Introduction

China is facing dual challenges related to air pollution and climate change1,2,3. To address the rising challenge of climate change, the government of China has pledged to reach its peak in carbon emissions before 2030 and achieve carbon neutrality by 20604. Despite substantial improvements over the past 30 years, China is still actively mitigating severe air pollution, which causes 1.4 million premature deaths annually5. Notably, around 80% of these deaths are linked to ambient particulate matter with a diameter of 2.5 micrometers or smaller (PM2.5), which exacerbates mortality risks associated with ischemic heart disease, cerebrovascular disease, chronic obstructive pulmonary disease, lung cancer, and lower respiratory infections6,7. China is currently prioritizing synergistic control to address both issues because they share common sources (e.g., fossil fuel combustion)4,8. Measures, including improving energy efficiency, adjusting industry structures, and increasing electrification from renewable energy, can all reduce air pollutant emissions and achieve health co-benefits.

Studies have shown that the implementation of climate policies can help improve air quality and reduce associated health losses9,10. The control over air pollution can also boost decarbonization11. For example, researchers have estimated that CO2 emission reductions following the Representative Concentration Pathways 4.5 scenario (RCP4.5, which depicts a moderately warming future) can help prevent 1.3 million premature deaths globally by 2050 by reducing air pollutant emissions simultaneously9. This number could double under scenarios in which more stringent CO2 emission control is applied12. However, climate policies may not adequately address air pollution-driven health losses, especially under unequal low-carbon development conditions that overlook less-developed regions13. Despite the promising air quality and health co-benefits attained from implementing climate policies, recent scenario analyses suggest that China cannot meet the World Health Organization air quality guidelines by implementing only climate policies11,14.

Coordinated solutions to dual challenges must be informed by location- and sector-specific health co-benefits. The extent of co-benefits should vary substantially depending on the proximity of sources to densely populated regions, source strength, end-of-pipe technology applied, and atmospheric conditions15,16. However, the location-specific source attribution of national health co-benefits is a key knowledge gap in scenario-based analyses, which quantify co-benefits only at aggregated levels11,14. Recent finer-resolution studies have revealed substantial spatial variation in health co-benefits to motivate spatially nuanced mitigation measures. However, limited by the high computational cost required for high-resolution analyses, these studies have focused only on limited sectors16,17,18.

In this study, we perform a detailed quantitative analysis to delineate the contributions of diverse sectors and their associated fuel/process combinations to CO2 emissions and the health damage related to air pollution in China. This analysis encompasses critical sectors such as energy production, industry, transportation, domestic activities (both residential and commercial sources), and agriculture. The health impact assessment focuses on nationwide premature mortality attributable to ambient PM2.5 exposure. Our coordinated source attribution analysis integrates coupled emission inventories for both air pollutants and CO2 with an advanced adjoint tool developed for a regional air quality model, the Community Multiscale Air Quality (CMAQ) model. This tool allows for the efficient calculation of backward sensitivities, thereby quantifying the impact of high-resolution emission changes on nationwide premature mortality counts. The analysis includes impact of emission from speciated primary PM2.5, which include organic carbon (OC), elementary carbon (EC), and other primary PM2.5 particles, as well as from principal precursors of secondary PM2.5, which include sulfur dioxide (SO2), nitrogen oxides (NOx), and ammonia (NH3). Moreover, we propose using the integrated costs – representing the aggregate of monetized social costs from health damage and climate change – as a comprehensive metric for prioritizing sectoral and spatial control targets.

Results

Substantial disparities in sectoral contributions to health damage and CO2 emissions

The source contribution of health-threatening air pollutants differs from that of CO2, although both mainly originate from fossil fuel combustion. An analysis of the coordinated emission inventories for CO2 and air pollutants for the year 2017 reveals that 86% of the total anthropogenic CO2 emissions were predominantly from the energy generation and industrial sectors. Notably, coal combustion was the major contributor, accounting for 97% and 50% of the total CO2 emissions in these two sectors, respectively (Fig. 1a). However, as a result of implementing multiple pre- and post-combustion control technologies19, coal combustion in these sectors contributed 17% of the primary PM2.5 emissions (Supplementary Fig. 1). Coal and solid-biomass combustion in the domestic sector contributed 38% of the total PM2.5 emissions and 22% of the total SO2 emissions. This substantial contribution is attributed to the restricted air supply, poor mixing, and insufficient emission control in household stoves, coupled with the use of low-quality fuels20. Adjoint-based source attribution indicates that more than a quarter of premature deaths attributable to ambient PM2.5 exposure originate from the domestic sector, despite its mere 4% contribution to CO2 emissions (Fig. 1b). In contrast, while the energy generation sector is an important contributor to CO2 emissions, it accounts for only 10% of premature deaths. The finding that the domestic sector is the leading contributor to health losses attributable to ambient PM2.5 exposure is consistent with findings in previous studies21,22. Certain subsectors contribute to health losses without being directly linked to CO2 emissions from fuel combustion. These include non-combustion industrial processes, agricultural fertilizer application and livestock management, and crop residue burning in the domestic sector (Supplementary Table 1). While the combustion of crop residues does emit CO2, this subsector is considered carbon neutral based on the assumption that the released CO2 is later reabsorbed through subsequent biomass regrowth23. As in our approach for other fuel types, life cycle emissions from biofuel processing and land use changes are not included24,25. We term these subsectors “non-synergistic subsectors”, which account for 23% of the total premature deaths attributable to the five sectors assessed.

Fig. 1: Disparities in source attributions of CO2 emissions and health damage.
figure 1

a Source contributions to CO2 emissions in China in 2017 based on the coordinated CO2 emission inventory with the air pollutants. The lower bar represents the contributions of different sectors, and the upper bar represents the contributions of different fuel/process categories. b Source contributions to premature deaths attributable to long-term ambient PM2.5 exposure in China in 2017 based on the Adjoint simulation of the Community Multiscale Air Quality modeling. The left bar represents the contributions of different sectors, and the right bar represents the contributions of different fuel/process categories. c Comparison of impacts on health damage and CO2 emissions for each subsector. The corresponding monetized impacts are also shown on the minor axis based on a uniform statistical life value (1.33 million US dollars per premature death) and social cost of carbon (100 US dollars per ton of CO2 emission). Source data are provided as a Source Data file.

Full size image

Seven subsectors were identified as significant contributors to health damage, each of which was attributable for more than 100,000 premature deaths (Fig. 1c). These subsectors include bituminous coal combustion in the domestic, energy generation, and industrial sectors; emissions from diesel-powered vehicles; and activities in hydraulic cement production, iron and steel production, as well as agricultural fertilizer application and livestock management. While these subsectors comparably contribute to health damage, their contributions to CO2 emissions differ substantially. For instance, agricultural activities significantly contribute to PM2.5-related health damage from NH3 emissions without being major contributors to CO2 emissions directly. This discrepancy is also observed in the analysis of CO2 emissions from other subsectors. For example, the CO2 emissions from domestic bituminous coal combustion are 52–93% lower than those from the remaining five subsectors. In terms of the speciated emission contribution to health damage, more than half of the health damage from hydraulic cement and iron production, as well as domestic bituminous coal combustion, is attributed to primary PM2.5 emissions. In contrast, health damage from bituminous coal combustion in the energy generation and industrial sectors, as well as from diesel vehicles, is predominantly attributable to SO2 and NOx emissions (Supplementary Fig. 2). The impact of various fuel types and industrial processes on premature deaths within a sector also shows substantial disparities. For example, diesel-powered vehicles, which are responsible for 210,000 premature deaths, contribute eightfold more to health damage than do gasoline-powered vehicles, which emphasizes the great importance of diesel vehicle electrification in reducing air pollution and associated health risks. Overall, the ratio of source contributions to health damage versus to CO2 emissions varies extensively across subsectors, ranging from 0.094 in the industrial combustion of dry natural gas to 22 in the domestic combustion of unorganized waste. This noteworthy variation in source profiles for health damage and CO2 emissions underscores that strategies aimed at reducing air pollution-related health damage may not always align with those targeting decarbonization. This necessitates the formulation of coordinated, synergistic control plans to effectively balance the two objectives.

Spatial heterogeneity in contributions to health damage and CO2 emissions

The integration of high-resolution health damage attribution data with the coordinate CO2 emission inventory also enables the delineation of spatial heterogeneities in contributions to health damage and CO2 emissions. By calculating the ratio of the percentage contributions to the nationwide PM2.5-related health damage versus to CO2 emissions at the gridded or regional levels, we can assess the relative significance of the contributions. In densely populated areas (Supplementary Fig. 3), including the eastern and central regions and the Sichuan Basin, the contribution to health damage notably surpassed that to CO2 emissions (Fig. 2a). The highest ratio was observed in Hubei (3.9), followed by Henan and Chongqing (3.8 and 3.7, respectively), whereas the lowest ratio was recorded in Xizang (0.11). Spatial heterogeneity in the source contributions to health damage and CO2 emissions is also evident within individual sectors. For instance, in the energy generation sector, Inner Mongolia contributes marginally more to CO2 emissions (3.8%) than Shandong (3.5%), yet its impact on health damage (0.75%) is significantly lower than Shandong’s (1.3%) (Fig. 2b). Similarly, in the industrial sector, Guangdong’s CO2 emissions surpass those of Anhui and Hebei (18%), yet its health damage contribution is merely half of that observed in the latter provinces. This spatial variability is influenced not only by population density but also by the intensity of air pollutant emissions (Supplementary Fig. 4). For example, Sichuan, with its lower coal quality leading to higher emissions of PM2.5 and SO2 per ton of CO2 emission, exhibits the most pronounced disparity in the energy generation sector. Similarly, the elevated PM2.5 emissions associated with clinker production in the hydraulic cement manufacturing make Anhui and Hebei the regions with the highest discrepancy in the industrial sector.

Fig. 2: Disparities in spatial distributions of CO2 emissions and health damage.
figure 2

a Ratio between the gridded contribution to PM2.5 exposure-related health damage and to CO2 emissions. The provincial boundary shapefile is obtained from Harvard Dataverse (https://doi.org/10.7910/DVN/DBJ3BX) and is publicly available under the Creative Commons CC0 Public Domain Dedication. b Comparison of provincial contributions to health damage and CO2 emissions in each sector. The red dots represent the ratio of provincial contributions to health damage versus to CO2 emissions. The provinces are ranked from large to small in population size. c Correlation between population density (pop) and the ratio of each city’s contributions to health damage versus to CO2 emissions (ratio). The dot color represents the domestic solid fuel usage fraction, which is defined as the fraction of energy consumed as solid fuels in the domestic sector out of the total province-level energy consumption. The solid line and dashed line represent the correlations for cities with domestic solid fuel consumption above and less than 1%, respectively, of the total provincial energy consumption. Source data are provided as a Source Data file.

Full size image

The spatial distribution of the health damage to CO2 emissions ratio is significantly influenced by population density. A log-linear regression analysis at the city level shows positive correlation between population density and the ratio of each city’s percentage contribution to nationwide PM2.5-related health damage versus its contribution to CO2 emissions (Fig. 2c). These findings are consistent with previous studies showing that the regional health impact of ambient air pollution and the effectiveness of mitigation measures are more important in densely populated regions26,27. Additionally, the source profile, particularly the domestic energy structure, contributes to the spatial variations in the ratio. The color-coded scatter plot in Fig. 2c illustrates the impact of reliance on solid fuels for domestic energy needs on the correlation between the ratio of contributions to health damage versus that to CO2 emissions and population density. Notably, this correlation intensifies in cities with a greater dependence on solid fuels for domestic energy, explaining more than half of the variance in city-level ratios. Specifically, a 1% increase in population density correlates with a 0.47% increase in the ratio. Conversely, in cities where domestic solid fuel consumption constitutes less than 1% of total provincial energy usage, population density accounts for merely 4% of the ratio variance, with a 1% population increase leading to a modest 0.12% increase in the ratio. The higher ratio in Beijing than in other megacities, including Shanghai, Guangzhou, and Shenzhen, further demonstrates the critical role of domestic solid fuel consumption (Supplementary Fig. 5). Although Beijing has already decommissioned its energy-intensive industries and coal-fired power plants, it is the only city among the four megacities whose contribution to health damage exceeds its contribution to CO2 emissions. This increase in Beijing’s contribution to health damage was primarily attributed to the consumption of 1.8 Mt of raw coal by rural residents in 201728.

Monetized social costs of CO2 emissions and PM2.5 exposure-related health damage

We further monetized health damage and climate impacts by applying a uniform value of statistical life (VSL) of 1.33 million US dollars per statistical death and a direct social cost of carbon (SCC) of 100 US dollars per ton of CO2 emission (Methods). The monetized costs for health damage and climate impacts attributable to different subsectors are depicted along the minor axis in Fig. 1c. For the 36 synergistic subsectors, the monetized health damage and CO2-related climate impacts fall along the one-by-one line on a logarithmic scale, with the monetized health damage exceeding the corresponding monetized climate impacts for half of the subsectors. This finding indicates that the social costs due to climate change are comparable to the monetized concurrent health losses attributable to air pollution. The VSL value employed in this study, following the Organisation for Economic Co-operation and Development (OECD) recommended benchmark, is 2–3 times lower than the estimates recommended by the United States Environmental Protection Agency (USEPA). If higher VSL estimates were used (Supplementary Fig. 6), monetized health damage would surpass climate impacts for the majority of subsectors. This alignment, or in certain cases, the excess of monetized health damage related to climate impacts, emphasizes the near-term health benefits as an important incentive for CO2 emission reduction12,29 and justifies more ambitious decarbonization plans.

We further conducted a comparison between the monetized health damage and climate impacts with sectoral gross domestic product (GDP), assigning production-based emissions to 42 economic sectors by using the China multi-regional input-output model table for 201730 (Supplementary Note 1, Supplementary Table 2). On average, the integrated costs of the 42 economic sectors equal 20% of the total GDP. For most of the economic sectors, the integrated costs are lower than 5% of the sectoral GDP (Supplementary Fig. 7). However, in four sectors, the integrated costs exceed the sectoral GDP. These sectors include the production and distribution of electric power and heat power; the manufacture of non-metallic mineral products, the processing of petroleum, coking, and nuclear fuel; and the smelting and processing of metals. This comparison further underscores the critical need for decarbonizing electricity generators and energy-intensive industrial processes.

Integrating costs from health damage modifies control priorities

Hence, we suggest employing a unified indicator, calculated as the sum of social costs from CO2-related climate change and PM2.5 exposure-related health damage. This indicator is proposed for harmonizing sectoral and spatial control objectives in co-optimal mitigation strategies. The model estimates that in 2017, the integrated costs for seven subsectors exceeded 100 billion US dollars. The highest integrated costs were associated with bituminous coal combustion in the energy generation sector, followed by industrial consumption of bituminous coal, diesel-powered vehicles, hydraulic cement production, iron and steel production, domestic consumption of bituminous coal, and agricultural fertilizer application and livestock management (Supplementary Table 1). Except for the last non-synergistic subsector, the other six subsectors all heavily relied on coal or diesel fuel. The social cost from climate change accounted for more than two-thirds of the integrated benefits from bituminous coal combustion in the energy generation sector, while the social cost from health damage accounted for 47%–89% of the integrated costs in the other five synergistic subsectors. Mitigation measures oriented toward air pollution control have been applied in these five subsectors, including household fuel switching and after-treatment technologies adopted for reducing exhaust emissions31,32. Their important contribution to the integrated costs encourages the implementation of synergistic control measures to mitigate CO2 and air pollutant emissions. Because of their substantial contribution to health damage, the ranking position of domestic subsectors reliant on solid fuels increased in the integrated cost-based ranking (Supplementary Table 1). In addition to the domestic bituminous coal combustion, indoor crop residue burning and brush wood burning were positioned 9th and 10th, respectively, out of the 53 subsectors assessed in terms of their contribution to integrated costs. However, decarbonization-oriented mitigation measures may not address these two subsectors due to their negligible contribution to CO2 emissions.

Spatially, the distributions of social costs from health damage and climate change, as well as the integrated costs, were uneven (Fig. 3c). The distributions of the contributions of climate change to social costs are skewed toward less populated regions, while the distributions of social costs from health damage are skewed toward densely populated regions (Fig. 3d). The incorporation of health co-benefits amplified the contribution of high-population regions to integrated benefits because of the proximity of air pollution and exposed populations. The areas with high integrated costs were concentrated in populated areas where greater amounts of air pollutants and CO2 emissions were emitted from fuel combustion to support intensive human activities. The grids that contributed more than 1 billion US dollars in integrated costs covered only 11% of the land but hosted 60% of the population of China. Spatial analysis at 36 × 36 km resolution identified densely populated cities as the primary contributors to integrated costs, including Chongqing, Zhengzhou, Shanghai, Wuhan, and six cities in the Beijing, Tianjin, and Hebei (BTH) city cluster. Among these populated cities, monetized health costs typically comprised 51%–84% of the integrated costs.

Fig. 3: Spatial distribution of social costs from PM2.5 exposure-related health damage, CO2-related climate change, and integrated costs.
figure 3

a Spatial distribution of PM2.5 exposure-related health damage in China at a 36 km by 36 km resolution in 2017. b Spatial distribution of social costs from CO2-related climate change. c Spatial distribution of the integrated costs, which equal the sum of the monetized health damage and social cost from CO2-related climate change. The provincial boundary shapefile used in a-c is obtained from Harvard Dataverse (https://doi.org/10.7910/DVN/DBJ3BX) and is publicly available under the Creative Commons CC0 Public Domain Dedication. d Quantile‒quantile plot of the social cost distribution against the population distribution. The three lines represent comparisons of social costs from CO2-related climate change, PM2.5-related health damage, and integrated costs. Source data are provided as a Source Data file.

Full size image

Consequently, the spatial priorities for synergistic control identified based on the integrated costs differ from those based solely on the social cost of CO2-related climate change. In addition to the emission amount, the integrated costs depend on the population density, energy and sector structure, and atmospheric conditions. Regarding provinces, the rankings based on the social costs from climate change were high in heavy industrial and energy-supplying provinces, including Jiangsu, Guangdong, and Inner Mongolia (Fig. 4). Because of the lower contribution to social costs from health damage, the rankings for these provinces decreased when they were sorted by integrated costs. The provincial contributions to social costs from health damage were lower in Jiangsu and Guangdong Provinces, China’s most developed industrial hubs, because these provinces have made greater efforts to upgrade their industrial structure to curb emissions33. In Inner Mongolia, the contribution to health damage is lower because of the lower population density, despite the heavy reliance on coal. In contrast, because of the disproportionate regional contribution to health damage in densely populated areas, the rankings based on integrated costs prioritized inland provinces with high population densities. The gap between the rankings based on the integrated costs and social costs from climate change is the largest in Chongqing, one of the most densely populated cities.

Fig. 4: Sectoral contributions to social cost from CO2-related climate change and integrated cost in each province in mainland China.
figure 4

The provinces were ranked in descending order based on their contributions to the social cost of CO2-related climate change (left column) and the integrated cost (right column). The sectoral contributions in each province are plotted in the bar chart. The provinces are color coded based on their predominant subsectors. Source data are provided as a Source Data file.

Full size image

In six provinces, namely, Henan, Liaoning, Tianjin, Beijing, Hebei, and Shanxi, the primary subsector contributing to integrated costs diverges from the leading subsector contributing to social costs from CO2-related climate change. In Henan and Liaoning, the predominant subsector shifts from being bituminous coal combustion in the energy generation sector to being bituminous coal combustion in industrial boilers, and in Tianjin, the predominant sector changes to iron and steel production. Although the energy generation sector is the most studied sector for low-carbon pathways16,34, the combined integrated costs from bituminous coal combustion and five energy-intensive processes (iron and steel, coke, brick, hydraulic cement, and lime production) dominated in 26 out of the 31 provinces. Thus, programs to decarbonize industrial boilers and phase out energy-intensive and highly polluting industrial processes are the keys to optimizing low-carbon development in most provinces. For Beijing, Hebei, and Shanxi, the top contributing subsectors shift to domestic bituminous coal combustion, a shift attributed to its substantial contribution to health damage.

Discussion

The substantial contribution of monetized health damage to integrated costs suggests that sectoral control priorities for addressing air pollution may not align with the primary targets of decarbonization plans. Notably, the combustion of bituminous coal for energy generation significantly contributes to the social costs of CO2-related climate change, and its health co-benefits are well studied16,34. However, in terms of health damage attributable to PM2.5 exposure, its contribution is comparatively less than that of industrial and domestic bituminous coal combustion and diesel vehicle emissions. Our assessment indicates that phasing out existing coal-fired power plants in mainland China in 2017 would reduce national losses from health damage by only 5% due to their minor role in reducing air pollutant emissions. Moreover, the health co-benefits of decarbonizing large-scale boilers could diminish if coal-fired power plants extensively adopt ultralow emission control technologies35. Scenario analyses based on Nationally Determined Contributions (NDCs) also reveal that climate policies alone are insufficient to achieve the air quality standards necessary for public health protection11,14. Conversely, despite a minimal contribution to CO2 emissions, fuel switches in the domestic sector are prioritized when health damage is integrated into the social cost assessment. The social cost of health damage significantly contributes to the integrated costs in regions such as Beijing, Hebei, and Shanxi. The clean winter heating plan in Northern China launched in 2017, which targets coal-fired household stoves in Beijing, Tianjin, and 26 surrounding cities, has reduced domestic coal combustion in these areas36. Domestic solid fuel combustion also contributes more than 20% of the total integrated costs in northern provinces such as Jilin, Heilongjiang, and Gansu, justifying expanded residential fuel switch programs in less-developed northern regions.

A comparison of integrated costs at 36 km by 36 km resolution facilitates the development of nuanced decarbonization strategies tailored to regional, provincial, and city levels. The spatial control priorities based on integrated costs differ from those guided solely by monetized climate impacts. Unlike long-term greenhouse gases, PM2.5 and its precursors, which have shorter lifetimes, significantly impact health depending on the proximity of the source37. Including monetized health damage in integrated costs highlights the need to focus on emission control in densely populated areas such as Shanghai, Chongqing, Zhengzhou, and the BHT city cluster. Urbanization offers opportunities for climate change mitigation through improved energy efficiency and public transportation38 but also concentrates emissions, increasing air pollution-related health risks in cities39. A lower dependency of health damage on population density in cities with minimal domestic solid fuel consumption endorses initiatives such as the clean heating campaign to mitigate urbanization’s impact on air quality. These findings support the integration of the Low-Carbon Cities Program into China’s NDC and its global application40,41, emphasizing population density in pilot program selection for enhanced health co-benefits. Neglecting health impacts in mitigation investments, such as carbon trading, could lead to prioritizing CO2 emission reductions in less-developed western regions because reducing emissions is more cost-effective in regions with outdated combustion and control technologies42.

The sectoral and spatial disparities in the source contributions to health damage and CO2 emissions identified in our study underscore the necessity for integrated assessments at a high spatial resolution to inform balanced strategies that address both air pollution and climate change mitigation. CMAQ Adjoint’s ability to assess the marginal benefits of emission reductions on a granular scale enables high-resolution source attribution, accounting for emission profiles, population density, and atmospheric conditions. The Sichuan Basin, for instance, is highlighted as an important contributor to health damage but not to CO2 emissions, illustrating the nuanced insights provided by adjoint-based source attribution, which accounts for the complex interplay of factors, including population density and terrain, which can exacerbate PM2.5 accumulation and secondary aerosol formation43.

Our results are subject to several limitations and uncertainties. First, limited by the availability of up-to-date high-resolution emission inventories with detailed source information, our source attribution analysis is based on adjoint sensitivities simulated for the year 2017. Considering China’s recent advancements in air pollution control, such as the adoption of ultra-low emission standards in power plants and the clean heating campaign, changes in the emission profile could alter the source attribution results. Second, our health impact assessment focuses exclusively on premature deaths due to ambient PM2.5 exposure, and excludes indoor exposures. The contribution of emissions from domestic solid fuel combustion to health damage is likely to be more pronounced when considering indoor exposure, given that both emission and exposure predominantly occur within indoor environments44,45,46. Third, the integrated cost assessment is sensitive to the monetized values of health damage and climate impacts. There is considerable variability in both the VSL and the SCC estimates, spanning several orders of magnitude. The global SCC used in this study is greater than the country-level SCC, and reflects the worldwide impact of CO2 emissions47. This implies that the relative importance of health damage, and thus the focus on mitigating pollution from heavily polluted sources, would be more pronounced when considering local social costs. Future studies constraining the monetized estimates of VSL and SCC are fundamental to ensuring robust and informed policymaking.

Methods

Model configuration

We used CMAQ v.5.0 with its recently developed adjoint model (CMAQ Adjoint) to simulate ambient PM2.5 concentrations and the sensitivity of population health loss to related air pollutant emissions in China in 2017. The CB05-AERO5 chemical mechanism was used in the CMAQ simulations. The CMAQ Adjoint implements the full adjoint of CMAQ v.5.0, including the discrete adjoint for gaseous chemistry, aerosol formation, cloud chemistry and dynamics, and diffusion and the continuous adjoint for advection48. Its detailed treatment of multiphase reactions for size-resolved aerosols enables adjoint sensitivity analysis of PM2.5-related health impacts. CMAQ has been widely applied and validated for simulating ambient PM2.5 concentrations and assessing source impacts across various regions, including China49,50. The CMAQ Adjoint has also been successfully applied in source attribution for air pollution-related health impacts and inverse modeling51,52.

The meteorological fields used to drive the CMAQ and CMAQ Adjoint simulations were downscaled from 0.5° × 0.5° global weather forecast products from the National Centers for Environmental Prediction Global Forecast System using the Weather Research Forecasting Model v.3.4.153,54. The 2017 AiMa emission inventory was used as the emission input55. The inventory provides constrained bottom-up emission data using ground measurements and satellite observations. It has been widely used in previous studies and air quality forecasting services in China53,56. The study domain covers China at a 36 km by 36 km horizontal resolution (Supplementary Fig. 8), and 13 vertical layers that extend ~16 km above the ground.

We evaluate the performance of the CMAQ model by comparing the simulated and observed PM2.5 concentrations at the 1504 monitoring sites across China. The comparison showed satisfying model performance according to the recommended benchmarks, with a Pearson correlation coefficient (r) of 0.76 (Supplementary Fig. 9)57. Furthermore, the adjoint model performed well when validated against forward sensitivities calculated using the finite difference and complex variable methods48.

CMAQ Adjoint analysis for health impacts

The adjoint model calculates the sensitivity of all sources by propagating the associated forcing to the receptors backward through time and space for a defined scalar cost function of the concentration field. Based on the adjoint equations derived from the atmospheric diffusion equations, the forcing term can be computed as the gradient of the cost function with respect to concentrations48,58. In this study, we defined a cost function, J, which represents the nationwide premature death attributable to long-term ambient PM2.5 exposure in China in 2017 as shown in Eqs. (1)–(3), following the integrated exposure-response (IER) models in global burden of disease7.

$$J={\sum }_{i,g}{B}_{i}\cdot po{p}_{g}\cdot \left[1-R{R}_{i}{({z}_{g})}^{-1}\right]$$
(1)
$$R{R}_{i}{({z}_{g})}^{-1}=1+{\alpha }_{i}\cdot \left(1-{e}^{-{\beta }_{i}\cdot {z}_{g}^{{\gamma }_{i}}}\right)$$
(2)
$${z}_{g}=\,\max ({z}_{g}-{z}_{cf},0)$$
(3)

where Bi is the regional background mortality rate for each of the five disease-burden causes, including ischemic heart disease, cerebrovascular disease (stroke), chronic obstructive pulmonary disease, and lung cancer for adults older than 25 years and acute lower respiratory infections for children under five59; popg is the gridded population data in mainland China60; and RRi is the IER function for evaluating the relative risk of disease i, as defined by the parameters αi, βi, and γi, and the counterfactual PM2.5 concentration, zcf7. The background mortality rates and IER model parameters for each cause are listed in Supplementary Table 3. The parameter zg is the annual mean PM2.5 concentration in grid g obtained from the CMAQ 1-year forward simulation.

Highly resolved emission inventory

The AiMa emission inventory categorizes emissions into coarse sectors, including electric, industrial, domestic, transportation, agricultural, solvent usage, fugitive dust, and biomass burning. To quantify the emissions of air pollutants and CO2 in the detailed subsectors, we coupled anthropogenic emissions in the AiMa inventory with the detailed bottom-up emission inventories generated by the Global Emission Modeling System (GEMS). The GEMS inventories feature detailed sector and fuel information and local and updated emission factors61,62. The emissions of each sector were apportioned into subcategories based on the relative contribution of each subcategory derived from the GEMS inventory. Energy generation, industrial, domestic, transportation, and agricultural activities were divided into 6, 24, 15, 5, and 3 subcategories, respectively. The detailed subcategories for each sector are listed in Supplementary Table 1.

Source attribution analysis for health damage

We quantify the health damage attributable to major anthropogenic sectors and their subsectors at a monthly resolution by combining the sectoral air pollution emission amount and the adjoint sensitivity as Eq. (4):

$${M}_{{{{{{\rm{g}}}}}},m,s}={\sum }_{p}\frac{\partial J}{\partial {E}_{p,g,m}}\cdot \Delta {E}_{p,g,m,s}$$
(4)

where \(\frac{\partial J}{\partial {E}_{p,g,m}}\) represents the marginal change in the annual premature death attributable to ambient PM2.5 exposure in China in 2017 per unit change in emissions of air pollutant p in grid g and month m. The total health damage from sector s equals the sum of premature deaths attributable to the emissions of each primary PM2.5 species and its precursors, including OC, EC, other primary PM2.5, SO2, NOx, and NH3. The contributions from volatile organic compounds emissions were also not discussed because of their minor contributions to PM2.5-related health damage (Supplementary Fig. 10). The monthly averaged adjoint sensitivities for each air pollutant species \((\frac{\partial J}{\partial {E}_{p,g,m}})\) were calculated as emission-weighted averages from the three-dimensional daily adjoint sensitivity outputs from the CMAQ Adjoint.

Monetized assessment and integrated costs

The social cost of PM2.5 exposure-related health damage can be evaluated as the monetized value of premature deaths based on the economic VSL. The VSL in China in 2017 (VSLc,2017) was calculated using the benefit-transfer approach as Eq. (5):

$$VS{L}_{c,2017}=VS{L}_{OECD}\cdot {\left(\frac{{Y}_{c,2017}}{{Y}_{OECD}}\right)}^{e}$$
(5)

where VSLOECD is the baseline VSL estimated for OECD countries, which equals US$3.83 million; YOECD is the average GDP per capita for the corresponding OECD countries in 201163; Yc,2017 is the GDP per capita of China in 2017 (adjusted to 2011 US dollars at purchasing power rates); and e represents the income elasticity of the VSL and is assumed to be 1.2, as suggested by the World Bank63. For ethical reasons, the uniform VSLc,2017 was applied to all grids, and monetized health impacts were assessed by multiplying the premature deaths with the VSL. The VSL values derived from the OECD benchmarks represent the lower end of the VSL estimates. The corresponding social cost of CO2-related climate change for each subsector can be evaluated based on the CO2 emission amount and SCC. Global values of the SCC for 2020 were employed to measure the direct social benefits, which account for the climate change damage avoided worldwide for a ton of CO2 emission reduction in 2020. The latest SCC estimate at a 2.5% discount rate from the USEPA was adopted and adjusted to the 2011 value, which equals 100 US dollars per ton of CO2 emission64. The SCC estimate is close to a recent expert elicitation estimate (approximately 80–100 US dollars when outliers were trimmed)65. We further evaluate the integrated costs as the sum of the social costs of health damage and climate change.

The impacts of VSL and SCC uncertainties on the monetized assessment are evaluated by perturbing the adopted values. A high estimate of VSL was calculated using the baseline VSL suggested by the USEPA, the average GDP per capita for the U.S., and an income elasticity of 0.566. The VSL was calculated to be 4.36 million US dollars in 2011. Another VSL estimate based on a contingent valuation study in six representative cities in China (0.66 million in 2011 US dollars) is viewed as a low estimate of the VSL67. The SCC is perturbed from 10 to 1000 US dollars per ton of CO2 emission64 for uncertainty assessment.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.