Analysis of carbon capture strategies for refineries considering extraordinary future economic and policy uncertainties

Abstract

The shifting transportation landscape, driven primarily by the electrification of the vehicle fleet and tightening carbon regulations, is reshaping the market for liquid fuels and challenging many existing refineries. Strategic decisions about carbon mitigation investments, such as carbon capture and storage (CCS), are complicated by uncertainty in evolving fuel markets and regulatory frameworks. This study assesses the cost-effectiveness and the opportunity of deploying CCS within the refining sector in the United States amidst a transitioning transportation fuel market and increasingly stringent carbon pricing policies. By integrating a carbon capture module into the Petroleum Refinery Life Cycle Inventory Model (PRELIM), we assess greenhouse gas (GHG) emissions under various fuel demand scenarios and examine the role of CCS in mitigation. Our findings indicate that the annual GHG emissions from U.S. refineries may decline from 212 MtCO₂eq per year in 2019 to 111 MtCO₂eq per year in 2050 due to demand shifts, while CCS could further reduce emissions to 47.6 MtCO₂eq per year. Fluid Catalytic Crackers (FCCs) are the most significant emissions source, with a potential reduction of 18.5 MtCO₂eq per year at a CO₂ avoidance cost as low as $83.5 per t. We also conduct a real options (RO) analysis to explore how refineries might respond to fuel market changes and carbon pricing. Results suggest hydroskimming refineries could be phased out by the early 2030s due to market and policy pressures, while medium and deep conversion refineries may accelerate CCS deployment by the 2040s. Hence, policy should enable CCS deployment as a transitional mitigation strategy while maintaining consistency with long-term decarbonization pathways.

---

Sustainability spotlight

Rising electric vehicle (EV) adoption is accelerating the shift away from petroleum-derived fuels. Together with increasingly stringent carbon pricing, this trend challenges the refining sector by raising questions about whether refineries can remain profitable under uncertain fuel demand while meeting tighter emissions targets. This study examines how fuel demand shifts affect greenhouse gas (GHG) emissions, the cost-effectiveness of carbon capture in the U.S. refining sector, and refinery responses to market and policy change. Our findings suggest that CCS may function either as a transitional mitigation strategy or as a means of extending refinery operating lifetimes, depending on the timing and strength of market and policy signals. Policy must therefore balance supply security with long-term decarbonization.

Introduction

Petroleum refineries convert crude oil into different high-value products (e.g., gasoline, diesel, jet fuel, fuel oil) that fuel the economy today. For example, energy sourced from petroleum contributed 38% (37.3 Exajoules) of total primary energy consumed in the U.S. in 2023, with most (i.e., ∼70%) being consumed by the transportation sector.¹ The refining sector is also the third largest greenhouse gas (GHG) emitting industry globally, accounting for 6% of global GHG emissions.² Efforts to restrict global temperature rise to a maximum of 1.5 °C above pre-industrial levels,³ have created unprecedented pressure on refineries to remain economically competitive while simultaneously reducing their carbon footprint, adapting to dynamic shifts in demand for their products and input crude supply options. As an example, according to the National Renewable Energy Laboratory (NREL)'s Electrification Future Study (EFS),⁴ the demand for gasoline in U.S. could drop from ∼19.7 Gigajoules per year (GJ per year) in 2019 to a range of about ∼3.97 to ∼15.6 GJ per year depending on the extent of personal vehicle fleet electrification (i.e., electric vehicles gaining market share over internal combustion engines). Another example is that crude quality (e.g., API gravity and sulfur content) fluctuates based on refiners' access to cheaper crudes and their ability to process these crudes into preferred products, and this creates uncertainties in decarbonizing the refining sector.⁵ At the same time, the U.S. have pursued policies aimed at reducing GHG emissions, with ambitions varying across administrations.^6–8 The timing and stringency of current and potential future climate policies create unprecedented uncertainty as refiners, as well as the sectors upstream and downstream in the supply chain, make critical decisions over the next decade.⁹

The strategies that refineries could employ to respond to these pressures and their consequences are complex. Refiners can change their operations (e.g., change the flows of crude fractions between different process units) to adapt to changes in product demand and crude quality changes.¹⁰ For instance, if there is a shift towards heavier crude in the market, a refiner could react in a few different ways. One option would be to simply process crude with a lower average API gravity. This would result in reduced naphtha yield from the atmospheric distillation tower (AT) and thereby decrease the volume of gasoline produced. This is fine if there is a corresponding decrease in demand for gasoline. However, if demand for gasoline remains constant and capacity permits, the refiner could change the operating conditions of the AT and vacuum distillation unit (VDU) to direct more atmospheric gas oil (AGO) and vacuum gas oil (VGO) (which tends to be more plentiful in heavier crude oil) to the fluid catalytic cracking (FCC) unit, thus maintaining overall gasoline output and maximize the refinery profit.¹¹ In the case where the demand for transportation fuels, such as gasoline, declines while the demand for other products does not drop proportionally, refiners could direct the naphtha fraction from different process units (e.g., AT, Coker, FCC, etc.) towards alternative uses, such as steam cracking for petrochemical feedstock production.¹² Alternatively, refiners could also lower overall crude consumption and therefore their production to respond to the declining demand. However, lowering the utilization rate could reduce the refinery's efficiency and make it less competitive with other refineries in the market, and could ultimately lead to a complete shutdown.¹³ These potential future conditions and reactions at an individual refinery will not only affect their profitability, they will also change the refinery's overall emissions profile.¹⁰ Beyond operational changes, refineries might also consider new investments to reduce GHG emissions in the face of new and increasingly stringent environmental regulations.¹³ In the case of a stricter carbon pricing scenario, refiners could reduce their carbon footprint by implementing mitigation measures, such as carbon capture and storage (CCS) to lessen the financial impact of carbon pricing on refinery margins. CCS has been widely discussed as a promising emissions mitigation solution despite substantial financial commitment.^14–20 However, the CO₂ formed throughout the refinery is dispersed and exhibits heterogeneity in terms of volume and CO₂ concentration. This results in considerable variability in both emissions reduction potential and the cost of carbon capture.^21,22 Moreover, the decision to invest in carbon capture technology within a refinery is not only influenced by the costs associated with its implementation but also by the uncertainties due to a dynamic fuel market and future climate policy stringency. Climate policies are also subject to frequent and difficult to predict changes driven by government and public opinion, momentum of international negotiations, changing technology landscape, and global progress towards emissions reduction targets.²³

Given the simultaneous challenges of navigating a dynamic fuel market and uncertain carbon policies, refiners face a great deal of risk when faced with emissions reduction decisions, such as implementing CCS which requires a high upfront capital investment. Current literature often examines the decarbonization pathways by considering one or a small number of uncertainties, but typically without fully accounting for their combined influence. For example, Young et al.²⁴ and Sun et al.²⁰ evaluated the GHG emissions reduction for U.S. refining sector under contemporary market conditions (i.e., 2017 & 2023), but did not account for future fuel market dynamics. Similarly, Kim¹⁰ evaluated the role that refinery changes can play in affecting emissions reduction potential in an uncertain future; however, other potential responses, such as refinery shutdowns, in the face of future uncertainties were not considered. Hence, this research bridges the gap in existing literature by considering the influences of multiple factors in a consistent way, including potential shifts in the transportation fuel market as the economy transitions towards decarbonization, timing and stringency of carbon pricing, and their combined effects on U.S. refinery GHG emissions and strategic decision-making.

To fill the above knowledge gap, this study seeks to accomplish three objectives. First, we investigate how uncertainties in future refinery product demand at the regional level (Petroleum Administration for Defense Districts, PADD)—driven by factors such as the rate of decline in gasoline demand due to the increasing penetration of electric vehicles (EVs)—could significantly affect U.S. refinery GHG emissions. Second, we examine the distribution of CO₂ avoidance costs across different process units, refinery configurations, and PADD regions to understand the economic feasibility of carbon capture, as well as its implications for investment prioritization and policy design. Third, we use real option (RO) analysis, investigate the potential impact of selected actions (i.e., continue operating refineries without abating emissions, deploying a carbon capture unit within the refinery, shut down the refinery) that different refinery archetypes (i.e., a “typical” refinery of different types) might take when uncertainties about market conditions and climate policy stringency are considered. This study offers insights into how different refinery types may respond to a number of changing future conditions and uncertainty, helping refiners make informed decisions and highlighting where policy may be needed to achieve outcomes that individual refineries might not pursue on their own.

Methods

PRELIM and carbon capture module

Petroleum Refinery Life Cycle Inventory Model (PRELIM) is an Excel spreadsheet-based life cycle inventory model to evaluate the gate-to-gate environmental impacts of refineries' operations. The model inputs include specifications of crude assays such as the American Petroleum Institute gravity (API gravity) and sulfur content, refinery configurations (i.e., the different combinations of process units) and product slates (i.e., the composition of refinery products). A carbon capture module is integrated within PRELIM to explore both the emissions reduction potential and the economic performance of implementing carbon capture technologies in refineries (SI, Fig. S1).

PRELIM considers a total of 7 different combinations of process units (i.e., configuration) for petroleum refineries (SI, Table S1), with the addition of a carbon capture and a compression & dehydration unit in the carbon capture module. It also accounts for four types of utilities – electricity, heat, steam, and hydrogen – that drive refinery operations. GHG emissions from the refinery operations are estimated by applying emissions factors to the total utilities consumed. The strategy employed in the carbon capture module is post-combustion with a many-to-one setup (i.e., individual absorbers are assigned to each furnace with a common regenerator). Oxy-fuel combustion is also modeled in the carbon capture module because it is anticipated to be a potentially less costly option for FCC units.²⁵ However, it is not utilized in this study because it has not yet been commercially implemented on FCC units. Additionally, employing oxy-fuel combustion for the FCC and post-combustion on other units would reduce the economy-of-scale benefit of the many-to-one model. Steam and electricity consumption, due to absorbent regeneration and fluid circulation, is calculated based on CO₂ concentration in flue gas. The specific steam consumption (i.e., MJ steam per kg CO₂) uses literature-averaged values for various flue gas flows,^26–32 while specific electricity consumption (MJ electricity per kg CO₂) is determined by a literature-based correlation.²⁷ The carbon capture module assumes a multi-stage centrifugal CO₂ compression with inter-stage cooling for dehydration. Detailed information could be found in Method 2 (SI).

Transportation fuel demand and refinery product slate

This study uses the 2019 transportation fuel demand in the U.S. as a baseline, as documented by the Energy Information Agency (EIA).³³ The EIA's production data for refinery products at the regional level³⁴ are used to define the 2019 PADD-level product slates, which are used to set the target output of PRELIM. The model optimizes refinery operations to achieve that product slates in each PADD. Transportation fuel consumption is assumed using the projections from EFS, which estimate demand from 2017 to 2050 under different EV adoption scenarios in each PADD. To illustrate the impact of potential changes in fuel consumption, two specific years – 2035 and 2050 – have been selected as timestamps. To translate the projected transportation fuel consumption in these years to refinery products slates, the proportion of the refinery product that is transportation fuel (see Table S2, SI) as well as the proportion of transportation fuel produced within the PADD (see Table S3, SI) are held at historical averages according to EIA data.^1,35–38 While we know that these proportions will also change in the future, how they will change is highly uncertain and forecasting these dynamics is outside the scope of current work. A detailed procedure for generating refinery product slates is outlined in SI (Method 2).

Crude qualities and crude proxies

The 2019 crude qualities⁵ (i.e., API gravity and sulfur content) for different PADDs as well as the volume of crudes assays imported from different country for individual refineries³⁹ in the U.S. are documented by EIA and thus will be used to blend the crude proxies based on the volume proportion of each crude assay. Future crude quality in each PADD is represented using a set of bounded shifts in API gravity and sulfur content relative to the 2019 baseline.⁵ We define two paired scenarios: light versus heavy crude with respect to API gravity, and sweet versus sour crude with respect to sulfur content, to evaluate potential differences in refining-sector GHG emissions. In the light scenario, API gravity is assumed to be higher than the 2019 baseline, whereas in the heavy scenario it is lower. Similarly, sulfur content is assumed to be lower than the baseline in the sweet scenario and higher in the sour scenario. The magnitude of these deviations increases over time, such that changes in the near term (e.g., 2035) are smaller than those in the longer term (e.g., 2050) (SI, Table S4 and Fig. S2). These stylized ranges are intended to span plausible conditions above and below present-day averages and to provide long-term inputs for scenario analysis, rather than to predict specific crude slates processed in U.S. refineries. The wider range considered for 2050 reflects the longer adjustment period over which larger shifts in crude quality could occur, recognizing that actual future crude composition will depend on supply-side, geopolitical, and resource depletion dynamics not explicitly modeled here. A crude proxy is then blended, using the approach developed by Cooney et al.,⁴⁰ with domestic crudes and the top 10 imported crudes in each PADD to achieve the projected API gravity & sulfur content. A detailed procedure for projecting future crude qualities and blending crude proxies is presented in SI (Method 2).

Refinery throughput

This study applies an approach developed by Kim,¹⁰ which adjusts the configuration level crude throughput to minimize the difference between the multi-linear regression model-derived regional Standardized Refinery Complexity Index (SRCI) and the volume-weighted regional SRCI under multiple constraints. The individual refinery's SRCI is originated from the Nelson Complexity Index (NCI) and is able to rank refineries based on their complexity, considering secondary conversion capacity and unit complexity.^41–43 Similar to NCI, higher SRCIs indicate a refinery's ability to process heavier crude and produce high-value products like gasoline and diesel. The configurational SRCI consolidates individual refineries' SRCI using their volume shares of crude throughput among refineries with the same configuration in each PADD from 1999 to 2019 (SI, Table S5). Regional SRCI is the volume-weighted average of configurational SRCI, which offers a standardized metric for assessing and comparing regional refinery complexity. Using a multi-linear regression model (eqn (1) and (2)), refineries can estimate regional volume-weighted SRCI based on API gravity, gasoline, diesel, and jet yields (data collected from the EIA^5,44 and Oil & Gas Journal⁴⁵). Table S6 (SI) summarizes coefficients for the multi-linear regression model. This allows refineries to project future regional SRCI, helping them assess the complexity required for future crudes and product slates to meet market demands. A detailed procedure to estimate the refinery throughput is presented in SI (Method 2).

SRCI^Target_i = a₁ × API_i + a₂ × gas_i + a₃ × diesel_i + a₄ × jet_i (1)

The objective function thus could be expressed as shown in eqn (3) and is optimized using MATLAB's fmincon solver.

Minimize: f₀(x) = |SRCI^Target_i − SRCI^VWAP_i| (3)

where i denotes the PADD number, j denotes the refinery configuration, a₁, a₂, a₃, a₄ are the regression coefficients, SRCI^Target_i is the multi-linear regression modeled target Standardized Refinery Complexity Index, SRCI^VWAP_i is the regional volume-weighted Standardized Refinery Complexity Index, SRCI_Config,i is the configurational Standardized Refinery Complexity Index, x is the crude throughput for each refinery configuration in each PADD, API is the regional crude input API gravity, gas, diesel, jet are the regional gasoline yields, diesel yields and jet yields respectively.

It is important to note that the SRCI framework is calibrated using historical refinery data reflecting a transportation fuels – oriented operating regime. As such, it summarizes observed historical relationships rather than serving as a predictive tool for fundamentally different future operating conditions. Structural shifts in refinery configuration, product slates, or decarbonization strategies may fall outside the historical range on which the framework is based, and these transitions would not be fully captured without model adaptation and additional data to support validation under emerging operating paradigms. A detailed explanation is presented in SI (Method 2).

CO₂ avoidance cost

The CO₂ avoidance cost, defined as the ratio of the annualized cost, including capital and operating expenditures, of implementing carbon capture, transport and storage to the annual amount of CO₂ avoided (the CO₂ stored that would have ended up in the atmosphere without CCS), allows refiners contemplating investments in mitigation (such as CCS) to make comparisons between options. The method of avoiding the cost of carbon capture is adopted from Gale et al.²¹ and is shown in eqn (4). Since the new-built refinery with carbon capture is not considered in this study, the selection of eqn (4) is thus rational because it is suited for retrofitting carbon capture in existing facilities, according to Roussanaly et al.²²where CC denotes the scenario with carbon capture deployed on certain process units, NoCC denotes the scenarios with no carbon capture deployed on certain process units, CapEx refers to the capital expenditure of the carbon capture facility, OpEx refers to the operating expenditure, µ is the discount rate, n is the lifespan of the carbon capture facility, GHG refers to the GHG emissions of the refinery.

Emissions from refineries, with or without carbon capture, are calculated using PRELIM and its integrated carbon capture module. In this study, we evaluated a total of 35 carbon capture scenarios on different process units for each configuration in different PADDs (SI, Method 1). This module also enables the estimation of annualized costs through a bottom-up analysis that considers both CapEx and OpEx of the carbon capture unit. CapEx estimation includes the costs of essential equipment, construction, engineering, procurement, and construction (EPC) service costs, as well as others (e.g., owner's cost, spare parts cost, etc.). Some assumed parameters for estimating CapEx can be found in Table S7 (SI). Notably, all costs are adjusted to the year 2019.

OpEx encompasses annual fixed and variable operation and maintenance (O&M) costs and assumed parameters can be found in Table S8 (SI). Notably, this calculation of OpEx does not consider the potential cost implications of a carbon price, nor does it consider the sales of captured CO₂ as a commodity.

The costs associated with CO₂ transportation and storage are included in this study. However, a rather simple approach is taken due to the lack of scenario design for future CO₂ infrastructure and storage sites. Such a design would necessitate a rigorous analysis of future CO₂ transportation pipeline and storage site networks, which is beyond the scope of this study. Instead, we assume a preliminary case study, where CO₂ transport and storage costs are parameterized using a representative case and applied uniformly across regions to provide order-of-magnitude estimates for long-term strategic analysis. Specifically, we use the Scalable Infrastructure Model for Carbon Capture and Storage (SimCCS)⁴⁶ platform to generate an approximate cost for CO₂ transportation and storage in the Southeastern U.S. The estimated costs for CO₂ transportation were $6.31 per t for capturing 100 MtCO₂ per year. Similarly, the estimated costs for CO₂ storage were $4.37 per t for capturing 100 MtCO₂ per year. The potential estimated CO₂ leakage rate could range between 11.3–173 kt CO₂ per year, based on parameters (e.g., failure rate, leakage occurrences, pipeline length, etc.) documented by Onyebuchi et al.⁴⁷ and Jensen et al.⁴⁸ In this study, we estimate an average of a total of 75.6 kt CO₂ per year, which accounts for 0.0756% of total CO₂ storage rate (i.e., 100 MtCO₂ per year).

A detailed description of the procedure for estimating the CO₂ avoidance cost is summarized in SI (Method 2).

Real option analysis

While the CO₂ avoidance cost indicates the minimum carbon price needed to justify deploying carbon capture, it does not fully capture the complexities and uncertainties in market conditions and carbon pricing that refineries face. In contrast, RO analysis is able to account for the managerial flexibility required to adapt and revise later decisions in response to uncertain market developments,⁴⁹ thus being a valuable tool for informing investment decisions for refiners. This study applies a binomial lattice tree model^50,51 to evaluate the potential strategic options the refiner might consider when facing an uncertain carbon price. Three individual refinery archetypes with different configurations, which are hydroskimming, “medium conversion: FCC & gas oil hydrocracker (GOHC)” and “deep conversion: FCC & GOHC” refineries are considered in the RO analysis. These refinery archetypes could choose between three strategic options, which are (1) to continue operating the refinery without emissions abatement (i.e., “continue operation”); (2) deploy carbon capture in one or multiple process units (i.e., “deploy carbon capture”); and (3) shut down the refinery entirely (i.e., “shutdown refinery”). These options are evaluated over a 31-year timeline, providing refiners the opportunity to make decisions at each annual interval from 2019 to 2050, consistent with the flexibility of an American option. It should also be noted that PRELIM specifically models the crude throughput, product slates, and GHG emissions for selected benchmark years – 2019, 2035, and 2050. For the intervening years not directly modeled by PRELIM, it is assumed that changes in crude throughput, product slates, and GHG emissions occur linearly on a yearly basis. The underlying asset value of each refinery will be determined using the present value of future discounted cash flow (DCF), which will be initially determined based on projected prices of crude oil, refinery products, and utilities, as forecasted in the Annual Energy Outlook (AEO) by the EIA⁵² and discounted by a rate of 12%. Carbon prices are assumed to follow a stochastic process (geometric Brownian motion, GBM). The uncertainty associated with prices in this model is handled by adopting a hybrid approach in which carbon prices are modeled stochastically while other uncertain variables (i.e., crude, fuel and utility prices) are examined through scenario analysis. This structure isolates the option value associated with carbon price uncertainty (the focus of this study), while avoiding the need to specify a fully parameterized joint stochastic process for multiple price variables that are strongly correlated in practice and could otherwise lead to overstated uncertainty if modeled independently. For a specified carbon price trajectory, a single binomial lattice is constructed and the expected present value at each node is determined using backward induction.⁵¹ The optimal response is selected as the option that achieves the highest expected present value among the three possible responses (see Method 3, SI). The expected present value of the shutdown response is set to 0 and is chosen if the expected present values for both continuing operations and deploying carbon capture are negative. Risk-neutral probability and backward induction are utilized to calculate each node's expected present value and corresponding options, where the risk-free interest rate is set as 2.40%, the average U.S. 20-year treasury bond rate in 2019.⁵³ A 12% discount rate is used only to compute the initial underlying asset value as the present value of future cash flows, representing the company's perspective on the project's economics based on the perfect foresight of future price projections. This risk-adjusted discount rate is not applied during the backward induction stage of the real-options analysis, because the project's risk profile becomes endogenous to the investment timing decision and may differ across options. A detailed explanation of this rationale is provided in Method 3 in the SI. Monte Carlo simulation is then utilized for each lattice for 10 000 runs to illustrate the likelihood (frequency) of choosing different options that should be taken in different years. The Real Options model and Monte Carlo simulations are implemented in Microsoft Excel using Visual Basic for Applications (VBA). We also construct two alternative stepwise carbon price trajectories – one with earlier increases and one with delayed increases – to assess the effect of timing on refinery decision-making under uncertainty. Detailed information on implementing the RO analysis is presented in Method 3 in SI. Detailed information on each individual refinery is provided in Table S9 in SI.

Results and discussion

U.S. refining sector emissions under different carbon capture and transportation fuel demand scenarios

Fig. 1 shows possible changes in GHG emissions from the U.S. refining sector under various scenarios for future transportation fuel demand, market shifts in crude quality and application of carbon capture technologies. NREL's EFS are used to explore the impact of EV adoption in the U.S. on fuel demand and in turn, the impact on refinery GHG emissions. For example, the high electrification scenario in EFS⁴ (referred to in this study as an LD or low petroleum fuel demand) projected a decline in U.S. transportation fuel demand from 5.54 to 3.47 Giga barrels per year (Gbbl per year) from 2019 to 2035 and 2.10 Gbbl per year by 2050. We hereby categorize future fuel demands projected by EFS as LD, MD (i.e., medium petroleum fuel demand) and HD (i.e., high petroleum fuel demand) scenarios for different years (i.e., 2035 and 2050), as shown in Table 1. The two black-dashed vertical lines in Fig. 1 show the bounds of possible demand in 2035 and 2060 across all the EFS cases. The 2019 fuel demand estimated by the EIA is used as the baseline petroleum fuel demand (BL) and is indicated by the red-dashed vertical line in Fig. 1. All scenarios and their corresponding fuel demands are summarized in Table 1, with further details on the EFS provided in Table S10 (SI).

Fig. 1 Estimated annual GHG emissions from U.S. refineries at different transportation fuel demand with and without carbon capture. No carbon capture refers to the case that no carbon capture unit is deployed to any processes in the refinery. Minimum carbon capture implementation refers to the case where monoethanolamine-based (MEA-based) post-combustion carbon capture units are deployed on all FCC regenerators in U.S. refineries. Maximum carbon capture refers to the case where MEA-based post-combustion carbon capture units are deployed on all process furnaces/boilers as well as FCC regenerators in U.S. refineries. Red dashed vertical line refers to the volume (i.e., 5.54 Gbbl per year) of transportation fuel (i.e., gasoline, jet fuel, diesel, fuel oil) supplied to U.S. market in 2019. X-Axis values correspond to those dots within the two vertical black dashed line are estimated volume (2.10 to 4.71 Gbbl per year) of transportation fuel demand using NREL's EFS data.

Table 1 Transportation fuel demand scenarios considered in the study

Year	Fuel demand scenarios	Scenario names	Proportion of EV in U.S. vehicle fleet, %	Estimated U.S. liquid transportation fuel demand, Gbbl per year	Estimated total U.S. refinery production, Gbbl per year	Data source
2019	Baseline petroleum fuel demand	BL	0.503	5.54	7.20	EIA^1
2035	High petroleum fuel demand	HD35	10.9	4.71	6.80	EFS^4
2050	High petroleum fuel demand	HD50	13.6	4.58	6.87	EFS^4
2035	Medium petroleum fuel demand	MD35	42.3	3.74	5.76	EFS^4
2050	Medium petroleum fuel demand	MD50	64.4	3.04	5.13	EFS^4
2035	Low petroleum fuel demand	LD35	46.7	3.47	5.32	EFS^4
2050	Low petroleum fuel demand	LD50	82.3	2.10	3.69	EFS^4

---

Changes in fuel demand will directly impact refinery production levels and, consequently, crude input requirements (see Methods). For example, under the LD scenario, total annual production from U.S. refineries is estimated to decline from 7.20 Gbbl per year in 2019 to 5.32 Gbbl per year in 2035 and further to 3.69 Gbbl per year by 2050 (SI, Fig. S3). Correspondingly, total crude input is expected to decrease from 16.8 million barrels per day (MMbbl per day) in 2019 to 13.6 MMbbl per day in 2035 and 12.2 MMbbl per day in 2050 (SI, Fig. S4).

We then use PRELIM to estimate the GHG emissions for each combination of transportation fuel demand and crude oil input. The exact date that a level of demand is realized in these scenarios is not the focus of this study. Instead, Fig. 1 presents the relationship between fuel demand, GHG emissions based on crude input consumed, and the impact of possible carbon capture deployment options for the refining sector. For the LD scenario, as the transportation fuel demand decreases from 5.54 Gbbl per year to 3.47 Gbbl per year and then further declines to 2.10 Gbbl per year, U.S. refinery GHG emissions are projected to decrease by 24.5%, from 212 to 158 million metric tons of CO₂ equivalent per year (MtCO₂eq per year) between 2019 and 2035, followed by a further decline to 111 MtCO₂eq per year by 2050. Implementing CCS could further decrease GHG emissions from 111 MtCO₂eq per year to 47.6 MtCO₂eq per year (i.e., a reduction of 57.3% or 63.9 MtCO₂eq per year), assuming maximum adoption of CCS. It should be noted that maximum carbon capture refers to the case where post-combustion carbon capture is deployed on all process furnaces/boilers as well as the FCC regenerator within the refinery, while minimum carbon capture refers only to deploying post-combustion carbon capture on the FCC regenerator. It is important to note that the absorption unit for steam methane reforming (SMR) is assumed to be placed after the furnace, where natural gas and tail gas from the hydrogen purification unit are mixed and combusted. In HD35 and HD50 scenarios, fuel demand is estimated to decline from 5.54 Gbbl per year to 4.71 Gbbl per year and further decrease to 4.58 Gbbl per year. Despite this downward trend, refinery GHG emissions first decrease by 4.50 MtCO₂eq per year and then increase by 2.64 MtCO₂eq per year, driven by a slight rise in refinery production from 6.80 Gbbl per year to 6.87 Gbbl per year (Fig. S3, SI). This apparent discrepancy is primarily due to differing rates of change in product demand and production across regions between HD35 and HD50 (Table S11, SI). Further details are provided in Result 1 (SI). Moreover, implementing maximum carbon capture could reduce GHG emissions by 120–122 MtCO₂eq per year, representing a 56.6–57.4% reduction compared to the BL scenario. In contrast, in LD35 and LD50 scenarios, maximum carbon capture, combined with reduced production, results in an annual emissions decrease of 145–164 MtCO₂eq per year, achieving a 68.4–77.6% reduction compared to the BL scenario. These results demonstrate that carbon capture can reduce GHG emissions even with lower refinery production. As depicted in Fig. S5 (SI), the deployment of carbon capture technologies could ensure that the refining sector stays on an emissions reduction target trajectory until the mid-2030s, regardless of changes in fuel demand. However, beyond this time period, additional measures would be needed to achieve zero emissions for the refining sector and a net-zero target for the entire supply chain. This is due to the fact even with decreasing demand for liquid transportation fuels and deployment of carbon capture, a remaining 47.6–67.0 MtCO₂eq per year of GHG emissions would still need to be addressed. Such additional measures could include transitioning to biomass feedstocks, utilizing low-carbon electricity, adopting green/pink hydrogen.^10,20 Simplifying assumptions such as holding the local production share of fuel consumed in a PADD at historic averages and limiting the variation in future crude quality (using only “high” and “low” bounding cases) limit the resulting variability of the GHG emissions from the U.S. refining sector predicted by the model. Because the model does not endogenously reallocate production across regions, it may underrepresent the full scope of future product mix adjustments. However, these assumptions enable us to isolate and evaluate the dominant driver of emissions change (i.e., declining fuel demand), while maintaining transparency in regional accounting. Similarly, our simplified assumptions on the potential range of crude qualities have a secondary influence on aggregate emissions outcomes: the difference in annual U.S. refining-sector GHG emissions between heavy/sour and light/sweet crude scenarios in 2050 is approximately 6 MtCO₂eq per year, corresponding to about 5.4% of total refining-sector emissions (SI, Fig. S6).

Lastly, even if the refinery emissions were reduced to zero, any emissions from upstream and combustion of liquid fuels (e.g., in internal combustion engines) will result in emissions that also need to be mitigated to achieve “net zero” across this supply chain. Although the LD50 scenario (i.e., 2.10 Gbbl per year) and aggressive deployment of carbon capture can reduce the sectoral emissions by 77.6%, it still results in 47.6 MtCO₂eq per year of emissions from the refining stage, 109 MtCO₂eq per year of emissions from the crude upstream activities (i.e., extraction and transportation) and 827 MtCO₂eq per year of emissions when these fuels are combusted (SI, Fig. S7). The decrease in GHG emissions in downstream fuel combustion is primarily driven by the reduction in GHG emissions from gasoline combustion. In the BL scenario, gasoline accounted for 56.8% of total fuel combustion GHG emissions, contributing 1184 MtCO₂eq per year out of a total of 2087 MtCO₂eq per year. In the LD50 scenario, gasoline contributes only 31.3% of total fuel combustion emissions, amounting to 259 MtCO₂eq per year out of a total of 827 MtCO₂eq per year (SI, Fig. S8). Note that PRELIM does not have an emissions factor inventory for the direct combustion of refinery-derived fuels, nor does it have emissions factor inventory for crude upstream activities. Hence, we adopted those emissions factors for downstream fuel combustion from the EIA⁵⁴ and emissions factor for crude upstream activities from Sun et al.²⁰ Consequently, further exploration of reductions across the supply chain is needed to achieve the societal net-zero targets.

Emissions reduction potential and CO₂ avoidance cost of carbon capture for U.S. refineries

Fig. 2 shows that the CO₂ avoidance cost associated with CO₂ capture from 8 different individual/grouped (i.e., AT & VDU, CNR, FCC, Coker, hydrotreaters, SMR, steam boiler and other units) process units in U.S. refineries varies by more than an order of magnitude (from $83.5 per t to exceeding $1360 per t CO₂ avoided), under an MD50 scenario when the crude proxy (i.e., a mix of different crudes from different oil fields) with low API gravity and high sulfur content (i.e., heavy/sour crude proxy) is processed. The cost would be different if the crude proxy had high API gravity and low sulfur content (i.e., light/sweet crude proxy), as less energy is generally required for refineries to process such crude compared to a heavy/sour crude proxy. Comparing these costs to current U.S. carbon prices (e.g., $75.0 per t CO₂ in 2025 from California's Low Carbon Fuel Standard), it is unlikely to see carbon capture deployed on any of these units as the current price is not only insufficient but also highly volatile, reducing the incentive for long-term investment. As shown in Fig. 2A and B, if carbon prices were to increase to $100 per t and $150 per t, 11.6 MtCO₂eq per year (12.3%) and 53.0 MtCO₂eq per year (56.4%) of the total potential avoided emissions (94.0 MtCO₂eq per year), respectively, could be achieved. A carbon price of $400 per t is required to unlock 93.8 MtCO₂eq per year of emissions reductions, representing 99.8% of the total mitigation potential. For comparison, Canada's federal carbon price is scheduled to reach CAD 170 per t CO₂ by 2030 (∼122 $ per t CO₂).⁵⁵ At this level, the model indicates that FCC units at medium and deep conversion refineries and SMR units at deep conversion refineries in PADD 3 would deploy CCS, achieving 16.6 MtCO₂eq per year (17.6%) of emissions reductions. These are associated with very high-cost carbon capture deployment cases within hydroskimming refineries, such as steam boilers.

Fig. 2 (A) CO₂ avoidance cost curve of carbon capture and storage in different process units in U.S. refineries under MD50 scenario with heavy/sour crude proxies input. (B) CO₂ avoidance cost curve of carbon capture in different PADD regions in U.S. refineries under MD50 scenario with heavy/sour crude proxies input. (C) Distribution of CO₂ avoidance cost of carbon capture in different refinery configurations grouped by PADD regions. Each rectangle in (A) and (B) represents an individual/grouped process unit in different refinery configurations in different PADD regions. X-Axis for (A) and (B) are the cumulative annual avoided CO₂ by mass (MtCO₂eq per year). X-Axis for (C) represents different refinery configurations from 0 to 6 based on PRELIM classification (SI, Table S1). Y-Axis upper limits are set to be 250, 250 and 400 $ per tCO₂ respectively. The detailed CO₂ avoidance cost under different fuel demand and crude quality scenarios is presented in Dataset S1 in SI. ADU/VDU – atmospheric distillation unit & vacuum distillation unit. CNR – catalytic naphtha reformer. FCC/RFCC – fluid catalytic cracker & residue fluid catalytic cracker. Hydrotreaters – naphtha hydrotreater (NHT), kerosene hydrotreater (KHT), diesel hydrotreater (DHT), FCC post hydrotreater (FCC PHT), Coker naphtha hydrotreater (Coker NHT) & residue hydrotreater (RHT). SMR – steam methane reforming.

From a process unit perspective, Fig. 2A suggests that FCC units should be prioritized for deploying carbon capture in U.S. refineries due to the higher emissions reduction potential and lower avoidance cost. Specifically, applying carbon capture to FCC units can achieve a reduction potential of 18.5 MtCO₂eq per year, accounting for 8.73% of the total U.S. refining emissions, while also attaining an avoidance cost ranging from $83.5 per t to $242 per t under the MD50 scenario. FCC units facilitate the conversion of heavy components of crude oil into valuable lighter products such as transportation fuels.⁵⁶ They are identified as one of the most emissions-intensive units, with an estimated volume-weighted average emissions intensity of 31.6 kg CO₂eq per bbl of crude fraction processed in the U.S., as shown in Table S12, SI. Such a high emissions intensity results from the combustion of deposited coke on catalysts, which generates the necessary heat to operate the unit at high temperatures (480 to 540 °C) as well as substantial electricity consumption.⁵⁶ Plus, refineries equipped with FCCs are the dominant GHG emissions contributors (i.e., 84.6% to 93.4%) in the U.S. from 2019 to 2050 (SI, Table S13). In addition, the cost-effectiveness of carbon capture on FCC units arises from their higher CO₂ concentration (i.e., 16.9 vol%) in the exhaust flue gas, compared to flue gas from other process units (i.e., 6–12.5 vol%) (SI, Table S14). This higher concentration allows for the use of a smaller absorber and thus lower energy use, leading to reductions in both capital and operation & maintenance costs.

Notably, refiners could achieve lower CO₂ avoidance costs by pooling flue gases from multiple process units into a single absorber and regenerator. This approach would mix gas streams from multiple process units of varying concentrations (from less than 13 to 17 vol%), increasing the total amount of CO₂ captured. For instance, in a deep conversion refinery in PADD 3, the capture of CO₂ from both FCC and SMR units yields an avoidance cost of $91.2 per t, with a total avoided CO₂ of 10.1 MtCO₂eq per year. Although this cost is higher than that from capturing solely from FCC units ($83.5 per t with 6.22 MtCO₂eq per year avoided), it remains lower than the cost of capturing exclusively from SMR units, which stands at $107 per t with 3.79 MtCO₂eq per year avoided, while achieving a higher total avoided CO₂ compared to capturing from either unit alone, as detailed in Dataset S2, SI.

We estimate avoidance costs from SMR units across the U.S. span a wide range depending on the stream: $82.8–811 per t for PSA inlet, $84.6–894 per t for PSA outlet, and $107–853 per t for furnace outlet. This is generally higher than costs reported previously (19.8 to 153$ per t CO₂ avoided^21,57–65) for two reasons: (1) we consider the transportation & storage cost for CCS as well as the leakage rate of CO₂ storage (this adds ∼11$ per t in avoidance costs, see Method); (2) we consider the cost for the interconnection pipelines for carbon capture, which are often excluded from literature estimates and could take up to 52% of total capital expenditure (CapEx) under high CO₂ volume flows. It is also notable that our SMR avoidance cost estimates vary widely across refinery configurations due to differences in hydrogen demand and CO₂ flow rates. In refineries with low H₂ use, such as hydroskimming or FCC-only configurations, poor economies of scale result in very high capture costs (e.g., up to $2800 per t), while hydrogen-intensive sites like deep conversion refineries with hydrocrackers consistently show SMR as one of the lowest-cost options. A detailed explanation of these costing results is presented in Result 2 (SI). This variation is not typically considered in previous estimates where the ideal applications of carbon capture are typically modeled (e.g., capturing CO₂ from streams with higher concentration or partial pressure).

Fig. 2B also suggests substantial regional and configuration-level heterogeneity in CCS avoidance costs. Specifically, deploying carbon capture in deep conversion refineries (configuration 6) in PADD 3 could achieve a lower avoidance cost relative to others (i.e., an average of $129 per t for configuration 6 vs. $169 per t for other configurations and $327 per t for other regions) due in large part due to the large volume of CO₂ avoided reducing the marginal cost of capital expenditure. These refineries (in PADD 3 alone) are estimated to account for 28.0% of the total crude processed in the U.S., leading to 30.1% of total GHG emissions under an MD50 scenario. This makes them the largest contributor to GHG emissions compared to refineries in other configurations. PADD 3 is also estimated to have the lowest construction cost compared to other regions, owing to the lower construction cost location factor (SI, Table S15) in the U.S. Gulf Coast region. Note that the cost of CO₂ transportation via pipeline and storage has also been considered in Fig. 2. For the scenario involving 100 MtCO₂ per year, the estimated unit costs are $4.37 per t for storage and $6.31 per t for transportation as mentioned in Methods. Past studies show the cost for CO₂ transportation and storage could vary in different regions, ranging from 10–22 $ per t.⁶⁶ A sensitivity analysis was conducted to evaluate the impact of this variability on the overall avoidance cost (SI, Result 3). The results indicate that CO₂ transportation costs via pipeline are not a primary driver of avoidance cost relative to capture costs, although they remain non-negligible. Accordingly, while capture costs dominate, refineries should still account for location-specific transport and storage costs, alongside carbon capture and carbon pricing, when making investment decisions.

Likely reactions of different refinery archetypes under uncertain future market and carbon pricing conditions

In the face of uncertain future regional transportation fuel markets and carbon pricing conditions, refineries will react in different ways (e.g., continue business as usual operations, shut down, invest in GHGs mitigation technologies). Fig. 3A illustrates that under various uncertain conditions – including fluctuating prices of crude oil, refinery products, utilities, and carbon dioxide, as well as changing fuel demand, medium and deep conversion refineries are unlikely to shut down (see Method 3, SI). For example, the likelihood of a typical medium conversion refinery being shut down under a medium fuel demand scenario (i.e., BL to MD35/50) is only 14.6% by 2042, indicating that a medium refinery is resilient to market fluctuations and is likely to remain economically viable within the modeled time horizon. A hydroskimming refinery archetype, however, is more sensitive to changing conditions, exhibiting a modeled shutdown probability exceeding 97% by around 2031 under all fuel demand scenarios (99.9%, 97.1%, 99.6% for low, medium and high fuel demand scenarios, respectively). The probability of a hydroskimming refinery shutting down escalates from 0% in 2027 to 21% by 2028 and is expected to reach 99.9% by 2029 under a scenario of low fuel demand (i.e., BL to LD35/50). These trends indicate that the expected present values (Dataset S3, SI) for the refinery will likely turn negative from 2030 onward under assumed carbon price trajectories, underscoring its economic vulnerability and increased likelihood of becoming unprofitable sooner than other refinery types.

Fig. 3 (A) Likelihood of three choices using real options analysis for three refinery types under three transportation fuel demand scenarios. (B) simulated carbon price trajectories assumed to follow stochastic process. The low/medium/high fuel demand scenarios refer to the fuel demand transition from the BL scenario to LD35/50, MD35/50 and HD35/50 scenarios, respectively. Carbon prices are assumed to follow a stochastic process (GBM), and could range from $27.8 per t to $584 per t. A single binomial lattice is generated based on a specific carbon price trajectory. A Monte Carlo simulation is then employed, sampling various carbon price trajectories that adhere to the GBM process to create a set of binomial lattices (i.e., 10 000 runs). The likelihood of each response at a specific time is calculated by dividing the number of nodes executing the response by the total number of nodes at that time (see Method 3, SI). Hydroskimming refers to hydroskimming configuration (0). Medium conversion refers to medium conversion: FCC & GO-HC (3). Deep conversion refers to deep conversion: FCC & GO-HC (6).

This analysis also shows that the expected present value maximizing choice is likely to be for medium and deep conversion refineries to adopt CCS today to allow them to continue operations into the 2040s, if an increasing carbon price is expected. Moreover, the modeled results indicate that, under the assumed carbon-price trajectories, a typical medium conversion refinery exhibits stronger near-term economic incentives for CCS adoption relative to other refinery archetypes. For example, the likelihood of a medium conversion refinery deploying carbon capture technology is estimated to increase to 50% by 2022 and then peak at 89.2% around 2032 under a low fuel demand scenario which is attributed to a more favorable expected present value, influenced by carbon pricing, from 2022 to 2032. This suggests that carbon capture remains a competitive option for refineries under certain carbon price trajectories, where the financial incentives outweigh the costs associated with the technology. The real-options results suggest a relatively narrow economic window for CCS investment under the assumed carbon-price process and price projections used in this study, particularly during the early periods of increasing carbon prices. Under these conditions, the window for CCS deployment is wider for the medium-conversion refinery archetype than for deep-conversion refinery archetype. For example, a deep conversion refinery under a medium fuel demand scenario is likely to find CCS increasingly appealing from the early 2020s to 2030, with the likelihood of adoption rising from 50.0% in 2022 to a peak of 84.2% by 2030. However, this interest is projected to decline rapidly, with the probability of deploying CCS dropping sharply to 10.7% by 2032. In contrast, under the medium fuel demand scenario, a typical medium conversion refinery would find CCS increasingly appealing from the early 2020s to the early 2040s, as the probability of adoption climbs from 50.0% in 2022 to 96.5% by 2041. Overtime, due to evolving market and regulatory pressures, interest in investing in CCS is expected to gradually decline, while the likelihood of shutting down the refinery increases. Specifically, the probability of deploying CCS decreases to 12.6% by 2049, while the probability of refinery shutdown rises to 87.4% during the same period. It is important to note that this analysis only considered three possible actions. Other technologies and strategies could become profitable options in place of shutting the refinery down.

Additionally, a sensitivity analysis is conducted to assess how the likelihood of the three actions for a medium conversion refinery varied under a medium fuel demand scenario (SI, Fig. S9). This sensitivity analysis varied five key variables: risk-free interest rate, crack spread (i.e., the margin a refiner realizes while procuring crude oil and simultaneously selling the products),⁶⁷ carbon prices, CCS utility costs, and CCS capital costs, each considered under high, reference, and low scenarios. The risk-free interest rate ranges from 1% to 7%, based on historical U.S. Treasury bond rates. Crack spread scenarios are derived from the AEO's low, reference, and high oil price cases to reflect fuel price dynamics.⁵² Carbon prices are varied by ±50% from the baseline trajectory used in the main RO analysis, while CCS utility and capital costs are also adjusted by ±50% to represent potential technological advances or setbacks. The specific variations applied to each variable are summarized in Table S16 (SI).

Among these factors, crack spread has the largest impact on the likelihood of shutting down a medium conversion refinery as shown in Fig. S9 (SI). Under the low crack spread scenario, the medium conversion refinery archetype is likely to consider closure as early as the early 2030s, with the probability of shutting down rising sharply from 11.3% in 2028 to 96.2% by 2030. In contrast, under the high crack spread scenario, the refinery is more likely to consider CCS deployment instead, as the likelihood of shutdown remains negligible – 0.03% by 2045 – before rising modestly to 19.8% by 2050. These results suggest that successfully reducing transportation fuel demand (e.g., through widespread EV adoption) could shift market conditions toward a lower crack spread, thereby increasing the likelihood of refinery closure. Conversely, a high crack spread could prolong the operation of medium conversion refineries and reduce the incentive to deploy CCS. While our model does not directly capture the complexities of market dynamics (i.e., correlating the EV demand to crude and fuel prices), this relatively simple model still identifies a potential risk of unintended outcomes – such as delayed decarbonization – when fuel demand reduction is paired with complementary policy measures (e.g., crude supply interventions, refining subsidies). Carbon pricing is another key driver influencing a refiner's decisions. A high carbon price encourages earlier CCS deployment in the medium conversion refinery archetype, with a 33.3% probability of adoption emerging in the early 2020s, compared to 50.0% under the reference case. However, if CCS investment is delayed beyond the mid-2040s, the economic pressure from high carbon prices increases the risk of shutdown, with the probability exceeding 50% during that period. Conversely, a low carbon price reduces the incentive to deploy CCS – dropping the probability to just 0.305% by 2030 – and prolongs refinery operations, keeping shutdown likelihood as low as 0.140% by the mid-2040s. Still, even with low carbon prices, refinery shutdowns may occur after 2050, driven by deteriorating market conditions such as declining fuel demand and narrower crack spreads. The results for the two additional carbon price trajectory scenarios are presented in SI Result 4.

Discussion

In this work, we evaluate how transitions in transportation fuel demand and the quality of crude mix consumed in refineries in the U.S. can impact refinery operations and sectoral GHG emissions. We estimate that ∼212 MtCO₂eq per year of GHG emissions (i.e., including natural gas extraction and power generation emissions generated by the utilities consumed by refineries) can be attributed to the U.S. refining sector in 2019. If carbon capture is employed extensively, over 126 MtCO₂eq per year (i.e., 59.1% of reduction) of GHG emissions can be mitigated. As a comparison, Sun et al.²⁰ reported emissions of 196 MtCO₂eq per year from U.S. refineries in 2019 and argued that CCS could reduce 67.4–90% of on-site CO₂ emissions. Yao et al.¹⁸ compared two carbon capture strategies in the U.S. refining sector: (1) post-combustion with amine scrubbing and (2) pre-combustion coupled with oxy-fuel combustion capture. They reported a total reduction potential of 100 MtCO₂eq per year in 2014, with pre-combustion coupled with oxy-fuel combustion being more cost-effective compared to post-combustion carbon capture. Our findings on total refining sector GHG emissions and reductions achievable through CCS are consistent with these estimates, but also reveal the strong interaction between transportation fuel demand and refining sector GHG emissions. Specifically, demand contraction alone can reduce emissions by 48% (to 111 MtCO₂eq per year), while full deployment of carbon capture can further decrease emissions to 47.6 MtCO₂eq per year, equivalent to a 57.3% reduction. It should also be noted that the premises of the decreasing petroleum fuel is the increasing electrification of the vehicle fleet. However, the future EV demand is highly uncertain due to policy uncertainty⁶⁸ and critical mineral supply chain constraints.⁶⁹

On the other hand, we conclude that deploying carbon capture in FCC units could be economically attractive to refiners as it could potentially avoiding GHG emissions by 18.5 MtCO₂eq per year, accounting for 8.73% of total U.S. refineries' GHG emissions, while also attaining a cheaper avoidance cost ranging from $83.5 per t to $242 per t under an MD50 scenario compared to other process units. Previous studies by van Straelen et al.¹⁴ and Gale et al.²¹ examined the complexity of deploying carbon capture in refineries as carbon sources scattered throughout the refinery and the estimated CO₂ avoidance cost vary widely. They concluded that refinery process units with medium to high CO₂ concentrations had the most favorable avoidance cost, while capturing from small emitters with low concentrations was less economically effective, which our study agree. We also conclude that capturing CO₂ from pooled flue gases with different CO₂ concentrations from multiple process units could be attractive to refiners. We find that capturing CO₂ from all furnaces/boilers along with the FCC regenerator is achievable at $125 per t, which is close to the estimated cost by Gale et al.²¹ at around $160–210 per t CO₂. However, we find that our estimated avoidance costs for SMR (i.e., $107–853 per t) are higher and span a wider range than those reported in previous studies (i.e., 19.8–153$ per t).^21,57–65 The higher costs primarily reflect differences in assumptions regarding cost for CO₂ transportation and storage, interconnection infrastructure, and capture location, while the wider range arises from variation in refinery configurations and regions—an aspect not examined in earlier analyses.

We are also the first, to our knowledge, to employ RO analysis to examine the strategic decisions of continuing operation, deploying carbon capture and shutting down that could be made by U.S. refineries. RO analysis extends financial option pricing methods to evaluate investment decisions under uncertainty, enabling the valuation of flexibility and strategic choices in managing real assets.^50,70,71 Multiple studies have also been conducted using RO in the past to investigate the condition and timing or optimal low-carbon technology investment in the energy sector.^72–75 Reinelt et al.⁷² developed a stochastic dynamic programming model to evaluate strategies for aging pulverized coal plants under uncertain carbon regulations, including continuing operations or building new facilities with advanced technologies (e.g., natural gas combined cycle plant, or integrated gasification combined cycle plant), with or without CCS. Their results showed that a preferred market (i.e., low natural gas prices) with higher carbon taxes favors investing in carbon mitigating technologies. McKeller et al.⁷³ used a quadrinomial tree model and stochastic dynamic programming to evaluate nine alternative uses for Alberta oil sands coke, including hydrogen production and electricity generation, with or without CCS. Their findings showed that higher carbon taxes shift preference from sending the coke overseas for electricity generation without mitigating technologies to having the coke consumed by the nearest market with the application of CCS, which offers superior financial and emissions performance under high carbon pricing scenarios. Similarly, our study concludes that, under the modeled crack spread and carbon price scenarios, favorable market conditions (i.e., high crack spread), with higher carbon costs, would encourage the deployment of CCS. However, we also find that certain refinery configurations, such as hydroskimming, are highly sensitive to market fluctuations and carbon pricing policies, making them more exposed to the risk of shutdown by the early 2030s, while medium- and deep-conversion refineries are more likely to consider deploying carbon capture beginning in the early 2020s and potentially continuing through the 2040s under the assumed transportation fuel demand scenarios and modeled carbon price trajectories. Notably, the modeled results indicate that the window of opportunity for taking action is relatively narrow: the probability of refinery shutdown begins to increase markedly by the 2040s under these assumptions.

Future work opportunities

This study relies on a number of simplifying assumptions that facilitate long-term scenario analysis but limit the scope of the results and point to several avenues for future research. Namely, the analysis is intended to examine refinery-level emissions mitigation potential and strategic investment responses under stylized demand, technology, and policy scenarios, rather than to predict endogenous market outcomes such as inter-regional trade reallocation, equilibrium fuel prices, or strategic interactions among refineries. The simplified assumption of fixed historical local production shares could be relaxed in future work by integrating an explicit multi-region partial equilibrium framework that endogenizes trade and product allocation decisions, allowing for exploration of variability in product slates and associated GHG emissions under alternative market and policy conditions. It should be noted, however, that this will come at the expense of the detailed technology representation used in the current work. The simplified representation of future crude qualities could be extended by incorporating supply-driven constraints and alternative crude-availability scenarios that reflect potential shifts in aggregate API gravity and sulfur content due to resource depletion, geopolitical disruptions, or changes in global production portfolios, thereby allowing the framework to explore how different crude-quality pathways influence refining-sector emissions under alternative market and policy conditions. Region-specific CCS network modeling could be incorporated to represent spatial variation in transport distances, storage availability, and infrastructure build-out, thereby enabling the framework to examine how regional differences in CO₂ transport and storage systems influence deployment feasibility and total mitigation costs. Price simulation could be extended to incorporate multivariate stochastic processes or explicitly modeled correlations among carbon prices, fuel prices, and other economic variables, enabling the framework to examine how joint uncertainty shapes investment timing and risk exposure. Finally, the analysis is based on stylized refinery archetypes and scenario assumptions, and results should be interpreted as conditional insights rather than deterministic or causal forecasts. Future work could employ agent-based modeling to capture interactions among refineries and assess how facility-level decisions, such as shutdowns or CCS investments, propagate through the sector.

Conclusions

Policies and market trends that reduce transportation fuel demand, such as the electrification of vehicle fleets, have the potential to lower GHG emissions in the refining sector by 45% from 2019 to 2050. However, the anticipated reduction in fuel demand resulting from high electric vehicle deployment will be insufficient to meet the emissions reduction target (e.g., net-zero by 2050). Therefore, if this target is to be met, additional efforts are required, including the deployment of CCS. CCS could potentially reduce remaining refining GHG emissions by 77.6% under a low fuel demand scenario in 2050 (i.e., LD50) compared to 2019 levels. Furthermore, even under a high fuel demand scenario in 2050 (i.e., HD50), carbon capture could reduce 58.0% in refining emissions, resulting in sectoral GHG emissions of 122 MtCO₂ per year. This reduction can help the refining sector in the mid-transition towards deep decarbonization targets but could also signal to a refinery that they hesitate to invest more or shut down once they invest in carbon capture. However, the analysis suggests that mitigation strategy such as CCS alone is not enough to fully decarbonize the supply chain under the modeled transportation fuel demand scenarios. Further measures, such as carbon dioxide removal (CDR), therefore will be required.

The competitiveness of deploying carbon capture on certain process units was evaluated depending on their emissions reduction potential and CO₂ avoidance costs. Flue gas streams with higher CO₂ concentration and volume (e.g., FCC, CNR, AT, SMR) are likely to be prioritized for carbon capture. Deploying carbon capture in some refinery configurations in certain regions is cheaper than others (e.g., FCC, SMR from deep conversion refinery in PADD 3) and thus is also likely to be considered for early deployment.

In the face of uncertain future transportation fuel demand and carbon pricing stringency, refineries must navigate a complex decision-making landscape. Within the modeled refinery archetypes, hydroskimming refinery, which are particularly sensitive to carbon pricing and declining fuel demand, are estimated – under the modeled carbon price trajectories – to face a 99% likelihood of shutdown by the early 2030s, as these factors critically undermine their financial viability. Conversely, the medium- and deep-conversion refinery archetypes exhibit greater resilience, with no immediate signals to shut down, and the modeled results suggest that they may consider deploying carbon capture technologies as early as 3 years after the start of the RO simulation to enhance their operational sustainability. The RO analysis suggests a narrow but critical window under the modeled assumptions, during which refineries – particularly medium conversion archetypes – can take strategic action to maintain economic viability during a low-carbon transition. Sensitivity analysis shows that both crack spread and carbon pricing impact the likelihood and timing of key decisions, such as deploying CCS or shutting down operations. The key insight is that refinery decisions are highly sensitive to both the magnitude and the timing of these signals; if mistimed, refineries may make decisions that are misaligned with long-term policy goals.

In general, we conceptualize CCS in refineries as serving both a transitional mitigation function and, under certain conditions, a means of extending operational lifetimes. On the one hand, CCS can reduce refinery-level emissions and, under declining fuel demand scenarios, help maintain emissions trajectories consistent with near-term decarbonization pathways by mitigating a large share of operational emissions and enabling refineries to continue providing essential fuel services while replacement low-carbon systems are deployed. On the other hand, although CCS deployment may enhance near-term economic viability by lowering carbon cost exposure and stabilizing margins under assumed carbon price trajectories, these reductions are largely confined to captured process emissions and do not eliminate residual refinery emissions or upstream and downstream supply chain emissions. As such, CCS alone does not ensure long-term competitiveness in a deeply decarbonized system. The real-options analysis indicates whether CCS functions primarily as a bridge strategy or as a mechanism that prolongs carbon-intensive operations depends on the magnitude and timing of market and policy signals. Declining demand may also push some refineries toward minimum viable operating scales, increasing the risk of abrupt shutdowns that could disrupt supply and create safety or system-level challenges.⁷⁶ Measures that reduce investment risk can shift the timing and scale of CCS deployment. These include investment-based instruments such as the U.S. 45Q tax credit (i.e., Credit for Carbon Oxide Sequestration) under the Inflation Reduction Act, which provides incentives for entities that capture CO₂,⁷⁷ as well as price-based approaches such as carbon price corridors (or “price collars”), which establish upper and lower bounds on carbon prices (e.g., mechanisms under the German ETS).⁷⁸ In a North American context, policies that strengthen effective carbon price floors – such as proposed increases to the Alberta Technology Innovation and Emissions Reduction (TIER) benchmark price – may serve a similar function.⁷⁹ In addition, long-term contracting mechanisms, including carbon contracts for difference (CCfDs) with governments and project developers, can also mitigate investment uncertainty by stabilizing expected carbon revenues over project lifetimes.⁸⁰ However, such interventions must be designed to balance the objective of a secure supply of refinery products that are expected to change in unprecedented ways over the coming two decades and long-term decarbonization goals. In this context, enabling near-term abatement through CCS should be balanced with coordinated demand-side transitions, upstream decarbonization, and complementary mitigation pathways, so that the pace of declining transportation fuel demand is aligned with supply-side adjustment and transitional emissions reductions at the refinery level do not delay system-wide alignment with long-term climate goals.

Author contributions

F. L.: conceptualization, data curation, formal analysis, investigation, methodology, software, validation, visualization, writing – original draft. L. J.: resources, software, writing – review & editing. S. M: validation, writing – review & editing. S. H.: writing – review & editing. J. B.: conceptualization, funding acquisition, project administration, software, resources, supervision and writing – review & editing.

Conflicts of interest

There are no conflicts to declare.

Data availability

The PRELIM model supporting this study is an unpublished version and will be made accessible upon request. The real option model, code and subset data (including the supplementary data sets and figure plotting dataset) supporting this study is included in public repository: https://doi.org/10.5281/zenodo.16052925. The Monte Carlo simulation results for real option can be found in https://doi.org/10.5281/zenodo.15938756.

Supplementary information (SI) is available. See DOI: https://doi.org/10.1039/d5su00853k.

Acknowledgements

We thank Hyunsoo Kim, Dr Ren Bo and Dr Tinu Abraham for providing subject matter expertise. Financial support from the Aramco Services Company (MI, USA) and the Natural Sciences and Engineering Research Council (NSERC) of Canada is gratefully acknowledged. ChatGPT-4 is used for code debugging and for improving grammar and spelling in this paper. The authors are solely responsible for the data and the contents of the paper.

References

1. Total Energy Monthly Data, U.S. Energy Information Administration (EIA), https://www.eia.gov/totalenergy/data/monthly/index.php, accessed 27 April 2023.
2. C. Watson, 2011-2021 Greenhouse Gas Reporting Program Industrial Profile: Petroleum Refineries Sector, EPA.
3. V. Masson-Delmotte, H.-O. Pörtner, J. Skea, P. Zhai, D. Roberts, P. R. Shukla, A. Pirani, R. Pidcock, Y. Chen, E. Lonnoy, W. Moufouma-Okia, C. Péan, S. Connors, J. B. R. Matthews, X. Zhou, M. I. Gomis, T. Maycock, M. Tignor and T. Waterfield, An IPCC Special Report on the Impacts of Global Warming of 1.5 °C above Pre-industrial Levels and Related Global Greenhouse Gas Emission Pathways, in the Context of Strengthening the Global Response to the Threat of Climate Change, Sustainable Development, and Efforts to Eradicate Poverty.
4. T. T. Mai, P. Jadun, J. S. Logan, C. A. McMillan, M. Muratori, D. C. Steinberg, L. J. Vimmerstedt, B. Haley, R. Jones and B. Nelson, Electrification Futures Study: Scenarios of Electric Technology Adoption and Power Consumption for the United States, 2018.
5. U.S. Refinery Crude Oil Input Qualities, https://www.eia.gov/dnav/pet/pet_pnp_crq_dcu_nus_m.htm, accessed 27 April 2023.
6. D. of S. U.S., The United States' Nationally Determined Contribution: Reducing Greenhouse Gases in the United States: A 2030 Emissions Target, 2021.
7. J. Bistline, G. Blanford, M. Brown, D. Burtraw, M. Domeshek, J. Farbes, A. Fawcett, A. Hamilton, J. Jenkins, R. Jones, B. King, H. Kolus, J. Larsen, A. Levin, M. Mahajan, C. Marcy, E. Mayfield, J. McFarland, H. McJeon, R. Orvis, N. Patankar, K. Rennert, C. Roney, N. Roy, G. Schivley, D. Steinberg, N. Victor, S. Wenzel, J. Weyant, R. Wiser, M. Yuan and A. Zhao, Science, 2023, 380, 1324–1327.
8. Congress Climate History, https://www.c2es.org/content/congress-climate-history/, accessed 25 February 2025.
9. T. Hale, T. Wetzer, S. K. Abebe, M. Allen, A. Amel-Zadeh, J. Armour, K. Axelsson, B. Caldecott, L. Dias, S. Fankhauser, B. Franta, C. Hepburn, K. Mbeva, L. Rajamani, S. Smith and R. Stuart-Smith, Nat. Clim. Change, 2024, 14, 306–308.
10. H. Kim, PhD thesis, University of Calgary, 2023.
11. W. Li, C.-W. Hui and A. Li, Comput. Chem. Eng., 2005, 29, 2010–2028.
12. J. Fallurin, T. J. Conway and T. J. Kirk, Oil Refining Emissions Cut Points: Downstream Decarbonization in the Decisive Decade, RMI, 2022.
13. S. Jiang, J. Appl. Syst. Anal., 2012, 39, 1505–1530.
14. J. van Straelen, F. Geuzebroek, N. Goodchild, G. Protopapas and L. Mahony, Int. J. Greenhouse Gas Control, 2010, 4, 316–320.
15. K. Heal and T. Kemp, Energy Procedia, 2013, 37, 7046–7055.
16. D. Johansson, P.-Å. Franck and T. Berntsson, Energy Convers. Manage., 2013, 66, 127–142.
17. K. Motazedi, J. P. Abella and J. A. Bergerson, Environ. Sci. Technol., 2017, 51, 1918–1928.
18. Y. Yao, J. Marano, W. R. Morrow and E. Masanet, Int. J. Greenhouse Gas Control, 2018, 74, 87–98.
19. H. Pilorgé, N. McQueen, D. Maynard, P. Psarras, J. He, T. Rufael and J. Wilcox, Environ. Sci. Technol., 2020, 54, 7524–7532.
20. P. Sun, V. Cappello, A. Elgowainy, P. Vyawahare, O. Ma, K. Podkaminer, N. Rustagi, M. Koleva and M. Melaina, Environ. Sci. Technol., 2023, 57(3), 1411–1424.
21. J. Gale, ReCAP Project Understanding Cost of Retrofitting, IEAGHG, Cheltenham, 2017, 2017-TR8.
22. S. Roussanaly, N. Berghout, T. Fout, M. Garcia, S. Gardarsdottir, S. M. Nazir, A. Ramirez and E. S. Rubin, Int. J. Greenhouse Gas Control, 2021, 106, 103263.
23. S. Fuss, D. J. A. Johansson, J. Szolgayova and M. Obersteiner, Energy Policy, 2009, 37, 733–743.
24. B. Young, M. Krynock, D. Carlson, T. R. Hawkins, J. Marriott, B. Morelli, M. Jamieson, G. Cooney and T. J. Skone, Int. J. Greenhouse Gas Control, 2019, 91, 102821.
25. F. Güleç, W. Meredith and C. E. Snape, Front. Energy Res., 2020, 8, 62.
26. H. Li, M. Ditaranto and D. Berstad, Energy, 2011, 36, 1124–1133.
27. D. G. Chapel, C. L. Mariz and J. Ernest, Recovery of CO₂ from Flue Gases: Commercial Trends, presented at the Canadian Society of Chemical Engineers Annual Meeting, Saskatoon, Saskatchewan, Canada, 4–6 October 1999, paper no. 340.
28. T. Mimura, H. Simayoshi, T. Suda, M. Iijima and S. Mituoka, Energy Convers. Manage., 1997, 38, S57–S62.
29. O. Bolland and H. Undrum, Adv. Environ. Res., 2003, 7, 901–911.
30. R. Digne, F. Feugnet and A. Gomez, Oil Gas Sci. Technol., 2014, 69, 1081–1089.
31. L. F. de Mello, R. D. M. Pimenta, G. T. Moure, O. R. C. Pravia, L. Gearhart, P. B. Milios and T. Melien, Energy Procedia, 2009, 1, 117–124.
32. W. Chung and J. H. Lee, Ind. Eng. Chem. Res., 2020, 59, 18951–18964.
33. 4-Week Avg U.S. Product Supplied of Petroleum Products (Thousand Barrels per Day), https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=PET%26s=WRPUPUS2%26f=4, accessed 6 May 2024.
34. U.S. Refinery and Blender Net Production, https://www.eia.gov/dnav/pet/pet_pnp_refp_dc_nus_mbbl_m.htm, accessed 27 April 2023.
35. East Coast (PADD 1), Net Receipts of Crude Oil and Petroleum Products by Pipeline, Tanker, Barge and Rail, https://www.eia.gov/dnav/pet/pet_move_netr_d_r10-z0p_VNR_mbbl_m.htm, accessed 27 April 2023.
36. East Coast (PADD 1), Imports of Crude Oil and Petroleum Products for Processing, https://www.eia.gov/dnav/pet/pet_move_imp2_dc_R10-Z00_mbbl_m.htm, accessed 27 April 2023.
37. East Coast (PADD 1), Total Crude Oil and Products Exports, https://www.eia.gov/dnav/pet/pet_move_expcp_a2_r10_ep00_eex_mbbl_m.htm, accessed 27 April 2023.
38. U.S. Refinery Stocks, https://www.eia.gov/dnav/pet/pet_stoc_ref_dc_nus_mbbl_m.htm, accessed 27 April 2023.
39. Company Level Imports, https://www.eia.gov/petroleum/imports/companylevel/, accessed 27 April 2023.
40. G. Cooney, M. Jamieson, J. Marriott, J. Bergerson, A. Brandt and T. J. Skone, Environ. Sci. Technol., 2017, 51, 977–987.
41. W. L. Nelson, Guide to Refinery Operating Cost (Process Costimating), 1976.
42. W. L. Nelson, How the Nelson Refinery Constraction Cost Indexes Evolved, 1976.
43. W. L. Nelson, Here's How Operating Cost Indexes Are Computed, 1977.
44. U.S. Refinery Yield, https://www.eia.gov/dnav/pet/pet_pnp_pct_dc_nus_pct_m.htm, accessed 27 April 2023.
45. 2021 Worldwide Refining Capacity Summary, https://www.ogj.com/ogj-survey-downloads/worldwide-refining/document/14233364/2021-worldwide-refining-capacity-summary, accessed 27 April 2023.
46. R. S. Middleton, S. P. Yaw, B. A. Hoover and K. M. Ellett, Environ. Modell. Software, 2020, 124, 104560.
47. V. E. Onyebuchi, A. Kolios, D. P. Hanak, C. Biliyok and V. Manovic, Renewable Sustainable Energy Rev., 2018, 81, 2563–2583.
48. M. D. Jensen, P. Pei, A. C. Snyder, L. V. Heebink, L. S. Botnen, C. D. Gorecki, E. N. Steadman and J. A. Harju, Energy Fuels, 2013, 27, 4175–4182.
49. L. Trigeorgis, Financ. Manag., 1993, 22, 202.
50. T. E. Copeland and V. Antikarov, Real Options: A Practitioner's Guide, Texere, 2001.
51. L. E. Brandão and J. S. Dyer, Ann. Oper. Res., 2005, 135, 21–39.
52. U.S. Energy Information Administration, Annual Energy Outlook 2022, 2022.
53. Resource Center, https://home.treasury.gov/resource-center/data-chart-center/interest-rates/TextView, accessed 20 October 2025.
54. U.S. Energy Information Administration – EIA – Independent Statistics and Analysis, https://www.eia.gov/environment/emissions/co2_vol_mass.php, accessed 4 July 2024.
55. E. and C. C. Canada, The federal carbon pollution pricing benchmark, https://www.canada.ca/en/environment-climate-change/services/climate-change/pricing-pollution-how-it-will-work/carbon-pollution-pricing-federal-benchmark-information.html, accessed 20 February 2026.
56. J. H. Gary and G. E. Handwerk, Petroleum Refining: Technology and Economics, M. Dekker, New York, 4th edn, 2001.
57. D. Mullen, L. Herraiz, J. Gibbins and M. Lucquiaud, Int. J. Greenhouse Gas Control, 2023, 126, 103904.
58. Y. Khojasteh Salkuyeh, B. A. Saville and H. L. MacLean, Int. J. Hydrogen Energy, 2017, 42, 18894–18909.
59. J. C. Meerman, E. S. Hamborg, T. van Keulen, A. Ramírez, W. C. Turkenburg and A. P. C. Faaij, Int. J. Greenhouse Gas Control, 2012, 9, 160–171.
60. I. Lindsay, C. Lowe, S. Reddy, M. Bhakta and S. Balkenende, Energy Procedia, 2009, 1, 4095–4102.
61. M. H. Ali Khan, R. Daiyan, P. Neal, N. Haque, I. MacGill and R. Amal, Int. J. Hydrogen Energy, 2021, 46, 22685–22706.
62. G. Lozza and P. Chiesa, J. Eng. Gas Turbines Power, 2002, 124, 89–95.
63. F. Pruvost, S. Cloete, C. Arnaiz Del Pozo and A. Zaabout, Energy Convers. Manage., 2022, 274, 116458.
64. D. Bonaquist, Analysis of CO2 Emissions, Reductions, and Capture for Large-Scale Hydrogen Production Plants, Praxair, 2010.
65. R. G. Lemus and J. M. Martínez Duart, Int. J. Hydrogen Energy, 2010, 35, 3929–3936.
66. T. Grant, D. Morgan and K. Gerdes, Carbon Dioxide Transport and Storage Costs in NETL Studies, National Energy Technology Laboratory, 2013.
67. M. Exchange, Crack Spread Handbook, NewYork: New York Mercantile Exchange, 1989.
68. J. Archsmith, E. Muehlegger and D. S. Rapson, Environ. Energy Policy Econ., 2022, 3, 71–110.
69. L. Woodley, C. Y. See, P. Cook, M. Yeo, D. S. Palmer, L. Huh, S. Wang and A. Nunes, Nat. Commun., 2024, 15, 6813.
70. F. Black and M. Scholes, J. Polit. Econ., 1973, 81, 637–654.
71. J. C. Cox, S. A. Ross and M. Rubinstein, J. Financ. Econ., 1979, 7, 229–263.
72. P. S. Reinelt and D. W. Keith, Energy J., 2007, 28, 101–128.
73. J. M. McKellar, J. A. Bergerson, J. Kettunen and H. L. MacLean, Environ. Sci. Technol., 2013, 47, 5979–5987.
74. J.-L. Fan, M. Xu, L. Yang, X. Zhang and F. Li, Energy Policy, 2019, 132, 1229–1240.
75. M. Zhang, L. Liu, Q. Wang and D. Zhou, Energy Econ., 2020, 92, 104954.
76. J. Lappen and E. Grubert, Science, 2026, 391, 449–452.
77. U. S. Congress, Inflation Reduction Act of 2022, Public Law 117–169, 136 Stat. 1818, 2022.
78. Emissions trading systems, https://www.oecd.org/en/publications/effective-carbon-rates-2025_a5a5d71f-en/full-report/emissions-trading-systems-a-variety-of-designs_b98e7b5c.html, accessed 17 March 2026.
79. Government of Alberta, Standard for Developing Benchmarks under the Technology Innovation and Emissions Reduction Regulation, Alberta Environment and Protected Areas, 2024.
80. International Emissions Trading Association (IETA), Carbon Contracts for Difference 101, International Emissions Trading Association, 2023.

---