Air quality and health co-benefits of China’s carbon dioxide emissions peaking before 2030

Recent evidence shows that carbon emissions in China are likely to peak ahead of 2030. However, the social and economic impacts of such an early carbon peak have rarely been assessed. Here we focus on the economic costs and health benefits of different carbon mitigation pathways, considering both possible socio-economic futures and varying ambitions of climate policies. We find that an early peak before 2030 in line with the 1.5 °C target could avoid ~118,000 and ~614,000 PM2.5 attributable deaths under the Shared Socioeconomic Pathway 1, in 2030 and 2050, respectively. Under the 2 °C target, carbon mitigation costs could be more than offset by health co-benefits in 2050, bringing a net benefit of $393–$3,017 billion (in 2017 USD value). This study not only provides insight into potential health benefits of an early peak in China, but also suggests that similar benefits may result from more ambitious climate targets in other countries.

for the RCPs adopted in this study are listed in Supplementary Table 9.
To characterize socioeconomic challenges to mitigation and adaptation in a

SSP-RCP framework and attainability of scenarios
In fact, to be useful for climate policy analysis, the scenarios should include both climate policies and others that are not directly related to climate. All those policies controlled by non-climate objectives will either have a substantial impact on climate policy related outcomes or be substantially impacted by climate policy itself 6 . The scenario matrix SSP-RCP is defined by combinations of distinct SSPs and climate forcing outcomes (as characterized by the RCPs).
The SSP-RCP framework facilitates the coupling of multiple socioeconomic reference pathways with climate model products using the representative concentration pathways, allowing for improved assessment of climate impacts, adaptation and mitigation 6 .
The RCPs and SSPs are paired by imposing a set of climate policies, called Shared Climate Policy Assumptions (SPAs), on the SSP baseline. Each SSP has its own SPA consistent with the narrative from which that SSP was developed. These SPAs describe the policy environment in both the near and long term. The long-term mitigation target as determined by the long-term forcing in an RCP will be a central part of the quantitative information given in the SPAs. In principle, the target could be specified in a number of ways, ranging from a global temperature target, to a climate forcing target to a cumulative emissions budget for the entire world. It may also include some constraints on the pathway of climate forcing or global emissions 7 .
The set of climate policy assumptions will have strong implications for the outcome of the scenario analysis, including energy structure, carbon trajectories, etc. Then, by comparing scenarios with and without climate policies, the impact of climate policy can be isolated from the other factors that are changing simultaneously (e.g., population, income, income distribution, land-use).
Studies have explored the feasibility of limiting the end-of-century radiative forcing to 2.6 W/m 2 or 1.9 W/m 2 under the five SSPs, using six integrated assessment models (IAMs) 8 . Some, but not all, SSPs are amenable to pathways to 2°C and 1.5°C. Successful 1.9 W/m 2 scenarios are characterized by a rapid shift away from traditional fossil-fuel use towards large-scale low-carbon energy supplies, reduced energy use, and carbon-dioxide removal.
The SSP1 assumptions include sustainable consumption patterns, low population growth, energy efficiency improving faster than historically, rapid deployment of renewable energy, and global cooperation. With rapid technology diffusion and effective global climate policy from 2020 onwards, all participating models were able to create scenarios in line with an end-of-century forcing target of 1.9 W/m 2 . As for SSP5, the ability to successfully deploy negative emissions technologies and the potential to replace technologies with significant amounts of residual CO2 emissions appear a key determining factor in making it possible for models to counterbalance the otherwise high energy and resource intensity assumed by the SSP5 narrative. Under these assumptions, the SSP5-RCP1.9 scenario has been successfully simulated in GCAM model 9 and REMIND 7 .

CO 2 mitigation pathways
Carbon mitigation can be achieved through a wide portfolio of measures in the energy, industry and land-use sectors, which are the main sources of carbon emissions in China (Supplementary Table 2).
Energy structure adjustment is an important way to reduce carbon emissions. The structural changes include replacement of carbon-intensive fossil fuels by cleaner and renewable energy on the supply side, while refer to energy conservation, efficiency improvements and also the electrification on the demand side. According to Supplementary Fig. 2, changes in primary energy consumption between SSPs baseline scenarios, i.e. SSP_REF, are far less than that brought by the implementation of climate policies. Under both the REF and RCP2.6 scenarios, the proportion of fossil fuels in primary energy consumption would keep above 80% and only decrease slightly after reaching the peak around 2020 or 2030. Though the share of coal and oil will decline slightly, increases in natural gas use would play the major role in the growth of fossil fuels consumption before 2030. Under the RCP2.6 scenarios, the proportion of oil and natural gas will remain stable at ~20% and ~13%, respectively, after 2030, and the proportion of coal will decline rapidly, from 56~65% in 2030 to 23~35% in 2050. Under the RCP1.9 scenarios, China could possibly fulfill the NDC commitment of improving the share of non-fossil fuels to ~20% in the total primary energy consumption in 2030, with more ambitious climate target.
Meanwhile, the preference of cleaner alternatives varies between the SSP scenarios under the climate targets. For the SSP1 scenarios, coal is largely replaced by clean and renewable energy, i.e., solar and wind energy, especially under the 1.5°C target. Yet, the SSP2 and SSP5 scenarios mainly rely on biomass and nuclear energy as non-fossil energy supply to achieve ambitious goals.
Aside from energy structural changes, application of CCS technologies is another critical way to achieve temperature control objectives ( Supplementary Fig. 3 CO2 emissions from fossil and biomass fuels use in power generation, industry and other sources will be collected and stored by CCS technology by 2050.

Mitigation of non-CO 2 GHG emissions
The non-CO2 GHGs in GCAM include methane (CH4), nitrous oxide (N2O) and fluorinated gases (CFCs, HFCs, SF6, PFCs), which are initialized from the CEDS inventory 10 Where, A is activity level (e.g., output of a technology), Fto is the emissions factor for base-year emissions per unit activity, Cprice is the carbon emission price (i.e., the price of carbon emission rights).

Air pollutant emissions in SSP-RCP framework from GCAM
The projections of carbon and air pollutant emission in this study are obtained through  The long-term radiation forcing target in an RCP will be imposed on the SSPs baselines as quantitative constraints, which is the most significant difference between  Table 12). Thus, the phase-out of solid fuels and coal-fired power plants are fully considered from the perspective of carbon mitigation, rather than the control target of air pollution.  25 . We also used the background field and default dust options in WRF-Chem.

Modeling the effect of climate policies on PM 2.5 exposure
In the ideal case, exposure can be estimated using a perturbation in emissions with compared to a reference case to evaluate the impacts of emission change on human exposure 26 . For example, a great number of researches explored the air quality changes and related health impacts caused lockdown measurements during the COVID-19 pandemic, a unique natural experiment 27,28 .
Aside from measured concentrations, air quality modeling studies like WRF-Chem were widely applied among them, to analyze the impact of policies or measures on air pollution 27,29 . We also employ this sensitivity analysis to evaluate the impact of climate policies on the ambient PM2. 5  Similar methods have been widely adopted in earlier studies to quantify the impact of polices addressing climate change 31,32 and other assessments like trade-driven air quality changes 33 . We should note that though it is not perfect, such methodology offers an effective means for the purpose of our analysis.

Model evaluation and uncertainties
The WRF-Chem model applied in this study has been widely used for modeling air The model simulation shows consistent spatial distribution characteristics of station observations, also with the satellite derived PM2.5 product from Ma et al. 36 ( Supplementary Fig. 12). The WRF-Chem simulation would overestimate in some areas, like Sichuan Basin, but still within a reasonable range. Also, the seasonal variation has also been well captured (Supplementary Fig. 13).
In addition, the ground-level satellite-retrieved PM2.5 estimates we applied are

Agriculture and ammonia emissions in future scenarios
In China, both livestock and fertilizer application play significant roles 40 In GCAM, historical agricultural production and consumption of crops and livestock from the Food and Agriculture Organization (FAO) are used for calibration 9 .
As for projections in scenarios, livestock production is closely related to food demand, while crop productions have complex interactions with bioenergy and afforestation.
Here are detailed explanations for livestock husbandry and agricultural planting.

i) Livestock production
In GCAM, per capita food demand is estimated using the evolution of demand As GCAM operates by determining a set of prices that ensure supply is equal to demand for all time steps, the total food demand would determine the total food supply.
Food supply would affect the pattern of agricultural production, and in turn affect related emissions. For livestock, GCAM uses logit sharing approach to make economic choices 44,45 , and calculate the shares for different production systems and feed sources.
Assumed logit parameters are used to dictate substitution between different commodities.

ii) Agricultural planting
Aside from food demand, crop production in the climate mitigation scenarios also has complex interactions with bioenergy and afforestation.
As shown in Supplementary Figure Table 14).
Increasing nitrogen use efficiency of fertilizer application could contribute to reductions in both NH3 and N2O emissions 47,48 . Our results also suggest the importance of nitrogen management in agriculture. 43 / 57

Impacts of ammonia emission change
The PM2.5 concentration of the SSP1_RCP2.6 scenario is slightly higher than that of SSP1_REF in 2030, as shown in Supplementary Fig. 8a. The positive deviations are concentrated eastern and southern China, especially in provinces Henan, Hubei and Hunan, with an approximate scope of 0-2 μg/m 3 . which is mainly caused by the higher ammonia emissions.
As the model configuration are kept definitely same for all scenarios, the total emissions for various scenarios from GCAM is the only variable. According to Fig. 2 and Supplementary Table 15, annual total emissions of carbon and other pollutants are reduced in SSP1_RCP2.6 scenario from SSP1_REF_2030, except ammonia (NH3). The increased NH3 emission is mainly from land use sector, with higher production of energy crops for biomass fuels, under efforts to mitigate 9 . As shown in Supplementary   Fig. 8b, the NH3 emission in SSP1_RCP2.6_2030 is higher than that in SSP1_REF_2030 in eastern China, provinces with large populations and agricultural outputs like Henan, Shandong and Jiangsu 49 . The spatial pattern is similar to the difference of simulated PM2.5 annual mean concentration between the two scenarios ( Supplementary Fig. 8a).
Ammonia is an important precursor for the formation of secondary inorganic particles in the atmosphere. After being discharged into the atmosphere, ammonia gas reacts with sulfuric acid generated by SO2 and nitric acid generated by NOx, and then produce secondary inorganic particles such as ammonium sulfate and ammonium nitrate. Ammonia emission contributes about 29.8% of the annual mean concentration of PM2.5 in China, especially nitrate and ammonium with a high contribution rate of 99.5% while only 4.2% for sulfate 50 . Previous research has found that emissions of NH3 contribute to nearly 30% of China's annual average concentration of PM2.5, and an increasing NH3 emission would hinder PM2.5 mitigation by enhancing ammonium nitrate formation, a key species of secondary PM2.5 51 .
The Primary emission of PM2.5 components in the SSP1_RCP2.6_2030 scenario is lower than the SSP1_REF_2030 scenario, as demonstrated by black carbon in Supplementary Fig. 14a. In contrast, the nitrate and ammonium are about respectively 0.2~1 μg /m 3 and 0.2~0.6 μg/m 3 higher than that in the SSP1_REF_2030 scenario ( Supplementary Fig. 14). In addition, the spatial distributions of the differences in nitrate and ammonium is highly consistent with the difference in ammonia emissions ( Supplementary Fig. 8). The higher annual average PM2.5 concentration in scenario SSP1_RCP2.6_2030 can be considered as the contribution of increase in secondary particulate matter caused by increased ammonia emissions. Here, the co-benefits of carbon mitigation on air pollution is offset by enhanced secondary PM2.5 due to increased ammonia emissions.

Evaluation of scenarios employing measures to control the NH 3
To explore the potential of employing various additional measures to control NH3 for air quality improvement, additional air quality modeling experiments of "reduced NH3 emissions" have been designed.
We choose the scenarios of SSP1_RCP2.6 and SSP1_RCP1.9 in 2030 to evaluate the potential of employing various additional measures to control NH3 (Supplementary   Table 16), since SSP1 is the pathways which could bring great environmental and health benefits for China at an affordable cost in the long-run.
As ammonia emission rate decreases in the SSP1_RCP2.6_cutNH3 scenario, the PM2.5 concentration declines a lot ( Supplementary Fig. 15). As shown in Supplementary

Supplementary Note 4: Decomposition of changes in PM 2.5related deaths
To understand the huge rise of the PM2.5-induced health burden in scenarios from 2030 to 2050, here we calculated the contributions from four key factors that might lead to the elevated health burden estimation, i.e., population size, age structure, PM2.5 exposure and mortality rate unrelated to ambient PM2.5 exposure.

Methodology of the decomposition:
We adopted the "Standardization and decomposition of rates" decomposition method developed by Das Gupta 52 and added better estimates in grid-scale 53 . The decomposition algorithm tests the sensitivities from an accumulative change in the input parameters in assessing PM2.5-related deaths.
The sequence of changing parameters can be described as: 1. the increase in population size (population growth, denoted as "PG"), 2. the change in age structure (population aging, denoted as "PA "), 3. the change in PM2.5 exposure (denoted as "EXP"), 4. the change in the rate of mortality which is unrelated to ambient PM2.5 exposure (i.e., due to changes in access to care, treatment and other risk factors, denoted as "ORF)" calculations for each step of the changed factor illustrated below: baseline mortality rate for non-communicable diseases and lower respiratory infections; RR and PWRR stand for the relative risk and population-weighted relative risk, respectively. The difference between every consecutive step is an estimate of the contribution of each factor: As shown in Supplementary Fig. 16 Where, i and j stand for disease and age group; , , are fitted parameters of the concentration response functions for a given disease provided by Burnett et al. 54 .

ii) IER
We also use the integrated exposure-response (IER) functions to assess the health impacts of PM2.5 55 Where g and j stand for grid cell and age group; M is the number of deaths attributable to PM2.5 exposure; P is the population; I is the national incidence rate for an endpoint; RR is the relative risk. RR is derived from the IER functions: Where , is the RR of health endpoint i and age group j of a given annual average PM2.5 concentration C in grid cell g (μg/m 3 ); C0 is the theoretical minimum risk exposure level, which was set to 2.4-5.9 μg/m 3 according to GBD 2017 study 59 ; , and are parameters of the IER functions for each health endpoint.
We use a set of 1,000 IER parameters (n=1000) for the five health endpoints and age stratum provided by Cohen et al. 56 to estimate uncertainties. We defined the 95% confidence intervals from the 2.5 and 97.5 percentiles of the RR probability distribution, and the mean values of RR were used to represent our results.
To compare with the results of IER, we use GEMM to calculate the mortality attribute to PM2.5 of five health endpoints including COPD, IHD, LC, ALRI and stroke

The impact of the value of statistical life (VSL)
Scaling approach is one of the three types of VSL estimation methods (direct estimation, meta-analysis, and scaling) and has been widely used in developing countries without Supplementary Figure 20 presents different VSL data based on various baseline studies and various coefficients. We find that when using the OECD baseline VSL, the estimated VSL in China were almost one order of magnitude higher than the results using the data from local VSL studies, and the impact of the baseline VSL was far bigger than the influence caused by the economic elasticity coefficient. The estimated VSL in this study is on the same order of magnitude as Liang et al. 67 . Overall, moderate VSLs were estimated based on the calculation method used in this study.

Implicit cost savings in air pollution by climate polices.
The decarbonized energy system under climate policies would indeed bring coreductions in air pollutant emissions, thus induce significant cost savings for air pollution control, or in other words, avoided expenses for air pollution control. It could be an additional co-benefit of climate policies, which has not been covered in our costbenefit analysis.
To assess this part of co-benefit, we estimate the annual avoided abatement cost for air pollution control in each scenario. It was calculated based on the air pollutant reduction ratios from the 2010 level (Supplementary Table 20) and the marginal abatement cost curves from Zhang et al. 68 , which describe the relationship between marginal cost and reduction ratio. The curves were retrieved from a linear programming algorithm model, International Control Cost Estimate Tool, and updated with cost data for applications of 56 types of end-of-pipe technologies and five types of renewable energy in 10 major sectors 68 . Both the end-of-pipe measures and energy structure adjustment were applied in the development of the marginal abatement cost curves.
The annual avoided abatement costs for air pollution control in each scenario are shown in Supplementary Fig. 21, using a discount rate of 5%. The abatement cost reduction for VOCs from 2010 level is rather significant in the climate policy scenarios, which is mainly contributed by the particularly expensive marginal abatement costs as estimated. In addition, the avoided air pollution control costs for primary PM are not estimated here, since primary PM are not included in projected emissions by GCAM, excluding BC and OC which, however, lack corresponding cost estimates. What's more, according to the bottom-up marginal cost curve in Zhang et al. 68 , the magnitude of the primary PM cost curve is at the same order with SO2, much smaller than other 3 pollutants. Thus, the avoided cost here would be underestimated to some extent in this aspect.
Then, the annual cost savings in air pollution control by climate mitigation were estimated by comparing RCPs scenarios with the corresponding REF (Supplementary We have compared the annual cost results with those in existing studies 71 , and found them all in the order of hundreds billion US dollars, for example mainly in