The increasing awareness of possibly irreversible damages of global warming has pushed both public opinion and governments toward the support of increasingly stringent climate mitigation measures. However, the path toward climate change policies is correlated to both technological and economic challenges (IPCC, 2014). Among typical mitigation strategies, Carbon Capture and Storage (CCS) and Carbon Dioxide Removal (CDR) represent valuable alternatives to renewable energy sources. According to existing studies IPCC (2014), the CCS potential will have an important role in reducing the carbon intensity of electricity. CCS might thus represent a considerable share of emission reduction in the energy sector (Metz et al., 2005; Finkenrath, 2011; NAS, 2015; GCCSI, 2016).

CCS and CDR can also help reduce the costs of mitigation. Nonetheless, their development has not been as fast as expected in the last decades (Davidson et al., 2017). This low deployment can be associated with the absence of adequate incentives, lack of public acceptance, and to technological uncertainties associated with CCS (IEA, 2016; Lipponen et al., 2017). Among these barriers, carbon leakage from stored CO₂ could counteract the usefulness of carbon sequestration to help limiting global temperature increase (van der Zwaan and Smekens, 2009). In this paper, we analyze the impact of CO₂ leakage on the propsects of CCS in the power sector and the economic costs of mitigation¹.

We evaluate different policy provisions to help take leakage into account. Several problems can arise with leakage. Estimating its size is difficult and costly, since monitoring techniques have focused on small scale case studies so far (Romanak et al., 2012; Dethlefsen et al., 2013). A second source of uncertainty is related to the economic liability of leaked emissions (Wilson et al., 2003; Imbus et al., 2013). Finally, leakage might depend on the stringency of the climate policy. To address these questions, we use an integrated assessment model to examine three main dimensions: climate targets, leakage rates and policy provisions to counteract it. We consider the 2 and 1.5°C temperature increase targets by 2100, a range of possible leakage rates consistent with the literature, and different cases of pricing and liability of carbon leakage in the atmosphere. On this last point, we analyze whether leaked CO₂ is (a) not monitored nor taken into account in the carbon budget, (b) taken into account for the chosen carbon budget, but not priced at the carbon price, e.g., due to institutional or technological constraints, or (c) taken into account in the carbon budget and priced at the carbon price. These three cases allow us to disentangle the importance of monitoring and pricing leaked emissions.

We use an integrated assessment model (IAM) to simulate the impact of leakage on the set of mitigation strategies. IAMs are tools which are routinely used to evaluate global climate policies. Currently, many integrated assessment models use aggregated storage cost and availability curves, notably the ones from Hendriks et al. (2004). For this exercise, we disaggregated the storage according to different types of storage including their respective potential and costs. Moreover, transportation costs also vary according to the storage site considered. Finally, we added leakage from the different storage sites and assessed a set of scenarios capturing different climate policies, leakage rates, and options to consider leaked carbon emissions. We use the WITCH (World Induced Technical Change Hybrid) integrated assessment model in this study.

WITCH is a global integrated assessment model with two main distinguishing features: a regional game-theoretic setup, and an endogenous treatment of technological innovation for energy conservation and decarbonization (Emmerling et al., 2016). A top-down inter-temporal Ramsey-type optimal growth model is hard linked with a representation of the energy sector described in a bottom-up fashion, hence the hybrid denomination. The regional and intertemporal dimensions of the model make it possible to differentiate and assess the optimal response to several climate and energy policies across regions and over time. The non-cooperative nature of international relationships is explicitly accounted for via an iterative algorithm which yields the open-loop Nash equilibrium between the simultaneous activity of a set of representative regions. Regional strategic actions interrelate through greenhouse gas (GHG) emissions, dependence on exhaustible natural resources, trade of fossil fuels and carbon permits, and technological research and development (R&D) spillovers. R&D investments are directed toward either energy efficiency improvements or development of carbon-free breakthrough technologies. Such innovation accumulates over time and spills across countries in the form of knowledge stocks and flows. R&D investments, along with investments in energy technologies and the final goods sector, are endogenously determined in the intertemporal optimization. Within the energy sector, for new renewable energy sources (wind and solar), battery development, and advanced bio-fuels, learning is also taken into account through one or two factor learning curves, which determine future capital costs. The competition for land use between agriculture, forestry, and bioenergy, which are the main land-based production sectors, is described through a soft link with a land use and forestry model (GLOBIOM, Global Biosphere Management Model). A climate model (MAGICC) is used to compute climate variables from GHG emission levels, and an air pollution model (FASST) is linked to compute air pollutant concentrations.

Concerning CCS in the model, we consider four coal technology options (including the possibility of retrofitting existing plants), one gas and one biomass technology with carbon capture. The model includes seven types of storage (saline aquifers, EOR sites, depleted oil and gas fields, all either onshore or offshore, and onshore ECBM sites), each characterized by a maximum available capacity, a storage cost and an average distance from power plants. Apart from storage costs for the different types, all values are regionally differentiated. Finally, we account for an average specific transport cost dependent on the distance calibrated as ctr′ = 0.006667$/tCO₂km (Rubin et al., 2015). The total cost of captured CO₂ transport and storage C_t&s[$/year] is therefore evaluated according to the following equation, where the dimensions are time (t), regions (n), and type of storage (k_st):

Here, Q_st [GtCO₂/year] represents the yearly quantity of CO₂ captured by CCS plants, l_tr [km] represents the average distance, and c_st [$/tCO₂] the storage cost. We consider an annual leakage rate λ_lk of between 0.0%/year and up to 0.1%/year and include leaked emissions in the model. The cumulative amount of CO₂ stored CUM_st [GtCO₂] is therefore calculated based on annual stored values, considering that the model is run at a time step of 5 years, and including possible leakage every time period:

Here, λ_lk stands for the leakage rate, or the percentage of CO₂ stored in the previous year that is lost due to leakage and emitted in the atmosphere. This set of equations allow us to represent the transport and storage chain as a single cost function, differentiated across regions. The cost function for each storage type follows a step increase in function of the cumulative stored quantity, where each step means a switch from a cheap but replete storage type to the immediate next, more expensive site. Finally, annual leaked emissions Q_leak are computed as follows:

It should be noted that Q_leak in period t is accounted for based on CUM_st in t, which is not including the emission captured in the same period, but only until t − 1. This is to represent delay in leakage.

Based on this model implementation of storage, transportation, and leakage of CO₂, we explore a set of 31 scenarios to capture the following dimensions: leakage rates, climate targets, and policy provisions. We implement different leakage rates (LR) starting from zero leakage, 0.01%, 0.05%, and up to 0.1% per year.

Secondly, we consider different stringency of climate policies, represented by carbon budgets covering total CO₂ emissions from fossil fuels and industrial processes from 2010 to 2100. In addition to the business as usual (BAU) case without a future climate policy, we consider cases of 550, 1,000, and 1,600 GtCO₂ corresponding to roughly 1.5, 2.0, and 2.5°C of global warming in 2100, according to the definition in the IPCC AR5 report (Edenhofer et al., 2014; Vuuren et al., 2017)⁵. When running these scenarios, the model sets a constraint on emitted CO₂ equal to the budget and solves finding the cost optimal solution for attaining the target.

Finally, we differentiate the economic and policy treatment of carbon leakage emissions. In particular, we consider whether or not (a) leaked emissions are priced (through a carbon tax or the price of emission permits) or not, and (b) the leaked emissions are included in the carbon budget of the policy maker. Four cases are possible based on these distinctions: In the first case (NN), leakage is not taken into account in the policy target nor priced, e.g., due to technical, institutional, or political barriers. This case allows us to assess the climatic impact of leakage if it is not taken into account for climate targets, nor in emission pricing schemes. The other limiting case, where pricing and monitoring are effective (YY), constitutes the first best case where the actual climate target is attained, and leakage is treated the same way as other carbon emissions and priced at the marginal cost of abatement. The two remaining cases represent different institutional, economic, and technological situations: In the YN case, leakage is anticipated for the climate policy goal, while due to monitoring or institutional constraints, the source cannot be taxed or held accountable. Hence, in this case, other mitigation options are required to counteract leakage emissions. Hypothetically, in the fourth situation (NY), on the other hand, pricing leakage emissions is possible and implemented, but the climate policy does not take leakage into account a situation which is not realistic and hence we do not consider it here.

As summarized in Table 1, we consider three different policy prescriptions (NN, YN, YY) for three different carbon budgets (1600, 1000, 550 GtCO₂) and three leakage rates (0.01, 0.05, 0.1 %/year, total: 27 scenarios), in addition to the four scenarios without leakage (BAU, 1,600, 1,000, 550).

The different sets of scenarios show different patterns in terms of emissions, CCS deployment, and economic costs. Firstly, looking at overall greenhouse gas emissions, Figure 1 shows the no leakage and high leakage scenarios and compares them to the scenarios of the AR5 database that are consistent with the two degree target (Edenhofer et al., 2014). Overall, emissions are only mildly affected by carbon leakage at the global scale, and the mitigation pathways are dominated by the stringency of the carbon budgets. In particular in the 550 scenarios, total emissions become negative toward the end of the century. The difference between cases where leakage (set as 0.1%/year of stored CO₂) is well accounted for (YY) and when it is not (NN) is small compared to total emissions, though still visible.

Looking closer at the leaked emissions, one can see that leakage can contribute to emissions as shown in Figure 2. The amount of leakage over this century in the NN and YN scenarios ranges from 2.5 GtCO₂ (for a leakage rate of 0.01%/year) to around 25 GtCO₂ for a leakage rate of 0.1%/year. Moreover, while it is quite similar for the different climate targets, it shows the highest values always for the 1,600 GtCO₂ scenario, where fossil-based CCS is widely deployed. Comparing the leakage to the amount of captured emissions for the NN and YN scenarios (around 700–750 GtCO₂), we get around 0.5% of leakage until 2100 for the low leakage rate, and 3% for the high leakage rate cases. These values are virtually unchanged in the YN and NN scenarios, where leakage is not priced and hence does not affect CCS deployment. However, in the YY scenario, when the effect and cost of leakage are fully accounted for, the model responds with a reduction in leakage, linked to a reduction in CCS technology adoption and captured emissions. For the highest leakage rate of 0.1%, captured emissions are lowered by about 5–10 GtCO₂ across the different climate targets. The percentage of leaked emission on the total captured between 2015 and 2100 is however similar to the previous scenarios. If compared to Deng et al. (2017), our results show higher percentage of emission leaking over the century given similar leakage rates. This is due to their assumption that most of leaked emissions do not reach the surface, but are priced and thus have a negative impact on CCS deployment.

Leaked emissions have an impact on the climate, in terms of CO₂ concentrations and global temperature increase, shown in Figure 3. In absolute terms, variations in global mean temperature increase in 2100 are relatively small, of the order of magnitude of 0.01°C when leaked emissions are not monitored. For the YN and YY scenarios where leakage is accounted in the carbon budget target, the results show a small temperature decrease with increasing leakage rate. This can be explained due to different timing of emissions, notably due to the early shift to zero carbon technologies replacing CCS. While overall the temperature effect is relatively small, it still implies further exacerbation of global warming when leakage is not accounted in the budget, which might be relevant for the most stringent scenarios.

Figure 4 shows that the reduction in CCS (in terms of capacity reduction by 2100) is linked to the leakage rate, and to whether it is priced and accounted for in the carbon budget: only in the case where the costs of leaked emissions are accounted for in the economy through carbon pricing (YY) is CCS substantially reduced. Therefore, in the scenarios where countries do not pay for their leaked emission, CCS is not affected, both in the case where seepage is considered in the carbon budget or not (YN and NN scenarios). This result can be explaind since leakage can not be directly linked to the storage owners and the capturing power plant. The CCS reduction is particularly high for fossil fuel based CCS where the reduction reaches between 10 and up to 35 per cent of the capacity without leakage. Bio-energy with carbon capture and storage (BECCS), on the other hand, shows reductions lower than 10%. That is, for biomass CCS, leakage seems to provide a less important barrier, even with a leakage rate of 0.1% per year, especially in the most stringent 1.5°C scenario. Moreover, it is interesting to note that the ordering across stringency of the climate targets is reversed for fossil fuel and biomass based CCS: The more stringent the scenario considered, the lower the impact on BECCS and the higher the reduction of fossil based CCS. This result is in line with the intermediary role of fossil fuel based CCS found e.g., in Rogelj et al. (2015), van der Zwaan and Smekens (2009) and Deng et al. (2017), even if these two latter studies did not consider stringent scenario such as the 1.5°C (550 GtCO₂).

In terms of economic costs of carbon leakage, we first look at the implied carbon price required to meet the different climate targets. Figure 5 shows the increase in carbon price with respect to the no leakage cases, noticeable for scenarios where leakage is included in the carbon budget. The standard carbon prices in the three scenarios in the year 2020 to implement the carbon budgets of 1,600, 1,000, and 550 GtCO₂ are 79, 164, and 318 $/tCO_2eq respectively, and grow at a rate of 5% per year⁶. Comparing to these default scenarios the leakage cases, first note that the NN scenario does not show any change in the carbon price as leakage increases, since it is not considered for the policy design. When seepage affects the carbon budget available for each climate policy, it becomes necessary to mitigate this effect using other technological strategies more expensive than CCS. This results in an increase of carbon price in the YN and YY scenarios, which ranges from 2.5 to 7.5%. When leakage is not priced (YN), it still leads to a higher carbon price due to the reduced global carbon budget, albeit to a lesser extent, resulting from higher mitigation effort based on the most cost-effective mitigation options available. When it is also priced, the cost-effective potential of CCS is significantly reduced, resulting in higher use of more expensive mitigation options such as energy efficiency improvements or renewable energy sources. Across carbon budgets, it should be noted that, although the relative variation in the carbon tax is similar, in absolute value it differs significantly.

The previous results showed how uncontrolled seepage would affect global climate and how, even well monitored leakage might be binding for CCS development and would lead, globally, to a more expensive energy system. Now we focus on the regionally differentiated modeling results, focusing here on the most stringent scenario (550) and using the 0.1% leakage rate, while for smaller rates the results scale down almost linearly as shown before.

As shown in Figure 6, the cumulative leaked emissions are not evenly distributed across regions: China, the United States, Canada/Japan (cajaz), and Latin America (laca) countries are expected to extensively use CCS power plants, and therefore are projected to leak more than 2 GtCO₂, with China reaching 5 GtCO₂ by 2100 (see Table 2 for the description of the regions).

Given the relatively high carbon price required to achieve the stringent climate policy targets, leakage can imply substantial additional costs of emissions, according to the carbon price in place. In the aforementioned regions with high projected leakage potential, this amounts to values between 75 and 200 billion USD over the century, with exception of 550 billion for China, see the left panel in Figure 7 (all values reported there refer to NPV cumulative values from 2015 to 2100, discounted at a 5% discount rate). In the scenario 550, NN with high leakage, the (discounted) yearly value of leaked emissions in 2100 reaches up to 12 billion USD in China and about 7 billion USD in Latin America. Globally, the yearly leakage in 2100 of 0.78 GtCO₂ amounts to a discounted value of 54 billion USD. Given the relatively small carbon budget consistent with the 1.5 degree scenario, the additional 25 GtCO₂ of leaked carbon emissions associated with the high leakage rate (0.1%/year) scenario represent a significant cost, with particular economic implications for some regions. Moreover, we can compare which regions have to bear the additional costs if, while initially not being priced, now leaked emissions are accordingly priced and the climate target is the same, i.e., moving from scenario YN to YY. The right part of Figure 7 shows the additional cost of leakage emissions when they are priced at the global carbon price (YY) compared to the case where they are not (YN): almost all regions show a negative difference, meaning that they reduce expenses when leakage is well regulated. In Canada/Japan (cajaz), including the leakage costs in the economy does not lead to a large CCS reduction. Therefore, the higher carbon price in the YY scenario results in higher costs for the country. This occurs mainly because the use of BECCS late in the century seems essential for these countries. In countries where CCS is only marginally profitable, its optimal deployment is reduced facing leakage, and hence the required carbon price is slightly higher, while leakage is significantly reduced. Since the latter effect dominates, those countries gain in terms of the value of carbon.

These are the value of losses that would not be payed by companies or countries in absence of regulation. But they can also be seen as a waste of money for a country which is investing in climate mitigation policies, and the cost of re-abating leaked emission. Moreover, other costs, like local or climate change damages are not accounted in this estimation, therefore the real economic loss might be even higher. Note that in both cases the carbon price increases with leakage rate, as more expensive low carbon technologies are installed to compensate seepage. However the YY scenario represents a perfect regulation system where CCS owners pay for the leaked emissions, so that use of CCS is reduced and the carbon price increases further. This behavior is considered more convenient than continuing using CCS and paying for leakage. The difference in total leakage costs can be considered as the regional gain or loss if leakage is well regulated or not. Moreover, we also compare the GDP of all scenarios to analyze the changes in policy costs with respect to the BAU scenario. Table 3 reports the policy cost for the aforementioned scenarios. We confirm the above mentioned trend of policy cost⁷ increasing with leakage rate and from zero leakage case to YY setting. Overall, policy costs increase in the range of 0.1–0.2% in the YN case due to the higher mitigation effort needed. If moreover leakage is priced, they increase by a further 0.1-0.2 percentage points. For instance, in the stringent 550, YY scenario and for a leakage rate of 0.1%, the costs of staying below 1.5° increase from 8.5% to almost 8.9% due to leakage.

The purpose of this work is to expand the assessment of leakage impact on CCS deployment and climate policies. We consider different institutional and economic settings reproducing issues in monitoring and paying for possible leakage. Furthermore, we perform analysis over leakage rates, and over three different climate scenarios, including the 1.5 and 2°C temperature increase target by 2100, particularly relevant after the Paris agreement in 2015. The results show that carbon leakage can lead to up to 25 GtCO₂ of additional emissions throughout the twenty-first century for a leakage rate of 0.1% per year, which represents about 3% of total captured emissions. Considering a more optimistic leakage rate (0.01%), only 0.5% of injected emissions would leak by 2100. If accounted for in the carbon budget and priced, CCS deployment is expected to be lowered by up to 35% (fossil) and 10% (BECCS) for high leakage rates. This means that CCS remains an important technology for mitigation in the power sector, notably coal and gas based in less stringent scenarios, and biomass fueled for the 1.5°C scenario. Due to more early-on abatement, considering leakage leads to slightly lower global warming in the long run. If not taken into consideration nor priced, on the other hand, it leads to an around 0.01–0.02 degrees higher global mean temperature. Overall, policy costs increase by about 0.2–0.4 percentage points (of GDP loss) due to considered leakage. In terms of regional distribution, China, Latin America, the U.S., and Canada have the highest leakage amount to be expected by 2100. The associated economic value of this quantity ranges across regions from 70 to more than 200 billion USD. Finally, we demonstrated how appropriate monitoring and accounting of leakage imply a reduction in use of CCS and also economic saving for most countries.

AV contributed to developing the idea, methodology, model design and execution, writing and revising the manuscript; JE contributed to developing the idea and methodology, supervised model design and execution, writing and revising the manuscript; MT have supervised the work, providing suggestions on the research scope and revisions.