Leakage risks of geologic CO2 storage and the impacts on the global energy system and climate change mitigation

This study investigated how subsurface and atmospheric leakage from geologic CO2 storage reservoirs could impact the deployment of Carbon Capture and Storage (CCS) in the global energy system. The Leakage Risk Monetization Model was used to estimate the costs of leakage for representative CO2 injection scenarios, and these costs were incorporated into the Global Change Assessment Model. Worst-case scenarios of CO2 leakage risk, which assume that all leakage pathway permeabilities are extremely high, were simulated. Even with this extreme assumption, the associated costs of monitoring, treatment, containment, and remediation resulted in minor shifts in the global energy system. For example, the reduction in CCS deployment in the electricity sector was 3% for the “high” leakage scenario, with replacement coming from fossil fuel and biomass without CCS, nuclear power, and renewable energy. In other words, the impact on CCS deployment under a realistic leakage scenario is likely to be negligible. We also quantified how the resulting shifts will impact atmospheric CO2 concentrations. Under a carbon tax that achieves an atmospheric CO2 concentration of 480 ppm in 2100, technology shifts due to leakage costs would increase this concentration by less than 5 ppm. It is important to emphasize that this increase does not result from leaked CO2 that reaches the land surface, which is minimal due to secondary trapping in geologic strata above the storage reservoir. The overall conclusion is that leakage risks and associated costs will likely not interfere with the effectiveness of policies for climate change mitigation.


Introduction
Meaningful CO 2 emissions reductions achieved through CO 2 capture, utilization, and storage (CCUS) will require the injection of captured CO 2 into sedimentary formations where it is "stored" within the rock matrix for hundreds to thousands of years.(1) Geologic CO 2 storage requires an overlying impervious caprock formation to contain the buoyant CO 2 .(2, 3) But the integrity of this caprock may not be perfect and some of the CO 2 , or the brine it displaces, may migrate out of the storage formation through leakage pathways (e.g., wells, faults, and fractures) that perforate the caprock.(4, 5) Numerous studies have investigated the physical causes and consequences of this leakage, including: characterizing leakage pathways and caprock integrity;(615) simulating horizontal and vertical fluid migration through stratigraphic sequences of sedimentary basins;(1619) investigating the outcomes of leakage into overlying formations, groundwater, the unsaturated zone, and to the atmosphere;(2024) developing approaches to verify storage and detect the movement and leakage of CO 2 ;(25) and remediating leakage by natural or engineered approaches. (26,27) Those investigations have yielded essential understanding of geophysical aspects of storage integrity, but leakage events are not all the same in terms of financial implications. As we and others have shown, (28,29) there are unique costs associated with the possible scenarios including CO 2 leaking but secured by secondary trapping, CO 2 leaking and interfering with valuable resources, CO 2 reaching potable aquifer resources, and CO 2 migrating all the way to the atmosphere. In order to effectively compare storage reservoirs and injection locations, a single monetary metric is needed to interpret leakage predictions. Operators, regulators, policymakers, and the public require reliable financial riskbased information for site selection, liability, compensation, and expected conformance with regulations.(30) Frameworks for geologic CO 2 storage risk assessment have been developed, (3133) but they do not monetize the consequences of imperfect storage and leakage remediation.
In this paper we introduce the Leakage Risk Monetization Model (LRiMM), a new approach that integrates and extends two pieces of our prior work. One piece established a method to conduct probabilistic simulations of injection and leakage to produce an expectation of the extent to which leakage could migrate vertically and horizontally through the overlying formations in the hydrostratigraphic sequence.(34) The other piece established the Leakage Impact Valuation (LIV) method to estimate the economic costs of a single leakage event.(28) LRiMM extends that work by estimating geospatial probabilities and extents of leakage and the associated economic costs. The resulting monetized leakage risk (MLR) serves to bound uncertainty about the performance of individual sites, and establishes the basis for modeling of the regional, national, and global deployment potential of CCUS. LRiMM also provides information about potential harms and financial burdens to specific stakeholders, thereby establishing mechanisms to compensate stakeholders who may be negatively affected but unable to directly benefit from the activity.
To demonstrate LRiMM, we compare two case studies of CO 2 injection into a deep saline aquifer in the Michigan sedimentary basin. Specifically, we investigate (a) the magnitude of MLR relative to CO 2 capture costs and with regard to specific stakeholders, (b) the effect of secondary trapping, whereby overlying sedimentary layers trap CO 2 that has leaked through the primary seal of the storage formation, (c) the effects of geospatial proximity of injection wells, leakage pathways, and other subsurface resources, and (d) how the ability to detect and remediate leakage would reduce MLR and stakeholder exposure. 

The Leakage Risk Monetization Model: LRiMM
LRiMM is designed to characterize leakage risk from perspectives ranging from an individual CO 2 injection project to regional planning at the scale of sedimentary basins. As shown in Figure 1, LRiMM uses 3D geospatial data on hydrostratigraphic units, other subsurface activities, and potential leakage pathways that intersect the impermeable strata. These data are used in geophysical simulations to produce a probabilistic characterization of the spatial patterns and temporal evolution of leakage. In principle, the LRiMM framework can use any geophysical model that simulates sitespecific fluid migration, plume evolution in aquifers, and leakage through any type of pathway. Here, we used a semianalytical model of CO 2 injection, brine displacement, and upconing and leakage through active or abandoned wells.(16, 35) The use of a computationally tractable model enables simulation of multiple injection and leakage scenarios, in which parameters with critical uncertainties (storage reservoir permeability κ aq , storage reservoir porosity ϕ aq , pathway leakage permeability κ leak ) are treated probabilistically. As implemented for this work, a 3D recording grid was constructed throughout the domain, which was used to track for each model condition at each time t (a) the fraction of the thickness of an aquifer that contains CO 2 (0 ≤ h CO2 ≤ 1), and (b) the increase in brine pressure (P i,t ). The resulting probabilities of CO 2 or brine leakage were determined by calculating the fraction of simulations in which CO 2 is present (p CO2,t ), and in which there is an increase in brine pressure (p Pi,t ), respectively. These probabilities establish the likelihood that CO 2 will be present or an increase in brine pressure will occur at a recorded location at time t. (See the Supporting Information for more details.) Additionally, the total mass of CO 2 accumulated in each aquifer (M t ) was quantified. LRiMM differentiates between primary leakage out of the storage reservoir and secondary leakage through pathways above the storage formation. LRIMM classifies leakage into four categories of outcomes, each with unique economic costs.
(28)CO 2 or displaced brine could: (o1) leak out of the storage reservoir but remain secure through secondary trapping, (o2) interfere with existing subsurface activities, (o3) migrate into potable groundwater, or (o4) reach the land surface. (1)where weights are determined from the average fraction of the thickness of an aquifer occupied by CO 2 , (2)and the average increase in the brine pressure above hydrostatic conditions, P h , (3)at each recorded location. (See the Supporting Information for more details.) We defined MLR t as the expected cost of leakage, R t , divided by the cumulative amount of CO 2 injected up to that point in time, t.
(4)where r is the injection rate and the weight W 3,t incorporates the potential for multiple leaky wells (n) in a grid cell at a recorded location, In this paper, we use z = 0.5 under the assumption that there are cost efficiencies for addressing leaky wells as well density increases. The MLR in eq 4 is determined for each stakeholder and for each aquifer over all grid cells. The results are summed to determine the total MLR.
While the MLR in eq 4 assumes no remediation of leakage, Above Zone Monitoring Intervals are dictated in rules governing CO 2 injection for storage,(36) and leakage that is detected may require remediation. The amount of CO 2 that leaks over time is a priori uncertain, and thus interventions to fix leaks will occur at indeterminate points in the future based on the amount of CO 2 that has leaked and is detected. For a given point in time, LRiMM determines a probability of intervention (q t (L)) that depends on the quantity of CO 2 (L) that is detected in aquifers overlying the injection reservoir. This probability distribution could be a function of the monitoring technology, but here we assume perfect detection and remediation technologies so that q t (L) represents the probability that a detectable amount of CO 2 has accumulated outside of the reservoir and intervention occurs to perfectly remediate that leakage. This probability of intervention is calculated by determining the portion of the simulations where at least L has accumulated in the overlying aquifer (see the Supporting Information for more details.) This probability is used to modify R t , and the MLR arising from unabated leakage to produce the InterventionAdjusted Risk (IAR) or the InterventionAdjusted MLR (6b)where t indexes years since the beginning of injection. CO 2 injection begins at t = 0 and thus q 0 (.) = 0. Eqs 6a and 6b show that IAR and IAMLR are forwardlooking projections; if q t (L) is the probability of perfectly remediating leakage after L is detected outside the reservoir, for leakage risk to exist at time t, L would not have been detected and leakage remediated before t, when < t τ . 

Application to the Michigan Sedimentary Basin
The Mt. Simon Sandstone is a primary candidate for CO 2 storage in the Michigan Sedimentary   Each case study simulates injection of 9.5 MtCO 2 /yr over 30 years into the Mt. Simon sandstone.
The CO 2 is presumed to be captured from the James De Young (JDY) power plant (0.6 MtCO 2 /yr) and the J.H. Campbell power plant (8.9 MtCO 2 /yr), 10 km to the northwest of JDY,(39) assuming 85% capacity factors and 90% capture efficiencies. Location 1 involves injection at the JDY site at a depth of 1600 m, and was chosen because eight UIC Class I disposal wells terminate in the Mt.
Simon within 1740 m of this site. These wells are potential leakage pathways, and six of them are active and thus locations where CO 2 or displaced brine could interfere with a subsurface activity.
Location 2 is 50 km north of JDY, where the Mt. Simon is deeper, CO 2 is injected at a depth of 1,950 m, and the nearest well that extends into the Mount Simon is ∼20 km to the southeast.
In general, leakage from the Mt. Simon that gets into the overlying Galesville formation may be contained within that formation (i.e., secondary trapping in the Galesville) and not migrate to shallower units (e.g., TrentonBlack River/St. Peter) if the resistance to horizontal flow-a result of the pressure perturbation from injection into stratigraphic nature of sedimentary basins-is less than the resistance to vertical flow; or, if the resistance to vertical flow through permeable leakage pathways is less than the resistance to horizontal flow, leakage may continue upward through the same pathways into shallower formations, or encounter new leakage pathways that do not extend into the storage reservoir (i.e., secondary leakage from the Galesville). Since wells extend from the surface into the subsurface, fluids leaking upward from a deep CO 2 reservoir will likely encounter more wells-and more opportunities for interference with other subsurface activities-if these leaked fluids migrate shallower into the sequence.
The injection and leakage approach we employed has been used in previous studies, where an alternating aquiferaquitard sequence was constructed from the hydrostratigraphic units.   Over time, leakage migrates into the TrentonBlack River/St. Peter and further up to the Traverse Dundee/SilurianDevonian, some of which occurs through wells that terminate in the TrentonBlack River/St Peter. The MLR for this secondary leakage is small relative to MLR for primary leakage out of the Mt. Simon: ∼ $0.08/tCO 2 in the first year of this secondary leakage (year 3) and reaches a maximum of $0.34/tCO 2 in year 9. In year 5, leakage may encounter oil and gas production in the TraverseDundee/SilurianDevonian. The LIV estimates ($2010 USD) for an interference with oil production are $2.2 million (I L 2 ) and $89 million (I H 2 ), and with natural gas production are $4.4 million (I L 2 ) and $100 million (I H 2 .(28) But the MLR for this interference is at most $0.001/tCO 2 over the remaining 25 years because secondary trapping reduces the amount of leaked fluids that enter the TraverseDundee/SilurianDevonian and the expected P i continues to be small relative to P h . On average, 91% of the total MLR over time is due to primary leakage out of the Mt. Simon through the UIC Class I wells, or interference with the waste disposal by these wells. Figure 5 shows the MLR for location 2, where the closest well that penetrates into the Mt. Simon is far outside the CO 2 plume. The pressure perturbation in the Mt. Simon drives brine through this and other primary leakage pathways, but P i in overlying units is small relative to P h . As a consequence, the MLR at location 2 reaches $0.04/tCO 2 -more than 2 orders of magnitude below the maximum MLR at location 1, and at most equal to the portion of the MLR at location 1 due to interference with waste disposal. From the standpoint of MLR, location 2 is a better choice than location 1. These results assume that the real discount rate is zero (nominal discount rate inflation rate of economic costs from LIV = 0), and thus Figure 4 and Figure 5 show the present value of the MLR.

MLR for Two Injection Sites
These MLR curves will rotate toward higher estimates of MLR if the real discount rate is negative, and rotate toward lower estimates of MLR if the real discount rate is positive. If the real discount rate is about −1.2%, the MLR at location 1 is roughly constant between year 8 and year 30, at $2.97/tCO 2 -$3.04/tCO 2 . The MLR at location 2 is roughly constant at $0.016/tCO 2 -$0.018/tCO 2 if the real discount rate is ∼3%. The results for the portions of the MLR that are due to primary leakage out of the storage reservoir, secondary leakage out of overlying aquifers, or interference with other activities, are robust to differences in the real discount rate because these differences will equally affect all values at the same point in time. (See the Supporting Information for the results of different real discount rates.) Figure 4 and Figure 5 show that the distribution of MLR over the subsurface units varies by location.
MLR decreases shallower in the sequence at location 1 but increases at location 2. Even though the probability of nonnative fluids being present in an aquifer and the degree of alteration of the subsurface environment due to leakage both decrease higher in the sequence,(34) the increase in the number of secondary leakage pathways and costincurring activities in shallower aquifers can result in higher portions of MLR being attributable to the shallower aquifers. For the MLR of secondary leakage out of the TraverseDundee/SilurianDevonian at location 2 to increase over time, the number of wells that leak out of the TraverseDundee/SilurianDevonian must increase more than this decrease in pressure.

MLR That Accounts for Intervention
The MLR profiles in Figure 4 and Figure 5 incorporate the assumption that CO 2 injection continues regardless of the occurrence of leakage, and that leakage that occurs will continue unabated. Figure 6 shows the IAMLR for location 1, where only the closest well was modeled as a leakage pathway and the unabated MLR is similar to the case where all wells are modeled as leakage pathways. Three IAMLR curves are shown in Figure 6, based on the detection of CO 2 that has leaked into the Galesville aquifer. The possibility of intervention to remediate leakage substantially reduces MLR, even for technologies that can only detect large amounts of leaked CO 2 . In fact, within five years of injection and leakage, the IAMLR is less than $1.00/tCO 2 -below the leakage risk attributable to the shallower aquifers-and decreases to less than $0.10/tCO 2 within five more years. The substantial decrease in MLR occurs in part because location 1 is sited in close proximity to leakage pathways.
Costincurring impacts of leakage occur early, and the probability of detectable leakage increases over the first few years. For location 2, the MLR and the IAMLR are identical because CO 2 does not leak from the reservoir.

Breakdown by Stakeholder Group
The MLR may be important for stakeholders who could require fees based on the amount of CO 2 injected (e.g., Regulators). Other stakeholders with financial exposure to leakage will likely care more about their expected costs, in part because they may have to use financial resources that are not tied to the amount of CO 2 that has been injected.(28) Figure 7 shows the expected costs due to leakage (R t ) at location 1 for a range of stakeholders. The R t for Storage Operators, who may offset these costs with revenue from storing CO 2 , is $71 million in Year 1, which grows to $130 million in Year 5 (assuming that the real discount rate is zero). In contrast, the R t for Surface Owners, who may not have offsetting revenue, is $1.1 million in Year 1 and doubles to $2.2 million in Year 5 if the real discount rate is zero, but the IAR t for detecting 1000 tCO 2 is $180,000 in Year 5. Similarly, the R t for Groundwater Users increases from $59,000 in Year 1 to $100,000 in Year 5 with a zero real discount rate, but the IAR t is $8,000. In general, nonzero real discount rates will affect the present value of estimated costs from LIV, but conclusions from relative comparisons at the same point in time will not be affected, assuming that individual components of cost estimates change at same rate at a moment in time. (The Supporting Information contains results where the real discount rate was varied.)

Implications
LRiMM facilitates assessments of the extent to which estimated leakage incurs financial costs, and can (1) compare individual locations; (2) inform mechanisms for the liability and financial responsibility of injection operators; (3) provide information to governments who may assume liability in the postclosure period; and (4) be applied basinwide to rank the suitability of storage locations and derive and refine resource supply curves. LRiMM also provides regulators and policymakers with a methodology to develop more efficient and equitable rules for addressing leakage. All told, monetizing leakage risk and addressing the inequity between who is responsible for leakage and who bears the costs may motivate efficient deployment of CCUS.
The case studies presented here demonstrate that CO 2 injection sites should be located with consideration of MLR, which is a product of proximity to potential primary and secondary leakage pathways-through which leakage may occur and incur costs-and other subsurface activitieswhere interference may incur costs. Our application of LRiMM to potential injection locations within the Michigan sedimentary basin permits several generalizable lessons: 1. MLR will vary over time, but it is likely to be orders of magnitude below the costs of CO2 storage in well-sited locations. (48) 2. The distribution of MLR within sedimentary units varies by depth. Fewer wells penetrate the deeper aquifers where CO2 is likely to be injected. The deeper the injection, the lower is the probability and amount of leaked fluids migrating to shallower units both due to fewer leakage pathways and due to secondary trapping as the number of aquifer-aquitard sequences that must be traversed increases. (34) 3. Although MLR can be substantially reduced by relocating injection sites away from leakage pathways and other subsurface activities, even CO2-brine plumes injected at the most ideal locations within sedimentary basins will likely intersect with potential leakage pathways. The primary determinant of MLR will be the proximity to overlying valuable resources. 4. Our estimated costs of leakage suggest that while Storage Operators will likely have adequate financial resources, leakage may impose costs on other stakeholders who are less likely to have the needed financial resources. 5. The ability to intervene and remediate leakage substantially reduces MLR. Improved methods to reliably detect leaked CO2 would also reduce the financial exposure to leakage because intervention could occur sooner and avoid future impacts that may be costlier if they grow faster than the discount rate.
The U.S. Department of Energy has set a goal that at most 1% of injected CO 2 can leak from a storage reservoir(34) and the U.S. Environmental Protection Agency defines CO 2 leakage within the Underground Injection Control program Class VI rules as any CO 2 that is detected outside of the injection formation.(36) A truer measure of the reliability of CO 2 storage may not rest in an assessment of how much remains in the storage reservoir, but rather where and to what degree leaked CO 2 incurs impacts and costs, and how these externalities may be unevenly imposed on stakeholders. 

Supporting Information
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.5b05329.