Journal of Rock Mechanics and Geotechnical Engineering.

2011, 3 (1): 39–56

Risk assessment of CO2 injection processes and storage in

carboniferous formations: a review

Manchao He1, Sousa Luis1, 2, Sousa Rita3, Gomes Ana2, Vargas Jr. Eurípedes4, Na Zhang1
1
State Key Laboratory for Geomechanics and Deep Underground Engineering, China University of Mining and Technology, Beijing, 100083, China
2
University of Porto, Porto, 4200-465, Portugal
3
Massachusetts Institute of Technology, Cambridge, MA 02139, USA
4
Catholic University of Rio de Janeiro, Rio de Janeiro, CEP 22451-900, Brazil
Received 15 January 2011; received in revised form 7 February 2011; accepted 10 February 2011

Abstract: Over the last decades, people from almost all over the world have realized that it is necessary to quickly develop
strategies for the control and reduction of greenhouse gases (GHG) emissions. Among various GHGs, carbon dioxide (CO2) is
the most abundant GHG. Its underground storage involves less risk and lower levels of dangerousness. The paper briefly
describes the most effective technologies available in the market for background processes to storage (capture and transport)
CO2, as well as the more secure solutions for its storage, in particular for the geological storage in carboniferous formations.
This paper also outlines the methodologies for the risk assessment involved in storage of CO2, with a particular focus on cases
where the injection is made into unminable coal seams and in abandoned coal mines. Methodologies used for risk analysis are
described in detail with particular emphasis on Bayesian network (BN). Some applications regarding the risk assessment of
CO2 injection processes and CO2 storage in carboniferous formations and contamination of aquifers are presented and
analyzed. Finally, based on the applications of BN, several conclusions are drawn.
Key words: risk assessment; underground storage of CO2; coal mines; monitoring

1 Introduction
 Terrestrial
sequestration
Atmospheric CO2
There are several ways of mitigating greenhouse CO2
O2 Ethanol
gases (GHG) emissions to the atmosphere. The storage Coal and biomass plant
Cement/steel/
of large quantities of carbon in geological formations refineries,etc.
is presented today as one of the most effective methods Industrial Fuel CO2
capture
with visible results. Carbon dioxide (CO2) capture and uses and food
Electricity
products
storage (CCS) are a process consisting of the CO2 generation CO2
es
separation of CO2 from industrial and energy-related ip e lin
Rp CO2
EO
sources. Figure 1 brings together, in schematic form, CO2
Seal
the main sources and some of the possible storage sites. Geological CO2 displaces methane from coal
sequestration Oil Seal
Storage of CO 2 in deep, onshore and offshore CO2 CO2 stored in depleted oil/gas reservoirs
geological formations uses many technologies CO2 displaces trapped oil (enhanced oil recovery)
CO2 stored in saline formations
developed by oil and gas industries, and it has been Seal
proved to be economically feasible under specific Seal

conditions in oil and gas fields and saline formations Fig.1 Processes of capturing and storing CO2 1.
1. CO2 can also be stored in carboniferous formations,
either in unminable coal seams or in abandoned coal characterized and properly managed sites. Injecting
mines. CO2 can be safely injected and stored at well CO2 in deep geological formations can store it
underground for a long period of time. At the depth of
800–1 000 m underground, CO2 has a liquid-like
Doi: 10.3724/SP.J.1235.2011.00039

Corresponding author. Tel: +351-966012385;
density that permits the potential for an efficient use of
E-mail: ribeiro.e.sousa@gmail.com underground reservoirs in porous sedimentary rocks.
40 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

Figure 2 illustrates the options for storing CO2 in Kyoto Protocol. The use of coal as a source of energy
deep underground geological formations [1]. Other is attractive due to its abundance and its low price.
geological options, which may serve as storage sites, However, research and development in technologies
include caverns in basalt, organic-rich shale and salt. for renewable energies, energy efficiency, CCS, etc.,
should also be considered in emerging countries. The
situation in China can be characterized by a large
number of abandoned coal mines of about one
Ocean

CO2-ECBM
thousand. Therefore, the storage of CO2 in abandoned

Deep unmineable coal seams

storage

Depleted oil and gas reservoirs coal mines can be a viable option 2.
CO2 EOR

Geological storage requires to construct facilities to

Deep saline aquifers

capture large emission sources of CO2, such as the

power for electricity production or cement, steel,
ethanol, etc.. The captured CO2 is then transported by
pipelines or in ships to underground storage sites. Most
of the mechanisms related to this technology are not
Fig.2 Options for storing CO2 in deep underground geological new, since they are already employed in oil industry,
formations 1. or by contractors for management and distribution of
natural gas, some industries in the food sector, etc..
As it can be observed in Fig.3, China became the Currently, capturing CO2 is costly and energy
largest emitter of CO2 in 2007. In 2006 it reached a consuming. The costs obviously depend on the
peak of 6.53  106 t per year 2. Despite being the dimensions of the industrial unit and the type of fuel
largest emitter and one of the fastest growing countries, used. There are four basic systems for capturing CO2
China yet releases much less GHG per capita than any from fossil fuels andor biomass 1.
developed country. The environmental impacts from geological storage
of CO2 can be integrated into two types, i.e. local
7 2004 2005 2006 environmental effects and global effects on the
6
CO2 emission (106 t)

2007 2008 atmosphere. Global effects may be viewed as

5
4 uncertainty in the effectiveness of CO2 storage. Local
3 hazards arise from the causes such as the direct effects
2
of elevated gas-phase CO2 concentrations on the
1
0 shallow surface or near surface, the effects of dissolved
China USA EU-27 Russia India Japan
Country CO2 in groundwater, and the effects induced by fluids
Fig.3 CO2 emissions from the consumption of energy. displacement of the injected CO2.
There are different potential escaping routes for CO2
China has the major resources of coal in the world. injected into the carboniferous formations. Risk
The situation of emission of CO2 is represented in assessment should be an integral element of the risk
Fig.4. China and other developing countries, such as management activities, such as the site selection, site
India, are not submitted to the limits imposed by the
characterization, storage system, design, monitoring
and remediation if necessary. A possible methodology
to assess risks in these situations is Bayesian network
(BN). BN is a graphical representation of knowledge
for reasoning under uncertainty, and it becomes a
popular representation for encoding uncertain expert
knowledge in expert system. BN can be used at any
stage of a risk analysis, and provides a good tool for
decision analysis, including prior analysis, posterior
CO2 sources analysis and pre-posterior analysis. Furthermore, they
Gas basins
Oil basins can be extended to influence diagrams, including
Coal regions
Deep saline formations decision and utility nodes to explicitly model a
Fig.4 CO2 emission situations in China. decision problem.
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 41

In 2010, the State Key Laboratory of Geomechanics (3) The heterogeneity of the mass taken as a whole
and Deep Underground Engineering, China University (stratigraphic heterogeneity, existence of discontinuities,
of Mining and Technology (Beijing), was selected to etc.).
conduct a project on the risk assessment of CO2 (4) Knowledge of the existence of abandoned
injection and sequestration in carboniferous reservoirs injection/pumping wells nearby.
by the State Administration of Foreign Experts Affairs, (5) The adequacy of the injection system.
China. The importance of the project is related to the (6) Changing biogeochemistry.
fact that China is the major producer of coal in the (7) Geomechanical weathering (generation of cracks
world. Therefore, there are several possibilities for and fractures).
selecting appropriate sites for reservoirs, even in (8) Methods of abandonment of the wells when the
abandoned coal mines. Coal formations contain cleats reservoir reaches the limit.
that impart some permeability to the system. Between Duguid et al. 7 suggested that, as one of the first
cleats, coal has a large number of micropores, into requirements to be met by a site candidate for the
which gas molecules can diffuse and be tightly reservoir, it was to have several layers of sealing. Thus
absorbed. Gaseous CO2 injected through wells will the system is redundant and it is possible to make early
flow through the cleat system, and diffuse in the coal detection of potential problems. If CO2 escapes, the
matrix and be absorbed onto the coal micropore system gives an indication to the authorities. If the
problem is not resolved, the secondary layers of
surfaces. If CO2 is injected into coal seams, it can
protection is in charge of retaining leakage.
displace gas methane, enhancing coal bed methane
In accordance to Ref.1, the commercial projects of
recovery.
CO2 storage in large scale should be adopted if it is
This paper reviews the literatures published on
assumed that the location is well chosen, designed,
geological storage of CO2 in deep saline aquifers and
operated and monitored. The data available from
carboniferous formations, including abandoned coal
existing projects suggest that it is very likely that the
mines with special emphasis on the problematic risk
fraction of stored CO2 trapped in the first 100 years is
assessment.
over 99%, and it is possible that the fraction of stored
CO2 trapped in the first 1 000 years is over 99%.
2 Injection and safety storage 2.2 Risks associated with the earlier stages of
storage
2.1 Introduction Various stages leading up to the storage itself cause
CO2 is a common constituent of the atmosphere, the changes in the state of stress and strain of the rock
non-toxic. However, high concentrations can be mass. In turn, flow paths may be generated, through
dangerous 3. An uncontrolled release of CO2 from an which CO2 can escape due to the discontinuities (pre-
underground reservoir will not have long-term effects existing or not), such as faults or other fractures.
once the CO2 is diluted in air or water, as that happens Associated with the existence of faults, seismic
in cases of highly toxic or nuclear waste. Thus, slow episodes may occur, which may bring more risks to the
migration of gas toward the surface is not a direct CCS project.
threat to humans. However, high concentrations can be To understand the influence of entire storage system
attained by a sudden release or other processes. Due to on the rock mass, it is necessary to study each phase
the high density of CO2 in relation to air in the case of separately. Different phases 8 that may be considered
leakage of large volumes, depressions or enclosures are as follows: (1) drilling and completion of wells; (2)
can be created near the earth’s surface, causing loss of formation dewatering and methane production; and (3)
consciousness or asphyxiation to humans who are in CO2 injection with or without secondary production of
the vicinity 4. methane.
The main risks of geological storage of CO2 vary Wellbore stability is a geomechanical problem that
from place to place, mainly depending on such factors can be encountered during drilling. Rock failure and
1, 5, 6: displacements associated with wellbore instability
(1) The configuration of the storage facility, generate potential leakage paths. These drilling issues
including the geological characteristics of the stratum and the main causes of instabilities are analyzed in
selected. detail in Ref.9. The risk of leakage will be minimized
(2) The heterogeneity of the sealing caprock. by cementing the case. Two constructive methods are
42 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

conventionally used in the execution of wells: cased

hole wells and open-hole cavity wells. The risks
associated with the two methods are analyzed in detail
in Refs.8, 9.
If CO2 is expected to be stored in unminable coal
seams, never considered by the mining industry (due to
great depths, the lack of profitability of the project, or
poor safety conditions for workers), it is necessary to
carry out wells with withdrawal of water, and possibly
advantageous to the extraction of methane adsorbed on
coal if intended to store CO2 during drilling. Fig.7 Diagrams of geomechanical problems, consequences,
risks and minimization measures for wells partially cemented
Injection of CO2 for enhancing methane production with cavities.
and sequestration will increase pore pressures in the
coal seam. If pore pressures exceed pre-development and wells partially cemented with cavities, respectively.
levels, there is a risk that slips would occur. This is More details regarding other situations are referred to
conceptually illustrated in Fig.5 [9]. Ref.10.
2.3 Risks associated with the storage
v = 1 The geological storage of CO2 means that CO2 will
be retained for hundreds or thousands of years.
Therefore, it is necessary to carefully evaluate all
potential escape mechanisms. The mechanisms that
may occur in unminable coal seams and abandoned
b = 3
mines are presented in Ref.5. In terms of risk, the
abandoned mines require major rehabilitation work,
checking the conditions for sealing wells and shafts,
and the removal of all materials that might react with
CO2. The existence of wells abandoned or not in
vicinity of reservoir is an important issue to be
analyzed in terms of safety. Figure 8 makes a summary
of some possible leakages of CO2.
Fig.5 Displacements in the fault of a reservoir 9

The causes for geomechanical problems and their

consequences, and the risks and their factors are
summarized in Figs.6 and 7, for wells totally cemented

Fig.8 Potential escape mechanisms 9.

The assessment of risks associated with the storage

of CO2 in unminable coal seams requires to identify
the processes of CO2 leakage and the probability of
occurrence, the escape rate over time, and the
Fig.6 Diagrams for wells totally cemented of geomechanical
implications for a safe long-term storage. A
problems, consequences, risks and factors. quantitative assessment of uncertainties and risks
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 43

associated can only be achieved if the parameters of

the reservoir and the physical processes involved are
well known.
The risk assessment is done by random selection of
input parameters, followed by analysis of results,
assigning a risk value, and ultimately the production of
statistics for the risk profile. This approach can be
implemented by applying the Monte Carlo method or
using BN among other methods. It is often based on
the assumption that the reservoir properties are random
and independent of each other. Other researchers 11, Fig.10 Simplified representation of the geometry of an
12 considered the relationships among the parameters, abandoned mine 4.
the uncertainty and variability of the data, the
Moreover, the complex geometry of a coal mine can
uncertainties of model parameters and the uncertainties
also be translated in a simplified manner by a sealed
associated with risk scenarios considered.
container vertical upwards, according to an idealization
In general, the CO2 retained by adsorption on the
of Piessens and Dusar [4]. In a coal mine, CO2 can be
surface of coal is remained in the deposit, even without
stored in the voids, dissolved in water, or adsorbed on
caprocks, unless the pressure in the coal mine is
the coal matrix. However, coal mines suitable for CO2
reduced through mining. If the pressure drops suddenly,
sequestration should not be flooded. So either it is a
any excess CO2 from coal can flow freely according to
mine without entrance for water (good sealing strata),
one of the mechanisms described previously in the
or, in the most likely case, the CO2 will be disposed of
Fig.8. Therefore, it is necessary to ensure that after
under high pressure. Note that in the first situation, the
storing CO2, the coal is never mined 9.
initial pressure of such reservoirs will be low (near
Figure 9 presents a scheme of storing CO2 in
atmospheric pressure), which means that the initial
carboniferous formations with enhanced coal-bed
state of pressure is at great unbalance with the
methane recovery (ECBMR) [5].
hydrostatic gradients. In the second situation, it is
necessary to ensure that the sealing caprocks, despite
being deformed due to the pressurization of the cavity,
are able to resist this pressure without open cracks or
cause sliding along existing faults 13. Figure 11
presents a schematic diagram of three different ways of
storing CO2 in a coal mine.

Fig.9 Storing CO2 in carboniferous formations with ECBMR 5.

An abandoned coal mine when used as a reservoir

can be seen with a very long curved gallery, as shown
in Fig.10 4. The storage capacity is much greater than
that of unminable coal seams. This is due partly to the
large area of contact between coal and CO2, which
enhances the adsorption phenomenon, but mainly due
to the large void volume constituted by the massive Fig.11 Schematic diagram of three different ways of storing
CO2 in a coal mine 4.
volume mined. The use of abandoned coal mines for
CO2 storage is consequently a good option, particularly The existence of pumping wells or injection of
in China 2, where the number of abandoned mines is fluids is a major source of potential escape problems of
relatively large. CO2. The wells are linear infrastructure that makes the
44 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

connection between surface and underground and masonry is represented. Path (f) shows another
reservoirs, crossing all rock strata, even the most way of leakage between the cement and the strata
impervious. An eventual path to the leakage of CO2 is surrounding the well.
then created. The sealing caprock of the well, the walls
of well, the annular area of interface with the walls, the
first layer of cement case and the involved rock mass
3 Associated risks
are the main elements that should be carefully
analyzed. 3.1 General description
In the presence of water, CO2 becomes carbonic acid, Risk assessment and mitigation strategies are
which can affect the integrity of the casing cement, or developed with the goal of avoiding major problems
even the first cement layer that lies between the walls described above. There are many definitions for risk
and the rock mass. Thus the resistance of the cement assessment. More generally, for an undesirable event E
can be affected. In order to prevent this degradation, an with different consequences, vulnerability levels are
extra thick wall and the introduction of additives to the associated and the risk 14 can be defined as
cement should be considered 5. Figure 12 shows R  P[ E ]P[C | E ]u[C ] (1)
potential escape paths of CO 2 along injecting or where R is the risk; P[ E ] is the hazard, i.e. the
pumping wells. In abandoned wells, the types of probability of the event; P[C | E ] is the vulnerability
escape mechanisms along the walls are similar to those of event E; and u[C ] is the utility of consequence C.
in the wells still in operation. Path (a) in Fig.12 focuses More generally, for different failure events Ej, with
on the flow through the interface of the well casing and which different consequences and hence vulnerability
cement layer on the inside face of the coating. Since levels are associated, expected risk 15 can be defined
both materials are very permeable, runoff is very as
focused in the vertical direction. In path (b), there is an
E[ R]   P[ E j ]P[u (Ci ) | E j ]u (Ci ) (2)
escape mechanism similar to path (a), but it is only j j

between the casing and the cement that leads to the where P[u (Ci ) | E j ] is the vulnerability to the failure
closing hole. In path (c), the mechanism of percolation mode j, P[ E j ] is the probability of failure mode j, and
of CO2 through the cement seal is illustrated. In paths u (Ci ) is the utility of consequence i.
(d) and (e), flow crossing the final layer of concrete For risk evaluation, it is necessary to identify the
tools or models to be used to represent this existing
(a)
knowledge and to perform risk and decision analyses.
Well casing
Cement fill
Risk assessment and risk management for CCS
systems require an evaluation of the hazard and the
(b) assessment of the likelihood of the harmful effects.
Formation Cement
well plug Risk assessment starts with the hazard identification,
rock
which refers to the identification of the major possible
(c) hazards, and focuses on the likelihood of extent of
damage. After the hazard identification, risk
characterization is followed, which involves a detailed
assessment of each hazard in order to evaluate the risk
(d)
associated with each hazard 16.
Based on studies presented in several publications 1,
8, 15, 16, nine hazard identification scenarios are
(e) characterized (Table 1). Once the risks associated with
each hazard are identified, the decision-makers can
develop a basis for their evaluation and the time
(f) necessarily to develop and carry actions to reduce the
risks 16.
3.2 Leakage of CO2 from pipelines or pumping
stations and shipping
Fig.12 Potential escape pathways along wells 9. CO2 from power plants or other industrial facilities
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 45

Table 1 Hazard identification scenarios.

tankers and terminals are clearly much less at risk from
Hazard Description fire, but there is an asphyxiation risk if collision
H1 Leakage of CO2 from pipelines or pumping stations ruptures a tank. The risk can be minimized by making
H2 Leakage of CO2 from shipping sure that high standards of construction are applied. An
H3 Slow and steady leakage of CO2 from geological storage accident with a liquid CO2 tanker might release
H4 Fast and large discharge of CO2 from geological storage
liquefied gas to the surface of the sea. CO2 would
H5 Leakage from geological storage to groundwater
behave differently from liquefied natural gas (LNG)
H6 Leakage of CO2 from geological storage to fossil fuel assets
because liquid CO2 in a tanker is not as cold as LNG.
H7 Leakage of CO2 that eliminates the benefits of geological storage
H8 Induced fracturing or seismicity Its interaction with the sea could be complex. Some of
H9 Leakage from abandoned coal mines the gas would dissolve in the sea, but some would be
released into the atmosphere. With little wind and
can be transported to storage sites by pipelines. For temperature inversion, CO2 gas might lead to
any transportation option, there are calculable and asphyxiation and stop the engines. The risk can be
perceivable risks. CO2 pipelines provide a direct route minimized by carefully planning routes and ensuring
to harmful human exposure or harmful impacts on high standards of training and management 1.
animals and plants by producing a local high 3.3 Slow and fast leakage of CO2 from geological
concentration of CO2 and generating exposures storage
sufficient to harm or kill people, plants and animals 1. Leakage of CO2 from the geological reservoir can
While important risk precautions can be taken to produce two types of hazards, depending on how slow
minimize the likelihood of a major pipeline rupture. or fast the leakage is 16. For slow and steady leakage
The long-distance CO2 pipelines existing in USA are of CO2 from geological storage, the release is too small
illustrated in Fig.13. Special emphasis on the Cortez to cause significant deaths or injuries. However, the
pipeline (808 km long), the Sheep Mountain pipeline leakage can cause local problems including human
(660 km long) and the Weyburn pipeline (330 km long) fatalities. For fast and large discharge of CO2 from
can be made. Measures taken to minimize the risks geological storage, it can cause large-scale fatalities,
from CO2 pipelines 1, 16 include: (1) to localize although the occurrence is rare. An example is the
pipelines away from populous areas; (2) to avoid disaster occurred in 1986 at Lake Nyos in Cameroon.
pipelines near populated valleys where leaking CO2 About 1 700 persons and 3 500 cattle were killed when
the lake released a large amount of CO2. Possible
could accumulate to dangerous levels; (3) to monitor
actions and measures for these hazards can be referred
pipelines against corrosion and to monitor regularly for
to in detail in Ref.16.
leaks; and (4) to install safety valves to shut off the
3.4 Leakage from geological storage to groundwater
pipeline in case of a large leak.
CO2 migration from a storage reservoir to the
surface potentially affects the shallow groundwater
used for potable water and industrial and agricultural
needs. Dissolved CO2 forms carbonic acid alters the
pH value of the solution and potentially causes indirect
effects including mobilization of metals (toxic),
sulphate or chloride. It possibly gives the water an odd
odor, color or taste. In the worst case, contamination
might reach dangerous levels, excluding the use of
groundwater for drinking or irrigation 1.
Among other measures to minimizing the leakage
Fig.13 CO2 pipelines in North America 1.
for the groundwater 17, it is relevant to develop
appropriate inspection methodologies coupled with the
Leakage of CO2 from shipping can occur in different use of dynamic BN for risk analysis 15.
ways, namely, through collision, foundering, stranding 3.5 Leakage of CO2 from geological storage to fossil
and fire 1. The accidents can occur due to the poor fuel assets
maintenance of ships, and the crew by inadequately Underground injection of CO2 at high pressures can
trained people, system failures and human errors. CO2 lead to seepage of fossil deposits through faults and
46 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

other discontinuities or not well-sealed wells. The

contamination of the fossil reservoir induces a severe 4 Preventing risks by monitoring
economic risk since the contamination decreases the
value of the fossil fuel. The probability of this hazard In order to prevent potential risks, monitoring is
occurrence is similar to that of CO2 leakage to the needed. Measurements of certain parameters should be
groundwater. made to assess the behavior of the CO2 system. The
Actions to reduce risks of leakage to fossil fuels monitoring results must be compared with the ones
include: (1) to select reservoir sites that are likely to predicted by modeling and risk analysis. The models
retain CO2 for at least a thousand years; and (2) to can be updated after careful interpretation of a set of
select sites that are far away from fossil fuel assets. observed results.
3.6 Leakage of CO2 eliminating the benefits of Monitoring is performed for various purposes 1,
geological storage and induced seismicity including: (1) to ensure and document the volume of
Leakage from a reservoir returns CO2 into the CO2 injected into wells, specifically to monitor the
atmosphere. The sequestration of CO2 is intended to conditions of the injection well and to measure the
last for a long period of time. Then, when CO2 leaks at rates of injection, as well as the pressures on the top of
a fast rate, the benefits of the geological storage are the well and in the formation; (2) to verify the amount
eliminated and additional costs are incurred. Some of injected CO2 stored by different mechanisms; (3) to
actions can be performed in the case of inadequately optimize the efficiencies of the storage project through
sealed wells. Wells can be monitored to ensure that the knowledge of the volume storage, the most
they are adequately sealed and additional activities can appropriate injection pressures and the need for drilling
be performed to better seal the wells. new wells; (4) to demonstrate, with appropriate
Geological carbon sequestration into porous rock monitoring techniques, that CO2 is still contained in
masses at a high pressure can induce fracturing and the intended storage formations; (5) to detect leakages
movements along faults. The resultant stresses can and to provide a early-warning of any occurrence, so
fracture the surrounding rock. This may pose two types that the situation can be remedied by appropriate
mitigation measures; (6) to know the integrity of wells
of risks: (1) brittle failure and associated microseismicity
that are being used or are abandoned; (7) to calibrate
that provide pathways for CO2 migration; and (2) fault
and verify models for determining the performance;
activation that can induce earthquakes large enough to
and (8) to detect the microsismicity associated with the
cause damage 1. So far, only moderate earthquakes
storage projects.
have occurred due to injection. Eventual actions to
Before CO2 storage, it is necessary to measure most
reduce risks induced by fracturing or seismicity are
relevant parameters to be controlled and to characterize
referred to Ref.16.
the site, in order to know the initial situation (baseline)
3.7 Leakage from abandoned coal mines
that will be used in future comparisons. It is convenient
In coal mines, slow migration towards the surface is
to perform several in-situ tests over different seasons,
not a direct threat to human and nature. However, high
since some properties have a natural variability. This
concentrations can be reached by a sudden or
need is particularly felt when the remote sensors are
temporary release of CO2. Because CO2 is much
denser than air, it could be up to high concentrations in used, for example, the seismic sensors. This is
depressions and confined areas near the surface and particularly true for seismic and other remote-sensing
cause problems to human, which is a known risk that technologies, where the identification of saturation of
happens in volcanic lakes. Leakage may also occur fluids with CO2 is based on comparative analysis.
along infrastructure, case of wells, and faults. The Monitoring the initial situation is also a prerequisite for
effect of active faults on sealing properties of the geochemical analysis, where anomalies. relative to
overburden is an important safety issue and it should background concentrations 9, 17.
be considered. A technical obstacle for injection of Measurement of CO2 injection is a common practice
CO2 into the abandoned coal mines is the low initial in oil and gas fields, and the instruments for this
reservoir pressure. purpose are available in the market. Measurements are
More details on the feasibility of CO2 sequestration made by gauges at the wellhead injection or in the
in coal mines and eventual actions to be considered to vicinity of the injection tube. The accuracy of
reduce the risk are referred to Ref.13. measurements depends on a number of factors 1. For
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 47

CO2, the accurate estimation of the density is very data provide information on the state of the CO2
important for improving the measurement accuracy. (supercritical, liquid or gaseous) and precise values of
Small changes in temperature, pressure and the quantity of CO2 injected. This information may be
composition can have large effects on the density. used for verification and possible updating of the
Measurements of injection pressure at the surface and model adopted.
in the rock formations are also usually performed. Figure 14 presents a methodology that can be used
Gauges are installed in most injection wells through by monitoring for the long-term integrity analysis of a
holes on the surface piping near the wellhead. well in terms of risk evaluation.
Measurements of pressure in the well are routine. A
wide variety of pressure sensors are available and
adequate to monitor pressures at the wellhead or in the
rock formations. The data are continuously available.
The surface pressure gauges are often linked to shut-off
valves that will stop or reduce the injection pressure to a
certain limit if the pressure exceeds a pre-determinated
safe value, or if there is a drop in pressure as a result of
a leakage  1  . Fiber-optic pressure sensors and
temperature sensors are available. These systems should
provide more reliable results, as well as better control of
the well. The current state of technology is more Fig.14 General methodology for integrity analysis of a well.
sufficient to meet the needs of monitoring rates of
injection, and the pressures on the top of the hole. The way that CO2 distributes and moves under-
Combining with temperature measurements, the ground can be monitored in several ways. Table 2 [1]

Table 2 Summary of direct and indirect techniques that can be used to monitor CO2 storage projects 1.

Measurement Measurement Example

technique parameters applications
(1) Travel time; (1) Tracing movement of CO2 in the storage formation;
Introduced and natural
(2) Partitioning of CO2 into brine or oil; (2) Quantifying solubility trapping;
tracers
(3) Identification of sources of CO2 (3) Tracing leakage
(1) CO2, HCO3 , CO32 ;
(1) Quantifying solubility and mineral trapping;
Water composition (2) Major ions; (2) Quantifying CO2-water-rock interactions;
(3) Trace elements; (3) Detecting leakage into shallow groundwater aquifers
(4) Salinity
(1) Formation pressure; (1) Control of formation pressure below fracture gradient;
Subsurface pressure (2) Annulus pressure; (2) Wellbore and injection tube condition;
(3) Groundwater aquifer pressure (3) Leakage out of the storage formation
(1) Brine salinity; (1) Tracing CO2 movement in and above storage formation;
Well logs (2) Sonic velocity; (2) Tracking migration of brine into shallow aquifers;
(3) CO2 saturation (3) Calibrating seismic velocities for 3D seismic surveys
(1) P- and S-wave velocities;
Time-lapse 3D seismic
(2) Reflection horizons; Tracing CO2 movement in and above storage formation
imaging
(3) Seismic amplitude attenuation
Vertical seismic profiling (1) P- and S-wave velocities;
(1) Detecting detailed distribution of CO2 in the storage formation;
and crosswell seismic (2) Reflection horizons;
(2) Detecting leakage through faults and fractures
imaging (3) Seismic amplitude attenuation
Location, magnitude and source characteristics (1) Development of microfractures in formation or caprock;
Passive seismic monitoring
of seismic events (2) CO2 migration paths
Electrical and (1) Formation conductivity; (1) Tracking movement of CO2 in and above the storage formation;
electromagnetic techniques (2) Electromagnetic induction (2) Detecting migration of brine into shallow aquifers
Time-lapse gravity (1) Detect CO2 movement in or above storage formation;
Density changes caused by fluid displacements
measurements (2) CO2 mass balance in the subsurface
(1) Tilt;
(1) Detect geomechanical effects on storage formation and caprock;
Land surface deformation (2) Vertical and horizontal displacements using
(2) Locate CO2 migration pathways
interferometry and GPS
Visible and infrared imaging
Hyperspectral imaging of land surface Detect vegetative stress
from satellite or planes
CO2 land surface flux
CO2 fluxes between the land surface and
monitoring using flux Detect, locate and quantify CO2 releases
atmosphere
chambers or eddycovariance
(1) Detect elevated levels of CO2;
(1) Soil gas composition;
Soil gas sampling (2) Identify source of elevated soil gas CO2;
(2) Isotopic analysis of CO2
(3) Evaluate ecosystems impacts
48 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

summarizes various techniques and their applications

to CO2 storage projects. The applicability is different 5 Development of methodologies for
from place to place and from reservoir to reservoir. A
risk evaluation
case study of monitoring is conducted at Sleipner gas
field in the middle of the North Sea. One Mt CO2 has
5.1 General
been injected at this reservoir per year since September
There are a number of models available for data
1996 1, 17. CO2 is injected into salt water containing
analysis and representation, including event trees, rule-
sand layer, called Utsira formation, 1 000 m below sea
based systems, fuzzy-rule based systems, artificial
bottom. In 1999, the project started to monitor CO2
neural networks, and BN. There are also several
behavior and has established a baseline for the first
techniques for data analysis such as classification,
seismic survey. The project is being carried out in
density estimation, regression and clustering 15.
three phases (Phase 0, 1 and 2). The last phase
Knowledge representation systems (or knowledge
involves data interpretation including monitoring and
based systems) and decision analysis techniques were
model verification. The transport of CO2 plume in the
both developed to facilitate and improve the decision-
storage formation has been monitored successfully by
making process. Knowledge representation systems
seismic time-lapse surveys (Figs.15 and 16). Work at
use various computational techniques of artificial
Sleipner demonstrates that conventional time-lapse P-
intelligence for representation of human knowledge
wave seismic data can be a successful monitoring tool
and inference. Decision analysis uses decision theory
for CO2 injected into a saline aquifer with CO2
and principles supplemented by judgment psychology
accumulation 18.
19. Both are emerged from research done in the
1940s, regarding development of techniques for
Sleipner A Utsira formation problem solving and decision making. More recently,
Sleipner T Norway there has been a resurgence of interest by many
Sleipner
Licence artificial intelligence researchers in the application of
Gas from west
Sleipner
Scotland probability theory, decision theory and analysis to
CO2 injection well several problems, resulting in the development of BN
CO2
Utsira formation
(800–1 000 m
and influence diagrams, an extension of BN designed
in depth) to include decision variables and utilities.
East Sleipner 5.2 BN
-Production and injection wells
Over the last decade, BN has become a popular
representation for encoding uncertain expert
knowledge in expert systems 20. BN can be used at
East Sleipner field
any stage of a risk analysis, and may substitute both
Fig.15 Simplified diagram of Sleipner CO2 storage project 1. fault trees and event trees in logical tree analysis.
While common causes or more general dependency
CO2 injection in the Utsira formation 2001 1999 phenomena pose significant complications on the
Sleipner A classical fault tree analysis, this is not the case with BN.
They are in fact designed to facilitate the modeling of
such dependencies. BN provides a strong tool for
Thickness maps of the
most extensive layer decision analysis, including prior analysis, posterior
2001 1999 1996 analysis and pre-posterior analysis. Furthermore, they
can be extended to influence diagrams, including
decision and utility nodes in order to explicitly model a
decision-making problem 21.
Bright seismic reflections A BN is a graphical representation of knowledge for
2001 perspective view of CO2 accumulations indicate thin layers of CO2
reasoning under uncertainty. It is a concise
Fig.16 Repeated seismic surveys and position of injected CO2 representation of the joint probability of the domain
17. that is being represented by the random variables. It is
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 49

a graph 22 that consists of: (1) a set of random of a variable (or subset of variables) given the
variables that make up the nodes of the network; (2) a observation:
set of directed links between nodes (these links reflect P( A, e )
P( A | e )  (5)
cause-effect relations within the domain); (3) each   P( X1 ,, X k , A, e)
variable has a finite set of mutually exclusive states; (4) X1 Xk A

the variables together with the direct links form a where e is the vector of all the evidence.
direct acyclic graph (DAG); and (5) attached to each 5.2.1 Inference for BN
random variable A with parents B1, B2, . . . , Bn, there is There are two main groups of inference algorithms:
a conditional probability table P(A | B1, B2, . . . , Bn), exact inference method and approximate inference
algorithm. The most common and exact inference
except for the variables in the root nodes. The root
method is the variable elimination algorithm that
nodes have prior probabilities.
consists of eliminating (by integration or summation)
Figure 17 is an illustration of a simple BN. The
the non-query, non-observed variables one by one by
arrows going from one variable to another reflect the
summing over the product. The approximate inference
relations between variables. In this example, the arrow
algorithms are used when exact inference may be
from C to B1 means that C has a direct influence on B1.
computationally infeasible, such as that in temporal
models (dynamic BN), where the structure of the
network is very repetitive, or in highly connected
networks.
(1) Dynamic Bayesian network (DBN)
DBN is the BN that represents sequences of
variables. It is often applied to temporal data such as
speech recognition, visual tracking, and financial
forecasting; however, it is also used in sequence data
analysis, e.g. Biosequence analysis, text processing
Fig.17 An illustration of a simple BN.
among others. It is mostly used for the problems such
as classification, state estimation, fault diagnosis,
Specifically, a BN is a graphical and concise
prediction, etc..
representation of a joint probability distribution over A specific case of a DBN is presented in Fig.18.
all the variables, taking into account that some This DBN represents a hidden Markov model (HMM),
variables are conditionally independent. The simplest where each state Xi generates an observation Yi. The
conditional independence relationship encoded in BN structure and the variables are repeated over time.
is that a node is independent of its ancestors, given its
parents, i.e. a node only depends on its direct parents.
Thus, the joint probability of a BN over the variables U =
{A1, A2, . . . , An} can be represent by the chain rule:
n
P(U )   P( Ai parents ( Ai )) (3) Fig.18 DBN representing a HMM.
i

where parents (Ai) is the parent set of Ai. In order to represent such DBN, we need: (a) initial
Since a BN defines a model for variables in a certain distribution P( X 1 ) ; (b) transition model, i.e. transition
domain, its relationships can be used to answer probability distributions P( X i  1| X i ) ; and (c) sensor
probabilistic queries about them. The most common model P(Yi | X i ) .
types of queries are as follows: (2) Inference in DBN
(1) A priori probability distribution of a variable: The problem of inference in DBN is NP-hard. There
P( A)    P( X 1 , , X k , A) (4) are several algorithms divided into two groups, i.e.
X1 Xk exact inference algorithm and approximate algorithm.
where A is the query-variable; and X i (i = 1, 2, . . . , k) For exact algorithm, we need: (a) forwards-
is the remaining variables of the network. backwards smoothing algorithm (on any discrete-state
(2) Posterior distribution of variables given evidence DBN); (b) the frontier algorithm; (c) the interface
(observation). This query consists of updating the state algorithm; and (d) Kalman filtering and smoothing.
50 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

For approximate algorithm, we need: (a) the Boyen-

Koller (BK) algorithm; (b) factored frontier (FF)
algorithm; (c) loopy propagation algorithm (LBP).; (d)
Kalman filtering and smoother; (e) stochastic sampling
algorithm; (f) importance sampling or MCMC; (g)
particle filtering (PF); and (h) influence diagrams
(decision graph).
BN can serve as a model of a part of the world, and
the relations in the model reflect causal impact among
events. However, the reason we are building models is
to use them when making decisions (i.e. the
probabilities provided by the network are used to
Fig.19 Influence diagram.
support some kinds of decision-makings). Decision
graph and influence diagram are both an “extension” device” represents the fact that a warning alarm may
of BN. In addition to nodes for representing random be issued or not. The decision node represents the
variables, influence diagrams also provide node types decision evacuating a population or not. The utility
for modeling alternatives and utilities. Besides the node (“consequences”) represents the consequences
chance nodes that denote random variables and (expressed in utilities of the decision) in combination
correspond to the only node type available in belief with the occurrence or not of the threat. The warning
networks, the decision nodes are also modeled. A device issuing an alarm depends directly on the
decision node indicates a decision facing the decision- possibility of occurrence of the threat. The decision of
maker (similar to decision nodes in decision trees) and evacuating the population or not will depend directly
contains all alternatives available to the decision- on the warning device issuing an alarm. Finally, the
maker at that point. The third node type provided by consequences will depend on the decision taken and
these diagrams is the utility node. These nodes whether or not the threat actually happens.
represent the utility function of the decision-maker. In There are mainly four types of connections for
the utility nodes, utilities are associated with each of
structural influence in a decision graph. They are
the possible outcomes of the decision problem
represented in Fig.20.
modeled by the influence diagram.
Direct links between nodes represent influences.
Links between two chance nodes have the same
semantics as in the belief networks. Other links in an
influence diagram may also represent a temporal
relation between the nodes involved. For example, a (a)

link from a decision node to a utility node indicates

that not only the choice of action influences the utility,
but also the decision precedes the outcome in time.
Influence diagrams are useful in structuring a
decision problem. While, for example, decision trees (b)

are more effective in presenting the details of a

decision problem, influence diagrams more clearly
show the factors that influence a decision. Figure 19
illustrates a simplified scheme of an influence diagram. (c)

It is composed of two chance nodes (“threat” and

“warning device”), one decision node (“decision”) and
a utility node (“consequence”)). In this specific
example, the chance node “threat” can represent the
occurrence or not of a natural threat (for example, a (d)

tsunami or a hurricane). The chance node “warning Fig.20 Influence diagram connections.
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 51

The first one (Fig.20(a)) is used when a decision

(decision 1) affects the probabilities of event 1, i.e. 6 Application of BN
decision 1 is relevant for event 1. In Fig.20(b), the
outcome of event 1 affects the probabilities of event 2,
In this section, the examples of BN and DBN are
i.e. event 1 is relevant for event 2. This is a typical BN
presented to illustrate their potential use for risk
without decision included. The type of connection in analysis in CO2 injection processes. The first example
Fig.20(c) is used when decision 1 occurs before is developed for a situation where one wants to
decision 2, i.e. decisions 1 and 2 are sequential. Finally, determine whether or not it is beneficial to inject CO2
Fig.20(d) represents a connection used when decision in carboniferous formations at a certain location. This
1 occurs after event 1. In this case, the outcome of example, is based on hazards H3 and H4 defined
event 1 is known when making decision 1. previously. In this example, the decision-maker is
Besides the structural influences described in Fig.20, looking at different mitigation measures (for reducing
there are also value (utilities) influences such as the the leakage of CO2), assessing the risk of each option
ones illustrated in Fig.21. and choosing the one that can minimize it. Finally, an
example of a DBN is presented to illustrate the use of
DBN coupling with results of a monitoring system.
6.1 Risk analysis for storage of CO2
For the risk analysis due to CO2 injection in
carboniferous formations, a BN is developed, as
presented in Fig.22. The involved variables are
(a) associated with:

(b)
Fig.21 Value influences.

In Fig.21(a), the value (utility) depends on the

(uncertain) event, for example, a manufacturing cost
depends on the (uncertain) availability of a certain
input. In the second value influence (Fig.21(b)), a
decision influences the value (utility). For example, a
manager’s decision influences the profit of a plant. Fig.22 BN for risk analysis of storage of CO2.
5.2.2 Inference for influence diagrams
The inference process in an influence diagram (1) Sedimentary strata conditions over the
consists of computing the expected utility associated carboniferous formations. Three values are adopted for
with different decisions or strategies. As in BN, there the formations: good, bad and very bad.
are two groups of algorithms that can be used to make (2) Coal seams characteristics. Three distinct values
inference in an influence diagram exactly and are taken: good, bad and very bad.
approximately. The most basic way to solve an (3) Combined characteristics due to the association
influence diagram is to unfold it into a decision tree of sedimentary strata and coal seams. The values are
and solve it. However, if one wants to take advantage attributed in function of the properties defined to both
of the structure of an influence diagram and encoded formations.
conditional independences, one of the most common (4) Geomechanical characteristics of the wells. Two
issues is the variable elimination algorithm for values are adopted for the shaft: good and bad in
influence diagrams, which has many similarities to the function of the existing corrosion.
variable elimination technique described for BN. For (5) Corrosion of the well. Two levels are considered:
more details, it can be referred to Refs.23, 24. level 1 (reasonable) and level 2 (bad).
52 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

(6) Existence of faults. Two hypotheses are

considered: yes and no. Table 7 Escape of CO2.
(7) Escape of CO2. For this situation, the value is Leakage CO2 Faults
Value

considered for the combined characteristics of both High Medium Low

Yes 0.7 0.3 0
formations involved (coal seams and sedimentary Yes
No 0.5 0.4 0.1
strata), the existence of wells and faults, and course Yes 0.05 0.05 0.9
whether CO2 is injected or not. No
No 0.01 0.03 0.96
(8) Injection of CO2. For this situation, two distinct
values (yes or no) are considered. Table 8 Damage values.
Injection of CO2 Damage Utilities
(9) Utilities (consequences). For the utilities, the
High 40
calculated result permits to be concluded whether the Yes Medium 20
rehabilitation measures are adopted or not. Low 20
(10) The calculated risk depends on the escape of High 0
No Medium 0
CO2 or not, and the existence of faults. The following
Low 0
three values are adopted: high, average and low.
In Tables 3 to 9, the local and conditional pro-
Table 9 Utilities associated with different scenarios and
babilities associated with each variable of the BN are decisions with CO2 injection.
represented. The quantification can be based on expert Combined Well Faults
judgment or available data, or a combination of both. characteristics characteristics Yes No
Good (Yes) 0.1 0.9
In this case, all the values are given for illustrative Good (No) 0.01 0.99
purposes. Average
Bad (Yes) 0.6 0.4
Bad (No) 0.2 0.8
Table 3 Sedimentary strata and coal seam characteristics. Good (Yes) 0.3 0.7
Value Good (No) 0.2 0.8
Item Bad
Good Bad Very bad Bad (Yes) 0.7 0.3
Sedimentary strata Bad (No) 0.6 0.4
0.333 0.333 0.333 Good (Yes) 0.6 0.4
characteristics
Good (No) 0.5 0.5
Coal seam Very bad
0.333 0.333 0.333 Bad (Yes) 0.8 0.2
characteristics
Bad (No) 0.7 0.3

Table 4 Combined characteristics of sedimentary strata and coal

seams. Applications were performed through the software
Sedimentary strata Coal seams Combined characteristics Genie (http:genie.sis.pitt.edudownloads.html). Two
characteristics characteristics Good Bad Very bad hypotheses (A and B) were considered, as assigned in
Good 1 0 0
Table 10.
Good Bad 0 1 0
Very bad 0 0.7 0.3
Table 10 Different hypotheses considered in the BN.
Good 0 1 0
Bad Bad 0 1 0 Sedimentary Coal Existence of
Hypothesis Wells Corrosion
Very bad 0 0.3 0.7 strata seams faults
Good 0 0.3 0.7 A Good Good Good — —
Very Bad Bad 0 0.3 0.7 B Good Bad — Level 2 —
Very bad 0 0 1

These are two different hypotheses that we consider

Table 5 Characteristics of the wells.
and want to assess. The risk associated with each
Value
Corrosion hypothesis (A or B) are calculated to make a decision
Good Bad
on whether or not CO2 at that location is injected.
Level 1 0.3 0.7
Level 2 0.6 0.4 For hypothesis A, Fig.23 shows the induced diagram
with probabilities calculations. The results demonstrate
Table 6 Corrosion of wells. clearly that is it beneficial to inject CO2 in the coal
Corrosion Value seams. For the hypothesis B, Fig.24 shows that the BN
Level 1 0.5
diagram recommends not to inject CO2 in the coal
Level 2 0.5
seams.
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 53

problematic. The water will be acidified, which allows

it to degrade geological formations, and the water
saturated with CO2 is not suitable for drinking.
In order to deal with this situation, the BN is
developed. DBN is adopted to evaluate the situation of
contamination of aquifers due to the leakage of CO2.
Two different models are built. One, a prediction
model, is employed to model and predict the CO2
leakage and the influence in the contamination on the
aquifer, based on water quality measurements, as
described in Fig.26, for different instants of time (slice
0 until slice n). The other, a decision model, is based
Fig.23 Diagram for hypothesis A.
on decision graphs indicated in Fig.27. The decision is
made on the optimal remedial measures solution for
the problem that can pass through the decision without
injecting CO2 any more.

CO2 leakage CO2 leakage (t = 1) CO2 leakage (t = 2)

Contamination Contamination
Contamination
level of aquifer (t = 1) level of aquifer (t = 2)
level of aquifer

Fig.24 Diagram for hypothesis B.

Water quality Water quality Water quality
measurement measurement (t = 1) measurement (t = 2)
Another BN is presented in Fig.25 when the active
faults are considered. The consequences in this Slice 0 Slice 1 Slice 2

situation can imply the existence of induced earthquakes.

CO2 leakage (t = 3) CO2 leakage (t = 4) CO2 leakage (t = 5)

Contamination Contamination Contamination

level of aquifer (t = 3) level of aquifer (t = 4) level of aquifer (t = 5)

Water quality Water quality Water quality

measurement (t = 3) measurement (t = 4) measurement (t = 5)

Slice n2 Slice n1 Slice n

Fig.26 Modeling the contamination of the aquifer by leakage of
CO2.

Remedial Leakage
measures rate of CO2 (t)

Fig.25 BN for risk analysis of storage of CO2 with the existence Contamination
of active faults. level of aquifer (t)

6.2 Contamination of aquifers by CO2

Water quality
Cost
Contamination of aquifers corresponds to the hazard Consequences measurement (t)
H5—leakage from geological storage to groundwater,
Fig.27 Decision model based on the water quality
according to different hazards defined in Table 1. CO2 measurements.
injected into the ground will be dissolved into water,
including pore water between grains or minerals in the The way that the model works is listed as follows
geological formations. Dissolution into water can be (Fig.28):
54 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

CO2 leakage CO2 leakage (t = 1) CO2 leakage (t = 2) CO2 leakage (t = 3) CO2 leakage (t = 4) CO2 leakage (t = 5)

Contamination Contamination Contamination Contamination Contamination

Contamination
level of aquifer (t = 1) level of aquifer (t = 2) level of aquifer (t = 3) level of aquifer (t = 4) level of aquifer (t = 5)
level of aquifer

Water quality Water quality Water quality Water quality Water quality Water quality
measurement measurement (t = 1) measurement (t = 2) measurement (t = 3) measurement (t = 4) measurement (t = 5)

Slice 0 Slice 1 Slice 2 Slice n2 Slice n1 Slice n

(a) Step 1: evidence (water measurement = good).

CO2 leakage CO2 leakage (t = 1) CO2 leakage (t = 2) CO2 leakage (t = 3) CO2 leakage (t = 4) CO2 leakage (t = 5)

Contamination Contamination Contamination Contamination Contamination

Contamination
level of aquifer (t = 1) level of aquifer (t = 2) level of aquifer (t = 3) level of aquifer (t = 4) level of aquifer (t = 5)
level of aquifer

Water quality Water quality Water quality Water quality Water quality Water quality
measurement measurement (t = 1) measurement (t = 2) measurement (t = 3) measurement (t = 4) measurement (t = 5)

Slice 0 Slice 1 Slice 2 Slice n2 Slice n1 Slice n

(b) Step 2: propagation of evidence in the same time slice. DBN results: P (contamination level = high) = 0.09; P (CO2 leakage rate = high) = 0.10.

CO2 leakage CO2 leakage (t = 1) CO2 leakage (t = 2) CO2 leakage (t = 3) CO2 leakage (t = 4) CO2 leakage (t = 5)

Contamination Contamination Contamination Contamination Contamination

Contamination
level of aquifer (t = 1) level of aquifer (t = 2) level of aquifer (t = 3) level of aquifer (t = 4) level of aquifer (t = 5)
level of aquifer

Water quality Water quality Water quality Water quality Water quality Water quality
measurement measurement (t = 1) measurement (t = 2) measurement (t = 3) measurement (t = 4) measurement (t = 5)

Slice 0 Slice 1 Slice 2 Slice n2 Slice n1 Slice n

(c) Step 3: propagation of evidence into the future time slices.

Fig.28 Illustration of the steps of the prediction model.

(1) Step 1: observation (water quality measurement)

(contamination level)

1.0
0.8 No
is made at time t0 and enters into the network (in grey).
0.6
Prob

(2) Step 2: the evidence is propagated through the 0.4 High

network at time t0, and the probability of leakage is 0.2 Medium
determined. 0.0
0 1 2 3 4 5 6 7 8 9
(3) Step 3: the evidence is propagated through into Time slice
(a)
the future, and the probability of leakage in the next 1.0 No
leakage rate)

slice of time is determined. 0.8

Prob (CO2

0.6
Once the prediction model has been employed, one 0.4 High
can use its results (Fig.29) to determine the optimal 0.2
remedial measure, which can be invalid if no remedial 0.0
0 1 2 3 4 5 6 7 8 9
measure is considered, by minimizing the risk. Figure 30 Time slice
(b)
shows the decision model with evidence (coming from Fig.29 Results of the execution of BN of Fig.26 with one
the prediction model) entered into the network. observation at time t0.
Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56 55

(1) In the risk management, BN is a powerful tool in

Remedial Leakage
rate of CO2 (t) the decision analysis, including priori and posteriori
measures
analyses.
(2) BN presents the extension of influence diagrams,
Contamination
level of aquifer (t) including the uses of decision nodes and utilities nodes.
(3) BN allows combining the knowledge of experts
and available data through statistical methods.
(4) The beneficial use of DBN in decision processes
with time is very relevant to the application made.
Water quality
Cost Consequences measure (t) The developed models just show how a technique
like BN can be used to assess risk in CO2 sequestration
problems. All the numbers are given for illustrative
Fig.30 The results of the prediction model are entered into the
decision model as evidence (in grey). purposes. The structure of the BN, however, comes
from expert knowledge (based on different hazard
The results of the execution of this model are presented scenarios). The framework itself works (prediction and
in Fig.31. The results show that the best decision given decision models). It has been already applied to a
the water measurement at time t0 is not to apply a tunnel project case (Porto Metro, Portugal), but it can
remedial measure. These steps are then repeated for be applied to any problem where one has observations
each slice of time. in time and space and wants to assess risk and
minimize it by making the “optimal” decision, that in
Remedial measures Leakage
Yes 172.7 rate of CO2 (t) CO2 case means to apply a mitigation measure or not.
No 17.92
Contamination
Acknowledgments
level of aquifer (t)
The authors want to express their acknowledgments
to the support from the State Administration of Foreign
Experts Affairs, China, for the research project “Risk
Water quality
Cost Consequences measure (t) assessment of CO2 injection processes in carboniferous
formations” at State Key Laboratory for Geomechanics
Fig.31 Execution of the decision model of Fig.27. and Deep Underground Engineering, China University
of Science and Technology (Beijing).
7 Conclusions
References

This paper describes briefly the most effective

[1] IPCC. Carbone dioxide and storage. [S.l.]: Cambridge University
technologies for CCS projects. Geological carbon
Press, 2005.
sequestration presents the possibility to reduce
[2] He M, Sun X. Considerations on CO2 storage in abandoned coal
emissions of CO2 into the atmosphere at a low cost
compared to many other options. China has mines in China. In: International Workshop on CO2 Storage in

exceptional conditions to store CO2 in carboniferous Carboniferous Formations and Abandoned Coal Mines. Beijing: [s.n.],

formations, particularly in abandoned coal mines. 2011 (to be published).

Geological carbon sequestration entails risk that [3] International Energy Agency (IEA). Carbon dioxide chemistry and
may be large and significant. However, risks can be properties. http://www.ieagreen.org.uk. 2002.
limited or reduced. Development of methodologies for [4] Piessens K, Dusar M. CO2-sequestration in abandoned coal mines.
risk evaluation based on BN has been made and some
Brussels: Royal Belgian Institute for Natural Sciences, Geological
relevant applications have been performed with
Survey of Belgium, 2003.
particular emphasis on the development of DBN for
[5] International Energy Agency (IEA). CO2 capture and storage—a key
the hazards related to the leakage from geological
storage to groundwater. carbon abatement option. Paris: OECD Publishing, 2008.

Based on the applications of BN, several [6] Vargas E. Simulation of storage processes in geological media. Rio de

conclusions can be drawn: Janeiro: Catholic University of Rio de Janeiro, 2008 (in Portuguese).
CO2 leakage (t =
3)

56 Manchao He et al. / J Rock Mech Geotech Eng. 2011, 3 (1): 39–56

[7] Duguid A, Couëslan M, Tombari J, et al. MMV technologies for [14] Einstein H. Landslide risk: systematic approaches to assessment and

effective and efficient monitoring of geologic carbon capture and management. In: Cruden A, Fell R ed. Landslide Risk Assessment.

storage projects. New Orleans: GWPC/US EPA CO2 MMV Rotterdam: A. A. Balkema, 1997: 25–50.

Workshop, 2007. [15] Sousa R. Risk analysis for tunneling projects. Cambridge: MIT Press,

[8] Myer L R. Geomechanical risks in coal bed carbon dioxide 2010: 589.

sequestration. Berkeley: Lawrence Berkeley National Laboratory, [16] Price P, McKone T, Sohn, M. Carbon sequestration risks and risk

Earth Sciences Division, 2003. management. Berkeley: Lawrence Berkeley National Laboratory,

[9] Gomes A. CO2 injection processes in carboniferous formations. MS Environment Energy Technologies Division, 2008.

Thesis. Porto: University of Porto, 2010 (in Portuguese). [17] Solomon S. Carbon dioxide storage: geological security and

[10] Sousa L, Sousa R. Risk associated to storage in carboniferous environmental issues—case study on the Sleipner gas field in Norway.

formations: application of Bayesian Networks. In: International Oslo: The Bellone Foundation, 2006.

Workshop on CO2 Storage in Carboniferous Formations and [18] Eiken O, Brevik I, Arts R, et al. Seismic monitoring of CO2 injected

Abandoned Coal Mines. Beijing: [s.n.], 2011 (to be published). into a marine aquifer. In: SEG Calgary 2000 International Conference

[11] Elsworth D. Complex process couplings in systems pushed far-far- and 70th Annual Meeting. Calgary: [s.n.], 2000.

from equilibrium: applications to deep geologic sequestration and [19] Henrion M, Breese J, Horvitz E. Decision analysis and expert systems.
energy recovery. In: International Workshop on CO2 Storage in AI Magazine, 1991, 12 (4): 64-91.
Carboniferous Formations and Abandoned Coal Mines. Beijing: [s.n.],
[20] Heckerman D. A tutorial on learning with Bayesian networks. Data
2011 (to be published).
Mining and Knowledge Discovery, 1997, (1): 79–119.
[12] Liu J. Multiphysics of coal-gas interactions. Something old,
[21] Faber M. Risk and safety in civil surveying and environmental
something new and something very new. In: International Workshop
engineering. Switzerland: Swiss Federal Institute of Technology, 2005.
on CO2 Storage in Carboniferous Formations and Abandoned Coal
[22] Russel S, Norvig P. Artificial intelligence: a modern approach. 2nd ed.
Mines. Beijing: [s.n.], 2011 (to be published).
[S.l.]: Prentice Hall, 2003.
[13] Piessens K. CO2 storage in abandoned coal mines. pressure constraints.
[23] Jordan M. Learning in graphical models. Cambridge: MIT Press, 1998.
In: International Workshop on CO2 Storage in Carboniferous

Formations and Abandoned Coal Mines. Beijing: [s.n.], 2011 (to be [24] Jensen F. Bayesian networks and decision graphs. London: Taylor and

published). Francis, 2001.