Integrating climate and physical constraints into assessments of net capture from direct air capture facilities

Significance

Technologies to remove carbon dioxide directly from the atmosphere are expected to play a role in achieving emissions targets. It is known that process performance is affected by site-specific factors, including ambient environmental conditions and the emissions intensity of the energy supply. Here, we integrate climate and emissions intensity data with a process model to make quantified global predictions of performance for a leading removal technology. Our results highlight the importance of climate and low-carbon energy supply, but also the importance of international cooperation where there is a mismatch between carbon dioxide removal ambitions and suitable climates. Similar assessments are needed for other removal technologies, where the impact of environmental conditions and energy requirements might vary.

Abstract

Limiting climate change to targets enshrined in the Paris Agreement will require both deep decarbonization of the energy system and the deployment of carbon dioxide removal at potentially large scale (gigatons of annual removal). Nations are pursuing direct air capture to compensate for inertia in the expansion of low-carbon energy systems, decarbonize hard-to-abate sectors, and address legacy emissions. Global assessments of this technology have failed to integrate factors that affect net capture and removal cost, including ambient conditions like temperature and humidity, as well as emission factors of electricity and natural gas systems. We present an integrated assessment of the global deployment potential of this technology. Employing a chemical process model, climate data, grid emission factors, and fugitive methane emission factors, we predict critical performance metrics, including carbon dioxide capture rates, and water-, energy-, and emissions-intensity of capture. Our results support investors and policy makers as they site facilities and develop credible policy instruments to support expansion.

Despite increased investment in low-carbon energy generation, progress toward the deep decarbonization of the global energy system remains slow relative to the ambitious net-zero emission targets proposed by many nations and the climate stabilization targets enshrined in global accords like the Paris Agreement (1). As a result, integrated assessment models of the global climate and energy system increasingly deploy carbon dioxide (CO₂) removal (CDR) to limit overshoot of climate targets. There is now broad agreement that limiting global warming to 2 ^°C or cooler will require such CDR (2), in addition to emissions reductions.

Conventional land-based CDR, such as afforestation, forms the majority of CDR committed through nationally determined contributions under the Paris Agreement. However, emissions reductions in the latter half of the century from conventional CDR are expected to decline as sinks become saturated (3). Novel, as opposed to conventional, CDR includes technologies of varied technological and commercial readiness, including the Direct Air Capture (DAC) of CO₂ from ambient air. Within the Intergovernmental Panel on Climate Change’s (IPCC) Sixth Assessment Report, DAC deployment is modeled between 0 and 39 GtCO₂/y by 2,100, with cost estimates of 84 to 386 USD/tCO₂ (2). These wide ranges reflect both the range of mitigation pathways (not all of which include DAC) and the technology readiness level of DAC. In submitted long-term low-emission development strategies (LT-LEDs), novel CDR comprises 0.6 GtCO₂/y by 2050—a substantial figure that nonetheless represents a shortfall of 0.31 to 2.9 GtCO₂/y for pathways representative of low, medium, and high levels of CDR, respectively (3).

Deploying DAC at the scale required might prove to be more expensive and energy intensive than models assume: Global assessments of DAC performance have failed to integrate key factors that affect the net capture rate of the technology, chief among which are ambient environmental conditions like temperature, humidity, and pressure that affect the performance and cost of capture. In addition, different jurisdictions have different emission factors for both their electricity and their natural gas systems, affecting net capture rates further. Modeling exists for the continental United States but does not consider pressure, variability in CO₂ concentration, or the full range of temperature and humidity conditions (4). Besides our previous modeling of an all-electric plant within a Canadian context (5), no analysis has yet integrated these constraints to faithfully represent the likely performance of DAC in the real world. This is key to making prudent investment and policy decisions, especially as nations begin to envision a role for DAC in their net-zero emission plans (6).

Here, we present results from an integrated assessment of the global deployment potential of a leading liquid-solvent DAC technology. Integrating a chemical process model of this system, climate data, grid emission factors, and fugitive methane emissions, we predict both capture rates (relative to a nominal 1 MtCO₂ gross capture facility operating at 21 ^°C and 1 atm) and the water-, energy-, and emissions-intensity of capture. Our results support investors as they site the first generation of facilities; both investors and policymakers as they consider the credibility of revenues provided through various policy instruments, like tax credits; and policy makers as they explore border carbon adjustments and cross-border carbon offsetting schemes for CDR.

Results

Environmental Dependence of Capture Rates and Water Loss.

Fig. 1 shows the effect of ambient environmental conditions on the performance of a nominal 1 MtCO₂ (gross capture) facility. Performance is quantified in Fig. 1A by the gross capture rate—the total amount of CO₂ extracted from the atmosphere, not including emissions captured from the combustion of CH₄—and in Fig. 1B by the make-up water required after evaporative losses in the contactor and capture of H₂O from CH₄ combustion are taken into account.

Fig. 1.

Global hourly climate reanalysis data (for the year 2019) was supplied as input to a nominal 1 MtCO₂ capture plant. (A) Sum of hourly capture rates predicted by the model, as a function of air temperature, pressure, humidity, and CO₂ concentration. White areas represent freezing conditions throughout the year. (B) Annual net water loss, normalized by annual capture rate. Water loss is the difference of evaporative losses from the contactor and water gained through combustion of CH₄ in the calciner.

Capture rates are strongly driven by temperature and, to a lesser extent, pressure. The aqueous liquid solvent requires temperatures above freezing to operate, and when above freezing, reaction kinetics are improved by higher temperatures. Note that the predicted capture rate does not linearly decrease to zero with temperature at freezing: Even at 1 ^°C (assuming 400 ppm CO₂, and 90 kPa air pressure), the hourly capture rate corresponds to 0.66 MtCO₂ of annual capture. Because of this, capture rates strongly favor subtropical and tropical latitudes, though more temperate coastal regions, where the ocean moderates winter temperatures, may also be suitable. Within the warmer latitudes, lower-elevation areas are preferable, as higher elevations are cooler and the lower pressure reduces the solubility of CO₂ in the solvent.

Water losses are strongly affected by temperature, humidity, and pressure (see ref. 5 for a more detailed treatment); we see the greatest losses in hot, dry regions like the Middle East, North Africa, and Australia, up to an annual average of approximately 20 tH₂O/tCO₂. Typically, these regions are already water-stressed, though they also enjoy high solar resource potential and high capture rates, which would otherwise make them attractive for DAC.

Differences in the total energy intensity of capture are largely driven by altitude, and the global distribution is plotted in SI Appendix, Fig. S1. Pressure decreases at altitude lead to higher fan loads at the contactor, as a larger volume of air must pass through the contactor for the same mass flow rate, as well as a greater pressure difference across the compressor. For most regions, the energy intensity of capture is predicted between 1.98 and 2.05 MWh/tCO₂, representing the 5th and 95th percentiles, respectively.

Net Capture Rates.

In Fig. 2, we plot the net capture rate as defined in (Eq. 5). This rate only includes emissions associated with the energy required for the capture process. Notably, any emissions associated with make-up water (531 t/h at nominal conditions) or calcium carbonate (3.4 t/h at nominal conditions) (7), have not been included, though emissions from makeup water for one example are given in the Discussion below. Embodied emissions from the construction of the facility are not included in the net capture rate but have been estimated at $9.6\times {10}^{-4}$ tCO_2e/tCO₂ (8).

Fig. 2.

Net capture rates after accounting for energy emission factors. (A, B) use national averages. (C) uses national natural gas emission factors and regional grid emission factors. (D) uses provincial or state-level grid emission factors, and national natural gas emission factors, except for the three westernmost Canadian provinces of British Columbia, Alberta, and Saskatchewan, where more detailed methane emissions data are available. (E) uses national averages.

Regional Shares of Capture and Energy-Based Emissions.

Fig. 3 shows the gross capture rate, grid emissions, and natural gas supply chain emissions that constitute the net capture rate, assuming that a nominal 1 MtCO₂ facility is located where regional microclimates are most advantageous (a comprehensive figure showing each country is included in SI Appendix). This breakdown emphasizes that a DAC project must have a suitable climate, low-carbon electricity supply, and tight natural gas supply to be considered for development. Except at low air pressures, where fan and compression loads are higher, process loads are split at approximately 24% electrical power and 76% natural gas power for the plant configuration in this study.

Fig. 3.

Decomposing the effect on net capture of climate, grid emissions, and upstream methane emissions. Horizontal bars indicate capture rates (positive) and related emissions (negative) for a 1 MtCO₂ nominal capture facility. Yellow bars represent gross capture and green bars represent net capture if the plant is located where the climate is most favorable for capture within the region identified. Red markers indicate the geographically averaged net capture rate throughout the region.

Discussion

Where integrated assessment models include DAC, DAC has been found to contain mitigation costs (2), presumably by placing an upper bound on the cost of emissions abatement. Our model allows for a more accurate weighing of the cost of DAC versus other mitigation options, by taking into account the effects of regional climate on capture rate and, to a coarser extent, methane and grid emissions intensities.

For the most part, climate will have a minimal impact on the energy intensity of capture (and hence operating expenses), but will impact the cost of mitigation through the efficiency of capital cost utilization. Emissions associated with energy supply will directly increase the cost of mitigation, as energy use scales linearly with capture rate. For example, the best annual gross capture rate in Texas, United States, is predicted at 1.04 MtCO₂/y. Texas is set to be the site of the first full-scale liquid-solvent DAC facility, with a nameplate capacity of 0.5 MtCO₂/y [as additional context, out of all parties to the United Nations Framework Convention on Climate Change (UNFCCC), the United States forecasts the highest levels of negative emissions by 2050, at 500 MtCO₂e/y (9)]. Assuming dedicated zero-emission electrical power, and natural gas only supplying the calciner, upstream methane leakage rates at the US average of 1.33% represent emissions of 0.04 MtCO₂e/y—a 4% loss of capture efficiency. Substituting values for capital expenses of 1B USD, natural gas at 2.59 USD/GJ [Texas’ 2023 annual average industrial price (10)], electricity at 69 USD/MWh [the globally averaged levelized cost of energy for solar PV with storage (11)], a discount rate of 5%, and a capital amortization period of 25 y—yielding a capital recovery factor (CRF) of 0.071 (Eq. 6)—we get an annual levelized cost of carbon removal (LCOR) of 122 USD/tCO₂e (Eq. 7), representing a 4% increase compared to a zero-emissions scenario.

There is a high degree of uncertainty in natural gas emission factors, especially for jurisdictions that use a generic emission factor, with uncertainties ranging from ± 5.2% to ± 500% (12). Where measurements do exist, there can also be significant variability in leakage rates from basin to basin, such as in the United States (0.64% to 3.87%, after attributing emissions to natural gas or oil production) (13). For the Texas example above, basin-to-basin variability in emissions is equivalent to 1.8% to 12.2% of the captured carbon.

Our model also provides an estimate of evaporative losses for conditions outside of the ranges provided by developers (2 to 9 tH₂O/tCO₂ for a temperature range of 20 to 30 ^°C and relative humidity range of 35 to 75%) (7), and more specific than the range of 0 to 50 tH₂O/tCO₂ referenced by the IPCC (2), which is based on the moisture carrying capacity of dry air and a significantly higher volume flow rate (14). For example, the United Arab Emirates (UAE) has committed to 9.5 MtCO₂e/y of negative emissions via DAC by 2050 (15). Based on 2019 data, if using liquid-solvent DAC, we predict an average net water loss of 12.6 tH₂O/tCO₂ given the UAE’s climate, or 120 MtH₂O/y. This can be compared to the 2019 water consumption of the city of Abu Dhabi of 492 MtH₂O (16). If we assume makeup water is supplied by desalination at 3.0 kWh/m³ (17) and using the UAE grid emission factor of 0.418 kgCO₂e/kWh, associated emissions from makeup water total 0.15 MtCO₂e, or 1.6% of emissions captured. Desalination also has other large environmental impacts beyond process emissions.

Our integrated assessment highlights the importance of two other results. First, the difference across borders is stark: There are numerous examples in Fig. 2 where, although climate does not change appreciably across an international border, the net capture rate does. Examples include North versus South Sudan, France versus Germany, Czechia versus Slovakia, or Manitoba versus Saskatchewan in Canada. This further highlights the importance of low-carbon electricity supply systems and tight control of methane fugitive emissions from natural gas systems (e.g., Germany’s electricity supply system is more carbon-intensive than France’s; Slovakia’s fugitive methane emission data are higher than any other European nation’s). Second, DAC highlights the importance of establishing credible international arrangements for carbon offsetting. Countries that plan on deploying DAC, including Canada, the United Kingdom, and Switzerland, might not be able to deploy liquid solvent DAC cost-effectively. They will therefore need to consider either financially supporting the deployment of these technologies in subtropical and tropical latitudes, or consider the suitability of other DAC technologies, including solid sorbent systems. All nations must radically scale up their low-carbon energy supply in the long term. In the near term, nations must weigh the carbon intensity of their planned electricity and methane supply systems. Considering these factors alone can assist investors and policy makers in making better upstream choices when investing in climate mitigation. Renewable or low-carbon energy potential, and CO₂ storage potential, are other region-specific characteristics that are critical for assessing the viability of expanding DAC deployment.

One limitation of this analysis is its focus on one leading DAC technology that has approached commercial maturity. Other liquid solvent-based removal technologies exist, as do solid sorbent technologies that employ adsorption rather than absorption. The effect of ambient environmental conditions on the performance of these systems differs from the liquid-solvent process modeled here. The energy requirements of other systems also vary. For example, solid sorbent systems tend to operate at much lower temperatures than the 900 ^°3C required in this process for the calciner.

Without substantial process modifications, other solvent-based technologies might suffer from similar evaporative losses to those described in this analysis. Work on solid sorbent technologies, including sorbents that are designed for cold climates, shows that although they capture more CO₂ from ambient air at cold temperatures (18), the adsorption step also requires more time to complete. This leads to fewer capture cycles annually and could reduce overall capture, though more work is needed to evaluate these capture dynamics.

Materials and Methods

Process Model.

The process model is based on a liquid-solvent horizontal-flow contactor, sized for a 1 Mt capture facility. Process modeling focuses on the contactor, where ambient air is driven through a packed column, and atmospheric CO₂ dissolves into and reacts with an aqueous solution of potassium hydroxide (KOH).

Modeling is restricted to mass and energy balances across the remaining process steps. Our abstraction of the process is described in detail elsewhere (5). Instead of assuming that all loads are electrical, as in ref. 5, here we assume a natural gas-fired calciner, with natural gas modeled as methane, CH₄ (results for the all-electric configuration are included in SI Appendix and data). A natural-gas fired calciner represents lower technological risk (8), and makes this work more comparable to other scholarship based on the original design described elsewhere (7). Accordingly, an air separation unit (ASU) supplying oxygen to the combustion reaction is now included as an additional electrical load, and water is included as a product of methane combustion (Eq. 1). All calciner inputs are assumed to be heated to an operating temperature of 900 ^°C, heat recovery on output streams is assumed at 50%, and combustion enthalpy $\mathrm{\Delta }{H}_{\text{combustion}}$ is modeled at 890 kJ/mol-CH₄ (19). Additional correlations for the heat capacity of water, methane, oxygen, and higher-temperature rate constants for the reaction of KOH with CO₂ are included in SI Appendix.

$$\begin{array}{cc}\hfill C{H}_4+2O_2& ⟶C{O}_2+2H_{2}O\hfill \end{array}$$ [1]

We use an ASU load of 238 kWh/tO₂ as described by Keith et al. (7), and which agrees with other estimates described elsewhere (8). Oxygen mass flow rates are stoichiometrically balanced for methane combustion. The methane molar flow rate ${\dot{n}}_{C{H}_4}$ is balanced with the CO₂ capture rate such that 90% of the combustion energy is available for calcination (Eq. 2).

$$\begin{array}{cc}\hfill 0.9& =\frac{{\dot{n}}_{C{O}_2}\times \mathrm{\Delta }{H}_{\text{calcination}}}{{\dot{n}}_{C{H}_4}\times \mathrm{\Delta }{H}_{\text{combustion}}}\hfill \end{array}$$ [2]

The process is simulated for a 4-dimensional array of environmental variables, assuming steady-state conditions with solvent inlet temperatures at ambient air temperature. The environmental variables comprise air temperature between 1 and 52 ^°C at 1 ^°C intervals, relative humidity between 0 and 100% at 1% intervals, air pressure between 48 and 107 kPa at 1 kPa intervals, and CO₂ concentration between 395 and 447 parts per million (ppm) at 1 ppm intervals.

Climate Data.

To demonstrate the effect of regional climate on capture performance, we consider gridded global reanalysis data for 2019 from the Copernicus Atmosphere Monitoring Service (CAMS). CO₂ concentrations are from the global greenhouse gas reanalysis (EGG4) dataset (20), available at 3 h and 0.75 degree resolution, while air temperature, humidity, and pressure data are from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5) dataset (21), available at 1 h and 0.25 degree resolution.

For each climate data point, we find the simulation results for the nearest 4-tuple of environmental variables, and then sum the hourly results at each location to produce annual performance metrics (CO₂ data was up-sampled without interpolation to match the resolution of the other climate data). Capture performance is measured in terms of the amount of CO₂ captured from the air, net water loss, and energy consumed.

Energy Emission Factors.

DAC is energy intensive, and we explore the sensitivity of the net carbon removal rate to the emissions intensity of the energy supply. For each annual performance data point, we apply grid and natural gas emission rates to the electrical and natural gas energy consumption, respectively.

National, as well as state and provincial grid-emission factor data for the United States and Canada, were compiled by Carbon Footprint (22). For China, we use the 2014 revised emission factors calculated by Tao (23) for the six regional power grids.

Most natural gas upstream emission factor data are national estimates from Cooper (12). We repeat Cooper in noting that there is a high level of uncertainty in these estimates, in large part because the majority of jurisdictions use a generic IPCC emission factor (2.21%) in the absence of other data. For regions where emissions data are incomplete or missing, we used the same generic emission factor. For the United States, we use the average emission factor of 1.33% found by Sherwin (13) (Sherwin reports the total emission fraction of 2.95%; 1.33% represents the weighted share attributed to natural gas production versus oil production). For Canadian jurisdictions, we use the national estimate of 0.64%, except where provincial emission factor data are available, as is the case for Saskatchewan [2.44% (24)], Alberta [1.58% (25)], and British Columbia [0.38% (25)].

The annual net capture rate (Eq. 5) reported here is the difference between the gross capture rate and the emissions associated with the carbon intensity of the electrical grid (Eq. 3) and upstream emissions from natural gas supply chains (Eq. 4), where EF-grid (tCO₂/MWh) and EF-CH₄ (% loss) are the respective emission factors, P_e is the electrical energy intensity of capture (MWh/tCO₂), NG is the CH₄ intensity of capture (tCH₄/tCO₂), and GWP is the global warming potential of methane (tCO₂/tCH₄). We use a GWP of 29.8, corresponding to the latest guidance within the IPCC Sixth Assessment Report for a 100 y analysis (26).

$$\text{Grid emissions}=\left(\frac{\mathrm{\text{EF-grid}}}{P_e}\right)\times \text{Capture rate}$$ [3]

$${\text{CH}}_4\text{emissions}=\left(\frac{{\text{EF-CH}}_4}{100\%}\times \text{NG}\times \text{GWP}\right)\times \text{Capture rate}$$ [4]

$$\text{Net capture}=\text{Capture rate}-\text{Grid emissions}-{\text{CH}}_4\text{emissions}$$ [5]

LCOR.

For the Texas example given above, we assume a discount rate r of 0.05 and capital amortization period n of 25 y. By (Eq. 6) this yields a CRF of 0.071. To calculate the LCOR (Eq. 7), capital (CapEx) and operating (OpEx) expenses are in USD per year, and the capture rate and emissions are in tCO₂e per year, yielding an LCOR in USD per tCO₂e.

$$\text{CRF}=\frac{r}{1-{\left(1+r\right)}^{-n}}$$ [6]

$$\text{LCOR}=\frac{\text{CapEx}\times \text{CRF}+\text{OpEx}}{\text{Capture Rate}-\text{emissions}}$$ [7]

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

The process model’s code is published under an open license via CC BY-NC-SA 4.0. (27). All other data are included in the manuscript and/or SI Appendix.