Environ Earth Sci (2014) 71:2739–2752
DOI 10.1007/s12665-013-2653-z
ORIGINAL ARTICLE
Non-parametric simulations-based conditional stochastic
predictions of geologic heterogeneities and leakage potentials
for hypothetical CO2 sequestration sites
Weon Shik Han • Kue-Young Kim •
Sungwook Choung • Jina Jeong •
Na-Hyun Jung • Eungyu Park
Received: 19 August 2012 / Accepted: 30 June 2013 / Published online: 16 July 2013
Ó Springer-Verlag Berlin Heidelberg 2013
Abstract The present study focuses on understanding the
leakage potentials of the stored supercritical CO2 plume
through caprocks generated in geostatistically created
heterogeneous media. For this purpose, two hypothetical
cases with different geostatistical features were developed,
and two conditional geostatistical simulation models (i.e.,
sequential indicator simulation or SISIM and generalized
coupled Markov chain or GCMC) were applied for the
stochastic characterizations of the heterogeneities. Then,
predictive CO2 plume migration simulations based on
stochastic realizations were performed and summarized. In
the geostatistical simulations, the results from the GCMC
model showed better performance than those of the SISIM
model for the strongly non-stationary case, while SISIM
models showed reasonable performance for the weakly
non-stationary case in terms of low-permeability lenses
characterization. In the subsequent predictive simulations
Electronic supplementary material The online version of this
article (doi:10.1007/s12665-013-2653-z) contains supplementary
material, which is available to authorized users.
W. S. Han  N.-H. Jung
Department of Geosciences, University
of Wisconsin-Milwaukee, Milwaukee, USA
K.-Y. Kim
Korea Institute of Geoscience and Mineral Resources,
Daejeon, Republic of Korea
S. Choung
Division of Advanced Nuclear Engineering, Pohang University
of Science and Technology, Pohang, Republic of Korea
J. Jeong  E. Park (&)
Department of Geology, Kyungpook National University,
1370 Sangyeok-dong, Buk-gu, Daegu, Republic of Korea
e-mail: egpark@knu.ac.kr
of CO2 plume migration, the observations in the geosta-
tistical simulations were confirmed and the GCMC-based
predictions showed underestimations in CO2 leakage in the
stationary case, while the SISIM-based predictions showed
considerable overestimations in the non-stationary case.
The overall results suggest that: (1) proper characterization
of low-permeability layering is significantly important in
the prediction of CO2 plume behavior, especially for the
leakage potential of CO2 and (2) appropriate geostatistical
techniques must be selectively employed considering the
degree of stationarity of the targeting fields to minimize the
uncertainties in the predictions.
Keywords Geologic CO2 sequestration  Geostatistics 
Leakage  Sequential indicator  Markov chain
Introduction
Prior to commercial-scale CO2 injection, various preceding
assessments from laboratory- to pilot-scale verifications
and numerical simulations have been conducted by inter-
national scientific and engineering communities in pursuit
of (1) safe and secure storage (Audigane et al. 2007; Han
et al. 2010b; Juanes et al. 2006), (2) short- and long-term
monitoring (Bateson et al. 2008; Emberley et al. 2004;
Lewicki et al. 2005; Newell et al. 2008; Wells et al. 2007),
and (3) risk assessment (Kopp et al. 2010; Rohmer and
Bouc 2010; Siirila et al. 2012; Smith et al. 2011). The
majority of these studies reached similar conclusions,
suggesting that sound characterizations of the subsurface
heterogeneity are the most important requirement for
constructing conceptual models and conducting associated
analyses (Ambrose et al. 2008; Han et al. 2010a; Hovorka
et al. 2004; Jahangiri and Zhang 2011; Lengler et al. 2010;
123
2740
Li et al. 2011). Nevertheless, the detailed characterization
of such complex subsurface heterogeneities at the site-wide
scale is often constrained due to issues related to the project
resources and budgets. For this reason, investigators often
rely on limited numbers of hard data collected within the
boundaries of the sites (e.g., withdrawn borehole core or
logging data). To overcome such limitations at the site-
wide scale, geostatistics and the associated simulation
methods could be applied. The uncertainties associated
with subsurface characterizations play important roles in
the sound predictions, which suggest incorporation of field-
collected borehole petrophysical and geophysical survey
data with the stochastic analyses techniques. Recently,
researchers in the field of geological CO2 sequestration
have started to acknowledge the importance of the geo-
statistical simulations to better grasp the CO2 plume dis-
tribution, trapping, and leakage through preferential
pathways in geological heterogeneity (Table 1).
A number of stochastic geostatistical simulation models
are applicable including sequential Gaussian simulation
(SGSIM), turning band simulation (TB), sequential indi-
cator simulation (SISIM), and Markov chain-based SISIM
(Carle and Fogg 1996; Deutsch and Journel 1998; Man-
toglou and Wilson 1982). Recently, the indicator-based
generalized coupled Markov chain (GCMC) model has
been proposed as a viable multidimensional option where
the directional asymmetry distribution of the subsurface
media can be handled efficiently (Park 2010; Park et al.
2007).
The aforementioned simulation tools can be divided into
parametric (e.g., TB, SGSIM) and non-parametric (e.g.,
SISIM, GCMC) groups. The parametric simulation meth-
ods assume that subsurface media include a non-extreme
value and, concurrently, the property transitions in space
are relatively smooth (Wen and Kung 1993). The optimal
sites for geological CO2 sequestration should not have
smooth changes in the media properties. Rather, they are
expected to have abrupt changes in permeability (k) or
porosity (/), in which a higher k and / formation lies
under a lower k and / layer serving as a capping formation.
Therefore, the geologic setting with two extreme para-
metric distributions is characterized as an optimal site for
geologic CO2 sequestration. In such a case, indicator-based
non-parametric simulations are recommended because the
methods do not assume Gaussian or smooth transitions of
the media properties.
The present study emphasizes understanding the use-
fulness of the stochastic and non-parametric geostatistical
characterization due to the nature of abrupt transitions in
targeted and capping formation. For this purpose, two
representative non-parametric geostatistical simulation
models, SISIM, and the recently developed model of
GCMC were chosen. To compare the heterogeneity fields
123
Environ Earth Sci (2014) 71:2739–2752
developed from these two geostatistical models, two
hypothetical cases describing the CO2 injection formation
under the low-k caprock were designed.
Non-parametric geostatistical models: generalized
coupled Markov chain and sequential indicator
simulations
Often the sequence of sedimentary rock strata is shifted in
time. From a geostatistical point of view, the variable
sequence of the sedimentary profiles may be described as
having non-stationarity. Also, the lithologic transition in
the positive and the negative vertical directions are mutu-
ally incompatible and terminated in the different litholog-
ical sequences, in general. Park (2010) defined these
observations as expressing directional non-stationarity. For
the characterizations of a deep sedimentary basin, the
lithological non-stationarity and the directional non-sta-
tionarity in the transitions are equally important and,
therefore, must be considered.
Figures 1a and 2a show distributions of three rocks in
two deterministic cases (cases 1 and 2), in which the rock
with medium k (RT1: 10-14 m2) occupies the upper half
and the rock with high k (RT2: 10-13 m2) occupies the
lower half of the domain. The low-k rocks (RT3: 10-16 m2)
with elongated lens shape are distributed around the inter-
face between the high- and medium-k rocks. The model
domains for two cases are identical with 4,000 m 9 100 m
size and are uniformly discretized by 40 m 9 1 m grids.
The primary difference between cases 1 and 2 is the size
of the elongated lenses and their distribution patterns. In
case 1, representing a weakly non-stationary field, the
distributions of low-k lenses were concentrated around the
mid-x and -y locations, where only four wide-ranged lenses
existed (Fig. 1a). On the other hand, in case 2 representing
a stationary field, the scales of the low-k lenses were short,
and they were distributed across the entire range along the
x-direction. In these heterogeneity fields, artificial bore-
holes are assigned in which the x-positions are randomly
selected (Figs. 1b, 2b). To further mimic actual borehole
drilling, the maximum depths of the boreholes are also
chosen randomly. For case 1, totally 10 boreholes are
collected for conditioning data and, for case 2, 15 bore-
holes are used considering the increased heterogeneity of
the structures.
Table 2 shows the one-step transition probabilities (TPs)
matrices and the approximated mean lengths (lk ) of three
rocks along the positive/negative x- and y-directions
obtained from the nearest neighborhood interpolation from
the collected borehole data in case 1 (Fig. 1b). The lk of
each rock is acquired from Carle and Fogg (1996) and Park
(2010);
Environ Earth Sci (2014) 71:2739–2752 2741

Table 1 Summary of previous studies adapting geostatistical techniques in geologic CO2 sequestration related field
Geostatiscal approaches Heterogeneity Approaches Flow simulator Study objectives
field

Han et al. Sequential Gaussian (GSLIB: Uncorrelated Multiple CMG’s GEM (CMG 2007) Understanding CO2 trapping
(2010a) Deutsch and Journel 1998) random field realizations mechanisms (mobile,
Various residually, dissolution-, and
corrected mineral-trapped CO2)
random field
Flett et al. Sequential Gaussian and Binary Single CHEARS Understanding CO2 trapping
(2007) sequential indicator uncorrelated realization mechanisms (mobile,
(GSLIB: Deutsch and random field residual-, and dissolution-
Journel 1998) trapped CO2)
Doughty and Markov chain (TProGS: Fluvial deltaic Single TOUGH2 (Pruess et al. Understanding spatial CO2
Pruess Carle and Fogg 1996, 1997) Frio formations realization 1999) plume distribution
(2004)
Jahangiri and Sequential Gaussian (GSLIB: Uncorrelated Multiple Eclipse 300 (Schlumberger Understanding spatial CO2
Zhang Deutsch and Journel 1998) random field realizations 2008) plume distribution
(2011) Various
corrected
random fields
Lengler et al. Sequential Gaussian Uncorrelated Multiple TOUGH2 (Pruess et al. Understanding CO2 trapping
(2010) (Pebesma 2004) random field realizations 1999) mechanisms (mobile and
Various dissolution-trapped CO2)
corrected
random fields
Pau et al. Uncorrelated In-house code and Understanding the onset time
(2010) random field TOUGH-MP (Zhang of dissolved CO2
et al. 2008) convection
Ide et al. Sequential Gaussian Correlated Single ECLIPSE 100 Understanding spatial CO2
(2007) random field realization (Schlumberger 2005) plume distribution
Zhang et al. Sequential Gaussian (GSLIB: Correlated Multiple ITOUGH (Finsterle, 2004) Testing the methodology of
(2011) Deutsch and Journel 1998) random field realizations estimating residually
trapped CO2 using a single-
well injection withdrawal
Farajzadeh Fourier transform (Bruining Correlated Multiple In-house code Identifying three flow
et al. (2011) et al. 1997) random field realizations regimes (fingering,
dispersive, and channeling)
for dissolved CO2
convection
Kumar et al. Fourier transform (Jennings Uncorrelated Single CMG’s GEM (CMG 2007) Understanding CO2 trapping
(2005) et al. 2000) random field realization mechanisms (mobile,
residually, and dissolution-
trapped CO2)

  
lk ¼ 1 op0kk oh h¼0  Dx=ð1  pkk Þ ðk ¼ 1; 2; 3Þ;
where p0 kk is continuous and pkk is the one-step recurrent
TP from rock type k to k. Along the positive and negative
x-directions, the directional non-stationarity is not severe
and, therefore, the diagonal components of the positive and
negative TPs are almost identical. In the horizontal direc-
tion, RT1 and RT2 do not communicate in both ways,
while RT2 has the largest self-TP and RT3 has the small-
est. The differences between the positive and negative TPs
are significant for the y-direction due to the directional non-
stationarity. Therefore, the diagonal as well as off-diagonal
components of the TPs from the positive and negative
ð1Þ y-directions differ considerably. As one of the most obvi-
ous examples, the TP from RT2 to RT1 is zero to positive
y-direction, while the TP from RT1 to RT2 is zero to the
negative y-direction. In case 1, the lk of RT1 and RT2 are
7,214.5 and 8,650.7 m, respectively, which are about twice
the domain length along the x-direction. On the other hand,
the lk of RT3 is 926 m, similar to the actual mean length of
RT3 measured from the original data (=877.3 m) in
Fig. 1a. In the y-direction, there is a discrepancy in the lk of
RT1 and RT2 measured along the positive and negative
directions. The estimated lk of RT1 and RT2 in the positive
y-direction are 41.1 and 135.4 m, respectively, while those

123
2742 Environ Earth Sci (2014) 71:2739–2752

Fig. 1 a Hypothetical and

deterministic k field of case 1,
which includes upper formation
(RT1: 10-14 m2, red color),
lower formation (RT2:
10-13 m2, green color), and
low-k lenses (RT3: 10-16 m2,
blue color). Case 1
characterizes the heterogeneous
k field where the low-k lenses
extend laterally long. b The
collected k values at the
randomly penetrated boreholes

Fig. 2 a Hypothetical and

deterministic k field of case 2,
which includes upper formation
(RT1: 10-14 m2, red color),
lower formation (RT2:
10-13 m2, green color), and
low-k lenses (RT3: 10-16 m2,
blue color). Case 2
characterizes the heterogeneous
k field where the low-k lenses
extend laterally short. b The
collected k values at the
randomly penetrated boreholes

in the negative y-direction are 255.1 and 35.5 m, respec-
tively. These inconsistencies indicate that there is strong
directional non-stationarity in the vertical y-direction. The
embedded RT3 shows an identical lk of 4.0 m in both of the
directions.
Table 3 shows the one-step TPs matrixes and the lk in
case 2 data. Similar to case 1, RT1 and RT2 do not com-
municate with each other in both the positive and negative
x-directions. Only slight differences occur in both the
diagonal and the off-diagonal components of the positive
and negative TPs. In the y-direction, the overall
components considerably differ, except for the component
of RT3 to RT3. The most notable difference is that the
positive TP from RT1 to RT2 is given as zero, while the
negative TP from RT2 to RT1 is also zero. The lk of RT1
and RT2 to the x-direction are 9,069.3 and 5,609 m,
respectively, and that of RT3 is 469 m. The discrepancy in
the lk of RT1 and RT2 measured along the positive and
negative directions is not as great as that for case 1. The
estimated lk of RT1 and RT2 in the positive y-direction are
25.1 and 27.4 m, respectively, while those in the negative
y-direction are 51.6 and 17.5 m, respectively. In case 1, the

123
Environ Earth Sci (2014) 71:2739–2752 2743

Table 2 Geostatistical parameters in case 1 Table 3 Geostatistical parameters in case 2

RT1 RT2 RT3 RT1 RT2 RT3

TPM along positive x-direction TPM along positive x-direction

RT1 0.9994 0.0 6.160E-04 RT1 0.999 0.0 9.977E-04
RT2 0.0 0.9998 1.576E-04 RT2 0.0 0.9983 1.675E-03
RT3 5.759E-03 5.040E-03 0.9892 RT3 8.94E-03 1.405E-02 0.977
TPM along negative x-direction TPM along negative x-direction
RT1 0.9996 0.0 4.107E-04 RT1 0.9989 0.0 1.103E-03
RT2 0.0 0.9996 3.678E-04 RT2 0.0 0.9982 1.783E-03
RT3 8.639E-03 2.16E-03 0.9892 RT3 8.102E-03 1.322E-02 0.9787
TPM along positive y-direction TPM along positive y-direction
RT1 0.9757 1.367E-02 1.065E-02 RT1 0.9601 1.341E-02 2.65E-02
RT2 0.0 0.9926 7.387E-03 RT2 0.0 0.9635 3.652E-02
RT3 5.4E-02 0.1951 0.7509 RT3 0.154 0.343 0.5028
TPM along negative y-direction TPM along negative y-direction
RT1 0.9961 0.0 3.921E-03 RT1 0.9806 0.0 1.937E-02
RT2 1.399E-02 0.9718 1.42E-02 RT2 1.38E-02 0.9427 4.349E-02
RT3 0.1497 9.935E-02 0.7509 RT3 0.2152 0.282 0.5028
Mean length along x-direction Mean length along x-direction
lix (m) 7,214.5 8,650.7 926.0 lix (m) 9,069.3 5,609.0 469.0
Mean length along positive y-direction Mean length along positive y-direction
1 i
lx (m) 41.1 135.4 4.0 1 i
ly (m) 25.1 27.4 2.0
Mean length along negative y-direction Mean length along negative y-direction
2 i
ly (m) 255.1 35.5 4.0 2 i
ly (m) 51.6 17.5 2.0

differences between lk in RT1 and RT2 along the y-direc-
tions are from 3.3 to 7.2 times in total. However, in case 2,
the differences are a minimum of 1.1 times and a maximum
of 2.9 times. The smaller differences in the y-directional lk
of case 2 compared to case 1 can be attributed to the
reduced non-stationarity and it may be concluded that the
geospatial regularities are increased in case 2. The
embedded RT3 in case 2 also shows identical lk , similarly
for case 1, of 2.0 m in both directions.
Based on the conditioning borehole logging data and the
geospatial statistics (i.e., TPs matrixes, lk , and lithological
proportions), a total of 100 realizations are generated using
SISIM (Deutsch and Journel 1998) and GCMC (Park
2010), respectively. From the superposition of the 100
realizations, the ensemble probabilities of each rock
occurrence are computed in Figs. 1, 2, 3, and 4 (ESM
only). For the conversion of the correlation scales (kk ) of
each rock from the lk , the equation of kk ¼ xð1  pk Þlk
with x = 1.5 by assuming a spherical lithological structure
is applied (Ritzi 2000) and the values are used as the input
model parameters of SISIM.
Figure 1 (ESM only) shows the ensemble probability
(EP) maps from the GCMC realizations of case 1. To
explain the predictability and the unpredictability of EP
maps, the fitness and the unfitness measures introduced by
Park (2010, Eq. (6)) are computed. The EP predicts the
actual RT3 distribution correctly with 0.76 and incorrectly
with 0.026. Figure 2 (ESM only) shows the EP maps
generated from the SISIM realization of case 1. In the input
data, ordinary kriging was used as an estimator, and the
employed semi-variogram models were all spherical,
which have no nugget effects and the sills are uniformly
1.0. Due to the higher entropy at the lithological interface
of SISIM (Park 2010; Park et al. 2007), the EP distributions
of the rocks show blurring around the rock parcels (Fig. 2a
and b, ESM only). Note that the almost symmetric EP
distribution of RT3 is peculiar in SISIM (Fig. 2c, ESM
only) due to the incapability of handling asymmetric rock
juxtaposition of indicator semi-variogram. Employing the
fitness and the unfitness measures, the EP maps correctly
predict the distribution of RT3 with 0.37 and incorrectly
with 0.028. These values indicate the low quality of the
SISIM predictions compared to GCMC. In addition, the
underestimation of the structural extension of RT3 to the
x- and y-direction in the SISIM is also distinctive relative to
the GCMC (Figs. 1c and 2c, ESM only).
Similarly, Fig. 3a–c (ESM only) shows the EP maps of
GCMC realizations in case 2 as shown in Fig. 2a. The

123
2744 Environ Earth Sci (2014) 71:2739–2752

distribution of the RT3 is predicted correctly with 0.69 in Table 4. In all simulations, supercritical CO2 was
fitness, and incorrectly with 0.029 unfitness value. Com- injected at a depth of 1,080 m and in the middle of the
parison of the EP map in Fig. 3c (ESM only) with the model (x = 2,000 m) during a 2-year period with a con-
deterministic k field shown in Fig. 2a, however, indicates stant rate of 0.5 million metric tons per year. This injection
that the RT3 extension predicted from GCMC seems to be rate and period were chosen to approximate the injection
slightly exaggerated, which may lead to overestimation of size of a medium-scale CO2 sequestration demonstration.
the performance of low-k lenses as capping formation. The primary drainage curves of CO2 and brine in relative
Finally, Figs 4a–c (ESM only) shows the EP maps from the permeabilities were calculated from Parker et al. (1987)
SISIM of case 2. The almost symmetric probability dis- and the associated imbibition curves of CO2 from Land
tribution of RT3 can be confirmed from Fig. 4c (ESM (1968). In addition, the capillary pressure is calculated
only) similar to the SISIM predictions in case 1. The fitness from Parker et al. (1987). Their mathematical expressions
and the unfitness measures were 0.37 and 0.037, respec- are described in Table 5. Finally, all simulations were
tively, which indicates a lower prediction quality than that performed with CMG’s GEMTM simulator, a multidimen-
of the GCMC counterpart. sional, finite-difference, isothermal, compositional simu-
lator (CMG 2007). A detailed description of the GEM
simulator is discussed in previous studies of subsurface
Numerical model settings CO2 plume behavior (Bachu et al. 2009; Bryant et al. 2008;
Ghaderi et al. 2011; Kumar et al. 2005; Saadatpoor et al.
With two hypothetical and deterministic k distributions 2010; Wong et al. 2007).
(Figs. 1, 2) and their stochastic realizations, two-dimen-
sional cross-sectional models were designed. The models
represent the candidate formation for geologic CO2 Results and discussion
sequestration with depths ranging from 1,000 m at the top
to 1,100 m at the bottom. For initial conditions, the model CO2 injection simulations on two hypothetical
is fully saturated with brine and assigned a hydrostatic and deterministic cases
pressure (P = qwgh) and temperature (T = 10 ? 0.025 h)
ranging from 10.0 to 11.0 MPa and from 35 to 40 °C, Figure 3a–d shows the snapshots of CO2 mass distributions
respectively. As such, the formation pressure and temper- at years 2, 50, 100, and 200 from the deterministic CO2
ature are above critical conditions (7.38 MPa and 31.1 °C) injection simulation on case 1. The purpose of the simu-
for supercritical CO2 (Span and Wagner 1996). The lation is to build a benchmark model for predictive simu-
detailed parameters of the base case model are summarized lations based on geostatistical realizations as discussed in
the ‘‘CO2 injection simulations on conditional stochastic
fields’’.
Table 4 Model parameters describing a generalized two-dimensional
The intrinsic driving force for the injected CO2 move-
model
ment was the buoyancy originating from the density dif-
Number of elements x-Direction: 400, y-direction: 1, ference between the supercritical-phase CO2 and brine,
z-direction: 100
resulting in a CO2 plume migration opposite to the gravi-
Size of each element (m) Dx = 100, Dy = 2,000, Dz = 1
tational direction (Fig. 3a). In the simulation, when the
Initial pressure condition Hydrostatic gradient from top (10.0)
(MPa) to bottom (12.0)
plume reached the low-k lens (RT3), the shape of the CO2
plume changed to an inversely shaped cone due to the
Initial temperature condition Geothermal gradient from top (35) to
(°C) bottom (40) lateral spreading and retarded vertical movement of CO2
Permeability: k (m2) RT1: kv: 10-14 m2 and kh: 10-15 m2 within the low-k lens. Two different transports controlling
RT2: kv: 10-13 m2 and kh: 10-14 m2 the CO2 plume distribution around and within the low-
RT3: kv: 10-16 m2 and kh: 10-17 m2 k lenses are observed. First, when an upward migrating
Porosity: / 0.1 CO2 plume encountered a low-k lens, the CO2 plume
Brine density: qbr (kg/m3) 1,100 accumulated and spread under the low-k lens, displacing
CO2 density: qCO2 (kg/m3) *690
the brine. Immediately, the CO2 plume migrated around the
low-k lens because it was difficult to penetrate the low-
Maximum residual CO2 0.1
saturation: Smax k lens before enough pressure was built up. Due to this
gr
Boundary conditions Top and bottom boundaries: no flow
reason, only a certain amount of CO2 penetrated through
the low-k lens (year 2 of Fig. 3a). The CO2 plume kept
Lateral boundaries: constant pressure
expanding underneath the low-k lens until it reached the
Simulation period 200 years
lateral margin of the low-k lens where the CO2 plume

123
Environ Earth Sci (2014) 71:2739–2752 2745

Table 5 Mathematical representations of relative permeability, cap- the low-k lens was so slow that substantial amounts of CO2
illary pressure curves, and hysteresis effect were accumulated within the low-k lenses.
Drainage: n = 4.2, Sr = 0.1 Figure 4a–d shows the distribution of CO2 mass in case
krCO2 and krbr br Sr
S þS S 2 where the scales of the low-k lenses are shorter than those
m ¼ 1  1=n; Sbr ¼ S1S r
; St ¼ br 1S
CO2
r
r

  m 2 of case 1, and these lenses are distributed across all ranges

0:5 1=m
krbr ¼ Sbr 1  1  Sbr along the x-direction. Although the geospatial structure and

0:5   
1=m m
 
1=m m 2
the distribution of the low-k lenses differ in cases 1 and 2,
krCO2 ¼ St  Sbr 1  Sbr  1  St the CO2 transport mechanisms are essentially the same as
krCO2 : Relative permeability of CO2; krbr : Relative those previously described. More CO2 escaped in case 2,
permeability of brine because the lateral extents of the low-k lenses are shorter
SCO2 : Saturation of CO2; Sbr : Relative permeability of than those of case 1. In case 2, even before the cessation of
brine CO2 injection, the CO2 plume has already reached the
Sr : Irreducible saturation of brine; n: Van Genuchten upper formation (RT1) by detouring around the low-
curve shape parameter
1 1
k lenses (Fig. 4a). Consequently, the CO2 arrival time at
Imbibition: Smax ¼ Sgrh  S1gh þ 1
krCO2
gr the top of the model domain was shortened where the CO2
SCO2 Sgrh
SCO2 ðshiftedÞ ¼ Sgh Sgrh Sgh plume arrived earlier than year 50 (Fig. 4b), but in case 1
krCO2 ðSCO2 Þ ¼ krCO2 ðSCO2 ðshiftedÞÞ the arrival time was over 100 years (Fig. 3c).
Smax
For further quantitative characterization, mass fraction
gr : Maximum residual saturation; Sgh : SCO2 when
the shift to imbibition occurs of supercritical CO2 within each rock type and that of CO2
Sgrh : Sgr corresponding to Sgh dissolved in water were calculated as time evolved
Drainage and n = 4.2, Sr = 0.1, a = 5.2 9 10-4, b = 2
(Fig. 5a, b). Immediately after starting the CO2 injection,
imbibition:  1=n the CO2 mass fraction in the lower formation (RT2) started
1 1=m
Pc Pc ¼ ab Sbr 1 to decrease until the saturation of mobile CO2 reached the
Pc : Capillary pressure; a; b; n: Van Genuchten curve threshold of residual CO2 saturation. When all the mobile
shape parameter CO2 had disappeared so that only immobile CO2 remained
in the lower formation, the CO2 mass fraction reached a
constant value of approximately 0.18 in both cases. Spe-
cifically, the reduction rate of the CO2 mass fraction in case
migrated upward. The escaping CO2 plume at the margin 1 is slower than that in case 2. The different reduction rates
of the low-k lens was impeded again when the plume are closely related to the length of the low-k lenses (RT3).
encountered a subsequent low-k lens (Fig. 3b). Henceforth, For example, because the lengths of the low-k lenses are
the CO2 plume repeated the horizontal spreading until it short in case 2, CO2 can easily detour the low-k lenses,
reached the margins of the low-k lenses. In contrast to the resulting in large amounts of CO2 mass being able to reach
fast-flowing CO2 plume, by detouring around the low-k the upper formation in a relatively short time of less than
lenses, certain amounts of CO2 directly penetrated across 2 years (Fig. 5a). In case 1, with the long length of low-
the low-k lenses at a condition in which the cumulative k lenses, the CO2 mass arrived at the upper formation
pressure underneath the low-k lenses was large enough to 20 years later.
overcome the capillary entry pressure of the low-k lenses The changes of CO2 mass within the low-k lenses were
(Fig. 3b–d). The penetration rate of the CO2 plume through governed by the CO2 mass flowing in and out of these

(a) (b)

(c) (d)

Fig. 3 CO2 mass distribution (kg) in the deterministic case 1 domain shown in Fig. 1a at: a 2 years, b 50 years, c 100 years, and d 200 years

123
2746 Environ Earth Sci (2014) 71:2739–2752

(a) (b)

(c) (d)

Fig. 4 CO2 mass distribution (kg) in the deterministic case 2 domain shown in Fig. 2a at: a 2 years, b 50 years, c 100 years, and d 200 years

Fig. 5 Evolution of 1 1
supercritical- and dissolved- (a) (b)
Case 1
phase CO2 mass fraction: a CO2 Case 2
mass in upper (RT1) and lower 0.8 Lower 0.8
formation (RT2)
(RT2) formations and b CO2

Mass fraction of CO2

Mass fraction of CO2

mass in low-k lenses (RT3) and

dissolved in brine 0.6 0.6 Dissolved
in brine

0.4 0.4
Upper
formation (RT1)
Low k
0.2 0.2 lenses (RT3)

0 0
0 40 80 120 160 200 0 40 80 120 160 200
Time (year) Time (year)

lenses. When the CO2 mass flowing in the low-k lenses was
greater than that flowing out, the CO2 mass fraction within
the low-k lenses increased, but when the CO2 mass flowing
in was smaller than that flowing out, the CO2 mass fraction
decreased (Fig. 5b). Overall, the low-k lenses with longer
extent (case 1) preserved more CO2. Finally, CO2 dissolved
in brine is approximately the same in both cases until year
70, but after this year the amount of CO2 dissolved in brine
for case 2 is decreased distinctively. The reduction of the
CO2 dissolution rate in case 2 occurred due to the early
arrival of the CO2 plume at the topmost boundary where
CO2 and brine interactions are reduced due to the large
amount of CO2 accumulation (Fig. 4c, d). Additionally, the
residence time of CO2 mass in low-k lenses was greater in
Case 1 than Case 2 (Figs. 3c, 4c). Thus, CO2 in Case 1 will
have more time to interact with brine before it reaches the
topmost layer, resulting in the increase in the amount of
dissolved CO2 in brine.
In these two cases possessing different degrees of non-
stationarity, the CO2 mass distributions at three formations
showed significant discrepancy with time, suggesting that
the geostatistical configuration, structures, and patterns of
the capping formation play important roles in the CO2
storage and migration characteristics. Therefore, to tackle
further detail, multiple realizations from GCMC and SISIM
after conditioning the borehole logging data were
employed in cases 1 and 2. Then, numerical CO2 plume
migration models based on randomly selected 50 realiza-
tions are transiently simulated over 200 years.
CO2 injection simulations on conditional stochastic
fields
Case 1
Figure 6a, c shows the permeability distributions of case 1
from randomly selected GCMC and SISIM realizations,
respectively. From two sequences of simulations, dissimi-
larity of the spatial variation in CO2 leakage is observed at
100 years (Fig. 6b, d). The most peculiar difference is the
geometry of CO2 plume distribution, the leakage pathways,
and the amount of CO2 plume arrived at the topmost
boundary of the model domain. As seen in the GCMC
simulations (Fig. 6b), relatively large CO2 masses were
accumulated in the injection formation (RT2) with large
lateral extent. In the upper (RT1) formation, the CO2 plume
leakage occurred locally only to the left of the injection
borehole. Due to the localized leakage processes, CO2 mass

123
Environ Earth Sci (2014) 71:2739–2752
Fig. 6 a Representative k field chosen from GCMG realizations in
case 1. b CO2 mass distribution (kg) at 100 years, simulated in the
chosen GCMC k field. c Representative k field chosen from SISIM
was highly concentrated only at a few locations of the
topmost boundary. In contrast, the CO2 plume predicted
from SISIM realization leaked uniformly though RT1
formation (Fig. 6d). Preliminarily judging from the single
realizations at GCMC and SISIM, the CO2 migration
simulation based on an SISIM realization overestimated
the leakage of the CO2 plume to the RT1 formation. This
overestimation can also be foreseen by some degree in the
analysis of the ensemble probabilities map of SISIM
(Fig. 2c, ESM only) where the sizes of the low-k lenses
(RT3) were underestimated compared to those of the
GCMC simulation (Fig. 1c, ESM only). However, it should
be noted that the localized CO2 accumulation reaching the
topmost layer is less for SISIM than GCMC prediction.
In Fig. 7, the mass fraction evolutions of supercritical
CO2 within RT1, RT2, and RT3 formations and dissolved
CO2 in brine from GCMC and SISIM realizations of case 1
are plotted. The solid-black lines represent the determin-
istic simulations shown in Figs. 3 and 5, and the dashed-
black lines show the arithmetic means calculated from
GCMC and SISIM realizations. The pink- and blue-shaded
areas represent 99 and 80 % prediction intervals, respec-
tively. In the GCMC realizations, the deterministic CO2
mass fraction evolutions of the upper formation (RT1),
lower formation (RT2), low-k lenses (RT3), and dissolved
CO2 in brine fall within the 80 % prediction intervals. In
detail, the deterministic lines do not coincide with the
simulated mean dashed lines. Rather, the CO2 mass frac-
tion from multiple realizations showed a certain range of
variation relative to the deterministic prediction, which
may be biased. The simulated mean of the CO2 mass
fraction in RT1 was positively biased compared to the
deterministic line, while those in RT3 and the dissolved
CO2 in brine were slightly underestimated in their deter-
ministic predictions. The overestimation of the CO2 mass
fraction in RT1 directly links to the underestimation of
RT3 because the performance of low-k lenses is critical to
determine the amount of CO2 mass leaving the lower half
of the simulated domain. The dissolved CO2 in brine is
2747
realizations in case 1. d CO2 mass distribution (kg) at 100 years,
simulated in the chosen SISIM k field
dependent on the detouring length of the migration path-
ways, which determines the probability of the CO2 plume
contacting the CO2-unsaturated brine. For this reason, the
slight underestimation of the dissolved CO2 in brine rela-
tive to the deterministic prediction implies that the pre-
dicted migration length of the CO2 plume in the GCMC
prediction is shorter than that of the deterministic
prediction.
In contrast to the transient evolution of CO2 mass
fraction through the GCMC realizations, the SISIM real-
izations showed significant over- or underestimations
subjected to deterministic predictions (Fig. 7). As dis-
cussed earlier in the ensemble probability distributions in
GCMC and SISIM (Figs. 1c, 2c, ESM only), the extent of
the low-k lenses (RT3) was shorter in the SISIM predic-
tions than in the GCMC predictions. For this reason, the
CO2 mass fractions preserved in RT2 and RT3 are signif-
icantly underestimated in the SISIM-based simulations.
Especially in the later time, most curves of the determin-
istic CO2 mass fraction deviate outside the 99 % prediction
intervals, except the CO2 mass fraction in RT2. The
asymptotic levels of CO2 mass fractions in RT2 predicted
from GCMC, SISIM, and the deterministic simulations
were coincident after 100 years, suggesting that the amount
of residually trapped CO2 in RT2 was not sensitive to the
configurations or distributions of the low-k lenses (RT3).
To obtain further quantitative comparisons of the sta-
tistics in CO2 leakage masses, histograms representing the
CO2 mass fraction within RT1 are plotted from GCMC-
and SISIM-based simulations, progressively from years 1
through 200 (Fig. 5, ESM only). In the GCMC-based
simulations, until year 10, the occurrence of CO2 mass
fractions in RT1 was small and only a few predictions
showed meaningful CO2 leakage. For this reason, the his-
togram of CO2 mass fraction showed a log-normal or an
exponential distribution up to year 20. Between year 50
and 100, the histograms evolved from the log-normal to the
normal distribution, suggesting that vertical CO2 move-
ment is distinct. The histograms of the SISIM-based
123
2748
Fig. 7 Mass fraction evolution of supercritical-phase CO2 predicted
from 50 GCMC and 50 SISIM realizations in case 1. Dark solid and
dashed lines represent the evolution of CO2 mass fraction from the
predictions are similar to those of the GCMC to some
degree. However, the log-normal shaped histogram was
maintained only until year 2. Overall, the variances of
histograms predicted from the SISIM are significantly
larger than those of the GCMC, and the mode of the his-
togram is not as distinctive as the GCMC counterpart in the
mid to later simulation time. Clearly, heterogeneity pre-
diction from two different stochastic approaches controls
the rate of CO2 plume migration to RT1; the leakage rate
was faster in SISIM than in GCMC.
The red- and blue-dashed lines indicate the CO2 mass
fraction in RT1 predicted from the deterministic and the
mode of multiple realizations, respectively (Fig. 5, ESM
only). The disparity between the red- and blue-dashed lines
implies the degree of accuracy predicted from the sto-
chastic simulations. As seen, most deterministic values are
located close to the predictive mode of the GCMG-based
simulations, but significantly deviate from the mode of the
SISIM-based simulations.
Case 2
Two representative realizations of case 2 are plotted in
Fig. 8a and c. Figure 8b, d is the associated CO2 plume
distributions at year 100, and the benchmark deterministic
prediction of the CO2 plume distribution in the same year is
plotted in Fig. 4c. The upward flux of the CO2 mass from the
GCMC and SISIM realizations reasonably matches with the
deterministic prediction. At the topmost boundary, CO2
mass was more concentrated in GCMC simulations. Nev-
ertheless, it can be noted that geostatistical predictions from
both GCMC and SISIM work better in Case 2 than in Case 1.
123
Environ Earth Sci (2014) 71:2739–2752
deterministic case and the means calculated from 50 GCMC and
SISIM realizations, respectively. Pink and blue-shaded areas repre-
sent 99 and 80 % prediction intervals, respectively
Figure 9 shows the CO2 mass fraction evolutions at the
upper formation (RT1), lower formation (RT2), low-
k lenses (RT3), and dissolved CO2 in brine from case 2. In
the GCMC-based simulations, the mean of realized simu-
lations slightly deviates from the deterministic curves,
especially at the early time. Specifically, the CO2 mass
fractions from the realized simulations were underesti-
mated relative to the deterministic curve in RT1 and,
consequently, both RT2 and RT3 were overestimated. At
the later simulation time, the deterministic curves mostly
fall within the 80 % prediction intervals and the degree of
the over- or underestimation becomes less.
In contrast, the SISIM-based simulations show a wide
range of prediction variations, resulting in both the 80 and
99 % prediction intervals being noticeably wider than
those of the GCMC-based simulations. Nevertheless, the
means of the SISIM-based simulations generally match
with the deterministic prediction better than those of the
GCMC-based simulations. The superior predictions of
mean CO2 mass fraction by the SISIM-based simulations
can be attributed to the better correlation scale represented
by the SISIM ensemble probability map of RT3 (Fig. 4c,
ESM only). In addition, a larger variance of SISIM-based
simulations can be foreseen in the diffused ensemble
probability distribution of RT3. From this analysis, the
SISIM-based prediction could be more suited for the field
conditions where heterogeneity structures show stationarity
similar to the synthetic case 2. However, the large vari-
ances associated with SISIM-based predictions may cause
additional problems in practice.
Finally, histograms representing the CO2 mass fraction
in RT1 are drawn from GCMC and SISIM (Fig. 6, ESM
Environ Earth Sci (2014) 71:2739–2752
Fig. 8 a Representative k field chosen from GCMG realizations in
case 2. b CO2 mass distribution (kg) at 100 years, simulated in the
chosen GCMC k field. c Representative k field chosen from SISIM
Fig. 9 Mass fraction evolution of supercritical-phase CO2 predicted
from 50 GCMC and 50 SISIM realizations in case 2. Dark solid and
dashed lines represent the evolution of CO2 mass fraction from the
only). In the histogram of the GCMC-based predictions of
case 2, a similar transition from a log-normal to a normal
distribution as that in case 1 is observed. Compared to case
1, this transition occurs relatively quickly. The earlier
transition in case 2 indicates the earlier arrival of the CO2
plume to RT1, which is due to the relatively short corre-
lation length of the low-k lenses in the x-direction. The
histograms of the SISIM-based predictions slightly differ
from those of the GCMC-based simulation. The histogram
seems to be closer to the mixture of a log-normal and a
normal to a uniform distribution rather than a single log-
normal distribution in the early simulation time. Also, the
variance of the histograms is considerably wider than that
of GCMC-based predictions. The non-regularity of the
transient histogram continued to the end of the simulations.
2749
realizations in case 2. d CO2 mass distribution (kg) at 100 years,
simulated in the chosen SISIM k field
deterministic case and the means calculated from 50 GCMC and 50
SISIM realizations, respectively. Pink and blue-shaded areas repre-
sent 99 and 80 % prediction intervals, respectively
Even though the means of the distribution match with the
deterministic prediction better, considerable uncertainty is
associated with SISIM-based prediction due to the non-
regular distribution and higher standard deviation of the
predictions.
Estimation of CO2 leakage potentials at the topmost layer
The CO2 mass that accumulated at the topmost model
domain 200 years later was quantified from deterministic
cases and the ensembles of the Monte Carlo simulations
based on GCMC and SISIM of Cases 1 and 2. In Fig. 7a
(ESM only), the CO2 mass distribution (black) estimated
from the deterministic case 1 showed two modes at x = 1.8
and 2.2 km with the localized CO2 accumulation. The
123
2750
ensemble of CO2 mass prediction from GCMC (blue)
showed dispersed leakage potential and consequently the
integrated CO2 mass under the curve was overestimated
relative to the deterministic prediction. The modes of
GCMC ensembles were at x = 1.6, 2.2, and 2.7 km, which
are similar to the modes of the deterministic estimation. In
contrast, the ensemble mass predicted from SISIM (red)
shows highly dispersed leakage potential implying the fast
arrival of CO2 mass at the broad area. CO2 mass leakage
potential follows a bell-shaped distribution with single
mode at x = 1.8 km. Overall, it may be concluded that the
leakage potential from GCMC simulation shows better
predictability compared to that from SISIM in Case 1.
In Case 2 representing stationary field, more CO2 mass
evidently accumulated at the topmost model domain
compared to Case 1 (Fig. 7b, ESM only). In the deter-
ministic estimation from Case 2, there exist clear modes at
x = 1.2, 1.7, and 2.4 km where the mode at x = 1.7 km is
the most peculiar. The ensemble predictions from both
GCMC and SISIM are similar to the deterministic one. The
predicted mode of the ensemble based on GCMC is at
x = 1.6 km and that of SISIM at x = 1.7 km, which
indicates the slightly better predictability of SISIM at the
topmost layer.
Summary and conclusions
The presented study focused on understanding the uncer-
tainties of two non-parametric geostatistical applications
(GCMC and SISIM) in subsurface heterogeneity and their
associated effects on CO2 plume migration predictions.
Two hypothetical models were developed. In case 1, rep-
resenting a weakly non-stationary field, only four wide-
ranging low-k lenses existed around the mid-x and -y loca-
tions. On the other hand, in case 2 representing a stationary
field, the scales of the low-k lenses were relatively short
compared to those in case 1, and were distributed across
almost the entire range along the x-direction.
In case 1, SISIM underestimated the horizontal extent of
the low-k lenses relative to the GCMC prediction, resulting
in more CO2 leakage to the upper formation. For this
reason, the GCMC-based simulations represented the
deterministic prediction better than those from the SISIM
realizations. Histogram analysis of CO2 mass fractions
within RT1 indicates that the histogram follows a log-
normal distribution in the early period, but later evolves to
normal distributions. The transition from log-normal to
normal distribution occurred quickly in the SISIM real-
izations. In case 2, the GCMC models slightly overesti-
mated the vertical and horizontal extensions of the low-
k lenses, causing the deviation of the predicted means in
the GCMC realizations from the deterministic curve,
123
Environ Earth Sci (2014) 71:2739–2752
especially at the early time. In contrast, the SISIM-pre-
dicted means generally show good agreement with the
deterministic prediction, but show a wide ranges of pre-
diction variances. Both simulations suggested that the sizes
of the low-k lenses were critical to determine the amount of
CO2 leaking to RT1. However, the amount of residually
trapped CO2 remaining in RT2 was fairly consistent in all
realizations and was less sensitive to the variations of the
low-k lens sizes or structures. Finally, CO2 mass dissolved
in brine was governed by the length of the migration
pathway.
Throughout the comprehensive ‘‘Monte Carlo’’-typed
analyses, the study suggests that geostatistical prediction is
a useful tool for characterizations of subsurface heteroge-
neities where the information is significantly limited and
the uncertainties affect critically various decisions on the
geologic CO2 sequestration sites. In the situation, every
conceivable tools as well as supportive information need to
be incorporated to limit the uncertainty. In the present
study, considerations on geostatistical models are high-
lighted by assuming that only a few borehole data are
available. In real practices, geostatistical subsurface esti-
mations are generally adopted. There could be more dif-
ferent types of data including geophysical information,
geological inferences of the experts, etc., which might limit
the uncertainties due to the absence of direct information
and result in improved predictions on the subsurface.
Nevertheless, the present study will provide useful infor-
mation as a worst case where the information on subsurface
is extremely limited, in which a choice of geostatistical
models could be sensitive to the final decision depending
on their capability of handling spatial non-stationarity in a
suitable manner.
Acknowledgments This work was jointly supported by ‘‘Technol-
ogy development for CO2 geological storage demonstration through
participating in the Canadian projects (2011T100100331)’’ of the
Korea National Oil Corporation (KNOC) and ‘‘Block funding projects
(GP2012-030)’’ of Korea Institute of Geoscience and Mineral
Resources (KIGAM) transferred from the Korea Institute of Energy
Technology Evaluation and Planning (KETEP) grants.
References
Ambrose WA, Lakshminarasimhan S, Holtz MH, Núñez-López S,
Hovorka SD, Duncan I (2008) Geologic factors controlling CO2
storage capacity and permanence: case studies based on expe-
rience with heterogeneity in oil and gas reservoirs applied to
CO2 storage. Environ Geol 54:1619–1633. doi:10.1007/s00254-
007-0940-2
Audigane P, Gaus I, Czernichowski-Lauriol I, Pruess K, Xu T (2007)
Two-dimensional reactive transport modeling of CO2 injection
in a saline aquifer at the Sleipner site, North Sea. Am J Sci
307:974–1008. doi:10.2475/07.2007.02
Environ Earth Sci (2014) 71:2739–2752
Bachu S, Pooladi-Darvish M, Hong H (2009) Chromatographic
partitioning of impurities (H2S) contained in a CO2 stream
injected into a deep saline aquifer: Part 2. Effects of flow
conditions. Int J Greenhouse Gas Contr 3:468–473. doi:
10.1016/j.ijggc.2009.01.002
Bateson L, Vellico M, Beaubien SE, Pearce JM, Annunziatellis A,
Ciotoli G, Coren F, Lombardi S, Marsh S (2008) The application
of remote-sensing techniques to monitor CO2-storage sites for
surface leakage: method development and testing at Latera
(Italy) where naturally produced CO2 is leaking to the
atmosphere. Int J Greenhouse Gas Contr 2:388–400. doi:
10.1016/j.ijggc.2007.12.005
Bruining J, van Batenburg DW, Lake LW, Yang AP (1997) Flexible
spectral methods for the generation of random fields with power-
law semivariograms. Math Geol 29:823–848. doi:10.1007/
bf02768904
Bryant SL, Lakshminarasimhan S, Pope GA (2008) Buoyancy-
dominated multiphase flow and its effect on geological seques-
tration of CO2. SPE J 13:447–454. doi:10.2118/99938-PA
Carle SF, Fogg GE (1996) Transition probability based indicator
geostatistics. Math Geol 28:453–477. doi:10.1007/BF02083656
CMG (2007) User’s guide GEM, Advanced Compositional Reservoir
Simulator (version 2007). Computer Modeling Group Ltd,
Calgary
Deutsch CV, Journel AG (1998) GSLIB: Geostatistical Software
Library and User’s Guide, 2nd edn. Oxford University, New
York
Doughty C, Pruess K (2004) Modeling supercritical carbon dioxide
injection in heterogeneous porous media. Vadose Zone J
3:837–847. doi:10.2113/3.3.837
Emberley S, Hutcheon I, Shevalier M, Durocher K, Gunter WD,
Perkins EH (2004) Geochemical monitoring of fluid–rock
interaction and CO2 storage at the Weyburn CO2-injection
enhanced oil recovery site, Saskatchewan, Canada. Energy
29:1393–1401. doi:10.1016/j.energy.2004.03.073
Farajzadeh R, Ranganathan P, Zitha PLJ, Bruining J (2011) The effect
of heterogeneity on the character of density-driven natural
convection of CO2 overlying a brine layer. Adv Water Resour
34:327–329. doi:10.1016/j.advwatres.2010.12.012
Finsterle S (2004) Multiphase inverse modeling: review and
iTOUGH2 applications. Vadose Zone J 3:747–762
Flett M, Gurton R, Weir G (2007) Heterogeneous saline formations
for carbon dioxide disposal: impact of varying heterogeneity on
containment and trapping. J Petrol Sci Eng 57:106–118. doi:
10.1016/j.petrol.2006.08.016
Ghaderi SM, Keith DW, Lavoie R, Leonenko Y (2011) Evolution of
hydrogen sulfide in sour saline aquifers during carbon dioxide
sequestration. Int J Greenhouse Gas Contr 5:347–355. doi:
10.1016/j.ijggc.2010.09.008
Han WS, Lee S-Y, Lu C, McPherson BJ (2010a) Effects of
permeability on CO2 trapping mechanisms and buoyancy-driven
CO2 migration in saline formations. Water Resour Res
46:W07510. doi:10.1029/2009WR007850
Han WS, McPherson BJ, Lichtner PC, Wang FP (2010b) Evaluation
of trapping mechanisms in geologic CO2 sequestration: case
study of SACROC northern platform, a 35-year CO2 injection
site. Am J Sci 310:282–324. doi:10.2475/04.2010.03
Hovorka SD, Doughty C, Benson SM, Pruess K, Knox PR (2004) The
impact of geologic heterogeneity on CO2 storage in brine
formations: a case study from the Texas Gulf Coast. In: Baines
SJ, Worden RH (eds) Geological storage of carbon dioxide.
Special Publication, Geological Society of London, pp 147–163
Ide ST, Jessen K, Orr FM (2007) Storage of CO2 in saline aquifers:
effects of gravity, viscous, and capillary forces on amount and
timing of trapping. Int J Greenhouse Gas Contr 1:481–491. doi:
10.1016/S1750-5836(07)00091-6
2751
Jahangiri HR, Zhang D (2011) Effect of spatial heterogeneity on
plume distribution and dilution during CO2 sequestration. Int J
Greenhouse Gas Contr 5:281–293. doi:10.1016/j.ijggc.
2010.10.003
Juanes R, Spiteri EJ, Orr FM, Blunt MJ (2006) Impact of relative
permeability hysteresis on geologic CO2 storage. Water Resour
Res 42:W12418. doi:10.1029/2005WR004806
Kopp A, Binning PJ, Johannsen K, Helmig R, Class H (2010) A
contribution to risk analysis for leakage through abandoned wells
in geologic CO2 storage. Adv Water Resour 33:867–879. doi:
10.1016/j.advwatres.2010.05.001
Kumar A, Ozah R, Noh M, Pope GA, Bryant S, Sepehrnoori K, Lake
LW (2005) Reservoir simulation of CO2 storage in deep saline
aquifers. SPE J 10:336–348. doi:10.2118/89343-PA
Lengler U, De LM, Kühn M (2010) The impact of heterogeneity on
the distribution of CO2: numerical simulation of CO2 storage at
Ketzin. Int J Greenhouse Gas Contr 4:1016–1025. doi:10.1016/
j.ijggc.2010.07.004
Land CS (1968) Calculation of imbibition relative permeability for
two- and three-phase flow from rock properties. SPE J
8:149–156. doi:10.2118/1942-PA
Lewicki JL, Hilley GE, Oldenburg CM (2005) An improved strategy
to detect CO2 leakage for verification of geologic carbon
sequestration. Geophys Res Lett 32:L19403. doi:10.1029/
2005GL024281
Li S, Zhang Y, Zhang X (2011) A study of conceptual model
uncertainty in large-scale CO2 storage simulation. Water Resour
Res 47:W05534. doi:10.1029/2010WR009707
Mantoglou A, Wilson JL (1982) The turning bands method for
simulation of random fields using line generation by a spectral
method. Water Resour Res 18:1379–1394. doi:10.1029/WR
018i005p01379
Newell DL, Kaszuba JP, Viswanathan HS, Pawar RJ, Carpenter T
(2008) Significance of carbonate buffers in natural waters
reacting with supercritical CO2: implication for monitoring,
measuring, and verification (MMV) of geologic carbon seques-
tration. Geophys Res Lett 35:L23403. doi:10.1029/2008GL
035615
Park E (2010) A multidimensional, generalized coupled Markov
chain model for surface and subsurface characterization. Water
Resour Res 45: W11509. doi:10.1029/2009WR008355
Park E, Elfeki AMM, Song Y, Kim K (2007) Generalized coupled
Markov Chain model for characterizing categorical variables in
soil mapping. Soil Sci Soc Am J 71:909–917. doi:10.2136/
sssaj2005.0386
Parker JC, Lenhard RJ, Kuppusamy T (1987) A parametric model for
constitutive properties governing multiphase flow in porous
media. Water Resour Res 23:618–624. doi:10.1029/WR023i004
p00618
Pau GSH, Bell JB, Pruess K, Almgren AS, Lijewski MJ, Zhang K
(2010) High-resolution simulation and characterization of den-
sity-driven flow in CO2 storage in saline aquifers. Adv Water
Resour 33:443–455. doi:10.1016/j.advwatres.2010.01.009
Pruess K, Oldenburg CM, Moridis G (1999) Tough2 user’s guide.
Version 2.0. Lawrence Berkeley National Laboratory, Berkeley,
California
Ritzi RW (2000) Behavior of indicator variograms and transition
probabilities in relation to the variance in lengths of hydrofacies.
Water Resour Res 36:3375–3381. doi:10.1029/2000WR900139
Rohmer J, Bouc O (2010) A response surface methodology to address
uncertainties in cap rock failure assessment for CO2 geological
storage in deep aquifers. Int J Greenhouse Gas Contr 4:198–208.
doi:10.1016/j.ijggc.2009.12.001
Saadatpoor E, Bryant SL, Sepehrnoori K (2010) New trapping
mechanism in carbon sequestration. Transp Porous Med
82:3–17. doi:10.1007/s11242-009-9446-6
123
2752
Schlumberger (2005) Eclipse reference manual
Schlumberger (2008) Eclipse reference manual 2008A
Siirila ER, Navarre-Sitchler AK, Maxwell RM, McCray JE (2012) A
quantitative methodology to assess the risks to human health
from CO2 leakage into groundwater. Adv Water Resour
36:146–164. doi:10.1016/j.advwaters.2010.11.005
Smith J, Durucan S, Korre A, Shi J-Q (2011) Carbon dioxide storage
risk assessment: analysis of caprock fracture network connec-
tivity. Int J Greenhouse Gas Contr 5:226–240. doi:10.1016/
j.ijggc.2010.10.002
Span R, Wagner W (1996) A new equation of state for carbon dioxide
covering the fluid region from the triple-point temperature to
1100 K at pressure up to 800 MPa. J Phys Chem Ref Data
25:1509–1596. doi:10.1063/1.555991
Wells AW, Diehl JR, Bromhal G, Strazisar BR, Wilson TH, White
CM (2007) The use of tracers to assess leakage from the
sequestration of CO2 in a depleted oil reservoir, New Mexico,
USA. Appl Geochem 22:996–1016. doi:10.1016/j.apgeochem.
2007.01.002
123
Environ Earth Sci (2014) 71:2739–2752
Wen X-H, Kung C (1993) Stochastic simulation of solute transport in
heterogeneous formations: a comparison of parametric and
nonparametric geostatistical approaches. Ground Water 31:
953–965. doi:10.1111/j.1745-6584.1993.tb00869.x
Wong S, Law D, Deng X, Robinson J, Kadatz B, Gunter WD,
Jianping Y, Sanli F, Zhiqiang F (2007) Enhanced coalbed
methane and CO2 storage in anthracitic coals—Micro-pilot test
at South Qinshui, Shanxi, China. Int J Greenhouse Gas Contr
1:215–222. doi:10.1016/S1750-5836(06)00005-3
Zhang K, Wu Y, Pruess K (2008) User’s guide for TOUGH2-MP-a
massively parallel version of the TOUGH2 code. Lawrence
Berkeley National Laboratory, Berkeley
Zhang Y, Freifeld B, Finsterle S, Leahy M, Ennis-King J, Paterson L,
Dance T (2011) Single-well experimental design for studying
residual trapping of supercrtical carbon dioxide. Int J Green-
house Gas Contr 5:88–98. doi:10.1016/j.ijggc.2010.06.011