全球变化知识资源中心

Because human activities emit greenhouse gases (GHGs) and conventional air pollutants from common sources, policy designed to reduce GHGs can have co-benefits for air quality that may offset some or all of the near-term costs of GHG mitigation. We present a systems approach to quantify air quality co-benefits of US policies to reduce GHG (carbon) emissions. We assess health-related benefits from reduced ozone and particulate matter (PM_2.5) by linking three advanced models, representing the full pathway from policy to pollutant damages. We also examine the sensitivity of co-benefits to key policy-relevant sources of uncertainty and variability. We find that monetized human health benefits associated with air quality improvements can offset 26–1,050% of the cost of US carbon policies. More flexible policies that minimize costs, such as cap-and-trade standards, have larger net co-benefits than policies that target specific sectors (electricity and transportation). Although air quality co-benefits can be comparable to policy costs for present-day air quality and near-term US carbon policies, potential co-benefits rapidly diminish as carbon policies become more stringent.

Climate change and regional air quality are major sustainability challenges. Ground level ozone (O₃) and particulate matter (PM_2.5, particulate matter with diameter ≤ 2.5 μm) are linked to respiratory diseases and premature death^{1, 2}. Despite regulatory efforts, 232 and 118 US counties exceeded national O₃ and PM_2.5 standards, respectively, in 2011 (refs 3, 4). Concurrently, changing climate is becoming a global health issue, as increasing temperatures and changing weather patterns threaten human well-being⁵.

The Intergovernmental Panel on Climate Change (IPCC) noted that GHG emissions controls can have near-term health co-benefits from reduced air pollution, which may offset a substantial fraction of mitigation costs⁶. Nemet et al.⁷ summarized 37 peer-reviewed co-benefits estimates, finding a range from US$2-196/tCO₂ and mean US$47/tCO₂, with highest values in developing countries. This range reflects co-benefit variability across different study methods, technologies, spatial scales and societies.

Air quality co-benefits estimates are additional to climate benefits from reduced CO₂ emission. In assessing the Social Cost of Carbon (SCC), the US Interagency Working Group estimated marginal damages of CO₂ emitted in 2020 at 43 US$/tonne (2007 US$ using 3% discounting; refs 8, 9). These monetized impacts of CO₂ emissions include, but are not limited to, reduced agricultural yields, coastal flooding, and increased frequency and severity of weather events¹⁰.

Air pollution and climate change are elements of a coupled social and technical system. Comprehensively assessing potential co-benefits of climate policies to air pollution and associated human impacts, considering variability and uncertainty, requires combining approaches from several disciplines tracing the entire pathway from policies to impacts. First, climate policies influence economic activities and associated emissions of both GHGs and conventional air pollutants. Unlike for GHGs, spatial distribution of air pollutant emissions matters. O₃ and PM_2.5 formation is nonlinear, and pollutant distribution also impacts population exposure; predicting these requires advanced atmospheric modelling. Atmospheric concentrations must then be linked to human health outcomes through exposure-response calculations. Costs are then derived from economic analyses.

Previous literature has addressed aspects of this system in both physical and societal dimensions, using models to simulate complexities and interactions¹¹. In atmospheric chemistry, most analyses using comprehensive chemical models have treated policies exogenously, with associated fixed costs^{12, 13}. Economic studies^{14, 15, 16, 17, 18, 19} have focused extensively on drivers of cost variation. However, these studies often use simplified methods linking emissions to concentrations and impacts, neglecting full atmospheric complexity.

It has been noted that the co-benefits literature has had little policy traction^{7, 11}; one reason given is lack of comprehensive analysis from the full set of disciplines underlying both cost and benefit analysis. With information on how assumptions and uncertainties from various fields combine to influence benefits-per-ton estimates, decision-makers can identify the robustness of policies to variation in drivers of both cost and benefit. As full quantitative uncertainty analysis of all factors is computationally impossible, methods are needed to selectively address the most policy-relevant uncertainties.

Here, we illustrate a systems-level approach to analysing how climate policies influence air quality, focusing on US emissions of O₃ and PM_2.5 precursors to 2030. We assess costs and air-quality-related benefits of three potential national-scale climate policies. We examine the entire pathway linking climate policies, economic sector responses, emissions, regional air quality, human health and related economic impacts, using advanced models at every stage. We first simulate climate policies in the United States Regional Energy Policy (USREP) model. Resulting economic constraints lead to economic output changes that vary by policy, economic sector and US region²⁰. Changed economic output is used to scale emissions inventories, and the Comprehensive Air Quality Model with Extensions (CAMx; ref. 21) projects resulting ambient pollutant levels. Finally, the Environmental Benefits Mapping and Analysis (BenMAP) program calculates changes in population exposure to pollution, resulting changes to human mortality and morbidity, and corresponding monetized benefits²². Using these coupled models, we capture important economic and atmospheric complexities and nonlinearities. We also conduct a policies-to-impacts sensitivity analysis to quantify policy-relevant uncertainties and variabilities: economic growth, technology costs, baseline emissions assumption and representation of health responses.

We first present results for economic costs and emissions, O₃ and PM_2.5 concentrations, and health and economic benefits for three carbon reduction policies for a base case. We then show our policies-to-impacts sensitivity analysis, examining variation of base case results with uncertainty and variability in key assumptions. Costs and benefits for all scenarios described below are reported inTable 1 and presented graphically in Supplementary Fig. 3.

Full table

We show results of sensitivity to policy stringency, baseline precursor emissions, and economic assumptions and parameters. Although additional uncertainties and variability exist along the policies-to-impacts pathway, as discussed further below, we select these as major influences on policy-relevant variation. Costs per tonne CO₂ are presented in Supplementary Table 2.

Sensitivity to stringency of policy.

We tested a cap-and-trade policy (CAT-High) that achieves approximately twice the CO₂ reduction of our base case. Figure 2 shows the percentage of policy costs covered by co-benefits versus total carbon reduction relative to 2006 for the CAT base and CAT-High scenarios. The benefit-cost ratio of CAT-High decreases with increasing policy stringency as cheaper controls are exhausted.

Policy cases (CAT, CES and TRN) and sensitivity scenarios shown for: 2005 and 2012 criteria pollutant emissions inventory and CAT-High sensitivity run. Red vertical line shows the approximate reduction target consistent with a 2 °C temperature increase limit³⁹.

Air quality co-benefits of carbon emissions policies would probably offset much of their economic cost⁷. We estimate that human health benefits associated with air quality improvements offset 26–1,050% of costs depending on the flexibility of the carbon policy. More flexible policies such as CAT are less costly than those that impose reductions from specific sectors (CES or TRN), as the latter fail to equalize marginal abatement costs across economic activities²⁶. We find that this flexibility has a relatively larger influence on cost, and a smaller influence on co-benefits.

Using our policies-to-impacts sensitivity analysis, we identify important qualifications to our base case conclusions, going beyond the insights of previous work. Median benefits of CAT aggregated at the national scale exceeded its low costs in all simulated sensitivity scenarios. Although carbon policies that target economic sectors known to contribute substantially to poor air quality (electricity and transportation sectors) have somewhat larger benefits, these policies are also more costly versus CAT. More stringent CAT policy (CAT-High) exceeded the benefits of these sector-specific approaches. However, increasing stringency of CAT policy leads to a smaller fraction of costs offset by co-benefits.

A key finding from our sensitivity analysis is that policy and economic assumptions had a larger impact on policy costs than on median co-benefits both across and within different scenarios. Standard atmospheric science approaches have largely omitted rigorous accounting for cost uncertainty; our analysis suggests this can be the most important policy-relevant uncertain term. This suggests that, for a variety of carbon policy choices, including subsidies that influence the cost of renewables and technologies, net co-benefit is driven by costs rather than benefits.

Large-scale pollutant emissions reductions unrelated to carbon policy will probably decrease human health co-benefits. We saw a 16–24% decrease in human health co-benefits by changing baseline emissions, sublinear relative to emissions decreases. Although our 2012 inventory takes into account present air quality regulations, our 2030 predictions do not account for further regulatory action. In particular, the Mercury and Air Toxics Standards (MATS) are projected to reduce SO₂ emissions from the power sector by >40% by 2016 (ref. 27). As the emissions baseline does not account for MATS SO₂ reductions, co-benefits associated with PM_2.5 may be biased high. Uncertainty in emissions changes due to future policy is not well understood²⁸, and although our emissions runs partially address this, future analyses could apply endogenous pollution abatement costs in economic models²⁹.

Uncertainty in concentration-response functions (crfs) and Value of a Statistical Life (VSL) has a large influence on the magnitude of benefits, but we show that their variation can be comparable to other assumptions along the policies-to-impacts chain (such as economic modelling assumptions). Previous work has found assumptions associated with crfs to be a larger source of uncertainty than assumptions associated with VSL (ref. 30). Uncertainty associated with both the crfs and VSL will be constant across scenarios and sensitivities, so although these assumptions will change the benefit to cost ratio, they do not change how policies compare to each other. In our benefits analysis, we used BenMAP, for consistency with regulatory analyses. Alternatively, representing air quality impacts in a computable general equilibrium model can assess their economy-wide welfare implications (mortality impacts on labour supply, and morbidity impacts on demand for health services), and can capture the response of these impacts to changing prices and policy constraints^{31, 32, 33}.

Although we have illustrated several policy-relevant uncertainties and variabilities along the policies-to-impacts pathway, there are numerous aspects that we have not quantitatively assessed. Year-to-year meteorological variability can change the distribution and formation of PM_2.5 and O₃. Climate changes can affect air quality: rising temperatures will probably increase O₃ formation on the order of increases predicted here by changing emissions^{34, 35}. However, health benefits are dominated by PM_2.5 (refs 36, 37), and the influence of climate change on PM_2.5 distribution and health impacts remains difficult to quantify³⁸. Uncertainty quantification in modelling transport and chemistry of pollutant formation is also limited both by model fidelity and scientific knowledge, particularly with respect to the formation of PM_2.5. Further applications of our approach, however, could incorporate future quantitative estimates of these influences in a policy-relevant way. For example, although meteorological variability may change the absolute level of air pollution co-benefits, it may not affect selected policies differently. In this way, our approach is distinct from traditional uncertainty and sensitivity analysis.

There are additional uncertainties which cannot be captured within this framework. Regional atmospheric models such as CAMx cannot capture sub-grid scale variability. If maximum reductions occur in areas of high population density with large spatial gradients (for example, primary PM_2.5 from vehicles), the corresponding scenario could have larger benefits. Secondary PM_2.5, however, is not sensitive to resolution on the scale of most regional modelling³⁶.

Whereas our model assumes that sectoral reductions will be homogeneous across all sources within each sector, different sources within each sector will react differently. For example, if carbon policy reduces coal-fired power plant output, some individual plants may close while others operate normally. This can affect the spatial distribution of calculated benefits.

Our approach suggests several insights for decision-makers considering co-benefits of different climate policies. We find co-benefits comparable with policy costs for existing air quality and realistic climate policy goals in the US, suggesting that substantial co-benefits for CO₂ reduction are not limited to developing regions.

A US carbon cap would have a measurable, positive impact on regional air quality relative to BAU, similar in magnitude to policy specifically targeting O₃ and PM_2.5. Air pollution reductions relative to BAU estimated here are comparable to the 1.4% and 38% reductions of NO_x and SO₂ proposed by the US EPA for 2014 and recently upheld by the US Supreme Court. However, EPA estimates of air quality policy costs (US$800 mil, ref. 24) are an order of magnitude smaller than carbon policy costs. This result should also be interpreted with caution, as our BAU case does not include air quality improvement.

We find diminishing relative co-benefits with both baseline emissions improvements and increasing climate policy stringency. Figure 2 suggests that very stringent climate policies—necessary to meet a 2° global target, estimated as ≥80% reductions³⁹—may be offset to a much lower degree by air quality co-benefits. Macro-economic costs of CO₂ abatement are increasing more than proportionally to the abatement target. This means that although initial policy actions can be motivated based on air pollution co-benefits, this strategy has important limits. Whereas we conduct national scale analysis, state and regional air quality decision-makers assess policies based on both regional cost-benefit assessment and regulatory attainment status. Benefits in individual regions largely follow PM_2.5 concentration changes, but policy makers might be more concerned with O₃ from a regulatory standpoint. Costs can vary substantially across regions, reflecting, among other things, regional disparities in energy intensity (energy consumption/GDP) and electricity generation fuel mix⁴⁰. Regional benefit/cost ratios may thus vary, creating winners and losers. Also, carbon policies may be applied at state or regional, rather than national scale. Future work could examine these regional differences.

The co-benefits estimates presented here should not be interpreted as a comprehensive benefit-cost analysis for the considered carbon policies. Co-benefits are additional to (and of larger magnitude than) estimates of the SCC. We focus on O₃ and PM_2.5; additional air quality improvements such as reductions in mercury and other air toxics could have co-benefits not captured here. Furthermore policies themselves may have additional health and economic benefits: for example, reduced vehicle transportation use may improve health by encouraging walking and bicycling. Our results suggest, however, that cost-benefit analyses of climate policy that omit regional air pollution could greatly underestimate benefits.

The MIT US regional energy policy (USREP) model.

USREP is a recursive-dynamic computable general equilibrium (CGE) model of the US economy designed to analyse energy and GHG policies. USREP has been widely used to investigate energy and climate policy, including interactions with tax policy, and effects on economic growth, efficiency and distribution^{20, 40, 41, 42}. USREP is described in detail in Rausch et al.^{40, 42}, and further model details are presented in the Supplementary Information. We conduct simulations from 2006 to 2030, with a five-year timestep. CO₂ emissions grow to 6,200 million metric tonnes (mmt) under the Business-As-Usual (BAU) case. Economic output and CO₂ emission under each policy is archived from USREP by region and sector (see Supplementary Information for details).

The three USREP scenarios used for sensitivity analysis vary baseline economic growth and CO₂ emissions, cost of renewables, and fuel efficiency improvements. High-Base has a 14% greater CO₂ emission under business-as-usual in 2030 (7,100 mmt). Low-Cost reduces the cost of wind technologies by 15% compared to the base case, and improves fuel efficiency by increasing flexibility in the model to substitute fuel inputs with powertrain capital in private vehicles. The High-Cost scenario increases wind technology cost by 15% compared to the base case and decreases the flexibility in the model to substitute fuel inputs with powertrain capital in private vehicles.

Future emissions projections.

We match USREP economic sectors and regions to individual emissions sources by matching Standard Classification Codes used in US EPA national emission inventories to USREP sector categories. The 2030 scaled emissions inventory is prepared for regional modelling using the Sparse Matrix Operator Kernel Emissions (SMOKE) emission preprocessing program⁴³. SMOKE creates gridded and speciated hourly emissions files for input to CAMx. Supplementary Table 4 shows benefits per short ton of SO₂ and NO_x reduced for three base case scenarios. We assume emissions