Effects of global climate mitigation on regional air quality and health

Climate mitigation can bring air quality and health co-benefits. How these health impacts might be distributed across countries remains unclear. Here we use a coupled climate–energy–health model to assess the country-varying health effects of a global carbon price across nearly 30,000 future states of the world (SOWs). As a carbon price lowers fossil fuel use, our analysis suggests consistent reductions in ambient fine particulate matter (PM2.5) levels and associated mortality risks in countries that currently suffer most from air pollution. For a few less-polluted countries, however, a carbon price can increase the mortality risks under some of the considered SOWs due to emissions increases from bioenergy use and land-use changes. These potential health co-harms are largely driven in our model by the scale and method of deforestation. A robust and quantitative understanding of these distributional outcomes requires improved representations of relevant deep uncertainties, country-specific characteristics and cross-sector interactions. More efficient and targeted climate mitigation policies require an improved understanding of how the associated air quality and health benefits will be distributed. This study assesses, at the country level, the health effects of a global carbon price under different future scenarios.

Climate mitigation can bring air quality and health co-benefits.How these health impacts might be distributed across countries remains unclear.
Here we use a coupled climate-energy-health model to assess the country-varying health effects of a global carbon price across nearly 30,000 future states of the world (SOWs).As a carbon price lowers fossil fuel use, our analysis suggests consistent reductions in ambient fine particulate matter (PM 2.5 ) levels and associated mortality risks in countries that currently suffer most from air pollution.For a few less-polluted countries, however, a carbon price can increase the mortality risks under some of the considered SOWs due to emissions increases from bioenergy use and land-use changes.These potential health co-harms are largely driven in our model by the scale and method of deforestation.A robust and quantitative understanding of these distributional outcomes requires improved representations of relevant deep uncertainties, country-specific characteristics and cross-sector interactions.
Reducing fossil fuel combustion decreases emissions of carbon dioxide and toxic air pollutants.As a result, climate mitigation efforts are expected to bring health co-benefits by improving air quality 1 .However, the distribution of these health impacts across countries remains poorly understood.Globally, the pollution and health impacts are already unevenly distributed at present.Half of all deaths attributable to the exposure to ambient fine particulate matter (technically PM 2.5 ) currently occur in China and India 2 due to high pollution levels combined with the large size of the exposed population.The future health burden in these countries may decrease as air pollution-control policies are further tightened to reduce pollution but may also increase if ageing trends increase the population's vulnerability to air pollution 3,4 .
The air quality and health implications of climate mitigation depend on a range of country-specific characteristics in the energy and social aspects.Reducing fossil fuel combustion often lowers pollution exposure 1,5 , but the magnitude of the air quality co-benefits is affected by the energy mixes.For instance, coal currently accounts for 60% of primary energy use (in EJ) in China but only 12% in the United States 6 , leading to greater health benefits from coal phase-out in China than the United States 7,8 .In addition, the health effects are influenced by socio-demographic patterns that determine the size and vulnerability of the exposed population.For instance, the elderly population (age 65 or greater, which is more vulnerable to air pollution exposure) is 131 million in China (9.5% of the national total population) as compared with 47 million in the United States (15%) (ref.3).Understanding the differential regional health impacts of climate mitigation thus requires careful consideration of the energy and socio-demographic trends in each region 1,9 .
Another factor complicating the relationship between climate mitigation and reduced mortality is the potential emergence of new sources of air pollution 10 .For example, climate mitigation pathways may involve large-scale production and consumption of bioenergy 11 .This can increase the emissions of PM 2.5 from biomass combustion in end-use sectors 12 and the emissions of ammonia (NH 3 ) from upstream agricultural activities to produce bioenergy crops 13,14 .Besides the direct emissions from bioenergy production and consumption, Article https://doi.org/10.1038/s41893-023-01133-5action: US$28, US$69 and US$117 ton −1 CO 2 in 2030, 2050 and 2100, respectively (Fig. 2a).The near-term price level broadly reflects the global policy ambition for the next decade by adding up countries' existing Nationally Determined Contributions [30][31][32] .The longer-term price level considers only a moderate increase over time and is set at the same magnitude as the highest carbon price that has been observed so far (that is, US$103 ton −1 in Sweden and Switzerland, inflation-adjusted 22 ).
Compared to the SOWs with no carbon price, we estimate that this carbon price trajectory reduces the global average temperature by 0.1 °C (based on ensemble median; range 0.1-0.2°C across the considered SOWs) in 2050 and by 0.6 °C (range: 0.4-1.1 °C) in 2100 (Fig. 2b).It suggests that even moderate global mitigation contributes to reduced warming and climate damage.However, with the considered carbon price, the ensemble median end-of-century global mean temperature is estimated to be 2.8 °C (range: 2.2-4.3 °C; N = 14,180) higher than the pre-industrial level (or radiative forcing level at 4.9 W m −2 ; range: 3.6-8.0W m −2 ).It indicates that achieving 2° or stricter climate targets would require more stringent global action beyond current ambitions.
Consistent with prior studies 5,28,29,33 , we find that pricing carbon improves global air quality and reduces average PM 2.5 -attributable death rates.Globally, imposing this carbon price reduces the ensemble median PM 2.5 -attributable death rate by 7% (or 69 deaths per million people; range: 38-108) in 2050 and 11% (or 137 deaths per million people; range: 35-358) in 2100 (Fig. 2c).This corresponds to 0.4 (range: 0.2-0.7) million avoided deaths in 2050 and 0.9 (range: 0.2-2.9) million avoided deaths in 2100 or an annual average reduction of 0.5 (range: 0.2-1.1) million deaths from 2015 to 2100.Our findings are broadly consistent with prior studies, though many of them considered more ambitious climate mitigation scenarios and therefore found a larger magnitude of the co-benefits (broad range across studies: 0.8-1.8 million) (refs.5,28,29).In addition, our assessment of future PM 2.5 concentrations (and the associated health impacts) consider only the changes in precursor air pollutant emissions but not the effects of changing meteorological conditions under a changing climate.

Reduced PM 2.5 -related health inequities across regions
Cross-country inequity can be defined and operationalized in different ways.Here we define the distribution of impacts as equitable when people in all regions face similar health outcomes.A policy action, such as pricing carbon, is equity-improving when it brings increased benefits to regions that currently suffer worse health outcomes than other regions.The results in the main text focus on one metric for the health outcome, that is, PM 2.5 -attributable death rates, which measures the health risks.The results for the other health outcomes (for example, PM 2.5 exposure level and the number of PM 2.5 -attributable deaths) and other equity definitions (for example, based on country income and age groups) are presented in Supplementary Figs.3-6.To demonstrate how the distributional effects may evolve over time, the main results below are for mid-century (for example, year 2050); Supplementary Figs. 7  and 8 provide results for more near-term (for example, year 2030) and longer-term (for example, year 2100) time periods.
Across all considered SOWs, regional inequities in pollution and health persist throughout the century (Fig. 3a).The future PM 2.5 -attributable death rate remains higher in developing countries and emerging economies that are currently exposed to higher levels of air pollution.For example, in the SOWs without a carbon price, India and other South Asian nations have the highest PM 2.5 -attributable death rates in 2050 with an ensemble median exceeding 1,500 deaths per million people.In contrast, the lowest projected death rates occur in Australia, Canada and Northern Europe, with an ensemble median less than 200 PM 2.5 -attributable deaths per million people.
Pricing carbon reduces, but does not eliminate, the regional inequities (Fig. 3b).The health benefits associated with the considered carbon price trajectory are generally greater for regions where the bioenergy-heavy futures may also result in increased land competition 15 , leading to indirect emissions from land-use changes (for example, organic carbon (OC) emissions from burning forests 16 ).This illustrates the complexities resulting from the multi-sector and multi-regional linkages of the global socio-economic systems.
How will global climate mitigation affect regional air quality and health in the twenty-first century?A central challenge in tackling this question is that changes in energy and socio-economic patterns, which drive future pollution exposure and population vulnerability, are highly complex and uncertain (Fig. 1a provides a generic illustration of the complex system dynamics).These uncertainties pose conceptual and methodological difficulties for assessing future air pollution effects and identifying key conditions that result in more or less equitable impact distributions.Here we address this challenge by considering many plausible future states of the world (SOWs) through an exploratory modelling approach 17 .We use a coupled energy-climate-health modelling framework, linking a leading integrated assessment model (Global Change Analysis Model, GCAM 18 ) with a reduced-form air pollution model 19 and a country-level health-impact assessment module 20 , to assess the effects of a global carbon price in nearly 30,000 potential SOWs from 2015 to 2100.We use a carbon price as a proxy for climate mitigation action because it is the most economically efficient way to achieve global decarbonization 21 and has been adopted by many countries and subnational regions to mitigate sectoral or economy-wide emissions 22 .
This study advances the previous literature in three main ways.First, we use an exploratory ensemble approach 23,24 to sample a wide range of socio-economic and technological uncertainties and to characterize how these uncertainties propagate through a highly interconnected, multi-sector system to impact air quality and health.We expand on the prior use of a small number of narrative-based scenarios (for example, the Shared Socioeconomic Pathways (SSPs) 25 ) to drastically increase the combinations of uncertain factors being considered.By evaluating a wider range of potential future SOWs, our a posteriori ensemble-based exploratory approach could achieve a clearer picture of the full range of the plausible health outcomes 23,26,27 .It also facilitates the identification of factors (or combinations of factors) that consistently contribute to health co-benefits and co-harms.
Second, we demonstrate possible pathways by which carbon pricing could result in unintended health co-harms under certain SOWs (Fig. 1b).Many studies have already demonstrated the direct air quality-related health co-benefits from lowering fossil fuel uses to mitigate climate change 5,28,29 .However, climate mitigation may induce changes in emerging low-carbon technologies and land uses through indirect mechanisms, potentially counteracting the health benefits from lowering fossil fuels.These plausible pathways for health co-harms have not been identified in previous studies that focus on a relatively limited number of scenarios.Our large-scale ensemble approach enables a more careful analysis of potential locations and future conditions where co-harms might occur.
Third, we expand on prior co-benefit studies by focusing on the distributional outcomes across countries.Shifting from aggregate impacts to distributions is crucial to analysing potential inequities and incorporating equity considerations into the global efforts to address climate change, air pollution and health challenges.Here we focus on the regional distributions of health-relevant metrics, including pollution exposure, mortality risks and the improvements from carbon pricing.By developing an integrated modelling framework, we trace how a global carbon price might drive the future distributional outcomes in health, considering the complex interactions of energy system changes, land-use changes, air pollutant emissions, human exposure and vulnerability.

Climate change mitigation lowering health risks
We impose an increasing carbon price trajectory on global energy-sector CO 2 emissions to approximate a moderate ambition level for climate Article https://doi.org/10.1038/s41893-023-01133-5We use the GCAM model 18 to sample future SOWs and implement the carbon price (total ensemble size: N = 28,706).We estimate the effects of air pollutant emissions on ambient PM 2.5 concentrations using the TM5-FASST model 19 .The health-impact assessment further uses the projected population and age structure from the IIASA database 56 and the baseline mortality rates from the IFs model 68 .More details are presented in Methods section.

Article
https://doi.org/10.1038/s41893-023-01133-5PM 2.5 -attributable death rates are presently high (Figs.2d and 3).For instance, based on the 2050 ensemble median, pricing carbon lowers the PM 2.5 -attributable death rate by 113-293 deaths per million per year (or 9.3-12.5%) in India and other South Asian nations.In comparison, for regions with lower health risk at present, the reduction is only 0.7-0.9PM 2.5 -attributable deaths per million per year (or 0.2-0.4%) in Australia, the United States and Northern Europe.
Our results suggest that pricing carbon provides a promising avenue to narrowing current pollution and health inequities.This core insight holds for all considered future time periods (Supplementary Figs.7 and 8 for 2030 and 2100 results).It also holds for alternative health outcomes that are used such as PM 2.5 exposure level and PM 2.5 -attributable deaths (Supplementary Figs. 3 and 4).In addition, considering other definitions for equity, we find supporting evidence that pricing carbon is likely to improve distributional equity by bringing greater health benefits to lower-income countries and to elderly populations (Supplementary Figs. 5 and 6).

Competing health pathways from carbon pricing
What causes these differential regional health effects of a global carbon price?Our analysis framework models conceptual pathways through which a carbon price could result in both co-benefits and co-harms (Figs.1a,b).Health co-benefits can be driven by a reduction in air pollutant emissions from fossil fuel combustion, which is the dominant impact in regions that currently suffer from high pollution and health risks.Health co-harms can result from increasing emissions from bioenergy use and land-use changes associated with bioenergy production, which is more prominent in the regions with cleaner air at present but may expand bioenergy production in the future.Comparing the SOWs in our ensemble, the relative importance of these two pathways contributes to the regional variations in how a global carbon price affects regional emissions and pollution exposure.We discuss these linkages in turn.First, imposing the carbon price lowers fossil fuel combustion and increases renewable and bioenergy uses across all world regions (Fig. 4a).Yet the extent of these changes depends on the current energy structures and projected technology costs.For instance, in 2050, the carbon price lowers the share of coal in the primary energy mix by 14 percentage points in India (reducing from 53% to 39% based on ensemble median; range across all SOWs: 11-17 percentage points) but only 5 percentage points in Canada (reducing from 12% to 7% based on ensemble median; range across all SOWs: 3-7 percentage points).This is consistent with the observation that India currently relies more heavily on coal 34 .The carbon price hence leads to a greater reduction in India's coal use in the model.In comparison, we find the increases in bioenergy shares are comparable across countries (for example, increases by 2-5 percentage points across the eight selected world regions, based on the ensemble medians).Here the small regional variations are largely driven by limited cross-region differences in bioenergy shares in current energy mixes and in future bioenergy supply curves assumed in the model.
How the changes in energy use affect air pollutant emissions depends on which sectors are affected and the stringency of pollution regulation in relevant sectors (Fig. 4b).For instance, carbon pricing leads to similar percentage reductions in coal share in Southeast Asia and the United States.Yet, the resulting reduction in per capita sulfur dioxide (SO 2 ) emissions is smaller in the United States due to more stringent pollution-control policies on existing coal facilities 35 .
Several types of precursor emission collectively influence the concentrations of ambient PM 2.5 .In most countries, the substantial reduction in SO 2 and nitrogen oxides (NO x ) emissions from fossil fuel uses is the dominant factor contributing to lower PM 2.5 concentrations, despite the slight increases in PM 2.5 emissions from the combustion of bioenergy.However, in some countries (for example, Canada and Russia), we find a substantial increase in OC emissions during the time frame of 2030-2060 under most of the considered SOWs, leading to a net increase in PM 2.5 concentrations (Supplementary Fig. 21 provides the trends of OC emissions from deforestation in Canada and Russia).The elevated OC emissions are an outcome of increased biomass production that intensifies land competition and increases the deforestation of the unmanaged forests (per capita land-use changes in Fig. 5a).These co-harm results depend on a range of model assumptions related to the bioenergy supply chain, energy-land interactions and deforestation practices (discussed in the next section).Finally, regional socio-demographic characteristics affect population vulnerability, influencing health outcomes.For instance, imposing a carbon price results in larger relative increases in Canadian PM 2.5 -attributable death rates than the associated PM 2.5 exposure levels.This is consistent with the combined effect of two factors.First, the nonlinear concentration-response relationships result in greater increases in mortality risks from one unit increase in PM 2.5 exposure in locations such as Canada where the air is already relatively clean (PM 2.5 concentrations without the carbon price are in Supplementary Fig. 15).Second, the increased population ageing in advanced economies such as Canada drives an increase in an elderly population that are more vulnerable to pollution exposure (Supplementary Table 9 provides the projected age structures in each region).

Drivers of potential health co-harms
Health co-harms can arise through changes in bioenergy production and the induced changes in land use and deforestation.To understand key factors and model assumptions that drive the modelled co-harms, we perform a model diagnostic analysis to systematically assess the model behaviours and the realism of key assumptions.Below we summarize the key insights related to the magnitude, location and timing of the health co-harms.A detailed assessment of all relevant model assumptions is presented in Supplementary Table 3.A simple feasibility check of key technology-related model outcomes is shown in Supplementary Section 5.
First, the deforestation method assumed in the model is a key assumption driving the OC emissions increases from deforestation and therefore the magnitude of health co-harms (Fig. 5).The model   applies historical emissions factors for global deforestation based on the Global Fire Emissions Database 36 , implicitly assuming most future deforestation would occur through slash and burn and thus emitting substantial amounts of OC.Yet, other methods of deforestation, such as clear-cutting, have become more prevalent and may be the preferred method to convert forests into bio-cropland 37 .We therefore conduct a sensitivity analysis on the deforestation method by evaluating an extreme case where all deforestation activities occur through clear-cutting and emit no OC.This assumption would avoid the increase in OC emissions from land use and deforestation, eliminating the health co-harms under all SOWs considered in our ensemble (Fig. 5c).This sensitivity analysis identifies that the deforestation method is a key factor in determining the magnitude of the potential health co-harms.
Second, location of the health co-harms is primarily affected by the model assumptions on the bioenergy supply chain and energyland interactions.Imposing a carbon price increases bioenergy use in most world regions.Yet, to meet the rising global demand, where bioenergy production would increase the most depends on two key model assumptions.First, biomass is largely assumed in the model to be globally traded (more in Supplementary Fig. 16).This assumption allows bioenergy to be produced in and exported from countries where it is cheapest to do so.Yet the real-world bioenergy markets are much more fragmented due to a range of logistical and market challenges 38 .Second, the land competition between bioenergy, food and forest is modelled based on the expected profitability of growing and selling bio-crops relative to other land-use options.Because countries such as Canada and Russia are assumed to have large areas of unmanaged forests in the model (Supplementary Fig. 17), converting forests to bioenergy land in these countries is more economically attractive compared to other countries (Fig. 5a).
Finally, the health co-harms are likely to be a more relevant concern for the first half of the century than the second half.For most countries with potential co-harms, we find the proportion of SOWs with health co-harms gradually increases from now to mid-century but then slowly disappears towards the end of the century.For example, we observe health co-harms (a higher PM  98% and 50% in 2050, then gradually decrease to 5% and <1% in 2100, respectively (more in Supplementary Fig. 22).The increase in the first half of the century is largely a result of increasing bioenergy use and production over time.The decrease in the second half of the century is largely an outcome of reduced land-use competition.Especially in the SOWs that assume large agricultural productivity improvements, less cropland is needed to meet food demand in the long-term future, lessening the land competition between food and bioenergy and the scale and emissions impacts from deforestation.

Discussion
Our analysis highlights the complexity of the system dynamics through which global climate mitigation can influence the regional distribution of pollution and health effects.While the direct health co-benefits from reducing fossil fuel use are well documented 1,5 , we demonstrate possible ways climate mitigation might increase air pollutant emissions and health risks in some regions 39 .The key pathway for co-harms identified in our study is that carbon pricing can increase PM 2.5 emissions, both from direct bioenergy combustion and, in a handful of countries, also from indirect land-use changes such as deforestation.Prior studies have also found intensified land-use competition in future mitigation scenarios that rely heavily on bioenergy 15 .While those studies demonstrated emerging risks for food security 40 and water stress 41 , our results suggest that unintended consequences can also occur for air quality and health.It underscores the importance of a comprehensive assessment for the sustainability implications of large-scale mitigation responses to climate change.
Examining the pathways for health co-harms are particularly relevant for countries that have reduced their coal dependence and where bioenergy might be pursued as a key decarbonization strategy.Prior studies have shown that the potential for health co-benefits from fossil reduction is often smaller in advanced economies than in the Global South countries due to already stringent pollution standards on existing fossil-based facilities 35,42 .Considering the potential energy-land interactions, our analysis suggests that health co-benefits from fossil reduction will become less prominent as countries advance towards In a, 'biomass' represents lands to grow bioenergy crops, and 'food purpose' represents lands for food crops and pasture.In b and c, we show the sensitivity analysis assuming deforestation occurs through open burning (main results as in Fig. 4d) versus clear-cutting.The box plots show the ensemble median, quartiles and range (N = 13,936 pairs of SOWs).As representative regions, we include two emerging markets that suffer from the highest pollution and health risks at present (China and India), two lower-income regions that may experience rapid economic growth and increasing pollution risks in the future (sub-Saharan Africa and Southeast Asia), two middle-income regions with vast areas of forest resources (Russia and Brazil) and two developed countries with cleaner air and large land areas (the United States and Canada).Supplementary Figs. 13 and 14 provide the results for 2030 and 2100.

Article
https://doi.org/10.1038/s41893-023-01133-5decarbonization, while the potential health co-harms from the mitigation actions will become an increasingly important consideration.Across all considered SOWs, we consistently find greater decreases in PM 2.5 -attributable death rates in countries currently facing higher health risks, such as China and India.For countries that may experience potential co-harms such as Canada, Russia and the United States, our analysis also identifies possible strategies (for example, changing deforestation method) that can eliminate the negative health impacts.This demonstrates that pricing carbon can improve global air quality while simultaneously reducing the current health inequity across countries.Such equity-improving outcomes can be expected under a wide range of plausible futures that vary in socio-economic trends, energy demand and technology costs and agricultural and land-use patterns.
Our analysis of how pricing carbon could affect health risks samples future uncertainties by considering a large ensemble of future SOWs.However, the links between large-scale climate mitigation, air pollution and the distribution of associated health impacts are still shrouded in considerable uncertainties not considered in this analysis.Our study does not resolve the effects of a wide range of additional uncertainties relating to model parameterization and structure 43 .For example, assuming that deforestation occurs through clear-cutting, rather than burning, effectively eliminates co-harms from our ensemble.Due to the complexity of the coupled multi-region, multi-sector system, our exploratory study takes a crucial initial step towards identifying the model assumptions and combination of uncertainties that might bias the conclusions about the health implications and distributional outcomes.Fruitful avenues for future research include an exploratory ensemble approach combined with scenario discovery methods and additional sensitivity analyses to identify future conditions or development trajectories resulting in a higher likelihood of health co-harms 26,27 .These methods can improve our understanding of failure modes of potential mitigation strategies and help to identify policy portfolios that are more robust to complex dynamics and deep uncertainties.
Our study is still silent on many important questions.For example, how can more refined strategies help to better navigate the complex landscape of climate, economics and health?A globally uniform carbon price has been used widely in models to represent climate policy, largely due to its simplicity and the appealing theoretical advantage as the most economically efficient way to achieve global decarbonization.However, real-world policies are more diverse and fragmented 44 .Regulations and sector-based measures are widely and typically adopted and nearly everywhere have a bigger impact on emissions abatement than directly pricing carbon 45,46 .Representing various types and combinations of policy instruments is particularly important in today's climate policy context as hundreds of countries now experiment with ways to reach net-zero emissions by mid-century.We hypothesize that these different policy designs would have different distributional consequences.For instance, compared to a subsidy on rooftop solar systems, electrifying the transport sector may bring greater benefits to populations living near major roads, who are often disproportionately minorities and people of lower socio-economic status 47 .In addition, the health co-harms identified in our analysis may also be mitigated by imposing land-conservation policies along with a carbon price on energy-sector emissions 48 .
A second open question is how much-needed improvements in the representations of health drivers, exposures and outcomes impact the conclusion.For instance, bioenergy is an important technology driver for the health co-harms observed in our study.Yet, our modelling approach considers only 12 land types for 384 land regions worldwide.A detailed, subnational representation of land-use patterns is essential to identify suitable land for bioenergy production and model the competition between different land-use purposes 49,50 .Assessing the disparities across socio-demographic groups, both for exposure and health outcomes, also requires fine-scale pollution simulation and health-impact assessment.While some studies are moving in this direction 51,52 , research that quantifies these linkages at decision-relevant resolutions is still largely in its infancy.These efforts can help in the search for decarbonization strategies that can simultaneously reduce adverse health impacts and associated inequities.
Our study lays the foundation for future efforts to address these open questions and advance our scientific understanding of the coupled energy-land-energy systems.Our work also has important policy implications.We assess the key drivers for the country-varying health effects of climate mitigation and identify potential cross-sector linkages (for example, between energy and land) that may lead to different distributional impacts.These insights are critically important, both for the international community and individual countries, to incorporate health and equity considerations into their climate mitigation efforts.

Construction of state of the world ensemble
We construct a large-scale exploratory ensemble of plausible future SOWs using a leading global-scale, process-based integrated assessment model, GCAM v5.4 18 (Supplementary Table 1 provides the uncertainties sampled in the ensemble and respective sampling strategies).GCAM is a multi-sector model with technology-rich representations of five systems and their interactions: energy, water, agriculture and land use (AGLU), economy and climate systems 18 .On the basis of varying input assumptions on socio-economic drivers, technology costs and policy actions, GCAM simulates the behaviours and interactions between these systems and projects future patterns at five-year intervals in a partial equilibrium economic modelling framework.For the GCAM version used in this study (v5.4), the energy and economy sectors are modelled for 32 world regions, the land system is divided into 384 subregions and the climate/physical Earth system is simulated by a reduced-form climate model, Hector 53 , at the global scale.
For the representation of the decision lever, we consider a simple policy design, that is, whether a globally uniform carbon price trajectory (Fig. 2a) is implemented from 2020 to 2100.This policy representation reflects the most economically efficient way to reach global decarbonization.Pricing carbon through a tax or a cap-and-trade system has also been widely adopted in many countries and regions to mitigate CO 2 emissions 22 .Our near-term carbon price levels are broadly consistent with the stringency of current carbon markets.We are aware that the real-world climate ambition and carbon price levels vary greatly across regions.For instance, the current carbon prices are US$9.20 ton −1 in China's national emissions trading system, US$13.89 in the Regional Greenhouse Gas Initiative, US$30.82 ton −1 in California and US$86.53 ton −1 the EU emissions trading system 22 .Meanwhile, most low-income and lower-to-middle-income countries do not have carbon prices 22 .In addition, the recent policy pledges from major economies to reach net-zero emissions by mid-century could substantially strengthen policy stringency in the decades to come.An improved representation of realistic policy choices is therefore an important area for future work 54,55 .
In this study, we sample seven broad categories of future uncertainties in socio-economic, technological and land-use aspects (details in the Supplementary Table 1), including: (1) population and gross domestic product (GDP) because demographics and economic development are fundamental drivers of future human activities, carbon emissions and health burdens; (2) price elasticity of energy demand because the demand for energy influences greenhouse gas (GHG) and air pollutant emissions from end-use and energy-supply sectors; (3) agricultural productivity and income elasticity for food because food demand and agricultural activities are key determinants for future land-use change and related emissions; (4) fossil fuel extraction costs because they affect the effectiveness of carbon price in driving down the demand; (5) low-emissions energy costs because they affect the competitiveness of low-carbon energy relative to fossil-based Article https://doi.org/10.1038/s41893-023-01133-5technologies; (6) carbon capture and sequestration (CCS) deployment costs because they determine the scale of CCS deployment and associated emissions under a carbon price; (7) water resource availability because it is a key limiting factor for energy and agricultural activities and therefore affects relevant GHG and air pollutant emissions.
We use a full factorial experimental design across these seven factors to encompass a wide range of plausible futures 23 .Among the seven, the first four (that is, socio-economics, energy demand, agricultural and land use, fossil fuel extraction costs) are sampled by considering five sets of assumptions that reflect the storylines of SSPs 56 .For the other three factors, we sample the future water run-offs using varying levels of groundwater level and reservoir capacity; and we sample the future competitiveness of low-emissions energy technologies and cost of CCS technology using varying levels of projected costs.The quantitative assumptions for different SSPs and technology costs are reported in refs.23 and 57 .
Combining one decision lever and seven types of uncertainty, we experimented with 30,000 SOWs (that is, 15,000 pairs with/without a carbon price).However, some SOWs do not yield feasible solutions.For example, the socio-economic assumption following SSP5 (fossil-fuelled development) is not compatible with AGLU assumption following SSP3 (regional rivalry).This is because SSP3 assumptions for AGLU include low agricultural technology development, restricted trade, lack of land-use regulations, but low agricultural productivity are formidable obstacles to achieving high-level socio-economic developments in SSP5.As a result, we have 14,526 model-solved SOWs without a carbon price and 14,180 model-solved SOWs with a carbon price.Between these two groups, we further pair up the SOWs with the same assumptions for other uncertainties and identify 13,936 pairs of SOWs that only differ in the decision lever.Supplementary Table 2 provides a comprehensive list of the numbers of solved SOWs categorized by the decision lever (carbon price) and each ensemble design factor.
Because GCAM is an open-source model (http://jgcri.github.io/gcam-doc/index.html),here we include only a short description of key energy and land-use assumptions that are particularly relevant for this study.In GCAM, the global trade of agricultural commodities is modelled using the Armington approach 58 , which assumes that products are differentiated by source and consumers treat goods from different countries as imperfect substitutes.The competition between imports and domestic production is governed by a logit sharing function in each regional market.Global trades of fossil fuels and bioenergy are also modelled using the Armington approach.To avoid unrestricted land conversion to bioenergy production, we apply the default GCAM assumption that protects 90% of all non-commercial lands (that is, non-commercial pasture and forest, grassland and shrubland) in each geographic land unit 59 .Renewable technologies, such as wind, solar and geothermal, are not traded.The market shares of different fuels/ technologies are governed by their relative or absolute cost difference through logit formulations 60,61 .The share weight parameters in the logit functions are resource-specific and calibrated using historical data.

Projection of GHG emissions
We project future emissions of annual total GHGs for 32 GCAM regions by technology and fuel choice.We estimate CO 2 emissions from fossil fuel and limestone uses by multiplying GCAM-projected production and consumption activities with the technology-specific emissions factors estimated from the Carbon Dioxide Information Analysis Center, which is a global inventory of historical carbon emissions from 1751 to 2017 62 .CO 2 emissions from land-use and land-cover change are estimated based on the areas of land-use change and the carbon intensity of each land-use type 63 .
When a carbon price is imposed, higher-carbon technologies become more expensive whereas lower-carbon technologies become more cost competitive.This cost difference (along with other noncost-related assumptions) determine the relative contribution of these technologies and fuel choices in meeting the demand in each economic sector such as electricity, transport, industry, residential and agricultural.
We also calculate the emissions of non-CO 2 GHGs, including methane, nitrous oxide and fluorinated gases by multiplying relevant activities with the emissions factors from the US Environmental Protection Agency 2019 64 .When the carbon price is imposed, the changes in non-CO 2 GHG emissions are proportional to the changes in activity level, except for the reductions in emissions intensity that are adjusted based on the exogenously assumed marginal abatement cost curves: where t stands for a five-year-period, E t is the non-CO 2 GHG emissions, A t is the activity level, EF t is emissions factor, MAC is the marginal abatement cost curve and E price t is the carbon price level.The regional MAC curves consider a wide range of various technologies and are derived based on the US Environmental Protection Agency 2019 database 64 .

Assessment of air pollutant emissions
We estimate the emissions of five types of air pollutant, including NH 3 , NO x , SO 2 , black carbon and OC, for 32 GCAM regions by technology and fuel choice.The emissions are calculated by multiplying relevant activities projected by the model with the respective emissions factors derived from historical data 18 .To account for the tightening of air pollution control policies over time, the future emissions factors are adjusted based on a declining trend with increasing income 65 .The technology mix is also adjusted over time by assuming a higher penetration rate of less polluting units 57,65 .Both adjustments vary across five SSPs.Specifically, for each pollutant emitted from each activity type: where t stands for a five-year period, E t is the air pollutant emissions, A t is the activity level and EF t is activity-specific emissions factor.EmCtrl represents the percent reduction in emissions factor as a result of emissions control, which is a function of per capita GDP, pcGDP t :

Steepness
where steepness is a technology-and air pollutant species-specific exogenous factor based on empirical evidence that determines the extent to which the changes in per capita GDP affect the stringency of emissions controls.

Assessment of climate outcomes
We model the climate system using the Hector model 53 , which interacts with the other parts of GCAM at every five-year time step.Hector is a reduced-form global climate carbon-cycle model, representing the most essential global-scale Earth system processes.The inputs to Hector are global total GHG emissions aggregated across all GCAM sectors and regions.Then Hector reports global average radiative forcing and temperature changes.

Assessment of ambient PM 2.5 concentrations
To assess the ambient PM 2.5 concentrations from precursor emissions, we use the Tracer Model version 5-FAst Scenario Screening Tool (TM5-FASST) 19 , a reduced-form source-receptor model for 56 world regions.TM5-FASST is derived from the Tracer Model version 5-Chemical Transport Model (TM5-CTM), a full chemical-transport model for which the nonlinear changes in pollution formation and wind transport are being considered 66 .The performance of TM5-FASST was evaluated in a prior publication 28 and demonstrates satisfying model capabilities in estimating ambient PM 2.5 concentrations.

Article
https://doi.org/10.1038/s41893-023-01133-5 To map from GCAM to TM5-FASST regions, we first downscale the emissions for 32 GCAM regions to 178 countries (Supplementary Table 4 provides GCAM sector mapping) by sector and for five types of precursor emission, using the country-to-region ratios based on the Emission Database for Global Atmospheric Research (EDGAR) data 67 (Supplementary Table 5 provides EDGAR sector mapping).We then re-aggregate country-level emissions to the 56 TM5-FASST regions.
For each year and state of the world, we estimate the PM 2.5 concentrations using the changes relative to 2000 as the base year, assuming a linear relationship between emissions and PM 2.5 concentrations and additivity across all types of emission and region.Specifically, the following equation is used: where C(y) and C base (y) are the ambient PM 2.5 concentration in receptor region y in a future year of interest and in 2000, respectively.E i (x) and E i,base (x) are the emissions of the air pollutant type i from a source region x in a future year of interest and in 2000, respectively.A i [x,y] is the source-receptor coefficient, capturing how the emissions of precursor air pollutant type i in source region x would influence the ambient PM 2.5 concentrations in receptor region y.n x is the total number of source regions whose emissions affect the ambient PM 2.5 concentration in receptor region y, plus two additional sources, shipping and aviation that are not tied to a particular location.i is the index for the type of precursor emission, which includes NH 3 , NO x , SO 2 , BC and particulate organic matter that are estimated from GCAM.n i is the total number of precursors that form ambient PM 2.5 .The unit of the PM 2.5 concentration is μg m −3 , and the units of the emissions are kTonne per year.
Because TM5-FASST model uses the year 2000 as the base year, the values for E i,base (x) are taken from the representative concentration pathway database for the year 2000 at 1° × 1° resolution; 19 using 2000 emissions as input, C base (y) is estimated using a full chemical-transport model TM5-CTM 66 , also at a global 1° × 1° resolution.The values in the source-receptor matrix A are derived from a series of perturbation runs that increase the precursor emissions by 20%, by precursor type and source region and assess the implications on PM 2.5 concentrations in each receptor region.
Although the simplicity of the TM5-FASST model enables our assessment of nearly 30,000 SOWs for many years, there are caveats when using this model for assessing future pollution levels.For instance, it does not consider how a changing climate might influence the pollution formation and transport in the future.It also assumes linear relationships between precursor emissions and resulting PM 2.5 concentrations, which simplifies the chemical and physical processes in the atmosphere.To evaluate the performance of TM5-FASST for modelling future pollution levels, we compare the projected PM 2.5 concentrations and health risks using TM5-FASST with the results from the Earth System Model Version 4.1 (ESM4), a high-resolution chemistry-carbon-climate model developed by Geophysical Fluid Dynamics Laboratory.The results from the two models are broadly consistent with each other, suggesting that the TM5-FASST model produces reasonable estimates for PM 2.5 concentrations (more in Supplementary Section 1.4).

Assessment of PM 2.5 -attributable deaths
Following the approach in the Global Burden of Disease (GBD) Study 2 , we consider six diseases that have been found to be associated with long-term exposure to ambient PM 2.5 , namely chronic obstructive pulmonary disease, diabetes mellitus type II, ischaemic heart disease (IHD), lung cancer, lower respiratory infections and stroke.
For each of the five-year age groups from 0 to 95+ in each of the 178 countries, we calculate the premature deaths attributable to each of the considered six diseases using the following equation: where y 0 is the age-and disease-specific baseline mortality rate; Pop is the size of the exposed population in each age group; AF is the attributable fraction, which changes with varying exposure levels to PM 2.5 concentration (c) in each region.
Below we describe the data source and calculation methods for each parameter.More detailed information about the input data is provided in Supplementary Table 6.
For population (Pop), we use age-specific population projections from the International Institute for Applied Systems Analysis (IIASA) SSP database 56 .The population projections are at the country level, with five-year intervals from 2010 to 2100, and vary across the five SSPs.
For baseline mortality rates (y 0 ), we use the age-specific baseline mortality rates for each country projected by the International Futures (IFs) model v7.64 68 , which also varies across the five SSPs.The baseline mortality rates from IFs are projected based on the GDP per capita and education-attainment level and calibrated using the GBD 2004 data for cardiovascular diseases, diabetes, malignant neoplasms, respiratory diseases and respiratory infections.We map IFs-reported rates onto the six considered diseases: for IHD and stroke, we use the rates for total cardiovascular disease from IFs and multiply by the shares of IHD and stroke in total cardiovascular-disease-related deaths; for lung cancer, we use the rates for malignant neoplasms; for chronic obstructive pulmonary disease, we use the rates for respiratory disease; for lower respiratory infections, we use the rates for respiratory infections; and for diabetes mellitus type II, we use the rates for diabetes.To check the validity of this mapping method, we compared the disease-specific baseline mortality rates calculated using our methods with the rates reported by the GBD study and found them to be largely consistent (Supplementary Table 7 provides the comparison).
For AF, we calculate the AFs for each disease and age group using the following equation: where c is the annual mean PM 2.5 concentration in each country and RR is the disease-specific relative risk.The annual mean PM 2.5 concentration c is simulated by TM5-FASST for 56 regions.We further assume all countries within the same TM5-FASST region have the same exposure level.The relative risks are obtained from the GBD study 2019 3 and derived from the Integrated Exposure-Response model 20 for the six types of disease for the PM 2.5 exposure levels from 0 to 600 μg m −3 .The RRs are age specific for IHD and stroke (from 25 to 95+ at five-year intervals) and are for all age groups for the other four diseases.

Assessment of the distributional and equity implications
We consider different measurements of equitable distribution and equity-improving distribution that vary in two dimensions, the metric of health outcomes and the definition of health equity.Below we include a brief summary with more details summarized in Supplementary Table 8.
Regarding the metric of health outcomes, our main results focus is on health risks measured by the PM 2.5 -attributable death rate.We also consider two alternative metrics: health exposure measured by PM 2.5 concentrations (Supplementary Fig. 3) and health burden measured by PM 2.5 -attributable deaths (Supplementary Fig. 4).
Regarding the definition of health equity, our main results focus on regional variations and consider 'equity-improving' outcome as regions that currently face higher health risks benefit more from carbon pricing.We further consider two alternative definitions that focus on variations across different country income groups and global age groups.Here 'equity-improving' outcome requires lower-income regions or elderly populations as more vulnerable regions/groups benefit more

Fig. 1 |
Fig. 1 | Pathways for a global carbon price to influence climate, health and equity outcomes.a, Conceptual mental model of relevant influences, feedbacks and system interactions.b, Pathways captured by our integrated climate-energy-health model and the uncertainties sampled in our exploratory ensemble.We use the GCAM model 18 to sample future SOWs and implement

dFig. 2 |
Fig.2| Impacts of a global carbon price on future global average temperature and regional distribution of PM 2.5 -attributable death rates.a, The carbon price trajectory from 2015 to 2100 considered in this study; the black dot highlights the price level in 2050 (US$69 ton −1 CO 2 ).b,c, The global average temperature increase relative to the 1850 level and the global annual PM 2.5attributable death rates, including the median and ranges of the SOWs with and without a carbon price (N = 14,180 and 14,526, respectively; the different sample sizes are because some combinations of input assumptions result in infeasible solutions; Supplementary Section 1.2).The borders of the belts are the ranges across all SOWs in each projected year.The box plots on the far right show the ensemble distributions in 2050 and 2100.d, Changes in PM 2.5 -attributable death rate in 2050 due to the carbon price (N = 13,936; limiting to the pairs of SOWs that have feasible solutions in both cases).'Consistent effects' indicate the same direction of effects (that is, co-benefits or co-harms) across all the SOWs, whereas 'Potential effects' show mixed effects across SOWs.The thicker border lines show the 32 GCAM regions (except Antarctica) for which the energy/land activities and associated emissions are simulated, whereas the lighter border lines show 178 regions and countries for which the health-impact assessments are performed69 .Supplementary Figs. 9 and 10 provide the spatial distribution for 2030 and 2100.

Fig. 3 |
Fig. 3 | Distribution of PM 2.5 -attributable death rates across regions in 2050.a, PM 2.5 -attributable death rates without a carbon price.b, Changes in the PM 2.5attributable death rates as a result of a carbon price.The circles and error bars represent the ensemble medians and ranges for 31 world regions consistent with GCAM regions (excluding Taiwan due to lack of data; N = 13,936 for a and b).The Article https://doi.org/10.1038/s41893-023-01133-5

Fig. 4 |
Fig. 4 | Regional changes in the health drivers, exposures and risks in response to the considered global carbon price in 2050.a, The changes (by percentage points) in shares of coal, biomass and renewables in the primary energy mix (Supplementary Fig.18provides results for the 32 GCAM regions).b, The changes in OC and SO 2 emissions per capita per year (Supplementary Fig.19provides results for the 32 GCAM regions).c, The changes in annual average PM 2.5 concentrations (Supplementary Fig.20provides global regionallevel distributions).d, The changes in PM 2.5 -attributable death rates.The box

cFig. 5 |
Fig. 5 | Regional changes in land use, OC emissions and health risks as a result of the considered global carbon price in 2050.a, Changes in land-use types (by percentage points).b, Changes in OC emissions in the residential sector and the agriculture and land-use sectors.c, Changes in PM 2.5 -attributable death rates.In a, 'biomass' represents lands to grow bioenergy crops, and 'food purpose' represents lands for food crops and pasture.In b and c, we show the sensitivity analysis assuming deforestation occurs through open burning (main results as in Fig.4d) versus clear-cutting.The box plots show the ensemble median, 2.5 -attributable death rate associated with carbon pricing) in 20% of all SOWs for Canada in 2025 and 15% of all SOWs for Russia in 2035.These proportions increase to