Environmental trade-offs of direct air capture technologies in climate change mitigation toward 2100

Overview

In this study, we adapt a cradle-to-grave LCA framework that evaluates temporally- and regionally-explicit environmental impacts of DACCS technologies in future electricity systems as projected by climate change mitigation scenarios³⁷. The prospective framework aligns the temporal dimensions of the foreground technology learning and the background system dynamics. The life cycle impacts for the respective DACCS technologies are calculated using the Python-coded LCA framework Brightway2⁴⁸ and life cycle inventory (LCI) data  from the ecoinvent database3.6⁴². The (background) electricity system context is provided by TIMER, the energy module of the IMAGE3.2 Integrated Assessment Model (IAM)³¹. TIMER develops regionally- and temporally-explicit projections for electricity mix, generation efficiency, and electricity-associated emissions, and these outputs are incorporated into another python-coded framework (Wurst)³⁷ to update the electricity-related LCI data in the ecoinvent database, which is then used by Brightway2 to calculate the impacts per DACCS technology and time-step. The calculations are performed for 10-year timesteps from 2020 to 2100.

Models

IMAGE 3.2 is an IAM framework developed to describe the relationships between humans and natural systems and the impacts of these relationships on the provision of ecosystem services to sustain human development³¹. The energy module of IMAGE 3.2, TIMER, is a recursive dynamic (i.e., no-foresight) energy system model representing the global energy system, disaggregated across 26 global regions, with projections till 2100³¹. It includes fossil and renewable primary energy carriers (coal, heavy/light oil, natural gas, modern/traditional biomass, nuclear, concentrated/PV solar, onshore/offshore wind, hydropower, and geothermal). Primary energy carriers can be converted to secondary and final energy carriers (solids, liquids, electricity, hydrogen, heat) to provide energy services for different end-use sectors (heavy industry, transport, residential, services, chemicals and other). The model projects future (useful) energy demand for each end-use sector (industry, transport, residential, commercial, other) based on relationships between energy services and activity, the latter of which is related to economic growth. For each demand sector, secondary energy carriers (including solid and liquid biofuels) compete based on relative costs with each other to meet the useful energy demand. The energy system representation of the IMAGE model does include demand elasticity with carbon prices. This is represented via two distinct mechanisms: (i) Investment in energy efficiency, and (ii) reduced demand in energy services (i.e., reducing consumption and foregoing activities and amenities which demand energy/emissions). The former is represented via technological options (i.e., invest in insulation, more efficient technologies, etc.) and the latter is represented based on econometric data. Energy prices are based on supply curves of energy carriers^49,50. For non-renewable sources, these are formulated in terms of cumulative extraction; while for renewable sources, these are formulated in terms of annual production^51,52,53.

Brightway2 is an open-source framework for LCA calculations in Python⁴⁸. It consists of several modules that handle data import, managing and accessing data, calculating, and analyzing LCA results. The combination of a modular structure, the interactivity of Python, and tunable calculation pathways allows for flexibility and user-defined functionalities in conducting LCA studies and offers new possibilities compared to existing LCA tools.

Wurst is also a Python-based software that enables the systematic modification of LCI databases with external scenario data³⁷. Wurst supports several generic modification types, including changing material efficiency, emissions, relative shares of markets inputs, and separating a global dataset into multiple regions. The current version of Wurst focuses on modifying the ecoinvent LCI database using IMAGE scenario data. More detailed information regarding modification steps of Wurst are discussed in the “LCI database modifications with climate scenario data” section.

Scenario description

The Shared Socioeconomic Pathways – Middle of the Road baseline scenario (SSP2-baseline) projections assume no climate policy whatsoever, thus acting as a counterfactual to which policy efforts can be compared. The RCP1.9 scenarios project the required effort needed to meet a climate target, defined as an emission budget consistent with a 1.5 °C global mean temperature increase. These scenarios also include current climate policy, per region, as defined by the NDCs⁵⁴. For the RCP1.9 scenarios, the IMAGE model determines the additional effort needed to meet the 1.5 °C target, represented by emission price projection across all GHG emission sources (fossil fuels, industry, and land use), applied globally, resulting in a cost-effective mitigation pathway. The emission price can reduce emissions via two mechanisms: (i) the increase in aggregate energy costs promotes investments in energy efficiency, (ii) by attaching this price to the carbon content of primary energy carriers, and it affects their competitiveness at meeting final energy demand services, thus promoting cleaner energy carriers. The application of an emission price makes DACCS competitive as it is assumed that sequestered carbon is renumerated, thus overcoming capital and variable costs (which in turn are affected by the projected cost of energy supply and technological learning). We present two RCP1.9 variations (SSP2-RCP1.9 w/ DACCS and SSP2-RCP1.9 w/o DACCS) to determine the impact of DACCS availability on climate change mitigation strategies. Regional cost-effectiveness in DACCS depends on capital and O&M costs (including endogenous learning-by doing reductions), electricity price, and CO₂ transport and storage costs linked to storage potential limitations⁵⁵. A single DACCS technology (with technology parameters and cost data based on plant capacity of 1 Mt CO₂/year) is included in IMAGE, represented by aggregate of different solvent-based technologies summarized in previous studies^8,56,57, but we assume that the DACCS deployment result estimated by IMAGE will represents the total deployment of a wide range of DACCS technologies (including both solvent- and sorbent-based DACCS). In IMAGE, it is assumed that DACCS is not available before 2030, and its global growth rate is limited to 1 GtCO₂/year. This growth rate limit is a binding constraint in the projection once DACCS becomes cost effective, while in the long-term storage potential limitation may limit its further expansion. DACCS becomes cost effective when emission prices exceed approximately $300/tCO₂. This emission price is surpassed in 2050 for both SSP2-RCP1.9 w/ DACCS and SSP2-RCP1.9 w/o DACCS. In the long-term, the application of DACCS limits the growth of the emission price, projected to be $423/tCO₂ and $885/tCO₂ 2100 for SSP2-RCP1.9 w/ DACCS and SSP2-RCP1.9 w/o DACCS respectively. By calculating the differences of electricity generation and the associated environmental impacts between the two RCP1.9 variations, we can also evaluate the effect of DACCS deployment on the electricity and energy demand systems.

Technology assumptions and details of DACCS systems

We focus on two types of DACCS technologies: a solvent-based and a sorbent-based DACCS, which rely on different capture and release mechanisms to remove CO₂ from the atmosphere.

Solvent-based DACCS applies aqueous hydroxide solutions (potassium hydroxide, sodium hydroxide) to capture atmospheric CO₂ via a chemical reaction^7,8,9,10. Here, we assume the solvent-based DACCS uses potassium hydroxide solution for CO₂ capture. In an air contactor, the potassium hydroxide solution reacts with CO₂ and forms potassium carbonate, which then, in a separate reactor, reacts with calcium hydroxide and generates calcium carbonate. The calcium carbonate precipitates, and potassium hydroxide solution can be regenerated and recycled back to the air contactor. The precipitated calcium carbonate is collected, dried, and then calcined under high temperature (about 900 °C) heat, which is typically provided by natural gas combustion in pure oxygen, to release the CO₂. The CO₂ released from calcium carbonate and the CO₂ generated by natural gas combustion are mixed and collected for further storage¹⁰. The high temperature heat requirements limit the heat supply options for solvent-based DACCS. In this study, we consider natural and renewable gas (biomethane) as the two heat options for the solvent-based DACCS (Supplementary Fig. 1). Other proposed methods include electric resistance heating and electrochemical regeneration, which were not studied here.

Sorbent-based DACCS typically uses amine materials bonded to a wide range of porous solid supports for CO₂ capture^11,12,13,14. Here, we considered the use of amine-based silica as the solid sorbent²⁴. The process consists of two main steps that operate cyclically: adsorption and desorption. In the adsorption step, a fan blows air through the air contactor, and the CO₂ in the air reacts with the sorbent and binds to it. When the solid sorbent has been saturated with CO₂, the desorption step will start in the air collector. Before heat is supplied, a vacuum is pulled to remove residual air from the contactor and decrease the temperature required for regeneration. Then, heat at about 100 °C will be supplied into the air contactor to desorb the CO₂. The collected CO₂ will then go through a cooling unit, where extra moisture can be removed through condensation and CO₂ will be brought to ambient temperature. In the desorption step, the temperature of heat is about 80–120 °C, so a wide variety of thermal energy sources (natural gas, heat pump, geothermal heat, and waste heat) can be used as the heat supply. Here, we model heat pump (with coefficient of performance of 2.5²⁴) and renewable gas (biomethane) as the two main options (Supplementary Fig. 1).

CO₂ transport and storage: Once the CO₂ is released from either process, we assume the CO₂ flow will be compressed through a compressor to 11 MPa and then transported through a pipeline to the storage site. The length of the transport pipeline is assumed to be 50 km. At the storage site, the CO₂ will be further compressed to 15 MPa and injected into a geological reservoir through wells with the depth of 3 km each. Here, the CO₂ will be permanently stored as supercritical phase⁵⁸(Supplementary Fig. 1).

Life cycle assessment

The system boundary starts at the air inlet with a CO₂ concentration of 415 ppm, and is followed by CO₂ capture, regeneration, compression, transport, and ends with geological storage. Our analysis also accounts for upstream emissions due to indirect energy demands for the construction of energy conversion technologies, fuel production and handling. The functional unit is capturing and sequestering one metric tonne (1t) of atmospheric CO₂ by DACCS technologies. The LCI data of the two studied DACCS technologies and subsequent compression and storage were collected from literature or estimated through bottoms-up materials requirements analysis (with the assumed plant capacities of 1 Mt CO₂ and 0.1 Mt CO₂ per year for solvent- and sorbent-based DACCS respectively), which are discussed in detail in Supplementary Note 1. The LCI data are assumed to represent the status quo material and energy consumptions over the life cycle of the two selected DACCS technologies. ReCiPe 2016 v1.1 hierarchist perspective is used as the characterization method to convert emissions and natural resource extractions to environmental impact categories at mid-point level⁵⁹.

In this study, when we compare the environmental impacts of DACCS under different electricity decarbonization pathways (SSP2-baseline vs. SSP2-RCP1.9 w/ DACCS), the results are calculated based on static LCI data of DACCS that represent their current material and energy uses without considering technology learning. Then, we also calculated another set of LCA results for DACCS under SSP2-RCP1.9 w/ DACCS scenario based on dynamic LCI data that are estimated using learning curve approach, so it captures the effects of both background electricity decarbonization and foreground technology learning. By comparing the LCA results of DACCS calculated using static and dynamic LCI data under SSP2-RCP1.9 w/ DACCS scenario, we can evaluate and compare the effects of background electricity decarbonization and foreground technology learning on the environmental impacts of DACCS.

Technology learning of DACCS systems

The learning curve approach has been used as an empirical method to study the unit cost reduction over time with cumulative production increases for a wide range of manufacturing⁶⁰ and energy technologies⁶¹. The learning effect can be characterized by various mechanisms, including technology advancement, increased labor productivity, economies-of-scale, and improved material and energy efficiency. The learning curve approach has also been acknowledged as one critical means to explore the future expected life cycle impacts of present-day emerging technologies^62,63. Here, we apply the one-factor learning curve approach to inform our prospective LCA. While the two technologies under investigation are presently operating in pilot- or demonstration scale, we assume a commercial-scale operation for both and apply constant learning rates, affecting the future life cycle material and energy consumption. Yet, for both technologies assessed herein, these learning effect on material and energy consumption are missing in the published literature. Thus, we assumed changes of material and energy consumption proportional to the changes per unit cost for the DACCS technologies.

It has been shown that the capital costs of solvent- and sorbent-based DACCS are likely to follow different learning rates given their different design characteristics. The solvent-based DACCS is site-built and large-scale, benefitting from economies-of-scale, but it is also less likely to incorporate rapid design or manufacturing improvement, while sorbent-based DACCS is based on standardized and modular units, and these units can be mass-produced and deployed, which enables fast iteration and learning⁶⁴. Therefore, we assumed the average learning rates of 10% and 15% for the material and energy consumption that are related to capital investments for solvent- and sorbent-based DACCS, respectively. Then, as for the material and energy consumption related to operational costs, we assumed average learning rates of 2.5% for both solvent- and sorbent-based DACCS, respectively. We also consider variation ranges for the learning rates to reflect their uncertainty (Supplementary Table 10), these variation ranges are used to develop a sensitivity analysis to understand how the speed of learning affects the environmental impacts of DACCS. Furthermore, to avoid unrealistic reductions of material and energy consumption under technology learning, we also set up minimum material and energy use factors of both DACCS technologies based on expert estimations. As for the solvent-based DACCS, the lower bound of material and energy uses related to capital and operational costs cannot be lower than 44% and 50% of their original amounts, respectively, and the sorbent-based DACCS, the lower bound of material and energy uses related to capital and operational costs cannot be lower than 18% and 50% of their original amounts in 2020, respectively. To incorporate the minimum material and energy use factors into the learning curve formula, we adjusted the learning curve formula into the following Eq. 1:

In Eq. 1, ({X}_{0}) represents the initial DAC deployment capacity at year (0); ({X}_{t}) represents the cumulative DAC deployment capacity at year (t). For a specific material or energy item (i), ({{LR}}_{i}) represents the learning rate of the item (i); ({D}_{i,0}) typically represents the unit consumption of the material or energy item (i) at year (0) (corresponding to the initial CO₂ capture ({X}_{0})). Here our goal is to calculate the material and energy use factors (instead of actual unit consumption) under technology learning, so we normalize the ({D}_{i,0}) to be 1; ({D}_{i,t}) is also a normalized material and energy use factors of item (i) at year (t) (corresponding to the cumulative CO₂ capture ({X}_{t})); ({D}_{i,{min }}) represents the minimum material and energy use factors of item (i).

Finally, we assume that solvent- and sorbent-based DACCS each account for half of the global cumulative capacity of DACCS (IMAGE model outputs), respectively. Then, we estimated material and energy use factors for both solvent- and sorbent-based DACCS from 2020 to 2100 based on their cumulative capacity, and the results are presented in Supplementary Table 11. By multiplying the material and energy use factors at a specific year to the actual unit material and energy consumption at the initial year, we can get the actual unit material and energy consumption in that specific year. Assumptions on technology learning rates and minimum material and energy use factors of solvent- and sorbent-based DACCS are discussed in detail in Supplementary Note 2.

LCI database modifications with climate scenario data

The ecoinvent database²³ is the most widely used LCI database which offers fully interlinked unit process supply chains for products presented in the database. It covers all relevant environmental flows, material and energy inputs, and products of around 18,000 activities, where researchers can collect data about the supply chain to form a comprehensive background system in an LCA study. However, since the data in ecoinvent are usually collected in a specific year, the database describes the material and energy flows among processes based on an existing supply chain system. Therefore, the ecoinvent database is limited in conducting prospective LCA studies, which assess the environmental impacts associated to future technologies or emerging technologies that evolve over time.

Here, to evaluate the environmental impacts of DACCS technologies in a context of a changing background electricity system, we adapt an open-source approach (Wurst)³⁷ that systematically integrates the IMAGE projections on electricity mix, generation efficiency, and electricity-associated emissions with the ecoinvent database, and change the parameters in electricity-related activity data in the ecoinvent database. Due to the differences of generation technologies between IMAGE and ecoinvent database, we develop a matching list to map the available technologies for both data sources (Supplementary Note 3). More detailed information regarding parameter modification for ecoinvent database using Wurst can be found in a previous study³⁷. After the parameter modification, we developed 27 versions of ecoinvent databases, which correspond to 9 different years from 2020 to 2100 under the SSP2-baseline, SSP2-RCP1.9 w/ DACCS, and SSP2-RCP1.9 w/o DACCS scenarios.

Limitations

In this study, we modify the background LCI database using IMAGE projections of grid mix, generation efficiency and emissions of thermal power plants (fossil-based sources, biomass, and nuclear), while the renewable sources and their efficiency levels are based on existing available technologies. Technological innovation has been observed for renewable (especially solar^65,66 and wind⁶⁷) and energy storage^68,69 technologies, and they will continue to evolve as they are more widely applied in the energy system. Therefore, to better evaluate the prospective environmental impacts of energy-intensive technologies, such as DACCS, under specific climate contexts, the analysis framework could be expanded to consider the advancement, particularly in material efficiency or circularity of variable renewable energy and storage technologies in the background electricity system.

Previous studies looking at the technology learning of DACCS have focused on cost reductions^64,70,71. Publicly available, empirical studies that reveal how material and energy inputs change as DACCS scales could not be identified. Given this limited data availability, we assume the material and energy inputs of DACCS follow the same learning rates as the associated cost projections. In reality, technology learning rates are likely to vary depending on processes and physical input types^29,72. Future LCA studies aiming to quantify the effects of technology learning on environmental impacts might be able to rely on more detailed learning data of specific physical inputs. In addition, learning rates of emerging technologies tend to change with technology-readiness-levels (TRL)^73,74,75. Prospective analyses of emerging technologies ideally reflect this by applying a multi-factor learning curve approach, differentiating between the varying learning rates at different TRL. The technologies analyzed herein operate at demonstration scale (TRL-7) while we apply a single-factor, constant learning rate, postulating learning-by-doing improvements at commercial scale (TRL-9). The learning rate at commercial scale is a research frontier and presently unknown. Yet, at the scale of our analysis, a respective differentiation is unlikely to add accuracy or insight. The uncertainty with respect to the specific single learning rate at commercial scale is captured by testing how different learning rates affect our results. Using a single-factor learning curve approach, we thus attribute the cost change and its related material and energy consumption to the cumulative installed capacity of DACCS over time, limiting our capability of revealing the correlation between technology progress and other factors, such as (prior) R&D expenditure⁷⁶.

The life cycle impact assessment step relates emissions and resource use to environmental impacts through characterization factors. The framework we adapted here applied global or European scale characterization factors. While location-generic characterization factors are suitable for global impacts such as climate change impact, they may lead to large uncertainty for quantifying non-global impacts, such as acidification⁷⁷, eutrophication⁷⁸, and ecotoxicity⁷⁹, which are typically affected by regional meteorological, hydrological, soil conditions and the sensitivity of ecosystems to emissions. While country-dependent characterization models and factors have been developed for these impact categories, they have not yet been incorporated into the LCA framework applied in this study. Further methodological improvements are needed to enhance the capability of the existing framework for conducting regional impact assessments.

This study shows the environmental impacts of DACCS could have different trajectories depending on the background energy system, so it is important to keep monitoring those environmental metrics or even considering them in the decision-making process. Future research could explore the feasibility of incorporating life cycle environmental metrics into IAMs for better environmental impact assessment. State-of-the art IAMs typically include some environment-related metrics, such as GHG emission, land and water use as constraints, but they lack many other environmental impact dimensions. For example, metal consumption could be an important metric given the increasing penetration of renewable and battery storage in the energy system, which are resource intensive. Furthermore, life cycle environmental metrics capture the emissions from all life cycle phases (e.g., construction, transport, operation, and end-of-life, etc.), and IAMs evaluate the interrelationship among different sectors. Therefore, the integration of life cycle environmental metrics and IAMs should carefully allocate the emissions of different life cycle phases to the corresponding sectors/energy carriers in IAM to avoid double counting⁸⁰.

Comments are closed.