You are on page 1of 17

REE189968 DOI: 10.

2118/189968-PA Date: 22-March-18 Stage: Page: 1 Total Pages: 17

Rock Typing in Eagle Ford, Barnett,


and Woodford Formations
Ishank Gupta, Chandra Rai, Carl Sondergeld, and Deepak Devegowda, University of Oklahoma

Summary
Shales are the most commonly found sedimentary rocks on Earth. Most US shale plays are massive, with different maturity regions and
varying prospects. There has been a paradigm shift in the understanding of shale anisotropy and microstructures in the last decade, with
a high focus on the identification of sweet spots and optimal reservoir-quality zones. Rock typing is one of the sought-after techniques
to achieve this objective, and it has become an integrated part of the unconventional-characterization work flow.
In this work, rock typing was performed with an integrated work flow using laboratory petrophysical measurements. The rock types
were derived with machine-learning clustering algorithms—namely, K-means and self-organizing maps (SOM). The integrated work
flow was applied in three different shale plays: Eagle Ford, Barnett, and Woodford.
Three different rock types were identified. In general, Rock Type 1 had the highest porosity and total organic carbon (TOC), indica-
tive of highest storage and source-rock potential, respectively. Rock Type 1 was also the key rock type controlling the production. Rock
Type 2 had intermediate porosity and TOC, whereas Rock Type 3 had the lowest porosity and TOC.
Next, core-derived rock types had to be scaled up to logs. Support vector machines (SVM), a classification algorithm, was used for
scaling up. It was trained with a data set consisting of depths at which both core and log data were available. Different logs such as
gamma ray, resistivity, neutron, and density were used for scaling up. Finally, a rock-type ratio (RTR) was defined from rock-type logs
based on fraction of Rock Type 1 over gross thickness. The ratio thus developed was found to have a strong correlation with normalized
oil-equivalent production rate.
In total, 22 wells with core data were considered for rock typing in the three shale plays. The rock types were scaled up to 95 wells
at a cumulative over a 20,000-ft depth interval. The work flow shown in this paper can easily be extended to other data sets in other
plays. The manual approach, on the other hand, can be prohibitively time-consuming.

Introduction
Over the years, numerous methods have been developed for rock typing in conventional reservoirs. Most of these methods were based
on core- or log-derived permeability/porosity crossplots. These methods were particularly successful in sandstones and carbonates
because of a large dynamic range of permeability and porosity values in these formations. Although well logs do not provide permeabil-
ity estimates directly, several correlations were developed to obtain them indirectly from other logs (Timur 1968; Coates and Dumanoir
1974; Thomeer 1983). If nuclear-magnetic-resonance (NMR) logs are available, Timur-Coates and Schlumberger Doll Research (SDR)
models also can be used to obtain permeability estimates. Pittman (1992) suggested using r35 cutoffs based on mercury-injection
experiments to determine different rock types. Amaefule et al. (1993) introduced rock-quality index (RQI) by modifying the Kozeny-
Carman equation. He also proposed flow-zone indicator (FZI) cutoffs to define the different rock types.
Walls (1982), Randolph et al. (1984), Davies et al. (1993), and Rushing et al. (2008) found commonly used rock-typing methods
inadequate for tight reservoirs. They believed that other attributes such as rock texture and composition, core-based descriptions, and
clay mineralogy are also important for rock typing in tight reservoirs.
Kale (2009) and Kale et al. (2010) conducted rock typing for Barnett, and they included several petrophysical parameters such as
TOC, mineralogy, helium porosity, and mercury-injection capillary pressure (MICP). Sondhi (2011), Gupta (2012), Aranibar et al.
(2013), Li et al. (2015), and Gupta et al. (2017) have presented similar approaches to rock typing in unconventional reservoirs.
This study focuses on Barnett, Woodford, and Eagle Ford shales. In this paper, as a first step, core-based rock typing was conducted
with several clustering algorithms—namely, SOM (Kohonen and Honkela 2007) and K-means (MacQueen 1967). To scale up rock
types from cores to logs, a supervised classification technique referred to as SVM (Cortes and Vapnik 1995) was used. Finally, the
scaled-up rock-type logs were correlated with production data to test the robustness of the rock-typing exercise.

Laboratory-Core Measurements
The laboratory petrophysical measurements commonly used for rock typing are helium porosity, mineralogy, TOC, source rock analysis
(SRA) (S1 and S2 values), and MICP. The laboratory measurements were performed for 263 samples (in Eagle Ford), 211 samples (in
Barnett), and 411 samples (in Woodford).

Porosity. The helium porosity-measurement procedure is outlined in Karastathis (2007). The methodology is based on the use of
Boyle’s law to determine grain volume. Bulk volume is determined by mercury immersion. Finally, porosity is calculated from the
knowledge of grain and bulk volumes.

Mineralogy. The fourier-transform infrared spectroscopy (FTIR) (Sondergeld and Rai 1993; Ballard 2007) technique is used to deter-
mine mineralogy of the samples. The technique uses an in-house algorithm to invert the absorbance spectrum for 16 different minerals,
namely, quartz, calcite, dolomite, illite, smectite, kaolinite, chlorite, pyrite, orthoclase, oglioclase, mixed clays, albite, anhydrite, sider-
ite, apatite, and aragonite. The composition is reported in terms of weight percentage.

Copyright V
C 2018 Society of Petroleum Engineers

Original SPE manuscript received for review 6 March 2017. Revised manuscript received for review 20 October 2017. Paper (SPE 189968) peer approved 27 October 2017.

2018 SPE Reservoir Evaluation & Engineering 1

ID: jaganm Time: 12:47 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 2 Total Pages: 17

TOC. TOC was measured with the dry-pyrolysis technique (Law 1999). Three factors that determine the source-rock potential are
source-rock richness, source-rock quality, and source-rock maturity. TOC is indicative of the richness of the source rock. A higher TOC
may indicate higher source-rock potential.

SRA. SRA analysis is performed by a pyrolysis flame-ionization detection (PFID) technique. At the first stage, the sample is heated at
the rate of 25 C /min up to 300 C. During this stage, volatile hydrocarbons are released which are measured as a S1 peak. In the next
stage, sample is heated to a higher temperature of 550 C. During this stage, kerogen in the sample is pyrolyzed and recorded as S2
peak. The CO2 released during this process is recorded as S3 peak. The residual carbon is recorded as S4 peak. Generally, S1 peak cor-
responds to movable hydrocarbons, and S2 represents immovable hydrocarbons. More details are available in Tissot and
Welte (1984).

MICP. MICP is measured by a Micromeritics AutoPore IV machine. The pressure steps are varied from 5 to 60,000 psia. The mercury
intrusion and extrusion from the sample are measured as a function of pressure. The incremental and cumulative intrusion and extrusion
curves are recorded and interpreted to assess the connectivity of the sample along with other data such as helium porosity.

Geological and Petrophysical Overview


This section describes the geological and petrophysical background for different shale plays: Eagle Ford, Barnett, and Woodford. This
information is critical because it provides constraints on the rock typing and also helps one understand different rock types in terms of
depositional processes.

Eagle Ford. Eagle Ford is one of the most prominent shale plays in the US. It is late-cretaceous in age with more than 1.5 billion bbl
of oil, and 4.2 Tcf gas produced so far. It varies in thickness between 150 and 450 ft (Callantine 2010). The depths of Eagle Ford for-
mation vary from 7,000 to 12,000 ft (CLR 2010). Eagle Ford is slightly overpressured with pressure gradient varying from 0.4 to
0.7 psi/ft. (CLR 2010). It has four different basins: Maverick, Hawkville, San Marcos, and East Texas. Traditionally, some experts do
not consider the East Texas Basin as a part of Eagle Ford play because of its very high clay content compared with the rest of the
Eagle Ford.
Breyer et al. (2013) discuss the depositional environment for Eagle Ford formation. They suggested that the sediment influx hap-
pened from the northeast direction. East Texas Basin in the northeastern part of the Eagle Ford shale play mainly had deltaic deposits.
This is the primary reason for higher clay fraction in this basin. The central, western, and southern parts of the play were dominated by
marine shelf and slope deposits. The marine deposits are very rich in carbonates. The slope deposits (in the south) are deeper compared
with shelf deposits (in the north). The slope deposits have a higher maturity compared with the shelf deposits. Thus, as one goes north,
maturity decreases, and there is a transition from gas to oil (Tuttle 2010).
EOG (2010) reported an isopach thickness map for Eagle Ford formation. They reported that western and southern parts of the play
have a higher thickness (going up to 350 ft), whereas northern and eastern parts of the play have lower thickness (up to 60 ft). The thick-
ness increases as one goes seaward. TOC is another important factor governing the prospect of a shale play. Generally, lower Eagle
Ford is richer in TOC. The TOC varies from 1 to 6% in upper Eagle Ford and from 2 to 12% in lower Eagle Ford (Tian 2014). The
thicker parts of the Eagle Ford play are associated with higher TOC.
The wells with the core and log data and which were used for rock typing are shown in Fig. 1. Twelve wells were available that had
the core data (shown as red bubbles). They spread throughout the Eagle Ford play but were mainly limited to the condensate window.
Core data were available for 263 depth points. Of these 12 wells, triple combo logs were available in only 3 wells, whereas three more
had gamma ray and resistivity. Thus, gamma ray and resistivity logs were used for scaling up from core to log level. In addition, 17
wells (shown as black bubbles) were taken which had no core data but had the required logs—namely, gamma ray and resistivity. The
rock-type logs were created in these wells to get large sample size for correlation of production data with the rock-type logs.

Barnett. Barnett is one of the prominent shale gas plays in the US. It is a Mississippian shale in the Delaware and Fort Worth basins in
North Texas (Pollastro et al. 2007). It has produced more than 69 million bbl oil and 19.2 Tcf gas so far. It varies in thickness between
100 and 700 ft (Kinley 2008). The depths of the Barnett formation vary between 7,000 and 18,000 ft. (Kinley 2008). Barnett is slightly
overpressured, with an average pressure gradient of 0.52 psi/ft. (Slatt 2008).
Barnett shale play lies in the Fort Worth Basin that was formed during the Paleozoic. Barnett was deposited in a back-arch setting
(Bruner and Smosna 2011). Barnett play consists of upper Barnett, lower Barnett, and Forestburg limestone. Lower Barnett lies directly
over a regional angular unconformity. The Forestburg limestone separates the upper and lower Barnett shale members. It is quite thick
in the north (200 ft) and thins as one goes south (few feet). Toward the south, it becomes difficult to distinguish upper Barnett from
lower Barnett. Bruner and Smosna (2011) also reported an isopach map for the Barnett. The northeastern and eastern portions of the
play have the highest thickness.
Sarmiento et al. (2013) discussed TOC and vitrinite reflectance data for Barnett. They showed that most of the play (including the
thick eastern and northeastern parts) has low TOC. The Southern part of the play has the highest TOC. But the maturity data (vitrinite
reflectance, % Ro) showed that most of the high TOC region lies in an immature window. The thick eastern and northeastern parts of
the basin with lower TOCs lie in the gas-maturity window. This explains the high gas potential of the Barnett play.
Singh (2008) identified 10 different lithofacies in Barnett. They were, namely, siliceous noncalcareous mudstone, siliceous calcare-
ous mudstone low calcite, siliceous calcareous mudstone high calcite, silty-shale, phosphatic deposits, limy mudstone, dolomitic mud-
stone, calcareous laminae, concretions, and fossiliferous deposits. Out of these, Lithofacies 1, 2, 3, and 6 are dominant and are
responsible for the majority of the petrophysical variation. Most of the cored interval also consists of these four lithofacies. Lithofacies
1 and 2 are associated with high TOC and high porosity. On the other extreme, Lithofacies 6 is very tight with little porosity and very
low TOC.
The wells with the core and log data and which were used for rock typing are shown in Fig. 2. Three wells that had core data were
available. They lie in an overmature, gas-rich, stratigraphically thick part of the Barnett shale play (shown as red bubbles). Core data
were available for 211 depth points. Of these three wells, only gamma ray and resistivity logs were available for two wells. Thus,
gamma ray and resistivity logs were used for scaling up from core to log level. In addition, 44 wells (shown as black bubbles) were
taken which had no core data but had the required logs—namely, gamma ray and resistivity. The rock-type logs were created in these
wells for correlation with production data.

2 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:47 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 3 Total Pages: 17

Eagle Ford Shale Play


W12

W10, W11

East Texas Basin

W7, W8, W9

W5, W6
W2
W4
San Marcos Basin

W3

W1

Hawkville Basin

Fig. 1—Wells with core and log data for rock typing in Eagle Ford. 12 wells had core data (shown as red bubbles). They spread
throughout the Eagle Ford play but were mainly limited to the condensate window. Core data were available for 263 depth points.
Additional wells (shown as black bubbles) were taken for correlation of rock types with production data. They did not have core
data but had the required logs.

Woodford. The Woodford shale in the Anadarko, Arkoma, and Ardmore basins in Oklahoma and Texas is a Devonian/Mississippian
shale. The shale has produced more than 87 million bbl oil and 4.6 Tcf gas so far. The thickness of the shale varies between 150 and
400 ft (CLR 2010). The depths of Woodford formation vary between 4,800 and 10,000 ft (CLR 2010). Woodford formation is overpres-
sured with pressure gradient varying between 0.60 and 0.65 psi/ft (CLR 2010).
Woodford shale was deposited in the Devonian, approximately 360 million years ago (CLR 2010). It was deposited in the marine
environment as an organic-rich shale. The low-oxygen environment facilitated preservation of oil-prone organic matter. During the
early Pennsylvanian, plate collision resulted in the formation of Anadarko, Ardmore, and Arkoma basins, followed by rapid subsidence
and sedimentation, during the late Pennsylvanian. It was during this period that most of the overlying sandstone reservoirs were depos-
ited. By the early Permian period, oil generation and migration into overlying conventional reservoirs had started.
The Anadarko basin has the thickest Woodford formation among the three basins (Caldwell and Johnson 2013). Comer (2005)
reported TOC and vitrinite reflectance data for the Woodford Shale. The data show that the Anadarko basin has the highest TOC, and it
lies in the oil-maturity window. Thus, Anadarko basin is likely to have the best oil production in Woodford Shale play. The South Cen-
tral Oklahoma Oil Province (SCOOP) and STACK (Sooner Trend Anadarko Basin Canadian and Kingfisher Counties), the most prolific
Woodford shale regions, lie in the Anadarko basin.
The common lithologies found in the Woodford Formation are black shale, chert, sandstone, siltstone, and dolostone. The most pro-
ductive lithologies are siliceous and include cherts and cherty black shales. Siliceous formations in Woodford are highly brittle, and nat-
ural fractures are also seen (Gupta 2012). Chert and quartz in Woodford have different sources and distributions. Quartz is detrital in
origin, whereas chert is biogenic and represents siliceous microorganisms—namely, Radiolaria. Chert deposits are organic-rich, and
where they are thermally mature, they form optimal exploration targets (Schieber 1996; Gupta 2012; Gupta et al. 2013).
The wells with both core and log data and which were used for rock typing are shown in Fig. 3. Seven wells were available which
had the core data. Most of these wells were in the Anadarko basin (shown as red bubbles). Core data were available for 411 depth
points. Triple combo logs were available in all seven wells. Thus, gamma ray, resistivity, neutron, and density logs were used for scal-
ing up from core to log level. In addition, 12 wells (shown as black bubbles) were taken that had no core data but had triple combo
logs. The rock-type logs were created in these wells for correlation with production data.

Rock-Typing Methodology
The work flow used for rock typing is given in Fig. 4. The key petrophysical parameters used for rock typing are porosity, TOC, and
mineralogical compositions. These data were selected because these represent the most commonly performed measurements in the lab-
oratory. These measurements have high accuracy and lower associated errors. Finally, they explain the maximum variance in the data
and are sufficient to distinguish different rock types.

2018 SPE Reservoir Evaluation & Engineering 3

ID: jaganm Time: 12:47 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 4 Total Pages: 17

Barnett Shale Play

W13

W14
W15

Fort Worth Basin

Fig. 2—Wells with core and log data for rock typing in Barnett; three wells (shown as red bubbles) were available which mainly lie
in over mature, gas-rich, stratigraphically thick part of the Barnett Shale play. Core data were available for 211 depth points. An
additional 44 wells (shown as black bubbles) were taken for correlation of rock types with production data. They did not have core
data but had the required logs.

Generating Clusters With Principal Component Analysis (PCA), K-means, and SOM. Rock typing involves grouping rock sam-
ples that share certain characteristics. Although this may be easily performed if the inputs are composed of two or three measurements,
it can be challenging to accomplish when the input data have high dimensionality. PCA (Hotelling 1933) can be used to reduce data
dimensionality. The idea behind PCA is to identify directions in multidimensional space that contain most of the variations observed in
the data. Principal components are a linear combination of different input parameters (TOC, porosity, and mineralogy in this study).
Mathematically, the principal components are the eigenvector of the covariance matrix of the original attributes. In general, the first
few principal components explain most of the variance in the data. Fig. 5a shows a plot of principal components vs. variance in the Eagle
Ford data. In this dataset, the first three principal components explain 90% of the variance in the data. Thus, instead of using five varia-
bles (such as TOC, porosity, quartz, clays, and carbonates) for clustering, it is sufficient if the first three principal components are used.
The other petrophysical measurements such as SRA and MICP data were not used as variables for clustering to avoid multicollinear-
ity in the data that affects clustering efficiency. For instance, SRA S1 values are highly correlated with TOC. Similarly, MICP pore vol-
ume (PV) is highly correlated with helium porosity. However, the data were evaluated later for different rock types to include them in
the integrated rock-typing process.
Clustering algorithms such as K-means and SOM (Chon and Park 2008) are then used with selected principal components to define
rock types. Before clustering, the determination of the optimal number of clusters is performed. The optimal number of clusters was
chosen to be three in this study because of the analysis presented in Fig. 5b, which plots intracluster variance (red curve) and interclus-
ter variance (green curve) as a function of the number of clusters. Intracluster variance [sum of squares within (SSW)] refers to the aver-
age variance within a cluster. Intercluster variance [sum of squares between (SSB)] refers to the distinctions between the clusters. The
point at which they start flattening out defines the optimal number of clusters. Increasing the number of clusters beyond this point does
not significantly improve the definition of rock types.
K-means clustering to identify rock types is a multistep process. In Step 1, the number of clusters is determined. This number can be
chosen by the user based on experience or decided using the SSW-SSB method explained earlier. In Step 2, the cluster centroids are
randomly assigned. In Step 3, points closest to a centroid are assigned to that group. This process is basically equivalent to reducing the
SSW variance. In Step 4, centroids are reassigned as the mean of all the data in a specific group. This process is repeated until conver-
gence is achieved. Fig. 5a shows the original data, whereas Fig. 5c shows the results of the clustering procedure for three clusters.
SOM is another form of clustering technique that is comparable to neural net analysis (Chon and Park 2008). In SOM, it is not
required to predefine the number of clusters. A 2D training map is created from the input data. The training map has nodes, and each
node has the same number of vectors as the input data. In this specific case, each data point or training map node has three vectors

4 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:47 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 5 Total Pages: 17

representing the three principal components. Next, a node is randomly selected from the training map and compared with input-data grid.
The data point on the grid that is closest to the training node is defined best-matching unit (BMU). The neighborhood of the BMU is mul-
tiplied by a weight factor to transform that area comparable to the training node. This process is repeated numerous times until conver-
gence is achieved. The convergence is achieved after the clusters develop on the data grid and do not change with subsequent iterations.

Woodford Shale Play

W19
Anadarko Basin

W18

W16, W17 Arkoma Basin

W20

W21, W22
Eddy, NM

Ardmore Basin

Fig. 3—Wells with core and log data for rock typing in Woodford; seven wells had core data (shown as red bubbles) and most were
in Anadarko basin. Core data were available for 411 depth points. Additional 12 wells (shown as black bubbles) had triple combo
logs but no core data. Rock-type logs were populated in these 12 wells for correlation with production data.

Core Data—Porosity, Mineralogy, Total


Organic Carbon

Clustering Algorithms—K-Means, Self-


Organizing Maps

Test Clusters—Source-Rock Analysis,


Nanoindentation data

Upscale rock types—Support Vector


Machines

Correlating Clusters to Production data

Fig. 4—This figure illustrates the work flow for rock typing. In Step 1, petrophysical data from the laboratory are collected. In Step
2, rock types are defined by clustering. Next, the clusters are quality checked. Then, rock types are scaled up to logs. Finally, rock-
type logs are analyzed with production data to ascertain the key rock type controlling the production.

The three clusters identified from SOM are shown in Fig. 6a. The characteristics of the three clusters are shown in the rose diagram
(or pie diagram) in Fig. 6b. Fig. 6 shows the results from Eagle Ford. From SOM clustering, Rock Type 1 (RT1) has the highest porosity
and highest TOC. It also shows that Rock Type 3 (RT3) is associated with high clays, high porosity, but low TOC. After comparing,
both SOM and K-means clustering provided very similar results. Table 1 shows the characteristics of different rock types obtained
from K-means and SOM clustering techniques for all three shale plays: Eagle Ford, Barnett, and Woodford.

2018 SPE Reservoir Evaluation & Engineering 5

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 6 Total Pages: 17

Clusters

600 800 1000 1200


3

2.0
1.5
Variances

SSB, SSW
4
1.0
0.5 3
2

2 2

200 400
PC3 1

PC2
0 0
0

–1
–3 –2 –2
Comp 1 Comp 2 Comp 3 Comp 4 Comp 5 2 4 6 8 10 12 14 1 2 1
PC1
Principal Components Number of Clusters
(a) (b) (c)

Fig. 5—K-means clustering schematic. (a) Principal component analysis. It is performed to reduce the dimensionality of the clus-
tering problem. The key principal components are taken for rock typing which explain the maximum variance in the data. (b) Deter-
mination of optimal number of clusters requires analysis of the intracluster variance (red curve) and intercluster variance (green
curve). Intracluster variance refers to the average variance within a cluster. Intercluster variance refers to the distinctions between
the clusters. The point at which the curves start flattening out defines the optimal number of clusters (in this case, rock types). (c)
Three clusters obtained using K-means plotted in three dimensions. Each axis is one principal component.

RT1 RT3

RT1 RT3

RT1

RT2

Porosity Quartz Carbonates RT2


TOC Clays

(a) (b)

Fig. 6—(a) Clusters created on an SOM map, (b) rose diagram (or pie diagram) from SOM. Rose diagram shows the characteristics of
different rock types/clusters. On comparing, the three clusters obtained from SOM are very similar to those obtained from K-means.

Eagle Ford TOC (wt%) Porosity (vol%) Carbonates (wt%) Clays (wt%) Quartz (wt%) S1 (mg/g rock)
Rock Type 1 3.5–5.5 7.0–9.0 55–70 10–30 0.0–6.0 2.0–6.0
Rock Type 2 1.5–3.0 4.5–6.5 70–85 5–15 0.0–4.0 1.5–3.5
Rock Type 3 1.5–3.0 7.0–9.0 15–35 45–70 0.0–4.0 1.0–2.0
Woodford TOC (wt%) Porosity (vol%) Carbonates (wt%) Clays (wt%) Quartz (wt%) S1 (mg/g rock)
Rock Type 1 4.0–9.0 5.5–9.0 10–20 18–30 35–50 1.0–5.0
Rock Type 2 3.0–5.0 5.5–7.0 5–15 42–55 15–30 0.0–4.0
Rock Type 3 0.5–4.5 2.0–4.0 65–90 0–10 0–10 0.0–2.0
Barnett TOC (wt%) Porosity (vol%) Carbonates (wt%) Clays (wt%) Quartz (wt%) S1 (mg/g rock)
Rock Type 1 3.5–5.5 6.0–8.0 5–15 32–40 28–36 0.7–1.1
Rock Type 2 2.5–4.5 6.0–8.0 7–18 42–52 12–20 0.3–0.8
Rock Type 3 1.5–3.5 3.5–6.0 30–55 15–30 11–20 0.3–0.7

Table 1—Characteristics of different rock types (data represent 25–75 percentile).

Scaling Up Rock Types With Classification Algorithms. Rock types were scaled up from petrophysical measurements on plugs to
logs by using SVM (Steinwart and Christmann 2008). SVM technique identifies boundaries between different clusters to separate them.
These boundaries are called hyperplanes. Examples of hyperplanes are shown in Fig. 7. Here, blue- and red-colored points belong to

6 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 7 Total Pages: 17

different clusters. For the rock-typing application, the SVM model was trained with a test dataset. Different log suites were used for dif-
ferent plays, depending on the data availability (gamma ray and resistivity in Eagle Ford, Barnett, and triple combo in Woodford). A
test dataset was prepared for each play that included depth points where both log and core data were available. A large part of the test
dataset was used to train the model, and then a prediction was made on a small portion of the test dataset. The efficiency or accuracy of
the classifier was gauged by the predictions of the rock type from the SVM classifier against those obtained with core data in the test
dataset. The SVM model with radial hyperplane gave the best results, and therefore, it was used for prediction. The trained model was
then used to predict rock types for the remaining depths in cored wells and uncored wells. The distribution of rock types along the well-
bore was then correlated with the production data to identify the key rock type controlling the production.

x2

x2

x1

x1

x3

(a) (b)

Fig. 7—Examples of hyperplane in SVM. The separation between different clusters is termed as hyperplane, and its geometry can
be linear, polynomial, or radial. (a) Shows an example of radial hyperplane. (b) Shows an example of linear hyperplane.

Results and Discussion


Core-Derived Rock Typing. The core-based rock types derived with clustering techniques were quality-checked using a priori petro-
physical knowledge. Higher values of TOC, S1, and porosity indicate higher storage and source-rock potential.
Brittleness is another key petrophysical property governing fracture propagation and well productivity in unconventional reservoirs.
Quartz is brittle, whereas clay minerals are ductile. Eq. 1 or Eq. 2, given by Jarvie et al. (2007) and Wang and Gale (2009), respectively,
was proposed to estimate brittleness index from mineralogy. However, recent publications have shown that mineralogy alone is insuffi-
cient, and a host of other parameters such as elastic and strength mechanical properties, anisotropy, pore pressure, stress regime, and
others are required for accurate estimation of brittleness and ductility (Sone and Zoback 2013). Nevertheless, it is not grossly inaccurate
to say that quartz-rich rocks are more amenable to fracturing compared with clay-rich rocks (Sone and Zoback 2013):
Qz
BIJarvie et al: ð2007Þ ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð1Þ
Qz þ Ca þ Cly
Qz þ Dol
BIWang and Gale ð2009Þ ¼ : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ð2Þ
Qz þ Ca þ Cly þ Dol
Eagle Ford. The parameters governing storage and source potential for different rock types in Eagle Ford are shown in Fig. 8.
Clearly, Rock Type 1 has the highest porosity, TOC, and S1 values. Thus, Rock Type 1 has the highest storage and source potential.
The average mineral content for the different rock types in Eagle Ford is shown in Fig. 9. All rock types have little quartz and thus can-
not be differentiated based on quartz content. However, Rock Type 3 has the highest clay percentage. Wells 10, 11, and 12 are rich in
Rock Type 3. These wells lie in detrital deltaic deposits. This explains why these wells are different and clay-rich compared with other
wells in carbonate-rich marine shore/shelf deposits. The Rock Type 3 has high porosity but very little TOC/S1. Thus, Rock Type 3 has
poor source-rock potential.

15 8
15

6
S1 (mg/gm rock)
Porosity (%)

10
10
TOC (%)

5
5
2

0 0 0
1 2 3 1 2 3 1 2 3
Rock Type Rock Type Rock Type

Fig. 8—Parameters governing storage and source potential in Eagle Ford. Clearly, Rock Type 1 has the highest porosity, TOC, and
S1 values. Porosity is a direct indicator of storage potential. High amount of TOC and S1 values generally suggests higher source-
rock potential. Thus, Rock Type 1 has the highest storage and source potential.

2018 SPE Reservoir Evaluation & Engineering 7

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 8 Total Pages: 17

100
RT1 RT2 RT3 78
80
64
57

Weight (%)
60

40
26
20
20 10
4 2 3
0
Quartz Clays Carbonates

Fig. 9—Average mineral content for different rock types in Eagle Ford. Rock Type 3 has high clay content. Wells rich in Rock Type
3 lie in East Texas Basin. This rock type has high porosity but poor source-rock potential. The minerals such as feldspar, pyrite,
and others and organic matter such as kerogen are not included in the statistics. Thus, for any rock type, the sum of quartz, clays,
and carbonates is less than 100%.

MICP data identified three distinct capillary pressure curves for three different rock types. The incremental and cumulative Hg intru-
sion plots were normalized by helium PV. The plots for the three rock types are shown in Fig. 10. Normalization of the mercury PV
with helium PV results in normalized intrusion scale between zero and one and helps determine connectivity of the sample.

Intrusion and Extrusion (RT1)


Normalized Incremental Hg

Normalized Cumulative Hg
0.030 0.70
0.025 0.60
Intrusion (RT1)

Extrusion
0.020 0.50
0.40
0.015
0.30
0.010
0.20
0.005 0.10
Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)
Intrusion and Extrusion (RT2)
Normalized Incremental Hg

Normalized Cumulative Hg

0.030 0.70

0.025 0.60
Intrusion (RT2)

0.50
0.020
0.40
0.015
0.30
0.010 Extrusion
0.20
0.005 0.10
Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)
Intrusion and Extrusion (RT3)
Normalized Incremental Hg

Normalized Cumulative Hg

0.030 0.70
0.025 0.60
Intrusion (RT3)

Extrusion
0.020 0.50
0.40
0.015
0.30
0.010
0.20
0.005 0.10
Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)

Fig. 10—Representative normalized incremental and cumulative mercury-intrusion plots for the three rock types in EF. It shows
that the connected PV decreases as we go from Rock Type 1 to Rock Type 3 and Rock Type 2. Rock Types 1 and 3 had a similar
range of helium porosity, but Rock Type 3 has lower connectivity than Rock Type 1. This is likely caused by its clay-rich nature,
where clays deposited at pore lining and pore throats reduce the connectivity.

The mercury-to-helium PV ratio, determined from normalized cumulative intrusion plot, varies between 0.6 and 0.75. The ratio
varies between 0.5 and 0.65 for Rock Type 3 samples and between 0.3 and 0.4 for Rock Type 2 samples. Thus, the connectivity
decreases as we go from Rock Type 1 to Rock Type 3 and Rock Type 2. It is interesting to note that both rock Types 1 and 3 had a sim-
ilar range of helium porosity, but Rock Type 3 has lower connectivity than Rock Type 1. This is likely caused by dispersed clay type
that deposits at pore lining and pore throats and, thereby, degrades connectivity.

8 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 9 Total Pages: 17

The shape of the capillary pressure curve also clearly distinguishes the three rock types. In Rock Type 2 samples, the normalized
incremental intrusion curve increases monotonously without reaching a plateau or an inflection point even at 60,000 psia. At 60,000
psia, equivalent pore size that the mercury could pass through is 3 nm. This shape is characteristic of very tight rocks when the domi-
nant pore size may be smaller than 3 nm.
In Rock Type 1 and Rock Type 3 samples, the capillary pressure curve exhibits a distinct maximum before 60,000 psia. The inflec-
tion point refers to the dominant pore throat. A larger value of the dominant pore throat means higher permeability. For Rock Type 1
samples, the average dominant pore-throat size was 13 nm, whereas for Rock Type 3 samples, it was 5 nm. Thus, Rock Type 1 samples
had the highest permeability. As discussed previously, Rock Type 3 samples may be affected by the presence of clay.
Barnett. The parameters governing storage and source potential for different rock types in Barnett are shown in Fig. 11. Rock Type
1 has the highest storage and source potential. The average mineral content for the different rock types in the Barnett is shown in
Fig. 12. Rock Type 1 has high-quartz content compared with the other two rock types, whereas Rock Type 2 is richest in clays. Rock
Type 1, with high storage and source-rock potential, is expected to have the largest impact on production.

12 8 1.4

10

S1 (mg/gm rock)
Porosity (%)

6
TOC (%)
8 1.0

6 4
0.6
4
2
2

0
1 2 3 1 2 3 1 2 3
Rock Type Rock Type Rock Type

Fig. 11—Parameters governing storage and source potential in Barnett. Clearly, Rock Type 1 has the highest storage and source-
rock potential.

60
RT1 RT2 RT3 49
50
43
Weight (%)

40 37
32
30
22
20 17 16 13
11
10

0
Quartz Clays Carbonates

Fig. 12—Average mineral content for different rock types in Barnett. Rock Type 1 has high quartz content compared with other two
rock types. Rock Type 2 has the highest clay percentage, whereas Rock Type 3 is richest in carbonates. Thus, different rock types
have different mineralogical compositions.

MICP data are shown in Fig. 13. In Rock Type 1 samples, the cumulative intrusion plot shows that the ratio of mercury-to-helium
volume varies between 0.65 and 0.80. In Rock Type 2 samples, this ratio varies between 0.40 and 0.60, and it varies between 0.30 and
0.45 for Rock Type 3 samples. Thus, it shows that the connected PV decreases as we go from Rock Type 1 to Rock Type 2 and Rock
Type 3. It is interesting to note that both rock Types 1 and 2 had a similar range of helium porosity, but Rock Type 2 has much lower
connectivity than Rock Type 1. Again, this is likely caused by dispersed clay type that deposits at pore lining and pore throats, and
thereby degrades connectivity.
The shape of the capillary-pressure curve in Rock Type 3 samples is characteristic of very tight rocks where the dominant pore size
may be smaller than 3 nm. Considerable hysteresis between saturating and desaturating curves is evident in both Rock Type 1 and Rock
Type 2, implying that there is real Hg intrusion into the samples. However, in Rock Type 3, cumulative intrusion curves show that satu-
rating and desaturating curves almost overlie. Lack of hysteresis is a sign of false intrusion/blank effect caused by sample compression
at high pressures. Samples exhibiting this type curve have high-calcite content and very low porosity.
For Rock Type 1 samples, the average dominant pore-throat size was 8 nm, whereas, for Rock Type 2 samples, it was 4 nm. Thus,
Rock Type 1 samples had the highest permeability. As discussed previously, Rock Type 2 samples may be affected by presence of clay.
Woodford. The parameters governing storage and source potential for different rock types in Woodford are shown in Fig. 14.
Again, Rock Type 1 has the highest porosity, TOC, and S1 values. MICP data and average mineral content for different rock types in
the Woodford are shown in Figs. 15 and 16, respectively. Rock Type 1 has high-quartz content. Again, it is expected to have the largest
impact on production because of its high storage and source-rock potential.
In Rock Type 1 samples, the cumulative intrusion plot shows that the ratio of mercury-to-helium volume varies between 0.65 and
0.80. In Rock Type 2 samples, this ratio varies between 0.50 and 0.65, and it varies between 0.40 and 0.55 for Rock Type 3 samples.
Thus, the connected PV decreases as we go from Rock Type 1 to Rock Type 2 and Rock Type 3.

2018 SPE Reservoir Evaluation & Engineering 9

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 10 Total Pages: 17

Intrusion and Extrusion (RT1)


Normalized Incremental Hg

Normalized Cumulative Hg
0.045 0.80
0.040 0.70
Extrusion

Intrusion (RT1)
0.035 0.60
0.030
0.50
0.025
0.40
0.020
0.30
0.015
0.010 0.20
0.005 0.10
Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)

Intrusion and Extrusion (RT2)


Normalized Incremental Hg

0.025

Normalized Cumulative Hg
0.80
0.70
0.020
Intrusion (RT2)

0.60
0.015 0.50
0.40
0.010 0.30 Extrusion
0.20
0.005
0.10
Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)

Intrusion and Extrusion (RT3)


Normalized Incremental Hg

0.016
Normalized Cumulative Hg
0.80
0.014 0.70
Intrusion (RT3)

0.012 0.60
0.010 0.50
0.008 0.40
0.006 0.30
0.004 0.20
Extrusion
0.002 0.10
Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)

Fig. 13—Representative normalized incremental and cumulative mercury-intrusion plots for the three rock types in Barnett. The
connected PV decreases as we go from Rock Type 1 to Rock Type 2 and Rock Type 3. Rock Types 1 and 2 had similar range of he-
lium porosity, but Rock Type 2 has lower connectivity than Rock Type 1. This is likely because of its clay-rich nature, where clays
deposited at pore lining and pore throats reduce the connectivity.

20 15
15
S1 (mg/gm rock)

15
Porosity (%)

10 10
TOC (%)

10

5 5
5

0 0 0
1 2 3 1 2 3 1 2 3
Rock Type Rock Type Rock Type

Fig. 14—Parameters governing storage and source potential in Woodford. Again, Rock Type 1 has the highest storage and source-
rock potential.

The shape of the capillary pressure curve shows that Rock Type 3 samples are characterized by high-carbonate percentage and have
a higher grain density. For Rock Type 1 samples, the average dominant pore-throat size was 6 nm, whereas, for Rock Type 2 samples,
it was 4 nm. Thus, Rock Type 1 samples had the highest permeability. As discussed previously, Rock Type 2 samples may be affected
by the presence of clay.

Extending Core-Based Classification to Well Logs. Next, the core-based classification was extended to the well-log data. This is nec-
essary to determine rock-type logs for uncored wells and uncored intervals in cored wells. Rock types were scaled up with SVM.

10 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 11 Total Pages: 17

Intrusion and Extrusion (RT1)


Normalized Incremental Hg

Normalized Cumulative Hg
0.060 0.90
0.80
0.050
0.70

Intrusion (RT1)
Extrusion
0.040 0.60
0.50
0.030
0.40
0.020 0.30
0.20
0.010
0.10 Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)

Intrusion and Extrusion (RT2)


Normalized Incremental Hg

Normalized Cumulative Hg
0.060 0.90
0.80
0.050
0.70
Intrusion (RT2)

0.040 0.60
0.50 Extrusion
0.030
0.40
0.020 0.30
0.20
0.010 Intrusion
0.10
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)

Intrusion and Extrusion (RT3)


Normalized Incremental Hg

0.060 Normalized Cumulative Hg 0.90


0.80
0.050
0.70
Intrusion (RT3)

0.040 0.60
0.50
0.030
0.40
0.020 0.30
0.20 Extrusion
0.010
0.10
Intrusion
0.000 0.00
1 10 100 1,000 10,000 100,000 1 10 100 1,000 10,000 100,000
Pressure (psi) Pressure (psi)

Fig. 15—MICP data for Woodford. The connected PV decreases as we go from Rock Type 1 to Rock Type 2 and Rock Type 3. Lower
connectivity in Rock Type 2, in spite of high helium porosity, is likely caused by high clay content.

100

RT1 RT2 RT3


77
80
Weight (%)

60
48
44
40
25 25

20 14
7 9
6

0
Quartz Clays Carbonates

Fig. 16—Average mineral content for different rock types in Woodford. Rock Type 1 has high quartz content. Rock Type 2 has the high-
est clay percentage, whereas Rock Type 3 is rich in carbonates. Thus, different rock types have different mineralogical compositions.

Eagle Ford. In Eagle Ford, gamma ray and resistivity logs were available for 6 of the 12 wells that had the core data. Thus, these
two logs were used for scaling up. The distribution of gamma ray and resistivity for different rock types is shown in Fig. 17. Rock Type
1 and Rock Type 3 both show high gamma ray possibly caused by high TOC and high clays, respectively. Also, Rock Type 1 shows a
high resistivity resulting from high oil saturation, and Rock Type 3 shows a lower resistivity resulting from high water saturation and
high clays. Thus, core and log data are consistent with each other. The rock-type logs for two sample wells (W10 on left, W6 on right)
in the Eagle Ford are shown in Fig. 18. W10 is in East Texas basin and is thus, very rich in Rock Type 3. W6 is in San Marcos basin
and is rich in Rock Type 1. W6 is expected to have a better production rate than W10.

2018 SPE Reservoir Evaluation & Engineering 11

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 12 Total Pages: 17

200
100.0

Gamma Ray (GAPI)


150

Resistivity (Ω·m)
20.0

100
5.0

50 1.0

0 0.2
1 2 3 1 2 3
Rock Type Rock Type

Fig. 17—Gamma ray and resistivity distribution for different rock types in the Eagle Ford from the depth points at which both core
and log data were available. Rock Type 1 and Rock Type 3 both show high gamma ray, possibly caused by high TOC and high
clays, respectively. Also, Rock Type 1 shows a high resistivity caused by high oil saturation, but Rock Type 3 shows a lower resis-
tivity caused by high water saturation and high clays. Thus, core and log data are consistent with each other, and logs can be
used for scaling up.

Sand - Shale (ND) RDEEP Sand - Shale (ND) RDEEP

Reference RHOB RT Reference RT


GR/Baseline (ft) GR/Baseline ILD (RT3) 4
(ft) 1.95 q/cc 2.95 (RT3) 4
GR NPHIL RDEEP RT3 GR ILD RT3
1:480 1:480 0 GAPI 150 0.2 Ω·m 200 0 0.2
0 GAPI 150 0.45 v/v –0.15 0.2 Ω·m 200 0 0.2

105.893 0.3203 2.572 2.42 EagleFord EagleFord

8,000 Rock Types


Neutron GR
Density
Resistivity

13,600

8,100

13,700

GR
Resistivity

8,200
Buda Buda Buda
False Buda False Buda False Buda False Buda

RT1 RT2 RT3

Fig. 18—Rock-type logs for two sample wells (W10 on left, W6 on right) in Eagle Ford. W10 is in East Texas Basin and is very rich
in Rock Type 3. W6 is in San Marcos Basin and is rich in Rock Type 1.

Barnett. In Barnett, again, gamma ray and resistivity were the only logs available in two of the three wells that had the core data.
Thus, these two logs were used for scaling up. Fig. 19 shows the distribution of gamma ray and resistivity for different rock types.
Rock Type 1 has low density, high gamma ray, and high neutron porosity, consistent with high TOC and high porosity measured in the
laboratory. Rock Type 3, on the other hand, has the highest density and lowest neutron porosity, consistent with high carbonates in labo-
ratory-measured mineralogy. The rock-type logs for two sample wells (W13 on left, W14 on right) are shown in Fig. 20. Of lower and
upper Barnett, lower Barnett is richer in Rock Type 1. Thus, lower Barnett has higher TOC and quartz compared with upper Barnett.
Woodford. In Woodford, triple combo logs were available for all 7 wells that had core data. Thus, gamma ray, neutron, density, and
resistivity logs were used for scaling up. Fig. 21 shows the distribution of gamma ray, neutron porosity, and density logs for different
rock types. The logs were consistent with the core data, and were found adequate for scaling up the rock types. Fig. 22 shows the rock-
type logs for two sample wells (W16 on left, W18 on right).

12 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 13 Total Pages: 17

300
200.0

Gamma Ray (GAPI)

Resistivity (Ω·m)
200
20.0

5.0
100

1.0
50

0 0.2
1 2 3 1 2 3
Rock Type Rock Type

Fig. 19—Gamma ray and resistivity distribution for different rock types in Barnett. Rock Type 1 show high gamma ray and resistiv-
ity caused by high TOC.

Reference RT Reference RT
GR/Baseline ILD GR/Baseline ILD
(ft) (RockType) 4
(ft) (RockType) 4
GR ILD RockType GR ILD RockType
1:2000 0 GAPI 200 0.2 Ω·m 200 0 0.2 1:2000 Ω·m
0 GAPI 150 0.2 200 0 0.2

Marble Falls Marble Falls Marble Falls


Marble Falls Marble Falls Marble Falls

7,500
Upper Barnett Upper Barnett Upper Barnett
7,000 38.36 38.914

7,750
Upper Barnett Upper Barnett Upper Barnett
7,250
Forestburg LM Forestburg LM Forestburg LM
Forestburg LM Forestburg LM Forestburg LM

8,000 Lower Barnett Lower Barnett Lower Barnett


Lower Barnett Lower Barnett Lower Barnett

7,500

8,250

7,750 Ordovician Ordovician

Ordovician Ordovician

8,500

Fig. 20—Rock-type logs for two sample wells (W13 on left, W14 on right) in Barnett. Of lower and upper Barnett, lower Barnett is
richer in Rock Type 1. Thus, lower Barnett has higher TOC and quartz compared with upper Barnett.

Relating Rock Types to Production Data. In the three shale plays, Rock Type 1 has the highest storage and source-rock potential.
Thus, Rock Type 1 is expected to be the key driver of the production. A Rock Type 1 ratio (RTR) was created by dividing the Rock
Type 1 thickness with the gross thickness (i.e., RT1 þ RT2 þ RT3) for all the wells. This was then correlated with normalized produc-
tion. A positive correlation between the two was seen in all three shale plays that bolstered the confidence in the validity of the rock-
typing exercise.
Eagle Ford. The spatial location of the wells for which rock types were scaled up is shown in Fig. 1. Red bubbles represent wells
with cores, and black bubbles represent additional wells that did not have the core data. All the wells were horizontal wells, and their
lateral length varied from 600 to 6,000 ft. The production was normalized by the lateral lengths. The comparison of RTR with normal-
ized production is shown in Fig. 23. Normalized production here refers to the first 24 months’ cumulative barrel of oil equivalent
(BOE) normalized by the lateral lengths. A strong positive correlation was seen, suggesting that Rock Type 1 is the key rock type con-
trolling the production. There are some outliers that are expected, caused by different completion treatments and other factors.

2018 SPE Reservoir Evaluation & Engineering 13

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 14 Total Pages: 17

400

Neutron Porosity (Fraction)


2.8 0.30

Gamma Ray (GAP)


Density (gm/cm3) 2.7 300

2.6 0.20
200
2.5
0.10
2.4 100

2.3
0 0.00
1 2 3 1 2 3 1 2 3
Rock Type Rock Type Rock Type

Fig. 21—Gamma ray, density, and neutron logs distribution for different rock types in Woodford. Rock Type 1 has low-density, high
gamma ray, and high neutron porosity that are consistent with high TOC and high laboratory-measured porosity. Rock Type 3, on
the other hand, has highest density and lowest neutron porosity, which are consistent with high carbonates in laboratory-meas-
ured mineralogy.

Sand-Shale (ND) Sand-Shale (ND)

Reference GR/Baseline NPHI RT Reference GR/Baseline RHOB SHIFT RT


–0.15 CFCF 0.45
AT90 (RockType) 4
AT90
1.95 G/C3 2.95 (RockType) 4
(ft) GR RHOB AT90 RockType (ft) GR AT90 RockType
NPHI
1:250 0 GAPI 400 1.95 G/C 2.95 0.2 Ω·m 2000 0
3
0.2 1:250 0 GAPI 400 –0.15 CFCF 0.45 0.2 Ω·m 2000 0 0.2
Woodford Shale Woodford Shale Woodford Shale Woodford Shale

16,150 Woodford Shale Woodford Shale Woodford Shale Woodford Shale

12,500

16,175
12,525

16,200 84.883 0.0888 164.914


Middle_WF Middle_WF Middle_WF Middle_WF 2.5171
12,550

16,225
12,575

Middle_WF Middle_WF Middle_WF Middle_WF


16,250
12,600

16,275
Lower_WF Lower_WF Lower_WF Lower_WF
12,625

16,300
12,650

16,325
12,675

HuntonLM HuntonLM HuntonLM HuntonLM HuntonLM HuntonLM HuntonLM HuntonLM


16,350

Fig. 22—Rock-type logs for two sample wells (W16 on left, W18 on right) in Woodford.

Barnett. The spatial location of wells for which rock types were scaled up is shown in Fig. 2. All the wells were vertical wells.
Fig. 24 shows the comparison of RTR with normalized production. Normalized production here refers to the first 24 months of cumula-
tive gas normalized by the zone thickness. A strong positive correlation was seen, suggesting that Rock Type 1 is the key rock type con-
trolling the production.
There are multiple trends evident in Fig. 24. The set of wells lying along the green trend line shows high productivity, whereas wells
lying along the red trend line show relatively lower productivity. The wells with high productivity are from different counties: Denton,
Wise, and Parker. The commonality among these high-productivity wells is that they were all completed by one operator. The wells
lying along the red trend line were completed by other operators. Thus, it appears that the multiple trends in Fig. 24 can be attributed to
different completion practices used by various operators.
Woodford. The same exercise was conducted in Woodford wells. Fig. 3 shows the location of the wells. Some of the wells were
horizontal; some were vertical. Fig. 25 shows the comparison of RTR with normalized production. Vertical well production was nor-
malized by zone thickness and horizontal-well production by lateral length. A positive correlation between normalized production and
RTR is evident.

14 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 15 Total Pages: 17

1,000

BOE/Per Feet Lateral (bbl/ft)


100

10

1
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
RTR [RT1/(RT1+RT2+RT3)]

Fig. 23—Normalized production correlated with the rock-type ratio (RTR) in Eagle Ford. A strong positive correlation is evident,
suggesting Rock Type 1 is the key rock type controlling the production.

1000
Cumulative Gas (24 Months)

800
Per Feet (bbl/ft)

600

400

200

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6
RTR [RT1/(RT1+RT2+RT3)]

Fig. 24—Normalized production correlated with the RTR in Barnett. A strong, positive correlation is evident, suggesting Rock Type
1 is the key rock type controlling the production. The set of wells lying along the green trend line shows high productivity, whereas
wells lying along the red trend line show relatively lower productivity. The commonality among these high-productivity wells is
that they were all completed by one operator. Thus, it appears that the multiple trends in the figure can be attributed to different
completion practices used by various operators.
BOE/Per Feet Lateral (bbl/ft)

100

80

60

40

20

0
0.0 0.2 0.4 0.6 0.8 1.0
RTR [RT1/(RT1+RT2+RT3)]
(a)

1,000
BOE/Per Feet (bbl/ft)

800

600

400

200

0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
RTR [RT1/(RT1+RT2+RT3)]
(b)

Fig. 25—Normalized production correlated with the RTR in Woodford. (a) Correlation plot for horizontal wells. (b) Correlation plot
for vertical wells. A positive correlation is evident on both the plots, suggesting that Rock Type 1 is the key rock type controlling
the production.

2018 SPE Reservoir Evaluation & Engineering 15

ID: jaganm Time: 12:48 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 16 Total Pages: 17

Conclusions
The current paper develops an integrated work flow for rock typing by using laboratory measurements and well logs. The work flow
correlates the core-based rock types with available logs and generates rock-type logs. These logs show a strong correlation with the pro-
duction from the well. Data-mining algorithms, such as K-means, SOM, SVM, and others, are very powerful in handling large amounts
of data and finding meaningful associations between different data types. The rock types were scaled up to 95 wells at over a 20,000-ft
depth interval. The additional benefit is that the work flow is mainly automated, making the rock-typing exercise rapid.
Rock Type 1 is the best reservoir rock, whereas Rock Type 3 is the worst. Rock Type 1 has the highest storage, source-rock poten-
tial, and connectivity. Rock Type 1 can aid in identifying the sweet spots and optimal-reservoir-quality zones.
Some other applications of the rock typing that are routinely conducted in conventional reservoirs and that will benefit unconven-
tional reservoirs are 3D reservoir modeling, sweet-spot identification along with seismic data, new well locations, and improved volu-
metric estimates.

Nomenclature
CO2 ¼ Carbon-di-oxide
Hg ¼ mercury
r35 ¼ Dominant Pore Throat radius at 35% Mercury saturation, mm
Ro ¼ Vitrinite reflectance, percent

Acknowledgment
We would like to thank Devon, Cimarex, Shell, ConocoPhillips, BHP, Pioneer, Apache, and Drilling Info for providing the cores and
logs to carry out the study. Their generous contribution has made this study possible.

References
Amaefule, J. O., Altunbay, M., Tiab, D. et al. 1993. Enhanced Reservoir Description: Using Core and Log Data to Identify Hydraulic (Flow) Units and
Predict Permeability in Uncored Intervals/Wells. Presented at the SPE Annual Technical Conference and Exhibition, Houston, 3–6 October. SPE-
26436-MS. https://doi.org/10.2118/26436-MS.
Aranibar, A., Saneifar, M., and Heidari, Z. 2013. Petrophysical Rock Typing in Organic-Rich Source Rocks Using Well Logs. Presented at the Uncon-
ventional Resources Technology Conference, Denver, 12–14 August. URTeC 1619574. https://doi.org/10.1190/URTeC2013-117.
Ballard, B. D. 2007. Quantitative Mineralogy of Reservoir Rocks Using Fourier Transform Infrared Spectroscopy. Presented at the SPE international Stu-
dent Paper Contest, Anaheim, California, USA, 11–14 November. SPE-113023-STU. https://doi.org/10.2118/113023-STU.
Breyer, J. A., Denne, R., Funk, J. et al. 2013. Stratigraphy and Sedimentary Facies of the Eagle Ford Shale (Cretaceous) Between the Maverick Basin
and the San Marcos Arch, Texas. Presented at the AAPG Annual Convention and Exhibition, Pittsburgh, Pennsylvania, USA, 19–22 May.
Bruner, K. and Smosna, R. 2011. A Comparative Study of the Mississippian Barnett Shale, Fort Worth Basin, and Devonian Marcellus Shale, Appala-
chian Basin. DOE/NETL-2011/1478, US DOE, April 2011.
Caldwell, C. D. and Johnson, P. G. 2013. Anadarko Woodford Shale: Improving Production by Understanding Lithologies/Mechanical Stratigraphy and
Optimizing Completion Design. Oral presentation at the AAPG Education Directorate Woodford Shale Forum, Oklahoma City, Oklahoma, USA, 11
April 2013.
Callantine, S. 2010. Eagle Ford Shale Play. Investor Report, Chesapeake Energy, Oklahoma City, Oklahoma, USA, August 2010.
Chon, T. S., and Park, Y. S. 2008. Self-Organizing Map. In Encyclopedia of Ecology, ed. Sven Erik Jorgensen and Brian D. Fath. Oxford: Academic
Press, pp. 3203–3210. https://doi.org/10.1016/B978-008045405-4.00907-1.
CLR. 2010. Anadarko Woodford: The SCOOP, Internal Report, Continental Resources Limited, Oklahoma City, Oklahoma, USA.
Coates, G. R. and Dumanoir, J. L. 1974. A New Approach to Improved Log-Derived Permeability. The Log Analyst (January–February 1974).
Comer, J. B. 2005. Facies Distribution and Hydrocarbon Production Potential of Woodford Shale in the Southern Midcontinent. Presented at the Unconven-
tional Energy Resources in the Southern Midcontinent Symposium, Norman, Oklahoma, USA, Oklahoma Geological Survey, Circular 110, pp. 51–62.
Cortes, C. and Vapnik, V. 1995. Support-Vector Networks. Machine Learning 20 (3): 273–297. https://doi.org/10.1007/BF00994018.
Davies, D. K., Williams, B. P. J., and Vessell, R. K. 1993. Reservoir Geometry and Internal Permeability Distribution in Fluvial, Tight, Gas Sandstones,
Travis Peak Formation, Texas. SPE Res Eng 8 (1): 7–12. SPE-21850-PA. https://doi.org/10.2118/21850-PA.
EOG Resources. 2010. South Texas Eagle Ford. Internal Report. Houston: EOG Resources.
Gupta, N. 2012. Multi-scale Characterization of the Woodford shale in West-central Oklahoma From Scanning Electron Microscope to 3D Seismic. PhD
thesis, Oklahoma University, Norman, Oklahoma.
Gupta, N., Rai, C., and Sondergeld, C. 2013. Petrophysical Characterization of Woodford Shale. Petrophysics 54 (4): 368–382. SPWLA-2013-v54n4-A4.
Gupta, I., Rai, C., Sondergeld, C. et al. 2017. Rock Typing in Wolfcamp Formation. Presented at the 58th Annual SPWLA Symposium, Oklahoma City,
Oklahoma, USA, 17–21 June. SPWLA-2017-D.
Hotelling, H. 1933. Analysis of a Complex of Statistical Variables Into Principal Components. Journal of Educational Psychology 24: 417–441, 498–520.
Jarvie, D. M., Hill, R. J., Ruble, T. E. et al. 2007. Unconventional Shale-gas Systems: The Mississippian Barnett Shale of North-Central Texas as One
Model for Thermogenic Shale-Gas Assessment. AAPG Bull. 91: 475–499. https://doi.org/10.1306/12190606068.
Kale, S. 2009. Petrophysical Characterization of Barnett Shale Play, MS thesis, Oklahoma University, Norman, Oklahoma, USA.
Kale, S., Rai, C., and Sondergeld, C. 2010. Rock Typing in Gas Shales. Presented at the SPE Annual Technical Conference, Florence, Italy, 19–22 Sep-
tember. SPE 134539-MS. https://doi.org/10.2118/134539-MS.
Karastathis A. 2007. Petrophysical Measurements on Tight Gas Shale, MS thesis, University of Oklahoma, Norman, Oklahoma, USA.
Kinley, T. 2008. Hydrocarbon Potential of the Barnett Shale (Mississippian), Delaware Basin, West Texas, and Southeastern New Mexico. AAPG 92:
967–991. https://doi.org/10.1306/03240807121.
Kohonen, T. and Honkela, T. 2007. Kohonen Network. Scholarpedia. https://doi.org/10.4249/scholarpedia.1568.
Law, C. 1999. Evaluating Source Rocks, Chap. 6, AAPG special volumes. In Petroleum Geology/Handbook of Petroleum Geology: Exploring for Oil
and Gas Traps, ed. E. A. Beaumont and N. H. Foster. pp. 6-1–6-41.
Li, H., Hart, B., Dawson, M. et al. 2015. Characterizing the Middle Bakken: Laboratory Measurement and Rock Typing of the Middle Bakken Formation.
Presented at the Unconventional Technology Conference, San Antonio, Texas, 20–22 July. SPE-178676-MS. https://doi.org/10.2118/178676-MS.
MacQueen, J. B. 1967. Some Methods for Classification and Analysis of Multivariate Observations. Proc., 5th Berkeley Symposium on Mathematical
Statistics and Probability, 21 June–18 July 1965. University of California Press. pp. 281–297.

16 2018 SPE Reservoir Evaluation & Engineering

ID: jaganm Time: 12:49 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054


REE189968 DOI: 10.2118/189968-PA Date: 22-March-18 Stage: Page: 17 Total Pages: 17

Pittman E. D. 1992. Relationship of Porosity and Permeability to Various Parameters Derived From Mercury Injection–Capillary Pressure Curves for
Sandstones. AAPG Bull. 76 (2): 191–198. https://doi.org/10.1306/BDFF87A4-1718-11D7-8645000102C1865D.
Pollastro, R. M., Jarvie, D. M., Hill, R. J. et al. 2007. Geologic Framework of the Mississippian Barnett Shale, Barnett-Paleozoic Total Petroleum Sys-
tem, Bend Arch–Fort Worth Basin, Texas. AAPG Bull. 91 (4): 405–436. https://doi.org/10.1306/10300606008.
Randolph, P. L., Soeder, D. J., and Chowdiah, P. 1984. Porosity and Permeability of Tight Sands. Presented at the Unconventional Gas Recovery Sympo-
sium, Pittsburgh, Pennsylvania, USA, 13–15 May. SPE-12836-MS. https://doi.org/10.2118/12836-MS.
Rushing, J. A., Newsham, K. E., and Blasingame, T. A. 2008. Rock Typing: Keys to Understanding Productivity in Tight Gas Sands. Presented at the
Unconventional Reservoirs Conference, Keystone, Colorado, USA, 10–12 February. SPE-114164-MS. https://doi.org/10.2118/114164-MS.
Sarmiento, M. F. R., Ducros, M., Carpentier, B. et al. 2013. Quantitative Evaluation of TOC, Organic Porosity and Gas Retention Distribution in a Gas
Shale Play Using Petroleum System Modeling: Application to the Barnett Shale. Marine and Petroleum Geology 45: 315–330. https://doi.org/
10.1016/j.marpetgeo.2013.04.003.
Schieber, J. 1996. Early Diagenetic Silica Deposition in Algal Cysts and Spores; A Source of Sand in Black Shales. Journal of Sedimentary Research 66
(1): 175–183. https://doi.org/10.1306/D42682ED-2B26-11D7-8648000102C1865D.
Singh P. 2008. Lithofacies and Sequence Stratigraphic Framework of the Barnett Shale, Northeast Texas. PhD dissertation, University of Oklahoma,
Norman, Oklahoma, USA.
Slatt, R., Singh P., Borges, G. et al. 2008. Reservoir Characterization of Unconventional Gas Shales: Example From the Barnet Shale, Texas. Presented
at the AAPG Annual Convention, San Antonio, Texas, 20–23 April. Search and Discovery Article #30075.
Sondergeld, C. H. and Rai, C. S. 1993. A New Concept of Quantitative Core Characterization. The Leading Edge 12 (7): 774–779. https://doi.org/
10.1190/1.1436968.
Sondhi, N. 2011. Petrophysical Characterization of the Eagle Ford Shale, MS thesis, Oklahoma University, Norman, Oklahoma, USA.
Sone, H. and Zoback, M. D. 2013. Mechanical Properties of Shale-Gas Reservoir Rocks—Part 2: Ductile Creep, Brittle Strength, and Their Relation to
the Elastic Modulus. Geophysics 78 (5): D393–D402. https://doi.org/10.1190/GEO2013-0051.1.
Steinwart, I. and Christmann, A. 2008. Support Vector Machines, Book, Information Science and Statistics.
Thomeer, J. H. 1983. Air Permeability as a Function of Three Pore Network Parameters. J Pet Technol 35 (4): 809–814. SPE-10922-PA. https://doi.org/
10.2118/10922-PA.
Tian, Y. 2014. Occurrence of Multiple Fluid Phases Across a Basin, in the Same Shale Gas Formation–Eagle Ford Shale Example. MS thesis, Texas
A&M, College Station, Texas.
Timur, A. 1968. An Investigation of Permeability, Porosity, and Residual Water Saturation Relationship for Sandstone Reservoirs. The Log Analyst. 9
(4). SPWLA-1968-vIXn4a2.
Tissot, B., P. and Welte, D., H. 1984. Petroleum Formation and Occurrence, second edition. New York: Springer-Verlag, 699 p.
Tuttle, S. 2010. Oil Resource Plays–Examples and Technology. Geological Review, Internal Report, Petrohawk Energy Corporation, Houston, Texas
(September 2010).
Walls, J. D. 1982. Tight Gas Sands Permeability, Pore Structure, and Clay. J Pet Technol 34 (11): 2708–2714. SPE-9871-PA. https://doi.org/10.2118/
9871-PA.
Wang, F. P. and Gale, J. F. W. 2009. Screening Criteria for Shale-gas Systems: Gulf Coast Association of Geological Societies. Trans. 59: 779–794.

Ishank Gupta earned a BS degree in petroleum engineering from University of Petroleum and Energy Studies in 2009. He worked
as a reservoir engineer in Schlumberger from 2010 to 2015. Gupta then finished his MS degree in petroleum engineering from the
University of Oklahoma. He is currently a PhD candidate at the University of Oklahoma. Gupta’s research interests are numerical
simulation, reservoir characterization, and unconventional petrophysics.
Chandra Rai completed his MS degree in geophysics from the Indian School of Mines in 1971. He also completed his PhD degree
in geology and geophysics from the University of Hawaii in 1977. Rai worked as a technology director for Amoco for 18 years. He
is currently director of Mewbourne School of Petroleum and Geology Engineering department and professor at the University of
Oklahoma. Rai teaches petrophysics, petrophysics laboratory, seismic reservoir modeling, unconventional reservoirs, and well
logging. His research interests include rock physics, petrophysics, and geomechanics with an emphasis on unconventional shale
gas and oil reservoirs.
Carl Sondergeld completed his MA degree in geology from Queens College in 1972. He also completed his PhD degree in geo-
physics from Cornell University in 1977. Sondergeld is currently a professor and holds the Curtis Mewbourne Chair at the University
of Oklahoma. He teaches petrophysics, petrophysics laboratory, technical communications, seismic reservoir modeling, uncon-
ventional reservoirs, introduction to petroleum engineering, and well logging. Sondergeld coaches the SPE Petrobowl team and
carries out research in the areas of rock physics, petrophysics, and geomechanics, with an emphasis on unconventional shale
gas and oil reservoirs.
Deepak Devegowda completed his BS degree in electrical and electronics engineering from the Indian Institute of Technology.
He also completed his PhD degree in petroleum engineering from Texas A&M University in 2008. Devegowda is currently an asso-
ciate professor of petroleum engineering at the University of Oklahoma. His research interests include unconventional oil and
gas reservoir engineering, high-resolution reservoir characterization, and production optimization.

2018 SPE Reservoir Evaluation & Engineering 17

ID: jaganm Time: 12:49 I Path: S:/REE#/Vol00000/170054/Comp/APPFile/SA-REE#170054

You might also like