Costs to achieve target net emissions reductions in the US electric sector using direct air capture

Sarang D Supekar; Tae-Hwan Lim; Steven J Skerlos

doi:10.1088/1748-9326/ab30aa

1. Introduction

Carbon dioxide removal from the atmosphere using 'negative emission technologies' (NETs) using a combination of physical, chemical, or biological processes has been deemed essential to contain the increase in the average global temperature over pre-industrial times to 1.5 °C–2 °C (temperature anomaly) by the end of the century (Gasser et al 2015, Rogelj et al 2018). The criticality of NETs in achieving these climate targets given humanity's rapidly diminishing global carbon budget (Tong et al 2019) has led to calls for more in-depth evaluations of individual NETs (Fuss et al 2016, Field and Mach 2017, van Vuuren et al 2017), particularly with regards to their scalability and systems-level impacts on the environment, economy, and society. The vast majority of published studies have focused on gigatonne-scale carbon dioxide removal approaches tied to biogeochemical and biogeophysical cycles such as bioenergy with carbon capture and sequestration and terrestrial carbon management (Minx et al 2018). There is also a growing research and commercial interest in the removal of atmospheric CO₂ through a process known as direct air capture (DAC), though the scalability and systems-level impacts of DAC as a gigatonne-scale NET remain poorly understood according to a recent report prepared by the US National Academies outlining research agendas for NETs (National Academies of Sciences, Engineering and Medicine 2018).

In this paper we ask if, when, and to what extent could DAC, acting as a purely backstop measure, help achieve a 70% reduction in CO₂ emissions by 2050 relative to 2010. The 70% target corresponds to the greenhouse gas (GHG) emissions reduction stipulated in the IPCC's Fifth Assessment Report (IPCC 2014) for the resulting CO₂ concentration pathways to have a >66% likelihood to keep the temperature anomaly to 2 °C by 2100. This paper specifically focuses on the US electricity generation sector, the emissions from which are responsible for about 30% (US Environmental Protection Agency 2016) of US GHG emissions and about 4% (World Resources Institute 2017) of the world's GHG emissions. The focus remains on CO₂ from fossil fuel combustion since it accounts for more than 98% (US Environmental Protection Agency 2016) of the electric sector GHG emissions (CH₄, N₂O, and SF₆ comprise the rest).

DAC here refers to the capture of atmospheric CO₂ using materials with preferential affinity for CO₂ over the other air gases, followed by storage/sequestration of the captured CO₂ (see section 1 in the supplementary material is available online at stacks.iop.org/ERL/14/084013/mmedia for the DAC process considered in this paper). Since the seminal work on this technology in the 1990s (Lackner et al 1999) knowledge about the chemistry, engineering, and costs of DAC has advanced considerably (Fuss et al 2018, Minx et al 2018, Nemet et al 2018). Sanz-Pérez et al (2016) provide a comprehensive review of the current state of the art in DAC technologies. Irrespective of the process chemistry employed for CO₂ capture from ambient air, DAC is an energy-consuming NET with heat and electricity requirements ranging between 7.6–15 GJ_th/tCO₂ (Goeppert et al 2012, Boot-Handford et al 2014) and 0.7–2.4 GJ_e/tCO₂, respectively (House et al 2009). CO₂ emissions from energy sources powering DAC plants will require major changes to the electric grid, which in turn will impact the optimal deployment of DAC plants, creating a recursive CO₂ feedback between DAC plants and their energy sources. This feedback between DAC plants and their energy sources greatly influences the systems-level private cost to society of achieving target emissions reductions. Section 6 in the supplementary material provides a detailed discussion and mathematical explanation of the recursive feedback.

Modeling these time-dependent systems-level interactions is foundational for evaluating the effectiveness of DAC as a viable large-scale NET. Table 1 summarizes major studies published on DAC, and shows that the vast majority of them explicitly or implicitly assume that energy sources powering DAC plants would be low-carbon or carbon-neutral. To the best of our knowledge, the major studies that address the dynamic nature of the DAC-energy supply interaction are by: Chen and Tavoni (2013) who examine the global CO₂ removal potential using DAC through the year 2100 via the WITCH integrated assessment model (IAM) Bosetti et al (2006), Kriegler et al (2013) who compare the potential of BECCS against DAC using the ReMIND IAM Leimbach et al (2010), Creutzig et al (2019) who examine the collective deployment of DAC and BECCS globally through 2100; and Wohland et al (2018) and Breyer et al (2019) who respectively examine the potential for DAC powered by excess renewable generation in Europe and Maghreb region.

Table 1. Classification of the literature based on the system boundary scope of the analysis and carbon intensity assumptions of energy supply powering DAC plants. Note that some studies feature in multiple classifications.

Treatment of carbon intensity of energy sources powering DAC	Study
Explicit assumption of low-carbon/carbon neutral energy supply	Baciocchi et al (2006)^a, Lackner (2009)^a, Holmes and Keith (2012)^a, Goldberg et al (2013)^a, Goldberg and Lackner (2015)^a, Buck (2016)^b^,^c^,^d, Fuss et al (2016)^b^,^c^,^d, Geng et al (2016)^a, Smith et al (2016)^b^,^c^,^d, Sinha et al (2017)^a, Zhang et al (2017)^a, Keith et al (2018)^a
Implicit assumption of low-carbon/carbon neutral energy supply based on timing of DAC deployment after 2050	Keith et al (2006)^b^,^c^,^d, Stolaroff et al (2008)^a, House et al (2011)^a, Socolow et al (2011)^a, Zeman (2014)^a, Marcucci et al (2017)^b^,^c^,^d, Parra et al (2017)^a, Rockström et al (2017)^b^,^c^,^d
Cost and/or energy analysis leading to low-carbon/carbon neutral energy supply assumption	Keith et al (2006)^a, Zeman (2007)^a, House et al (2011)^a, Simon et al (2011)^a, Kulkarni and Sholl (2012)^a, Mazzotti et al (2013)^a, Zeman (2014)^a, Pritchard et al (2015)^a, van der Giesen et al (2017)^a, National Academies of Sciences, Engineering, and Medicine (2018)^a
Carbon intensity of electricity supply endogenously calculated in the model	Chen and Tavoni (2013)^b^,^c^,^d, Kriegler et al (2013)^b^,^c^,^d, Wohland et al (2018)^b^,^e^,^f, Breyer et al (2019)^b^,^e^,^f, Creutzig et al (2019)^b^,^c^,^d, This study^b^,^e^,^f

^aPlant-level analysis. ^bSystems-level analysis. ^cGlobal scope. ^dUses IAM. ^eRegional/sectoral scope. ^fUses bottom-up model or some other simulation model.

In this paper, we supplement this previous work by considering only the US electric sector with an annual time resolution, and narrowly studying the potential role for DAC in CO₂ emissions reduction pathways leading up to 2050 using a non-IAM-based model.

Our overarching research question is: should the US electric sector rely on DAC to help achieve a 70% reduction target? To address this question, this paper quantifies systems-level costs, deployment scale, and associated changes needed in the electric supply to accommodate DAC as a strategy for meeting CO₂ emissions targets within reasonable economic and practical bounds. Hence we intend to contribute to ongoing scientific dialogue on the prudence of deploying new low-carbon energy sources to power NETs as opposed to directing those resources to replace existing fossil fuel-based energy sources—a comparison that is deemed crucial to cost-effective climate change mitigation by the US National Academies report on NETs (National Academies of Sciences, Engineering, and Medicine 2018).

Since the analysis is geared towards technological transformations pursuant to a sector-specific emissions goal, we follow guidance from Ackerman et al (2009) and use an engineering-based 'least-cost' model (also known as a bottom-up model). Our use of this computationally simplified yet technologically rich representation allows us to supplement the IAM-based approach adopted by Chen and Tavoni (2013) and Kriegler et al (2013), as well as the stylized simulation model-based approach adopted by Wohland et al (2018). It also allows us to analyze a large range of uncertainty scenarios that capture uncertainties in technology costs, fuel prices, emission factors, and electricity demand. We begin by briefly describing a previously published stock-and-flow model of the US electric sector from which this work builds, and then focus on the addition of DAC to the model as required to answer our research question. Then we discuss key input data and assumptions employed in the analysis. Finally, we discuss the model results, limitations, and implications of our findings.

2. Methodology

The US electric utilities sector (including combined heat and power units) is represented using the LETSACT model, which has been published previously (Supekar and Skerlos 2017). LETSACT is a linear programming optimization model that contains a stock-and-flow representation of US electricity generating units (EGUs). It contains 13 generation technologies: pulverized coal and natural gas combined cycle with and without carbon capture and storage (CCS), gas turbine, petroleum, biomass, nuclear, conventional hydroelectric, on-shore wind, solar photovoltaic, solar thermal, and geothermal. The last six technologies on this list, along with coal and natural gas CCS, are referred to here as 'low-carbon' EGU technologies.

The initial stock of EGUs as a function of EGU technology and age is populated using the US EPA's NEEDS database (US Environmental Protection Agency 2015). Stocks are updated at 1 year intervals by changing EGU additions and retirements for each of the 13 technologies. Annual EGU additions, retirements, and stocks, collectively referred to as a 'technology trajectory', are decision variables in an optimization problem. The objective of the optimization problem is to minimize the net present value (NPV) of the total capital, operating, and retirement costs of EGUs over the analysis time horizon. The constraints include achieving electricity demand equal to supply, and a CO₂ emissions budget corresponding to a 70% reduction in emissions by 2050 relative to 2010. The CO₂ emissions budget is calculated as the area under the curve defined by a straight-line emission trajectory corresponding to the emissions reduction target (see figure S5). We refer to this emissions target (approximately 50 Gt CO₂) as the '70% reduction target' or the '2050 emissions budget'.

Equations (1)–(4) describe the optimization problem formulation, where X^new, X^ret, and X^stockare the decision variables representing EGU additions, early retirements, and total stocks in MWh. The sets N, T, and Y contain EGU technologies, ages, years in the analysis time horizon, respectively. The model characterizes EGU costs and emissions on a per MWh generation basis. The coefficients c^new, c^stock, c^ret represent the unit costs of building and operating a MWh of new generation, retiring a MWh of existing generation, and operating a MWh of existing generation respectively. Similarly, e^new and e^stock represent the emissions per MWh of new and existing generation. The discount rate is given by r, which is assumed as 7% in this analysis. E is the emissions budget corresponding to the area under the straight-line emissions reduction trajectory, and D is the total non-DAC related electricity demand that is treated as exogenous

$\begin{eqnarray}&&\begin{array}{l}\text{Minimize}\,f\left({X}^{new},{X}^{stock},{X}^{ret}\right)\\ \,=\displaystyle \sum _{k\in Y}\displaystyle \sum _{i\in N}\displaystyle \frac{{c}_{ik}^{new}{X}_{ik}^{new}}{{\left(1+r\right)}^{k-1}}\\ \,+\displaystyle \sum _{k\in Y}\displaystyle \sum _{i\in N}\displaystyle \sum _{j\in T}\displaystyle \frac{{c}_{ijk}^{ret}{X}_{ijk}^{ret}+{c}_{ijk}^{stock}{X}_{ijk}^{stock}}{{\left(1+r\right)}^{k-1}}.\end{array}\end{eqnarray} \tag{ 1 }$

Subject to

$\begin{eqnarray}&&{X}_{ik}^{new},{X}_{ijk}^{stock},{X}_{ijk}^{ret}\geqslant 0\,\forall i\in N,j\in T,k\in Y,\end{eqnarray} \tag{ 2 }$

$\begin{eqnarray}&&\displaystyle \sum _{i\in N}{X}_{ik}^{new}+\displaystyle \sum _{i\in N}\displaystyle \sum _{j\in T}{X}_{ijk}^{stock}\geqslant {D}_{k}\,\forall k\in Y,\end{eqnarray} \tag{ 3 }$

$\begin{eqnarray}&&\displaystyle \sum _{k\in Y}\displaystyle \sum _{i\in N}{e}_{ik}^{new}{X}_{ik}^{new}+\displaystyle \sum _{k\in Y}\displaystyle \sum _{i\in N}\displaystyle \sum _{j\in T}{e}_{ijk}^{stock}{X}_{ijk}^{stock}\leqslant E.\end{eqnarray} \tag{ 4 }$

Retirement in this analysis refers to the decommissioning of a unit of generation prematurely before its typical expected plant life. This 'early retirement' could be an outcome of functional or economic obsolescence as a result of competing technologies or regulations. Early retirement is treated as separate from and in addition to the decommissioning of generation capacity at the end of its expected life. For instance, coal plants are built to typically serve for 60 years, and thus if the model chooses to decommission 1 MWh of coal generation after it reaches only 30 years of life, this would be treated as early retirement. The unit retirement cost (c^ret) is treated as the remaining capital liability, if any, on a unit of generation beyond its assumed financing period of 20 years. The retirement cost shown in equation (1) is therefore a function of both age and year of the EGU. Key details of the model in the context of this paper are provided in section 2 of the supplementary material. Additional information about the LETSACT model and its mathematical framework can be found in Supekar and Skerlos (2017).

2.1. DAC representation in LETSACT

DAC plants are net consumers of heat and electricity and remove CO₂ from the atmosphere. This means that they add non-negative values to the right hand side of equations (3) and (4), which in turn affects decision variables on the left hand sides of those equations. To maintain the linearity of the model which is vital for minimizing computation time, we incorporate DAC into LETSACT by treating it as an EGU that consumes heat to produce negative useful power output and emit negative CO₂. To achieve this, we take energy and cost parameters reported in the literature (Socolow et al 2011, Keith et al 2018, Grant et al 2018) for a single aqueous alkaline sorbent-based DAC plant, and convert the parameters to a per MWh of electricity consumed basis.

For instance, the KOH + calcium caustic recovery loop-based DAC plant described by Keith et al (2018) has a 1 Mt per year of CO₂ removal capacity, and requires 366 kWh of electricity and 5.25 GJ of heat per tonne of CO₂. This DAC plant would be equivalent to an EGU with –46.5 MW capacity (assuming a 90% capacity factor based on Keith et al 2018) that operates with a heat rate of about –13596 Btu/kWh as shown in equations (5) and (6). The 694 million USD capital cost and 26 and 18 USD/tCO₂ of O&M and transportation costs of the DAC plant are then converted to their equivalent EGU basis as shown in equations (7) and (8)

$\begin{eqnarray}&&\begin{array}{l}{\rm{D}}{\rm{A}}{{\rm{C}}}_{Capacity}^{{\rm{E}}{\rm{G}}{\rm{U}}}=\displaystyle \frac{-0.366\left(\tfrac{{\rm{M}}{\rm{W}}{\rm{h}}}{{\rm{t}}{\rm{C}}{{\rm{O}}}_{{\rm{2}}}}\right)\times {10}^{6}\left(\tfrac{{\rm{t}}{\rm{C}}{{\rm{O}}}_{{\rm{2}}}}{{\rm{y}}{\rm{e}}{\rm{a}}{\rm{r}}}\right)}{365\left(\frac{{\rm{d}}{\rm{a}}{\rm{y}}{\rm{s}}}{{\rm{y}}{\rm{e}}{\rm{a}}{\rm{r}}}\right)\times 24\left(\frac{{\rm{h}}{\rm{o}}{\rm{u}}{\rm{r}}{\rm{s}}}{{\rm{d}}{\rm{a}}{\rm{y}}}\right)\times 0.9}\\ \,\,\,\,\,\,=\,-46.5\left({\rm{M}}{\rm{W}}\right),\end{array}\end{eqnarray} \tag{ 5 }$

$\begin{eqnarray}&&\begin{array}{l}{\rm{D}}{\rm{A}}{{\rm{C}}}_{HeatRate}^{{\rm{E}}{\rm{G}}{\rm{U}}}=\displaystyle \frac{5.25\left(\tfrac{{\rm{G}}{\rm{J}}}{{\rm{t}}{\rm{C}}{{\rm{O}}}_{{\rm{2}}}}\right)\times 947\,817\left(\frac{{\rm{B}}{\rm{t}}{\rm{u}}}{{\rm{G}}{\rm{J}}}\right)}{-366\left(\tfrac{{\rm{k}}{\rm{W}}{\rm{h}}}{{\rm{t}}{\rm{C}}{{\rm{O}}}_{{\rm{2}}}}\right)}\\ \,\,\,\,\,\,=\,-13\,596\left(\frac{{\rm{B}}{\rm{t}}{\rm{u}}}{{\rm{k}}{\rm{W}}{\rm{h}}}\right),\end{array}\end{eqnarray} \tag{ 6 }$

$\begin{eqnarray}&&{\rm{D}}{\rm{A}}{{\rm{C}}}_{CapCost}^{{\rm{E}}{\rm{G}}{\rm{U}}}=\left|\displaystyle \frac{694\left({\rm{m}}{\rm{i}}{\rm{l}}{\rm{l}}{\rm{i}}{\rm{o}}{\rm{n}}\$\right)}{-46.5\left({\rm{M}}{\rm{W}}\right)}\right|=14\,925\left(\tfrac{\$}{{\rm{k}}{\rm{W}}}\right),\end{eqnarray} \tag{ 7 }$

$\begin{eqnarray}\begin{array}{l}{\rm{D}}{\rm{A}}{{\rm{C}}}_{O\& M+Transport}^{{\rm{E}}{\rm{G}}{\rm{U}}}=\left|\displaystyle \frac{44\left(\tfrac{\$}{{\rm{t}}{{\rm{CO}}}_{{\rm{2}}}}\right)}{-0.366\left(\tfrac{{\rm{M}}{\rm{W}}{\rm{h}}}{{\rm{t}}{{\rm{CO}}}_{{\rm{2}}}}\right)}\right|\\ \,\,\,\,\,\,\,\,=\,120.2\left(\tfrac{\$}{{\rm{M}}{\rm{W}}{\rm{h}}}\right).\end{array}\end{eqnarray} \tag{ 8 }$

For every Mt of CO₂ by the DAC plant, 0.413 Mt CO₂ is captured from natural gas combustion in the DAC plant's calciner as per the process described in the literature. We note that DAC costs and energy use are assumed to increase linearly with DAC deployement—that is, the costs and energy use of ten DAC plants with 1 Mt CO₂ capacity is treated as equal to one DAC plant with 10 Mt CO₂ capacity. Thus, potential cost savings through economies of scale and energy savings through better thermal integration are not considered in this study.

In addition to ensuring structural consistency with the LETSACT framework, representation of DAC as an EGU captures the feedbacks between the DAC plants, the electric grid, and net CO₂ removal, and also quantifies the system-level costs associated with building and operating DAC plants together with the costs of EGU additions and retirements necessary for effective CO₂ removal. The capacity and timing of DAC deployment is thus determined in concert with changes in the electricity supply as required to stay within the CO₂ emissions budget at least cost. As such, the analysis views DAC with CO₂ storage as a measure that may act in conjunction with preventive mitigation-based supply-side transformations to low-carbon/carbon-free EGUs.

2.2. Business-as-usual (BAU) and climate action

We reference 'BAU' when the LETSACT model is run without the emissions constraint given by equation (4) to minimize the NPV of costs of meeting electricity demand. We reference the case with the emissions constraint as 'climate action.' By definition, climate action is a more aggressive CO₂ emissions timeline than BAU. Both BAU and climate action cases are subject to a constraint to emulate nation-wide implementation of the Renewable Portfolio Standard, which is imposed as an inequality constraint requiring at least 15% of the total generation to come from renewables by 2025. The difference in NPV of the cost difference between BAU and climate action, under identical input parameters, is defined as the 'climate action cost'.

The term 'delayed climate action' is referenced when BAU continues beyond the model start year (2015). When climate action is delayed until a certain year, the model forces the BAU technology and emissions trajectory from 2015 to the year before the year in which climate action is initiated. The year in which climate action is initiated is called the 'climate action year,' and a set of climate action years is collectively to referred to as a 'climate action timeframe.' Table S5 provides a glossary of the terms introduced here and used throughout the rest of the paper, and figure S6 provides visual guidance for contextualizing the terminologies. The model treats 2050 emissions targets as infeasible without DAC and using preventive mitigation alone either when the primal-dual interior-point method used to solve the optimization problem returns an infeasible solution.

2.3. Uncertainty scenarios

Given large uncertainties in costs, emission characteristics, and demand over the analysis timeframe, we perform a sensitivity analysis by varying all key model inputs corresponding to technology and fuel costs, emissions, and demand at three levels—low, nominal, and high. The 3 parameter categories and 3 parameter levels give rise to 3³ = 27 scenarios for which least-cost EGU additions, retirements, stocks, emissions, and costs are calculated at each climate action year. As with EGU technologies, the sensitivity analysis includes uncertainties in DAC costs, emissions, energy demands at low, nominal, and high levels obtained from the literature. For each uncertainty scenario, we run the optimization model for all climate action years from 2015 to 2035. Single values expressed in the paper for quantities emanating from the 27 uncertainty scenarios reflect medians, and ranges reflect the first and ninth deciles, unless specified otherwise.

Section 2 in the supplemetary material provides values and data sources for the uncertainty parameters. The supplementary material also contain links to result files containing costs, emissions, DAC deployment, EGU additions and retirements, and other important variables calculated for different uncertainty scenarios and climate action years as generated by the least-cost model. Links to codes used for the analysis and to generate these results are also included.

3. Results and discussion

The model estimates that continued BAU emissions through 2050 would total 57.9–62.2 Gt CO₂ as shown by the gray lines in figure 1(a). This is significantly higher than the 50 Gt CO₂ budget. Despite falling short of the 2050 CO₂ reduction target, the model projects BAU emissions to follow a downward trend as shown in figure 1(b), with an estimated emissions reduction of 30%–44% relative to 2010. This is because even without a 2050 emissions constraint, least-cost technology trajectories under BAU project that coal EGUs would be largely replaced with natural gas combined cycle EGUs, with the median fraction of total generation from coal falling to about 4% by 2050.

3.1. Preventive climate action

The analysis finds that initiating climate action within the next decade could still achieve 2050 CO₂ targets using preventive mitigation alone, provided BAU emissions during any period of climate inaction follow the downward trend projected by the model. Without DAC (or other NETs), preventive climate action would thus be likely impossible if BAU emissions were allowed to continue beyond 2030. However, any delays in initiating timely preventive climate action starting in 2020 will result in progressively higher costs, as shown in figure 2(a) in which the distribution of the total climate action costs through 2050 across various uncertainty scenarios is plotted as a function of the climate action year. For instance, delays in initiating climate action increase the median climate action cost through 2050 from about 135 billion USD for climate action year 2020 to 175 billion USD and 320 billion USD for climate action years 2025 and 2030, respectively.

**Figure 2.** (a) Climate action cost; (b) total generation retired early through 2050; and (c) total new generation added through 2050 as a function of climate action year. Approximated Gaussian density distribution for each quantity in the left panels in (a)–(c) is shown in their respective right panels.
Download figure:
Standard image High-resolution image

The higher cost of delays relative to climate action starting in 2020 result from higher retirement of fossil fuel-based EGUs to compensate for excess BAU emissions. Figure 2(b) shows EGU retirements through 2050 increasing with delays in climate action. Total (median) EGU retirements through 2050 increase from 1570 TWh for climate action year 2020 to 1830 TWh for climate action year 2025 (16% increase), and 2350 TWh for climate action year 2030 (50% increase). New generation added to compensate for the higher retirements and meet the overall electricity demand accounts for the largest fraction of the increase in climate action costs. Median total EGU additions through 2050 increase from 4190 TWh for climate action year 2020 to 4275 TWh for climate action year 2025 (2% increase), and 4780 TWh for climate action year 2030 (14% increase). Figure 2(c) shows the distribution of EGU additions through 2050 as a function of climate action year.

3.2. DAC-based climate action

Although DAC is available to be deployed at any time, the least-cost model prefers preventive mitigation over DAC unless preventive mitigation alone without NETs becomes infeasible. This confirms the premise of this study that DAC would serve as a backstop technology option towards achieving climate targets. The model shows that about 7% of the uncertainty scenarios would require DAC starting in climate action year 2031 as shown in figure 1(c). This fraction rises to 80% of the uncertainty scenarios by climate action year 2035. The median DAC capacity installed by 2050 under DAC-based climate action increases from 0.2 Gt CO₂/year in climate action year 2031 to 0.8 Gt CO₂/year in climate action year 2035. The median CO₂ storage through 2050, indicated by the bubble sizes in figure 1(c), is estimated to be 2.5 Gt CO₂ for climate action year 2031, and 9.9 Gt CO₂ for climate action year 2035. These values include CO₂ captured from the DAC plant calciner. The total CO₂ storage potential in the US is estimated to be between 413 and 448 Gt (see section 4.2 in the supplementary material for more on this estimate) according to a recent NETL study (Grant et al 2017). It is therefore unlikely that CO₂ storage for DAC would exceed the total CO₂ storage capacity.

Figures 1(b) and (c) further show that DAC deployment would be preceded by a brief period during which emissions fall significantly. Median emissions intensity of the electricity supply before DAC is deployed is found to be about 25 kg CO₂/MWh, although the 90th percentile value is as high as 93 kg CO₂/MWh. For reference, the average emissions intensity of utility-scale electricity supply in 2017 was about 460 kg CO₂/MWh (US Energy Information Administration 2018). The emissions drop is achieved by retiring most of the remaining fossil fuel EGUs in the fleet and replacing them with low-carbon EGUs to minimize the cost of offsetting further CO₂ emissions from energy sources powering the DAC plants.

Not only is the total EGU retirement under DAC-based climate action considerably higher than preventive mitigation-based climate action as shown in figure 2(c), but the average age of retired EGUs is considerably lower in DAC-based climate action as shown in figure 3(a). In fact, figure 3(b) shows that DAC-based climate action would lead to the early retirement of several of the same EGUs as preventive climate action—only sooner and in greater numbers supplemented by the retirement of much younger EGUs as clearly seen in the peak around age 21—immediately after the capital liability of plants is paid off by age 20. Figure S7 illustrates this phenomenon for a single uncertainty scenario to provide clarity on this crucial point. In other words, we find that in order for DAC-based climate action to cost-effectively achieve 2050 emission targets, the EGU turnover expected under timely preventive climate action cannot be avoided. Rather, deferring this EGU turnover would hasten what would otherwise have been be a more gradual turnover under timely preventive climate action, and the private costs to society would balloon.

**Figure 3.** (a) Generation-weighted average age of retired EGUs as a function of climate action year, where solid line represents the median, and bands represent first and ninth decile values across uncertainty scenarios; and (b) retired generation as a function of EGU age for all uncertainty scenarios.
Download figure:
Standard image High-resolution image

Figure 4 illustrates EGU turnover for a specific scenario. It shows how timely preventive action would lead to a relatively gradual reduction in coal generation and eventual phase out before 2040. Emissions prevented from coal EGUs, together with considerable deployment of renewable EGUs would in fact allow the operation of significantly more efficient natural gas EGUs without CCS and still meet the 2050 CO₂ budget, as shown in figure 4(b). However, continued BAU emissions from delays in climate action would quickly require much higher emissions reduction rates. This could lead to phase out of coal earlier while also limiting the amount of natural gas generation possible, as shown in figure 4(c). These observations also hold true for the uncertainty scenarios that rely on NG-CCS and renewables for low-carbon electricity (roughly 25% of all uncertainty scenarios feature natural gas EGUs with CCS; see figure S8 for an example of one such technology trajectory).

The hastened turnover under delayed DAC-based climate action would significantly increase costs as shown in figure 2(a). The median climate action cost for DAC-based climate action starting 2035 is found to be 1005 billion USD through 2050. Comparing this to the median cost of climate action without DAC starting in 2020 discussed earlier, 135 billion USD, we find that DAC-based climate action will be costlier by an order of magnitude or more, and the higher costs would be deferred to future generations.

DAC-based climate action would also draw a considerable amount of electricity from utilities. The median electricity demand for powering DAC plants is estimated to be 5% of the total electricity supply by 2050, although this demand could be as high as 15% of the total supply if a DAC capacity of more than 2 Gt CO₂/year is deployed. Further, every tonne of CO₂ captured also requires 5.25–8.1 GJ of heat, which means that every 0.1 Gt CO₂/year of DAC capacity installed would create an annual heat energy demand of about 0.5–0.75 quadrillion Btus (quads) in addition to the electricity demand. For context, the total natural gas use for electric power in the US in 2018 was estimated to be about 10.75 quads (US Energy Information Administration 2019). If this heat is provided by natural gas, as has been assumed in this study based on available process data in the literature, DAC would also create a significant demand for non-renewable energy. To mitigate this, alternate sources of high-temperature heat for the DAC process such as biomass-derived fuels and electro-fuels warrant exploration for their process compatibility, overall energy balance, and system-level impacts.

The impact of discount rate on the results was examined by running the model with 4% and 10% discount rates. When the future is discounted at a lower rate of 4%, the model places greater emphasis on preventive mitigation without DAC through higher EGU early retirement and replacement with low-carbon EGUs than the 7% case. The resulting BAU emissions are lower, and therefore DAC deployment is lower than the 7% case. The reverse trend is seen when the future is discounted at a higher rate of 10%. BAU emissions are found to be higher in the 10% case than the 7% case, and therefore DAC deployment is higher and needed earlier than the 7% case. Results for these additional discount rates are shown in section 10.1 of the supplementary material.

3.3. Model limitations

The model used in the study does not pose constraints on the rate of turnover of EGUs. Other potentially important factors excluded from the analysis include 'outgassing' of CO₂ from the oceans (Tokarska and Zickfeld 2015); emissions from the construction of DAC plants (de Jonge et al 2019) and additional EGUs; and non-CO₂ GHG emissions from the electricity generation, transmission, and distribution systems. Any of these factors could effectively reduce the available equivalent CO₂ budget through 2050. To compensate for a reduced CO₂ budget, significantly higher rates of CO₂ removal than those estimated in this study would be needed, which in turn would need even greater numbers of EGU retirements and additions—a nonlinear feedback loop which has been shown in this study to have a nonlinear effect on costs. On the other hand, the model estimates that the median coal and natural gas use could significantly decrease compared to BAU (see figure S9), which would lead to substantially lower upstream emissions from fuel supply chains (such as fugitive methane emissions) that are excluded from the analysis and can increase the equivalent CO₂ budget.

While quantification of the impacts of these factors on the CO₂ budget falls outside the scope of the analysis, we ran the model under two additional emission targets—60% by 2050 and 80% by 2050—to understand the sensitivity of the results and conclusions to this parameter. The results show that for a CO₂ reduction target of 60% by 2050, DAC may be needed in about 20% of the uncertainty scenarios by climate action year 2035. The installed DAC capacity in this case is also found to be lower than the 70% reduction case as would be expected for a lower emission target. For the 80% reduction target, DAC may be needed as early as 2028 and at much higher installed capacity, with more than 50% of the uncertainty scenarios requiring DAC to achieve the emission target by climate action year 2031. Results from these additional emission targets are provided in section 10.2 of the supplementary material.

Climate action costs could be higher than those estimated in this analysis if lost revenues are included in the valuation of EGUs retired early as suggested by some (Pomykacz and Olmsted 2014). Additionally, the effects on the system capital and operating cost from inclusion of factors such as spinning reserves, ancillary services, energy storage, and transmission constraints that may be captured by a dispatch and unit commitment model are not modeled in this study. Costs associated with expanding the transmission and distribution network to support the high penetration of low-carbon EGUs are also excluded. As pointed out by studies in the literature (Kroposki 2017, Heuberger and Mac Dowell 2018), these costs could be substantial and could further increase the climate action costs of DAC-based interventions. On the other hand, as found by Wohland et al (2018), there may also exist synergies within the electric dispatch system whereby DAC could in fact reduce curtailment of renewables and thus decrease overall system costs compared to values estimated in this study. A systematic inclusion of these factors into the least-cost model used in this study should be part of the future work on DAC-based climate action.

4. Conclusions

This analysis finds that continued BAU emissions beyond 2030 would necessitate coupling preventive mitigation with 0.2–1.4 Gt CO₂/year of DAC or similar NET as a backstop technology measure to offset historic emissions in order to achieve a 70% reduction in CO₂ emissions by 2050. This is despite a somewhat optimistic decline in emissions projected under BAU by the least-cost model used in the analysis. Any DAC-based climate action would involve the retirement of many of the same fossil fuel-based EGUs that would be retired under preventive climate action starting now, and necessitate additional retirement of newer and more efficient natural gas-based EGUs (without carbon capture). To compensate for such extensive early retirement of EGUs and meet projected electricity demand, significantly more new low-carbon EGUs would need to be deployed in addition to building DAC plants.

The key conclusions of the study that—(1) DAC is far from a substitute for preventive climate action since EGU turnover expected under preventive climate action would be a pre-requisite for effective CO₂ removal with DAC, and (2) although modest CO₂ removal rates could be achieved with DAC, preventive mitigation would be less expensive, would afford a more gradual turnover of fossil fuel-based EGUs, and require significantly fewer low-carbon EGUs to achieve the 2050 CO₂ budget—remain robust to the uncertainties surrounding costs, electricity demand, and other factors examined in this analysis.

Acknowledgments

This material is based upon work supported by the Beyond Carbon Neutral initiative at the University of Michigan Energy Institute and under seed grant # U052191.

Data availability statement:

Data supporting the findings of this study are included within the article. Publicly accessible links to the input data files, software code, and result files can be found in the supplementary material.

Costs to achieve target net emissions reductions in the US electric sector using direct air capture

Article metrics

Submit

Author e-mails

Author affiliations

Author notes

ORCID iDs

Dates

Peer review information

Abstract

1. Introduction