The model used in this study is an online coupled meteorology–chemistry model, the Weather Research and Forecasting model coupled to Chemistry (WRF-Chem v3.7.1), which can simulate meteorological fields and concentrations of gases and aerosols simultaneously (Grell et al., 2005; Skamarock et al., 2008). Figure S1 in the Supplement shows the simulated domain that covers most regions of China with a horizontal resolution of 27 km and grid points of 167 (west–east) × 167 (south–north). The model contains 32 vertical levels extending from the surface to 50 hPa, with the first 16 layers located below 2 km to resolve fine boundary layer processes. The black line in Fig. S1 represents eastern China (22–41.5° N, 102–123° E), and four heavily polluted regions are also selected for analysis, including BTH (36.0–41.5° N, 113–119.5° E), YRD (29.5–32.5° N, 118–122° E), PRD (21–23.5° N, 112–116° E), and SCB (27.5–31.5° N, 102.5–107.5° E).

The National Center for Environmental Prediction (NCEP) Final (FNL) analysis dataset, with a spatial resolution of 1° × 1° and 6 h temporal resolution, is used to provide the meteorological initial and lateral boundary conditions. The chemical initial and boundary conditions for the WRF-Chem model are taken from the outputs of the Community Atmosphere Model with Chemistry (CAM-Chem).

The Carbon Bond Mechanism Z (CBM-Z) is applied as the gas-phase chemical mechanism (Zaveri and Peters, 1999), and the full eight-bin MOSAIC (Model for Simulating Aerosol Interactions and Chemistry) module with aqueous chemistry is used to simulate aerosol evolution (Zaveri et al., 2008). In the MOSAIC module, aerosols are assumed to be internally mixed into eight bins (0.039–0.078, 0.078–0.156, 0.156–0.312, 0.312–0.625, 0.625–1.25, 1.25–2.5, 2.5–5.0, and 5.0–10 µm), and each bin considers all major aerosol species, such as sulfate (SO42-), nitrate (NO3-), ammonium (NH4+), black carbon (BC), organic carbon (OC), and other inorganic mass. The impacts of aerosols on photolysis rates are calculated using the Fast-J scheme (Wild et al., 2000). The following physical parameterizations are used in WRF-Chem. The rapid radiative transfer model for general circulation models (RRTMG) is used to treat both shortwave and longwave radiation in the atmosphere (Iacono et al., 2008). The Purdue Lin microphysics scheme (Lin et al., 1983) and the Grell 3D ensemble scheme (Grell, 1993) are used to describe the cloud microphysical processes and cumulus convective processes. The Noah land surface scheme (Chen and Dudhia, 2001) and the Monin–Obukhov surface scheme (Foken, 2006) are used to simulate land–atmosphere interactions. The planetary boundary layer is characterized by the Yonsei University PBL scheme (Hong et al., 2006). The main physical and chemical schemes used in this study are summarized in Table S1 in the Supplement.

In this study, the Multi-resolution Emission Inventory model for Climate and air pollution research (MEIC; http://www.meicmodel.org/, last access: 21 March 2024) data from 2013 and 2017 are used as the anthropogenic emissions of particles and gases (Zheng et al., 2018). Biogenic emissions are calculated using the online Model of Emissions of Gases and Aerosols from Nature (MEGAN) developed by Guenther et al. (2006).

Seven sensitivity experiments were designed (Table 1). Here are the detailed descriptions. 1.

BASE_17E17M is the baseline experiment coupled with the interactions between aerosol and radiation, which include the impacts of API and ARF. Both the meteorological field and anthropogenic emissions are from 2017.

Figure 1 presents the detailed schematic overview of designed numerical experiments. As shown in Fig. 1, the differences between BASE_17E17M and BASE_13E13M (BASE_17E17M minus BASE_13E13M) represent the changed O3 (ΔO3) due to variations in meteorology and anthropogenic emissions from 2013 to 2017. The differences between BASE_13E17M and BASE_13E13M (BASE_13E17M minus BASE_13E13M) show the impact of changed meteorological conditions on O3 (ΔO3_MET) from 2013 to 2017. The differences between BASE_17E17M and BASE_13E17M (BASE_17E17M minus BASE_13E17M) indicate the impact of anthropogenic emission reductions on O3 (ΔO3_EMI) from 2013 to 2017.

The impacts of aerosol–radiation interaction (ARI) on O3 under different anthropogenic emission scenarios (i.e., strong anthropogenic emission levels in 2013 and weaker anthropogenic emission levels in 2017) can be analyzed as the differences between BASE_17E17M and NOALL_17E17M (BASE_17E17M minus NOALL_17E17M, denoted as ΔO3_ARI17E) and BASE_13E17M and NOALL_13E17M (BASE_13E17M minus NOALL_13E17M, denoted as ΔO3_ARI13E). The term ΔO3_ARI17E means that the impact of ARI on O3 under the conditions of both the meteorological field and anthropogenic emissions is from 2017, and ΔO3_ARI13E means the effect of ARI on O3 with the meteorological field used in 2017 and anthropogenic emissions applied in 2013. In order to quantify the impacts caused by the decreased anthropogenic emissions from 2013 to 2017, the impacts of changed meteorological variables should be removed by fixing the meteorological fields in 2017 using sensitivity experiments. Thus, the impact of weakened ARI due to decreased anthropogenic emissions from 2013 to 2017 on O3 (denoted as ΔO3_ΔARI_EMI) can be quantified using the differences between ΔO3_ARI17E and ΔO3_ARI13E. Similarly, the impacts of weakened API (denoted as ΔO3_ΔAPI_EMI) and ARF (denoted as ΔO3_ΔARF_EMI) due to decreased anthropogenic emissions on O3 can also be estimated using the differences between BASE_17E17M minus NOAPI_17E17M (denoted as ΔO3_API17E) and BASE_13E17M minus NOAPI_13E17M (denoted as ΔO3_API13E), as well as between NOAPI_17E17M minus NOALL_17E17M (denoted as ΔO3_ARF17E) and NOAPI_13E17M minus NOALL_13E17M (denoted as ΔO3_ARF13E). Detailed descriptions can be found in Fig. 1.

Simulation periods are integrated from 30 May to 30 June (denoted as summer) and 29 November to 31 December (denoted as winter) in both 2013 and 2017. To avoid potential deviations caused by long-term model integration, each simulation is re-initialized every 8 d, with the first 40 h as the model spin-up. The complete simulation includes five model cycles. Simulation results from the BASE_17E17M case during summer and winter are used to evaluate the model performance. If not otherwise specified, the time in this paper is the local time, and the synergetic impacts of ARF and API are equal to the impact of ARI (i.e., ARI = ARF + API).

Meteorological observations of temperature (T2), relative humidity (RH2), wind speed (WS10), and wind direction (WD10) provided by the NOAA National Climatic Data Center (https://www.ncei.noaa.gov/, last access: 21 March 2024) were used to validate the meteorological performance of the model. In this study, 353 meteorological stations are selected, and the locations are shown as red dots in Fig. S1. Observed surface PM2.5, O3, and NO2 concentrations in eastern China are obtained from the China National Environmental Monitoring Centre, which can be downloaded from https://quotsoft.net/air/ (last access: 21 March 2024). To ensure the data quality, a single site with at least 500 actual observations during the simulated period was used for model evaluation. A total of 1296 sites, as shown in Fig. 2a, were obtained. Photolysis rates of nitrogen dioxide (NO2) (J[NO2]) measured at the Peking University site (39.99° N, 116.31° E) were also used to evaluate the model performance.

Process analysis techniques, i.e., integrated process rate (IPR) analysis, can be used in grid-based Eulerian models (e.g., WRF-Chem) to obtain contributions of each physical–chemical process to variations in pollutant concentrations. Eulerian models utilize the numerical technique of operator splitting to resolve continuity equations for each species into several simple ordinary differential equations or partial differential equations that only contain the influence of one or two processes (Gipson, 1999).

In order to quantitatively elucidate individual contributions of physical and chemical processes to O3 concentration changes due to weakened ARI, the integrated process rate (IPR) methodology is applied in this study. IPR analysis is an advanced tool to evaluate the key process for O3 concentration variation (Shu et al., 2016; Zhu et al., 2021; Yang et al., 2022). In this study, the IPR analysis tracks the hourly (e.g., one time step) contribution to O3 concentration variation from four main processes: vertical mixing (VMIX), net chemical production (CHEM), horizontal advection (ADVH), and vertical advection (ADVZ). VMIX is initiated by a turbulent process and is closely related to PBL development, which influences O3 vertical gradients. CHEM represents the net O3 chemical production (chemical production minus chemical consumption). ADVH and ADVZ represent transport by wind. We define ADV as the sum of ADVH plus ADVZ.

Figure S2 shows the time series of observed and simulated T2, RH2, WS10, and WD10 averaged over the 353 meteorological stations in China during summer and winter in 2017. Statistical performance of simulated meteorological parameters compared to ground-based observations is shown in Table 2. Simulations track well with observed T2, with a correlation coefficient (R) of 0.99 and 0.92, but underestimate T2 with a mean bias (MB) of -1.0 and -2.0 K in summer and winter, respectively. Simulated RH2 agrees reasonably well with observations, with R of 0.97 and 0.87, and small normalized mean biases (NMBs) are found in summer and winter with values of 3.2 % and 3.5 %, respectively. WS10 is slightly overpredicted, with an MB of 1.6–2.1 m s-1. The R and root mean square error (RMSE) of WS10 are 0.77–0.82 and 1.6–2.1 m s-1. A large bias in wind speed could be partly caused by unresolved topographical features (Jimenez and Dudhia, 2012). The NMB of WD10 ranges from -3.9 % to -2.6 % and R ranges from 0.40 to 0.69. As shown in Fig. S3, the predicted J[NO2] matches the observations well, with R of 0.93–0.94 and NMB of 4.8 %–12.3 %. In general, the simulated meteorological variables agree fairly well with the observations.

Figure 2 shows the spatiotemporal variations in observed and simulated near-surface PM2.5, O3, and NO2 concentrations averaged over eastern China during summer and winter in 2017. As demonstrated in Fig. 2a1 and c1, the WRF-Chem model reproduces the spatial distribution of observed PM2.5 reasonably well, with high values over large city clusters. The predicted O3 concentrations can also reproduce the spatial variation in the observed concentrations (Fig. 2a2 and c2). NO2 is an important precursor of O3 and aerosol; thus a good performance for NO2 is necessary. From Fig. 2a3 and c3, the model can reproduce the spatial distribution of observed NO2 well. Although the distributions of simulated air pollutants are in good agreement with the observations, biases still exist, which may be due to the uncertainty in the emission inventories. Figure 2b1–b3 and d1–d3 show the temporal profiles of observed and simulated surface-layer air pollutants averaged over monitoring sites and the grid cell containing the monitoring site in eastern China. The statistical metrics are also shown in Table 2. As shown in Fig. 2b1 and d1, the model tracks well with the diurnal variation in PM2.5 over eastern China, with R of 0.63 and 0.80. However, the model slightly underestimates the concentrations of PM2.5, with MB of -6.3 and -10.1 µgm-3 in summer and winter, respectively. Simulated O3 agrees reasonably well with observations, with R of 0.90 and 0.86, and small MB is found in summer and winter with values of -0.6 and 2.8 ppb, respectively. The model tracks the daily variation in observed NO2 reasonably well, with R of 0.73 and 0.83. However, the model slightly underestimates the NO2 versus measurements, with MB of -1.5 and -4.5 ppb in summer and winter, respectively. In general, the WRF-Chem model can reproduce the features of observed meteorology and air pollutants well over eastern China.

Figure 3 demonstrates the spatial distribution of changed summer (left) and winter (right) surface PM2.5 (panels a, b) and MDA8 O3 (panels c, d) from 2013 to 2017. As shown in Fig. 3a and b, the observed concentrations of PM2.5 in eastern China are significantly reduced in both summer (-16.2 µgm-3) and winter (-56.0 µgm-3), and these changes can be captured by the model well (-14.3 µgm-3 for summer and -49.8 µgm-3 for winter). Therefore, the model can reproduce the observed decrease in PM2.5 levels from 2013 to 2017. As shown in Fig. 3c and d, the model reproduces the seasonal patterns of changed surface MDA8 O3 reasonably well over eastern China during summer and winter from 2013 to 2017. In summer, both the observations and simulations showed the increased (decreased) MDA8 O3 in YRD (PRD and SCB), while the model could not simulate the positive changes in MDA8 O3 over BTH; the potential reason may be that this study did not consider the effect of changes in heterogeneous aerosol reactions. Li et al. (2019) found that the weakened uptake of HO2 on aerosol surfaces was the main reason for the O3 increase over BTH. In contrast to the changes in summer, observed MDA8 O3 in winter generally increased over eastern China, which can be reproduced well by the model.

The strategy of the clean air action decreased the anthropogenic emission of NOx, but the changes in anthropogenic VOC emissions were non-obvious (Fig. S4), which might influence the sensitive O3 formation regime and the O3 concentration. Figure 4 shows the spatial distributions of changed summer and winter MDA8 O3 concentrations from 2013 to 2017 due to changed anthropogenic emissions alone and due to changed meteorological conditions alone. As shown in Fig. 4a, the concentration of summer MDA8 O3 from 2013 to 2017 increased in city clusters but decreased in rural regions. This discrepancy might be explained by the ozone formation regimes in urban areas typically being VOC-limited, while rural areas are typically NOx-limited during summer (Li et al., 2019; Wang et al., 2019). Contrary to the phenomenon in summer, decreased anthropogenic emissions lead to a uniform increase in winter MDA8 O3 over the whole of eastern China (Fig. 4c). These different spatial variation characteristics in summer and winter may be explained by the different ozone formation regimes in winter (VOC-limited) and summer (NOx-limited) (Fig. S5; Jin and Holloway, 2015). From Fig. 4b and d, the impacts of changed meteorological conditions on MDA8 O3 varied by region, ranging from -24.9 (-14.0) to 17.0 (7.3) ppb in summer (winter). Focusing on the four developed city clusters, compared to 2013, the meteorological conditions in the summer of 2017 promoted the generation of O3 in the YRD region (Fig. 8a3) but suppressed the generation of O3 in the BTH (Fig. 8a2), PRD (Fig. 8a4), and SCB (Fig. 8a5) regions. In PRD and SCB, the changes in MDA8 O3 due to meteorology have an even greater impact than those due to emission changes, which highlights the significant role of meteorology in summer O3 variations.

Figures S6a (S7a) and S6b (S7b) present the spatial distribution of the impacts of ARF, API, and ARI on surface MDA8 O3 concentrations in summer (winter) under different anthropogenic emission conditions in 2017 and 2013. As shown in Fig. S6, summer MDA8 O3 is significantly reduced over eastern China: ARF, API, and ARI decrease the surface MDA8 O3 concentrations by 0.23 (0.59), 1.09 (1.54), and 1.32 (2.13) ppb, respectively, under low (high) anthropogenic emission conditions in 2017 (2013). The changes in MDA8 O3 concentrations due to aerosol–radiation interaction under low emission conditions are weaker than those under high emission conditions. This is because the concentration of aerosols in 2013 was higher than that in 2017, and thus its impact on meteorological conditions and J[NO2] was greater (Fig. S8). As shown in Fig. S7a, 2017 winter MDA8 O3 concentrations were decreased by ARF, API, and ARI by 0.38 ppb (-0.9 %), 1.59 ppb (-4.1 %), and 1.96 ppb (-5.1 %), respectively. Compared to the impacts under relatively high anthropogenic emission conditions in 2013, the reduction in surface MDA8 O3 concentrations caused by ARF, API, and ARI is also greater, with values of 0.62 ppb (-1.6 %), 1.98 ppb (-5.4 %), and 2.59 ppb (-7.1 %), respectively. Both API and ARF reduce O3 concentrations, and the reduction in O3 caused by API is greater than that caused by ARF in both summer and winter.

Further, the significant reduction in PM2.5 due to the clean air action (Fig. S9) led to an increase in O3 concentrations as there were weakened effects of aerosols on O3. Therefore, this study further quantifies the effects of ΔO3_ΔARF_EMI, ΔO3_ΔAPI_EMI, and ΔO3_ΔARI_EMI (ΔO3_ΔARI_EMI =ΔO3_ΔARF_EMI + ΔO3_ΔAPI_EMI) on O3 air quality. As shown in Fig. 5a1–a3, the surface MDA8 O3 in summer increased over most of eastern China due to ΔO3_ΔARF_EMI, ΔO3_ΔAPI_EMI, and ΔO3_ΔARI_EMI. The largest increases in MDA8 O3 concentrations due to ΔO3_ΔARF_EMI and ΔO3_ΔAPI_EMI were found in the four developed city clusters, with increases larger than 4 ppb. Overall, ΔO3_ΔARF_EMI, ΔO3_ΔAPI_EMI, and ΔO3_ΔARI_EMI led to increases in surface MDA8 O3 by 0.36, 0.45, and 0.81 ppb, respectively, averaged over eastern China during summer. As shown in Fig. 5b1–b3, the ΔO3_ΔARF_EMI, ΔO3_ΔAPI_EMI, and ΔO3_ΔARI_EMI can also cause an increase of 0.24, 0.39, and 0.63 ppb, respectively, in winter MDA8 O3 concentrations. In general, weakened aerosol–radiation interaction due to reduced anthropogenic emissions from 2013 to 2017 can exacerbate ozone pollution in both summer and winter.

In order to explore the mechanism of the impacts of ΔO3_ΔARI_EMI on MDA8 O3, we resolve the changed O3 into the contributions from chemical and physical processes. Figure 6 presents the accumulated changes in O3 and each process contribution by ΔO3_ΔAPI_EMI, ΔO3_ΔARF_EMI, and ΔO3_ΔARI_EMI from 09:00 to 16:00 LST during summer and winter. As shown in Fig. 6, enhanced chemical production is the dominant process leading to the increase in O3 concentrations over eastern China and the four city clusters in both summer and winter. The leading factor of enhancement in O3 over BTH is inconsistent with that over eastern China, and the enhancement of O3 concentration over BTH is mainly due to ΔO3_ΔARF_EMI. But the leading factor of enhancement in O3 over SCB is consistent with that over eastern China; the enhancement of O3 concentration is mainly due to ΔO3_ΔAPI_EMI in both summer and winter. Moreover, the enhancement of O3 concentration in BTH, YRD, and PRD is mainly due to ΔO3_ΔARF_EMI during winter, which is opposite to that of eastern China. The leading factors for the increase in O3 concentration in different city clusters are different. The enhancement of O3 concentration in most areas was caused by ΔO3_ΔAPI_EMI, whereas the increase in O3 concentration in the BTH, YRD, and PRD areas is dominated by ΔO3_ΔARF_EMI in winter. In general, the weakened aerosol–radiation interaction caused by emission reduction promotes the chemical production of O3 and increases the O3 concentrations over eastern China in summer and winter.

In order to explore the reason for the increase in O3 chemical production, we further analyzed the variation in HOx (HO + HO2) concentration from 2013 to 2017. As the aerosol concentration decreases, its influence on solar radiation is weakened and photolysis is enhanced, leading to an increase in HOx levels. It can be seen from Fig. S10 that the concentration of HOx increases in both winter and summer. The increase in HOx promotes the conversion of NO to NO2, which leads to the accumulation of O3 concentration.

Figure 7 shows the changed summer and winter surface-layer MDA8 O3 concentrations caused by anthropogenic emission reduction from 2013 to 2017 with (ΔO3_EMI) and without (ΔO3_NOARI) ARI, including the effects of weakened ARI on the effectiveness of emission reduction for O3 air quality (ΔO3_ΔARI_EMI, which is also equal to ΔO3_EMI minus ΔO3_NOARI). As shown in Fig. 7a1 and b1, the surface-layer MDA8 O3 concentrations increased mainly in urban areas during summer and increased uniformly in winter due to anthropogenic emission reduction from 2013 to 2017 without the impact of ARI. When the effect of ARI is considered, the concentrations of MDA8 O3 increased more than when ARI was not taken into account (Fig. 7a2 and b2). The consequences of weakened ARI as a result of anthropogenic emission reduction on MDA8 O3 concentrations are shown in Fig. 7a3 and b3. From Fig. 7a3 and b3 we can see that the concentrations of MDA8 O3 increased in both summer and winter over eastern China. Therefore, ΔO3_ΔARI_EMI creates the superimposed impact on the effectiveness of anthropogenic emission reduction for the increased MDA8 O3 concentrations from 2013 to 2017 over eastern China. However, during summer, the worsened O3 air quality due to weakened ARI can only be found in scattered city clusters (e.g., BTH, YRD, and PRD in Fig. 7a3). During winter, it caused increased MDA8 O3 concentrations over nearly the whole of eastern China (Fig. 7b3).

We also average the observed MDA8 O3 concentrations of monitoring sites in the urban areas and the simulation value for the grid cell containing the monitoring site to further examine the impacts of changed meteorological conditions, anthropogenic emissions, and ARI on O3 levels in densely populated urban areas (Fig. 8). Given that most of the monitoring stations with 5 years of continuous observations are located in urban areas, these monitoring stations and the grid cells containing the monitoring stations can be considered urban areas in this study (Liu and Wang, 2020b). As shown in Fig. 8a1 and b1, the changes in observed MDA8 O3 over urban areas in eastern China from 2013 to 2017 can be captured well by WRF-Chem in both summer and winter. In summer, changed meteorological conditions from 2013 to 2017 had little impact on the variations in MDA8 O3 over the urban areas, while the contribution of emission reductions to increased MDA8 O3 is significant. In winter, changed meteorological conditions were unfavorable for the increase in MDA8 O3 from 2013 to 2017, indicating that the worsened ozone pollution was driven by the changed anthropogenic emissions. What is more, ΔO3_ΔARI_EMI had a significant effect on the increased MDA8 O3 in summer from 2013 to 2017, with a value of +1.77 ppb (87.6 %), but its impacts in winter were smaller at only +0.42 ppb (11.8 %), which is consistent with the results in Li et al. (2021). The increased MDA8 O3 concentration over urban areas in summer caused by O3_ΔARI_EMI in this study was 1.77 ppb compared to the 2.12 ppb increase caused by weakened aerosol heterogeneous reactions quantified by Liu and Wang (2020b). Meanwhile, the contributions of ΔO3_ΔAPI_EMI and ΔO3_ΔARF_EMI to the increase in O3 concentration averaged over urban areas in eastern China were almost the same in summer (0.79 vs. 0.98) and winter (0.20 vs. 0.22). In general, we find that the enhancement of O3 concentrations in both summer and winter was mainly caused by reduced anthropogenic emissions. Furthermore, the contributions of ΔO3_ΔAPI_EMI and ΔO3_ΔARF_EMI to the increases in O3 concentrations from 2013 to 2017 over urban areas were almost the same during summer and winter.

1.

The CBMZ gas-phase chemistry coupled with MOSAIC aerosol module (CBMZ–MOSAIC for short) used in this study does not include secondary organic aerosol (SOA); thus, we applied three additional chemical mechanisms that consider SOA, namely RADM2 gas-phase chemistry coupled with the MADE/SORGAM aerosol module (RADM2–MADE/SORGAM for short), CBMZ gas-phase chemistry coupled with the MADE/SORGAM aerosol module (CBMZ–MADE/SORGAM for short), and MOZART gas-phase chemistry coupled with the MOSAIC aerosol module (MOZART–MOSAIC for short), to test the impact of ARI on O3 with and without SOA for the BASE_17E17M scenario.

Figure S11 shows the temporal variations in observed and simulated PM2.5 and O3 concentrations over eastern China for the three additional chemical mechanisms. Compared with the observed PM2.5 (O3) concentrations, MOZART–MOSAIC showed the best performance in December 2017, with R of 0.73 (0.79) and NMB of -18.7 % (-20.5 %). Therefore, we used this mechanism to further simulate the air pollutant concentrations during the period of June 2017. As shown in Fig. S11 (a4, b4), the temporal variations in observed PM2.5 (O3) can be captured by this mechanism well, with R of 0.56 (0.91) and NMB of -1.7 % (-20.3 %).

Finally, we investigated the effect of ARI on O3 from the results of CBMZ–MOSAIC (the mechanism applied in this paper, which does not include SOA) and MOZART–MOSAIC (this mechanism includes SOA and performs the best compared with RADM2–MADE/SORGAM and CBMZ–MADE/SORGAM). As shown in Fig. S12, summer (winter) MDA8 O3 is significantly reduced over eastern China. ARI reduces the surface MDA8 O3 concentrations by 1.32 (1.96) and 1.85 (1.60) ppb in CBMZ–MOSAIC and MOZART–MOSAIC, respectively. The O3 reductions are of comparable magnitude in these two schemes. Therefore, we can conclude that although the CBMZ–MOSAIC mechanism applied in this paper does not take into account the formation of SOA and its associated effects, the aerosol radiative effects on O3 concentrations, not only in the pattern of spatiotemporal distribution but also in the order of magnitude, are consistent with the results when the SOA simulation mechanism is considered.

As shown in Fig. S13, the mean SOAs simulated by RADM2–MADE/SORGAM, CBMZ–MADE/SORGAM, and MOZART–MOSAIC are 0.29, 0.45, and 0.94 µgm-3, accounting for 3.4 %, 3.8 %, and 4.4 % of PM2.5 concentrations in winter 2017, respectively. From Fig. S14, the mean SOA simulated in MOZART–MOSAIC is 0.90 µgm-3, accounting for 9.1 % of PM2.5 in summer 2017. Model-simulated SOA concentrations are generally underestimated by most current chemical transport models (Zhang et al., 2015; Zhao et al., 2015). The low SOA concentrations simulated by the model can be explained by low emission of biogenic and anthropogenic VOCs (key precursors of SOA), but a thorough investigation of this underestimation is outside the scope of this paper and will be discussed in our future work.