocks
(e.g., across major growth fault systems). For these areas,
both the trend curve and the application at the individual well
could require further analysis.

The results after calibration (Figure 4, right track) show a


remarkable fit to the density curve, with errors of less than
2%. When discrepancies are evident, the predicted curve
provides information that can be used in editing or evaluating
the logs. In particular, the zones between 7872 to 7912 and
from 8100 to 8160 ft show major differences. These gas
zones have a low density that both Gardners and the trendanalysis method does not properly estimate. In the modified
Gardners equation, we would have to introduce a porosity
term and incorporate the fluid density. Similarly, the trend
curves used in the trend-analysis method were dominated by
wet sands. A separate correction needs to be applied to
account for gas. Fortunately, the gas sands are the zones
that typically have density logs.

References
Castagna, J. P., Batzle, M. L., and Kan, T. K., 1993, Rock
Physics -- the link between rock properties and AVO
response, in
Castagna, J. P. and Backus, M. M., ed., Offset-dependent
reflectivity-- theory and practice of AVO analysis, Soc. Expl.
Geophys.
Gardner, G. H. F., Gardner, L. W., and Gregory, A. R., 1974,
Formation velocity and density -- the diagnostic basics for
stratigraphic traps: Geophysics, 39, 770-780.

The synthetic data for the trend-analysis density estimate


(Figure 7) shows good agreement with the observed density
log model (Figure 1). This holds true even over the laminated
sand-shale region and over the gas zone. While the density
dominates the model at the laminated region, the velocity
seems to dominate at the gas zone with the density contrast
only contributing slightly.

Hilterman, F., 1990, Is AVO the seismic signature of


lithology? A case history of Ship Shoal - South addition,
Leading Edge, 9 no. 6, 15-22.

1099

Anda mungkin juga menyukai