Incorporating depth-dependency in quantitative interpretation through statistical rock physics

Sagar Ronghe and Anne Locke, DownUnder GeoSolutions

Introduction

Quantitative interpretation calibrates elastic properties (P-impedance, Vp/Vs) derived through

simultaneous inversion of seismic angle stacks to a rock physics model in order to characterise the
formation. The rock physics interpretation framework comprises probability density functions (PDFs)
that define the range of elastic properties associated with different lithology and fluid types. The
PDFs are often calculated from cross-plot samples of upscaled elastic well logs. Simultaneous
inversion results, sometimes from large vertical intervals, are then compared against the cross-plot
derived PDFs to predict the occurrence of different kinds of lithologies and fluids. This approach has
two fundamental problems: cross-plot samples may not be representative of the complete possible
range of elastic properties associated with the lithologies and fluids of interest; and the implications of
depth trends on elastic properties is ignored. In this paper we highlight a workflow that uses
statistical rock physics to understand the depth-dependent population behaviour of the rock types, and
incorporates depth dependency in the criteria for predicting lithology and fluid distributions from
seismic AVA and elastic property derivatives.

Building a rock physics interpretation framework using the extents of recorded or synthesised elastic
logs has significant limitations. Formation lithologies and fluids of interest may have been intersected
over narrow depths relative to the logged intervals. The length of acquired elastic logs may be short.
Elastic properties vary with depth, usually as a function of overburden pressure. Elastic log
measurements of lithologies and fluids are therefore only valid at the depths that they are encountered.
Changes in structural geology away from the wells may cause lithologies of interest to occur at depths
outside the logged intervals and have elastic properties that fall outside the logged ranges of values.
Using elastic log measurements of intersected lithologies and fluids to characterise formations outside
the logged intervals is inaccurate.

Wells may be few in number and preferentially drilled to intersect specific lithologies. Cross-plots of
wireline log samples may not be representative of the complete range of elastic properties possible for
the lithology at the intersected depth. Calibrating seismic derived elastic properties with well logs
needs to overcome differences in resolution. This is usually undertaken by filtering the well logs to
the bandwidth of the derived elastic properties before cross-plotting. Filtering reduces the density of
samples on the cross-plot. Using a filtered subset of samples risks establishing interpretation criteria
that do not appropriately reflect the properties of the formation of interest.

Statistical rock physics

Statistical rock physics provides an understanding of the population behaviour (the complete likely
range of elastic property responses) of lithology and fluid combinations as a function of end-member
rock types, fluid content, reservoir quality and depth. The workflow, described below, comprises
analysis (end-member picking and trending) and modelling (statistical sampling of the end-member
trend information). A depth-dependent interpretation framework can be created in elastic property
space and in interface reflectivity space (Lamont, et al., 2008, Thompson et al., 2011).

Statistical rock physics analysis involves picking end-member lithology types from logs and
establishing end-member trends (Figure 1). End-members are the cleanest logged examples of any
lithology defined using distinct elastic properties. End-member intervals (picks) are selected on logs
and their elastic properties are upscaled to form single values. The data points are cross-plotted, and
end-member trends are established for each lithology type. The end-member trends, along with their
standard deviation corridors, specify the relationships between the elastic properties, between elastic
properties and reservoir porosity, and show the effect of depth.

2nd EAGE Conference on Reservoir Geoscience

2 – 4 December 2019, Hanoi, Vietnam
Statistical rock physics modelling samples the rock property distributions defined by end-member
trends and associated standard deviations to derive large numbers of data points representative of the
population behaviour for each lithology and fluid mixture at any depth. Prior to sampling, reservoir
and non-reservoir trends are proportionally mixed to represent formation characteristics consistent
with petrophysical evaluations; and fluid mixture properties are obtained for these lithologies using
Gassmann fluid substitution. The statistically sampled data are cross-plotted for elastic properties
(e.g. P-impedance versus Vp/Vs) and are characterised using PDFs for each lithology and fluid
mixture at depth increments. The result is a depth-dependent interpretation framework in elastic
property space (Figure 2). Similarly, a depth-dependent interpretation framework can be established
in interface property space by sampling the contrasts between lithology and fluid mixtures for
reflectivity. Cross-plots are created for reflectivity (e.g. near [10 deg] vs far [30 deg] reflectivity) and
are summarised using PDFs (Figure 3). In reflectivity models, PDFs explain AVA behaviour. PDFs
will plot in different regions of near vs far reflectivity space for the different classes of AVA. PDFs
can span multiple classes. The PDFs enable an understanding of the range of seismic AVA responses
possible from an interface given the statistical rock physics model.

Figure 1 End-member picks and trends. (a) Two intervals representing different end-member picks
are highlighted on well logs in orange and purple. (b) An end-member pick is upscaled to form a
single point on cross-plots. End-member pick distributions are used to define end-member trends and
standard deviations for each end-member lithology.

Discussion

Depth trends affect interpretation criteria in both elastic property and interface reflectivity space. For
the dataset displayed in Figure 2, a number of changes can be observed with increasing depth: the
properties of the individual PDFs change; the size of the PDFs reduce due to decrease in reflectivity;
and the associations between different PDFs vary. At 700 m TVDBML the Brine Sandstone PDF is
completely overlain by the Claystone PDF, whereas the PDFs associated with the hydrocarbon sands
show little overlap with the other PDFs. In contrast, at 1,900 m TVDBML, the Brine Sandstone PDF
shows increased separation from the Claystone PDF and hydrocarbon sand PDFs show increased
overlap with the other two. In between these depths is a continuum of change. This demonstrates that
the range of elastic properties that can classify a particular lithology or fluid type at a depth are invalid
in classifying the same lithology or fluid at a different depth.

Similar effects are seen in interface reflectivity space. For the dataset in Figure 3, at 1,000 m
TVDBML, the PDF representative of the Claystone / Gas Sand interface has a strong Class II or Class
III AVA signature. This transitions to subtle Class III to Class IV between 3,000 and 4,000 m
TVDBML. At 6,000 m TVDBML, the response is a strong Class IV. Note also how the associations
between the PDFs change with depth and the implications of decreasing reflectivity with depth. It is
therefore important not just to know the likely range of AVA responses associated with the top of a
gas sand, but also the depths at which these responses are expected. It is crucially important to

2nd EAGE Conference on Reservoir Geoscience

2 – 4 December 2019, Hanoi, Vietnam
incorporate and apply depth dependency in the interpretation criteria for both simultaneous inversion
results as well as seismic AVA.

Figure 2 A depth-dependent elastic property interpretation framework is displayed as P-impedance

versus Vp/Vs cross-plots at various depths, each overlain by PDFs for mixtures of lithology and fluid.
Each PDF is displayed using a mean point and a 2 standard deviation contour of the property
distribution. The PDFs change in properties, size and associations with increasing depth.

Reflectivity @ 1000 mTVDBML Reflectivity @ 2000 mTVDBML Reflectivity @ 3000 mTVDBML

Claystone
___ ____ ___
Brine
Claystone
___ ____ ___
Claystone
Class I

Class II
Claystone
__ ___ ___ ______ _ Class II
Low Sat. Gas
(Sg = 10%) Class III
Claystone
__ ___ ___ __
Gas
(Sg = 70%) Class II / III

Reflectivity @ 4000 mTVDBML Reflectivity @ 5000 mTVDBML Reflectivity @ 6000 mTVDBML

Claystone
___ ____ ___
Claystone

Claystone
___ ____ ___
Brine

Claystone
___ ____ ___ _____
Low Sat. Gas
(Sg = 1 0% )

Claystone
___ ____ ___
Gas
(Sg = 60%)

Class IV

Figure 3 A depth-dependent interface reflectivity interpretation framework is displayed as Near (10

deg) versus Far (30 deg) cross-plots at various depths, each overlain by PDFs for interface contrasts
between lithology and fluid mixtures. PDFs change AVA character, size and associations with
increasing depth.

2nd EAGE Conference on Reservoir Geoscience

2 – 4 December 2019, Hanoi, Vietnam
The statistical rock physics workflow overcomes limitations of traditional rock physics approaches.
Depth trends can be extrapolated outside the logged intervals if appropriate, with due consideration to
changes in structure and stratigraphy. This means that, unlike well log cross-plot based methods, it is
possible to derive interpretation criteria outside the wells extents, with some caveats. Also, the
differences in bandwidth between well and seismic data can be addressed by mixing end-member
lithology and fluid trends in proportions representative of the resolution of the seismic data. The
resulting modelled PDFs retain the population behaviour of the end-member lithologies and are in
agreement with seismic. The loss of information inherent in cross-plotting frequency filtered logs to
bridge the resolution gap between well and seismic data does not feature in a statistical rock physics
model.

Conclusion

Cross-plotting of filtered well logs to define the framework for quantitative interpretation of AVA and
inversion products is not appropriate as it reduces data density and ignores depth dependency.
Statistical rock physics overcomes these limitations by defining the population behaviour of
lithologies and fluids as a function of depth. The rock physics model can be extended and
interpretation criteria established outside the logged intervals. Lithology and fluid trends can be
mixed to define PDFs at seismic resolution. This does not require any data decimation that loses
information. Interpretation of simultaneous inversion results and seismic AVA must incorporate
depth dependency.

Acknowledgements

Thank you to DownUnder GeoSolutions for permission to present this work.

References

Lamont, M.G., Thompson, T.A., and Bevilaqua, C. [2008] Drilling success as a result of probabilistic
lithology and fluid prediction: a case study in the Carnarvon Basin, W.A. APPEA Journal, 48(1).

Thompson, T., Lamont, M., Bevilacqua, C., and Hendrick, N. [2011] Fit for Purpose Seismic
Reservoir Characterisation. Petroleum Geology Conference and Exhibition, Kuala Lumpur, Extended
Abstracts.

2nd EAGE Conference on Reservoir Geoscience

2 – 4 December 2019, Hanoi, Vietnam