Spwla 2012 195

SPWLA 53rd Annual Logging Symposium, June 16-20, 2012
MULTISCALE WORKFLOW FOR RESERVOIR SIMULATION
Downloaded from http://onepetro.org/SPWLAALS/proceedings-pdf/SPWLA12/All-SPWLA12/SPWLA-2012-195/1610152/spwla-2012-195.pdf/1 by China Petroleum & Chemical Corp (SINOPEC) user on 08 February 2023
Neil F. Hurley, Weishu Zhao, and Tuanfeng Zhang, Schlumberger
Copyright 2012, held jointly by the Society of Petrophysicists and Well Log Analysts (SPWLA) and the submitting authors
This paper was prepared for presentation at the SPWLA 53rd Annual Logging Symposium held in Cartagena, Colombia, June 16-20, 2012.
__________________________________________________________________
ABSTRACT
Digital rock models can be used to perform multiscale simulations in heterogeneous rocks. Models range in size
from pore (nm to micron) to borehole (mm to m) to interwell (10‟s to 100‟s of m) to full-field scale (10‟s to 100‟s of
km). At each scale, petrophysical and dynamic properties are calculated from digital rock models at the adjacent
smaller scale, then used to populate simulations at the next larger scale. This process, a form of upscaling, converts
fine-scale models to coarse simulation grids while preserving key rock properties at all scales.
Digital rock models of this study employ multi-point statistics (MPS) and representative element volumes
(REV‟s). MPS is a geostatistical modeling approach that creates simulations of spatial geological and reservoir
property fields, conditioned to honor known results, such as those measured in wellbores or rock samples. REV‟s
are the smallest volumes that can be modeled to capture rock heterogeneity, within specified limits of variance of
the measured property, such as porosity or permeability.
Core plugs for pore-scale modeling are strategically chosen using whole-core minipermeability grids and
conventional CT scans. Pore-scale models employ tools such as transmitted laser scanning fluorescent microscopy,
microCTscans, and focused ion beam-scanning electron microscopy to image pores in 3D. Analyses must show that
an REV has been sampled for each petrophysical facies before segmentation into binary images should occur. MPS
is used to create realistic models of arbitrary size and shape. Pore models are directly used for flow modeling, or
converted to pore-network models. Computed numerical SCAL (Special Core AnaLysis) properties are validated
with laboratory-derived data, and used to populate borehole-scale models.
Borehole-scale models combine gridded minipermeability maps and conventional CTscans of whole core with
electrical borehole images to create 3D numerical cores. SCAL properties determined from pore-scale models are
distributed for each petrophysical facies in numerical cores. Once again, analyses should show that an REV has been
sampled before flow modeling can occur. Effective SCAL properties computed from various MPS realizations are
used to populate interwell-scale models.
At the interwell scale, seismic attributes and variogram statistics from logging while drilling (LWD) are used to
populate digital rock models. Effective properties computed from flow simulations in grid blocks are used to
populate full-field scale models. At the full-field scale, outcrop analogs, sequence stratigraphy, forward stratigraphic
models, diagenetic models, and basin-scale models are combined to improve flow simulations.
INTRODUCTION
Reservoir modeling is the process of building a representation of a reservoir that incorporates all characteristics that
pertain to its ability to store and produce hydrocarbons. Reservoir models are subdivided into static and dynamic
components. Static models are fine-scale simulations of rock properties, such as porosity, permeability, capillary
pressure, fractures, faults, seismic attributes, and parameters that do not change significantly with time. Dynamic
models are coarser simulations that incorporate fluid dynamic properties that change with time, such as pressure and
flow rates of oil, gas, and water.
Reservoir modeling covers at least 14 orders of magnitude (Figure 1), ranging from pore (nm to micron) to
borehole (mm to m) to interwell (10‟s to 100‟s of m) to full-field scale (10‟s of km). Reservoir rocks are complex
and heterogeneous at all scales. Multiscale simulation is a major goal of the petroleum industry, and many
1
Fig 1 Upscaling problem in heterogeneous rocks. A typical grid block used in a reservoir simulator is 250 x 250 x 1
m in size (right). Borehole-scale numerical cores represent rock volumes on the cubic-meter scale (center). Core
plugs, microCTscans, and confocal scans represent even smaller volumes (left). The question is, how representative
are these small volumes when compared to coarse grid blocks, or full fields?
upscaling approaches have been proposed. Most of these conventional approaches start with fine-scale reservoir
models that are coarsened to models that fluid-flow simulators can handle. The biggest challenge in this type of
upscaling occurs because it is commonly difficult to preserve essential geologic heterogeneities in the resulting
coarse models.
The approach described here provides an improved understanding of fluid-flow paths, fluid saturations, and
recovery factors in hydrocarbon reservoirs. The workflow is suitable for any lithology - carbonates, sandstones, or
shales. At each scale, petrophysical properties are calculated, transferred up to the next scale, and used to populate
coarser digital rock models. Work is currently mature at the pore and borehole scale, and more conceptual at the
interwell and full-field scale.
BACKGROUND TECHNOLOGY
Heterogeneity and Upscaling
Heterogeneity is defined as the variation in properties as a function of location within a reservoir or formation.
Many reservoirs are heterogeneous because mineralogy, grain type, grain size, depositional environment, porosity,
permeability, natural fractures, faults, channels, and other attributes vary from place to place. Heterogeneity
2
causes problems in formation evaluation and reservoir simulation because reservoirs occupy enormous volumes,
but there is limited core and log control.
A geocellular model is a layered, gridded 3D model. Layers can have zero thickness, as in the case of bed pinch outs
or truncations. Layers can be as thin as the spacing of log measurements, or they can be thicker, to reflect the known
thickness of rock layers. Geocellular models capture geologic-scale heterogeneities, and commonly have millions of
cells.
Upscaling is the process of converting rock properties from fine scales to coarser scales. Upscaling algorithms
assign suitable values of porosity, permeability, and other flow functions to each coarser grid block (for example,
see Lasseter et al., 1986; Christie, 1996; Durlofsky, 2003). Upscaling is necessary because reservoir simulators
cannot handle the large number of cells in typical geocellular models.
Digital Rock Modeling
This paper describes a workflow for multiscale reservoir simulation, based upon digital rock models. Such models
are built from cores, well logs, and seismic data. Our goal is to create 3D models to digitally represent pore space
and heterogeneous rock fabrics at all scales. This approach can be extended to shales and other lithologies. Fractures
and faults can be included at various scales.
Recurring themes of digital rock modeling include multi-point statistics and representative element volumes.
Multipoint statistics - Multipoint (or multiple-point) statistical methods (MPS) are a family of spatial statistical
interpolation algorithms proposed in the 1990s used to generate conditional simulations of discrete variable fields,
such as geological facies, through training images (Guardiano and Srivastava, 1993). MPS generates realistic
models that can be constrained by different types of data. Unlike the conventional 2-point or variogram-based
geostatistical approaches, MPS uses a training image to quantify complex depositional patterns that exist in studied
reservoirs. These training patterns are then reproduced in the final MPS models with conditioning to local data
collected from the reservoirs. Therefore, MPS allows modelers to use their prior geological interpretations as
conceptual models (training images) in the reservoir modeling process and evaluate the uncertainty associated with
prior interpretations by using different training images.
In addition to categorical variables, MPS can deal with continuously variable training images, such as spatial
distribution of porosity. Two families of MPS algorithms are available to handle these different types of training
images: Snesim for categorical variables, and Filtersim for continuous variables. Strebelle (2002) proposed an
efficient Snesim algorithm that introduced the concept of a search tree to store all replicates of patterns found
within a template over the training image. This makes Snesim code several orders of magnitude faster than the
original algorithm proposed by Guardiano and Srivastava (1993). Filtersim, developed by Zhang (2006), applies a
set of local filters to the training image, which can be either categorical or continuous, to group local patterns into
pattern classes. Pattern simulation then proceeds on the basis of that classification.
Snesim and Filtersim algorithms honor absolute, or “hard” constraints from data acquired in wells or outcrops, and
other interpreted trend maps of the reservoir under study. Training images are the main driver of any MPS
approach. An issue raised implicitly by current MPS algorithms is how to generate training images. Training
images seek to model or reproduce real geological features and should as much as possible be derived from
existing geologically meaningful images. This requires research on statistical and image-processing methods that
will allow use of images from any source: hand-drawn sketches, aerial photographs, satellite images, seismic
volumes, geological object-based models, physical-scale models, or geological process-based models.
In this paper, MPS is used to create simulations of spatial geological and reservoir property fields. These
conditional simulations use known results, such as those measured in wellbores or rock samples, as fixed or hard
data that are absolutely honored during modeling.
3
Representative element volumes - Representative element volumes (REV) provide a new way to deal with
heterogeneity and upscaling issues in reservoir modeling (for example, see Qi, 2007). In summary, REV is the
smallest volume that can be modeled to yield consistent results, within acceptable limits of variance of a modeled
property, such as porosity or permeability. Using this approach, we can upscale rock properties from fine to coarse
scales. How? We determine the smallest volume that needs to be modeled, run the flow model, and use the results
in the next larger-scale simulations. Once we have modeled the REV, we do not need to model larger volumes
because we have captured heterogeneity for the particular rock type at that scale.
Using the REV approach, the porous medium is replaced by “a fictitious continuum: a structureless substance, to
any point of which we can assign kinematic and dynamic variables and parameters that are continuous functions of
the spatial coordinates of the point and of time” (Bear, 1972). Note that the REV for porosity may differ from the
REV for permeability or other parameters. Also, the REV for static vs. dynamic properties may vary. In practice, the
best method is to use the largest REV determined for various parameters.
How can we determine REV for a rock property, such as porosity? Conceptually, one could model a large volume,
subsample that volume, and compute variance in porosity as a function of subsample volume. This could be done at
any scale, ranging from pores to boreholes to grid blocks. Figure 2 shows a modeled volume at the interwell scale.
Smaller sub-volumes, for example, 1, 10, and 20 m3, could be extracted from the modeled volume, and their
porosities could be determined. All sub-volumes, regardless of scale, should be independent, non-overlapping
volumes. If porosity variance is less than a chosen cutoff, for example +/-5%, then that volume can be used as the
REV. For the purpose of flow modeling, the REV is sufficient to yield representative results.
Fig. 2 Non-overlapping subvolumes in a borehole- to interwell-scale model. REV can be computed from the
variance in a petrophysical property, such as porosity, as a function of subvolume size.
4
The generalized approach for REV determination for any rock property is:
 Model a large block with rock properties of interest.
 Randomly select a subsample of a given small size within the block.
 Randomly select another non-overlapping subsample of the same size.
 Repeat this process many times.
 Increase subsample size, and sample many similar objects.
 Plot rock property vs. subsample size, to see how variance from the sample mean decreases as a function of
subsample size.
 When variance is within acceptable limits (for example, ±5%), this is the REV for the rock-property under
study.
WORKFLOW FOR GENERATION OF 3D DIGITAL ROCK MODELS
This section documents the step-by-step procedure to create digital rock models and upscale them from pore to
borehole to interwell to full-field scale. Variations to this procedure may be necessary, depending on available data.
Preliminary Work
Before a company delves into multiscale reservoir simulation, they should define Reservoir Rock Types (RRT‟s)
and perform conventional log analyses.
 Describe cores to identify facies, rock fabrics, and RRT‟s. A given field will generally have 5-15 RRT‟s. This
exercise works best if RRT‟s are based on lithofacies combined with petrophysical properties (for example,
porosity, permeability, mercury injection capillary pressure, nuclear magnetic resonance).
 Acquire conventional porosity and permeability analyses for whole-core samples and/or core plugs.
 Evaluate and integrate existing core analyses, core descriptions, and RRT‟s.
 Perform conventional log analyses in the wells.
 Interpret borehole images and other openhole logs, and compare them to geological facies observed in core. If
the correlation is good, use an approach, such as neural networks, to distribute facies along the length of the
borehole(s).
Pore- and Borehole-Scale Models
A major goal of pore-scale digital rock modeling is to build 3D models that use multipoint statistics (MPS), as
applied to laser scanning fluorescent microscopy and other high-resolution pore-imaging techniques, such as
microCTscans. Our workflow uses the following tools and technologies:
Transmitted laser scanning fluorescent microscopy, also known as confocal microscopy (Fredrich, 1999), provides
high-resolution (250-nm) 3D pore models to quantitatively capture microporosity. From this we compute pore-
body and pore-throat size distributions and simulated capillary pressure curves.
5
MicroCTscans use x-ray computed tomography (CT) on small samples (commonly 5- to 10-mm diameter core
plugs) to detect pore bodies, with resolutions that are typically 1 to 5 microns in size (Knackstedt et al., 2004).
Software converts pore images into pore-network models, with resulting pore-body and pore-throat size
distributions.
NanoCTscans use x-ray computed tomography (CT) on very small samples (commonly 60-micron diameter core
plugs) to detect pore bodies with resolutions that are typically 50 to 60 nm in size. Software converts pore images
into pore-network models, with resulting pore-body and pore-throat size distributions.
Focused ion beam-scanning electron microscopy (FIB-SEM) uses ion-beam thinning to create multiple 2D serial
sections used to build 3D models of sub-micron scale pores. Resolution is typically on the 5 to 10 nm scale, and
imaged volumes are typically 10x10x10 microns in size.
Mercury injection capillary pressure (MICP) involves progressive injection of mercury into a cleaned sample,
commonly a plug, at steadily increasing pressures (Jennings, 1987; Pittman, 1992). At each increased pressure,
pore throats of a particular size are invaded by mercury. Pore-throat size distributions are generally shown as
histograms, computed from MICP results. Note that MICP is not useful for pore throats larger than 100 microns
because those throats are filled at very low injection pressures. The ideal pore-throat size for MICP is 0.1 to 100
microns.
Nuclear magnetic resonance (NMR), summarized by Coates et al. (1999), is based on the interaction of hydrogen
nuclei (protons) with a magnetic field and pulses of radio-frequency signals. The NMR transverse relaxation time
distribution ( distribution) relates to pore-size distribution in the rock. NMR results can be used to partition
porosity into micro, meso, and macroporosity (Ramamoorthy et al., 2008). Such results can be used to constrain
pore-scale digital rock models.
Petrophysical facies can be represented by areas enclosed by minipermeability contours on core-slab faces
(Bourke, 1993; Dehghani et al., 1999). Such regions commonly also have characteristic signatures on borehole-
image logs, such as vugs, resistive patches, and conductive patches (Zhang et al., 2009). Conductive patches, which
correspond to regions of enhanced porosity and permeability, provide flow continuity between vugs. Such
conductive or resistive patches have complex 3D shapes.
Workflows generate the following products:
Pore-network models are computed for representative element volumes (REV‟s) of individual petrophysical facies.
Modeled results include capillary pressures, resistivity indices, permeabilities, and relative permeabilities. These
numerical SCAL values, especially when validated with laboratory measurements, are used to populate borehole-
scale models.
A major goal of borehole-scale digital rock modeling is to build flow models of heterogeneous rocks using cores and
borehole images. Our workflow employs the following tools and technologies:
Conventional CTscans (1- or 2-mm spacing) are run and processed into 3D core images. These are used as MPS
training images for numerical cores.
Fullbore images are 360-degree views of the borehole wall generated by “filling in the gaps” between the pads of
electrical borehole image logs using MPS (Hurley and Zhang, 2011). They image rock heterogeneity in the near-
borehole volume, and are used as hard data to constrain MPS models of numerical cores.
Minipermeameters capture permeability variation on rock-slab surfaces, and are used to: (a) segment digital rock
models into appropriate petrophysical facies, (b) validate those facies to absolute permeability values, and (c)
identify subvolumes for more detailed sampling.
6
Numerical cores use conventional CTscans of cores, borehole image logs, and MPS to generate 3D models wherein
every cell has its own porosity, permeability, capillary-pressure, and relative-permeability attributes. Such models
quantitatively capture rock heterogeneity at the borehole scale (Zhang et al., 2009).
Fluid-flow models of numerical cores are used to compute effective relative permeabilities, residual oil saturations,
and recovery factors for REV‟s of key rock types.
The following steps comprise advanced digital core analysis, in the form of numerical core and flow simulation
(Figures 3 and 4):
 Choose 1- to 3-ft (0.3- to 1.0-m) whole-core or slabbed-core intervals from each key RRT. Preferably, these
intervals will be from well(s) with electrical borehole image logs. Added benefits will come from wells with
elemental spectroscopy and nuclear magnetic resonance logs. These logs will be useful for mineralogy and
porosity partitioning (macro, meso, microporosity), respectively.
 Acquire conventional CTscans (1- or 2-mm step distance) for chosen 1-to 3-ft (0.3- to 1.0-m) intervals.
 Acquire gridded minipermeability data (0.5- to 1.0-cm grid spacing) for rocks that have been CTscanned.
Calibrate the minipermeability device to core plugs that have a wide range of permeabilities (for example, 0.1 to
3,000 md). Compute calibrated absolute permeabilities for each grid point. Contour minipermeability data. This
step will lead to: (a) proper segmentation of the whole-core CTscan, (b) validation to lab-measured
permeabilities, and (c) identification of subvolumes for more detailed sampling.
 Register minipermeability data to slab face of conventional CTscans. Use this combination to choose subsample
areas suitable for milling or core plugging. Drill strategically chosen core plugs of distinctive petrophysical
facies, and submit them for: (a) thin (30 micron) or thick (5,000 micron) sections for scanning laser fluorescent
microscopy and microCTscan, and (b) laboratory porosity, permeability, and SCAL (MICP, k rel). Preferably,
these data will be acquired at reservoir conditions. Such results will be used as end-point rock properties for
petrophysical facies. This is a critical part of the laboratory validation step.
 Thin or thick sections must be vacuum-pressure impregnated with fluorescent-dyed (for example, Rhodamine
B) epoxy. Use transmitted laser scanning fluorescent microscopy to verify that an REV has been sampled for
each petrophysical facies.
 Use REV‟s imaged with transmitted laser scanning fluorescent microscopy to segment microCTscans. This step
is especially important if significant amounts of porosity lie below the resolution threshold of the microCTscan
device.
 Use MPS to build 3D pore-network models. Such models are used to compute numerical SCAL (MICP, relative
permeability, resistivity index), which will be used to populate larger-scale numerical cores. If thick sections are
made, run microCTscan with resolution of 5 microns or better. MPS uses confocal scans as training images and
microCTscans as hard data to build composite “total porosity” models for each petrophysical facies.
 If electrical borehole image logs are available, generate fullbore images for each 3-ft (1-m) interval,
encompassing the CTscanned cored intervals. Create 3D numerical cores from borehole images and CTscans
using MPS.
 Use numerical SCAL results from pore-scale models to populate borehole-scale digital rock models, i.e.,
numerical cores.
7
Fig. 3 Pore-scale digital rock models. In practice, we image whole cores with conventional CTscans, using closely
spaced slices (1- or 2-mm). Minipermeability grids (0.5- or 1-cm) on slab faces allow us to segment cores into sub-
regions, validated with lab-measured permeabilities. Strategically chosen core plugs, sampled within the context of
CTscan and minipermeability results, represent petrophysical facies. Submicron-sized pores of each petrophyisical
facies are imaged for their corresponding REVs using laser scanning fluorescent (confocal) microscopy and
microCTscans. Numerical SCAL, computed from pore models or pore-network models, carry us to the next larger
scale, i.e., the borehole.
 Confirm that numerical cores are REV‟s for the particular RRT.
 Determine effective porosities, permeabilities, capillary pressures, relative permeabilities, saturations, and
recovery factors in near-borehole volumes using conventional fluid-flow simulators. If possible, validate
computed SCAL determined from numerical cores with whole-core SCAL run in the lab.
 Run multiple realizations, or run realizations for different porosity ranges. Integrate reservoir pressure, fluid
properties, and other field data to run flow simulations.
Interwell-Scale Modeling
A major goal of interwell-scale digital rock modeling is to use variogram statistics from LWD data, seismic
attributes, and cross-well geophysics to capture interwell heterogeneity. Flow properties are provided by numerical
cores. We use the following tools and technologies:
Logging while drilling (LWD) data, especially azimuthal density logs, are acquired for geosteering in horizontal
wells. Outcrop studies show that cyclic variations (known as hole effect; Pyrcz and Deutsch, 2003) occur in
porosity and permeability transects along particular stratigraphic horizons. If these cycles occur in reservoirs,
variogram statistics from LWD data can be used to provide heterogeneous „structure‟ to help populate models of
the interwell region.
8
Fig. 4 Borehole-scale digital rock models. Numerical cores (right) are created using discrete-variable MPS
algorithms. Integer values are assigned to each petrophysical facies, for example, dense rock matrix (0), vugs (1),
resistive patches (2), and conductive patches (3). Conventional CTscans (left) of whole-core rock samples are
used as MPS training images, i.e., they are the quantitative templates used to guide the modeling of 3D textures at
the borehole scale. Fullbore images (middle), derived from electrical or acoustic borehole images, surround
numerical cores with cylindrical envelopes that condition the models. Each numerical core (right) absolutely
honors the heterogeneity of the digital rock samples and the fullbore image data. Subvolumes can be sampled to
confirm that REV’s are being modeled for a given RRT.
Numerical cores are gridded into models used for fluid-flow simulation. For each petrophysical facies, capillary
pressure and relative permeability curves are provided by numerical SCAL. Bulk, or system-scale, properties are
computed from flow-model results for the numerical cores: effective residual oil saturations, capillary pressures,
recovery factors, and relative permeabilities. These properties are used to populate digital rock models at the next
scale, i.e., interwell-scale grid blocks.
Numerical cores are run for particular reservoir rock types as multiple realizations, or as models with selected
porosity ranges. Numerical SCAL predicted from REV‟s of these models can be used to populate interwell-scale
digital rock models.
Variograms are geostatistical tools used to depict spatial variance within groups of data, plotted as a function of
distance between data points. When variograms are constructed from log data in horizontal wells, they can be used
to map spatial variability at the interwell scale.
Seismic attributes, such as amplitude and acoustic impedance, commonly relate to porosity. Such attributes can be
used as soft data to constrain MPS models of the interwell volume.
Cross-well geophysics, such as EM (electromagnetics) or seismic tomography, can be used to constrain property
variations in the interwell volume.
9
Fluid-flow models of numerical cores are used to populate interwell-scale flow models with effective relative
permeabilities, residual oil saturations, and recovery factors for REV‟s of key rock types.
The following steps are required to distribute facies and construct the interwell-scale model (Figure 5):
 Determine the stratigraphic profile intersected by horizontal wells that have LWD density data. Compute
variogram statistics for intervals where the well stayed within the same stratigraphic layer. Look for the hole
effect, an indication of spatial cyclicity (Pyrcz and Deutsch, 2003), build variograms, and construct
geostatistical maps of porosity for the interwell area.
 Populate porosity voxels in interwell areas using openhole and LWD well logs, combined with digital SCAL
determined from numerical cores in the appropriate RRT‟s and with the appropriate porosity ranges.
 Confirm that interwell volumes are REV‟s for the particular RRT.
 Run multiple realizations of numerical cores, or create numerical cores for particular porosity ranges. Populate
geostatistical maps with numerical SCAL data from numerical cores.
 Map porosity heterogeneity in interwell volumes using seismic attributes, if available. Use seismic data to
provide input to 3D facies distribution models. If seismic attributes correlate to porosity, use this as soft data to
constrain geostatistical maps of interwell regions.
 Constrain rock properties in the interwell area using cross-well geophysical data, such as EM (electromagnetics)
and seismic tomography, if available.
 Based on porosity models, permeability heterogeneity can be predicted using a technique such as collocated
cosimulation with porosity being used as a secondary variable.
 Compute capillary pressures, relative permeabilities, saturations, and recovery factors for each grid block using
conventional flow simulators.
Full-Field Scale Modeling
A major goal of full-field scale digital rock modeling is to build static models of the reservoir using cores, well logs,
outcrop analogs, and sequence stratigraphy. Variogram-based geostatistical models, MPS, or forward stratigraphic
models can be used to help populate interwell regions. We use the following tools and technologies:
Sequence stratigraphy is an interdisciplinary field of study that combines seismic, log, fossil, and outcrop data at
local, regional and global scales. Basin-filling sedimentary deposits are interpreted within the framework of
sedimentation and sea-level changes, caused by either tectonic or eustatic effects, or both. The approach is used to
correlate strata and predict stratigraphy in relatively unknown areas. Sequence stratigraphy promotes an
understanding of the evolution of basins, while it allows for the interpretation of potential source rocks and reservoir
rocks in both frontier areas and more mature hydrocarbon provinces (Neal et al., 1993).
Outcrops are bodies of rock exposed at the surface of the earth. Outcrops with analogous lithologies and
depositional environments to subsurface reservoirs can be used to help construct static models.
Forward stratigraphic models are used to construct realistic 3D stratal patterns based on hydraulic principles that
apply to sediment transport. Such models can deal with subaqueous and eolian transport of siliciclastic material,
and organic growth of carbonates and vegetation-related sediments, such as coals (for example, see Sedsim, 2012).
10
Fig. 5 Interwell-scale digital rock models. Variogram statistics from horizontal wells drilled within particular rock
layers can be combined with seismic attributes and cross-well geophysics to capture heterogeneity at the interwell
scale. Multiple realizations of digital rock models, or multiple samples at various porosity values, are used to
populate interwell volumes with numerical SCAL properties.
Isopach maps depict thickness of particular rock layers, or thickness of rocks with given petrophysical properties,
such as porosity.
Facies proportion curves are generated from core descriptions and logs. They are used to estimate and constrain
the relative amounts of each facies in a layer of a geomodel.
Seismic attributes, such as amplitude and acoustic impedance, commonly relate to porosity. Such attributes can be
used as soft data to constrain MPS models of the interwell volume.
Diagenetic models are used to simulate cementation, compaction, and other diagenetic processes that accompany
burial of sediment. Approaches range from reaction transport models, based upon thermodynamics and kinetics, to
models based on texture, composition, and burial history of the original sediment (for example, see Geocosm,
2012).
11
Basin-scale models are used to model petroleum systems, commonly at scales much larger than oil fields. A
petroleum system is defined as the combination of geologic elements and processes needed to generate and store
hydrocarbons. Elements and processes include charge, trap, and timing of hydrocarbon generation, migration, and
loss (for example, see Petromod, 2012).
The following steps are recommended to upscale from interwell to full-field simulations (Figure 6):
 Use results for effective numerical SCAL for interwell volumes to populate full-field simulations, built from
static models with independently distributed RRT‟s within a log-correlation and/or sequence-stratigraphic
framework.
Fig. 6 Workflow for full-field simulation using forward stratigraphic modeling, isopach maps, facies proportion
curves, and multi-point statistics to honor well data. Sequence stratigraphy, outcrop analogs, and seismic attributes
are used to develop the static model. Diagenetic models are used to modify petrophysical properties during early
and late deposition and burial.
12
 Use outcrop analogs, if available, to constrain facies types, facies associations, and lateral correlations in the
subsurface static model. These outcrop analogs can be used to assist MPS modeling in generating training
images.
 Use basin-scale models, if available, to place the field within a regional context in terms of source rock,
reservoir, trap, and seal. Burial history, and its effect on diagenesis are important considerations.
 Create spreadsheets of log tops, facies, and sequence boundaries for each described core in a given field.
Compute facies proportion curves for each cored well. Determine facies associations in the vertical succession
using Markov chain analysis, or a similar approach. Construct isopach maps of key facies and/or parameter
thicknesses.
 Use seismic surfaces or conceptual models to provide paleotopography for the forward stratigraphic model
(FSM). Create FSM‟s to visualize scenarios for hydraulic transport or organic growth of sediments. Match
thickness of sequences and the gross interval of interest. Use FSM time-step adjustments to roughly match layer
thicknesses, facies proportions, and facies associations observed in cored wells.
 Create pseudowells at arbitrary locations within the field. Use FSM “lithology” or “grain size” as a proxy for
facies from core description. Adjust the FSM using core descriptions and log interpretations for: (a) layer
thicknesses, (b) facies proportions, and (c) facies associations.
 Use the adjusted FSM as a facies probability cube (soft data) for MPS simulation. Training images will be
layered models, with comparable thicknesses, facies proportions, and facies associations to described cores and
logs in the wells. If necessary, due to variable strata architecture, different training images can be used in
different regions of the field.
 Use diagenetic models to depict changes in porosity and permeability as a function of cementation and
compaction due to burial.
CONCLUSIONS
Reservoir simulation covers at least 14 orders of magnitude, ranging from pore (nm to micron) to borehole (mm to
m) to interwell (10‟s to 100‟s of m) to full-field scale (10‟s to 100‟s of km). Reservoir rocks are complex and
heterogeneous at all scales. Multiscale simulation is a major goal of the petroleum industry, and many upscaling
approaches have been proposed.
This paper summarizes integrated, practical workflows to create digital rock models at the pore, borehole, grid-
block, and full-field scale. The first step involves core description and log interpretation to identify reservoir rock
types (RRT‟s) for a given field. Petrophysics-based RRT‟s, combined with lithofacies, yield the best results.
At all scales, multi-point statistics (MPS) and representative element volumes (REV‟s) are used to generate models
and ensure that sufficient volumes are modeled to capture rock heterogeneity. Multi-point statistics (MPS) creates
simulations of spatial geological and reservoir property fields. These conditional simulations use known results,such
as those measured in wellbores or rock samples, as fixed or hard data that are absolutely honored during modeling.
Representative element volumes (REV) provide a new application in reservoir modeling, based on techniques used
in groundwater hydrology. In summary, REV is the smallest volume that can be modeled to yield consistent results,
within acceptable limits of variance of the modeled property, such as porosity or permeability. REV can be
quantitatively determined by subsampling modeled volumes of increasing size, and finding the volume where
variance is below a chosen cutoff in a particular parameter, such as porosity or permeability.
Pore- and borehole-scale models are created for each key RRT. Pore-scale models employ such tools such as laser
scanning fluorescent microscopy, microCTscans, focused ion beam-scanning electron microscopy, mercury
13
injection capillary pressure, nuclear magnetic resonance, and pore-network models to image pore structure.
Predicted numerical SCAL is transferred up one level to populate borehole-scale numerical cores, based on
minipermeability grids, conventional CTscans, and electrical borehole image logs.
Multiple realizations of numerical cores, or realizations at different porosity ranges are used to populate interwell-
scale models and evaluate uncertainties. Seismic attributes, cross-well geophysics, and variogram statistics from
LWD data provide independent constraints on rock heterogeneity in the interwell volumes.
Interwell-scale simulations are used to populate full-field simulations, within the context of static models
constructed using log correlations, sequence stratigraphy, outcrop analogs, forward stratigraphic models, and
diagenetic models. Basin-scale models are used to provide petroleum systems information for a given field.
REFERENCES
Bear, J., 1972, Dynamics of fluids in porous media: Elsevier, New York, 746 p.
Bourke, L. T., 1993, Core permeability imaging: It‟s relevance to conventional core characterization and potential
application to wireline measurement: Marine and Petroleum Geology, v. 10, p. 318-324.
Christie, M. A., 1996, Upscaling for reservoir simulation: JPT, v. 48, No. 11, p. 1004-1010.
Coates, G.R., Xiao, L. and Prammer, M. G., 1999, NMR logging: Principles and applications: Halliburton Energy
Services, USA, 234 p.
Dehghani, K., Harris, P. M., Edwards, K. A., and Dees, W. T., 1999, Modeling a vuggy carbonate reservoir: AAPG
Bulletin, v. 83, p. 19-42.
Durlofsky, L. J., 2003, Upscaling of geocellular models for reservoir flow simulation: A review of recent progress:
7th International Forum on Reservoir Simulation, Buhl/Baden-Baden, Germany, June 23-27, p. 58.
Fredrich, J. T., 1999, 3D imaging of porous media using laser scanning confocal microscopy with application to
microscale transport processes: Physics and Chemistry of the Earth, Part A: Solid Earth and Geodesy, v. 24,
Issue 7, p. 551-561.
Geocosm, 2012, http://www.geocosm.net/, accessed April 19.
Guardiano, F., and Srivastava, R.M. 1993, Multivariate geostatistics: Beyond bivariate moments: Geostatistics-
Troia, A. Soares, Dordrecht, Netherlands, Kluwer Academic Publications, v. 1, p. 133- 144.
Hurley, N. F., and Zhang, T., 2011, Method to generate full-bore images using borehole images and multipoint
statistics: SPE Reservoir Evaluation and Engineering, p. 204-214.
Jennings, J. B., 1987, Capillary pressure techniques: Application to exploration and development geology: AAPG
Bulletin, v. 71, p. 1196-1209.
Knackstedt, M. A., Arns, C. H., Sakellariou, A., Senden, T. J., Sheppard, A. P., Sok, R. M., Pinczewski,W. V., and
Bunn, G. F., 2004, Digital core laboratory: Properties of reservoir core derived from 3D images: SPE Preprint
87009, Presented at the Asia-Pacific Conference on Integrated Modeling for Asset Management, March 29-30.
Lasseter, T. J., Waggoner, J. R., and Lake, L. W., 1986, Reservoir heterogeneities and their influence on ultimate
recovery, in Lake, L. W., and Carroll, H. B., Jr., eds., Reservoir Characterization: Academic Press, Orlando,
Florida, p. 545-559.
14
Neal, J., Risch, D., and Vail, P., 1993, Sequence stratigraphy—A global theory for local success: Oilfield Review,
January issue, p. 51-62.
Petromod, 2012, http://www.slb.com/services/software/geo/petromod_modeling_software.aspx, accessed April 19.
Pittman, E. D., 1992, Relationship of porosity and permeability to various parameters derived from mercury
injection-capillary pressure curves for sandstone: AAPG Bulletin, v. 76, p. 191-198.
Pyrcz, M. J., and Deutsch, C.V., 2003, The whole story on the hole effect, in Searston, S. (ed.) Geostatistical
Association of Australasia, Newsletter 18, May, 16 p.
Qi, D., 2007, Upscaling theory and application techniques for reservoir simulation: Lambert Academic Publishing,
Saarbrücken, Germany, p. 244.
Ramamoorthy, R., Boyd, A., Neville, T. J., Seleznev, N., Sun, H., Flaum, C., and Ma, J., 2008, A new workflow for
petrophysical and textural evaluation of carbonate reservoirs: SPWLA Preprint, 49th Annual Logging
Symposium, May 25-28, 15 p.
Sedsim, 2012, https://wiki.csiro.au/confluence/display/seabedchange/Sedsim+Model, accessed April 19.
Strebelle, S. 2002, Conditional simulation of complex geological structures using multiple point statistics:
Mathematical Geology, v. 34, p. 1-22.
Whitaker, F. F., and Xiao, Y., 2010, Reactive transport modeling of early burial dolomitization of carbonate
platforms by geothermal convection: AAPG Bulletin, v. 94, p. 889 917.
Zhang, T. 2006, Filter-based training image pattern classification for spatial pattern simulation: Unpublished Ph.D.
dissertation, Stanford University, Palo Alto, CA.
Zhang, T., Hurley, N. F., and Zhao, W., 2009, Numerical modeling of heterogeneous carbonates and multi-scale
dynamics: Presented at the SPWLA 50th Annual Logging Symposium, The Woodlands, Texas, June 21-24.
ABOUT THE AUTHORS
Neil Hurley is a Scientific Advisor at Schlumberger-Doll Research Center, Cambridge, MA, USA. His research
interests include borehole images and carbonate reservoir characterization. He worked for Conoco Inc. and
Marathon Oil Company from 1978-1996, and he is Emeritus Professor, retired from the Colorado School of Mines,
2006. He earned his BS degrees in geology and petroleum engineering from the University of Southern California in
1976, his MS degree in geology from the University of Wisconsin in 1978, and his PhD degree in geology from the
University of Michigan in 1986.
Weishu Zhao is a Research Scientist at Schlumberger Dhahran Carbonate Research Center, Dhahran, Saudi Arabia.
His research interests include transport phenomena in porous media, pore-scale modeling of multiphase flow and
reservoir simulation. He earned his MS degree (2002) and PhD degree (2006) in chemical engineering at the
University of Waterloo, Ontario, Canada.
Tuanfeng Zhang is a Senior Research Scientist at Schlumberger-Doll Research, Cambridge, MA, USA. His research
interests include geomodeling, reservoir modeling and characterization, geostatistics, spatial statistics and its
applications. He was an associate professor in Xian Petroleum Institute, China, from 1995 to 2000. Later, he studied
at Stanford University from 2000 to 2006. He earned his MS degree in petroleum engineering in 2002 and his PhD
in geological and environmental sciences in 2006.
15

Spwla 2012 195

Uploaded by

Copyright:

Available Formats

Spwla 2012 195

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Spwla 2012 195

Uploaded by

Copyright:

Available Formats

SPWLA 53rd Annual Logging Symposium, June 16-20, 2012

MULTISCALE WORKFLOW FOR RESERVOIR SIMULATION

Heterogeneity and Upscaling

Digital Rock Modeling

 Randomly select a subsample of a given small size within the block.

 Randomly select another non-overlapping subsample of the same size.

 Repeat this process many times.

 Increase subsample size, and sample many similar objects.

WORKFLOW FOR GENERATION OF 3D DIGITAL ROCK MODELS

 Perform conventional log analyses in the wells.

Pore- and Borehole-Scale Models

Workflows generate the following products:

Workflows generate the following products:

Workflows generate the following products:

Full-Field Scale Modeling

Geocosm, 2012, http://www.geocosm.net/, accessed April 19.

Sedsim, 2012, https://wiki.csiro.au/confluence/display/seabedchange/Sedsim+Model, accessed April 19.

ABOUT THE AUTHORS

You might also like