You are on page 1of 5

Investigating Thomas-Stieber model for property estimation of thin-bedded shaly-sand

Downloaded 03/21/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

reservoirs
Piyapa Dejtrakulwong*, Stanford Rock Physics Laboratory, Stanford University
Tapan Mukerji, Stanford Center for Reservoir Forecasting, Stanford University
Gary Mavko, Stanford Rock Physics Laboratory, Stanford University

Summary possible to deterministically infer fractions and property of


sand within the mixing.
This paper focuses on investigating the Thomas-Stieber
model and its application for evaluating proportions of One assumption made in this model is that there are only
dispersed clay in shaly sands in thin laminated sand-shale two types of rocks alternating within the lamination. In
reservoirs. The goal of this study is to incorporate natural other words, the interpretation is limited only to mixing
variations and uncertainties which can be found in real between sand and shale or dirty sand and shale. However,
observations into the model and to point out some pitfalls laminations consisting of only two alternating elements
in the interpretation when these variations are neglected. may not always be present. For example, both the sand and
For a case of bimodal mixing (i.e. sand and pure shale), this shale properties within the interval of interest can vary due
model can be used to infer sand fraction and its porosity. to variations in amount of clay content present in each.
However, when the sand beds within the interval of interest Dawson et al. (2008) characterized shale samples from
consists of a range of properties (e.g. clay content), the submarine fan in deep-water settings and reported different
model cannot deterministically estimate property of each types of shale due to compositions, fabric and sealing
individual sand bed. Instead, it gives an average property properties. Furthermore, in real applications one needs to
of all the sand beds within the stack. The uncertainties of estimate the two endpoint properties which are not certain.
the input parameters of the model can be used in stochastic It is essential to incorporate these variations into the
methods to quantify uncertainty of our interpretation. We Thomas-Stieber model in order to quantify uncertainty in
use Monte Carlo simulations propagated through the our interpretation. In this paper, we explore various mixing
Thomas-Stieber model to estimate the posterior scenarios and stochastically apply the model to synthetic
distributions of interpreted sand fractions in a Bayesian examples.
framework.
Thomas-Stieber model
Introduction
For simplicity, this model is based on an assumption that
We use geophysical measurements at various scales to infer the main factor which decreases sand porosity is shale
the properties and structure of the subsurface. These content.
measured data often represent average properties of
multiple thin layers when heterogeneities are below the
resolution of the measurement tools. The lack of
knowledge about the existence and properties of thin
laminations could mislead the interpretation and thus affect
our evaluation of the reservoir. For example, a direct
application of isotropic Gassmann’s equation without
accounting for the sub-resolution sand-shale lamination
would be erroneous due to the fact that fluid substitution is
likely to be applicable only in permeable sand beds
(Katahara, 2004; Skelt, 2004).

Thomas and Stieber (1975) provided a simple model


exploring how porosity could vary with shale volume
depending on the configuration of shale in the sand-shale
sequences (e.g. laminated, dispersed or structural). Thus,
by inputting properties of the “pure” sand and shale end-
members this model can be used to infer the shale
distribution within the rock and therefore provide evidence Figure 1: An illustration of shale configuration in
of fine-scaled lamination. In some simple cases, it is also sand. Marked letters (A to D) correspond to points on
the ternary diagram in Figure 2.

SEG Houston 2009 International Exposition and Annual Meeting 1965


Property estimation of thin-bedded reservoirs

Three different ways that shale can be distributed in sand segment AE represents lamination between clean sand and
include laminated, dispersed, and structural (Figure 1). pure shale with varying Vlam in the mixture. Line segment
Downloaded 03/21/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

AD represents the dispersed mode of mixing. Total


Notations and relations between total porosity and the bulk porosity continues to decrease with increasing volume of
volume of shale (Vsh) added to the mixture can be dispersed shale until it reaches point D, where the original
established as follows: sand pore space is completely filled with shale. Thus, the
1. Clean sand and pure shale porosities are ϕs and ϕsh other line segments (BE, CE and DE) correspond to
respectively. lamination between dirty sand and shale. This model is
2. When shale is dispersed into the original sand pore derived initially using gamma ray responses (Thomas and
space without disturbing the sand pack, total porosity Stieber, 1975). This ternary diagram is also equivalent to
linearly decreases with Vsh. The relation follows ϕT = ϕs an ideal bimodal mixing (i.e. sand grains and clay particles)
– (1 – Vsh) ϕsh . (Marion, 1990). Given point F, one can infer that the point
3. When Vsh = ϕs , the original pore space of sand is represents lamination between 60% dirty sand and 40%
completely filled with shale. At this point, ϕT = ϕs ϕsh . shale. The “dirty” sand has total porosity of 13% which is
4. If we continue to add shale into the mixture, it is equivalent to volume of dispersed shale (Vdis) of 0.2.
equivalent to replacing voidless sand grains by shale
with porosity. As a result, total porosity linearly Examples
increases with Vsh. In other words, in the region Vsh >
ϕs , we have ϕT = Vsh ϕsh . In this section, we show various scenarios of mixing and
the results of property estimation using the Thomas-Stieber
The formulation following (1) to (4) completes the model. First, we construct vertical sequences by randomly
dispersed-mixing category. We will specifically refer to drawing either a sand or shale layer to insert into the
Vsh as Vdis in this case. sequence. Then, each layer is assigned a value for Vclay,
specified in Table 1 for each case. Using the Thomas-
5. In the lamination domain, the total porosity is simply a Stieber model, we obtain the corresponding porosity.
weighted average of porosity of all the end-members. These values are considered to be the fine-scale properties.
For clean sand laminated with pure shale, the relation is We then upscale both Vclay and porosity using a 15-point
the following: ϕT = (1 – Vlam) ϕs + Vlam ϕsh , where moving average to obtain data points. If the inference from
Vlam is the bulk volume of shale within the lamination. the model works, the results for the simulated data should
show an exact match to the up-scaled true values. Table 1
The derived mathematical model can be graphically shown shows the parameters used for each scenario. In all cases,
as a ternary diagram (Figure 2). This diagram can be we use the same ternary diagram both for generating and
constructed using the properties of clean sand and pure inferring the properties.
shale end-members as input parameters.
Case True property in the Input for the model
lamination
Vclay of sand Vclay of Porosity Porosity
shale at at
Vsh=0 Vsh=1
1 0.1 1 0.4 0.6
2 0.1 0.8 0.4 0.6
3 N(0.1,0.0025) 1 0.4 0.6
4 N(0.1,0.0025) 0.8 0.4 0.6
5 N(0.1,0.0025) N(0.8,0.0025) 0.4 0.6
6 N(0.1,0.0025) 0.8 0.4 0.6
N(0.2,0.0025)
Table 1: Parameter settings for each case. In cases 3 – 6, the
notation N(x,y) stands for normal distribution with mean x and
variance y.
Figure 2: A graphical display of Thomas-Stieber model relating
The results of case 1 and 2 are shown in Figure 3. In case
total porosity and Vsh.
1, we see an exact match of the estimated and the true
values as expected. However, when the properties of the
Points A and E correspond to clean sand and pure shale
shale lamination change (i.e. silty shale instead of pure
points whose Vsh values are 0 and 1 respectively. The line
shale), our sand fraction estimation could be overestimated,

SEG Houston 2009 International Exposition and Annual Meeting 1966


Property estimation of thin-bedded reservoirs

as shown in case 2. This example shows the potential values. However, we need to keep in mind that these
pitfall of misinterpreting the result as it appears to have a properties have already been up-scaled (i.e. averaged over
Downloaded 03/21/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

larger amount of dirtier sand than it actually does. several sand layers). Therefore, in this case we can only
obtain an “average” property of the sand in the lamination.
If the shale properties within the lamination do not
correspond to the pure shale (case 4) it further complicates
the situation and the results show even larger scatter than
observed in case 2. When the shale properties also vary
between layers (as in case 5), the relation between the true
and estimated sand fraction is not simple as other cases.

Uncertainty from the input parameters

So far, we have not included the uncertainties of the input


Figure 3: Comparison between true and estimated sand properties from parameters (i.e. the sand and shale end-members) into the
case 1 and 2. model. In this section, we perform forward modeling using
Monte Carlo simulations and present results in forms of
posterior distributions of the sand property inferred from
the Thomas-Stieber model for a given data point.

Using the decomposition of conditional probability, we


have the following relation:
P ( A | D ) = ∫ P ( A, E | D )dE = ∫ P ( A | E , D) P ( E | D )dE , [1]
where A is the property of interest (e.g. sand fractions)
D is a given data point
Figure 4: Comparison between true and estimated sand properties from E represents the two end-members’ properties.
case 3 and 4.
For simplicity, we assume that P(E | D) is known. For any
data point acquired, we have some ideas about its location,
geological setting, depth, etc. Thus, we can use this
information to constrain and estimate the distribution of the
endpoints’ properties. It means both distributions (but not
the specific values) of porosity of clean sand and pure shale
are known and, in our case, both are assumed to be
Gaussian and independent.
Figure 5: Comparison between true and estimated sand properties from
case 5.
We first separately draw the two endpoints from Gaussian
distributions. In the example shown here, we assume the
porosity distribution of the shale has larger uncertainty than
the sand. Then, we use the Thomas-Stieber model to infer
the values of sand fractions and Vdis in the sand for a given
data point. Drawing multiple realizations, we obtain
distributions of sand fractions and Vdis for that particular
data point (Figure 7). Results are shown in Figures 8 and 9.
We observe that the distributions of the estimated sand
fractions for all three data points have a similar shape with
narrow spreads. However, the distribution of the estimated
Figure 6: Comparison between true and estimated sand properties from Vdis at point 3 appears to have larger uncertainty than
case 6. those of the other two points. A factor that contributes to
the larger uncertainty in the result at point 3 is its location
on the ternary plot. Point 3 appears to be closer to the shale
endpoint which is assumed to be more uncertain than the
From Figure 4, we see that in case 3 we can still obtain the
other endpoint. Therefore, by the nature or geometric of
correct sand fractions. Furthermore, the estimated sand
this ternary diagram, the estimated Vdis distribution at
property at each sampling point also matches the true

SEG Houston 2009 International Exposition and Annual Meeting 1967


Property estimation of thin-bedded reservoirs

point 3 is greatly influenced by the variation of the input


parameter.
Downloaded 03/21/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Conclusion

The Thomas Stieber model can be used in evaluating thin-


bedded shaly-sand reservoirs whose bed thicknesses are too
thin to be resolved by conventional tools. This study aims
to integrate natural variations and uncertainties into the
model in order to quantify uncertainty in our interpretation
of the thin-bedded reservoirs and to reduce risk. We show
several mixing scenarios and compare the estimation using
the model with the true values. It is clear that a small
variation in the sand and shale properties can lead to
misinterpretation which will greatly affect our evaluation of
the thin-bedded reservoirs. The uncertainties in the input
parameters for Thomas-Stieber model should also be
considered. We show examples of posterior distributions
of properties using Monte Carlo simulations.
Figure 7: Example of multiple realizations of the ternary
In real application, this type of analysis will help us to diagrams generated from a set of two endpoints. Three data
points are considered.
better understand how the natural variations and
uncertainty in data or input parameters propagate through
models and ultimately to help us quantify uncertainty in our
reservoir evaluation.

Acknowledgements

This work was supported by the Stanford Rock Physics and


Borehole Geophysics project and the Stanford Center for
Reservoir Forecasting.

Figure 8: Posterior distributions for estimated sand fractions of


the three data points shown in Figure 7.

Figure 9: Posterior distributions for estimated volume of


dispersed shale in the sand of the three data points shown in
Figure 7.

SEG Houston 2009 International Exposition and Annual Meeting 1968


EDITED REFERENCES
Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2009
SEG Technical Program Expanded Abstracts have been copy edited so that references provided with the online metadata for
each paper will achieve a high degree of linking to cited sources that appear on the Web.
Downloaded 03/21/15 to 169.230.243.252. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

REFERENCES
Dawson, W. C., W. R. Almon, K. Dempster, and S. J. Sutton, 2008, Shale variability in deep-marine depositional systems:
Implications for seal character – subsurface and outcrop examples: American Association of Petroleum Geologists, Search
and Discovery Article 50128, http://www.searchanddiscovery.com/documents/2008/08144dawson/ndx_dawson.pdf,
accessed 3 March 2009.
Katahara, K., 2004, Fluid substitution in laminated shaly sands: 74th Annual International Meeting, SEG, Expanded Abstracts,
1718–1721.
Marion, D., 1990, Acoustical, mechanical and transport properties of sediments and granular materials: Ph.D. Thesis, Stanford
University.
Skelt, C., 2004, Fluid substitution in laminated sands: The Leading Edge, 23, 485–488.
Thomas, E. C., and S. J. Stieber, 1975, The distribution of shale in sandstones and its effect upon porosity: 16th Annual Logging
Symposium, SPWLA, Paper T.

SEG Houston 2009 International Exposition and Annual Meeting 1969

You might also like