Introduction

Enhanced concentrations of ozone (O3) and/or fine particulate matter (PM2.5, particles with aerodynamic diameters <2.5 μm) adversely affect human health1,2,3, and have been a wide concern in densely populated megacities4,5,6. They frequently co-occur during summer (June-August), which is due in part to stagnant meteorological conditions accompanied with high solar radiation and temperature under which high concentrations of nitrogen oxides (NOx) and volatile organic compounds (VOCs) enhance PM2.5 and O3 formation1,4.

Previous studies have demonstrated a positive relationship between the maximum daily 8-h average (MDA8) O3 and the daily 24-h average (hereafter: DA24 in this study for simplification) PM2.5 concentrations in polluted regions during summertime5,6. A maximum turning point (MTP) in the relationship of MDA8 O3 versus DA24 PM2.5 was observed at about 50~60 µg m-3 of DA24 PM2.5 for Chinese megacity-clusters6. The MDA8 O3 was linearly and positively correlated with DA24 PM2.5 when DA24 PM2.5 below MTP while it remained relatively stable despite the increasing PM2.5 above the MTP. These changes have been attributed to the scavenging of hydroperoxyl (HO2) and/or nitrate radicals (NO3) by high concentrations of PM2.5 that inhibited the photochemical production of O37,8,9.

New York City (NYC) and Beijing are two megacities that have been extensively studied during the last two decades10,11,12. Strict emission control policies have been implemented in NYC since the 1970s with amendments in 1990 (https://www.epa.gov/clean-air-act-overview) and in Beijing from 201313, leading to substantial decreases in PM2.5. Maximum O3 concentrations have reduced more slowly than PM2.5 in NYC14, while they have increased in Beijing15,16,17,18. This discrepancy between PM2.5 and O3 following the imposition of emission controls at these two locations requires further investigation. To explore these, we analyzed the relationship of MDA8 O3 and DA24 PM2.5 by using 19 years of surface measurements in NYC and 6 years of measurements in Beijing, along with related aerosol chemical composition measurements. The features of the O3-PM2.5 relationships and their responses to emission controls in the two megacities are compared, empirically fit using a non-linear power function, and used as a basis for model calculations aimed at developing a future strategy for controlling PM2.5 and O3 together in Beijing and other Chinese megacity-clusters.

Results

Dependence of the NYC O3–PM2.5 relationship on aerosol composition

Based on the magnitude of DA24 PM2.5 concentration, the summertime periods of 2001-2019 were separated into 4 subperiods (Fig. 1a, SPNY1: 2001-2003, SPNY2: 2004-2008; SPNY3: 2009–2013, SPNY4: 2014-2019. See Data availability for the data sources). The standard deviation of the annual summertime average DA24 PM2.5 in each period was below 1 µg m-3. The relationship between MDA8 O3 and DA24 PM2.5 for each subperiod was developed (Fig. 1b) following the approach in Li et al.6 and Buysse et al.9 with the PM2.5 data being binned in increments of 5 μg m−3, and these relationships were fitted using a non-linear function including (1) a positive linear part reflecting their co-occurrence mainly due to their common precursors, e.g., NOx and VOCs19 and (2) a negative power function part reflecting the O3 formation suppression by PM2.5 through the uptake of HO2/NO2 by PM2.57,8,9 and the reduced photolysis rates with PM2.5 increasing20 (Method 1). As shown in Fig. 1b, the O3-PM2.5 linear slope increased from around 2.0 during SPNY1 and SPNY2 to ~3.86 during SPNY4, which corresponds to lower PM2.5 mass concentration at the same O3 pollution level comparing SPNY4 to SPNY1/SPNY2, and also matched the discrepancy of the trends MDA8 O3 and DA24 PM2.5 extreme concentrations (the top5%) of each year (Supplementary Fig. 1a). The extreme concentrations decreased from 2001 to 2019 at a rate of 1.1 ppb yr-1 for MDA8 O3 and 1.9 µg m-3 yr-1 for DA24 PM2.5 (Supplementary Fig. 1a), corresponding to a total reduction of 22% and 62% for O3 and PM2.5, respectively. A larger O3-PM2.5 linear slope indicates a weaker control effect for O3 than for PM2.5.

Fig. 1: O3 and PM2.5 in NYC.
figure 1

a The time series of the annual summertime average DA24 PM2.5 in NYC for each subperiod specified in the main text; b NYC O3-PM2.5 relationship for the different subperiods with the non-linear fitting; c the aerosol mass fraction for each subperiod in NYC based on the AMS measurement from the representative year (Black dash box indicates SAP compounds. The number in each column indicates the SAP mass fraction); d the VOC, NOx, SO2, and primary PM2.5 emissions of NY state in 2001, 2005, 2011, 2017.

Given that sulfate related inorganic and primary organic chemical components of PM2.5 (e.g., sulfate, ammonium from ammonium sulfate, primary organic aerosol, hereafter SAP as the sum of these three components) do not share the NOx and VOCs precursors with O3, we divided the PM2.5 composition into two groups: (1) the SAP, and (2) the non-SAP - with NOx and VOCs as precursors (non-SAP: the sum of secondary organic aerosol (SOA), nitrate and nitrate-related-ammonium). The measured historical PM1 chemical composition in the NYC metropolitan area (Method 2) shows a reduced mass fraction of SAP from 51% in SPNY1 to 29% in SPNY4 (Fig. 1c), consistent with the increased O3-PM2.5 linear slope (Fig. 1b). By excluding the influence of this change in SAP mass fraction, the differences in O3-PM2.5 slopes among the four subperiods were much smaller (Supplementary Fig. 1b). This demonstrates that the reduced SAP mass fraction can explain the increased O3-PM2.5 slope in NYC during the last two decades. The increased fraction of NOx + VOCs in the total emissions (PM2.5, SO2, VOCs and NOx) from 69% in 2001 to 85% in 2017 (Fig. 1d) means higher NOx and VOCs emissions per ton of total emissions, which will lead to the relatively high concentration of the non-SAP and O3 precursors (NOx and VOCs). Greater relative levels of these precursors will contribute to relatively more formation of non-SAP and O3, and therefore relatively less formation of SAP, resulting in a reduced SAP mass fraction and also higher O3-PM2.5 slopes.

The power function coefficient also increased over the past 19 years, from 0.08 in SPNY1 to 0.23 in SPNY4, indicating the O3 formation suppression by PM2.5 increased for the same PM2.5 level from SPNY1 to SPNY4. It should be noted that there were only four points for the SPNY4 fit in Fig. 1b, which could affect the fitted power function coefficient of 0.23. The accuracy of this coefficient was verified for this period by redoing the fitting using PM2.5 mass increments of 2 μg m−3 (0.23, Supplementary Fig. 1c). The larger fitting coefficient may imply a higher deposition rate of O3, reactive VOCs, HO2, and/or NOx onto PM2.5 with a higher non-SAP compared to the same mass concentration with a higher SAP fraction. However, due to the sparsity of detailed measurements and/or directly related confirming studies, it is hard to verify these hypotheses. Due to the simplicity of the non-linear fitting, the empirical power function coefficient determined here may not totally represent the O3 suppression effect. More studies are warranted for verifying this phenomenon and exploring these explanations further.

Increased O3–PM2.5 relationship in Beijing

PM2.5 pollution in Beijing has significantly improved since the clean air action in 2013, with strict emission controls mainly for SO2 and primary particle emissions in the Beijing-Tianjin-Hebei (BTH) region (Fig. 2a; see Supplementary Fig. 2 for the location of BTH), yet the region continues to experience elevated O3 pollution15,16,17,18. In addition to the fact that emissions of NOx and VOCs have changed less than PM2.5 and SO2 (a reduction of 12% for NOx, 0% for VOCs, 28% for PM2.5, and 39% for SO2 comparing 2016 to 2014, Fig. 2a) and remain the major contributor for continued high O3 concentrations16,17,21, some alternative explanations for O3 enhancement include (1) meteorological variability, including elevated temperature, reduced relative humidity (RH) and cloud cover, etc15,17,22, (2) higher HO2 concentrations7 and/or increased photolysis rates23,24,25 due to the reduction of particle concentration, and (3) reduced nitrogen oxides (NOx) emissions26. Meanwhile, these emission control policies may also result in a changed summertime O3-PM2.5 relationship such as observed in NYC, which is the focus of this study. To test for this relationship, we separated the time period for Beijing with existing MDA8 O3 and DA24 PM2.5 data (2014-2019) into two subperiods based on the PM2.5 concentration – subperiod 1 (hereafter SPBJ1, 2014-2016) and subperiod 2 (hereafter SPBJ2, 2017-2019, BJ for Beijing), and the standard deviation of the annual summertime average DA24 PM2.5 in each period was below 8 µg m-3 (Fig. 2b). For each O3-PM2.5 relationship during SPBJ1 or SPBJ2 (Fig. 2c), in contrast to NYC, there was a fitted MTP with PM2.5 mass concentration of about 140 µg m-3 for SPBJ1 (about 83 µg m-3 for SPBJ2). However, the existence of the MTP and its location greatly depended on the non-linear fitting, and further studies based on more model simulations and surface observations will help to verify the MTP and reduce the uncertainty of its location caused by the simplicity of the non-linear fitting in this study. Comparing the O3-PM2.5 relationship of Beijing between SPBJ1 vs. SPBJ2, consistent with the NYC pattern, (1) the O3-PM2.5 linear slope increased from SPBJ1 to SPBJ2, (2) the power function coefficient increased from 0.02 in SPBJ1 to 0.05 SPBJ2 indicating the enhanced O3 suppression by PM2.5, and (3) the O3-PM2.5 relationship in SPBJ1 and SPBJ2 were similar after adjusting for the variability of measured SAP mass fraction in Beijing (Fig. 2d for aerosol composition mass fraction, Supplementary Fig. 3 for O3-PM2.5 non-SAP relationship, and Method 2 for the aerosol composition measurements). The enhanced (VOCs + NOx) emission fraction due to the control policies of SPBJ2 is the dominant reason for the increased O3-PM2.5 linear slopes (Fig. 2a), as in NYC. Considering Beijing extreme air pollution episodes as frequently being influenced by regional transport from BTH27, these O3-PM2.5 relationships in Beijing also represent a regional phenomenon in the BTH region (Fig. 2e) and reflect a complicating consideration in the current regional emission abatements with greater SO2 and PM2.5 reduction.

Fig. 2: O3 and PM2.5, and related emissions in Beijing and BTH.
figure 2

a The 2014 and 2016 MEIC anthropogenic emissions in BTH; b the time series of the annual summertime average DA24 PM2.5 in Beijing for each subperiod specified in the main text; c the O3-PM2.5 relationship of Beijing for SP1 and SP2 with the non-linear fitting; d the averaged aerosol mass fraction for each subperiod in Beijing. (Black dash box indicates SAP compounds. The number in each column indicates the SAP mass fraction); e The O3 vs PM2.5 relationship of main urban cities in BTH for SP1 and SP2 with the non-linear fitting (The error bar indicates 25-75% data range); f The O3 vs PM2.5 relationship for different simultaneous emission abatement scenarios based on CMAQ simulations with NOx and HO2 uptake for 2017.

Regional equal percentage emission reductions

Beijing’s PM2.5 maximums during SPBJ2 are located near the MTP for this period. This implies that it is possible to reduce O3 and PM2.5 together following their relationship line (Fig. 2c, the red line) without encountering the O3 enhancement that may occur with PM2.5 reduction alone. Community Multiscale Air Quality (CMAQ) model simulations (Fig. 2f and Supplementary Fig. 4) verified this possibility by keeping the fraction of each specie emission (PM2.5, SO2, VOCs and NOx) constant while imposing emission reductions, a strategy named “regional equal percentage emission reductions”, which also avoids further increased O3-PM2.5 linear slope observed under the controls targeting mainly SO2 and PM2.5 emissions. As mentioned above, the increased O3-PM2.5 linear slope indicates less O3 reduction than PM2.5 as in NYC. Under regional equal percentage emission reductions, the required BTH regional emission reductions based on the 2019 annual emissions were about (1) 42% to achieve the 2019 Beijing top5% DA24 PM2.5 reaching the concentration levels of the 2001 NYC top5% DA24 PM2.5 (39 µg m-3, hereafter “Goal 1”), (2) 53% to reach the MDA8 O3 concentration below China’s O3 standards (75 ppb, which corresponds to DA24 PM2.5 about 29 µg m-3 for Beijing, hereafter “Goal 2”), and (3) 70% to reach the 2019 NYC DA24 PM2.5 concentration levels (15 µg m-3, hereafter “Goal 3”) (Method 4), respectively.

The emission related O3-PM2.5 relationship variations were also observed in two other Chinese megacity clusters, namely the Yangtze River Delta (YRD) and the Pearl River Delta (PRD) (Fig. 3; see Supplementary Fig. 2 for the locations). The O3-PM2.5 linear slope of YRD increased in recent years (Fig. 3a), consistent with the larger reductions in SO2 and primary PM2.5 than non-SAP species (Fig. 3b), with the enhanced O3 formation suppression by PM2.5. To reach the above three goals in Shanghai, the emissions in YRD need to be reduced by 28%, 32%, and 59%, respectively on the basis of the 2019 YRD annual emissions and the Shanghai top5% DA24 PM2.5 (Method 4). In comparison, the O3-PM2.5 linear slope of PRD was only slightly increased (Fig. 3c), due to the small changes in emission ratios of SO2 and PM2.5 (Fig. 3d), and the PM2.5 suppression effect stayed the same. Meanwhile, the DA24 PM2.5 extreme concentration in Guangzhou (representative of PRD) in 2019 was 36.2 µg m-3, which was already below the Goal 1 concentration level. However, additional regional and synchronous emission reductions in PRD by 14% and 69% (Method 4), respectively are needed to meet the other two goals (“Goal 2 and 3”).

Fig. 3: O3 and PM2.5, and related emissions in YRD and PRD.
figure 3

a The O3 vs PM2.5 relationship of main urban cities in YRD for SP1 (2014–2016) and SP2 (2017–2019) with the non-linear fitting (The error bar indicates 25–75% data range); b the 2014 and 2016 MEIC anthropogenic emissions in YRD (including provinces of Shanghai, Jiangsu and Zhejiang); c the O3 vs PM2.5 relationship of main urban cities in PRD for SP1 (2014–2016) and SP2 (2017–2019) with the non-linear fitting (The error bar indicates 25–75% data range); d the 2014 and 2016 MEIC anthropogenic emissions in PRD (including Guangzhou province).

Discussion

Under current emission policies for NYC and Beijing with larger reductions in the emissions of SO2 and PM2.5 than those of VOCs and NOx, we obtained an increased O3-PM2.5 linear slope due to the reduced SAP mass fraction, indicating a weaker O3 reduction than for PM2.5. Additional contribution of the increased O3-PM2.5 linear slope can also come from the higher NOx emission reduction than VOCs (a reduction of 12% for NOx, 0% for VOCs comparing 2016 to 2014, Fig. 2a), a period during which Beijing was a VOC-limited regime26. Regional equal percentage emission reductions could be an achievable way based on CMAQ simulation, to avoid further increased O3-PM2.5 linear slope as are expected under the continuation of current control policies in BTH. We suggest a 42% equal percentage reduction of the BTH emissions as a goal for the 2019 Beijing top5% DA24 PM2.5 dropping to the 2001 NYC top5% DA24 PM2.5 concentration level (39 µg m-3). Other equal percentage emission controls are needed for other regions to meet pollution reduction goals. We note that equal percentage emission reductions represent only one possible control scenario, and this may not represent an optimal solution Further model simulations with different combinations of NOx and VOCs emission reductions should be used for future studies to explore the response of the O3-PM2.5 relationship to (1) a range of scenarios, (2) a more accurate non-linear function, and (3) the existence of MTP and its uncertainty, etc.

The suppression of O3 formation has been described here by a power function whose strength depends on the PM2.5 level as shown in Figs. 1 and 2. Mechanistic details of this suppression are not yet known, but there is an indication that it does depend on the chemical composition of the suppressing aerosol. However, it is notable that there seems to have some connection between the variation of the fitted power function coefficient for each city or regions with the O3–NOx–VOC sensitivity. These changes in the fitted power function coefficient during 2014-2019 at BTH (0.02 to 0.06), at YRD (0.02 to 0.07), at PRD (0.06 to 0.08), and during 2001-2019 at NYC (0.08 to 0.23), are consistent with changes in the regime of BTH and YRD transforming from VOC-limited to transitional/weak-VOC-limited, PRD staying at similar transitional/weak-VOC-limited26,28, and NYC from transitional/weak-VOC-limited to the transitional/weak-NOx-limited29. Thus, it is possible that the power function coefficient varies following the O3–NOx–VOC sensitivity, with regions with a power function coefficient of 0.02 in the VOC-limited regime, 0.06 in the transitional/weak-VOC-limited regime, and 0.23 in the transitional/weak-NOx-limited regime. All of these need to be further verified through more detailed studies. Finally, the emission reduction on NOx (66% reduction) and VOCs (65% reduction) from NY state comparing 2017 to 2001 could also provide useful guidance for the Chinese megacity clusters, with focusing on the (1) NOx emission reduction from vehicle emissions, energy generation combustion sources, industrial processes, and (2) VOCs emission reduction from vehicle emissions, solvent utilization, fuel combustion (Supplementary Fig. 5).

Methods

M1. Non-linear fitting of the O3-PM2.5 relationship

Equation (1) was used in this study to fit the non-linear relationships between O3 and PM2.5, which contains (a) a positive linear part to reflect the O3/PM2.5 co-occurrence6 without any interaction between these two, (b) a negative power function part reflecting the O3 formation suppression by PM2.5, i.e., the uptake of HO2/NO2 by PM2.57,8,9 and/or the reduced photolysis rates with PM2.5 increasing20, and (c) a constant. The power function exponent was set to 5/3, based on the considerations of (1) the uptake coefficients of the radicals related to the aerosol surface concentration, proportional to the 2/3 power of PM2.5 mass concentration, and (2) the radical concentrations were simply assumed to relate with the O3 concentration, proportional to the PM2.5 mass concentration as mentioned in (a). We ignore the possibility of very high NO concentrations in O3 formation suppression in this study, based on the linear relationship between MAD8 O3 and its time-related odd oxidant (Ox = O3 + NO2) for NYC and Beijing (Supplementary Fig. 6).

$$O_3 = aPM_{2.5} + b(PM_{2.5})^{5/3} + c$$
(1)

Where O3 is its mixing ratio, PM2.5 is the its mass concentration, a is the slope of the linear part, b is power function coefficient, and c is the constant. a, b, and c could be fitted through the non-linear fitting of the O3-PM2.5 relationship. It should be noted that Eq. (1) only represents a very simplified solution for the non-linear O3-PM2.5 relationship and will cause some uncertainty for the results. Further studies related to the mechanism are warrantied to explore a more accurate function.

M2. Aerosol chemical composition measurements

The Aerodyne Aerosol Mass Spectrometer (AMS) was used to obtain the chemical composition of non-refractory particulate matter <1 μm in diameter (NR-PM1). In this study, we assumed that the aerosol composition of PM1 is similar to PM2.5 based on (1) the current application of AMS for PM1 and PM2.5 chemical composition measurements in Beijing with a similar mass fraction30 and (2) the dominant contribution (>70%) of PM1 to PM2.5 in NYC31,32 with similar contribution (around 0.7) from each main composition (e.g. sulfate, nitrate, ammonium) of PM1 to PM2.5 comparing the measurements from an AMS to by a Particle-into-Liquid Sampler (PILS) coupled with two Metrohm Compact 761 Ion Chromatography (IC) systems (PILS-IC)31. Also, we ignored primary black carbon due to the inability of the standard AMS to detect this species. For NYC, AMS field measurements were made during the summers of 200133, 200931,34, 201135, and 201810,32 in NYC or the surrounding area. The aerosol mass fractions of (1) 2001 measurements were used for the subperiod of 2001-2003, (2) the average of 2001 and 2009 for the subperiod of 2004-2008, (3) 2011 for the subperiod of 2009-2013, and (4) 2018 for the subperiod of 2014-2019. For Beijing, AMS measurements were made at the tower branch of the Institute of Atmospheric Physics, a typical urban site located between the north 3rd and 4th ring road in Beijing36 during 06/07-07/08 in 2014 and 06/01-06/29 in 2017, which were used for the subperiods of 2014-2016 and 2017-2019, respectively. It should be noted that some uncertainties can be caused due to the assumptions used, i.e., the PM2.5 mass fraction based on PM1 measurement, the neglect of primary black carbon, and using a single year for the whole subperiod. Due to the limited measurements, these uncertainties are difficult to fully quantify in this study, and further work could benefit from additional continuous PM2.5 composition and other related measurements. The measured OA organic mass spectra have been applied by the Positive Matrix Factorization (PMF) analysis to separate into different OA factors/subtypes37, such as the oxidized organic aerosols (OOAs) as a surrogate of SOA, and the hydrocarbon-like OA (HOA) as a surrogate of POA38,39.

M3. Community Multiscale Air Quality model (CMAQ)

The Community Multiscale Air Quality Model (CMAQ) version 5.2 was applied in the current study, coupled with SAPRC-07 mechanism and updated AERO6 aerosol module. The updates including the heterogeneous loss of NO2, SO2, glyoxal, and methylglyoxal to form nitrate, sulfate, and SOA. The reactive surface uptake coefficient of NO2, glyoxal, and methylglyoxal followed Ying et al. (2014)40 and Ying et al. (2015)41, respectively and the heterogeneous formation of sulfate from surface-controlled reactive uptake of SO2 followed Hu et al (2016)42. In addition, the uptake of HO2 onto aerosol surfaces was considered, and an uptake coefficient (γHO2) of 0.2 was used based on previous studies6,43,44,45. The γHO2 uptake coefficient is related to the aerosol composition, temperature and RH46, and Tan et al. (2020)47 showed the impact of aerosol HO2 uptake effect is much less if a γHO2 of 0.08 was applied in the model, which highlights the large uncertainty for the aerosol HO2 uptake effect from model simulation using a constant γHO2. Since our model is mainly used to simulate the effects of emission reductions, rather than focusing on explaining the reason for the increased ozone in China as Li et al (2019)7 did, and there is good agreement between the observations and the model simulation, it is reasonable to apply γHO2 of 0.2 for this study. The model was applied to simulate O3 and PM2.5 formation during June-August 2017 using a 36 km × 36 km horizontal domain that covers China and surrounding countries in East Asia. The meteorological fields were generated by the Weather Research and Forecasting Model (WRF v4.0.) with a 3D nudging of winds, temperature, and water vapor above the PBL based on the NCEP ADP Global Upper Air Observational Weather Data (https://rda.ucar.edu/datasets/ds351.0/#!access). For anthropogenic emissions, the monthly Multi-resolution Emission Inventory for China (MEICv1.3, https://www.meicmodel.org) and the Regional Emission inventory in Asia (REASv3.1, https://www.nies.go.jp/REAS/) were applied to China and the rest of the domain, respectively. The resolution of both inventories was 0.25°x0.25°. The 2016 MEIC anthropogenic emissions was used for 2017 summertime simulation. Biogenic emissions were generated by the Model for Emissions of Gases and Aerosols from Nature (MEGAN) v2.1, with the leaf area index (LAI) from the 8-day Moderate Resolution Imaging Spectroradiometer (MODIS) LAI product (MOD15A2) and the plant function types (PFTs) from the Global Community Land Model (CLM 3.0). Open burning emissions were based on the Fire Inventory from NCAR (FINN). Dust and sea salt emissions were generated via inline processing during CMAQ simulations. Lightning-induced NOx production was not included in the current study. Initial boundary conditions were based on the default vertical distribution of concentrations provided by CMAQ that represent clean continental conditions with a fixed background O3 concentration ranges from 30 to 70 ppb. The first 3 days were taken as spin-up days and results were excluded from the analysis. Three days were used based on the consideration that a spin-up time of 48 h would be enough for the chemical components study in the planetary boundary layer48. The comparison between the CMAQ simulations and observations for the 24 cities in the BTH and its nearby region are shown in Supplementary Figs. 79. In general, the model can well capture the temporal variations of both O3 and PM2.5 in all the cities, with the performance statistics conforming to the recommended benchmarks by Emery et al. (2017)49 as shown in Supplementary Table 1. Besides the base case with the default emissions of 2016, other cases were designed to represent the possible changes of anthropogenic emissions in the future, including proportional reductions in all the anthropogenic sectors by 25%, 50%, and 75%, respectively.

M4. Synchronous emission reductions estimation

The relationship of top5% DA24 PM2.5 concentration with an emission reduction ratio of Beijing (71.6 µg m−3/100% MEIC reduction, Supplementary Fig. 10a) and other cities (75.4 µg m−3/100% MEIC reduction for Shanghai and 34.3 µg m−3/100% MEIC reduction for Guangzhou, Supplementary Fig. 10b, c) from the model simulation was used to estimate the synchronous emission abatements. Meanwhile, the ratios of observed top5% DA24 PM2.5 to the simulated values (1.2 for Beijing, 1.0 for Shanghai, and 0.9 for Guangzhou) were used to correct the above estimates, assuming they were within the simulation uncertainty. The corrected estimates were 85.9 µg m-3/100% MEIC reduction for Beijing, 75.4 µg m-3/100% MEIC reduction for Shanghai, and 30.9 µg m-3/100% MEIC reduction for Guangzhou, respectively. The observed 2019 top5% DA24 PM2.5 concentration of Beijing (74.8 µg m-3), Shanghai (59.8 µg m-3) and Guangzhou (36.2 µg m-3) were used as the references. The MDA8 O3 concentration below China’s O3 standards (about 75ppb) related DA PM2.5 concentration for Beijing (about 29 µg m-3), Shanghai (about 36 µg m-3), and Guangzhou (about 32 µg m-3) were estimated based on the O3-PM2.5 relations from Figs. 2c, 3a and 3c.